Re: [PATCH] x86/kvm: Don't enable IRQ when IRQ enabled in kvm_wait

[Paolo Bonzini]

On 14/08/21 05:51, Lai Jiangshan wrote:
> From: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>
>
> Commit f4e61f0c9add3 ("x86/kvm: Fix broken irq restoration in kvm_wait")
> replaced "local_irq_restore() when IRQ enabled" with "local_irq_enable()
> when IRQ enabled" to suppress a warnning.
>
> Although there is no similar debugging warnning for doing local_irq_enable()
> when IRQ enabled as doing local_irq_restore() in the same IRQ situation.  But
> doing local_irq_enable() when IRQ enabled is no less broken as doing
> local_irq_restore() and we'd better avoid it.
>
> Cc: Mark Rutland <mark.rutland@xxxxxxx>
> Cc: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> Signed-off-by: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>
> ---
>
> The original debugging warnning was introduced in commit 997acaf6b4b5
> ("lockdep: report broken irq restoration").  I think a similar debugging
> check and warnning should also be added to "local_irq_enable() when IRQ
> enabled" and even maybe "local_irq_disable() when IRQ disabled" to detect
> something this:
>
>      | local_irq_save(flags);
>      | local_irq_disable();
>      | local_irq_restore(flags);
>      | local_irq_enable();
>
> Or even we can do the check in lockdep+TRACE_IRQFLAGS:
>
> In lockdep_hardirqs_on_prepare(), lockdep_hardirqs_enabled() was checked
> (and exit) before checking DEBUG_LOCKS_WARN_ON(!irqs_disabled()), so lockdep
> can't give any warning for these kind of situations.  If we did the check
> in lockdep, we would have found the problem before, and we don't need
> 997acaf6b4b5.
>
> Any thought? Mark? Peter?
>
>   arch/x86/kernel/kvm.c | 5 +++--
>   1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
> index a26643dc6bd6..b656456c3a94 100644
> --- a/arch/x86/kernel/kvm.c
> +++ b/arch/x86/kernel/kvm.c
> @@ -884,10 +884,11 @@ static void kvm_wait(u8 *ptr, u8 val)
>   	} else {
>   		local_irq_disable();
>   
> +		/* safe_halt() will enable IRQ */
>   		if (READ_ONCE(*ptr) == val)
>   			safe_halt();
> -
> -		local_irq_enable();
> +		else
> +			local_irq_enable();
>   	}
>   }
>   

Queued, thanks.

paolo