/*
 * This file is part of the coreboot project.
 *
 * Copyright (C) 2005-2007 Advanced Micro Devices, Inc.
 * Copyright (C) 2008 Carl-Daniel Hailfinger
 * Copyright (C) 2015 Timothy Pearson <tpearson@raptorengineeringinc.com>, Raptor Engineering
 *
 * 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/amd/mtrr.h>

#define CacheSize		CONFIG_DCACHE_RAM_SIZE
#define CacheBase		CONFIG_DCACHE_RAM_BASE
#define CacheSizeBSPStack	CONFIG_DCACHE_BSP_STACK_SIZE
#define CacheSizeBSPSlush	CONFIG_DCACHE_BSP_STACK_SLUSH

/* For CAR with Fam10h. */
#define CacheSizeAPStack	CONFIG_DCACHE_AP_STACK_SIZE

#define MSR_MCFG_BASE		0xC0010058
#define MSR_BU_CFG2		0xC001102A

#define jmp_if_not_k8(x)	comisd	%xmm2, %xmm1; jae x
#define jmp_if_k8(x)		comisd	%xmm2, %xmm1; jb x
#define jmp_if_not_fam15h(x)	comisd	%xmm3, %xmm1; jb x
#define jmp_if_fam15h(x)	comisd	%xmm3, %xmm1; jae x

#define CPUID_MASK		0x0ff00f00
#define CPUID_VAL_FAM10_ROTATED	0x0f000010
#define CPUID_VAL_FAM15_ROTATED	0x0f000060

/*
 * XMM map:
 *   xmm1: CPU family
 *   xmm2: Fam10h comparison value
 *   xmm3: Fam15h comparison value
 *   xmm4: Backup EBX
 *   xmm5: Coreboot init detect
 */

	/* Save the BIST result. */
	movl	%eax, %ebp

	/*
	 * For normal part %ebx already contain cpu_init_detected
	 * from fallback call.
	 */

cache_as_ram_setup:
	post_code(0xa0)

	/* Enable SSE. */
	movl	%cr4, %eax
	orl	$(3 << 9), %eax
	movl	%eax, %cr4

	/* Figure out the CPU family. */
	cvtsi2sd %ebx, %xmm4
	movl	$0x01, %eax
	cpuid
	/* Base family is bits 8..11, extended family is bits 20..27. */
	andl	$CPUID_MASK, %eax
	/* Reorder bits for easier comparison by value. */
	roll	$0x10, %eax
	cvtsi2sd %eax, %xmm1
	movl	$CPUID_VAL_FAM10_ROTATED, %eax
	cvtsi2sd %eax, %xmm2
	movl	$CPUID_VAL_FAM15_ROTATED, %eax
	cvtsi2sd %eax, %xmm3
	cvtsd2si %xmm4, %ebx

	/* Check if cpu_init_detected. */
	movl	$MTRR_DEF_TYPE_MSR, %ecx
	rdmsr
	andl	$MTRR_DEF_TYPE_EN, %eax
	movl	%eax, %ebx	/* We store the status. */
	cvtsi2sd %ebx, %xmm5

	jmp_if_k8(CAR_FAM10_out_post_errata)

	/*
	 * For GH, CAR need to set DRAM Base/Limit registers to direct that
	 * to node0.
	 * Only BSP needed, for other nodes set during HT/memory init.
	 * So we need to check if it is BSP.
	 */
	movl	$0x1b, %ecx
	rdmsr
	bt	$8, %eax	/* BSP */
	jnc	CAR_FAM10_out

	/* Enable RT tables on BSP. */
	movl	$0x8000c06c, %eax
	movw	$0xcf8, %dx
	outl	%eax, %dx
	addw	$4, %dx
	inl	%dx, %eax
	btr	$0, %eax
	outl	%eax, %dx

	/* Setup temporary DRAM map: [0,16M) bit 0-23. */
	movl	$0x8000c144, %eax
	movw	$0xcf8, %dx
	outl	%eax, %dx
	addw	$4, %dx
	movl	$0, %eax
	outl	%eax, %dx

	movl	$0x8000c140, %eax
	movw	$0xcf8, %dx
	outl	%eax, %dx
	addw	$4, %dx
	movl	$3, %eax
	outl	%eax, %dx

CAR_FAM10_out:

	jmp_if_fam15h(CAR_FAM10_errata_applied)
	/*
	 * Errata 193: Disable clean copybacks to L3 cache to allow cached ROM.
	 * Re-enable it in after RAM is initialized and before CAR is disabled.
	 */
	movl	$MSR_BU_CFG2, %ecx
	rdmsr
	bts	$15, %eax	/* Set bit 15 in EDX:EAX (bit 15 in EAX). */
	wrmsr

	/* Erratum 343, RevGuide for Fam10h, Pub#41322 Rev. 3.33 */
	movl	$MSR_BU_CFG2, %ecx
	rdmsr
	bts	$35-32, %edx	/* Set bit 35 in EDX:EAX (bit 3 in EDX). */
	wrmsr

CAR_FAM10_errata_applied:

#if CONFIG_MMCONF_SUPPORT
   #if (CONFIG_MMCONF_BASE_ADDRESS > 0xFFFFFFFF)
   #error "MMCONF_BASE_ADDRESS too big"
   #elif (CONFIG_MMCONF_BASE_ADDRESS & 0xFFFFF)
   #error "MMCONF_BASE_ADDRESS not 1MB aligned"
   #endif
	movl	$0, %edx
	movl	$((CONFIG_MMCONF_BASE_ADDRESS) | (1 << 0)), %eax
   #if (CONFIG_MMCONF_BUS_NUMBER == 1)
   #elif (CONFIG_MMCONF_BUS_NUMBER == 2)
	orl	$(1 << 2), %eax
   #elif (CONFIG_MMCONF_BUS_NUMBER == 4)
	orl	$(2 << 2), %eax
   #elif (CONFIG_MMCONF_BUS_NUMBER == 8)
	orl	$(3 << 2), %eax
   #elif (CONFIG_MMCONF_BUS_NUMBER == 16)
	orl	$(4 << 2), %eax
   #elif (CONFIG_MMCONF_BUS_NUMBER == 32)
	orl	$(5 << 2), %eax
   #elif (CONFIG_MMCONF_BUS_NUMBER == 64)
	orl	$(6 << 2), %eax
   #elif (CONFIG_MMCONF_BUS_NUMBER == 128)
	orl	$(7 << 2), %eax
   #elif (CONFIG_MMCONF_BUS_NUMBER == 256)
	orl	$(8 << 2), %eax
   #else
	#error "bad MMCONF_BUS_NUMBER value"
   #endif
	movl	$MSR_MCFG_BASE, %ecx
	wrmsr
#endif

CAR_FAM10_out_post_errata:

	/* Fam15h APIC IDs do not depend on NB config bit 54 */
	jmp_if_not_fam15h(skip_nb54_set)
	movl	$0xc001001f, %ecx	/* NB_CFG_MSR */
	rdmsr
	bts	$(54 - 32), %edx	/* Set NB config bit 54 */
	wrmsr

skip_nb54_set:
	/* On Fam15h CPUs each compute unit's MTRRs are shared between two cores */
	jmp_if_not_fam15h(skip_cu_check)

	/* Get the initial APIC ID. */
	movl	$1, %eax
	cpuid
	movl	%ebx, %eax

	/* Restore init detect */
	cvtsd2si %xmm5, %ebx

	/* Determine if this is the second core to start in a compute unit; if so, wait for first core start, clear init detect and skip MTRR init */
	bt	$24, %eax
	jnc	skip_cu_check		/* First core in the compute unit jumps to skip_cu_check */

	/* Determine if this is the second core to start in a compute unit; if so, clear init detect and skip MTRR init */
	/* Busywait until the first core sets up the MTRRs */
check_init_detect_1:
	/* Check if cpu_init_detected. */
	movl	$MTRR_DEF_TYPE_MSR, %ecx
	rdmsr
	andl	$MTRR_DEF_TYPE_EN, %eax
	cmp	$0x00000000, %eax
	je	check_init_detect_1	/* First core has not yet started */

check_init_detect_2:
	movl	$SYSCFG_MSR, %ecx
	rdmsr
	andl	$(SYSCFG_MSR_MtrrFixDramEn | SYSCFG_MSR_MtrrVarDramEn), %eax
	cmp	$0x00000000, %eax
	je	check_init_detect_2	/* First core has not yet started */

	/* First core has now started */
	movl	$0x00000000, %ebx	/* Clear init detect flag */
	cvtsi2sd %ebx, %xmm5
	jmp	fam10_mtrr_setup_complete

skip_cu_check:

	jmp_if_not_fam15h(CAR_FAM15_errata_applied)

	/* Erratum 714, RevGuide for Fam15h, Pub#48063 Rev. 3.24 */
	movl	$MSR_BU_CFG2, %ecx
	rdmsr
	bts	$8, %eax	/* Set bit 8 in EDX:EAX (bit 8 in EAX). */
	wrmsr

CAR_FAM15_errata_applied:

	/* Set MtrrFixDramModEn for clear fixed MTRR. */
enable_fixed_mtrr_dram_modify:
	movl	$SYSCFG_MSR, %ecx
	rdmsr
	andl	$(~(SYSCFG_MSR_MtrrFixDramEn | SYSCFG_MSR_MtrrVarDramEn)), %eax
	orl	$SYSCFG_MSR_MtrrFixDramModEn, %eax
	wrmsr

	/* Clear all MTRRs. */
	xorl	%edx, %edx
	movl	$all_mtrr_msrs, %esi

clear_fixed_var_mtrr:
	lodsl	(%esi), %eax
	testl	%eax, %eax
	jz	clear_fixed_var_mtrr_out

	movl	%eax, %ecx
	xorl	%eax, %eax
	wrmsr

	jmp	clear_fixed_var_mtrr
clear_fixed_var_mtrr_out:

/*
 * 0x06 is the WB IO type for a given 4k segment.
 * 0x1e is the MEM IO type for a given 4k segment (K10 and above).
 * segs is the number of 4k segments in the area of the particular
 *      register we want to use for CAR.
 * reg is the register where the IO type should be stored.
 */
.macro extractmask segs, reg
.if \segs <= 0
	/*
	 * The xorl here is superfluous because at the point of first execution
	 * of this macro, %eax and %edx are cleared. Later invocations of this
	 * macro will have a monotonically increasing segs parameter.
	 */
	xorl	\reg, \reg
.else
	jmp_if_k8(1f)

.if \segs == 1
	movl	$0x1e000000, \reg /* WB MEM type */
.elseif \segs == 2
	movl	$0x1e1e0000, \reg /* WB MEM type */
.elseif \segs == 3
	movl	$0x1e1e1e00, \reg /* WB MEM type */
.elseif \segs >= 4
	movl	$0x1e1e1e1e, \reg /* WB MEM type */
.endif
	jmp 2f
1:
.if \segs == 1
	movl	$0x06000000, \reg /* WB IO type */
.elseif \segs == 2
	movl	$0x06060000, \reg /* WB IO type */
.elseif \segs == 3
	movl	$0x06060600, \reg /* WB IO type */
.elseif \segs >= 4
	movl	$0x06060606, \reg /* WB IO type */
.endif
2:
.endif /* if \segs <= 0 */
.endm

/*
 * carsize is the cache size in bytes we want to use for CAR.
 * windowoffset is the 32k-aligned window into CAR size.
 */
.macro simplemask carsize, windowoffset
	.set gas_bug_workaround,(((\carsize - \windowoffset) >> 12) - 4)
	extractmask gas_bug_workaround, %eax
	.set gas_bug_workaround,(((\carsize - \windowoffset) >> 12))
	extractmask gas_bug_workaround, %edx
	/*
	 * Without the gas bug workaround, the entire macro would consist
	 * only of the two lines below:
	 *   extractmask (((\carsize - \windowoffset) >> 12) - 4), %eax
	 *   extractmask (((\carsize - \windowoffset) >> 12)), %edx
	 */
.endm

#if IS_ENABLED(CONFIG_CPU_AMD_MODEL_10XXX)
  #if CacheSize > 0x80000
  #error Invalid CAR size, must be at most 128k (processor limit is 512k).
  #endif
#else
  #if CacheSize > 0x10000
  #error Invalid CAR size, must be at most 64k.
  #endif
#endif
#if CacheSize < 0x1000
#error Invalid CAR size, must be at least 4k. This is a processor limitation.
#endif
#if (CacheSize & (0x1000 - 1))
#error Invalid CAR size, is not a multiple of 4k. This is a processor limitation.
#endif

#if CacheSize > 0x8000
	/* Enable caching for 32K-64K using fixed MTRR. */
	movl	$MTRR_FIX_4K_C0000, %ecx
	simplemask CacheSize, 0x8000
	wrmsr
#endif

#if CacheSize > 0x10000
	/* Enable caching for 64K-96K using fixed MTRR. */
	movl	$MTRR_FIX_4K_D0000, %ecx
	simplemask CacheSize, 0x10000
	wrmsr
#endif

#if CacheSize > 0x18000
	/* Enable caching for 96K-128K using fixed MTRR. */
	movl	$MTRR_FIX_4K_D8000, %ecx
	simplemask CacheSize, 0x18000
	wrmsr
#endif

	/* Enable caching for 0-32K using fixed MTRR. */
	movl	$MTRR_FIX_4K_C8000, %ecx
	simplemask CacheSize, 0
	wrmsr

	jmp_if_fam15h(fam15_skip_dram_mtrr_setup)

	/* Enable memory access for first MBs using top_mem. */
	movl	$TOP_MEM, %ecx
	xorl	%edx, %edx
	movl	$(((CONFIG_RAMTOP) + TOP_MEM_MASK) & ~TOP_MEM_MASK) , %eax
	wrmsr

fam15_skip_dram_mtrr_setup:

#if CONFIG_XIP_ROM_SIZE

	/* Enable write base caching so we can do execute in place (XIP)
	 * on the flash ROM.
	 */
	movl	$MTRR_PHYS_BASE(1), %ecx
	xorl	%edx, %edx
	/*
	 * IMPORTANT: The following calculation _must_ be done at runtime. See
	 * http://www.coreboot.org/pipermail/coreboot/2010-October/060855.html
	 */
	movl	$copy_and_run, %eax
	andl	$(~(CONFIG_XIP_ROM_SIZE - 1)), %eax
	orl	$MTRR_TYPE_WRBACK, %eax
	wrmsr

	movl	$MTRR_PHYS_MASK(1), %ecx
	movl	$0xff, %edx /* (1 << (CONFIG_CPU_ADDR_BITS - 32)) - 1 for K8 (CONFIG_CPU_ADDR_BITS = 40) */
	jmp_if_k8(wbcache_post_fam10_setup)
	movl	$0xffff, %edx /* (1 << (CONFIG_CPU_ADDR_BITS - 32)) - 1 for FAM10 (CONFIG_CPU_ADDR_BITS = 48) */
wbcache_post_fam10_setup:
	movl	$(~(CONFIG_XIP_ROM_SIZE - 1) | MTRR_PHYS_MASK_VALID), %eax
	wrmsr
#endif /* CONFIG_XIP_ROM_SIZE */

	/* Set the default memory type and enable fixed and variable MTRRs. */
	movl	$MTRR_DEF_TYPE_MSR, %ecx
	xorl	%edx, %edx
	movl	$(MTRR_DEF_TYPE_EN | MTRR_DEF_TYPE_FIX_EN), %eax
	wrmsr

	/* Enable the MTRRs and IORRs in SYSCFG. */
	movl	$SYSCFG_MSR, %ecx
	rdmsr
	orl	$(SYSCFG_MSR_MtrrVarDramEn | SYSCFG_MSR_MtrrFixDramEn), %eax
	wrmsr

fam10_mtrr_setup_complete:
	post_code(0xa1)

	/* Disable conversion of INVD to WBINVD (INVDWBINVD = 0) */
	mov	$0xc0010015, %ecx
	rdmsr
	btr	$4, %eax
	wrmsr

jmp_if_not_fam15h(fam15_car_msr_setup_complete)
	/* Disable streaming store (DisSS = 1) */
	mov	$0xc0011020, %ecx
	rdmsr
	bts	$28, %eax
	wrmsr

	/* Disable speculative ITLB reloads (DisSpecTlbRld = 1) */
	mov	$0xc0011021, %ecx
	rdmsr
	bts	$9, %eax
	wrmsr

	/* Disable speculative DTLB reloads (DisSpecTlbRld = 1) and set DisHwPf = 1 */
	mov	$0xc0011022, %ecx
	rdmsr
	bts	$4, %eax
	bts	$13, %eax
	wrmsr

	/* Disable CR0 combining (CombineCr0Cd = 0) */
	mov	$0xc001102b, %ecx
	rdmsr
	btr	$49-32, %edx
	wrmsr
fam15_car_msr_setup_complete:

	/* Enable cache. */
	movl	%cr0, %eax
	andl	$(~(CR0_CacheDisable | CR0_NoWriteThrough)), %eax
	movl	%eax, %cr0

	jmp_if_not_k8(CAR_skip_k8_errata_part1)

	/* Set DisFillP on BSP. */
	movl	$0x8000c068, %eax
	movw	$0xcf8, %dx
	outl	%eax, %dx
	addw	$4, %dx
	inl	%dx, %eax
	bts	$10, %eax
	outl	%eax, %dx

CAR_skip_k8_errata_part1:

	jmp_if_k8(fam10_end_part1)

	/* So we need to check if it is BSP. */
	movl	$0x1b, %ecx
	rdmsr
	bt	$8, %eax	/* BSP */
	jnc	CAR_FAM10_ap
fam10_end_part1:

	post_code(0xa2)

	/* Read the range with lodsl. */
	cld
	movl	$CacheBase, %esi
	movl	$(CacheSize >> 2), %ecx
	rep	lodsl

	/* Clear the range. */
	movl	$CacheBase, %edi
	movl	$(CacheSize >> 2), %ecx
	xorl	%eax, %eax
	rep	stosl

	jmp_if_not_k8(CAR_skip_k8_errata_part2)

	/* Clear DisFillP on BSP. */
	movl	$0x8000c068, %eax
	movw	$0xcf8, %dx
	outl	%eax, %dx
	addw	$4, %dx
	inl	%dx, %eax
	btr	$10, %eax
	outl	%eax, %dx

CAR_skip_k8_errata_part2:

	/* Set up the stack pointer. */
	movl	$(CacheBase + CacheSize), %eax
	movl	%eax, %esp

	/* Poison the lower stack boundary */
	movl	$((CacheBase + CacheSize) - CacheSizeBSPStack), %eax
	movl	$0xdeadbeef, (%eax)

	post_code(0xa3)

	jmp	CAR_FAM10_ap_out
CAR_FAM10_ap:
	/*
	 * Need to set stack pointer for AP.
	 * It will be from:
	 *   CacheBase + (CacheSize - (CacheSizeBSPStack + CacheSizeBSPSlush))
	 *   - (NodeID << CoreIDbits + CoreID) * CacheSizeAPStack
	 * The spacing between the BSP stack and the top of the AP
	 * stacks is purposefully set larger (an extra CacheSizeBSPSlush
	 * worth of unused space) than necessary to aid debugging when
	 * additional stack variables are added by future developers.
	 * The extra space will allow BSP overruns to be caught by
	 * the warning logic and easily fixed instead of crashing the
	 * system with no obvious clues of what went wrong.
	 *
	 * So, need to get the NodeID and CoreID at first.
	 * If NB_CFG bit 54 is set just use initial APIC ID, otherwise need
	 * to reverse it.
	 */

	/* Get the coreid bits at first. */
	movl	$0x80000008, %eax
	cpuid
	shrl	$12, %ecx
	andl	$0x0f, %ecx
	movl	%ecx, %edi

	/* Get the initial APIC ID. */
	movl	$1, %eax
	cpuid
	shrl	$24, %ebx

	/* Get the nb cfg bit 54. */
	movl	$0xc001001f, %ecx	/* NB_CFG_MSR */
	rdmsr
	movl	%edi, %ecx		/* CoreID bits */
	bt	$(54 - 32), %edx
	jc	roll_cfg

	/* Fam10h NB config bit 54 was not set */
	rolb	%cl, %bl
roll_cfg:
	jmp_if_not_fam15h(ap_apicid_ready)
	cmp	$0x5, %ecx
	jne	ap_apicid_ready

	/* This is a multi-node CPU
	 * Adjust the maximum APIC ID to a more reasonable value
	 * given that no 32-core Family 15h processors exist
	 */
	movl	%ebx, %ecx
	and	$0x0f, %ecx		/* Get lower 4 bits of CPU number */
	and	$0x60, %ebx		/* Get node ID */
	shrl	$0x1, %ebx		/* Shift node ID part of APIC ID down by 1 */
	or	%ecx, %ebx		/* Recombine node ID and CPU number */

ap_apicid_ready:

	/* Calculate stack pointer using adjusted APIC ID stored in ebx */
	movl	$CacheSizeAPStack, %eax
	mull	%ebx
	movl	$(CacheBase + (CacheSize - (CacheSizeBSPStack + CacheSizeBSPSlush))), %esp
	subl	%eax, %esp

	/* Restore init detect */
	cvtsd2si %xmm5, %ebx

	post_code(0xa4)

CAR_FAM10_ap_out:

	post_code(0xa5)

	/* Disable SSE. */
	movl	%cr4, %eax
	andl	$~(3 << 9), %eax
	movl	%eax, %cr4

	post_code(0xa6)

	/* Restore the BIST result. */
	movl	%ebp, %eax

	/* We need to set EBP? No need. */
	movl	%esp, %ebp
	pushl	%ebx		/* Init detected. */
	pushl	%eax		/* BIST */

	post_code(0xa7)

	call	cache_as_ram_main

	/* We will not go back. */

	post_code(0xaf)		/* Should never see this POST code. */

	.globl cache_as_ram_switch_stack

cache_as_ram_switch_stack:
	/* Return address.  */
	popl	%eax
	/* New stack.  */
	popl	%eax
	movl	%eax, %esp
	call	cache_as_ram_new_stack

all_mtrr_msrs:
	/* fixed MTRR MSRs */
	.long	MTRR_FIX_64K_00000
	.long	MTRR_FIX_16K_80000
	.long	MTRR_FIX_16K_A0000
	.long	MTRR_FIX_4K_C0000
	.long	MTRR_FIX_4K_C8000
	.long	MTRR_FIX_4K_D0000
	.long	MTRR_FIX_4K_D8000
	.long	MTRR_FIX_4K_E0000
	.long	MTRR_FIX_4K_E8000
	.long	MTRR_FIX_4K_F0000
	.long	MTRR_FIX_4K_F8000

	/* var MTRR MSRs */
	.long	MTRR_PHYS_BASE(0)
	.long	MTRR_PHYS_MASK(0)
	.long	MTRR_PHYS_BASE(1)
	.long	MTRR_PHYS_MASK(1)
	.long	MTRR_PHYS_BASE(2)
	.long	MTRR_PHYS_MASK(2)
	.long	MTRR_PHYS_BASE(3)
	.long	MTRR_PHYS_MASK(3)
	.long	MTRR_PHYS_BASE(4)
	.long	MTRR_PHYS_MASK(4)
	.long	MTRR_PHYS_BASE(5)
	.long	MTRR_PHYS_MASK(5)
	.long	MTRR_PHYS_BASE(6)
	.long	MTRR_PHYS_MASK(6)
	.long	MTRR_PHYS_BASE(7)
	.long	MTRR_PHYS_MASK(7)

	/* Variable IORR MTRR MSRs */
	.long	IORRBase_MSR(0)
	.long	IORRMask_MSR(0)
	.long	IORRBase_MSR(1)
	.long	IORRMask_MSR(1)

	/* Top of memory MTRR MSRs */
	.long	TOP_MEM
	.long	TOP_MEM2

	.long	0x000 /* NULL, end of table */

cache_as_ram_setup_out: