drm/i915/selftest: Synchronise with the GPU timestamp

Wait for the GPU to wake up from the semaphore before measuring the
time, so that we coordinate the sampling on both the CPU and GPU for
more accurate comparisons.

v2: Switch to local_irq_disable() as once suggested by Mika.

Reported-by: Bruce Chang <yu.bruce.chang@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: CQ Tang <cq.tang@intel.com>
Reviewed-by: Bruce Chang <yu.bruce.chang@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210205112912.22978-1-chris@chris-wilson.co.uk
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

diff --git a/drivers/gpu/drm/i915/gt/selftest_engine_pm.c b/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
index 41dc1a5..2c89862 100644 (file)
--- a/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
+++ b/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
@@ -110,13 +110,15 @@ static int __measure_timestamps(struct intel_context *ce,
                cpu_relax();
 
        /* Run the request for a 100us, sampling timestamps before/after */
-       preempt_disable();
-       *dt = local_clock();
+       local_irq_disable();
        write_semaphore(&sema[2], 0);
+       while (READ_ONCE(sema[1]) == 0) /* wait for the gpu to catch up */
+               cpu_relax();
+       *dt = local_clock();
        udelay(100);
        *dt = local_clock() - *dt;
        write_semaphore(&sema[2], 1);
-       preempt_enable();
+       local_irq_enable();
 
        if (i915_request_wait(rq, 0, HZ / 2) < 0) {
                i915_request_put(rq);