Re: [Intel-gfx] [PATCH v3 2/4] drm/i915: No TLB invalidation on wedged or suspended GT

[Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>]

On 02/10/2023 18:24, Jonathan Cavitt wrote:
> From: Fei Yang <fei.yang@intel.com>
> 
> In case of GT is suspended or wedged, 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@intel.com>
> Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
> CC: John Harrison <john.c.harrison@intel.com>
> ---
>   drivers/gpu/drm/i915/gt/uc/intel_guc.h        |  1 +
>   .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 35 +++++++++++++++----
>   drivers/gpu/drm/i915/i915_driver.c            |  9 +++++
>   3 files changed, 39 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
> index 5fc5e67f870cc..0cdc7ca66861c 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 3478fa73180ab..2f194cadbe553 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> @@ -32,6 +32,7 @@
>   
>   #include "i915_drv.h"
>   #include "i915_reg.h"
> +#include "i915_irq.h"
>   #include "i915_trace.h"
>   
>   /**
> @@ -1803,13 +1804,20 @@ static void wake_up_tlb_invalidate(struct intel_guc_tlb_wait *wait)
>   	wake_up(&wait->wq);
>   }
>   
> -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;
> +
> +	xa_for_each(&guc->tlb_lookup, i, wait)
> +		wake_up_tlb_invalidate(wait);
> +}
> +
> +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! */
> @@ -1840,8 +1848,7 @@ void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stall
>   	 * The full GT reset will have cleared the TLB caches and flushed the
>   	 * G2H message queue; we can release all the blocked waiters.
>   	 */
> -	xa_for_each(&guc->tlb_lookup, i, wait)
> -		wake_up_tlb_invalidate(wait);
> +	wake_up_all_tlb_invalidate(guc);
>   }
>   
>   static void guc_cancel_context_requests(struct intel_context *ce)
> @@ -1937,6 +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);
>   }
>   
>   void intel_guc_submission_reset_finish(struct intel_guc *guc)
> @@ -4748,6 +4761,14 @@ static long must_wait_woken(struct wait_queue_entry *wq_entry, long timeout)
>   	return timeout;
>   }
>   
> +static bool intel_gt_is_enabled(const struct intel_gt *gt)
> +{
> +	/* Check if GT is wedged or suspended */
> +	if (intel_gt_is_wedged(gt) || !intel_irqs_enabled(gt->i915))
> +		return false;
> +	return true;
> +}

Name still sucks but at least it is now hidden, okay.

> +
>   static int guc_send_invalidate_tlb(struct intel_guc *guc, u32 type)
>   {
>   	struct intel_guc_tlb_wait _wq, *wq = &_wq;
> @@ -4765,7 +4786,8 @@ static int guc_send_invalidate_tlb(struct intel_guc *guc, u32 type)
>   	};
>   	u32 size = ARRAY_SIZE(action);
>   
> -	if (!intel_guc_ct_enabled(&guc->ct))
> +	if (!intel_guc_ct_enabled(&guc->ct) ||
> +	    !intel_gt_is_enabled(gt))
>   		return -EINVAL;
>   
>   	init_waitqueue_head(&_wq.wq);
> @@ -4807,7 +4829,8 @@ static int guc_send_invalidate_tlb(struct intel_guc *guc, u32 type)
>   	 * queued in CT buffer.
>   	 */
>   #define OUTSTANDING_GUC_TIMEOUT_PERIOD  (HZ * 2)
> -	if (!must_wait_woken(&wait, OUTSTANDING_GUC_TIMEOUT_PERIOD)) {
> +	if (!must_wait_woken(&wait, OUTSTANDING_GUC_TIMEOUT_PERIOD) &&
> +	    intel_gt_is_enabled(gt)) {

Order of conditions is okay? Makes sense to first wait and only then 
check if "gt is enabled"?

>   		gt_err(gt,
>   		       "TLB invalidation response timed out for seqno %u\n", seqno);
>   		err = -ETIME;
> diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c
> index 78501a83ba109..f5175103ea900 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/uc/intel_guc.h"
>   
>   #include "pxp/intel_pxp.h"
>   #include "pxp/intel_pxp_debugfs.h"
> @@ -1092,6 +1093,9 @@ static int i915_drm_suspend(struct drm_device *dev)
>   	intel_dp_mst_suspend(dev_priv);
>   
>   	intel_runtime_pm_disable_interrupts(dev_priv);
> +
> +	wake_up_all_tlb_invalidate(&to_gt(dev_priv)->uc.guc);
> +
>   	intel_hpd_cancel_work(dev_priv);
>   
>   	intel_suspend_encoders(dev_priv);
> @@ -1263,6 +1267,11 @@ static int i915_drm_resume(struct drm_device *dev)
>   
>   	intel_gvt_resume(dev_priv);
>   
> +	if (INTEL_GUC_SUPPORTS_TLB_INVALIDATION(&to_gt(dev_priv)->uc.guc)) {
> +		intel_guc_invalidate_tlb_full(&to_gt(dev_priv)->uc.guc);
> +		intel_guc_invalidate_tlb(&to_gt(dev_priv)->uc.guc);
> +	}
> +
>   	enable_rpm_wakeref_asserts(&dev_priv->runtime_pm);
>   
>   	return 0;

Suspend resume bits feel like should be moved into 
intel_gt_susped|resume_* in order to have proper component organisation.

If that works, it is likely those already have for_each_gt loops so 3/4 
perhaps could be redundant too.

Regards,

Tvrtko