Re: [PATCH v4 4/7] drm/i915/gt: Fix selftests on PREEMPT_RT

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Incomplete implementation of PREEMPT_RT fixes for i915 selftests, leaving preemption-disabled busy-waits and delays that cause deadlocks and latency spikes.
--

commit c008dff1b2cefd9ee50ca924f83ff7c237a6fb3a
Author: Maarten Lankhorst <dev@lankhorst.se>

drm/i915/gt: Fix selftests on PREEMPT_RT

This commit modifies the i915 engine PM selftest to use migrate_disable()
instead of preempt_disable() around delay functions. This helps prevent 
latency issues on PREEMPT_RT kernels.

> diff --git a/drivers/gpu/drm/i915/gt/selftest_engine_pm.c b/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
> index 10e556a7eac45..c1eff9edd8a5e 100644
> --- a/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
> +++ b/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
[ ... ]
> @@ -316,11 +316,11 @@ static int live_engine_busy_stats(void *arg)
>  		}
>  
>  		ENGINE_TRACE(engine, "measuring busy time\n");
> -		preempt_disable();
> +		migrate_disable();
>  		de = intel_engine_get_busy_time(engine, &t[0]);
>  		mdelay(100);
>  		de = ktime_sub(intel_engine_get_busy_time(engine, &t[1]), de);
> -		preempt_enable();
> +		migrate_enable();

[Severity: High]
This isn't a bug introduced by this patch, but since this series addresses
PREEMPT_RT issues in i915 selftests, are there similar preemption-disabled
delays and waits remaining in other selftests that should be addressed?

For example, looking at live_rps_clock_interval() in
drivers/gpu/drm/i915/gt/selftest_rps.c, there is a block that disables
preemption for at least 1 millisecond:

    preempt_disable();
    cycles_[i] = -intel_uncore_read_fw(gt->uncore, GEN6_RP_CUR_UP_EI);
    dt_[i] = ktime_get();
    
    udelay(1000);
    
    cycles_[i] += intel_uncore_read_fw(gt->uncore, GEN6_RP_CUR_UP_EI);
    dt_[i] = ktime_sub(ktime_get(), dt_[i]);
    preempt_enable();

Could this cause a severe latency spike on PREEMPT_RT?

Additionally, in drivers/gpu/drm/i915/selftests/i915_request.c,
measure_completion() disables preemption while busy-waiting for a callback:

    preempt_disable();
    semaphore_set(sema, i);
    while (!READ_ONCE(cb.seen))
        cpu_relax();
        
    elapsed[i - 1] = ENGINE_READ_FW(ce->engine, RING_TIMESTAMP);
    preempt_enable();

If the threaded handler that executes this callback is scheduled on the same
CPU on a PREEMPT_RT kernel, would this create a hard deadlock since the
busy-wait loop will never yield to the handler?

>  		dt = ktime_sub(t[1], t[0]);
>  		if (100 * de < 95 * dt || 95 * de > 100 * dt) {
>  			pr_err("%s: reported %lldns [%d%%] busyness while spinning [for %lldns]\n",

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260706114823.752313-1-dev@lankhorst.se?part=4