On 12/21/2022 8:30 PM, John.C.Harrison@xxxxxxxxx wrote:
From: John Harrison <John.C.Harrison@xxxxxxxxx>
The CI results for the 'fast request' patch set (enables error return
codes for fire-and-forget H2G messages) hit an issue with the KMD
sending context submission requests on an invalid context. That was
caused by a fault injection probe failing the context creation of a
kernel context. However, there was no return code checking on any of
the kernel context registration paths. So the driver kept going and
tried to use the kernel context for the record defaults process.
Fix that by checking for errors and aborting as appropriate when
creating kernel contexts. While at it, clean up some other submission
init related failure cleanup paths. Also, rename guc_init_lrc_mapping
to guc_init_submission as the former name hasn't been valid in a long
time.
IMO we should specify that this isn't a bug in the current code. The
only way this can fail if we don't inject the failure is if the GuC is
dead and/or is not seeing our messages, in which case we'll eventually
correctly wedge. We only need this to get ready to switch to fast
request messages.
Signed-off-by: John Harrison <John.C.Harrison@xxxxxxxxx>
---
.../gpu/drm/i915/gt/uc/intel_guc_submission.c | 74 ++++++++++++++-----
.../gpu/drm/i915/gt/uc/intel_guc_submission.h | 2 +-
drivers/gpu/drm/i915/gt/uc/intel_uc.c | 7 +-
3 files changed, 62 insertions(+), 21 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
index 53f7f599cde3a..4682ec1dbd9c0 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -1431,21 +1431,25 @@ static int guc_action_enable_usage_stats(struct intel_guc *guc)
return intel_guc_send(guc, action, ARRAY_SIZE(action));
}
-static void guc_init_engine_stats(struct intel_guc *guc)
+static int guc_init_engine_stats(struct intel_guc *guc)
{
struct intel_gt *gt = guc_to_gt(guc);
intel_wakeref_t wakeref;
+ int ret;
mod_delayed_work(system_highpri_wq, &guc->timestamp.work,
guc->timestamp.ping_delay);
with_intel_runtime_pm(>->i915->runtime_pm, wakeref) {
- int ret = guc_action_enable_usage_stats(guc);
+ ret = guc_action_enable_usage_stats(guc);
- if (ret)
- drm_err(>->i915->drm,
- "Failed to enable usage stats: %d!\n", ret);
+ if (ret) {
+ cancel_delayed_work_sync(&guc->timestamp.work);
+ drm_err(>->i915->drm, "Failed to enable usage stats: %d!\n", ret);
+ }
}
+
+ return ret;
}
void intel_guc_busyness_park(struct intel_gt *gt)
@@ -4101,9 +4105,11 @@ static void guc_set_default_submission(struct intel_engine_cs *engine)
engine->submit_request = guc_submit_request;
}
-static inline void guc_kernel_context_pin(struct intel_guc *guc,
- struct intel_context *ce)
+static inline int guc_kernel_context_pin(struct intel_guc *guc,
+ struct intel_context *ce)
{
+ int ret;
+
/*
* Note: we purposefully do not check the returns below because
* the registration can only fail if a reset is just starting.
@@ -4111,16 +4117,24 @@ static inline void guc_kernel_context_pin(struct intel_guc *guc,
* isn't happening and even it did this code would be run again.
*/
- if (context_guc_id_invalid(ce))
- pin_guc_id(guc, ce);
+ if (context_guc_id_invalid(ce)) {
+ int ret = pin_guc_id(guc, ce);
+
+ if (ret < 0)
+ return ret;
+ }
if (!test_bit(CONTEXT_GUC_INIT, &ce->flags))
guc_context_init(ce);
- try_context_registration(ce, true);
+ ret = try_context_registration(ce, true);
+ if (ret)
+ unpin_guc_id(guc, ce);
+
+ return ret;
}
-static inline void guc_init_lrc_mapping(struct intel_guc *guc)
+static inline int guc_init_submission(struct intel_guc *guc)
{
struct intel_gt *gt = guc_to_gt(guc);
struct intel_engine_cs *engine;
@@ -4147,9 +4161,17 @@ static inline void guc_init_lrc_mapping(struct intel_guc *guc)
struct intel_context *ce;
list_for_each_entry(ce, &engine->pinned_contexts_list,
- pinned_contexts_link)
- guc_kernel_context_pin(guc, ce);
+ pinned_contexts_link) {
+ int ret = guc_kernel_context_pin(guc, ce);
+
+ if (ret) {
+ /* No point in trying to clean up as i915 will wedge on failure */
+ return ret;
+ }
nit: no need for {} here
+ }
}
+
+ return 0;
}
static void guc_release(struct intel_engine_cs *engine)
@@ -4391,9 +4413,10 @@ static int guc_init_global_schedule_policy(struct intel_guc *guc)
return ret;
}
-void intel_guc_submission_enable(struct intel_guc *guc)
+int intel_guc_submission_enable(struct intel_guc *guc)
{
struct intel_gt *gt = guc_to_gt(guc);
+ int ret;
/* Enable and route to GuC */
if (GRAPHICS_VER(gt->i915) >= 12)
@@ -4401,16 +4424,31 @@ void intel_guc_submission_enable(struct intel_guc *guc)
GUC_SEM_INTR_ROUTE_TO_GUC |
GUC_SEM_INTR_ENABLE_ALL);
- guc_init_lrc_mapping(guc);
- guc_init_engine_stats(guc);
- guc_init_global_schedule_policy(guc);
+ ret = guc_init_submission(guc);
+ if (ret)
+ goto fail;
+
+ ret = guc_init_engine_stats(guc);
+ if (ret)
+ goto fail;
+
+ ret = guc_init_global_schedule_policy(guc);
+ if (ret)
+ goto fail;
+
+ return 0;
+
+fail:
+ intel_guc_submission_disable(guc);
We usually avoid calling the _disable() function in the error path of
the _enable() and instead do onion unwind. AFAICS the only thing you
need to disable is the semaphore routing, so might be worth moving that
to an helper. Something like:
void guc_route_semaphores(struct intel_guc *guc, bool to_guc)
{
struct intel_gt *gt = guc_to_gt(guc);
u32 val;
if (GRAPHICS_VER(gt->i915) < 12)
return;
if (to_guc)
val = GUC_SEM_INTR_ROUTE_TO_GUC |
GUC_SEM_INTR_ENABLE_ALL;
else
val = 0;
intel_uncore_write(gt->uncore, GEN12_GUC_SEM_INTR_ENABLES, val); );
}
And then call this from the enable/disable functions.
+ return ret;
}
+/* Note: By the time we're here, GuC may have already been reset */
void intel_guc_submission_disable(struct intel_guc *guc)
{
struct intel_gt *gt = guc_to_gt(guc);
- /* Note: By the time we're here, GuC may have already been reset */
+ cancel_delayed_work_sync(&guc->timestamp.work);
is this only added here because you now call it from the enable? if so,
better to cancel the work directly as onion unwind. If we do need it for
other use-cases (error path from slpc init failure?) it might be worth
mentioning it as an extra fix in the commit message.
Daniele
/* Disable and route to host */
if (GRAPHICS_VER(gt->i915) >= 12)
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.h b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.h
index 5a95a9f0a8e31..c57b29cdb1a64 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.h
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.h
@@ -15,7 +15,7 @@ struct intel_engine_cs;
void intel_guc_submission_init_early(struct intel_guc *guc);
int intel_guc_submission_init(struct intel_guc *guc);
-void intel_guc_submission_enable(struct intel_guc *guc);
+int intel_guc_submission_enable(struct intel_guc *guc);
void intel_guc_submission_disable(struct intel_guc *guc);
void intel_guc_submission_fini(struct intel_guc *guc);
int intel_guc_preempt_work_create(struct intel_guc *guc);
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_uc.c b/drivers/gpu/drm/i915/gt/uc/intel_uc.c
index 9a8a1abf71d7f..8e338b3321a93 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_uc.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_uc.c
@@ -537,8 +537,11 @@ static int __uc_init_hw(struct intel_uc *uc)
else
intel_huc_auth(huc);
- if (intel_uc_uses_guc_submission(uc))
- intel_guc_submission_enable(guc);
+ if (intel_uc_uses_guc_submission(uc)) {
+ ret = intel_guc_submission_enable(guc);
+ if (ret)
+ goto err_log_capture;
+ }
if (intel_uc_uses_guc_slpc(uc)) {
ret = intel_guc_slpc_enable(&guc->slpc);
