[mirror_edk2.git] / UefiCpuPkg / Library / MpInitLib / X64 / MpFuncs.nasm

;------------------------------------------------------------------------------ ;\r
; Copyright (c) 2015 - 2017, Intel Corporation. All rights reserved.<BR>\r
; This program and the accompanying materials\r
; are licensed and made available under the terms and conditions of the BSD License\r
; which accompanies this distribution.  The full text of the license may be found at\r
; http://opensource.org/licenses/bsd-license.php.\r
;\r
; THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,\r
; WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.\r
;\r
; Module Name:\r
;\r
;   MpFuncs.nasm\r
;\r
; Abstract:\r
;\r
;   This is the assembly code for MP support\r
;\r
;-------------------------------------------------------------------------------\r
\r
%include "MpEqu.inc"\r
extern ASM_PFX(InitializeFloatingPointUnits)\r
\r
DEFAULT REL\r
\r
SECTION .text\r
\r
;-------------------------------------------------------------------------------------\r
;RendezvousFunnelProc  procedure follows. All APs execute their procedure. This\r
;procedure serializes all the AP processors through an Init sequence. It must be\r
;noted that APs arrive here very raw...ie: real mode, no stack.\r
;ALSO THIS PROCEDURE IS EXECUTED BY APs ONLY ON 16 BIT MODE. HENCE THIS PROC\r
;IS IN MACHINE CODE.\r
;-------------------------------------------------------------------------------------\r
global ASM_PFX(RendezvousFunnelProc)\r
ASM_PFX(RendezvousFunnelProc):\r
RendezvousFunnelProcStart:\r
; At this point CS = 0x(vv00) and ip= 0x0.\r
; Save BIST information to ebp firstly\r
\r
BITS 16\r
    mov        ebp, eax                        ; Save BIST information\r
\r
    mov        ax, cs\r
    mov        ds, ax\r
    mov        es, ax\r
    mov        ss, ax\r
    xor        ax, ax\r
    mov        fs, ax\r
    mov        gs, ax\r
\r
    mov        si,  BufferStartLocation\r
    mov        ebx, [si]\r
\r
    mov        di,  ModeOffsetLocation\r
    mov        eax, [di]\r
    mov        di,  CodeSegmentLocation\r
    mov        edx, [di]\r
    mov        di,  ax\r
    sub        di,  02h\r
    mov        [di],dx                         ; Patch long mode CS\r
    sub        di,  04h\r
    add        eax, ebx\r
    mov        [di],eax                        ; Patch address\r
\r
    mov        si, GdtrLocation\r
o32 lgdt       [cs:si]\r
\r
    mov        si, IdtrLocation\r
o32 lidt       [cs:si]\r
\r
    mov        si, EnableExecuteDisableLocation\r
    cmp        byte [si], 0\r
    jz         SkipEnableExecuteDisableBit\r
\r
    ;\r
    ; Enable execute disable bit\r
    ;\r
    mov        ecx, 0c0000080h             ; EFER MSR number\r
    rdmsr                                  ; Read EFER\r
    bts        eax, 11                     ; Enable Execute Disable Bit\r
    wrmsr                                  ; Write EFER\r
\r
SkipEnableExecuteDisableBit:\r
\r
    mov        di,  DataSegmentLocation\r
    mov        edi, [di]                   ; Save long mode DS in edi\r
\r
    mov        si, Cr3Location             ; Save CR3 in ecx\r
    mov        ecx, [si]\r
\r
    xor        ax,  ax\r
    mov        ds,  ax                     ; Clear data segment\r
\r
    mov        eax, cr0                    ; Get control register 0\r
    or         eax, 000000003h             ; Set PE bit (bit #0) & MP\r
    mov        cr0, eax\r
\r
    mov        eax, cr4\r
    bts        eax, 5\r
    mov        cr4, eax\r
\r
    mov        cr3, ecx                    ; Load CR3\r
\r
    mov        ecx, 0c0000080h             ; EFER MSR number\r
    rdmsr                                  ; Read EFER\r
    bts        eax, 8                      ; Set LME=1\r
    wrmsr                                  ; Write EFER\r
\r
    mov        eax, cr0                    ; Read CR0\r
    bts        eax, 31                     ; Set PG=1\r
    mov        cr0, eax                    ; Write CR0\r
\r
    jmp        0:strict dword 0  ; far jump to long mode\r
BITS 64\r
LongModeStart:\r
    mov        eax, edi\r
    mov        ds,  ax\r
    mov        es,  ax\r
    mov        ss,  ax\r
\r
    mov        esi, ebx\r
    lea        edi, [esi + InitFlagLocation]\r
    cmp        qword [edi], 1       ; ApInitConfig\r
    jnz        GetApicId\r
\r
    ; AP init\r
    mov        edi, esi\r
    add        edi, LockLocation\r
    mov        rax, NotVacantFlag\r
\r
TestLock:\r
    xchg       qword [edi], rax\r
    cmp        rax, NotVacantFlag\r
    jz         TestLock\r
\r
    lea        ecx, [esi + NumApsExecutingLocation]\r
    inc        dword [ecx]\r
    mov        ebx, [ecx]\r
\r
Releaselock:\r
    mov        rax, VacantFlag\r
    xchg       qword [edi], rax\r
    ; program stack\r
    mov        edi, esi\r
    add        edi, StackSizeLocation\r
    mov        eax, dword [edi]\r
    mov        ecx, ebx\r
    inc        ecx\r
    mul        ecx                               ; EAX = StackSize * (CpuNumber + 1)\r
    mov        edi, esi\r
    add        edi, StackStartAddressLocation\r
    add        rax, qword [edi]\r
    mov        rsp, rax\r
    jmp        CProcedureInvoke\r
\r
GetApicId:\r
    mov        eax, 0\r
    cpuid\r
    cmp        eax, 0bh\r
    jb         NoX2Apic             ; CPUID level below CPUID_EXTENDED_TOPOLOGY\r
\r
    mov        eax, 0bh\r
    xor        ecx, ecx\r
    cpuid\r
    test       ebx, 0ffffh\r
    jz         NoX2Apic             ; CPUID.0BH:EBX[15:0] is zero\r
\r
    ; Processor is x2APIC capable; 32-bit x2APIC ID is already in EDX\r
    jmp        GetProcessorNumber\r
\r
NoX2Apic:\r
    ; Processor is not x2APIC capable, so get 8-bit APIC ID\r
    mov        eax, 1\r
    cpuid\r
    shr        ebx, 24\r
    mov        edx, ebx\r
\r
GetProcessorNumber:\r
    ;\r
    ; Get processor number for this AP\r
    ; Note that BSP may become an AP due to SwitchBsp()\r
    ;\r
    xor         ebx, ebx\r
    lea         eax, [esi + CpuInfoLocation]\r
    mov         edi, [eax]\r
\r
GetNextProcNumber:\r
    cmp         dword [edi], edx                      ; APIC ID match?\r
    jz          ProgramStack\r
    add         edi, 20\r
    inc         ebx\r
    jmp         GetNextProcNumber    \r
\r
ProgramStack:\r
    mov         rsp, qword [edi + 12]\r
\r
CProcedureInvoke:\r
    push       rbp               ; Push BIST data at top of AP stack\r
    xor        rbp, rbp          ; Clear ebp for call stack trace\r
    push       rbp\r
    mov        rbp, rsp\r
\r
    mov        rax, qword [esi + InitializeFloatingPointUnitsAddress]\r
    sub        rsp, 20h\r
    call       rax               ; Call assembly function to initialize FPU per UEFI spec\r
    add        rsp, 20h\r
\r
    mov        edx, ebx          ; edx is NumApsExecuting\r
    mov        ecx, esi\r
    add        ecx, LockLocation ; rcx is address of exchange info data buffer\r
\r
    mov        edi, esi\r
    add        edi, ApProcedureLocation\r
    mov        rax, qword [edi]\r
\r
    sub        rsp, 20h\r
    call       rax               ; Invoke C function\r
    add        rsp, 20h\r
    jmp        $                 ; Should never reach here\r
\r
RendezvousFunnelProcEnd:\r
\r
;-------------------------------------------------------------------------------------\r
;  AsmRelocateApLoop (MwaitSupport, ApTargetCState, PmCodeSegment, TopOfApStack, CountTofinish);\r
;-------------------------------------------------------------------------------------\r
global ASM_PFX(AsmRelocateApLoop)\r
ASM_PFX(AsmRelocateApLoop):\r
AsmRelocateApLoopStart:\r
    mov        rax, [rsp + 40]   ; CountTofinish\r
    lock dec   dword [rax]       ; (*CountTofinish)--\r
    mov        rsp, r9\r
    push       rcx\r
    push       rdx\r
\r
    lea        rsi, [PmEntry]    ; rsi <- The start address of transition code\r
\r
    push       r8\r
    push       rsi\r
    DB         0x48\r
    retf\r
BITS 32\r
PmEntry:\r
    mov        eax, cr0\r
    btr        eax, 31           ; Clear CR0.PG\r
    mov        cr0, eax          ; Disable paging and caches\r
\r
    mov        ebx, edx          ; Save EntryPoint to rbx, for rdmsr will overwrite rdx\r
    mov        ecx, 0xc0000080\r
    rdmsr\r
    and        ah, ~ 1           ; Clear LME\r
    wrmsr\r
    mov        eax, cr4\r
    and        al, ~ (1 << 5)    ; Clear PAE\r
    mov        cr4, eax\r
\r
    pop        edx\r
    add        esp, 4\r
    pop        ecx,\r
    add        esp, 4\r
    cmp        cl, 1              ; Check mwait-monitor support\r
    jnz        HltLoop\r
    mov        ebx, edx           ; Save C-State to ebx\r
MwaitLoop:\r
    mov        eax, esp           ; Set Monitor Address\r
    xor        ecx, ecx           ; ecx = 0\r
    xor        edx, edx           ; edx = 0\r
    monitor\r
    mov        eax, ebx           ; Mwait Cx, Target C-State per eax[7:4]\r
    shl        eax, 4\r
    mwait\r
    jmp        MwaitLoop\r
HltLoop:\r
    cli\r
    hlt\r
    jmp        HltLoop\r
BITS 64\r
AsmRelocateApLoopEnd:\r
\r
;-------------------------------------------------------------------------------------\r
;  AsmGetAddressMap (&AddressMap);\r
;-------------------------------------------------------------------------------------\r
global ASM_PFX(AsmGetAddressMap)\r
ASM_PFX(AsmGetAddressMap):\r
    lea        rax, [ASM_PFX(RendezvousFunnelProc)]\r
    mov        qword [rcx], rax\r
    mov        qword [rcx +  8h], LongModeStart - RendezvousFunnelProcStart\r
    mov        qword [rcx + 10h], RendezvousFunnelProcEnd - RendezvousFunnelProcStart\r
    lea        rax, [ASM_PFX(AsmRelocateApLoop)]\r
    mov        qword [rcx + 18h], rax\r
    mov        qword [rcx + 20h], AsmRelocateApLoopEnd - AsmRelocateApLoopStart\r
    ret\r
\r
;-------------------------------------------------------------------------------------\r
;AsmExchangeRole procedure follows. This procedure executed by current BSP, that is\r
;about to become an AP. It switches its stack with the current AP.\r
;AsmExchangeRole (IN   CPU_EXCHANGE_INFO    *MyInfo, IN   CPU_EXCHANGE_INFO    *OthersInfo);\r
;-------------------------------------------------------------------------------------\r
global ASM_PFX(AsmExchangeRole)\r
ASM_PFX(AsmExchangeRole):\r
    ; DO NOT call other functions in this function, since 2 CPU may use 1 stack\r
    ; at the same time. If 1 CPU try to call a function, stack will be corrupted.\r
\r
    push       rax\r
    push       rbx\r
    push       rcx\r
    push       rdx\r
    push       rsi\r
    push       rdi\r
    push       rbp\r
    push       r8\r
    push       r9\r
    push       r10\r
    push       r11\r
    push       r12\r
    push       r13\r
    push       r14\r
    push       r15\r
\r
    mov        rax, cr0\r
    push       rax\r
\r
    mov        rax, cr4\r
    push       rax\r
\r
    ; rsi contains MyInfo pointer\r
    mov        rsi, rcx\r
\r
    ; rdi contains OthersInfo pointer\r
    mov        rdi, rdx\r
\r
    ;Store EFLAGS, GDTR and IDTR regiter to stack\r
    pushfq\r
    sgdt       [rsi + 16]\r
    sidt       [rsi + 26]\r
\r
    ; Store the its StackPointer\r
    mov        [rsi + 8], rsp\r
\r
    ; update its switch state to STORED\r
    mov        byte [rsi], CPU_SWITCH_STATE_STORED\r
\r
WaitForOtherStored:\r
    ; wait until the other CPU finish storing its state\r
    cmp        byte [rdi], CPU_SWITCH_STATE_STORED\r
    jz         OtherStored\r
    pause\r
    jmp        WaitForOtherStored\r
\r
OtherStored:\r
    ; Since another CPU already stored its state, load them\r
    ; load GDTR value\r
    lgdt       [rdi + 16]\r
\r
    ; load IDTR value\r
    lidt       [rdi + 26]\r
\r
    ; load its future StackPointer\r
    mov        rsp, [rdi + 8]\r
\r
    ; update the other CPU's switch state to LOADED\r
    mov        byte [rdi], CPU_SWITCH_STATE_LOADED\r
\r
WaitForOtherLoaded:\r
    ; wait until the other CPU finish loading new state,\r
    ; otherwise the data in stack may corrupt\r
    cmp        byte [rsi], CPU_SWITCH_STATE_LOADED\r
    jz         OtherLoaded\r
    pause\r
    jmp        WaitForOtherLoaded\r
\r
OtherLoaded:\r
    ; since the other CPU already get the data it want, leave this procedure\r
    popfq\r
\r
    pop        rax\r
    mov        cr4, rax\r
\r
    pop        rax\r
    mov        cr0, rax\r
\r
    pop        r15\r
    pop        r14\r
    pop        r13\r
    pop        r12\r
    pop        r11\r
    pop        r10\r
    pop        r9\r
    pop        r8\r
    pop        rbp\r
    pop        rdi\r
    pop        rsi\r
    pop        rdx\r
    pop        rcx\r
    pop        rbx\r
    pop        rax\r
\r
    ret\r
Commit	Line	Data
	1	;------------------------------------------------------------------------------ ;\r
	2	; Copyright (c) 2015 - 2017, Intel Corporation. All rights reserved.<BR>\r
	3	; This program and the accompanying materials\r
	4	; are licensed and made available under the terms and conditions of the BSD License\r
	5	; which accompanies this distribution. The full text of the license may be found at\r
	6	; http://opensource.org/licenses/bsd-license.php.\r
	7	;\r
	8	; THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,\r
	9	; WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.\r
	10	;\r
	11	; Module Name:\r
	12	;\r
	13	; MpFuncs.nasm\r
	14	;\r
	15	; Abstract:\r
	16	;\r
	17	; This is the assembly code for MP support\r
	18	;\r
	19	;-------------------------------------------------------------------------------\r
	20	\r
	21	%include "MpEqu.inc"\r
	22	extern ASM_PFX(InitializeFloatingPointUnits)\r
	23	\r
	24	DEFAULT REL\r
	25	\r
	26	SECTION .text\r
	27	\r
	28	;-------------------------------------------------------------------------------------\r
	29	;RendezvousFunnelProc procedure follows. All APs execute their procedure. This\r
	30	;procedure serializes all the AP processors through an Init sequence. It must be\r
	31	;noted that APs arrive here very raw...ie: real mode, no stack.\r
	32	;ALSO THIS PROCEDURE IS EXECUTED BY APs ONLY ON 16 BIT MODE. HENCE THIS PROC\r
	33	;IS IN MACHINE CODE.\r
	34	;-------------------------------------------------------------------------------------\r
	35	global ASM_PFX(RendezvousFunnelProc)\r
	36	ASM_PFX(RendezvousFunnelProc):\r
	37	RendezvousFunnelProcStart:\r
	38	; At this point CS = 0x(vv00) and ip= 0x0.\r
	39	; Save BIST information to ebp firstly\r
	40	\r
	41	BITS 16\r
	42	mov ebp, eax ; Save BIST information\r
	43	\r
	44	mov ax, cs\r
	45	mov ds, ax\r
	46	mov es, ax\r
	47	mov ss, ax\r
	48	xor ax, ax\r
	49	mov fs, ax\r
	50	mov gs, ax\r
	51	\r
	52	mov si, BufferStartLocation\r
	53	mov ebx, [si]\r
	54	\r
	55	mov di, ModeOffsetLocation\r
	56	mov eax, [di]\r
	57	mov di, CodeSegmentLocation\r
	58	mov edx, [di]\r
	59	mov di, ax\r
	60	sub di, 02h\r
	61	mov [di],dx ; Patch long mode CS\r
	62	sub di, 04h\r
	63	add eax, ebx\r
	64	mov [di],eax ; Patch address\r
	65	\r
	66	mov si, GdtrLocation\r
	67	o32 lgdt [cs:si]\r
	68	\r
	69	mov si, IdtrLocation\r
	70	o32 lidt [cs:si]\r
	71	\r
	72	mov si, EnableExecuteDisableLocation\r
	73	cmp byte [si], 0\r
	74	jz SkipEnableExecuteDisableBit\r
	75	\r
	76	;\r
	77	; Enable execute disable bit\r
	78	;\r
	79	mov ecx, 0c0000080h ; EFER MSR number\r
	80	rdmsr ; Read EFER\r
	81	bts eax, 11 ; Enable Execute Disable Bit\r
	82	wrmsr ; Write EFER\r
	83	\r
	84	SkipEnableExecuteDisableBit:\r
	85	\r
	86	mov di, DataSegmentLocation\r
	87	mov edi, [di] ; Save long mode DS in edi\r
	88	\r
	89	mov si, Cr3Location ; Save CR3 in ecx\r
	90	mov ecx, [si]\r
	91	\r
	92	xor ax, ax\r
	93	mov ds, ax ; Clear data segment\r
	94	\r
	95	mov eax, cr0 ; Get control register 0\r
	96	or eax, 000000003h ; Set PE bit (bit #0) & MP\r
	97	mov cr0, eax\r
	98	\r
	99	mov eax, cr4\r
	100	bts eax, 5\r
	101	mov cr4, eax\r
	102	\r
	103	mov cr3, ecx ; Load CR3\r
	104	\r
	105	mov ecx, 0c0000080h ; EFER MSR number\r
	106	rdmsr ; Read EFER\r
	107	bts eax, 8 ; Set LME=1\r
	108	wrmsr ; Write EFER\r
	109	\r
	110	mov eax, cr0 ; Read CR0\r
	111	bts eax, 31 ; Set PG=1\r
	112	mov cr0, eax ; Write CR0\r
	113	\r
	114	jmp 0:strict dword 0 ; far jump to long mode\r
	115	BITS 64\r
	116	LongModeStart:\r
	117	mov eax, edi\r
	118	mov ds, ax\r
	119	mov es, ax\r
	120	mov ss, ax\r
	121	\r
	122	mov esi, ebx\r
	123	lea edi, [esi + InitFlagLocation]\r
	124	cmp qword [edi], 1 ; ApInitConfig\r
	125	jnz GetApicId\r
	126	\r
	127	; AP init\r
	128	mov edi, esi\r
	129	add edi, LockLocation\r
	130	mov rax, NotVacantFlag\r
	131	\r
	132	TestLock:\r
	133	xchg qword [edi], rax\r
	134	cmp rax, NotVacantFlag\r
	135	jz TestLock\r
	136	\r
	137	lea ecx, [esi + NumApsExecutingLocation]\r
	138	inc dword [ecx]\r
	139	mov ebx, [ecx]\r
	140	\r
	141	Releaselock:\r
	142	mov rax, VacantFlag\r
	143	xchg qword [edi], rax\r
	144	; program stack\r
	145	mov edi, esi\r
	146	add edi, StackSizeLocation\r
	147	mov eax, dword [edi]\r
	148	mov ecx, ebx\r
	149	inc ecx\r
	150	mul ecx ; EAX = StackSize * (CpuNumber + 1)\r
	151	mov edi, esi\r
	152	add edi, StackStartAddressLocation\r
	153	add rax, qword [edi]\r
	154	mov rsp, rax\r
	155	jmp CProcedureInvoke\r
	156	\r
	157	GetApicId:\r
	158	mov eax, 0\r
	159	cpuid\r
	160	cmp eax, 0bh\r
	161	jb NoX2Apic ; CPUID level below CPUID_EXTENDED_TOPOLOGY\r
	162	\r
	163	mov eax, 0bh\r
	164	xor ecx, ecx\r
	165	cpuid\r
	166	test ebx, 0ffffh\r
	167	jz NoX2Apic ; CPUID.0BH:EBX[15:0] is zero\r
	168	\r
	169	; Processor is x2APIC capable; 32-bit x2APIC ID is already in EDX\r
	170	jmp GetProcessorNumber\r
	171	\r
	172	NoX2Apic:\r
	173	; Processor is not x2APIC capable, so get 8-bit APIC ID\r
	174	mov eax, 1\r
	175	cpuid\r
	176	shr ebx, 24\r
	177	mov edx, ebx\r
	178	\r
	179	GetProcessorNumber:\r
	180	;\r
	181	; Get processor number for this AP\r
	182	; Note that BSP may become an AP due to SwitchBsp()\r
	183	;\r
	184	xor ebx, ebx\r
	185	lea eax, [esi + CpuInfoLocation]\r
	186	mov edi, [eax]\r
	187	\r
	188	GetNextProcNumber:\r
	189	cmp dword [edi], edx ; APIC ID match?\r
	190	jz ProgramStack\r
	191	add edi, 20\r
	192	inc ebx\r
	193	jmp GetNextProcNumber \r
	194	\r
	195	ProgramStack:\r
	196	mov rsp, qword [edi + 12]\r
	197	\r
	198	CProcedureInvoke:\r
	199	push rbp ; Push BIST data at top of AP stack\r
	200	xor rbp, rbp ; Clear ebp for call stack trace\r
	201	push rbp\r
	202	mov rbp, rsp\r
	203	\r
	204	mov rax, qword [esi + InitializeFloatingPointUnitsAddress]\r
	205	sub rsp, 20h\r
	206	call rax ; Call assembly function to initialize FPU per UEFI spec\r
	207	add rsp, 20h\r
	208	\r
	209	mov edx, ebx ; edx is NumApsExecuting\r
	210	mov ecx, esi\r
	211	add ecx, LockLocation ; rcx is address of exchange info data buffer\r
	212	\r
	213	mov edi, esi\r
	214	add edi, ApProcedureLocation\r
	215	mov rax, qword [edi]\r
	216	\r
	217	sub rsp, 20h\r
	218	call rax ; Invoke C function\r
	219	add rsp, 20h\r
	220	jmp $ ; Should never reach here\r
	221	\r
	222	RendezvousFunnelProcEnd:\r
	223	\r
	224	;-------------------------------------------------------------------------------------\r
	225	; AsmRelocateApLoop (MwaitSupport, ApTargetCState, PmCodeSegment, TopOfApStack, CountTofinish);\r
	226	;-------------------------------------------------------------------------------------\r
	227	global ASM_PFX(AsmRelocateApLoop)\r
	228	ASM_PFX(AsmRelocateApLoop):\r
	229	AsmRelocateApLoopStart:\r
	230	mov rax, [rsp + 40] ; CountTofinish\r
	231	lock dec dword [rax] ; (*CountTofinish)--\r
	232	mov rsp, r9\r
	233	push rcx\r
	234	push rdx\r
	235	\r
	236	lea rsi, [PmEntry] ; rsi <- The start address of transition code\r
	237	\r
	238	push r8\r
	239	push rsi\r
	240	DB 0x48\r
	241	retf\r
	242	BITS 32\r
	243	PmEntry:\r
	244	mov eax, cr0\r
	245	btr eax, 31 ; Clear CR0.PG\r
	246	mov cr0, eax ; Disable paging and caches\r
	247	\r
	248	mov ebx, edx ; Save EntryPoint to rbx, for rdmsr will overwrite rdx\r
	249	mov ecx, 0xc0000080\r
	250	rdmsr\r
	251	and ah, ~ 1 ; Clear LME\r
	252	wrmsr\r
	253	mov eax, cr4\r
	254	and al, ~ (1 << 5) ; Clear PAE\r
	255	mov cr4, eax\r
	256	\r
	257	pop edx\r
	258	add esp, 4\r
	259	pop ecx,\r
	260	add esp, 4\r
	261	cmp cl, 1 ; Check mwait-monitor support\r
	262	jnz HltLoop\r
	263	mov ebx, edx ; Save C-State to ebx\r
	264	MwaitLoop:\r
	265	mov eax, esp ; Set Monitor Address\r
	266	xor ecx, ecx ; ecx = 0\r
	267	xor edx, edx ; edx = 0\r
	268	monitor\r
	269	mov eax, ebx ; Mwait Cx, Target C-State per eax[7:4]\r
	270	shl eax, 4\r
	271	mwait\r
	272	jmp MwaitLoop\r
	273	HltLoop:\r
	274	cli\r
	275	hlt\r
	276	jmp HltLoop\r
	277	BITS 64\r
	278	AsmRelocateApLoopEnd:\r
	279	\r
	280	;-------------------------------------------------------------------------------------\r
	281	; AsmGetAddressMap (&AddressMap);\r
	282	;-------------------------------------------------------------------------------------\r
	283	global ASM_PFX(AsmGetAddressMap)\r
	284	ASM_PFX(AsmGetAddressMap):\r
	285	lea rax, [ASM_PFX(RendezvousFunnelProc)]\r
	286	mov qword [rcx], rax\r
	287	mov qword [rcx + 8h], LongModeStart - RendezvousFunnelProcStart\r
	288	mov qword [rcx + 10h], RendezvousFunnelProcEnd - RendezvousFunnelProcStart\r
	289	lea rax, [ASM_PFX(AsmRelocateApLoop)]\r
	290	mov qword [rcx + 18h], rax\r
	291	mov qword [rcx + 20h], AsmRelocateApLoopEnd - AsmRelocateApLoopStart\r
	292	ret\r
	293	\r
	294	;-------------------------------------------------------------------------------------\r
	295	;AsmExchangeRole procedure follows. This procedure executed by current BSP, that is\r
	296	;about to become an AP. It switches its stack with the current AP.\r
	297	;AsmExchangeRole (IN CPU_EXCHANGE_INFO MyInfo, IN CPU_EXCHANGE_INFO OthersInfo);\r
	298	;-------------------------------------------------------------------------------------\r
	299	global ASM_PFX(AsmExchangeRole)\r
	300	ASM_PFX(AsmExchangeRole):\r
	301	; DO NOT call other functions in this function, since 2 CPU may use 1 stack\r
	302	; at the same time. If 1 CPU try to call a function, stack will be corrupted.\r
	303	\r
	304	push rax\r
	305	push rbx\r
	306	push rcx\r
	307	push rdx\r
	308	push rsi\r
	309	push rdi\r
	310	push rbp\r
	311	push r8\r
	312	push r9\r
	313	push r10\r
	314	push r11\r
	315	push r12\r
	316	push r13\r
	317	push r14\r
	318	push r15\r
	319	\r
	320	mov rax, cr0\r
	321	push rax\r
	322	\r
	323	mov rax, cr4\r
	324	push rax\r
	325	\r
	326	; rsi contains MyInfo pointer\r
	327	mov rsi, rcx\r
	328	\r
	329	; rdi contains OthersInfo pointer\r
	330	mov rdi, rdx\r
	331	\r
	332	;Store EFLAGS, GDTR and IDTR regiter to stack\r
	333	pushfq\r
	334	sgdt [rsi + 16]\r
	335	sidt [rsi + 26]\r
	336	\r
	337	; Store the its StackPointer\r
	338	mov [rsi + 8], rsp\r
	339	\r
	340	; update its switch state to STORED\r
	341	mov byte [rsi], CPU_SWITCH_STATE_STORED\r
	342	\r
	343	WaitForOtherStored:\r
	344	; wait until the other CPU finish storing its state\r
	345	cmp byte [rdi], CPU_SWITCH_STATE_STORED\r
	346	jz OtherStored\r
	347	pause\r
	348	jmp WaitForOtherStored\r
	349	\r
	350	OtherStored:\r
	351	; Since another CPU already stored its state, load them\r
	352	; load GDTR value\r
	353	lgdt [rdi + 16]\r
	354	\r
	355	; load IDTR value\r
	356	lidt [rdi + 26]\r
	357	\r
	358	; load its future StackPointer\r
	359	mov rsp, [rdi + 8]\r
	360	\r
	361	; update the other CPU's switch state to LOADED\r
	362	mov byte [rdi], CPU_SWITCH_STATE_LOADED\r
	363	\r
	364	WaitForOtherLoaded:\r
	365	; wait until the other CPU finish loading new state,\r
	366	; otherwise the data in stack may corrupt\r
	367	cmp byte [rsi], CPU_SWITCH_STATE_LOADED\r
	368	jz OtherLoaded\r
	369	pause\r
	370	jmp WaitForOtherLoaded\r
	371	\r
	372	OtherLoaded:\r
	373	; since the other CPU already get the data it want, leave this procedure\r
	374	popfq\r
	375	\r
	376	pop rax\r
	377	mov cr4, rax\r
	378	\r
	379	pop rax\r
	380	mov cr0, rax\r
	381	\r
	382	pop r15\r
	383	pop r14\r
	384	pop r13\r
	385	pop r12\r
	386	pop r11\r
	387	pop r10\r
	388	pop r9\r
	389	pop r8\r
	390	pop rbp\r
	391	pop rdi\r
	392	pop rsi\r
	393	pop rdx\r
	394	pop rcx\r
	395	pop rbx\r
	396	pop rax\r
	397	\r
	398	ret\r