diff options
Diffstat (limited to 'arch/alpha')
-rw-r--r-- | arch/alpha/SConscript | 487 | ||||
-rw-r--r-- | arch/alpha/alpha_linux_process.cc | 151 | ||||
-rw-r--r-- | arch/alpha/alpha_memory.cc | 38 | ||||
-rw-r--r-- | arch/alpha/alpha_memory.hh | 5 | ||||
-rw-r--r-- | arch/alpha/alpha_tru64_process.cc | 236 | ||||
-rw-r--r-- | arch/alpha/ev5.cc | 80 | ||||
-rw-r--r-- | arch/alpha/faults.cc | 77 | ||||
-rw-r--r-- | arch/alpha/faults.hh | 153 | ||||
-rw-r--r-- | arch/alpha/isa/branch.isa | 259 | ||||
-rw-r--r-- | arch/alpha/isa/decoder.isa | 802 | ||||
-rw-r--r-- | arch/alpha/isa/fp.isa | 300 | ||||
-rw-r--r-- | arch/alpha/isa/int.isa | 128 | ||||
-rw-r--r-- | arch/alpha/isa/main.isa | 435 | ||||
-rw-r--r-- | arch/alpha/isa/mem.isa | 702 | ||||
-rw-r--r-- | arch/alpha/isa/pal.isa | 271 | ||||
-rw-r--r-- | arch/alpha/isa/unimp.isa | 165 | ||||
-rw-r--r-- | arch/alpha/isa/unknown.isa | 52 | ||||
-rw-r--r-- | arch/alpha/isa/util.isa | 112 | ||||
-rw-r--r-- | arch/alpha/isa_desc | 2737 | ||||
-rw-r--r-- | arch/alpha/isa_traits.hh | 6 | ||||
-rw-r--r-- | arch/alpha/pseudo_inst.cc | 256 | ||||
-rw-r--r-- | arch/alpha/pseudo_inst.hh | 54 | ||||
-rw-r--r-- | arch/alpha/vptr.hh | 114 |
23 files changed, 4173 insertions, 3447 deletions
diff --git a/arch/alpha/SConscript b/arch/alpha/SConscript new file mode 100644 index 000000000..8bf408c06 --- /dev/null +++ b/arch/alpha/SConscript @@ -0,0 +1,487 @@ +# -*- mode:python -*- + +# Copyright (c) 2004-2005 The Regents of The University of Michigan +# All rights reserved. +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions are +# met: redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer; +# redistributions in binary form must reproduce the above copyright +# notice, this list of conditions and the following disclaimer in the +# documentation and/or other materials provided with the distribution; +# neither the name of the copyright holders nor the names of its +# contributors may be used to endorse or promote products derived from +# this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +import os +import sys +from os.path import isdir + +# This file defines how to build a particular configuration of M5 +# based on variable settings in the 'env' build environment. + +# Import build environment variable from SConstruct. +Import('env') + +################################################### +# +# Define needed sources. +# +################################################### + +# Base sources used by all configurations. +arch_base_sources = Split(''' + arch/alpha/decoder.cc + arch/alpha/alpha_o3_exec.cc + arch/alpha/fast_cpu_exec.cc + arch/alpha/simple_cpu_exec.cc + arch/alpha/full_cpu_exec.cc + arch/alpha/faults.cc + arch/alpha/isa_traits.cc + ''') + +# base/circlebuf.cc +# base/copyright.cc +# base/cprintf.cc +# base/embedfile.cc +# base/fast_alloc.cc +# base/fifo_buffer.cc +# base/hostinfo.cc +# base/hybrid_pred.cc +# base/inifile.cc +# base/intmath.cc +# base/match.cc +# base/misc.cc +# base/output.cc +# base/pollevent.cc +# base/range.cc +# base/random.cc +# base/sat_counter.cc +# base/socket.cc +# base/statistics.cc +# base/str.cc +# base/time.cc +# base/trace.cc +# base/traceflags.cc +# base/userinfo.cc +# base/compression/lzss_compression.cc +# base/loader/aout_object.cc +# base/loader/ecoff_object.cc +# base/loader/elf_object.cc +# base/loader/object_file.cc +# base/loader/symtab.cc +# base/stats/events.cc +# base/stats/statdb.cc +# base/stats/visit.cc +# base/stats/text.cc +# +# cpu/base.cc +# cpu/base_dyn_inst.cc +# cpu/exec_context.cc +# cpu/exetrace.cc +# cpu/pc_event.cc +# cpu/static_inst.cc +# cpu/o3/2bit_local_pred.cc +# cpu/o3/alpha_dyn_inst.cc +# cpu/o3/alpha_cpu.cc +# cpu/o3/alpha_cpu_builder.cc +# cpu/o3/bpred_unit.cc +# cpu/o3/btb.cc +# cpu/o3/commit.cc +# cpu/o3/decode.cc +# cpu/o3/fetch.cc +# cpu/o3/free_list.cc +# cpu/o3/cpu.cc +# cpu/o3/iew.cc +# cpu/o3/inst_queue.cc +# cpu/o3/ldstq.cc +# cpu/o3/mem_dep_unit.cc +# cpu/o3/ras.cc +# cpu/o3/rename.cc +# cpu/o3/rename_map.cc +# cpu/o3/rob.cc +# cpu/o3/sat_counter.cc +# cpu/o3/store_set.cc +# cpu/o3/tournament_pred.cc +# cpu/fast/cpu.cc +# cpu/sampler/sampler.cc +# cpu/simple/cpu.cc +# cpu/trace/reader/mem_trace_reader.cc +# cpu/trace/reader/ibm_reader.cc +# cpu/trace/reader/itx_reader.cc +# cpu/trace/reader/m5_reader.cc +# cpu/trace/opt_cpu.cc +# cpu/trace/trace_cpu.cc +# +# encumbered/cpu/full/bpred.cc +# encumbered/cpu/full/commit.cc +# encumbered/cpu/full/cpu.cc +# encumbered/cpu/full/create_vector.cc +# encumbered/cpu/full/cv_spec_state.cc +# encumbered/cpu/full/dd_queue.cc +# encumbered/cpu/full/dep_link.cc +# encumbered/cpu/full/dispatch.cc +# encumbered/cpu/full/dyn_inst.cc +# encumbered/cpu/full/execute.cc +# encumbered/cpu/full/fetch.cc +# encumbered/cpu/full/floss_reasons.cc +# encumbered/cpu/full/fu_pool.cc +# encumbered/cpu/full/inst_fifo.cc +# encumbered/cpu/full/instpipe.cc +# encumbered/cpu/full/issue.cc +# encumbered/cpu/full/ls_queue.cc +# encumbered/cpu/full/machine_queue.cc +# encumbered/cpu/full/pipetrace.cc +# encumbered/cpu/full/readyq.cc +# encumbered/cpu/full/reg_info.cc +# encumbered/cpu/full/rob_station.cc +# encumbered/cpu/full/spec_memory.cc +# encumbered/cpu/full/spec_state.cc +# encumbered/cpu/full/storebuffer.cc +# encumbered/cpu/full/writeback.cc +# encumbered/cpu/full/iq/iq_station.cc +# encumbered/cpu/full/iq/iqueue.cc +# encumbered/cpu/full/iq/segmented/chain_info.cc +# encumbered/cpu/full/iq/segmented/chain_wire.cc +# encumbered/cpu/full/iq/segmented/iq_seg.cc +# encumbered/cpu/full/iq/segmented/iq_segmented.cc +# encumbered/cpu/full/iq/segmented/seg_chain.cc +# encumbered/cpu/full/iq/seznec/iq_seznec.cc +# encumbered/cpu/full/iq/standard/iq_standard.cc +# encumbered/mem/functional/main.cc +# +# mem/base_hier.cc +# mem/base_mem.cc +# mem/hier_params.cc +# mem/mem_cmd.cc +# mem/mem_debug.cc +# mem/mem_req.cc +# mem/memory_interface.cc +# mem/bus/base_interface.cc +# mem/bus/bus.cc +# mem/bus/bus_bridge.cc +# mem/bus/bus_bridge_master.cc +# mem/bus/bus_bridge_slave.cc +# mem/bus/bus_interface.cc +# mem/bus/dma_bus_interface.cc +# mem/bus/dma_interface.cc +# mem/bus/master_interface.cc +# mem/bus/slave_interface.cc +# mem/cache/base_cache.cc +# mem/cache/cache.cc +# mem/cache/cache_builder.cc +# mem/cache/coherence/coherence_protocol.cc +# mem/cache/coherence/uni_coherence.cc +# mem/cache/miss/blocking_buffer.cc +# mem/cache/miss/miss_queue.cc +# mem/cache/miss/mshr.cc +# mem/cache/miss/mshr_queue.cc +# mem/cache/prefetch/base_prefetcher.cc +# mem/cache/prefetch/prefetcher.cc +# mem/cache/prefetch/tagged_prefetcher.cc +# mem/cache/tags/base_tags.cc +# mem/cache/tags/cache_tags.cc +# mem/cache/tags/fa_lru.cc +# mem/cache/tags/iic.cc +# mem/cache/tags/lru.cc +# mem/cache/tags/repl/gen.cc +# mem/cache/tags/repl/repl.cc +# mem/cache/tags/split.cc +# mem/cache/tags/split_lru.cc +# mem/cache/tags/split_lifo.cc +# mem/functional/functional.cc +# mem/timing/base_memory.cc +# mem/timing/memory_builder.cc +# mem/timing/simple_mem_bank.cc +# mem/trace/itx_writer.cc +# mem/trace/mem_trace_writer.cc +# mem/trace/m5_writer.cc +# +# python/pyconfig.cc +# python/embedded_py.cc +# +# sim/builder.cc +# sim/configfile.cc +# sim/debug.cc +# sim/eventq.cc +# sim/main.cc +# sim/param.cc +# sim/profile.cc +# sim/root.cc +# sim/serialize.cc +# sim/sim_events.cc +# sim/sim_exit.cc +# sim/sim_object.cc +# sim/startup.cc +# sim/stat_context.cc +# sim/stat_control.cc +# sim/trace_context.cc +# ''') + +# MySql sources +arch_mysql_sources = Split(''' + ''') +# base/mysql.cc +# base/stats/mysql.cc +# ''') + +# Full-system sources +arch_full_system_sources = Split(''' + arch/alpha/alpha_memory.cc + arch/alpha/arguments.cc + arch/alpha/ev5.cc + arch/alpha/osfpal.cc + arch/alpha/stacktrace.cc + arch/alpha/vtophys.cc + ''') + +# base/crc.cc +# base/inet.cc +# base/remote_gdb.cc +# +# cpu/intr_control.cc +# cpu/profile.cc +# +# dev/alpha_console.cc +# dev/baddev.cc +# dev/simconsole.cc +# dev/disk_image.cc +# dev/etherbus.cc +# dev/etherdump.cc +# dev/etherint.cc +# dev/etherlink.cc +# dev/etherpkt.cc +# dev/ethertap.cc +# dev/ide_ctrl.cc +# dev/ide_disk.cc +# dev/io_device.cc +# dev/ns_gige.cc +# dev/pciconfigall.cc +# dev/pcidev.cc +# dev/pcifake.cc +# dev/pktfifo.cc +# dev/platform.cc +# dev/sinic.cc +# dev/simple_disk.cc +# dev/tsunami.cc +# dev/tsunami_cchip.cc +# dev/isa_fake.cc +# dev/tsunami_io.cc +# dev/tsunami_pchip.cc +# dev/uart.cc +# dev/uart8250.cc +# +# kern/kernel_binning.cc +# kern/kernel_stats.cc +# kern/system_events.cc +# kern/freebsd/freebsd_system.cc +# kern/linux/linux_syscalls.cc +# kern/linux/linux_system.cc +# kern/linux/printk.cc +# kern/tru64/dump_mbuf.cc +# kern/tru64/printf.cc +# kern/tru64/tru64_events.cc +# kern/tru64/tru64_syscalls.cc +# kern/tru64/tru64_system.cc +# +# mem/functional/memory_control.cc +# mem/functional/physical.cc +# +# sim/system.cc +# ''') + +# turbolaser encumbered sources +arch_turbolaser_sources = Split(''' + ''') +# encumbered/dev/dma.cc +# encumbered/dev/etherdev.cc +# encumbered/dev/scsi.cc +# encumbered/dev/scsi_ctrl.cc +# encumbered/dev/scsi_disk.cc +# encumbered/dev/scsi_none.cc +# encumbered/dev/tlaser_clock.cc +# encumbered/dev/tlaser_ipi.cc +# encumbered/dev/tlaser_mbox.cc +# encumbered/dev/tlaser_mc146818.cc +# encumbered/dev/tlaser_node.cc +# encumbered/dev/tlaser_pcia.cc +# encumbered/dev/tlaser_pcidev.cc +# encumbered/dev/tlaser_serial.cc +# encumbered/dev/turbolaser.cc +# encumbered/dev/uart8530.cc +# ''') + +# Syscall emulation (non-full-system) sources +arch_syscall_emulation_sources = Split(''' + arch/alpha/alpha_common_syscall_emul.cc + arch/alpha/alpha_linux_process.cc + arch/alpha/alpha_tru64_process.cc + ''') +# cpu/memtest/memtest.cc +# encumbered/eio/eio.cc +# encumbered/eio/exolex.cc +# encumbered/eio/libexo.cc +# sim/process.cc +# sim/syscall_emul.cc +# ''') + +#targetarch_files = Split(''' +# alpha_common_syscall_emul.hh +# alpha_linux_process.hh +# alpha_memory.hh +# alpha_tru64_process.hh +# aout_machdep.h +# arguments.hh +# byte_swap.hh +# ecoff_machdep.h +# ev5.hh +# faults.hh +# isa_fullsys_traits.hh +# isa_traits.hh +# osfpal.hh +# pseudo_inst.hh +# stacktrace.hh +# vptr.hh +# vtophys.hh +# ''') + +#for f in targetarch_files: +# env.Command('targetarch/' + f, 'arch/alpha/' + f, +# '''echo '#include "arch/alpha/%s"' > $TARGET''' % f) + + +# Set up complete list of sources based on configuration. +sources = arch_base_sources + +if env['FULL_SYSTEM']: + sources += arch_full_system_sources + if env['ALPHA_TLASER']: + sources += arch_turbolaser_sources +else: + sources += arch_syscall_emulation_sources + +if env['USE_MYSQL']: + sources += arch_mysql_sources + +for opt in env.ExportOptions: + env.ConfigFile(opt) + +################################################### +# +# Special build rules. +# +################################################### + +# base/traceflags.{cc,hh} are generated from base/traceflags.py. +# $TARGET.base will expand to "<build-dir>/base/traceflags". +# env.Command(Split('base/traceflags.hh base/traceflags.cc'), +# 'base/traceflags.py', +# 'python $SOURCE $TARGET.base') + +# several files are generated from arch/$TARGET_ISA/isa_desc. +#env.Command(Split('''decoder.cc +# decoder.hh +# alpha_o3_exec.cc +# fast_cpu_exec.cc +# simple_cpu_exec.cc +# full_cpu_exec.cc'''), +# Split('''isa_desc +# ../isa_parser.py'''), +# '$SRCDIR/arch/isa_parser.py $SOURCE $TARGET.dir arch/alpha') + + +# libelf build is described in its own SConscript file. +# SConscript-local is the per-config build, which just copies some +# header files into a place where they can be found. +# SConscript('libelf/SConscript-local', exports = 'env', duplicate=0) +# SConscript('python/SConscript', exports = ['env'], duplicate=0) + +# This function adds the specified sources to the given build +# environment, and returns a list of all the corresponding SCons +# Object nodes (including an extra one for date.cc). We explicitly +# add the Object nodes so we can set up special dependencies for +# date.cc. +# def make_objs(sources, env): +# objs = [env.Object(s) for s in sources] +# # make date.cc depend on all other objects so it always gets +# # recompiled whenever anything else does +# date_obj = env.Object('base/date.cc') +# base/traceflags.{cc,hh} are generated from base/traceflags.py. +# $TARGET.base will expand to "<build-dir>/base/traceflags". +# env.Command(Split('base/traceflags.hh base/traceflags.cc'), +# 'base/traceflags.py', +# 'python $SOURCE $TARGET.base') +# +# Split('''arch/alpha/isa_desc +# arch/isa_parser.py'''), +# env.Depends(date_obj, objs) +# objs.append(date_obj) +# return objs + +################################################### +# +# Define binaries. Each different build type (debug, opt, etc.) gets +# a slightly different build environment. +# +################################################### + +# Include file paths are rooted in this directory. SCons will +# automatically expand '.' to refer to both the source directory and +# the corresponding build directory to pick up generated include +# files. +# env.Append(CPPPATH='.') + +# Debug binary +# debugEnv = env.Copy(OBJSUFFIX='.do') +# debugEnv.Label = 'debug' +# debugEnv.Append(CCFLAGS=Split('-g -gstabs+ -O0')) +# debugEnv.Append(CPPDEFINES='DEBUG') +# tlist = debugEnv.Program(target = 'm5.debug', +# source = make_objs(sources, debugEnv)) +# debugEnv.M5Binary = tlist[0] + +# Optimized binary +# optEnv = env.Copy() +# optEnv.Label = 'opt' +# optEnv.Append(CCFLAGS=Split('-g -O5')) +# tlist = optEnv.Program(target = 'm5.opt', +# source = make_objs(sources, optEnv)) +# optEnv.M5Binary = tlist[0] + +# "Fast" binary +# fastEnv = env.Copy(OBJSUFFIX='.fo') +# fastEnv.Label = 'fast' +# fastEnv.Append(CCFLAGS=Split('-O5')) +# fastEnv.Append(CPPDEFINES='NDEBUG') +# fastEnv.Program(target = 'm5.fast.unstripped', +# source = make_objs(sources, fastEnv)) +# tlist = fastEnv.Command(target = 'm5.fast', +# source = 'm5.fast.unstripped', +# action = 'strip $SOURCE -o $TARGET') +# fastEnv.M5Binary = tlist[0] + +# Profiled binary +# profEnv = env.Copy(OBJSUFFIX='.po') +# profEnv.Label = 'prof' +# profEnv.Append(CCFLAGS=Split('-O5 -g -pg'), LINKFLAGS='-pg') +# tlist = profEnv.Program(target = 'm5.prof', +# source = make_objs(sources, profEnv)) +# profEnv.M5Binary = tlist[0] +# +# envList = [debugEnv, optEnv, fastEnv, profEnv] +# +# Return('envList') +Return('sources') diff --git a/arch/alpha/alpha_linux_process.cc b/arch/alpha/alpha_linux_process.cc index 83b0b5e5a..fb5e32e63 100644 --- a/arch/alpha/alpha_linux_process.cc +++ b/arch/alpha/alpha_linux_process.cc @@ -42,6 +42,7 @@ #include "sim/process.hh" #include "sim/sim_events.hh" +#include "arch/alpha/isa_traits.hh" #include "arch/alpha/alpha_common_syscall_emul.hh" #include "sim/syscall_emul.hh" #include "sim/root.hh" // for curTick & ticksPerSecond @@ -70,6 +71,15 @@ class Linux { typedef uint32_t gid_t; //@} +#if BSD_HOST + typedef struct stat hst_stat; + typedef struct stat hst_stat64; +#else + typedef struct stat hst_stat ; + typedef struct stat64 hst_stat64; +#endif + + //@{ /// open(2) flag values. static const int TGT_O_RDONLY = 00000000; //!< O_RDONLY @@ -138,7 +148,7 @@ class Linux { uint64_t st_mtime_nsec; uint64_t tgt_st_ctime; uint64_t st_ctime_nsec; - int64_t __unused[3]; + int64_t ___unused[3]; }; /// Length of strings in struct utsname (plus 1 for null char). @@ -169,18 +179,18 @@ class Linux { /// Resource enumeration for getrlimit(). enum rlimit_resources { - RLIMIT_CPU = 0, - RLIMIT_FSIZE = 1, - RLIMIT_DATA = 2, - RLIMIT_STACK = 3, - RLIMIT_CORE = 4, - RLIMIT_RSS = 5, - RLIMIT_NOFILE = 6, - RLIMIT_AS = 7, - RLIMIT_VMEM = 7, - RLIMIT_NPROC = 8, - RLIMIT_MEMLOCK = 9, - RLIMIT_LOCKS = 10 + TGT_RLIMIT_CPU = 0, + TGT_RLIMIT_FSIZE = 1, + TGT_RLIMIT_DATA = 2, + TGT_RLIMIT_STACK = 3, + TGT_RLIMIT_CORE = 4, + TGT_RLIMIT_RSS = 5, + TGT_RLIMIT_NOFILE = 6, + TGT_RLIMIT_AS = 7, + TGT_RLIMIT_VMEM = 7, + TGT_RLIMIT_NPROC = 8, + TGT_RLIMIT_MEMLOCK = 9, + TGT_RLIMIT_LOCKS = 10 }; /// Limit struct for getrlimit/setrlimit. @@ -207,9 +217,9 @@ class Linux { //@{ /// For getrusage(). - static const int RUSAGE_SELF = 0; - static const int RUSAGE_CHILDREN = -1; - static const int RUSAGE_BOTH = -2; + static const int TGT_RUSAGE_SELF = 0; + static const int TGT_RUSAGE_CHILDREN = -1; + static const int TGT_RUSAGE_BOTH = -2; //@} /// For getrusage(). @@ -235,59 +245,98 @@ class Linux { /// Helper function to convert a host stat buffer to a target stat /// buffer. Also copies the target buffer out to the simulated /// memory space. Used by stat(), fstat(), and lstat(). +#if !BSD_HOST static void - copyOutStatBuf(FunctionalMemory *mem, Addr addr, struct stat *host) + copyOutStatBuf(FunctionalMemory *mem, Addr addr, hst_stat *host) { TypedBufferArg<Linux::tgt_stat> tgt(addr); - tgt->st_dev = host->st_dev; - tgt->st_ino = host->st_ino; - tgt->st_mode = host->st_mode; - tgt->st_nlink = host->st_nlink; - tgt->st_uid = host->st_uid; - tgt->st_gid = host->st_gid; - tgt->st_rdev = host->st_rdev; - tgt->st_size = host->st_size; - tgt->st_atimeX = host->st_atime; - tgt->st_mtimeX = host->st_mtime; - tgt->st_ctimeX = host->st_ctime; - tgt->st_blksize = host->st_blksize; - tgt->st_blocks = host->st_blocks; + tgt->st_dev = htog(host->st_dev); + tgt->st_ino = htog(host->st_ino); + tgt->st_mode = htog(host->st_mode); + tgt->st_nlink = htog(host->st_nlink); + tgt->st_uid = htog(host->st_uid); + tgt->st_gid = htog(host->st_gid); + tgt->st_rdev = htog(host->st_rdev); + tgt->st_size = htog(host->st_size); + tgt->st_atimeX = htog(host->st_atime); + tgt->st_mtimeX = htog(host->st_mtime); + tgt->st_ctimeX = htog(host->st_ctime); + tgt->st_blksize = htog(host->st_blksize); + tgt->st_blocks = htog(host->st_blocks); tgt.copyOut(mem); } +#else + // Third version for bsd systems which no longer have any support for + // the old stat() call and stat() is actually a stat64() + static void + copyOutStatBuf(FunctionalMemory *mem, Addr addr, hst_stat64 *host) + { + TypedBufferArg<Linux::tgt_stat> tgt(addr); + + tgt->st_dev = htog(host->st_dev); + tgt->st_ino = htog(host->st_ino); + tgt->st_mode = htog(host->st_mode); + tgt->st_nlink = htog(host->st_nlink); + tgt->st_uid = htog(host->st_uid); + tgt->st_gid = htog(host->st_gid); + tgt->st_rdev = htog(host->st_rdev); + tgt->st_size = htog(host->st_size); + tgt->st_atimeX = htog(host->st_atime); + tgt->st_mtimeX = htog(host->st_mtime); + tgt->st_ctimeX = htog(host->st_ctime); + tgt->st_blksize = htog(host->st_blksize); + tgt->st_blocks = htog(host->st_blocks); + + tgt.copyOut(mem); + } +#endif + // Same for stat64 static void - copyOutStat64Buf(FunctionalMemory *mem, Addr addr, struct stat64 *host) + copyOutStat64Buf(FunctionalMemory *mem, int fd, Addr addr, hst_stat64 *host) { TypedBufferArg<Linux::tgt_stat64> tgt(addr); - // XXX byteswaps - tgt->st_dev = host->st_dev; + // fd == 1 checks are because libc does some checks + // that the stdout is interactive vs. a file + // this makes it work on non-linux systems + if (fd == 1) + tgt->st_dev = htog((uint64_t)0xA); + else + tgt->st_dev = htog((uint64_t)host->st_dev); // XXX What about STAT64_HAS_BROKEN_ST_INO ??? - tgt->st_ino = host->st_ino; - tgt->st_rdev = host->st_rdev; - tgt->st_size = host->st_size; - tgt->st_blocks = host->st_blocks; - - tgt->st_mode = host->st_mode; - tgt->st_uid = host->st_uid; - tgt->st_gid = host->st_gid; - tgt->st_blksize = host->st_blksize; - tgt->st_nlink = host->st_nlink; - tgt->tgt_st_atime = host->st_atime; - tgt->tgt_st_mtime = host->st_mtime; - tgt->tgt_st_ctime = host->st_ctime; -#ifdef STAT_HAVE_NSEC - tgt->st_atime_nsec = host->st_atime_nsec; - tgt->st_mtime_nsec = host->st_mtime_nsec; - tgt->st_ctime_nsec = host->st_ctime_nsec; + tgt->st_ino = htog((uint64_t)host->st_ino); + if (fd == 1) + tgt->st_rdev = htog((uint64_t)0x880d); + else + tgt->st_rdev = htog((uint64_t)host->st_rdev); + tgt->st_size = htog((int64_t)host->st_size); + tgt->st_blocks = htog((uint64_t)host->st_blocks); + + if (fd == 1) + tgt->st_mode = htog((uint32_t)0x2190); + else + tgt->st_mode = htog((uint32_t)host->st_mode); + tgt->st_uid = htog((uint32_t)host->st_uid); + tgt->st_gid = htog((uint32_t)host->st_gid); + tgt->st_blksize = htog((uint32_t)host->st_blksize); + tgt->st_nlink = htog((uint32_t)host->st_nlink); + tgt->tgt_st_atime = htog((uint64_t)host->st_atime); + tgt->tgt_st_mtime = htog((uint64_t)host->st_mtime); + tgt->tgt_st_ctime = htog((uint64_t)host->st_ctime); +#if defined(STAT_HAVE_NSEC) + tgt->st_atime_nsec = htog(host->st_atime_nsec); + tgt->st_mtime_nsec = htog(host->st_mtime_nsec); + tgt->st_ctime_nsec = htog(host->st_ctime_nsec); #else tgt->st_atime_nsec = 0; tgt->st_mtime_nsec = 0; tgt->st_ctime_nsec = 0; #endif + tgt.copyOut(mem); } @@ -355,7 +404,7 @@ class Linux { // I don't think this exactly matches the HW FPCR fpcr.copyIn(xc->mem); DPRINTFR(SyscallVerbose, "osf_setsysinfo(SSI_IEEE_FP_CONTROL): " - " setting FPCR to 0x%x\n", *(uint64_t*)fpcr); + " setting FPCR to 0x%x\n", gtoh(*(uint64_t*)fpcr)); return 0; } diff --git a/arch/alpha/alpha_memory.cc b/arch/alpha/alpha_memory.cc index 8dda4d9c4..615ce92a4 100644 --- a/arch/alpha/alpha_memory.cc +++ b/arch/alpha/alpha_memory.cc @@ -303,7 +303,7 @@ AlphaITB::fault(Addr pc, ExecContext *xc) const } -Fault +Fault * AlphaITB::translate(MemReqPtr &req) const { InternalProcReg *ipr = req->xc->regs.ipr; @@ -312,7 +312,7 @@ AlphaITB::translate(MemReqPtr &req) const // strip off PAL PC marker (lsb is 1) req->paddr = (req->vaddr & ~3) & PAddrImplMask; hits++; - return No_Fault; + return NoFault; } if (req->flags & PHYSICAL) { @@ -322,7 +322,7 @@ AlphaITB::translate(MemReqPtr &req) const if (!validVirtualAddress(req->vaddr)) { fault(req->vaddr, req->xc); acv++; - return ITB_Acv_Fault; + return ItbAcvFault; } @@ -339,7 +339,7 @@ AlphaITB::translate(MemReqPtr &req) const AlphaISA::mode_kernel) { fault(req->vaddr, req->xc); acv++; - return ITB_Acv_Fault; + return ItbAcvFault; } req->paddr = req->vaddr & PAddrImplMask; @@ -360,7 +360,7 @@ AlphaITB::translate(MemReqPtr &req) const if (!pte) { fault(req->vaddr, req->xc); misses++; - return ITB_Fault_Fault; + return ItbPageFault; } req->paddr = (pte->ppn << AlphaISA::PageShift) + @@ -371,7 +371,7 @@ AlphaITB::translate(MemReqPtr &req) const // instruction access fault fault(req->vaddr, req->xc); acv++; - return ITB_Acv_Fault; + return ItbAcvFault; } hits++; @@ -380,11 +380,11 @@ AlphaITB::translate(MemReqPtr &req) const // check that the physical address is ok (catch bad physical addresses) if (req->paddr & ~PAddrImplMask) - return Machine_Check_Fault; + return MachineCheckFault; checkCacheability(req); - return No_Fault; + return NoFault; } /////////////////////////////////////////////////////////////////////// @@ -493,7 +493,7 @@ AlphaDTB::fault(MemReqPtr &req, uint64_t flags) const } } -Fault +Fault * AlphaDTB::translate(MemReqPtr &req, bool write) const { RegFile *regs = &req->xc->regs; @@ -511,7 +511,7 @@ AlphaDTB::translate(MemReqPtr &req, bool write) const fault(req, write ? MM_STAT_WR_MASK : 0); DPRINTF(TLB, "Alignment Fault on %#x, size = %d", req->vaddr, req->size); - return Alignment_Fault; + return AlignmentFault; } if (pc & 0x1) { @@ -530,7 +530,7 @@ AlphaDTB::translate(MemReqPtr &req, bool write) const MM_STAT_ACV_MASK); if (write) { write_acv++; } else { read_acv++; } - return DTB_Fault_Fault; + return DtbPageFault; } // Check for "superpage" mapping @@ -547,7 +547,7 @@ AlphaDTB::translate(MemReqPtr &req, bool write) const fault(req, ((write ? MM_STAT_WR_MASK : 0) | MM_STAT_ACV_MASK)); if (write) { write_acv++; } else { read_acv++; } - return DTB_Acv_Fault; + return DtbAcvFault; } req->paddr = req->vaddr & PAddrImplMask; @@ -575,7 +575,7 @@ AlphaDTB::translate(MemReqPtr &req, bool write) const fault(req, (write ? MM_STAT_WR_MASK : 0) | MM_STAT_DTB_MISS_MASK); if (write) { write_misses++; } else { read_misses++; } - return (req->flags & VPTE) ? Pdtb_Miss_Fault : Ndtb_Miss_Fault; + return (req->flags & VPTE) ? (Fault *)PDtbMissFault : (Fault *)NDtbMissFault; } req->paddr = (pte->ppn << AlphaISA::PageShift) + @@ -588,25 +588,25 @@ AlphaDTB::translate(MemReqPtr &req, bool write) const MM_STAT_ACV_MASK | (pte->fonw ? MM_STAT_FONW_MASK : 0)); write_acv++; - return DTB_Fault_Fault; + return DtbPageFault; } if (pte->fonw) { fault(req, MM_STAT_WR_MASK | MM_STAT_FONW_MASK); write_acv++; - return DTB_Fault_Fault; + return DtbPageFault; } } else { if (!(pte->xre & MODE2MASK(mode))) { fault(req, MM_STAT_ACV_MASK | (pte->fonr ? MM_STAT_FONR_MASK : 0)); read_acv++; - return DTB_Acv_Fault; + return DtbAcvFault; } if (pte->fonr) { fault(req, MM_STAT_FONR_MASK); read_acv++; - return DTB_Fault_Fault; + return DtbPageFault; } } } @@ -619,11 +619,11 @@ AlphaDTB::translate(MemReqPtr &req, bool write) const // check that the physical address is ok (catch bad physical addresses) if (req->paddr & ~PAddrImplMask) - return Machine_Check_Fault; + return MachineCheckFault; checkCacheability(req); - return No_Fault; + return NoFault; } AlphaISA::PTE & diff --git a/arch/alpha/alpha_memory.hh b/arch/alpha/alpha_memory.hh index 788923434..849063f59 100644 --- a/arch/alpha/alpha_memory.hh +++ b/arch/alpha/alpha_memory.hh @@ -32,6 +32,7 @@ #include <map> #include "arch/alpha/isa_traits.hh" +#include "arch/alpha/faults.hh" #include "base/statistics.hh" #include "mem/mem_req.hh" #include "sim/sim_object.hh" @@ -93,7 +94,7 @@ class AlphaITB : public AlphaTLB AlphaITB(const std::string &name, int size); virtual void regStats(); - Fault translate(MemReqPtr &req) const; + Fault * translate(MemReqPtr &req) const; }; class AlphaDTB : public AlphaTLB @@ -119,7 +120,7 @@ class AlphaDTB : public AlphaTLB AlphaDTB(const std::string &name, int size); virtual void regStats(); - Fault translate(MemReqPtr &req, bool write) const; + Fault * translate(MemReqPtr &req, bool write) const; }; #endif // __ALPHA_MEMORY_HH__ diff --git a/arch/alpha/alpha_tru64_process.cc b/arch/alpha/alpha_tru64_process.cc index b7a1c7d59..5c24adad9 100644 --- a/arch/alpha/alpha_tru64_process.cc +++ b/arch/alpha/alpha_tru64_process.cc @@ -28,7 +28,7 @@ #include <sys/types.h> #include <sys/stat.h> -#if defined(__OpenBSD__) +#if defined(__OpenBSD__) || defined(__APPLE__) || defined(__FreeBSD__) #include <sys/param.h> #include <sys/mount.h> #else @@ -41,6 +41,7 @@ #include <string.h> // for memset() #include <unistd.h> +#include "arch/alpha/isa_traits.hh" #include "arch/alpha/alpha_common_syscall_emul.hh" #include "arch/alpha/alpha_tru64_process.hh" #include "base/trace.hh" @@ -259,15 +260,15 @@ class Tru64 { /// Resource enumeration for getrlimit(). enum rlimit_resources { - RLIMIT_CPU = 0, - RLIMIT_FSIZE = 1, - RLIMIT_DATA = 2, - RLIMIT_STACK = 3, - RLIMIT_CORE = 4, - RLIMIT_RSS = 5, - RLIMIT_NOFILE = 6, - RLIMIT_AS = 7, - RLIMIT_VMEM = 7 + TGT_RLIMIT_CPU = 0, + TGT_RLIMIT_FSIZE = 1, + TGT_RLIMIT_DATA = 2, + TGT_RLIMIT_STACK = 3, + TGT_RLIMIT_CORE = 4, + TGT_RLIMIT_RSS = 5, + TGT_RLIMIT_NOFILE = 6, + TGT_RLIMIT_AS = 7, + TGT_RLIMIT_VMEM = 7 }; /// Limit struct for getrlimit/setrlimit. @@ -319,9 +320,9 @@ class Tru64 { //@{ /// For getrusage(). - static const int RUSAGE_THREAD = 1; - static const int RUSAGE_SELF = 0; - static const int RUSAGE_CHILDREN = -1; + static const int TGT_RUSAGE_THREAD = 1; + static const int TGT_RUSAGE_SELF = 0; + static const int TGT_RUSAGE_CHILDREN = -1; //@} /// For getrusage(). @@ -541,19 +542,19 @@ class Tru64 { { TypedBufferArg<T> tgt(addr); - tgt->st_dev = host->st_dev; - tgt->st_ino = host->st_ino; - tgt->st_mode = host->st_mode; - tgt->st_nlink = host->st_nlink; - tgt->st_uid = host->st_uid; - tgt->st_gid = host->st_gid; - tgt->st_rdev = host->st_rdev; - tgt->st_size = host->st_size; - tgt->st_atimeX = host->st_atime; - tgt->st_mtimeX = host->st_mtime; - tgt->st_ctimeX = host->st_ctime; - tgt->st_blksize = host->st_blksize; - tgt->st_blocks = host->st_blocks; + tgt->st_dev = htog(host->st_dev); + tgt->st_ino = htog(host->st_ino); + tgt->st_mode = htog(host->st_mode); + tgt->st_nlink = htog(host->st_nlink); + tgt->st_uid = htog(host->st_uid); + tgt->st_gid = htog(host->st_gid); + tgt->st_rdev = htog(host->st_rdev); + tgt->st_size = htog(host->st_size); + tgt->st_atimeX = htog(host->st_atime); + tgt->st_mtimeX = htog(host->st_mtime); + tgt->st_ctimeX = htog(host->st_ctime); + tgt->st_blksize = htog(host->st_blksize); + tgt->st_blocks = htog(host->st_blocks); tgt.copyOut(mem); } @@ -567,17 +568,19 @@ class Tru64 { { TypedBufferArg<T> tgt(addr); -#if defined(__OpenBSD__) +#if defined(__OpenBSD__) || defined(__APPLE__) || defined(__FreeBSD__) tgt->f_type = 0; #else - tgt->f_type = host->f_type; + tgt->f_type = htog(host->f_type); #endif - tgt->f_bsize = host->f_bsize; - tgt->f_blocks = host->f_blocks; - tgt->f_bfree = host->f_bfree; - tgt->f_bavail = host->f_bavail; - tgt->f_files = host->f_files; - tgt->f_ffree = host->f_ffree; + tgt->f_bsize = htog(host->f_bsize); + tgt->f_blocks = htog(host->f_blocks); + tgt->f_bfree = htog(host->f_bfree); + tgt->f_bavail = htog(host->f_bavail); + tgt->f_files = htog(host->f_files); + tgt->f_ffree = htog(host->f_ffree); + + // Is this as string normally? memcpy(&tgt->f_fsid, &host->f_fsid, sizeof(host->f_fsid)); tgt.copyOut(mem); @@ -622,19 +625,19 @@ class Tru64 { { TypedBufferArg<Tru64::pre_F64_stat> tgt(addr); - tgt->st_dev = host->st_dev; - tgt->st_ino = host->st_ino; - tgt->st_mode = host->st_mode; - tgt->st_nlink = host->st_nlink; - tgt->st_uid = host->st_uid; - tgt->st_gid = host->st_gid; - tgt->st_rdev = host->st_rdev; - tgt->st_size = host->st_size; - tgt->st_atimeX = host->st_atime; - tgt->st_mtimeX = host->st_mtime; - tgt->st_ctimeX = host->st_ctime; - tgt->st_blksize = host->st_blksize; - tgt->st_blocks = host->st_blocks; + tgt->st_dev = htog(host->st_dev); + tgt->st_ino = htog(host->st_ino); + tgt->st_mode = htog(host->st_mode); + tgt->st_nlink = htog(host->st_nlink); + tgt->st_uid = htog(host->st_uid); + tgt->st_gid = htog(host->st_gid); + tgt->st_rdev = htog(host->st_rdev); + tgt->st_size = htog(host->st_size); + tgt->st_atimeX = htog(host->st_atime); + tgt->st_mtimeX = htog(host->st_mtime); + tgt->st_ctimeX = htog(host->st_ctime); + tgt->st_blksize = htog(host->st_blksize); + tgt->st_blocks = htog(host->st_blocks); tgt.copyOut(mem); } @@ -673,21 +676,21 @@ class Tru64 { case Tru64::GSI_MAX_CPU: { TypedBufferArg<uint32_t> max_cpu(xc->getSyscallArg(1)); - *max_cpu = process->numCpus(); + *max_cpu = htog((uint32_t)process->numCpus()); max_cpu.copyOut(xc->mem); return 1; } case Tru64::GSI_CPUS_IN_BOX: { TypedBufferArg<uint32_t> cpus_in_box(xc->getSyscallArg(1)); - *cpus_in_box = process->numCpus(); + *cpus_in_box = htog((uint32_t)process->numCpus()); cpus_in_box.copyOut(xc->mem); return 1; } case Tru64::GSI_PHYSMEM: { TypedBufferArg<uint64_t> physmem(xc->getSyscallArg(1)); - *physmem = 1024 * 1024; // physical memory in KB + *physmem = htog((uint64_t)1024 * 1024); // physical memory in KB physmem.copyOut(xc->mem); return 1; } @@ -695,15 +698,15 @@ class Tru64 { case Tru64::GSI_CPU_INFO: { TypedBufferArg<Tru64::cpu_info> infop(xc->getSyscallArg(1)); - infop->current_cpu = 0; - infop->cpus_in_box = process->numCpus(); - infop->cpu_type = 57; - infop->ncpus = process->numCpus(); - int cpumask = (1 << process->numCpus()) - 1; - infop->cpus_present = infop->cpus_running = cpumask; - infop->cpu_binding = 0; - infop->cpu_ex_binding = 0; - infop->mhz = 667; + infop->current_cpu = htog(0); + infop->cpus_in_box = htog(process->numCpus()); + infop->cpu_type = htog(57); + infop->ncpus = htog(process->numCpus()); + uint64_t cpumask = (1 << process->numCpus()) - 1; + infop->cpus_present = infop->cpus_running = htog(cpumask); + infop->cpu_binding = htog(0); + infop->cpu_ex_binding = htog(0); + infop->mhz = htog(667); infop.copyOut(xc->mem); return 1; @@ -711,7 +714,7 @@ class Tru64 { case Tru64::GSI_PROC_TYPE: { TypedBufferArg<uint64_t> proc_type(xc->getSyscallArg(1)); - *proc_type = 11; + *proc_type = htog((uint64_t)11); proc_type.copyOut(xc->mem); return 1; } @@ -727,7 +730,7 @@ class Tru64 { case Tru64::GSI_CLK_TCK: { TypedBufferArg<uint64_t> clk_hz(xc->getSyscallArg(1)); - *clk_hz = 1024; + *clk_hz = htog((uint64_t)1024); clk_hz.copyOut(xc->mem); return 1; } @@ -823,7 +826,7 @@ class Tru64 { // just pass basep through uninterpreted. TypedBufferArg<int64_t> basep(tgt_basep); basep.copyIn(xc->mem); - long host_basep = (off_t)*basep; + long host_basep = (off_t)htog((int64_t)*basep); int host_result = getdirentries(fd, host_buf, tgt_nbytes, &host_basep); // check for error @@ -857,7 +860,7 @@ class Tru64 { delete [] host_buf; - *basep = host_basep; + *basep = htog((int64_t)host_basep); basep.copyOut(xc->mem); return tgt_buf_ptr - tgt_buf; @@ -878,14 +881,14 @@ class Tru64 { // Note that we'll advance PC <- NPC before the end of the cycle, // so we need to restore the desired PC into NPC. // The current regs->pc will get clobbered. - regs->npc = sc->sc_pc; + regs->npc = htog(sc->sc_pc); for (int i = 0; i < 31; ++i) { - regs->intRegFile[i] = sc->sc_regs[i]; - regs->floatRegFile.q[i] = sc->sc_fpregs[i]; + regs->intRegFile[i] = htog(sc->sc_regs[i]); + regs->floatRegFile.q[i] = htog(sc->sc_fpregs[i]); } - regs->miscRegs.fpcr = sc->sc_fpcr; + regs->miscRegs.fpcr = htog(sc->sc_fpcr); return 0; } @@ -908,15 +911,15 @@ class Tru64 { TypedBufferArg<Tru64::tbl_sysinfo> elp(xc->getSyscallArg(2)); const int clk_hz = one_million; - elp->si_user = curTick / (Clock::Frequency / clk_hz); - elp->si_nice = 0; - elp->si_sys = 0; - elp->si_idle = 0; - elp->wait = 0; - elp->si_hz = clk_hz; - elp->si_phz = clk_hz; - elp->si_boottime = seconds_since_epoch; // seconds since epoch? - elp->si_max_procs = process->numCpus(); + elp->si_user = htog(curTick / (Clock::Frequency / clk_hz)); + elp->si_nice = htog(0); + elp->si_sys = htog(0); + elp->si_idle = htog(0); + elp->wait = htog(0); + elp->si_hz = htog(clk_hz); + elp->si_phz = htog(clk_hz); + elp->si_boottime = htog(seconds_since_epoch); // seconds since epoch? + elp->si_max_procs = htog(process->numCpus()); elp.copyOut(xc->mem); return 0; } @@ -951,9 +954,10 @@ class Tru64 { // if the user chose an address, just let them have it. Otherwise // pick one for them. - if (argp->address == 0) { - argp->address = process->next_thread_stack_base; - int stack_size = (argp->rsize + argp->ysize + argp->gsize); + if (htog(argp->address) == 0) { + argp->address = htog(process->next_thread_stack_base); + int stack_size = (htog(argp->rsize) + htog(argp->ysize) + + htog(argp->gsize)); process->next_thread_stack_base -= stack_size; argp.copyOut(xc->mem); } @@ -977,14 +981,14 @@ class Tru64 { attrp.copyIn(xc->mem); - if (attrp->nxm_version != NXM_LIB_VERSION) { + if (gtoh(attrp->nxm_version) != NXM_LIB_VERSION) { cerr << "nxm_task_init: thread library version mismatch! " << "got " << attrp->nxm_version << ", expected " << NXM_LIB_VERSION << endl; abort(); } - if (attrp->flags != Tru64::NXM_TASK_INIT_VP) { + if (gtoh(attrp->flags) != Tru64::NXM_TASK_INIT_VP) { cerr << "nxm_task_init: bad flag value " << attrp->flags << " (expected " << Tru64::NXM_TASK_INIT_VP << ")" << endl; abort(); @@ -1011,10 +1015,10 @@ class Tru64 { // now initialize a config_info struct and copy it out to user space TypedBufferArg<Tru64::nxm_config_info> config(config_addr); - config->nxm_nslots_per_rad = process->numCpus(); - config->nxm_nrads = 1; // only one RAD in our system! - config->nxm_slot_state = slot_state_addr; - config->nxm_rad[0] = rad_state_addr; + config->nxm_nslots_per_rad = htog(process->numCpus()); + config->nxm_nrads = htog(1); // only one RAD in our system! + config->nxm_slot_state = htog(slot_state_addr); + config->nxm_rad[0] = htog(rad_state_addr); config.copyOut(xc->mem); @@ -1023,6 +1027,8 @@ class Tru64 { slot_state_size); for (int i = 0; i < process->numCpus(); ++i) { // CPU 0 is bound to the calling process; all others are available + // XXX this code should have an endian conversion, but I don't think + // it works anyway slot_state[i] = (i == 0) ? Tru64::NXM_SLOT_BOUND : Tru64::NXM_SLOT_AVAIL; } @@ -1040,24 +1046,24 @@ class Tru64 { rad_state->nxm_uniq_offset = attrp->nxm_uniq_offset; for (int i = 0; i < process->numCpus(); ++i) { Tru64::nxm_sched_state *ssp = &rad_state->nxm_ss[i]; - ssp->nxm_u.sigmask = 0; - ssp->nxm_u.sig = 0; - ssp->nxm_u.flags = 0; - ssp->nxm_u.cancel_state = 0; + ssp->nxm_u.sigmask = htog(0); + ssp->nxm_u.sig = htog(0); + ssp->nxm_u.flags = htog(0); + ssp->nxm_u.cancel_state = htog(0); ssp->nxm_u.nxm_ssig = 0; - ssp->nxm_bits = 0; + ssp->nxm_bits = htog(0); ssp->nxm_quantum = attrp->nxm_quantum; ssp->nxm_set_quantum = attrp->nxm_quantum; - ssp->nxm_sysevent = 0; + ssp->nxm_sysevent = htog(0); if (i == 0) { uint64_t uniq = xc->regs.miscRegs.uniq; - ssp->nxm_u.pth_id = uniq + attrp->nxm_uniq_offset; - ssp->nxm_u.nxm_active = uniq | 1; + ssp->nxm_u.pth_id = htog(uniq + gtoh(attrp->nxm_uniq_offset)); + ssp->nxm_u.nxm_active = htog(uniq | 1); } else { - ssp->nxm_u.pth_id = 0; - ssp->nxm_u.nxm_active = 0; + ssp->nxm_u.pth_id = htog(0); + ssp->nxm_u.nxm_active = htog(0); } } @@ -1066,7 +1072,7 @@ class Tru64 { // // copy pointer to shared config area out to user // - *configptr_ptr = config_addr; + *configptr_ptr = htog(config_addr); configptr_ptr.copyOut(xc->mem); // Register this as a valid address range with the process @@ -1083,13 +1089,13 @@ class Tru64 { { memset(&ec->regs, 0, sizeof(ec->regs)); - ec->regs.intRegFile[ArgumentReg0] = attrp->registers.a0; - ec->regs.intRegFile[27/*t12*/] = attrp->registers.pc; - ec->regs.intRegFile[StackPointerReg] = attrp->registers.sp; + ec->regs.intRegFile[ArgumentReg0] = gtoh(attrp->registers.a0); + ec->regs.intRegFile[27/*t12*/] = gtoh(attrp->registers.pc); + ec->regs.intRegFile[StackPointerReg] = gtoh(attrp->registers.sp); ec->regs.miscRegs.uniq = uniq_val; - ec->regs.pc = attrp->registers.pc; - ec->regs.npc = attrp->registers.pc + sizeof(MachInst); + ec->regs.pc = gtoh(attrp->registers.pc); + ec->regs.npc = gtoh(attrp->registers.pc) + sizeof(MachInst); ec->activate(); } @@ -1106,7 +1112,7 @@ class Tru64 { // get attribute args attrp.copyIn(xc->mem); - if (attrp->version != NXM_LIB_VERSION) { + if (gtoh(attrp->version) != NXM_LIB_VERSION) { cerr << "nxm_thread_create: thread library version mismatch! " << "got " << attrp->version << ", expected " << NXM_LIB_VERSION << endl; @@ -1131,28 +1137,28 @@ class Tru64 { rad_state_size); rad_state.copyIn(xc->mem); - uint64_t uniq_val = attrp->pthid - rad_state->nxm_uniq_offset; + uint64_t uniq_val = gtoh(attrp->pthid) - gtoh(rad_state->nxm_uniq_offset); - if (attrp->type == Tru64::NXM_TYPE_MANAGER) { + if (gtoh(attrp->type) == Tru64::NXM_TYPE_MANAGER) { // DEC pthreads seems to always create one of these (in // addition to N application threads), but we don't use it, // so don't bother creating it. // This is supposed to be a port number. Make something up. - *kidp = 99; + *kidp = htog(99); kidp.copyOut(xc->mem); return 0; - } else if (attrp->type == Tru64::NXM_TYPE_VP) { + } else if (gtoh(attrp->type) == Tru64::NXM_TYPE_VP) { // A real "virtual processor" kernel thread. Need to fork // this thread on another CPU. Tru64::nxm_sched_state *ssp = &rad_state->nxm_ss[thread_index]; - if (ssp->nxm_u.nxm_active != 0) + if (gtoh(ssp->nxm_u.nxm_active) != 0) return (int) Tru64::KERN_NOT_RECEIVER; ssp->nxm_u.pth_id = attrp->pthid; - ssp->nxm_u.nxm_active = uniq_val | 1; + ssp->nxm_u.nxm_active = htog(uniq_val | 1); rad_state.copyOut(xc->mem); @@ -1172,6 +1178,8 @@ class Tru64 { fatal(""); } + // XXX This should have an endian conversion but I think this code + // doesn't work anyway slot_state[thread_index] = Tru64::NXM_SLOT_BOUND; slot_state.copyOut(xc->mem); @@ -1187,7 +1195,7 @@ class Tru64 { // This is supposed to be a port number, but we'll try // and get away with just sticking the thread index // here. - *kidp = thread_index; + *kidp = htog(thread_index); kidp.copyOut(xc->mem); return 0; @@ -1319,9 +1327,9 @@ class Tru64 { lockp.copyIn(xc->mem); - if (*lockp == 0) { + if (gtoh(*lockp) == 0) { // lock is free: grab it - *lockp = 1; + *lockp = htog(1); lockp.copyOut(xc->mem); } else { // lock is busy: disable until free @@ -1374,9 +1382,9 @@ class Tru64 { lockp.copyIn(xc->mem); - if (*lockp == 0) { + if (gtoh(*lockp) == 0) { // lock is free: grab it - *lockp = 1; + *lockp = htog(1); lockp.copyOut(xc->mem); return 0; } else { @@ -1433,7 +1441,7 @@ class Tru64 { // user is supposed to acquire lock before entering lockp.copyIn(xc->mem); - assert(*lockp != 0); + assert(gtoh(*lockp) != 0); m5_unlock_mutex(lock_addr, process, xc); diff --git a/arch/alpha/ev5.cc b/arch/alpha/ev5.cc index 125affd03..72f48bfb2 100644 --- a/arch/alpha/ev5.cc +++ b/arch/alpha/ev5.cc @@ -76,7 +76,7 @@ AlphaISA::initCPU(RegFile *regs) // CPU comes up with PAL regs enabled swap_palshadow(regs, true); - regs->pc = regs->ipr[IPR_PAL_BASE] + fault_addr[Reset_Fault]; + regs->pc = regs->ipr[IPR_PAL_BASE] + fault_addr(ResetFault); regs->npc = regs->pc + sizeof(MachInst); } @@ -84,25 +84,15 @@ AlphaISA::initCPU(RegFile *regs) // // alpha exceptions - value equals trap address, update with MD_FAULT_TYPE // -Addr -AlphaISA::fault_addr[Num_Faults] = { - 0x0000, /* No_Fault */ - 0x0001, /* Reset_Fault */ - 0x0401, /* Machine_Check_Fault */ - 0x0501, /* Arithmetic_Fault */ - 0x0101, /* Interrupt_Fault */ - 0x0201, /* Ndtb_Miss_Fault */ - 0x0281, /* Pdtb_Miss_Fault */ - 0x0301, /* Alignment_Fault */ - 0x0381, /* DTB_Fault_Fault */ - 0x0381, /* DTB_Acv_Fault */ - 0x0181, /* ITB_Miss_Fault */ - 0x0181, /* ITB_Fault_Fault */ - 0x0081, /* ITB_Acv_Fault */ - 0x0481, /* Unimplemented_Opcode_Fault */ - 0x0581, /* Fen_Fault */ - 0x2001, /* Pal_Fault */ - 0x0501, /* Integer_Overflow_Fault: maps to Arithmetic_Fault */ +const Addr +AlphaISA::fault_addr(Fault * fault) +{ + //Check for the system wide faults + if(fault == NoFault) return 0x0000; + else if(fault == MachineCheckFault) return 0x0401; + else if(fault == AlignmentFault) return 0x0301; + //Deal with the alpha specific faults + return ((AlphaFault*)fault)->vect; }; const int AlphaISA::reg_redir[AlphaISA::NumIntRegs] = { @@ -168,7 +158,7 @@ AlphaISA::processInterrupts(CPU *cpu) if (ipl && ipl > ipr[IPR_IPLR]) { ipr[IPR_ISR] = summary; ipr[IPR_INTID] = ipl; - cpu->trap(Interrupt_Fault); + cpu->trap(InterruptFault); DPRINTF(Flow, "Interrupt! IPLR=%d ipl=%d summary=%x\n", ipr[IPR_IPLR], ipl, summary); } @@ -187,25 +177,25 @@ AlphaISA::zeroRegisters(CPU *cpu) } void -ExecContext::ev5_trap(Fault fault) +ExecContext::ev5_trap(Fault * fault) { - DPRINTF(Fault, "Fault %s at PC: %#x\n", FaultName(fault), regs.pc); - cpu->recordEvent(csprintf("Fault %s", FaultName(fault))); + DPRINTF(Fault, "Fault %s at PC: %#x\n", fault->name, regs.pc); + cpu->recordEvent(csprintf("Fault %s", fault->name)); assert(!misspeculating()); kernelStats->fault(fault); - if (fault == Arithmetic_Fault) + if (fault == ArithmeticFault) panic("Arithmetic traps are unimplemented!"); AlphaISA::InternalProcReg *ipr = regs.ipr; // exception restart address - if (fault != Interrupt_Fault || !inPalMode()) + if (fault != InterruptFault || !inPalMode()) ipr[AlphaISA::IPR_EXC_ADDR] = regs.pc; - if (fault == Pal_Fault || fault == Arithmetic_Fault /* || - fault == Interrupt_Fault && !inPalMode() */) { + if (fault == PalFault || fault == ArithmeticFault /* || + fault == InterruptFault && !inPalMode() */) { // traps... skip faulting instruction ipr[AlphaISA::IPR_EXC_ADDR] += 4; } @@ -213,22 +203,22 @@ ExecContext::ev5_trap(Fault fault) if (!inPalMode()) AlphaISA::swap_palshadow(®s, true); - regs.pc = ipr[AlphaISA::IPR_PAL_BASE] + AlphaISA::fault_addr[fault]; + regs.pc = ipr[AlphaISA::IPR_PAL_BASE] + AlphaISA::fault_addr(fault); regs.npc = regs.pc + sizeof(MachInst); } void -AlphaISA::intr_post(RegFile *regs, Fault fault, Addr pc) +AlphaISA::intr_post(RegFile *regs, Fault * fault, Addr pc) { InternalProcReg *ipr = regs->ipr; - bool use_pc = (fault == No_Fault); + bool use_pc = (fault == NoFault); - if (fault == Arithmetic_Fault) + if (fault == ArithmeticFault) panic("arithmetic faults NYI..."); // compute exception restart address - if (use_pc || fault == Pal_Fault || fault == Arithmetic_Fault) { + if (use_pc || fault == PalFault || fault == ArithmeticFault) { // traps... skip faulting instruction ipr[IPR_EXC_ADDR] = regs->pc + 4; } else { @@ -238,20 +228,20 @@ AlphaISA::intr_post(RegFile *regs, Fault fault, Addr pc) // jump to expection address (PAL PC bit set here as well...) if (!use_pc) - regs->npc = ipr[IPR_PAL_BASE] + fault_addr[fault]; + regs->npc = ipr[IPR_PAL_BASE] + fault_addr(fault); else regs->npc = ipr[IPR_PAL_BASE] + pc; // that's it! (orders of magnitude less painful than x86) } -Fault +Fault * ExecContext::hwrei() { uint64_t *ipr = regs.ipr; if (!inPalMode()) - return Unimplemented_Opcode_Fault; + return UnimplementedOpcodeFault; setNextPC(ipr[AlphaISA::IPR_EXC_ADDR]); @@ -265,11 +255,11 @@ ExecContext::hwrei() } // FIXME: XXX check for interrupts? XXX - return No_Fault; + return NoFault; } uint64_t -ExecContext::readIpr(int idx, Fault &fault) +ExecContext::readIpr(int idx, Fault * &fault) { uint64_t *ipr = regs.ipr; uint64_t retval = 0; // return value, default 0 @@ -363,12 +353,12 @@ ExecContext::readIpr(int idx, Fault &fault) case AlphaISA::IPR_DTB_IAP: case AlphaISA::IPR_ITB_IA: case AlphaISA::IPR_ITB_IAP: - fault = Unimplemented_Opcode_Fault; + fault = UnimplementedOpcodeFault; break; default: // invalid IPR - fault = Unimplemented_Opcode_Fault; + fault = UnimplementedOpcodeFault; break; } @@ -380,14 +370,14 @@ ExecContext::readIpr(int idx, Fault &fault) int break_ipl = -1; #endif -Fault +Fault * ExecContext::setIpr(int idx, uint64_t val) { uint64_t *ipr = regs.ipr; uint64_t old; if (misspeculating()) - return No_Fault; + return NoFault; switch (idx) { case AlphaISA::IPR_PALtemp0: @@ -533,7 +523,7 @@ ExecContext::setIpr(int idx, uint64_t val) case AlphaISA::IPR_ITB_PTE_TEMP: case AlphaISA::IPR_DTB_PTE_TEMP: // read-only registers - return Unimplemented_Opcode_Fault; + return UnimplementedOpcodeFault; case AlphaISA::IPR_HWINT_CLR: case AlphaISA::IPR_SL_XMIT: @@ -635,11 +625,11 @@ ExecContext::setIpr(int idx, uint64_t val) default: // invalid IPR - return Unimplemented_Opcode_Fault; + return UnimplementedOpcodeFault; } // no error... - return No_Fault; + return NoFault; } /** diff --git a/arch/alpha/faults.cc b/arch/alpha/faults.cc index 3aecf029d..e05b3fe59 100644 --- a/arch/alpha/faults.cc +++ b/arch/alpha/faults.cc @@ -28,34 +28,53 @@ #include "arch/alpha/faults.hh" -namespace { - const char * - fault_name[Num_Faults] = { - "none", - "reset", - "mchk", - "arith", - "interrupt", - "dtb_miss_single", - "dtb_miss_double", - "unalign", - "dfault", - "dfault", - "itbmiss", - "itbmiss", - "iaccvio", - "opdec", - "fen", - "pal", - }; -} +ResetFaultType * const ResetFault = + new ResetFaultType("reset", 1, 0x0001); +ArithmeticFaultType * const ArithmeticFault = + new ArithmeticFaultType("arith", 3, 0x0501); +InterruptFaultType * const InterruptFault = + new InterruptFaultType("interrupt", 4, 0x0101); +NDtbMissFaultType * const NDtbMissFault = + new NDtbMissFaultType("dtb_miss_single", 5, 0x0201); +PDtbMissFaultType * const PDtbMissFault = + new PDtbMissFaultType("dtb_miss_double", 6, 0x0281); +DtbPageFaultType * const DtbPageFault = + new DtbPageFaultType("dfault", 8, 0x0381); +DtbAcvFaultType * const DtbAcvFault = + new DtbAcvFaultType("dfault", 9, 0x0381); +ItbMissFaultType * const ItbMissFault = + new ItbMissFaultType("itbmiss", 10, 0x0181); +ItbPageFaultType * const ItbPageFault = + new ItbPageFaultType("itbmiss", 11, 0x0181); +ItbAcvFaultType * const ItbAcvFault = + new ItbAcvFaultType("iaccvio", 12, 0x0081); +UnimplementedOpcodeFaultType * const UnimplementedOpcodeFault = + new UnimplementedOpcodeFaultType("opdec", 13, 0x0481); +FloatEnableFaultType * const FloatEnableFault = + new FloatEnableFaultType("fen", 14, 0x0581); +PalFaultType * const PalFault = + new PalFaultType("pal", 15, 0x2001); +IntegerOverflowFaultType * const IntegerOverflowFault = + new IntegerOverflowFaultType("intover", 16, 0x0501); -const char * -FaultName(int index) -{ - if (index < 0 || index >= Num_Faults) - return 0; - - return fault_name[index]; -} +Fault ** ListOfFaults[] = { + (Fault **)&NoFault, + (Fault **)&ResetFault, + (Fault **)&MachineCheckFault, + (Fault **)&ArithmeticFault, + (Fault **)&InterruptFault, + (Fault **)&NDtbMissFault, + (Fault **)&PDtbMissFault, + (Fault **)&AlignmentFault, + (Fault **)&DtbPageFault, + (Fault **)&DtbAcvFault, + (Fault **)&ItbMissFault, + (Fault **)&ItbPageFault, + (Fault **)&ItbAcvFault, + (Fault **)&UnimplementedOpcodeFault, + (Fault **)&FloatEnableFault, + (Fault **)&PalFault, + (Fault **)&IntegerOverflowFault, + }; +int NumFaults = sizeof(ListOfFaults) / sizeof(Fault **); diff --git a/arch/alpha/faults.hh b/arch/alpha/faults.hh index bbac7cbf2..06605861a 100644 --- a/arch/alpha/faults.hh +++ b/arch/alpha/faults.hh @@ -26,32 +26,135 @@ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifndef __FAULTS_HH__ -#define __FAULTS_HH__ - -enum Fault { - No_Fault, - Reset_Fault, // processor reset - Machine_Check_Fault, // machine check (also internal S/W fault) - Arithmetic_Fault, // FP exception - Interrupt_Fault, // external interrupt - Ndtb_Miss_Fault, // DTB miss - Pdtb_Miss_Fault, // nested DTB miss - Alignment_Fault, // unaligned access - DTB_Fault_Fault, // DTB page fault - DTB_Acv_Fault, // DTB access violation - ITB_Miss_Fault, // ITB miss - ITB_Fault_Fault, // ITB page fault - ITB_Acv_Fault, // ITB access violation - Unimplemented_Opcode_Fault, // invalid/unimplemented instruction - Fen_Fault, // FP not-enabled fault - Pal_Fault, // call_pal S/W interrupt - Integer_Overflow_Fault, - Fake_Mem_Fault, - Num_Faults // number of faults +#ifndef __ALPHA_FAULTS_HH__ +#define __ALPHA_FAULTS_HH__ + +#include "sim/faults.hh" +#include "arch/isa_traits.hh" //For the Addr type + +class AlphaFault : public Fault +{ + public: + AlphaFault(char * newName, int newId, Addr newVect) + : Fault(newName, newId), vect(newVect) + {;} + + TheISA::Addr vect; }; -const char * -FaultName(int index); +extern class ResetFaultType : public AlphaFault +{ + public: + ResetFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const ResetFault; + +extern class ArithmeticFaultType : public AlphaFault +{ + public: + ArithmeticFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const ArithmeticFault; + +extern class InterruptFaultType : public AlphaFault +{ + public: + InterruptFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const InterruptFault; + +extern class NDtbMissFaultType : public AlphaFault +{ + public: + NDtbMissFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const NDtbMissFault; + +extern class PDtbMissFaultType : public AlphaFault +{ + public: + PDtbMissFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const PDtbMissFault; + +extern class DtbPageFaultType : public AlphaFault +{ + public: + DtbPageFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const DtbPageFault; + +extern class DtbAcvFaultType : public AlphaFault +{ + public: + DtbAcvFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const DtbAcvFault; + +extern class ItbMissFaultType : public AlphaFault +{ + public: + ItbMissFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const ItbMissFault; + +extern class ItbPageFaultType : public AlphaFault +{ + public: + ItbPageFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const ItbPageFault; + +extern class ItbAcvFaultType : public AlphaFault +{ + public: + ItbAcvFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const ItbAcvFault; + +extern class UnimplementedOpcodeFaultType : public AlphaFault +{ + public: + UnimplementedOpcodeFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const UnimplementedOpcodeFault; + +extern class FloatEnableFaultType : public AlphaFault +{ + public: + FloatEnableFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const FloatEnableFault; + +extern class PalFaultType : public AlphaFault +{ + public: + PalFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const PalFault; + +extern class IntegerOverflowFaultType : public AlphaFault +{ + public: + IntegerOverflowFaultType(char * newName, int newId, Addr newVect) + : AlphaFault(newName, newId, newVect) + {;} +} * const IntegerOverflowFault; + +extern Fault ** ListOfFaults[]; +extern int NumFaults; #endif // __FAULTS_HH__ diff --git a/arch/alpha/isa/branch.isa b/arch/alpha/isa/branch.isa new file mode 100644 index 000000000..e9c790c53 --- /dev/null +++ b/arch/alpha/isa/branch.isa @@ -0,0 +1,259 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output header {{ + + /** + * Base class for instructions whose disassembly is not purely a + * function of the machine instruction (i.e., it depends on the + * PC). This class overrides the disassemble() method to check + * the PC and symbol table values before re-using a cached + * disassembly string. This is necessary for branches and jumps, + * where the disassembly string includes the target address (which + * may depend on the PC and/or symbol table). + */ + class PCDependentDisassembly : public AlphaStaticInst + { + protected: + /// Cached program counter from last disassembly + mutable Addr cachedPC; + /// Cached symbol table pointer from last disassembly + mutable const SymbolTable *cachedSymtab; + + /// Constructor + PCDependentDisassembly(const char *mnem, MachInst _machInst, + OpClass __opClass) + : AlphaStaticInst(mnem, _machInst, __opClass), + cachedPC(0), cachedSymtab(0) + { + } + + const std::string & + disassemble(Addr pc, const SymbolTable *symtab) const; + }; + + /** + * Base class for branches (PC-relative control transfers), + * conditional or unconditional. + */ + class Branch : public PCDependentDisassembly + { + protected: + /// Displacement to target address (signed). + int32_t disp; + + /// Constructor. + Branch(const char *mnem, MachInst _machInst, OpClass __opClass) + : PCDependentDisassembly(mnem, _machInst, __opClass), + disp(BRDISP << 2) + { + } + + Addr branchTarget(Addr branchPC) const; + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; + + /** + * Base class for jumps (register-indirect control transfers). In + * the Alpha ISA, these are always unconditional. + */ + class Jump : public PCDependentDisassembly + { + protected: + + /// Displacement to target address (signed). + int32_t disp; + + public: + /// Constructor + Jump(const char *mnem, MachInst _machInst, OpClass __opClass) + : PCDependentDisassembly(mnem, _machInst, __opClass), + disp(BRDISP) + { + } + + Addr branchTarget(ExecContext *xc) const; + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + +output decoder {{ + Addr + Branch::branchTarget(Addr branchPC) const + { + return branchPC + 4 + disp; + } + + Addr + Jump::branchTarget(ExecContext *xc) const + { + Addr NPC = xc->readPC() + 4; + uint64_t Rb = xc->readIntReg(_srcRegIdx[0]); + return (Rb & ~3) | (NPC & 1); + } + + const std::string & + PCDependentDisassembly::disassemble(Addr pc, + const SymbolTable *symtab) const + { + if (!cachedDisassembly || + pc != cachedPC || symtab != cachedSymtab) + { + if (cachedDisassembly) + delete cachedDisassembly; + + cachedDisassembly = + new std::string(generateDisassembly(pc, symtab)); + cachedPC = pc; + cachedSymtab = symtab; + } + + return *cachedDisassembly; + } + + std::string + Branch::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + std::stringstream ss; + + ccprintf(ss, "%-10s ", mnemonic); + + // There's only one register arg (RA), but it could be + // either a source (the condition for conditional + // branches) or a destination (the link reg for + // unconditional branches) + if (_numSrcRegs > 0) { + printReg(ss, _srcRegIdx[0]); + ss << ","; + } + else if (_numDestRegs > 0) { + printReg(ss, _destRegIdx[0]); + ss << ","; + } + +#ifdef SS_COMPATIBLE_DISASSEMBLY + if (_numSrcRegs == 0 && _numDestRegs == 0) { + printReg(ss, 31); + ss << ","; + } +#endif + + Addr target = pc + 4 + disp; + + std::string str; + if (symtab && symtab->findSymbol(target, str)) + ss << str; + else + ccprintf(ss, "0x%x", target); + + return ss.str(); + } + + std::string + Jump::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + std::stringstream ss; + + ccprintf(ss, "%-10s ", mnemonic); + +#ifdef SS_COMPATIBLE_DISASSEMBLY + if (_numDestRegs == 0) { + printReg(ss, 31); + ss << ","; + } +#endif + + if (_numDestRegs > 0) { + printReg(ss, _destRegIdx[0]); + ss << ","; + } + + ccprintf(ss, "(r%d)", RB); + + return ss.str(); + } +}}; + +def template JumpOrBranchDecode {{ + return (RA == 31) + ? (StaticInst<AlphaISA> *)new %(class_name)s(machInst) + : (StaticInst<AlphaISA> *)new %(class_name)sAndLink(machInst); +}}; + +def format CondBranch(code) {{ + code = 'bool cond;\n' + code + '\nif (cond) NPC = NPC + disp;\n'; + iop = InstObjParams(name, Name, 'Branch', CodeBlock(code), + ('IsDirectControl', 'IsCondControl')) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + decode_block = BasicDecode.subst(iop) + exec_output = BasicExecute.subst(iop) +}}; + +let {{ +def UncondCtrlBase(name, Name, base_class, npc_expr, flags): + # Declare basic control transfer w/o link (i.e. link reg is R31) + nolink_code = 'NPC = %s;\n' % npc_expr + nolink_iop = InstObjParams(name, Name, base_class, + CodeBlock(nolink_code), flags) + header_output = BasicDeclare.subst(nolink_iop) + decoder_output = BasicConstructor.subst(nolink_iop) + exec_output = BasicExecute.subst(nolink_iop) + + # Generate declaration of '*AndLink' version, append to decls + link_code = 'Ra = NPC & ~3;\n' + nolink_code + link_iop = InstObjParams(name, Name + 'AndLink', base_class, + CodeBlock(link_code), flags) + header_output += BasicDeclare.subst(link_iop) + decoder_output += BasicConstructor.subst(link_iop) + exec_output += BasicExecute.subst(link_iop) + + # need to use link_iop for the decode template since it is expecting + # the shorter version of class_name (w/o "AndLink") + + return (header_output, decoder_output, + JumpOrBranchDecode.subst(nolink_iop), exec_output) +}}; + +def format UncondBranch(*flags) {{ + flags += ('IsUncondControl', 'IsDirectControl') + (header_output, decoder_output, decode_block, exec_output) = \ + UncondCtrlBase(name, Name, 'Branch', 'NPC + disp', flags) +}}; + +def format Jump(*flags) {{ + flags += ('IsUncondControl', 'IsIndirectControl') + (header_output, decoder_output, decode_block, exec_output) = \ + UncondCtrlBase(name, Name, 'Jump', '(Rb & ~3) | (NPC & 1)', flags) +}}; + + diff --git a/arch/alpha/isa/decoder.isa b/arch/alpha/isa/decoder.isa new file mode 100644 index 000000000..37b15416b --- /dev/null +++ b/arch/alpha/isa/decoder.isa @@ -0,0 +1,802 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +decode OPCODE default Unknown::unknown() { + + format LoadAddress { + 0x08: lda({{ Ra = Rb + disp; }}); + 0x09: ldah({{ Ra = Rb + (disp << 16); }}); + } + + format LoadOrNop { + 0x0a: ldbu({{ Ra.uq = Mem.ub; }}); + 0x0c: ldwu({{ Ra.uq = Mem.uw; }}); + 0x0b: ldq_u({{ Ra = Mem.uq; }}, ea_code = {{ EA = (Rb + disp) & ~7; }}); + 0x23: ldt({{ Fa = Mem.df; }}); + 0x2a: ldl_l({{ Ra.sl = Mem.sl; }}, mem_flags = LOCKED); + 0x2b: ldq_l({{ Ra.uq = Mem.uq; }}, mem_flags = LOCKED); + 0x20: MiscPrefetch::copy_load({{ EA = Ra; }}, + {{ fault = xc->copySrcTranslate(EA); }}, + inst_flags = [IsMemRef, IsLoad, IsCopy]); + } + + format LoadOrPrefetch { + 0x28: ldl({{ Ra.sl = Mem.sl; }}); + 0x29: ldq({{ Ra.uq = Mem.uq; }}, pf_flags = EVICT_NEXT); + // IsFloating flag on lds gets the prefetch to disassemble + // using f31 instead of r31... funcitonally it's unnecessary + 0x22: lds({{ Fa.uq = s_to_t(Mem.ul); }}, + pf_flags = PF_EXCLUSIVE, inst_flags = IsFloating); + } + + format Store { + 0x0e: stb({{ Mem.ub = Ra<7:0>; }}); + 0x0d: stw({{ Mem.uw = Ra<15:0>; }}); + 0x2c: stl({{ Mem.ul = Ra<31:0>; }}); + 0x2d: stq({{ Mem.uq = Ra.uq; }}); + 0x0f: stq_u({{ Mem.uq = Ra.uq; }}, {{ EA = (Rb + disp) & ~7; }}); + 0x26: sts({{ Mem.ul = t_to_s(Fa.uq); }}); + 0x27: stt({{ Mem.df = Fa; }}); + 0x24: MiscPrefetch::copy_store({{ EA = Rb; }}, + {{ fault = xc->copy(EA); }}, + inst_flags = [IsMemRef, IsStore, IsCopy]); + } + + format StoreCond { + 0x2e: stl_c({{ Mem.ul = Ra<31:0>; }}, + {{ + uint64_t tmp = write_result; + // see stq_c + Ra = (tmp == 0 || tmp == 1) ? tmp : Ra; + }}, mem_flags = LOCKED); + 0x2f: stq_c({{ Mem.uq = Ra; }}, + {{ + uint64_t tmp = write_result; + // If the write operation returns 0 or 1, then + // this was a conventional store conditional, + // and the value indicates the success/failure + // of the operation. If another value is + // returned, then this was a Turbolaser + // mailbox access, and we don't update the + // result register at all. + Ra = (tmp == 0 || tmp == 1) ? tmp : Ra; + }}, mem_flags = LOCKED); + } + + format IntegerOperate { + + 0x10: decode INTFUNC { // integer arithmetic operations + + 0x00: addl({{ Rc.sl = Ra.sl + Rb_or_imm.sl; }}); + 0x40: addlv({{ + uint32_t tmp = Ra.sl + Rb_or_imm.sl; + // signed overflow occurs when operands have same sign + // and sign of result does not match. + if (Ra.sl<31:> == Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>) + fault = IntegerOverflowFault; + Rc.sl = tmp; + }}); + 0x02: s4addl({{ Rc.sl = (Ra.sl << 2) + Rb_or_imm.sl; }}); + 0x12: s8addl({{ Rc.sl = (Ra.sl << 3) + Rb_or_imm.sl; }}); + + 0x20: addq({{ Rc = Ra + Rb_or_imm; }}); + 0x60: addqv({{ + uint64_t tmp = Ra + Rb_or_imm; + // signed overflow occurs when operands have same sign + // and sign of result does not match. + if (Ra<63:> == Rb_or_imm<63:> && tmp<63:> != Ra<63:>) + fault = IntegerOverflowFault; + Rc = tmp; + }}); + 0x22: s4addq({{ Rc = (Ra << 2) + Rb_or_imm; }}); + 0x32: s8addq({{ Rc = (Ra << 3) + Rb_or_imm; }}); + + 0x09: subl({{ Rc.sl = Ra.sl - Rb_or_imm.sl; }}); + 0x49: sublv({{ + uint32_t tmp = Ra.sl - Rb_or_imm.sl; + // signed overflow detection is same as for add, + // except we need to look at the *complemented* + // sign bit of the subtrahend (Rb), i.e., if the initial + // signs are the *same* then no overflow can occur + if (Ra.sl<31:> != Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>) + fault = IntegerOverflowFault; + Rc.sl = tmp; + }}); + 0x0b: s4subl({{ Rc.sl = (Ra.sl << 2) - Rb_or_imm.sl; }}); + 0x1b: s8subl({{ Rc.sl = (Ra.sl << 3) - Rb_or_imm.sl; }}); + + 0x29: subq({{ Rc = Ra - Rb_or_imm; }}); + 0x69: subqv({{ + uint64_t tmp = Ra - Rb_or_imm; + // signed overflow detection is same as for add, + // except we need to look at the *complemented* + // sign bit of the subtrahend (Rb), i.e., if the initial + // signs are the *same* then no overflow can occur + if (Ra<63:> != Rb_or_imm<63:> && tmp<63:> != Ra<63:>) + fault = IntegerOverflowFault; + Rc = tmp; + }}); + 0x2b: s4subq({{ Rc = (Ra << 2) - Rb_or_imm; }}); + 0x3b: s8subq({{ Rc = (Ra << 3) - Rb_or_imm; }}); + + 0x2d: cmpeq({{ Rc = (Ra == Rb_or_imm); }}); + 0x6d: cmple({{ Rc = (Ra.sq <= Rb_or_imm.sq); }}); + 0x4d: cmplt({{ Rc = (Ra.sq < Rb_or_imm.sq); }}); + 0x3d: cmpule({{ Rc = (Ra.uq <= Rb_or_imm.uq); }}); + 0x1d: cmpult({{ Rc = (Ra.uq < Rb_or_imm.uq); }}); + + 0x0f: cmpbge({{ + int hi = 7; + int lo = 0; + uint64_t tmp = 0; + for (int i = 0; i < 8; ++i) { + tmp |= (Ra.uq<hi:lo> >= Rb_or_imm.uq<hi:lo>) << i; + hi += 8; + lo += 8; + } + Rc = tmp; + }}); + } + + 0x11: decode INTFUNC { // integer logical operations + + 0x00: and({{ Rc = Ra & Rb_or_imm; }}); + 0x08: bic({{ Rc = Ra & ~Rb_or_imm; }}); + 0x20: bis({{ Rc = Ra | Rb_or_imm; }}); + 0x28: ornot({{ Rc = Ra | ~Rb_or_imm; }}); + 0x40: xor({{ Rc = Ra ^ Rb_or_imm; }}); + 0x48: eqv({{ Rc = Ra ^ ~Rb_or_imm; }}); + + // conditional moves + 0x14: cmovlbs({{ Rc = ((Ra & 1) == 1) ? Rb_or_imm : Rc; }}); + 0x16: cmovlbc({{ Rc = ((Ra & 1) == 0) ? Rb_or_imm : Rc; }}); + 0x24: cmoveq({{ Rc = (Ra == 0) ? Rb_or_imm : Rc; }}); + 0x26: cmovne({{ Rc = (Ra != 0) ? Rb_or_imm : Rc; }}); + 0x44: cmovlt({{ Rc = (Ra.sq < 0) ? Rb_or_imm : Rc; }}); + 0x46: cmovge({{ Rc = (Ra.sq >= 0) ? Rb_or_imm : Rc; }}); + 0x64: cmovle({{ Rc = (Ra.sq <= 0) ? Rb_or_imm : Rc; }}); + 0x66: cmovgt({{ Rc = (Ra.sq > 0) ? Rb_or_imm : Rc; }}); + + // For AMASK, RA must be R31. + 0x61: decode RA { + 31: amask({{ Rc = Rb_or_imm & ~ULL(0x17); }}); + } + + // For IMPLVER, RA must be R31 and the B operand + // must be the immediate value 1. + 0x6c: decode RA { + 31: decode IMM { + 1: decode INTIMM { + // return EV5 for FULL_SYSTEM and EV6 otherwise + 1: implver({{ +#if FULL_SYSTEM + Rc = 1; +#else + Rc = 2; +#endif + }}); + } + } + } + +#if FULL_SYSTEM + // The mysterious 11.25... + 0x25: WarnUnimpl::eleven25(); +#endif + } + + 0x12: decode INTFUNC { + 0x39: sll({{ Rc = Ra << Rb_or_imm<5:0>; }}); + 0x34: srl({{ Rc = Ra.uq >> Rb_or_imm<5:0>; }}); + 0x3c: sra({{ Rc = Ra.sq >> Rb_or_imm<5:0>; }}); + + 0x02: mskbl({{ Rc = Ra & ~(mask( 8) << (Rb_or_imm<2:0> * 8)); }}); + 0x12: mskwl({{ Rc = Ra & ~(mask(16) << (Rb_or_imm<2:0> * 8)); }}); + 0x22: mskll({{ Rc = Ra & ~(mask(32) << (Rb_or_imm<2:0> * 8)); }}); + 0x32: mskql({{ Rc = Ra & ~(mask(64) << (Rb_or_imm<2:0> * 8)); }}); + + 0x52: mskwh({{ + int bv = Rb_or_imm<2:0>; + Rc = bv ? (Ra & ~(mask(16) >> (64 - 8 * bv))) : Ra; + }}); + 0x62: msklh({{ + int bv = Rb_or_imm<2:0>; + Rc = bv ? (Ra & ~(mask(32) >> (64 - 8 * bv))) : Ra; + }}); + 0x72: mskqh({{ + int bv = Rb_or_imm<2:0>; + Rc = bv ? (Ra & ~(mask(64) >> (64 - 8 * bv))) : Ra; + }}); + + 0x06: extbl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))< 7:0>; }}); + 0x16: extwl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<15:0>; }}); + 0x26: extll({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<31:0>; }}); + 0x36: extql({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8)); }}); + + 0x5a: extwh({{ + Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<15:0>; }}); + 0x6a: extlh({{ + Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<31:0>; }}); + 0x7a: extqh({{ + Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>); }}); + + 0x0b: insbl({{ Rc = Ra< 7:0> << (Rb_or_imm<2:0> * 8); }}); + 0x1b: inswl({{ Rc = Ra<15:0> << (Rb_or_imm<2:0> * 8); }}); + 0x2b: insll({{ Rc = Ra<31:0> << (Rb_or_imm<2:0> * 8); }}); + 0x3b: insql({{ Rc = Ra << (Rb_or_imm<2:0> * 8); }}); + + 0x57: inswh({{ + int bv = Rb_or_imm<2:0>; + Rc = bv ? (Ra.uq<15:0> >> (64 - 8 * bv)) : 0; + }}); + 0x67: inslh({{ + int bv = Rb_or_imm<2:0>; + Rc = bv ? (Ra.uq<31:0> >> (64 - 8 * bv)) : 0; + }}); + 0x77: insqh({{ + int bv = Rb_or_imm<2:0>; + Rc = bv ? (Ra.uq >> (64 - 8 * bv)) : 0; + }}); + + 0x30: zap({{ + uint64_t zapmask = 0; + for (int i = 0; i < 8; ++i) { + if (Rb_or_imm<i:>) + zapmask |= (mask(8) << (i * 8)); + } + Rc = Ra & ~zapmask; + }}); + 0x31: zapnot({{ + uint64_t zapmask = 0; + for (int i = 0; i < 8; ++i) { + if (!Rb_or_imm<i:>) + zapmask |= (mask(8) << (i * 8)); + } + Rc = Ra & ~zapmask; + }}); + } + + 0x13: decode INTFUNC { // integer multiplies + 0x00: mull({{ Rc.sl = Ra.sl * Rb_or_imm.sl; }}, IntMultOp); + 0x20: mulq({{ Rc = Ra * Rb_or_imm; }}, IntMultOp); + 0x30: umulh({{ + uint64_t hi, lo; + mul128(Ra, Rb_or_imm, hi, lo); + Rc = hi; + }}, IntMultOp); + 0x40: mullv({{ + // 32-bit multiply with trap on overflow + int64_t Rax = Ra.sl; // sign extended version of Ra.sl + int64_t Rbx = Rb_or_imm.sl; + int64_t tmp = Rax * Rbx; + // To avoid overflow, all the upper 32 bits must match + // the sign bit of the lower 32. We code this as + // checking the upper 33 bits for all 0s or all 1s. + uint64_t sign_bits = tmp<63:31>; + if (sign_bits != 0 && sign_bits != mask(33)) + fault = IntegerOverflowFault; + Rc.sl = tmp<31:0>; + }}, IntMultOp); + 0x60: mulqv({{ + // 64-bit multiply with trap on overflow + uint64_t hi, lo; + mul128(Ra, Rb_or_imm, hi, lo); + // all the upper 64 bits must match the sign bit of + // the lower 64 + if (!((hi == 0 && lo<63:> == 0) || + (hi == mask(64) && lo<63:> == 1))) + fault = IntegerOverflowFault; + Rc = lo; + }}, IntMultOp); + } + + 0x1c: decode INTFUNC { + 0x00: decode RA { 31: sextb({{ Rc.sb = Rb_or_imm< 7:0>; }}); } + 0x01: decode RA { 31: sextw({{ Rc.sw = Rb_or_imm<15:0>; }}); } + 0x32: ctlz({{ + uint64_t count = 0; + uint64_t temp = Rb; + if (temp<63:32>) temp >>= 32; else count += 32; + if (temp<31:16>) temp >>= 16; else count += 16; + if (temp<15:8>) temp >>= 8; else count += 8; + if (temp<7:4>) temp >>= 4; else count += 4; + if (temp<3:2>) temp >>= 2; else count += 2; + if (temp<1:1>) temp >>= 1; else count += 1; + if ((temp<0:0>) != 0x1) count += 1; + Rc = count; + }}, IntAluOp); + + 0x33: cttz({{ + uint64_t count = 0; + uint64_t temp = Rb; + if (!(temp<31:0>)) { temp >>= 32; count += 32; } + if (!(temp<15:0>)) { temp >>= 16; count += 16; } + if (!(temp<7:0>)) { temp >>= 8; count += 8; } + if (!(temp<3:0>)) { temp >>= 4; count += 4; } + if (!(temp<1:0>)) { temp >>= 2; count += 2; } + if (!(temp<0:0> & ULL(0x1))) count += 1; + Rc = count; + }}, IntAluOp); + + format FailUnimpl { + 0x30: ctpop(); + 0x31: perr(); + 0x34: unpkbw(); + 0x35: unpkbl(); + 0x36: pkwb(); + 0x37: pklb(); + 0x38: minsb8(); + 0x39: minsw4(); + 0x3a: minub8(); + 0x3b: minuw4(); + 0x3c: maxub8(); + 0x3d: maxuw4(); + 0x3e: maxsb8(); + 0x3f: maxsw4(); + } + + format BasicOperateWithNopCheck { + 0x70: decode RB { + 31: ftoit({{ Rc = Fa.uq; }}, FloatCvtOp); + } + 0x78: decode RB { + 31: ftois({{ Rc.sl = t_to_s(Fa.uq); }}, + FloatCvtOp); + } + } + } + } + + // Conditional branches. + format CondBranch { + 0x39: beq({{ cond = (Ra == 0); }}); + 0x3d: bne({{ cond = (Ra != 0); }}); + 0x3e: bge({{ cond = (Ra.sq >= 0); }}); + 0x3f: bgt({{ cond = (Ra.sq > 0); }}); + 0x3b: ble({{ cond = (Ra.sq <= 0); }}); + 0x3a: blt({{ cond = (Ra.sq < 0); }}); + 0x38: blbc({{ cond = ((Ra & 1) == 0); }}); + 0x3c: blbs({{ cond = ((Ra & 1) == 1); }}); + + 0x31: fbeq({{ cond = (Fa == 0); }}); + 0x35: fbne({{ cond = (Fa != 0); }}); + 0x36: fbge({{ cond = (Fa >= 0); }}); + 0x37: fbgt({{ cond = (Fa > 0); }}); + 0x33: fble({{ cond = (Fa <= 0); }}); + 0x32: fblt({{ cond = (Fa < 0); }}); + } + + // unconditional branches + format UncondBranch { + 0x30: br(); + 0x34: bsr(IsCall); + } + + // indirect branches + 0x1a: decode JMPFUNC { + format Jump { + 0: jmp(); + 1: jsr(IsCall); + 2: ret(IsReturn); + 3: jsr_coroutine(IsCall, IsReturn); + } + } + + // Square root and integer-to-FP moves + 0x14: decode FP_SHORTFUNC { + // Integer to FP register moves must have RB == 31 + 0x4: decode RB { + 31: decode FP_FULLFUNC { + format BasicOperateWithNopCheck { + 0x004: itofs({{ Fc.uq = s_to_t(Ra.ul); }}, FloatCvtOp); + 0x024: itoft({{ Fc.uq = Ra.uq; }}, FloatCvtOp); + 0x014: FailUnimpl::itoff(); // VAX-format conversion + } + } + } + + // Square root instructions must have FA == 31 + 0xb: decode FA { + 31: decode FP_TYPEFUNC { + format FloatingPointOperate { +#if SS_COMPATIBLE_FP + 0x0b: sqrts({{ + if (Fb < 0.0) + fault = ArithmeticFault; + Fc = sqrt(Fb); + }}, FloatSqrtOp); +#else + 0x0b: sqrts({{ + if (Fb.sf < 0.0) + fault = ArithmeticFault; + Fc.sf = sqrt(Fb.sf); + }}, FloatSqrtOp); +#endif + 0x2b: sqrtt({{ + if (Fb < 0.0) + fault = ArithmeticFault; + Fc = sqrt(Fb); + }}, FloatSqrtOp); + } + } + } + + // VAX-format sqrtf and sqrtg are not implemented + 0xa: FailUnimpl::sqrtfg(); + } + + // IEEE floating point + 0x16: decode FP_SHORTFUNC_TOP2 { + // The top two bits of the short function code break this + // space into four groups: binary ops, compares, reserved, and + // conversions. See Table 4-12 of AHB. There are different + // special cases in these different groups, so we decode on + // these top two bits first just to select a decode strategy. + // Most of these instructions may have various trapping and + // rounding mode flags set; these are decoded in the + // FloatingPointDecode template used by the + // FloatingPointOperate format. + + // add/sub/mul/div: just decode on the short function code + // and source type. All valid trapping and rounding modes apply. + 0: decode FP_TRAPMODE { + // check for valid trapping modes here + 0,1,5,7: decode FP_TYPEFUNC { + format FloatingPointOperate { +#if SS_COMPATIBLE_FP + 0x00: adds({{ Fc = Fa + Fb; }}); + 0x01: subs({{ Fc = Fa - Fb; }}); + 0x02: muls({{ Fc = Fa * Fb; }}, FloatMultOp); + 0x03: divs({{ Fc = Fa / Fb; }}, FloatDivOp); +#else + 0x00: adds({{ Fc.sf = Fa.sf + Fb.sf; }}); + 0x01: subs({{ Fc.sf = Fa.sf - Fb.sf; }}); + 0x02: muls({{ Fc.sf = Fa.sf * Fb.sf; }}, FloatMultOp); + 0x03: divs({{ Fc.sf = Fa.sf / Fb.sf; }}, FloatDivOp); +#endif + + 0x20: addt({{ Fc = Fa + Fb; }}); + 0x21: subt({{ Fc = Fa - Fb; }}); + 0x22: mult({{ Fc = Fa * Fb; }}, FloatMultOp); + 0x23: divt({{ Fc = Fa / Fb; }}, FloatDivOp); + } + } + } + + // Floating-point compare instructions must have the default + // rounding mode, and may use the default trapping mode or + // /SU. Both trapping modes are treated the same by M5; the + // only difference on the real hardware (as far a I can tell) + // is that without /SU you'd get an imprecise trap if you + // tried to compare a NaN with something else (instead of an + // "unordered" result). + 1: decode FP_FULLFUNC { + format BasicOperateWithNopCheck { + 0x0a5, 0x5a5: cmpteq({{ Fc = (Fa == Fb) ? 2.0 : 0.0; }}, + FloatCmpOp); + 0x0a7, 0x5a7: cmptle({{ Fc = (Fa <= Fb) ? 2.0 : 0.0; }}, + FloatCmpOp); + 0x0a6, 0x5a6: cmptlt({{ Fc = (Fa < Fb) ? 2.0 : 0.0; }}, + FloatCmpOp); + 0x0a4, 0x5a4: cmptun({{ // unordered + Fc = (!(Fa < Fb) && !(Fa == Fb) && !(Fa > Fb)) ? 2.0 : 0.0; + }}, FloatCmpOp); + } + } + + // The FP-to-integer and integer-to-FP conversion insts + // require that FA be 31. + 3: decode FA { + 31: decode FP_TYPEFUNC { + format FloatingPointOperate { + 0x2f: decode FP_ROUNDMODE { + format FPFixedRounding { + // "chopped" i.e. round toward zero + 0: cvttq({{ Fc.sq = (int64_t)trunc(Fb); }}, + Chopped); + // round to minus infinity + 1: cvttq({{ Fc.sq = (int64_t)floor(Fb); }}, + MinusInfinity); + } + default: cvttq({{ Fc.sq = (int64_t)nearbyint(Fb); }}); + } + + // The cvtts opcode is overloaded to be cvtst if the trap + // mode is 2 or 6 (which are not valid otherwise) + 0x2c: decode FP_FULLFUNC { + format BasicOperateWithNopCheck { + // trap on denorm version "cvtst/s" is + // simulated same as cvtst + 0x2ac, 0x6ac: cvtst({{ Fc = Fb.sf; }}); + } + default: cvtts({{ Fc.sf = Fb; }}); + } + + // The trapping mode for integer-to-FP conversions + // must be /SUI or nothing; /U and /SU are not + // allowed. The full set of rounding modes are + // supported though. + 0x3c: decode FP_TRAPMODE { + 0,7: cvtqs({{ Fc.sf = Fb.sq; }}); + } + 0x3e: decode FP_TRAPMODE { + 0,7: cvtqt({{ Fc = Fb.sq; }}); + } + } + } + } + } + + // misc FP operate + 0x17: decode FP_FULLFUNC { + format BasicOperateWithNopCheck { + 0x010: cvtlq({{ + Fc.sl = (Fb.uq<63:62> << 30) | Fb.uq<58:29>; + }}); + 0x030: cvtql({{ + Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29); + }}); + + // We treat the precise & imprecise trapping versions of + // cvtql identically. + 0x130, 0x530: cvtqlv({{ + // To avoid overflow, all the upper 32 bits must match + // the sign bit of the lower 32. We code this as + // checking the upper 33 bits for all 0s or all 1s. + uint64_t sign_bits = Fb.uq<63:31>; + if (sign_bits != 0 && sign_bits != mask(33)) + fault = IntegerOverflowFault; + Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29); + }}); + + 0x020: cpys({{ // copy sign + Fc.uq = (Fa.uq<63:> << 63) | Fb.uq<62:0>; + }}); + 0x021: cpysn({{ // copy sign negated + Fc.uq = (~Fa.uq<63:> << 63) | Fb.uq<62:0>; + }}); + 0x022: cpyse({{ // copy sign and exponent + Fc.uq = (Fa.uq<63:52> << 52) | Fb.uq<51:0>; + }}); + + 0x02a: fcmoveq({{ Fc = (Fa == 0) ? Fb : Fc; }}); + 0x02b: fcmovne({{ Fc = (Fa != 0) ? Fb : Fc; }}); + 0x02c: fcmovlt({{ Fc = (Fa < 0) ? Fb : Fc; }}); + 0x02d: fcmovge({{ Fc = (Fa >= 0) ? Fb : Fc; }}); + 0x02e: fcmovle({{ Fc = (Fa <= 0) ? Fb : Fc; }}); + 0x02f: fcmovgt({{ Fc = (Fa > 0) ? Fb : Fc; }}); + + 0x024: mt_fpcr({{ FPCR = Fa.uq; }}); + 0x025: mf_fpcr({{ Fa.uq = FPCR; }}); + } + } + + // miscellaneous mem-format ops + 0x18: decode MEMFUNC { + format WarnUnimpl { + 0x8000: fetch(); + 0xa000: fetch_m(); + 0xe800: ecb(); + } + + format MiscPrefetch { + 0xf800: wh64({{ EA = Rb & ~ULL(63); }}, + {{ xc->writeHint(EA, 64, memAccessFlags); }}, + mem_flags = NO_FAULT, + inst_flags = [IsMemRef, IsDataPrefetch, + IsStore, MemWriteOp]); + } + + format BasicOperate { + 0xc000: rpcc({{ +#if FULL_SYSTEM + /* Rb is a fake dependency so here is a fun way to get + * the parser to understand that. + */ + Ra = xc->readIpr(AlphaISA::IPR_CC, fault) + (Rb & 0); + +#else + Ra = curTick; +#endif + }}); + + // All of the barrier instructions below do nothing in + // their execute() methods (hence the empty code blocks). + // All of their functionality is hard-coded in the + // pipeline based on the flags IsSerializing, + // IsMemBarrier, and IsWriteBarrier. In the current + // detailed CPU model, the execute() function only gets + // called at fetch, so there's no way to generate pipeline + // behavior at any other stage. Once we go to an + // exec-in-exec CPU model we should be able to get rid of + // these flags and implement this behavior via the + // execute() methods. + + // trapb is just a barrier on integer traps, where excb is + // a barrier on integer and FP traps. "EXCB is thus a + // superset of TRAPB." (Alpha ARM, Sec 4.11.4) We treat + // them the same though. + 0x0000: trapb({{ }}, IsSerializing, No_OpClass); + 0x0400: excb({{ }}, IsSerializing, No_OpClass); + 0x4000: mb({{ }}, IsMemBarrier, MemReadOp); + 0x4400: wmb({{ }}, IsWriteBarrier, MemWriteOp); + } + +#if FULL_SYSTEM + format BasicOperate { + 0xe000: rc({{ + Ra = xc->readIntrFlag(); + xc->setIntrFlag(0); + }}, IsNonSpeculative); + 0xf000: rs({{ + Ra = xc->readIntrFlag(); + xc->setIntrFlag(1); + }}, IsNonSpeculative); + } +#else + format FailUnimpl { + 0xe000: rc(); + 0xf000: rs(); + } +#endif + } + +#if FULL_SYSTEM + 0x00: CallPal::call_pal({{ + if (!palValid || + (palPriv + && xc->readIpr(AlphaISA::IPR_ICM, fault) != AlphaISA::mode_kernel)) { + // invalid pal function code, or attempt to do privileged + // PAL call in non-kernel mode + fault = UnimplementedOpcodeFault; + } + else { + // check to see if simulator wants to do something special + // on this PAL call (including maybe suppress it) + bool dopal = xc->simPalCheck(palFunc); + + if (dopal) { + AlphaISA::swap_palshadow(&xc->xcBase()->regs, true); + xc->setIpr(AlphaISA::IPR_EXC_ADDR, NPC); + NPC = xc->readIpr(AlphaISA::IPR_PAL_BASE, fault) + palOffset; + } + } + }}, IsNonSpeculative); +#else + 0x00: decode PALFUNC { + format EmulatedCallPal { + 0x00: halt ({{ + SimExit(curTick, "halt instruction encountered"); + }}, IsNonSpeculative); + 0x83: callsys({{ + xc->syscall(); + }}, IsNonSpeculative); + // Read uniq reg into ABI return value register (r0) + 0x9e: rduniq({{ R0 = Runiq; }}); + // Write uniq reg with value from ABI arg register (r16) + 0x9f: wruniq({{ Runiq = R16; }}); + } + } +#endif + +#if FULL_SYSTEM + format HwLoad { + 0x1b: decode HW_LDST_QUAD { + 0: hw_ld({{ EA = (Rb + disp) & ~3; }}, {{ Ra = Mem.ul; }}, L); + 1: hw_ld({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem.uq; }}, Q); + } + } + + format HwStore { + 0x1f: decode HW_LDST_COND { + 0: decode HW_LDST_QUAD { + 0: hw_st({{ EA = (Rb + disp) & ~3; }}, + {{ Mem.ul = Ra<31:0>; }}, L); + 1: hw_st({{ EA = (Rb + disp) & ~7; }}, + {{ Mem.uq = Ra.uq; }}, Q); + } + + 1: FailUnimpl::hw_st_cond(); + } + } + + format HwMoveIPR { + 0x19: hw_mfpr({{ + // this instruction is only valid in PAL mode + if (!xc->inPalMode()) { + fault = UnimplementedOpcodeFault; + } + else { + Ra = xc->readIpr(ipr_index, fault); + } + }}); + 0x1d: hw_mtpr({{ + // this instruction is only valid in PAL mode + if (!xc->inPalMode()) { + fault = UnimplementedOpcodeFault; + } + else { + xc->setIpr(ipr_index, Ra); + if (traceData) { traceData->setData(Ra); } + } + }}); + } + + format BasicOperate { + 0x1e: hw_rei({{ xc->hwrei(); }}, IsSerializing); + + // M5 special opcodes use the reserved 0x01 opcode space + 0x01: decode M5FUNC { + 0x00: arm({{ + AlphaPseudo::arm(xc->xcBase()); + }}, IsNonSpeculative); + 0x01: quiesce({{ + AlphaPseudo::quiesce(xc->xcBase()); + }}, IsNonSpeculative); + 0x10: ivlb({{ + AlphaPseudo::ivlb(xc->xcBase()); + }}, No_OpClass, IsNonSpeculative); + 0x11: ivle({{ + AlphaPseudo::ivle(xc->xcBase()); + }}, No_OpClass, IsNonSpeculative); + 0x20: m5exit_old({{ + AlphaPseudo::m5exit_old(xc->xcBase()); + }}, No_OpClass, IsNonSpeculative); + 0x21: m5exit({{ + AlphaPseudo::m5exit(xc->xcBase(), R16); + }}, No_OpClass, IsNonSpeculative); + 0x30: initparam({{ Ra = xc->xcBase()->cpu->system->init_param; }}); + 0x40: resetstats({{ + AlphaPseudo::resetstats(xc->xcBase(), R16, R17); + }}, IsNonSpeculative); + 0x41: dumpstats({{ + AlphaPseudo::dumpstats(xc->xcBase(), R16, R17); + }}, IsNonSpeculative); + 0x42: dumpresetstats({{ + AlphaPseudo::dumpresetstats(xc->xcBase(), R16, R17); + }}, IsNonSpeculative); + 0x43: m5checkpoint({{ + AlphaPseudo::m5checkpoint(xc->xcBase(), R16, R17); + }}, IsNonSpeculative); + 0x50: m5readfile({{ + R0 = AlphaPseudo::readfile(xc->xcBase(), R16, R17, R18); + }}, IsNonSpeculative); + 0x51: m5break({{ + AlphaPseudo::debugbreak(xc->xcBase()); + }}, IsNonSpeculative); + 0x52: m5switchcpu({{ + AlphaPseudo::switchcpu(xc->xcBase()); + }}, IsNonSpeculative); + 0x53: m5addsymbol({{ + AlphaPseudo::addsymbol(xc->xcBase(), R16, R17); + }}, IsNonSpeculative); + + } + } +#endif +} diff --git a/arch/alpha/isa/fp.isa b/arch/alpha/isa/fp.isa new file mode 100644 index 000000000..c718c5524 --- /dev/null +++ b/arch/alpha/isa/fp.isa @@ -0,0 +1,300 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output exec {{ + /// Check "FP enabled" machine status bit. Called when executing any FP + /// instruction in full-system mode. + /// @retval Full-system mode: NoFault if FP is enabled, FenFault + /// if not. Non-full-system mode: always returns NoFault. +#if FULL_SYSTEM + inline Fault * checkFpEnableFault(%(CPU_exec_context)s *xc) + { + Fault * fault = NoFault; // dummy... this ipr access should not fault + if (!EV5::ICSR_FPE(xc->readIpr(AlphaISA::IPR_ICSR, fault))) { + fault = FloatEnableFault; + } + return fault; + } +#else + inline Fault * checkFpEnableFault(%(CPU_exec_context)s *xc) + { + return NoFault; + } +#endif +}}; + +output header {{ + /** + * Base class for general floating-point instructions. Includes + * support for various Alpha rounding and trapping modes. Only FP + * instructions that require this support are derived from this + * class; the rest derive directly from AlphaStaticInst. + */ + class AlphaFP : public AlphaStaticInst + { + public: + /// Alpha FP rounding modes. + enum RoundingMode { + Chopped = 0, ///< round toward zero + Minus_Infinity = 1, ///< round toward minus infinity + Normal = 2, ///< round to nearest (default) + Dynamic = 3, ///< use FPCR setting (in instruction) + Plus_Infinity = 3 ///< round to plus inifinity (in FPCR) + }; + + /// Alpha FP trapping modes. + /// For instructions that produce integer results, the + /// "Underflow Enable" modes really mean "Overflow Enable", and + /// the assembly modifier is V rather than U. + enum TrappingMode { + /// default: nothing enabled + Imprecise = 0, ///< no modifier + /// underflow/overflow traps enabled, inexact disabled + Underflow_Imprecise = 1, ///< /U or /V + Underflow_Precise = 5, ///< /SU or /SV + /// underflow/overflow and inexact traps enabled + Underflow_Inexact_Precise = 7 ///< /SUI or /SVI + }; + + protected: + /// Map Alpha rounding mode to C99 constants from <fenv.h>. + static const int alphaToC99RoundingMode[]; + + /// Map enum RoundingMode values to disassembly suffixes. + static const char *roundingModeSuffix[]; + /// Map enum TrappingMode values to FP disassembly suffixes. + static const char *fpTrappingModeSuffix[]; + /// Map enum TrappingMode values to integer disassembly suffixes. + static const char *intTrappingModeSuffix[]; + + /// This instruction's rounding mode. + RoundingMode roundingMode; + /// This instruction's trapping mode. + TrappingMode trappingMode; + + /// Have we warned about this instruction's unsupported + /// rounding mode (if applicable)? + mutable bool warnedOnRounding; + + /// Have we warned about this instruction's unsupported + /// trapping mode (if applicable)? + mutable bool warnedOnTrapping; + + /// Constructor + AlphaFP(const char *mnem, MachInst _machInst, OpClass __opClass) + : AlphaStaticInst(mnem, _machInst, __opClass), + roundingMode((enum RoundingMode)FP_ROUNDMODE), + trappingMode((enum TrappingMode)FP_TRAPMODE), + warnedOnRounding(false), + warnedOnTrapping(false) + { + } + + int getC99RoundingMode(uint64_t fpcr_val) const; + + // This differs from the AlphaStaticInst version only in + // printing suffixes for non-default rounding & trapping modes. + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; + +}}; + + +output decoder {{ + int + AlphaFP::getC99RoundingMode(uint64_t fpcr_val) const + { + if (roundingMode == Dynamic) { + return alphaToC99RoundingMode[bits(fpcr_val, 59, 58)]; + } + else { + return alphaToC99RoundingMode[roundingMode]; + } + } + + std::string + AlphaFP::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + std::string mnem_str(mnemonic); + +#ifndef SS_COMPATIBLE_DISASSEMBLY + std::string suffix(""); + suffix += ((_destRegIdx[0] >= FP_Base_DepTag) + ? fpTrappingModeSuffix[trappingMode] + : intTrappingModeSuffix[trappingMode]); + suffix += roundingModeSuffix[roundingMode]; + + if (suffix != "") { + mnem_str = csprintf("%s/%s", mnemonic, suffix); + } +#endif + + std::stringstream ss; + ccprintf(ss, "%-10s ", mnem_str.c_str()); + + // just print the first two source regs... if there's + // a third one, it's a read-modify-write dest (Rc), + // e.g. for CMOVxx + if (_numSrcRegs > 0) { + printReg(ss, _srcRegIdx[0]); + } + if (_numSrcRegs > 1) { + ss << ","; + printReg(ss, _srcRegIdx[1]); + } + + // just print the first dest... if there's a second one, + // it's generally implicit + if (_numDestRegs > 0) { + if (_numSrcRegs > 0) + ss << ","; + printReg(ss, _destRegIdx[0]); + } + + return ss.str(); + } + + const int AlphaFP::alphaToC99RoundingMode[] = { + FE_TOWARDZERO, // Chopped + FE_DOWNWARD, // Minus_Infinity + FE_TONEAREST, // Normal + FE_UPWARD // Dynamic in inst, Plus_Infinity in FPCR + }; + + const char *AlphaFP::roundingModeSuffix[] = { "c", "m", "", "d" }; + // mark invalid trapping modes, but don't fail on them, because + // you could decode anything on a misspeculated path + const char *AlphaFP::fpTrappingModeSuffix[] = + { "", "u", "INVTM2", "INVTM3", "INVTM4", "su", "INVTM6", "sui" }; + const char *AlphaFP::intTrappingModeSuffix[] = + { "", "v", "INVTM2", "INVTM3", "INVTM4", "sv", "INVTM6", "svi" }; +}}; + +// FP instruction class execute method template. Handles non-standard +// rounding modes. +def template FloatingPointExecute {{ + Fault * %(class_name)s::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + if (trappingMode != Imprecise && !warnedOnTrapping) { + warn("%s: non-standard trapping mode not supported", + generateDisassembly(0, NULL)); + warnedOnTrapping = true; + } + + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; +#if USE_FENV + if (roundingMode == Normal) { + %(code)s; + } else { + fesetround(getC99RoundingMode(xc->readFpcr())); + %(code)s; + fesetround(FE_TONEAREST); + } +#else + if (roundingMode != Normal && !warnedOnRounding) { + warn("%s: non-standard rounding mode not supported", + generateDisassembly(0, NULL)); + warnedOnRounding = true; + } + %(code)s; +#endif + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + +// FP instruction class execute method template where no dynamic +// rounding mode control is needed. Like BasicExecute, but includes +// check & warning for non-standard trapping mode. +def template FPFixedRoundingExecute {{ + Fault * %(class_name)s::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + if (trappingMode != Imprecise && !warnedOnTrapping) { + warn("%s: non-standard trapping mode not supported", + generateDisassembly(0, NULL)); + warnedOnTrapping = true; + } + + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + %(code)s; + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + +def template FloatingPointDecode {{ + { + AlphaStaticInst *i = new %(class_name)s(machInst); + if (FC == 31) { + i = makeNop(i); + } + return i; + } +}}; + +// General format for floating-point operate instructions: +// - Checks trapping and rounding mode flags. Trapping modes +// currently unimplemented (will fail). +// - Generates NOP if FC == 31. +def format FloatingPointOperate(code, *opt_args) {{ + iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args) + decode_block = FloatingPointDecode.subst(iop) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + exec_output = FloatingPointExecute.subst(iop) +}}; + +// Special format for cvttq where rounding mode is pre-decoded +def format FPFixedRounding(code, class_suffix, *opt_args) {{ + Name += class_suffix + iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args) + decode_block = FloatingPointDecode.subst(iop) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + exec_output = FPFixedRoundingExecute.subst(iop) +}}; + diff --git a/arch/alpha/isa/int.isa b/arch/alpha/isa/int.isa new file mode 100644 index 000000000..049437f8c --- /dev/null +++ b/arch/alpha/isa/int.isa @@ -0,0 +1,128 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output header {{ + /** + * Base class for integer immediate instructions. + */ + class IntegerImm : public AlphaStaticInst + { + protected: + /// Immediate operand value (unsigned 8-bit int). + uint8_t imm; + + /// Constructor + IntegerImm(const char *mnem, MachInst _machInst, OpClass __opClass) + : AlphaStaticInst(mnem, _machInst, __opClass), imm(INTIMM) + { + } + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + +output decoder {{ + std::string + IntegerImm::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + std::stringstream ss; + + ccprintf(ss, "%-10s ", mnemonic); + + // just print the first source reg... if there's + // a second one, it's a read-modify-write dest (Rc), + // e.g. for CMOVxx + if (_numSrcRegs > 0) { + printReg(ss, _srcRegIdx[0]); + ss << ","; + } + + ss << (int)imm; + + if (_numDestRegs > 0) { + ss << ","; + printReg(ss, _destRegIdx[0]); + } + + return ss.str(); + } +}}; + + +def template RegOrImmDecode {{ + { + AlphaStaticInst *i = + (IMM) ? (AlphaStaticInst *)new %(class_name)sImm(machInst) + : (AlphaStaticInst *)new %(class_name)s(machInst); + if (RC == 31) { + i = makeNop(i); + } + return i; + } +}}; + +// Primary format for integer operate instructions: +// - Generates both reg-reg and reg-imm versions if Rb_or_imm is used. +// - Generates NOP if RC == 31. +def format IntegerOperate(code, *opt_flags) {{ + # If the code block contains 'Rb_or_imm', we define two instructions, + # one using 'Rb' and one using 'imm', and have the decoder select + # the right one. + uses_imm = (code.find('Rb_or_imm') != -1) + if uses_imm: + orig_code = code + # base code is reg version: + # rewrite by substituting 'Rb' for 'Rb_or_imm' + code = re.sub(r'Rb_or_imm', 'Rb', orig_code) + # generate immediate version by substituting 'imm' + # note that imm takes no extenstion, so we extend + # the regexp to replace any extension as well + imm_code = re.sub(r'Rb_or_imm(\.\w+)?', 'imm', orig_code) + + # generate declaration for register version + cblk = CodeBlock(code) + iop = InstObjParams(name, Name, 'AlphaStaticInst', cblk, opt_flags) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + exec_output = BasicExecute.subst(iop) + + if uses_imm: + # append declaration for imm version + imm_cblk = CodeBlock(imm_code) + imm_iop = InstObjParams(name, Name + 'Imm', 'IntegerImm', imm_cblk, + opt_flags) + header_output += BasicDeclare.subst(imm_iop) + decoder_output += BasicConstructor.subst(imm_iop) + exec_output += BasicExecute.subst(imm_iop) + # decode checks IMM bit to pick correct version + decode_block = RegOrImmDecode.subst(iop) + else: + # no imm version: just check for nop + decode_block = OperateNopCheckDecode.subst(iop) +}}; diff --git a/arch/alpha/isa/main.isa b/arch/alpha/isa/main.isa new file mode 100644 index 000000000..42fb29404 --- /dev/null +++ b/arch/alpha/isa/main.isa @@ -0,0 +1,435 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output header {{ +#include <sstream> +#include <iostream> +#include <iomanip> + +#include "config/ss_compatible_fp.hh" +#include "cpu/static_inst.hh" +#include "arch/alpha/faults.hh" +#include "mem/mem_req.hh" // some constructors use MemReq flags +}}; + +output decoder {{ +#include "base/cprintf.hh" +#include "base/fenv.hh" +#include "base/loader/symtab.hh" +#include "config/ss_compatible_fp.hh" +#include "cpu/exec_context.hh" // for Jump::branchTarget() + +#include <math.h> +}}; + +output exec {{ +#include <math.h> + +#if FULL_SYSTEM +#include "sim/pseudo_inst.hh" +#endif +#include "base/fenv.hh" +#include "config/ss_compatible_fp.hh" +#include "cpu/base.hh" +#include "cpu/exetrace.hh" +#include "sim/sim_exit.hh" +}}; + +//////////////////////////////////////////////////////////////////// +// +// Namespace statement. Everything below this line will be in the +// AlphaISAInst namespace. +// + + +namespace AlphaISA; + +//////////////////////////////////////////////////////////////////// +// +// Bitfield definitions. +// + +// Universal (format-independent) fields +def bitfield OPCODE <31:26>; +def bitfield RA <25:21>; +def bitfield RB <20:16>; + +// Memory format +def signed bitfield MEMDISP <15: 0>; // displacement +def bitfield MEMFUNC <15: 0>; // function code (same field, unsigned) + +// Memory-format jumps +def bitfield JMPFUNC <15:14>; // function code (disp<15:14>) +def bitfield JMPHINT <13: 0>; // tgt Icache idx hint (disp<13:0>) + +// Branch format +def signed bitfield BRDISP <20: 0>; // displacement + +// Integer operate format(s>; +def bitfield INTIMM <20:13>; // integer immediate (literal) +def bitfield IMM <12:12>; // immediate flag +def bitfield INTFUNC <11: 5>; // function code +def bitfield RC < 4: 0>; // dest reg + +// Floating-point operate format +def bitfield FA <25:21>; +def bitfield FB <20:16>; +def bitfield FP_FULLFUNC <15: 5>; // complete function code + def bitfield FP_TRAPMODE <15:13>; // trapping mode + def bitfield FP_ROUNDMODE <12:11>; // rounding mode + def bitfield FP_TYPEFUNC <10: 5>; // type+func: handiest for decoding + def bitfield FP_SRCTYPE <10: 9>; // source reg type + def bitfield FP_SHORTFUNC < 8: 5>; // short function code + def bitfield FP_SHORTFUNC_TOP2 <8:7>; // top 2 bits of short func code +def bitfield FC < 4: 0>; // dest reg + +// PALcode format +def bitfield PALFUNC <25: 0>; // function code + +// EV5 PAL instructions: +// HW_LD/HW_ST +def bitfield HW_LDST_PHYS <15>; // address is physical +def bitfield HW_LDST_ALT <14>; // use ALT_MODE IPR +def bitfield HW_LDST_WRTCK <13>; // HW_LD only: fault if no write acc +def bitfield HW_LDST_QUAD <12>; // size: 0=32b, 1=64b +def bitfield HW_LDST_VPTE <11>; // HW_LD only: is PTE fetch +def bitfield HW_LDST_LOCK <10>; // HW_LD only: is load locked +def bitfield HW_LDST_COND <10>; // HW_ST only: is store conditional +def signed bitfield HW_LDST_DISP <9:0>; // signed displacement + +// HW_REI +def bitfield HW_REI_TYP <15:14>; // type: stalling vs. non-stallingk +def bitfield HW_REI_MBZ <13: 0>; // must be zero + +// HW_MTPR/MW_MFPR +def bitfield HW_IPR_IDX <15:0>; // IPR index + +// M5 instructions +def bitfield M5FUNC <7:0>; + +def operand_types {{ + 'sb' : ('signed int', 8), + 'ub' : ('unsigned int', 8), + 'sw' : ('signed int', 16), + 'uw' : ('unsigned int', 16), + 'sl' : ('signed int', 32), + 'ul' : ('unsigned int', 32), + 'sq' : ('signed int', 64), + 'uq' : ('unsigned int', 64), + 'sf' : ('float', 32), + 'df' : ('float', 64) +}}; + +def operands {{ + # Int regs default to unsigned, but code should not count on this. + # For clarity, descriptions that depend on unsigned behavior should + # explicitly specify '.uq'. + 'Ra': ('IntReg', 'uq', 'RA', 'IsInteger', 1), + 'Rb': ('IntReg', 'uq', 'RB', 'IsInteger', 2), + 'Rc': ('IntReg', 'uq', 'RC', 'IsInteger', 3), + 'Fa': ('FloatReg', 'df', 'FA', 'IsFloating', 1), + 'Fb': ('FloatReg', 'df', 'FB', 'IsFloating', 2), + 'Fc': ('FloatReg', 'df', 'FC', 'IsFloating', 3), + 'Mem': ('Mem', 'uq', None, ('IsMemRef', 'IsLoad', 'IsStore'), 4), + 'NPC': ('NPC', 'uq', None, ( None, None, 'IsControl' ), 4), + 'Runiq': ('ControlReg', 'uq', 'Uniq', None, 1), + 'FPCR': (' ControlReg', 'uq', 'Fpcr', None, 1), + # The next two are hacks for non-full-system call-pal emulation + 'R0': ('IntReg', 'uq', '0', None, 1), + 'R16': ('IntReg', 'uq', '16', None, 1), + 'R17': ('IntReg', 'uq', '17', None, 1), + 'R18': ('IntReg', 'uq', '18', None, 1) +}}; + +//////////////////////////////////////////////////////////////////// +// +// Basic instruction classes/templates/formats etc. +// + +output header {{ +// uncomment the following to get SimpleScalar-compatible disassembly +// (useful for diffing output traces). +// #define SS_COMPATIBLE_DISASSEMBLY + + /** + * Base class for all Alpha static instructions. + */ + class AlphaStaticInst : public StaticInst<AlphaISA> + { + protected: + + /// Make AlphaISA register dependence tags directly visible in + /// this class and derived classes. Maybe these should really + /// live here and not in the AlphaISA namespace. + enum DependenceTags { + FP_Base_DepTag = AlphaISA::FP_Base_DepTag, + Fpcr_DepTag = AlphaISA::Fpcr_DepTag, + Uniq_DepTag = AlphaISA::Uniq_DepTag, + IPR_Base_DepTag = AlphaISA::IPR_Base_DepTag + }; + + /// Constructor. + AlphaStaticInst(const char *mnem, MachInst _machInst, + OpClass __opClass) + : StaticInst<AlphaISA>(mnem, _machInst, __opClass) + { + } + + /// Print a register name for disassembly given the unique + /// dependence tag number (FP or int). + void printReg(std::ostream &os, int reg) const; + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + +output decoder {{ + void + AlphaStaticInst::printReg(std::ostream &os, int reg) const + { + if (reg < FP_Base_DepTag) { + ccprintf(os, "r%d", reg); + } + else { + ccprintf(os, "f%d", reg - FP_Base_DepTag); + } + } + + std::string + AlphaStaticInst::generateDisassembly(Addr pc, + const SymbolTable *symtab) const + { + std::stringstream ss; + + ccprintf(ss, "%-10s ", mnemonic); + + // just print the first two source regs... if there's + // a third one, it's a read-modify-write dest (Rc), + // e.g. for CMOVxx + if (_numSrcRegs > 0) { + printReg(ss, _srcRegIdx[0]); + } + if (_numSrcRegs > 1) { + ss << ","; + printReg(ss, _srcRegIdx[1]); + } + + // just print the first dest... if there's a second one, + // it's generally implicit + if (_numDestRegs > 0) { + if (_numSrcRegs > 0) + ss << ","; + printReg(ss, _destRegIdx[0]); + } + + return ss.str(); + } +}}; + +// Declarations for execute() methods. +def template BasicExecDeclare {{ + Fault * execute(%(CPU_exec_context)s *, Trace::InstRecord *) const; +}}; + +// Basic instruction class declaration template. +def template BasicDeclare {{ + /** + * Static instruction class for "%(mnemonic)s". + */ + class %(class_name)s : public %(base_class)s + { + public: + /// Constructor. + %(class_name)s(MachInst machInst); + + %(BasicExecDeclare)s + }; +}}; + +// Basic instruction class constructor template. +def template BasicConstructor {{ + inline %(class_name)s::%(class_name)s(MachInst machInst) + : %(base_class)s("%(mnemonic)s", machInst, %(op_class)s) + { + %(constructor)s; + } +}}; + +// Basic instruction class execute method template. +def template BasicExecute {{ + Fault * %(class_name)s::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + %(code)s; + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + +// Basic decode template. +def template BasicDecode {{ + return new %(class_name)s(machInst); +}}; + +// Basic decode template, passing mnemonic in as string arg to constructor. +def template BasicDecodeWithMnemonic {{ + return new %(class_name)s("%(mnemonic)s", machInst); +}}; + +// The most basic instruction format... used only for a few misc. insts +def format BasicOperate(code, *flags) {{ + iop = InstObjParams(name, Name, 'AlphaStaticInst', CodeBlock(code), flags) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + decode_block = BasicDecode.subst(iop) + exec_output = BasicExecute.subst(iop) +}}; + + + +//////////////////////////////////////////////////////////////////// +// +// Nop +// + +output header {{ + /** + * Static instruction class for no-ops. This is a leaf class. + */ + class Nop : public AlphaStaticInst + { + /// Disassembly of original instruction. + const std::string originalDisassembly; + + public: + /// Constructor + Nop(const std::string _originalDisassembly, MachInst _machInst) + : AlphaStaticInst("nop", _machInst, No_OpClass), + originalDisassembly(_originalDisassembly) + { + flags[IsNop] = true; + } + + ~Nop() { } + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + + %(BasicExecDeclare)s + }; +}}; + +output decoder {{ + std::string Nop::generateDisassembly(Addr pc, + const SymbolTable *symtab) const + { +#ifdef SS_COMPATIBLE_DISASSEMBLY + return originalDisassembly; +#else + return csprintf("%-10s (%s)", "nop", originalDisassembly); +#endif + } + + /// Helper function for decoding nops. Substitute Nop object + /// for original inst passed in as arg (and delete latter). + inline + AlphaStaticInst * + makeNop(AlphaStaticInst *inst) + { + AlphaStaticInst *nop = new Nop(inst->disassemble(0), inst->machInst); + delete inst; + return nop; + } +}}; + +output exec {{ + Fault * + Nop::execute(%(CPU_exec_context)s *, Trace::InstRecord *) const + { + return NoFault; + } +}}; + +// integer & FP operate instructions use Rc as dest, so check for +// Rc == 31 to detect nops +def template OperateNopCheckDecode {{ + { + AlphaStaticInst *i = new %(class_name)s(machInst); + if (RC == 31) { + i = makeNop(i); + } + return i; + } +}}; + +// Like BasicOperate format, but generates NOP if RC/FC == 31 +def format BasicOperateWithNopCheck(code, *opt_args) {{ + iop = InstObjParams(name, Name, 'AlphaStaticInst', CodeBlock(code), + opt_args) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + decode_block = OperateNopCheckDecode.subst(iop) + exec_output = BasicExecute.subst(iop) +}}; + +// Integer instruction templates, formats, etc. +##include "m5/arch/alpha/isa/int.isa" + +// Floating-point instruction templates, formats, etc. +##include "m5/arch/alpha/isa/fp.isa" + +// Memory instruction templates, formats, etc. +##include "m5/arch/alpha/isa/mem.isa" + +// Branch/jump instruction templates, formats, etc. +##include "m5/arch/alpha/isa/branch.isa" + +// PAL instruction templates, formats, etc. +##include "m5/arch/alpha/isa/pal.isa" + +// Unimplemented instruction templates, formats, etc. +##include "m5/arch/alpha/isa/unimp.isa" + +// Unknown instruction templates, formats, etc. +##include "m5/arch/alpha/isa/unknown.isa" + +// Execution utility functions +##include "m5/arch/alpha/isa/util.isa" + +// The actual decoder +##include "m5/arch/alpha/isa/decoder.isa" diff --git a/arch/alpha/isa/mem.isa b/arch/alpha/isa/mem.isa new file mode 100644 index 000000000..df26a534d --- /dev/null +++ b/arch/alpha/isa/mem.isa @@ -0,0 +1,702 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output header {{ + /** + * Base class for general Alpha memory-format instructions. + */ + class Memory : public AlphaStaticInst + { + protected: + + /// Memory request flags. See mem_req_base.hh. + unsigned memAccessFlags; + /// Pointer to EAComp object. + const StaticInstPtr<AlphaISA> eaCompPtr; + /// Pointer to MemAcc object. + const StaticInstPtr<AlphaISA> memAccPtr; + + /// Constructor + Memory(const char *mnem, MachInst _machInst, OpClass __opClass, + StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, + StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr) + : AlphaStaticInst(mnem, _machInst, __opClass), + memAccessFlags(0), eaCompPtr(_eaCompPtr), memAccPtr(_memAccPtr) + { + } + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + + public: + + const StaticInstPtr<AlphaISA> &eaCompInst() const { return eaCompPtr; } + const StaticInstPtr<AlphaISA> &memAccInst() const { return memAccPtr; } + }; + + /** + * Base class for memory-format instructions using a 32-bit + * displacement (i.e. most of them). + */ + class MemoryDisp32 : public Memory + { + protected: + /// Displacement for EA calculation (signed). + int32_t disp; + + /// Constructor. + MemoryDisp32(const char *mnem, MachInst _machInst, OpClass __opClass, + StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, + StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr) + : Memory(mnem, _machInst, __opClass, _eaCompPtr, _memAccPtr), + disp(MEMDISP) + { + } + }; + + + /** + * Base class for a few miscellaneous memory-format insts + * that don't interpret the disp field: wh64, fetch, fetch_m, ecb. + * None of these instructions has a destination register either. + */ + class MemoryNoDisp : public Memory + { + protected: + /// Constructor + MemoryNoDisp(const char *mnem, MachInst _machInst, OpClass __opClass, + StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, + StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr) + : Memory(mnem, _machInst, __opClass, _eaCompPtr, _memAccPtr) + { + } + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + + +output decoder {{ + std::string + Memory::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + return csprintf("%-10s %c%d,%d(r%d)", mnemonic, + flags[IsFloating] ? 'f' : 'r', RA, MEMDISP, RB); + } + + std::string + MemoryNoDisp::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + return csprintf("%-10s (r%d)", mnemonic, RB); + } +}}; + +def format LoadAddress(code) {{ + iop = InstObjParams(name, Name, 'MemoryDisp32', CodeBlock(code)) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + decode_block = BasicDecode.subst(iop) + exec_output = BasicExecute.subst(iop) +}}; + + +def template LoadStoreDeclare {{ + /** + * Static instruction class for "%(mnemonic)s". + */ + class %(class_name)s : public %(base_class)s + { + protected: + + /** + * "Fake" effective address computation class for "%(mnemonic)s". + */ + class EAComp : public %(base_class)s + { + public: + /// Constructor + EAComp(MachInst machInst); + + %(BasicExecDeclare)s + }; + + /** + * "Fake" memory access instruction class for "%(mnemonic)s". + */ + class MemAcc : public %(base_class)s + { + public: + /// Constructor + MemAcc(MachInst machInst); + + %(BasicExecDeclare)s + }; + + public: + + /// Constructor. + %(class_name)s(MachInst machInst); + + %(BasicExecDeclare)s + + %(InitiateAccDeclare)s + + %(CompleteAccDeclare)s + }; +}}; + + +def template InitiateAccDeclare {{ + Fault * initiateAcc(%(CPU_exec_context)s *, Trace::InstRecord *) const; +}}; + + +def template CompleteAccDeclare {{ + Fault * completeAcc(uint8_t *, %(CPU_exec_context)s *, Trace::InstRecord *) const; +}}; + + +def template LoadStoreConstructor {{ + /** TODO: change op_class to AddrGenOp or something (requires + * creating new member of OpClass enum in op_class.hh, updating + * config files, etc.). */ + inline %(class_name)s::EAComp::EAComp(MachInst machInst) + : %(base_class)s("%(mnemonic)s (EAComp)", machInst, IntAluOp) + { + %(ea_constructor)s; + } + + inline %(class_name)s::MemAcc::MemAcc(MachInst machInst) + : %(base_class)s("%(mnemonic)s (MemAcc)", machInst, %(op_class)s) + { + %(memacc_constructor)s; + } + + inline %(class_name)s::%(class_name)s(MachInst machInst) + : %(base_class)s("%(mnemonic)s", machInst, %(op_class)s, + new EAComp(machInst), new MemAcc(machInst)) + { + %(constructor)s; + } +}}; + + +def template EACompExecute {{ + Fault * + %(class_name)s::EAComp::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + %(code)s; + + if (fault == NoFault) { + %(op_wb)s; + xc->setEA(EA); + } + + return fault; + } +}}; + +def template LoadMemAccExecute {{ + Fault * + %(class_name)s::MemAcc::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + EA = xc->getEA(); + + if (fault == NoFault) { + fault = xc->read(EA, (uint%(mem_acc_size)d_t&)Mem, memAccessFlags); + %(code)s; + } + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + + +def template LoadExecute {{ + Fault * %(class_name)s::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + %(ea_code)s; + + if (fault == NoFault) { + fault = xc->read(EA, (uint%(mem_acc_size)d_t&)Mem, memAccessFlags); + %(memacc_code)s; + } + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + + +def template LoadInitiateAcc {{ + Fault * %(class_name)s::initiateAcc(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_src_decl)s; + %(op_rd)s; + %(ea_code)s; + + if (fault == NoFault) { + fault = xc->read(EA, (uint%(mem_acc_size)d_t &)Mem, memAccessFlags); + } + + return fault; + } +}}; + + +def template LoadCompleteAcc {{ + Fault * %(class_name)s::completeAcc(uint8_t *data, + %(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_src_decl)s; + %(op_dest_decl)s; + + memcpy(&Mem, data, sizeof(Mem)); + + if (fault == NoFault) { + %(memacc_code)s; + } + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + + +def template StoreMemAccExecute {{ + Fault * + %(class_name)s::MemAcc::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + uint64_t write_result = 0; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + EA = xc->getEA(); + + if (fault == NoFault) { + %(code)s; + } + + if (fault == NoFault) { + fault = xc->write((uint%(mem_acc_size)d_t&)Mem, EA, + memAccessFlags, &write_result); + if (traceData) { traceData->setData(Mem); } + } + + if (fault == NoFault) { + %(postacc_code)s; + } + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + + +def template StoreExecute {{ + Fault * %(class_name)s::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + uint64_t write_result = 0; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + %(ea_code)s; + + if (fault == NoFault) { + %(memacc_code)s; + } + + if (fault == NoFault) { + fault = xc->write((uint%(mem_acc_size)d_t&)Mem, EA, + memAccessFlags, &write_result); + if (traceData) { traceData->setData(Mem); } + } + + if (fault == NoFault) { + %(postacc_code)s; + } + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + +def template StoreInitiateAcc {{ + Fault * %(class_name)s::initiateAcc(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + uint64_t write_result = 0; + + %(fp_enable_check)s; + %(op_src_decl)s; + %(op_dest_decl)s; + %(op_rd)s; + %(ea_code)s; + + if (fault == NoFault) { + %(memacc_code)s; + } + + if (fault == NoFault) { + fault = xc->write((uint%(mem_acc_size)d_t&)Mem, EA, + memAccessFlags, &write_result); + if (traceData) { traceData->setData(Mem); } + } + + return fault; + } +}}; + + +def template StoreCompleteAcc {{ + Fault * %(class_name)s::completeAcc(uint8_t *data, + %(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Fault * fault = NoFault; + uint64_t write_result = 0; + + %(fp_enable_check)s; + %(op_dest_decl)s; + + memcpy(&write_result, data, sizeof(write_result)); + + if (fault == NoFault) { + %(postacc_code)s; + } + + if (fault == NoFault) { + %(op_wb)s; + } + + return fault; + } +}}; + + +def template MiscMemAccExecute {{ + Fault * %(class_name)s::MemAcc::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + EA = xc->getEA(); + + if (fault == NoFault) { + %(code)s; + } + + return NoFault; + } +}}; + +def template MiscExecute {{ + Fault * %(class_name)s::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + Addr EA; + Fault * fault = NoFault; + + %(fp_enable_check)s; + %(op_decl)s; + %(op_rd)s; + %(ea_code)s; + + if (fault == NoFault) { + %(memacc_code)s; + } + + return NoFault; + } +}}; + +def template MiscInitiateAcc {{ + Fault * %(class_name)s::initiateAcc(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + panic("Misc instruction does not support split access method!"); + return NoFault; + } +}}; + + +def template MiscCompleteAcc {{ + Fault * %(class_name)s::completeAcc(uint8_t *data, + %(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + panic("Misc instruction does not support split access method!"); + + return NoFault; + } +}}; + +// load instructions use Ra as dest, so check for +// Ra == 31 to detect nops +def template LoadNopCheckDecode {{ + { + AlphaStaticInst *i = new %(class_name)s(machInst); + if (RA == 31) { + i = makeNop(i); + } + return i; + } +}}; + + +// for some load instructions, Ra == 31 indicates a prefetch (not a nop) +def template LoadPrefetchCheckDecode {{ + { + if (RA != 31) { + return new %(class_name)s(machInst); + } + else { + return new %(class_name)sPrefetch(machInst); + } + } +}}; + + +let {{ +def LoadStoreBase(name, Name, ea_code, memacc_code, mem_flags, inst_flags, + postacc_code = '', base_class = 'MemoryDisp32', + decode_template = BasicDecode, exec_template_base = ''): + # Make sure flags are in lists (convert to lists if not). + mem_flags = makeList(mem_flags) + inst_flags = makeList(inst_flags) + + # add hook to get effective addresses into execution trace output. + ea_code += '\nif (traceData) { traceData->setAddr(EA); }\n' + + # generate code block objects + ea_cblk = CodeBlock(ea_code) + memacc_cblk = CodeBlock(memacc_code) + postacc_cblk = CodeBlock(postacc_code) + + # Some CPU models execute the memory operation as an atomic unit, + # while others want to separate them into an effective address + # computation and a memory access operation. As a result, we need + # to generate three StaticInst objects. Note that the latter two + # are nested inside the larger "atomic" one. + + # generate InstObjParams for EAComp object + ea_iop = InstObjParams(name, Name, base_class, ea_cblk, inst_flags) + + # generate InstObjParams for MemAcc object + memacc_iop = InstObjParams(name, Name, base_class, memacc_cblk, inst_flags) + # in the split execution model, the MemAcc portion is responsible + # for the post-access code. + memacc_iop.postacc_code = postacc_cblk.code + + # generate InstObjParams for InitiateAcc, CompleteAcc object + # The code used depends on the template being used + if (exec_template_base == 'Load'): + initiateacc_cblk = CodeBlock(ea_code + memacc_code) + completeacc_cblk = CodeBlock(memacc_code + postacc_code) + elif (exec_template_base == 'Store'): + initiateacc_cblk = CodeBlock(ea_code + memacc_code) + completeacc_cblk = CodeBlock(postacc_code) + else: + initiateacc_cblk = '' + completeacc_cblk = '' + + initiateacc_iop = InstObjParams(name, Name, base_class, initiateacc_cblk, + inst_flags) + + completeacc_iop = InstObjParams(name, Name, base_class, completeacc_cblk, + inst_flags) + + if (exec_template_base == 'Load'): + initiateacc_iop.ea_code = ea_cblk.code + initiateacc_iop.memacc_code = memacc_cblk.code + completeacc_iop.memacc_code = memacc_cblk.code + completeacc_iop.postacc_code = postacc_cblk.code + elif (exec_template_base == 'Store'): + initiateacc_iop.ea_code = ea_cblk.code + initiateacc_iop.memacc_code = memacc_cblk.code + completeacc_iop.postacc_code = postacc_cblk.code + + # generate InstObjParams for unified execution + cblk = CodeBlock(ea_code + memacc_code + postacc_code) + iop = InstObjParams(name, Name, base_class, cblk, inst_flags) + + iop.ea_constructor = ea_cblk.constructor + iop.ea_code = ea_cblk.code + iop.memacc_constructor = memacc_cblk.constructor + iop.memacc_code = memacc_cblk.code + iop.postacc_code = postacc_cblk.code + + if mem_flags: + s = '\n\tmemAccessFlags = ' + string.join(mem_flags, '|') + ';' + iop.constructor += s + memacc_iop.constructor += s + + # select templates + memAccExecTemplate = eval(exec_template_base + 'MemAccExecute') + fullExecTemplate = eval(exec_template_base + 'Execute') + initiateAccTemplate = eval(exec_template_base + 'InitiateAcc') + completeAccTemplate = eval(exec_template_base + 'CompleteAcc') + + # (header_output, decoder_output, decode_block, exec_output) + return (LoadStoreDeclare.subst(iop), LoadStoreConstructor.subst(iop), + decode_template.subst(iop), + EACompExecute.subst(ea_iop) + + memAccExecTemplate.subst(memacc_iop) + + fullExecTemplate.subst(iop) + + initiateAccTemplate.subst(initiateacc_iop) + + completeAccTemplate.subst(completeacc_iop)) +}}; + + +def format LoadOrNop(memacc_code, ea_code = {{ EA = Rb + disp; }}, + mem_flags = [], inst_flags = []) {{ + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name, ea_code, memacc_code, mem_flags, inst_flags, + decode_template = LoadNopCheckDecode, + exec_template_base = 'Load') +}}; + + +// Note that the flags passed in apply only to the prefetch version +def format LoadOrPrefetch(memacc_code, ea_code = {{ EA = Rb + disp; }}, + mem_flags = [], pf_flags = [], inst_flags = []) {{ + # declare the load instruction object and generate the decode block + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name, ea_code, memacc_code, mem_flags, inst_flags, + decode_template = LoadPrefetchCheckDecode, + exec_template_base = 'Load') + + # Declare the prefetch instruction object. + + # Make sure flag args are lists so we can mess with them. + mem_flags = makeList(mem_flags) + pf_flags = makeList(pf_flags) + inst_flags = makeList(inst_flags) + + pf_mem_flags = mem_flags + pf_flags + ['NO_FAULT'] + pf_inst_flags = inst_flags + ['IsMemRef', 'IsLoad', + 'IsDataPrefetch', 'MemReadOp'] + + (pf_header_output, pf_decoder_output, _, pf_exec_output) = \ + LoadStoreBase(name, Name + 'Prefetch', ea_code, + 'xc->prefetch(EA, memAccessFlags);', + pf_mem_flags, pf_inst_flags, exec_template_base = 'Misc') + + header_output += pf_header_output + decoder_output += pf_decoder_output + exec_output += pf_exec_output +}}; + + +def format Store(memacc_code, ea_code = {{ EA = Rb + disp; }}, + mem_flags = [], inst_flags = []) {{ + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name, ea_code, memacc_code, mem_flags, inst_flags, + exec_template_base = 'Store') +}}; + + +def format StoreCond(memacc_code, postacc_code, + ea_code = {{ EA = Rb + disp; }}, + mem_flags = [], inst_flags = []) {{ + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name, ea_code, memacc_code, mem_flags, inst_flags, + postacc_code, exec_template_base = 'Store') +}}; + + +// Use 'MemoryNoDisp' as base: for wh64, fetch, ecb +def format MiscPrefetch(ea_code, memacc_code, + mem_flags = [], inst_flags = []) {{ + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name, ea_code, memacc_code, mem_flags, inst_flags, + base_class = 'MemoryNoDisp', exec_template_base = 'Misc') +}}; + + diff --git a/arch/alpha/isa/pal.isa b/arch/alpha/isa/pal.isa new file mode 100644 index 000000000..b68a7c19f --- /dev/null +++ b/arch/alpha/isa/pal.isa @@ -0,0 +1,271 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output header {{ + /** + * Base class for emulated call_pal calls (used only in + * non-full-system mode). + */ + class EmulatedCallPal : public AlphaStaticInst + { + protected: + + /// Constructor. + EmulatedCallPal(const char *mnem, MachInst _machInst, + OpClass __opClass) + : AlphaStaticInst(mnem, _machInst, __opClass) + { + } + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + +output decoder {{ + std::string + EmulatedCallPal::generateDisassembly(Addr pc, + const SymbolTable *symtab) const + { +#ifdef SS_COMPATIBLE_DISASSEMBLY + return csprintf("%s %s", "call_pal", mnemonic); +#else + return csprintf("%-10s %s", "call_pal", mnemonic); +#endif + } +}}; + +def format EmulatedCallPal(code, *flags) {{ + iop = InstObjParams(name, Name, 'EmulatedCallPal', CodeBlock(code), flags) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + decode_block = BasicDecode.subst(iop) + exec_output = BasicExecute.subst(iop) +}}; + +output header {{ + /** + * Base class for full-system-mode call_pal instructions. + * Probably could turn this into a leaf class and get rid of the + * parser template. + */ + class CallPalBase : public AlphaStaticInst + { + protected: + int palFunc; ///< Function code part of instruction + int palOffset; ///< Target PC, offset from IPR_PAL_BASE + bool palValid; ///< is the function code valid? + bool palPriv; ///< is this call privileged? + + /// Constructor. + CallPalBase(const char *mnem, MachInst _machInst, + OpClass __opClass); + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + +output decoder {{ + inline + CallPalBase::CallPalBase(const char *mnem, MachInst _machInst, + OpClass __opClass) + : AlphaStaticInst(mnem, _machInst, __opClass), + palFunc(PALFUNC) + { + // From the 21164 HRM (paraphrased): + // Bit 7 of the function code (mask 0x80) indicates + // whether the call is privileged (bit 7 == 0) or + // unprivileged (bit 7 == 1). The privileged call table + // starts at 0x2000, the unprivielged call table starts at + // 0x3000. Bits 5-0 (mask 0x3f) are used to calculate the + // offset. + const int palPrivMask = 0x80; + const int palOffsetMask = 0x3f; + + // Pal call is invalid unless all other bits are 0 + palValid = ((machInst & ~(palPrivMask | palOffsetMask)) == 0); + palPriv = ((machInst & palPrivMask) == 0); + int shortPalFunc = (machInst & palOffsetMask); + // Add 1 to base to set pal-mode bit + palOffset = (palPriv ? 0x2001 : 0x3001) + (shortPalFunc << 6); + } + + std::string + CallPalBase::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + return csprintf("%-10s %#x", "call_pal", palFunc); + } +}}; + +def format CallPal(code, *flags) {{ + iop = InstObjParams(name, Name, 'CallPalBase', CodeBlock(code), flags) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + decode_block = BasicDecode.subst(iop) + exec_output = BasicExecute.subst(iop) +}}; + +//////////////////////////////////////////////////////////////////// +// +// hw_ld, hw_st +// + +output header {{ + /** + * Base class for hw_ld and hw_st. + */ + class HwLoadStore : public Memory + { + protected: + + /// Displacement for EA calculation (signed). + int16_t disp; + + /// Constructor + HwLoadStore(const char *mnem, MachInst _machInst, OpClass __opClass, + StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, + StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr); + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + + +output decoder {{ + inline + HwLoadStore::HwLoadStore(const char *mnem, MachInst _machInst, + OpClass __opClass, + StaticInstPtr<AlphaISA> _eaCompPtr, + StaticInstPtr<AlphaISA> _memAccPtr) + : Memory(mnem, _machInst, __opClass, _eaCompPtr, _memAccPtr), + disp(HW_LDST_DISP) + { + memAccessFlags = 0; + if (HW_LDST_PHYS) memAccessFlags |= PHYSICAL; + if (HW_LDST_ALT) memAccessFlags |= ALTMODE; + if (HW_LDST_VPTE) memAccessFlags |= VPTE; + if (HW_LDST_LOCK) memAccessFlags |= LOCKED; + } + + std::string + HwLoadStore::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { +#ifdef SS_COMPATIBLE_DISASSEMBLY + return csprintf("%-10s r%d,%d(r%d)", mnemonic, RA, disp, RB); +#else + // HW_LDST_LOCK and HW_LDST_COND are the same bit. + const char *lock_str = + (HW_LDST_LOCK) ? (flags[IsLoad] ? ",LOCK" : ",COND") : ""; + + return csprintf("%-10s r%d,%d(r%d)%s%s%s%s%s", + mnemonic, RA, disp, RB, + HW_LDST_PHYS ? ",PHYS" : "", + HW_LDST_ALT ? ",ALT" : "", + HW_LDST_QUAD ? ",QUAD" : "", + HW_LDST_VPTE ? ",VPTE" : "", + lock_str); +#endif + } +}}; + +def format HwLoad(ea_code, memacc_code, class_ext, *flags) {{ + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name + class_ext, ea_code, memacc_code, + mem_flags = [], inst_flags = flags, + base_class = 'HwLoadStore', exec_template_base = 'Load') +}}; + + +def format HwStore(ea_code, memacc_code, class_ext, *flags) {{ + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name + class_ext, ea_code, memacc_code, + mem_flags = [], inst_flags = flags, + base_class = 'HwLoadStore', exec_template_base = 'Store') +}}; + + +def format HwStoreCond(ea_code, memacc_code, postacc_code, class_ext, + *flags) {{ + (header_output, decoder_output, decode_block, exec_output) = \ + LoadStoreBase(name, Name + class_ext, ea_code, memacc_code, + postacc_code, mem_flags = [], inst_flags = flags, + base_class = 'HwLoadStore') +}}; + + +output header {{ + /** + * Base class for hw_mfpr and hw_mtpr. + */ + class HwMoveIPR : public AlphaStaticInst + { + protected: + /// Index of internal processor register. + int ipr_index; + + /// Constructor + HwMoveIPR(const char *mnem, MachInst _machInst, OpClass __opClass) + : AlphaStaticInst(mnem, _machInst, __opClass), + ipr_index(HW_IPR_IDX) + { + } + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + +output decoder {{ + std::string + HwMoveIPR::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + if (_numSrcRegs > 0) { + // must be mtpr + return csprintf("%-10s r%d,IPR(%#x)", + mnemonic, RA, ipr_index); + } + else { + // must be mfpr + return csprintf("%-10s IPR(%#x),r%d", + mnemonic, ipr_index, RA); + } + } +}}; + +def format HwMoveIPR(code) {{ + iop = InstObjParams(name, Name, 'HwMoveIPR', CodeBlock(code), + ['IprAccessOp']) + header_output = BasicDeclare.subst(iop) + decoder_output = BasicConstructor.subst(iop) + decode_block = BasicDecode.subst(iop) + exec_output = BasicExecute.subst(iop) +}}; + + diff --git a/arch/alpha/isa/unimp.isa b/arch/alpha/isa/unimp.isa new file mode 100644 index 000000000..ce8197708 --- /dev/null +++ b/arch/alpha/isa/unimp.isa @@ -0,0 +1,165 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output header {{ + /** + * Static instruction class for unimplemented instructions that + * cause simulator termination. Note that these are recognized + * (legal) instructions that the simulator does not support; the + * 'Unknown' class is used for unrecognized/illegal instructions. + * This is a leaf class. + */ + class FailUnimplemented : public AlphaStaticInst + { + public: + /// Constructor + FailUnimplemented(const char *_mnemonic, MachInst _machInst) + : AlphaStaticInst(_mnemonic, _machInst, No_OpClass) + { + // don't call execute() (which panics) if we're on a + // speculative path + flags[IsNonSpeculative] = true; + } + + %(BasicExecDeclare)s + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; + + /** + * Base class for unimplemented instructions that cause a warning + * to be printed (but do not terminate simulation). This + * implementation is a little screwy in that it will print a + * warning for each instance of a particular unimplemented machine + * instruction, not just for each unimplemented opcode. Should + * probably make the 'warned' flag a static member of the derived + * class. + */ + class WarnUnimplemented : public AlphaStaticInst + { + private: + /// Have we warned on this instruction yet? + mutable bool warned; + + public: + /// Constructor + WarnUnimplemented(const char *_mnemonic, MachInst _machInst) + : AlphaStaticInst(_mnemonic, _machInst, No_OpClass), warned(false) + { + // don't call execute() (which panics) if we're on a + // speculative path + flags[IsNonSpeculative] = true; + } + + %(BasicExecDeclare)s + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + +output decoder {{ + std::string + FailUnimplemented::generateDisassembly(Addr pc, + const SymbolTable *symtab) const + { + return csprintf("%-10s (unimplemented)", mnemonic); + } + + std::string + WarnUnimplemented::generateDisassembly(Addr pc, + const SymbolTable *symtab) const + { +#ifdef SS_COMPATIBLE_DISASSEMBLY + return csprintf("%-10s", mnemonic); +#else + return csprintf("%-10s (unimplemented)", mnemonic); +#endif + } +}}; + +output exec {{ + Fault * + FailUnimplemented::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + panic("attempt to execute unimplemented instruction '%s' " + "(inst 0x%08x, opcode 0x%x)", mnemonic, machInst, OPCODE); + return UnimplementedOpcodeFault; + } + + Fault * + WarnUnimplemented::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + if (!warned) { + warn("instruction '%s' unimplemented\n", mnemonic); + warned = true; + } + + return NoFault; + } +}}; + + +def format FailUnimpl() {{ + iop = InstObjParams(name, 'FailUnimplemented') + decode_block = BasicDecodeWithMnemonic.subst(iop) +}}; + +def format WarnUnimpl() {{ + iop = InstObjParams(name, 'WarnUnimplemented') + decode_block = BasicDecodeWithMnemonic.subst(iop) +}}; + +output header {{ + /** + * Static instruction class for unknown (illegal) instructions. + * These cause simulator termination if they are executed in a + * non-speculative mode. This is a leaf class. + */ + class Unknown : public AlphaStaticInst + { + public: + /// Constructor + Unknown(MachInst _machInst) + : AlphaStaticInst("unknown", _machInst, No_OpClass) + { + // don't call execute() (which panics) if we're on a + // speculative path + flags[IsNonSpeculative] = true; + } + + %(BasicExecDeclare)s + + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const; + }; +}}; + diff --git a/arch/alpha/isa/unknown.isa b/arch/alpha/isa/unknown.isa new file mode 100644 index 000000000..e7f8bc8db --- /dev/null +++ b/arch/alpha/isa/unknown.isa @@ -0,0 +1,52 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output decoder {{ + std::string + Unknown::generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + return csprintf("%-10s (inst 0x%x, opcode 0x%x)", + "unknown", machInst, OPCODE); + } +}}; + +output exec {{ + Fault * + Unknown::execute(%(CPU_exec_context)s *xc, + Trace::InstRecord *traceData) const + { + panic("attempt to execute unknown instruction " + "(inst 0x%08x, opcode 0x%x)", machInst, OPCODE); + return UnimplementedOpcodeFault; + } +}}; + +def format Unknown() {{ + decode_block = 'return new Unknown(machInst);\n' +}}; + diff --git a/arch/alpha/isa/util.isa b/arch/alpha/isa/util.isa new file mode 100644 index 000000000..9fbbf6636 --- /dev/null +++ b/arch/alpha/isa/util.isa @@ -0,0 +1,112 @@ +// -*- mode:c++ -*- + +// Copyright (c) 2003-2005 The Regents of The University of Michigan +// All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer; +// redistributions in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution; +// neither the name of the copyright holders nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +output exec {{ + + /// Return opa + opb, summing carry into third arg. + inline uint64_t + addc(uint64_t opa, uint64_t opb, int &carry) + { + uint64_t res = opa + opb; + if (res < opa || res < opb) + ++carry; + return res; + } + + /// Multiply two 64-bit values (opa * opb), returning the 128-bit + /// product in res_hi and res_lo. + inline void + mul128(uint64_t opa, uint64_t opb, uint64_t &res_hi, uint64_t &res_lo) + { + // do a 64x64 --> 128 multiply using four 32x32 --> 64 multiplies + uint64_t opa_hi = opa<63:32>; + uint64_t opa_lo = opa<31:0>; + uint64_t opb_hi = opb<63:32>; + uint64_t opb_lo = opb<31:0>; + + res_lo = opa_lo * opb_lo; + + // The middle partial products logically belong in bit + // positions 95 to 32. Thus the lower 32 bits of each product + // sum into the upper 32 bits of the low result, while the + // upper 32 sum into the low 32 bits of the upper result. + uint64_t partial1 = opa_hi * opb_lo; + uint64_t partial2 = opa_lo * opb_hi; + + uint64_t partial1_lo = partial1<31:0> << 32; + uint64_t partial1_hi = partial1<63:32>; + uint64_t partial2_lo = partial2<31:0> << 32; + uint64_t partial2_hi = partial2<63:32>; + + // Add partial1_lo and partial2_lo to res_lo, keeping track + // of any carries out + int carry_out = 0; + res_lo = addc(partial1_lo, res_lo, carry_out); + res_lo = addc(partial2_lo, res_lo, carry_out); + + // Now calculate the high 64 bits... + res_hi = (opa_hi * opb_hi) + partial1_hi + partial2_hi + carry_out; + } + + /// Map 8-bit S-floating exponent to 11-bit T-floating exponent. + /// See Table 2-2 of Alpha AHB. + inline int + map_s(int old_exp) + { + int hibit = old_exp<7:>; + int lobits = old_exp<6:0>; + + if (hibit == 1) { + return (lobits == 0x7f) ? 0x7ff : (0x400 | lobits); + } + else { + return (lobits == 0) ? 0 : (0x380 | lobits); + } + } + + /// Convert a 32-bit S-floating value to the equivalent 64-bit + /// representation to be stored in an FP reg. + inline uint64_t + s_to_t(uint32_t s_val) + { + uint64_t tmp = s_val; + return (tmp<31:> << 63 // sign bit + | (uint64_t)map_s(tmp<30:23>) << 52 // exponent + | tmp<22:0> << 29); // fraction + } + + /// Convert a 64-bit T-floating value to the equivalent 32-bit + /// S-floating representation to be stored in memory. + inline int32_t + t_to_s(uint64_t t_val) + { + return (t_val<63:62> << 30 // sign bit & hi exp bit + | t_val<58:29>); // rest of exp & fraction + } +}}; + diff --git a/arch/alpha/isa_desc b/arch/alpha/isa_desc deleted file mode 100644 index c998b1a0a..000000000 --- a/arch/alpha/isa_desc +++ /dev/null @@ -1,2737 +0,0 @@ -// -*- mode:c++ -*- - -//Copyright (c) 2003, 2004, 2005 -//The Regents of The University of Michigan -//All Rights Reserved - -//This code is part of the M5 simulator, developed by Nathan Binkert, -//Erik Hallnor, Steve Raasch, and Steve Reinhardt, with contributions -//from Ron Dreslinski, Dave Greene, Lisa Hsu, Kevin Lim, Ali Saidi, -//and Andrew Schultz. - -//Permission is granted to use, copy, create derivative works and -//redistribute this software and such derivative works for any purpose, -//so long as the copyright notice above, this grant of permission, and -//the disclaimer below appear in all copies made; and so long as the -//name of The University of Michigan is not used in any advertising or -//publicity pertaining to the use or distribution of this software -//without specific, written prior authorization. - -//THIS SOFTWARE IS PROVIDED AS IS, WITHOUT REPRESENTATION FROM THE -//UNIVERSITY OF MICHIGAN AS TO ITS FITNESS FOR ANY PURPOSE, AND WITHOUT -//WARRANTY BY THE UNIVERSITY OF MICHIGAN OF ANY KIND, EITHER EXPRESS OR -//IMPLIED, INCLUDING WITHOUT LIMITATION THE IMPLIED WARRANTIES OF -//MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE REGENTS OF -//THE UNIVERSITY OF MICHIGAN SHALL NOT BE LIABLE FOR ANY DAMAGES, -//INCLUDING DIRECT, SPECIAL, INDIRECT, INCIDENTAL, OR CONSEQUENTIAL -//DAMAGES, WITH RESPECT TO ANY CLAIM ARISING OUT OF OR IN CONNECTION -//WITH THE USE OF THE SOFTWARE, EVEN IF IT HAS BEEN OR IS HEREAFTER -//ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. - -//////////////////////////////////////////////////////////////////// -// -// Alpha ISA description file. -// -//////////////////////////////////////////////////////////////////// - - -//////////////////////////////////////////////////////////////////// -// -// Output include file directives. -// - -output header {{ -#include <sstream> -#include <iostream> -#include <iomanip> - -#include "config/ss_compatible_fp.hh" -#include "cpu/static_inst.hh" -#include "mem/mem_req.hh" // some constructors use MemReq flags -}}; - -output decoder {{ -#include "base/cprintf.hh" -#include "base/fenv.hh" -#include "base/loader/symtab.hh" -#include "config/ss_compatible_fp.hh" -#include "cpu/exec_context.hh" // for Jump::branchTarget() - -#include <math.h> -}}; - -output exec {{ -#include <math.h> - -#if FULL_SYSTEM -#include "arch/alpha/pseudo_inst.hh" -#endif -#include "base/fenv.hh" -#include "config/ss_compatible_fp.hh" -#include "cpu/base.hh" -#include "cpu/exetrace.hh" -#include "sim/sim_exit.hh" -}}; - -//////////////////////////////////////////////////////////////////// -// -// Namespace statement. Everything below this line will be in the -// AlphaISAInst namespace. -// - - -namespace AlphaISA; - -//////////////////////////////////////////////////////////////////// -// -// Bitfield definitions. -// - -// Universal (format-independent) fields -def bitfield OPCODE <31:26>; -def bitfield RA <25:21>; -def bitfield RB <20:16>; - -// Memory format -def signed bitfield MEMDISP <15: 0>; // displacement -def bitfield MEMFUNC <15: 0>; // function code (same field, unsigned) - -// Memory-format jumps -def bitfield JMPFUNC <15:14>; // function code (disp<15:14>) -def bitfield JMPHINT <13: 0>; // tgt Icache idx hint (disp<13:0>) - -// Branch format -def signed bitfield BRDISP <20: 0>; // displacement - -// Integer operate format(s>; -def bitfield INTIMM <20:13>; // integer immediate (literal) -def bitfield IMM <12:12>; // immediate flag -def bitfield INTFUNC <11: 5>; // function code -def bitfield RC < 4: 0>; // dest reg - -// Floating-point operate format -def bitfield FA <25:21>; -def bitfield FB <20:16>; -def bitfield FP_FULLFUNC <15: 5>; // complete function code - def bitfield FP_TRAPMODE <15:13>; // trapping mode - def bitfield FP_ROUNDMODE <12:11>; // rounding mode - def bitfield FP_TYPEFUNC <10: 5>; // type+func: handiest for decoding - def bitfield FP_SRCTYPE <10: 9>; // source reg type - def bitfield FP_SHORTFUNC < 8: 5>; // short function code - def bitfield FP_SHORTFUNC_TOP2 <8:7>; // top 2 bits of short func code -def bitfield FC < 4: 0>; // dest reg - -// PALcode format -def bitfield PALFUNC <25: 0>; // function code - -// EV5 PAL instructions: -// HW_LD/HW_ST -def bitfield HW_LDST_PHYS <15>; // address is physical -def bitfield HW_LDST_ALT <14>; // use ALT_MODE IPR -def bitfield HW_LDST_WRTCK <13>; // HW_LD only: fault if no write acc -def bitfield HW_LDST_QUAD <12>; // size: 0=32b, 1=64b -def bitfield HW_LDST_VPTE <11>; // HW_LD only: is PTE fetch -def bitfield HW_LDST_LOCK <10>; // HW_LD only: is load locked -def bitfield HW_LDST_COND <10>; // HW_ST only: is store conditional -def signed bitfield HW_LDST_DISP <9:0>; // signed displacement - -// HW_REI -def bitfield HW_REI_TYP <15:14>; // type: stalling vs. non-stallingk -def bitfield HW_REI_MBZ <13: 0>; // must be zero - -// HW_MTPR/MW_MFPR -def bitfield HW_IPR_IDX <15:0>; // IPR index - -// M5 instructions -def bitfield M5FUNC <7:0>; - -def operand_types {{ - 'sb' : ('signed int', 8), - 'ub' : ('unsigned int', 8), - 'sw' : ('signed int', 16), - 'uw' : ('unsigned int', 16), - 'sl' : ('signed int', 32), - 'ul' : ('unsigned int', 32), - 'sq' : ('signed int', 64), - 'uq' : ('unsigned int', 64), - 'sf' : ('float', 32), - 'df' : ('float', 64) -}}; - -def operands {{ - # Int regs default to unsigned, but code should not count on this. - # For clarity, descriptions that depend on unsigned behavior should - # explicitly specify '.uq'. - 'Ra': IntRegOperandTraits('uq', 'RA', 'IsInteger', 1), - 'Rb': IntRegOperandTraits('uq', 'RB', 'IsInteger', 2), - 'Rc': IntRegOperandTraits('uq', 'RC', 'IsInteger', 3), - 'Fa': FloatRegOperandTraits('df', 'FA', 'IsFloating', 1), - 'Fb': FloatRegOperandTraits('df', 'FB', 'IsFloating', 2), - 'Fc': FloatRegOperandTraits('df', 'FC', 'IsFloating', 3), - 'Mem': MemOperandTraits('uq', None, - ('IsMemRef', 'IsLoad', 'IsStore'), 4), - 'NPC': NPCOperandTraits('uq', None, ( None, None, 'IsControl' ), 4), - 'Runiq': ControlRegOperandTraits('uq', 'Uniq', None, 1), - 'FPCR': ControlRegOperandTraits('uq', 'Fpcr', None, 1), - # The next two are hacks for non-full-system call-pal emulation - 'R0': IntRegOperandTraits('uq', '0', None, 1), - 'R16': IntRegOperandTraits('uq', '16', None, 1) -}}; - -//////////////////////////////////////////////////////////////////// -// -// Basic instruction classes/templates/formats etc. -// - -output header {{ -// uncomment the following to get SimpleScalar-compatible disassembly -// (useful for diffing output traces). -// #define SS_COMPATIBLE_DISASSEMBLY - - /** - * Base class for all Alpha static instructions. - */ - class AlphaStaticInst : public StaticInst<AlphaISA> - { - protected: - - /// Make AlphaISA register dependence tags directly visible in - /// this class and derived classes. Maybe these should really - /// live here and not in the AlphaISA namespace. - enum DependenceTags { - FP_Base_DepTag = AlphaISA::FP_Base_DepTag, - Fpcr_DepTag = AlphaISA::Fpcr_DepTag, - Uniq_DepTag = AlphaISA::Uniq_DepTag, - IPR_Base_DepTag = AlphaISA::IPR_Base_DepTag - }; - - /// Constructor. - AlphaStaticInst(const char *mnem, MachInst _machInst, - OpClass __opClass) - : StaticInst<AlphaISA>(mnem, _machInst, __opClass) - { - } - - /// Print a register name for disassembly given the unique - /// dependence tag number (FP or int). - void printReg(std::ostream &os, int reg) const; - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -output decoder {{ - void - AlphaStaticInst::printReg(std::ostream &os, int reg) const - { - if (reg < FP_Base_DepTag) { - ccprintf(os, "r%d", reg); - } - else { - ccprintf(os, "f%d", reg - FP_Base_DepTag); - } - } - - std::string - AlphaStaticInst::generateDisassembly(Addr pc, - const SymbolTable *symtab) const - { - std::stringstream ss; - - ccprintf(ss, "%-10s ", mnemonic); - - // just print the first two source regs... if there's - // a third one, it's a read-modify-write dest (Rc), - // e.g. for CMOVxx - if (_numSrcRegs > 0) { - printReg(ss, _srcRegIdx[0]); - } - if (_numSrcRegs > 1) { - ss << ","; - printReg(ss, _srcRegIdx[1]); - } - - // just print the first dest... if there's a second one, - // it's generally implicit - if (_numDestRegs > 0) { - if (_numSrcRegs > 0) - ss << ","; - printReg(ss, _destRegIdx[0]); - } - - return ss.str(); - } -}}; - -// Declarations for execute() methods. -def template BasicExecDeclare {{ - Fault execute(%(CPU_exec_context)s *, Trace::InstRecord *) const; -}}; - -// Basic instruction class declaration template. -def template BasicDeclare {{ - /** - * Static instruction class for "%(mnemonic)s". - */ - class %(class_name)s : public %(base_class)s - { - public: - /// Constructor. - %(class_name)s(MachInst machInst); - - %(BasicExecDeclare)s - }; -}}; - -// Basic instruction class constructor template. -def template BasicConstructor {{ - inline %(class_name)s::%(class_name)s(MachInst machInst) - : %(base_class)s("%(mnemonic)s", machInst, %(op_class)s) - { - %(constructor)s; - } -}}; - -// Basic instruction class execute method template. -def template BasicExecute {{ - Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - Fault fault = No_Fault; - - %(fp_enable_check)s; - %(op_decl)s; - %(op_rd)s; - %(code)s; - - if (fault == No_Fault) { - %(op_wb)s; - } - - return fault; - } -}}; - -// Basic decode template. -def template BasicDecode {{ - return new %(class_name)s(machInst); -}}; - -// Basic decode template, passing mnemonic in as string arg to constructor. -def template BasicDecodeWithMnemonic {{ - return new %(class_name)s("%(mnemonic)s", machInst); -}}; - -// The most basic instruction format... used only for a few misc. insts -def format BasicOperate(code, *flags) {{ - iop = InstObjParams(name, Name, 'AlphaStaticInst', CodeBlock(code), flags) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - decode_block = BasicDecode.subst(iop) - exec_output = BasicExecute.subst(iop) -}}; - - - -//////////////////////////////////////////////////////////////////// -// -// Nop -// - -output header {{ - /** - * Static instruction class for no-ops. This is a leaf class. - */ - class Nop : public AlphaStaticInst - { - /// Disassembly of original instruction. - const std::string originalDisassembly; - - public: - /// Constructor - Nop(const std::string _originalDisassembly, MachInst _machInst) - : AlphaStaticInst("nop", _machInst, No_OpClass), - originalDisassembly(_originalDisassembly) - { - flags[IsNop] = true; - } - - ~Nop() { } - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - - %(BasicExecDeclare)s - }; -}}; - -output decoder {{ - std::string Nop::generateDisassembly(Addr pc, - const SymbolTable *symtab) const - { -#ifdef SS_COMPATIBLE_DISASSEMBLY - return originalDisassembly; -#else - return csprintf("%-10s (%s)", "nop", originalDisassembly); -#endif - } - - /// Helper function for decoding nops. Substitute Nop object - /// for original inst passed in as arg (and delete latter). - inline - AlphaStaticInst * - makeNop(AlphaStaticInst *inst) - { - AlphaStaticInst *nop = new Nop(inst->disassemble(0), inst->machInst); - delete inst; - return nop; - } -}}; - -output exec {{ - Fault - Nop::execute(%(CPU_exec_context)s *, Trace::InstRecord *) const - { - return No_Fault; - } -}}; - -// integer & FP operate instructions use Rc as dest, so check for -// Rc == 31 to detect nops -def template OperateNopCheckDecode {{ - { - AlphaStaticInst *i = new %(class_name)s(machInst); - if (RC == 31) { - i = makeNop(i); - } - return i; - } -}}; - -// Like BasicOperate format, but generates NOP if RC/FC == 31 -def format BasicOperateWithNopCheck(code, *opt_args) {{ - iop = InstObjParams(name, Name, 'AlphaStaticInst', CodeBlock(code), - opt_args) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - decode_block = OperateNopCheckDecode.subst(iop) - exec_output = BasicExecute.subst(iop) -}}; - - -//////////////////////////////////////////////////////////////////// -// -// Integer operate instructions -// - -output header {{ - /** - * Base class for integer immediate instructions. - */ - class IntegerImm : public AlphaStaticInst - { - protected: - /// Immediate operand value (unsigned 8-bit int). - uint8_t imm; - - /// Constructor - IntegerImm(const char *mnem, MachInst _machInst, OpClass __opClass) - : AlphaStaticInst(mnem, _machInst, __opClass), imm(INTIMM) - { - } - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -output decoder {{ - std::string - IntegerImm::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - std::stringstream ss; - - ccprintf(ss, "%-10s ", mnemonic); - - // just print the first source reg... if there's - // a second one, it's a read-modify-write dest (Rc), - // e.g. for CMOVxx - if (_numSrcRegs > 0) { - printReg(ss, _srcRegIdx[0]); - ss << ","; - } - - ss << (int)imm; - - if (_numDestRegs > 0) { - ss << ","; - printReg(ss, _destRegIdx[0]); - } - - return ss.str(); - } -}}; - - -def template RegOrImmDecode {{ - { - AlphaStaticInst *i = - (IMM) ? (AlphaStaticInst *)new %(class_name)sImm(machInst) - : (AlphaStaticInst *)new %(class_name)s(machInst); - if (RC == 31) { - i = makeNop(i); - } - return i; - } -}}; - -// Primary format for integer operate instructions: -// - Generates both reg-reg and reg-imm versions if Rb_or_imm is used. -// - Generates NOP if RC == 31. -def format IntegerOperate(code, *opt_flags) {{ - # If the code block contains 'Rb_or_imm', we define two instructions, - # one using 'Rb' and one using 'imm', and have the decoder select - # the right one. - uses_imm = (code.find('Rb_or_imm') != -1) - if uses_imm: - orig_code = code - # base code is reg version: - # rewrite by substituting 'Rb' for 'Rb_or_imm' - code = re.sub(r'Rb_or_imm', 'Rb', orig_code) - # generate immediate version by substituting 'imm' - # note that imm takes no extenstion, so we extend - # the regexp to replace any extension as well - imm_code = re.sub(r'Rb_or_imm(\.\w+)?', 'imm', orig_code) - - # generate declaration for register version - cblk = CodeBlock(code) - iop = InstObjParams(name, Name, 'AlphaStaticInst', cblk, opt_flags) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - exec_output = BasicExecute.subst(iop) - - if uses_imm: - # append declaration for imm version - imm_cblk = CodeBlock(imm_code) - imm_iop = InstObjParams(name, Name + 'Imm', 'IntegerImm', imm_cblk, - opt_flags) - header_output += BasicDeclare.subst(imm_iop) - decoder_output += BasicConstructor.subst(imm_iop) - exec_output += BasicExecute.subst(imm_iop) - # decode checks IMM bit to pick correct version - decode_block = RegOrImmDecode.subst(iop) - else: - # no imm version: just check for nop - decode_block = OperateNopCheckDecode.subst(iop) -}}; - - -//////////////////////////////////////////////////////////////////// -// -// Floating-point instructions -// -// Note that many FP-type instructions which do not support all the -// various rounding & trapping modes use the simpler format -// BasicOperateWithNopCheck. -// - -output exec {{ - /// Check "FP enabled" machine status bit. Called when executing any FP - /// instruction in full-system mode. - /// @retval Full-system mode: No_Fault if FP is enabled, Fen_Fault - /// if not. Non-full-system mode: always returns No_Fault. -#if FULL_SYSTEM - inline Fault checkFpEnableFault(%(CPU_exec_context)s *xc) - { - Fault fault = No_Fault; // dummy... this ipr access should not fault - if (!EV5::ICSR_FPE(xc->readIpr(AlphaISA::IPR_ICSR, fault))) { - fault = Fen_Fault; - } - return fault; - } -#else - inline Fault checkFpEnableFault(%(CPU_exec_context)s *xc) - { - return No_Fault; - } -#endif -}}; - -output header {{ - /** - * Base class for general floating-point instructions. Includes - * support for various Alpha rounding and trapping modes. Only FP - * instructions that require this support are derived from this - * class; the rest derive directly from AlphaStaticInst. - */ - class AlphaFP : public AlphaStaticInst - { - public: - /// Alpha FP rounding modes. - enum RoundingMode { - Chopped = 0, ///< round toward zero - Minus_Infinity = 1, ///< round toward minus infinity - Normal = 2, ///< round to nearest (default) - Dynamic = 3, ///< use FPCR setting (in instruction) - Plus_Infinity = 3 ///< round to plus inifinity (in FPCR) - }; - - /// Alpha FP trapping modes. - /// For instructions that produce integer results, the - /// "Underflow Enable" modes really mean "Overflow Enable", and - /// the assembly modifier is V rather than U. - enum TrappingMode { - /// default: nothing enabled - Imprecise = 0, ///< no modifier - /// underflow/overflow traps enabled, inexact disabled - Underflow_Imprecise = 1, ///< /U or /V - Underflow_Precise = 5, ///< /SU or /SV - /// underflow/overflow and inexact traps enabled - Underflow_Inexact_Precise = 7 ///< /SUI or /SVI - }; - - protected: - /// Map Alpha rounding mode to C99 constants from <fenv.h>. - static const int alphaToC99RoundingMode[]; - - /// Map enum RoundingMode values to disassembly suffixes. - static const char *roundingModeSuffix[]; - /// Map enum TrappingMode values to FP disassembly suffixes. - static const char *fpTrappingModeSuffix[]; - /// Map enum TrappingMode values to integer disassembly suffixes. - static const char *intTrappingModeSuffix[]; - - /// This instruction's rounding mode. - RoundingMode roundingMode; - /// This instruction's trapping mode. - TrappingMode trappingMode; - - /// Have we warned about this instruction's unsupported - /// rounding mode (if applicable)? - mutable bool warnedOnRounding; - - /// Have we warned about this instruction's unsupported - /// trapping mode (if applicable)? - mutable bool warnedOnTrapping; - - /// Constructor - AlphaFP(const char *mnem, MachInst _machInst, OpClass __opClass) - : AlphaStaticInst(mnem, _machInst, __opClass), - roundingMode((enum RoundingMode)FP_ROUNDMODE), - trappingMode((enum TrappingMode)FP_TRAPMODE), - warnedOnRounding(false), - warnedOnTrapping(false) - { - } - - int getC99RoundingMode(uint64_t fpcr_val) const; - - // This differs from the AlphaStaticInst version only in - // printing suffixes for non-default rounding & trapping modes. - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; - -}}; - - -output decoder {{ - int - AlphaFP::getC99RoundingMode(uint64_t fpcr_val) const - { - if (roundingMode == Dynamic) { - return alphaToC99RoundingMode[bits(fpcr_val, 59, 58)]; - } - else { - return alphaToC99RoundingMode[roundingMode]; - } - } - - std::string - AlphaFP::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - std::string mnem_str(mnemonic); - -#ifndef SS_COMPATIBLE_DISASSEMBLY - std::string suffix(""); - suffix += ((_destRegIdx[0] >= FP_Base_DepTag) - ? fpTrappingModeSuffix[trappingMode] - : intTrappingModeSuffix[trappingMode]); - suffix += roundingModeSuffix[roundingMode]; - - if (suffix != "") { - mnem_str = csprintf("%s/%s", mnemonic, suffix); - } -#endif - - std::stringstream ss; - ccprintf(ss, "%-10s ", mnem_str.c_str()); - - // just print the first two source regs... if there's - // a third one, it's a read-modify-write dest (Rc), - // e.g. for CMOVxx - if (_numSrcRegs > 0) { - printReg(ss, _srcRegIdx[0]); - } - if (_numSrcRegs > 1) { - ss << ","; - printReg(ss, _srcRegIdx[1]); - } - - // just print the first dest... if there's a second one, - // it's generally implicit - if (_numDestRegs > 0) { - if (_numSrcRegs > 0) - ss << ","; - printReg(ss, _destRegIdx[0]); - } - - return ss.str(); - } - - const int AlphaFP::alphaToC99RoundingMode[] = { - FE_TOWARDZERO, // Chopped - FE_DOWNWARD, // Minus_Infinity - FE_TONEAREST, // Normal - FE_UPWARD // Dynamic in inst, Plus_Infinity in FPCR - }; - - const char *AlphaFP::roundingModeSuffix[] = { "c", "m", "", "d" }; - // mark invalid trapping modes, but don't fail on them, because - // you could decode anything on a misspeculated path - const char *AlphaFP::fpTrappingModeSuffix[] = - { "", "u", "INVTM2", "INVTM3", "INVTM4", "su", "INVTM6", "sui" }; - const char *AlphaFP::intTrappingModeSuffix[] = - { "", "v", "INVTM2", "INVTM3", "INVTM4", "sv", "INVTM6", "svi" }; -}}; - -// FP instruction class execute method template. Handles non-standard -// rounding modes. -def template FloatingPointExecute {{ - Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - if (trappingMode != Imprecise && !warnedOnTrapping) { - warn("%s: non-standard trapping mode not supported", - generateDisassembly(0, NULL)); - warnedOnTrapping = true; - } - - Fault fault = No_Fault; - - %(fp_enable_check)s; - %(op_decl)s; - %(op_rd)s; -#if USE_FENV - if (roundingMode == Normal) { - %(code)s; - } else { - fesetround(getC99RoundingMode(xc->readFpcr())); - %(code)s; - fesetround(FE_TONEAREST); - } -#else - if (roundingMode != Normal && !warnedOnRounding) { - warn("%s: non-standard rounding mode not supported", - generateDisassembly(0, NULL)); - warnedOnRounding = true; - } - %(code)s; -#endif - - if (fault == No_Fault) { - %(op_wb)s; - } - - return fault; - } -}}; - -// FP instruction class execute method template where no dynamic -// rounding mode control is needed. Like BasicExecute, but includes -// check & warning for non-standard trapping mode. -def template FPFixedRoundingExecute {{ - Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - if (trappingMode != Imprecise && !warnedOnTrapping) { - warn("%s: non-standard trapping mode not supported", - generateDisassembly(0, NULL)); - warnedOnTrapping = true; - } - - Fault fault = No_Fault; - - %(fp_enable_check)s; - %(op_decl)s; - %(op_rd)s; - %(code)s; - - if (fault == No_Fault) { - %(op_wb)s; - } - - return fault; - } -}}; - -def template FloatingPointDecode {{ - { - AlphaStaticInst *i = new %(class_name)s(machInst); - if (FC == 31) { - i = makeNop(i); - } - return i; - } -}}; - -// General format for floating-point operate instructions: -// - Checks trapping and rounding mode flags. Trapping modes -// currently unimplemented (will fail). -// - Generates NOP if FC == 31. -def format FloatingPointOperate(code, *opt_args) {{ - iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args) - decode_block = FloatingPointDecode.subst(iop) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - exec_output = FloatingPointExecute.subst(iop) -}}; - -// Special format for cvttq where rounding mode is pre-decoded -def format FPFixedRounding(code, class_suffix, *opt_args) {{ - Name += class_suffix - iop = InstObjParams(name, Name, 'AlphaFP', CodeBlock(code), opt_args) - decode_block = FloatingPointDecode.subst(iop) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - exec_output = FPFixedRoundingExecute.subst(iop) -}}; - -//////////////////////////////////////////////////////////////////// -// -// Memory-format instructions: LoadAddress, Load, Store -// - -output header {{ - /** - * Base class for general Alpha memory-format instructions. - */ - class Memory : public AlphaStaticInst - { - protected: - - /// Memory request flags. See mem_req_base.hh. - unsigned memAccessFlags; - /// Pointer to EAComp object. - const StaticInstPtr<AlphaISA> eaCompPtr; - /// Pointer to MemAcc object. - const StaticInstPtr<AlphaISA> memAccPtr; - - /// Constructor - Memory(const char *mnem, MachInst _machInst, OpClass __opClass, - StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, - StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr) - : AlphaStaticInst(mnem, _machInst, __opClass), - memAccessFlags(0), eaCompPtr(_eaCompPtr), memAccPtr(_memAccPtr) - { - } - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - - public: - - const StaticInstPtr<AlphaISA> &eaCompInst() const { return eaCompPtr; } - const StaticInstPtr<AlphaISA> &memAccInst() const { return memAccPtr; } - }; - - /** - * Base class for memory-format instructions using a 32-bit - * displacement (i.e. most of them). - */ - class MemoryDisp32 : public Memory - { - protected: - /// Displacement for EA calculation (signed). - int32_t disp; - - /// Constructor. - MemoryDisp32(const char *mnem, MachInst _machInst, OpClass __opClass, - StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, - StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr) - : Memory(mnem, _machInst, __opClass, _eaCompPtr, _memAccPtr), - disp(MEMDISP) - { - } - }; - - - /** - * Base class for a few miscellaneous memory-format insts - * that don't interpret the disp field: wh64, fetch, fetch_m, ecb. - * None of these instructions has a destination register either. - */ - class MemoryNoDisp : public Memory - { - protected: - /// Constructor - MemoryNoDisp(const char *mnem, MachInst _machInst, OpClass __opClass, - StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, - StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr) - : Memory(mnem, _machInst, __opClass, _eaCompPtr, _memAccPtr) - { - } - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - - -output decoder {{ - std::string - Memory::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - return csprintf("%-10s %c%d,%d(r%d)", mnemonic, - flags[IsFloating] ? 'f' : 'r', RA, MEMDISP, RB); - } - - std::string - MemoryNoDisp::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - return csprintf("%-10s (r%d)", mnemonic, RB); - } -}}; - -def format LoadAddress(code) {{ - iop = InstObjParams(name, Name, 'MemoryDisp32', CodeBlock(code)) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - decode_block = BasicDecode.subst(iop) - exec_output = BasicExecute.subst(iop) -}}; - - -def template LoadStoreDeclare {{ - /** - * Static instruction class for "%(mnemonic)s". - */ - class %(class_name)s : public %(base_class)s - { - protected: - - /** - * "Fake" effective address computation class for "%(mnemonic)s". - */ - class EAComp : public %(base_class)s - { - public: - /// Constructor - EAComp(MachInst machInst); - - %(BasicExecDeclare)s - }; - - /** - * "Fake" memory access instruction class for "%(mnemonic)s". - */ - class MemAcc : public %(base_class)s - { - public: - /// Constructor - MemAcc(MachInst machInst); - - %(BasicExecDeclare)s - }; - - public: - - /// Constructor. - %(class_name)s(MachInst machInst); - - %(BasicExecDeclare)s - }; -}}; - -def template LoadStoreConstructor {{ - /** TODO: change op_class to AddrGenOp or something (requires - * creating new member of OpClass enum in op_class.hh, updating - * config files, etc.). */ - inline %(class_name)s::EAComp::EAComp(MachInst machInst) - : %(base_class)s("%(mnemonic)s (EAComp)", machInst, IntAluOp) - { - %(ea_constructor)s; - } - - inline %(class_name)s::MemAcc::MemAcc(MachInst machInst) - : %(base_class)s("%(mnemonic)s (MemAcc)", machInst, %(op_class)s) - { - %(memacc_constructor)s; - } - - inline %(class_name)s::%(class_name)s(MachInst machInst) - : %(base_class)s("%(mnemonic)s", machInst, %(op_class)s, - new EAComp(machInst), new MemAcc(machInst)) - { - %(constructor)s; - } -}}; - - -def template EACompExecute {{ - Fault - %(class_name)s::EAComp::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - Addr EA; - Fault fault = No_Fault; - - %(fp_enable_check)s; - %(op_decl)s; - %(op_rd)s; - %(code)s; - - if (fault == No_Fault) { - %(op_wb)s; - xc->setEA(EA); - } - - return fault; - } -}}; - -def template MemAccExecute {{ - Fault - %(class_name)s::MemAcc::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - Addr EA; - Fault fault = No_Fault; - - %(fp_enable_check)s; - %(op_decl)s; - %(op_nonmem_rd)s; - EA = xc->getEA(); - - if (fault == No_Fault) { - %(op_mem_rd)s; - %(code)s; - } - - if (fault == No_Fault) { - %(op_mem_wb)s; - } - - if (fault == No_Fault) { - %(postacc_code)s; - } - - if (fault == No_Fault) { - %(op_nonmem_wb)s; - } - - return fault; - } -}}; - - -def template LoadStoreExecute {{ - Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - Addr EA; - Fault fault = No_Fault; - - %(fp_enable_check)s; - %(op_decl)s; - %(op_nonmem_rd)s; - %(ea_code)s; - - if (fault == No_Fault) { - %(op_mem_rd)s; - %(memacc_code)s; - } - - if (fault == No_Fault) { - %(op_mem_wb)s; - } - - if (fault == No_Fault) { - %(postacc_code)s; - } - - if (fault == No_Fault) { - %(op_nonmem_wb)s; - } - - return fault; - } -}}; - - -def template PrefetchExecute {{ - Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - Addr EA; - Fault fault = No_Fault; - - %(fp_enable_check)s; - %(op_decl)s; - %(op_nonmem_rd)s; - %(ea_code)s; - - if (fault == No_Fault) { - xc->prefetch(EA, memAccessFlags); - } - - return No_Fault; - } -}}; - -// load instructions use Ra as dest, so check for -// Ra == 31 to detect nops -def template LoadNopCheckDecode {{ - { - AlphaStaticInst *i = new %(class_name)s(machInst); - if (RA == 31) { - i = makeNop(i); - } - return i; - } -}}; - - -// for some load instructions, Ra == 31 indicates a prefetch (not a nop) -def template LoadPrefetchCheckDecode {{ - { - if (RA != 31) { - return new %(class_name)s(machInst); - } - else { - return new %(class_name)sPrefetch(machInst); - } - } -}}; - - -let {{ -def LoadStoreBase(name, Name, ea_code, memacc_code, postacc_code = '', - base_class = 'MemoryDisp32', flags = [], - decode_template = BasicDecode, - exec_template = LoadStoreExecute): - # Segregate flags into instruction flags (handled by InstObjParams) - # and memory access flags (handled here). - - # Would be nice to autogenerate this list, but oh well. - valid_mem_flags = ['LOCKED', 'NO_FAULT', 'EVICT_NEXT', 'PF_EXCLUSIVE'] - mem_flags = [f for f in flags if f in valid_mem_flags] - inst_flags = [f for f in flags if f not in valid_mem_flags] - - # add hook to get effective addresses into execution trace output. - ea_code += '\nif (traceData) { traceData->setAddr(EA); }\n' - - # generate code block objects - ea_cblk = CodeBlock(ea_code) - memacc_cblk = CodeBlock(memacc_code) - postacc_cblk = CodeBlock(postacc_code) - - # Some CPU models execute the memory operation as an atomic unit, - # while others want to separate them into an effective address - # computation and a memory access operation. As a result, we need - # to generate three StaticInst objects. Note that the latter two - # are nested inside the larger "atomic" one. - - # generate InstObjParams for EAComp object - ea_iop = InstObjParams(name, Name, base_class, ea_cblk, inst_flags) - - # generate InstObjParams for MemAcc object - memacc_iop = InstObjParams(name, Name, base_class, memacc_cblk, inst_flags) - # in the split execution model, the MemAcc portion is responsible - # for the post-access code. - memacc_iop.postacc_code = postacc_cblk.code - - # generate InstObjParams for unified execution - cblk = CodeBlock(ea_code + memacc_code + postacc_code) - iop = InstObjParams(name, Name, base_class, cblk, inst_flags) - - iop.ea_constructor = ea_cblk.constructor - iop.ea_code = ea_cblk.code - iop.memacc_constructor = memacc_cblk.constructor - iop.memacc_code = memacc_cblk.code - iop.postacc_code = postacc_cblk.code - - if mem_flags: - s = '\n\tmemAccessFlags = ' + string.join(mem_flags, '|') + ';' - iop.constructor += s - memacc_iop.constructor += s - - # (header_output, decoder_output, decode_block, exec_output) - return (LoadStoreDeclare.subst(iop), LoadStoreConstructor.subst(iop), - decode_template.subst(iop), - EACompExecute.subst(ea_iop) - + MemAccExecute.subst(memacc_iop) - + exec_template.subst(iop)) -}}; - - -def format LoadOrNop(ea_code, memacc_code, *flags) {{ - (header_output, decoder_output, decode_block, exec_output) = \ - LoadStoreBase(name, Name, ea_code, memacc_code, flags = flags, - decode_template = LoadNopCheckDecode) -}}; - - -// Note that the flags passed in apply only to the prefetch version -def format LoadOrPrefetch(ea_code, memacc_code, *pf_flags) {{ - # declare the load instruction object and generate the decode block - (header_output, decoder_output, decode_block, exec_output) = \ - LoadStoreBase(name, Name, ea_code, memacc_code, - decode_template = LoadPrefetchCheckDecode) - - # Declare the prefetch instruction object. - - # convert flags from tuple to list to make them mutable - pf_flags = list(pf_flags) + ['IsMemRef', 'IsLoad', 'IsDataPrefetch', 'MemReadOp', 'NO_FAULT'] - - (pf_header_output, pf_decoder_output, _, pf_exec_output) = \ - LoadStoreBase(name, Name + 'Prefetch', ea_code, '', - flags = pf_flags, exec_template = PrefetchExecute) - - header_output += pf_header_output - decoder_output += pf_decoder_output - exec_output += pf_exec_output -}}; - - -def format Store(ea_code, memacc_code, *flags) {{ - (header_output, decoder_output, decode_block, exec_output) = \ - LoadStoreBase(name, Name, ea_code, memacc_code, flags = flags) -}}; - - -def format StoreCond(ea_code, memacc_code, postacc_code, *flags) {{ - (header_output, decoder_output, decode_block, exec_output) = \ - LoadStoreBase(name, Name, ea_code, memacc_code, postacc_code, - flags = flags) -}}; - - -// Use 'MemoryNoDisp' as base: for wh64, fetch, ecb -def format MiscPrefetch(ea_code, memacc_code, *flags) {{ - (header_output, decoder_output, decode_block, exec_output) = \ - LoadStoreBase(name, Name, ea_code, memacc_code, flags = flags, - base_class = 'MemoryNoDisp') -}}; - - -//////////////////////////////////////////////////////////////////// -// -// Control transfer instructions -// - -output header {{ - - /** - * Base class for instructions whose disassembly is not purely a - * function of the machine instruction (i.e., it depends on the - * PC). This class overrides the disassemble() method to check - * the PC and symbol table values before re-using a cached - * disassembly string. This is necessary for branches and jumps, - * where the disassembly string includes the target address (which - * may depend on the PC and/or symbol table). - */ - class PCDependentDisassembly : public AlphaStaticInst - { - protected: - /// Cached program counter from last disassembly - mutable Addr cachedPC; - /// Cached symbol table pointer from last disassembly - mutable const SymbolTable *cachedSymtab; - - /// Constructor - PCDependentDisassembly(const char *mnem, MachInst _machInst, - OpClass __opClass) - : AlphaStaticInst(mnem, _machInst, __opClass), - cachedPC(0), cachedSymtab(0) - { - } - - const std::string & - disassemble(Addr pc, const SymbolTable *symtab) const; - }; - - /** - * Base class for branches (PC-relative control transfers), - * conditional or unconditional. - */ - class Branch : public PCDependentDisassembly - { - protected: - /// Displacement to target address (signed). - int32_t disp; - - /// Constructor. - Branch(const char *mnem, MachInst _machInst, OpClass __opClass) - : PCDependentDisassembly(mnem, _machInst, __opClass), - disp(BRDISP << 2) - { - } - - Addr branchTarget(Addr branchPC) const; - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; - - /** - * Base class for jumps (register-indirect control transfers). In - * the Alpha ISA, these are always unconditional. - */ - class Jump : public PCDependentDisassembly - { - protected: - - /// Displacement to target address (signed). - int32_t disp; - - public: - /// Constructor - Jump(const char *mnem, MachInst _machInst, OpClass __opClass) - : PCDependentDisassembly(mnem, _machInst, __opClass), - disp(BRDISP) - { - } - - Addr branchTarget(ExecContext *xc) const; - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -output decoder {{ - Addr - Branch::branchTarget(Addr branchPC) const - { - return branchPC + 4 + disp; - } - - Addr - Jump::branchTarget(ExecContext *xc) const - { - Addr NPC = xc->readPC() + 4; - uint64_t Rb = xc->readIntReg(_srcRegIdx[0]); - return (Rb & ~3) | (NPC & 1); - } - - const std::string & - PCDependentDisassembly::disassemble(Addr pc, - const SymbolTable *symtab) const - { - if (!cachedDisassembly || - pc != cachedPC || symtab != cachedSymtab) - { - if (cachedDisassembly) - delete cachedDisassembly; - - cachedDisassembly = - new std::string(generateDisassembly(pc, symtab)); - cachedPC = pc; - cachedSymtab = symtab; - } - - return *cachedDisassembly; - } - - std::string - Branch::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - std::stringstream ss; - - ccprintf(ss, "%-10s ", mnemonic); - - // There's only one register arg (RA), but it could be - // either a source (the condition for conditional - // branches) or a destination (the link reg for - // unconditional branches) - if (_numSrcRegs > 0) { - printReg(ss, _srcRegIdx[0]); - ss << ","; - } - else if (_numDestRegs > 0) { - printReg(ss, _destRegIdx[0]); - ss << ","; - } - -#ifdef SS_COMPATIBLE_DISASSEMBLY - if (_numSrcRegs == 0 && _numDestRegs == 0) { - printReg(ss, 31); - ss << ","; - } -#endif - - Addr target = pc + 4 + disp; - - std::string str; - if (symtab && symtab->findSymbol(target, str)) - ss << str; - else - ccprintf(ss, "0x%x", target); - - return ss.str(); - } - - std::string - Jump::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - std::stringstream ss; - - ccprintf(ss, "%-10s ", mnemonic); - -#ifdef SS_COMPATIBLE_DISASSEMBLY - if (_numDestRegs == 0) { - printReg(ss, 31); - ss << ","; - } -#endif - - if (_numDestRegs > 0) { - printReg(ss, _destRegIdx[0]); - ss << ","; - } - - ccprintf(ss, "(r%d)", RB); - - return ss.str(); - } -}}; - -def template JumpOrBranchDecode {{ - return (RA == 31) - ? (StaticInst<AlphaISA> *)new %(class_name)s(machInst) - : (StaticInst<AlphaISA> *)new %(class_name)sAndLink(machInst); -}}; - -def format CondBranch(code) {{ - code = 'bool cond;\n' + code + '\nif (cond) NPC = NPC + disp;\n'; - iop = InstObjParams(name, Name, 'Branch', CodeBlock(code), - ('IsDirectControl', 'IsCondControl')) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - decode_block = BasicDecode.subst(iop) - exec_output = BasicExecute.subst(iop) -}}; - -let {{ -def UncondCtrlBase(name, Name, base_class, npc_expr, flags): - # Declare basic control transfer w/o link (i.e. link reg is R31) - nolink_code = 'NPC = %s;\n' % npc_expr - nolink_iop = InstObjParams(name, Name, base_class, - CodeBlock(nolink_code), flags) - header_output = BasicDeclare.subst(nolink_iop) - decoder_output = BasicConstructor.subst(nolink_iop) - exec_output = BasicExecute.subst(nolink_iop) - - # Generate declaration of '*AndLink' version, append to decls - link_code = 'Ra = NPC & ~3;\n' + nolink_code - link_iop = InstObjParams(name, Name + 'AndLink', base_class, - CodeBlock(link_code), flags) - header_output += BasicDeclare.subst(link_iop) - decoder_output += BasicConstructor.subst(link_iop) - exec_output += BasicExecute.subst(link_iop) - - # need to use link_iop for the decode template since it is expecting - # the shorter version of class_name (w/o "AndLink") - - return (header_output, decoder_output, - JumpOrBranchDecode.subst(nolink_iop), exec_output) -}}; - -def format UncondBranch(*flags) {{ - flags += ('IsUncondControl', 'IsDirectControl') - (header_output, decoder_output, decode_block, exec_output) = \ - UncondCtrlBase(name, Name, 'Branch', 'NPC + disp', flags) -}}; - -def format Jump(*flags) {{ - flags += ('IsUncondControl', 'IsIndirectControl') - (header_output, decoder_output, decode_block, exec_output) = \ - UncondCtrlBase(name, Name, 'Jump', '(Rb & ~3) | (NPC & 1)', flags) -}}; - - -//////////////////////////////////////////////////////////////////// -// -// PAL calls -// - -output header {{ - /** - * Base class for emulated call_pal calls (used only in - * non-full-system mode). - */ - class EmulatedCallPal : public AlphaStaticInst - { - protected: - - /// Constructor. - EmulatedCallPal(const char *mnem, MachInst _machInst, - OpClass __opClass) - : AlphaStaticInst(mnem, _machInst, __opClass) - { - } - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -output decoder {{ - std::string - EmulatedCallPal::generateDisassembly(Addr pc, - const SymbolTable *symtab) const - { -#ifdef SS_COMPATIBLE_DISASSEMBLY - return csprintf("%s %s", "call_pal", mnemonic); -#else - return csprintf("%-10s %s", "call_pal", mnemonic); -#endif - } -}}; - -def format EmulatedCallPal(code, *flags) {{ - iop = InstObjParams(name, Name, 'EmulatedCallPal', CodeBlock(code), flags) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - decode_block = BasicDecode.subst(iop) - exec_output = BasicExecute.subst(iop) -}}; - -output header {{ - /** - * Base class for full-system-mode call_pal instructions. - * Probably could turn this into a leaf class and get rid of the - * parser template. - */ - class CallPalBase : public AlphaStaticInst - { - protected: - int palFunc; ///< Function code part of instruction - int palOffset; ///< Target PC, offset from IPR_PAL_BASE - bool palValid; ///< is the function code valid? - bool palPriv; ///< is this call privileged? - - /// Constructor. - CallPalBase(const char *mnem, MachInst _machInst, - OpClass __opClass); - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -output decoder {{ - inline - CallPalBase::CallPalBase(const char *mnem, MachInst _machInst, - OpClass __opClass) - : AlphaStaticInst(mnem, _machInst, __opClass), - palFunc(PALFUNC) - { - // From the 21164 HRM (paraphrased): - // Bit 7 of the function code (mask 0x80) indicates - // whether the call is privileged (bit 7 == 0) or - // unprivileged (bit 7 == 1). The privileged call table - // starts at 0x2000, the unprivielged call table starts at - // 0x3000. Bits 5-0 (mask 0x3f) are used to calculate the - // offset. - const int palPrivMask = 0x80; - const int palOffsetMask = 0x3f; - - // Pal call is invalid unless all other bits are 0 - palValid = ((machInst & ~(palPrivMask | palOffsetMask)) == 0); - palPriv = ((machInst & palPrivMask) == 0); - int shortPalFunc = (machInst & palOffsetMask); - // Add 1 to base to set pal-mode bit - palOffset = (palPriv ? 0x2001 : 0x3001) + (shortPalFunc << 6); - } - - std::string - CallPalBase::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - return csprintf("%-10s %#x", "call_pal", palFunc); - } -}}; - -def format CallPal(code, *flags) {{ - iop = InstObjParams(name, Name, 'CallPalBase', CodeBlock(code), flags) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - decode_block = BasicDecode.subst(iop) - exec_output = BasicExecute.subst(iop) -}}; - -//////////////////////////////////////////////////////////////////// -// -// hw_ld, hw_st -// - -output header {{ - /** - * Base class for hw_ld and hw_st. - */ - class HwLoadStore : public Memory - { - protected: - - /// Displacement for EA calculation (signed). - int16_t disp; - - /// Constructor - HwLoadStore(const char *mnem, MachInst _machInst, OpClass __opClass, - StaticInstPtr<AlphaISA> _eaCompPtr = nullStaticInstPtr, - StaticInstPtr<AlphaISA> _memAccPtr = nullStaticInstPtr); - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - - -output decoder {{ - inline - HwLoadStore::HwLoadStore(const char *mnem, MachInst _machInst, - OpClass __opClass, - StaticInstPtr<AlphaISA> _eaCompPtr, - StaticInstPtr<AlphaISA> _memAccPtr) - : Memory(mnem, _machInst, __opClass, _eaCompPtr, _memAccPtr), - disp(HW_LDST_DISP) - { - memAccessFlags = 0; - if (HW_LDST_PHYS) memAccessFlags |= PHYSICAL; - if (HW_LDST_ALT) memAccessFlags |= ALTMODE; - if (HW_LDST_VPTE) memAccessFlags |= VPTE; - if (HW_LDST_LOCK) memAccessFlags |= LOCKED; - } - - std::string - HwLoadStore::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { -#ifdef SS_COMPATIBLE_DISASSEMBLY - return csprintf("%-10s r%d,%d(r%d)", mnemonic, RA, disp, RB); -#else - // HW_LDST_LOCK and HW_LDST_COND are the same bit. - const char *lock_str = - (HW_LDST_LOCK) ? (flags[IsLoad] ? ",LOCK" : ",COND") : ""; - - return csprintf("%-10s r%d,%d(r%d)%s%s%s%s%s", - mnemonic, RA, disp, RB, - HW_LDST_PHYS ? ",PHYS" : "", - HW_LDST_ALT ? ",ALT" : "", - HW_LDST_QUAD ? ",QUAD" : "", - HW_LDST_VPTE ? ",VPTE" : "", - lock_str); -#endif - } -}}; - -def format HwLoadStore(ea_code, memacc_code, class_ext, *flags) {{ - (header_output, decoder_output, decode_block, exec_output) = \ - LoadStoreBase(name, Name + class_ext, ea_code, memacc_code, - flags = flags, base_class = 'HwLoadStore') -}}; - - -def format HwStoreCond(ea_code, memacc_code, postacc_code, class_ext, *flags) {{ - (header_output, decoder_output, decode_block, exec_output) = \ - LoadStoreBase(name, Name + class_ext, ea_code, memacc_code, - postacc_code, flags = flags, base_class = 'HwLoadStore') -}}; - - -output header {{ - /** - * Base class for hw_mfpr and hw_mtpr. - */ - class HwMoveIPR : public AlphaStaticInst - { - protected: - /// Index of internal processor register. - int ipr_index; - - /// Constructor - HwMoveIPR(const char *mnem, MachInst _machInst, OpClass __opClass) - : AlphaStaticInst(mnem, _machInst, __opClass), - ipr_index(HW_IPR_IDX) - { - } - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -output decoder {{ - std::string - HwMoveIPR::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - if (_numSrcRegs > 0) { - // must be mtpr - return csprintf("%-10s r%d,IPR(%#x)", - mnemonic, RA, ipr_index); - } - else { - // must be mfpr - return csprintf("%-10s IPR(%#x),r%d", - mnemonic, ipr_index, RA); - } - } -}}; - -def format HwMoveIPR(code) {{ - iop = InstObjParams(name, Name, 'HwMoveIPR', CodeBlock(code), - ['IprAccessOp']) - header_output = BasicDeclare.subst(iop) - decoder_output = BasicConstructor.subst(iop) - decode_block = BasicDecode.subst(iop) - exec_output = BasicExecute.subst(iop) -}}; - - -//////////////////////////////////////////////////////////////////// -// -// Unimplemented instructions -// - -output header {{ - /** - * Static instruction class for unimplemented instructions that - * cause simulator termination. Note that these are recognized - * (legal) instructions that the simulator does not support; the - * 'Unknown' class is used for unrecognized/illegal instructions. - * This is a leaf class. - */ - class FailUnimplemented : public AlphaStaticInst - { - public: - /// Constructor - FailUnimplemented(const char *_mnemonic, MachInst _machInst) - : AlphaStaticInst(_mnemonic, _machInst, No_OpClass) - { - // don't call execute() (which panics) if we're on a - // speculative path - flags[IsNonSpeculative] = true; - } - - %(BasicExecDeclare)s - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; - - /** - * Base class for unimplemented instructions that cause a warning - * to be printed (but do not terminate simulation). This - * implementation is a little screwy in that it will print a - * warning for each instance of a particular unimplemented machine - * instruction, not just for each unimplemented opcode. Should - * probably make the 'warned' flag a static member of the derived - * class. - */ - class WarnUnimplemented : public AlphaStaticInst - { - private: - /// Have we warned on this instruction yet? - mutable bool warned; - - public: - /// Constructor - WarnUnimplemented(const char *_mnemonic, MachInst _machInst) - : AlphaStaticInst(_mnemonic, _machInst, No_OpClass), warned(false) - { - // don't call execute() (which panics) if we're on a - // speculative path - flags[IsNonSpeculative] = true; - } - - %(BasicExecDeclare)s - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -output decoder {{ - std::string - FailUnimplemented::generateDisassembly(Addr pc, - const SymbolTable *symtab) const - { - return csprintf("%-10s (unimplemented)", mnemonic); - } - - std::string - WarnUnimplemented::generateDisassembly(Addr pc, - const SymbolTable *symtab) const - { -#ifdef SS_COMPATIBLE_DISASSEMBLY - return csprintf("%-10s", mnemonic); -#else - return csprintf("%-10s (unimplemented)", mnemonic); -#endif - } -}}; - -output exec {{ - Fault - FailUnimplemented::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - panic("attempt to execute unimplemented instruction '%s' " - "(inst 0x%08x, opcode 0x%x)", mnemonic, machInst, OPCODE); - return Unimplemented_Opcode_Fault; - } - - Fault - WarnUnimplemented::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - if (!warned) { - warn("instruction '%s' unimplemented\n", mnemonic); - warned = true; - } - - return No_Fault; - } -}}; - - -def format FailUnimpl() {{ - iop = InstObjParams(name, 'FailUnimplemented') - decode_block = BasicDecodeWithMnemonic.subst(iop) -}}; - -def format WarnUnimpl() {{ - iop = InstObjParams(name, 'WarnUnimplemented') - decode_block = BasicDecodeWithMnemonic.subst(iop) -}}; - -output header {{ - /** - * Static instruction class for unknown (illegal) instructions. - * These cause simulator termination if they are executed in a - * non-speculative mode. This is a leaf class. - */ - class Unknown : public AlphaStaticInst - { - public: - /// Constructor - Unknown(MachInst _machInst) - : AlphaStaticInst("unknown", _machInst, No_OpClass) - { - // don't call execute() (which panics) if we're on a - // speculative path - flags[IsNonSpeculative] = true; - } - - %(BasicExecDeclare)s - - std::string - generateDisassembly(Addr pc, const SymbolTable *symtab) const; - }; -}}; - -//////////////////////////////////////////////////////////////////// -// -// Unknown instructions -// - -output decoder {{ - std::string - Unknown::generateDisassembly(Addr pc, const SymbolTable *symtab) const - { - return csprintf("%-10s (inst 0x%x, opcode 0x%x)", - "unknown", machInst, OPCODE); - } -}}; - -output exec {{ - Fault - Unknown::execute(%(CPU_exec_context)s *xc, - Trace::InstRecord *traceData) const - { - panic("attempt to execute unknown instruction " - "(inst 0x%08x, opcode 0x%x)", machInst, OPCODE); - return Unimplemented_Opcode_Fault; - } -}}; - -def format Unknown() {{ - decode_block = 'return new Unknown(machInst);\n' -}}; - -//////////////////////////////////////////////////////////////////// -// -// Utility functions for execute methods -// - -output exec {{ - - /// Return opa + opb, summing carry into third arg. - inline uint64_t - addc(uint64_t opa, uint64_t opb, int &carry) - { - uint64_t res = opa + opb; - if (res < opa || res < opb) - ++carry; - return res; - } - - /// Multiply two 64-bit values (opa * opb), returning the 128-bit - /// product in res_hi and res_lo. - inline void - mul128(uint64_t opa, uint64_t opb, uint64_t &res_hi, uint64_t &res_lo) - { - // do a 64x64 --> 128 multiply using four 32x32 --> 64 multiplies - uint64_t opa_hi = opa<63:32>; - uint64_t opa_lo = opa<31:0>; - uint64_t opb_hi = opb<63:32>; - uint64_t opb_lo = opb<31:0>; - - res_lo = opa_lo * opb_lo; - - // The middle partial products logically belong in bit - // positions 95 to 32. Thus the lower 32 bits of each product - // sum into the upper 32 bits of the low result, while the - // upper 32 sum into the low 32 bits of the upper result. - uint64_t partial1 = opa_hi * opb_lo; - uint64_t partial2 = opa_lo * opb_hi; - - uint64_t partial1_lo = partial1<31:0> << 32; - uint64_t partial1_hi = partial1<63:32>; - uint64_t partial2_lo = partial2<31:0> << 32; - uint64_t partial2_hi = partial2<63:32>; - - // Add partial1_lo and partial2_lo to res_lo, keeping track - // of any carries out - int carry_out = 0; - res_lo = addc(partial1_lo, res_lo, carry_out); - res_lo = addc(partial2_lo, res_lo, carry_out); - - // Now calculate the high 64 bits... - res_hi = (opa_hi * opb_hi) + partial1_hi + partial2_hi + carry_out; - } - - /// Map 8-bit S-floating exponent to 11-bit T-floating exponent. - /// See Table 2-2 of Alpha AHB. - inline int - map_s(int old_exp) - { - int hibit = old_exp<7:>; - int lobits = old_exp<6:0>; - - if (hibit == 1) { - return (lobits == 0x7f) ? 0x7ff : (0x400 | lobits); - } - else { - return (lobits == 0) ? 0 : (0x380 | lobits); - } - } - - /// Convert a 32-bit S-floating value to the equivalent 64-bit - /// representation to be stored in an FP reg. - inline uint64_t - s_to_t(uint32_t s_val) - { - uint64_t tmp = s_val; - return (tmp<31:> << 63 // sign bit - | (uint64_t)map_s(tmp<30:23>) << 52 // exponent - | tmp<22:0> << 29); // fraction - } - - /// Convert a 64-bit T-floating value to the equivalent 32-bit - /// S-floating representation to be stored in memory. - inline int32_t - t_to_s(uint64_t t_val) - { - return (t_val<63:62> << 30 // sign bit & hi exp bit - | t_val<58:29>); // rest of exp & fraction - } -}}; - -//////////////////////////////////////////////////////////////////// -// -// The actual decoder specification -// - -decode OPCODE default Unknown::unknown() { - - format LoadAddress { - 0x08: lda({{ Ra = Rb + disp; }}); - 0x09: ldah({{ Ra = Rb + (disp << 16); }}); - } - - format LoadOrNop { - 0x0a: ldbu({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.ub; }}); - 0x0c: ldwu({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.uw; }}); - 0x0b: ldq_u({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem.uq; }}); - 0x23: ldt({{ EA = Rb + disp; }}, {{ Fa = Mem.df; }}); - 0x2a: ldl_l({{ EA = Rb + disp; }}, {{ Ra.sl = Mem.sl; }}, LOCKED); - 0x2b: ldq_l({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.uq; }}, LOCKED); - 0x20: copy_load({{EA = Ra;}}, - {{fault = xc->copySrcTranslate(EA);}}, - IsMemRef, IsLoad, IsCopy); - } - - format LoadOrPrefetch { - 0x28: ldl({{ EA = Rb + disp; }}, {{ Ra.sl = Mem.sl; }}); - 0x29: ldq({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.uq; }}, EVICT_NEXT); - // IsFloating flag on lds gets the prefetch to disassemble - // using f31 instead of r31... funcitonally it's unnecessary - 0x22: lds({{ EA = Rb + disp; }}, {{ Fa.uq = s_to_t(Mem.ul); }}, - PF_EXCLUSIVE, IsFloating); - } - - format Store { - 0x0e: stb({{ EA = Rb + disp; }}, {{ Mem.ub = Ra<7:0>; }}); - 0x0d: stw({{ EA = Rb + disp; }}, {{ Mem.uw = Ra<15:0>; }}); - 0x2c: stl({{ EA = Rb + disp; }}, {{ Mem.ul = Ra<31:0>; }}); - 0x2d: stq({{ EA = Rb + disp; }}, {{ Mem.uq = Ra.uq; }}); - 0x0f: stq_u({{ EA = (Rb + disp) & ~7; }}, {{ Mem.uq = Ra.uq; }}); - 0x26: sts({{ EA = Rb + disp; }}, {{ Mem.ul = t_to_s(Fa.uq); }}); - 0x27: stt({{ EA = Rb + disp; }}, {{ Mem.df = Fa; }}); - 0x24: copy_store({{EA = Rb;}}, - {{fault = xc->copy(EA);}}, - IsMemRef, IsStore, IsCopy); - } - - format StoreCond { - 0x2e: stl_c({{ EA = Rb + disp; }}, {{ Mem.ul = Ra<31:0>; }}, - {{ - uint64_t tmp = Mem_write_result; - // see stq_c - Ra = (tmp == 0 || tmp == 1) ? tmp : Ra; - }}, LOCKED); - 0x2f: stq_c({{ EA = Rb + disp; }}, {{ Mem.uq = Ra; }}, - {{ - uint64_t tmp = Mem_write_result; - // If the write operation returns 0 or 1, then - // this was a conventional store conditional, - // and the value indicates the success/failure - // of the operation. If another value is - // returned, then this was a Turbolaser - // mailbox access, and we don't update the - // result register at all. - Ra = (tmp == 0 || tmp == 1) ? tmp : Ra; - }}, LOCKED); - } - - format IntegerOperate { - - 0x10: decode INTFUNC { // integer arithmetic operations - - 0x00: addl({{ Rc.sl = Ra.sl + Rb_or_imm.sl; }}); - 0x40: addlv({{ - uint32_t tmp = Ra.sl + Rb_or_imm.sl; - // signed overflow occurs when operands have same sign - // and sign of result does not match. - if (Ra.sl<31:> == Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>) - fault = Integer_Overflow_Fault; - Rc.sl = tmp; - }}); - 0x02: s4addl({{ Rc.sl = (Ra.sl << 2) + Rb_or_imm.sl; }}); - 0x12: s8addl({{ Rc.sl = (Ra.sl << 3) + Rb_or_imm.sl; }}); - - 0x20: addq({{ Rc = Ra + Rb_or_imm; }}); - 0x60: addqv({{ - uint64_t tmp = Ra + Rb_or_imm; - // signed overflow occurs when operands have same sign - // and sign of result does not match. - if (Ra<63:> == Rb_or_imm<63:> && tmp<63:> != Ra<63:>) - fault = Integer_Overflow_Fault; - Rc = tmp; - }}); - 0x22: s4addq({{ Rc = (Ra << 2) + Rb_or_imm; }}); - 0x32: s8addq({{ Rc = (Ra << 3) + Rb_or_imm; }}); - - 0x09: subl({{ Rc.sl = Ra.sl - Rb_or_imm.sl; }}); - 0x49: sublv({{ - uint32_t tmp = Ra.sl - Rb_or_imm.sl; - // signed overflow detection is same as for add, - // except we need to look at the *complemented* - // sign bit of the subtrahend (Rb), i.e., if the initial - // signs are the *same* then no overflow can occur - if (Ra.sl<31:> != Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>) - fault = Integer_Overflow_Fault; - Rc.sl = tmp; - }}); - 0x0b: s4subl({{ Rc.sl = (Ra.sl << 2) - Rb_or_imm.sl; }}); - 0x1b: s8subl({{ Rc.sl = (Ra.sl << 3) - Rb_or_imm.sl; }}); - - 0x29: subq({{ Rc = Ra - Rb_or_imm; }}); - 0x69: subqv({{ - uint64_t tmp = Ra - Rb_or_imm; - // signed overflow detection is same as for add, - // except we need to look at the *complemented* - // sign bit of the subtrahend (Rb), i.e., if the initial - // signs are the *same* then no overflow can occur - if (Ra<63:> != Rb_or_imm<63:> && tmp<63:> != Ra<63:>) - fault = Integer_Overflow_Fault; - Rc = tmp; - }}); - 0x2b: s4subq({{ Rc = (Ra << 2) - Rb_or_imm; }}); - 0x3b: s8subq({{ Rc = (Ra << 3) - Rb_or_imm; }}); - - 0x2d: cmpeq({{ Rc = (Ra == Rb_or_imm); }}); - 0x6d: cmple({{ Rc = (Ra.sq <= Rb_or_imm.sq); }}); - 0x4d: cmplt({{ Rc = (Ra.sq < Rb_or_imm.sq); }}); - 0x3d: cmpule({{ Rc = (Ra.uq <= Rb_or_imm.uq); }}); - 0x1d: cmpult({{ Rc = (Ra.uq < Rb_or_imm.uq); }}); - - 0x0f: cmpbge({{ - int hi = 7; - int lo = 0; - uint64_t tmp = 0; - for (int i = 0; i < 8; ++i) { - tmp |= (Ra.uq<hi:lo> >= Rb_or_imm.uq<hi:lo>) << i; - hi += 8; - lo += 8; - } - Rc = tmp; - }}); - } - - 0x11: decode INTFUNC { // integer logical operations - - 0x00: and({{ Rc = Ra & Rb_or_imm; }}); - 0x08: bic({{ Rc = Ra & ~Rb_or_imm; }}); - 0x20: bis({{ Rc = Ra | Rb_or_imm; }}); - 0x28: ornot({{ Rc = Ra | ~Rb_or_imm; }}); - 0x40: xor({{ Rc = Ra ^ Rb_or_imm; }}); - 0x48: eqv({{ Rc = Ra ^ ~Rb_or_imm; }}); - - // conditional moves - 0x14: cmovlbs({{ Rc = ((Ra & 1) == 1) ? Rb_or_imm : Rc; }}); - 0x16: cmovlbc({{ Rc = ((Ra & 1) == 0) ? Rb_or_imm : Rc; }}); - 0x24: cmoveq({{ Rc = (Ra == 0) ? Rb_or_imm : Rc; }}); - 0x26: cmovne({{ Rc = (Ra != 0) ? Rb_or_imm : Rc; }}); - 0x44: cmovlt({{ Rc = (Ra.sq < 0) ? Rb_or_imm : Rc; }}); - 0x46: cmovge({{ Rc = (Ra.sq >= 0) ? Rb_or_imm : Rc; }}); - 0x64: cmovle({{ Rc = (Ra.sq <= 0) ? Rb_or_imm : Rc; }}); - 0x66: cmovgt({{ Rc = (Ra.sq > 0) ? Rb_or_imm : Rc; }}); - - // For AMASK, RA must be R31. - 0x61: decode RA { - 31: amask({{ Rc = Rb_or_imm & ~ULL(0x17); }}); - } - - // For IMPLVER, RA must be R31 and the B operand - // must be the immediate value 1. - 0x6c: decode RA { - 31: decode IMM { - 1: decode INTIMM { - // return EV5 for FULL_SYSTEM and EV6 otherwise - 1: implver({{ -#if FULL_SYSTEM - Rc = 1; -#else - Rc = 2; -#endif - }}); - } - } - } - -#if FULL_SYSTEM - // The mysterious 11.25... - 0x25: WarnUnimpl::eleven25(); -#endif - } - - 0x12: decode INTFUNC { - 0x39: sll({{ Rc = Ra << Rb_or_imm<5:0>; }}); - 0x34: srl({{ Rc = Ra.uq >> Rb_or_imm<5:0>; }}); - 0x3c: sra({{ Rc = Ra.sq >> Rb_or_imm<5:0>; }}); - - 0x02: mskbl({{ Rc = Ra & ~(mask( 8) << (Rb_or_imm<2:0> * 8)); }}); - 0x12: mskwl({{ Rc = Ra & ~(mask(16) << (Rb_or_imm<2:0> * 8)); }}); - 0x22: mskll({{ Rc = Ra & ~(mask(32) << (Rb_or_imm<2:0> * 8)); }}); - 0x32: mskql({{ Rc = Ra & ~(mask(64) << (Rb_or_imm<2:0> * 8)); }}); - - 0x52: mskwh({{ - int bv = Rb_or_imm<2:0>; - Rc = bv ? (Ra & ~(mask(16) >> (64 - 8 * bv))) : Ra; - }}); - 0x62: msklh({{ - int bv = Rb_or_imm<2:0>; - Rc = bv ? (Ra & ~(mask(32) >> (64 - 8 * bv))) : Ra; - }}); - 0x72: mskqh({{ - int bv = Rb_or_imm<2:0>; - Rc = bv ? (Ra & ~(mask(64) >> (64 - 8 * bv))) : Ra; - }}); - - 0x06: extbl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))< 7:0>; }}); - 0x16: extwl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<15:0>; }}); - 0x26: extll({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<31:0>; }}); - 0x36: extql({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8)); }}); - - 0x5a: extwh({{ - Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<15:0>; }}); - 0x6a: extlh({{ - Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<31:0>; }}); - 0x7a: extqh({{ - Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>); }}); - - 0x0b: insbl({{ Rc = Ra< 7:0> << (Rb_or_imm<2:0> * 8); }}); - 0x1b: inswl({{ Rc = Ra<15:0> << (Rb_or_imm<2:0> * 8); }}); - 0x2b: insll({{ Rc = Ra<31:0> << (Rb_or_imm<2:0> * 8); }}); - 0x3b: insql({{ Rc = Ra << (Rb_or_imm<2:0> * 8); }}); - - 0x57: inswh({{ - int bv = Rb_or_imm<2:0>; - Rc = bv ? (Ra.uq<15:0> >> (64 - 8 * bv)) : 0; - }}); - 0x67: inslh({{ - int bv = Rb_or_imm<2:0>; - Rc = bv ? (Ra.uq<31:0> >> (64 - 8 * bv)) : 0; - }}); - 0x77: insqh({{ - int bv = Rb_or_imm<2:0>; - Rc = bv ? (Ra.uq >> (64 - 8 * bv)) : 0; - }}); - - 0x30: zap({{ - uint64_t zapmask = 0; - for (int i = 0; i < 8; ++i) { - if (Rb_or_imm<i:>) - zapmask |= (mask(8) << (i * 8)); - } - Rc = Ra & ~zapmask; - }}); - 0x31: zapnot({{ - uint64_t zapmask = 0; - for (int i = 0; i < 8; ++i) { - if (!Rb_or_imm<i:>) - zapmask |= (mask(8) << (i * 8)); - } - Rc = Ra & ~zapmask; - }}); - } - - 0x13: decode INTFUNC { // integer multiplies - 0x00: mull({{ Rc.sl = Ra.sl * Rb_or_imm.sl; }}, IntMultOp); - 0x20: mulq({{ Rc = Ra * Rb_or_imm; }}, IntMultOp); - 0x30: umulh({{ - uint64_t hi, lo; - mul128(Ra, Rb_or_imm, hi, lo); - Rc = hi; - }}, IntMultOp); - 0x40: mullv({{ - // 32-bit multiply with trap on overflow - int64_t Rax = Ra.sl; // sign extended version of Ra.sl - int64_t Rbx = Rb_or_imm.sl; - int64_t tmp = Rax * Rbx; - // To avoid overflow, all the upper 32 bits must match - // the sign bit of the lower 32. We code this as - // checking the upper 33 bits for all 0s or all 1s. - uint64_t sign_bits = tmp<63:31>; - if (sign_bits != 0 && sign_bits != mask(33)) - fault = Integer_Overflow_Fault; - Rc.sl = tmp<31:0>; - }}, IntMultOp); - 0x60: mulqv({{ - // 64-bit multiply with trap on overflow - uint64_t hi, lo; - mul128(Ra, Rb_or_imm, hi, lo); - // all the upper 64 bits must match the sign bit of - // the lower 64 - if (!((hi == 0 && lo<63:> == 0) || - (hi == mask(64) && lo<63:> == 1))) - fault = Integer_Overflow_Fault; - Rc = lo; - }}, IntMultOp); - } - - 0x1c: decode INTFUNC { - 0x00: decode RA { 31: sextb({{ Rc.sb = Rb_or_imm< 7:0>; }}); } - 0x01: decode RA { 31: sextw({{ Rc.sw = Rb_or_imm<15:0>; }}); } - 0x32: ctlz({{ - uint64_t count = 0; - uint64_t temp = Rb; - if (temp<63:32>) temp >>= 32; else count += 32; - if (temp<31:16>) temp >>= 16; else count += 16; - if (temp<15:8>) temp >>= 8; else count += 8; - if (temp<7:4>) temp >>= 4; else count += 4; - if (temp<3:2>) temp >>= 2; else count += 2; - if (temp<1:1>) temp >>= 1; else count += 1; - if ((temp<0:0>) != 0x1) count += 1; - Rc = count; - }}, IntAluOp); - - 0x33: cttz({{ - uint64_t count = 0; - uint64_t temp = Rb; - if (!(temp<31:0>)) { temp >>= 32; count += 32; } - if (!(temp<15:0>)) { temp >>= 16; count += 16; } - if (!(temp<7:0>)) { temp >>= 8; count += 8; } - if (!(temp<3:0>)) { temp >>= 4; count += 4; } - if (!(temp<1:0>)) { temp >>= 2; count += 2; } - if (!(temp<0:0> & ULL(0x1))) count += 1; - Rc = count; - }}, IntAluOp); - - format FailUnimpl { - 0x30: ctpop(); - 0x31: perr(); - 0x34: unpkbw(); - 0x35: unpkbl(); - 0x36: pkwb(); - 0x37: pklb(); - 0x38: minsb8(); - 0x39: minsw4(); - 0x3a: minub8(); - 0x3b: minuw4(); - 0x3c: maxub8(); - 0x3d: maxuw4(); - 0x3e: maxsb8(); - 0x3f: maxsw4(); - } - - format BasicOperateWithNopCheck { - 0x70: decode RB { - 31: ftoit({{ Rc = Fa.uq; }}, FloatCvtOp); - } - 0x78: decode RB { - 31: ftois({{ Rc.sl = t_to_s(Fa.uq); }}, - FloatCvtOp); - } - } - } - } - - // Conditional branches. - format CondBranch { - 0x39: beq({{ cond = (Ra == 0); }}); - 0x3d: bne({{ cond = (Ra != 0); }}); - 0x3e: bge({{ cond = (Ra.sq >= 0); }}); - 0x3f: bgt({{ cond = (Ra.sq > 0); }}); - 0x3b: ble({{ cond = (Ra.sq <= 0); }}); - 0x3a: blt({{ cond = (Ra.sq < 0); }}); - 0x38: blbc({{ cond = ((Ra & 1) == 0); }}); - 0x3c: blbs({{ cond = ((Ra & 1) == 1); }}); - - 0x31: fbeq({{ cond = (Fa == 0); }}); - 0x35: fbne({{ cond = (Fa != 0); }}); - 0x36: fbge({{ cond = (Fa >= 0); }}); - 0x37: fbgt({{ cond = (Fa > 0); }}); - 0x33: fble({{ cond = (Fa <= 0); }}); - 0x32: fblt({{ cond = (Fa < 0); }}); - } - - // unconditional branches - format UncondBranch { - 0x30: br(); - 0x34: bsr(IsCall); - } - - // indirect branches - 0x1a: decode JMPFUNC { - format Jump { - 0: jmp(); - 1: jsr(IsCall); - 2: ret(IsReturn); - 3: jsr_coroutine(IsCall, IsReturn); - } - } - - // Square root and integer-to-FP moves - 0x14: decode FP_SHORTFUNC { - // Integer to FP register moves must have RB == 31 - 0x4: decode RB { - 31: decode FP_FULLFUNC { - format BasicOperateWithNopCheck { - 0x004: itofs({{ Fc.uq = s_to_t(Ra.ul); }}, FloatCvtOp); - 0x024: itoft({{ Fc.uq = Ra.uq; }}, FloatCvtOp); - 0x014: FailUnimpl::itoff(); // VAX-format conversion - } - } - } - - // Square root instructions must have FA == 31 - 0xb: decode FA { - 31: decode FP_TYPEFUNC { - format FloatingPointOperate { -#if SS_COMPATIBLE_FP - 0x0b: sqrts({{ - if (Fb < 0.0) - fault = Arithmetic_Fault; - Fc = sqrt(Fb); - }}, FloatSqrtOp); -#else - 0x0b: sqrts({{ - if (Fb.sf < 0.0) - fault = Arithmetic_Fault; - Fc.sf = sqrt(Fb.sf); - }}, FloatSqrtOp); -#endif - 0x2b: sqrtt({{ - if (Fb < 0.0) - fault = Arithmetic_Fault; - Fc = sqrt(Fb); - }}, FloatSqrtOp); - } - } - } - - // VAX-format sqrtf and sqrtg are not implemented - 0xa: FailUnimpl::sqrtfg(); - } - - // IEEE floating point - 0x16: decode FP_SHORTFUNC_TOP2 { - // The top two bits of the short function code break this - // space into four groups: binary ops, compares, reserved, and - // conversions. See Table 4-12 of AHB. There are different - // special cases in these different groups, so we decode on - // these top two bits first just to select a decode strategy. - // Most of these instructions may have various trapping and - // rounding mode flags set; these are decoded in the - // FloatingPointDecode template used by the - // FloatingPointOperate format. - - // add/sub/mul/div: just decode on the short function code - // and source type. All valid trapping and rounding modes apply. - 0: decode FP_TRAPMODE { - // check for valid trapping modes here - 0,1,5,7: decode FP_TYPEFUNC { - format FloatingPointOperate { -#if SS_COMPATIBLE_FP - 0x00: adds({{ Fc = Fa + Fb; }}); - 0x01: subs({{ Fc = Fa - Fb; }}); - 0x02: muls({{ Fc = Fa * Fb; }}, FloatMultOp); - 0x03: divs({{ Fc = Fa / Fb; }}, FloatDivOp); -#else - 0x00: adds({{ Fc.sf = Fa.sf + Fb.sf; }}); - 0x01: subs({{ Fc.sf = Fa.sf - Fb.sf; }}); - 0x02: muls({{ Fc.sf = Fa.sf * Fb.sf; }}, FloatMultOp); - 0x03: divs({{ Fc.sf = Fa.sf / Fb.sf; }}, FloatDivOp); -#endif - - 0x20: addt({{ Fc = Fa + Fb; }}); - 0x21: subt({{ Fc = Fa - Fb; }}); - 0x22: mult({{ Fc = Fa * Fb; }}, FloatMultOp); - 0x23: divt({{ Fc = Fa / Fb; }}, FloatDivOp); - } - } - } - - // Floating-point compare instructions must have the default - // rounding mode, and may use the default trapping mode or - // /SU. Both trapping modes are treated the same by M5; the - // only difference on the real hardware (as far a I can tell) - // is that without /SU you'd get an imprecise trap if you - // tried to compare a NaN with something else (instead of an - // "unordered" result). - 1: decode FP_FULLFUNC { - format BasicOperateWithNopCheck { - 0x0a5, 0x5a5: cmpteq({{ Fc = (Fa == Fb) ? 2.0 : 0.0; }}, - FloatCmpOp); - 0x0a7, 0x5a7: cmptle({{ Fc = (Fa <= Fb) ? 2.0 : 0.0; }}, - FloatCmpOp); - 0x0a6, 0x5a6: cmptlt({{ Fc = (Fa < Fb) ? 2.0 : 0.0; }}, - FloatCmpOp); - 0x0a4, 0x5a4: cmptun({{ // unordered - Fc = (!(Fa < Fb) && !(Fa == Fb) && !(Fa > Fb)) ? 2.0 : 0.0; - }}, FloatCmpOp); - } - } - - // The FP-to-integer and integer-to-FP conversion insts - // require that FA be 31. - 3: decode FA { - 31: decode FP_TYPEFUNC { - format FloatingPointOperate { - 0x2f: decode FP_ROUNDMODE { - format FPFixedRounding { - // "chopped" i.e. round toward zero - 0: cvttq({{ Fc.sq = (int64_t)trunc(Fb); }}, - Chopped); - // round to minus infinity - 1: cvttq({{ Fc.sq = (int64_t)floor(Fb); }}, - MinusInfinity); - } - default: cvttq({{ Fc.sq = (int64_t)nearbyint(Fb); }}); - } - - // The cvtts opcode is overloaded to be cvtst if the trap - // mode is 2 or 6 (which are not valid otherwise) - 0x2c: decode FP_FULLFUNC { - format BasicOperateWithNopCheck { - // trap on denorm version "cvtst/s" is - // simulated same as cvtst - 0x2ac, 0x6ac: cvtst({{ Fc = Fb.sf; }}); - } - default: cvtts({{ Fc.sf = Fb; }}); - } - - // The trapping mode for integer-to-FP conversions - // must be /SUI or nothing; /U and /SU are not - // allowed. The full set of rounding modes are - // supported though. - 0x3c: decode FP_TRAPMODE { - 0,7: cvtqs({{ Fc.sf = Fb.sq; }}); - } - 0x3e: decode FP_TRAPMODE { - 0,7: cvtqt({{ Fc = Fb.sq; }}); - } - } - } - } - } - - // misc FP operate - 0x17: decode FP_FULLFUNC { - format BasicOperateWithNopCheck { - 0x010: cvtlq({{ - Fc.sl = (Fb.uq<63:62> << 30) | Fb.uq<58:29>; - }}); - 0x030: cvtql({{ - Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29); - }}); - - // We treat the precise & imprecise trapping versions of - // cvtql identically. - 0x130, 0x530: cvtqlv({{ - // To avoid overflow, all the upper 32 bits must match - // the sign bit of the lower 32. We code this as - // checking the upper 33 bits for all 0s or all 1s. - uint64_t sign_bits = Fb.uq<63:31>; - if (sign_bits != 0 && sign_bits != mask(33)) - fault = Integer_Overflow_Fault; - Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29); - }}); - - 0x020: cpys({{ // copy sign - Fc.uq = (Fa.uq<63:> << 63) | Fb.uq<62:0>; - }}); - 0x021: cpysn({{ // copy sign negated - Fc.uq = (~Fa.uq<63:> << 63) | Fb.uq<62:0>; - }}); - 0x022: cpyse({{ // copy sign and exponent - Fc.uq = (Fa.uq<63:52> << 52) | Fb.uq<51:0>; - }}); - - 0x02a: fcmoveq({{ Fc = (Fa == 0) ? Fb : Fc; }}); - 0x02b: fcmovne({{ Fc = (Fa != 0) ? Fb : Fc; }}); - 0x02c: fcmovlt({{ Fc = (Fa < 0) ? Fb : Fc; }}); - 0x02d: fcmovge({{ Fc = (Fa >= 0) ? Fb : Fc; }}); - 0x02e: fcmovle({{ Fc = (Fa <= 0) ? Fb : Fc; }}); - 0x02f: fcmovgt({{ Fc = (Fa > 0) ? Fb : Fc; }}); - - 0x024: mt_fpcr({{ FPCR = Fa.uq; }}); - 0x025: mf_fpcr({{ Fa.uq = FPCR; }}); - } - } - - // miscellaneous mem-format ops - 0x18: decode MEMFUNC { - format WarnUnimpl { - 0x8000: fetch(); - 0xa000: fetch_m(); - 0xe800: ecb(); - } - - format MiscPrefetch { - 0xf800: wh64({{ EA = Rb & ~ULL(63); }}, - {{ xc->writeHint(EA, 64, memAccessFlags); }}, - IsMemRef, IsDataPrefetch, IsStore, MemWriteOp, - NO_FAULT); - } - - format BasicOperate { - 0xc000: rpcc({{ -#if FULL_SYSTEM - /* Rb is a fake dependency so here is a fun way to get - * the parser to understand that. - */ - Ra = xc->readIpr(AlphaISA::IPR_CC, fault) + (Rb & 0); - -#else - Ra = curTick; -#endif - }}); - - // All of the barrier instructions below do nothing in - // their execute() methods (hence the empty code blocks). - // All of their functionality is hard-coded in the - // pipeline based on the flags IsSerializing, - // IsMemBarrier, and IsWriteBarrier. In the current - // detailed CPU model, the execute() function only gets - // called at fetch, so there's no way to generate pipeline - // behavior at any other stage. Once we go to an - // exec-in-exec CPU model we should be able to get rid of - // these flags and implement this behavior via the - // execute() methods. - - // trapb is just a barrier on integer traps, where excb is - // a barrier on integer and FP traps. "EXCB is thus a - // superset of TRAPB." (Alpha ARM, Sec 4.11.4) We treat - // them the same though. - 0x0000: trapb({{ }}, IsSerializing, No_OpClass); - 0x0400: excb({{ }}, IsSerializing, No_OpClass); - 0x4000: mb({{ }}, IsMemBarrier, MemReadOp); - 0x4400: wmb({{ }}, IsWriteBarrier, MemWriteOp); - } - -#if FULL_SYSTEM - format BasicOperate { - 0xe000: rc({{ - Ra = xc->readIntrFlag(); - xc->setIntrFlag(0); - }}, IsNonSpeculative); - 0xf000: rs({{ - Ra = xc->readIntrFlag(); - xc->setIntrFlag(1); - }}, IsNonSpeculative); - } -#else - format FailUnimpl { - 0xe000: rc(); - 0xf000: rs(); - } -#endif - } - -#if FULL_SYSTEM - 0x00: CallPal::call_pal({{ - if (!palValid || - (palPriv - && xc->readIpr(AlphaISA::IPR_ICM, fault) != AlphaISA::mode_kernel)) { - // invalid pal function code, or attempt to do privileged - // PAL call in non-kernel mode - fault = Unimplemented_Opcode_Fault; - } - else { - // check to see if simulator wants to do something special - // on this PAL call (including maybe suppress it) - bool dopal = xc->simPalCheck(palFunc); - - if (dopal) { - AlphaISA::swap_palshadow(&xc->xcBase()->regs, true); - xc->setIpr(AlphaISA::IPR_EXC_ADDR, NPC); - NPC = xc->readIpr(AlphaISA::IPR_PAL_BASE, fault) + palOffset; - } - } - }}, IsNonSpeculative); -#else - 0x00: decode PALFUNC { - format EmulatedCallPal { - 0x00: halt ({{ - SimExit(curTick, "halt instruction encountered"); - }}, IsNonSpeculative); - 0x83: callsys({{ - xc->syscall(); - }}, IsNonSpeculative); - // Read uniq reg into ABI return value register (r0) - 0x9e: rduniq({{ R0 = Runiq; }}); - // Write uniq reg with value from ABI arg register (r16) - 0x9f: wruniq({{ Runiq = R16; }}); - } - } -#endif - -#if FULL_SYSTEM - format HwLoadStore { - 0x1b: decode HW_LDST_QUAD { - 0: hw_ld({{ EA = (Rb + disp) & ~3; }}, {{ Ra = Mem.ul; }}, L); - 1: hw_ld({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem.uq; }}, Q); - } - - 0x1f: decode HW_LDST_COND { - 0: decode HW_LDST_QUAD { - 0: hw_st({{ EA = (Rb + disp) & ~3; }}, - {{ Mem.ul = Ra<31:0>; }}, L); - 1: hw_st({{ EA = (Rb + disp) & ~7; }}, - {{ Mem.uq = Ra.uq; }}, Q); - } - - 1: FailUnimpl::hw_st_cond(); - } - } - - format HwMoveIPR { - 0x19: hw_mfpr({{ - // this instruction is only valid in PAL mode - if (!xc->inPalMode()) { - fault = Unimplemented_Opcode_Fault; - } - else { - Ra = xc->readIpr(ipr_index, fault); - } - }}); - 0x1d: hw_mtpr({{ - // this instruction is only valid in PAL mode - if (!xc->inPalMode()) { - fault = Unimplemented_Opcode_Fault; - } - else { - xc->setIpr(ipr_index, Ra); - if (traceData) { traceData->setData(Ra); } - } - }}); - } - - format BasicOperate { - 0x1e: hw_rei({{ xc->hwrei(); }}, IsSerializing); - - // M5 special opcodes use the reserved 0x01 opcode space - 0x01: decode M5FUNC { - 0x00: arm({{ - AlphaPseudo::arm(xc->xcBase()); - }}, IsNonSpeculative); - 0x01: quiesce({{ - AlphaPseudo::quiesce(xc->xcBase()); - }}, IsNonSpeculative); - 0x10: ivlb({{ - AlphaPseudo::ivlb(xc->xcBase()); - }}, No_OpClass, IsNonSpeculative); - 0x11: ivle({{ - AlphaPseudo::ivle(xc->xcBase()); - }}, No_OpClass, IsNonSpeculative); - 0x20: m5exit_old({{ - AlphaPseudo::m5exit_old(xc->xcBase()); - }}, No_OpClass, IsNonSpeculative); - 0x21: m5exit({{ - AlphaPseudo::m5exit(xc->xcBase()); - }}, No_OpClass, IsNonSpeculative); - 0x30: initparam({{ Ra = xc->xcBase()->cpu->system->init_param; }}); - 0x40: resetstats({{ - AlphaPseudo::resetstats(xc->xcBase()); - }}, IsNonSpeculative); - 0x41: dumpstats({{ - AlphaPseudo::dumpstats(xc->xcBase()); - }}, IsNonSpeculative); - 0x42: dumpresetstats({{ - AlphaPseudo::dumpresetstats(xc->xcBase()); - }}, IsNonSpeculative); - 0x43: m5checkpoint({{ - AlphaPseudo::m5checkpoint(xc->xcBase()); - }}, IsNonSpeculative); - 0x50: m5readfile({{ - AlphaPseudo::readfile(xc->xcBase()); - }}, IsNonSpeculative); - 0x51: m5break({{ - AlphaPseudo::debugbreak(xc->xcBase()); - }}, IsNonSpeculative); - 0x52: m5switchcpu({{ - AlphaPseudo::switchcpu(xc->xcBase()); - }}, IsNonSpeculative); - 0x53: m5addsymbol({{ - AlphaPseudo::addsymbol(xc->xcBase()); - }}, IsNonSpeculative); - - } - } -#endif -} diff --git a/arch/alpha/isa_traits.hh b/arch/alpha/isa_traits.hh index 2570d12f0..a6e34acbb 100644 --- a/arch/alpha/isa_traits.hh +++ b/arch/alpha/isa_traits.hh @@ -29,10 +29,14 @@ #ifndef __ARCH_ALPHA_ISA_TRAITS_HH__ #define __ARCH_ALPHA_ISA_TRAITS_HH__ -#include "arch/alpha/faults.hh" +namespace LittleEndianGuest {} +using namespace LittleEndianGuest; + +//#include "arch/alpha/faults.hh" #include "base/misc.hh" #include "config/full_system.hh" #include "sim/host.hh" +#include "sim/faults.hh" class FastCPU; class FullCPU; diff --git a/arch/alpha/pseudo_inst.cc b/arch/alpha/pseudo_inst.cc deleted file mode 100644 index e105b3cc8..000000000 --- a/arch/alpha/pseudo_inst.cc +++ /dev/null @@ -1,256 +0,0 @@ -/* - * Copyright (c) 2003-2005 The Regents of The University of Michigan - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer; - * redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution; - * neither the name of the copyright holders nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -#include <errno.h> -#include <fcntl.h> -#include <unistd.h> -#include <cstdio> - -#include <string> - -#include "arch/alpha/pseudo_inst.hh" -#include "arch/alpha/vtophys.hh" -#include "cpu/base.hh" -#include "cpu/sampler/sampler.hh" -#include "cpu/exec_context.hh" -#include "kern/kernel_stats.hh" -#include "sim/param.hh" -#include "sim/serialize.hh" -#include "sim/sim_exit.hh" -#include "sim/stat_control.hh" -#include "sim/stats.hh" -#include "sim/system.hh" -#include "sim/debug.hh" -#include "targetarch/vptr.hh" - -using namespace std; - -extern Sampler *SampCPU; - -using namespace Stats; - -namespace AlphaPseudo -{ - bool doStatisticsInsts; - bool doCheckpointInsts; - bool doQuiesce; - - void - arm(ExecContext *xc) - { - xc->kernelStats->arm(); - } - - void - quiesce(ExecContext *xc) - { - if (!doQuiesce) - return; - - xc->suspend(); - xc->kernelStats->quiesce(); - } - - void - ivlb(ExecContext *xc) - { - xc->kernelStats->ivlb(); - } - - void - ivle(ExecContext *xc) - { - } - - void - m5exit_old(ExecContext *xc) - { - SimExit(curTick, "m5_exit_old instruction encountered"); - } - - void - m5exit(ExecContext *xc) - { - Tick delay = xc->regs.intRegFile[16]; - Tick when = curTick + delay * Clock::Int::ns; - SimExit(when, "m5_exit instruction encountered"); - } - - void - resetstats(ExecContext *xc) - { - if (!doStatisticsInsts) - return; - - Tick delay = xc->regs.intRegFile[16]; - Tick period = xc->regs.intRegFile[17]; - - Tick when = curTick + delay * Clock::Int::ns; - Tick repeat = period * Clock::Int::ns; - - using namespace Stats; - SetupEvent(Reset, when, repeat); - } - - void - dumpstats(ExecContext *xc) - { - if (!doStatisticsInsts) - return; - - Tick delay = xc->regs.intRegFile[16]; - Tick period = xc->regs.intRegFile[17]; - - Tick when = curTick + delay * Clock::Int::ns; - Tick repeat = period * Clock::Int::ns; - - using namespace Stats; - SetupEvent(Dump, when, repeat); - } - - void - addsymbol(ExecContext *xc) - { - Addr addr = xc->regs.intRegFile[16]; - char symb[100]; - CopyString(xc, symb, xc->regs.intRegFile[17], 100); - std::string symbol(symb); - - DPRINTF(Loader, "Loaded symbol: %s @ %#llx\n", symbol, addr); - - xc->system->kernelSymtab->insert(addr,symbol); - } - - void - dumpresetstats(ExecContext *xc) - { - if (!doStatisticsInsts) - return; - - Tick delay = xc->regs.intRegFile[16]; - Tick period = xc->regs.intRegFile[17]; - - Tick when = curTick + delay * Clock::Int::ns; - Tick repeat = period * Clock::Int::ns; - - using namespace Stats; - SetupEvent(Dump|Reset, when, repeat); - } - - void - m5checkpoint(ExecContext *xc) - { - if (!doCheckpointInsts) - return; - - Tick delay = xc->regs.intRegFile[16]; - Tick period = xc->regs.intRegFile[17]; - - Tick when = curTick + delay * Clock::Int::ns; - Tick repeat = period * Clock::Int::ns; - - Checkpoint::setup(when, repeat); - } - - void - readfile(ExecContext *xc) - { - const string &file = xc->cpu->system->params->readfile; - if (file.empty()) { - xc->regs.intRegFile[0] = ULL(0); - return; - } - - Addr vaddr = xc->regs.intRegFile[16]; - uint64_t len = xc->regs.intRegFile[17]; - uint64_t offset = xc->regs.intRegFile[18]; - uint64_t result = 0; - - int fd = ::open(file.c_str(), O_RDONLY, 0); - if (fd < 0) - panic("could not open file %s\n", file); - - if (::lseek(fd, offset, SEEK_SET) < 0) - panic("could not seek: %s", strerror(errno)); - - char *buf = new char[len]; - char *p = buf; - while (len > 0) { - int bytes = ::read(fd, p, len); - if (bytes <= 0) - break; - - p += bytes; - result += bytes; - len -= bytes; - } - - close(fd); - CopyIn(xc, vaddr, buf, result); - delete [] buf; - xc->regs.intRegFile[0] = result; - } - - class Context : public ParamContext - { - public: - Context(const string §ion) : ParamContext(section) {} - void checkParams(); - }; - - Context context("pseudo_inst"); - - Param<bool> __quiesce(&context, "quiesce", - "enable quiesce instructions", - true); - Param<bool> __statistics(&context, "statistics", - "enable statistics pseudo instructions", - true); - Param<bool> __checkpoint(&context, "checkpoint", - "enable checkpoint pseudo instructions", - true); - - void - Context::checkParams() - { - doQuiesce = __quiesce; - doStatisticsInsts = __statistics; - doCheckpointInsts = __checkpoint; - } - - void debugbreak(ExecContext *xc) - { - debug_break(); - } - - void switchcpu(ExecContext *xc) - { - if (SampCPU) - SampCPU->switchCPUs(); - } -} diff --git a/arch/alpha/pseudo_inst.hh b/arch/alpha/pseudo_inst.hh deleted file mode 100644 index 0e7462a56..000000000 --- a/arch/alpha/pseudo_inst.hh +++ /dev/null @@ -1,54 +0,0 @@ -/* - * Copyright (c) 2003-2005 The Regents of The University of Michigan - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer; - * redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution; - * neither the name of the copyright holders nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -class ExecContext; - -namespace AlphaPseudo -{ - /** - * @todo these externs are only here for a hack in fullCPU::takeOver... - */ - extern bool doStatisticsInsts; - extern bool doCheckpointInsts; - extern bool doQuiesce; - - void arm(ExecContext *xc); - void quiesce(ExecContext *xc); - void ivlb(ExecContext *xc); - void ivle(ExecContext *xc); - void m5exit(ExecContext *xc); - void m5exit_old(ExecContext *xc); - void resetstats(ExecContext *xc); - void dumpstats(ExecContext *xc); - void dumpresetstats(ExecContext *xc); - void m5checkpoint(ExecContext *xc); - void readfile(ExecContext *xc); - void debugbreak(ExecContext *xc); - void switchcpu(ExecContext *xc); - void addsymbol(ExecContext *xc); -} diff --git a/arch/alpha/vptr.hh b/arch/alpha/vptr.hh deleted file mode 100644 index d1bda4061..000000000 --- a/arch/alpha/vptr.hh +++ /dev/null @@ -1,114 +0,0 @@ -/* - * Copyright (c) 2004-2005 The Regents of The University of Michigan - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer; - * redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution; - * neither the name of the copyright holders nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -#ifndef __ARCH_ALPHA_VPTR_HH__ -#define __ARCH_ALPHA_VPTR_HH__ - -#include "arch/alpha/vtophys.hh" - -class ExecContext; - -template <class T> -class VPtr -{ - public: - typedef T Type; - - private: - ExecContext *xc; - Addr ptr; - - public: - ExecContext *GetXC() const { return xc; } - Addr GetPointer() const { return ptr; } - - public: - explicit VPtr(ExecContext *_xc, Addr p = 0) : xc(_xc), ptr(p) { } - template <class U> - VPtr(const VPtr<U> &vp) : xc(vp.GetXC()), ptr(vp.GetPointer()) {} - ~VPtr() {} - - bool operator!() const - { - return ptr == 0; - } - - VPtr<T> operator+(int offset) - { - VPtr<T> ptr(*this); - ptr += offset; - - return ptr; - } - - const VPtr<T> &operator+=(int offset) - { - ptr += offset; - assert((ptr & (AlphaISA::PageBytes - 1)) + sizeof(T) - < AlphaISA::PageBytes); - - return *this; - } - - const VPtr<T> &operator=(Addr p) - { - assert((p & (AlphaISA::PageBytes)) + sizeof(T) < AlphaISA::PageBytes); - ptr = p; - - return *this; - } - - template <class U> - const VPtr<T> &operator=(const VPtr<U> &vp) - { - xc = vp.GetXC(); - ptr = vp.GetPointer(); - - return *this; - } - - operator T *() - { - void *addr = vtomem(xc, ptr, sizeof(T)); - return (T *)addr; - } - - T *operator->() - { - void *addr = vtomem(xc, ptr, sizeof(T)); - return (T *)addr; - } - - T &operator*() - { - void *addr = vtomem(xc, ptr, sizeof(T)); - return *(T *)addr; - } -}; - -#endif // __ARCH_ALPHA_VPTR_HH__ |