On Thu, May 20, 2021 at 08:35:14AM +0100, Matthew Auld wrote:
From: Chris Wilson chris@chris-wilson.co.uk
The first tracepoint for a request is trace_dma_fence_init called before we have associated the request with a device. The tracepoint uses fence->ops->get_driver_name() as a pretty name, and as we try to report the device name this oopses as it is then NULL. Support the early tracepoint by reporting the DRIVER_NAME instead of the actual device name.
Note that rq->engine remains during the course of request recycling (SLAB_TYPESAFE_BY_RCU). For the physical engines, the pointer remains valid, however a virtual engine may be destroyed after the request is retired. If we process a preempt-to-busy completed request along the virtual engine, we should make sure we mark the request as no longer belonging to the virtual engine to remove the dangling pointers from the tracepoint.
Why can't we assign the request beforehand? The idea behind these tracepoints is that they actually match up, if trace_dma_fence_init is different, then we're breaking that. -Daniel
Fixes: 855e39e65cfc ("drm/i915: Initialise basic fence before acquiring seqno") Signed-off-by: Chris Wilson chris@chris-wilson.co.uk Cc: Tvrtko Ursulin tvrtko.ursulin@linux.intel.com Cc: Chintan M Patel chintan.m.patel@intel.com Cc: Andi Shyti andi.shyti@intel.com Cc: stable@vger.kernel.org # v5.7+ Signed-off-by: Matthew Auld matthew.auld@intel.com
.../drm/i915/gt/intel_execlists_submission.c | 20 ++++++++++++++----- drivers/gpu/drm/i915/i915_request.c | 7 ++++++- 2 files changed, 21 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c index de124870af44..75604e927d34 100644 --- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c +++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c @@ -3249,6 +3249,18 @@ static struct list_head *virtual_queue(struct virtual_engine *ve) return &ve->base.execlists.default_priolist.requests; }
+static void +virtual_submit_completed(struct virtual_engine *ve, struct i915_request *rq) +{
- GEM_BUG_ON(!__i915_request_is_complete(rq));
- GEM_BUG_ON(rq->engine != &ve->base);
- __i915_request_submit(rq);
- /* Remove the dangling pointer to the stale virtual engine */
- WRITE_ONCE(rq->engine, ve->siblings[0]);
+}
static void rcu_virtual_context_destroy(struct work_struct *wrk) { struct virtual_engine *ve = @@ -3265,8 +3277,7 @@ static void rcu_virtual_context_destroy(struct work_struct *wrk)
old = fetch_and_zero(&ve->request); if (old) {
GEM_BUG_ON(!__i915_request_is_complete(old));__i915_request_submit(old);
virtual_submit_completed(ve, old); i915_request_put(old);@@ -3538,13 +3549,12 @@ static void virtual_submit_request(struct i915_request *rq)
/* By the time we resubmit a request, it may be completed */ if (__i915_request_is_complete(rq)) {
__i915_request_submit(rq);
virtual_submit_completed(ve, rq);goto unlock; }
if (ve->request) { /* background completion from preempt-to-busy */
GEM_BUG_ON(!__i915_request_is_complete(ve->request));__i915_request_submit(ve->request);
virtual_submit_completed(ve, ve->request);diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index 970d8f4986bb..aa124adb1051 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -61,7 +61,12 @@ static struct i915_global_request {
static const char *i915_fence_get_driver_name(struct dma_fence *fence) {
- return dev_name(to_request(fence)->engine->i915->drm.dev);
- struct i915_request *rq = to_request(fence);
- if (unlikely(!rq->engine)) /* not yet attached to any device */
return DRIVER_NAME;- return dev_name(rq->engine->i915->drm.dev);
}
static const char *i915_fence_get_timeline_name(struct dma_fence *fence)
2.26.3