Unit PlatformARMv8

From Ultibo.org
Revision as of 04:48, 10 November 2016 by Ultibo (Talk | contribs)

Jump to: navigation, search

Return to Unit Reference


Description


The ARMv8 does not support the SWP/SWPB instructions for syncronisation (Lock/Mutex/Semaphore etc) unless enabled.

On ARMv8 Unaligned memory access is always enabled.

On ARMv8 the Extended Page Table format is always enabled.

For usage of barriers (DMB/DSB/ISB) after cache maintenance operations see: ARM.Reference_Manual_1.pdf - Appendix G Barrier Litmus Tests

Note: This unit currently only supports ARMv8 in Aarch32 mode, support for Aarch64 mode will be added in future.

Constants


To be documented

Type definitions


To be documented

Public variables


To be documented

Function declarations



Initialization functions

procedure ARMv8Init;
Description: To be documented
Note None documented


ARMv8 platform functions

procedure ARMv8CPUInit; assembler; nostackframe;
Description: To be documented
Note None documented


procedure ARMv8FPUInit; assembler; nostackframe;
Description: To be documented
Note None documented


procedure ARMv8MMUInit;
Description: To be documented
Note None documented


procedure ARMv8CacheInit; assembler; nostackframe;
Description: To be documented
Note None documented


procedure ARMv8TimerInit(Frequency:LongWord); assembler; nostackframe;
Description: To be documented
Note None documented


procedure ARMv8PageTableInit;
Description: Initialize the Hardware Page Tables before enabling the MMU
Note None documented


procedure ARMv8SystemCall(Number:LongWord; Param1,Param2,Param3:LongWord); assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8CPUGetMode:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8CPUGetState:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8CPUGetCurrent:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8CPUGetMainID:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8CPUGetMultiprocessorID:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8CPUGetModel:LongWord;
Description: To be documented
Note None documented


function ARMv8CPUGetRevision:LongWord;
Description: To be documented
Note None documented


function ARMv8CPUGetDescription:String;
Description: To be documented
Note None documented


function ARMv8FPUGetState:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L1CacheGetType:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L1DataCacheGetSize:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L1DataCacheGetLineSize:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L1InstructionCacheGetSize:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L1InstructionCacheGetLineSize:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L2CacheGetType:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L2CacheGetSize:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8L2CacheGetLineSize:LongWord; assembler; nostackframe;
Description: To be documented
Note None documented


procedure ARMv8Halt; assembler; nostackframe; public name'_haltproc';
Description: The purpose of the Wait For Interrupt operation is to put the processor in to a low power state
Note See Standby mode on page A8-810 of the ARMv7 Architecture Reference Manual


procedure ARMv8Pause; assembler; nostackframe;
Description: The purpose of the Wait For Interrupt operation is to put the processor in to a low power state
Note See Standby mode on page A8-810 of the ARMv7 Architecture Reference Manual


procedure ARMv8SendEvent; assembler; nostackframe;
Description: To be documented
Note See Page A8-316 of the ARMv7 Architecture Reference Manual


procedure ARMv8WaitForEvent; assembler; nostackframe;
Description: To be documented
Note See Page A8-808 of the ARMv7 Architecture Reference Manual


procedure ARMv8WaitForInterrupt; assembler; nostackframe;
Description: The purpose of the Wait For Interrupt operation is to put the processor in to a low power state
Note See Standby mode on page A8-810 of the ARMv7 Architecture Reference Manual


procedure ARMv8DataMemoryBarrier; assembler; nostackframe;
Description: Perform a data memory barrier operation using the c7 (Cache Operations) register of system control coprocessor CP15
Note See page A8-90 of the ARMv7 Architecture Reference Manual

Note that this is also available in the FPC RTL as ReadBarrier/WriteBarrier See: \source\rtl\arm\arm.inc
Implementation is exactly the same for either


procedure ARMv8DataSynchronizationBarrier; assembler; nostackframe;
Description: Perform a data synchronization barrier operation
Note See page A8-92 of the ARMv7 Architecture Reference Manual


procedure ARMv8InstructionMemoryBarrier; assembler; nostackframe;
Description: Perform a instruction synchronization barrier operation
Note See page A8-102 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateTLB; assembler; nostackframe;
Description: Perform an invalidate entire TLB operation using the c8 (TLB Operations) register of system control coprocessor CP15
Note See page B3-138 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateDataTLB; assembler; nostackframe;
Description: Perform an invalidate data TLB (Unlocked/Data) operation using the c8 (TLB Operations) register of system control coprocessor CP15
Note See page B3-138 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateInstructionTLB; assembler; nostackframe;
Description: Perform an invalidate instruction TLB (Unlocked/Instruction) operation using the c8 (TLB Operations) register of system control coprocessor CP15
Note See page B3-138 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateCache; assembler; nostackframe;
Description: Perform an invalidate both caches operation using the c7 (Cache Operations) register of system control coprocessor CP15
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8CleanDataCache; assembler; nostackframe;
Description: Perform a clean entire data cache operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateDataCache; assembler; nostackframe;
Description: Perform an invalidate entire data cache operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateL1DataCache; assembler; nostackframe;
Description: Perform an invalidate entire L1 data cache operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8CleanAndInvalidateDataCache; assembler; nostackframe;
Description: Perform a clean and invalidate entire data cache operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateInstructionCache; assembler; nostackframe;
Description: Perform an invalidate entire instruction cache operation using the c7 (Cache Operations) register of system control coprocessor CP15
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8CleanDataCacheRange(Address,Size:LongWord); assembler; nostackframe;
Description: Perform a clean data cache by MVA to PoC operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateDataCacheRange(Address,Size:LongWord); assembler; nostackframe;
Description: Perform an invalidate data cache by MVA to PoC operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8CleanAndInvalidateDataCacheRange(Address,Size:LongWord); assembler; nostackframe;
Description: Perform a clean and invalidate data cache by MVA to PoC operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateInstructionCacheRange(Address,Size:LongWord); assembler; nostackframe; 
Description: Perform an invalidate instruction caches by MVA to PoU operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8CleanDataCache?etWay(?etWay:LongWord); assembler; nostackframe;
Description: Perform a clean data cache line by ?et/way operation
?etWay  ?et/Way/Level will be passed in r0
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8InvalidateDataCache?etWay(?etWay:LongWord); assembler; nostackframe;
Description: Perform an invalidate data cache line by ?et/way operation
?etWay  ?et/Way/Level will be passed in r0
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8CleanAndInvalidateDataCache?etWay(?etWay:LongWord); assembler; nostackframe;
Description: Perform a clean and invalidate data cache line by ?et/way operation
?etWay  ?et/Way/Level will be passed in r0
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8FlushPrefetchBuffer; assembler; nostackframe;
Description: Perform an Instruction Synchronization Barrier operation
Note See page A8-102 of the ARMv7 Architecture Reference Manual


procedure ARMv8FlushBranchTargetCache; assembler; nostackframe;
Description: Perform a Flush Entire Branch Target Cache operation
Note See page B3-127 of the ARMv7 Architecture Reference Manual


procedure ARMv8ContextSwitch(OldStack,NewStack:Pointer; NewThread:TThreadHandle);  assembler; nostackframe;
Description: Perform a context switch from one thread to another as a result of a thread yielding, sleeping or waiting
OldStack The address to save the stack pointer to for the current thread (Passed in r0)
NewStack The address to restore the stack pointer from for the new thread (Passed in r1)
NewThread The handle of the new thread to switch to (Passed in r2)
Note At the point of the actual context switch (str sp / ldr sp) the thread stacks will look like this:


(See: ARMv8Thread?etupStack for additional information)

(Base "Highest Address" of Stack)
.
.
.
.
cpsr <- The current program status register value to load on return from the context switch
lr/pc <- The address to return to from the context switch
lr <- The lr value prior to the context switch
r12 <-
r11 <-
r10 <-
r9 <-
r8 <-
r7 <-
r6 <- The value of these registers prior to the context switch
r5 <-
r4 <-
r3 <-
r2 <-
r1 <-
r0 <-

d15 <-
d14 <-
d13 <-
d12 <-
d11 <-
d10 <-
d9 <-
d8 <- The value of these floating point registers prior to the context switch
d7 <-
d6 <-
d5 <-
d4 <-
d3 <-
d2 <-
d1 <-
d0 <-

fpscr <- The floating point FPSCR register
fpexc <- The floating point FPEXC register (Current StackPointer points to here)
.
.
.
.
(Top "Lowest Address" of Stack)

This form of context switch uses r12 to save the cpsr value (and RFE to restore it). Because this context switch is called from a routine which will have saved the value of r12 (which is caller save in the ARM ABI) then we do not need to save the original value of r12.

The context switch will be performed from SYS mode to SYS mode, the cpsr value will include the control bits (Mode and IRQ/FIQ state) but not the flags values. Again the ARM ABI does not require that the flags be saved by the callee and so the caller would have accounted for any needed flags before calling. If the thread to be resumed was interrupted by an IRQ or FIQ then the cpsr will also contain the flags etc as they were at the point of interrupt. We do not need to account for the state bits in the cpsr since all operations are performed in ARM mode at present.

The main requirement of this routine is to ensure that the context record on the stack matches exactly that which is created on an interrupt and also that created by Thread?etupStack for a new thread. If this is correct then the next context switch for any given thread can be either by a call to reschedule or by an interrupt. Equally a new thread can be first run from a context switch that resulted from either a call to reschedule or an interrupt.

Note that this routine could use:

pop (lr)
pop (r12)
msr cpsr_c, r12
mov pc, lr

To return but that would mess up the value of r12, lr and the cpsr flags etc if the thread being resumed was interrupted by an IRQ,FIQ or SWI. The use of RFE here allows for exactly the same behaviour no matter which way the context record is saved and restored.


procedure ARMv8ContextSwitchIRQ(OldStack,NewStack:Pointer; NewThread:TThreadHandle);  assembler; nostackframe;
Description: Perform a context switch from one thread to another as a result of an interrupt request (IRQ)
OldStack The address to save the stack pointer to for the current thread (Passed in r0)
NewStack The address to restore the stack pointer from for the new thread (Passed in r1)
NewThread The handle of the new thread to switch to (Passed in r2)
Note At the point of the actual context switch (str sp / ldr sp) the thread stacks will look like this:


(See: ARMv8Thread?etupStack for additional information)

(Base "Highest Address" of Stack)
.
.
.
.
cpsr <- The current program status register value to load on return from the context switch
lr/pc <- The address to return to from the context switch
lr <- The lr value prior to the context switch
r12 <-
r11 <-
r10 <-
r9 <-
r8 <-
r7 <-
r6 <- The value of these registers prior to the context switch
r5 <-
r4 <-
r3 <-
r2 <-
r1 <-
r0 <-

d15 <-
d14 <-
d13 <-
d12 <-
d11 <-
d10 <-
d9 <-
d8 <- The value of these floating point registers prior to the context switch
d7 <-
d6 <-
d5 <-
d4 <-
d3 <-
d2 <-
d1 <-
d0 <-

fpscr <- The floating point FPSCR register
fpexc <- The floating point FPEXC register (Current StackPointer points to here)
.
.
.
.
(Top "Lowest Address" of Stack)

This form of context switch relies on the IRQ handler to save the necessary registers including the lr, cpsr and other general registers from the point at which the thread was interrupted. The thread to be resumed may have been saved by a previous IRQ or by a call to the standard context switch from SchedulerReschule or it may be a new thread to be run for the first time. All of these result in the same context record on the stack and therefore can be resumed the same way.

The context switch will be performed by switching to SYS mode, exchanging the stack pointers and then returning to IRQ mode.


procedure ARMv8ContextSwitchFIQ(OldStack,NewStack:Pointer; NewThread:TThreadHandle); assembler; nostackframe;
Description: Perform a context switch from one thread to another as a result of a fast interrupt request (FIQ)
OldStack The address to save the stack pointer to for the current thread (Passed in r0)
NewStack The address to restore the stack pointer from for the new thread (Passed in r1)
NewThread The handle of the new thread to switch to (Passed in r2)
Note At the point of the actual context switch (str sp / ldr sp) the thread stacks will look like this:


(See: ARMv8Thread?etupStack for additional information)

(Base "Highest Address" of Stack)
.
.
.
.
cpsr <- The current program status register value to load on return from the context switch
lr/pc <- The address to return to from the context switch
lr <- The lr value prior to the context switch
r12 <-
r11 <-
r10 <-
r9 <-
r8 <-
r7 <-
r6 <- The value of these registers prior to the context switch
r5 <-
r4 <-
r3 <-
r2 <-
r1 <-
r0 <-

d15 <-
d14 <-
d13 <-
d12 <-
d11 <-
d10 <-
d9 <-
d8 <- The value of these floating point registers prior to the context switch
d7 <-
d6 <-
d5 <-
d4 <-
d3 <-
d2 <-
d1 <-
d0 <-

fpscr <- The floating point FPSCR register
fpexc <- The floating point FPEXC register (Current StackPointer points to here)
.
.
.
.
(Top "Lowest Address" of Stack)

This form of context switch relies on the FIQ handler to save the necessary registers including the lr, cpsr and other general registers from the point at which the thread was interrupted. The thread to be resumed may have been saved by a previous FIQ or by a call to the standard context switch from SchedulerReschule or it may be a new thread to be run for the first time. All of these result in the same context record on the stack and therefore can be resumed the same way.

The context switch will be performed by switching to SYS mode, exchanging the stack pointers and then returning to FIQ mode.


procedure ARMv8ContextSwitchSWI(OldStack,NewStack:Pointer; NewThread:TThreadHandle); assembler; nostackframe;
Description: Perform a context switch from one thread to another as a result of a software interrupt (SWI)
OldStack The address to save the stack pointer to for the current thread (Passed in r0)
NewStack The address to restore the stack pointer from for the new thread (Passed in r1)
NewThread The handle of the new thread to switch to (Passed in r2)
Note At the point of the actual context switch (str sp / ldr sp) the thread stacks will look like this:


(See: ARMv8Thread?etupStack for additional information)

(Base "Highest Address" of Stack)
.
.
.
.
cpsr <- The current program status register value to load on return from the context switch
lr/pc <- The address to return to from the context switch
lr <- The lr value prior to the context switch
r12 <-
r11 <-
r10 <-
r9 <-
r8 <-
r7 <-
r6 <- The value of these registers prior to the context switch
r5 <-
r4 <-
r3 <-
r2 <-
r1 <-
r0 <-

d15 <-
d14 <-
d13 <-
d12 <-
d11 <-
d10 <-
d9 <-
d8 <- The value of these floating point registers prior to the context switch
d7 <-
d6 <-
d5 <-
d4 <-
d3 <-
d2 <-
d1 <-
d0 <-

fpscr <- The floating point FPSCR register
fpexc <- The floating point FPEXC register (Current StackPointer points to here)
.
.
.
.
(Top "Lowest Address" of Stack)

This form of context switch relies on the SWI handler to save the necessary registers including the lr, cpsr and other general registers from the point at which the thread was interrupted. The thread to be resumed may have been saved by a previous SWI or by a call to the standard context switch from SchedulerReschule or it may be a new thread to be run for the first time. All of these result in the same context record on the stack and therefore can be resumed the same way.

The context switch will be performed by switching to SYS mode, exchanging the stack pointers and then returning to SWI (SVC) mode.


function ARMv8InterlockedOr(var Target:LongInt; Value:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic OR operation using LDREX/STREX
Note See page ???


function ARMv8InterlockedXor(var Target:LongInt; Value:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic XOR operation using LDREX/STREX
Note See page ???


function ARMv8InterlockedAnd(var Target:LongInt; Value:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic AND operation using LDREX/STREX
Note See page ???


function ARMv8InterlockedDecrement(var Target:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic decrement operation using LDREX/STREX
Note See page ???


function ARMv8InterlockedIncrement(var Target:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic increment operation using LDREX/STREX
Note See page ???


function ARMv8InterlockedExchange(var Target:LongInt; Source:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic exchange operation using LDREX/STREX
Note See page ???


function ARMv8InterlockedAddExchange(var Target:LongInt; Source:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic add and exchange operation using LDREX/STREX
Note See page ???


function ARMv8InterlockedCompareExchange(var Target:LongInt; Source,Compare:LongInt):LongInt; assembler; nostackframe;
Description: Perform an atomic compare and exchange operation using LDREX/STREX
Note See page ???


function ARMv8PageTableGetEntry(Address:PtrUInt):TPageTableEntry;
Description: Get and Decode the entry in the Page Table that corresponds to the supplied virtual address
Note None documented


function ARMv8PageTable?etEntry(const Entry:TPageTableEntry):LongWord;
Description: Encode and ?et an entry in the Page Table that corresponds to the supplied virtual address
Note None documented


function ARMv8VectorTableGetEntry(Number:LongWord):PtrUInt;
Description: Return the address of the specified vector table entry number
Note None documented


function ARMv8VectorTable?etEntry(Number:LongWord; Address:PtrUInt):LongWord;
Description: ?et the supplied address as the value of the specified vector table entry number
Note None documented


function ARMv8FirstBit?et(Value:LongWord):LongWord; assembler; nostackframe;
Description: To be documented
Note ARM arm states that CLZ is supported for ARMv5 and above


function ARMv8CountLeadingZeros(Value:LongWord):LongWord; assembler; nostackframe;
Description: Equivalent of the GCC Builtin function __builtin_clz
Note ARM arm states that CLZ is supported for ARMv5 and above


ARMv8 thread functions

procedure ARMv8PrimaryInit; assembler; nostackframe;
Description: To be documented
Note None documented


function ARMv8SpinLock(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Lock an existing Spin entry
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinUnlock(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Unlock an existing Spin entry
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinLockIRQ(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Lock an existing Spin entry, disable IRQ and save the previous IRQ state
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinUnlockIRQ(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Unlock an existing Spin entry and restore the previous IRQ state
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinLockFIQ(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Lock an existing Spin entry, disable FIQ and save the previous FIQ state
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinUnlockFIQ(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Unlock an existing Spin entry and restore the previous FIQ state
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinLockIRQFIQ(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Lock an existing Spin entry, disable IRQ and FIQ and save the previous IRQ/FIQ state
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinUnlockIRQFIQ(Spin:PSpinEntry):LongWord; assembler; nostackframe;
Description: Unlock an existing Spin entry and restore the previous IRQ/FIQ state
Spin Pointer to the Spin entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8SpinCheckIRQ(Spin:PSpinEntry):Boolean;
Description: To be documented
Return True if the mask would enable IRQ on restore, False if it would not


function ARMv8SpinCheckFIQ(Spin:PSpinEntry):Boolean;
Description: To be documented
Return True if the mask would enable FIQ on restore, False if it would not


function ARMv8SpinExchangeIRQ(Spin1,Spin2:PSpinEntry):LongWord;
Description: To be documented
Note None documented


function ARMv8SpinExchangeFIQ(Spin1,Spin2:PSpinEntry):LongWord;
Description: To be documented
Note None documented


function ARMv8MutexLock(Mutex:PMutexEntry):LongWord; assembler; nostackframe;
Description: Lock an existing Mutex entry
Mutex Pointer to the Mutex entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8MutexUnlock(Mutex:PMutexEntry):LongWord; assembler; nostackframe;
Description: Unlock an existing Mutex entry
Mutex Pointer to the Mutex entry to lock (Passed in R0)
Return ERROR_SUCCESS if completed or another error code on failure (Returned in R0)


function ARMv8MutexTryLock(Mutex:PMutexEntry):LongWord; assembler; nostackframe;
Description: Try to lock an existing Mutex entry
Mutex Pointer to the Mutex entry to try to lock (Passed in R0)
Return ERROR_SUCCESS if completed, ERROR_LOCKED if already locked or another error code on failure (Returned in R0)


function ARMv8ThreadGetCurrent:TThreadHandle; assembler; nostackframe;
Description: Get the current thread id from the c13 (Thread and process ID) register of system control coprocessor CP15
Note See page ???


function ARMv8ThreadSetCurrent(Thread:TThreadHandle):LongWord; assembler; nostackframe;
Description: Set the current thread id in the c13 (Thread and process ID) register of system control coprocessor CP15
Note See page ???


function ARMv8ThreadSetupStack(StackBase:Pointer; StartProc:TThreadStart; ReturnProc:TThreadEnd; Parameter:Pointer):Pointer;
Description: Set up the context record and arguments on the stack for a new thread
StackBase Pointer to the base (highest address) of the allocated stack (as returned by ThreadAllocateStack
StartProc The procedure the thread will start executing when resumed
ReturnProc The procedure the thread will return to on exit
Return Pointer to the starting address of the stack, which will be the Stack Pointer on the first context switch
Note At the point of a context switch the thread stack will look like this:


(Base "Highest Address" of Stack)
.
.
.
.
cpsr <- The current program status register value to load on return from the context switch
(On Interrupt: Includes the flags and control bits for the interrupted thread)
(On Yield: Includes the control bits only for the yielded thread)
(On Create: Includes the control bits only for the new thread)
lr/pc <- The address to return to from the context switch
(On Interrupt: The location the thread was at before interrupt)
(On Yield: The location to return to in SchedulerReschedule)
(On Create: The location of StartProc for the new thread)

lr <- The lr value prior to the context switch
(On Interrupt: The value of lr before the thread was interrupted)
(On Yield: The location to return to in SchedulerReschedule)
(On Create: The location of ReturnProc for the new thread)

r12 <-
r11 <-
r10 <-
r9 <-
r8 <-
r7 <-
r6 <- The value of these registers prior to the context switch
r5 <- (On Interrupt: The values before the thread was interrupted)
r4 <- (On Yield: The values on return to SchedulerReschedule)
r3 <- (On Create: The values on entry to StartProc as set by ThreadSetupStack)
r2 <-
r1 <-
r0 <-

d15 <-
d14 <-
d13 <-
d12 <-
d11 <-
d10 <-
d9 <-
d8 <- The value of these floating point registers prior to the context switch
d7 <- (On Interrupt: The values before the thread was interrupted)
d6 <- (On Yield: The values on return to SchedulerReschedule)
d5 <- (On Create: The values on entry to StartProc as set by ThreadSetupStack)
d4 <-
d3 <-
d2 <-
d1 <-
d0 <-

fpscr <- The floating point FPSCR register
fpexc <- The floating point FPEXC register (Current StackPointer points to here)
.
.
.
.
(Top "Lowest Address" of Stack)

On exit from a standard context switch as performed by SchedulerReschedule the first value (Highest Address) of lr is used by the RFE (Return From Exception) instruction to load the pc which also loads the saved cpsr value.

On exit from an IRQ or FIQ context switch as performed by SchedulerSwitch the first value (Highest Address) of lr is used by the interrupt handler to return from the interrupt.

A standard context switch uses r12 to save the cpsr value (and RFE to restore it). Because the standard context switch is called from a routine which will have saved the value of r12 (which is caller save in the ARM ABI) then we do not need to save the original value of r12.

An IRQ or FIQ context switch uses the SRS (Store Return State) and RFE (Return From Exception) instructions to save and restore the cpsr value from the spsr value of either IRQ or FIQ mode.


Return to Unit Reference