x86: chromebook_link: Implement CAR support (cache as RAM)

Add support for CAR so that we have memory to use prior to DRAM init.
On link there is a total of 128KB of CAR available, although some is
used for the memory reference code.

Signed-off-by: Simon Glass <sjg@chromium.org>
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index aa7cf77..f27d731 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -34,6 +34,22 @@
 
 endchoice
 
+config RAMBASE
+	hex
+	default 0x100000
+
+config RAMTOP
+	hex
+	default 0x200000
+
+config XIP_ROM_SIZE
+	hex
+	default 0x10000
+
+config CPU_ADDR_BITS
+	int
+	default 36
+
 config ROM_SIZE
 	hex
 	default 0x800000
diff --git a/arch/x86/cpu/ivybridge/car.S b/arch/x86/cpu/ivybridge/car.S
index 0480813..dca68e4 100644
--- a/arch/x86/cpu/ivybridge/car.S
+++ b/arch/x86/cpu/ivybridge/car.S
@@ -12,9 +12,167 @@
  */
 
 #include <common.h>
+#include <asm/mtrr.h>
+#include <asm/post.h>
+#include <asm/processor-flags.h>
 
-	/* Note: ebp must not be touched in this code */
+#define MTRR_PHYS_BASE_MSR(reg) (0x200 + 2 * (reg))
+#define MTRR_PHYS_MASK_MSR(reg) (0x200 + 2 * (reg) + 1)
+
+#define CACHE_AS_RAM_SIZE	CONFIG_DCACHE_RAM_SIZE
+#define CACHE_AS_RAM_BASE	CONFIG_DCACHE_RAM_BASE
+
+/* Cache 4GB - MRC_SIZE_KB for MRC */
+#define CACHE_MRC_BYTES	((CONFIG_CACHE_MRC_SIZE_KB << 10) - 1)
+#define CACHE_MRC_BASE		(0xFFFFFFFF - CACHE_MRC_BYTES)
+#define CACHE_MRC_MASK		(~CACHE_MRC_BYTES)
+
+#define CPU_PHYSMASK_HI	(1 << (CONFIG_CPU_ADDR_BITS - 32) - 1)
+
+#define NOEVICTMOD_MSR	0x2e0
+
+	/*
+	 * Note: ebp must not be touched in this code as it holds the BIST
+	 * value (built-in self test). We preserve this value until it can
+	 * be written to global_data when CAR is ready for use.
+	 */
 .globl car_init
 car_init:
-	/* TODO: Add cache-as-RAM init here */
+	post_code(POST_CAR_START)
+
+	/* Send INIT IPI to all excluding ourself */
+	movl	$0x000C4500, %eax
+	movl	$0xFEE00300, %esi
+	movl	%eax, (%esi)
+
+	post_code(POST_CAR_SIPI)
+	/* Zero out all fixed range and variable range MTRRs */
+	movl	$mtrr_table, %esi
+	movl	$((mtrr_table_end - mtrr_table) / 2), %edi
+	xorl	%eax, %eax
+	xorl	%edx, %edx
+clear_mtrrs:
+	movw	(%esi), %bx
+	movzx	%bx, %ecx
+	wrmsr
+	add	$2, %esi
+	dec	%edi
+	jnz	clear_mtrrs
+
+	post_code(POST_CAR_MTRR)
+	/* Configure the default memory type to uncacheable */
+	movl	$MTRRdefType_MSR, %ecx
+	rdmsr
+	andl	$(~0x00000cff), %eax
+	wrmsr
+
+	post_code(POST_CAR_UNCACHEABLE)
+	/* Set Cache-as-RAM base address */
+	movl	$(MTRR_PHYS_BASE_MSR(0)), %ecx
+	movl	$(CACHE_AS_RAM_BASE | MTRR_TYPE_WRBACK), %eax
+	xorl	%edx, %edx
+	wrmsr
+
+	post_code(POST_CAR_BASE_ADDRESS)
+	/* Set Cache-as-RAM mask */
+	movl	$(MTRR_PHYS_MASK_MSR(0)), %ecx
+	movl	$(~(CACHE_AS_RAM_SIZE - 1) | MTRRphysMaskValid), %eax
+	movl	$CPU_PHYSMASK_HI, %edx
+	wrmsr
+
+	post_code(POST_CAR_MASK)
+
+	/* Enable MTRR */
+	movl	$MTRRdefType_MSR, %ecx
+	rdmsr
+	orl	$MTRRdefTypeEn, %eax
+	wrmsr
+
+	/* Enable cache (CR0.CD = 0, CR0.NW = 0) */
+        movl	%cr0, %eax
+	andl	$(~(X86_CR0_CD | X86_CR0_NW)), %eax
+	invd
+	movl	%eax, %cr0
+
+	/* enable the 'no eviction' mode */
+	movl    $NOEVICTMOD_MSR, %ecx
+	rdmsr
+	orl     $1, %eax
+	andl    $~2, %eax
+	wrmsr
+
+       /* Clear the cache memory region. This will also fill up the cache */
+	movl	$CACHE_AS_RAM_BASE, %esi
+	movl	%esi, %edi
+	movl	$(CACHE_AS_RAM_SIZE / 4), %ecx
+	xorl	%eax, %eax
+	rep	stosl
+
+	/* enable the 'no eviction run' state */
+	movl    $NOEVICTMOD_MSR, %ecx
+	rdmsr
+	orl     $3, %eax
+	wrmsr
+
+	post_code(POST_CAR_FILL)
+	/* Enable Cache-as-RAM mode by disabling cache */
+	movl	%cr0, %eax
+	orl	$X86_CR0_CD, %eax
+	movl	%eax, %cr0
+
+	/* Enable cache for our code in Flash because we do XIP here */
+	movl	$MTRR_PHYS_BASE_MSR(1), %ecx
+	xorl	%edx, %edx
+	movl    $car_init_ret, %eax
+	andl    $(~(CONFIG_XIP_ROM_SIZE - 1)), %eax
+	orl	$MTRR_TYPE_WRPROT, %eax
+	wrmsr
+
+	movl	$MTRR_PHYS_MASK_MSR(1), %ecx
+	movl	$CPU_PHYSMASK_HI, %edx
+	movl	$(~(CONFIG_XIP_ROM_SIZE - 1) | MTRRphysMaskValid), %eax
+	wrmsr
+
+	post_code(POST_CAR_ROM_CACHE)
+#ifdef CONFIG_CACHE_MRC_BIN
+	/* Enable caching for ram init code to run faster */
+	movl	$MTRR_PHYS_BASE_MSR(2), %ecx
+	movl	$(CACHE_MRC_BASE | MTRR_TYPE_WRPROT), %eax
+	xorl	%edx, %edx
+	wrmsr
+	movl	$MTRR_PHYS_MASK_MSR(2), %ecx
+	movl	$(CACHE_MRC_MASK | MTRRphysMaskValid), %eax
+	movl	$CPU_PHYSMASK_HI, %edx
+	wrmsr
+#endif
+
+	post_code(POST_CAR_MRC_CACHE)
+	/* Enable cache */
+	movl	%cr0, %eax
+	andl	$(~(X86_CR0_CD | X86_CR0_NW)), %eax
+	movl	%eax, %cr0
+
+	post_code(POST_CAR_CPU_CACHE)
+
+	/* All CPUs need to be in Wait for SIPI state */
+wait_for_sipi:
+	movl	(%esi), %eax
+	bt	$12, %eax
+	jc	wait_for_sipi
+
+	/* return */
 	jmp	car_init_ret
+
+mtrr_table:
+	/* Fixed MTRRs */
+	.word 0x250, 0x258, 0x259
+	.word 0x268, 0x269, 0x26A
+	.word 0x26B, 0x26C, 0x26D
+	.word 0x26E, 0x26F
+	/* Variable MTRRs */
+	.word 0x200, 0x201, 0x202, 0x203
+	.word 0x204, 0x205, 0x206, 0x207
+	.word 0x208, 0x209, 0x20A, 0x20B
+	.word 0x20C, 0x20D, 0x20E, 0x20F
+	.word 0x210, 0x211, 0x212, 0x213
+mtrr_table_end:
diff --git a/arch/x86/cpu/ivybridge/cpu.c b/arch/x86/cpu/ivybridge/cpu.c
index 1cbdaaf..5863811 100644
--- a/arch/x86/cpu/ivybridge/cpu.c
+++ b/arch/x86/cpu/ivybridge/cpu.c
@@ -12,6 +12,7 @@
 
 #include <common.h>
 #include <asm/cpu.h>
+#include <asm/post.h>
 #include <asm/processor.h>
 
 DECLARE_GLOBAL_DATA_PTR;
@@ -20,6 +21,7 @@
 {
 	int ret;
 
+	post_code(POST_CPU_INIT);
 	timer_set_base(rdtsc());
 
 	ret = x86_cpu_init_f();
diff --git a/arch/x86/include/asm/mtrr.h b/arch/x86/include/asm/mtrr.h
new file mode 100644
index 0000000..5f05a48
--- /dev/null
+++ b/arch/x86/include/asm/mtrr.h
@@ -0,0 +1,121 @@
+/*
+ * Copyright (c) 2014 Google, Inc
+ *
+ * From Coreboot file of the same name
+ *
+ * SPDX-License-Identifier:	GPL-2.0+
+ */
+
+#ifndef _ASM_MTRR_H
+#define _ASM_MTRR_H
+
+/*  These are the region types  */
+#define MTRR_TYPE_UNCACHEABLE 0
+#define MTRR_TYPE_WRCOMB     1
+/*#define MTRR_TYPE_         2*/
+/*#define MTRR_TYPE_         3*/
+#define MTRR_TYPE_WRTHROUGH  4
+#define MTRR_TYPE_WRPROT     5
+#define MTRR_TYPE_WRBACK     6
+#define MTRR_NUM_TYPES       7
+
+#define MTRRcap_MSR     0x0fe
+#define MTRRdefType_MSR 0x2ff
+
+#define MTRRdefTypeEn		(1 << 11)
+#define MTRRdefTypeFixEn	(1 << 10)
+
+#define SMRRphysBase_MSR 0x1f2
+#define SMRRphysMask_MSR 0x1f3
+
+#define MTRRphysBase_MSR(reg) (0x200 + 2 * (reg))
+#define MTRRphysMask_MSR(reg) (0x200 + 2 * (reg) + 1)
+
+#define MTRRphysMaskValid	(1 << 11)
+
+#define NUM_FIXED_RANGES 88
+#define RANGES_PER_FIXED_MTRR 8
+#define MTRRfix64K_00000_MSR 0x250
+#define MTRRfix16K_80000_MSR 0x258
+#define MTRRfix16K_A0000_MSR 0x259
+#define MTRRfix4K_C0000_MSR 0x268
+#define MTRRfix4K_C8000_MSR 0x269
+#define MTRRfix4K_D0000_MSR 0x26a
+#define MTRRfix4K_D8000_MSR 0x26b
+#define MTRRfix4K_E0000_MSR 0x26c
+#define MTRRfix4K_E8000_MSR 0x26d
+#define MTRRfix4K_F0000_MSR 0x26e
+#define MTRRfix4K_F8000_MSR 0x26f
+
+#if !defined(__ASSEMBLER__)
+
+/*
+ * The MTRR code has some side effects that the callers should be aware for.
+ * 1. The call sequence matters. x86_setup_mtrrs() calls
+ *    x86_setup_fixed_mtrrs_no_enable() then enable_fixed_mtrrs() (equivalent
+ *    of x86_setup_fixed_mtrrs()) then x86_setup_var_mtrrs(). If the callers
+ *    want to call the components of x86_setup_mtrrs() because of other
+ *    rquirements the ordering should still preserved.
+ * 2. enable_fixed_mtrr() will enable both variable and fixed MTRRs because
+ *    of the nature of the global MTRR enable flag. Therefore, all direct
+ *    or indirect callers of enable_fixed_mtrr() should ensure that the
+ *    variable MTRR MSRs do not contain bad ranges.
+ * 3. If CONFIG_CACHE_ROM is selected an MTRR is allocated for enabling
+ *    the caching of the ROM. However, it is set to uncacheable (UC). It
+ *    is the responsiblity of the caller to enable it by calling
+ *    x86_mtrr_enable_rom_caching().
+ */
+void x86_setup_mtrrs(void);
+/*
+ * x86_setup_var_mtrrs() parameters:
+ * address_bits - number of physical address bits supported by cpu
+ * above4gb - 2 means dynamically detect number of variable MTRRs available.
+ *            non-zero means handle memory ranges above 4GiB.
+ *            0 means ignore memory ranges above 4GiB
+ */
+void x86_setup_var_mtrrs(unsigned int address_bits, unsigned int above4gb);
+void enable_fixed_mtrr(void);
+void x86_setup_fixed_mtrrs(void);
+/* Set up fixed MTRRs but do not enable them. */
+void x86_setup_fixed_mtrrs_no_enable(void);
+int x86_mtrr_check(void);
+/* ROM caching can be used after variable MTRRs are set up. Beware that
+ * enabling CONFIG_CACHE_ROM will eat through quite a few MTRRs based on
+ * one's IO hole size and WRCOMB resources. Be sure to check the console
+ * log when enabling CONFIG_CACHE_ROM or adding WRCOMB resources. Beware that
+ * on CPUs with core-scoped MTRR registers such as hyperthreaded CPUs the
+ * rom caching will be disabled if all threads run the MTRR code. Therefore,
+ * one needs to call x86_mtrr_enable_rom_caching() after all threads of the
+ * same core have run the MTRR code. */
+#if CONFIG_CACHE_ROM
+void x86_mtrr_enable_rom_caching(void);
+void x86_mtrr_disable_rom_caching(void);
+/* Return the variable range MTRR index of the ROM cache. */
+long x86_mtrr_rom_cache_var_index(void);
+#else
+static inline void x86_mtrr_enable_rom_caching(void) {}
+static inline void x86_mtrr_disable_rom_caching(void) {}
+static inline long x86_mtrr_rom_cache_var_index(void) { return -1; }
+#endif /* CONFIG_CACHE_ROM */
+
+#endif
+
+#if !defined(CONFIG_RAMTOP)
+# error "CONFIG_RAMTOP not defined"
+#endif
+
+#if ((CONFIG_XIP_ROM_SIZE & (CONFIG_XIP_ROM_SIZE - 1)) != 0)
+# error "CONFIG_XIP_ROM_SIZE is not a power of 2"
+#endif
+
+#if ((CONFIG_CACHE_ROM_SIZE & (CONFIG_CACHE_ROM_SIZE - 1)) != 0)
+# error "CONFIG_CACHE_ROM_SIZE is not a power of 2"
+#endif
+
+#define CACHE_ROM_BASE	(((1 << 20) - (CONFIG_CACHE_ROM_SIZE >> 12)) << 12)
+
+#if (CONFIG_RAMTOP & (CONFIG_RAMTOP - 1)) != 0
+# error "CONFIG_RAMTOP must be a power of 2"
+#endif
+
+#endif
diff --git a/arch/x86/include/asm/post.h b/arch/x86/include/asm/post.h
index 3371185..b203dc9 100644
--- a/arch/x86/include/asm/post.h
+++ b/arch/x86/include/asm/post.h
@@ -13,9 +13,18 @@
 /* post codes which represent various stages of init */
 #define POST_START		0x1e
 #define POST_CAR_START		0x1f
-
+#define POST_CAR_SIPI		0x20
+#define POST_CAR_MTRR		0x21
+#define POST_CAR_UNCACHEABLE	0x22
+#define POST_CAR_BASE_ADDRESS	0x23
+#define POST_CAR_MASK		0x24
+#define POST_CAR_FILL		0x25
+#define POST_CAR_ROM_CACHE	0x26
+#define POST_CAR_MRC_CACHE	0x27
+#define POST_CAR_CPU_CACHE	0x28
 #define POST_START_STACK	0x29
 #define POST_START_DONE		0x2a
+#define POST_CPU_INIT		0x2b
 
 /* Output a post code using al - value must be 0 to 0xff */
 #ifdef __ASSEMBLY__
@@ -23,6 +32,8 @@
 	movb	$value, %al; \
 	outb	%al, $POST_PORT
 #else
+#include <asm/io.h>
+
 static inline void post_code(int code)
 {
 	outb(code, POST_PORT);