Re: [RFC PATCH v2 09/45] arm64: entry: Avoid unnecessary local_irq_disable() on kernel exit
From: Jinjie Ruan <hidden>
Date: 2026-07-28 03:19:02
在 2026/7/28 0:34, Vladimir Murzin 写道:
Currently, when exiting to kernel mode, we attempt involuntary preemption. The preemption logic expects IRQs to be disabled, which is why we call local_irq_disable() before attempting preemption. However, depending on the context, local_irq_disable() may be unnecessary: - __el1_irq(), the non-NMI EL1 IRQ path, already has IRQs disabled, so local_irq_disable() is redundant. - irqentry_exit_to_kernel_mode_preempt() immediately returns when exiting from an NMI-like context, so calling local_irq_disable() beforehand is unnecessary work. Split arm64_exit_to_kernel_mode() into preempt, non-preempt, and dispatch parts so that we can avoid this extra work where it is not needed.
Make sense for me, in an NMI-like context, irqentry_exit_to_kernel_mode_preempt() and local_irq_disable() are both unnecessary.
quoted hunk ↗ jump to hunk
Signed-off-by: Vladimir Murzin <redacted> --- arch/arm64/kernel/entry-common.c | 30 +++++++++++++++++++++++++----- 1 file changed, 25 insertions(+), 5 deletions(-)diff --git a/arch/arm64/kernel/entry-common.c b/arch/arm64/kernel/entry-common.c index 2be42d7f4eaa..72c03ccea59f 100644 --- a/arch/arm64/kernel/entry-common.c +++ b/arch/arm64/kernel/entry-common.c@@ -52,16 +52,36 @@ static noinstr irqentry_state_t arm64_enter_from_kernel_mode(struct pt_regs *reg * After this function returns it is not safe to call regular kernel code, * instrumentable code, or any code which may trigger an exception. */ -static void noinstr arm64_exit_to_kernel_mode(struct pt_regs *regs, - irqentry_state_t state) +static void noinstr __arm64_exit_to_kernel_mode(struct pt_regs *regs, + irqentry_state_t state) { - local_irq_disable(); - irqentry_exit_to_kernel_mode_preempt(regs, state); local_daif_mask(); mte_check_tfsr_exit(); irqentry_exit_to_kernel_mode_after_preempt(regs, state); } +/* + * We are returning from the context which allows involuntary kernel preemption
Should we add the irq disabled status for preemption? We are returning from the context which allows involuntary kernel preemption "with irq disabled" ? Reviewed-by: Jinjie Ruan <redacted>
quoted hunk ↗ jump to hunk
+ */ +static void noinstr arm64_exit_to_kernel_mode_preempt(struct pt_regs *regs, + irqentry_state_t state) +{ + irqentry_exit_to_kernel_mode_preempt(regs, state); + __arm64_exit_to_kernel_mode(regs, state); +} + +static void noinstr arm64_exit_to_kernel_mode(struct pt_regs *regs, + irqentry_state_t state) +{ + if (!regs_irqs_disabled(regs)) { + local_irq_disable(); + arm64_exit_to_kernel_mode_preempt(regs, state); + return; + } + + __arm64_exit_to_kernel_mode(regs, state); +} + static __always_inline void arm64_syscall_enter_from_user_mode(struct pt_regs *regs) { enter_from_user_mode(regs);@@ -510,7 +530,7 @@ static __always_inline void __el1_irq(struct pt_regs *regs, do_interrupt_handler(regs, handler); irq_exit_rcu(); - arm64_exit_to_kernel_mode(regs, state); + arm64_exit_to_kernel_mode_preempt(regs, state); } static void noinstr el1_interrupt(struct pt_regs *regs, void (*handler)(struct pt_regs *))