kernel/arch/thread: implement for riscv64

Change-Id: I3effa598626b32c29606299cd0edee390d430baf
Reviewed-on: https://review.haiku-os.org/c/haiku/+/4066
Reviewed-by: Alex von Gluck IV <kallisti5@unixzen.com>
This commit is contained in:
X512 2021-06-06 22:15:18 +09:00 committed by Alex von Gluck IV
parent 7ef006297e
commit 4a32f48e70
5 changed files with 472 additions and 38 deletions

View File

@ -16,7 +16,7 @@ struct vregs {
ulong x[31];
ulong pc;
double f[32];
char fcsr;
ulong fcsr;
};
#endif /* defined(__RISCV64__) */

View File

@ -11,30 +11,24 @@
#include <arch/cpu.h>
#warning IMPLEMENT arch_thread.h
#ifdef __cplusplus
extern "C" {
#endif
void riscv64_push_iframe(struct iframe_stack* stack, struct iframe* frame);
void riscv64_pop_iframe(struct iframe_stack* stack);
struct iframe* riscv64_get_user_iframe(void);
static inline Thread*
arch_thread_get_current_thread(void)
{
#warning IMPLEMENT arch_thread_get_current_thread
Thread* t = NULL;
return t;
Thread* t;
asm volatile("mv %0, tp" : "=r" (t));
return t;
}
static inline void
arch_thread_set_current_thread(Thread* t)
{
#warning IMPLEMENT arch_thread_set_current_thread
asm volatile("mv tp, %0" : : "r" (t));
}

View File

@ -7,20 +7,65 @@
#include <kernel.h>
#define IFRAME_TRACE_DEPTH 4
struct iframe_stack {
struct iframe *frames[IFRAME_TRACE_DEPTH];
int32 index;
namespace BKernel {
struct Thread;
}
struct iframe {
uint64 ra;
uint64 t6;
uint64 sp;
uint64 gp;
uint64 tp;
uint64 t0;
uint64 t1;
uint64 t2;
uint64 t5;
uint64 s1;
uint64 a0;
uint64 a1;
uint64 a2;
uint64 a3;
uint64 a4;
uint64 a5;
uint64 a6;
uint64 a7;
uint64 s2;
uint64 s3;
uint64 s4;
uint64 s5;
uint64 s6;
uint64 s7;
uint64 s8;
uint64 s9;
uint64 s10;
uint64 s11;
uint64 t3;
uint64 t4;
uint64 fp;
uint64 epc;
};
// architecture specific thread info
struct arch_thread {
void *sp; // stack pointer
void *interrupt_stack;
struct arch_context {
uint64 ra; // 0
uint64 s[12]; // 12
uint64 sp; // 13
uint64 satp; // 14
};
// used to track interrupts on this thread
struct iframe_stack iframes;
struct fpu_context {
double f[32];
uint64 fcsr;
};
struct arch_thread {
BKernel::Thread* thread;
arch_context context;
fpu_context fpuContext;
iframe* userFrame;
};
struct arch_team {
@ -31,10 +76,25 @@ struct arch_team {
};
struct arch_fork_arg {
// gcc treats empty structures as zero-length in C, but as if they contain
// a char in C++. So we have to put a dummy in to be able to use the struct
// from both in a consistent way.
char dummy;
iframe frame;
};
#ifdef __cplusplus
extern "C" {
#endif
int arch_setjmp(arch_context* ctx);
void arch_longjmp(arch_context* ctx, int val);
void save_fpu(fpu_context* ctx);
void restore_fpu(fpu_context* ctx);
void arch_thread_entry();
void arch_enter_userspace(void *arg1, void *arg2, addr_t sp);
void arch_longjmp_iframe(iframe* frame);
#ifdef __cplusplus
}
#endif
#endif /* KERNEL_ARCH_RISCV64_THREAD_TYPES_H */

View File

@ -8,3 +8,155 @@
.text
FUNCTION(MSyscall):
ecall
ret
FUNCTION_END(MSyscall)
FUNCTION(arch_setjmp):
sd ra, 0*8(a0)
sd s0, 1*8(a0)
sd s1, 2*8(a0)
sd s2, 3*8(a0)
sd s3, 4*8(a0)
sd s4, 5*8(a0)
sd s5, 6*8(a0)
sd s6, 7*8(a0)
sd s7, 8*8(a0)
sd s8, 9*8(a0)
sd s9, 10*8(a0)
sd s10, 11*8(a0)
sd s11, 12*8(a0)
sd sp, 13*8(a0)
csrr t0, satp
sd t0, 14*8(a0)
li a0, 0
ret
FUNCTION_END(arch_setjmp)
FUNCTION(arch_longjmp):
ld ra, 0*8(a0)
ld s0, 1*8(a0)
ld s1, 2*8(a0)
ld s2, 3*8(a0)
ld s3, 4*8(a0)
ld s4, 5*8(a0)
ld s5, 6*8(a0)
ld s6, 7*8(a0)
ld s7, 8*8(a0)
ld s8, 9*8(a0)
ld s9, 10*8(a0)
ld s10, 11*8(a0)
ld s11, 12*8(a0)
ld sp, 13*8(a0)
ld t0, 14*8(a0)
csrw satp, t0
sfence.vma
seqz a0, a1
add a0, a0, a1 # a0 = (a1 == 0) ? 1 : a1
ret
FUNCTION_END(arch_longjmp)
FUNCTION(save_fpu):
fsd f0, 0*8(a0)
fsd f1, 1*8(a0)
fsd f2, 2*8(a0)
fsd f3, 3*8(a0)
fsd f4, 4*8(a0)
fsd f5, 5*8(a0)
fsd f6, 6*8(a0)
fsd f7, 7*8(a0)
fsd f8, 8*8(a0)
fsd f9, 9*8(a0)
fsd f10, 10*8(a0)
fsd f11, 11*8(a0)
fsd f12, 12*8(a0)
fsd f13, 13*8(a0)
fsd f14, 14*8(a0)
fsd f15, 15*8(a0)
fsd f16, 16*8(a0)
fsd f17, 17*8(a0)
fsd f18, 18*8(a0)
fsd f19, 19*8(a0)
fsd f20, 20*8(a0)
fsd f21, 21*8(a0)
fsd f22, 22*8(a0)
fsd f23, 23*8(a0)
fsd f24, 24*8(a0)
fsd f25, 25*8(a0)
fsd f26, 26*8(a0)
fsd f27, 27*8(a0)
fsd f28, 28*8(a0)
fsd f29, 29*8(a0)
fsd f30, 30*8(a0)
fsd f31, 31*8(a0)
frcsr t0
sd t0, 32*8(a0)
ret
FUNCTION_END(save_fpu)
FUNCTION(restore_fpu):
fld f0, 0*8(a0)
fld f1, 1*8(a0)
fld f2, 2*8(a0)
fld f3, 3*8(a0)
fld f4, 4*8(a0)
fld f5, 5*8(a0)
fld f6, 6*8(a0)
fld f7, 7*8(a0)
fld f8, 8*8(a0)
fld f9, 9*8(a0)
fld f10, 10*8(a0)
fld f11, 11*8(a0)
fld f12, 12*8(a0)
fld f13, 13*8(a0)
fld f14, 14*8(a0)
fld f15, 15*8(a0)
fld f16, 16*8(a0)
fld f17, 17*8(a0)
fld f18, 18*8(a0)
fld f19, 19*8(a0)
fld f20, 20*8(a0)
fld f21, 21*8(a0)
fld f22, 22*8(a0)
fld f23, 23*8(a0)
fld f24, 24*8(a0)
fld f25, 25*8(a0)
fld f26, 26*8(a0)
fld f27, 27*8(a0)
fld f28, 28*8(a0)
fld f29, 29*8(a0)
fld f30, 30*8(a0)
fld f31, 31*8(a0)
ld t0, 32*8(a0)
fscsr t0
ret
FUNCTION_END(restore_fpu)
FUNCTION(arch_thread_entry):
mv a0, s2
jalr s1
FUNCTION_END(arch_thread_entry)
FUNCTION(arch_enter_userspace):
mv sp, a2
sret
FUNCTION_END(arch_enter_userspace)
FUNCTION(arch_longjmp_iframe):
mv sp, a0
call SVecURet
FUNCTION_END(arch_longjmp_iframe)

View File

@ -6,11 +6,27 @@
#include <string.h>
#include <arch_cpu.h>
#include <arch_debug.h>
#include <arch/thread.h>
#include <boot/stage2.h>
#include <commpage.h>
#include <kernel.h>
#include <thread.h>
#include <team.h>
#include <tls.h>
#include <vm/vm_types.h>
#include <vm/VMAddressSpace.h>
#include "RISCV64VMTranslationMap.h"
extern "C" void SVecU();
extern "C" void RestoreUserRegs()
{
SetSscratch((addr_t)&thread_get_current_thread()->arch_info);
SetTp(thread_get_current_thread()->user_local_storage);
}
status_t
@ -34,33 +50,70 @@ arch_team_init_team_struct(Team *team, bool kernel)
status_t
arch_thread_init_thread_struct(Thread *thread)
{
// set up an initial state (stack & fpu)
//memcpy(&thread->arch_info, &sInitialState, sizeof(struct arch_thread));
thread->arch_info.thread = thread;
return B_OK;
}
static inline VMAddressSpace*
GetThreadAddressSpace(Thread* thread)
{
/*
if (thread->team == team_get_kernel_team())
return VMAddressSpace::Kernel();
*/
return thread->team->address_space;
}
void
arch_thread_init_kthread_stack(Thread* thread, void* _stack, void* _stackTop,
void (*function)(void*), const void* data)
{
#warning RISCV64: Implement thread init kthread
panic("arch_thread_init_kthread_stack(): Implement me!");
// dprintf("arch_thread_init_kthread_stack(%p(%s))\n", thread, thread->name);
memset(&thread->arch_info.context, 0, sizeof(arch_context));
thread->arch_info.context.sp = (addr_t)_stackTop;
thread->arch_info.context.s[0] = 0; // fp
thread->arch_info.context.s[1] = (addr_t)function;
thread->arch_info.context.s[2] = (addr_t)data;
thread->arch_info.context.ra = (addr_t)arch_thread_entry;
VMTranslationMap* map = GetThreadAddressSpace(thread)->TranslationMap();
thread->arch_info.context.satp = ((RISCV64VMTranslationMap*)map)->Satp();
}
status_t
arch_thread_init_tls(Thread *thread)
{
// TODO: Implement!
return B_OK;
addr_t tls[TLS_FIRST_FREE_SLOT];
thread->user_local_storage = thread->user_stack_base
+ thread->user_stack_size;
// initialize default TLS fields
memset(tls, 0, sizeof(tls));
tls[TLS_BASE_ADDRESS_SLOT] = thread->user_local_storage;
tls[TLS_THREAD_ID_SLOT] = thread->id;
tls[TLS_USER_THREAD_SLOT] = (addr_t)thread->user_thread;
return user_memcpy((void*)thread->user_local_storage, tls, sizeof(tls));
}
void
arch_thread_context_switch(Thread *from, Thread *to)
{
/*
dprintf("arch_thread_context_switch(%p(%s), %p(%s))\n", from, from->name,
to, to->name);
*/
// TODO: save/restore FPU only if needed
save_fpu(&from->arch_info.fpuContext);
if (arch_setjmp(&from->arch_info.context) == 0) {
arch_longjmp(&to->arch_info.context, 1);
} else {
restore_fpu(&from->arch_info.fpuContext);
}
}
@ -71,9 +124,25 @@ arch_thread_dump_info(void *info)
status_t
arch_thread_enter_userspace(Thread *thread, addr_t entry, void *arg1, void *arg2)
arch_thread_enter_userspace(Thread *thread, addr_t entry, void *arg1,
void *arg2)
{
panic("arch_thread_enter_uspace(): not yet implemented\n");
// dprintf("arch_thread_enter_uspace()\n");
disable_interrupts();
if (arch_setjmp(&thread->arch_info.context) == 0) {
SstatusReg status(Sstatus());
status.pie = (1 << modeS); // enable interrupts when enter userspace
status.spp = modeU;
SetSstatus(status.val);
SetStvec((addr_t)SVecU);
SetSepc(entry);
RestoreUserRegs();
arch_enter_userspace(arg1, arg2,
thread->user_stack_base + thread->user_stack_size);
} else {
panic("return from userspace");
}
return B_ERROR;
}
@ -81,7 +150,35 @@ arch_thread_enter_userspace(Thread *thread, addr_t entry, void *arg1, void *arg2
bool
arch_on_signal_stack(Thread *thread)
{
return false;
struct iframe* frame = thread->arch_info.userFrame;
if (frame == NULL) {
panic("arch_on_signal_stack(): No user iframe!");
return false;
}
return frame->sp >= thread->signal_stack_base
&& frame->sp < thread->signal_stack_base
+ thread->signal_stack_size;
}
static uint8*
get_signal_stack(Thread* thread, struct iframe* frame,
struct sigaction* action, size_t spaceNeeded)
{
// use the alternate signal stack if we should and can
if (
thread->signal_stack_enabled &&
(action->sa_flags & SA_ONSTACK) != 0 && (
frame->sp < thread->signal_stack_base ||
frame->sp >= thread->signal_stack_base + thread->signal_stack_size
)
) {
addr_t stackTop = thread->signal_stack_base
+ thread->signal_stack_size;
return (uint8*)ROUNDDOWN(stackTop - spaceNeeded, 16);
}
return (uint8*)ROUNDDOWN(frame->sp - spaceNeeded, 16);
}
@ -89,20 +186,132 @@ status_t
arch_setup_signal_frame(Thread *thread, struct sigaction *sa,
struct signal_frame_data *signalFrameData)
{
return B_ERROR;
// dprintf("arch_setup_signal_frame()\n");
iframe* frame = thread->arch_info.userFrame;
// fill signal context
signalFrameData->context.uc_mcontext.x[ 0] = frame->ra;
signalFrameData->context.uc_mcontext.x[ 1] = frame->sp;
signalFrameData->context.uc_mcontext.x[ 2] = frame->gp;
signalFrameData->context.uc_mcontext.x[ 3] = frame->tp;
signalFrameData->context.uc_mcontext.x[ 4] = frame->t0;
signalFrameData->context.uc_mcontext.x[ 5] = frame->t1;
signalFrameData->context.uc_mcontext.x[ 6] = frame->t2;
signalFrameData->context.uc_mcontext.x[ 7] = frame->fp;
signalFrameData->context.uc_mcontext.x[ 8] = frame->s1;
signalFrameData->context.uc_mcontext.x[ 9] = frame->a0;
signalFrameData->context.uc_mcontext.x[10] = frame->a1;
signalFrameData->context.uc_mcontext.x[11] = frame->a2;
signalFrameData->context.uc_mcontext.x[12] = frame->a3;
signalFrameData->context.uc_mcontext.x[13] = frame->a4;
signalFrameData->context.uc_mcontext.x[14] = frame->a5;
signalFrameData->context.uc_mcontext.x[15] = frame->a6;
signalFrameData->context.uc_mcontext.x[16] = frame->a7;
signalFrameData->context.uc_mcontext.x[17] = frame->s2;
signalFrameData->context.uc_mcontext.x[18] = frame->s3;
signalFrameData->context.uc_mcontext.x[19] = frame->s4;
signalFrameData->context.uc_mcontext.x[20] = frame->s5;
signalFrameData->context.uc_mcontext.x[21] = frame->s6;
signalFrameData->context.uc_mcontext.x[22] = frame->s7;
signalFrameData->context.uc_mcontext.x[23] = frame->s8;
signalFrameData->context.uc_mcontext.x[24] = frame->s9;
signalFrameData->context.uc_mcontext.x[25] = frame->s10;
signalFrameData->context.uc_mcontext.x[26] = frame->s11;
signalFrameData->context.uc_mcontext.x[27] = frame->t3;
signalFrameData->context.uc_mcontext.x[28] = frame->t4;
signalFrameData->context.uc_mcontext.x[29] = frame->t5;
signalFrameData->context.uc_mcontext.x[30] = frame->t6;
signalFrameData->context.uc_mcontext.pc = frame->epc;
// TODO: don't assume that kernel code don't use FPU
save_fpu((fpu_context*)&signalFrameData->context.uc_mcontext.f[0]);
// end of fill signal context
signal_get_user_stack(frame->sp, &signalFrameData->context.uc_stack);
/*
dprintf(" thread->signal_stack_enabled: %d\n",
thread->signal_stack_enabled);
if (thread->signal_stack_enabled) {
dprintf(" signal stack: 0x%" B_PRIxADDR " - 0x%" B_PRIxADDR "\n",
thread->signal_stack_base,
thread->signal_stack_base + thread->signal_stack_size
);
}
*/
uint8* userStack = get_signal_stack(thread, frame, sa,
sizeof(*signalFrameData));
// dprintf(" user stack: 0x%" B_PRIxADDR "\n", (addr_t)userStack);
status_t res = user_memcpy(userStack, signalFrameData,
sizeof(*signalFrameData));
if (res < B_OK)
return res;
addr_t commpageAdr = (addr_t)thread->team->commpage_address;
// dprintf(" commpageAdr: 0x%" B_PRIxADDR "\n", commpageAdr);
addr_t signalHandlerAddr;
ASSERT(user_memcpy(&signalHandlerAddr,
&((addr_t*)commpageAdr)[COMMPAGE_ENTRY_RISCV64_SIGNAL_HANDLER],
sizeof(signalHandlerAddr)) >= B_OK);
signalHandlerAddr += commpageAdr;
frame->ra = frame->epc;
frame->sp = (addr_t)userStack;
frame->epc = signalHandlerAddr;
frame->a0 = frame->sp;
// WriteTrapInfo();
return B_OK;
}
int64
arch_restore_signal_frame(struct signal_frame_data* signalFrameData)
{
return 0;
// dprintf("arch_restore_signal_frame()\n");
iframe* frame = thread_get_current_thread()->arch_info.userFrame;
frame->ra = signalFrameData->context.uc_mcontext.x[ 0];
frame->sp = signalFrameData->context.uc_mcontext.x[ 1];
frame->gp = signalFrameData->context.uc_mcontext.x[ 2];
frame->tp = signalFrameData->context.uc_mcontext.x[ 3];
frame->t0 = signalFrameData->context.uc_mcontext.x[ 4];
frame->t1 = signalFrameData->context.uc_mcontext.x[ 5];
frame->t2 = signalFrameData->context.uc_mcontext.x[ 6];
frame->fp = signalFrameData->context.uc_mcontext.x[ 7];
frame->s1 = signalFrameData->context.uc_mcontext.x[ 8];
frame->a0 = signalFrameData->context.uc_mcontext.x[ 9];
frame->a1 = signalFrameData->context.uc_mcontext.x[10];
frame->a2 = signalFrameData->context.uc_mcontext.x[11];
frame->a3 = signalFrameData->context.uc_mcontext.x[12];
frame->a4 = signalFrameData->context.uc_mcontext.x[13];
frame->a5 = signalFrameData->context.uc_mcontext.x[14];
frame->a6 = signalFrameData->context.uc_mcontext.x[15];
frame->a7 = signalFrameData->context.uc_mcontext.x[16];
frame->s2 = signalFrameData->context.uc_mcontext.x[17];
frame->s3 = signalFrameData->context.uc_mcontext.x[18];
frame->s4 = signalFrameData->context.uc_mcontext.x[19];
frame->s5 = signalFrameData->context.uc_mcontext.x[20];
frame->s6 = signalFrameData->context.uc_mcontext.x[21];
frame->s7 = signalFrameData->context.uc_mcontext.x[22];
frame->s8 = signalFrameData->context.uc_mcontext.x[23];
frame->s9 = signalFrameData->context.uc_mcontext.x[24];
frame->s10 = signalFrameData->context.uc_mcontext.x[25];
frame->s11 = signalFrameData->context.uc_mcontext.x[26];
frame->t3 = signalFrameData->context.uc_mcontext.x[27];
frame->t4 = signalFrameData->context.uc_mcontext.x[28];
frame->t5 = signalFrameData->context.uc_mcontext.x[29];
frame->t6 = signalFrameData->context.uc_mcontext.x[30];
frame->epc = signalFrameData->context.uc_mcontext.pc;
restore_fpu((fpu_context*)&signalFrameData->context.uc_mcontext.f[0]);
return frame->a0;
}
void
arch_check_syscall_restart(Thread *thread)
{
panic("arch_check_syscall_restart(): not yet implemented\n");
}
@ -114,6 +323,15 @@ arch_check_syscall_restart(Thread *thread)
void
arch_store_fork_frame(struct arch_fork_arg *arg)
{
/*
dprintf("arch_store_fork_frame()\n");
dprintf(" arg: %p\n", arg);
dprintf(" userFrame: %p\n",
thread_get_current_thread()->arch_info.userFrame);
*/
memcpy(&arg->frame, thread_get_current_thread()->arch_info.userFrame,
sizeof(iframe));
arg->frame.a0 = 0; // fork return value
}
@ -128,5 +346,15 @@ arch_store_fork_frame(struct arch_fork_arg *arg)
void
arch_restore_fork_frame(struct arch_fork_arg *arg)
{
// dprintf("arch_restore_fork_frame(%p)\n", arg);
disable_interrupts();
if (arch_setjmp(&thread_get_current_thread()->arch_info.context) == 0) {
SstatusReg status(Sstatus());
status.pie = (1 << modeS); // enable interrupts when enter userspace
status.spp = modeU;
SetSstatus(status.val);
arch_longjmp_iframe(&arg->frame);
} else {
panic("return from userspace");
}
}