linux/arch/powerpc/kernel/syscall_64.c
Nicholas Piggin 01eb01877f powerpc/64s: Fix restore_math unnecessarily changing MSR
Before returning to user, if there are missing FP/VEC/VSX bits from the
user MSR then those registers had been saved and must be restored again
before use. restore_math will decide whether to restore immediately, or
skip the restore and let fp/vec/vsx unavailable faults demand load the
registers.

Each time restore_math restores one of the FP/VSX or VEC register sets
is loaded, an 8-bit counter is incremented (load_fp and load_vec). When
these wrap to zero, restore_math no longer restores that register set
until after they are next demand faulted.

It's quite usual for those counters to have different values, so if one
wraps to zero and restore_math no longer restores its registers or user
MSR bit but the other is not zero yet does not need to be restored
(because the kernel is not frequently using the FPU), then restore_math
will be called and it will also not return in the early exit check.
This causes msr_check_and_set to test and set the MSR at every kernel
exit despite having no work to do.

This can cause workloads (e.g., a NULL syscall microbenchmark) to run
fast for a time while both counters are non-zero, then slow down when
one of the counters reaches zero, then speed up again after the second
counter reaches zero. The cost is significant, about 10% slowdown on a
NULL syscall benchmark, and the jittery behaviour is very undesirable.

Fix this by having restore_math test all conditions first, and only
update MSR if we will be loading registers.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-07-16 13:00:24 +10:00

392 lines
10 KiB
C

// SPDX-License-Identifier: GPL-2.0-or-later
#include <linux/err.h>
#include <asm/asm-prototypes.h>
#include <asm/book3s/64/kup-radix.h>
#include <asm/cputime.h>
#include <asm/hw_irq.h>
#include <asm/kprobes.h>
#include <asm/paca.h>
#include <asm/ptrace.h>
#include <asm/reg.h>
#include <asm/signal.h>
#include <asm/switch_to.h>
#include <asm/syscall.h>
#include <asm/time.h>
#include <asm/unistd.h>
typedef long (*syscall_fn)(long, long, long, long, long, long);
/* Has to run notrace because it is entered not completely "reconciled" */
notrace long system_call_exception(long r3, long r4, long r5,
long r6, long r7, long r8,
unsigned long r0, struct pt_regs *regs)
{
syscall_fn f;
if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
BUG_ON(irq_soft_mask_return() != IRQS_ALL_DISABLED);
trace_hardirqs_off(); /* finish reconciling */
if (IS_ENABLED(CONFIG_PPC_BOOK3S))
BUG_ON(!(regs->msr & MSR_RI));
BUG_ON(!(regs->msr & MSR_PR));
BUG_ON(!FULL_REGS(regs));
BUG_ON(regs->softe != IRQS_ENABLED);
kuap_check_amr();
account_cpu_user_entry();
#ifdef CONFIG_PPC_SPLPAR
if (IS_ENABLED(CONFIG_VIRT_CPU_ACCOUNTING_NATIVE) &&
firmware_has_feature(FW_FEATURE_SPLPAR)) {
struct lppaca *lp = local_paca->lppaca_ptr;
if (unlikely(local_paca->dtl_ridx != be64_to_cpu(lp->dtl_idx)))
accumulate_stolen_time();
}
#endif
/*
* This is not required for the syscall exit path, but makes the
* stack frame look nicer. If this was initialised in the first stack
* frame, or if the unwinder was taught the first stack frame always
* returns to user with IRQS_ENABLED, this store could be avoided!
*/
regs->softe = IRQS_ENABLED;
local_irq_enable();
if (unlikely(current_thread_info()->flags & _TIF_SYSCALL_DOTRACE)) {
/*
* We use the return value of do_syscall_trace_enter() as the
* syscall number. If the syscall was rejected for any reason
* do_syscall_trace_enter() returns an invalid syscall number
* and the test against NR_syscalls will fail and the return
* value to be used is in regs->gpr[3].
*/
r0 = do_syscall_trace_enter(regs);
if (unlikely(r0 >= NR_syscalls))
return regs->gpr[3];
r3 = regs->gpr[3];
r4 = regs->gpr[4];
r5 = regs->gpr[5];
r6 = regs->gpr[6];
r7 = regs->gpr[7];
r8 = regs->gpr[8];
} else if (unlikely(r0 >= NR_syscalls)) {
return -ENOSYS;
}
/* May be faster to do array_index_nospec? */
barrier_nospec();
if (unlikely(is_32bit_task())) {
f = (void *)compat_sys_call_table[r0];
r3 &= 0x00000000ffffffffULL;
r4 &= 0x00000000ffffffffULL;
r5 &= 0x00000000ffffffffULL;
r6 &= 0x00000000ffffffffULL;
r7 &= 0x00000000ffffffffULL;
r8 &= 0x00000000ffffffffULL;
} else {
f = (void *)sys_call_table[r0];
}
return f(r3, r4, r5, r6, r7, r8);
}
/*
* local irqs must be disabled. Returns false if the caller must re-enable
* them, check for new work, and try again.
*/
static notrace inline bool prep_irq_for_enabled_exit(void)
{
/* This must be done with RI=1 because tracing may touch vmaps */
trace_hardirqs_on();
/* This pattern matches prep_irq_for_idle */
__hard_EE_RI_disable();
if (unlikely(lazy_irq_pending_nocheck())) {
/* Took an interrupt, may have more exit work to do. */
__hard_RI_enable();
trace_hardirqs_off();
local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
return false;
}
local_paca->irq_happened = 0;
irq_soft_mask_set(IRQS_ENABLED);
return true;
}
/*
* This should be called after a syscall returns, with r3 the return value
* from the syscall. If this function returns non-zero, the system call
* exit assembly should additionally load all GPR registers and CTR and XER
* from the interrupt frame.
*
* The function graph tracer can not trace the return side of this function,
* because RI=0 and soft mask state is "unreconciled", so it is marked notrace.
*/
notrace unsigned long syscall_exit_prepare(unsigned long r3,
struct pt_regs *regs)
{
unsigned long *ti_flagsp = &current_thread_info()->flags;
unsigned long ti_flags;
unsigned long ret = 0;
kuap_check_amr();
regs->result = r3;
/* Check whether the syscall is issued inside a restartable sequence */
rseq_syscall(regs);
ti_flags = *ti_flagsp;
if (unlikely(r3 >= (unsigned long)-MAX_ERRNO)) {
if (likely(!(ti_flags & (_TIF_NOERROR | _TIF_RESTOREALL)))) {
r3 = -r3;
regs->ccr |= 0x10000000; /* Set SO bit in CR */
}
}
if (unlikely(ti_flags & _TIF_PERSYSCALL_MASK)) {
if (ti_flags & _TIF_RESTOREALL)
ret = _TIF_RESTOREALL;
else
regs->gpr[3] = r3;
clear_bits(_TIF_PERSYSCALL_MASK, ti_flagsp);
} else {
regs->gpr[3] = r3;
}
if (unlikely(ti_flags & _TIF_SYSCALL_DOTRACE)) {
do_syscall_trace_leave(regs);
ret |= _TIF_RESTOREALL;
}
again:
local_irq_disable();
ti_flags = READ_ONCE(*ti_flagsp);
while (unlikely(ti_flags & (_TIF_USER_WORK_MASK & ~_TIF_RESTORE_TM))) {
local_irq_enable();
if (ti_flags & _TIF_NEED_RESCHED) {
schedule();
} else {
/*
* SIGPENDING must restore signal handler function
* argument GPRs, and some non-volatiles (e.g., r1).
* Restore all for now. This could be made lighter.
*/
if (ti_flags & _TIF_SIGPENDING)
ret |= _TIF_RESTOREALL;
do_notify_resume(regs, ti_flags);
}
local_irq_disable();
ti_flags = READ_ONCE(*ti_flagsp);
}
if (IS_ENABLED(CONFIG_PPC_BOOK3S) && IS_ENABLED(CONFIG_PPC_FPU)) {
if (IS_ENABLED(CONFIG_PPC_TRANSACTIONAL_MEM) &&
unlikely((ti_flags & _TIF_RESTORE_TM))) {
restore_tm_state(regs);
} else {
unsigned long mathflags = MSR_FP;
if (cpu_has_feature(CPU_FTR_VSX))
mathflags |= MSR_VEC | MSR_VSX;
else if (cpu_has_feature(CPU_FTR_ALTIVEC))
mathflags |= MSR_VEC;
/*
* If userspace MSR has all available FP bits set,
* then they are live and no need to restore. If not,
* it means the regs were given up and restore_math
* may decide to restore them (to avoid taking an FP
* fault).
*/
if ((regs->msr & mathflags) != mathflags)
restore_math(regs);
}
}
if (unlikely(!prep_irq_for_enabled_exit())) {
local_irq_enable();
goto again;
}
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
local_paca->tm_scratch = regs->msr;
#endif
account_cpu_user_exit();
return ret;
}
#ifdef CONFIG_PPC_BOOK3S /* BOOK3E not yet using this */
notrace unsigned long interrupt_exit_user_prepare(struct pt_regs *regs, unsigned long msr)
{
#ifdef CONFIG_PPC_BOOK3E
struct thread_struct *ts = &current->thread;
#endif
unsigned long *ti_flagsp = &current_thread_info()->flags;
unsigned long ti_flags;
unsigned long flags;
unsigned long ret = 0;
if (IS_ENABLED(CONFIG_PPC_BOOK3S))
BUG_ON(!(regs->msr & MSR_RI));
BUG_ON(!(regs->msr & MSR_PR));
BUG_ON(!FULL_REGS(regs));
BUG_ON(regs->softe != IRQS_ENABLED);
/*
* We don't need to restore AMR on the way back to userspace for KUAP.
* AMR can only have been unlocked if we interrupted the kernel.
*/
kuap_check_amr();
local_irq_save(flags);
again:
ti_flags = READ_ONCE(*ti_flagsp);
while (unlikely(ti_flags & (_TIF_USER_WORK_MASK & ~_TIF_RESTORE_TM))) {
local_irq_enable(); /* returning to user: may enable */
if (ti_flags & _TIF_NEED_RESCHED) {
schedule();
} else {
if (ti_flags & _TIF_SIGPENDING)
ret |= _TIF_RESTOREALL;
do_notify_resume(regs, ti_flags);
}
local_irq_disable();
ti_flags = READ_ONCE(*ti_flagsp);
}
if (IS_ENABLED(CONFIG_PPC_BOOK3S) && IS_ENABLED(CONFIG_PPC_FPU)) {
if (IS_ENABLED(CONFIG_PPC_TRANSACTIONAL_MEM) &&
unlikely((ti_flags & _TIF_RESTORE_TM))) {
restore_tm_state(regs);
} else {
unsigned long mathflags = MSR_FP;
if (cpu_has_feature(CPU_FTR_VSX))
mathflags |= MSR_VEC | MSR_VSX;
else if (cpu_has_feature(CPU_FTR_ALTIVEC))
mathflags |= MSR_VEC;
/* See above restore_math comment */
if ((regs->msr & mathflags) != mathflags)
restore_math(regs);
}
}
if (unlikely(!prep_irq_for_enabled_exit())) {
local_irq_enable();
local_irq_disable();
goto again;
}
#ifdef CONFIG_PPC_BOOK3E
if (unlikely(ts->debug.dbcr0 & DBCR0_IDM)) {
/*
* Check to see if the dbcr0 register is set up to debug.
* Use the internal debug mode bit to do this.
*/
mtmsr(mfmsr() & ~MSR_DE);
mtspr(SPRN_DBCR0, ts->debug.dbcr0);
mtspr(SPRN_DBSR, -1);
}
#endif
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
local_paca->tm_scratch = regs->msr;
#endif
account_cpu_user_exit();
return ret;
}
void unrecoverable_exception(struct pt_regs *regs);
void preempt_schedule_irq(void);
notrace unsigned long interrupt_exit_kernel_prepare(struct pt_regs *regs, unsigned long msr)
{
unsigned long *ti_flagsp = &current_thread_info()->flags;
unsigned long flags;
unsigned long ret = 0;
unsigned long amr;
if (IS_ENABLED(CONFIG_PPC_BOOK3S) && unlikely(!(regs->msr & MSR_RI)))
unrecoverable_exception(regs);
BUG_ON(regs->msr & MSR_PR);
BUG_ON(!FULL_REGS(regs));
amr = kuap_get_and_check_amr();
if (unlikely(*ti_flagsp & _TIF_EMULATE_STACK_STORE)) {
clear_bits(_TIF_EMULATE_STACK_STORE, ti_flagsp);
ret = 1;
}
local_irq_save(flags);
if (regs->softe == IRQS_ENABLED) {
/* Returning to a kernel context with local irqs enabled. */
WARN_ON_ONCE(!(regs->msr & MSR_EE));
again:
if (IS_ENABLED(CONFIG_PREEMPT)) {
/* Return to preemptible kernel context */
if (unlikely(*ti_flagsp & _TIF_NEED_RESCHED)) {
if (preempt_count() == 0)
preempt_schedule_irq();
}
}
if (unlikely(!prep_irq_for_enabled_exit())) {
/*
* Can't local_irq_restore to replay if we were in
* interrupt context. Must replay directly.
*/
if (irqs_disabled_flags(flags)) {
replay_soft_interrupts();
} else {
local_irq_restore(flags);
local_irq_save(flags);
}
/* Took an interrupt, may have more exit work to do. */
goto again;
}
} else {
/* Returning to a kernel context with local irqs disabled. */
__hard_EE_RI_disable();
if (regs->msr & MSR_EE)
local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
}
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
local_paca->tm_scratch = regs->msr;
#endif
/*
* Don't want to mfspr(SPRN_AMR) here, because this comes after mtmsr,
* which would cause Read-After-Write stalls. Hence, we take the AMR
* value from the check above.
*/
kuap_restore_amr(regs, amr);
return ret;
}
#endif