[PATCH v4 1/3] drm/xe: Move LNL scheduling WA to xe_device.h

[PATCH v4 1/3] drm/xe: Move LNL scheduling WA to xe_device.h

[Nirmoy Das]

Move LNL scheduling WA to xe_device.h so this can be used in other
places without needing keep the same comment about removal of this WA
in the future. The WA, which flushes work or workqueues, is now wrapped
in macros and can be reused wherever needed.

Cc: Badal Nilawar <badal.nilawar@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
cc: <stable@vger.kernel.org> # v6.11+
Suggested-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
---
 drivers/gpu/drm/xe/xe_device.h | 14 ++++++++++++++
 drivers/gpu/drm/xe/xe_guc_ct.c | 11 +----------
 2 files changed, 15 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_device.h b/drivers/gpu/drm/xe/xe_device.h
index 4c3f0ebe78a9..f1fbfe916867 100644
--- a/drivers/gpu/drm/xe/xe_device.h
+++ b/drivers/gpu/drm/xe/xe_device.h
@@ -191,4 +191,18 @@ void xe_device_declare_wedged(struct xe_device *xe);
 struct xe_file *xe_file_get(struct xe_file *xef);
 void xe_file_put(struct xe_file *xef);
 
+/*
+ * Occasionally it is seen that the G2H worker starts running after a delay of more than
+ * a second even after being queued and activated by the Linux workqueue subsystem. This
+ * leads to G2H timeout error. The root cause of issue lies with scheduling latency of
+ * Lunarlake Hybrid CPU. Issue disappears if we disable Lunarlake atom cores from BIOS
+ * and this is beyond xe kmd.
+ *
+ * TODO: Drop this change once workqueue scheduling delay issue is fixed on LNL Hybrid CPU.
+ */
+#define LNL_FLUSH_WORKQUEUE(wq__) \
+	flush_workqueue(wq__)
+#define LNL_FLUSH_WORK(wrk__) \
+	flush_work(wrk__)
+
 #endif
diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c
index 1b5d8fb1033a..703b44b257a7 100644
--- a/drivers/gpu/drm/xe/xe_guc_ct.c
+++ b/drivers/gpu/drm/xe/xe_guc_ct.c
@@ -1018,17 +1018,8 @@ static int guc_ct_send_recv(struct xe_guc_ct *ct, const u32 *action, u32 len,
 
 	ret = wait_event_timeout(ct->g2h_fence_wq, g2h_fence.done, HZ);
 
-	/*
-	 * Occasionally it is seen that the G2H worker starts running after a delay of more than
-	 * a second even after being queued and activated by the Linux workqueue subsystem. This
-	 * leads to G2H timeout error. The root cause of issue lies with scheduling latency of
-	 * Lunarlake Hybrid CPU. Issue dissappears if we disable Lunarlake atom cores from BIOS
-	 * and this is beyond xe kmd.
-	 *
-	 * TODO: Drop this change once workqueue scheduling delay issue is fixed on LNL Hybrid CPU.
-	 */
 	if (!ret) {
-		flush_work(&ct->g2h_worker);
+		LNL_FLUSH_WORK(&ct->g2h_worker);
 		if (g2h_fence.done) {
 			xe_gt_warn(gt, "G2H fence %u, action %04x, done\n",
 				   g2h_fence.seqno, action[0]);
-- 
2.46.0

[PATCH v4 2/3] drm/xe/ufence: Flush xe ordered_wq in case of ufence timeout

[Nirmoy Das]

Flush xe ordered_wq in case of ufence timeout which is observed
on LNL and that points to recent scheduling issue with E-cores.

This is similar to the recent fix:
commit e51527233804 ("drm/xe/guc/ct: Flush g2h worker in case of g2h
response timeout") and should be removed once there is a E-core
scheduling fix for LNL.

v2: Add platform check(Himal)
    s/__flush_workqueue/flush_workqueue(Jani)
v3: Remove gfx platform check as the issue related to cpu
    platform(John)
v4: Use the Common macro(John) and print when the flush resolves
    timeout(Matt B)

Cc: Badal Nilawar <badal.nilawar@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: <stable@vger.kernel.org> # v6.11+
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2754
Suggested-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
---
 drivers/gpu/drm/xe/xe_wait_user_fence.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_wait_user_fence.c b/drivers/gpu/drm/xe/xe_wait_user_fence.c
index f5deb81eba01..5b4264ea38bd 100644
--- a/drivers/gpu/drm/xe/xe_wait_user_fence.c
+++ b/drivers/gpu/drm/xe/xe_wait_user_fence.c
@@ -155,6 +155,13 @@ int xe_wait_user_fence_ioctl(struct drm_device *dev, void *data,
 		}
 
 		if (!timeout) {
+			LNL_FLUSH_WORKQUEUE(xe->ordered_wq);
+			err = do_compare(addr, args->value, args->mask,
+					 args->op);
+			if (err <= 0) {
+				drm_dbg(&xe->drm, "LNL_FLUSH_WORKQUEUE resolved ufence timeout\n");
+				break;
+			}
 			err = -ETIME;
 			break;
 		}
-- 
2.46.0

[PATCH v4 3/3] drm/xe/guc/tlb: Flush g2h worker in case of tlb timeout

[Nirmoy Das]

Flush the g2h worker explicitly if TLB timeout happens which is
observed on LNL and that points to the recent scheduling issue with
E-cores on LNL.

This is similar to the recent fix:
commit e51527233804 ("drm/xe/guc/ct: Flush g2h worker in case of g2h
response timeout") and should be removed once there is E core
scheduling fix.

v2: Add platform check(Himal)
v3: Remove gfx platform check as the issue related to cpu
    platform(John)
    Use the common WA macro(John) and print when the flush
    resolves timeout(Matt B)

Cc: Badal Nilawar <badal.nilawar@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: <stable@vger.kernel.org> # v6.11+
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2687
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
---
 drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
index 773de1f08db9..0bdb3ba5220a 100644
--- a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
@@ -81,6 +81,15 @@ static void xe_gt_tlb_fence_timeout(struct work_struct *work)
 		if (msecs_to_jiffies(since_inval_ms) < tlb_timeout_jiffies(gt))
 			break;
 
+		LNL_FLUSH_WORK(&gt->uc.guc.ct.g2h_worker);
+		since_inval_ms = ktime_ms_delta(ktime_get(),
+						fence->invalidation_time);
+		if (msecs_to_jiffies(since_inval_ms) < tlb_timeout_jiffies(gt)) {
+			xe_gt_dbg(gt, "LNL_FLUSH_WORK resolved TLB invalidation fence timeout, seqno=%d recv=%d",
+				  fence->seqno, gt->tlb_invalidation.seqno_recv);
+			break;
+		}
+
 		trace_xe_gt_tlb_invalidation_fence_timeout(xe, fence);
 		xe_gt_err(gt, "TLB invalidation fence timeout, seqno=%d recv=%d",
 			  fence->seqno, gt->tlb_invalidation.seqno_recv);
-- 
2.46.0

Re: [PATCH v4 3/3] drm/xe/guc/tlb: Flush g2h worker in case of tlb timeout

[Matthew Auld]

On 29/10/2024 09:54, Nirmoy Das wrote:
> Flush the g2h worker explicitly if TLB timeout happens which is
> observed on LNL and that points to the recent scheduling issue with
> E-cores on LNL.
> 
> This is similar to the recent fix:
> commit e51527233804 ("drm/xe/guc/ct: Flush g2h worker in case of g2h
> response timeout") and should be removed once there is E core
> scheduling fix.
> 
> v2: Add platform check(Himal)
> v3: Remove gfx platform check as the issue related to cpu
>      platform(John)
>      Use the common WA macro(John) and print when the flush
>      resolves timeout(Matt B)
> 
> Cc: Badal Nilawar <badal.nilawar@intel.com>
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Matthew Auld <matthew.auld@intel.com>
> Cc: John Harrison <John.C.Harrison@Intel.com>
> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
> Cc: Lucas De Marchi <lucas.demarchi@intel.com>
> Cc: <stable@vger.kernel.org> # v6.11+
> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2687
> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
> ---
>   drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c | 9 +++++++++
>   1 file changed, 9 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
> index 773de1f08db9..0bdb3ba5220a 100644
> --- a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
> +++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
> @@ -81,6 +81,15 @@ static void xe_gt_tlb_fence_timeout(struct work_struct *work)
>   		if (msecs_to_jiffies(since_inval_ms) < tlb_timeout_jiffies(gt))
>   			break;
>   
> +		LNL_FLUSH_WORK(&gt->uc.guc.ct.g2h_worker);

I think here we are holding the pending lock, and g2h worker also wants 
to grab that same lock so this smells like potential deadlock. Also 
flush_work can sleep so I don't think is allowed under spinlock.

> +		since_inval_ms = ktime_ms_delta(ktime_get(),
> +						fence->invalidation_time);

I think invalidation_time is rather when we sent off the invalidation 
req, and we already check that above so if we get here then we know the 
timeout has expired for this fence, so checking again after the flush 
doesn't really help AFAICT.

I think we can just move the flush to before the loop and outside the 
lock, and then if the fence(s) gets signalled they will be removed from 
the list and then also won't be considered for timeout?

> +		if (msecs_to_jiffies(since_inval_ms) < tlb_timeout_jiffies(gt)) {
> +			xe_gt_dbg(gt, "LNL_FLUSH_WORK resolved TLB invalidation fence timeout, seqno=%d recv=%d",
> +				  fence->seqno, gt->tlb_invalidation.seqno_recv);
> +			break;
> +		}
> +
>   		trace_xe_gt_tlb_invalidation_fence_timeout(xe, fence);
>   		xe_gt_err(gt, "TLB invalidation fence timeout, seqno=%d recv=%d",
>   			  fence->seqno, gt->tlb_invalidation.seqno_recv);

Re: [PATCH v4 3/3] drm/xe/guc/tlb: Flush g2h worker in case of tlb timeout

[Nirmoy Das]

On 10/29/2024 11:59 AM, Matthew Auld wrote:
> On 29/10/2024 09:54, Nirmoy Das wrote:
>> Flush the g2h worker explicitly if TLB timeout happens which is
>> observed on LNL and that points to the recent scheduling issue with
>> E-cores on LNL.
>>
>> This is similar to the recent fix:
>> commit e51527233804 ("drm/xe/guc/ct: Flush g2h worker in case of g2h
>> response timeout") and should be removed once there is E core
>> scheduling fix.
>>
>> v2: Add platform check(Himal)
>> v3: Remove gfx platform check as the issue related to cpu
>>      platform(John)
>>      Use the common WA macro(John) and print when the flush
>>      resolves timeout(Matt B)
>>
>> Cc: Badal Nilawar <badal.nilawar@intel.com>
>> Cc: Matthew Brost <matthew.brost@intel.com>
>> Cc: Matthew Auld <matthew.auld@intel.com>
>> Cc: John Harrison <John.C.Harrison@Intel.com>
>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>> Cc: Lucas De Marchi <lucas.demarchi@intel.com>
>> Cc: <stable@vger.kernel.org> # v6.11+
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2687
>> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c | 9 +++++++++
>>   1 file changed, 9 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
>> index 773de1f08db9..0bdb3ba5220a 100644
>> --- a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
>> +++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
>> @@ -81,6 +81,15 @@ static void xe_gt_tlb_fence_timeout(struct work_struct *work)
>>           if (msecs_to_jiffies(since_inval_ms) < tlb_timeout_jiffies(gt))
>>               break;
>>   +        LNL_FLUSH_WORK(&gt->uc.guc.ct.g2h_worker);
>
> I think here we are holding the pending lock, and g2h worker also wants to grab that same lock so this smells like potential deadlock. Also flush_work can sleep so I don't think is allowed under spinlock.
Ah yes. It  ran BAT tests without any warning so I  didn't think much about this change :/

>
>> +        since_inval_ms = ktime_ms_delta(ktime_get(),
>> +                        fence->invalidation_time);
>
> I think invalidation_time is rather when we sent off the invalidation req, and we already check that above so if we get here then we know the timeout has expired for this fence, so checking again after the flush doesn't really help AFAICT.
>
> I think we can just move the flush to before the loop and outside the lock, and then if the fence(s) gets signalled they will be removed from the list and then also won't be considered for timeout?


I put the flush inside to get a log when it resolves a timeout. I will revert it and just do flush and not print anything.


Regards,

Nirmoy

>
>> +        if (msecs_to_jiffies(since_inval_ms) < tlb_timeout_jiffies(gt)) {
>> +            xe_gt_dbg(gt, "LNL_FLUSH_WORK resolved TLB invalidation fence timeout, seqno=%d recv=%d",
>> +                  fence->seqno, gt->tlb_invalidation.seqno_recv);
>> +            break;
>> +        }
>> +
>>           trace_xe_gt_tlb_invalidation_fence_timeout(xe, fence);
>>           xe_gt_err(gt, "TLB invalidation fence timeout, seqno=%d recv=%d",
>>                 fence->seqno, gt->tlb_invalidation.seqno_recv);