cpu/intel/car: Prepare for some POSTCAR_STAGE support

The file cache_as_ram_ht.inc is used across a variety
of CPUs and northbridges. We need to split it anyway
for future C_ENVIRONMENT_BOOTBLOCK and verstage work.

Split and rename the files, remove code that is globally
implemented in POSTCAR_STAGE framework already.

Change-Id: I2ba67772328fce3d5d1ae34c36aea8dcdcc56b87
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/26747
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>

8 files changed, 449 insertions, 5 deletions

diff --git a/src/cpu/intel/car/p4-netburst/cache_as_ram.S b/src/cpu/intel/car/p4-netburst/cache_as_ram.S
new file mode 100644
index 0000000000..58782b941d
--- /dev/null
+++ b/src/cpu/intel/car/p4-netburst/cache_as_ram.S
@@ -0,0 +1,376 @@
+/*
+ * This file is part of the coreboot project.
+ *
+ * Copyright (C) 2000,2007 Ronald G. Minnich <rminnich@gmail.com>
+ * Copyright (C) 2005 Tyan (written by Yinghai Lu for Tyan)
+ * Copyright (C) 2007-2008 coresystems GmbH
+ * Copyright (C) 2012 Kyösti Mälkki <kyosti.malkki@gmail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <cpu/x86/mtrr.h>
+#include <cpu/x86/cache.h>
+#include <cpu/x86/post_code.h>
+#include <cpu/x86/lapic_def.h>
+
+/* Macro to access Local APIC registers at default base. */
+#define LAPIC(x)		$(LAPIC_DEFAULT_BASE | LAPIC_ ## x)
+#define START_IPI_VECTOR	((CONFIG_AP_SIPI_VECTOR >> 12) & 0xff)
+
+#define CACHE_AS_RAM_SIZE CONFIG_DCACHE_RAM_SIZE
+#define CACHE_AS_RAM_BASE CONFIG_DCACHE_RAM_BASE
+
+.code32
+_cache_as_ram_setup:
+
+	/* Save the BIST result. */
+	movl	%eax, %ebp
+
+cache_as_ram:
+	post_code(0x20)
+
+	movl	$LAPIC_BASE_MSR, %ecx
+	rdmsr
+	andl	$LAPIC_BASE_MSR_BOOTSTRAP_PROCESSOR, %eax
+	jz	ap_init
+
+	/* Clear/disable fixed MTRRs */
+	mov	$fixed_mtrr_list_size, %ebx
+	xor	%eax, %eax
+	xor	%edx, %edx
+
+clear_fixed_mtrr:
+	add	$-2, %ebx
+	movzwl	fixed_mtrr_list(%ebx), %ecx
+	wrmsr
+	jnz	clear_fixed_mtrr
+
+	/* Figure put how many MTRRs we have, and clear them out */
+	mov	$MTRR_CAP_MSR, %ecx
+	rdmsr
+	movzb	%al, %ebx		/* Number of variable MTRRs */
+	mov	$MTRR_PHYS_BASE(0), %ecx
+	xor	%eax, %eax
+	xor	%edx, %edx
+
+clear_var_mtrr:
+	wrmsr
+	inc	%ecx
+	wrmsr
+	inc	%ecx
+	dec	%ebx
+	jnz	clear_var_mtrr
+	post_code(0x21)
+
+	/* Configure the default memory type to uncacheable. */
+	movl	$MTRR_DEF_TYPE_MSR, %ecx
+	rdmsr
+	andl	$(~0x00000cff), %eax
+	wrmsr
+
+	post_code(0x22)
+
+	/* Determine CPU_ADDR_BITS and load PHYSMASK high
+	 * word to %edx.
+	 */
+	movl	$0x80000000, %eax
+	cpuid
+	cmpl	$0x80000008, %eax
+	jc	addrsize_no_MSR
+	movl	$0x80000008, %eax
+	cpuid
+	movb	%al, %cl
+	sub	$32, %cl
+	movl	$1, %edx
+	shl	%cl, %edx
+	subl	$1, %edx
+	jmp	addrsize_set_high
+addrsize_no_MSR:
+	movl	$1, %eax
+	cpuid
+	andl	$(1 << 6 | 1 << 17), %edx	/* PAE or PSE36 */
+	jz	addrsize_set_high
+	movl	$0x0f, %edx
+
+	/* Preload high word of address mask (in %edx) for Variable
+	 * MTRRs 0 and 1 and enable local APIC at default base.
+	 */
+addrsize_set_high:
+	xorl	%eax, %eax
+	movl	$MTRR_PHYS_MASK(0), %ecx
+	wrmsr
+	movl	$MTRR_PHYS_MASK(1), %ecx
+	wrmsr
+	movl	$LAPIC_BASE_MSR, %ecx
+	not	%edx
+	movl	%edx, %ebx
+	rdmsr
+	andl	%ebx, %edx
+	andl	$(~LAPIC_BASE_MSR_ADDR_MASK), %eax
+	orl	$(LAPIC_DEFAULT_BASE | LAPIC_BASE_MSR_ENABLE), %eax
+	wrmsr
+
+bsp_init:
+
+	post_code(0x23)
+
+	/* Send INIT IPI to all excluding ourself. */
+	movl	LAPIC(ICR), %edi
+	movl	$(LAPIC_DEST_ALLBUT | LAPIC_INT_ASSERT | LAPIC_DM_INIT), %eax
+1:	movl	%eax, (%edi)
+	movl	$0x30, %ecx
+2:	pause
+	dec	%ecx
+	jnz	2b
+	movl	(%edi), %ecx
+	andl	$LAPIC_ICR_BUSY, %ecx
+	jnz	1b
+
+	post_code(0x24)
+
+	movl	$1, %eax
+	cpuid
+	btl	$28, %edx
+	jnc	sipi_complete
+	bswapl	%ebx
+	movzx	%bh, %edi
+	cmpb	$1, %bh
+	jbe	sipi_complete	/* only one LAPIC ID in package */
+
+	movl	$0, %eax
+	cpuid
+	movb	$1, %bl
+	cmpl	$4, %eax
+	jb	cores_counted
+	movl	$4, %eax
+	movl	$0, %ecx
+	cpuid
+	shr	$26, %eax
+	movb	%al, %bl
+	inc	%bl
+
+cores_counted:
+	movl	%edi, %eax
+	divb	%bl
+	cmpb	$1, %al
+	jbe	sipi_complete	/* only LAPIC ID of a core */
+
+	/* For a hyper-threading processor, cache must not be disabled
+	 * on an AP on the same physical package with the BSP.
+	 */
+
+hyper_threading_cpu:
+
+	/* delay 10 ms */
+	movl	$10000, %ecx
+1:	inb	$0x80, %al
+	dec	%ecx
+	jnz	1b
+
+	post_code(0x25)
+
+	/* Send Start IPI to all excluding ourself. */
+	movl	LAPIC(ICR), %edi
+	movl	$(LAPIC_DEST_ALLBUT | LAPIC_DM_STARTUP | START_IPI_VECTOR), %eax
+1:	movl	%eax, (%edi)
+	movl	$0x30, %ecx
+2:	pause
+	dec	%ecx
+	jnz	2b
+	movl	(%edi), %ecx
+	andl	$LAPIC_ICR_BUSY, %ecx
+	jnz	1b
+
+	/* delay 250 us */
+	movl	$250, %ecx
+1:	inb	$0x80, %al
+	dec	%ecx
+	jnz	1b
+
+	post_code(0x26)
+
+	/* Wait for sibling CPU to start. */
+1:	movl	$(MTRR_PHYS_BASE(0)), %ecx
+	rdmsr
+	andl	%eax, %eax
+	jnz	sipi_complete
+
+	movl	$0x30, %ecx
+2:	pause
+	dec	%ecx
+	jnz	2b
+	jmp	1b
+
+
+ap_init:
+	post_code(0x27)
+
+	/* Do not disable cache (so BSP can enable it). */
+	movl	%cr0, %eax
+	andl	$(~(CR0_CacheDisable | CR0_NoWriteThrough)), %eax
+	movl	%eax, %cr0
+
+	post_code(0x28)
+
+	/* MTRR registers are shared between HT siblings. */
+	movl	$(MTRR_PHYS_BASE(0)), %ecx
+	movl	$(1 << 12), %eax
+	xorl	%edx, %edx
+	wrmsr
+
+	post_code(0x29)
+
+ap_halt:
+	cli
+1:	hlt
+	jmp	1b
+
+
+
+sipi_complete:
+
+	post_code(0x2a)
+
+	/* Set Cache-as-RAM base address. */
+	movl	$(MTRR_PHYS_BASE(0)), %ecx
+	movl	$(CACHE_AS_RAM_BASE | MTRR_TYPE_WRBACK), %eax
+	xorl	%edx, %edx
+	wrmsr
+
+	/* Set Cache-as-RAM mask. */
+	movl	$(MTRR_PHYS_MASK(0)), %ecx
+	rdmsr
+	movl	$(~(CACHE_AS_RAM_SIZE - 1) | MTRR_PHYS_MASK_VALID), %eax
+	wrmsr
+
+	post_code(0x2b)
+
+	/* Enable MTRR. */
+	movl	$MTRR_DEF_TYPE_MSR, %ecx
+	rdmsr
+	orl	$MTRR_DEF_TYPE_EN, %eax
+	wrmsr
+
+	/* Enable L2 cache Write-Back (WBINVD and FLUSH#).
+	 *
+	 * MSR is set when DisplayFamily_DisplayModel is one of:
+	 * 06_0x, 06_17, 06_1C
+	 *
+	 * Description says this bit enables use of WBINVD and FLUSH#.
+	 * Should this be set only after the system bus and/or memory
+	 * controller can successfully handle write cycles?
+	 */
+
+#define EAX_FAMILY(a)	(a << 8)	/* for family <= 0fH */
+#define EAX_MODEL(a)	(((a & 0xf0) << 12) | ((a & 0xf) << 4))
+
+	movl	$1, %eax
+	cpuid
+	movl	%eax, %ebx
+	andl	$EAX_FAMILY(0x0f), %eax
+	cmpl	$EAX_FAMILY(0x06), %eax
+	jne	no_msr_11e
+	movl	%ebx, %eax
+	andl	$EAX_MODEL(0xff), %eax
+	cmpl	$EAX_MODEL(0x17), %eax
+	je	has_msr_11e
+	cmpl	$EAX_MODEL(0x1c), %eax
+	je	has_msr_11e
+	andl	$EAX_MODEL(0xf0), %eax
+	cmpl	$EAX_MODEL(0x00), %eax
+	jne	no_msr_11e
+has_msr_11e:
+	movl	$0x11e, %ecx
+	rdmsr
+	orl	$(1 << 8), %eax
+	wrmsr
+no_msr_11e:
+
+	post_code(0x2c)
+
+	/* Enable cache (CR0.CD = 0, CR0.NW = 0). */
+	movl	%cr0, %eax
+	andl	$(~(CR0_CacheDisable | CR0_NoWriteThrough)), %eax
+	invd
+	movl	%eax, %cr0
+
+	/* Clear the cache memory region. This will also fill up the cache. */
+	cld
+	xorl	%eax, %eax
+	movl	$CACHE_AS_RAM_BASE, %edi
+	movl	$(CACHE_AS_RAM_SIZE >> 2), %ecx
+	rep	stosl
+
+	post_code(0x2d)
+	/* Enable Cache-as-RAM mode by disabling cache. */
+	movl	%cr0, %eax
+	orl	$CR0_CacheDisable, %eax
+	movl	%eax, %cr0
+
+	/* Enable cache for our code in Flash because we do XIP here */
+	movl	$MTRR_PHYS_BASE(1), %ecx
+	xorl	%edx, %edx
+	/*
+	 * IMPORTANT: The following calculation _must_ be done at runtime. See
+	 * https://www.coreboot.org/pipermail/coreboot/2010-October/060855.html
+	 */
+	movl	$copy_and_run, %eax
+	andl	$(~(CONFIG_XIP_ROM_SIZE - 1)), %eax
+	orl	$MTRR_TYPE_WRPROT, %eax
+	wrmsr
+
+	movl	$MTRR_PHYS_MASK(1), %ecx
+	rdmsr
+	movl	$(~(CONFIG_XIP_ROM_SIZE - 1) | MTRR_PHYS_MASK_VALID), %eax
+	wrmsr
+
+	post_code(0x2e)
+	/* Enable cache. */
+	movl	%cr0, %eax
+	andl	$(~(CR0_CacheDisable | CR0_NoWriteThrough)), %eax
+	movl	%eax, %cr0
+
+	/* Setup the stack. */
+	movl	$(CONFIG_DCACHE_RAM_BASE + CONFIG_DCACHE_RAM_SIZE), %eax
+	movl	%eax, %esp
+
+	/* Restore the BIST result. */
+	movl	%ebp, %eax
+	movl	%esp, %ebp
+	pushl	%eax
+
+before_romstage:
+	post_code(0x2f)
+	/* Call romstage.c main function. */
+	call	romstage_main
+
+	/* Should never see this postcode */
+	post_code(POST_DEAD_CODE)
+
+.Lhlt:
+	hlt
+	jmp	.Lhlt
+
+fixed_mtrr_list:
+	.word	MTRR_FIX_64K_00000
+	.word	MTRR_FIX_16K_80000
+	.word	MTRR_FIX_16K_A0000
+	.word	MTRR_FIX_4K_C0000
+	.word	MTRR_FIX_4K_C8000
+	.word	MTRR_FIX_4K_D0000
+	.word	MTRR_FIX_4K_D8000
+	.word	MTRR_FIX_4K_E0000
+	.word	MTRR_FIX_4K_E8000
+	.word	MTRR_FIX_4K_F0000
+	.word	MTRR_FIX_4K_F8000
+fixed_mtrr_list_size = . - fixed_mtrr_list
+
+_cache_as_ram_setup_end:
diff --git a/src/cpu/intel/car/p4-netburst/exit_car.S b/src/cpu/intel/car/p4-netburst/exit_car.S
new file mode 100644
index 0000000000..3b991288ff
--- /dev/null
+++ b/src/cpu/intel/car/p4-netburst/exit_car.S
@@ -0,0 +1,45 @@
+/*
+ * This file is part of the coreboot project.
+ *
+ * Copyright (C) 2000,2007 Ronald G. Minnich <rminnich@gmail.com>
+ * Copyright (C) 2007-2008 coresystems GmbH
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <cpu/x86/mtrr.h>
+#include <cpu/x86/cache.h>
+#include <cpu/x86/post_code.h>
+
+.code32
+.global chipset_teardown_car
+
+chipset_teardown_car:
+	pop %esp
+
+	post_code(0x30)
+
+	/* Disable cache. */
+	movl	%cr0, %eax
+	orl	$CR0_CacheDisable, %eax
+	movl	%eax, %cr0
+
+	post_code(0x31)
+
+	/* Disable MTRR. */
+	movl	$MTRR_DEF_TYPE_MSR, %ecx
+	rdmsr
+	andl	$(~MTRR_DEF_TYPE_EN), %eax
+	wrmsr
+
+	post_code(0x32)
+
+	/* Return to caller. */
+	jmp	*%esp
diff --git a/src/cpu/intel/car/romstage.c b/src/cpu/intel/car/romstage.c
index 37e39428e2..c36e0468e7 100644
--- a/src/cpu/intel/car/romstage.c
+++ b/src/cpu/intel/car/romstage.c
@@ -23,7 +23,6 @@
 asmlinkage void *romstage_main(unsigned long bist)
 {
 	int i;
-	void *romstage_stack_after_car;
 	const int num_guards = 4;
 	const u32 stack_guard = 0xdeadbeef;
 	u32 *stack_base;
@@ -52,10 +51,13 @@ asmlinkage void *romstage_main(unsigned long bist)
 		printk(BIOS_DEBUG, "Smashed stack detected in romstage!\n");
 	}
 
-	/* Get the stack to use after cache-as-ram is torn down. */
-	romstage_stack_after_car = setup_stack_and_mtrrs();
+	if (!IS_ENABLED(CONFIG_POSTCAR_STAGE))
+		return setup_stack_and_mtrrs();
 
-	return romstage_stack_after_car;
+	platform_enter_postcar();
+
+	/* We do not return. */
+	return NULL;
 }
 
 asmlinkage void romstage_after_car(void)
diff --git a/src/cpu/intel/socket_441/Makefile.inc b/src/cpu/intel/socket_441/Makefile.inc
index dbf300b0d2..7c37019194 100644
--- a/src/cpu/intel/socket_441/Makefile.inc
+++ b/src/cpu/intel/socket_441/Makefile.inc
@@ -8,5 +8,10 @@ subdirs-y += ../microcode
 subdirs-y += ../hyperthreading
 subdirs-y += ../speedstep
 
+ifneq ($(CONFIG_POSTCAR_STAGE),y)
 cpu_incs-y += $(src)/cpu/intel/car/cache_as_ram_ht.inc
+else
+cpu_incs-y += $(src)/cpu/intel/car/p4-netburst/cache_as_ram.S
+postcar-y += ../car/p4-netburst/exit_car.S
+endif
 romstage-y += ../car/romstage.c
diff --git a/src/cpu/intel/socket_FCBGA559/Makefile.inc b/src/cpu/intel/socket_FCBGA559/Makefile.inc
index dbf300b0d2..7c37019194 100644
--- a/src/cpu/intel/socket_FCBGA559/Makefile.inc
+++ b/src/cpu/intel/socket_FCBGA559/Makefile.inc
@@ -8,5 +8,10 @@ subdirs-y += ../microcode
 subdirs-y += ../hyperthreading
 subdirs-y += ../speedstep
 
+ifneq ($(CONFIG_POSTCAR_STAGE),y)
 cpu_incs-y += $(src)/cpu/intel/car/cache_as_ram_ht.inc
+else
+cpu_incs-y += $(src)/cpu/intel/car/p4-netburst/cache_as_ram.S
+postcar-y += ../car/p4-netburst/exit_car.S
+endif
 romstage-y += ../car/romstage.c
diff --git a/src/cpu/intel/socket_LGA775/Makefile.inc b/src/cpu/intel/socket_LGA775/Makefile.inc
index ffcd1cb33e..7ff2f33f3f 100644
--- a/src/cpu/intel/socket_LGA775/Makefile.inc
+++ b/src/cpu/intel/socket_LGA775/Makefile.inc
@@ -13,6 +13,10 @@ subdirs-y += ../microcode
 subdirs-y += ../hyperthreading
 subdirs-y += ../speedstep
 
+ifneq ($(CONFIG_POSTCAR_STAGE),y)
 cpu_incs-y += $(src)/cpu/intel/car/cache_as_ram_ht.inc
-romstage-y += ../car/romstage.c
+else
+cpu_incs-y += $(src)/cpu/intel/car/p4-netburst/cache_as_ram.S
+postcar-y += ../car/p4-netburst/exit_car.S
+endif
 romstage-y += ../car/romstage.c
diff --git a/src/cpu/intel/socket_mPGA604/Makefile.inc b/src/cpu/intel/socket_mPGA604/Makefile.inc
index 98306d4fa6..717c5c2b6a 100644
--- a/src/cpu/intel/socket_mPGA604/Makefile.inc
+++ b/src/cpu/intel/socket_mPGA604/Makefile.inc
@@ -9,5 +9,10 @@ subdirs-y += ../../x86/smm
 subdirs-y += ../microcode
 subdirs-y += ../hyperthreading
 
+ifneq ($(CONFIG_POSTCAR_STAGE),y)
 cpu_incs-y += $(src)/cpu/intel/car/cache_as_ram_ht.inc
+else
+cpu_incs-y += $(src)/cpu/intel/car/p4-netburst/cache_as_ram.S
+postcar-y += ../car/p4-netburst/exit_car.S
+endif
 romstage-y += ../car/romstage.c
diff --git a/src/include/cpu/intel/romstage.h b/src/include/cpu/intel/romstage.h
index 3a9e98934b..eace57e558 100644
--- a/src/include/cpu/intel/romstage.h
+++ b/src/include/cpu/intel/romstage.h
@@ -22,6 +22,8 @@ void mainboard_romstage_entry(unsigned long bist);
  */
 void *setup_stack_and_mtrrs(void);
 
+void platform_enter_postcar(void);
+
 /* romstage_main is called from the cache-as-ram assembly file to prepare
  * CAR stack guards.*/
 asmlinkage void *romstage_main(unsigned long bist);