On 14/09/2023 20:24, Rob Clark wrote:
From: Rob Clark <robdclark@xxxxxxxxxxxx>
So, when you want to get a cmdstream trace of some deqp or piglit test,
but you happen to be running it on the same laptop with full desktop
env, the current dump-everything firehose of `cat $debugfs/dri/n/rd` is
quite a bit too much. Ptrace seemed kind of a natural way to control
it, ie. either run what you want to trace under gdb or hack in a
ptrace(PTRACE_TRACEME) call.
Signed-off-by: Rob Clark <robdclark@xxxxxxxxxxxx>
---
drivers/gpu/drm/msm/msm_gem_submit.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
I think this is a good idea, to be able to separate process being
debugged and all other tasks.
diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c
index a908373cf34b..a105ca1f2102 100644
--- a/drivers/gpu/drm/msm/msm_gem_submit.c
+++ b/drivers/gpu/drm/msm/msm_gem_submit.c
@@ -979,7 +979,8 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,
/* The scheduler owns a ref now: */
msm_gem_submit_get(submit);
- msm_rd_dump_submit(priv->rd, submit, NULL);
+ if (current->ptrace & PT_PTRACED)
+ msm_rd_dump_submit(priv->rd, submit, NULL);
pm_runtime_get_sync(&gpu->pdev->dev);
--
With best wishes
Dmitry
