aboutsummaryrefslogtreecommitdiff
path: root/src/vendorcode/amd/pi/00730F01/binaryPI/gcccar.inc
diff options
context:
space:
mode:
authorMarc Jones <marc.jones@se-eng.com>2015-07-19 15:20:17 -0600
committerMarc Jones <marc.jones@se-eng.com>2015-08-07 17:59:48 +0200
commit0b11bd0d028089288e52f9f08cdcfafc9910511b (patch)
treee561ef6110713f774cc71f95e5ad0f26302de85c /src/vendorcode/amd/pi/00730F01/binaryPI/gcccar.inc
parent9b9400dc90906fddadc0303994422ec011a8f6e7 (diff)
vendorcode: Move AMD sources from blobs to vendorcode
The AMD AGESA binaryPI sources were incorrectly committed to 3rdparty/blobs. Move them from blobs to vendorcode and fix Kconfig and Makefile.inc to match. Change-Id: I55a777553c1203464d7f7f4293b361fedcfa3283 Signed-off-by: Marc Jones <marc.jones@se-eng.com> Reviewed-on: http://review.coreboot.org/10982 Tested-by: build bot (Jenkins) Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Diffstat (limited to 'src/vendorcode/amd/pi/00730F01/binaryPI/gcccar.inc')
-rw-r--r--src/vendorcode/amd/pi/00730F01/binaryPI/gcccar.inc1311
1 files changed, 1311 insertions, 0 deletions
diff --git a/src/vendorcode/amd/pi/00730F01/binaryPI/gcccar.inc b/src/vendorcode/amd/pi/00730F01/binaryPI/gcccar.inc
new file mode 100644
index 0000000000..2c51eb1fe8
--- /dev/null
+++ b/src/vendorcode/amd/pi/00730F01/binaryPI/gcccar.inc
@@ -0,0 +1,1311 @@
+/*
+ * Copyright (c) 2012, Advanced Micro Devices, Inc.
+ * 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 Advanced Micro Devices, Inc. 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 ADVANCED MICRO DEVICES, INC. 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.
+ *
+ */
+
+/******************************************************************************
+* AMD Generic Encapsulated Software Architecture
+*
+* $Workfile:: GccCar.inc $Revision:: 32932 $
+*
+* Description: GccCar.inc - AGESA cache-as-RAM setup Include File for GCC complier
+*
+******************************************************************************/
+
+.altmacro
+
+BSP_STACK_BASE_ADDR = 0x30000 /* Base address for primary cores stack */
+BSP_STACK_SIZE = 0x10000 /* 64KB for BSP core */
+CORE0_STACK_BASE_ADDR = 0x80000 /* Base address for primary cores stack */
+CORE0_STACK_SIZE = 0x4000 /* 16KB for primary cores */
+CORE1_STACK_BASE_ADDR = 0x40000 /* Base address for AP cores */
+CORE1_STACK_SIZE = 0x1000 /* 4KB for each AP cores */
+
+APIC_BASE_ADDRESS = 0x0000001B
+ APIC_BSC = 8 /* Boot Strap Core */
+
+APIC_MSG_REG = 0x380 # Location of BSC message
+ APIC_MSG = 0x00DE00AD # Message data
+ APIC_INVD_ALL_DONE_MSG = 0x00AD00DE /* Indicate all cores have invalidated */
+APIC_CMD_LO_REG = 0x300 # APIC command low
+APIC_CMD_HI_REG = 0x310 # APIC command high
+// del CMD_REG_TO_READ_DATA = 0x00000338 # APIC command for remote read of APIC_MSG_REG
+ REMOTE_READ_STS = 0x00030000 # Remote read status mask
+ REMOTE_DELIVERY_PEND = 0x00010000 # Remote read is pending
+ REMOTE_DELIVERY_DONE = 0x00020000 # Remote read is complete
+ DELIVERY_STS_BIT = 12 #Delivery status valid bit
+APIC_ID_REG = 0x0020 # Local APIC ID offset
+ APIC20_APICID = 24
+APIC_REMOTE_READ_REG = 0x00C0 # Remote read offset
+
+# Flags can only run from bits 31 to 24. Bits 23:0 are in use.
+AMD_CU_NEED_TO_WAIT = 31
+AMD_CU_SEND_INVD_MSG = 30
+AMD_CU_RESTORE_ES = 29
+
+AMD_MTRR_VARIABLE_BASE0 = 0x0200
+AMD_MTRR_VARIABLE_BASE6 = 0x020C
+AMD_MTRR_VARIABLE_BASE7 = 0x020E
+ VMTRR_VALID = 11
+ MTRR_TYPE_WB = 0x06
+ MTRR_TYPE_WP = 0x05
+ MTRR_TYPE_WT = 0x04
+ MTRR_TYPE_UC = 0x00
+AMD_MTRR_VARIABLE_MASK7 = 0x020F
+AMD_MTRR_FIX64k_00000 = 0x0250
+AMD_MTRR_FIX16k_80000 = 0x0258
+AMD_MTRR_FIX16k_A0000 = 0x0259
+AMD_MTRR_FIX4k_C0000 = 0x0268
+AMD_MTRR_FIX4k_C8000 = 0x0269
+AMD_MTRR_FIX4k_D0000 = 0x026A
+AMD_MTRR_FIX4k_D8000 = 0x026B
+AMD_MTRR_FIX4k_E0000 = 0x026C
+AMD_MTRR_FIX4k_E8000 = 0x026D
+AMD_MTRR_FIX4k_F0000 = 0x026E
+AMD_MTRR_FIX4k_F8000 = 0x026F
+
+/* Reproduced from AGESA.h */
+AMD_AP_MTRR_FIX64k_00000 = 0x00000250
+AMD_AP_MTRR_FIX16k_80000 = 0x00000258
+AMD_AP_MTRR_FIX16k_A0000 = 0x00000259
+AMD_AP_MTRR_FIX4k_C0000 = 0x00000268
+AMD_AP_MTRR_FIX4k_C8000 = 0x00000269
+AMD_AP_MTRR_FIX4k_D0000 = 0x0000026A
+AMD_AP_MTRR_FIX4k_D8000 = 0x0000026B
+AMD_AP_MTRR_FIX4k_E0000 = 0x0000026C
+AMD_AP_MTRR_FIX4k_E8000 = 0x0000026D
+AMD_AP_MTRR_FIX4k_F0000 = 0x0000026E
+AMD_AP_MTRR_FIX4k_F8000 = 0x0000026F
+CPU_LIST_TERMINAL = 0xFFFFFFFF
+
+AMD_MTRR_DEFTYPE = 0x02FF
+ WB_DRAM_TYPE = 0x1E /* MemType - memory type */
+ MTRR_DEF_TYPE_EN = 11 /* MtrrDefTypeEn - variable and fixed MTRRs default enabled */
+ MTRR_DEF_TYPE_FIX_EN = 10 /* MtrrDefTypeEn - fixed MTRRs default enabled */
+
+HWCR = 0x0C0010015 /* Hardware Configuration */
+ INVD_WBINVD = 0x04 /* INVD to WBINVD conversion */
+
+IORR_BASE = 0x0C0010016 /* IO Range Regusters Base/Mask, 2 pairs */
+ /* uses 16h - 19h */
+TOP_MEM = 0x0C001001A /* Top of Memory */
+TOP_MEM2 = 0x0C001001D /* Top of Memory2 */
+
+LS_CFG = 0x0C0011020 /* Load-Store Configuration */
+ DIS_SS = 28 /* Family 10h,12h,15h:Disable Streng Store functionality */
+ DIS_STREAM_ST = 28 /* Family 14h:DisStreamSt - Disable Streaming Store functionality */
+
+IC_CFG = 0x0C0011021 /* Instruction Cache Config Register */
+ IC_DIS_SPEC_TLB_RLD = 9 /* Disable speculative TLB reloads */
+ DIS_IND = 14 /* Family 10-14h:Disable Indirect Branch Predictor */
+ DIS_I_CACHE = 14 /* Family 15h:DisICache - Disable Indirect Branch Predictor */
+
+DC_CFG = 0x0C0011022 /* Data Cache Configuration */
+ DC_DIS_SPEC_TLB_WALK = 4 /* Disable speculative table-walks */
+ DIS_HW_PF = 13 /* Hardware prefetches bit */
+DE_CFG = 0x0C0011029 /* Decode Configuration */
+ CL_FLUSH_SERIALIZE = 23 /* Family 12h,15h: CL Flush Serialization */
+
+BU_CFG2 = 0x0C001102A /* Family 10h: Bus Unit Configuration 2 */
+CU_CFG2 = 0x0C001102A /* Family 15h: Combined Unit Configuration 2 */
+ F10_CL_LINES_TO_NB_DIS = 15 /* ClLinesToNbDis - allows WP code to be cached in L2 */
+ IC_DIS_SPEC_TLB_WR = 35 /* IcDisSpecTlbWr - ITLB speculative writes */
+ F16_CL_LINES_TO_L2_DIS = 15 /* ClLinesToL2Dis */
+
+//del CU_CFG3 = 0x0C001102B /* Combined Unit Configuration 3 */
+//del COMBINE_CR0_CD = 49 /* Combine CR0.CD for both cores of a compute unit */
+//del L2I_CFG = 0x0C00110A0 /* L2I Configuration */
+ L2_RINSER_DIS = 20 /* L2 rinser disable */
+ PREFETCHER_DIS = 7 /* L2 prefetcher disable*/
+ CACHE_IC_ATTR_DIS = 3 /* Inserting IC attributes into the L2 disable */
+
+CR0_PE = 0 # Protection Enable
+CR0_NW = 29 # Not Write-through
+CR0_CD = 30 # Cache Disable
+CR0_PG = 31 # Paging Enable
+
+/* CPUID Functions */
+
+CPUID_MODEL = 1
+AMD_CPUID_FMF = 0x80000001 /* Family Model Features information */
+AMD_CPUID_L2Cache = 0X80000006 /* L2/L3 cache info */
+AMD_CPUID_APIC = 0x80000008 /* Long Mode and APIC info., core count */
+ APIC_ID_CORE_ID_SIZE = 12 /* ApicIdCoreIdSize bit position */
+
+NB_CFG = 0x0C001001F /* Northbridge Configuration Register */
+ INIT_APIC_ID_CPU_ID_LO = 54 /* InitApicIdCpuIdLo - is core# in high or low half of APIC ID? */
+ ENABLE_CF8_EXT_CFG = 46 /* EnableCf8ExtCfg - enable CF8 extended configuration cycles */
+
+MTRR_SYS_CFG = 0x0C0010010 /* System Configuration Register */
+ CHX_TO_DIRTY_DIS = 16 /* ChxToDirtyDis Change to dirty disable */
+ SYS_UC_LOCK_EN = 17 /* SysUcLockEn System lock command enable */
+ MTRR_FIX_DRAM_EN = 18 /* MtrrFixDramEn MTRR fixed RdDram and WrDram attributes enable */
+ MTRR_FIX_DRAM_MOD_EN = 19 /* MtrrFixDramModEn MTRR fixed RdDram and WrDram modification enable */
+ MTRR_VAR_DRAM_EN = 20 /* MtrrVarDramEn MTRR variable DRAM enable */
+ MTRR_TOM2_EN = 21 /* MtrrTom2En MTRR top of memory 2 enable */
+
+PERF_CONTROL3 = 0x0C0010003 /* Performance event control three */
+ PERF_CONTROL3_RESERVE_L = 0x00200000 /* Preserve the reserved bits */
+ PERF_CONTROL3_RESERVE_H = 0x0FCF0 /* Preserve the reserved bits */
+ CONFIG_EVENT_L = 0x0F0E2 /* All cores with level detection */
+ CONFIG_EVENT_H = 4 /* Increment count by number of event */
+ /* occured in clock cycle */
+ EVENT_ENABLE = 22 /* Enable the event */
+PERF_COUNTER3 = 0x0C0010007 /* Performance event counter three */
+
+COMPUTE_UNIT_STATUS = 0x08000C580 /* Compute Unit Status Register */
+ QUAD_CORE = 24 /* QuadCore four cores of a compute unit are enabled */
+ DUAL_CORE = 16 /* DualCore two cores of a compute unit are enabled */
+ TRIPLE_CORE = 8 /* TripleCore three cores of a compute unit are enabled */
+ CU_ENABLED = 0 /* Enabled at least one core of a compute unit is enabled */
+
+FUNC_3 = 3
+MCA_NB_CFG = 0x44 /* MCA NB Configuration */
+CPU_ERR_DIS = 6 /* CPU error response disable */
+PRODUCT_INFO_REG1 = 0x1FC /* Product Information Register 1 */
+
+# Local use flags, in upper most byte if ESI
+FLAG_UNKNOWN_FAMILY = 24 # Signals that the family# of the installed processor is not recognized
+FLAG_STACK_REENTRY = 25 # Signals that the environment has made a re-entry (2nd) call to set up the stack
+FLAG_IS_PRIMARY = 26 # Signals that this core is the primary within the comoute unit
+FLAG_CORE_NOT_IDENTIFIED = 27 # Signals that the cores/compute units of the installed processor is not recognized
+FLAG_FORCE_32K_STACK = 28 # Signals that to force 32KB stack size for BSP core
+CR0_MASK = ((1 << CR0_CD) | (1 << CR0_NW))
+MSR_MASK = ((1 << MTRR_DEF_TYPE_EN)+(1 << MTRR_DEF_TYPE_FIX_EN))
+
+/****************************************************************************
+ *
+ * CPU MACROS - PUBLIC
+ *
+ ****************************************************************************/
+.macro _WRMSR
+ .byte 0x0f, 0x30
+.endm
+
+.macro _RDMSR
+ .byte 0x0F, 0x32
+.endm
+
+.macro AMD_CPUID arg0
+ .ifb \arg0
+ mov $0x1, %eax
+ .byte 0x0F, 0x0A2 /* Execute instruction */
+ bswap %eax
+ xchg %ah, %al /* Ext model in al now */
+ rol $0x08, %eax /* Ext model in ah, model in al */
+ and $0x0FFCF, ax /* Keep 23:16, 7:6, 3:0 */
+ .else
+ mov \arg0, %eax
+ .byte 0x0F, 0x0A2
+ .endif
+.endm
+
+.macro MAKE_EXT_PCI_ADDR Seg, Bus, Dev, Func, Offset
+ mov $(1 << 31 | (Seg) << 28 | (((Offset) & (0x0F00)) >> 8) << 24 | (Bus) << 16 | (Dev) << 11 | (Func) << 8) | ((Offset) & (0xFC)), %eax
+.endm
+/****************************************************************************
+*
+* AMD_ENABLE_STACK_FAMILY_HOOK Macro - Stackless
+*
+* Set any family specific controls needed to enable the use of
+* cache as general storage before main memory is available.
+*
+* Inputs:
+* none
+* Outputs:
+* none
+ ****************************************************************************/
+.macro AMD_ENABLE_STACK_FAMILY_HOOK
+
+ AMD_ENABLE_STACK_FAMILY_HOOK_F16
+
+.endm
+
+/****************************************************************************
+*
+* AMD_DISABLE_STACK_FAMILY_HOOK Macro - Stackless
+*
+* Return any family specific controls to their 'standard'
+* settings for using cache with main memory.
+*
+* Inputs:
+* none
+* Outputs:
+* none
+ ****************************************************************************/
+.macro AMD_DISABLE_STACK_FAMILY_HOOK
+
+ AMD_DISABLE_STACK_FAMILY_HOOK_F16
+
+.endm
+
+/****************************************************************************
+*
+* GET_NODE_ID_CORE_ID Macro - Stackless
+*
+* Read family specific values to determine the node and core
+* numbers for the core executing this code.
+*
+* Inputs:
+* none
+* Outputs:
+* SI[7:0] = Core# (0..N, relative to node)
+* SI[15:8]= Node# (0..N)
+* SI[23:16]= reserved
+* SI[24]= flag: 1=Family Unrecognized
+* SI[25]= flag: 1=Interface re-entry call
+* SI[26]= flag: 1=Core is primary of compute unit
+* SI[31:27]= reserved, =0
+****************************************************************************/
+.macro GET_NODE_ID_CORE_ID
+ LOCAL node_core_exit
+
+ mov $-1, %si
+ GET_NODE_ID_CORE_ID_F16
+
+ /*
+ * Check for unrecognized Family
+ */
+ cmp $-1, %si # Has family (node/core) already been discovered?
+ jnz node_core_exit # Br if yes
+
+ mov $((1 << FLAG_UNKNOWN_FAMILY)+(1 << FLAG_IS_PRIMARY)), %esi # No, Set error code, Only let BSP continue
+
+ mov $APIC_BASE_ADDRESS, %ecx # MSR:0000_001B
+ _RDMSR
+ bt $APIC_BSC, %eax # Is this the BSC?
+ jc node_core_exit # Br if yes
+ hlt # Kill APs
+node_core_exit:
+
+.endm
+
+/*
+***************************************************************************
+ Family 16h MACROS
+***************************************************************************/
+/*--------------------------------------------------
+
+AMD_ENABLE_STACK_FAMILY_HOOK_F16 Macro - Stackless
+
+ Set any family specific controls needed to enable the use of
+ cache as general storage before main memory is available.
+
+Inputs:
+ ESI - node#, core#, flags from GET_NODE_ID_CORE_ID
+Outputs:
+ none
+Destroyed:
+ eax, ebx, ecx, edx
+
+Family 16h requirements (BKDG #48751 section 2.3.3):
+ * Paging must be disabled.
+ * MSRC001_0015[INVD_WBINVD]=0
+ * MSRC001_1020[DisSS]=1
+ * MSRC001_1021[DIS_SPEC_TLB_RLD]=1
+ * MSRC001_1022[DIS_SPEC_TLB_RLD]=1
+ * MSRC001_1022[DisHwPf]=1
+ * If MSRC001_102B[CombineCr0Cd] == 1 then MSRC001_102B[CombineCroCd] = 0
+ * No INVD or WBINVD, no exceptions, page faults or interrupts
+--------------------------------------------------
+
+ */
+
+.macro AMD_ENABLE_STACK_FAMILY_HOOK_F16
+ LOCAL fam16_enable_stack_hook_exit
+
+ AMD_CPUID $CPUID_MODEL
+ mov %eax, %ebx # Save revision info to EBX
+ shr $20, %eax # AL = cpu extended family
+ cmp $0x07, %al # Is this family 16h?
+ jnz fam16_enable_stack_hook_exit # Br if no
+
+ bt $FLAG_STACK_REENTRY , %esi # Check if stack has already been set
+ jc fam16_skipClearingBit4
+ mov $HWCR, %ecx # MSR C001_0015
+ _RDMSR
+ btr $INVD_WBINVD, %eax # disable INVD -> WBINVD conversion
+ _WRMSR
+
+fam16_skipClearingBit4:
+ mov $IC_CFG, %ecx # MSR:C001_1021
+ _RDMSR
+ bts $IC_DIS_SPEC_TLB_RLD, %eax # Turn on Disable speculative IC-TLB reloads bit
+ _WRMSR
+
+# mov %ebx, %eax # Restore revision info to EAX
+# shr $16, %eax
+# and $0x0F, %al # AL = cpu extended model
+
+ mov $DC_CFG, %ecx # MSR:C001_1022
+ _RDMSR
+ bts $DC_DIS_SPEC_TLB_WALK, %eax # Turn on Disable speculative DC-TLB reloads bit
+ bts $DIS_HW_PF, %eax # Turn on Disable hardware prefetches bit
+ _WRMSR # Remove KM in PI 1.1.0.0
+
+ mov $0x0C00110A0, %ecx # MSR:C001_10A0
+ _RDMSR
+ bts $L2_RINSER_DIS, %eax #Do not search for the repair single bit errors in the background
+ bts $PREFETCHER_DIS, %eax
+ bts $CACHE_IC_ATTR_DIS, %eax
+ _WRMSR
+
+fam16_enable_stack_hook_exit:
+.endm
+
+/*
+;
+; AMD_DISABLE_STACK_FAMILY_HOOK_F16 Macro - Stackless
+;
+; Return any family specific controls to their 'standard'
+; settings for using cache with main memory.
+;
+; Inputs:
+; ESI - [31:24] flags; [15:8]= Node#; [7:0]= core#
+; Outputs:
+; none
+; Destroyed:
+; eax, ebx, ecx, edx
+;
+; Family 16h requirements:
+; * INVD or WBINVD
+; * MSRC001_0015[INVD_WBINVD]=1
+; * MSRC001_1020[DisSS]=0
+; * MSRC001_1021[DIS_SPEC_TLB_RLD]=0
+; * MSRC001_1022[DIS_SPEC_TLB_RLD]=0
+; * MSRC001_1022[DIS_HW_PF]=0
+;---------------------------------------------------
+*/
+.macro AMD_DISABLE_STACK_FAMILY_HOOK_F16
+ LOCAL fam16_disable_stack_hook_exit
+ LOCAL fam16_disable_stack_remote_read_exit
+# LOCAL fam16_invd_done_remote_read_exit
+
+ AMD_CPUID $CPUID_MODEL
+ mov %eax, %ebx # Save revision info to EBX
+ shr $20, %eax # AL = cpu extended family
+ cmp $0x07, %al # Is this family 16h?
+ jnz fam16_disable_stack_hook_exit # Br if no
+
+ mov %ebx, %edi # Save revision info to EDI
+ AMD_CPUID $AMD_CPUID_APIC
+ mov %cl, %al # AL = number of cores - 1
+ shr $APIC_ID_CORE_ID_SIZE, %cx # CL = ApicIdCoreIdSize
+ mov $1, %bx
+ shl %cl, %bl # BL = theoretical number of cores on socket
+ dec %bx # BL = core number on socket mask
+ mov %bl, %ah # AH = core number on socket mask
+ mov %edi, %ebx # Restore revision info to EBX
+ mov %ax, %di # DI[15:8] = core number mask, DI[7:0] = number of cores - 1
+
+ and $0x0F00FF, %ebx
+ mov %ebx, %eax
+ shr $8, %eax
+ or %ax, %bx # Save Extended Model, Model and Stepping to BX
+ # [11:8] = Extended Model, [7:4] = Model, [3:0] = Stepping (bx=0000000000010100, ok)
+
+ mov $APIC_BASE_ADDRESS, %ecx
+ _RDMSR # dx=0 ax=fee00800
+ mov %bx, %dx # Save Extended Model, Model and Stepping to DX
+ shl $16, %edx #EDX[31:16] = Extended Model, Model and Stepping
+ mov %eax ,%ebx # EBX = LAPIC base
+ xor %ecx ,%ecx # Zero out CU flags
+ bts $AMD_CU_NEED_TO_WAIT, %ecx # Default to waiting
+ bts $AMD_CU_SEND_INVD_MSG, %ecx # Default to signaling
+ mov %cr0, %eax
+ bt $CR0_PE, %ax # Are we in protected mode?
+ # .if (!carry?)
+ jc 1f
+ bts $AMD_CU_RESTORE_ES, %ecx # Indicate ES restore is required
+ mov %es, %cx # Save ES segment register to CX
+ xor %ax, %ax
+ mov %ax, %es # Set ES to big real mode selector for 4GB access
+ # .endif
+
+1:
+ and $0x0F000, %bx # EBX = LAPIC base, offset 0
+ or $APIC_ID_REG, %bl #
+ mov %es:(%ebx), %eax # EAX[31:24] = APIC ID
+ shr $APIC20_APICID, %eax # AL = APIC ID
+ mov %al, %ah # AH = APIC ID
+ mov %di, %dx # DH = core mask
+ and %dh, %ah # AH = core number # ax=111 dx=01000F03
+
+ # .if (zero?)
+ jnz 1f
+ # Core 0 of a socket
+ btr $AMD_CU_SEND_INVD_MSG, %ecx # No need to signal after INVD
+ #.if (dl != 0)
+ cmp $0, %dl
+ jz 2f
+ # This socket has multiple cores
+ and $0xf000, %bx # EBX = LAPIC base, offset 0
+ or $APIC_MSG_REG, %bx
+ mov $APIC_MSG, %edi
+ mov %edi, %es:(%ebx) # Signal for non core 0s to complete CAR breakdown
+ jmp 1f
+ #.else
+2: btr $AMD_CU_NEED_TO_WAIT, %ecx # No need to wait on a single core CPU
+ #.endif
+ # .endif
+1:
+
+ bt $AMD_CU_NEED_TO_WAIT, %ecx #cx = c0000000
+ #.if (carry?)
+ jnc 1f
+ #.if (ah == dl)
+ cmp %dl, %ah
+ jnz 2f
+ # This is the highest numbered core on this socket -- wait on core 0
+ not %dh # Flip the mask to determine local core 0's APID ID
+ and %dh, %al # AL = target APIC ID # ax=310
+ jmp 3f
+2: #.else
+ # All other cores (including core 0) wait on the next highest core.
+ # In this way, cores will halt in a cascading fashion down to 0.
+ inc %al
+ #.endif
+3:
+ shl $APIC20_APICID, %eax
+ and $0x0F000, %bx
+ or $APIC_CMD_HI_REG, %bx
+ mov %eax, %es:(%ebx) # Set target APIC ID
+
+ # Use bits 23:16 as a timeout for unresponsive cores
+ ror $8, %ecx
+ mov $0xFF, %ch
+ stc
+
+ #.while (carry?)
+5: jnc 4f
+ and $0xF000, %bx #EBX = LAPIC base, offset 0
+ or $APIC_CMD_LO_REG, %bx # bx = 00000000FEE00300
+ mov $0x338, %eax
+ mov %eax, %es:(%ebx) #Fire remove read IPI
+ inc %ch #Pre increment the timeout
+ stc
+ #.while (carry?)
+7: jnc 6f
+ dec %ch #Check the timeout
+ jz fam16_disable_stack_remote_read_exit
+ mov %es:(%ebx), %eax # ax = 0000000000020338
+ bt $DELIVERY_STS_BIT, %eax
+ jmp 7b
+6: #.endw
+ stc
+ #.while (carry?)
+7: jnc 6f
+ mov %es:(%ebx), %eax
+ and $REMOTE_READ_STS, %eax
+ #.if (eax == REMOTE_DELIVERY_PEND)
+ cmp $REMOTE_DELIVERY_PEND, %eax
+ jnz 8f
+ dec %ch # Check the timeout
+ jz fam16_disable_stack_hook_exit # Branch if there is an unreponsive core
+ stc
+ jmp 9f
+8: #.else
+ clc
+9: #.endif
+ jmp 7b
+6: #.endw
+ #.if (eax == REMOTE_DELIVERY_DONE)
+ cmp $REMOTE_DELIVERY_DONE, %eax
+ jnz 6f
+ and $0x0F000, %bx #EBX = LAPIC base, offset 0
+ or $APIC_REMOTE_READ_REG, %bl
+ mov %es:(%ebx), %eax
+ #.if (eax == APIC_MSG)
+ cmp $APIC_MSG, %eax # ax=00000000FFC5BBB2
+ jnz 8f
+ clc
+ jmp 9f
+ #.else
+8: stc
+9: #.endif
+ jmp 7f
+6: #.else
+ dec %ch
+ jz fam16_disable_stack_remote_read_exit
+ stc
+7: #.endif
+ jmp 5b
+4: #.endw
+
+fam16_disable_stack_remote_read_exit:
+ rol $8, %ecx # Restore ECX
+
+1: #.endif
+
+ bt $AMD_CU_RESTORE_ES, %ecx
+ #.if (carry?)
+ jnc 1f
+ mov %cx, %es
+1:
+ mov %ecx, %edi
+ shr $16, %edx
+ mov %dx, %bx
+
+ #Handshaking complete. Continue tearing down CAR.
+
+ mov $IC_CFG, %ecx # MSR:C001_1021
+ _RDMSR
+ btr $IC_DIS_SPEC_TLB_RLD, %eax # Turn on speculative TLB reloads
+ _WRMSR
+
+ mov $DC_CFG, %ecx # MSR:C001_1022
+ _RDMSR
+ btr $DC_DIS_SPEC_TLB_WALK, %eax # Turn on speculative table-walks
+ #.if (bx != 0) # Is this rev A0?
+ #cmp $0, %bx
+ #jz 0f
+ btr $DIS_HW_PF, %eax # Turn on hardware prefetches
+ #.endif # End workaround for erratum 498
+ #0:
+ _WRMSR
+
+ mov $0x0C00110A0, %ecx #MSR:C001_10A0
+ _RDMSR
+ btr $PREFETCHER_DIS, %eax
+ btr $CACHE_IC_ATTR_DIS, %eax
+ _WRMSR
+
+ mov $BU_CFG2, %ecx
+ _RDMSR
+ btr $F16_CL_LINES_TO_L2_DIS, %eax
+ _WRMSR
+
+ mov $HWCR, %ecx # MSR:C001_0015h
+ _RDMSR
+ btr $INVD_WBINVD, %eax # Disable INVD -> WBINVD conversion
+ _WRMSR
+ wbinvd # Clear the cache tag RAMs
+ #invd
+
+ #Do Standard Family 16 work
+ mov $HWCR, %ecx # MSR:C001_0015h
+ _RDMSR
+ bts $INVD_WBINVD, %eax # Turn on Conversion of INVD to WBINVD
+ _WRMSR
+ #.endif # end
+
+ bt $AMD_CU_SEND_INVD_MSG, %edi
+ #.if (carry?)
+ jnc 1f
+ AMD_CPUID $AMD_CPUID_APIC
+ shr $APIC_ID_CORE_ID_SIZE, %cx # CL = ApicIdCoreIdSize
+ mov $1, %di
+ shl %cl, %di #DI = theoretical number of cores on socket
+ dec %di # DI = core number on socket mask
+ # Non core zero needs to signal to core 0 to proceed
+ mov $APIC_BASE_ADDRESS, %ecx
+ _RDMSR
+ mov %eax, %ebx # EBX = LAPIC base
+ and $0x0F000, %bx # EBX = LAPIC base, offset 0
+ or $APIC_MSG_REG, %bx
+ mov $APIC_MSG, %eax
+ mov %eax, %es:(%ebx) # Signal for core 0 to complete CAR breakdown
+
+ #TODO: Non core zero needs to wait for core zero to do INVD
+ #A handshake is required to ensure that all cores on a node invalidate in sync.
+ and $0x0F000, %bx
+ or $APIC_ID_REG, %bl
+ mov %es:(%ebx), %eax # EAX[31:24] = APIC ID
+ shr $APIC20_APICID, %eax # AL = APIC ID
+ mov %di, %dx # Use DL as core mask
+ not %dl
+ and %dl, %al # Remote read message from core zero
+ shl $APIC20_APICID, %eax
+
+ and $0x0F000, %bx
+ or $APIC_CMD_HI_REG, %bx
+ mov %eax, %es:(%ebx) #
+ # Use bits 23:16 as a timeout for unresponsive cores
+ ror $8, %ecx
+ mov $0xFF, %ch
+ stc
+
+ #.while (carry?)
+5: jnc 4f
+ and $0xF000, %bx #EBX = LAPIC base, offset 0
+ or $APIC_CMD_LO_REG, %bx # bx = 00000000FEE00300
+ mov $0x338, %eax
+ mov %eax, %es:(%ebx) #Fire remove read IPI
+ inc %ch #Pre increment the timeout
+ stc
+ #.while (carry?)
+7: jnc 6f
+ dec %ch #Check the timeout
+ jz fam16_invd_done_remote_read_exit
+ mov %es:(%ebx), %eax # ax = 0000000000020338
+ bt $DELIVERY_STS_BIT, %eax
+ jmp 7b
+6: #.endw
+ stc
+ #.while (carry?)
+7: jnc 6f
+ mov %es:(%ebx), %eax
+ and $REMOTE_READ_STS, %eax
+ #.if (eax == REMOTE_DELIVERY_PEND)
+ cmp $REMOTE_DELIVERY_PEND, %eax
+ jnz 8f
+ dec %ch # Check the timeout
+ jz fam16_invd_done_remote_read_exit # Branch if there is an unreponsive core
+ stc
+ jmp 9f
+8: #.else
+ clc
+9: #.endif
+ jmp 7b
+6: #.endw
+ #.if (eax == REMOTE_DELIVERY_DONE)
+ cmp $REMOTE_DELIVERY_DONE, %eax
+ jnz 6f
+ and $0x0F000, %bx #EBX = LAPIC base, offset 0
+ or $APIC_REMOTE_READ_REG, %bl
+ mov %es:(%ebx), %eax
+ #.if (eax == APIC_MSG)
+ cmp $APIC_MSG, %eax # ax=00000000FFC5BBB2
+ jnz 8f
+ clc
+ jmp 9f
+ #.else
+8: stc
+9: #.endif
+ jmp 7f
+6: #.else
+ dec %ch
+ jz fam16_invd_done_remote_read_exit
+ stc
+7: #.endif
+ jmp 5b
+4: #.endw
+
+ jmp 2f
+fam16_invd_done_remote_read_exit:
+1: #.else
+ mov $APIC_BASE_ADDRESS, %ecx
+ _RDMSR
+ mov %eax, %ebx
+ and $0x0F000, %bx
+ or $APIC_MSG_REG, %bx
+ mov $APIC_INVD_ALL_DONE_MSG, %edi
+ mov %edi, %es:(%ebx)
+2: #.endif
+
+ mov $0xc0011004, %ecx
+ _RDMSR
+ btr $30, %edx
+ _WRMSR
+
+fam16_disable_stack_hook_exit:
+.endm
+
+/*
+*/
+.macro GET_NODE_ID_CORE_ID_F16
+
+ LOCAL node_core_f16_exit
+ LOCAL node_core_f16_AP
+ LOCAL node_core_f16_shared
+ LOCAL node_core_f16_AP_not_TN
+
+#define F16_L2Size 1024
+#define F16_ShareCores 4
+#define F16_AllocMem 0
+#define F16_AllocExe 0
+#define F16_SzAddrBus 40
+#define F16_pad 0
+ cmp $-1, %si # Has node/core already been discovered?
+ jnz node_core_f16_exit # Br if yes
+
+ AMD_CPUID $CPUID_MODEL
+ shr $12, %eax # AL = cpu extended family
+ cmp $07, %ah # Is this family 16h?
+ jnz node_core_f16_exit # Br if no
+ shr $4, %al # AL = cpu extended model
+ shr $16, %ebx # BH = LocalApicId
+ mov %al, %bl # BL = cpu extended model
+
+ # LoadTableAddress(FAM16H_INFO_STRUCT)
+ # movd mm5, eax # load pointer to Family Info Struc
+
+ xor %esi, %esi # Assume BSC, clear local flags
+ mov $APIC_BASE_ADDRESS, %ecx # MSR:0000_001B
+ _RDMSR
+ bt $APIC_BSC, %eax # Is this the BSC?
+ jnc node_core_f16_AP # Br if no
+
+ # This is the BSP.
+ # Enable routing tables on BSP (just in case the HT init code has not yet enabled them)
+ mov $0x8000C06C, %eax # PCI address for D18F0x6C Link Initialization Control Register
+ mov $0x0CF8, %dx
+ out %eax, %dx
+ add $4, %dx
+ in %dx, %eax
+ btr $0, %eax # Set LinkInitializationControl[RouteTblDis] = 0
+ out %eax, %dx
+ jmp 6f #node_core_f16_shared #
+
+node_core_f16_AP:
+ mov %bl, %al # AL = cpu extended model
+ shr $8, %bx # BL = CPUID Fn0000_0001_EBX[LocalApicId]
+ #.if (al == 0)
+ cmp $3, %al # Is This ML?
+ jnz 5f
+ #.else
+4: mov %bx, %si
+ jmp 6f
+5: #node_core_f16_AP_not_KB
+ bts $FLAG_CORE_NOT_IDENTIFIED, %esi
+ #.endif
+ #.endif
+ #
+ # determine if this core shares MTRRs
+ #
+6: #node_core_f16_shared
+ mov $COMPUTE_UNIT_STATUS, %eax # Compute Unit Status
+ mov %si, %bx
+ shl $3, %bh # Move node# to PCI Dev# field
+ add %bh, %ah # Adjust for node number
+ mov $0x0CF8, %dx
+ out %eax, %dx
+ add $4, %dx
+ in %dx, %eax # [3:0]=Enabled# [19:16]=DualCore
+
+ # BL is MyCore#
+ mov $0x04, %cx # Use CH as 'first of pair' core#
+ #.while (cl > 0)
+ jmp 0f
+ 8:
+ bt $CU_ENABLED, %eax # Is pair enabled?
+ #.if (carry?) #
+ jnc 1f
+ mov $0x01, %bh # flag core as primary
+ #.break .if (ch == bl) # Does 1st match MyCore#?
+ cmp %bl, %ch
+ je 9f
+ #inc %ch
+ xor %bh, %bh # flag core as NOT primary
+ #
+ bt $DUAL_CORE, %eax
+ #.if (carry?)
+ jnc 5f
+ add $1, %eax
+ #.endif
+ 5:
+ bt $TRIPLE_CORE, %eax
+ jnc 5f
+ #.if (carry?)
+ add $2, %eax
+ #.endif
+ 5:
+ bt $QUAD_CORE, %eax
+ jnc 5f
+ #.if (carry?)
+ add $3, %eax
+ #.endif
+ 5:
+ #.break .if (ch >= bl) #TODO:
+ cmp %bl, %ch
+ jae 9f
+ inc %ch
+ #.endif
+ 1:
+ shr $1, %eax
+ dec %cl
+ 0:
+ cmp $0x0, %cl
+ ja 8b
+
+ #.endw
+ 9:
+
+ #.if (cl == 0)
+ or %cl, %cl
+ jne 1f
+ #Error - core# didn't match Compute Unit Status content
+ bts $FLAG_CORE_NOT_IDENTIFIED, %esi
+ bts $FLAG_IS_PRIMARY, %esi # Set Is_Primary for unknowns
+ #.endif
+ 1:
+ #.if (bh != 0) # Check state of primary for the matched core
+ or %bh, %bh
+ je 2f
+ bts $FLAG_IS_PRIMARY, %esi # Set shared flag into return value
+ #.endif
+ 2:
+
+node_core_f16_exit:
+
+.endm
+
+/*****************************************************************************
+* AMD_ENABLE_STACK: Setup a stack
+*
+* In:
+* EBX = Return address (preserved)
+*
+* Out:
+* SS:ESP - Our new private stack location
+*
+* EAX = AGESA_STATUS
+*
+* ECX = Stack size in bytes
+*
+* Requirements:
+* * This routine presently is limited to a max of 64 processor cores
+* Preserved:
+* ebx ebp
+* Destroyed:
+* eax, ecx, edx, edi, esi, ds, es, ss, esp
+* mmx0, mmx1
+*
+* Description:
+* Fixed MTRR address allocation to cores:
+* The BSP gets 64K of stack, Core0 of each node gets 16K of stack, all other cores get 4K.
+* There is a max of 1 BSP, 7 core0s and 56 other cores.
+* Although each core has it's own cache storage, they share the address space. Each core must
+* be assigned a private and unique address space for its stack. To support legacy systems,
+* the stack needs to be within the legacy address space (1st 1Meg). Room must also be reserved
+* for the other legacy elements (Interrupt vectors, BIOS ROM, video buffer, etc.)
+*
+* 80000h 40000h 00000h
+* +----------+----------+----------+----------+----------+----------+----------+----------+
+* 64K | | | | | | | | | 64K ea
+* ea +----------+----------+----------+----------+----------+----------+----------+----------+
+* | MTRR 0000_0250 MTRRfix64K_00000 |
+* +----------+----------+----------+----------+----------+----------+----------+----------+
+* | 7 , 6 | 5 , 4 | 3 , 2 | 1 , 0 | 0 | | | | <-node
+* |7..1,7..1 |7..1,7..1 |7..1,7..1 |7..1,7..1 | 0 | | | | <-core
+* +----------+----------+----------+----------+----------+----------+----------+----------+
+*
+* C0000h B0000h A0000h 90000h 80000h
+* +------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+
+*16K | | | | | | | | | | | | | | | | |
+* ea +------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+
+* | MTRR 0259 MTRRfix16K_A0000 | MTRR 0258 MTRRfix16K_80000 |
+* +------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+
+* | > Dis|play B|uffer | < | | | | | 7 | 6 | 5 | 4 | 3 | 2 | 1 | | <-node
+* | > T| e m |p o r |a r y | B u |f f e |r A |r e a<| 0 | 0 | 0 | 0 | 0 | 0 | 0 | | <-core
+* +------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+------+
+*
+* E0000h D0000h C0000h
+* +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+* 4K | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | 4K ea
+* ea +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+* | 026B MTRRfix4K_D8000 | 026A MTRRfix4K_D0000 | 0269 MTRRfix4K_C8000 | 0268 MTRRfix4K_C0000 |
+* +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+* | | | | | | | | | | | | | | | | | >| V| I| D| E| O| |B |I |O |S | |A |r |e |a<|
+* +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+*
+* 100000h F0000h E0000h
+* +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+* | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | 4K ea
+* +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+* | 026F MTRRfix4K_F8000 | 026E MTRRfix4K_F0000 | 026D MTRRfix4K_E8000 | 026C MTRRfix4K_E0000 |
+* +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+* | >|MA|IN| B|IO|S |RA|NG|E | | | | | | |< | >|EX|TE|ND|ED| B|IO|S |ZO|NE| | | | | |< |
+* +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+*****************************************************************************/
+.macro AMD_ENABLE_STACK
+
+# These are local labels. Declared so linker doesn't cause 'redefined label' errors
+ LOCAL SetupStack
+ LOCAL Real16bMode
+ LOCAL Protected32Mode
+ LOCAL ClearTheStack
+
+# Note that SS:ESP will be default stack. Note that this stack
+# routine will not be used after memory has been initialized. Because
+# of its limited lifetime, it will not conflict with typical PCI devices.
+ movd %ebx, %mm0 # Put return address in a safe place
+ movd %ebp, %mm1 # Save some other user registers
+
+ # get node id and core id of current executing core
+ GET_NODE_ID_CORE_ID # Sets ESI[23:16]=Shared core## SI[15,8]= Node## SI[7,0]= core# (relative to node)
+ # Note: ESI[31:24] are used for flags: Unrecognized Family, Is_Primary core, Stack already established
+
+ # determine if stack is already enabled. We are using the DefType MSR for this determination.
+ # It is =0 after reset; CAR setup sets it to enable the MTRRs
+ mov %cr0, %eax
+ test $CR0_MASK, %eax # Is cache disabled? (CD & NW bits)
+ jnz SetupStack # Jump if yes
+ mov $AMD_MTRR_DEFTYPE, %ecx # MSR:0000_02FF
+ _RDMSR
+ test $MSR_MASK, %eax # Are the default types enabled? (MTRR_DEF_TYPE_EN + MTRR_DEF_TYPE_FIX_EN)
+ jz SetupStack # Jump if no
+ or $FLAG_STACK_REENTRY, %esi # Bit25, indicate stack has already been initialized
+
+SetupStack:
+ # Set node to map the first 16MB to node 0# 0000_0000 to 00FF_FFFF as DRAM
+ mov %esi, %ebx # Get my Node/Core info
+ xor %bl, %bl
+ shl $3, %bh # Isolate my node#, match alignment for PCI Dev#
+ mov $0x8000C144, %eax # D18F1x44:DRAM Base/Limit# N is Base, N+4 is Limit
+ add %bh, %ah
+ mov %eax, %ebx # Save PCI address for Base/Limit pair
+
+ mov $0x0CF8, %dx
+ out %eax, %dx
+ add $4, %dx
+ xor %eax, %eax # Least Significant bit is AD24 so 0 sets mask of 00FF_FFFF (16MB)
+ out %eax, %dx # DRAM Limit = node0, no interleave
+
+ mov %ebx, %eax
+ sub $4, %eax # Now point to the Base register
+ mov $0x0CF8, %dx
+ out %eax, %dx
+ add $4, %dx
+ mov $0x00000003, %eax # Set the read and write enable bits
+ out %eax, %dx # DRAM Base = 0x0000, R/W
+
+ AMD_ENABLE_STACK_FAMILY_HOOK
+
+ # Init CPU MSRs for our init routines
+ mov $MTRR_SYS_CFG, %ecx # SYS_CFG
+ _RDMSR
+ bts $MTRR_FIX_DRAM_MOD_EN, %eax # Turn on modification enable bit
+ _WRMSR
+
+ mov %esi, %eax
+ bt $FLAG_STACK_REENTRY, %eax # Is this a 2nd entry?
+ #.if (!carry?) # On a re-entry, do not clear MTRRs or reset TOM; just reset the stack SS:ESP
+ jc 0f
+ bt $FLAG_IS_PRIMARY, %eax # Is this core the primary in a compute unit?
+ #.if (carry?) # Families using shared groups do not need to clear the MTRRs since that is done at power-on reset
+ # Note: Relying on MSRs to be cleared to 0's at reset for families w/shared cores
+ # Clear all variable and Fixed MTRRs for non-shared cores
+ jnc 0f
+ mov $AMD_MTRR_VARIABLE_BASE0, %ecx
+ xor %eax, %eax
+ xor %edx, %edx
+ #.while (cl != 10h) # Variable MTRRphysBase[n] and MTRRphysMask[n]
+ jmp 1f
+ 2:
+ _WRMSR
+ inc %cl
+ #.endw
+ 1:
+ cmp $0x10, %cl
+ jne 2b
+ mov $AMD_MTRR_FIX64k_00000, %cx # MSR:0000_0250
+ _WRMSR
+ mov $AMD_MTRR_FIX16k_80000, %cx # MSR:0000_0258
+ _WRMSR
+ mov $AMD_MTRR_FIX16k_A0000, %cx # MSR:0000_0259
+ _WRMSR
+ mov $AMD_MTRR_FIX4k_C0000, %cx # Fixed 4Ks: MTRRfix4K_C0000 to MTRRfix4K_F8000
+ #.while (cl != 70h)
+ jmp 3f
+ 4:
+ _WRMSR
+ inc %cl
+ #.endw
+ 3:
+ cmp $0x70, %cl
+ jne 4b
+ # Set TOP_MEM (C001_001A) for non-shared cores to 16M. This will be increased at heap init.
+ # - not strictly needed since the FixedMTRRs take presedence.
+ mov $(16 * 1024 * 1024), %eax
+ mov $TOP_MEM, %ecx # MSR:C001_001A
+ _WRMSR
+ #.endif # End Is_Primary
+ #.endif # End Stack_ReEntry
+ 0:
+ # Clear IORRs (C001_0016-19) and TOM2(C001_001D) for all cores
+ xor %eax, %eax
+ xor %edx, %edx
+ mov $IORR_BASE, %ecx # MSR:C001_0016 - 0019
+ #.while (cl != 1Ah)
+ jmp 1f
+ 2:
+ _WRMSR
+ inc %cl
+ #.endw
+ 1:
+ cmp $0x1A, %cl
+ jne 2b
+ mov $TOP_MEM2, %ecx # MSR:C001_001D
+ _WRMSR
+
+ # setup MTTRs for stacks
+ # A speculative read can be generated by a speculative fetch mis-aligned in a code zone
+ # or due to a data zone being interpreted as code. When a speculative read occurs outside a
+ # controlled region (intentionally used by software), it could cause an unwanted cache eviction.
+ # To prevent speculative reads from causing an eviction, the unused cache ranges are set
+ # to UC type. Only the actively used regions (stack, heap) are reflected in the MTRRs.
+ # Note: some core stack regions will share an MTRR since the control granularity is much
+ # larger than the allocated stack zone. The allocation algorithm must account for this 'extra'
+ # space covered by the MTRR when parseling out cache space for the various uses. In some cases
+ # this could reduce the amount of EXE cache available to a core. see cpuCacheInit.c
+ #
+ # Outcome of this block is that: (Note the MTRR map at the top of the file)
+ # ebp - start address of stack block
+ # ebx - [31:16] - MTRR MSR address
+ # - [15:8] - slot# in MTRR register
+ # - [7:0] - block size in #4K blocks
+ # review: ESI[31:24]=Flags; SI[15,8]= Node#; SI[7,0]= core# (relative to node)
+ #
+
+ mov %si, %ax # Load node, core
+ #.if (al == 0) # Is a core 0?
+ or %al, %al
+ jne 1f
+ #.if (ah == 0) # Is Node 0? (BSP)
+ or %ah, %ah
+ jne 2f
+ # Is BSP, assign a 64K stack
+ mov $((AMD_MTRR_FIX64k_00000 << 16) + (3 << 8) + (BSP_STACK_SIZE >> 12)), %ebx
+ mov $BSP_STACK_BASE_ADDR, %ebp
+ jmp 0f
+ #.else # node 1 to 7, core0
+ 2:
+ # Is a Core0 of secondary node, assign 16K stacks
+ mov $AMD_MTRR_FIX16k_80000, %bx
+ shl $16, %ebx #
+ mov %ah, %bh # Node# is used as slot#
+ mov $(CORE0_STACK_SIZE >> 12), %bl
+ mov %ah, %al # Base = (Node# * Size)#
+ mul %bl #
+ movzx %ax, %eax #
+ shl $12, %eax # Expand back to full byte count (* 4K)
+ add $CORE0_STACK_BASE_ADDR, %eax
+ mov %eax, %ebp
+ #.endif
+ jmp 0f
+ #.else #core 1 thru core 7
+ 1:
+ # Is core 1-7 of any node, assign 4K stacks
+ mov $8, %al # CoreIndex = ( (Node# * 8) ...
+ mul %ah #
+ mov %si, %bx #
+ add %bl, %al # ... + Core#)#
+
+ mov $AMD_MTRR_FIX64k_00000, %bx
+ shl $16, %ebx #
+ mov %al, %bh # Slot# = (CoreIndex / 16) + 4#
+ shr $4, %bh #
+ add $4, %bh #
+ mov $(CORE1_STACK_SIZE >> 12), %bl
+
+ mul %bl # Base = ( (CoreIndex * Size) ...
+ movzx %ax, %eax #
+ shl $12, %eax # Expand back to full byte count (* 4K)
+ add $CORE1_STACK_BASE_ADDR, %eax # ... + Base_Addr)#
+ mov %eax, %ebp
+ #.endif
+ 0:
+
+ # Now set the MTRR. Add this to already existing settings (don't clear any MTRR)
+ mov $WB_DRAM_TYPE, %edi # Load Cache type in 1st slot
+ mov %bh, %cl # ShiftCount = ((slot# ...
+ and $0x03, %cl # ... % 4) ...
+ shl $0x03, %cl # ... * 8)#
+ shl %cl, %edi # Cache type is now in correct position
+ ror $16, %ebx # Get the MTRR address
+ movzx %bx, %ecx #
+ rol $16, %ebx # Put slot# & size back in BX
+ _RDMSR # Read-modify-write the MSR
+ #.if (bh < 4) # Is value in lower or upper half of MSR?
+ cmp $4, %bh
+ jae 1f
+ or %edi, %eax #
+ jmp 0f
+ #.else
+ 1: #
+ or %edi, %edx #
+ #.endif #
+ 0:
+ _WRMSR #
+
+ # Enable MTRR defaults as UC type
+ mov $AMD_MTRR_DEFTYPE, %ecx # MSR:0000_02FF
+ _RDMSR # Read-modify-write the MSR
+ bts $MTRR_DEF_TYPE_EN, %eax # MtrrDefTypeEn
+ bts $MTRR_DEF_TYPE_FIX_EN, %eax # MtrrDefTypeFixEn
+ _WRMSR
+
+ # Close the modification window on the Fixed MTRRs
+ mov $MTRR_SYS_CFG, %ecx # MSR:0C001_0010
+ _RDMSR
+ bts $MTRR_FIX_DRAM_EN, %eax # MtrrFixDramEn
+ bts $MTRR_VAR_DRAM_EN, %eax # variable MTRR enable bit
+ btr $MTRR_FIX_DRAM_MOD_EN, %eax # Turn off modification enable bit
+ _WRMSR
+
+ # Enable caching in CR0
+ mov %cr0, %eax # Enable WT/WB cache
+ btr $CR0_PG, %eax # Make sure paging is disabled
+ btr $CR0_CD, %eax # Clear CR0 NW and CD
+ btr $CR0_NW, %eax
+ mov %eax, %cr0
+
+ # Use the Stack Base & size to calculate SS and ESP values
+ # review:
+ # esi[31:24]=Flags; esi[15,8]= Node#; esi[7,0]= core# (relative to node)
+ # ebp - start address of stack block
+ # ebx - [31:16] - MTRR MSR address
+ # - [15:8] - slot# in MTRR register
+ # - [7:0] - block size in #4K blocks
+ #
+ mov %ebp, %esp # Initialize the stack pointer
+ mov %esp, %edi # Copy the stack start to edi
+ movzx %bl, %bx
+ movzx %bx, %ebx # Clear upper ebx, don't need MSR addr anymore
+ shl $12, %ebx # Make size full byte count (* 4K)
+ add %ebx, %esp # Set the Stack Pointer as full linear address
+ sub $4, %esp
+ #
+ # review:
+ # esi[31:24]=Flags; esi[15,8]= Node#; esi[7,0]= core# (relative to node)
+ # edi - 32b start address of stack block
+ # ebx - size of stack block
+ # esp - 32b linear stack pointer
+ #
+
+ # Determine mode for SS base;
+ mov %cr0, %ecx # Check for 32-bit protect mode
+ bt $CR0_PE, %ecx #
+ #.if (!carry?) # PE=0 means real mode
+ jc Protected32Mode
+ mov %cs, %cx # PE=1
+ cmp $0x0D000, %cx # Check for CS
+ jb Protected32Mode # If CS < D000, it is a selector instead of a segment
+ # alter SS:ESP for 16b Real Mode:
+Real16bMode:
+ mov %edi, %eax
+ shr $4, %eax # Create a Real Mode segment for ss, ds, es
+ mov %ax, %ss
+ mov %ax, %ds
+ mov %ax, %es
+ shl $4, %eax
+ sub %eax, %edi # Adjust the clearing pointer for Seg:Offset mode
+ mov %ebx, %esp # Make SP an offset from SS
+ sub $4, %esp #
+ # .endif # endif
+ # #else
+ # Default is to use Protected 32b Mode
+ #.endif
+ ;
+Protected32Mode:
+ #
+ # Clear The Stack
+ # Now that we have set the location and the MTRRs, initialize the cache by
+ # reading then writing to zero all of the stack area.
+ # review:
+ # ss - Stack base
+ # esp - stack pointer
+ # ebx - size of stack block
+ # esi[31:24]=Flags; esi[15,8]= Node#; esi[7,0]= core# (relative to node)
+ # edi - address of start of stack block
+ #
+
+ClearTheStack: # Stack base is in SS, stack pointer is in ESP
+ shr $2, %ebx # ebx = stack block size in dwords
+ mov %bx, %cx #
+ # Check our flags - Don't clear an existing stack
+ #.if ( !(esi & 0FF000000h)) # Check our flags
+ test $(1 << FLAG_STACK_REENTRY), %esi
+ jne 1f
+ cld
+ mov %edi, %esi
+ rep lodsl (%esi) # Pre-load the range
+ xor %eax, %eax
+ mov %bx, %cx
+ mov %edi, %esi # Preserve base for push on stack
+ rep stosl (%edi) # Clear the range
+ movl $0x0ABCDDCBA, (%esp) # Put marker in top stack dword
+ shl $2, %ebx # Put stack size and base
+ push %ebx # in top of stack
+ push %esi
+
+ mov %ebx, %ecx # Return size of stack in bytes
+ xor %eax, %eax # eax = 0 : no error return code
+ jmp 0f
+ #.else
+ 1:
+ movzx %cx, %ecx
+ shl $2, %ecx # Return size of stack in bytes
+ mov %esi, %eax
+ shr $24, %eax # Keep the flags as part of the error report
+ or $0x40000000, %eax # eax = AGESA_WARNING (Stack has already been set up)
+ #.endif
+ 0:
+
+ movd %mm0, %ebx # Restore return address
+ movd %mm1, %ebp
+.endm
+
+/*****************************************************************************
+* AMD_DISABLE_STACK: Destroy the stack inside the cache. This routine
+* should only be executed on the BSP
+*
+* In:
+* none
+*
+* Out:
+* EAX = AGESA_SUCCESS
+*
+* Preserved:
+* ebx
+* Destroyed:
+* eax, ecx, edx, esp
+*****************************************************************************/
+.macro AMD_DISABLE_STACK
+
+ mov %ebx, %esp # Save return address
+
+ # get node/core/flags of current executing core
+ GET_NODE_ID_CORE_ID # Sets ESI[15,8]= Node#; ESI[7,0]= core# (relative to node)
+
+ # Turn on modification enable bit
+ mov $MTRR_SYS_CFG, %ecx # MSR:C001_0010
+ _RDMSR
+ bts $MTRR_FIX_DRAM_MOD_EN, %eax # Enable modifications
+ _WRMSR
+
+ # Set lower 640K MTRRs for Write-Back memory caching
+ mov $AMD_MTRR_FIX64k_00000, %ecx
+ mov $0x1E1E1E1E, %eax
+ mov %eax, %edx
+ _WRMSR # 0 - 512K = WB Mem
+ mov $AMD_MTRR_FIX16k_80000, %ecx
+ _WRMSR # 512K - 640K = WB Mem
+
+ # Turn off modification enable bit
+ mov $MTRR_SYS_CFG, %ecx # MSR:C001_0010
+ _RDMSR
+ btr $MTRR_FIX_DRAM_MOD_EN, %eax # Disable modification
+ _WRMSR
+
+ AMD_DISABLE_STACK_FAMILY_HOOK # Re-Enable 'normal' cache operations
+
+ mov %esp, %ebx
+ xor %eax, %eax
+
+.endm
+