sergiotarxz/nuttx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

arch/arm64: Save FPU context when a context switch occurs in SMP mode

Browse Source 
In SMP mode, the fpu owner may switch from core0 to core1,
so it is necessary to force saving the FPU context when a
context switch occurs.

This PR fixed the crash issue mentioned in #8799.

Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
This commit is contained in:
zhangyuan21 2023-04-26 20:34:01 +08:00 committed by Masayuki Ishikawa
parent 6c38c44af2
commit 6f8cb7edd3
3 changed files with 37 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
arch/arm64/src/common/arm64_arch.h
View File

@ -328,7 +328,6 @@ struct fpu_reg
__int128 q[32]; __int128 q[32];
uint32_t fpsr; uint32_t fpsr;
uint32_t fpcr; uint32_t fpcr;
uint64_t fpu_trap;
}; };
#endif #endif

43
arch/arm64/src/common/arm64_fpu.c
View File

@ -268,8 +268,6 @@ static int arm64_fpu_procfs_stat(const char *relpath, struct stat *buf)
void arm64_init_fpu(struct tcb_s *tcb) void arm64_init_fpu(struct tcb_s *tcb)
{ {
struct fpu_reg *fpu_reg;
if (tcb->pid < CONFIG_SMP_NCPUS) if (tcb->pid < CONFIG_SMP_NCPUS)
{ {
memset(&g_cpu_fpu_ctx[this_cpu()], 0, memset(&g_cpu_fpu_ctx[this_cpu()], 0,
@ -280,13 +278,11 @@ void arm64_init_fpu(struct tcb_s *tcb)
} }
memset(tcb->xcp.fpu_regs, 0, sizeof(struct fpu_reg)); memset(tcb->xcp.fpu_regs, 0, sizeof(struct fpu_reg));
fpu_reg = (struct fpu_reg *)tcb->xcp.fpu_regs;
fpu_reg->fpu_trap = 0;
} }
void arm64_destory_fpu(struct tcb_s * tcb) void arm64_destory_fpu(struct tcb_s *tcb)
{ {
struct tcb_s * owner; struct tcb_s *owner;
/* save current fpu owner's context */ /* save current fpu owner's context */
@ -314,10 +310,9 @@ void arm64_fpu_exit_exception(void)
{ {
} }
void arm64_fpu_trap(struct regs_context * regs) void arm64_fpu_trap(struct regs_context *regs)
{ {
struct tcb_s * owner; struct tcb_s *owner;
struct fpu_reg *fpu_reg;
UNUSED(regs); UNUSED(regs);
@ -359,25 +354,14 @@ void arm64_fpu_trap(struct regs_context * regs)
/* become new owner */ /* become new owner */
g_cpu_fpu_ctx[this_cpu()].fpu_owner = owner; g_cpu_fpu_ctx[this_cpu()].fpu_owner = owner;
fpu_reg = (struct fpu_reg *)owner->xcp.fpu_regs;
fpu_reg->fpu_trap = 0;
} }
void arm64_fpu_context_restore(void) void arm64_fpu_context_restore(void)
{ {
struct tcb_s *new_tcb = (struct tcb_s *)arch_get_current_tcb(); struct tcb_s *new_tcb = (struct tcb_s *)arch_get_current_tcb();
struct fpu_reg *fpu_reg = (struct fpu_reg *)new_tcb->xcp.fpu_regs;
arm64_fpu_access_trap_enable(); arm64_fpu_access_trap_disable();
if (fpu_reg->fpu_trap == 0)
{
/* FPU trap hasn't happened at this task */
arm64_fpu_access_trap_enable();
}
else
{
/* FPU trap has happened at this task */ /* FPU trap has happened at this task */
if (new_tcb == g_cpu_fpu_ctx[this_cpu()].fpu_owner) if (new_tcb == g_cpu_fpu_ctx[this_cpu()].fpu_owner)
@ -388,11 +372,26 @@ void arm64_fpu_context_restore(void)
{ {
arm64_fpu_access_trap_enable(); arm64_fpu_access_trap_enable();
} }
}
g_cpu_fpu_ctx[this_cpu()].switch_count++; g_cpu_fpu_ctx[this_cpu()].switch_count++;
} }
#ifdef CONFIG_SMP
void arm64_fpu_context_save(void)
{
struct tcb_s *tcb = (struct tcb_s *)arch_get_current_tcb();
if (tcb == g_cpu_fpu_ctx[this_cpu()].fpu_owner)
{
arm64_fpu_access_trap_disable();
arm64_fpu_save((struct fpu_reg *)tcb->xcp.fpu_regs);
ARM64_DSB();
g_cpu_fpu_ctx[this_cpu()].save_count++;
g_cpu_fpu_ctx[this_cpu()].fpu_owner = NULL;
}
}
#endif
void arm64_fpu_enable(void) void arm64_fpu_enable(void)
{ {
irqstate_t flags = up_irq_save(); irqstate_t flags = up_irq_save();

8
arch/arm64/src/common/arm64_vectors.S
View File

@ -121,6 +121,14 @@ SECTION_FUNC(text, arm64_context_switch)
cmp x1, #0x0 cmp x1, #0x0
beq restore_new beq restore_new
#if defined(CONFIG_ARCH_FPU) && defined(CONFIG_SMP)
stp x0, x1, [sp, #-16]!
stp xzr, x30, [sp, #-16]!
bl arm64_fpu_context_save
ldp xzr, x30, [sp], #16
ldp x0, x1, [sp], #16
#endif
/* Save the current SP_EL0 */ /* Save the current SP_EL0 */
mov x4, sp mov x4, sp
str x4, [x1, #8 * REG_SP_ELX] str x4, [x1, #8 * REG_SP_ELX]