diff options
| -rw-r--r-- | system/alpha/palcode/osfpal.s | 123 | ||||
| -rw-r--r-- | system/alpha/palcode/osfpal_cache_copy.s | 5257 | ||||
| -rw-r--r-- | system/alpha/palcode/osfpal_cache_copy_unaligned.s | 5253 |
3 files changed, 10631 insertions, 2 deletions
diff --git a/system/alpha/palcode/osfpal.s b/system/alpha/palcode/osfpal.s index 655b17cd7..3f42ae143 100644 --- a/system/alpha/palcode/osfpal.s +++ b/system/alpha/palcode/osfpal.s @@ -3143,7 +3143,7 @@ pal_restore_state: // mfpr r31, pt0 // pad last load to icsr write (in case of replay, icsr will be written anyway) //orig - mfpr r31, pt0 // " //orig + mfpr r31, pt0 // "" //orig mfpr r0, icsr // Get icsr //orig //orig ldah r2, <1@<icsr_v_sde-16>>(r31) // Get a one in SHADOW_ENABLE bit location ldah r2, (1<<(ICSR_V_SDE-16))(r31) // Get a one in SHADOW_ENABLE bit location //orig @@ -4707,7 +4707,8 @@ CallPal_OpcDecBE: CALL_PAL_UNPRIV(0x00BF) CallPal_OpcDecBF: - br r31, osfpal_calpal_opcdec + // MODIFIED BY EGH 2/25/04 + br r31, copypal_impl /*======================================================================*/ @@ -5121,3 +5122,121 @@ hw_rei_update_spe_10_: mfpr r9, pt_pcbb // Kernel mode br r31, pal_ic_flush // Pal restriction - must flush Icache if changing ICSR<SPE> #endif + +copypal_impl: + mov r16, r0 + ble r18, finished #if len <=0 we are finished + ldq_u r8, 0(r17) + xor r17, r16, r9 + and r9, 7, r9 + and r16, 7, r10 + bne r9, unaligned + beq r10, aligned + ldq_u r9, 0(r16) + addq r18, r10, r18 + mskqh r8, r17, r8 + mskql r9, r17, r9 + bis r8, r9, r8 +aligned: + subq r18, 1, r10 + bic r10, 7, r10 + and r18, 7, r18 + beq r10, aligned_done +loop: + stq_u r8, 0(r16) + ldq_u r8, 8(r17) + subq r10, 8, r10 + lda r16,8(r16) + lda r17,8(r17) + bne r10, loop +aligned_done: + bne r18, few_left + stq_u r8, 0(r16) + br r31, finished + few_left: + mskql r8, r18, r10 + ldq_u r9, 0(r16) + mskqh r9, r18, r9 + bis r10, r9, r10 + stq_u r10, 0(r16) + br r31, finished +unaligned: + addq r17, r18, r25 + cmpule r18, 8, r9 + bne r9, unaligned_few_left + beq r10, unaligned_dest_aligned + and r16, 7, r10 + subq r31, r10, r10 + addq r10, 8, r10 + ldq_u r9, 7(r17) + extql r8, r17, r8 + extqh r9, r17, r9 + bis r8, r9, r12 + insql r12, r16, r12 + ldq_u r13, 0(r16) + mskql r13, r16, r13 + bis r12, r13, r12 + stq_u r12, 0(r16) + addq r16, r10, r16 + addq r17, r10, r17 + subq r18, r10, r18 + ldq_u r8, 0(r17) +unaligned_dest_aligned: + subq r18, 1, r10 + bic r10, 7, r10 + and r18, 7, r18 + beq r10, unaligned_partial_left +unaligned_loop: + ldq_u r9, 7(r17) + lda r17, 8(r17) + extql r8, r17, r12 + extqh r9, r17, r13 + subq r10, 8, r10 + bis r12, r13, r13 + stq r13, 0(r16) + lda r16, 8(r16) + beq r10, unaligned_second_partial_left + ldq_u r8, 7(r17) + lda r17, 8(r17) + extql r9, r17, r12 + extqh r8, r17, r13 + bis r12, r13, r13 + subq r10, 8, r10 + stq r13, 0(r16) + lda r16, 8(r16) + bne r10, unaligned_loop +unaligned_partial_left: + mov r8, r9 +unaligned_second_partial_left: + ldq_u r8, -1(r25) + extql r9, r17, r9 + extqh r8, r17, r8 + bis r8, r9, r8 + bne r18, few_left + stq_u r8, 0(r16) + br r31, finished +unaligned_few_left: + ldq_u r9, -1(r25) + extql r8, r17, r8 + extqh r9, r17, r9 + bis r8, r9, r8 + insqh r8, r16, r9 + insql r8, r16, r8 + lda r12, -1(r31) + mskql r12, r18, r13 + cmovne r13, r13, r12 + insqh r12, r16, r13 + insql r12, r16, r12 + addq r16, r18, r10 + ldq_u r14, 0(r16) + ldq_u r25, -1(r10) + bic r14, r12, r14 + bic r25, r13, r25 + and r8, r12, r8 + and r9, r13, r9 + bis r8, r14, r8 + bis r9, r25, r9 + stq_u r9, -1(r10) + stq_u r8, 0(r16) +finished: + hw_rei diff --git a/system/alpha/palcode/osfpal_cache_copy.s b/system/alpha/palcode/osfpal_cache_copy.s new file mode 100644 index 000000000..3509b369a --- /dev/null +++ b/system/alpha/palcode/osfpal_cache_copy.s @@ -0,0 +1,5257 @@ +// modified to use the Hudson style "impure.h" instead of ev5_impure.sdl +// since we don't have a mechanism to expand the data structures.... pb Nov/95 + +// build_fixed_image: not sure what means +// real_mm to be replaced during rewrite +// remove_save_state remove_restore_state can be remooved to save space ?? + + +#include "ev5_defs.h" +#include "ev5_impure.h" +#include "ev5_alpha_defs.h" +#include "ev5_paldef.h" +#include "ev5_osfalpha_defs.h" +#include "fromHudsonMacros.h" +#include "fromHudsonOsf.h" +#include "dc21164FromGasSources.h" + +#ifdef SIMOS +#define DEBUGSTORE(c) nop +#else +#define DEBUGSTORE(c) \ + lda r13, c(zero) ; \ + bsr r25, debugstore +#endif + +#define DEBUG_EXC_ADDR()\ + bsr r25, put_exc_addr; \ + DEBUGSTORE(13) ; \ + DEBUGSTORE(10) + +#define egore 0 +#define acore 0 +#define beh_model 0 +#define ev5_p2 1 +#define ev5_p1 0 +#define ldvpte_bug_fix 1 +#define osf_chm_fix 0 + +// Do we want to do this?? pb +#define spe_fix 0 +// Do we want to do this?? pb +#define build_fixed_image 0 + +#define ev5_pass2 +#define enable_p4_fixups 0 +#define osf_svmin 1 +#define enable_physical_console 0 +#define fill_err_hack 0 +#define icflush_on_tbix 0 +#define max_cpuid 1 +#define perfmon_debug 0 +#define rawhide_system 0 +#define rax_mode 0 + + +// This is the fix for the user-mode super page references causing the machine to crash. +#if (spe_fix == 1) && (build_fixed_image==1) +#define hw_rei_spe br r31, hw_rei_update_spe +#else +#define hw_rei_spe hw_rei +#endif + + +// redefine a few of the distribution-code names to match the Hudson gas names. +// opcodes +#define ldqp ldq_p +#define stqp stq_p +#define ldlp ldl_p +#define stlp stl_p + +#define r0 $0 +#define r1 $1 +#define r2 $2 +#define r3 $3 +#define r4 $4 +#define r5 $5 +#define r6 $6 +#define r7 $7 +#define r8 $8 +#define r9 $9 +#define r10 $10 +#define r11 $11 +#define r12 $12 +#define r13 $13 +#define r14 $14 +#define r15 $15 +#define r16 $16 +#define r17 $17 +#define r18 $18 +#define r19 $19 +#define r20 $20 +#define r21 $21 +#define r22 $22 +#define r23 $23 +#define r24 $24 +#define r25 $25 +#define r26 $26 +#define r27 $27 +#define r28 $28 +#define r29 $29 +#define r30 $30 +#define r31 $31 + +// .title "EV5 OSF PAL" +// .ident "V1.18" +// +//**************************************************************************** +//* * +//* Copyright (c) 1992, 1993, 1994, 1995 * +//* by DIGITAL Equipment Corporation, Maynard, Mass. * +//* * +//* This software is furnished under a license and may be used and copied * +//* only in accordance with the terms of such license and with the * +//* inclusion of the above copyright notice. This software or any other * +//* copies thereof may not be provided or otherwise made available to any * +//* other person. No title to and ownership of the software is hereby * +//* transferred. * +//* * +//* The information in this software is subject to change without notice * +//* and should not be construed as a commitment by DIGITAL Equipment * +//* Corporation. * +//* * +//* DIGITAL assumes no responsibility for the use or reliability of its * +//* software on equipment which is not supplied by DIGITAL. * +//* * +//**************************************************************************** + +// .sbttl "Edit History" +//+ +// Who Rev When What +// ------------ --- ----------- -------------------------------- +// DB 0.0 03-Nov-1992 Start +// DB 0.1 28-Dec-1992 add swpctx +// DB 0.2 05-Jan-1993 Bug: PVC found mtpr dtb_CM -> virt ref bug +// DB 0.3 11-Jan-1993 rearrange trap entry points +// DB 0.4 01-Feb-1993 add tbi +// DB 0.5 04-Feb-1993 real MM, kludge reset flow, kludge swppal +// DB 0.6 09-Feb-1993 Bug: several stack pushers used r16 for pc (should be r14) +// DB 0.7 10-Feb-1993 Bug: pushed wrong PC (+8) on CALL_PAL OPCDEC +// Bug: typo on register number for store in wrunique +// Bug: rti to kern uses r16 as scratch +// Bug: callsys saving wrong value in pt_usp +// DB 0.8 16-Feb-1993 PVC: fix possible pt write->read bug in wrkgp, wrusp +// DB 0.9 18-Feb-1993 Bug: invalid_dpte_handler shifted pte twice +// Bug: rti stl_c could corrupt the stack +// Bug: unaligned returning wrong value in r17 (or should be and) +// DB 0.10 19-Feb-1993 Add draina, rd/wrmces, cflush, cserve, interrupt +// DB 0.11 23-Feb-1993 Turn caches on in reset flow +// DB 0.12 10-Mar-1993 Bug: wrong value for icsr for FEN in kern mode flow +// DB 0.13 15-Mar-1993 Bug: wrong value pushed for PC in invalid_dpte_handler if stack push tbmisses +// DB 0.14 23-Mar-1993 Add impure pointer paltemp, reshuffle some other paltemps to match VMS +// DB 0.15 15-Apr-1993 Combine paltemps for WHAMI and MCES +// DB 0.16 12-May-1993 Update reset +// New restriction: no mfpr exc_addr in cycle 1 of call_pal flows +// Bug: in wrmces, not clearing DPC, DSC +// Update swppal +// Add pal bugchecks, pal_save_state, pal_restore_state +// DB 0.17 24-May-1993 Add dfault_in_pal flow; fixup stack builder to have common state for pc/ps. +// New restriction: No hw_rei_stall in 0,1,2 after mtpr itb_asn +// DB 0.18 26-May-1993 PVC fixes +// JM 0.19 01-jul-1993 Bug: OSFPAL_CALPAL_OPCDEC, TRAP_OPCDEC -- move mt exc_addr after stores +// JM 0.20 07-jul-1993 Update cns_ and mchk_ names for impure.mar conversion to .sdl +// Bug: exc_addr was being loaded before stores that could dtb_miss in the following +// routines: TRAP_FEN,FEN_TO_OPCDEC,CALL_PAL_CALLSYS,RTI_TO_KERN +// JM 0.21 26-jul-1993 Bug: move exc_addr load after ALL stores in the following routines: +// TRAP_IACCVIO::,TRAP_OPCDEC::,TRAP_ARITH::,TRAP_FEN:: +// dfault_trap_cont:,fen_to_opcdec:,invalid_dpte_handler: +// osfpal_calpal_opcdec:,CALL_PAL_callsys::,TRAP_UNALIGN:: +// Bugs from PVC: trap_unalign - mt pt0 ->mf pt0 within 2 cycles +// JM 0.22 28-jul-1993 Add WRIPIR instruction +// JM 0.23 05-aug-1993 Bump version number for release +// JM 0.24 11-aug-1993 Bug: call_pal_swpipl - palshadow write -> hw_rei violation +// JM 0.25 09-sep-1993 Disable certain "hidden" pvc checks in call_pals; +// New restriction: No hw_rei_stall in 0,1,2,3,4 after mtpr itb_asn - affects HALT(raxmode), +// and SWPCTX +// JM 0.26 07-oct-1993 Re-implement pal_version +// JM 0.27 12-oct-1993 One more time: change pal_version format to conform to SRM +// JM 0.28 14-oct-1993 Change ic_flush routine to pal_ic_flush +// JM 0.29 19-oct-1993 BUG(?): dfault_in_pal: use exc_addr to check for dtbmiss,itbmiss check instead +// of mm_stat<opcode>. mm_stat contains original opcode, not hw_ld. +// JM 0.30 28-oct-1993 BUG: PVC violation - mf exc_addr in first cycles of call_pal in rti,retsys +// JM 0.31 15-nov-1993 BUG: WRFEN trashing r0 +// JM 0.32 21-nov-1993 BUG: dtb_ldq,itb_ldq (used in dfault_in_pal) not defined when real_mm=0 +// JM 0.33 24-nov-1993 save/restore_state - +// BUG: use ivptbr to restore mvptbr +// BUG: adjust hw_ld/st base/offsets to accomodate 10-bit offset limit +// CHANGE: Load 2 pages into dtb to accomodate compressed logout area/multiprocessors +// JM 0.34 20-dec-1993 BUG: set r11<mode> to kernel for ksnv halt case +// BUG: generate ksnv halt when tb miss on kernel stack accesses +// save exc_addr in r14 for invalid_dpte stack builder +// JM 0.35 30-dec-1993 BUG: PVC violation in trap_arith - mt exc_sum in shadow of store with mf exc_mask in +// the same shadow +// JM 0.36 6-jan-1994 BUG: fen_to_opcdec - savePC should be PC+4, need to save old PS, update new PS +// New palcode restiction: mt icsr<fpe,hwe> --> 3 bubbles to hw_rei --affects wrfen +// JM 0.37 25-jan-1994 BUG: PVC violations in restore_state - mt dc_mode/maf_mode ->mbox instructions +// Hide impure area manipulations in macros +// BUG: PVC violation in save and restore state-- move mt icsr out of shadow of ld/st +// Add some pvc_violate statements +// JM 0.38 1-feb-1994 Changes to save_state: save pt1; don't save r31,f31; update comments to reflect reality; +// Changes to restore_state: restore pt1, icsr; don't restore r31,f31; update comments +// Add code to ensure fen bit set in icsr before ldt +// conditionally compile rax_more_reset out. +// move ldqp,stqp macro definitions to ev5_pal_macros.mar and add .mcall's for them here +// move rax reset stuff to ev5_osf_system_pal.m64 +// JM 0.39 7-feb-1994 Move impure pointer to pal scratch space. Use former pt_impure for bc_ctl shadow +// and performance monitoring bits +// Change to save_state routine to save more iprs. +// JM 0.40 19-feb-1994 Change algorithm in save/restore_state routines; add f31,r31 back in +// JM 0.41 21-feb-1994 Add flags to compile out save/restore state (not needed in some systems) +// remove_save_state,remove_restore_state;fix new pvc violation in save_state +// JM 0.42 22-feb-1994 BUG: save_state overwriting r3 +// JM 0.43 24-feb-1994 BUG: save_state saving wrong icsr +// JM 0.44 28-feb-1994 Remove ic_flush from wr_tbix instructions +// JM 0.45 15-mar-1994 BUG: call_pal_tbi trashes a0 prior to range check (instruction order problem) +// New pal restriction in pal_restore_state: icsr<fpe>->floating instr = 3 bubbles +// Add exc_sum and exc_mask to pal_save_state (not restore) +// JM 0.46 22-apr-1994 Move impure pointer back into paltemp; Move bc_ctl shadow and pmctr_ctl into impure +// area. +// Add performance counter support to swpctx and wrperfmon +// JM 0.47 9-may-1994 Bump version # (for ev5_osf_system_pal.m64 sys_perfmon fix) +// JM 0.48 13-jun-1994 BUG: trap_interrupt --> put new ev5 ipl at 30 for all osfipl6 interrupts +// JM 0.49 8-jul-1994 BUG: In the unlikely (impossible?) event that the branch to pal_pal_bug_check is +// taken in the interrupt flow, stack is pushed twice. +// SWPPAL - update to support ECO 59 to allow 0 as a valid address +// Add itb flush to save/restore state routines +// Change hw_rei to hw_rei_stall in ic_flush routine. Shouldn't be necessary, but +// conforms to itbia restriction. +// Added enable_physical_console flag (for enter/exit console routines only) +// JM 0.50 29-jul-1994 Add code to dfault & invalid_dpte_handler to ignore exceptions on a +// load to r31/f31. changed dfault_fetch_err to dfault_fetch_ldr31_err and +// nmiss_fetch_err to nmiss_fetch_ldr31_err. +// JM 1.00 1-aug-1994 Add pass2 support (swpctx) +// JM 1.01 2-aug-1994 swppal now passes bc_ctl/bc_config in r1/r2 +// JM 1.02 15-sep-1994 BUG: swpctx missing shift of pme bit to correct position in icsr (pass2) +// Moved perfmon code here from system file. +// BUG: pal_perfmon - enable function not saving correct enables when pme not set (pass1) +// JM 1.03 3-oct-1994 Added (pass2 only) code to wrperfmon enable function to look at pme bit. +// JM 1.04 14-oct-1994 BUG: trap_interrupt - ISR read (and saved) before INTID -- INTID can change +// after ISR read, but we won't catch the ISR update. reverse order +// JM 1.05 17-nov-1994 Add code to dismiss UNALIGN trap if LD r31/F31 +// JM 1.06 28-nov-1994 BUG: missing mm_stat shift for store case in trap_unalign (new bug due to "dismiss" code) +// JM 1.07 1-dec-1994 EV5 PASS1,2,3 BUG WORKAROUND: Add flag LDVPTE_BUG_FIX. In DTBMISS_DOUBLE, branch to +// DTBMISS_SINGLE if not in palmode. +// JM 1.08 9-jan-1995 Bump version number for change to EV5_OSF_SYSTEM_PAL.M64 - ei_stat fix in mchk logout frame +// JM 1.09 2-feb-1995 Add flag "spe_fix" and accompanying code to workaround pre-pass4 bug: Disable Ibox +// superpage mode in User mode and re-enable in kernel mode. +// EV5_OSF_SYSTEM_PAL.M64 and EV5_PALDEF.MAR (added pt_misc_v_cm) also changed to support this. +// JM 1.10 24-feb-1995 Set ldvpte_bug_fix regardless of ev5 pass. set default to ev5_p2 +// ES 1.11 10-mar-1995 Add flag "osf_chm_fix" to enable dcache in user mode only to avoid +// cpu bug. +// JM 1.12 17-mar-1995 BUG FIX: Fix F0 corruption problem in pal_restore_state +// ES 1.13 17-mar-1995 Refine osf_chm_fix +// ES 1.14 20-mar-1995 Don't need as many stalls before hw_rei_stall in chm_fix +// ES 1.15 21-mar-1995 Add a stall to avoid a pvc violation in pal_restore_state +// Force pvc checking of exit_console +// ES 1.16 26-apr-1995 In the wrperfmon disable function, correct meaning of R17<2:0> to ctl2,ctl2,ctl0 +// ES 1.17 01-may-1995 In hw_rei_update_spe code, in the osf_chm fix, use bic and bis (self-correcting) +// instead of xor to maintain previous mode in pt_misc +// ES 1.18 14-jul-1995 In wrperfmon enable on pass2, update pmctr even if current process does +// not have pme set. The bits in icsr maintain the master enable state. +// In sys_reset, add icsr<17>=1 for ev56 byte/word eco enable +// +#define vmaj 1 +#define vmin 18 +#define vms_pal 1 +#define osf_pal 2 +#define pal_type osf_pal +#define osfpal_version_l ((pal_type<<16) | (vmaj<<8) | (vmin<<0)) +//- + +// .sbttl "PALtemp register usage" + +//+ +// The EV5 Ibox holds 24 PALtemp registers. This maps the OSF PAL usage +// for these PALtemps: +// +// pt0 local scratch +// pt1 local scratch +// pt2 entUna pt_entUna +// pt3 CPU specific impure area pointer pt_impure +// pt4 memory management temp +// pt5 memory management temp +// pt6 memory management temp +// pt7 entIF pt_entIF +// pt8 intmask pt_intmask +// pt9 entSys pt_entSys +// pt10 +// pt11 entInt pt_entInt +// pt12 entArith pt_entArith +// pt13 reserved for system specific PAL +// pt14 reserved for system specific PAL +// pt15 reserved for system specific PAL +// pt16 MISC: scratch ! WHAMI<7:0> ! 0 0 0 MCES<4:0> pt_misc, pt_whami, pt_mces +// pt17 sysval pt_sysval +// pt18 usp pt_usp +// pt19 ksp pt_ksp +// pt20 PTBR pt_ptbr +// pt21 entMM pt_entMM +// pt22 kgp pt_kgp +// pt23 PCBB pt_pcbb +// +//- + +// .sbttl "PALshadow register usage" +// +//+ +// +// EV5 shadows R8-R14 and R25 when in PALmode and ICSR<shadow_enable> = 1. +// This maps the OSF PAL usage of R8 - R14 and R25: +// +// r8 ITBmiss/DTBmiss scratch +// r9 ITBmiss/DTBmiss scratch +// r10 ITBmiss/DTBmiss scratch +// r11 PS +// r12 local scratch +// r13 local scratch +// r14 local scratch +// r25 local scratch +// +// +//- + +// .sbttl "ALPHA symbol definitions" +// _OSF_PSDEF GLOBAL +// _OSF_PTEDEF GLOBAL +// _OSF_VADEF GLOBAL +// _OSF_PCBDEF GLOBAL +// _OSF_SFDEF GLOBAL +// _OSF_MMCSR_DEF GLOBAL +// _SCBDEF GLOBAL +// _FRMDEF GLOBAL +// _EXSDEF GLOBAL +// _OSF_A0_DEF GLOBAL +// _MCESDEF GLOBAL + +// .sbttl "EV5 symbol definitions" + +// _EV5DEF +// _PALTEMP +// _MM_STAT_DEF +// _EV5_MM +// _EV5_IPLDEF + +// _HALT_CODES GLOBAL +// _MCHK_CODES GLOBAL + +// _PAL_IMPURE +// _PAL_LOGOUT + + + + +// .sbttl "PALcode configuration options" + +// There are a number of options that may be assembled into this version of +// PALcode. They should be adjusted in a prefix assembly file (i.e. do not edit +// the following). The options that can be adjusted cause the resultant PALcode +// to reflect the desired target system. + + +#define osfpal 1 // This is the PALcode for OSF. + +#ifndef rawhide_system + +#define rawhide_system 0 +#endif + + +#ifndef real_mm +// Page table translation vs 1-1 mapping +#define real_mm 1 +#endif + + +#ifndef rax_mode + +#define rax_mode 0 +#endif + +#ifndef egore +// End of reset flow starts a program at 200000(hex). +#define egore 1 +#endif + +#ifndef acore +// End of reset flow starts a program at 40000(hex). +#define acore 0 +#endif + + +// assume acore+egore+rax_mode lt 2 // Assertion checker + +#ifndef beh_model +// EV5 behavioral model specific code +#define beh_model 1 +#endif + +#ifndef init_cbox +// Reset flow init of Bcache and Scache +#define init_cbox 1 +#endif + +#ifndef disable_crd +// Decides whether the reset flow will disable +#define disable_crd 0 +#endif + + // correctable read interrupts via ICSR +#ifndef perfmon_debug +#define perfmon_debug 0 +#endif + +#ifndef icflush_on_tbix +#define icflush_on_tbix 0 +#endif + +#ifndef remove_restore_state +#define remove_restore_state 0 +#endif + +#ifndef remove_save_state +#define remove_save_state 0 +#endif + +#ifndef enable_physical_console +#define enable_physical_console 0 +#endif + +#ifndef ev5_p1 +#define ev5_p1 0 +#endif + +#ifndef ev5_p2 +#define ev5_p2 1 +#endif + +// assume ev5_p1+ev5_p2 eq 1 + +#ifndef ldvpte_bug_fix +#define ldvpte_bug_fix 1 // If set, fix ldvpte bug in dtbmiss_double flow. +#endif + +#ifndef spe_fix +// If set, disable super-page mode in user mode and re-enable +#define spe_fix 0 +#endif + // in kernel. Workaround for cpu bug. +#ifndef build_fixed_image +#define build_fixed_image 0 +#endif + + +#ifndef fill_err_hack +// If set, disable fill_error mode in user mode and re-enable +#define fill_err_hack 0 +#endif + + // in kernel. Workaround for cpu bug. + +// .macro hw_rei_spe +// .iif eq spe_fix, hw_rei +//#if spe_fix != 0 +// +// +//#define hw_rei_chm_count hw_rei_chm_count + 1 +// p4_fixup_label \hw_rei_chm_count +// .iif eq build_fixed_image, br r31, hw_rei_update_spe +// .iif ne build_fixed_image, hw_rei +//#endif +// +// .endm + +// Add flag "osf_chm_fix" to enable dcache in user mode only +// to avoid cpu bug. + +#ifndef osf_chm_fix +// If set, enable D-Cache in +#define osf_chm_fix 0 +#endif + +#if osf_chm_fix != 0 +// user mode only. +#define hw_rei_chm_count 0 +#endif + +#if osf_chm_fix != 0 + +#define hw_rei_stall_chm_count 0 +#endif + +#ifndef enable_p4_fixups + +#define enable_p4_fixups 0 +#endif + + // If set, do EV5 Pass 4 fixups +#if spe_fix == 0 + +#define osf_chm_fix 0 +#endif + +#if spe_fix == 0 + +#define enable_p4_fixups 0 +#endif + + // Only allow fixups if fix enabled + + //Turn off fill_errors and MEM_NEM in user mode +// .macro fill_error_hack ?L10_, ?L20_, ?L30_, ?L40_ +// //save r22,r23,r24 +// stqp r22, 0x150(r31) //add +// stqp r23, 0x158(r31) //contents +// stqp r24, 0x160(r31) //bit mask +// +// lda r22, 0x82(r31) +// ldah r22, 0x8740(r22) +// sll r22, 8, r22 +// ldlp r23, 0x80(r22) // r23 <- contents of CIA_MASK +// bis r23,r31,r23 +// +// lda r24, 0x8(r31) // r24 <- MEM_NEM bit +// beq r10, L10_ // IF user mode (r10<0> == 0) pal mode +// bic r23, r24, r23 // set fillerr_en bit +// br r31, L20_ // ELSE +//L10_: bis r23, r24, r23 // clear fillerr_en bit +//L20_: // ENDIF +// +// stlp r23, 0x80(r22) // write back the CIA_MASK register +// mb +// ldlp r23, 0x80(r22) +// bis r23,r31,r23 +// mb +// +// lda r22, 1(r31) // r22 <- 87.4000.0100 ptr to CIA_CTRL +// ldah r22, 0x8740(r22) +// sll r22, 8, r22 +// ldlp r23, 0(r22) // r23 <- contents of CIA_CTRL +// bis r23,r31,r23 +// +// +// lda r24, 0x400(r31) // r9 <- fillerr_en bit +// beq r10, L30_ // IF user mode (r10<0> == 0) pal mode +// bic r23, r24, r23 // set fillerr_en bit +// br r31, L40_ // ELSE +//L30_: bis r23, r24, r23 // clear fillerr_en bit +//L40_: // ENDIF +// +// stlp r23, 0(r22) // write back the CIA_CTRL register +// mb +// ldlp r23, 0(r22) +// bis r23,r31,r23 +// mb +// +// //restore r22,r23,r24 +// ldqp r22, 0x150(r31) +// ldqp r23, 0x158(r31) +// ldqp r24, 0x160(r31) +// +// .endm + +// multiprocessor support can be enabled for a max of n processors by +// setting the following to the number of processors on the system. +// Note that this is really the max cpuid. + +#ifndef max_cpuid +#define max_cpuid 8 +#endif + +#ifndef osf_svmin // platform specific palcode version number +#define osf_svmin 0 +#endif + + +#define osfpal_version_h ((max_cpuid<<16) | (osf_svmin<<0)) + +// .mcall ldqp // override macro64 definition with macro from library +// .mcall stqp // override macro64 definition with macro from library + + +// .psect _pal,mix +// huh pb pal_base: +// huh pb #define current_block_base . - pal_base + +// .sbttl "RESET - Reset Trap Entry Point" +//+ +// RESET - offset 0000 +// Entry: +// Vectored into via hardware trap on reset, or branched to +// on swppal. +// +// r0 = whami +// r1 = pal_base +// r2 = base of scratch area +// r3 = halt code +// +// +// Function: +// +//- + + .text 0 + . = 0x0000 + .globl Pal_Base +Pal_Base: + HDW_VECTOR(PAL_RESET_ENTRY) +Trap_Reset: + nop +#ifdef SIMOS + /* + * store into r1 + */ + br r1,sys_reset +#else + /* following is a srcmax change */ + + DEBUGSTORE(0x41) + /* The original code jumped using r1 as a linkage register to pass the base + of PALcode to the platform specific code. We use r1 to pass a parameter + from the SROM, so we hardcode the address of Pal_Base in platform.s + */ + br r31, sys_reset +#endif + + // Specify PAL version info as a constant + // at a known location (reset + 8). + + .long osfpal_version_l // <pal_type@16> ! <vmaj@8> ! <vmin@0> + .long osfpal_version_h // <max_cpuid@16> ! <osf_svmin@0> + .long 0 + .long 0 +pal_impure_start: + .quad 0 +pal_debug_ptr: + .quad 0 // reserved for debug pointer ; 20 +#if beh_model == 0 + + +#if enable_p4_fixups != 0 + + + .quad 0 + .long p4_fixup_hw_rei_fixup_table +#endif + +#else + + .quad 0 // + .quad 0 //0x0030 + .quad 0 + .quad 0 //0x0040 + .quad 0 + .quad 0 //0x0050 + .quad 0 + .quad 0 //0x0060 + .quad 0 +pal_enter_cns_address: + .quad 0 //0x0070 -- address to jump to from enter_console + .long <<sys_exit_console-pal_base>+1> //0x0078 -- offset to sys_exit_console (set palmode bit) +#endif + + + + +// .sbttl "IACCVIO- Istream Access Violation Trap Entry Point" + +//+ +// IACCVIO - offset 0080 +// Entry: +// Vectored into via hardware trap on Istream access violation or sign check error on PC. +// +// Function: +// Build stack frame +// a0 <- Faulting VA +// a1 <- MMCSR (1 for ACV) +// a2 <- -1 (for ifetch fault) +// vector via entMM +//- + + HDW_VECTOR(PAL_IACCVIO_ENTRY) +Trap_Iaccvio: + DEBUGSTORE(0x42) + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS + bge r25, TRAP_IACCVIO_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +TRAP_IACCVIO_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bic r14, 3, r16 // pass pc/va as a0 + + stq r17, osfsf_a1(sp) // a1 + or r31, mmcsr_c_acv, r17 // pass mm_csr as a1 + + stq r18, osfsf_a2(sp) // a2 + mfpr r13, pt_entmm // get entry point + + stq r11, osfsf_ps(sp) // save old ps + bis r12, r31, r11 // update ps + + stq r16, osfsf_pc(sp) // save pc + stq r29, osfsf_gp(sp) // save gp + + mtpr r13, exc_addr // load exc_addr with entMM + // 1 cycle to hw_rei + mfpr r29, pt_kgp // get the kgp + + subq r31, 1, r18 // pass flag of istream, as a2 + hw_rei_spe + + +// .sbttl "INTERRUPT- Interrupt Trap Entry Point" + +//+ +// INTERRUPT - offset 0100 +// Entry: +// Vectored into via trap on hardware interrupt +// +// Function: +// check for halt interrupt +// check for passive release (current ipl geq requestor) +// if necessary, switch to kernel mode +// push stack frame, update ps (including current mode and ipl copies), sp, and gp +// pass the interrupt info to the system module +// +//- + + + HDW_VECTOR(PAL_INTERRUPT_ENTRY) +Trap_Interrupt: + mfpr r13, ev5__intid // Fetch level of interruptor + mfpr r25, ev5__isr // Fetch interrupt summary register + + srl r25, isr_v_hlt, r9 // Get HLT bit + mfpr r14, ev5__ipl + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kern + blbs r9, sys_halt_interrupt // halt_interrupt if HLT bit set + + cmple r13, r14, r8 // R8 = 1 if intid .less than or eql. ipl + bne r8, sys_passive_release // Passive release is current rupt is lt or eq ipl + + and r11, osfps_m_mode, r10 // get mode bit + beq r10, TRAP_INTERRUPT_10_ // Skip stack swap in kernel + + mtpr r30, pt_usp // save user stack + mfpr r30, pt_ksp // get kern stack + +TRAP_INTERRUPT_10_: + lda sp, (0-osfsf_c_size)(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r11, osfsf_ps(sp) // save ps + stq r14, osfsf_pc(sp) // save pc + + stq r29, osfsf_gp(sp) // push gp + stq r16, osfsf_a0(sp) // a0 + +// pvc_violate 354 // ps is cleared anyway, if store to stack faults. + mtpr r31, ev5__ps // Set Ibox current mode to kernel + stq r17, osfsf_a1(sp) // a1 + + stq r18, osfsf_a2(sp) // a2 + subq r13, 0x11, r12 // Start to translate from EV5IPL->OSFIPL + + srl r12, 1, r8 // 1d, 1e: ipl 6. 1f: ipl 7. + subq r13, 0x1d, r9 // Check for 1d, 1e, 1f + + cmovge r9, r8, r12 // if .ge. 1d, then take shifted value + bis r12, r31, r11 // set new ps + + mfpr r12, pt_intmask + and r11, osfps_m_ipl, r14 // Isolate just new ipl (not really needed, since all non-ipl bits zeroed already) + +#ifdef SIMOS + /* + * Lance had space problems. We don't. + */ + extbl r12, r14, r14 // Translate new OSFIPL->EV5IPL + mfpr r29, pt_kgp // update gp + mtpr r14, ev5__ipl // load the new IPL into Ibox +#else +// Moved the following three lines to sys_interrupt to make room for debug +// extbl r12, r14, r14 // Translate new OSFIPL->EV5IPL +// mfpr r29, pt_kgp // update gp + +// mtpr r14, ev5__ipl // load the new IPL into Ibox +#endif + br r31, sys_interrupt // Go handle interrupt + + + +// .sbttl "ITBMISS- Istream TBmiss Trap Entry Point" + +//+ +// ITBMISS - offset 0180 +// Entry: +// Vectored into via hardware trap on Istream translation buffer miss. +// +// Function: +// Do a virtual fetch of the PTE, and fill the ITB if the PTE is valid. +// Can trap into DTBMISS_DOUBLE. +// This routine can use the PALshadow registers r8, r9, and r10 +// +//- + + HDW_VECTOR(PAL_ITB_MISS_ENTRY) +Trap_Itbmiss: +#if real_mm == 0 + + + // Simple 1-1 va->pa mapping + + nop // Pad to align to E1 + mfpr r8, exc_addr + + srl r8, page_offset_size_bits, r9 + sll r9, 32, r9 + + lda r9, 0x3301(r9) // Make PTE, V set, all KRE, URE, KWE, UWE + mtpr r9, itb_pte // E1 + + hw_rei_stall // Nital says I don't have to obey shadow wait rule here. +#else + + // Real MM mapping + nop + mfpr r8, ev5__ifault_va_form // Get virtual address of PTE. + + nop + mfpr r10, exc_addr // Get PC of faulting instruction in case of DTBmiss. + +pal_itb_ldq: + ld_vpte r8, 0(r8) // Get PTE, traps to DTBMISS_DOUBLE in case of TBmiss + mtpr r10, exc_addr // Restore exc_address if there was a trap. + + mfpr r31, ev5__va // Unlock VA in case there was a double miss + nop + + and r8, osfpte_m_foe, r25 // Look for FOE set. + blbc r8, invalid_ipte_handler // PTE not valid. + + nop + bne r25, foe_ipte_handler // FOE is set + + nop + mtpr r8, ev5__itb_pte // Ibox remembers the VA, load the PTE into the ITB. + + hw_rei_stall // + +#endif + + + + +// .sbttl "DTBMISS_SINGLE - Dstream Single TBmiss Trap Entry Point" + +//+ +// DTBMISS_SINGLE - offset 0200 +// Entry: +// Vectored into via hardware trap on Dstream single translation buffer miss. +// +// Function: +// Do a virtual fetch of the PTE, and fill the DTB if the PTE is valid. +// Can trap into DTBMISS_DOUBLE. +// This routine can use the PALshadow registers r8, r9, and r10 +//- + + HDW_VECTOR(PAL_DTB_MISS_ENTRY) +Trap_Dtbmiss_Single: +#if real_mm == 0 + // Simple 1-1 va->pa mapping + mfpr r8, va // E0 + srl r8, page_offset_size_bits, r9 + + sll r9, 32, r9 + lda r9, 0x3301(r9) // Make PTE, V set, all KRE, URE, KWE, UWE + + mtpr r9, dtb_pte // E0 + nop // Pad to align to E0 + + + + mtpr r8, dtb_tag // E0 + nop + + nop // Pad tag write + nop + + nop // Pad tag write + nop + + hw_rei +#else + mfpr r8, ev5__va_form // Get virtual address of PTE - 1 cycle delay. E0. + mfpr r10, exc_addr // Get PC of faulting instruction in case of error. E1. + +// DEBUGSTORE(0x45) +// DEBUG_EXC_ADDR() + // Real MM mapping + mfpr r9, ev5__mm_stat // Get read/write bit. E0. + mtpr r10, pt6 // Stash exc_addr away + +pal_dtb_ldq: + ld_vpte r8, 0(r8) // Get PTE, traps to DTBMISS_DOUBLE in case of TBmiss + nop // Pad MF VA + + mfpr r10, ev5__va // Get original faulting VA for TB load. E0. + nop + + mtpr r8, ev5__dtb_pte // Write DTB PTE part. E0. + blbc r8, invalid_dpte_handler // Handle invalid PTE + + mtpr r10, ev5__dtb_tag // Write DTB TAG part, completes DTB load. No virt ref for 3 cycles. + mfpr r10, pt6 + + // Following 2 instructions take 2 cycles + mtpr r10, exc_addr // Return linkage in case we trapped. E1. + mfpr r31, pt0 // Pad the write to dtb_tag + + hw_rei // Done, return +#endif + + + + +// .sbttl "DTBMISS_DOUBLE - Dstream Double TBmiss Trap Entry Point" + +//+ +// DTBMISS_DOUBLE - offset 0280 +// Entry: +// Vectored into via hardware trap on Double TBmiss from single miss flows. +// +// r8 - faulting VA +// r9 - original MMstat +// r10 - original exc_addr (both itb,dtb miss) +// pt6 - original exc_addr (dtb miss flow only) +// VA IPR - locked with original faulting VA +// +// Function: +// Get PTE, if valid load TB and return. +// If not valid then take TNV/ACV exception. +// +// pt4 and pt5 are reserved for this flow. +// +// +//- + + HDW_VECTOR(PAL_DOUBLE_MISS_ENTRY) +Trap_Dtbmiss_double: +#if ldvpte_bug_fix != 0 + mtpr r8, pt4 // save r8 to do exc_addr check + mfpr r8, exc_addr + blbc r8, Trap_Dtbmiss_Single //if not in palmode, should be in the single routine, dummy! + mfpr r8, pt4 // restore r8 +#endif + nop + mtpr r22, pt5 // Get some scratch space. E1. + // Due to virtual scheme, we can skip the first lookup and go + // right to fetch of level 2 PTE + sll r8, (64-((2*page_seg_size_bits)+page_offset_size_bits)), r22 // Clean off upper bits of VA + mtpr r21, pt4 // Get some scratch space. E1. + + srl r22, 61-page_seg_size_bits, r22 // Get Va<seg1>*8 + mfpr r21, pt_ptbr // Get physical address of the page table. + + nop + addq r21, r22, r21 // Index into page table for level 2 PTE. + + sll r8, (64-((1*page_seg_size_bits)+page_offset_size_bits)), r22 // Clean off upper bits of VA + ldqp r21, 0(r21) // Get level 2 PTE (addr<2:0> ignored) + + srl r22, 61-page_seg_size_bits, r22 // Get Va<seg1>*8 + blbc r21, double_pte_inv // Check for Invalid PTE. + + srl r21, 32, r21 // extract PFN from PTE + sll r21, page_offset_size_bits, r21 // get PFN * 2^13 for add to <seg3>*8 + + addq r21, r22, r21 // Index into page table for level 3 PTE. + nop + + ldqp r21, 0(r21) // Get level 3 PTE (addr<2:0> ignored) + blbc r21, double_pte_inv // Check for invalid PTE. + + mtpr r21, ev5__dtb_pte // Write the PTE. E0. + mfpr r22, pt5 // Restore scratch register + + mtpr r8, ev5__dtb_tag // Write the TAG. E0. No virtual references in subsequent 3 cycles. + mfpr r21, pt4 // Restore scratch register + + nop // Pad write to tag. + nop + + nop // Pad write to tag. + nop + + hw_rei + + + +// .sbttl "UNALIGN -- Dstream unalign trap" +//+ +// UNALIGN - offset 0300 +// Entry: +// Vectored into via hardware trap on unaligned Dstream reference. +// +// Function: +// Build stack frame +// a0 <- Faulting VA +// a1 <- Opcode +// a2 <- src/dst register number +// vector via entUna +//- + + HDW_VECTOR(PAL_UNALIGN_ENTRY) +Trap_Unalign: +/* DEBUGSTORE(0x47)*/ + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r8, ev5__mm_stat // Get mmstat --ok to use r8, no tbmiss + mfpr r14, exc_addr // get pc + + srl r8, mm_stat_v_ra, r13 // Shift Ra field to ls bits + blbs r14, pal_pal_bug_check // Bugcheck if unaligned in PAL + + blbs r8, UNALIGN_NO_DISMISS // lsb only set on store or fetch_m + // not set, must be a load + and r13, 0x1F, r8 // isolate ra + + cmpeq r8, 0x1F, r8 // check for r31/F31 + bne r8, dfault_fetch_ldr31_err // if its a load to r31 or f31 -- dismiss the fault + +UNALIGN_NO_DISMISS: + bis r11, r31, r12 // Save PS + bge r25, UNALIGN_NO_DISMISS_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +UNALIGN_NO_DISMISS_10_: + mfpr r25, ev5__va // Unlock VA + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + mtpr r25, pt0 // Stash VA + stq r18, osfsf_a2(sp) // a2 + + stq r11, osfsf_ps(sp) // save old ps + srl r13, mm_stat_v_opcode-mm_stat_v_ra, r25// Isolate opcode + + stq r29, osfsf_gp(sp) // save gp + addq r14, 4, r14 // inc PC past the ld/st + + stq r17, osfsf_a1(sp) // a1 + and r25, mm_stat_m_opcode, r17// Clean opocde for a1 + + stq r16, osfsf_a0(sp) // save regs + mfpr r16, pt0 // a0 <- va/unlock + + stq r14, osfsf_pc(sp) // save pc + mfpr r25, pt_entuna // get entry point + + + bis r12, r31, r11 // update ps + br r31, unalign_trap_cont + + + + +// .sbttl "DFAULT - Dstream Fault Trap Entry Point" + +//+ +// DFAULT - offset 0380 +// Entry: +// Vectored into via hardware trap on dstream fault or sign check error on DVA. +// +// Function: +// Ignore faults on FETCH/FETCH_M +// Check for DFAULT in PAL +// Build stack frame +// a0 <- Faulting VA +// a1 <- MMCSR (1 for ACV, 2 for FOR, 4 for FOW) +// a2 <- R/W +// vector via entMM +// +//- + HDW_VECTOR(PAL_D_FAULT_ENTRY) +Trap_Dfault: +// DEBUGSTORE(0x48) + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r13, ev5__mm_stat // Get mmstat + mfpr r8, exc_addr // get pc, preserve r14 + + srl r13, mm_stat_v_opcode, r9 // Shift opcode field to ls bits + blbs r8, dfault_in_pal + + bis r8, r31, r14 // move exc_addr to correct place + bis r11, r31, r12 // Save PS + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + and r9, mm_stat_m_opcode, r9 // Clean all but opcode + + cmpeq r9, evx_opc_sync, r9 // Is the opcode fetch/fetchm? + bne r9, dfault_fetch_ldr31_err // Yes, dismiss the fault + + //dismiss exception if load to r31/f31 + blbs r13, dfault_no_dismiss // mm_stat<0> set on store or fetchm + + // not a store or fetch, must be a load + srl r13, mm_stat_v_ra, r9 // Shift rnum to low bits + + and r9, 0x1F, r9 // isolate rnum + nop + + cmpeq r9, 0x1F, r9 // Is the rnum r31 or f31? + bne r9, dfault_fetch_ldr31_err // Yes, dismiss the fault + +dfault_no_dismiss: + and r13, 0xf, r13 // Clean extra bits in mm_stat + bge r25, dfault_trap_cont // no stack swap needed if cm=kern + + + mtpr r30, pt_usp // save user stack + bis r31, r31, r12 // Set new PS + + mfpr r30, pt_ksp + br r31, dfault_trap_cont + + + + + +// .sbttl "MCHK - Machine Check Trap Entry Point" + +//+ +// MCHK - offset 0400 +// Entry: +// Vectored into via hardware trap on machine check. +// +// Function: +// +//- + + HDW_VECTOR(PAL_MCHK_ENTRY) +Trap_Mchk: + DEBUGSTORE(0x49) + mtpr r31, ic_flush_ctl // Flush the Icache + br r31, sys_machine_check + + + + +// .sbttl "OPCDEC - Illegal Opcode Trap Entry Point" + +//+ +// OPCDEC - offset 0480 +// Entry: +// Vectored into via hardware trap on illegal opcode. +// +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + + HDW_VECTOR(PAL_OPCDEC_ENTRY) +Trap_Opcdec: + DEBUGSTORE(0x4a) +//simos DEBUG_EXC_ADDR() + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r14, exc_addr // get pc + blbs r14, pal_pal_bug_check // check opcdec in palmode + + bis r11, r31, r12 // Save PS + bge r25, TRAP_OPCDEC_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +TRAP_OPCDEC_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + addq r14, 4, r14 // inc pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_opdec, r16 // set a0 + + stq r11, osfsf_ps(sp) // save old ps + mfpr r13, pt_entif // get entry point + + stq r18, osfsf_a2(sp) // a2 + stq r17, osfsf_a1(sp) // a1 + + stq r29, osfsf_gp(sp) // save gp + stq r14, osfsf_pc(sp) // save pc + + bis r12, r31, r11 // update ps + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei, E1 + + mfpr r29, pt_kgp // get the kgp, E1 + + hw_rei_spe // done, E1 + + + + + + +// .sbttl "ARITH - Arithmetic Exception Trap Entry Point" + +//+ +// ARITH - offset 0500 +// Entry: +// Vectored into via hardware trap on arithmetic excpetion. +// +// Function: +// Build stack frame +// a0 <- exc_sum +// a1 <- exc_mask +// a2 <- unpred +// vector via entArith +// +//- + HDW_VECTOR(PAL_ARITH_ENTRY) +Trap_Arith: + DEBUGSTORE(0x4b) + and r11, osfps_m_mode, r12 // get mode bit + mfpr r31, ev5__va // unlock mbox + + bis r11, r31, r25 // save ps + mfpr r14, exc_addr // get pc + + nop + blbs r14, pal_pal_bug_check // arith trap from PAL + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + beq r12, TRAP_ARITH_10_ // if zero we are in kern now + + bis r31, r31, r25 // set the new ps + mtpr r30, pt_usp // save user stack + + nop + mfpr r30, pt_ksp // get kern stack + +TRAP_ARITH_10_: lda sp, 0-osfsf_c_size(sp) // allocate stack space + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + nop // Pad current mode write and stq + mfpr r13, ev5__exc_sum // get the exc_sum + + mfpr r12, pt_entarith + stq r14, osfsf_pc(sp) // save pc + + stq r17, osfsf_a1(sp) + mfpr r17, ev5__exc_mask // Get exception register mask IPR - no mtpr exc_sum in next cycle + + stq r11, osfsf_ps(sp) // save ps + bis r25, r31, r11 // set new ps + + stq r16, osfsf_a0(sp) // save regs + srl r13, exc_sum_v_swc, r16// shift data to correct position + + stq r18, osfsf_a2(sp) +// pvc_violate 354 // ok, but make sure reads of exc_mask/sum are not in same trap shadow + mtpr r31, ev5__exc_sum // Unlock exc_sum and exc_mask + + stq r29, osfsf_gp(sp) + mtpr r12, exc_addr // Set new PC - 1 bubble to hw_rei - E1 + + mfpr r29, pt_kgp // get the kern gp - E1 + hw_rei_spe // done - E1 + + + + + + +// .sbttl "FEN - Illegal Floating Point Operation Trap Entry Point" + +//+ +// FEN - offset 0580 +// Entry: +// Vectored into via hardware trap on illegal FP op. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + + HDW_VECTOR(PAL_FEN_ENTRY) +Trap_Fen: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r14, exc_addr // get pc + blbs r14, pal_pal_bug_check // check opcdec in palmode + + mfpr r13, ev5__icsr + nop + + bis r11, r31, r12 // Save PS + bge r25, TRAP_FEN_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +TRAP_FEN_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + srl r13, icsr_v_fpe, r25 // Shift FP enable to bit 0 + + + stq r16, osfsf_a0(sp) // save regs + mfpr r13, pt_entif // get entry point + + stq r18, osfsf_a2(sp) // a2 + stq r11, osfsf_ps(sp) // save old ps + + stq r29, osfsf_gp(sp) // save gp + bis r12, r31, r11 // set new ps + + stq r17, osfsf_a1(sp) // a1 + blbs r25,fen_to_opcdec // If FP is enabled, this is really OPCDEC. + + bis r31, osf_a0_fen, r16 // set a0 + stq r14, osfsf_pc(sp) // save pc + + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei -E1 + + mfpr r29, pt_kgp // get the kgp -E1 + + hw_rei_spe // done -E1 + +// FEN trap was taken, but the fault is really opcdec. + ALIGN_BRANCH +fen_to_opcdec: + addq r14, 4, r14 // save PC+4 + bis r31, osf_a0_opdec, r16 // set a0 + + stq r14, osfsf_pc(sp) // save pc + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kgp + hw_rei_spe // done + + + +// .sbttl "Misc handlers" + // Start area for misc code. +//+ +//dfault_trap_cont +// A dfault trap has been taken. The sp has been updated if necessary. +// Push a stack frame a vector via entMM. +// +// Current state: +// r12 - new PS +// r13 - MMstat +// VA - locked +// +//- + ALIGN_BLOCK +dfault_trap_cont: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r25, ev5__va // Fetch VA/unlock + + stq r18, osfsf_a2(sp) // a2 + and r13, 1, r18 // Clean r/w bit for a2 + + stq r16, osfsf_a0(sp) // save regs + bis r25, r31, r16 // a0 <- va + + stq r17, osfsf_a1(sp) // a1 + srl r13, 1, r17 // shift fault bits to right position + + stq r11, osfsf_ps(sp) // save old ps + bis r12, r31, r11 // update ps + + stq r14, osfsf_pc(sp) // save pc + mfpr r25, pt_entmm // get entry point + + stq r29, osfsf_gp(sp) // save gp + cmovlbs r17, 1, r17 // a2. acv overrides fox. + + mtpr r25, exc_addr // load exc_addr with entMM + // 1 cycle to hw_rei + mfpr r29, pt_kgp // get the kgp + + hw_rei_spe // done + +//+ +//unalign_trap_cont +// An unalign trap has been taken. Just need to finish up a few things. +// +// Current state: +// r25 - entUna +// r13 - shifted MMstat +// +//- + ALIGN_BLOCK +unalign_trap_cont: + mtpr r25, exc_addr // load exc_addr with entUna + // 1 cycle to hw_rei + + + mfpr r29, pt_kgp // get the kgp + and r13, mm_stat_m_ra, r18 // Clean Ra for a2 + + hw_rei_spe // done + + + +//+ +// dfault_in_pal +// Dfault trap was taken, exc_addr points to a PAL PC. +// r9 - mmstat<opcode> right justified +// r8 - exception address +// +// These are the cases: +// opcode was STQ -- from a stack builder, KSP not valid halt +// r14 - original exc_addr +// r11 - original PS +// opcode was STL_C -- rti or retsys clear lock_flag by stack write, +// KSP not valid halt +// r11 - original PS +// r14 - original exc_addr +// opcode was LDQ -- retsys or rti stack read, KSP not valid halt +// r11 - original PS +// r14 - original exc_addr +// opcode was HW_LD -- itbmiss or dtbmiss, bugcheck due to fault on page tables +// r10 - original exc_addr +// r11 - original PS +// +// +//- + ALIGN_BLOCK +dfault_in_pal: + DEBUGSTORE(0x50) + bic r8, 3, r8 // Clean PC + mfpr r9, pal_base + + mfpr r31, va // unlock VA +#if real_mm != 0 + // if not real_mm, should never get here from miss flows + + subq r9, r8, r8 // pal_base - offset + + lda r9, pal_itb_ldq-pal_base(r8) + nop + + beq r9, dfault_do_bugcheck + lda r9, pal_dtb_ldq-pal_base(r8) + + beq r9, dfault_do_bugcheck +#endif + +// +// KSP invalid halt case -- +ksp_inval_halt: + DEBUGSTORE(76) + bic r11, osfps_m_mode, r11 // set ps to kernel mode + mtpr r0, pt0 + + mtpr r31, dtb_cm // Make sure that the CM IPRs are all kernel mode + mtpr r31, ips + + mtpr r14, exc_addr // Set PC to instruction that caused trouble +//orig pvc_jsr updpcb, bsr=1 + bsr r0, pal_update_pcb // update the pcb + + lda r0, hlt_c_ksp_inval(r31) // set halt code to hw halt + br r31, sys_enter_console // enter the console + + ALIGN_BRANCH +dfault_do_bugcheck: + bis r10, r31, r14 // bugcheck expects exc_addr in r14 + br r31, pal_pal_bug_check + + + ALIGN_BLOCK +//+ +// dfault_fetch_ldr31_err - ignore faults on fetch(m) and loads to r31/f31 +// On entry - +// r14 - exc_addr +// VA is locked +// +//- +dfault_fetch_ldr31_err: + mtpr r11, ev5__dtb_cm + mtpr r11, ev5__ps // Make sure ps hasn't changed + + mfpr r31, va // unlock the mbox + addq r14, 4, r14 // inc the pc to skip the fetch + + mtpr r14, exc_addr // give ibox new PC + mfpr r31, pt0 // pad exc_addr write + + hw_rei + + + + ALIGN_BLOCK +//+ +// sys_from_kern +// callsys from kernel mode - OS bugcheck machine check +// +//- +sys_from_kern: + mfpr r14, exc_addr // PC points to call_pal + subq r14, 4, r14 + + lda r25, mchk_c_os_bugcheck(r31) // fetch mchk code + br r31, pal_pal_mchk + + +// .sbttl "Continuation of long call_pal flows" + ALIGN_BLOCK +//+ +// wrent_tbl +// Table to write *int in paltemps. +// 4 instructions/entry +// r16 has new value +// +//- +wrent_tbl: +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entint + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entarith + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entmm + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entif + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entuna + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entsys + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + ALIGN_BLOCK +//+ +// tbi_tbl +// Table to do tbi instructions +// 4 instructions per entry +//- +tbi_tbl: + // -2 tbia +//orig pvc_jsr tbi, dest=1 + mtpr r31, ev5__dtb_ia // Flush DTB + mtpr r31, ev5__itb_ia // Flush ITB + +#if icflush_on_tbix != 0 + + + br r31, pal_ic_flush // Flush Icache +#else + + hw_rei_stall +#endif + + nop // Pad table + + // -1 tbiap +//orig pvc_jsr tbi, dest=1 + mtpr r31, ev5__dtb_iap // Flush DTB + mtpr r31, ev5__itb_iap // Flush ITB + +#if icflush_on_tbix != 0 + + + br r31, pal_ic_flush // Flush Icache +#else + + hw_rei_stall +#endif + + nop // Pad table + + + // 0 unused +//orig pvc_jsr tbi, dest=1 + hw_rei // Pad table + nop + nop + nop + + + // 1 tbisi +//orig pvc_jsr tbi, dest=1 +#if icflush_on_tbix != 0 + + + + nop + br r31, pal_ic_flush_and_tbisi // Flush Icache + nop + nop // Pad table +#else + + nop + nop + mtpr r17, ev5__itb_is // Flush ITB + hw_rei_stall +#endif + + + + // 2 tbisd +//orig pvc_jsr tbi, dest=1 + mtpr r17, ev5__dtb_is // Flush DTB. + nop + + nop + hw_rei_stall + + + // 3 tbis +//orig pvc_jsr tbi, dest=1 + mtpr r17, ev5__dtb_is // Flush DTB +#if icflush_on_tbix != 0 + + + br r31, pal_ic_flush_and_tbisi // Flush Icache and ITB +#else + br r31, tbi_finish + ALIGN_BRANCH +tbi_finish: + mtpr r17, ev5__itb_is // Flush ITB + hw_rei_stall +#endif + + + + ALIGN_BLOCK +//+ +// bpt_bchk_common: +// Finish up the bpt/bchk instructions +//- +bpt_bchk_common: + stq r18, osfsf_a2(sp) // a2 + mfpr r13, pt_entif // get entry point + + stq r12, osfsf_ps(sp) // save old ps + stq r14, osfsf_pc(sp) // save pc + + stq r29, osfsf_gp(sp) // save gp + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kgp + + + hw_rei_spe // done + + + ALIGN_BLOCK +//+ +// rti_to_user +// Finish up the rti instruction +//- +rti_to_user: + mtpr r11, ev5__dtb_cm // set Mbox current mode - no virt ref for 2 cycles + mtpr r11, ev5__ps // set Ibox current mode - 2 bubble to hw_rei + + mtpr r31, ev5__ipl // set the ipl. No hw_rei for 2 cycles + mtpr r25, pt_ksp // save off incase RTI to user + + mfpr r30, pt_usp + hw_rei_spe // and back + + + ALIGN_BLOCK +//+ +// rti_to_kern +// Finish up the rti instruction +//- +rti_to_kern: + and r12, osfps_m_ipl, r11 // clean ps + mfpr r12, pt_intmask // get int mask + + extbl r12, r11, r12 // get mask for this ipl + mtpr r25, pt_ksp // save off incase RTI to user + + mtpr r12, ev5__ipl // set the new ipl. + or r25, r31, sp // sp + +// pvc_violate 217 // possible hidden mt->mf ipl not a problem in callpals + hw_rei + + ALIGN_BLOCK +//+ +// swpctx_cont +// Finish up the swpctx instruction +//- + +swpctx_cont: +#if ev5_p1 != 0 + + + bic r25, r24, r25 // clean icsr<FPE> + get_impure r8 // get impure pointer + + sll r12, icsr_v_fpe, r12 // shift new fen to pos + fix_impure_ipr r8 // adjust impure pointer + + restore_reg1 pmctr_ctl, r8, r8, ipr=1 // "ldqp" - get pmctr_ctl bits + srl r23, 32, r24 // move asn to low asn pos + + ldqp r14, osfpcb_q_mmptr(r16)// get new mmptr + srl r22, osfpcb_v_pme, r22 // get pme down to bit 0 + + or r25, r12, r25 // icsr with new fen + sll r24, itb_asn_v_asn, r12 + +#else + + bic r25, r24, r25 // clean icsr<FPE,PMP> + sll r12, icsr_v_fpe, r12 // shift new fen to pos + + ldqp r14, osfpcb_q_mmptr(r16)// get new mmptr + srl r22, osfpcb_v_pme, r22 // get pme down to bit 0 + + or r25, r12, r25 // icsr with new fen + srl r23, 32, r24 // move asn to low asn pos + + and r22, 1, r22 + sll r24, itb_asn_v_asn, r12 + + sll r22, icsr_v_pmp, r22 + nop + + or r25, r22, r25 // icsr with new pme +#endif + + sll r24, dtb_asn_v_asn, r24 + + subl r23, r13, r13 // gen new cc offset + mtpr r12, itb_asn // no hw_rei_stall in 0,1,2,3,4 + + mtpr r24, dtb_asn // Load up new ASN + mtpr r25, icsr // write the icsr + + sll r14, page_offset_size_bits, r14 // Move PTBR into internal position. + ldqp r25, osfpcb_q_usp(r16) // get new usp + + insll r13, 4, r13 // >> 32 +// pvc_violate 379 // ldqp can't trap except replay. only problem if mf same ipr in same shadow + mtpr r14, pt_ptbr // load the new ptbr + + mtpr r13, cc // set new offset + ldqp r30, osfpcb_q_ksp(r16) // get new ksp + +// pvc_violate 379 // ldqp can't trap except replay. only problem if mf same ipr in same shadow + mtpr r25, pt_usp // save usp + +#if ev5_p1 != 0 + + + blbc r8, no_pm_change // if monitoring all processes -- no need to change pm + + // otherwise, monitoring select processes - update pm + lda r25, 0x3F(r31) + cmovlbc r22, r31, r8 // if pme set, disable counters, otherwise use saved encodings + + sll r25, pmctr_v_ctl2, r25 // create ctl field bit mask + mfpr r22, ev5__pmctr + + and r8, r25, r8 // mask new ctl value + bic r22, r25, r22 // clear ctl field in pmctr + + or r8, r22, r8 + mtpr r8, ev5__pmctr + +no_pm_change: +#endif + + +#if osf_chm_fix != 0 + + + p4_fixup_hw_rei_stall // removes this section for Pass 4 by placing a hw_rei_stall here + +#if build_fixed_image != 0 + + + hw_rei_stall +#else + + mfpr r9, pt_pcbb // get FEN +#endif + + ldqp r9, osfpcb_q_fen(r9) + blbc r9, no_pm_change_10_ // skip if FEN disabled + + mb // ensure no outstanding fills + lda r12, 1<<dc_mode_v_dc_ena(r31) + mtpr r12, dc_mode // turn dcache on so we can flush it + nop // force correct slotting + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + + lda r8, 0(r31) // flood the dcache with junk data +no_pm_change_5_: ldqp r31, 0(r8) + lda r8, 0x20(r8) // touch each cache block + srl r8, 13, r9 + blbc r9, no_pm_change_5_ + + mb // ensure no outstanding fills + mtpr r31, dc_mode // turn the dcache back off + nop // force correct slotting + mfpr r31, pt0 // no hw_rei_stall in 0,1 +#endif + + +no_pm_change_10_: hw_rei_stall // back we go + + ALIGN_BLOCK +//+ +// swppal_cont - finish up the swppal call_pal +//- + +swppal_cont: + mfpr r2, pt_misc // get misc bits + sll r0, pt_misc_v_switch, r0 // get the "I've switched" bit + or r2, r0, r2 // set the bit + mtpr r31, ev5__alt_mode // ensure alt_mode set to 0 (kernel) + mtpr r2, pt_misc // update the chip + + or r3, r31, r4 + mfpr r3, pt_impure // pass pointer to the impure area in r3 +//orig fix_impure_ipr r3 // adjust impure pointer for ipr read +//orig restore_reg1 bc_ctl, r1, r3, ipr=1 // pass cns_bc_ctl in r1 +//orig restore_reg1 bc_config, r2, r3, ipr=1 // pass cns_bc_config in r2 +//orig unfix_impure_ipr r3 // restore impure pointer + lda r3, CNS_Q_IPR(r3) + RESTORE_SHADOW(r1,CNS_Q_BC_CTL,r3); + RESTORE_SHADOW(r1,CNS_Q_BC_CFG,r3); + lda r3, -CNS_Q_IPR(r3) + + or r31, r31, r0 // set status to success +// pvc_violate 1007 + jmp r31, (r4) // and call our friend, it's her problem now + + +swppal_fail: + addq r0, 1, r0 // set unknown pal or not loaded + hw_rei // and return + + +// .sbttl "Memory management" + + ALIGN_BLOCK +//+ +//foe_ipte_handler +// IFOE detected on level 3 pte, sort out FOE vs ACV +// +// on entry: +// with +// R8 = pte +// R10 = pc +// +// Function +// Determine TNV vs ACV vs FOE. Build stack and dispatch +// Will not be here if TNV. +//- + +foe_ipte_handler: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, foe_ipte_handler_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + + srl r8, osfpte_v_ure-osfpte_v_kre, r8 // move pte user bits to kern + nop + +foe_ipte_handler_10_: srl r8, osfpte_v_kre, r25 // get kre to <0> + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + or r10, r31, r14 // Save pc/va in case TBmiss or fault on stack + mfpr r13, pt_entmm // get entry point + + stq r16, osfsf_a0(sp) // a0 + or r14, r31, r16 // pass pc/va as a0 + + stq r17, osfsf_a1(sp) // a1 + nop + + stq r18, osfsf_a2(sp) // a2 + lda r17, mmcsr_c_acv(r31) // assume ACV + + stq r16, osfsf_pc(sp) // save pc + cmovlbs r25, mmcsr_c_foe, r17 // otherwise FOE + + stq r12, osfsf_ps(sp) // save ps + subq r31, 1, r18 // pass flag of istream as a2 + + stq r29, osfsf_gp(sp) + mtpr r13, exc_addr // set vector address + + mfpr r29, pt_kgp // load kgp + hw_rei_spe // out to exec + + ALIGN_BLOCK +//+ +//invalid_ipte_handler +// TNV detected on level 3 pte, sort out TNV vs ACV +// +// on entry: +// with +// R8 = pte +// R10 = pc +// +// Function +// Determine TNV vs ACV. Build stack and dispatch. +//- + +invalid_ipte_handler: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, invalid_ipte_handler_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + + srl r8, osfpte_v_ure-osfpte_v_kre, r8 // move pte user bits to kern + nop + +invalid_ipte_handler_10_: srl r8, osfpte_v_kre, r25 // get kre to <0> + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + or r10, r31, r14 // Save pc/va in case TBmiss on stack + mfpr r13, pt_entmm // get entry point + + stq r16, osfsf_a0(sp) // a0 + or r14, r31, r16 // pass pc/va as a0 + + stq r17, osfsf_a1(sp) // a1 + nop + + stq r18, osfsf_a2(sp) // a2 + and r25, 1, r17 // Isolate kre + + stq r16, osfsf_pc(sp) // save pc + xor r17, 1, r17 // map to acv/tnv as a1 + + stq r12, osfsf_ps(sp) // save ps + subq r31, 1, r18 // pass flag of istream as a2 + + stq r29, osfsf_gp(sp) + mtpr r13, exc_addr // set vector address + + mfpr r29, pt_kgp // load kgp + hw_rei_spe // out to exec + + + + + ALIGN_BLOCK +//+ +//invalid_dpte_handler +// INVALID detected on level 3 pte, sort out TNV vs ACV +// +// on entry: +// with +// R10 = va +// R8 = pte +// R9 = mm_stat +// PT6 = pc +// +// Function +// Determine TNV vs ACV. Build stack and dispatch +//- + + +invalid_dpte_handler: + mfpr r12, pt6 + blbs r12, tnv_in_pal // Special handler if original faulting reference was in PALmode + + bis r12, r31, r14 // save PC in case of tbmiss or fault + srl r9, mm_stat_v_opcode, r25 // shift opc to <0> + + mtpr r11, pt0 // Save PS for stack write + and r25, mm_stat_m_opcode, r25 // isolate opcode + + cmpeq r25, evx_opc_sync, r25 // is it FETCH/FETCH_M? + blbs r25, nmiss_fetch_ldr31_err // yes + + //dismiss exception if load to r31/f31 + blbs r9, invalid_dpte_no_dismiss // mm_stat<0> set on store or fetchm + + // not a store or fetch, must be a load + srl r9, mm_stat_v_ra, r25 // Shift rnum to low bits + + and r25, 0x1F, r25 // isolate rnum + nop + + cmpeq r25, 0x1F, r25 // Is the rnum r31 or f31? + bne r25, nmiss_fetch_ldr31_err // Yes, dismiss the fault + +invalid_dpte_no_dismiss: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + bge r25, invalid_dpte_no_dismiss_10_ // no stack swap needed if cm=kern + + srl r8, osfpte_v_ure-osfpte_v_kre, r8 // move pte user bits to kern + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +invalid_dpte_no_dismiss_10_: srl r8, osfpte_v_kre, r12 // get kre to <0> + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + or r10, r31, r25 // Save va in case TBmiss on stack + and r9, 1, r13 // save r/w flag + + stq r16, osfsf_a0(sp) // a0 + or r25, r31, r16 // pass va as a0 + + stq r17, osfsf_a1(sp) // a1 + or r31, mmcsr_c_acv, r17 // assume acv + + srl r12, osfpte_v_kwe-osfpte_v_kre, r25 // get write enable to <0> + stq r29, osfsf_gp(sp) + + stq r18, osfsf_a2(sp) // a2 + cmovlbs r13, r25, r12 // if write access move acv based on write enable + + or r13, r31, r18 // pass flag of dstream access and read vs write + mfpr r25, pt0 // get ps + + stq r14, osfsf_pc(sp) // save pc + mfpr r13, pt_entmm // get entry point + + stq r25, osfsf_ps(sp) // save ps + mtpr r13, exc_addr // set vector address + + mfpr r29, pt_kgp // load kgp + cmovlbs r12, mmcsr_c_tnv, r17 // make p2 be tnv if access ok else acv + + hw_rei_spe // out to exec + +//+ +// +// We come here if we are erring on a dtb_miss, and the instr is a +// fetch, fetch_m, of load to r31/f31. +// The PC is incremented, and we return to the program. +// essentially ignoring the instruction and error. +// +//- + ALIGN_BLOCK +nmiss_fetch_ldr31_err: + mfpr r12, pt6 + addq r12, 4, r12 // bump pc to pc+4 + + mtpr r12, exc_addr // and set entry point + mfpr r31, pt0 // pad exc_addr write + + hw_rei // + + ALIGN_BLOCK +//+ +// double_pte_inv +// We had a single tbmiss which turned into a double tbmiss which found +// an invalid PTE. Return to single miss with a fake pte, and the invalid +// single miss flow will report the error. +// +// on entry: +// r21 PTE +// r22 available +// VA IPR locked with original fault VA +// pt4 saved r21 +// pt5 saved r22 +// pt6 original exc_addr +// +// on return to tbmiss flow: +// r8 fake PTE +// +// +//- +double_pte_inv: + srl r21, osfpte_v_kre, r21 // get the kre bit to <0> + mfpr r22, exc_addr // get the pc + + lda r22, 4(r22) // inc the pc + lda r8, osfpte_m_prot(r31) // make a fake pte with xre and xwe set + + cmovlbc r21, r31, r8 // set to all 0 for acv if pte<kre> is 0 + mtpr r22, exc_addr // set for rei + + mfpr r21, pt4 // restore regs + mfpr r22, pt5 // restore regs + + hw_rei // back to tb miss + + ALIGN_BLOCK +//+ +//tnv_in_pal +// The only places in pal that ld or store are the +// stack builders, rti or retsys. Any of these mean we +// need to take a ksp not valid halt. +// +//- +tnv_in_pal: + + + br r31, ksp_inval_halt + + +// .sbttl "Icache flush routines" + + ALIGN_BLOCK +//+ +// Common Icache flush routine. +// +// +//- +pal_ic_flush: + nop + mtpr r31, ev5__ic_flush_ctl // Icache flush - E1 + nop + nop + +// Now, do 44 NOPs. 3RFB prefetches (24) + IC buffer,IB,slot,issue (20) + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 10 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 20 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 30 + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 40 + + nop + nop + +one_cycle_and_hw_rei: + nop + nop + + hw_rei_stall + +#if icflush_on_tbix != 0 + + + ALIGN_BLOCK + +//+ +// Common Icache flush and ITB invalidate single routine. +// ITBIS and hw_rei_stall must be in same octaword. +// r17 - has address to invalidate +// +//- +PAL_IC_FLUSH_AND_TBISI: + nop + mtpr r31, ev5__ic_flush_ctl // Icache flush - E1 + nop + nop + +// Now, do 44 NOPs. 3RFB prefetches (24) + IC buffer,IB,slot,issue (20) + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 10 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 20 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 30 + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 40 + + + nop + nop + + nop + nop + + // A quadword is 64 bits, so an octaword is 128 bits -> 16 bytes -> 4 instructions + // 44 nops plus 4 instructions before it is 48 instructions. + // Since this routine started on a 32-byte (8 instruction) boundary, + // the following 2 instructions will be in the same octword as required. +// ALIGN_BRANCH + mtpr r17, ev5__itb_is // Flush ITB + hw_rei_stall + +#endif + + ALIGN_BLOCK +//+ +//osfpal_calpal_opcdec +// Here for all opcdec CALL_PALs +// +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + +osfpal_calpal_opcdec: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r14, exc_addr // get pc + nop + + bis r11, r31, r12 // Save PS for stack write + bge r25, osfpal_calpal_opcdec_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +osfpal_calpal_opcdec_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + nop + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_opdec, r16 // set a0 + + stq r18, osfsf_a2(sp) // a2 + mfpr r13, pt_entif // get entry point + + stq r12, osfsf_ps(sp) // save old ps + stq r17, osfsf_a1(sp) // a1 + + stq r14, osfsf_pc(sp) // save pc + nop + + stq r29, osfsf_gp(sp) // save gp + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kgp + + + hw_rei_spe // done + + + + + +//+ +//pal_update_pcb +// Update the PCB with the current SP, AST, and CC info +// +// r0 - return linkage +//- + ALIGN_BLOCK + +pal_update_pcb: + mfpr r12, pt_pcbb // get pcbb + and r11, osfps_m_mode, r25 // get mode + beq r25, pal_update_pcb_10_ // in kern? no need to update user sp + mtpr r30, pt_usp // save user stack + stqp r30, osfpcb_q_usp(r12) // store usp + br r31, pal_update_pcb_20_ // join common +pal_update_pcb_10_: stqp r30, osfpcb_q_ksp(r12) // store ksp +pal_update_pcb_20_: rpcc r13 // get cyccounter + srl r13, 32, r14 // move offset + addl r13, r14, r14 // merge for new time + stlp r14, osfpcb_l_cc(r12) // save time + +//orig pvc_jsr updpcb, bsr=1, dest=1 + ret r31, (r0) + + + +#if remove_save_state == 0 + +// .sbttl "PAL_SAVE_STATE" +//+ +// +// Pal_save_state +// +// Function +// All chip state saved, all PT's, SR's FR's, IPR's +// +// +// Regs' on entry... +// +// R0 = halt code +// pt0 = r0 +// R1 = pointer to impure +// pt4 = r1 +// R3 = return addr +// pt5 = r3 +// +// register usage: +// r0 = halt_code +// r1 = addr of impure area +// r3 = return_address +// r4 = scratch +// +//- + + + ALIGN_BLOCK + .globl pal_save_state +pal_save_state: +// +// +// start of implementation independent save routine +// +// the impure area is larger than the addressibility of hw_ld and hw_st +// therefore, we need to play some games: The impure area +// is informally divided into the "machine independent" part and the +// "machine dependent" part. The state that will be saved in the +// "machine independent" part are gpr's, fpr's, hlt, flag, mchkflag (use (un)fix_impure_gpr macros). +// All others will be in the "machine dependent" part (use (un)fix_impure_ipr macros). +// The impure pointer will need to be adjusted by a different offset for each. The store/restore_reg +// macros will automagically adjust the offset correctly. +// + +// The distributed code is commented out and followed by corresponding SRC code. +// Beware: SAVE_IPR and RESTORE_IPR blow away r0(v0) + +//orig fix_impure_gpr r1 // adjust impure area pointer for stores to "gpr" part of impure area + lda r1, 0x200(r1) // Point to center of CPU segment +//orig store_reg1 flag, r31, r1, ipr=1 // clear dump area flag + SAVE_GPR(r31,CNS_Q_FLAG,r1) // Clear the valid flag +//orig store_reg1 hlt, r0, r1, ipr=1 + SAVE_GPR(r0,CNS_Q_HALT,r1) // Save the halt code + + mfpr r0, pt0 // get r0 back //orig +//orig store_reg1 0, r0, r1 // save r0 + SAVE_GPR(r0,CNS_Q_GPR+0x00,r1) // Save r0 + + mfpr r0, pt4 // get r1 back //orig +//orig store_reg1 1, r0, r1 // save r1 + SAVE_GPR(r0,CNS_Q_GPR+0x08,r1) // Save r1 + +//orig store_reg 2 // save r2 + SAVE_GPR(r2,CNS_Q_GPR+0x10,r1) // Save r2 + + mfpr r0, pt5 // get r3 back //orig +//orig store_reg1 3, r0, r1 // save r3 + SAVE_GPR(r0,CNS_Q_GPR+0x18,r1) // Save r3 + + // reason code has been saved + // r0 has been saved + // r1 has been saved + // r2 has been saved + // r3 has been saved + // pt0, pt4, pt5 have been lost + + // + // Get out of shadow mode + // + + mfpr r2, icsr // Get icsr //orig +//orig ldah r0, <1@<icsr_v_sde-16>>(r31) // Get a one in SHADOW_ENABLE bit location + ldah r0, (1<<(icsr_v_sde-16))(r31) + bic r2, r0, r0 // ICSR with SDE clear //orig + mtpr r0, icsr // Turn off SDE //orig + + mfpr r31, pt0 // SDE bubble cycle 1 //orig + mfpr r31, pt0 // SDE bubble cycle 2 //orig + mfpr r31, pt0 // SDE bubble cycle 3 //orig + nop //orig + + + // save integer regs R4-r31 +//orig #define t 4 +//orig .repeat 28 +//orig store_reg \t +//orig #define t t + 1 +//orig .endr + SAVE_GPR(r4,CNS_Q_GPR+0x20,r1) + SAVE_GPR(r5,CNS_Q_GPR+0x28,r1) + SAVE_GPR(r6,CNS_Q_GPR+0x30,r1) + SAVE_GPR(r7,CNS_Q_GPR+0x38,r1) + SAVE_GPR(r8,CNS_Q_GPR+0x40,r1) + SAVE_GPR(r9,CNS_Q_GPR+0x48,r1) + SAVE_GPR(r10,CNS_Q_GPR+0x50,r1) + SAVE_GPR(r11,CNS_Q_GPR+0x58,r1) + SAVE_GPR(r12,CNS_Q_GPR+0x60,r1) + SAVE_GPR(r13,CNS_Q_GPR+0x68,r1) + SAVE_GPR(r14,CNS_Q_GPR+0x70,r1) + SAVE_GPR(r15,CNS_Q_GPR+0x78,r1) + SAVE_GPR(r16,CNS_Q_GPR+0x80,r1) + SAVE_GPR(r17,CNS_Q_GPR+0x88,r1) + SAVE_GPR(r18,CNS_Q_GPR+0x90,r1) + SAVE_GPR(r19,CNS_Q_GPR+0x98,r1) + SAVE_GPR(r20,CNS_Q_GPR+0xA0,r1) + SAVE_GPR(r21,CNS_Q_GPR+0xA8,r1) + SAVE_GPR(r22,CNS_Q_GPR+0xB0,r1) + SAVE_GPR(r23,CNS_Q_GPR+0xB8,r1) + SAVE_GPR(r24,CNS_Q_GPR+0xC0,r1) + SAVE_GPR(r25,CNS_Q_GPR+0xC8,r1) + SAVE_GPR(r26,CNS_Q_GPR+0xD0,r1) + SAVE_GPR(r27,CNS_Q_GPR+0xD8,r1) + SAVE_GPR(r28,CNS_Q_GPR+0xE0,r1) + SAVE_GPR(r29,CNS_Q_GPR+0xE8,r1) + SAVE_GPR(r30,CNS_Q_GPR+0xF0,r1) + SAVE_GPR(r31,CNS_Q_GPR+0xF8,r1) + + // save all paltemp regs except pt0 + +//orig unfix_impure_gpr r1 // adjust impure area pointer for gpr stores +//orig fix_impure_ipr r1 // adjust impure area pointer for pt stores +//orig #define t 1 +//orig .repeat 23 +//orig store_reg \t , pal=1 +//orig #define t t + 1 +//orig .endr + + lda r1, -0x200(r1) // Restore the impure base address. + lda r1, CNS_Q_IPR(r1) // Point to the base of IPR area. + SAVE_IPR(pt0,CNS_Q_PT+0x00,r1) // the osf code didn't save/restore palTemp 0 ?? pboyle + SAVE_IPR(pt1,CNS_Q_PT+0x08,r1) + SAVE_IPR(pt2,CNS_Q_PT+0x10,r1) + SAVE_IPR(pt3,CNS_Q_PT+0x18,r1) + SAVE_IPR(pt4,CNS_Q_PT+0x20,r1) + SAVE_IPR(pt5,CNS_Q_PT+0x28,r1) + SAVE_IPR(pt6,CNS_Q_PT+0x30,r1) + SAVE_IPR(pt7,CNS_Q_PT+0x38,r1) + SAVE_IPR(pt8,CNS_Q_PT+0x40,r1) + SAVE_IPR(pt9,CNS_Q_PT+0x48,r1) + SAVE_IPR(pt10,CNS_Q_PT+0x50,r1) + SAVE_IPR(pt11,CNS_Q_PT+0x58,r1) + SAVE_IPR(pt12,CNS_Q_PT+0x60,r1) + SAVE_IPR(pt13,CNS_Q_PT+0x68,r1) + SAVE_IPR(pt14,CNS_Q_PT+0x70,r1) + SAVE_IPR(pt15,CNS_Q_PT+0x78,r1) + SAVE_IPR(pt16,CNS_Q_PT+0x80,r1) + SAVE_IPR(pt17,CNS_Q_PT+0x88,r1) + SAVE_IPR(pt18,CNS_Q_PT+0x90,r1) + SAVE_IPR(pt19,CNS_Q_PT+0x98,r1) + SAVE_IPR(pt20,CNS_Q_PT+0xA0,r1) + SAVE_IPR(pt21,CNS_Q_PT+0xA8,r1) + SAVE_IPR(pt22,CNS_Q_PT+0xB0,r1) + SAVE_IPR(pt23,CNS_Q_PT+0xB8,r1) + + // Restore shadow mode + mfpr r31, pt0 // pad write to icsr out of shadow of store (trap does not abort write) //orig + mfpr r31, pt0 //orig + mtpr r2, icsr // Restore original ICSR //orig + + mfpr r31, pt0 // SDE bubble cycle 1 //orig + mfpr r31, pt0 // SDE bubble cycle 2 //orig + mfpr r31, pt0 // SDE bubble cycle 3 //orig + nop //orig + + // save all integer shadow regs + +//orig #define t 8 +//orig .repeat 7 +//orig store_reg \t, shadow=1 +//orig #define t t + 1 +//orig .endr +//orig store_reg 25, shadow=1 + + SAVE_SHADOW( r8,CNS_Q_SHADOW+0x00,r1) // also called p0...p7 in the Hudson code + SAVE_SHADOW( r9,CNS_Q_SHADOW+0x08,r1) + SAVE_SHADOW(r10,CNS_Q_SHADOW+0x10,r1) + SAVE_SHADOW(r11,CNS_Q_SHADOW+0x18,r1) + SAVE_SHADOW(r12,CNS_Q_SHADOW+0x20,r1) + SAVE_SHADOW(r13,CNS_Q_SHADOW+0x28,r1) + SAVE_SHADOW(r14,CNS_Q_SHADOW+0x30,r1) + SAVE_SHADOW(r25,CNS_Q_SHADOW+0x38,r1) + +//orig store_reg exc_addr, ipr=1 // save ipr +//orig store_reg pal_base, ipr=1 // save ipr +//orig store_reg mm_stat, ipr=1 // save ipr +//orig store_reg va, ipr=1 // save ipr +//orig store_reg icsr, ipr=1 // save ipr +//orig store_reg ipl, ipr=1 // save ipr +//orig store_reg ps, ipr=1 // save ipr +//orig store_reg itb_asn, ipr=1 // save ipr +//orig store_reg aster, ipr=1 // save ipr +//orig store_reg astrr, ipr=1 // save ipr +//orig store_reg sirr, ipr=1 // save ipr +//orig store_reg isr, ipr=1 // save ipr +//orig store_reg ivptbr, ipr=1 // save ipr +//orig store_reg mcsr, ipr=1 // save ipr +//orig store_reg dc_mode, ipr=1 // save ipr + + SAVE_IPR(excAddr,CNS_Q_EXC_ADDR,r1) + SAVE_IPR(palBase,CNS_Q_PAL_BASE,r1) + SAVE_IPR(mmStat,CNS_Q_MM_STAT,r1) + SAVE_IPR(va,CNS_Q_VA,r1) + SAVE_IPR(icsr,CNS_Q_ICSR,r1) + SAVE_IPR(ipl,CNS_Q_IPL,r1) + SAVE_IPR(ips,CNS_Q_IPS,r1) + SAVE_IPR(itbAsn,CNS_Q_ITB_ASN,r1) + SAVE_IPR(aster,CNS_Q_ASTER,r1) + SAVE_IPR(astrr,CNS_Q_ASTRR,r1) + SAVE_IPR(sirr,CNS_Q_SIRR,r1) + SAVE_IPR(isr,CNS_Q_ISR,r1) + SAVE_IPR(iVptBr,CNS_Q_IVPTBR,r1) + SAVE_IPR(mcsr,CNS_Q_MCSR,r1) + SAVE_IPR(dcMode,CNS_Q_DC_MODE,r1) + +//orig pvc_violate 379 // mf maf_mode after a store ok (pvc doesn't distinguish ld from st) +//orig store_reg maf_mode, ipr=1 // save ipr -- no mbox instructions for +//orig // PVC violation applies only to +pvc$osf35$379: // loads. HW_ST ok here, so ignore + SAVE_IPR(mafMode,CNS_Q_MAF_MODE,r1) // MBOX INST->MF MAF_MODE IN 0,1,2 + + + //the following iprs are informational only -- will not be restored + +//orig store_reg icperr_stat, ipr=1 +//orig store_reg pmctr, ipr=1 +//orig store_reg intid, ipr=1 +//orig store_reg exc_sum, ipr=1 +//orig store_reg exc_mask, ipr=1 +//orig ldah r14, 0xfff0(r31) +//orig zap r14, 0xE0, r14 // Get Cbox IPR base +//orig nop // pad mf dcperr_stat out of shadow of last store +//orig nop +//orig nop +//orig store_reg dcperr_stat, ipr=1 + + SAVE_IPR(icPerr,CNS_Q_ICPERR_STAT,r1) + SAVE_IPR(PmCtr,CNS_Q_PM_CTR,r1) + SAVE_IPR(intId,CNS_Q_INT_ID,r1) + SAVE_IPR(excSum,CNS_Q_EXC_SUM,r1) + SAVE_IPR(excMask,CNS_Q_EXC_MASK,r1) + ldah r14, 0xFFF0(zero) + zap r14, 0xE0, r14 // Get base address of CBOX IPRs + NOP // Pad mfpr dcPerr out of shadow of + NOP // last store + NOP + SAVE_IPR(dcPerr,CNS_Q_DCPERR_STAT,r1) + + // read cbox ipr state + +//orig mb +//orig ldqp r2, ev5__sc_ctl(r14) +//orig ldqp r13, ld_lock(r14) +//orig ldqp r4, ev5__sc_addr(r14) +//orig ldqp r5, ev5__ei_addr(r14) +//orig ldqp r6, ev5__bc_tag_addr(r14) +//orig ldqp r7, ev5__fill_syn(r14) +//orig bis r5, r4, r31 +//orig bis r7, r6, r31 // make sure previous loads finish before reading stat registers which unlock them +//orig ldqp r8, ev5__sc_stat(r14) // unlocks sc_stat,sc_addr +//orig ldqp r9, ev5__ei_stat(r14) // may unlock ei_*, bc_tag_addr, fill_syn +//orig ldqp r31, ev5__ei_stat(r14) // ensures it is really unlocked +//orig mb + +#ifndef SIMOS + mb + ldq_p r2, scCtl(r14) + ldq_p r13, ldLock(r14) + ldq_p r4, scAddr(r14) + ldq_p r5, eiAddr(r14) + ldq_p r6, bcTagAddr(r14) + ldq_p r7, fillSyn(r14) + bis r5, r4, zero // Make sure all loads complete before + bis r7, r6, zero // reading registers that unlock them. + ldq_p r8, scStat(r14) // Unlocks scAddr. + ldq_p r9, eiStat(r14) // Unlocks eiAddr, bcTagAddr, fillSyn. + ldq_p zero, eiStat(r14) // Make sure it is really unlocked. + mb +#endif +//orig // save cbox ipr state +//orig store_reg1 sc_ctl, r2, r1, ipr=1 +//orig store_reg1 ld_lock, r13, r1, ipr=1 +//orig store_reg1 sc_addr, r4, r1, ipr=1 +//orig store_reg1 ei_addr, r5, r1, ipr=1 +//orig store_reg1 bc_tag_addr, r6, r1, ipr=1 +//orig store_reg1 fill_syn, r7, r1, ipr=1 +//orig store_reg1 sc_stat, r8, r1, ipr=1 +//orig store_reg1 ei_stat, r9, r1, ipr=1 +//orig //bc_config? sl_rcv? + + SAVE_SHADOW(r2,CNS_Q_SC_CTL,r1); + SAVE_SHADOW(r13,CNS_Q_LD_LOCK,r1); + SAVE_SHADOW(r4,CNS_Q_SC_ADDR,r1); + SAVE_SHADOW(r5,CNS_Q_EI_ADDR,r1); + SAVE_SHADOW(r6,CNS_Q_BC_TAG_ADDR,r1); + SAVE_SHADOW(r7,CNS_Q_FILL_SYN,r1); + SAVE_SHADOW(r8,CNS_Q_SC_STAT,r1); + SAVE_SHADOW(r9,CNS_Q_EI_STAT,r1); + +// restore impure base //orig +//orig unfix_impure_ipr r1 + lda r1, -CNS_Q_IPR(r1) + +// save all floating regs //orig + mfpr r0, icsr // get icsr //orig + or r31, 1, r2 // get a one //orig +//orig sll r2, #icsr_v_fpe, r2 // shift for fpu spot //orig + sll r2, icsr_v_fpe, r2 // Shift it into ICSR<FPE> position + or r2, r0, r0 // set FEN on //orig + mtpr r0, icsr // write to icsr, enabling FEN //orig + +// map the save area virtually +// orig mtpr r31, dtb_ia // clear the dtb +// orig srl r1, page_offset_size_bits, r0 // Clean off low bits of VA +// orig sll r0, 32, r0 // shift to PFN field +// orig lda r2, 0xff(r31) // all read enable and write enable bits set +// orig sll r2, 8, r2 // move to PTE location +// orig addq r0, r2, r0 // combine with PFN +// orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +// orig mtpr r1, dtb_tag // write TB tag + + mtpr r31, dtbIa // Clear all DTB entries + srl r1, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r1, dtbTag // Write the PTE and tag into the DTB + + +//orig // map the next page too - in case the impure area crosses a page boundary +//orig lda r4, 1@page_offset_size_bits(r1) // generate address for next page +//orig srl r4, page_offset_size_bits, r0 // Clean off low bits of VA +//orig sll r0, 32, r0 // shift to PFN field +//orig lda r2, 0xff(r31) // all read enable and write enable bits set +//orig sll r2, 8, r2 // move to PTE location +//orig addq r0, r2, r0 // combine with PFN +//orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +//orig mtpr r4, dtb_tag // write TB tag + + lda r4, (1<<va_s_off)(r1) // Generate address for next page + srl r4, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r4, dtbTag // Write the PTE and tag into the DTB + + sll r31, 0, r31 // stall cycle 1 // orig + sll r31, 0, r31 // stall cycle 2 // orig + sll r31, 0, r31 // stall cycle 3 // orig + nop // orig + +//orig // add offset for saving fpr regs +//orig fix_impure_gpr r1 + + lda r1, 0x200(r1) // Point to center of CPU segment + +// now save the regs - F0-F31 + +//orig #define t 0 +//orig .repeat 32 +//orig store_reg \t , fpu=1 +//orig #define t t + 1 +//orig .endr + + mf_fpcr f0 // original + + SAVE_FPR(f0,CNS_Q_FPR+0x00,r1) + SAVE_FPR(f1,CNS_Q_FPR+0x08,r1) + SAVE_FPR(f2,CNS_Q_FPR+0x10,r1) + SAVE_FPR(f3,CNS_Q_FPR+0x18,r1) + SAVE_FPR(f4,CNS_Q_FPR+0x20,r1) + SAVE_FPR(f5,CNS_Q_FPR+0x28,r1) + SAVE_FPR(f6,CNS_Q_FPR+0x30,r1) + SAVE_FPR(f7,CNS_Q_FPR+0x38,r1) + SAVE_FPR(f8,CNS_Q_FPR+0x40,r1) + SAVE_FPR(f9,CNS_Q_FPR+0x48,r1) + SAVE_FPR(f10,CNS_Q_FPR+0x50,r1) + SAVE_FPR(f11,CNS_Q_FPR+0x58,r1) + SAVE_FPR(f12,CNS_Q_FPR+0x60,r1) + SAVE_FPR(f13,CNS_Q_FPR+0x68,r1) + SAVE_FPR(f14,CNS_Q_FPR+0x70,r1) + SAVE_FPR(f15,CNS_Q_FPR+0x78,r1) + SAVE_FPR(f16,CNS_Q_FPR+0x80,r1) + SAVE_FPR(f17,CNS_Q_FPR+0x88,r1) + SAVE_FPR(f18,CNS_Q_FPR+0x90,r1) + SAVE_FPR(f19,CNS_Q_FPR+0x98,r1) + SAVE_FPR(f20,CNS_Q_FPR+0xA0,r1) + SAVE_FPR(f21,CNS_Q_FPR+0xA8,r1) + SAVE_FPR(f22,CNS_Q_FPR+0xB0,r1) + SAVE_FPR(f23,CNS_Q_FPR+0xB8,r1) + SAVE_FPR(f24,CNS_Q_FPR+0xC0,r1) + SAVE_FPR(f25,CNS_Q_FPR+0xC8,r1) + SAVE_FPR(f26,CNS_Q_FPR+0xD0,r1) + SAVE_FPR(f27,CNS_Q_FPR+0xD8,r1) + SAVE_FPR(f28,CNS_Q_FPR+0xE0,r1) + SAVE_FPR(f29,CNS_Q_FPR+0xE8,r1) + SAVE_FPR(f30,CNS_Q_FPR+0xF0,r1) + SAVE_FPR(f31,CNS_Q_FPR+0xF8,r1) + +//orig //switch impure offset from gpr to ipr--- +//orig unfix_impure_gpr r1 +//orig fix_impure_ipr r1 +//orig store_reg1 fpcsr, f0, r1, fpcsr=1 + + SAVE_FPR(f0,CNS_Q_FPCSR,r1) // fpcsr loaded above into f0 -- can it reach// pb + lda r1, -0x200(r1) // Restore the impure base address + +//orig // and back to gpr --- +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 + +//orig lda r0, cns_mchksize(r31) // get size of mchk area +//orig store_reg1 mchkflag, r0, r1, ipr=1 +//orig mb + + lda r1, CNS_Q_IPR(r1) // Point to base of IPR area again + // save this using the IPR base (it is closer) not the GRP base as they used...pb + lda r0, MACHINE_CHECK_SIZE(r31) // get size of mchk area + SAVE_SHADOW(r0,CNS_Q_MCHK,r1); + mb + +//orig or r31, 1, r0 // get a one +//orig store_reg1 flag, r0, r1, ipr=1 // set dump area flag +//orig mb + + lda r1, -CNS_Q_IPR(r1) // back to the base + lda r1, 0x200(r1) // Point to center of CPU segment + or r31, 1, r0 // get a one + SAVE_GPR(r0,CNS_Q_FLAG,r1) // // set dump area valid flag + mb + +//orig // restore impure area base +//orig unfix_impure_gpr r1 + lda r1, -0x200(r1) // Point to center of CPU segment + + mtpr r31, dtb_ia // clear the dtb //orig + mtpr r31, itb_ia // clear the itb //orig + +//orig pvc_jsr savsta, bsr=1, dest=1 + ret r31, (r3) // and back we go +#endif + + +#if remove_restore_state == 0 + + +// .sbttl "PAL_RESTORE_STATE" +//+ +// +// Pal_restore_state +// +// +// register usage: +// r1 = addr of impure area +// r3 = return_address +// all other regs are scratchable, as they are about to +// be reloaded from ram. +// +// Function: +// All chip state restored, all SRs, FRs, PTs, IPRs +// *** except R1, R3, PT0, PT4, PT5 *** +// +//- + ALIGN_BLOCK +pal_restore_state: + +//need to restore sc_ctl,bc_ctl,bc_config??? if so, need to figure out a safe way to do so. + +//orig // map the console io area virtually +//orig mtpr r31, dtb_ia // clear the dtb +//orig srl r1, page_offset_size_bits, r0 // Clean off low bits of VA +//orig sll r0, 32, r0 // shift to PFN field +//orig lda r2, 0xff(r31) // all read enable and write enable bits set +//orig sll r2, 8, r2 // move to PTE location +//orig addq r0, r2, r0 // combine with PFN +//orig +//orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +//orig mtpr r1, dtb_tag // write TB tag +//orig + + mtpr r31, dtbIa // Clear all DTB entries + srl r1, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r1, dtbTag // Write the PTE and tag into the DTB + + +//orig // map the next page too, in case impure area crosses page boundary +//orig lda r4, 1@page_offset_size_bits(r1) // generate address for next page +//orig srl r4, page_offset_size_bits, r0 // Clean off low bits of VA +//orig sll r0, 32, r0 // shift to PFN field +//orig lda r2, 0xff(r31) // all read enable and write enable bits set +//orig sll r2, 8, r2 // move to PTE location +//orig addq r0, r2, r0 // combine with PFN +//orig +//orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +//orig mtpr r4, dtb_tag // write TB tag - no virtual mbox instruction for 3 cycles + + lda r4, (1<<VA_S_OFF)(r1) // Generate address for next page + srl r4, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r4, dtbTag // Write the PTE and tag into the DTB + +//orig // save all floating regs +//orig mfpr r0, icsr // get icsr +//orig// assume ICSR_V_SDE gt <ICSR_V_FPE> // assertion checker +//orig or r31, <<1@<ICSR_V_SDE-ICSR_V_FPE>> ! 1>, r2 // set SDE and FPE +//orig sll r2, #icsr_v_fpe, r2 // shift for fpu spot +//orig or r2, r0, r0 // set FEN on +//orig mtpr r0, icsr // write to icsr, enabling FEN and SDE. 3 bubbles to floating instr. + + mfpr r0, icsr // Get current ICSR + bis zero, 1, r2 // Get a '1' + or r2, (1<<(icsr_v_sde-icsr_v_fpe)), r2 + sll r2, icsr_v_fpe, r2 // Shift bits into position + bis r2, r2, r0 // Set ICSR<SDE> and ICSR<FPE> + mtpr r0, icsr // Update the chip + + mfpr r31, pt0 // FPE bubble cycle 1 //orig + mfpr r31, pt0 // FPE bubble cycle 2 //orig + mfpr r31, pt0 // FPE bubble cycle 3 //orig + +//orig fix_impure_ipr r1 +//orig restore_reg1 fpcsr, f0, r1, fpcsr=1 +//orig mt_fpcr f0 +//orig +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 // adjust impure pointer offset for gpr access +//orig +//orig // restore all floating regs +//orig#define t 0 +//orig .repeat 32 +//orig restore_reg \t , fpu=1 +//orig#define t t + 1 +//orig .endr + + lda r1, 200(r1) // Point to base of IPR area again + RESTORE_FPR(f0,CNS_Q_FPCSR,r1) // can it reach?? pb + mt_fpcr f0 // original + + lda r1, 0x200(r1) // point to center of CPU segment + RESTORE_FPR(f0,CNS_Q_FPR+0x00,r1) + RESTORE_FPR(f1,CNS_Q_FPR+0x08,r1) + RESTORE_FPR(f2,CNS_Q_FPR+0x10,r1) + RESTORE_FPR(f3,CNS_Q_FPR+0x18,r1) + RESTORE_FPR(f4,CNS_Q_FPR+0x20,r1) + RESTORE_FPR(f5,CNS_Q_FPR+0x28,r1) + RESTORE_FPR(f6,CNS_Q_FPR+0x30,r1) + RESTORE_FPR(f7,CNS_Q_FPR+0x38,r1) + RESTORE_FPR(f8,CNS_Q_FPR+0x40,r1) + RESTORE_FPR(f9,CNS_Q_FPR+0x48,r1) + RESTORE_FPR(f10,CNS_Q_FPR+0x50,r1) + RESTORE_FPR(f11,CNS_Q_FPR+0x58,r1) + RESTORE_FPR(f12,CNS_Q_FPR+0x60,r1) + RESTORE_FPR(f13,CNS_Q_FPR+0x68,r1) + RESTORE_FPR(f14,CNS_Q_FPR+0x70,r1) + RESTORE_FPR(f15,CNS_Q_FPR+0x78,r1) + RESTORE_FPR(f16,CNS_Q_FPR+0x80,r1) + RESTORE_FPR(f17,CNS_Q_FPR+0x88,r1) + RESTORE_FPR(f18,CNS_Q_FPR+0x90,r1) + RESTORE_FPR(f19,CNS_Q_FPR+0x98,r1) + RESTORE_FPR(f20,CNS_Q_FPR+0xA0,r1) + RESTORE_FPR(f21,CNS_Q_FPR+0xA8,r1) + RESTORE_FPR(f22,CNS_Q_FPR+0xB0,r1) + RESTORE_FPR(f23,CNS_Q_FPR+0xB8,r1) + RESTORE_FPR(f24,CNS_Q_FPR+0xC0,r1) + RESTORE_FPR(f25,CNS_Q_FPR+0xC8,r1) + RESTORE_FPR(f26,CNS_Q_FPR+0xD0,r1) + RESTORE_FPR(f27,CNS_Q_FPR+0xD8,r1) + RESTORE_FPR(f28,CNS_Q_FPR+0xE0,r1) + RESTORE_FPR(f29,CNS_Q_FPR+0xE8,r1) + RESTORE_FPR(f30,CNS_Q_FPR+0xF0,r1) + RESTORE_FPR(f31,CNS_Q_FPR+0xF8,r1) + +//orig // switch impure pointer from gpr to ipr area -- +//orig unfix_impure_gpr r1 +//orig fix_impure_ipr r1 +//orig +//orig // restore all pal regs +//orig#define t 1 +//orig .repeat 23 +//orig restore_reg \t , pal=1 +//orig#define t t + 1 +//orig .endr + + lda r1, -0x200(r1) // Restore base address of impure area. + lda r1, CNS_Q_IPR(r1) // Point to base of IPR area. + RESTORE_IPR(pt0,CNS_Q_PT+0x00,r1) // the osf code didn't save/restore palTemp 0 ?? pboyle + RESTORE_IPR(pt1,CNS_Q_PT+0x08,r1) + RESTORE_IPR(pt2,CNS_Q_PT+0x10,r1) + RESTORE_IPR(pt3,CNS_Q_PT+0x18,r1) + RESTORE_IPR(pt4,CNS_Q_PT+0x20,r1) + RESTORE_IPR(pt5,CNS_Q_PT+0x28,r1) + RESTORE_IPR(pt6,CNS_Q_PT+0x30,r1) + RESTORE_IPR(pt7,CNS_Q_PT+0x38,r1) + RESTORE_IPR(pt8,CNS_Q_PT+0x40,r1) + RESTORE_IPR(pt9,CNS_Q_PT+0x48,r1) + RESTORE_IPR(pt10,CNS_Q_PT+0x50,r1) + RESTORE_IPR(pt11,CNS_Q_PT+0x58,r1) + RESTORE_IPR(pt12,CNS_Q_PT+0x60,r1) + RESTORE_IPR(pt13,CNS_Q_PT+0x68,r1) + RESTORE_IPR(pt14,CNS_Q_PT+0x70,r1) + RESTORE_IPR(pt15,CNS_Q_PT+0x78,r1) + RESTORE_IPR(pt16,CNS_Q_PT+0x80,r1) + RESTORE_IPR(pt17,CNS_Q_PT+0x88,r1) + RESTORE_IPR(pt18,CNS_Q_PT+0x90,r1) + RESTORE_IPR(pt19,CNS_Q_PT+0x98,r1) + RESTORE_IPR(pt20,CNS_Q_PT+0xA0,r1) + RESTORE_IPR(pt21,CNS_Q_PT+0xA8,r1) + RESTORE_IPR(pt22,CNS_Q_PT+0xB0,r1) + RESTORE_IPR(pt23,CNS_Q_PT+0xB8,r1) + + +//orig restore_reg exc_addr, ipr=1 // restore ipr +//orig restore_reg pal_base, ipr=1 // restore ipr +//orig restore_reg ipl, ipr=1 // restore ipr +//orig restore_reg ps, ipr=1 // restore ipr +//orig mtpr r0, dtb_cm // set current mode in mbox too +//orig restore_reg itb_asn, ipr=1 +//orig srl r0, itb_asn_v_asn, r0 +//orig sll r0, dtb_asn_v_asn, r0 +//orig mtpr r0, dtb_asn // set ASN in Mbox too +//orig restore_reg ivptbr, ipr=1 +//orig mtpr r0, mvptbr // use ivptbr value to restore mvptbr +//orig restore_reg mcsr, ipr=1 +//orig restore_reg aster, ipr=1 +//orig restore_reg astrr, ipr=1 +//orig restore_reg sirr, ipr=1 +//orig restore_reg maf_mode, ipr=1 // no mbox instruction for 3 cycles +//orig mfpr r31, pt0 // (may issue with mt maf_mode) +//orig mfpr r31, pt0 // bubble cycle 1 +//orig mfpr r31, pt0 // bubble cycle 2 +//orig mfpr r31, pt0 // bubble cycle 3 +//orig mfpr r31, pt0 // (may issue with following ld) + + // r0 gets the value of RESTORE_IPR in the macro and this code uses this side effect (gag) + RESTORE_IPR(excAddr,CNS_Q_EXC_ADDR,r1) + RESTORE_IPR(palBase,CNS_Q_PAL_BASE,r1) + RESTORE_IPR(ipl,CNS_Q_IPL,r1) + RESTORE_IPR(ips,CNS_Q_IPS,r1) + mtpr r0, dtbCm // Set Mbox current mode too. + RESTORE_IPR(itbAsn,CNS_Q_ITB_ASN,r1) + srl r0, 4, r0 + sll r0, 57, r0 + mtpr r0, dtbAsn // Set Mbox ASN too + RESTORE_IPR(iVptBr,CNS_Q_IVPTBR,r1) + mtpr r0, mVptBr // Set Mbox VptBr too + RESTORE_IPR(mcsr,CNS_Q_MCSR,r1) + RESTORE_IPR(aster,CNS_Q_ASTER,r1) + RESTORE_IPR(astrr,CNS_Q_ASTRR,r1) + RESTORE_IPR(sirr,CNS_Q_SIRR,r1) + RESTORE_IPR(mafMode,CNS_Q_MAF_MODE,r1) + STALL + STALL + STALL + STALL + STALL + + + // restore all integer shadow regs +//orig#define t 8 +//orig .repeat 7 +//orig restore_reg \t, shadow=1 +//orig#define t t + 1 +//orig .endr +//orig restore_reg 25, shadow=1 +//orig restore_reg dc_mode, ipr=1 // no mbox instructions for 4 cycles + + RESTORE_SHADOW( r8,CNS_Q_SHADOW+0x00,r1) // also called p0...p7 in the Hudson code + RESTORE_SHADOW( r9,CNS_Q_SHADOW+0x08,r1) + RESTORE_SHADOW(r10,CNS_Q_SHADOW+0x10,r1) + RESTORE_SHADOW(r11,CNS_Q_SHADOW+0x18,r1) + RESTORE_SHADOW(r12,CNS_Q_SHADOW+0x20,r1) + RESTORE_SHADOW(r13,CNS_Q_SHADOW+0x28,r1) + RESTORE_SHADOW(r14,CNS_Q_SHADOW+0x30,r1) + RESTORE_SHADOW(r25,CNS_Q_SHADOW+0x38,r1) + RESTORE_IPR(dcMode,CNS_Q_DC_MODE,r1) + + // + // Get out of shadow mode + // + + mfpr r31, pt0 // pad last load to icsr write (in case of replay, icsr will be written anyway) //orig + mfpr r31, pt0 // "" //orig + mfpr r0, icsr // Get icsr //orig +//orig ldah r2, <1@<icsr_v_sde-16>>(r31) // Get a one in SHADOW_ENABLE bit location + ldah r2, (1<<(ICSR_V_SDE-16))(r31) // Get a one in SHADOW_ENABLE bit location //orig + bic r0, r2, r2 // ICSR with SDE clear //orig + mtpr r2, icsr // Turn off SDE - no palshadow rd/wr for 3 bubble cycles //orig + + mfpr r31, pt0 // SDE bubble cycle 1 //orig + mfpr r31, pt0 // SDE bubble cycle 2 //orig + mfpr r31, pt0 // SDE bubble cycle 3 //orig + nop //orig + +//orig // switch impure pointer from ipr to gpr area -- +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 +//orig // restore all integer regs +//orig#define t 4 +//orig .repeat 28 +//orig restore_reg \t +//orig#define t t + 1 +//orig .endr + +// Restore GPRs (r0, r2 are restored later, r1 and r3 are trashed) ... + + lda r1, -CNS_Q_IPR(r1) // Restore base address of impure area + lda r1, 0x200(r1) // Point to center of CPU segment + + RESTORE_GPR(r4,CNS_Q_GPR+0x20,r1) + RESTORE_GPR(r5,CNS_Q_GPR+0x28,r1) + RESTORE_GPR(r6,CNS_Q_GPR+0x30,r1) + RESTORE_GPR(r7,CNS_Q_GPR+0x38,r1) + RESTORE_GPR(r8,CNS_Q_GPR+0x40,r1) + RESTORE_GPR(r9,CNS_Q_GPR+0x48,r1) + RESTORE_GPR(r10,CNS_Q_GPR+0x50,r1) + RESTORE_GPR(r11,CNS_Q_GPR+0x58,r1) + RESTORE_GPR(r12,CNS_Q_GPR+0x60,r1) + RESTORE_GPR(r13,CNS_Q_GPR+0x68,r1) + RESTORE_GPR(r14,CNS_Q_GPR+0x70,r1) + RESTORE_GPR(r15,CNS_Q_GPR+0x78,r1) + RESTORE_GPR(r16,CNS_Q_GPR+0x80,r1) + RESTORE_GPR(r17,CNS_Q_GPR+0x88,r1) + RESTORE_GPR(r18,CNS_Q_GPR+0x90,r1) + RESTORE_GPR(r19,CNS_Q_GPR+0x98,r1) + RESTORE_GPR(r20,CNS_Q_GPR+0xA0,r1) + RESTORE_GPR(r21,CNS_Q_GPR+0xA8,r1) + RESTORE_GPR(r22,CNS_Q_GPR+0xB0,r1) + RESTORE_GPR(r23,CNS_Q_GPR+0xB8,r1) + RESTORE_GPR(r24,CNS_Q_GPR+0xC0,r1) + RESTORE_GPR(r25,CNS_Q_GPR+0xC8,r1) + RESTORE_GPR(r26,CNS_Q_GPR+0xD0,r1) + RESTORE_GPR(r27,CNS_Q_GPR+0xD8,r1) + RESTORE_GPR(r28,CNS_Q_GPR+0xE0,r1) + RESTORE_GPR(r29,CNS_Q_GPR+0xE8,r1) + RESTORE_GPR(r30,CNS_Q_GPR+0xF0,r1) + RESTORE_GPR(r31,CNS_Q_GPR+0xF8,r1) + +//orig // switch impure pointer from gpr to ipr area -- +//orig unfix_impure_gpr r1 +//orig fix_impure_ipr r1 +//orig restore_reg icsr, ipr=1 // restore original icsr- 4 bubbles to hw_rei + + lda t0, -0x200(t0) // Restore base address of impure area. + lda t0, CNS_Q_IPR(t0) // Point to base of IPR area again. + RESTORE_IPR(icsr,CNS_Q_ICSR,r1) + +//orig // and back again -- +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 +//orig store_reg1 flag, r31, r1, ipr=1 // clear dump area valid flag +//orig mb + + lda t0, -CNS_Q_IPR(t0) // Back to base of impure area again, + lda t0, 0x200(t0) // and back to center of CPU segment + SAVE_GPR(r31,CNS_Q_FLAG,r1) // Clear the dump area valid flag + mb + +//orig // and back we go +//orig// restore_reg 3 +//orig restore_reg 2 +//orig// restore_reg 1 +//orig restore_reg 0 +//orig // restore impure area base +//orig unfix_impure_gpr r1 + + RESTORE_GPR(r2,CNS_Q_GPR+0x10,r1) + RESTORE_GPR(r0,CNS_Q_GPR+0x00,r1) + lda r1, -0x200(r1) // Restore impure base address + + mfpr r31, pt0 // stall for ldqp above //orig + + mtpr r31, dtb_ia // clear the tb //orig + mtpr r31, itb_ia // clear the itb //orig + +//orig pvc_jsr rststa, bsr=1, dest=1 + ret r31, (r3) // back we go //orig +#endif + + +//+ +// pal_pal_bug_check -- code has found a bugcheck situation. +// Set things up and join common machine check flow. +// +// Input: +// r14 - exc_addr +// +// On exit: +// pt0 - saved r0 +// pt1 - saved r1 +// pt4 - saved r4 +// pt5 - saved r5 +// pt6 - saved r6 +// pt10 - saved exc_addr +// pt_misc<47:32> - mchk code +// pt_misc<31:16> - scb vector +// r14 - base of Cbox IPRs in IO space +// MCES<mchk> is set +//- + + ALIGN_BLOCK + .globl pal_pal_bug_check_from_int +pal_pal_bug_check_from_int: + DEBUGSTORE(0x79) +//simos DEBUG_EXC_ADDR() + DEBUGSTORE(0x20) +//simos bsr r25, put_hex + lda r25, mchk_c_bugcheck(r31) + addq r25, 1, r25 // set flag indicating we came from interrupt and stack is already pushed + br r31, pal_pal_mchk + nop + +pal_pal_bug_check: + lda r25, mchk_c_bugcheck(r31) + +pal_pal_mchk: + sll r25, 32, r25 // Move mchk code to position + + mtpr r14, pt10 // Stash exc_addr + mtpr r14, exc_addr + + mfpr r12, pt_misc // Get MCES and scratch + zap r12, 0x3c, r12 + + or r12, r25, r12 // Combine mchk code + lda r25, scb_v_procmchk(r31) // Get SCB vector + + sll r25, 16, r25 // Move SCBv to position + or r12, r25, r25 // Combine SCBv + + mtpr r0, pt0 // Stash for scratch + bis r25, mces_m_mchk, r25 // Set MCES<MCHK> bit + + mtpr r25, pt_misc // Save mchk code!scbv!whami!mces + ldah r14, 0xfff0(r31) + + mtpr r1, pt1 // Stash for scratch + zap r14, 0xE0, r14 // Get Cbox IPR base + + mtpr r4, pt4 + mtpr r5, pt5 + + mtpr r6, pt6 + blbs r12, sys_double_machine_check // MCHK halt if double machine check + + br r31, sys_mchk_collect_iprs // Join common machine check flow + +// align_to_call_pal_section // Align to address of first call_pal entry point - 2000 + +// .sbttl "HALT - PALcode for HALT instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// GO to console code +// +//- + + .text 1 +// . = 0x2000 + CALL_PAL_PRIV(PAL_HALT_ENTRY) +call_pal_halt: +#if rax_mode == 0 + mfpr r31, pt0 // Pad exc_addr read + mfpr r31, pt0 + + mfpr r12, exc_addr // get PC + subq r12, 4, r12 // Point to the HALT + + mtpr r12, exc_addr + mtpr r0, pt0 + +//orig pvc_jsr updpcb, bsr=1 + bsr r0, pal_update_pcb // update the pcb + lda r0, hlt_c_sw_halt(r31) // set halt code to sw halt + br r31, sys_enter_console // enter the console + +#else // RAX mode + mb + mb + mtpr r9, ev5__dtb_asn // no Dstream virtual ref for next 3 cycles. + mtpr r9, ev5__itb_asn // E1. Update ITB ASN. No hw_rei for 5 cycles. + mtpr r8, exc_addr // no HW_REI for 1 cycle. + blbc r9, not_begin_case + mtpr r31, ev5__dtb_ia // clear DTB. No Dstream virtual ref for 2 cycles. + mtpr r31, ev5__itb_ia // clear ITB. + +not_begin_case: + nop + nop + + nop + nop // pad mt itb_asn ->hw_rei_stall + + hw_rei_stall +#endif + +// .sbttl "CFLUSH- PALcode for CFLUSH instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// R16 - contains the PFN of the page to be flushed +// +// Function: +// Flush all Dstream caches of 1 entire page +// The CFLUSH routine is in the system specific module. +// +//- + + CALL_PAL_PRIV(PAL_CFLUSH_ENTRY) +Call_Pal_Cflush: + br r31, sys_cflush + +// .sbttl "DRAINA - PALcode for DRAINA instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// Implicit TRAPB performed by hardware. +// +// Function: +// Stall instruction issue until all prior instructions are guaranteed to +// complete without incurring aborts. For the EV5 implementation, this +// means waiting until all pending DREADS are returned. +// +//- + + CALL_PAL_PRIV(PAL_DRAINA_ENTRY) +Call_Pal_Draina: + ldah r14, 0x100(r31) // Init counter. Value? + nop + +DRAINA_LOOP: + subq r14, 1, r14 // Decrement counter + mfpr r13, ev5__maf_mode // Fetch status bit + + srl r13, maf_mode_v_dread_pending, r13 + ble r14, DRAINA_LOOP_TOO_LONG + + nop + blbs r13, DRAINA_LOOP // Wait until all DREADS clear + + hw_rei + +DRAINA_LOOP_TOO_LONG: + br r31, call_pal_halt + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x0003) +CallPal_OpcDec03: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0004) +CallPal_OpcDec04: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0005) +CallPal_OpcDec05: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0006) +CallPal_OpcDec06: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0007) +CallPal_OpcDec07: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0008) +CallPal_OpcDec08: + br r31, osfpal_calpal_opcdec + +// .sbttl "CSERVE- PALcode for CSERVE instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Various functions for private use of console software +// +// option selector in r0 +// arguments in r16.... +// The CSERVE routine is in the system specific module. +// +//- + + CALL_PAL_PRIV(PAL_CSERVE_ENTRY) +Call_Pal_Cserve: + br r31, sys_cserve + +// .sbttl "swppal - PALcode for swppal instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// Vectored into via hardware PALcode instruction dispatch. +// R16 contains the new PAL identifier +// R17:R21 contain implementation-specific entry parameters +// +// R0 receives status: +// 0 success (PAL was switched) +// 1 unknown PAL variant +// 2 known PAL variant, but PAL not loaded +// +// +// Function: +// Swap control to another PAL. +//- + + CALL_PAL_PRIV(PAL_SWPPAL_ENTRY) +Call_Pal_Swppal: + cmpule r16, 255, r0 // see if a kibble was passed + cmoveq r16, r16, r0 // if r16=0 then a valid address (ECO 59) + + or r16, r31, r3 // set r3 incase this is a address + blbc r0, swppal_cont // nope, try it as an address + + cmpeq r16, 2, r0 // is it our friend OSF? + blbc r0, swppal_fail // nope, don't know this fellow + + br r2, CALL_PAL_SWPPAL_10_ // tis our buddy OSF + +// .global osfpal_hw_entry_reset +// .weak osfpal_hw_entry_reset +// .long <osfpal_hw_entry_reset-pal_start> +//orig halt // don't know how to get the address here - kludge ok, load pal at 0 + .long 0 // ?? hack upon hack...pb + +CALL_PAL_SWPPAL_10_: ldlp r3, 0(r2) // fetch target addr +// ble r3, swppal_fail ; if OSF not linked in say not loaded. + mfpr r2, pal_base // fetch pal base + + addq r2, r3, r3 // add pal base + lda r2, 0x3FFF(r31) // get pal base checker mask + + and r3, r2, r2 // any funky bits set? + cmpeq r2, 0, r0 // + + blbc r0, swppal_fail // return unknown if bad bit set. + br r31, swppal_cont + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x000B) +CallPal_OpcDec0B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x000C) +CallPal_OpcDec0C: + br r31, osfpal_calpal_opcdec + +// .sbttl "wripir- PALcode for wripir instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// r16 = processor number to interrupt +// +// Function: +// IPIR <- R16 +// Handled in system-specific code +// +// Exit: +// interprocessor interrupt is recorded on the target processor +// and is initiated when the proper enabling conditions are present. +//- + + CALL_PAL_PRIV(PAL_WRIPIR_ENTRY) +Call_Pal_Wrpir: + br r31, sys_wripir + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x000E) +CallPal_OpcDec0E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x000F) +CallPal_OpcDec0F: + br r31, osfpal_calpal_opcdec + +// .sbttl "rdmces- PALcode for rdmces instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// R0 <- ZEXT(MCES) +//- + + CALL_PAL_PRIV(PAL_RDMCES_ENTRY) +Call_Pal_Rdmces: + mfpr r0, pt_mces // Read from PALtemp + and r0, mces_m_all, r0 // Clear other bits + + hw_rei + +// .sbttl "wrmces- PALcode for wrmces instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// If {R16<0> EQ 1} then MCES<0> <- 0 (MCHK) +// If {R16<1> EQ 1} then MCES<1> <- 0 (SCE) +// If {R16<2> EQ 1} then MCES<2> <- 0 (PCE) +// MCES<3> <- R16<3> (DPC) +// MCES<4> <- R16<4> (DSC) +// +//- + + CALL_PAL_PRIV(PAL_WRMCES_ENTRY) +Call_Pal_Wrmces: + and r16, ((1<<mces_v_mchk) | (1<<mces_v_sce) | (1<<mces_v_pce)), r13 // Isolate MCHK, SCE, PCE + mfpr r14, pt_mces // Get current value + + ornot r31, r13, r13 // Flip all the bits + and r16, ((1<<mces_v_dpc) | (1<<mces_v_dsc)), r17 + + and r14, r13, r1 // Update MCHK, SCE, PCE + bic r1, ((1<<mces_v_dpc) | (1<<mces_v_dsc)), r1 // Clear old DPC, DSC + + or r1, r17, r1 // Update DPC and DSC + mtpr r1, pt_mces // Write MCES back + +#if rawhide_system == 0 + nop // Pad to fix PT write->read restriction +#else + blbs r16, RAWHIDE_clear_mchk_lock // Clear logout from lock +#endif + + nop + hw_rei + + + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x0012) +CallPal_OpcDec12: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0013) +CallPal_OpcDec13: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0014) +CallPal_OpcDec14: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0015) +CallPal_OpcDec15: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0016) +CallPal_OpcDec16: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0017) +CallPal_OpcDec17: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0018) +CallPal_OpcDec18: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0019) +CallPal_OpcDec19: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001A) +CallPal_OpcDec1A: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001B) +CallPal_OpcDec1B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001C) +CallPal_OpcDec1C: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001D) +CallPal_OpcDec1D: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001E) +CallPal_OpcDec1E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001F) +CallPal_OpcDec1F: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0020) +CallPal_OpcDec20: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0021) +CallPal_OpcDec21: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0022) +CallPal_OpcDec22: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0023) +CallPal_OpcDec23: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0024) +CallPal_OpcDec24: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0025) +CallPal_OpcDec25: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0026) +CallPal_OpcDec26: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0027) +CallPal_OpcDec27: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0028) +CallPal_OpcDec28: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0029) +CallPal_OpcDec29: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x002A) +CallPal_OpcDec2A: + br r31, osfpal_calpal_opcdec + +// .sbttl "wrfen - PALcode for wrfen instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// a0<0> -> ICSR<FPE> +// Store new FEN in PCB +// Final value of t0 (r1), t8..t10 (r22..r24) and a0 (r16) are UNPREDICTABLE +// +// Issue: What about pending FP loads when FEN goes from on->off???? +//- + + CALL_PAL_PRIV(PAL_WRFEN_ENTRY) +Call_Pal_Wrfen: + or r31, 1, r13 // Get a one + mfpr r1, ev5__icsr // Get current FPE + + sll r13, icsr_v_fpe, r13 // shift 1 to icsr<fpe> spot, e0 + and r16, 1, r16 // clean new fen + + sll r16, icsr_v_fpe, r12 // shift new fen to correct bit position + bic r1, r13, r1 // zero icsr<fpe> + + or r1, r12, r1 // Or new FEN into ICSR + mfpr r12, pt_pcbb // Get PCBB - E1 + + mtpr r1, ev5__icsr // write new ICSR. 3 Bubble cycles to HW_REI + stlp r16, osfpcb_q_fen(r12) // Store FEN in PCB. + + mfpr r31, pt0 // Pad ICSR<FPE> write. + mfpr r31, pt0 + + mfpr r31, pt0 +// pvc_violate 225 // cuz PVC can't distinguish which bits changed + hw_rei + + + CALL_PAL_PRIV(0x002C) +CallPal_OpcDec2C: + br r31, osfpal_calpal_opcdec + +// .sbttl "wrvptpr - PALcode for wrvptpr instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// vptptr <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRVPTPTR_ENTRY) +Call_Pal_Wrvptptr: + mtpr r16, ev5__mvptbr // Load Mbox copy + mtpr r16, ev5__ivptbr // Load Ibox copy + nop // Pad IPR write + nop + hw_rei + + CALL_PAL_PRIV(0x002E) +CallPal_OpcDec2E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x002F) +CallPal_OpcDec2F: + br r31, osfpal_calpal_opcdec + +// .sbttl "swpctx- PALcode for swpctx instruction" + +//+ +// +// Entry: +// hardware dispatch via callPal instruction +// R16 -> new pcb +// +// Function: +// dynamic state moved to old pcb +// new state loaded from new pcb +// pcbb pointer set +// old pcbb returned in R0 +// +// Note: need to add perf monitor stuff +//- + + CALL_PAL_PRIV(PAL_SWPCTX_ENTRY) +Call_Pal_Swpctx: + rpcc r13 // get cyccounter + mfpr r0, pt_pcbb // get pcbb + + ldqp r22, osfpcb_q_fen(r16) // get new fen/pme + ldqp r23, osfpcb_l_cc(r16) // get new asn + + srl r13, 32, r25 // move offset + mfpr r24, pt_usp // get usp + + stqp r30, osfpcb_q_ksp(r0) // store old ksp +// pvc_violate 379 // stqp can't trap except replay. only problem if mf same ipr in same shadow. + mtpr r16, pt_pcbb // set new pcbb + + stqp r24, osfpcb_q_usp(r0) // store usp + addl r13, r25, r25 // merge for new time + + stlp r25, osfpcb_l_cc(r0) // save time + ldah r24, (1<<(icsr_v_fpe-16))(r31) + + and r22, 1, r12 // isolate fen + mfpr r25, icsr // get current icsr + + ev5_pass2 lda r24, (1<<icsr_v_pmp)(r24) + br r31, swpctx_cont + +// .sbttl "wrval - PALcode for wrval instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// sysvalue <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRVAL_ENTRY) +Call_Pal_Wrval: + nop + mtpr r16, pt_sysval // Pad paltemp write + nop + nop + hw_rei + + +// .sbttl "rdval - PALcode for rdval instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- sysvalue +//- + + CALL_PAL_PRIV(PAL_RDVAL_ENTRY) +Call_Pal_Rdval: + nop + mfpr r0, pt_sysval + nop + hw_rei + +// .sbttl "tbi - PALcode for tbi instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// TB invalidate +// r16/a0 = TBI type +// r17/a1 = Va for TBISx instructions +//- + + CALL_PAL_PRIV(PAL_TBI_ENTRY) +Call_Pal_Tbi: + addq r16, 2, r16 // change range to 0-2 + br r23, CALL_PAL_tbi_10_ // get our address + +CALL_PAL_tbi_10_: cmpult r16, 6, r22 // see if in range + lda r23, tbi_tbl-CALL_PAL_tbi_10_(r23) // set base to start of table + sll r16, 4, r16 // * 16 + blbc r22, CALL_PAL_tbi_30_ // go rei, if not + + addq r23, r16, r23 // addr of our code +//orig pvc_jsr tbi + jmp r31, (r23) // and go do it + +CALL_PAL_tbi_30_: + hw_rei + nop + +// .sbttl "wrent - PALcode for wrent instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Update ent* in paltemps +// r16/a0 = Address of entry routine +// r17/a1 = Entry Number 0..5 +// +// r22, r23 trashed +//- + + CALL_PAL_PRIV(PAL_WRENT_ENTRY) +Call_Pal_Wrent: + cmpult r17, 6, r22 // see if in range + br r23, CALL_PAL_wrent_10_ // get our address + +CALL_PAL_wrent_10_: bic r16, 3, r16 // clean pc + blbc r22, CALL_PAL_wrent_30_ // go rei, if not in range + + lda r23, wrent_tbl-CALL_PAL_wrent_10_(r23) // set base to start of table + sll r17, 4, r17 // *16 + + addq r17, r23, r23 // Get address in table +//orig pvc_jsr wrent + jmp r31, (r23) // and go do it + +CALL_PAL_wrent_30_: + hw_rei // out of range, just return + +// .sbttl "swpipl - PALcode for swpipl instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- PS<IPL> +// PS<IPL> <- a0<2:0> (r16) +// +// t8 (r22) is scratch +//- + + CALL_PAL_PRIV(PAL_SWPIPL_ENTRY) +Call_Pal_Swpipl: + and r16, osfps_m_ipl, r16 // clean New ipl + mfpr r22, pt_intmask // get int mask + + extbl r22, r16, r22 // get mask for this ipl + bis r11, r31, r0 // return old ipl + + bis r16, r31, r11 // set new ps + mtpr r22, ev5__ipl // set new mask + + mfpr r31, pt0 // pad ipl write + mfpr r31, pt0 // pad ipl write + + hw_rei // back + +// .sbttl "rdps - PALcode for rdps instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- ps +//- + + CALL_PAL_PRIV(PAL_RDPS_ENTRY) +Call_Pal_Rdps: + bis r11, r31, r0 // Fetch PALshadow PS + nop // Must be 2 cycles long + hw_rei + +// .sbttl "wrkgp - PALcode for wrkgp instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// kgp <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRKGP_ENTRY) +Call_Pal_Wrkgp: + nop + mtpr r16, pt_kgp + nop // Pad for pt write->read restriction + nop + hw_rei + +// .sbttl "wrusp - PALcode for wrusp instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// usp <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRUSP_ENTRY) +Call_Pal_Wrusp: + nop + mtpr r16, pt_usp + nop // Pad possible pt write->read restriction + nop + hw_rei + +// .sbttl "wrperfmon - PALcode for wrperfmon instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// +// Function: +// Various control functions for the onchip performance counters +// +// option selector in r16 +// option argument in r17 +// returned status in r0 +// +// +// r16 = 0 Disable performance monitoring for one or more cpu's +// r17 = 0 disable no counters +// r17 = bitmask disable counters specified in bit mask (1=disable) +// +// r16 = 1 Enable performance monitoring for one or more cpu's +// r17 = 0 enable no counters +// r17 = bitmask enable counters specified in bit mask (1=enable) +// +// r16 = 2 Mux select for one or more cpu's +// r17 = Mux selection (cpu specific) +// <24:19> bc_ctl<pm_mux_sel> field (see spec) +// <31>,<7:4>,<3:0> pmctr <sel0>,<sel1>,<sel2> fields (see spec) +// +// r16 = 3 Options +// r17 = (cpu specific) +// <0> = 0 log all processes +// <0> = 1 log only selected processes +// <30,9,8> mode select - ku,kp,kk +// +// r16 = 4 Interrupt frequency select +// r17 = (cpu specific) indicates interrupt frequencies desired for each +// counter, with "zero interrupts" being an option +// frequency info in r17 bits as defined by PMCTR_CTL<FRQx> below +// +// r16 = 5 Read Counters +// r17 = na +// r0 = value (same format as ev5 pmctr) +// <0> = 0 Read failed +// <0> = 1 Read succeeded +// +// r16 = 6 Write Counters +// r17 = value (same format as ev5 pmctr; all counters written simultaneously) +// +// r16 = 7 Enable performance monitoring for one or more cpu's and reset counter to 0 +// r17 = 0 enable no counters +// r17 = bitmask enable & clear counters specified in bit mask (1=enable & clear) +// +//============================================================================= +//Assumptions: +//PMCTR_CTL: +// +// <15:14> CTL0 -- encoded frequency select and enable - CTR0 +// <13:12> CTL1 -- " - CTR1 +// <11:10> CTL2 -- " - CTR2 +// +// <9:8> FRQ0 -- frequency select for CTR0 (no enable info) +// <7:6> FRQ1 -- frequency select for CTR1 +// <5:4> FRQ2 -- frequency select for CTR2 +// +// <0> all vs. select processes (0=all,1=select) +// +// where +// FRQx<1:0> +// 0 1 disable interrupt +// 1 0 frequency = 65536 (16384 for ctr2) +// 1 1 frequency = 256 +// note: FRQx<1:0> = 00 will keep counters from ever being enabled. +// +//============================================================================= +// + CALL_PAL_PRIV(0x0039) +// unsupported in Hudson code .. pboyle Nov/95 +CALL_PAL_Wrperfmon: +#if perfmon_debug == 0 + // "real" performance monitoring code + cmpeq r16, 1, r0 // check for enable + bne r0, perfmon_en // br if requested to enable + + cmpeq r16, 2, r0 // check for mux ctl + bne r0, perfmon_muxctl // br if request to set mux controls + + cmpeq r16, 3, r0 // check for options + bne r0, perfmon_ctl // br if request to set options + + cmpeq r16, 4, r0 // check for interrupt frequency select + bne r0, perfmon_freq // br if request to change frequency select + + cmpeq r16, 5, r0 // check for counter read request + bne r0, perfmon_rd // br if request to read counters + + cmpeq r16, 6, r0 // check for counter write request + bne r0, perfmon_wr // br if request to write counters + + cmpeq r16, 7, r0 // check for counter clear/enable request + bne r0, perfmon_enclr // br if request to clear/enable counters + + beq r16, perfmon_dis // br if requested to disable (r16=0) + br r31, perfmon_unknown // br if unknown request +#else + + br r31, pal_perfmon_debug +#endif + +// .sbttl "rdusp - PALcode for rdusp instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- usp +//- + + CALL_PAL_PRIV(PAL_RDUSP_ENTRY) +Call_Pal_Rdusp: + nop + mfpr r0, pt_usp + hw_rei + + + CALL_PAL_PRIV(0x003B) +CallPal_OpcDec3B: + br r31, osfpal_calpal_opcdec + +// .sbttl "whami - PALcode for whami instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- whami +//- + CALL_PAL_PRIV(PAL_WHAMI_ENTRY) +Call_Pal_Whami: + nop + mfpr r0, pt_whami // Get Whami + extbl r0, 1, r0 // Isolate just whami bits + hw_rei + +// .sbttl "retsys - PALcode for retsys instruction" +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// 00(sp) contains return pc +// 08(sp) contains r29 +// +// Function: +// Return from system call. +// mode switched from kern to user. +// stacks swapped, ugp, upc restored. +// r23, r25 junked +//- + + CALL_PAL_PRIV(PAL_RETSYS_ENTRY) +Call_Pal_Retsys: + lda r25, osfsf_c_size(sp) // pop stack + bis r25, r31, r14 // touch r25 & r14 to stall mf exc_addr + + mfpr r14, exc_addr // save exc_addr in case of fault + ldq r23, osfsf_pc(sp) // get pc + + ldq r29, osfsf_gp(sp) // get gp + stl_c r31, -4(sp) // clear lock_flag + + lda r11, 1<<osfps_v_mode(r31)// new PS:mode=user + mfpr r30, pt_usp // get users stack + + bic r23, 3, r23 // clean return pc + mtpr r31, ev5__ipl // zero ibox IPL - 2 bubbles to hw_rei + + mtpr r11, ev5__dtb_cm // set Mbox current mode - no virt ref for 2 cycles + mtpr r11, ev5__ps // set Ibox current mode - 2 bubble to hw_rei + + mtpr r23, exc_addr // set return address - 1 bubble to hw_rei + mtpr r25, pt_ksp // save kern stack + + rc r31 // clear inter_flag +// pvc_violate 248 // possible hidden mt->mf pt violation ok in callpal + hw_rei_spe // and back + + + CALL_PAL_PRIV(0x003E) +CallPal_OpcDec3E: + br r31, osfpal_calpal_opcdec + +// .sbttl "rti - PALcode for rti instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// 00(sp) -> ps +// 08(sp) -> pc +// 16(sp) -> r29 (gp) +// 24(sp) -> r16 (a0) +// 32(sp) -> r17 (a1) +// 40(sp) -> r18 (a3) +//- + + CALL_PAL_PRIV(PAL_RTI_ENTRY) +#ifdef SIMOS + /* called once by platform_tlaser */ + .globl Call_Pal_Rti +#endif +Call_Pal_Rti: + lda r25, osfsf_c_size(sp) // get updated sp + bis r25, r31, r14 // touch r14,r25 to stall mf exc_addr + + mfpr r14, exc_addr // save PC in case of fault + rc r31 // clear intr_flag + + ldq r12, -6*8(r25) // get ps + ldq r13, -5*8(r25) // pc + + ldq r18, -1*8(r25) // a2 + ldq r17, -2*8(r25) // a1 + + ldq r16, -3*8(r25) // a0 + ldq r29, -4*8(r25) // gp + + bic r13, 3, r13 // clean return pc + stl_c r31, -4(r25) // clear lock_flag + + and r12, osfps_m_mode, r11 // get mode + mtpr r13, exc_addr // set return address + + beq r11, rti_to_kern // br if rti to Kern + br r31, rti_to_user // out of call_pal space + + +// .sbttl "Start the Unprivileged CALL_PAL Entry Points" +// .sbttl "bpt- PALcode for bpt instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- +// + .text 1 +// . = 0x3000 + CALL_PAL_UNPRIV(PAL_BPT_ENTRY) +Call_Pal_Bpt: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, CALL_PAL_bpt_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +CALL_PAL_bpt_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_bpt, r16 // set a0 + + stq r17, osfsf_a1(sp) // a1 + br r31, bpt_bchk_common // out of call_pal space + + +// .sbttl "bugchk- PALcode for bugchk instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- +// + CALL_PAL_UNPRIV(PAL_BUGCHK_ENTRY) +Call_Pal_Bugchk: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, CALL_PAL_bugchk_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +CALL_PAL_bugchk_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_bugchk, r16 // set a0 + + stq r17, osfsf_a1(sp) // a1 + br r31, bpt_bchk_common // out of call_pal space + + + CALL_PAL_UNPRIV(0x0082) +CallPal_OpcDec82: + br r31, osfpal_calpal_opcdec + +// .sbttl "callsys - PALcode for callsys instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Switch mode to kernel and build a callsys stack frame. +// sp = ksp +// gp = kgp +// t8 - t10 (r22-r24) trashed +// +//- +// + CALL_PAL_UNPRIV(PAL_CALLSYS_ENTRY) +Call_Pal_Callsys: + + and r11, osfps_m_mode, r24 // get mode + mfpr r22, pt_ksp // get ksp + + beq r24, sys_from_kern // sysCall from kern is not allowed + mfpr r12, pt_entsys // get address of callSys routine + +//+ +// from here on we know we are in user going to Kern +//- + mtpr r31, ev5__dtb_cm // set Mbox current mode - no virt ref for 2 cycles + mtpr r31, ev5__ps // set Ibox current mode - 2 bubble to hw_rei + + bis r31, r31, r11 // PS=0 (mode=kern) + mfpr r23, exc_addr // get pc + + mtpr r30, pt_usp // save usp + lda sp, 0-osfsf_c_size(r22)// set new sp + + stq r29, osfsf_gp(sp) // save user gp/r29 + stq r24, osfsf_ps(sp) // save ps + + stq r23, osfsf_pc(sp) // save pc + mtpr r12, exc_addr // set address + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kern gp/r29 + + hw_rei_spe // and off we go! + + + CALL_PAL_UNPRIV(0x0084) +CallPal_OpcDec84: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0085) +CallPal_OpcDec85: + br r31, osfpal_calpal_opcdec + +// .sbttl "imb - PALcode for imb instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Flush the writebuffer and flush the Icache +// +//- +// + CALL_PAL_UNPRIV(PAL_IMB_ENTRY) +Call_Pal_Imb: + mb // Clear the writebuffer + mfpr r31, ev5__mcsr // Sync with clear + nop + nop + br r31, pal_ic_flush // Flush Icache + + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_UNPRIV(0x0087) +CallPal_OpcDec87: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0088) +CallPal_OpcDec88: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0089) +CallPal_OpcDec89: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008A) +CallPal_OpcDec8A: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008B) +CallPal_OpcDec8B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008C) +CallPal_OpcDec8C: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008D) +CallPal_OpcDec8D: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008E) +CallPal_OpcDec8E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008F) +CallPal_OpcDec8F: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0090) +CallPal_OpcDec90: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0091) +CallPal_OpcDec91: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0092) +CallPal_OpcDec92: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0093) +CallPal_OpcDec93: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0094) +CallPal_OpcDec94: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0095) +CallPal_OpcDec95: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0096) +CallPal_OpcDec96: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0097) +CallPal_OpcDec97: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0098) +CallPal_OpcDec98: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0099) +CallPal_OpcDec99: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009A) +CallPal_OpcDec9A: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009B) +CallPal_OpcDec9B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009C) +CallPal_OpcDec9C: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009D) +CallPal_OpcDec9D: + br r31, osfpal_calpal_opcdec + +// .sbttl "rdunique - PALcode for rdunique instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- unique +// +//- +// + CALL_PAL_UNPRIV(PAL_RDUNIQUE_ENTRY) +CALL_PALrdunique_: + mfpr r0, pt_pcbb // get pcb pointer + ldqp r0, osfpcb_q_unique(r0) // get new value + + hw_rei + +// .sbttl "wrunique - PALcode for wrunique instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// unique <- a0 (r16) +// +//- +// +CALL_PAL_UNPRIV(PAL_WRUNIQUE_ENTRY) +CALL_PAL_Wrunique: + nop + mfpr r12, pt_pcbb // get pcb pointer + stqp r16, osfpcb_q_unique(r12)// get new value + nop // Pad palshadow write + hw_rei // back + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_UNPRIV(0x00A0) +CallPal_OpcDecA0: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A1) +CallPal_OpcDecA1: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A2) +CallPal_OpcDecA2: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A3) +CallPal_OpcDecA3: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A4) +CallPal_OpcDecA4: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A5) +CallPal_OpcDecA5: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A6) +CallPal_OpcDecA6: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A7) +CallPal_OpcDecA7: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A8) +CallPal_OpcDecA8: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A9) +CallPal_OpcDecA9: + br r31, osfpal_calpal_opcdec + + +// .sbttl "gentrap - PALcode for gentrap instruction" +//+ +// CALL_PAL_gentrap: +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + + CALL_PAL_UNPRIV(0x00AA) +// unsupported in Hudson code .. pboyle Nov/95 +CALL_PAL_gentrap: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, CALL_PAL_gentrap_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +CALL_PAL_gentrap_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_gentrap, r16// set a0 + + stq r17, osfsf_a1(sp) // a1 + br r31, bpt_bchk_common // out of call_pal space + + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_UNPRIV(0x00AB) +CallPal_OpcDecAB: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AC) +CallPal_OpcDecAC: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AD) +CallPal_OpcDecAD: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AE) +CallPal_OpcDecAE: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AF) +CallPal_OpcDecAF: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B0) +CallPal_OpcDecB0: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B1) +CallPal_OpcDecB1: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B2) +CallPal_OpcDecB2: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B3) +CallPal_OpcDecB3: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B4) +CallPal_OpcDecB4: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B5) +CallPal_OpcDecB5: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B6) +CallPal_OpcDecB6: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B7) +CallPal_OpcDecB7: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B8) +CallPal_OpcDecB8: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B9) +CallPal_OpcDecB9: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BA) +CallPal_OpcDecBA: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BB) +CallPal_OpcDecBB: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BC) +CallPal_OpcDecBC: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BD) +CallPal_OpcDecBD: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BE) +CallPal_OpcDecBE: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BF) +CallPal_OpcDecBF: + // MODIFIED BY EGH 2/25/04 + br r31, copypal_impl + + +/*======================================================================*/ +/* OSF/1 CALL_PAL CONTINUATION AREA */ +/*======================================================================*/ + + .text 2 + + . = 0x4000 + + +// .sbttl "Continuation of MTPR_PERFMON" + ALIGN_BLOCK +#if perfmon_debug == 0 + // "real" performance monitoring code +// mux ctl +perfmon_muxctl: + lda r8, 1(r31) // get a 1 + sll r8, pmctr_v_sel0, r8 // move to sel0 position + or r8, ((0xf<<pmctr_v_sel1) | (0xf<<pmctr_v_sel2)), r8 // build mux select mask + and r17, r8, r25 // isolate pmctr mux select bits + mfpr r0, ev5__pmctr + bic r0, r8, r0 // clear old mux select bits + or r0,r25, r25 // or in new mux select bits + mtpr r25, ev5__pmctr + + // ok, now tackle cbox mux selects + ldah r14, 0xfff0(r31) + zap r14, 0xE0, r14 // Get Cbox IPR base +//orig get_bc_ctl_shadow r16 // bc_ctl returned in lower longword +// adapted from ev5_pal_macros.mar + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + RESTORE_SHADOW(r16,CNS_Q_BC_CTL,r16); + + lda r8, 0x3F(r31) // build mux select mask + sll r8, bc_ctl_v_pm_mux_sel, r8 + + and r17, r8, r25 // isolate bc_ctl mux select bits + bic r16, r8, r16 // isolate old mux select bits + or r16, r25, r25 // create new bc_ctl + mb // clear out cbox for future ipr write + stqp r25, ev5__bc_ctl(r14) // store to cbox ipr + mb // clear out cbox for future ipr write + +//orig update_bc_ctl_shadow r25, r16 // r25=value, r16-overwritten with adjusted impure ptr +// adapted from ev5_pal_macros.mar + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + SAVE_SHADOW(r25,CNS_Q_BC_CTL,r16); + + br r31, perfmon_success + + +// requested to disable perf monitoring +perfmon_dis: + mfpr r14, ev5__pmctr // read ibox pmctr ipr +perfmon_dis_ctr0: // and begin with ctr0 + blbc r17, perfmon_dis_ctr1 // do not disable ctr0 + lda r8, 3(r31) + sll r8, pmctr_v_ctl0, r8 + bic r14, r8, r14 // disable ctr0 +perfmon_dis_ctr1: + srl r17, 1, r17 + blbc r17, perfmon_dis_ctr2 // do not disable ctr1 + lda r8, 3(r31) + sll r8, pmctr_v_ctl1, r8 + bic r14, r8, r14 // disable ctr1 +perfmon_dis_ctr2: + srl r17, 1, r17 + blbc r17, perfmon_dis_update // do not disable ctr2 + lda r8, 3(r31) + sll r8, pmctr_v_ctl2, r8 + bic r14, r8, r14 // disable ctr2 +perfmon_dis_update: + mtpr r14, ev5__pmctr // update pmctr ipr +//;the following code is not needed for ev5 pass2 and later, but doesn't hurt anything to leave in +// adapted from ev5_pal_macros.mar +//orig get_pmctr_ctl r8, r25 // pmctr_ctl bit in r8. adjusted impure pointer in r25 + mfpr r25, pt_impure + lda r25, CNS_Q_IPR(r25) + RESTORE_SHADOW(r8,CNS_Q_PM_CTL,r25); + + lda r17, 0x3F(r31) // build mask + sll r17, pmctr_v_ctl2, r17 // shift mask to correct position + and r14, r17, r14 // isolate ctl bits + bic r8, r17, r8 // clear out old ctl bits + or r14, r8, r14 // create shadow ctl bits +//orig store_reg1 pmctr_ctl, r14, r25, ipr=1 // update pmctr_ctl register +//adjusted impure pointer still in r25 + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r25); + + br r31, perfmon_success + + +// requested to enable perf monitoring +//;the following code can be greatly simplified for pass2, but should work fine as is. + + +perfmon_enclr: + lda r9, 1(r31) // set enclr flag + br perfmon_en_cont + +perfmon_en: + bis r31, r31, r9 // clear enclr flag + +perfmon_en_cont: + mfpr r8, pt_pcbb // get PCB base +//orig get_pmctr_ctl r25, r25 + mfpr r25, pt_impure + lda r25, CNS_Q_IPR(r25) + RESTORE_SHADOW(r25,CNS_Q_PM_CTL,r25); + + ldqp r16, osfpcb_q_fen(r8) // read DAT/PME/FEN quadword + mfpr r14, ev5__pmctr // read ibox pmctr ipr + srl r16, osfpcb_v_pme, r16 // get pme bit + mfpr r13, icsr + and r16, 1, r16 // isolate pme bit + + // this code only needed in pass2 and later +//orig sget_addr r12, 1<<icsr_v_pmp, r31 + lda r12, 1<<icsr_v_pmp(r31) // pb + bic r13, r12, r13 // clear pmp bit + sll r16, icsr_v_pmp, r12 // move pme bit to icsr<pmp> position + or r12, r13, r13 // new icsr with icsr<pmp> bit set/clear + ev5_pass2 mtpr r13, icsr // update icsr + +#if ev5_p1 != 0 + lda r12, 1(r31) + cmovlbc r25, r12, r16 // r16<0> set if either pme=1 or sprocess=0 (sprocess in bit 0 of r25) +#else + bis r31, 1, r16 // set r16<0> on pass2 to update pmctr always (icsr provides real enable) +#endif + + sll r25, 6, r25 // shift frequency bits into pmctr_v_ctl positions + bis r14, r31, r13 // copy pmctr + +perfmon_en_ctr0: // and begin with ctr0 + blbc r17, perfmon_en_ctr1 // do not enable ctr0 + + blbc r9, perfmon_en_noclr0 // enclr flag set, clear ctr0 field + lda r8, 0xffff(r31) + zapnot r8, 3, r8 // ctr0<15:0> mask + sll r8, pmctr_v_ctr0, r8 + bic r14, r8, r14 // clear ctr bits + bic r13, r8, r13 // clear ctr bits + +perfmon_en_noclr0: +//orig get_addr r8, 3<<pmctr_v_ctl0, r31 + LDLI(r8, (3<<pmctr_v_ctl0)) + and r25, r8, r12 //isolate frequency select bits for ctr0 + bic r14, r8, r14 // clear ctl0 bits in preparation for enabling + or r14,r12,r14 // or in new ctl0 bits + +perfmon_en_ctr1: // enable ctr1 + srl r17, 1, r17 // get ctr1 enable + blbc r17, perfmon_en_ctr2 // do not enable ctr1 + + blbc r9, perfmon_en_noclr1 // if enclr flag set, clear ctr1 field + lda r8, 0xffff(r31) + zapnot r8, 3, r8 // ctr1<15:0> mask + sll r8, pmctr_v_ctr1, r8 + bic r14, r8, r14 // clear ctr bits + bic r13, r8, r13 // clear ctr bits + +perfmon_en_noclr1: +//orig get_addr r8, 3<<pmctr_v_ctl1, r31 + LDLI(r8, (3<<pmctr_v_ctl1)) + and r25, r8, r12 //isolate frequency select bits for ctr1 + bic r14, r8, r14 // clear ctl1 bits in preparation for enabling + or r14,r12,r14 // or in new ctl1 bits + +perfmon_en_ctr2: // enable ctr2 + srl r17, 1, r17 // get ctr2 enable + blbc r17, perfmon_en_return // do not enable ctr2 - return + + blbc r9, perfmon_en_noclr2 // if enclr flag set, clear ctr2 field + lda r8, 0x3FFF(r31) // ctr2<13:0> mask + sll r8, pmctr_v_ctr2, r8 + bic r14, r8, r14 // clear ctr bits + bic r13, r8, r13 // clear ctr bits + +perfmon_en_noclr2: +//orig get_addr r8, 3<<pmctr_v_ctl2, r31 + LDLI(r8, (3<<pmctr_v_ctl2)) + and r25, r8, r12 //isolate frequency select bits for ctr2 + bic r14, r8, r14 // clear ctl2 bits in preparation for enabling + or r14,r12,r14 // or in new ctl2 bits + +perfmon_en_return: + cmovlbs r16, r14, r13 // if pme enabled, move enables into pmctr + // else only do the counter clears + mtpr r13, ev5__pmctr // update pmctr ipr + +//;this code not needed for pass2 and later, but does not hurt to leave it in + lda r8, 0x3F(r31) +//orig get_pmctr_ctl r25, r12 // read pmctr ctl; r12=adjusted impure pointer + mfpr r12, pt_impure + lda r12, CNS_Q_IPR(r12) + RESTORE_SHADOW(r25,CNS_Q_PM_CTL,r12); + + sll r8, pmctr_v_ctl2, r8 // build ctl mask + and r8, r14, r14 // isolate new ctl bits + bic r25, r8, r25 // clear out old ctl value + or r25, r14, r14 // create new pmctr_ctl +//orig store_reg1 pmctr_ctl, r14, r12, ipr=1 + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r12); // r12 still has the adjusted impure ptr + + br r31, perfmon_success + + +// options... +perfmon_ctl: + +// set mode +//orig get_pmctr_ctl r14, r12 // read shadow pmctr ctl; r12=adjusted impure pointer + mfpr r12, pt_impure + lda r12, CNS_Q_IPR(r12) + RESTORE_SHADOW(r14,CNS_Q_PM_CTL,r12); + +//orig get_addr r8, (1<<pmctr_v_killu) | (1<<pmctr_v_killp) | (1<<pmctr_v_killk), r31 // build mode mask for pmctr register + LDLI(r8, ((1<<pmctr_v_killu) | (1<<pmctr_v_killp) | (1<<pmctr_v_killk))) + mfpr r0, ev5__pmctr + and r17, r8, r25 // isolate pmctr mode bits + bic r0, r8, r0 // clear old mode bits + or r0, r25, r25 // or in new mode bits + mtpr r25, ev5__pmctr + +//;the following code will only be used in pass2, but should not hurt anything if run in pass1. + mfpr r8, icsr + lda r25, 1<<icsr_v_pma(r31) // set icsr<pma> if r17<0>=0 + bic r8, r25, r8 // clear old pma bit + cmovlbs r17, r31, r25 // and clear icsr<pma> if r17<0>=1 + or r8, r25, r8 + ev5_pass2 mtpr r8, icsr // 4 bubbles to hw_rei + mfpr r31, pt0 // pad icsr write + mfpr r31, pt0 // pad icsr write + +//;the following code not needed for pass2 and later, but should work anyway. + bis r14, 1, r14 // set for select processes + blbs r17, perfmon_sp // branch if select processes + bic r14, 1, r14 // all processes +perfmon_sp: +//orig store_reg1 pmctr_ctl, r14, r12, ipr=1 // update pmctr_ctl register + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r12); // r12 still has the adjusted impure ptr + br r31, perfmon_success + +// counter frequency select +perfmon_freq: +//orig get_pmctr_ctl r14, r12 // read shadow pmctr ctl; r12=adjusted impure pointer + mfpr r12, pt_impure + lda r12, CNS_Q_IPR(r12) + RESTORE_SHADOW(r14,CNS_Q_PM_CTL,r12); + + lda r8, 0x3F(r31) +//orig sll r8, pmctr_ctl_v_frq2, r8 // build mask for frequency select field +// I guess this should be a shift of 4 bits from the above control register structure .. pb +#define pmctr_ctl_v_frq2_SHIFT 4 + sll r8, pmctr_ctl_v_frq2_SHIFT, r8 // build mask for frequency select field + + and r8, r17, r17 + bic r14, r8, r14 // clear out old frequency select bits + + or r17, r14, r14 // or in new frequency select info +//orig store_reg1 pmctr_ctl, r14, r12, ipr=1 // update pmctr_ctl register + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r12); // r12 still has the adjusted impure ptr + + br r31, perfmon_success + +// read counters +perfmon_rd: + mfpr r0, ev5__pmctr + or r0, 1, r0 // or in return status + hw_rei // back to user + +// write counters +perfmon_wr: + mfpr r14, ev5__pmctr + lda r8, 0x3FFF(r31) // ctr2<13:0> mask + sll r8, pmctr_v_ctr2, r8 + +//orig get_addr r9, 0xFFFFFFFF, r31, verify=0 // ctr2<15:0>,ctr1<15:0> mask + LDLI(r9, (0xFFFFFFFF)) + sll r9, pmctr_v_ctr1, r9 + or r8, r9, r8 // or ctr2, ctr1, ctr0 mask + bic r14, r8, r14 // clear ctr fields + and r17, r8, r25 // clear all but ctr fields + or r25, r14, r14 // write ctr fields + mtpr r14, ev5__pmctr // update pmctr ipr + + mfpr r31, pt0 // pad pmctr write (needed only to keep PVC happy) + +perfmon_success: + or r31, 1, r0 // set success + hw_rei // back to user + +perfmon_unknown: + or r31, r31, r0 // set fail + hw_rei // back to user + +#else + +// end of "real code", start of debug code + +//+ +// Debug environment: +// (in pass2, always set icsr<pma> to ensure master counter enable is on) +// R16 = 0 Write to on-chip performance monitor ipr +// r17 = on-chip ipr +// r0 = return value of read of on-chip performance monitor ipr +// R16 = 1 Setup Cbox mux selects +// r17 = Cbox mux selects in same position as in bc_ctl ipr. +// r0 = return value of read of on-chip performance monitor ipr +// +//- +pal_perfmon_debug: + mfpr r8, icsr + lda r9, 1<<icsr_v_pma(r31) + bis r8, r9, r8 + mtpr r8, icsr + + mfpr r0, ev5__pmctr // read old value + bne r16, cbox_mux_sel + + mtpr r17, ev5__pmctr // update pmctr ipr + br r31, end_pm + +cbox_mux_sel: + // ok, now tackle cbox mux selects + ldah r14, 0xfff0(r31) + zap r14, 0xE0, r14 // Get Cbox IPR base +//orig get_bc_ctl_shadow r16 // bc_ctl returned + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + RESTORE_SHADOW(r16,CNS_Q_BC_CTL,r16); + + lda r8, 0x3F(r31) // build mux select mask + sll r8, BC_CTL_V_PM_MUX_SEL, r8 + + and r17, r8, r25 // isolate bc_ctl mux select bits + bic r16, r8, r16 // isolate old mux select bits + or r16, r25, r25 // create new bc_ctl + mb // clear out cbox for future ipr write + stqp r25, ev5__bc_ctl(r14) // store to cbox ipr + mb // clear out cbox for future ipr write +//orig update_bc_ctl_shadow r25, r16 // r25=value, r16-overwritten with adjusted impure ptr + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + SAVE_SHADOW(r25,CNS_Q_BC_CTL,r16); + +end_pm: hw_rei + +#endif + + +//;The following code is a workaround for a cpu bug where Istream prefetches to +//;super-page address space in user mode may escape off-chip. +#if spe_fix != 0 + + ALIGN_BLOCK +hw_rei_update_spe: + mfpr r12, pt_misc // get previous mode + srl r11, osfps_v_mode, r10 // isolate current mode bit + and r10, 1, r10 + extbl r12, 7, r8 // get previous mode field + and r8, 1, r8 // isolate previous mode bit + cmpeq r10, r8, r8 // compare previous and current modes + beq r8, hw_rei_update_spe_5_ + hw_rei // if same, just return + +hw_rei_update_spe_5_: + +#if fill_err_hack != 0 + + fill_error_hack +#endif + + mfpr r8, icsr // get current icsr value + ldah r9, (2<<(icsr_v_spe-16))(r31) // get spe bit mask + bic r8, r9, r8 // disable spe + xor r10, 1, r9 // flip mode for new spe bit + sll r9, icsr_v_spe+1, r9 // shift into position + bis r8, r9, r8 // enable/disable spe + lda r9, 1(r31) // now update our flag + sll r9, pt_misc_v_cm, r9 // previous mode saved bit mask + bic r12, r9, r12 // clear saved previous mode + sll r10, pt_misc_v_cm, r9 // current mode saved bit mask + bis r12, r9, r12 // set saved current mode + mtpr r12, pt_misc // update pt_misc + mtpr r8, icsr // update icsr + +#if osf_chm_fix != 0 + + + blbc r10, hw_rei_update_spe_10_ // branch if not user mode + + mb // ensure no outstanding fills + lda r12, 1<<dc_mode_v_dc_ena(r31) // User mode + mtpr r12, dc_mode // Turn on dcache + mtpr r31, dc_flush // and flush it + br r31, pal_ic_flush + +hw_rei_update_spe_10_: mfpr r9, pt_pcbb // Kernel mode + ldqp r9, osfpcb_q_Fen(r9) // get FEN + blbc r9, pal_ic_flush // return if FP disabled + mb // ensure no outstanding fills + mtpr r31, dc_mode // turn off dcache +#endif + + + br r31, pal_ic_flush // Pal restriction - must flush Icache if changing ICSR<SPE> +#endif + + +copypal_impl: + mov r16, r0 + and r16, 63, r8 + and r17, 63, r9 + bis r8, r9, r8 + bne r8, cache_copy_done + bic r18, 63, r8 + and r18, 63, r18 + beq r8, cache_copy_done + cache_loop: + ldf f17, 0(r16) + stf f17, 0(r16) + addq r17, 64, r17 + addq r16, 64, r16 + subq r8, 64, r8 + bne r8, cache_loop +cache_copy_done: + ble r18, finished #if len <=0 we are finished + ldq_u r8, 0(r17) + xor r17, r16, r9 + and r9, 7, r9 + and r16, 7, r10 + bne r9, unaligned + beq r10, aligned + ldq_u r9, 0(r16) + addq r18, r10, r18 + mskqh r8, r17, r8 + mskql r9, r17, r9 + bis r8, r9, r8 +aligned: + subq r18, 1, r10 + bic r10, 7, r10 + and r18, 7, r18 + beq r10, aligned_done +loop: + stq_u r8, 0(r16) + ldq_u r8, 8(r17) + subq r10, 8, r10 + lda r16,8(r16) + lda r17,8(r17) + bne r10, loop +aligned_done: + bne r18, few_left + stq_u r8, 0(r16) + br r31, finished + few_left: + mskql r8, r18, r10 + ldq_u r9, 0(r16) + mskqh r9, r18, r9 + bis r10, r9, r10 + stq_u r10, 0(r16) + br r31, finished +unaligned: + addq r17, r18, r25 + cmpule r18, 8, r9 + bne r9, unaligned_few_left + beq r10, unaligned_dest_aligned + and r16, 7, r10 + subq r31, r10, r10 + addq r10, 8, r10 + ldq_u r9, 7(r17) + extql r8, r17, r8 + extqh r9, r17, r9 + bis r8, r9, r12 + insql r12, r16, r12 + ldq_u r13, 0(r16) + mskql r13, r16, r13 + bis r12, r13, r12 + stq_u r12, 0(r16) + addq r16, r10, r16 + addq r17, r10, r17 + subq r18, r10, r18 + ldq_u r8, 0(r17) +unaligned_dest_aligned: + subq r18, 1, r10 + bic r10, 7, r10 + and r18, 7, r18 + beq r10, unaligned_partial_left +unaligned_loop: + ldq_u r9, 7(r17) + lda r17, 8(r17) + extql r8, r17, r12 + extqh r9, r17, r13 + subq r10, 8, r10 + bis r12, r13, r13 + stq r13, 0(r16) + lda r16, 8(r16) + beq r10, unaligned_second_partial_left + ldq_u r8, 7(r17) + lda r17, 8(r17) + extql r9, r17, r12 + extqh r8, r17, r13 + bis r12, r13, r13 + subq r10, 8, r10 + stq r13, 0(r16) + lda r16, 8(r16) + bne r10, unaligned_loop +unaligned_partial_left: + mov r8, r9 +unaligned_second_partial_left: + ldq_u r8, -1(r25) + extql r9, r17, r9 + extqh r8, r17, r8 + bis r8, r9, r8 + bne r18, few_left + stq_u r8, 0(r16) + br r31, finished +unaligned_few_left: + ldq_u r9, -1(r25) + extql r8, r17, r8 + extqh r9, r17, r9 + bis r8, r9, r8 + insqh r8, r16, r9 + insql r8, r16, r8 + lda r12, -1(r31) + mskql r12, r18, r13 + cmovne r13, r13, r12 + insqh r12, r16, r13 + insql r12, r16, r12 + addq r16, r18, r10 + ldq_u r14, 0(r16) + ldq_u r25, -1(r10) + bic r14, r12, r14 + bic r25, r13, r25 + and r8, r12, r8 + and r9, r13, r9 + bis r8, r14, r8 + bis r9, r25, r9 + stq_u r9, -1(r10) + stq_u r8, 0(r16) +finished: + hw_rei diff --git a/system/alpha/palcode/osfpal_cache_copy_unaligned.s b/system/alpha/palcode/osfpal_cache_copy_unaligned.s new file mode 100644 index 000000000..e60c3ccfc --- /dev/null +++ b/system/alpha/palcode/osfpal_cache_copy_unaligned.s @@ -0,0 +1,5253 @@ +// modified to use the Hudson style "impure.h" instead of ev5_impure.sdl +// since we don't have a mechanism to expand the data structures.... pb Nov/95 + +// build_fixed_image: not sure what means +// real_mm to be replaced during rewrite +// remove_save_state remove_restore_state can be remooved to save space ?? + + +#include "ev5_defs.h" +#include "ev5_impure.h" +#include "ev5_alpha_defs.h" +#include "ev5_paldef.h" +#include "ev5_osfalpha_defs.h" +#include "fromHudsonMacros.h" +#include "fromHudsonOsf.h" +#include "dc21164FromGasSources.h" + +#ifdef SIMOS +#define DEBUGSTORE(c) nop +#else +#define DEBUGSTORE(c) \ + lda r13, c(zero) ; \ + bsr r25, debugstore +#endif + +#define DEBUG_EXC_ADDR()\ + bsr r25, put_exc_addr; \ + DEBUGSTORE(13) ; \ + DEBUGSTORE(10) + +#define egore 0 +#define acore 0 +#define beh_model 0 +#define ev5_p2 1 +#define ev5_p1 0 +#define ldvpte_bug_fix 1 +#define osf_chm_fix 0 + +// Do we want to do this?? pb +#define spe_fix 0 +// Do we want to do this?? pb +#define build_fixed_image 0 + +#define ev5_pass2 +#define enable_p4_fixups 0 +#define osf_svmin 1 +#define enable_physical_console 0 +#define fill_err_hack 0 +#define icflush_on_tbix 0 +#define max_cpuid 1 +#define perfmon_debug 0 +#define rawhide_system 0 +#define rax_mode 0 + + +// This is the fix for the user-mode super page references causing the machine to crash. +#if (spe_fix == 1) && (build_fixed_image==1) +#define hw_rei_spe br r31, hw_rei_update_spe +#else +#define hw_rei_spe hw_rei +#endif + + +// redefine a few of the distribution-code names to match the Hudson gas names. +// opcodes +#define ldqp ldq_p +#define stqp stq_p +#define ldlp ldl_p +#define stlp stl_p + +#define r0 $0 +#define r1 $1 +#define r2 $2 +#define r3 $3 +#define r4 $4 +#define r5 $5 +#define r6 $6 +#define r7 $7 +#define r8 $8 +#define r9 $9 +#define r10 $10 +#define r11 $11 +#define r12 $12 +#define r13 $13 +#define r14 $14 +#define r15 $15 +#define r16 $16 +#define r17 $17 +#define r18 $18 +#define r19 $19 +#define r20 $20 +#define r21 $21 +#define r22 $22 +#define r23 $23 +#define r24 $24 +#define r25 $25 +#define r26 $26 +#define r27 $27 +#define r28 $28 +#define r29 $29 +#define r30 $30 +#define r31 $31 + +// .title "EV5 OSF PAL" +// .ident "V1.18" +// +//**************************************************************************** +//* * +//* Copyright (c) 1992, 1993, 1994, 1995 * +//* by DIGITAL Equipment Corporation, Maynard, Mass. * +//* * +//* This software is furnished under a license and may be used and copied * +//* only in accordance with the terms of such license and with the * +//* inclusion of the above copyright notice. This software or any other * +//* copies thereof may not be provided or otherwise made available to any * +//* other person. No title to and ownership of the software is hereby * +//* transferred. * +//* * +//* The information in this software is subject to change without notice * +//* and should not be construed as a commitment by DIGITAL Equipment * +//* Corporation. * +//* * +//* DIGITAL assumes no responsibility for the use or reliability of its * +//* software on equipment which is not supplied by DIGITAL. * +//* * +//**************************************************************************** + +// .sbttl "Edit History" +//+ +// Who Rev When What +// ------------ --- ----------- -------------------------------- +// DB 0.0 03-Nov-1992 Start +// DB 0.1 28-Dec-1992 add swpctx +// DB 0.2 05-Jan-1993 Bug: PVC found mtpr dtb_CM -> virt ref bug +// DB 0.3 11-Jan-1993 rearrange trap entry points +// DB 0.4 01-Feb-1993 add tbi +// DB 0.5 04-Feb-1993 real MM, kludge reset flow, kludge swppal +// DB 0.6 09-Feb-1993 Bug: several stack pushers used r16 for pc (should be r14) +// DB 0.7 10-Feb-1993 Bug: pushed wrong PC (+8) on CALL_PAL OPCDEC +// Bug: typo on register number for store in wrunique +// Bug: rti to kern uses r16 as scratch +// Bug: callsys saving wrong value in pt_usp +// DB 0.8 16-Feb-1993 PVC: fix possible pt write->read bug in wrkgp, wrusp +// DB 0.9 18-Feb-1993 Bug: invalid_dpte_handler shifted pte twice +// Bug: rti stl_c could corrupt the stack +// Bug: unaligned returning wrong value in r17 (or should be and) +// DB 0.10 19-Feb-1993 Add draina, rd/wrmces, cflush, cserve, interrupt +// DB 0.11 23-Feb-1993 Turn caches on in reset flow +// DB 0.12 10-Mar-1993 Bug: wrong value for icsr for FEN in kern mode flow +// DB 0.13 15-Mar-1993 Bug: wrong value pushed for PC in invalid_dpte_handler if stack push tbmisses +// DB 0.14 23-Mar-1993 Add impure pointer paltemp, reshuffle some other paltemps to match VMS +// DB 0.15 15-Apr-1993 Combine paltemps for WHAMI and MCES +// DB 0.16 12-May-1993 Update reset +// New restriction: no mfpr exc_addr in cycle 1 of call_pal flows +// Bug: in wrmces, not clearing DPC, DSC +// Update swppal +// Add pal bugchecks, pal_save_state, pal_restore_state +// DB 0.17 24-May-1993 Add dfault_in_pal flow; fixup stack builder to have common state for pc/ps. +// New restriction: No hw_rei_stall in 0,1,2 after mtpr itb_asn +// DB 0.18 26-May-1993 PVC fixes +// JM 0.19 01-jul-1993 Bug: OSFPAL_CALPAL_OPCDEC, TRAP_OPCDEC -- move mt exc_addr after stores +// JM 0.20 07-jul-1993 Update cns_ and mchk_ names for impure.mar conversion to .sdl +// Bug: exc_addr was being loaded before stores that could dtb_miss in the following +// routines: TRAP_FEN,FEN_TO_OPCDEC,CALL_PAL_CALLSYS,RTI_TO_KERN +// JM 0.21 26-jul-1993 Bug: move exc_addr load after ALL stores in the following routines: +// TRAP_IACCVIO::,TRAP_OPCDEC::,TRAP_ARITH::,TRAP_FEN:: +// dfault_trap_cont:,fen_to_opcdec:,invalid_dpte_handler: +// osfpal_calpal_opcdec:,CALL_PAL_callsys::,TRAP_UNALIGN:: +// Bugs from PVC: trap_unalign - mt pt0 ->mf pt0 within 2 cycles +// JM 0.22 28-jul-1993 Add WRIPIR instruction +// JM 0.23 05-aug-1993 Bump version number for release +// JM 0.24 11-aug-1993 Bug: call_pal_swpipl - palshadow write -> hw_rei violation +// JM 0.25 09-sep-1993 Disable certain "hidden" pvc checks in call_pals; +// New restriction: No hw_rei_stall in 0,1,2,3,4 after mtpr itb_asn - affects HALT(raxmode), +// and SWPCTX +// JM 0.26 07-oct-1993 Re-implement pal_version +// JM 0.27 12-oct-1993 One more time: change pal_version format to conform to SRM +// JM 0.28 14-oct-1993 Change ic_flush routine to pal_ic_flush +// JM 0.29 19-oct-1993 BUG(?): dfault_in_pal: use exc_addr to check for dtbmiss,itbmiss check instead +// of mm_stat<opcode>. mm_stat contains original opcode, not hw_ld. +// JM 0.30 28-oct-1993 BUG: PVC violation - mf exc_addr in first cycles of call_pal in rti,retsys +// JM 0.31 15-nov-1993 BUG: WRFEN trashing r0 +// JM 0.32 21-nov-1993 BUG: dtb_ldq,itb_ldq (used in dfault_in_pal) not defined when real_mm=0 +// JM 0.33 24-nov-1993 save/restore_state - +// BUG: use ivptbr to restore mvptbr +// BUG: adjust hw_ld/st base/offsets to accomodate 10-bit offset limit +// CHANGE: Load 2 pages into dtb to accomodate compressed logout area/multiprocessors +// JM 0.34 20-dec-1993 BUG: set r11<mode> to kernel for ksnv halt case +// BUG: generate ksnv halt when tb miss on kernel stack accesses +// save exc_addr in r14 for invalid_dpte stack builder +// JM 0.35 30-dec-1993 BUG: PVC violation in trap_arith - mt exc_sum in shadow of store with mf exc_mask in +// the same shadow +// JM 0.36 6-jan-1994 BUG: fen_to_opcdec - savePC should be PC+4, need to save old PS, update new PS +// New palcode restiction: mt icsr<fpe,hwe> --> 3 bubbles to hw_rei --affects wrfen +// JM 0.37 25-jan-1994 BUG: PVC violations in restore_state - mt dc_mode/maf_mode ->mbox instructions +// Hide impure area manipulations in macros +// BUG: PVC violation in save and restore state-- move mt icsr out of shadow of ld/st +// Add some pvc_violate statements +// JM 0.38 1-feb-1994 Changes to save_state: save pt1; don't save r31,f31; update comments to reflect reality; +// Changes to restore_state: restore pt1, icsr; don't restore r31,f31; update comments +// Add code to ensure fen bit set in icsr before ldt +// conditionally compile rax_more_reset out. +// move ldqp,stqp macro definitions to ev5_pal_macros.mar and add .mcall's for them here +// move rax reset stuff to ev5_osf_system_pal.m64 +// JM 0.39 7-feb-1994 Move impure pointer to pal scratch space. Use former pt_impure for bc_ctl shadow +// and performance monitoring bits +// Change to save_state routine to save more iprs. +// JM 0.40 19-feb-1994 Change algorithm in save/restore_state routines; add f31,r31 back in +// JM 0.41 21-feb-1994 Add flags to compile out save/restore state (not needed in some systems) +// remove_save_state,remove_restore_state;fix new pvc violation in save_state +// JM 0.42 22-feb-1994 BUG: save_state overwriting r3 +// JM 0.43 24-feb-1994 BUG: save_state saving wrong icsr +// JM 0.44 28-feb-1994 Remove ic_flush from wr_tbix instructions +// JM 0.45 15-mar-1994 BUG: call_pal_tbi trashes a0 prior to range check (instruction order problem) +// New pal restriction in pal_restore_state: icsr<fpe>->floating instr = 3 bubbles +// Add exc_sum and exc_mask to pal_save_state (not restore) +// JM 0.46 22-apr-1994 Move impure pointer back into paltemp; Move bc_ctl shadow and pmctr_ctl into impure +// area. +// Add performance counter support to swpctx and wrperfmon +// JM 0.47 9-may-1994 Bump version # (for ev5_osf_system_pal.m64 sys_perfmon fix) +// JM 0.48 13-jun-1994 BUG: trap_interrupt --> put new ev5 ipl at 30 for all osfipl6 interrupts +// JM 0.49 8-jul-1994 BUG: In the unlikely (impossible?) event that the branch to pal_pal_bug_check is +// taken in the interrupt flow, stack is pushed twice. +// SWPPAL - update to support ECO 59 to allow 0 as a valid address +// Add itb flush to save/restore state routines +// Change hw_rei to hw_rei_stall in ic_flush routine. Shouldn't be necessary, but +// conforms to itbia restriction. +// Added enable_physical_console flag (for enter/exit console routines only) +// JM 0.50 29-jul-1994 Add code to dfault & invalid_dpte_handler to ignore exceptions on a +// load to r31/f31. changed dfault_fetch_err to dfault_fetch_ldr31_err and +// nmiss_fetch_err to nmiss_fetch_ldr31_err. +// JM 1.00 1-aug-1994 Add pass2 support (swpctx) +// JM 1.01 2-aug-1994 swppal now passes bc_ctl/bc_config in r1/r2 +// JM 1.02 15-sep-1994 BUG: swpctx missing shift of pme bit to correct position in icsr (pass2) +// Moved perfmon code here from system file. +// BUG: pal_perfmon - enable function not saving correct enables when pme not set (pass1) +// JM 1.03 3-oct-1994 Added (pass2 only) code to wrperfmon enable function to look at pme bit. +// JM 1.04 14-oct-1994 BUG: trap_interrupt - ISR read (and saved) before INTID -- INTID can change +// after ISR read, but we won't catch the ISR update. reverse order +// JM 1.05 17-nov-1994 Add code to dismiss UNALIGN trap if LD r31/F31 +// JM 1.06 28-nov-1994 BUG: missing mm_stat shift for store case in trap_unalign (new bug due to "dismiss" code) +// JM 1.07 1-dec-1994 EV5 PASS1,2,3 BUG WORKAROUND: Add flag LDVPTE_BUG_FIX. In DTBMISS_DOUBLE, branch to +// DTBMISS_SINGLE if not in palmode. +// JM 1.08 9-jan-1995 Bump version number for change to EV5_OSF_SYSTEM_PAL.M64 - ei_stat fix in mchk logout frame +// JM 1.09 2-feb-1995 Add flag "spe_fix" and accompanying code to workaround pre-pass4 bug: Disable Ibox +// superpage mode in User mode and re-enable in kernel mode. +// EV5_OSF_SYSTEM_PAL.M64 and EV5_PALDEF.MAR (added pt_misc_v_cm) also changed to support this. +// JM 1.10 24-feb-1995 Set ldvpte_bug_fix regardless of ev5 pass. set default to ev5_p2 +// ES 1.11 10-mar-1995 Add flag "osf_chm_fix" to enable dcache in user mode only to avoid +// cpu bug. +// JM 1.12 17-mar-1995 BUG FIX: Fix F0 corruption problem in pal_restore_state +// ES 1.13 17-mar-1995 Refine osf_chm_fix +// ES 1.14 20-mar-1995 Don't need as many stalls before hw_rei_stall in chm_fix +// ES 1.15 21-mar-1995 Add a stall to avoid a pvc violation in pal_restore_state +// Force pvc checking of exit_console +// ES 1.16 26-apr-1995 In the wrperfmon disable function, correct meaning of R17<2:0> to ctl2,ctl2,ctl0 +// ES 1.17 01-may-1995 In hw_rei_update_spe code, in the osf_chm fix, use bic and bis (self-correcting) +// instead of xor to maintain previous mode in pt_misc +// ES 1.18 14-jul-1995 In wrperfmon enable on pass2, update pmctr even if current process does +// not have pme set. The bits in icsr maintain the master enable state. +// In sys_reset, add icsr<17>=1 for ev56 byte/word eco enable +// +#define vmaj 1 +#define vmin 18 +#define vms_pal 1 +#define osf_pal 2 +#define pal_type osf_pal +#define osfpal_version_l ((pal_type<<16) | (vmaj<<8) | (vmin<<0)) +//- + +// .sbttl "PALtemp register usage" + +//+ +// The EV5 Ibox holds 24 PALtemp registers. This maps the OSF PAL usage +// for these PALtemps: +// +// pt0 local scratch +// pt1 local scratch +// pt2 entUna pt_entUna +// pt3 CPU specific impure area pointer pt_impure +// pt4 memory management temp +// pt5 memory management temp +// pt6 memory management temp +// pt7 entIF pt_entIF +// pt8 intmask pt_intmask +// pt9 entSys pt_entSys +// pt10 +// pt11 entInt pt_entInt +// pt12 entArith pt_entArith +// pt13 reserved for system specific PAL +// pt14 reserved for system specific PAL +// pt15 reserved for system specific PAL +// pt16 MISC: scratch ! WHAMI<7:0> ! 0 0 0 MCES<4:0> pt_misc, pt_whami, pt_mces +// pt17 sysval pt_sysval +// pt18 usp pt_usp +// pt19 ksp pt_ksp +// pt20 PTBR pt_ptbr +// pt21 entMM pt_entMM +// pt22 kgp pt_kgp +// pt23 PCBB pt_pcbb +// +//- + +// .sbttl "PALshadow register usage" +// +//+ +// +// EV5 shadows R8-R14 and R25 when in PALmode and ICSR<shadow_enable> = 1. +// This maps the OSF PAL usage of R8 - R14 and R25: +// +// r8 ITBmiss/DTBmiss scratch +// r9 ITBmiss/DTBmiss scratch +// r10 ITBmiss/DTBmiss scratch +// r11 PS +// r12 local scratch +// r13 local scratch +// r14 local scratch +// r25 local scratch +// +// +//- + +// .sbttl "ALPHA symbol definitions" +// _OSF_PSDEF GLOBAL +// _OSF_PTEDEF GLOBAL +// _OSF_VADEF GLOBAL +// _OSF_PCBDEF GLOBAL +// _OSF_SFDEF GLOBAL +// _OSF_MMCSR_DEF GLOBAL +// _SCBDEF GLOBAL +// _FRMDEF GLOBAL +// _EXSDEF GLOBAL +// _OSF_A0_DEF GLOBAL +// _MCESDEF GLOBAL + +// .sbttl "EV5 symbol definitions" + +// _EV5DEF +// _PALTEMP +// _MM_STAT_DEF +// _EV5_MM +// _EV5_IPLDEF + +// _HALT_CODES GLOBAL +// _MCHK_CODES GLOBAL + +// _PAL_IMPURE +// _PAL_LOGOUT + + + + +// .sbttl "PALcode configuration options" + +// There are a number of options that may be assembled into this version of +// PALcode. They should be adjusted in a prefix assembly file (i.e. do not edit +// the following). The options that can be adjusted cause the resultant PALcode +// to reflect the desired target system. + + +#define osfpal 1 // This is the PALcode for OSF. + +#ifndef rawhide_system + +#define rawhide_system 0 +#endif + + +#ifndef real_mm +// Page table translation vs 1-1 mapping +#define real_mm 1 +#endif + + +#ifndef rax_mode + +#define rax_mode 0 +#endif + +#ifndef egore +// End of reset flow starts a program at 200000(hex). +#define egore 1 +#endif + +#ifndef acore +// End of reset flow starts a program at 40000(hex). +#define acore 0 +#endif + + +// assume acore+egore+rax_mode lt 2 // Assertion checker + +#ifndef beh_model +// EV5 behavioral model specific code +#define beh_model 1 +#endif + +#ifndef init_cbox +// Reset flow init of Bcache and Scache +#define init_cbox 1 +#endif + +#ifndef disable_crd +// Decides whether the reset flow will disable +#define disable_crd 0 +#endif + + // correctable read interrupts via ICSR +#ifndef perfmon_debug +#define perfmon_debug 0 +#endif + +#ifndef icflush_on_tbix +#define icflush_on_tbix 0 +#endif + +#ifndef remove_restore_state +#define remove_restore_state 0 +#endif + +#ifndef remove_save_state +#define remove_save_state 0 +#endif + +#ifndef enable_physical_console +#define enable_physical_console 0 +#endif + +#ifndef ev5_p1 +#define ev5_p1 0 +#endif + +#ifndef ev5_p2 +#define ev5_p2 1 +#endif + +// assume ev5_p1+ev5_p2 eq 1 + +#ifndef ldvpte_bug_fix +#define ldvpte_bug_fix 1 // If set, fix ldvpte bug in dtbmiss_double flow. +#endif + +#ifndef spe_fix +// If set, disable super-page mode in user mode and re-enable +#define spe_fix 0 +#endif + // in kernel. Workaround for cpu bug. +#ifndef build_fixed_image +#define build_fixed_image 0 +#endif + + +#ifndef fill_err_hack +// If set, disable fill_error mode in user mode and re-enable +#define fill_err_hack 0 +#endif + + // in kernel. Workaround for cpu bug. + +// .macro hw_rei_spe +// .iif eq spe_fix, hw_rei +//#if spe_fix != 0 +// +// +//#define hw_rei_chm_count hw_rei_chm_count + 1 +// p4_fixup_label \hw_rei_chm_count +// .iif eq build_fixed_image, br r31, hw_rei_update_spe +// .iif ne build_fixed_image, hw_rei +//#endif +// +// .endm + +// Add flag "osf_chm_fix" to enable dcache in user mode only +// to avoid cpu bug. + +#ifndef osf_chm_fix +// If set, enable D-Cache in +#define osf_chm_fix 0 +#endif + +#if osf_chm_fix != 0 +// user mode only. +#define hw_rei_chm_count 0 +#endif + +#if osf_chm_fix != 0 + +#define hw_rei_stall_chm_count 0 +#endif + +#ifndef enable_p4_fixups + +#define enable_p4_fixups 0 +#endif + + // If set, do EV5 Pass 4 fixups +#if spe_fix == 0 + +#define osf_chm_fix 0 +#endif + +#if spe_fix == 0 + +#define enable_p4_fixups 0 +#endif + + // Only allow fixups if fix enabled + + //Turn off fill_errors and MEM_NEM in user mode +// .macro fill_error_hack ?L10_, ?L20_, ?L30_, ?L40_ +// //save r22,r23,r24 +// stqp r22, 0x150(r31) //add +// stqp r23, 0x158(r31) //contents +// stqp r24, 0x160(r31) //bit mask +// +// lda r22, 0x82(r31) +// ldah r22, 0x8740(r22) +// sll r22, 8, r22 +// ldlp r23, 0x80(r22) // r23 <- contents of CIA_MASK +// bis r23,r31,r23 +// +// lda r24, 0x8(r31) // r24 <- MEM_NEM bit +// beq r10, L10_ // IF user mode (r10<0> == 0) pal mode +// bic r23, r24, r23 // set fillerr_en bit +// br r31, L20_ // ELSE +//L10_: bis r23, r24, r23 // clear fillerr_en bit +//L20_: // ENDIF +// +// stlp r23, 0x80(r22) // write back the CIA_MASK register +// mb +// ldlp r23, 0x80(r22) +// bis r23,r31,r23 +// mb +// +// lda r22, 1(r31) // r22 <- 87.4000.0100 ptr to CIA_CTRL +// ldah r22, 0x8740(r22) +// sll r22, 8, r22 +// ldlp r23, 0(r22) // r23 <- contents of CIA_CTRL +// bis r23,r31,r23 +// +// +// lda r24, 0x400(r31) // r9 <- fillerr_en bit +// beq r10, L30_ // IF user mode (r10<0> == 0) pal mode +// bic r23, r24, r23 // set fillerr_en bit +// br r31, L40_ // ELSE +//L30_: bis r23, r24, r23 // clear fillerr_en bit +//L40_: // ENDIF +// +// stlp r23, 0(r22) // write back the CIA_CTRL register +// mb +// ldlp r23, 0(r22) +// bis r23,r31,r23 +// mb +// +// //restore r22,r23,r24 +// ldqp r22, 0x150(r31) +// ldqp r23, 0x158(r31) +// ldqp r24, 0x160(r31) +// +// .endm + +// multiprocessor support can be enabled for a max of n processors by +// setting the following to the number of processors on the system. +// Note that this is really the max cpuid. + +#ifndef max_cpuid +#define max_cpuid 8 +#endif + +#ifndef osf_svmin // platform specific palcode version number +#define osf_svmin 0 +#endif + + +#define osfpal_version_h ((max_cpuid<<16) | (osf_svmin<<0)) + +// .mcall ldqp // override macro64 definition with macro from library +// .mcall stqp // override macro64 definition with macro from library + + +// .psect _pal,mix +// huh pb pal_base: +// huh pb #define current_block_base . - pal_base + +// .sbttl "RESET - Reset Trap Entry Point" +//+ +// RESET - offset 0000 +// Entry: +// Vectored into via hardware trap on reset, or branched to +// on swppal. +// +// r0 = whami +// r1 = pal_base +// r2 = base of scratch area +// r3 = halt code +// +// +// Function: +// +//- + + .text 0 + . = 0x0000 + .globl Pal_Base +Pal_Base: + HDW_VECTOR(PAL_RESET_ENTRY) +Trap_Reset: + nop +#ifdef SIMOS + /* + * store into r1 + */ + br r1,sys_reset +#else + /* following is a srcmax change */ + + DEBUGSTORE(0x41) + /* The original code jumped using r1 as a linkage register to pass the base + of PALcode to the platform specific code. We use r1 to pass a parameter + from the SROM, so we hardcode the address of Pal_Base in platform.s + */ + br r31, sys_reset +#endif + + // Specify PAL version info as a constant + // at a known location (reset + 8). + + .long osfpal_version_l // <pal_type@16> ! <vmaj@8> ! <vmin@0> + .long osfpal_version_h // <max_cpuid@16> ! <osf_svmin@0> + .long 0 + .long 0 +pal_impure_start: + .quad 0 +pal_debug_ptr: + .quad 0 // reserved for debug pointer ; 20 +#if beh_model == 0 + + +#if enable_p4_fixups != 0 + + + .quad 0 + .long p4_fixup_hw_rei_fixup_table +#endif + +#else + + .quad 0 // + .quad 0 //0x0030 + .quad 0 + .quad 0 //0x0040 + .quad 0 + .quad 0 //0x0050 + .quad 0 + .quad 0 //0x0060 + .quad 0 +pal_enter_cns_address: + .quad 0 //0x0070 -- address to jump to from enter_console + .long <<sys_exit_console-pal_base>+1> //0x0078 -- offset to sys_exit_console (set palmode bit) +#endif + + + + +// .sbttl "IACCVIO- Istream Access Violation Trap Entry Point" + +//+ +// IACCVIO - offset 0080 +// Entry: +// Vectored into via hardware trap on Istream access violation or sign check error on PC. +// +// Function: +// Build stack frame +// a0 <- Faulting VA +// a1 <- MMCSR (1 for ACV) +// a2 <- -1 (for ifetch fault) +// vector via entMM +//- + + HDW_VECTOR(PAL_IACCVIO_ENTRY) +Trap_Iaccvio: + DEBUGSTORE(0x42) + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS + bge r25, TRAP_IACCVIO_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +TRAP_IACCVIO_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bic r14, 3, r16 // pass pc/va as a0 + + stq r17, osfsf_a1(sp) // a1 + or r31, mmcsr_c_acv, r17 // pass mm_csr as a1 + + stq r18, osfsf_a2(sp) // a2 + mfpr r13, pt_entmm // get entry point + + stq r11, osfsf_ps(sp) // save old ps + bis r12, r31, r11 // update ps + + stq r16, osfsf_pc(sp) // save pc + stq r29, osfsf_gp(sp) // save gp + + mtpr r13, exc_addr // load exc_addr with entMM + // 1 cycle to hw_rei + mfpr r29, pt_kgp // get the kgp + + subq r31, 1, r18 // pass flag of istream, as a2 + hw_rei_spe + + +// .sbttl "INTERRUPT- Interrupt Trap Entry Point" + +//+ +// INTERRUPT - offset 0100 +// Entry: +// Vectored into via trap on hardware interrupt +// +// Function: +// check for halt interrupt +// check for passive release (current ipl geq requestor) +// if necessary, switch to kernel mode +// push stack frame, update ps (including current mode and ipl copies), sp, and gp +// pass the interrupt info to the system module +// +//- + + + HDW_VECTOR(PAL_INTERRUPT_ENTRY) +Trap_Interrupt: + mfpr r13, ev5__intid // Fetch level of interruptor + mfpr r25, ev5__isr // Fetch interrupt summary register + + srl r25, isr_v_hlt, r9 // Get HLT bit + mfpr r14, ev5__ipl + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kern + blbs r9, sys_halt_interrupt // halt_interrupt if HLT bit set + + cmple r13, r14, r8 // R8 = 1 if intid .less than or eql. ipl + bne r8, sys_passive_release // Passive release is current rupt is lt or eq ipl + + and r11, osfps_m_mode, r10 // get mode bit + beq r10, TRAP_INTERRUPT_10_ // Skip stack swap in kernel + + mtpr r30, pt_usp // save user stack + mfpr r30, pt_ksp // get kern stack + +TRAP_INTERRUPT_10_: + lda sp, (0-osfsf_c_size)(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r11, osfsf_ps(sp) // save ps + stq r14, osfsf_pc(sp) // save pc + + stq r29, osfsf_gp(sp) // push gp + stq r16, osfsf_a0(sp) // a0 + +// pvc_violate 354 // ps is cleared anyway, if store to stack faults. + mtpr r31, ev5__ps // Set Ibox current mode to kernel + stq r17, osfsf_a1(sp) // a1 + + stq r18, osfsf_a2(sp) // a2 + subq r13, 0x11, r12 // Start to translate from EV5IPL->OSFIPL + + srl r12, 1, r8 // 1d, 1e: ipl 6. 1f: ipl 7. + subq r13, 0x1d, r9 // Check for 1d, 1e, 1f + + cmovge r9, r8, r12 // if .ge. 1d, then take shifted value + bis r12, r31, r11 // set new ps + + mfpr r12, pt_intmask + and r11, osfps_m_ipl, r14 // Isolate just new ipl (not really needed, since all non-ipl bits zeroed already) + +#ifdef SIMOS + /* + * Lance had space problems. We don't. + */ + extbl r12, r14, r14 // Translate new OSFIPL->EV5IPL + mfpr r29, pt_kgp // update gp + mtpr r14, ev5__ipl // load the new IPL into Ibox +#else +// Moved the following three lines to sys_interrupt to make room for debug +// extbl r12, r14, r14 // Translate new OSFIPL->EV5IPL +// mfpr r29, pt_kgp // update gp + +// mtpr r14, ev5__ipl // load the new IPL into Ibox +#endif + br r31, sys_interrupt // Go handle interrupt + + + +// .sbttl "ITBMISS- Istream TBmiss Trap Entry Point" + +//+ +// ITBMISS - offset 0180 +// Entry: +// Vectored into via hardware trap on Istream translation buffer miss. +// +// Function: +// Do a virtual fetch of the PTE, and fill the ITB if the PTE is valid. +// Can trap into DTBMISS_DOUBLE. +// This routine can use the PALshadow registers r8, r9, and r10 +// +//- + + HDW_VECTOR(PAL_ITB_MISS_ENTRY) +Trap_Itbmiss: +#if real_mm == 0 + + + // Simple 1-1 va->pa mapping + + nop // Pad to align to E1 + mfpr r8, exc_addr + + srl r8, page_offset_size_bits, r9 + sll r9, 32, r9 + + lda r9, 0x3301(r9) // Make PTE, V set, all KRE, URE, KWE, UWE + mtpr r9, itb_pte // E1 + + hw_rei_stall // Nital says I don't have to obey shadow wait rule here. +#else + + // Real MM mapping + nop + mfpr r8, ev5__ifault_va_form // Get virtual address of PTE. + + nop + mfpr r10, exc_addr // Get PC of faulting instruction in case of DTBmiss. + +pal_itb_ldq: + ld_vpte r8, 0(r8) // Get PTE, traps to DTBMISS_DOUBLE in case of TBmiss + mtpr r10, exc_addr // Restore exc_address if there was a trap. + + mfpr r31, ev5__va // Unlock VA in case there was a double miss + nop + + and r8, osfpte_m_foe, r25 // Look for FOE set. + blbc r8, invalid_ipte_handler // PTE not valid. + + nop + bne r25, foe_ipte_handler // FOE is set + + nop + mtpr r8, ev5__itb_pte // Ibox remembers the VA, load the PTE into the ITB. + + hw_rei_stall // + +#endif + + + + +// .sbttl "DTBMISS_SINGLE - Dstream Single TBmiss Trap Entry Point" + +//+ +// DTBMISS_SINGLE - offset 0200 +// Entry: +// Vectored into via hardware trap on Dstream single translation buffer miss. +// +// Function: +// Do a virtual fetch of the PTE, and fill the DTB if the PTE is valid. +// Can trap into DTBMISS_DOUBLE. +// This routine can use the PALshadow registers r8, r9, and r10 +//- + + HDW_VECTOR(PAL_DTB_MISS_ENTRY) +Trap_Dtbmiss_Single: +#if real_mm == 0 + // Simple 1-1 va->pa mapping + mfpr r8, va // E0 + srl r8, page_offset_size_bits, r9 + + sll r9, 32, r9 + lda r9, 0x3301(r9) // Make PTE, V set, all KRE, URE, KWE, UWE + + mtpr r9, dtb_pte // E0 + nop // Pad to align to E0 + + + + mtpr r8, dtb_tag // E0 + nop + + nop // Pad tag write + nop + + nop // Pad tag write + nop + + hw_rei +#else + mfpr r8, ev5__va_form // Get virtual address of PTE - 1 cycle delay. E0. + mfpr r10, exc_addr // Get PC of faulting instruction in case of error. E1. + +// DEBUGSTORE(0x45) +// DEBUG_EXC_ADDR() + // Real MM mapping + mfpr r9, ev5__mm_stat // Get read/write bit. E0. + mtpr r10, pt6 // Stash exc_addr away + +pal_dtb_ldq: + ld_vpte r8, 0(r8) // Get PTE, traps to DTBMISS_DOUBLE in case of TBmiss + nop // Pad MF VA + + mfpr r10, ev5__va // Get original faulting VA for TB load. E0. + nop + + mtpr r8, ev5__dtb_pte // Write DTB PTE part. E0. + blbc r8, invalid_dpte_handler // Handle invalid PTE + + mtpr r10, ev5__dtb_tag // Write DTB TAG part, completes DTB load. No virt ref for 3 cycles. + mfpr r10, pt6 + + // Following 2 instructions take 2 cycles + mtpr r10, exc_addr // Return linkage in case we trapped. E1. + mfpr r31, pt0 // Pad the write to dtb_tag + + hw_rei // Done, return +#endif + + + + +// .sbttl "DTBMISS_DOUBLE - Dstream Double TBmiss Trap Entry Point" + +//+ +// DTBMISS_DOUBLE - offset 0280 +// Entry: +// Vectored into via hardware trap on Double TBmiss from single miss flows. +// +// r8 - faulting VA +// r9 - original MMstat +// r10 - original exc_addr (both itb,dtb miss) +// pt6 - original exc_addr (dtb miss flow only) +// VA IPR - locked with original faulting VA +// +// Function: +// Get PTE, if valid load TB and return. +// If not valid then take TNV/ACV exception. +// +// pt4 and pt5 are reserved for this flow. +// +// +//- + + HDW_VECTOR(PAL_DOUBLE_MISS_ENTRY) +Trap_Dtbmiss_double: +#if ldvpte_bug_fix != 0 + mtpr r8, pt4 // save r8 to do exc_addr check + mfpr r8, exc_addr + blbc r8, Trap_Dtbmiss_Single //if not in palmode, should be in the single routine, dummy! + mfpr r8, pt4 // restore r8 +#endif + nop + mtpr r22, pt5 // Get some scratch space. E1. + // Due to virtual scheme, we can skip the first lookup and go + // right to fetch of level 2 PTE + sll r8, (64-((2*page_seg_size_bits)+page_offset_size_bits)), r22 // Clean off upper bits of VA + mtpr r21, pt4 // Get some scratch space. E1. + + srl r22, 61-page_seg_size_bits, r22 // Get Va<seg1>*8 + mfpr r21, pt_ptbr // Get physical address of the page table. + + nop + addq r21, r22, r21 // Index into page table for level 2 PTE. + + sll r8, (64-((1*page_seg_size_bits)+page_offset_size_bits)), r22 // Clean off upper bits of VA + ldqp r21, 0(r21) // Get level 2 PTE (addr<2:0> ignored) + + srl r22, 61-page_seg_size_bits, r22 // Get Va<seg1>*8 + blbc r21, double_pte_inv // Check for Invalid PTE. + + srl r21, 32, r21 // extract PFN from PTE + sll r21, page_offset_size_bits, r21 // get PFN * 2^13 for add to <seg3>*8 + + addq r21, r22, r21 // Index into page table for level 3 PTE. + nop + + ldqp r21, 0(r21) // Get level 3 PTE (addr<2:0> ignored) + blbc r21, double_pte_inv // Check for invalid PTE. + + mtpr r21, ev5__dtb_pte // Write the PTE. E0. + mfpr r22, pt5 // Restore scratch register + + mtpr r8, ev5__dtb_tag // Write the TAG. E0. No virtual references in subsequent 3 cycles. + mfpr r21, pt4 // Restore scratch register + + nop // Pad write to tag. + nop + + nop // Pad write to tag. + nop + + hw_rei + + + +// .sbttl "UNALIGN -- Dstream unalign trap" +//+ +// UNALIGN - offset 0300 +// Entry: +// Vectored into via hardware trap on unaligned Dstream reference. +// +// Function: +// Build stack frame +// a0 <- Faulting VA +// a1 <- Opcode +// a2 <- src/dst register number +// vector via entUna +//- + + HDW_VECTOR(PAL_UNALIGN_ENTRY) +Trap_Unalign: +/* DEBUGSTORE(0x47)*/ + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r8, ev5__mm_stat // Get mmstat --ok to use r8, no tbmiss + mfpr r14, exc_addr // get pc + + srl r8, mm_stat_v_ra, r13 // Shift Ra field to ls bits + blbs r14, pal_pal_bug_check // Bugcheck if unaligned in PAL + + blbs r8, UNALIGN_NO_DISMISS // lsb only set on store or fetch_m + // not set, must be a load + and r13, 0x1F, r8 // isolate ra + + cmpeq r8, 0x1F, r8 // check for r31/F31 + bne r8, dfault_fetch_ldr31_err // if its a load to r31 or f31 -- dismiss the fault + +UNALIGN_NO_DISMISS: + bis r11, r31, r12 // Save PS + bge r25, UNALIGN_NO_DISMISS_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +UNALIGN_NO_DISMISS_10_: + mfpr r25, ev5__va // Unlock VA + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + mtpr r25, pt0 // Stash VA + stq r18, osfsf_a2(sp) // a2 + + stq r11, osfsf_ps(sp) // save old ps + srl r13, mm_stat_v_opcode-mm_stat_v_ra, r25// Isolate opcode + + stq r29, osfsf_gp(sp) // save gp + addq r14, 4, r14 // inc PC past the ld/st + + stq r17, osfsf_a1(sp) // a1 + and r25, mm_stat_m_opcode, r17// Clean opocde for a1 + + stq r16, osfsf_a0(sp) // save regs + mfpr r16, pt0 // a0 <- va/unlock + + stq r14, osfsf_pc(sp) // save pc + mfpr r25, pt_entuna // get entry point + + + bis r12, r31, r11 // update ps + br r31, unalign_trap_cont + + + + +// .sbttl "DFAULT - Dstream Fault Trap Entry Point" + +//+ +// DFAULT - offset 0380 +// Entry: +// Vectored into via hardware trap on dstream fault or sign check error on DVA. +// +// Function: +// Ignore faults on FETCH/FETCH_M +// Check for DFAULT in PAL +// Build stack frame +// a0 <- Faulting VA +// a1 <- MMCSR (1 for ACV, 2 for FOR, 4 for FOW) +// a2 <- R/W +// vector via entMM +// +//- + HDW_VECTOR(PAL_D_FAULT_ENTRY) +Trap_Dfault: +// DEBUGSTORE(0x48) + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r13, ev5__mm_stat // Get mmstat + mfpr r8, exc_addr // get pc, preserve r14 + + srl r13, mm_stat_v_opcode, r9 // Shift opcode field to ls bits + blbs r8, dfault_in_pal + + bis r8, r31, r14 // move exc_addr to correct place + bis r11, r31, r12 // Save PS + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + and r9, mm_stat_m_opcode, r9 // Clean all but opcode + + cmpeq r9, evx_opc_sync, r9 // Is the opcode fetch/fetchm? + bne r9, dfault_fetch_ldr31_err // Yes, dismiss the fault + + //dismiss exception if load to r31/f31 + blbs r13, dfault_no_dismiss // mm_stat<0> set on store or fetchm + + // not a store or fetch, must be a load + srl r13, mm_stat_v_ra, r9 // Shift rnum to low bits + + and r9, 0x1F, r9 // isolate rnum + nop + + cmpeq r9, 0x1F, r9 // Is the rnum r31 or f31? + bne r9, dfault_fetch_ldr31_err // Yes, dismiss the fault + +dfault_no_dismiss: + and r13, 0xf, r13 // Clean extra bits in mm_stat + bge r25, dfault_trap_cont // no stack swap needed if cm=kern + + + mtpr r30, pt_usp // save user stack + bis r31, r31, r12 // Set new PS + + mfpr r30, pt_ksp + br r31, dfault_trap_cont + + + + + +// .sbttl "MCHK - Machine Check Trap Entry Point" + +//+ +// MCHK - offset 0400 +// Entry: +// Vectored into via hardware trap on machine check. +// +// Function: +// +//- + + HDW_VECTOR(PAL_MCHK_ENTRY) +Trap_Mchk: + DEBUGSTORE(0x49) + mtpr r31, ic_flush_ctl // Flush the Icache + br r31, sys_machine_check + + + + +// .sbttl "OPCDEC - Illegal Opcode Trap Entry Point" + +//+ +// OPCDEC - offset 0480 +// Entry: +// Vectored into via hardware trap on illegal opcode. +// +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + + HDW_VECTOR(PAL_OPCDEC_ENTRY) +Trap_Opcdec: + DEBUGSTORE(0x4a) +//simos DEBUG_EXC_ADDR() + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r14, exc_addr // get pc + blbs r14, pal_pal_bug_check // check opcdec in palmode + + bis r11, r31, r12 // Save PS + bge r25, TRAP_OPCDEC_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +TRAP_OPCDEC_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + addq r14, 4, r14 // inc pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_opdec, r16 // set a0 + + stq r11, osfsf_ps(sp) // save old ps + mfpr r13, pt_entif // get entry point + + stq r18, osfsf_a2(sp) // a2 + stq r17, osfsf_a1(sp) // a1 + + stq r29, osfsf_gp(sp) // save gp + stq r14, osfsf_pc(sp) // save pc + + bis r12, r31, r11 // update ps + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei, E1 + + mfpr r29, pt_kgp // get the kgp, E1 + + hw_rei_spe // done, E1 + + + + + + +// .sbttl "ARITH - Arithmetic Exception Trap Entry Point" + +//+ +// ARITH - offset 0500 +// Entry: +// Vectored into via hardware trap on arithmetic excpetion. +// +// Function: +// Build stack frame +// a0 <- exc_sum +// a1 <- exc_mask +// a2 <- unpred +// vector via entArith +// +//- + HDW_VECTOR(PAL_ARITH_ENTRY) +Trap_Arith: + DEBUGSTORE(0x4b) + and r11, osfps_m_mode, r12 // get mode bit + mfpr r31, ev5__va // unlock mbox + + bis r11, r31, r25 // save ps + mfpr r14, exc_addr // get pc + + nop + blbs r14, pal_pal_bug_check // arith trap from PAL + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + beq r12, TRAP_ARITH_10_ // if zero we are in kern now + + bis r31, r31, r25 // set the new ps + mtpr r30, pt_usp // save user stack + + nop + mfpr r30, pt_ksp // get kern stack + +TRAP_ARITH_10_: lda sp, 0-osfsf_c_size(sp) // allocate stack space + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + nop // Pad current mode write and stq + mfpr r13, ev5__exc_sum // get the exc_sum + + mfpr r12, pt_entarith + stq r14, osfsf_pc(sp) // save pc + + stq r17, osfsf_a1(sp) + mfpr r17, ev5__exc_mask // Get exception register mask IPR - no mtpr exc_sum in next cycle + + stq r11, osfsf_ps(sp) // save ps + bis r25, r31, r11 // set new ps + + stq r16, osfsf_a0(sp) // save regs + srl r13, exc_sum_v_swc, r16// shift data to correct position + + stq r18, osfsf_a2(sp) +// pvc_violate 354 // ok, but make sure reads of exc_mask/sum are not in same trap shadow + mtpr r31, ev5__exc_sum // Unlock exc_sum and exc_mask + + stq r29, osfsf_gp(sp) + mtpr r12, exc_addr // Set new PC - 1 bubble to hw_rei - E1 + + mfpr r29, pt_kgp // get the kern gp - E1 + hw_rei_spe // done - E1 + + + + + + +// .sbttl "FEN - Illegal Floating Point Operation Trap Entry Point" + +//+ +// FEN - offset 0580 +// Entry: +// Vectored into via hardware trap on illegal FP op. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + + HDW_VECTOR(PAL_FEN_ENTRY) +Trap_Fen: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r14, exc_addr // get pc + blbs r14, pal_pal_bug_check // check opcdec in palmode + + mfpr r13, ev5__icsr + nop + + bis r11, r31, r12 // Save PS + bge r25, TRAP_FEN_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r12 // Set new PS + mfpr r30, pt_ksp + +TRAP_FEN_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + srl r13, icsr_v_fpe, r25 // Shift FP enable to bit 0 + + + stq r16, osfsf_a0(sp) // save regs + mfpr r13, pt_entif // get entry point + + stq r18, osfsf_a2(sp) // a2 + stq r11, osfsf_ps(sp) // save old ps + + stq r29, osfsf_gp(sp) // save gp + bis r12, r31, r11 // set new ps + + stq r17, osfsf_a1(sp) // a1 + blbs r25,fen_to_opcdec // If FP is enabled, this is really OPCDEC. + + bis r31, osf_a0_fen, r16 // set a0 + stq r14, osfsf_pc(sp) // save pc + + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei -E1 + + mfpr r29, pt_kgp // get the kgp -E1 + + hw_rei_spe // done -E1 + +// FEN trap was taken, but the fault is really opcdec. + ALIGN_BRANCH +fen_to_opcdec: + addq r14, 4, r14 // save PC+4 + bis r31, osf_a0_opdec, r16 // set a0 + + stq r14, osfsf_pc(sp) // save pc + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kgp + hw_rei_spe // done + + + +// .sbttl "Misc handlers" + // Start area for misc code. +//+ +//dfault_trap_cont +// A dfault trap has been taken. The sp has been updated if necessary. +// Push a stack frame a vector via entMM. +// +// Current state: +// r12 - new PS +// r13 - MMstat +// VA - locked +// +//- + ALIGN_BLOCK +dfault_trap_cont: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r25, ev5__va // Fetch VA/unlock + + stq r18, osfsf_a2(sp) // a2 + and r13, 1, r18 // Clean r/w bit for a2 + + stq r16, osfsf_a0(sp) // save regs + bis r25, r31, r16 // a0 <- va + + stq r17, osfsf_a1(sp) // a1 + srl r13, 1, r17 // shift fault bits to right position + + stq r11, osfsf_ps(sp) // save old ps + bis r12, r31, r11 // update ps + + stq r14, osfsf_pc(sp) // save pc + mfpr r25, pt_entmm // get entry point + + stq r29, osfsf_gp(sp) // save gp + cmovlbs r17, 1, r17 // a2. acv overrides fox. + + mtpr r25, exc_addr // load exc_addr with entMM + // 1 cycle to hw_rei + mfpr r29, pt_kgp // get the kgp + + hw_rei_spe // done + +//+ +//unalign_trap_cont +// An unalign trap has been taken. Just need to finish up a few things. +// +// Current state: +// r25 - entUna +// r13 - shifted MMstat +// +//- + ALIGN_BLOCK +unalign_trap_cont: + mtpr r25, exc_addr // load exc_addr with entUna + // 1 cycle to hw_rei + + + mfpr r29, pt_kgp // get the kgp + and r13, mm_stat_m_ra, r18 // Clean Ra for a2 + + hw_rei_spe // done + + + +//+ +// dfault_in_pal +// Dfault trap was taken, exc_addr points to a PAL PC. +// r9 - mmstat<opcode> right justified +// r8 - exception address +// +// These are the cases: +// opcode was STQ -- from a stack builder, KSP not valid halt +// r14 - original exc_addr +// r11 - original PS +// opcode was STL_C -- rti or retsys clear lock_flag by stack write, +// KSP not valid halt +// r11 - original PS +// r14 - original exc_addr +// opcode was LDQ -- retsys or rti stack read, KSP not valid halt +// r11 - original PS +// r14 - original exc_addr +// opcode was HW_LD -- itbmiss or dtbmiss, bugcheck due to fault on page tables +// r10 - original exc_addr +// r11 - original PS +// +// +//- + ALIGN_BLOCK +dfault_in_pal: + DEBUGSTORE(0x50) + bic r8, 3, r8 // Clean PC + mfpr r9, pal_base + + mfpr r31, va // unlock VA +#if real_mm != 0 + // if not real_mm, should never get here from miss flows + + subq r9, r8, r8 // pal_base - offset + + lda r9, pal_itb_ldq-pal_base(r8) + nop + + beq r9, dfault_do_bugcheck + lda r9, pal_dtb_ldq-pal_base(r8) + + beq r9, dfault_do_bugcheck +#endif + +// +// KSP invalid halt case -- +ksp_inval_halt: + DEBUGSTORE(76) + bic r11, osfps_m_mode, r11 // set ps to kernel mode + mtpr r0, pt0 + + mtpr r31, dtb_cm // Make sure that the CM IPRs are all kernel mode + mtpr r31, ips + + mtpr r14, exc_addr // Set PC to instruction that caused trouble +//orig pvc_jsr updpcb, bsr=1 + bsr r0, pal_update_pcb // update the pcb + + lda r0, hlt_c_ksp_inval(r31) // set halt code to hw halt + br r31, sys_enter_console // enter the console + + ALIGN_BRANCH +dfault_do_bugcheck: + bis r10, r31, r14 // bugcheck expects exc_addr in r14 + br r31, pal_pal_bug_check + + + ALIGN_BLOCK +//+ +// dfault_fetch_ldr31_err - ignore faults on fetch(m) and loads to r31/f31 +// On entry - +// r14 - exc_addr +// VA is locked +// +//- +dfault_fetch_ldr31_err: + mtpr r11, ev5__dtb_cm + mtpr r11, ev5__ps // Make sure ps hasn't changed + + mfpr r31, va // unlock the mbox + addq r14, 4, r14 // inc the pc to skip the fetch + + mtpr r14, exc_addr // give ibox new PC + mfpr r31, pt0 // pad exc_addr write + + hw_rei + + + + ALIGN_BLOCK +//+ +// sys_from_kern +// callsys from kernel mode - OS bugcheck machine check +// +//- +sys_from_kern: + mfpr r14, exc_addr // PC points to call_pal + subq r14, 4, r14 + + lda r25, mchk_c_os_bugcheck(r31) // fetch mchk code + br r31, pal_pal_mchk + + +// .sbttl "Continuation of long call_pal flows" + ALIGN_BLOCK +//+ +// wrent_tbl +// Table to write *int in paltemps. +// 4 instructions/entry +// r16 has new value +// +//- +wrent_tbl: +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entint + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entarith + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entmm + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entif + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entuna + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + +//orig pvc_jsr wrent, dest=1 + nop + mtpr r16, pt_entsys + + mfpr r31, pt0 // Pad for mt->mf paltemp rule + hw_rei + + ALIGN_BLOCK +//+ +// tbi_tbl +// Table to do tbi instructions +// 4 instructions per entry +//- +tbi_tbl: + // -2 tbia +//orig pvc_jsr tbi, dest=1 + mtpr r31, ev5__dtb_ia // Flush DTB + mtpr r31, ev5__itb_ia // Flush ITB + +#if icflush_on_tbix != 0 + + + br r31, pal_ic_flush // Flush Icache +#else + + hw_rei_stall +#endif + + nop // Pad table + + // -1 tbiap +//orig pvc_jsr tbi, dest=1 + mtpr r31, ev5__dtb_iap // Flush DTB + mtpr r31, ev5__itb_iap // Flush ITB + +#if icflush_on_tbix != 0 + + + br r31, pal_ic_flush // Flush Icache +#else + + hw_rei_stall +#endif + + nop // Pad table + + + // 0 unused +//orig pvc_jsr tbi, dest=1 + hw_rei // Pad table + nop + nop + nop + + + // 1 tbisi +//orig pvc_jsr tbi, dest=1 +#if icflush_on_tbix != 0 + + + + nop + br r31, pal_ic_flush_and_tbisi // Flush Icache + nop + nop // Pad table +#else + + nop + nop + mtpr r17, ev5__itb_is // Flush ITB + hw_rei_stall +#endif + + + + // 2 tbisd +//orig pvc_jsr tbi, dest=1 + mtpr r17, ev5__dtb_is // Flush DTB. + nop + + nop + hw_rei_stall + + + // 3 tbis +//orig pvc_jsr tbi, dest=1 + mtpr r17, ev5__dtb_is // Flush DTB +#if icflush_on_tbix != 0 + + + br r31, pal_ic_flush_and_tbisi // Flush Icache and ITB +#else + br r31, tbi_finish + ALIGN_BRANCH +tbi_finish: + mtpr r17, ev5__itb_is // Flush ITB + hw_rei_stall +#endif + + + + ALIGN_BLOCK +//+ +// bpt_bchk_common: +// Finish up the bpt/bchk instructions +//- +bpt_bchk_common: + stq r18, osfsf_a2(sp) // a2 + mfpr r13, pt_entif // get entry point + + stq r12, osfsf_ps(sp) // save old ps + stq r14, osfsf_pc(sp) // save pc + + stq r29, osfsf_gp(sp) // save gp + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kgp + + + hw_rei_spe // done + + + ALIGN_BLOCK +//+ +// rti_to_user +// Finish up the rti instruction +//- +rti_to_user: + mtpr r11, ev5__dtb_cm // set Mbox current mode - no virt ref for 2 cycles + mtpr r11, ev5__ps // set Ibox current mode - 2 bubble to hw_rei + + mtpr r31, ev5__ipl // set the ipl. No hw_rei for 2 cycles + mtpr r25, pt_ksp // save off incase RTI to user + + mfpr r30, pt_usp + hw_rei_spe // and back + + + ALIGN_BLOCK +//+ +// rti_to_kern +// Finish up the rti instruction +//- +rti_to_kern: + and r12, osfps_m_ipl, r11 // clean ps + mfpr r12, pt_intmask // get int mask + + extbl r12, r11, r12 // get mask for this ipl + mtpr r25, pt_ksp // save off incase RTI to user + + mtpr r12, ev5__ipl // set the new ipl. + or r25, r31, sp // sp + +// pvc_violate 217 // possible hidden mt->mf ipl not a problem in callpals + hw_rei + + ALIGN_BLOCK +//+ +// swpctx_cont +// Finish up the swpctx instruction +//- + +swpctx_cont: +#if ev5_p1 != 0 + + + bic r25, r24, r25 // clean icsr<FPE> + get_impure r8 // get impure pointer + + sll r12, icsr_v_fpe, r12 // shift new fen to pos + fix_impure_ipr r8 // adjust impure pointer + + restore_reg1 pmctr_ctl, r8, r8, ipr=1 // "ldqp" - get pmctr_ctl bits + srl r23, 32, r24 // move asn to low asn pos + + ldqp r14, osfpcb_q_mmptr(r16)// get new mmptr + srl r22, osfpcb_v_pme, r22 // get pme down to bit 0 + + or r25, r12, r25 // icsr with new fen + sll r24, itb_asn_v_asn, r12 + +#else + + bic r25, r24, r25 // clean icsr<FPE,PMP> + sll r12, icsr_v_fpe, r12 // shift new fen to pos + + ldqp r14, osfpcb_q_mmptr(r16)// get new mmptr + srl r22, osfpcb_v_pme, r22 // get pme down to bit 0 + + or r25, r12, r25 // icsr with new fen + srl r23, 32, r24 // move asn to low asn pos + + and r22, 1, r22 + sll r24, itb_asn_v_asn, r12 + + sll r22, icsr_v_pmp, r22 + nop + + or r25, r22, r25 // icsr with new pme +#endif + + sll r24, dtb_asn_v_asn, r24 + + subl r23, r13, r13 // gen new cc offset + mtpr r12, itb_asn // no hw_rei_stall in 0,1,2,3,4 + + mtpr r24, dtb_asn // Load up new ASN + mtpr r25, icsr // write the icsr + + sll r14, page_offset_size_bits, r14 // Move PTBR into internal position. + ldqp r25, osfpcb_q_usp(r16) // get new usp + + insll r13, 4, r13 // >> 32 +// pvc_violate 379 // ldqp can't trap except replay. only problem if mf same ipr in same shadow + mtpr r14, pt_ptbr // load the new ptbr + + mtpr r13, cc // set new offset + ldqp r30, osfpcb_q_ksp(r16) // get new ksp + +// pvc_violate 379 // ldqp can't trap except replay. only problem if mf same ipr in same shadow + mtpr r25, pt_usp // save usp + +#if ev5_p1 != 0 + + + blbc r8, no_pm_change // if monitoring all processes -- no need to change pm + + // otherwise, monitoring select processes - update pm + lda r25, 0x3F(r31) + cmovlbc r22, r31, r8 // if pme set, disable counters, otherwise use saved encodings + + sll r25, pmctr_v_ctl2, r25 // create ctl field bit mask + mfpr r22, ev5__pmctr + + and r8, r25, r8 // mask new ctl value + bic r22, r25, r22 // clear ctl field in pmctr + + or r8, r22, r8 + mtpr r8, ev5__pmctr + +no_pm_change: +#endif + + +#if osf_chm_fix != 0 + + + p4_fixup_hw_rei_stall // removes this section for Pass 4 by placing a hw_rei_stall here + +#if build_fixed_image != 0 + + + hw_rei_stall +#else + + mfpr r9, pt_pcbb // get FEN +#endif + + ldqp r9, osfpcb_q_fen(r9) + blbc r9, no_pm_change_10_ // skip if FEN disabled + + mb // ensure no outstanding fills + lda r12, 1<<dc_mode_v_dc_ena(r31) + mtpr r12, dc_mode // turn dcache on so we can flush it + nop // force correct slotting + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + mfpr r31, pt0 // no mbox instructions in 1,2,3,4 + + lda r8, 0(r31) // flood the dcache with junk data +no_pm_change_5_: ldqp r31, 0(r8) + lda r8, 0x20(r8) // touch each cache block + srl r8, 13, r9 + blbc r9, no_pm_change_5_ + + mb // ensure no outstanding fills + mtpr r31, dc_mode // turn the dcache back off + nop // force correct slotting + mfpr r31, pt0 // no hw_rei_stall in 0,1 +#endif + + +no_pm_change_10_: hw_rei_stall // back we go + + ALIGN_BLOCK +//+ +// swppal_cont - finish up the swppal call_pal +//- + +swppal_cont: + mfpr r2, pt_misc // get misc bits + sll r0, pt_misc_v_switch, r0 // get the "I've switched" bit + or r2, r0, r2 // set the bit + mtpr r31, ev5__alt_mode // ensure alt_mode set to 0 (kernel) + mtpr r2, pt_misc // update the chip + + or r3, r31, r4 + mfpr r3, pt_impure // pass pointer to the impure area in r3 +//orig fix_impure_ipr r3 // adjust impure pointer for ipr read +//orig restore_reg1 bc_ctl, r1, r3, ipr=1 // pass cns_bc_ctl in r1 +//orig restore_reg1 bc_config, r2, r3, ipr=1 // pass cns_bc_config in r2 +//orig unfix_impure_ipr r3 // restore impure pointer + lda r3, CNS_Q_IPR(r3) + RESTORE_SHADOW(r1,CNS_Q_BC_CTL,r3); + RESTORE_SHADOW(r1,CNS_Q_BC_CFG,r3); + lda r3, -CNS_Q_IPR(r3) + + or r31, r31, r0 // set status to success +// pvc_violate 1007 + jmp r31, (r4) // and call our friend, it's her problem now + + +swppal_fail: + addq r0, 1, r0 // set unknown pal or not loaded + hw_rei // and return + + +// .sbttl "Memory management" + + ALIGN_BLOCK +//+ +//foe_ipte_handler +// IFOE detected on level 3 pte, sort out FOE vs ACV +// +// on entry: +// with +// R8 = pte +// R10 = pc +// +// Function +// Determine TNV vs ACV vs FOE. Build stack and dispatch +// Will not be here if TNV. +//- + +foe_ipte_handler: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, foe_ipte_handler_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + + srl r8, osfpte_v_ure-osfpte_v_kre, r8 // move pte user bits to kern + nop + +foe_ipte_handler_10_: srl r8, osfpte_v_kre, r25 // get kre to <0> + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + or r10, r31, r14 // Save pc/va in case TBmiss or fault on stack + mfpr r13, pt_entmm // get entry point + + stq r16, osfsf_a0(sp) // a0 + or r14, r31, r16 // pass pc/va as a0 + + stq r17, osfsf_a1(sp) // a1 + nop + + stq r18, osfsf_a2(sp) // a2 + lda r17, mmcsr_c_acv(r31) // assume ACV + + stq r16, osfsf_pc(sp) // save pc + cmovlbs r25, mmcsr_c_foe, r17 // otherwise FOE + + stq r12, osfsf_ps(sp) // save ps + subq r31, 1, r18 // pass flag of istream as a2 + + stq r29, osfsf_gp(sp) + mtpr r13, exc_addr // set vector address + + mfpr r29, pt_kgp // load kgp + hw_rei_spe // out to exec + + ALIGN_BLOCK +//+ +//invalid_ipte_handler +// TNV detected on level 3 pte, sort out TNV vs ACV +// +// on entry: +// with +// R8 = pte +// R10 = pc +// +// Function +// Determine TNV vs ACV. Build stack and dispatch. +//- + +invalid_ipte_handler: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, invalid_ipte_handler_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + + srl r8, osfpte_v_ure-osfpte_v_kre, r8 // move pte user bits to kern + nop + +invalid_ipte_handler_10_: srl r8, osfpte_v_kre, r25 // get kre to <0> + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + or r10, r31, r14 // Save pc/va in case TBmiss on stack + mfpr r13, pt_entmm // get entry point + + stq r16, osfsf_a0(sp) // a0 + or r14, r31, r16 // pass pc/va as a0 + + stq r17, osfsf_a1(sp) // a1 + nop + + stq r18, osfsf_a2(sp) // a2 + and r25, 1, r17 // Isolate kre + + stq r16, osfsf_pc(sp) // save pc + xor r17, 1, r17 // map to acv/tnv as a1 + + stq r12, osfsf_ps(sp) // save ps + subq r31, 1, r18 // pass flag of istream as a2 + + stq r29, osfsf_gp(sp) + mtpr r13, exc_addr // set vector address + + mfpr r29, pt_kgp // load kgp + hw_rei_spe // out to exec + + + + + ALIGN_BLOCK +//+ +//invalid_dpte_handler +// INVALID detected on level 3 pte, sort out TNV vs ACV +// +// on entry: +// with +// R10 = va +// R8 = pte +// R9 = mm_stat +// PT6 = pc +// +// Function +// Determine TNV vs ACV. Build stack and dispatch +//- + + +invalid_dpte_handler: + mfpr r12, pt6 + blbs r12, tnv_in_pal // Special handler if original faulting reference was in PALmode + + bis r12, r31, r14 // save PC in case of tbmiss or fault + srl r9, mm_stat_v_opcode, r25 // shift opc to <0> + + mtpr r11, pt0 // Save PS for stack write + and r25, mm_stat_m_opcode, r25 // isolate opcode + + cmpeq r25, evx_opc_sync, r25 // is it FETCH/FETCH_M? + blbs r25, nmiss_fetch_ldr31_err // yes + + //dismiss exception if load to r31/f31 + blbs r9, invalid_dpte_no_dismiss // mm_stat<0> set on store or fetchm + + // not a store or fetch, must be a load + srl r9, mm_stat_v_ra, r25 // Shift rnum to low bits + + and r25, 0x1F, r25 // isolate rnum + nop + + cmpeq r25, 0x1F, r25 // Is the rnum r31 or f31? + bne r25, nmiss_fetch_ldr31_err // Yes, dismiss the fault + +invalid_dpte_no_dismiss: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + bge r25, invalid_dpte_no_dismiss_10_ // no stack swap needed if cm=kern + + srl r8, osfpte_v_ure-osfpte_v_kre, r8 // move pte user bits to kern + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +invalid_dpte_no_dismiss_10_: srl r8, osfpte_v_kre, r12 // get kre to <0> + lda sp, 0-osfsf_c_size(sp)// allocate stack space + + or r10, r31, r25 // Save va in case TBmiss on stack + and r9, 1, r13 // save r/w flag + + stq r16, osfsf_a0(sp) // a0 + or r25, r31, r16 // pass va as a0 + + stq r17, osfsf_a1(sp) // a1 + or r31, mmcsr_c_acv, r17 // assume acv + + srl r12, osfpte_v_kwe-osfpte_v_kre, r25 // get write enable to <0> + stq r29, osfsf_gp(sp) + + stq r18, osfsf_a2(sp) // a2 + cmovlbs r13, r25, r12 // if write access move acv based on write enable + + or r13, r31, r18 // pass flag of dstream access and read vs write + mfpr r25, pt0 // get ps + + stq r14, osfsf_pc(sp) // save pc + mfpr r13, pt_entmm // get entry point + + stq r25, osfsf_ps(sp) // save ps + mtpr r13, exc_addr // set vector address + + mfpr r29, pt_kgp // load kgp + cmovlbs r12, mmcsr_c_tnv, r17 // make p2 be tnv if access ok else acv + + hw_rei_spe // out to exec + +//+ +// +// We come here if we are erring on a dtb_miss, and the instr is a +// fetch, fetch_m, of load to r31/f31. +// The PC is incremented, and we return to the program. +// essentially ignoring the instruction and error. +// +//- + ALIGN_BLOCK +nmiss_fetch_ldr31_err: + mfpr r12, pt6 + addq r12, 4, r12 // bump pc to pc+4 + + mtpr r12, exc_addr // and set entry point + mfpr r31, pt0 // pad exc_addr write + + hw_rei // + + ALIGN_BLOCK +//+ +// double_pte_inv +// We had a single tbmiss which turned into a double tbmiss which found +// an invalid PTE. Return to single miss with a fake pte, and the invalid +// single miss flow will report the error. +// +// on entry: +// r21 PTE +// r22 available +// VA IPR locked with original fault VA +// pt4 saved r21 +// pt5 saved r22 +// pt6 original exc_addr +// +// on return to tbmiss flow: +// r8 fake PTE +// +// +//- +double_pte_inv: + srl r21, osfpte_v_kre, r21 // get the kre bit to <0> + mfpr r22, exc_addr // get the pc + + lda r22, 4(r22) // inc the pc + lda r8, osfpte_m_prot(r31) // make a fake pte with xre and xwe set + + cmovlbc r21, r31, r8 // set to all 0 for acv if pte<kre> is 0 + mtpr r22, exc_addr // set for rei + + mfpr r21, pt4 // restore regs + mfpr r22, pt5 // restore regs + + hw_rei // back to tb miss + + ALIGN_BLOCK +//+ +//tnv_in_pal +// The only places in pal that ld or store are the +// stack builders, rti or retsys. Any of these mean we +// need to take a ksp not valid halt. +// +//- +tnv_in_pal: + + + br r31, ksp_inval_halt + + +// .sbttl "Icache flush routines" + + ALIGN_BLOCK +//+ +// Common Icache flush routine. +// +// +//- +pal_ic_flush: + nop + mtpr r31, ev5__ic_flush_ctl // Icache flush - E1 + nop + nop + +// Now, do 44 NOPs. 3RFB prefetches (24) + IC buffer,IB,slot,issue (20) + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 10 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 20 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 30 + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 40 + + nop + nop + +one_cycle_and_hw_rei: + nop + nop + + hw_rei_stall + +#if icflush_on_tbix != 0 + + + ALIGN_BLOCK + +//+ +// Common Icache flush and ITB invalidate single routine. +// ITBIS and hw_rei_stall must be in same octaword. +// r17 - has address to invalidate +// +//- +PAL_IC_FLUSH_AND_TBISI: + nop + mtpr r31, ev5__ic_flush_ctl // Icache flush - E1 + nop + nop + +// Now, do 44 NOPs. 3RFB prefetches (24) + IC buffer,IB,slot,issue (20) + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 10 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 20 + + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 30 + nop + nop + nop + nop + + nop + nop + nop + nop + + nop + nop // 40 + + + nop + nop + + nop + nop + + // A quadword is 64 bits, so an octaword is 128 bits -> 16 bytes -> 4 instructions + // 44 nops plus 4 instructions before it is 48 instructions. + // Since this routine started on a 32-byte (8 instruction) boundary, + // the following 2 instructions will be in the same octword as required. +// ALIGN_BRANCH + mtpr r17, ev5__itb_is // Flush ITB + hw_rei_stall + +#endif + + ALIGN_BLOCK +//+ +//osfpal_calpal_opcdec +// Here for all opcdec CALL_PALs +// +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + +osfpal_calpal_opcdec: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + mfpr r14, exc_addr // get pc + nop + + bis r11, r31, r12 // Save PS for stack write + bge r25, osfpal_calpal_opcdec_10_ // no stack swap needed if cm=kern + + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +osfpal_calpal_opcdec_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + nop + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_opdec, r16 // set a0 + + stq r18, osfsf_a2(sp) // a2 + mfpr r13, pt_entif // get entry point + + stq r12, osfsf_ps(sp) // save old ps + stq r17, osfsf_a1(sp) // a1 + + stq r14, osfsf_pc(sp) // save pc + nop + + stq r29, osfsf_gp(sp) // save gp + mtpr r13, exc_addr // load exc_addr with entIF + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kgp + + + hw_rei_spe // done + + + + + +//+ +//pal_update_pcb +// Update the PCB with the current SP, AST, and CC info +// +// r0 - return linkage +//- + ALIGN_BLOCK + +pal_update_pcb: + mfpr r12, pt_pcbb // get pcbb + and r11, osfps_m_mode, r25 // get mode + beq r25, pal_update_pcb_10_ // in kern? no need to update user sp + mtpr r30, pt_usp // save user stack + stqp r30, osfpcb_q_usp(r12) // store usp + br r31, pal_update_pcb_20_ // join common +pal_update_pcb_10_: stqp r30, osfpcb_q_ksp(r12) // store ksp +pal_update_pcb_20_: rpcc r13 // get cyccounter + srl r13, 32, r14 // move offset + addl r13, r14, r14 // merge for new time + stlp r14, osfpcb_l_cc(r12) // save time + +//orig pvc_jsr updpcb, bsr=1, dest=1 + ret r31, (r0) + + + +#if remove_save_state == 0 + +// .sbttl "PAL_SAVE_STATE" +//+ +// +// Pal_save_state +// +// Function +// All chip state saved, all PT's, SR's FR's, IPR's +// +// +// Regs' on entry... +// +// R0 = halt code +// pt0 = r0 +// R1 = pointer to impure +// pt4 = r1 +// R3 = return addr +// pt5 = r3 +// +// register usage: +// r0 = halt_code +// r1 = addr of impure area +// r3 = return_address +// r4 = scratch +// +//- + + + ALIGN_BLOCK + .globl pal_save_state +pal_save_state: +// +// +// start of implementation independent save routine +// +// the impure area is larger than the addressibility of hw_ld and hw_st +// therefore, we need to play some games: The impure area +// is informally divided into the "machine independent" part and the +// "machine dependent" part. The state that will be saved in the +// "machine independent" part are gpr's, fpr's, hlt, flag, mchkflag (use (un)fix_impure_gpr macros). +// All others will be in the "machine dependent" part (use (un)fix_impure_ipr macros). +// The impure pointer will need to be adjusted by a different offset for each. The store/restore_reg +// macros will automagically adjust the offset correctly. +// + +// The distributed code is commented out and followed by corresponding SRC code. +// Beware: SAVE_IPR and RESTORE_IPR blow away r0(v0) + +//orig fix_impure_gpr r1 // adjust impure area pointer for stores to "gpr" part of impure area + lda r1, 0x200(r1) // Point to center of CPU segment +//orig store_reg1 flag, r31, r1, ipr=1 // clear dump area flag + SAVE_GPR(r31,CNS_Q_FLAG,r1) // Clear the valid flag +//orig store_reg1 hlt, r0, r1, ipr=1 + SAVE_GPR(r0,CNS_Q_HALT,r1) // Save the halt code + + mfpr r0, pt0 // get r0 back //orig +//orig store_reg1 0, r0, r1 // save r0 + SAVE_GPR(r0,CNS_Q_GPR+0x00,r1) // Save r0 + + mfpr r0, pt4 // get r1 back //orig +//orig store_reg1 1, r0, r1 // save r1 + SAVE_GPR(r0,CNS_Q_GPR+0x08,r1) // Save r1 + +//orig store_reg 2 // save r2 + SAVE_GPR(r2,CNS_Q_GPR+0x10,r1) // Save r2 + + mfpr r0, pt5 // get r3 back //orig +//orig store_reg1 3, r0, r1 // save r3 + SAVE_GPR(r0,CNS_Q_GPR+0x18,r1) // Save r3 + + // reason code has been saved + // r0 has been saved + // r1 has been saved + // r2 has been saved + // r3 has been saved + // pt0, pt4, pt5 have been lost + + // + // Get out of shadow mode + // + + mfpr r2, icsr // Get icsr //orig +//orig ldah r0, <1@<icsr_v_sde-16>>(r31) // Get a one in SHADOW_ENABLE bit location + ldah r0, (1<<(icsr_v_sde-16))(r31) + bic r2, r0, r0 // ICSR with SDE clear //orig + mtpr r0, icsr // Turn off SDE //orig + + mfpr r31, pt0 // SDE bubble cycle 1 //orig + mfpr r31, pt0 // SDE bubble cycle 2 //orig + mfpr r31, pt0 // SDE bubble cycle 3 //orig + nop //orig + + + // save integer regs R4-r31 +//orig #define t 4 +//orig .repeat 28 +//orig store_reg \t +//orig #define t t + 1 +//orig .endr + SAVE_GPR(r4,CNS_Q_GPR+0x20,r1) + SAVE_GPR(r5,CNS_Q_GPR+0x28,r1) + SAVE_GPR(r6,CNS_Q_GPR+0x30,r1) + SAVE_GPR(r7,CNS_Q_GPR+0x38,r1) + SAVE_GPR(r8,CNS_Q_GPR+0x40,r1) + SAVE_GPR(r9,CNS_Q_GPR+0x48,r1) + SAVE_GPR(r10,CNS_Q_GPR+0x50,r1) + SAVE_GPR(r11,CNS_Q_GPR+0x58,r1) + SAVE_GPR(r12,CNS_Q_GPR+0x60,r1) + SAVE_GPR(r13,CNS_Q_GPR+0x68,r1) + SAVE_GPR(r14,CNS_Q_GPR+0x70,r1) + SAVE_GPR(r15,CNS_Q_GPR+0x78,r1) + SAVE_GPR(r16,CNS_Q_GPR+0x80,r1) + SAVE_GPR(r17,CNS_Q_GPR+0x88,r1) + SAVE_GPR(r18,CNS_Q_GPR+0x90,r1) + SAVE_GPR(r19,CNS_Q_GPR+0x98,r1) + SAVE_GPR(r20,CNS_Q_GPR+0xA0,r1) + SAVE_GPR(r21,CNS_Q_GPR+0xA8,r1) + SAVE_GPR(r22,CNS_Q_GPR+0xB0,r1) + SAVE_GPR(r23,CNS_Q_GPR+0xB8,r1) + SAVE_GPR(r24,CNS_Q_GPR+0xC0,r1) + SAVE_GPR(r25,CNS_Q_GPR+0xC8,r1) + SAVE_GPR(r26,CNS_Q_GPR+0xD0,r1) + SAVE_GPR(r27,CNS_Q_GPR+0xD8,r1) + SAVE_GPR(r28,CNS_Q_GPR+0xE0,r1) + SAVE_GPR(r29,CNS_Q_GPR+0xE8,r1) + SAVE_GPR(r30,CNS_Q_GPR+0xF0,r1) + SAVE_GPR(r31,CNS_Q_GPR+0xF8,r1) + + // save all paltemp regs except pt0 + +//orig unfix_impure_gpr r1 // adjust impure area pointer for gpr stores +//orig fix_impure_ipr r1 // adjust impure area pointer for pt stores +//orig #define t 1 +//orig .repeat 23 +//orig store_reg \t , pal=1 +//orig #define t t + 1 +//orig .endr + + lda r1, -0x200(r1) // Restore the impure base address. + lda r1, CNS_Q_IPR(r1) // Point to the base of IPR area. + SAVE_IPR(pt0,CNS_Q_PT+0x00,r1) // the osf code didn't save/restore palTemp 0 ?? pboyle + SAVE_IPR(pt1,CNS_Q_PT+0x08,r1) + SAVE_IPR(pt2,CNS_Q_PT+0x10,r1) + SAVE_IPR(pt3,CNS_Q_PT+0x18,r1) + SAVE_IPR(pt4,CNS_Q_PT+0x20,r1) + SAVE_IPR(pt5,CNS_Q_PT+0x28,r1) + SAVE_IPR(pt6,CNS_Q_PT+0x30,r1) + SAVE_IPR(pt7,CNS_Q_PT+0x38,r1) + SAVE_IPR(pt8,CNS_Q_PT+0x40,r1) + SAVE_IPR(pt9,CNS_Q_PT+0x48,r1) + SAVE_IPR(pt10,CNS_Q_PT+0x50,r1) + SAVE_IPR(pt11,CNS_Q_PT+0x58,r1) + SAVE_IPR(pt12,CNS_Q_PT+0x60,r1) + SAVE_IPR(pt13,CNS_Q_PT+0x68,r1) + SAVE_IPR(pt14,CNS_Q_PT+0x70,r1) + SAVE_IPR(pt15,CNS_Q_PT+0x78,r1) + SAVE_IPR(pt16,CNS_Q_PT+0x80,r1) + SAVE_IPR(pt17,CNS_Q_PT+0x88,r1) + SAVE_IPR(pt18,CNS_Q_PT+0x90,r1) + SAVE_IPR(pt19,CNS_Q_PT+0x98,r1) + SAVE_IPR(pt20,CNS_Q_PT+0xA0,r1) + SAVE_IPR(pt21,CNS_Q_PT+0xA8,r1) + SAVE_IPR(pt22,CNS_Q_PT+0xB0,r1) + SAVE_IPR(pt23,CNS_Q_PT+0xB8,r1) + + // Restore shadow mode + mfpr r31, pt0 // pad write to icsr out of shadow of store (trap does not abort write) //orig + mfpr r31, pt0 //orig + mtpr r2, icsr // Restore original ICSR //orig + + mfpr r31, pt0 // SDE bubble cycle 1 //orig + mfpr r31, pt0 // SDE bubble cycle 2 //orig + mfpr r31, pt0 // SDE bubble cycle 3 //orig + nop //orig + + // save all integer shadow regs + +//orig #define t 8 +//orig .repeat 7 +//orig store_reg \t, shadow=1 +//orig #define t t + 1 +//orig .endr +//orig store_reg 25, shadow=1 + + SAVE_SHADOW( r8,CNS_Q_SHADOW+0x00,r1) // also called p0...p7 in the Hudson code + SAVE_SHADOW( r9,CNS_Q_SHADOW+0x08,r1) + SAVE_SHADOW(r10,CNS_Q_SHADOW+0x10,r1) + SAVE_SHADOW(r11,CNS_Q_SHADOW+0x18,r1) + SAVE_SHADOW(r12,CNS_Q_SHADOW+0x20,r1) + SAVE_SHADOW(r13,CNS_Q_SHADOW+0x28,r1) + SAVE_SHADOW(r14,CNS_Q_SHADOW+0x30,r1) + SAVE_SHADOW(r25,CNS_Q_SHADOW+0x38,r1) + +//orig store_reg exc_addr, ipr=1 // save ipr +//orig store_reg pal_base, ipr=1 // save ipr +//orig store_reg mm_stat, ipr=1 // save ipr +//orig store_reg va, ipr=1 // save ipr +//orig store_reg icsr, ipr=1 // save ipr +//orig store_reg ipl, ipr=1 // save ipr +//orig store_reg ps, ipr=1 // save ipr +//orig store_reg itb_asn, ipr=1 // save ipr +//orig store_reg aster, ipr=1 // save ipr +//orig store_reg astrr, ipr=1 // save ipr +//orig store_reg sirr, ipr=1 // save ipr +//orig store_reg isr, ipr=1 // save ipr +//orig store_reg ivptbr, ipr=1 // save ipr +//orig store_reg mcsr, ipr=1 // save ipr +//orig store_reg dc_mode, ipr=1 // save ipr + + SAVE_IPR(excAddr,CNS_Q_EXC_ADDR,r1) + SAVE_IPR(palBase,CNS_Q_PAL_BASE,r1) + SAVE_IPR(mmStat,CNS_Q_MM_STAT,r1) + SAVE_IPR(va,CNS_Q_VA,r1) + SAVE_IPR(icsr,CNS_Q_ICSR,r1) + SAVE_IPR(ipl,CNS_Q_IPL,r1) + SAVE_IPR(ips,CNS_Q_IPS,r1) + SAVE_IPR(itbAsn,CNS_Q_ITB_ASN,r1) + SAVE_IPR(aster,CNS_Q_ASTER,r1) + SAVE_IPR(astrr,CNS_Q_ASTRR,r1) + SAVE_IPR(sirr,CNS_Q_SIRR,r1) + SAVE_IPR(isr,CNS_Q_ISR,r1) + SAVE_IPR(iVptBr,CNS_Q_IVPTBR,r1) + SAVE_IPR(mcsr,CNS_Q_MCSR,r1) + SAVE_IPR(dcMode,CNS_Q_DC_MODE,r1) + +//orig pvc_violate 379 // mf maf_mode after a store ok (pvc doesn't distinguish ld from st) +//orig store_reg maf_mode, ipr=1 // save ipr -- no mbox instructions for +//orig // PVC violation applies only to +pvc$osf35$379: // loads. HW_ST ok here, so ignore + SAVE_IPR(mafMode,CNS_Q_MAF_MODE,r1) // MBOX INST->MF MAF_MODE IN 0,1,2 + + + //the following iprs are informational only -- will not be restored + +//orig store_reg icperr_stat, ipr=1 +//orig store_reg pmctr, ipr=1 +//orig store_reg intid, ipr=1 +//orig store_reg exc_sum, ipr=1 +//orig store_reg exc_mask, ipr=1 +//orig ldah r14, 0xfff0(r31) +//orig zap r14, 0xE0, r14 // Get Cbox IPR base +//orig nop // pad mf dcperr_stat out of shadow of last store +//orig nop +//orig nop +//orig store_reg dcperr_stat, ipr=1 + + SAVE_IPR(icPerr,CNS_Q_ICPERR_STAT,r1) + SAVE_IPR(PmCtr,CNS_Q_PM_CTR,r1) + SAVE_IPR(intId,CNS_Q_INT_ID,r1) + SAVE_IPR(excSum,CNS_Q_EXC_SUM,r1) + SAVE_IPR(excMask,CNS_Q_EXC_MASK,r1) + ldah r14, 0xFFF0(zero) + zap r14, 0xE0, r14 // Get base address of CBOX IPRs + NOP // Pad mfpr dcPerr out of shadow of + NOP // last store + NOP + SAVE_IPR(dcPerr,CNS_Q_DCPERR_STAT,r1) + + // read cbox ipr state + +//orig mb +//orig ldqp r2, ev5__sc_ctl(r14) +//orig ldqp r13, ld_lock(r14) +//orig ldqp r4, ev5__sc_addr(r14) +//orig ldqp r5, ev5__ei_addr(r14) +//orig ldqp r6, ev5__bc_tag_addr(r14) +//orig ldqp r7, ev5__fill_syn(r14) +//orig bis r5, r4, r31 +//orig bis r7, r6, r31 // make sure previous loads finish before reading stat registers which unlock them +//orig ldqp r8, ev5__sc_stat(r14) // unlocks sc_stat,sc_addr +//orig ldqp r9, ev5__ei_stat(r14) // may unlock ei_*, bc_tag_addr, fill_syn +//orig ldqp r31, ev5__ei_stat(r14) // ensures it is really unlocked +//orig mb + +#ifndef SIMOS + mb + ldq_p r2, scCtl(r14) + ldq_p r13, ldLock(r14) + ldq_p r4, scAddr(r14) + ldq_p r5, eiAddr(r14) + ldq_p r6, bcTagAddr(r14) + ldq_p r7, fillSyn(r14) + bis r5, r4, zero // Make sure all loads complete before + bis r7, r6, zero // reading registers that unlock them. + ldq_p r8, scStat(r14) // Unlocks scAddr. + ldq_p r9, eiStat(r14) // Unlocks eiAddr, bcTagAddr, fillSyn. + ldq_p zero, eiStat(r14) // Make sure it is really unlocked. + mb +#endif +//orig // save cbox ipr state +//orig store_reg1 sc_ctl, r2, r1, ipr=1 +//orig store_reg1 ld_lock, r13, r1, ipr=1 +//orig store_reg1 sc_addr, r4, r1, ipr=1 +//orig store_reg1 ei_addr, r5, r1, ipr=1 +//orig store_reg1 bc_tag_addr, r6, r1, ipr=1 +//orig store_reg1 fill_syn, r7, r1, ipr=1 +//orig store_reg1 sc_stat, r8, r1, ipr=1 +//orig store_reg1 ei_stat, r9, r1, ipr=1 +//orig //bc_config? sl_rcv? + + SAVE_SHADOW(r2,CNS_Q_SC_CTL,r1); + SAVE_SHADOW(r13,CNS_Q_LD_LOCK,r1); + SAVE_SHADOW(r4,CNS_Q_SC_ADDR,r1); + SAVE_SHADOW(r5,CNS_Q_EI_ADDR,r1); + SAVE_SHADOW(r6,CNS_Q_BC_TAG_ADDR,r1); + SAVE_SHADOW(r7,CNS_Q_FILL_SYN,r1); + SAVE_SHADOW(r8,CNS_Q_SC_STAT,r1); + SAVE_SHADOW(r9,CNS_Q_EI_STAT,r1); + +// restore impure base //orig +//orig unfix_impure_ipr r1 + lda r1, -CNS_Q_IPR(r1) + +// save all floating regs //orig + mfpr r0, icsr // get icsr //orig + or r31, 1, r2 // get a one //orig +//orig sll r2, #icsr_v_fpe, r2 // shift for fpu spot //orig + sll r2, icsr_v_fpe, r2 // Shift it into ICSR<FPE> position + or r2, r0, r0 // set FEN on //orig + mtpr r0, icsr // write to icsr, enabling FEN //orig + +// map the save area virtually +// orig mtpr r31, dtb_ia // clear the dtb +// orig srl r1, page_offset_size_bits, r0 // Clean off low bits of VA +// orig sll r0, 32, r0 // shift to PFN field +// orig lda r2, 0xff(r31) // all read enable and write enable bits set +// orig sll r2, 8, r2 // move to PTE location +// orig addq r0, r2, r0 // combine with PFN +// orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +// orig mtpr r1, dtb_tag // write TB tag + + mtpr r31, dtbIa // Clear all DTB entries + srl r1, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r1, dtbTag // Write the PTE and tag into the DTB + + +//orig // map the next page too - in case the impure area crosses a page boundary +//orig lda r4, 1@page_offset_size_bits(r1) // generate address for next page +//orig srl r4, page_offset_size_bits, r0 // Clean off low bits of VA +//orig sll r0, 32, r0 // shift to PFN field +//orig lda r2, 0xff(r31) // all read enable and write enable bits set +//orig sll r2, 8, r2 // move to PTE location +//orig addq r0, r2, r0 // combine with PFN +//orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +//orig mtpr r4, dtb_tag // write TB tag + + lda r4, (1<<va_s_off)(r1) // Generate address for next page + srl r4, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r4, dtbTag // Write the PTE and tag into the DTB + + sll r31, 0, r31 // stall cycle 1 // orig + sll r31, 0, r31 // stall cycle 2 // orig + sll r31, 0, r31 // stall cycle 3 // orig + nop // orig + +//orig // add offset for saving fpr regs +//orig fix_impure_gpr r1 + + lda r1, 0x200(r1) // Point to center of CPU segment + +// now save the regs - F0-F31 + +//orig #define t 0 +//orig .repeat 32 +//orig store_reg \t , fpu=1 +//orig #define t t + 1 +//orig .endr + + mf_fpcr f0 // original + + SAVE_FPR(f0,CNS_Q_FPR+0x00,r1) + SAVE_FPR(f1,CNS_Q_FPR+0x08,r1) + SAVE_FPR(f2,CNS_Q_FPR+0x10,r1) + SAVE_FPR(f3,CNS_Q_FPR+0x18,r1) + SAVE_FPR(f4,CNS_Q_FPR+0x20,r1) + SAVE_FPR(f5,CNS_Q_FPR+0x28,r1) + SAVE_FPR(f6,CNS_Q_FPR+0x30,r1) + SAVE_FPR(f7,CNS_Q_FPR+0x38,r1) + SAVE_FPR(f8,CNS_Q_FPR+0x40,r1) + SAVE_FPR(f9,CNS_Q_FPR+0x48,r1) + SAVE_FPR(f10,CNS_Q_FPR+0x50,r1) + SAVE_FPR(f11,CNS_Q_FPR+0x58,r1) + SAVE_FPR(f12,CNS_Q_FPR+0x60,r1) + SAVE_FPR(f13,CNS_Q_FPR+0x68,r1) + SAVE_FPR(f14,CNS_Q_FPR+0x70,r1) + SAVE_FPR(f15,CNS_Q_FPR+0x78,r1) + SAVE_FPR(f16,CNS_Q_FPR+0x80,r1) + SAVE_FPR(f17,CNS_Q_FPR+0x88,r1) + SAVE_FPR(f18,CNS_Q_FPR+0x90,r1) + SAVE_FPR(f19,CNS_Q_FPR+0x98,r1) + SAVE_FPR(f20,CNS_Q_FPR+0xA0,r1) + SAVE_FPR(f21,CNS_Q_FPR+0xA8,r1) + SAVE_FPR(f22,CNS_Q_FPR+0xB0,r1) + SAVE_FPR(f23,CNS_Q_FPR+0xB8,r1) + SAVE_FPR(f24,CNS_Q_FPR+0xC0,r1) + SAVE_FPR(f25,CNS_Q_FPR+0xC8,r1) + SAVE_FPR(f26,CNS_Q_FPR+0xD0,r1) + SAVE_FPR(f27,CNS_Q_FPR+0xD8,r1) + SAVE_FPR(f28,CNS_Q_FPR+0xE0,r1) + SAVE_FPR(f29,CNS_Q_FPR+0xE8,r1) + SAVE_FPR(f30,CNS_Q_FPR+0xF0,r1) + SAVE_FPR(f31,CNS_Q_FPR+0xF8,r1) + +//orig //switch impure offset from gpr to ipr--- +//orig unfix_impure_gpr r1 +//orig fix_impure_ipr r1 +//orig store_reg1 fpcsr, f0, r1, fpcsr=1 + + SAVE_FPR(f0,CNS_Q_FPCSR,r1) // fpcsr loaded above into f0 -- can it reach// pb + lda r1, -0x200(r1) // Restore the impure base address + +//orig // and back to gpr --- +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 + +//orig lda r0, cns_mchksize(r31) // get size of mchk area +//orig store_reg1 mchkflag, r0, r1, ipr=1 +//orig mb + + lda r1, CNS_Q_IPR(r1) // Point to base of IPR area again + // save this using the IPR base (it is closer) not the GRP base as they used...pb + lda r0, MACHINE_CHECK_SIZE(r31) // get size of mchk area + SAVE_SHADOW(r0,CNS_Q_MCHK,r1); + mb + +//orig or r31, 1, r0 // get a one +//orig store_reg1 flag, r0, r1, ipr=1 // set dump area flag +//orig mb + + lda r1, -CNS_Q_IPR(r1) // back to the base + lda r1, 0x200(r1) // Point to center of CPU segment + or r31, 1, r0 // get a one + SAVE_GPR(r0,CNS_Q_FLAG,r1) // // set dump area valid flag + mb + +//orig // restore impure area base +//orig unfix_impure_gpr r1 + lda r1, -0x200(r1) // Point to center of CPU segment + + mtpr r31, dtb_ia // clear the dtb //orig + mtpr r31, itb_ia // clear the itb //orig + +//orig pvc_jsr savsta, bsr=1, dest=1 + ret r31, (r3) // and back we go +#endif + + +#if remove_restore_state == 0 + + +// .sbttl "PAL_RESTORE_STATE" +//+ +// +// Pal_restore_state +// +// +// register usage: +// r1 = addr of impure area +// r3 = return_address +// all other regs are scratchable, as they are about to +// be reloaded from ram. +// +// Function: +// All chip state restored, all SRs, FRs, PTs, IPRs +// *** except R1, R3, PT0, PT4, PT5 *** +// +//- + ALIGN_BLOCK +pal_restore_state: + +//need to restore sc_ctl,bc_ctl,bc_config??? if so, need to figure out a safe way to do so. + +//orig // map the console io area virtually +//orig mtpr r31, dtb_ia // clear the dtb +//orig srl r1, page_offset_size_bits, r0 // Clean off low bits of VA +//orig sll r0, 32, r0 // shift to PFN field +//orig lda r2, 0xff(r31) // all read enable and write enable bits set +//orig sll r2, 8, r2 // move to PTE location +//orig addq r0, r2, r0 // combine with PFN +//orig +//orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +//orig mtpr r1, dtb_tag // write TB tag +//orig + + mtpr r31, dtbIa // Clear all DTB entries + srl r1, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r1, dtbTag // Write the PTE and tag into the DTB + + +//orig // map the next page too, in case impure area crosses page boundary +//orig lda r4, 1@page_offset_size_bits(r1) // generate address for next page +//orig srl r4, page_offset_size_bits, r0 // Clean off low bits of VA +//orig sll r0, 32, r0 // shift to PFN field +//orig lda r2, 0xff(r31) // all read enable and write enable bits set +//orig sll r2, 8, r2 // move to PTE location +//orig addq r0, r2, r0 // combine with PFN +//orig +//orig mtpr r0, dtb_pte // Load PTE and set TB valid bit +//orig mtpr r4, dtb_tag // write TB tag - no virtual mbox instruction for 3 cycles + + lda r4, (1<<VA_S_OFF)(r1) // Generate address for next page + srl r4, va_s_off, r0 // Clean off byte-within-page offset + sll r0, pte_v_pfn, r0 // Shift to form PFN + lda r0, pte_m_prot(r0) // Set all read/write enable bits + mtpr r0, dtbPte // Load the PTE and set valid + mtpr r4, dtbTag // Write the PTE and tag into the DTB + +//orig // save all floating regs +//orig mfpr r0, icsr // get icsr +//orig// assume ICSR_V_SDE gt <ICSR_V_FPE> // assertion checker +//orig or r31, <<1@<ICSR_V_SDE-ICSR_V_FPE>> ! 1>, r2 // set SDE and FPE +//orig sll r2, #icsr_v_fpe, r2 // shift for fpu spot +//orig or r2, r0, r0 // set FEN on +//orig mtpr r0, icsr // write to icsr, enabling FEN and SDE. 3 bubbles to floating instr. + + mfpr r0, icsr // Get current ICSR + bis zero, 1, r2 // Get a '1' + or r2, (1<<(icsr_v_sde-icsr_v_fpe)), r2 + sll r2, icsr_v_fpe, r2 // Shift bits into position + bis r2, r2, r0 // Set ICSR<SDE> and ICSR<FPE> + mtpr r0, icsr // Update the chip + + mfpr r31, pt0 // FPE bubble cycle 1 //orig + mfpr r31, pt0 // FPE bubble cycle 2 //orig + mfpr r31, pt0 // FPE bubble cycle 3 //orig + +//orig fix_impure_ipr r1 +//orig restore_reg1 fpcsr, f0, r1, fpcsr=1 +//orig mt_fpcr f0 +//orig +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 // adjust impure pointer offset for gpr access +//orig +//orig // restore all floating regs +//orig#define t 0 +//orig .repeat 32 +//orig restore_reg \t , fpu=1 +//orig#define t t + 1 +//orig .endr + + lda r1, 200(r1) // Point to base of IPR area again + RESTORE_FPR(f0,CNS_Q_FPCSR,r1) // can it reach?? pb + mt_fpcr f0 // original + + lda r1, 0x200(r1) // point to center of CPU segment + RESTORE_FPR(f0,CNS_Q_FPR+0x00,r1) + RESTORE_FPR(f1,CNS_Q_FPR+0x08,r1) + RESTORE_FPR(f2,CNS_Q_FPR+0x10,r1) + RESTORE_FPR(f3,CNS_Q_FPR+0x18,r1) + RESTORE_FPR(f4,CNS_Q_FPR+0x20,r1) + RESTORE_FPR(f5,CNS_Q_FPR+0x28,r1) + RESTORE_FPR(f6,CNS_Q_FPR+0x30,r1) + RESTORE_FPR(f7,CNS_Q_FPR+0x38,r1) + RESTORE_FPR(f8,CNS_Q_FPR+0x40,r1) + RESTORE_FPR(f9,CNS_Q_FPR+0x48,r1) + RESTORE_FPR(f10,CNS_Q_FPR+0x50,r1) + RESTORE_FPR(f11,CNS_Q_FPR+0x58,r1) + RESTORE_FPR(f12,CNS_Q_FPR+0x60,r1) + RESTORE_FPR(f13,CNS_Q_FPR+0x68,r1) + RESTORE_FPR(f14,CNS_Q_FPR+0x70,r1) + RESTORE_FPR(f15,CNS_Q_FPR+0x78,r1) + RESTORE_FPR(f16,CNS_Q_FPR+0x80,r1) + RESTORE_FPR(f17,CNS_Q_FPR+0x88,r1) + RESTORE_FPR(f18,CNS_Q_FPR+0x90,r1) + RESTORE_FPR(f19,CNS_Q_FPR+0x98,r1) + RESTORE_FPR(f20,CNS_Q_FPR+0xA0,r1) + RESTORE_FPR(f21,CNS_Q_FPR+0xA8,r1) + RESTORE_FPR(f22,CNS_Q_FPR+0xB0,r1) + RESTORE_FPR(f23,CNS_Q_FPR+0xB8,r1) + RESTORE_FPR(f24,CNS_Q_FPR+0xC0,r1) + RESTORE_FPR(f25,CNS_Q_FPR+0xC8,r1) + RESTORE_FPR(f26,CNS_Q_FPR+0xD0,r1) + RESTORE_FPR(f27,CNS_Q_FPR+0xD8,r1) + RESTORE_FPR(f28,CNS_Q_FPR+0xE0,r1) + RESTORE_FPR(f29,CNS_Q_FPR+0xE8,r1) + RESTORE_FPR(f30,CNS_Q_FPR+0xF0,r1) + RESTORE_FPR(f31,CNS_Q_FPR+0xF8,r1) + +//orig // switch impure pointer from gpr to ipr area -- +//orig unfix_impure_gpr r1 +//orig fix_impure_ipr r1 +//orig +//orig // restore all pal regs +//orig#define t 1 +//orig .repeat 23 +//orig restore_reg \t , pal=1 +//orig#define t t + 1 +//orig .endr + + lda r1, -0x200(r1) // Restore base address of impure area. + lda r1, CNS_Q_IPR(r1) // Point to base of IPR area. + RESTORE_IPR(pt0,CNS_Q_PT+0x00,r1) // the osf code didn't save/restore palTemp 0 ?? pboyle + RESTORE_IPR(pt1,CNS_Q_PT+0x08,r1) + RESTORE_IPR(pt2,CNS_Q_PT+0x10,r1) + RESTORE_IPR(pt3,CNS_Q_PT+0x18,r1) + RESTORE_IPR(pt4,CNS_Q_PT+0x20,r1) + RESTORE_IPR(pt5,CNS_Q_PT+0x28,r1) + RESTORE_IPR(pt6,CNS_Q_PT+0x30,r1) + RESTORE_IPR(pt7,CNS_Q_PT+0x38,r1) + RESTORE_IPR(pt8,CNS_Q_PT+0x40,r1) + RESTORE_IPR(pt9,CNS_Q_PT+0x48,r1) + RESTORE_IPR(pt10,CNS_Q_PT+0x50,r1) + RESTORE_IPR(pt11,CNS_Q_PT+0x58,r1) + RESTORE_IPR(pt12,CNS_Q_PT+0x60,r1) + RESTORE_IPR(pt13,CNS_Q_PT+0x68,r1) + RESTORE_IPR(pt14,CNS_Q_PT+0x70,r1) + RESTORE_IPR(pt15,CNS_Q_PT+0x78,r1) + RESTORE_IPR(pt16,CNS_Q_PT+0x80,r1) + RESTORE_IPR(pt17,CNS_Q_PT+0x88,r1) + RESTORE_IPR(pt18,CNS_Q_PT+0x90,r1) + RESTORE_IPR(pt19,CNS_Q_PT+0x98,r1) + RESTORE_IPR(pt20,CNS_Q_PT+0xA0,r1) + RESTORE_IPR(pt21,CNS_Q_PT+0xA8,r1) + RESTORE_IPR(pt22,CNS_Q_PT+0xB0,r1) + RESTORE_IPR(pt23,CNS_Q_PT+0xB8,r1) + + +//orig restore_reg exc_addr, ipr=1 // restore ipr +//orig restore_reg pal_base, ipr=1 // restore ipr +//orig restore_reg ipl, ipr=1 // restore ipr +//orig restore_reg ps, ipr=1 // restore ipr +//orig mtpr r0, dtb_cm // set current mode in mbox too +//orig restore_reg itb_asn, ipr=1 +//orig srl r0, itb_asn_v_asn, r0 +//orig sll r0, dtb_asn_v_asn, r0 +//orig mtpr r0, dtb_asn // set ASN in Mbox too +//orig restore_reg ivptbr, ipr=1 +//orig mtpr r0, mvptbr // use ivptbr value to restore mvptbr +//orig restore_reg mcsr, ipr=1 +//orig restore_reg aster, ipr=1 +//orig restore_reg astrr, ipr=1 +//orig restore_reg sirr, ipr=1 +//orig restore_reg maf_mode, ipr=1 // no mbox instruction for 3 cycles +//orig mfpr r31, pt0 // (may issue with mt maf_mode) +//orig mfpr r31, pt0 // bubble cycle 1 +//orig mfpr r31, pt0 // bubble cycle 2 +//orig mfpr r31, pt0 // bubble cycle 3 +//orig mfpr r31, pt0 // (may issue with following ld) + + // r0 gets the value of RESTORE_IPR in the macro and this code uses this side effect (gag) + RESTORE_IPR(excAddr,CNS_Q_EXC_ADDR,r1) + RESTORE_IPR(palBase,CNS_Q_PAL_BASE,r1) + RESTORE_IPR(ipl,CNS_Q_IPL,r1) + RESTORE_IPR(ips,CNS_Q_IPS,r1) + mtpr r0, dtbCm // Set Mbox current mode too. + RESTORE_IPR(itbAsn,CNS_Q_ITB_ASN,r1) + srl r0, 4, r0 + sll r0, 57, r0 + mtpr r0, dtbAsn // Set Mbox ASN too + RESTORE_IPR(iVptBr,CNS_Q_IVPTBR,r1) + mtpr r0, mVptBr // Set Mbox VptBr too + RESTORE_IPR(mcsr,CNS_Q_MCSR,r1) + RESTORE_IPR(aster,CNS_Q_ASTER,r1) + RESTORE_IPR(astrr,CNS_Q_ASTRR,r1) + RESTORE_IPR(sirr,CNS_Q_SIRR,r1) + RESTORE_IPR(mafMode,CNS_Q_MAF_MODE,r1) + STALL + STALL + STALL + STALL + STALL + + + // restore all integer shadow regs +//orig#define t 8 +//orig .repeat 7 +//orig restore_reg \t, shadow=1 +//orig#define t t + 1 +//orig .endr +//orig restore_reg 25, shadow=1 +//orig restore_reg dc_mode, ipr=1 // no mbox instructions for 4 cycles + + RESTORE_SHADOW( r8,CNS_Q_SHADOW+0x00,r1) // also called p0...p7 in the Hudson code + RESTORE_SHADOW( r9,CNS_Q_SHADOW+0x08,r1) + RESTORE_SHADOW(r10,CNS_Q_SHADOW+0x10,r1) + RESTORE_SHADOW(r11,CNS_Q_SHADOW+0x18,r1) + RESTORE_SHADOW(r12,CNS_Q_SHADOW+0x20,r1) + RESTORE_SHADOW(r13,CNS_Q_SHADOW+0x28,r1) + RESTORE_SHADOW(r14,CNS_Q_SHADOW+0x30,r1) + RESTORE_SHADOW(r25,CNS_Q_SHADOW+0x38,r1) + RESTORE_IPR(dcMode,CNS_Q_DC_MODE,r1) + + // + // Get out of shadow mode + // + + mfpr r31, pt0 // pad last load to icsr write (in case of replay, icsr will be written anyway) //orig + mfpr r31, pt0 // "" //orig + mfpr r0, icsr // Get icsr //orig +//orig ldah r2, <1@<icsr_v_sde-16>>(r31) // Get a one in SHADOW_ENABLE bit location + ldah r2, (1<<(ICSR_V_SDE-16))(r31) // Get a one in SHADOW_ENABLE bit location //orig + bic r0, r2, r2 // ICSR with SDE clear //orig + mtpr r2, icsr // Turn off SDE - no palshadow rd/wr for 3 bubble cycles //orig + + mfpr r31, pt0 // SDE bubble cycle 1 //orig + mfpr r31, pt0 // SDE bubble cycle 2 //orig + mfpr r31, pt0 // SDE bubble cycle 3 //orig + nop //orig + +//orig // switch impure pointer from ipr to gpr area -- +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 +//orig // restore all integer regs +//orig#define t 4 +//orig .repeat 28 +//orig restore_reg \t +//orig#define t t + 1 +//orig .endr + +// Restore GPRs (r0, r2 are restored later, r1 and r3 are trashed) ... + + lda r1, -CNS_Q_IPR(r1) // Restore base address of impure area + lda r1, 0x200(r1) // Point to center of CPU segment + + RESTORE_GPR(r4,CNS_Q_GPR+0x20,r1) + RESTORE_GPR(r5,CNS_Q_GPR+0x28,r1) + RESTORE_GPR(r6,CNS_Q_GPR+0x30,r1) + RESTORE_GPR(r7,CNS_Q_GPR+0x38,r1) + RESTORE_GPR(r8,CNS_Q_GPR+0x40,r1) + RESTORE_GPR(r9,CNS_Q_GPR+0x48,r1) + RESTORE_GPR(r10,CNS_Q_GPR+0x50,r1) + RESTORE_GPR(r11,CNS_Q_GPR+0x58,r1) + RESTORE_GPR(r12,CNS_Q_GPR+0x60,r1) + RESTORE_GPR(r13,CNS_Q_GPR+0x68,r1) + RESTORE_GPR(r14,CNS_Q_GPR+0x70,r1) + RESTORE_GPR(r15,CNS_Q_GPR+0x78,r1) + RESTORE_GPR(r16,CNS_Q_GPR+0x80,r1) + RESTORE_GPR(r17,CNS_Q_GPR+0x88,r1) + RESTORE_GPR(r18,CNS_Q_GPR+0x90,r1) + RESTORE_GPR(r19,CNS_Q_GPR+0x98,r1) + RESTORE_GPR(r20,CNS_Q_GPR+0xA0,r1) + RESTORE_GPR(r21,CNS_Q_GPR+0xA8,r1) + RESTORE_GPR(r22,CNS_Q_GPR+0xB0,r1) + RESTORE_GPR(r23,CNS_Q_GPR+0xB8,r1) + RESTORE_GPR(r24,CNS_Q_GPR+0xC0,r1) + RESTORE_GPR(r25,CNS_Q_GPR+0xC8,r1) + RESTORE_GPR(r26,CNS_Q_GPR+0xD0,r1) + RESTORE_GPR(r27,CNS_Q_GPR+0xD8,r1) + RESTORE_GPR(r28,CNS_Q_GPR+0xE0,r1) + RESTORE_GPR(r29,CNS_Q_GPR+0xE8,r1) + RESTORE_GPR(r30,CNS_Q_GPR+0xF0,r1) + RESTORE_GPR(r31,CNS_Q_GPR+0xF8,r1) + +//orig // switch impure pointer from gpr to ipr area -- +//orig unfix_impure_gpr r1 +//orig fix_impure_ipr r1 +//orig restore_reg icsr, ipr=1 // restore original icsr- 4 bubbles to hw_rei + + lda t0, -0x200(t0) // Restore base address of impure area. + lda t0, CNS_Q_IPR(t0) // Point to base of IPR area again. + RESTORE_IPR(icsr,CNS_Q_ICSR,r1) + +//orig // and back again -- +//orig unfix_impure_ipr r1 +//orig fix_impure_gpr r1 +//orig store_reg1 flag, r31, r1, ipr=1 // clear dump area valid flag +//orig mb + + lda t0, -CNS_Q_IPR(t0) // Back to base of impure area again, + lda t0, 0x200(t0) // and back to center of CPU segment + SAVE_GPR(r31,CNS_Q_FLAG,r1) // Clear the dump area valid flag + mb + +//orig // and back we go +//orig// restore_reg 3 +//orig restore_reg 2 +//orig// restore_reg 1 +//orig restore_reg 0 +//orig // restore impure area base +//orig unfix_impure_gpr r1 + + RESTORE_GPR(r2,CNS_Q_GPR+0x10,r1) + RESTORE_GPR(r0,CNS_Q_GPR+0x00,r1) + lda r1, -0x200(r1) // Restore impure base address + + mfpr r31, pt0 // stall for ldqp above //orig + + mtpr r31, dtb_ia // clear the tb //orig + mtpr r31, itb_ia // clear the itb //orig + +//orig pvc_jsr rststa, bsr=1, dest=1 + ret r31, (r3) // back we go //orig +#endif + + +//+ +// pal_pal_bug_check -- code has found a bugcheck situation. +// Set things up and join common machine check flow. +// +// Input: +// r14 - exc_addr +// +// On exit: +// pt0 - saved r0 +// pt1 - saved r1 +// pt4 - saved r4 +// pt5 - saved r5 +// pt6 - saved r6 +// pt10 - saved exc_addr +// pt_misc<47:32> - mchk code +// pt_misc<31:16> - scb vector +// r14 - base of Cbox IPRs in IO space +// MCES<mchk> is set +//- + + ALIGN_BLOCK + .globl pal_pal_bug_check_from_int +pal_pal_bug_check_from_int: + DEBUGSTORE(0x79) +//simos DEBUG_EXC_ADDR() + DEBUGSTORE(0x20) +//simos bsr r25, put_hex + lda r25, mchk_c_bugcheck(r31) + addq r25, 1, r25 // set flag indicating we came from interrupt and stack is already pushed + br r31, pal_pal_mchk + nop + +pal_pal_bug_check: + lda r25, mchk_c_bugcheck(r31) + +pal_pal_mchk: + sll r25, 32, r25 // Move mchk code to position + + mtpr r14, pt10 // Stash exc_addr + mtpr r14, exc_addr + + mfpr r12, pt_misc // Get MCES and scratch + zap r12, 0x3c, r12 + + or r12, r25, r12 // Combine mchk code + lda r25, scb_v_procmchk(r31) // Get SCB vector + + sll r25, 16, r25 // Move SCBv to position + or r12, r25, r25 // Combine SCBv + + mtpr r0, pt0 // Stash for scratch + bis r25, mces_m_mchk, r25 // Set MCES<MCHK> bit + + mtpr r25, pt_misc // Save mchk code!scbv!whami!mces + ldah r14, 0xfff0(r31) + + mtpr r1, pt1 // Stash for scratch + zap r14, 0xE0, r14 // Get Cbox IPR base + + mtpr r4, pt4 + mtpr r5, pt5 + + mtpr r6, pt6 + blbs r12, sys_double_machine_check // MCHK halt if double machine check + + br r31, sys_mchk_collect_iprs // Join common machine check flow + +// align_to_call_pal_section // Align to address of first call_pal entry point - 2000 + +// .sbttl "HALT - PALcode for HALT instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// GO to console code +// +//- + + .text 1 +// . = 0x2000 + CALL_PAL_PRIV(PAL_HALT_ENTRY) +call_pal_halt: +#if rax_mode == 0 + mfpr r31, pt0 // Pad exc_addr read + mfpr r31, pt0 + + mfpr r12, exc_addr // get PC + subq r12, 4, r12 // Point to the HALT + + mtpr r12, exc_addr + mtpr r0, pt0 + +//orig pvc_jsr updpcb, bsr=1 + bsr r0, pal_update_pcb // update the pcb + lda r0, hlt_c_sw_halt(r31) // set halt code to sw halt + br r31, sys_enter_console // enter the console + +#else // RAX mode + mb + mb + mtpr r9, ev5__dtb_asn // no Dstream virtual ref for next 3 cycles. + mtpr r9, ev5__itb_asn // E1. Update ITB ASN. No hw_rei for 5 cycles. + mtpr r8, exc_addr // no HW_REI for 1 cycle. + blbc r9, not_begin_case + mtpr r31, ev5__dtb_ia // clear DTB. No Dstream virtual ref for 2 cycles. + mtpr r31, ev5__itb_ia // clear ITB. + +not_begin_case: + nop + nop + + nop + nop // pad mt itb_asn ->hw_rei_stall + + hw_rei_stall +#endif + +// .sbttl "CFLUSH- PALcode for CFLUSH instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// R16 - contains the PFN of the page to be flushed +// +// Function: +// Flush all Dstream caches of 1 entire page +// The CFLUSH routine is in the system specific module. +// +//- + + CALL_PAL_PRIV(PAL_CFLUSH_ENTRY) +Call_Pal_Cflush: + br r31, sys_cflush + +// .sbttl "DRAINA - PALcode for DRAINA instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// Implicit TRAPB performed by hardware. +// +// Function: +// Stall instruction issue until all prior instructions are guaranteed to +// complete without incurring aborts. For the EV5 implementation, this +// means waiting until all pending DREADS are returned. +// +//- + + CALL_PAL_PRIV(PAL_DRAINA_ENTRY) +Call_Pal_Draina: + ldah r14, 0x100(r31) // Init counter. Value? + nop + +DRAINA_LOOP: + subq r14, 1, r14 // Decrement counter + mfpr r13, ev5__maf_mode // Fetch status bit + + srl r13, maf_mode_v_dread_pending, r13 + ble r14, DRAINA_LOOP_TOO_LONG + + nop + blbs r13, DRAINA_LOOP // Wait until all DREADS clear + + hw_rei + +DRAINA_LOOP_TOO_LONG: + br r31, call_pal_halt + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x0003) +CallPal_OpcDec03: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0004) +CallPal_OpcDec04: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0005) +CallPal_OpcDec05: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0006) +CallPal_OpcDec06: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0007) +CallPal_OpcDec07: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0008) +CallPal_OpcDec08: + br r31, osfpal_calpal_opcdec + +// .sbttl "CSERVE- PALcode for CSERVE instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Various functions for private use of console software +// +// option selector in r0 +// arguments in r16.... +// The CSERVE routine is in the system specific module. +// +//- + + CALL_PAL_PRIV(PAL_CSERVE_ENTRY) +Call_Pal_Cserve: + br r31, sys_cserve + +// .sbttl "swppal - PALcode for swppal instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// Vectored into via hardware PALcode instruction dispatch. +// R16 contains the new PAL identifier +// R17:R21 contain implementation-specific entry parameters +// +// R0 receives status: +// 0 success (PAL was switched) +// 1 unknown PAL variant +// 2 known PAL variant, but PAL not loaded +// +// +// Function: +// Swap control to another PAL. +//- + + CALL_PAL_PRIV(PAL_SWPPAL_ENTRY) +Call_Pal_Swppal: + cmpule r16, 255, r0 // see if a kibble was passed + cmoveq r16, r16, r0 // if r16=0 then a valid address (ECO 59) + + or r16, r31, r3 // set r3 incase this is a address + blbc r0, swppal_cont // nope, try it as an address + + cmpeq r16, 2, r0 // is it our friend OSF? + blbc r0, swppal_fail // nope, don't know this fellow + + br r2, CALL_PAL_SWPPAL_10_ // tis our buddy OSF + +// .global osfpal_hw_entry_reset +// .weak osfpal_hw_entry_reset +// .long <osfpal_hw_entry_reset-pal_start> +//orig halt // don't know how to get the address here - kludge ok, load pal at 0 + .long 0 // ?? hack upon hack...pb + +CALL_PAL_SWPPAL_10_: ldlp r3, 0(r2) // fetch target addr +// ble r3, swppal_fail ; if OSF not linked in say not loaded. + mfpr r2, pal_base // fetch pal base + + addq r2, r3, r3 // add pal base + lda r2, 0x3FFF(r31) // get pal base checker mask + + and r3, r2, r2 // any funky bits set? + cmpeq r2, 0, r0 // + + blbc r0, swppal_fail // return unknown if bad bit set. + br r31, swppal_cont + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x000B) +CallPal_OpcDec0B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x000C) +CallPal_OpcDec0C: + br r31, osfpal_calpal_opcdec + +// .sbttl "wripir- PALcode for wripir instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// r16 = processor number to interrupt +// +// Function: +// IPIR <- R16 +// Handled in system-specific code +// +// Exit: +// interprocessor interrupt is recorded on the target processor +// and is initiated when the proper enabling conditions are present. +//- + + CALL_PAL_PRIV(PAL_WRIPIR_ENTRY) +Call_Pal_Wrpir: + br r31, sys_wripir + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x000E) +CallPal_OpcDec0E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x000F) +CallPal_OpcDec0F: + br r31, osfpal_calpal_opcdec + +// .sbttl "rdmces- PALcode for rdmces instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// R0 <- ZEXT(MCES) +//- + + CALL_PAL_PRIV(PAL_RDMCES_ENTRY) +Call_Pal_Rdmces: + mfpr r0, pt_mces // Read from PALtemp + and r0, mces_m_all, r0 // Clear other bits + + hw_rei + +// .sbttl "wrmces- PALcode for wrmces instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// If {R16<0> EQ 1} then MCES<0> <- 0 (MCHK) +// If {R16<1> EQ 1} then MCES<1> <- 0 (SCE) +// If {R16<2> EQ 1} then MCES<2> <- 0 (PCE) +// MCES<3> <- R16<3> (DPC) +// MCES<4> <- R16<4> (DSC) +// +//- + + CALL_PAL_PRIV(PAL_WRMCES_ENTRY) +Call_Pal_Wrmces: + and r16, ((1<<mces_v_mchk) | (1<<mces_v_sce) | (1<<mces_v_pce)), r13 // Isolate MCHK, SCE, PCE + mfpr r14, pt_mces // Get current value + + ornot r31, r13, r13 // Flip all the bits + and r16, ((1<<mces_v_dpc) | (1<<mces_v_dsc)), r17 + + and r14, r13, r1 // Update MCHK, SCE, PCE + bic r1, ((1<<mces_v_dpc) | (1<<mces_v_dsc)), r1 // Clear old DPC, DSC + + or r1, r17, r1 // Update DPC and DSC + mtpr r1, pt_mces // Write MCES back + +#if rawhide_system == 0 + nop // Pad to fix PT write->read restriction +#else + blbs r16, RAWHIDE_clear_mchk_lock // Clear logout from lock +#endif + + nop + hw_rei + + + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_PRIV(0x0012) +CallPal_OpcDec12: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0013) +CallPal_OpcDec13: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0014) +CallPal_OpcDec14: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0015) +CallPal_OpcDec15: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0016) +CallPal_OpcDec16: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0017) +CallPal_OpcDec17: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0018) +CallPal_OpcDec18: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0019) +CallPal_OpcDec19: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001A) +CallPal_OpcDec1A: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001B) +CallPal_OpcDec1B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001C) +CallPal_OpcDec1C: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001D) +CallPal_OpcDec1D: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001E) +CallPal_OpcDec1E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x001F) +CallPal_OpcDec1F: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0020) +CallPal_OpcDec20: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0021) +CallPal_OpcDec21: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0022) +CallPal_OpcDec22: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0023) +CallPal_OpcDec23: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0024) +CallPal_OpcDec24: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0025) +CallPal_OpcDec25: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0026) +CallPal_OpcDec26: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0027) +CallPal_OpcDec27: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0028) +CallPal_OpcDec28: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x0029) +CallPal_OpcDec29: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x002A) +CallPal_OpcDec2A: + br r31, osfpal_calpal_opcdec + +// .sbttl "wrfen - PALcode for wrfen instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// a0<0> -> ICSR<FPE> +// Store new FEN in PCB +// Final value of t0 (r1), t8..t10 (r22..r24) and a0 (r16) are UNPREDICTABLE +// +// Issue: What about pending FP loads when FEN goes from on->off???? +//- + + CALL_PAL_PRIV(PAL_WRFEN_ENTRY) +Call_Pal_Wrfen: + or r31, 1, r13 // Get a one + mfpr r1, ev5__icsr // Get current FPE + + sll r13, icsr_v_fpe, r13 // shift 1 to icsr<fpe> spot, e0 + and r16, 1, r16 // clean new fen + + sll r16, icsr_v_fpe, r12 // shift new fen to correct bit position + bic r1, r13, r1 // zero icsr<fpe> + + or r1, r12, r1 // Or new FEN into ICSR + mfpr r12, pt_pcbb // Get PCBB - E1 + + mtpr r1, ev5__icsr // write new ICSR. 3 Bubble cycles to HW_REI + stlp r16, osfpcb_q_fen(r12) // Store FEN in PCB. + + mfpr r31, pt0 // Pad ICSR<FPE> write. + mfpr r31, pt0 + + mfpr r31, pt0 +// pvc_violate 225 // cuz PVC can't distinguish which bits changed + hw_rei + + + CALL_PAL_PRIV(0x002C) +CallPal_OpcDec2C: + br r31, osfpal_calpal_opcdec + +// .sbttl "wrvptpr - PALcode for wrvptpr instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// vptptr <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRVPTPTR_ENTRY) +Call_Pal_Wrvptptr: + mtpr r16, ev5__mvptbr // Load Mbox copy + mtpr r16, ev5__ivptbr // Load Ibox copy + nop // Pad IPR write + nop + hw_rei + + CALL_PAL_PRIV(0x002E) +CallPal_OpcDec2E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_PRIV(0x002F) +CallPal_OpcDec2F: + br r31, osfpal_calpal_opcdec + +// .sbttl "swpctx- PALcode for swpctx instruction" + +//+ +// +// Entry: +// hardware dispatch via callPal instruction +// R16 -> new pcb +// +// Function: +// dynamic state moved to old pcb +// new state loaded from new pcb +// pcbb pointer set +// old pcbb returned in R0 +// +// Note: need to add perf monitor stuff +//- + + CALL_PAL_PRIV(PAL_SWPCTX_ENTRY) +Call_Pal_Swpctx: + rpcc r13 // get cyccounter + mfpr r0, pt_pcbb // get pcbb + + ldqp r22, osfpcb_q_fen(r16) // get new fen/pme + ldqp r23, osfpcb_l_cc(r16) // get new asn + + srl r13, 32, r25 // move offset + mfpr r24, pt_usp // get usp + + stqp r30, osfpcb_q_ksp(r0) // store old ksp +// pvc_violate 379 // stqp can't trap except replay. only problem if mf same ipr in same shadow. + mtpr r16, pt_pcbb // set new pcbb + + stqp r24, osfpcb_q_usp(r0) // store usp + addl r13, r25, r25 // merge for new time + + stlp r25, osfpcb_l_cc(r0) // save time + ldah r24, (1<<(icsr_v_fpe-16))(r31) + + and r22, 1, r12 // isolate fen + mfpr r25, icsr // get current icsr + + ev5_pass2 lda r24, (1<<icsr_v_pmp)(r24) + br r31, swpctx_cont + +// .sbttl "wrval - PALcode for wrval instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// sysvalue <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRVAL_ENTRY) +Call_Pal_Wrval: + nop + mtpr r16, pt_sysval // Pad paltemp write + nop + nop + hw_rei + + +// .sbttl "rdval - PALcode for rdval instruction" + +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- sysvalue +//- + + CALL_PAL_PRIV(PAL_RDVAL_ENTRY) +Call_Pal_Rdval: + nop + mfpr r0, pt_sysval + nop + hw_rei + +// .sbttl "tbi - PALcode for tbi instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// TB invalidate +// r16/a0 = TBI type +// r17/a1 = Va for TBISx instructions +//- + + CALL_PAL_PRIV(PAL_TBI_ENTRY) +Call_Pal_Tbi: + addq r16, 2, r16 // change range to 0-2 + br r23, CALL_PAL_tbi_10_ // get our address + +CALL_PAL_tbi_10_: cmpult r16, 6, r22 // see if in range + lda r23, tbi_tbl-CALL_PAL_tbi_10_(r23) // set base to start of table + sll r16, 4, r16 // * 16 + blbc r22, CALL_PAL_tbi_30_ // go rei, if not + + addq r23, r16, r23 // addr of our code +//orig pvc_jsr tbi + jmp r31, (r23) // and go do it + +CALL_PAL_tbi_30_: + hw_rei + nop + +// .sbttl "wrent - PALcode for wrent instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Update ent* in paltemps +// r16/a0 = Address of entry routine +// r17/a1 = Entry Number 0..5 +// +// r22, r23 trashed +//- + + CALL_PAL_PRIV(PAL_WRENT_ENTRY) +Call_Pal_Wrent: + cmpult r17, 6, r22 // see if in range + br r23, CALL_PAL_wrent_10_ // get our address + +CALL_PAL_wrent_10_: bic r16, 3, r16 // clean pc + blbc r22, CALL_PAL_wrent_30_ // go rei, if not in range + + lda r23, wrent_tbl-CALL_PAL_wrent_10_(r23) // set base to start of table + sll r17, 4, r17 // *16 + + addq r17, r23, r23 // Get address in table +//orig pvc_jsr wrent + jmp r31, (r23) // and go do it + +CALL_PAL_wrent_30_: + hw_rei // out of range, just return + +// .sbttl "swpipl - PALcode for swpipl instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- PS<IPL> +// PS<IPL> <- a0<2:0> (r16) +// +// t8 (r22) is scratch +//- + + CALL_PAL_PRIV(PAL_SWPIPL_ENTRY) +Call_Pal_Swpipl: + and r16, osfps_m_ipl, r16 // clean New ipl + mfpr r22, pt_intmask // get int mask + + extbl r22, r16, r22 // get mask for this ipl + bis r11, r31, r0 // return old ipl + + bis r16, r31, r11 // set new ps + mtpr r22, ev5__ipl // set new mask + + mfpr r31, pt0 // pad ipl write + mfpr r31, pt0 // pad ipl write + + hw_rei // back + +// .sbttl "rdps - PALcode for rdps instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- ps +//- + + CALL_PAL_PRIV(PAL_RDPS_ENTRY) +Call_Pal_Rdps: + bis r11, r31, r0 // Fetch PALshadow PS + nop // Must be 2 cycles long + hw_rei + +// .sbttl "wrkgp - PALcode for wrkgp instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// kgp <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRKGP_ENTRY) +Call_Pal_Wrkgp: + nop + mtpr r16, pt_kgp + nop // Pad for pt write->read restriction + nop + hw_rei + +// .sbttl "wrusp - PALcode for wrusp instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// usp <- a0 (r16) +//- + + CALL_PAL_PRIV(PAL_WRUSP_ENTRY) +Call_Pal_Wrusp: + nop + mtpr r16, pt_usp + nop // Pad possible pt write->read restriction + nop + hw_rei + +// .sbttl "wrperfmon - PALcode for wrperfmon instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// +// Function: +// Various control functions for the onchip performance counters +// +// option selector in r16 +// option argument in r17 +// returned status in r0 +// +// +// r16 = 0 Disable performance monitoring for one or more cpu's +// r17 = 0 disable no counters +// r17 = bitmask disable counters specified in bit mask (1=disable) +// +// r16 = 1 Enable performance monitoring for one or more cpu's +// r17 = 0 enable no counters +// r17 = bitmask enable counters specified in bit mask (1=enable) +// +// r16 = 2 Mux select for one or more cpu's +// r17 = Mux selection (cpu specific) +// <24:19> bc_ctl<pm_mux_sel> field (see spec) +// <31>,<7:4>,<3:0> pmctr <sel0>,<sel1>,<sel2> fields (see spec) +// +// r16 = 3 Options +// r17 = (cpu specific) +// <0> = 0 log all processes +// <0> = 1 log only selected processes +// <30,9,8> mode select - ku,kp,kk +// +// r16 = 4 Interrupt frequency select +// r17 = (cpu specific) indicates interrupt frequencies desired for each +// counter, with "zero interrupts" being an option +// frequency info in r17 bits as defined by PMCTR_CTL<FRQx> below +// +// r16 = 5 Read Counters +// r17 = na +// r0 = value (same format as ev5 pmctr) +// <0> = 0 Read failed +// <0> = 1 Read succeeded +// +// r16 = 6 Write Counters +// r17 = value (same format as ev5 pmctr; all counters written simultaneously) +// +// r16 = 7 Enable performance monitoring for one or more cpu's and reset counter to 0 +// r17 = 0 enable no counters +// r17 = bitmask enable & clear counters specified in bit mask (1=enable & clear) +// +//============================================================================= +//Assumptions: +//PMCTR_CTL: +// +// <15:14> CTL0 -- encoded frequency select and enable - CTR0 +// <13:12> CTL1 -- " - CTR1 +// <11:10> CTL2 -- " - CTR2 +// +// <9:8> FRQ0 -- frequency select for CTR0 (no enable info) +// <7:6> FRQ1 -- frequency select for CTR1 +// <5:4> FRQ2 -- frequency select for CTR2 +// +// <0> all vs. select processes (0=all,1=select) +// +// where +// FRQx<1:0> +// 0 1 disable interrupt +// 1 0 frequency = 65536 (16384 for ctr2) +// 1 1 frequency = 256 +// note: FRQx<1:0> = 00 will keep counters from ever being enabled. +// +//============================================================================= +// + CALL_PAL_PRIV(0x0039) +// unsupported in Hudson code .. pboyle Nov/95 +CALL_PAL_Wrperfmon: +#if perfmon_debug == 0 + // "real" performance monitoring code + cmpeq r16, 1, r0 // check for enable + bne r0, perfmon_en // br if requested to enable + + cmpeq r16, 2, r0 // check for mux ctl + bne r0, perfmon_muxctl // br if request to set mux controls + + cmpeq r16, 3, r0 // check for options + bne r0, perfmon_ctl // br if request to set options + + cmpeq r16, 4, r0 // check for interrupt frequency select + bne r0, perfmon_freq // br if request to change frequency select + + cmpeq r16, 5, r0 // check for counter read request + bne r0, perfmon_rd // br if request to read counters + + cmpeq r16, 6, r0 // check for counter write request + bne r0, perfmon_wr // br if request to write counters + + cmpeq r16, 7, r0 // check for counter clear/enable request + bne r0, perfmon_enclr // br if request to clear/enable counters + + beq r16, perfmon_dis // br if requested to disable (r16=0) + br r31, perfmon_unknown // br if unknown request +#else + + br r31, pal_perfmon_debug +#endif + +// .sbttl "rdusp - PALcode for rdusp instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- usp +//- + + CALL_PAL_PRIV(PAL_RDUSP_ENTRY) +Call_Pal_Rdusp: + nop + mfpr r0, pt_usp + hw_rei + + + CALL_PAL_PRIV(0x003B) +CallPal_OpcDec3B: + br r31, osfpal_calpal_opcdec + +// .sbttl "whami - PALcode for whami instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- whami +//- + CALL_PAL_PRIV(PAL_WHAMI_ENTRY) +Call_Pal_Whami: + nop + mfpr r0, pt_whami // Get Whami + extbl r0, 1, r0 // Isolate just whami bits + hw_rei + +// .sbttl "retsys - PALcode for retsys instruction" +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// 00(sp) contains return pc +// 08(sp) contains r29 +// +// Function: +// Return from system call. +// mode switched from kern to user. +// stacks swapped, ugp, upc restored. +// r23, r25 junked +//- + + CALL_PAL_PRIV(PAL_RETSYS_ENTRY) +Call_Pal_Retsys: + lda r25, osfsf_c_size(sp) // pop stack + bis r25, r31, r14 // touch r25 & r14 to stall mf exc_addr + + mfpr r14, exc_addr // save exc_addr in case of fault + ldq r23, osfsf_pc(sp) // get pc + + ldq r29, osfsf_gp(sp) // get gp + stl_c r31, -4(sp) // clear lock_flag + + lda r11, 1<<osfps_v_mode(r31)// new PS:mode=user + mfpr r30, pt_usp // get users stack + + bic r23, 3, r23 // clean return pc + mtpr r31, ev5__ipl // zero ibox IPL - 2 bubbles to hw_rei + + mtpr r11, ev5__dtb_cm // set Mbox current mode - no virt ref for 2 cycles + mtpr r11, ev5__ps // set Ibox current mode - 2 bubble to hw_rei + + mtpr r23, exc_addr // set return address - 1 bubble to hw_rei + mtpr r25, pt_ksp // save kern stack + + rc r31 // clear inter_flag +// pvc_violate 248 // possible hidden mt->mf pt violation ok in callpal + hw_rei_spe // and back + + + CALL_PAL_PRIV(0x003E) +CallPal_OpcDec3E: + br r31, osfpal_calpal_opcdec + +// .sbttl "rti - PALcode for rti instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// 00(sp) -> ps +// 08(sp) -> pc +// 16(sp) -> r29 (gp) +// 24(sp) -> r16 (a0) +// 32(sp) -> r17 (a1) +// 40(sp) -> r18 (a3) +//- + + CALL_PAL_PRIV(PAL_RTI_ENTRY) +#ifdef SIMOS + /* called once by platform_tlaser */ + .globl Call_Pal_Rti +#endif +Call_Pal_Rti: + lda r25, osfsf_c_size(sp) // get updated sp + bis r25, r31, r14 // touch r14,r25 to stall mf exc_addr + + mfpr r14, exc_addr // save PC in case of fault + rc r31 // clear intr_flag + + ldq r12, -6*8(r25) // get ps + ldq r13, -5*8(r25) // pc + + ldq r18, -1*8(r25) // a2 + ldq r17, -2*8(r25) // a1 + + ldq r16, -3*8(r25) // a0 + ldq r29, -4*8(r25) // gp + + bic r13, 3, r13 // clean return pc + stl_c r31, -4(r25) // clear lock_flag + + and r12, osfps_m_mode, r11 // get mode + mtpr r13, exc_addr // set return address + + beq r11, rti_to_kern // br if rti to Kern + br r31, rti_to_user // out of call_pal space + + +// .sbttl "Start the Unprivileged CALL_PAL Entry Points" +// .sbttl "bpt- PALcode for bpt instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- +// + .text 1 +// . = 0x3000 + CALL_PAL_UNPRIV(PAL_BPT_ENTRY) +Call_Pal_Bpt: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, CALL_PAL_bpt_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +CALL_PAL_bpt_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_bpt, r16 // set a0 + + stq r17, osfsf_a1(sp) // a1 + br r31, bpt_bchk_common // out of call_pal space + + +// .sbttl "bugchk- PALcode for bugchk instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- +// + CALL_PAL_UNPRIV(PAL_BUGCHK_ENTRY) +Call_Pal_Bugchk: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, CALL_PAL_bugchk_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +CALL_PAL_bugchk_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_bugchk, r16 // set a0 + + stq r17, osfsf_a1(sp) // a1 + br r31, bpt_bchk_common // out of call_pal space + + + CALL_PAL_UNPRIV(0x0082) +CallPal_OpcDec82: + br r31, osfpal_calpal_opcdec + +// .sbttl "callsys - PALcode for callsys instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Switch mode to kernel and build a callsys stack frame. +// sp = ksp +// gp = kgp +// t8 - t10 (r22-r24) trashed +// +//- +// + CALL_PAL_UNPRIV(PAL_CALLSYS_ENTRY) +Call_Pal_Callsys: + + and r11, osfps_m_mode, r24 // get mode + mfpr r22, pt_ksp // get ksp + + beq r24, sys_from_kern // sysCall from kern is not allowed + mfpr r12, pt_entsys // get address of callSys routine + +//+ +// from here on we know we are in user going to Kern +//- + mtpr r31, ev5__dtb_cm // set Mbox current mode - no virt ref for 2 cycles + mtpr r31, ev5__ps // set Ibox current mode - 2 bubble to hw_rei + + bis r31, r31, r11 // PS=0 (mode=kern) + mfpr r23, exc_addr // get pc + + mtpr r30, pt_usp // save usp + lda sp, 0-osfsf_c_size(r22)// set new sp + + stq r29, osfsf_gp(sp) // save user gp/r29 + stq r24, osfsf_ps(sp) // save ps + + stq r23, osfsf_pc(sp) // save pc + mtpr r12, exc_addr // set address + // 1 cycle to hw_rei + + mfpr r29, pt_kgp // get the kern gp/r29 + + hw_rei_spe // and off we go! + + + CALL_PAL_UNPRIV(0x0084) +CallPal_OpcDec84: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0085) +CallPal_OpcDec85: + br r31, osfpal_calpal_opcdec + +// .sbttl "imb - PALcode for imb instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Flush the writebuffer and flush the Icache +// +//- +// + CALL_PAL_UNPRIV(PAL_IMB_ENTRY) +Call_Pal_Imb: + mb // Clear the writebuffer + mfpr r31, ev5__mcsr // Sync with clear + nop + nop + br r31, pal_ic_flush // Flush Icache + + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_UNPRIV(0x0087) +CallPal_OpcDec87: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0088) +CallPal_OpcDec88: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0089) +CallPal_OpcDec89: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008A) +CallPal_OpcDec8A: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008B) +CallPal_OpcDec8B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008C) +CallPal_OpcDec8C: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008D) +CallPal_OpcDec8D: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008E) +CallPal_OpcDec8E: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x008F) +CallPal_OpcDec8F: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0090) +CallPal_OpcDec90: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0091) +CallPal_OpcDec91: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0092) +CallPal_OpcDec92: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0093) +CallPal_OpcDec93: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0094) +CallPal_OpcDec94: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0095) +CallPal_OpcDec95: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0096) +CallPal_OpcDec96: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0097) +CallPal_OpcDec97: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0098) +CallPal_OpcDec98: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x0099) +CallPal_OpcDec99: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009A) +CallPal_OpcDec9A: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009B) +CallPal_OpcDec9B: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009C) +CallPal_OpcDec9C: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x009D) +CallPal_OpcDec9D: + br r31, osfpal_calpal_opcdec + +// .sbttl "rdunique - PALcode for rdunique instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// v0 (r0) <- unique +// +//- +// + CALL_PAL_UNPRIV(PAL_RDUNIQUE_ENTRY) +CALL_PALrdunique_: + mfpr r0, pt_pcbb // get pcb pointer + ldqp r0, osfpcb_q_unique(r0) // get new value + + hw_rei + +// .sbttl "wrunique - PALcode for wrunique instruction" +//+ +// +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// unique <- a0 (r16) +// +//- +// +CALL_PAL_UNPRIV(PAL_WRUNIQUE_ENTRY) +CALL_PAL_Wrunique: + nop + mfpr r12, pt_pcbb // get pcb pointer + stqp r16, osfpcb_q_unique(r12)// get new value + nop // Pad palshadow write + hw_rei // back + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_UNPRIV(0x00A0) +CallPal_OpcDecA0: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A1) +CallPal_OpcDecA1: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A2) +CallPal_OpcDecA2: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A3) +CallPal_OpcDecA3: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A4) +CallPal_OpcDecA4: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A5) +CallPal_OpcDecA5: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A6) +CallPal_OpcDecA6: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A7) +CallPal_OpcDecA7: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A8) +CallPal_OpcDecA8: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00A9) +CallPal_OpcDecA9: + br r31, osfpal_calpal_opcdec + + +// .sbttl "gentrap - PALcode for gentrap instruction" +//+ +// CALL_PAL_gentrap: +// Entry: +// Vectored into via hardware PALcode instruction dispatch. +// +// Function: +// Build stack frame +// a0 <- code +// a1 <- unpred +// a2 <- unpred +// vector via entIF +// +//- + + CALL_PAL_UNPRIV(0x00AA) +// unsupported in Hudson code .. pboyle Nov/95 +CALL_PAL_gentrap: + sll r11, 63-osfps_v_mode, r25 // Shift mode up to MS bit + mtpr r31, ev5__ps // Set Ibox current mode to kernel + + bis r11, r31, r12 // Save PS for stack write + bge r25, CALL_PAL_gentrap_10_ // no stack swap needed if cm=kern + + mtpr r31, ev5__dtb_cm // Set Mbox current mode to kernel - + // no virt ref for next 2 cycles + mtpr r30, pt_usp // save user stack + + bis r31, r31, r11 // Set new PS + mfpr r30, pt_ksp + +CALL_PAL_gentrap_10_: + lda sp, 0-osfsf_c_size(sp)// allocate stack space + mfpr r14, exc_addr // get pc + + stq r16, osfsf_a0(sp) // save regs + bis r31, osf_a0_gentrap, r16// set a0 + + stq r17, osfsf_a1(sp) // a1 + br r31, bpt_bchk_common // out of call_pal space + + +// .sbttl "CALL_PAL OPCDECs" + + CALL_PAL_UNPRIV(0x00AB) +CallPal_OpcDecAB: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AC) +CallPal_OpcDecAC: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AD) +CallPal_OpcDecAD: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AE) +CallPal_OpcDecAE: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00AF) +CallPal_OpcDecAF: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B0) +CallPal_OpcDecB0: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B1) +CallPal_OpcDecB1: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B2) +CallPal_OpcDecB2: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B3) +CallPal_OpcDecB3: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B4) +CallPal_OpcDecB4: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B5) +CallPal_OpcDecB5: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B6) +CallPal_OpcDecB6: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B7) +CallPal_OpcDecB7: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B8) +CallPal_OpcDecB8: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00B9) +CallPal_OpcDecB9: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BA) +CallPal_OpcDecBA: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BB) +CallPal_OpcDecBB: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BC) +CallPal_OpcDecBC: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BD) +CallPal_OpcDecBD: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BE) +CallPal_OpcDecBE: + br r31, osfpal_calpal_opcdec + + CALL_PAL_UNPRIV(0x00BF) +CallPal_OpcDecBF: + // MODIFIED BY EGH 2/25/04 + br r31, copypal_impl + + +/*======================================================================*/ +/* OSF/1 CALL_PAL CONTINUATION AREA */ +/*======================================================================*/ + + .text 2 + + . = 0x4000 + + +// .sbttl "Continuation of MTPR_PERFMON" + ALIGN_BLOCK +#if perfmon_debug == 0 + // "real" performance monitoring code +// mux ctl +perfmon_muxctl: + lda r8, 1(r31) // get a 1 + sll r8, pmctr_v_sel0, r8 // move to sel0 position + or r8, ((0xf<<pmctr_v_sel1) | (0xf<<pmctr_v_sel2)), r8 // build mux select mask + and r17, r8, r25 // isolate pmctr mux select bits + mfpr r0, ev5__pmctr + bic r0, r8, r0 // clear old mux select bits + or r0,r25, r25 // or in new mux select bits + mtpr r25, ev5__pmctr + + // ok, now tackle cbox mux selects + ldah r14, 0xfff0(r31) + zap r14, 0xE0, r14 // Get Cbox IPR base +//orig get_bc_ctl_shadow r16 // bc_ctl returned in lower longword +// adapted from ev5_pal_macros.mar + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + RESTORE_SHADOW(r16,CNS_Q_BC_CTL,r16); + + lda r8, 0x3F(r31) // build mux select mask + sll r8, bc_ctl_v_pm_mux_sel, r8 + + and r17, r8, r25 // isolate bc_ctl mux select bits + bic r16, r8, r16 // isolate old mux select bits + or r16, r25, r25 // create new bc_ctl + mb // clear out cbox for future ipr write + stqp r25, ev5__bc_ctl(r14) // store to cbox ipr + mb // clear out cbox for future ipr write + +//orig update_bc_ctl_shadow r25, r16 // r25=value, r16-overwritten with adjusted impure ptr +// adapted from ev5_pal_macros.mar + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + SAVE_SHADOW(r25,CNS_Q_BC_CTL,r16); + + br r31, perfmon_success + + +// requested to disable perf monitoring +perfmon_dis: + mfpr r14, ev5__pmctr // read ibox pmctr ipr +perfmon_dis_ctr0: // and begin with ctr0 + blbc r17, perfmon_dis_ctr1 // do not disable ctr0 + lda r8, 3(r31) + sll r8, pmctr_v_ctl0, r8 + bic r14, r8, r14 // disable ctr0 +perfmon_dis_ctr1: + srl r17, 1, r17 + blbc r17, perfmon_dis_ctr2 // do not disable ctr1 + lda r8, 3(r31) + sll r8, pmctr_v_ctl1, r8 + bic r14, r8, r14 // disable ctr1 +perfmon_dis_ctr2: + srl r17, 1, r17 + blbc r17, perfmon_dis_update // do not disable ctr2 + lda r8, 3(r31) + sll r8, pmctr_v_ctl2, r8 + bic r14, r8, r14 // disable ctr2 +perfmon_dis_update: + mtpr r14, ev5__pmctr // update pmctr ipr +//;the following code is not needed for ev5 pass2 and later, but doesn't hurt anything to leave in +// adapted from ev5_pal_macros.mar +//orig get_pmctr_ctl r8, r25 // pmctr_ctl bit in r8. adjusted impure pointer in r25 + mfpr r25, pt_impure + lda r25, CNS_Q_IPR(r25) + RESTORE_SHADOW(r8,CNS_Q_PM_CTL,r25); + + lda r17, 0x3F(r31) // build mask + sll r17, pmctr_v_ctl2, r17 // shift mask to correct position + and r14, r17, r14 // isolate ctl bits + bic r8, r17, r8 // clear out old ctl bits + or r14, r8, r14 // create shadow ctl bits +//orig store_reg1 pmctr_ctl, r14, r25, ipr=1 // update pmctr_ctl register +//adjusted impure pointer still in r25 + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r25); + + br r31, perfmon_success + + +// requested to enable perf monitoring +//;the following code can be greatly simplified for pass2, but should work fine as is. + + +perfmon_enclr: + lda r9, 1(r31) // set enclr flag + br perfmon_en_cont + +perfmon_en: + bis r31, r31, r9 // clear enclr flag + +perfmon_en_cont: + mfpr r8, pt_pcbb // get PCB base +//orig get_pmctr_ctl r25, r25 + mfpr r25, pt_impure + lda r25, CNS_Q_IPR(r25) + RESTORE_SHADOW(r25,CNS_Q_PM_CTL,r25); + + ldqp r16, osfpcb_q_fen(r8) // read DAT/PME/FEN quadword + mfpr r14, ev5__pmctr // read ibox pmctr ipr + srl r16, osfpcb_v_pme, r16 // get pme bit + mfpr r13, icsr + and r16, 1, r16 // isolate pme bit + + // this code only needed in pass2 and later +//orig sget_addr r12, 1<<icsr_v_pmp, r31 + lda r12, 1<<icsr_v_pmp(r31) // pb + bic r13, r12, r13 // clear pmp bit + sll r16, icsr_v_pmp, r12 // move pme bit to icsr<pmp> position + or r12, r13, r13 // new icsr with icsr<pmp> bit set/clear + ev5_pass2 mtpr r13, icsr // update icsr + +#if ev5_p1 != 0 + lda r12, 1(r31) + cmovlbc r25, r12, r16 // r16<0> set if either pme=1 or sprocess=0 (sprocess in bit 0 of r25) +#else + bis r31, 1, r16 // set r16<0> on pass2 to update pmctr always (icsr provides real enable) +#endif + + sll r25, 6, r25 // shift frequency bits into pmctr_v_ctl positions + bis r14, r31, r13 // copy pmctr + +perfmon_en_ctr0: // and begin with ctr0 + blbc r17, perfmon_en_ctr1 // do not enable ctr0 + + blbc r9, perfmon_en_noclr0 // enclr flag set, clear ctr0 field + lda r8, 0xffff(r31) + zapnot r8, 3, r8 // ctr0<15:0> mask + sll r8, pmctr_v_ctr0, r8 + bic r14, r8, r14 // clear ctr bits + bic r13, r8, r13 // clear ctr bits + +perfmon_en_noclr0: +//orig get_addr r8, 3<<pmctr_v_ctl0, r31 + LDLI(r8, (3<<pmctr_v_ctl0)) + and r25, r8, r12 //isolate frequency select bits for ctr0 + bic r14, r8, r14 // clear ctl0 bits in preparation for enabling + or r14,r12,r14 // or in new ctl0 bits + +perfmon_en_ctr1: // enable ctr1 + srl r17, 1, r17 // get ctr1 enable + blbc r17, perfmon_en_ctr2 // do not enable ctr1 + + blbc r9, perfmon_en_noclr1 // if enclr flag set, clear ctr1 field + lda r8, 0xffff(r31) + zapnot r8, 3, r8 // ctr1<15:0> mask + sll r8, pmctr_v_ctr1, r8 + bic r14, r8, r14 // clear ctr bits + bic r13, r8, r13 // clear ctr bits + +perfmon_en_noclr1: +//orig get_addr r8, 3<<pmctr_v_ctl1, r31 + LDLI(r8, (3<<pmctr_v_ctl1)) + and r25, r8, r12 //isolate frequency select bits for ctr1 + bic r14, r8, r14 // clear ctl1 bits in preparation for enabling + or r14,r12,r14 // or in new ctl1 bits + +perfmon_en_ctr2: // enable ctr2 + srl r17, 1, r17 // get ctr2 enable + blbc r17, perfmon_en_return // do not enable ctr2 - return + + blbc r9, perfmon_en_noclr2 // if enclr flag set, clear ctr2 field + lda r8, 0x3FFF(r31) // ctr2<13:0> mask + sll r8, pmctr_v_ctr2, r8 + bic r14, r8, r14 // clear ctr bits + bic r13, r8, r13 // clear ctr bits + +perfmon_en_noclr2: +//orig get_addr r8, 3<<pmctr_v_ctl2, r31 + LDLI(r8, (3<<pmctr_v_ctl2)) + and r25, r8, r12 //isolate frequency select bits for ctr2 + bic r14, r8, r14 // clear ctl2 bits in preparation for enabling + or r14,r12,r14 // or in new ctl2 bits + +perfmon_en_return: + cmovlbs r16, r14, r13 // if pme enabled, move enables into pmctr + // else only do the counter clears + mtpr r13, ev5__pmctr // update pmctr ipr + +//;this code not needed for pass2 and later, but does not hurt to leave it in + lda r8, 0x3F(r31) +//orig get_pmctr_ctl r25, r12 // read pmctr ctl; r12=adjusted impure pointer + mfpr r12, pt_impure + lda r12, CNS_Q_IPR(r12) + RESTORE_SHADOW(r25,CNS_Q_PM_CTL,r12); + + sll r8, pmctr_v_ctl2, r8 // build ctl mask + and r8, r14, r14 // isolate new ctl bits + bic r25, r8, r25 // clear out old ctl value + or r25, r14, r14 // create new pmctr_ctl +//orig store_reg1 pmctr_ctl, r14, r12, ipr=1 + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r12); // r12 still has the adjusted impure ptr + + br r31, perfmon_success + + +// options... +perfmon_ctl: + +// set mode +//orig get_pmctr_ctl r14, r12 // read shadow pmctr ctl; r12=adjusted impure pointer + mfpr r12, pt_impure + lda r12, CNS_Q_IPR(r12) + RESTORE_SHADOW(r14,CNS_Q_PM_CTL,r12); + +//orig get_addr r8, (1<<pmctr_v_killu) | (1<<pmctr_v_killp) | (1<<pmctr_v_killk), r31 // build mode mask for pmctr register + LDLI(r8, ((1<<pmctr_v_killu) | (1<<pmctr_v_killp) | (1<<pmctr_v_killk))) + mfpr r0, ev5__pmctr + and r17, r8, r25 // isolate pmctr mode bits + bic r0, r8, r0 // clear old mode bits + or r0, r25, r25 // or in new mode bits + mtpr r25, ev5__pmctr + +//;the following code will only be used in pass2, but should not hurt anything if run in pass1. + mfpr r8, icsr + lda r25, 1<<icsr_v_pma(r31) // set icsr<pma> if r17<0>=0 + bic r8, r25, r8 // clear old pma bit + cmovlbs r17, r31, r25 // and clear icsr<pma> if r17<0>=1 + or r8, r25, r8 + ev5_pass2 mtpr r8, icsr // 4 bubbles to hw_rei + mfpr r31, pt0 // pad icsr write + mfpr r31, pt0 // pad icsr write + +//;the following code not needed for pass2 and later, but should work anyway. + bis r14, 1, r14 // set for select processes + blbs r17, perfmon_sp // branch if select processes + bic r14, 1, r14 // all processes +perfmon_sp: +//orig store_reg1 pmctr_ctl, r14, r12, ipr=1 // update pmctr_ctl register + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r12); // r12 still has the adjusted impure ptr + br r31, perfmon_success + +// counter frequency select +perfmon_freq: +//orig get_pmctr_ctl r14, r12 // read shadow pmctr ctl; r12=adjusted impure pointer + mfpr r12, pt_impure + lda r12, CNS_Q_IPR(r12) + RESTORE_SHADOW(r14,CNS_Q_PM_CTL,r12); + + lda r8, 0x3F(r31) +//orig sll r8, pmctr_ctl_v_frq2, r8 // build mask for frequency select field +// I guess this should be a shift of 4 bits from the above control register structure .. pb +#define pmctr_ctl_v_frq2_SHIFT 4 + sll r8, pmctr_ctl_v_frq2_SHIFT, r8 // build mask for frequency select field + + and r8, r17, r17 + bic r14, r8, r14 // clear out old frequency select bits + + or r17, r14, r14 // or in new frequency select info +//orig store_reg1 pmctr_ctl, r14, r12, ipr=1 // update pmctr_ctl register + SAVE_SHADOW(r14,CNS_Q_PM_CTL,r12); // r12 still has the adjusted impure ptr + + br r31, perfmon_success + +// read counters +perfmon_rd: + mfpr r0, ev5__pmctr + or r0, 1, r0 // or in return status + hw_rei // back to user + +// write counters +perfmon_wr: + mfpr r14, ev5__pmctr + lda r8, 0x3FFF(r31) // ctr2<13:0> mask + sll r8, pmctr_v_ctr2, r8 + +//orig get_addr r9, 0xFFFFFFFF, r31, verify=0 // ctr2<15:0>,ctr1<15:0> mask + LDLI(r9, (0xFFFFFFFF)) + sll r9, pmctr_v_ctr1, r9 + or r8, r9, r8 // or ctr2, ctr1, ctr0 mask + bic r14, r8, r14 // clear ctr fields + and r17, r8, r25 // clear all but ctr fields + or r25, r14, r14 // write ctr fields + mtpr r14, ev5__pmctr // update pmctr ipr + + mfpr r31, pt0 // pad pmctr write (needed only to keep PVC happy) + +perfmon_success: + or r31, 1, r0 // set success + hw_rei // back to user + +perfmon_unknown: + or r31, r31, r0 // set fail + hw_rei // back to user + +#else + +// end of "real code", start of debug code + +//+ +// Debug environment: +// (in pass2, always set icsr<pma> to ensure master counter enable is on) +// R16 = 0 Write to on-chip performance monitor ipr +// r17 = on-chip ipr +// r0 = return value of read of on-chip performance monitor ipr +// R16 = 1 Setup Cbox mux selects +// r17 = Cbox mux selects in same position as in bc_ctl ipr. +// r0 = return value of read of on-chip performance monitor ipr +// +//- +pal_perfmon_debug: + mfpr r8, icsr + lda r9, 1<<icsr_v_pma(r31) + bis r8, r9, r8 + mtpr r8, icsr + + mfpr r0, ev5__pmctr // read old value + bne r16, cbox_mux_sel + + mtpr r17, ev5__pmctr // update pmctr ipr + br r31, end_pm + +cbox_mux_sel: + // ok, now tackle cbox mux selects + ldah r14, 0xfff0(r31) + zap r14, 0xE0, r14 // Get Cbox IPR base +//orig get_bc_ctl_shadow r16 // bc_ctl returned + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + RESTORE_SHADOW(r16,CNS_Q_BC_CTL,r16); + + lda r8, 0x3F(r31) // build mux select mask + sll r8, BC_CTL_V_PM_MUX_SEL, r8 + + and r17, r8, r25 // isolate bc_ctl mux select bits + bic r16, r8, r16 // isolate old mux select bits + or r16, r25, r25 // create new bc_ctl + mb // clear out cbox for future ipr write + stqp r25, ev5__bc_ctl(r14) // store to cbox ipr + mb // clear out cbox for future ipr write +//orig update_bc_ctl_shadow r25, r16 // r25=value, r16-overwritten with adjusted impure ptr + mfpr r16, pt_impure + lda r16, CNS_Q_IPR(r16) + SAVE_SHADOW(r25,CNS_Q_BC_CTL,r16); + +end_pm: hw_rei + +#endif + + +//;The following code is a workaround for a cpu bug where Istream prefetches to +//;super-page address space in user mode may escape off-chip. +#if spe_fix != 0 + + ALIGN_BLOCK +hw_rei_update_spe: + mfpr r12, pt_misc // get previous mode + srl r11, osfps_v_mode, r10 // isolate current mode bit + and r10, 1, r10 + extbl r12, 7, r8 // get previous mode field + and r8, 1, r8 // isolate previous mode bit + cmpeq r10, r8, r8 // compare previous and current modes + beq r8, hw_rei_update_spe_5_ + hw_rei // if same, just return + +hw_rei_update_spe_5_: + +#if fill_err_hack != 0 + + fill_error_hack +#endif + + mfpr r8, icsr // get current icsr value + ldah r9, (2<<(icsr_v_spe-16))(r31) // get spe bit mask + bic r8, r9, r8 // disable spe + xor r10, 1, r9 // flip mode for new spe bit + sll r9, icsr_v_spe+1, r9 // shift into position + bis r8, r9, r8 // enable/disable spe + lda r9, 1(r31) // now update our flag + sll r9, pt_misc_v_cm, r9 // previous mode saved bit mask + bic r12, r9, r12 // clear saved previous mode + sll r10, pt_misc_v_cm, r9 // current mode saved bit mask + bis r12, r9, r12 // set saved current mode + mtpr r12, pt_misc // update pt_misc + mtpr r8, icsr // update icsr + +#if osf_chm_fix != 0 + + + blbc r10, hw_rei_update_spe_10_ // branch if not user mode + + mb // ensure no outstanding fills + lda r12, 1<<dc_mode_v_dc_ena(r31) // User mode + mtpr r12, dc_mode // Turn on dcache + mtpr r31, dc_flush // and flush it + br r31, pal_ic_flush + +hw_rei_update_spe_10_: mfpr r9, pt_pcbb // Kernel mode + ldqp r9, osfpcb_q_Fen(r9) // get FEN + blbc r9, pal_ic_flush // return if FP disabled + mb // ensure no outstanding fills + mtpr r31, dc_mode // turn off dcache +#endif + + + br r31, pal_ic_flush // Pal restriction - must flush Icache if changing ICSR<SPE> +#endif + + +copypal_impl: + mov r16, r0 + bic r18, 63, r8 + and r18, 63, r18 + beq r8, cache_copy_done +cache_loop: + ldf f17, 0(r16) + stf f17, 0(r16) + addq r17, 64, r17 + addq r16, 64, r16 + subq r8, 64, r8 + bne r8, cache_loop +cache_copy_done: + ble r18, finished #if len <=0 we are finished + ldq_u r8, 0(r17) + xor r17, r16, r9 + and r9, 7, r9 + and r16, 7, r10 + bne r9, unaligned + beq r10, aligned + ldq_u r9, 0(r16) + addq r18, r10, r18 + mskqh r8, r17, r8 + mskql r9, r17, r9 + bis r8, r9, r8 +aligned: + subq r18, 1, r10 + bic r10, 7, r10 + and r18, 7, r18 + beq r10, aligned_done +loop: + stq_u r8, 0(r16) + ldq_u r8, 8(r17) + subq r10, 8, r10 + lda r16,8(r16) + lda r17,8(r17) + bne r10, loop +aligned_done: + bne r18, few_left + stq_u r8, 0(r16) + br r31, finished + few_left: + mskql r8, r18, r10 + ldq_u r9, 0(r16) + mskqh r9, r18, r9 + bis r10, r9, r10 + stq_u r10, 0(r16) + br r31, finished +unaligned: + addq r17, r18, r25 + cmpule r18, 8, r9 + bne r9, unaligned_few_left + beq r10, unaligned_dest_aligned + and r16, 7, r10 + subq r31, r10, r10 + addq r10, 8, r10 + ldq_u r9, 7(r17) + extql r8, r17, r8 + extqh r9, r17, r9 + bis r8, r9, r12 + insql r12, r16, r12 + ldq_u r13, 0(r16) + mskql r13, r16, r13 + bis r12, r13, r12 + stq_u r12, 0(r16) + addq r16, r10, r16 + addq r17, r10, r17 + subq r18, r10, r18 + ldq_u r8, 0(r17) +unaligned_dest_aligned: + subq r18, 1, r10 + bic r10, 7, r10 + and r18, 7, r18 + beq r10, unaligned_partial_left +unaligned_loop: + ldq_u r9, 7(r17) + lda r17, 8(r17) + extql r8, r17, r12 + extqh r9, r17, r13 + subq r10, 8, r10 + bis r12, r13, r13 + stq r13, 0(r16) + lda r16, 8(r16) + beq r10, unaligned_second_partial_left + ldq_u r8, 7(r17) + lda r17, 8(r17) + extql r9, r17, r12 + extqh r8, r17, r13 + bis r12, r13, r13 + subq r10, 8, r10 + stq r13, 0(r16) + lda r16, 8(r16) + bne r10, unaligned_loop +unaligned_partial_left: + mov r8, r9 +unaligned_second_partial_left: + ldq_u r8, -1(r25) + extql r9, r17, r9 + extqh r8, r17, r8 + bis r8, r9, r8 + bne r18, few_left + stq_u r8, 0(r16) + br r31, finished +unaligned_few_left: + ldq_u r9, -1(r25) + extql r8, r17, r8 + extqh r9, r17, r9 + bis r8, r9, r8 + insqh r8, r16, r9 + insql r8, r16, r8 + lda r12, -1(r31) + mskql r12, r18, r13 + cmovne r13, r13, r12 + insqh r12, r16, r13 + insql r12, r16, r12 + addq r16, r18, r10 + ldq_u r14, 0(r16) + ldq_u r25, -1(r10) + bic r14, r12, r14 + bic r25, r13, r25 + and r8, r12, r8 + and r9, r13, r9 + bis r8, r14, r8 + bis r9, r25, r9 + stq_u r9, -1(r10) + stq_u r8, 0(r16) +finished: + hw_rei |