[mirror_edk2.git] / UefiCpuPkg / Library / MpInitLib / Ia32 / MpFuncs.nasm

;------------------------------------------------------------------------------ ;\r
; Copyright (c) 2015 - 2018, 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
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
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        si,  DataSegmentLocation\r
    mov        edx, [si]\r
\r
    ;\r
    ; Get start address of 32-bit code in low memory (<1MB)\r
    ;\r
    mov        edi, ModeTransitionMemoryLocation\r
\r
    mov        si, GdtrLocation\r
o32 lgdt       [cs:si]\r
\r
    mov        si, IdtrLocation\r
o32 lidt       [cs:si]\r
\r
    ;\r
    ; Switch to protected mode\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
    ; Switch to 32-bit code in executable memory (>1MB)\r
o32 jmp far    [cs:di]\r
\r
;\r
; Following code may be copied to memory with type of EfiBootServicesCode.\r
; This is required at DXE phase if NX is enabled for EfiBootServicesCode of\r
; memory.\r
;\r
BITS 32\r
Flat32Start:                                   ; protected mode entry point\r
    mov        ds, dx\r
    mov        es, dx\r
    mov        fs, dx\r
    mov        gs, dx\r
    mov        ss, dx\r
\r
    mov        esi, ebx\r
\r
    ; Increment the number of APs executing here as early as possible\r
    ; This is decremented in C code when AP is finished executing\r
    mov        edi, esi\r
    add        edi, NumApsExecutingLocation\r
    lock inc   dword [edi]\r
\r
    mov         edi, esi\r
    add         edi, EnableExecuteDisableLocation\r
    cmp         byte [edi], 0\r
    jz          SkipEnableExecuteDisable\r
\r
    ;\r
    ; Enable IA32 PAE execute disable\r
    ;\r
\r
    mov         ecx, 0xc0000080\r
    rdmsr\r
    bts         eax, 11\r
    wrmsr\r
\r
    mov         edi, esi\r
    add         edi, Cr3Location\r
    mov         eax, dword [edi]\r
    mov         cr3, eax\r
\r
    mov         eax, cr4\r
    bts         eax, 5\r
    mov         cr4, eax\r
\r
    mov         eax, cr0\r
    bts         eax, 31\r
    mov         cr0, eax\r
\r
SkipEnableExecuteDisable:\r
    mov        edi, esi\r
    add        edi, InitFlagLocation\r
    cmp        dword [edi], 1       ; 1 == ApInitConfig\r
    jnz        GetApicId\r
\r
    ; AP init\r
    mov        edi, esi\r
    add        edi, LockLocation\r
    mov        eax, NotVacantFlag\r
\r
TestLock:\r
    xchg       [edi], eax\r
    cmp        eax, NotVacantFlag\r
    jz         TestLock\r
\r
    mov        ecx, esi\r
    add        ecx, ApIndexLocation\r
    inc        dword [ecx]\r
    mov        ebx, [ecx]\r
\r
Releaselock:\r
    mov        eax, VacantFlag\r
    xchg       [edi], eax\r
\r
    mov        edi, esi\r
    add        edi, StackSizeLocation\r
    mov        eax, [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        eax, [edi]\r
    mov        esp, eax\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         [edi], edx                       ; APIC ID match?\r
    jz          ProgramStack\r
    add         edi, 20\r
    inc         ebx\r
    jmp         GetNextProcNumber\r
\r
ProgramStack:\r
    mov         esp, [edi + 12]\r
\r
CProcedureInvoke:\r
    push       ebp               ; push BIST data at top of AP stack\r
    xor        ebp, ebp          ; clear ebp for call stack trace\r
    push       ebp\r
    mov        ebp, esp\r
\r
    mov        eax, ASM_PFX(InitializeFloatingPointUnits)\r
    call       eax               ; Call assembly function to initialize FPU per UEFI spec\r
\r
    push       ebx               ; Push ApIndex\r
    mov        eax, esi\r
    add        eax, LockLocation\r
    push       eax               ; push address of exchange info data buffer\r
\r
    mov        edi, esi\r
    add        edi, ApProcedureLocation\r
    mov        eax, [edi]\r
\r
    call       eax               ; Invoke C function\r
\r
    jmp        $                 ; Never reach here\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        eax, esp\r
    mov        esp, [eax + 16]     ; TopOfApStack\r
    push       dword [eax]         ; push return address for stack trace\r
    push       ebp\r
    mov        ebp, esp\r
    mov        ebx, [eax + 8]      ; ApTargetCState\r
    mov        ecx, [eax + 4]      ; MwaitSupport\r
    mov        eax, [eax + 20]     ; CountTofinish\r
    lock dec   dword [eax]         ; (*CountTofinish)--\r
    cmp        cl,  1              ; Check mwait-monitor support\r
    jnz        HltLoop\r
MwaitLoop:\r
    cli\r
    mov        eax, esp\r
    xor        ecx, ecx\r
    xor        edx, edx\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
AsmRelocateApLoopEnd:\r
\r
;-------------------------------------------------------------------------------------\r
;  AsmGetAddressMap (&AddressMap);\r
;-------------------------------------------------------------------------------------\r
global ASM_PFX(AsmGetAddressMap)\r
ASM_PFX(AsmGetAddressMap):\r
    pushad\r
    mov        ebp,esp\r
\r
    mov        ebx,  [ebp + 24h]\r
    mov        dword [ebx], RendezvousFunnelProcStart\r
    mov        dword [ebx +  4h], Flat32Start - RendezvousFunnelProcStart\r
    mov        dword [ebx +  8h], RendezvousFunnelProcEnd - RendezvousFunnelProcStart\r
    mov        dword [ebx + 0Ch], AsmRelocateApLoopStart\r
    mov        dword [ebx + 10h], AsmRelocateApLoopEnd - AsmRelocateApLoopStart\r
    mov        dword [ebx + 14h], Flat32Start - RendezvousFunnelProcStart\r
\r
    popad\r
    ret\r
\r
;-------------------------------------------------------------------------------------\r
;AsmExchangeRole procedure follows. This procedure executed by current BSP, that is\r
;about to become an AP. It switches it'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
    pushad\r
    mov        ebp,esp\r
\r
    ; esi contains MyInfo pointer\r
    mov        esi, [ebp + 24h]\r
\r
    ; edi contains OthersInfo pointer\r
    mov        edi, [ebp + 28h]\r
\r
    ;Store EFLAGS, GDTR and IDTR register to stack\r
    pushfd\r
    mov        eax, cr4\r
    push       eax       ; push cr4 firstly\r
    mov        eax, cr0\r
    push       eax\r
\r
    sgdt       [esi + 8]\r
    sidt       [esi + 14]\r
\r
    ; Store the its StackPointer\r
    mov        [esi + 4],esp\r
\r
    ; update its switch state to STORED\r
    mov        byte [esi], CPU_SWITCH_STATE_STORED\r
\r
WaitForOtherStored:\r
    ; wait until the other CPU finish storing its state\r
    cmp        byte [edi], 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       [edi + 8]\r
\r
    ; load IDTR value\r
    lidt       [edi + 14]\r
\r
    ; load its future StackPointer\r
    mov        esp, [edi + 4]\r
\r
    ; update the other CPU's switch state to LOADED\r
    mov        byte [edi], 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 [esi], 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
    pop        eax\r
    mov        cr0, eax\r
    pop        eax\r
    mov        cr4, eax\r
    popfd\r
\r
    popad\r
    ret\r
Commit	Line	Data
d94e5f67	1	;------------------------------------------------------------------------------ ;\r
a7bbe9d2	2	; Copyright (c) 2015 - 2018, Intel Corporation. All rights reserved.<BR>\r
d94e5f67 JF	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	SECTION .text\r
	25	\r
	26	;-------------------------------------------------------------------------------------\r
	27	;RendezvousFunnelProc procedure follows. All APs execute their procedure. This\r
	28	;procedure serializes all the AP processors through an Init sequence. It must be\r
	29	;noted that APs arrive here very raw...ie: real mode, no stack.\r
	30	;ALSO THIS PROCEDURE IS EXECUTED BY APs ONLY ON 16 BIT MODE. HENCE THIS PROC\r
	31	;IS IN MACHINE CODE.\r
	32	;-------------------------------------------------------------------------------------\r
	33	global ASM_PFX(RendezvousFunnelProc)\r
	34	ASM_PFX(RendezvousFunnelProc):\r
	35	RendezvousFunnelProcStart:\r
	36	; At this point CS = 0x(vv00) and ip= 0x0.\r
	37	BITS 16\r
	38	mov ebp, eax ; save BIST information\r
	39	\r
	40	mov ax, cs\r
	41	mov ds, ax\r
	42	mov es, ax\r
	43	mov ss, ax\r
	44	xor ax, ax\r
	45	mov fs, ax\r
	46	mov gs, ax\r
	47	\r
	48	mov si, BufferStartLocation\r
	49	mov ebx, [si]\r
	50	\r
d94e5f67 JF	51	mov si, DataSegmentLocation\r
	52	mov edx, [si]\r
	53	\r
f32bfe6d JW	54	;\r
	55	; Get start address of 32-bit code in low memory (<1MB)\r
	56	;\r
	57	mov edi, ModeTransitionMemoryLocation\r
	58	\r
d94e5f67 JF	59	mov si, GdtrLocation\r
	60	o32 lgdt [cs:si]\r
	61	\r
	62	mov si, IdtrLocation\r
	63	o32 lidt [cs:si]\r
	64	\r
f32bfe6d JW	65	;\r
	66	; Switch to protected mode\r
	67	;\r
8396e2dd JF	68	mov eax, cr0 ; Get control register 0\r
8396e2dd JF	69	or eax, 000000003h ; Set PE bit (bit #0) & MP\r
d94e5f67 JF	70	mov cr0, eax\r
d94e5f67 JF	71	\r
f32bfe6d JW	72	; Switch to 32-bit code in executable memory (>1MB)\r
	73	o32 jmp far [cs:di]\r
	74	\r
	75	;\r
	76	; Following code may be copied to memory with type of EfiBootServicesCode.\r
	77	; This is required at DXE phase if NX is enabled for EfiBootServicesCode of\r
	78	; memory.\r
	79	;\r
d94e5f67 JF	80	BITS 32\r
	81	Flat32Start: ; protected mode entry point\r
	82	mov ds, dx\r
	83	mov es, dx\r
	84	mov fs, dx\r
	85	mov gs, dx\r
	86	mov ss, dx\r
	87	\r
	88	mov esi, ebx\r
5c66d125	89	\r
0594ec41 ED	90	; Increment the number of APs executing here as early as possible\r
	91	; This is decremented in C code when AP is finished executing\r
	92	mov edi, esi\r
	93	add edi, NumApsExecutingLocation\r
	94	lock inc dword [edi]\r
	95	\r
5c66d125 JF	96	mov edi, esi\r
	97	add edi, EnableExecuteDisableLocation\r
	98	cmp byte [edi], 0\r
	99	jz SkipEnableExecuteDisable\r
	100	\r
	101	;\r
	102	; Enable IA32 PAE execute disable\r
	103	;\r
	104	\r
	105	mov ecx, 0xc0000080\r
	106	rdmsr\r
	107	bts eax, 11\r
	108	wrmsr\r
	109	\r
	110	mov edi, esi\r
	111	add edi, Cr3Location\r
	112	mov eax, dword [edi]\r
	113	mov cr3, eax\r
	114	\r
	115	mov eax, cr4\r
	116	bts eax, 5\r
	117	mov cr4, eax\r
	118	\r
	119	mov eax, cr0\r
	120	bts eax, 31\r
	121	mov cr0, eax\r
	122	\r
	123	SkipEnableExecuteDisable:\r
845c5be1 JF	124	mov edi, esi\r
	125	add edi, InitFlagLocation\r
	126	cmp dword [edi], 1 ; 1 == ApInitConfig\r
	127	jnz GetApicId\r
5c66d125	128	\r
845c5be1	129	; AP init\r
d94e5f67 JF	130	mov edi, esi\r
	131	add edi, LockLocation\r
	132	mov eax, NotVacantFlag\r
	133	\r
	134	TestLock:\r
	135	xchg [edi], eax\r
	136	cmp eax, NotVacantFlag\r
	137	jz TestLock\r
	138	\r
845c5be1	139	mov ecx, esi\r
37676b9f	140	add ecx, ApIndexLocation\r
845c5be1 JF	141	inc dword [ecx]\r
	142	mov ebx, [ecx]\r
	143	\r
	144	Releaselock:\r
	145	mov eax, VacantFlag\r
	146	xchg [edi], eax\r
d94e5f67	147	\r
d94e5f67 JF	148	mov edi, esi\r
	149	add edi, StackSizeLocation\r
	150	mov eax, [edi]\r
845c5be1 JF	151	mov ecx, ebx\r
	152	inc ecx\r
	153	mul ecx ; EAX = StackSize * (CpuNumber + 1)\r
d94e5f67 JF	154	mov edi, esi\r
	155	add edi, StackStartAddressLocation\r
	156	add eax, [edi]\r
	157	mov esp, eax\r
845c5be1 JF	158	jmp CProcedureInvoke\r
	159	\r
	160	GetApicId:\r
	161	mov eax, 0\r
	162	cpuid\r
	163	cmp eax, 0bh\r
1cbd8330 LE	164	jb NoX2Apic ; CPUID level below CPUID_EXTENDED_TOPOLOGY\r
	165	\r
	166	mov eax, 0bh\r
	167	xor ecx, ecx\r
	168	cpuid\r
	169	test ebx, 0ffffh\r
	170	jz NoX2Apic ; CPUID.0BH:EBX[15:0] is zero\r
	171	\r
	172	; Processor is x2APIC capable; 32-bit x2APIC ID is already in EDX\r
	173	jmp GetProcessorNumber\r
	174	\r
	175	NoX2Apic:\r
845c5be1 JF	176	; Processor is not x2APIC capable, so get 8-bit APIC ID\r
	177	mov eax, 1\r
	178	cpuid\r
	179	shr ebx, 24\r
	180	mov edx, ebx\r
845c5be1	181	\r
845c5be1 JF	182	GetProcessorNumber:\r
	183	;\r
	184	; Get processor number for this AP\r
	185	; Note that BSP may become an AP due to SwitchBsp()\r
	186	;\r
	187	xor ebx, ebx\r
	188	lea eax, [esi + CpuInfoLocation]\r
	189	mov edi, [eax]\r
d94e5f67	190	\r
845c5be1 JF	191	GetNextProcNumber:\r
	192	cmp [edi], edx ; APIC ID match?\r
	193	jz ProgramStack\r
dd3fa0cd	194	add edi, 20\r
845c5be1	195	inc ebx\r
7367cc6c	196	jmp GetNextProcNumber\r
d94e5f67	197	\r
845c5be1 JF	198	ProgramStack:\r
845c5be1 JF	199	mov esp, [edi + 12]\r
7367cc6c	200	\r
d94e5f67 JF	201	CProcedureInvoke:\r
	202	push ebp ; push BIST data at top of AP stack\r
	203	xor ebp, ebp ; clear ebp for call stack trace\r
	204	push ebp\r
	205	mov ebp, esp\r
	206	\r
	207	mov eax, ASM_PFX(InitializeFloatingPointUnits)\r
	208	call eax ; Call assembly function to initialize FPU per UEFI spec\r
	209	\r
37676b9f	210	push ebx ; Push ApIndex\r
d94e5f67 JF	211	mov eax, esi\r
	212	add eax, LockLocation\r
	213	push eax ; push address of exchange info data buffer\r
	214	\r
	215	mov edi, esi\r
	216	add edi, ApProcedureLocation\r
	217	mov eax, [edi]\r
	218	\r
8396e2dd	219	call eax ; Invoke C function\r
d94e5f67	220	\r
8396e2dd	221	jmp $ ; Never reach here\r
d94e5f67 JF	222	RendezvousFunnelProcEnd:\r
d94e5f67 JF	223	\r
76157021	224	;-------------------------------------------------------------------------------------\r
9f91cb01	225	; AsmRelocateApLoop (MwaitSupport, ApTargetCState, PmCodeSegment, TopOfApStack, CountTofinish);\r
76157021 JF	226	;-------------------------------------------------------------------------------------\r
	227	global ASM_PFX(AsmRelocateApLoop)\r
	228	ASM_PFX(AsmRelocateApLoop):\r
	229	AsmRelocateApLoopStart:\r
bf2786dc JF	230	mov eax, esp\r
	231	mov esp, [eax + 16] ; TopOfApStack\r
	232	push dword [eax] ; push return address for stack trace\r
	233	push ebp\r
	234	mov ebp, esp\r
	235	mov ebx, [eax + 8] ; ApTargetCState\r
	236	mov ecx, [eax + 4] ; MwaitSupport\r
9f91cb01 JF	237	mov eax, [eax + 20] ; CountTofinish\r
9f91cb01 JF	238	lock dec dword [eax] ; (*CountTofinish)--\r
bf2786dc	239	cmp cl, 1 ; Check mwait-monitor support\r
76157021 JF	240	jnz HltLoop\r
76157021 JF	241	MwaitLoop:\r
a7bbe9d2	242	cli\r
76157021 JF	243	mov eax, esp\r
	244	xor ecx, ecx\r
	245	xor edx, edx\r
	246	monitor\r
bf2786dc	247	mov eax, ebx ; Mwait Cx, Target C-State per eax[7:4]\r
76157021 JF	248	shl eax, 4\r
	249	mwait\r
	250	jmp MwaitLoop\r
	251	HltLoop:\r
	252	cli\r
	253	hlt\r
	254	jmp HltLoop\r
76157021 JF	255	AsmRelocateApLoopEnd:\r
76157021 JF	256	\r
d94e5f67 JF	257	;-------------------------------------------------------------------------------------\r
	258	; AsmGetAddressMap (&AddressMap);\r
	259	;-------------------------------------------------------------------------------------\r
	260	global ASM_PFX(AsmGetAddressMap)\r
	261	ASM_PFX(AsmGetAddressMap):\r
	262	pushad\r
	263	mov ebp,esp\r
	264	\r
	265	mov ebx, [ebp + 24h]\r
	266	mov dword [ebx], RendezvousFunnelProcStart\r
	267	mov dword [ebx + 4h], Flat32Start - RendezvousFunnelProcStart\r
	268	mov dword [ebx + 8h], RendezvousFunnelProcEnd - RendezvousFunnelProcStart\r
f7f85d83 JF	269	mov dword [ebx + 0Ch], AsmRelocateApLoopStart\r
f7f85d83 JF	270	mov dword [ebx + 10h], AsmRelocateApLoopEnd - AsmRelocateApLoopStart\r
f32bfe6d	271	mov dword [ebx + 14h], Flat32Start - RendezvousFunnelProcStart\r
d94e5f67 JF	272	\r
	273	popad\r
	274	ret\r
	275	\r
	276	;-------------------------------------------------------------------------------------\r
	277	;AsmExchangeRole procedure follows. This procedure executed by current BSP, that is\r
	278	;about to become an AP. It switches it'stack with the current AP.\r
	279	;AsmExchangeRole (IN CPU_EXCHANGE_INFO MyInfo, IN CPU_EXCHANGE_INFO OthersInfo);\r
	280	;-------------------------------------------------------------------------------------\r
	281	global ASM_PFX(AsmExchangeRole)\r
	282	ASM_PFX(AsmExchangeRole):\r
	283	; DO NOT call other functions in this function, since 2 CPU may use 1 stack\r
	284	; at the same time. If 1 CPU try to call a function, stack will be corrupted.\r
	285	pushad\r
	286	mov ebp,esp\r
	287	\r
	288	; esi contains MyInfo pointer\r
	289	mov esi, [ebp + 24h]\r
	290	\r
	291	; edi contains OthersInfo pointer\r
	292	mov edi, [ebp + 28h]\r
	293	\r
	294	;Store EFLAGS, GDTR and IDTR register to stack\r
	295	pushfd\r
	296	mov eax, cr4\r
	297	push eax ; push cr4 firstly\r
	298	mov eax, cr0\r
	299	push eax\r
	300	\r
	301	sgdt [esi + 8]\r
	302	sidt [esi + 14]\r
	303	\r
	304	; Store the its StackPointer\r
	305	mov [esi + 4],esp\r
	306	\r
	307	; update its switch state to STORED\r
	308	mov byte [esi], CPU_SWITCH_STATE_STORED\r
	309	\r
	310	WaitForOtherStored:\r
	311	; wait until the other CPU finish storing its state\r
	312	cmp byte [edi], CPU_SWITCH_STATE_STORED\r
	313	jz OtherStored\r
	314	pause\r
	315	jmp WaitForOtherStored\r
	316	\r
	317	OtherStored:\r
	318	; Since another CPU already stored its state, load them\r
	319	; load GDTR value\r
	320	lgdt [edi + 8]\r
	321	\r
	322	; load IDTR value\r
	323	lidt [edi + 14]\r
	324	\r
	325	; load its future StackPointer\r
	326	mov esp, [edi + 4]\r
	327	\r
	328	; update the other CPU's switch state to LOADED\r
	329	mov byte [edi], CPU_SWITCH_STATE_LOADED\r
	330	\r
	331	WaitForOtherLoaded:\r
	332	; wait until the other CPU finish loading new state,\r
	333	; otherwise the data in stack may corrupt\r
	334	cmp byte [esi], CPU_SWITCH_STATE_LOADED\r
	335	jz OtherLoaded\r
336	pause\r
337	jmp WaitForOtherLoaded\r
338	\r
339	OtherLoaded:\r
340	; since the other CPU already get the data it want, leave this procedure\r
341	pop eax\r
342	mov cr0, eax\r
343	pop eax\r
344	mov cr4, eax\r
345	popfd\r
346	\r
347	popad\r
348	ret\r