The arm64 entry code only preempts a kernel context upon a return from a regular IRQ exception. The generic entry code may preempt a kernel context for any exception return where irqentry_exit() is used, and so may preempt other exceptions such as faults. In preparation for moving arm64 over to the generic entry code, align arm64 with the generic behaviour by calling arm64_preempt_schedule_irq() from exit_to_kernel_mode(). To make this possible, arm64_preempt_schedule_irq() and dynamic/raw_irqentry_exit_cond_resched() are moved earlier in the file, with no changes. As Mark pointed out, this change will have the following 2 key impact: - " We'll preempt even without taking a "real" interrupt. That shouldn't result in preemption that wasn't possible before, but it does change the probability of preempting at certain points, and might have a performance impact, so probably warrants a benchmark." - " 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 | 92 ++++++++++++++++---------------- 1 file changed, 46 insertions(+), 46 deletions(-) diff --git a/arch/arm64/kernel/entry-common.c b/arch/arm64/kernel/entry-common.c index 4f92664fd46c..7c7aa5711a39 100644 --- a/arch/arm64/kernel/entry-common.c +++ b/arch/arm64/kernel/entry-common.c @@ -76,6 +76,49 @@ static noinstr arm64_irqentry_state_t enter_from_kernel_mode(struct pt_regs *reg return state; } +static inline bool arm64_preempt_schedule_irq(void) +{ + /* + * DAIF.DA are cleared at the start of IRQ/FIQ handling, and when GIC + * priority masking is used the GIC irqchip driver will clear DAIF.IF + * using gic_arch_enable_irqs() for normal IRQs. If anything is set in + * DAIF we must have handled an NMI, so skip preemption. + */ + if (system_uses_irq_prio_masking() && read_sysreg(daif)) + return false; + + /* + * Preempting a task from an IRQ means we leave copies of PSTATE + * on the stack. cpufeature's enable calls may modify PSTATE, but + * resuming one of these preempted tasks would undo those changes. + * + * Only allow a task to be preempted once cpufeatures have been + * enabled. + */ + if (!system_capabilities_finalized()) + return false; + + return true; +} + +void raw_irqentry_exit_cond_resched(void) +{ + if (!preempt_count()) { + if (need_resched() && arm64_preempt_schedule_irq()) + preempt_schedule_irq(); + } +} + +#ifdef CONFIG_PREEMPT_DYNAMIC +DEFINE_STATIC_KEY_TRUE(sk_dynamic_irqentry_exit_cond_resched); +void dynamic_irqentry_exit_cond_resched(void) +{ + if (!static_branch_unlikely(&sk_dynamic_irqentry_exit_cond_resched)) + return; + raw_irqentry_exit_cond_resched(); +} +#endif + /* * Handle IRQ/context state management when exiting to kernel mode. * After this function returns it is not safe to call regular kernel code, @@ -98,6 +141,9 @@ static __always_inline void __exit_to_kernel_mode(struct pt_regs *regs, return; } + if (IS_ENABLED(CONFIG_PREEMPTION)) + irqentry_exit_cond_resched(); + trace_hardirqs_on(); } else { if (state.exit_rcu) @@ -285,31 +331,6 @@ static void noinstr arm64_exit_el1_dbg(struct pt_regs *regs, lockdep_hardirqs_on(CALLER_ADDR0); } -static inline bool arm64_preempt_schedule_irq(void) -{ - /* - * DAIF.DA are cleared at the start of IRQ/FIQ handling, and when GIC - * priority masking is used the GIC irqchip driver will clear DAIF.IF - * using gic_arch_enable_irqs() for normal IRQs. If anything is set in - * DAIF we must have handled an NMI, so skip preemption. - */ - if (system_uses_irq_prio_masking() && read_sysreg(daif)) - return false; - - /* - * Preempting a task from an IRQ means we leave copies of PSTATE - * on the stack. cpufeature's enable calls may modify PSTATE, but - * resuming one of these preempted tasks would undo those changes. - * - * Only allow a task to be preempted once cpufeatures have been - * enabled. - */ - if (!system_capabilities_finalized()) - return false; - - return true; -} - static void do_interrupt_handler(struct pt_regs *regs, void (*handler)(struct pt_regs *)) { @@ -661,24 +682,6 @@ static __always_inline void __el1_pnmi(struct pt_regs *regs, arm64_exit_nmi(regs, state); } -void raw_irqentry_exit_cond_resched(void) -{ - if (!preempt_count()) { - if (need_resched() && arm64_preempt_schedule_irq()) - preempt_schedule_irq(); - } -} - -#ifdef CONFIG_PREEMPT_DYNAMIC -DEFINE_STATIC_KEY_TRUE(sk_dynamic_irqentry_exit_cond_resched); -void dynamic_irqentry_exit_cond_resched(void) -{ - if (!static_branch_unlikely(&sk_dynamic_irqentry_exit_cond_resched)) - return; - raw_irqentry_exit_cond_resched(); -} -#endif - static __always_inline void __el1_irq(struct pt_regs *regs, void (*handler)(struct pt_regs *)) { @@ -688,9 +691,6 @@ static __always_inline void __el1_irq(struct pt_regs *regs, do_interrupt_handler(regs, handler); irq_exit_rcu(); - if (IS_ENABLED(CONFIG_PREEMPTION)) - irqentry_exit_cond_resched(); - exit_to_kernel_mode(regs, state); } static void noinstr el1_interrupt(struct pt_regs *regs, -- 2.34.1