[Intel-gfx] [PATCH v8 4/7] drm/i915: No TLB invalidation on suspended GT

Mon Oct 9 09:31:32 UTC 2023

On Mon, 09 Oct 2023, Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com> wrote:
> On 06/10/2023 19:20, Jonathan Cavitt wrote:
>> In case of GT is suspended, don't allow submission of new TLB invalidation
>> request and cancel all pending requests. The TLB entries will be
>> invalidated either during GuC reload or on system resume.
>> 
>> Signed-off-by: Fei Yang <fei.yang at intel.com>
>> Signed-off-by: Jonathan Cavitt <jonathan.cavitt at intel.com>
>> CC: John Harrison <john.c.harrison at intel.com>
>> ---
>>   drivers/gpu/drm/i915/gt/intel_gt.h            |  1 +
>>   drivers/gpu/drm/i915/gt/uc/intel_guc.h        |  1 +
>>   .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 23 +++++++++++++++----
>>   drivers/gpu/drm/i915/gt/uc/intel_uc.c         |  8 +++++++
>>   drivers/gpu/drm/i915/i915_driver.c            |  2 ++
>>   5 files changed, 31 insertions(+), 4 deletions(-)
>> 
>> diff --git a/drivers/gpu/drm/i915/gt/intel_gt.h b/drivers/gpu/drm/i915/gt/intel_gt.h
>> index 970bedf6b78a7..4e3bb221d2f4d 100644
>> --- a/drivers/gpu/drm/i915/gt/intel_gt.h
>> +++ b/drivers/gpu/drm/i915/gt/intel_gt.h
>> @@ -179,4 +179,5 @@ enum i915_map_type intel_gt_coherent_map_type(struct intel_gt *gt,
>>   void intel_gt_bind_context_set_ready(struct intel_gt *gt);
>>   void intel_gt_bind_context_set_unready(struct intel_gt *gt);
>>   bool intel_gt_is_bind_context_ready(struct intel_gt *gt);
>> +
>
> Zap this.
>
>>   #endif /* __INTEL_GT_H__ */
>> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
>> index 88deb43bbdc48..add74000e621e 100644
>> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc.h
>> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
>> @@ -536,4 +536,5 @@ void intel_guc_dump_time_info(struct intel_guc *guc, struct drm_printer *p);
>>   
>>   int intel_guc_sched_disable_gucid_threshold_max(struct intel_guc *guc);
>>   
>> +void wake_up_all_tlb_invalidate(struct intel_guc *guc);
>>   #endif
>> 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 0c5ccda1b3e87..4ed6e9e759007 100644
>> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
>> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
>> @@ -1796,13 +1796,23 @@ static void __guc_reset_context(struct intel_context *ce, intel_engine_mask_t st
>>   	intel_context_put(parent);
>>   }
>>   
>> -void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stalled)
>> +void wake_up_all_tlb_invalidate(struct intel_guc *guc)
>>   {
>>   	struct intel_guc_tlb_wait *wait;
>> +	unsigned long i;
>> +
>> +	if (!HAS_GUC_TLB_INVALIDATION(guc_to_gt(guc)->i915))
>> +		return;
>> +	xa_for_each(&guc->tlb_lookup, i, wait) {
>> +		wake_up(&wait->wq);
>> +	}
>
> Drop the curlies?
>
>> +}
>> +
>> +void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stalled)
>> +{
>>   	struct intel_context *ce;
>>   	unsigned long index;
>>   	unsigned long flags;
>> -	unsigned long i;
>>   
>>   	if (unlikely(!guc_submission_initialized(guc))) {
>>   		/* Reset called during driver load? GuC not yet initialised! */
>> @@ -1838,8 +1848,7 @@ void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stall
>>   	 * invalidations on GT reset, and there's a large window of time
>>   	 * between the GT reset and GuC becoming available.
>>   	 */
>> -	xa_for_each(&guc->tlb_lookup, i, wait)
>> -		wake_up(&wait->wq);
>> +	wake_up_all_tlb_invalidate(guc);
>>   }
>>   
>>   static void guc_cancel_context_requests(struct intel_context *ce)
>> @@ -1935,6 +1944,12 @@ void intel_guc_submission_cancel_requests(struct intel_guc *guc)
>>   
>>   	/* GuC is blown away, drop all references to contexts */
>>   	xa_destroy(&guc->context_lookup);
>> +
>> +	/*
>> +	 * Wedged GT won't respond to any TLB invalidation request. Simply
>> +	 * release all the blocked waiters.
>> +	 */
>> +	wake_up_all_tlb_invalidate(guc);
>
> This goes to the following patch I think.
>
>>   }
>>   
>>   void intel_guc_submission_reset_finish(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 98b103375b7ab..8d6a4d8ce61bb 100644
>> --- a/drivers/gpu/drm/i915/gt/uc/intel_uc.c
>> +++ b/drivers/gpu/drm/i915/gt/uc/intel_uc.c
>> @@ -688,6 +688,9 @@ void intel_uc_suspend(struct intel_uc *uc)
>>   	/* flush the GSC worker */
>>   	intel_gsc_uc_flush_work(&uc->gsc);
>>   
>> +	if (HAS_GUC_TLB_INVALIDATION(guc_to_gt(guc)->i915))
>> +		wake_up_all_tlb_invalidate(guc);
>> +
>>   	if (!intel_guc_is_ready(guc)) {
>>   		guc->interrupts.enabled = false;
>>   		return;
>> @@ -736,6 +739,11 @@ static int __uc_resume(struct intel_uc *uc, bool enable_communication)
>>   
>>   	intel_gsc_uc_resume(&uc->gsc);
>>   
>> +	if (HAS_GUC_TLB_INVALIDATION(gt->i915)) {
>> +		intel_guc_invalidate_tlb_engines(guc);
>> +		intel_guc_invalidate_tlb_guc(guc);
>> +	}
>> +
>>   	return 0;
>>   }
>>   
>> diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c
>> index ccbb2834cde07..85ac9400c2dc7 100644
>> --- a/drivers/gpu/drm/i915/i915_driver.c
>> +++ b/drivers/gpu/drm/i915/i915_driver.c
>> @@ -72,6 +72,7 @@
>>   #include "gt/intel_gt.h"
>>   #include "gt/intel_gt_pm.h"
>>   #include "gt/intel_rc6.h"
>> +#include "gt/intel_tlb.h"

Does not belong in this patch.

>>   
>>   #include "pxp/intel_pxp.h"
>>   #include "pxp/intel_pxp_debugfs.h"
>> @@ -1093,6 +1094,7 @@ static int i915_drm_suspend(struct drm_device *dev)
>>   	intel_dp_mst_suspend(dev_priv);
>>   
>>   	intel_runtime_pm_disable_interrupts(dev_priv);
>> +
>
> Zap this too please.
>
>>   	intel_hpd_cancel_work(dev_priv);
>>   
>>   	intel_suspend_encoders(dev_priv);
>
> Patch got smaller and will be smaller still which is neat! If it is 
> still functionally correct it is a win-win.
>
> Regards,
>
> Tvrtko

-- 
Jani Nikula, Intel