[Intel-gfx] [PATCH] drm/i915: Enable debugobjects for request validation
Chris Wilson
chris at chris-wilson.co.uk
Sun Jan 28 09:28:49 UTC 2018
Use debugobjects to track and validate the lifecycle of a struct
drm_i915_gem_request.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/Kconfig.debug | 14 +++
drivers/gpu/drm/i915/i915_gem_request.c | 158 ++++++++++++++++++++++++++++++--
2 files changed, 165 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/i915/Kconfig.debug b/drivers/gpu/drm/i915/Kconfig.debug
index 108d21f34777..326be039e39c 100644
--- a/drivers/gpu/drm/i915/Kconfig.debug
+++ b/drivers/gpu/drm/i915/Kconfig.debug
@@ -27,6 +27,7 @@ config DRM_I915_DEBUG
select DRM_DEBUG_MM if DRM=y
select DRM_DEBUG_MM_SELFTEST
select SW_SYNC # signaling validation framework (igt/syncobj*)
+ select DRM_I915_REQUEST_DEBUG_OBJECTS
select DRM_I915_SW_FENCE_DEBUG_OBJECTS
select DRM_I915_SELFTEST
default n
@@ -64,6 +65,19 @@ config DRM_I915_TRACE_GEM
If in doubt, say "N".
+config DRM_I915_REQUEST_DEBUG_OBJECTS
+ bool "Enable additional driver debugging for request objects"
+ depends on DRM_I915
+ select DEBUG_OBJECTS
+ default n
+ help
+ Choose this option to turn on extra driver debugging that may affect
+ performance but will catch some internal issues.
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
config DRM_I915_SW_FENCE_DEBUG_OBJECTS
bool "Enable additional driver debugging for fence objects"
depends on DRM_I915
diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index 06ec27a68f5c..001a02d2b603 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -30,14 +30,128 @@
#include "i915_drv.h"
+enum {
+ DEBUG_REQUEST_INACTIVE = 0,
+ DEBUG_REQUEST_WAITING,
+ DEBUG_REQUEST_READY,
+ DEBUG_REQUEST_EXECUTING,
+ DEBUG_REQUEST_RETIRED,
+};
+
+#if IS_ENABLED(CONFIG_KASAN)
+static inline struct drm_i915_gem_request *
+__request_alloc(struct drm_i915_private *i915, gfp_t gfp)
+{
+ return kmalloc(sizeof(struct drm_i915_gem_request), gfp);
+}
+
+static inline void
+__request_free(struct drm_i915_gem_request *rq)
+{
+ dma_fence_free(&rq->fence);
+}
+#else
+static inline struct drm_i915_gem_request *
+__request_alloc(struct drm_i915_private *i915, gfp_t gfp)
+{
+ return kmem_cache_alloc(i915->requests, gfp);
+}
+
+static inline void
+__request_free(struct drm_i915_gem_request *rq)
+{
+ kmem_cache_free(rq->i915->requests, rq);
+}
+#endif
+
+#if IS_ENABLED(CONFIG_DRM_I915_REQUEST_DEBUG_OBJECTS)
+
+static struct debug_obj_descr i915_request_debug_descr = {
+ .name = "i915_request",
+};
+
+static inline void debug_request_init(struct drm_i915_gem_request *rq)
+{
+ debug_object_init(rq, &i915_request_debug_descr);
+}
+
+static inline void debug_request_activate(struct drm_i915_gem_request *rq)
+{
+ debug_object_activate(rq, &i915_request_debug_descr);
+}
+
+static inline void debug_request_set_state(struct drm_i915_gem_request *rq,
+ int old, int new)
+{
+ debug_object_active_state(rq, &i915_request_debug_descr, old, new);
+}
+
+static inline void debug_request_deactivate(struct drm_i915_gem_request *rq)
+{
+ debug_object_deactivate(rq, &i915_request_debug_descr);
+}
+
+static inline void debug_request_destroy(struct drm_i915_gem_request *rq)
+{
+ debug_object_destroy(rq, &i915_request_debug_descr);
+}
+
+static inline void debug_request_free(struct drm_i915_gem_request *rq)
+{
+ debug_object_free(rq, &i915_request_debug_descr);
+ smp_wmb(); /* flush the change in state before reallocation */
+}
+
+static inline void debug_request_assert(struct drm_i915_gem_request *rq)
+{
+ debug_object_assert_init(rq, &i915_request_debug_descr);
+}
+
+#else
+
+static inline void debug_request_init(struct drm_i915_gem_request *rq)
+{
+}
+
+static inline void debug_request_activate(struct drm_i915_gem_request *rq)
+{
+}
+
+static inline void debug_request_set_state(struct drm_i915_gem_request *rq,
+ int old, int new)
+{
+}
+
+static inline void debug_request_deactivate(struct drm_i915_gem_request *rq)
+{
+}
+
+static inline void debug_request_destroy(struct drm_i915_gem_request *rq)
+{
+}
+
+static inline void debug_request_free(struct drm_i915_gem_request *rq)
+{
+}
+
+static inline void debug_request_assert(struct drm_i915_gem_request *rq)
+{
+}
+
+#endif
+
static const char *i915_fence_get_driver_name(struct dma_fence *fence)
{
+ debug_request_assert(to_request(fence));
return "i915";
}
static const char *i915_fence_get_timeline_name(struct dma_fence *fence)
{
- /* The timeline struct (as part of the ppgtt underneath a context)
+ debug_request_assert(to_request(fence));
+
+ /*
+ * The timeline struct (as part of the ppgtt underneath a context)
* may be freed when the request is no longer in use by the GPU.
* We could extend the life of a context to beyond that of all
* fences, possibly keeping the hw resource around indefinitely,
@@ -53,11 +167,14 @@ static const char *i915_fence_get_timeline_name(struct dma_fence *fence)
static bool i915_fence_signaled(struct dma_fence *fence)
{
+ debug_request_assert(to_request(fence));
return i915_gem_request_completed(to_request(fence));
}
static bool i915_fence_enable_signaling(struct dma_fence *fence)
{
+ debug_request_assert(to_request(fence));
+
if (i915_fence_signaled(fence))
return false;
@@ -69,6 +186,8 @@ static signed long i915_fence_wait(struct dma_fence *fence,
bool interruptible,
signed long timeout)
{
+ debug_request_assert(to_request(fence));
+
return i915_wait_request(to_request(fence), interruptible, timeout);
}
@@ -76,7 +195,12 @@ static void i915_fence_release(struct dma_fence *fence)
{
struct drm_i915_gem_request *req = to_request(fence);
- /* The request is put onto a RCU freelist (i.e. the address
+ debug_request_set_state(req,
+ DEBUG_REQUEST_RETIRED, DEBUG_REQUEST_INACTIVE);
+ debug_request_deactivate(req);
+
+ /*
+ * The request is put onto a RCU freelist (i.e. the address
* is immediately reused), mark the fences as being freed now.
* Otherwise the debugobjects for the fences are only marked as
* freed when the slab cache itself is freed, and so we would get
@@ -84,7 +208,9 @@ static void i915_fence_release(struct dma_fence *fence)
*/
i915_sw_fence_fini(&req->submit);
- kmem_cache_free(req->i915->requests, req);
+ debug_request_free(req);
+
+ __request_free(req);
}
const struct dma_fence_ops i915_fence_ops = {
@@ -387,6 +513,8 @@ static void i915_gem_request_retire(struct drm_i915_gem_request *request)
GEM_BUG_ON(!request->i915->gt.active_requests);
trace_i915_gem_request_retire(request);
+ debug_request_set_state(request,
+ DEBUG_REQUEST_EXECUTING, DEBUG_REQUEST_RETIRED);
spin_lock_irq(&engine->timeline->lock);
list_del_init(&request->link);
@@ -484,6 +612,8 @@ void __i915_gem_request_submit(struct drm_i915_gem_request *request)
lockdep_assert_held(&engine->timeline->lock);
trace_i915_gem_request_execute(request);
+ debug_request_set_state(request,
+ DEBUG_REQUEST_READY, DEBUG_REQUEST_EXECUTING);
/* Transfer from per-context onto the global per-engine timeline */
timeline = engine->timeline;
@@ -532,6 +662,9 @@ void __i915_gem_request_unsubmit(struct drm_i915_gem_request *request)
GEM_BUG_ON(!irqs_disabled());
lockdep_assert_held(&engine->timeline->lock);
+ debug_request_set_state(request,
+ DEBUG_REQUEST_EXECUTING, DEBUG_REQUEST_READY);
+
/* Only unwind in reverse order, required so that the per-context list
* is kept in seqno/ring order.
*/
@@ -584,6 +717,9 @@ submit_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)
switch (state) {
case FENCE_COMPLETE:
trace_i915_gem_request_submit(request);
+ debug_request_set_state(request,
+ DEBUG_REQUEST_WAITING,
+ DEBUG_REQUEST_READY);
/*
* We need to serialize use of the submit_request() callback with its
* hotplugging performed during an emergency i915_gem_set_wedged().
@@ -688,8 +824,8 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
*
* Do not use kmem_cache_zalloc() here!
*/
- req = kmem_cache_alloc(dev_priv->requests,
- GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN);
+ req = __request_alloc(dev_priv,
+ GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN);
if (unlikely(!req)) {
/* Ratelimit ourselves to prevent oom from malicious clients */
ret = i915_gem_wait_for_idle(dev_priv,
@@ -709,13 +845,15 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
kmem_cache_shrink(dev_priv->requests);
rcu_barrier(); /* Recover the TYPESAFE_BY_RCU pages */
- req = kmem_cache_alloc(dev_priv->requests, GFP_KERNEL);
+ req = __request_alloc(dev_priv, GFP_KERNEL);
if (!req) {
ret = -ENOMEM;
goto err_unreserve;
}
}
+ debug_request_init(req);
+
req->timeline = i915_gem_context_lookup_timeline(ctx, engine);
GEM_BUG_ON(req->timeline == engine->timeline);
@@ -784,7 +922,7 @@ i915_gem_request_alloc(struct intel_engine_cs *engine,
GEM_BUG_ON(!list_empty(&req->priotree.signalers_list));
GEM_BUG_ON(!list_empty(&req->priotree.waiters_list));
- kmem_cache_free(dev_priv->requests, req);
+ __request_free(req);
err_unreserve:
unreserve_engine(engine);
err_unpin:
@@ -1063,6 +1201,10 @@ void __i915_add_request(struct drm_i915_gem_request *request, bool flush_caches)
engine->schedule(request, request->ctx->priority);
local_bh_disable();
+ debug_request_activate(request);
+ debug_request_set_state(request,
+ DEBUG_REQUEST_INACTIVE,
+ DEBUG_REQUEST_WAITING);
i915_sw_fence_commit(&request->submit);
local_bh_enable(); /* Kick the execlists tasklet if just scheduled */
}
@@ -1204,6 +1346,7 @@ long i915_wait_request(struct drm_i915_gem_request *req,
!!(flags & I915_WAIT_LOCKED));
#endif
GEM_BUG_ON(timeout < 0);
+ debug_request_assert(req);
if (i915_gem_request_completed(req))
return timeout;
@@ -1319,6 +1462,7 @@ long i915_wait_request(struct drm_i915_gem_request *req,
remove_wait_queue(errq, &reset);
remove_wait_queue(&req->execute, &exec);
trace_i915_gem_request_wait_end(req);
+ debug_request_assert(req);
return timeout;
}
--
2.15.1
More information about the Intel-gfx
mailing list