| Message ID | 20241025100700.3714552-8-ruanjinjie@huawei.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | arm64: entry: Convert to generic entry | expand |
On Fri, Oct 25, 2024 at 06:06:48PM +0800, Jinjie Ruan wrote: > Only if irqs are enabled when the interrupt trapped, there may be > a chance to reschedule after the interrupt has been handled, so move > arm64_preempt_schedule_irq() into regs_irqs_disabled() check false > if block. > > As Mark pointed out, this change will have the following key impact: > > "We will not preempt when taking interrupts from a region of kernel > code where IRQs are enabled but RCU is not watching, matching the > behaviour of the generic entry code. > > This has the potential to introduce livelock if we can ever have a > screaming interrupt in such a region, so we'll need to go figure out > whether that's actually a problem. > > Having this as a separate patch will make it easier to test/bisect > for that specifically." > > Suggested-by: Mark Rutland <mark.rutland@arm.com> > Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com> This should be folded into the prior patch. Mark. > --- > arch/arm64/kernel/entry-common.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/arch/arm64/kernel/entry-common.c b/arch/arm64/kernel/entry-common.c > index e0380812d71e..b57f6dc66115 100644 > --- a/arch/arm64/kernel/entry-common.c > +++ b/arch/arm64/kernel/entry-common.c > @@ -114,8 +114,6 @@ static void __sched arm64_preempt_schedule_irq(void) > static void noinstr exit_to_kernel_mode(struct pt_regs *regs, > irqentry_state_t state) > { > - arm64_preempt_schedule_irq(); > - > mte_check_tfsr_exit(); > > lockdep_assert_irqs_disabled(); > @@ -129,6 +127,8 @@ static void noinstr exit_to_kernel_mode(struct pt_regs *regs, > return; > } > > + arm64_preempt_schedule_irq(); > + > trace_hardirqs_on(); > } else { > if (state.exit_rcu) > -- > 2.34.1 >
diff --git a/arch/arm64/kernel/entry-common.c b/arch/arm64/kernel/entry-common.c index e0380812d71e..b57f6dc66115 100644 --- a/arch/arm64/kernel/entry-common.c +++ b/arch/arm64/kernel/entry-common.c @@ -114,8 +114,6 @@ static void __sched arm64_preempt_schedule_irq(void) static void noinstr exit_to_kernel_mode(struct pt_regs *regs, irqentry_state_t state) { - arm64_preempt_schedule_irq(); - mte_check_tfsr_exit(); lockdep_assert_irqs_disabled(); @@ -129,6 +127,8 @@ static void noinstr exit_to_kernel_mode(struct pt_regs *regs, return; } + arm64_preempt_schedule_irq(); + trace_hardirqs_on(); } else { if (state.exit_rcu)
Only if irqs are enabled when the interrupt trapped, there may be a chance to reschedule after the interrupt has been handled, so move arm64_preempt_schedule_irq() into regs_irqs_disabled() check false if block. As Mark pointed out, this change will have the following key impact: "We will not preempt when taking interrupts from a region of kernel code where IRQs are enabled but RCU is not watching, matching the behaviour of the generic entry code. This has the potential to introduce livelock if we can ever have a screaming interrupt in such a region, so we'll need to go figure out whether that's actually a problem. Having this as a separate patch will make it easier to test/bisect for that specifically." Suggested-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com> --- arch/arm64/kernel/entry-common.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)