Re: [PATCH v1] arm: clocksource: Check if timer is enabled for timer irq

[Marc Zyngier]

On Thu, 25 May 2023 17:03:11 +0100,
Ayan Kumar Halder <ayankuma@xxxxxxx> wrote:
> 
> Hi Marc,
> 
> Apologies, this got lost in my mailbox.
> 
> On 11/08/2022 10:49, Marc Zyngier wrote:
> > On Thu, 11 Aug 2022 10:36:20 +0100,
> > Ayan Kumar Halder <ayankuma@xxxxxxx> wrote:
> >> Refer ARM DDI 0487G.b, CNTP_CTL_EL0,
> >> ISTATUS, bit [2] - When the value of the ENABLE bit is 1, ISTATUS
> >> indicates whether the timer condition is met.
> >> 
> >> Thus, one need to check ENABLE bit along with ISTATUS, to confirm
> >> whether the timer condition is met. Further as the doc says,
> >> "When the value of the ENABLE bit is 0, the ISTATUS field is UNKNOWN."
> >> 
> >> Signed-off-by: Ayan Kumar Halder <ayankuma@xxxxxxx>
> >> ---
> >> 
> >> Please refer to https://lore.kernel.org/all/6cfcd4fa-3afd-1c70-6a70-9df557ee1811@xxxxxxx/T/
> >> for the previous discussion on this issue on xen-devel mailing list.
> >> 
> >>   drivers/clocksource/arm_arch_timer.c | 2 +-
> >>   1 file changed, 1 insertion(+), 1 deletion(-)
> >> 
> >> diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
> >> index 9ab8221ee3c6..96921772814c 100644
> >> --- a/drivers/clocksource/arm_arch_timer.c
> >> +++ b/drivers/clocksource/arm_arch_timer.c
> >> @@ -647,7 +647,7 @@ static __always_inline irqreturn_t timer_handler(const int access,
> >>   	unsigned long ctrl;
> >>     	ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, evt);
> >> -	if (ctrl & ARCH_TIMER_CTRL_IT_STAT) {
> >> +	if ((ctrl & ARCH_TIMER_CTRL_IT_STAT) && (ctrl & ARCH_TIMER_CTRL_ENABLE)) {
> >>   		ctrl |= ARCH_TIMER_CTRL_IT_MASK;
> >>   		arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, evt);
> >>   		evt->event_handler(evt);
> > And how can the timer be disabled if we're in the interrupt handler?
> 
> I am not very sure about this.
> 
> Is it possible for a pending interrupt to arrive just after the timer
> has been disabled ?

I can't see how. This is a level interrupt, so disabling it at the
source also retires the pending bit from the GIC. And if what you
describe actually happens, this is only a spurious interrupt and most
probably a slightly broken interrupt controller implementation.

Overall, this change doesn't make much sense, unless you can show me a
HW implementation that is broken enough that this leads to actual
problems.

	M.

-- 
Without deviation from the norm, progress is not possible.