Re: [patch V4 part 3 12/29] x86/entry/common: Provide idtentry_enter/exit()
From: Andy Lutomirski
Date: Mon May 11 2020 - 00:34:56 EST
On Tue, May 5, 2020 at 7:15 AM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>
> Provide functions which handle the low level entry and exit similiar to
> enter/exit from user mode.
>
> +
> +/**
> + * idtentry_exit - Common code to handle return from exceptions
> + * @regs: Pointer to pt_regs (exception entry regs)
> + *
> + * Depending on the return target (kernel/user) this runs the necessary
> + * preemption and work checks if possible and reguired and returns to
> + * the caller with interrupts disabled and no further work pending.
> + *
> + * This is the last action before returning to the low level ASM code which
> + * just needs to return to the appropriate context.
> + *
> + * Invoked by all exception/interrupt IDTENTRY handlers which are not
> + * returning through the paranoid exit path (all except NMI, #DF and the IST
> + * variants of #MC and #DB).
The paranoid-exit bit is not really relevant. The important part is
which stack we're on. See below.
> + */
> +void noinstr idtentry_exit(struct pt_regs *regs)
> +{
> + lockdep_assert_irqs_disabled();
How about:
#ifdef CONFIG_DEBUG_ENTRY
WARN_ON_ONCE(!on_thread_stack());
#endif
> +
> + /* Check whether this returns to user mode */
> + if (user_mode(regs)) {
> + prepare_exit_to_usermode(regs);
> + } else if (regs->flags & X86_EFLAGS_IF) {
> + /* Check kernel preemption, if enabled */
> + if (IS_ENABLED(CONFIG_PREEMPTION)) {
> + /*
> + * This needs to be done very carefully.
> + * idtentry_enter() invoked rcu_irq_enter(). This
> + * needs to undone before scheduling.
> + *
> + * Preemption is disabled inside of RCU idle
> + * sections. When the task returns from
> + * preempt_schedule_irq(), RCU is still watching.
> + *
> + * rcu_irq_exit_preempt() has additional state
> + * checking if CONFIG_PROVE_RCU=y
> + */
> + if (!preempt_count()) {
> + instr_begin();
> + rcu_irq_exit_preempt();
> + if (need_resched())
> + preempt_schedule_irq();
This is an excellent improvement. Thanks!
> + /* Covers both tracing and lockdep */
> + trace_hardirqs_on();
> + instr_end();
> + return;
> + }
> + }
> + instr_begin();
> + /* Tell the tracer that IRET will enable interrupts */
> + trace_hardirqs_on_prepare();
Why is trace_hardirqs_on() okay above but not here? Is it that we
know we weren't RCU-quiescent if we had preemption and IF on? But
even this code path came from an IF-on context. I'm confused. Maybe
some comments as to why this case seems to be ordered so differently
from the !preempt_count() case would be helpful.
> + lockdep_hardirqs_on_prepare(CALLER_ADDR0);
> + instr_end();
> + rcu_irq_exit();
> + lockdep_hardirqs_on(CALLER_ADDR0);
> + } else {
> + /* IRQ flags state is correct already. Just tell RCU */
> + rcu_irq_exit();
> + }
> +}
> --- a/arch/x86/include/asm/idtentry.h
> +++ b/arch/x86/include/asm/idtentry.h
> @@ -7,6 +7,9 @@
>
> #ifndef __ASSEMBLY__
>
> +void idtentry_enter(struct pt_regs *regs);
> +void idtentry_exit(struct pt_regs *regs);
> +
> /**
> * DECLARE_IDTENTRY - Declare functions for simple IDT entry points
> * No error code pushed by hardware
>