Skip to contentSkip to conversation

drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL

RFC PATCH Superseded Tales A. Mendonça 6 participants 30 messages Aug 4 – Aug 13

Subsystems (2)

v1 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

cover drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor buffer unmaps etc.) hitches for the full window.

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no exec queue or job to blame - leaves a devcoredump with the GuC log and CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into a sub-500ms hiccup for the majority of occurrences; a minority of severe episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
 drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 131 +++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
 4 files changed, 243 insertions(+), 4 deletions(-)
Signature · 1 line
2.55.0
#1 lore raw

RFC PATCH 1/3 drm/xe: Capture devcoredump on TLB invalidation timeout

TLB invalidation timeouts currently leave no record of the firmware state behind: there is no exec queue or job to blame, so nothing calls xe_devcoredump() and the GuC log content at the time of the hang is lost.

Add xe_devcoredump_gt(), a variant of xe_devcoredump() for hangs that are not tied to an exec queue or job. It captures the GuC log and CT state of the affected GT, reusing the existing snapshot machinery and the "only first snapshot" policy, and hook it up to the TLB invalidation timeout path.

This was instrumental in diagnosing GuC TLB invalidation ack stalls on ARL (see Link), where the invalidation request is consumed from the H2G CTB immediately but the ack G2H only arrives ~2.3s later, after the timeout has already fired.

---
 drivers/gpu/drm/xe/xe_devcoredump.c | 68 +++++++++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_devcoredump.h |  6 +++
 drivers/gpu/drm/xe/xe_tlb_inval.c   | 20 +++++++++
 3 files changed, 94 insertions(+)
drivers/gpu/drm/xe/xe_devcoredump.c +68 -0
index 5f2b90b18f9..0ccaed176a4 100644
@@ -403,6 +403,74 @@ void xe_devcoredump(struct xe_exec_queue *q, struct xe_sched_job *job, const cha
 	mutex_unlock(&coredump->lock);
 }
 
+static void devcoredump_snapshot_gt(struct xe_devcoredump *coredump,
+				    struct xe_gt *gt)
+{
+	struct xe_devcoredump_snapshot *ss = &coredump->snapshot;
+	struct xe_guc *guc = &gt->uc.guc;
+	bool cookie;
+
+	ss->snapshot_time = ktime_get_real();
+	ss->boot_time = ktime_get_boottime();
+
+	strscpy(ss->process_name, "no process");
+
+	ss->gt = gt;
+	INIT_WORK(&ss->work, xe_devcoredump_deferred_snap_work);
+
+	/* keep going if fw fails as we still want to save the SW data */
+	CLASS(xe_force_wake, fw_ref)(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+
+	cookie = dma_fence_begin_signalling();
+
+	ss->guc.log = xe_guc_log_snapshot_capture(&guc->log, true);
+	ss->guc.ct = xe_guc_ct_snapshot_capture(&guc->ct);
+
+	queue_work(system_dfl_wq, &ss->work);
+
+	dma_fence_end_signalling(cookie);
+}
+
+/**
+ * xe_devcoredump_gt - Take GT-level snapshots and initialize coredump device.
+ * @gt: The GT where the issue was detected.
+ * @fmt: Printf format + args to describe the reason for the core dump
+ *
+ * Variant of xe_devcoredump() for hangs that are not tied to an exec queue
+ * or job, e.g. TLB invalidation timeouts. Captures the GuC log and CT state
+ * of @gt so the firmware side of the hang can be inspected. Skipped if a
+ * coredump is already captured, same as xe_devcoredump().
+ */
+__printf(2, 3)
+void xe_devcoredump_gt(struct xe_gt *gt, const char *fmt, ...)
+{
+	struct xe_device *xe = gt_to_xe(gt);
+	struct xe_devcoredump *coredump = &xe->devcoredump;
+	va_list varg;
+
+	mutex_lock(&coredump->lock);
+
+	if (coredump->captured) {
+		drm_dbg(&xe->drm, "Multiple hangs are occurring, but only the first snapshot was taken\n");
+		mutex_unlock(&coredump->lock);
+		return;
+	}
+
+	coredump->captured = true;
+
+	va_start(varg, fmt);
+	coredump->snapshot.reason = kvasprintf(GFP_ATOMIC, fmt, varg);
+	va_end(varg);
+
+	devcoredump_snapshot_gt(coredump, gt);
+
+	drm_info(&xe->drm, "Xe device coredump has been created\n");
+	drm_info(&xe->drm, "Check your /sys/class/drm/card%d/device/devcoredump/data\n",
+		 xe->drm.primary->index);
+
+	mutex_unlock(&coredump->lock);
+}
+
 static void xe_driver_devcoredump_fini(void *arg)
 {
 	struct drm_device *drm = arg;
drivers/gpu/drm/xe/xe_devcoredump.h +6 -0
index 5391a80a4d1..f071bd11f24 100644
@@ -11,10 +11,12 @@
 struct drm_printer;
 struct xe_device;
 struct xe_exec_queue;
+struct xe_gt;
 struct xe_sched_job;
 
 #ifdef CONFIG_DEV_COREDUMP
 void xe_devcoredump(struct xe_exec_queue *q, struct xe_sched_job *job, const char *fmt, ...);
+void xe_devcoredump_gt(struct xe_gt *gt, const char *fmt, ...);
 int xe_devcoredump_init(struct xe_device *xe);
 #else
 static inline void xe_devcoredump(struct xe_exec_queue *q,
@@ -23,6 +25,10 @@ static inline void xe_devcoredump(struct xe_exec_queue *q,
 {
 }
 
+static inline void xe_devcoredump_gt(struct xe_gt *gt, const char *fmt, ...)
+{
+}
+
 static inline int xe_devcoredump_init(struct xe_device *xe)
 {
 	return 0;
drivers/gpu/drm/xe/xe_tlb_inval.c +20 -0
index bbd21d39306..833fb92cd3e 100644
@@ -5,6 +5,7 @@
 
 #include <drm/drm_managed.h>
 
+#include "xe_devcoredump.h"
 #include "xe_device_types.h"
 #include "xe_force_wake.h"
 #include "xe_gt_stats.h"
@@ -29,6 +30,12 @@
 
 #define FENCE_STACK_BIT		DMA_FENCE_FLAG_USER_BITS
 
+/* The frontend is only ever embedded in a GT */
+static struct xe_gt *tlb_inval_to_gt(struct xe_tlb_inval *tlb_inval)
+{
+	return container_of(tlb_inval, struct xe_gt, tlb_inval);
+}
+
 static void xe_tlb_inval_fence_fini(struct xe_tlb_inval_fence *fence)
 {
 	if (WARN_ON_ONCE(!fence->tlb_inval))
@@ -73,6 +80,7 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 	struct xe_device *xe = tlb_inval->xe;
 	struct xe_tlb_inval_fence *fence, *next;
 	long timeout_delay = tlb_inval->ops->timeout_delay(tlb_inval);
+	int timedout_seqno = 0;
 
 	tlb_inval->ops->flush(tlb_inval);
 
@@ -90,6 +98,8 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 			"TLB invalidation fence timeout, seqno=%d recv=%d",
 			fence->seqno, tlb_inval->seqno_recv);
 
+		timedout_seqno = fence->seqno;
+
 		fence->base.error = -ETIME;
 		xe_tlb_inval_fence_signal(fence);
 	}
@@ -97,6 +107,16 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 		queue_delayed_work(tlb_inval->timeout_wq, &tlb_inval->fence_tdr,
 				   timeout_delay);
 	spin_unlock_irq(&tlb_inval->pending_lock);
+
+	/*
+	 * Capture the GuC log and CT state so the firmware side of the hang
+	 * can be inspected; there is no queue or job to blame here. Must be
+	 * outside pending_lock as the capture takes sleeping locks.
+	 */
+	if (timedout_seqno)
+		xe_devcoredump_gt(tlb_inval_to_gt(tlb_inval),
+				  "TLB invalidation fence timeout, seqno=%d recv=%d",
+				  timedout_seqno, tlb_inval->seqno_recv);
 }
 
 /**
Signature · 1 line
2.55.0
#2 lore raw

Re: [RFC PATCH 1/3] drm/xe: Capture devcoredump on TLB invalidation timeout

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Pre-existing issues:
- [High] Missing synchronization in xe_tlb_inval_reset() during device teardown leaves the fence_tdr worker running after xe_device is freed, leading to Use-After-Free.
- [High] Devcoredump snapshot worker ss->work is not explicitly canceled during device teardown if it hasn't run yet, leading to Use-After-Free. --

--- Patch [1]: [RFC PATCH 1/3] drm/xe: Capture devcoredump on TLB invalidation timeout --- commit cb8081669b45d5d264497cf1169cfc3a90d4aadd
Author: Tales A. Mendonça <[email protected]>

drm/xe: Capture devcoredump on TLB invalidation timeout

Adds a new xe_devcoredump_gt variant to capture GuC logs and CT state for GT-level hangs, particularly hooking it into the TLB invalidation timeout path.

Quoted 3 lines
drivers/gpu/drm/xe/xe_devcoredump.c +0 -0

[ ... ]

Quoted 19 lines
drivers/gpu/drm/xe/xe_devcoredump.c +15 -0
@@ -403,6 +403,74 @@ void xe_devcoredump(struct xe_exec_queue *q, struct xe_sched_job *job, const cha
 	mutex_unlock(&coredump->lock);
 }
 
+static void devcoredump_snapshot_gt(struct xe_devcoredump *coredump,
+				    struct xe_gt *gt)
+{
+	struct xe_devcoredump_snapshot *ss = &coredump->snapshot;
+	struct xe_guc *guc = &gt->uc.guc;
+	bool cookie;
+
+	ss->snapshot_time = ktime_get_real();
+	ss->boot_time = ktime_get_boottime();
+
+	strscpy(ss->process_name, "no process");
+
+	ss->gt = gt;
+	INIT_WORK(&ss->work, xe_devcoredump_deferred_snap_work);
+

drivers/gpu/drm/xe/xe_devcoredump.c void xe_devcoredump(struct xe_exec_queue *q, struct xe_sched_job *job, const cha · line 430

Quoted 10 lines
drivers/gpu/drm/xe/xe_devcoredump.c +10 -0
+	/* keep going if fw fails as we still want to save the SW data */
+	CLASS(xe_force_wake, fw_ref)(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+
+	cookie = dma_fence_begin_signalling();
+
+	ss->guc.log = xe_guc_log_snapshot_capture(&guc->log, true);
+	ss->guc.ct = xe_guc_ct_snapshot_capture(&guc->ct);
+
+	queue_work(system_dfl_wq, &ss->work);
+

[Severity: High]
This is a pre-existing issue, but is it possible for this deferred work to cause a use-after-free if the device is unbound before the worker runs?

If the device is unbound before xe_devcoredump_deferred_snap_work() executes, xe_driver_devcoredump_fini() calls dev_coredump_put(). However, since the devcoredump device hasn't been registered yet, dev_coredump_put() does nothing and xe_devcoredump_free() is not called:

drivers/gpu/drm/xe/xe_devcoredump.c:xe_driver_devcoredump_fini() {
        ...
        dev_coredump_put(drm->dev);
}

Since there is no explicit cancel_work_sync(&coredump->snapshot.work) when the core dump hasn't been created, does this leave the worker queued on system_dfl_wq to eventually wake up and dereference freed device memory?

Quoted 3 lines
drivers/gpu/drm/xe/xe_devcoredump.c +3 -0
+	dma_fence_end_signalling(cookie);
+}
+

[ ... ]

Quoted 3 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +0 -0

[ ... ]

Quoted 4 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +0 -0
@@ -97,6 +107,16 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 		queue_delayed_work(tlb_inval->timeout_wq, &tlb_inval->fence_tdr,
 				   timeout_delay);
 	spin_unlock_irq(&tlb_inval->pending_lock);

drivers/gpu/drm/xe/xe_tlb_inval.c static void xe_tlb_inval_fence_timeout(struct work_struct *work) · line 119

Quoted 10 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +10 -0
+
+	/*
+	 * Capture the GuC log and CT state so the firmware side of the hang
+	 * can be inspected; there is no queue or job to blame here. Must be
+	 * outside pending_lock as the capture takes sleeping locks.
+	 */
+	if (timedout_seqno)
+		xe_devcoredump_gt(tlb_inval_to_gt(tlb_inval),
+				  "TLB invalidation fence timeout, seqno=%d recv=%d",
+				  timedout_seqno, tlb_inval->seqno_recv);

[Severity: High]
This is a pre-existing issue, but can this lead to a use-after-free during device teardown?

Looking at xe_tlb_inval_reset() in drivers/gpu/drm/xe/xe_tlb_inval.c, it cancels the timeout worker using cancel_delayed_work() rather than cancel_delayed_work_sync():

xe_tlb_inval_reset() {
        ...
        cancel_delayed_work(&tlb_inval->fence_tdr);
        ...
}

Since the cancellation is not synchronous and the worker is scheduled on system_wq, the teardown can proceed and free the device memory while xe_tlb_inval_fence_timeout() is still running.

If the worker drops pending_lock and executes this new call to xe_devcoredump_gt(), won't it access the already freed tlb_inval outside of the lock?

Quoted 1 line
drivers/gpu/drm/xe/xe_tlb_inval.c +0 -0
 }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
maintainer
#3 lore raw

Re: [RFC PATCH 1/3] drm/xe: Capture devcoredump on TLB invalidation timeout

In reply to Tales A. Mendonça
Quoted 11 lines · Tales A. Mendonça

TLB invalidation timeouts currently leave no record of the firmware state behind: there is no exec queue or job to blame, so nothing calls xe_devcoredump() and the GuC log content at the time of the hang is lost.

Add xe_devcoredump_gt(), a variant of xe_devcoredump() for hangs that are not tied to an exec queue or job. It captures the GuC log and CT state of the affected GT, reusing the existing snapshot machinery and the "only first snapshot" policy, and hook it up to the TLB invalidation timeout path.

Quoted 5 lines

This was instrumental in diagnosing GuC TLB invalidation ack stalls on ARL (see Link), where the invalidation request is consumed from the H2G CTB immediately but the ack G2H only arrives ~2.3s later, after the timeout has already fired.

Thanks for doing this. A couple suggestions.

Quoted 46 lines
---
 drivers/gpu/drm/xe/xe_devcoredump.c | 68 +++++++++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_devcoredump.h |  6 +++
 drivers/gpu/drm/xe/xe_tlb_inval.c   | 20 +++++++++
 3 files changed, 94 insertions(+)
drivers/gpu/drm/xe/xe_devcoredump.c +30 -0
index 5f2b90b18f9..0ccaed176a4 100644
@@ -403,6 +403,74 @@ void xe_devcoredump(struct xe_exec_queue *q, struct xe_sched_job *job, const cha
 	mutex_unlock(&coredump->lock);
 }
 
+static void devcoredump_snapshot_gt(struct xe_devcoredump *coredump,
+				    struct xe_gt *gt)
+{
+	struct xe_devcoredump_snapshot *ss = &coredump->snapshot;
+	struct xe_guc *guc = &gt->uc.guc;
+	bool cookie;
+
+	ss->snapshot_time = ktime_get_real();
+	ss->boot_time = ktime_get_boottime();
+
+	strscpy(ss->process_name, "no process");
+
+	ss->gt = gt;
+	INIT_WORK(&ss->work, xe_devcoredump_deferred_snap_work);
+
+	/* keep going if fw fails as we still want to save the SW data */
+	CLASS(xe_force_wake, fw_ref)(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+
+	cookie = dma_fence_begin_signalling();
+
+	ss->guc.log = xe_guc_log_snapshot_capture(&guc->log, true);
+	ss->guc.ct = xe_guc_ct_snapshot_capture(&guc->ct);
+
+	queue_work(system_dfl_wq, &ss->work);
+
+	dma_fence_end_signalling(cookie);
+}
+
+/**
+ * xe_devcoredump_gt - Take GT-level snapshots and initialize coredump device.

drivers/gpu/drm/xe/xe_devcoredump.c void xe_devcoredump(struct xe_exec_queue *q, struct xe_sched_job *job, const cha · line 445

Quoted 10 lines
drivers/gpu/drm/xe/xe_devcoredump.c +10 -0
+ * @gt: The GT where the issue was detected.
+ * @fmt: Printf format + args to describe the reason for the core dump
+ *
+ * Variant of xe_devcoredump() for hangs that are not tied to an exec queue
+ * or job, e.g. TLB invalidation timeouts. Captures the GuC log and CT state
+ * of @gt so the firmware side of the hang can be inspected. Skipped if a
+ * coredump is already captured, same as xe_devcoredump().
+ */
+__printf(2, 3)
+void xe_devcoredump_gt(struct xe_gt *gt, const char *fmt, ...)

I'd unify these functions with the existing xe_devcoredump/devcoredump_snapshot by adding a GT argument to both and teaching those functions that 'q' can be NULL. Then replace s/xe_devcoredump/__xe_devcoredump/ and add wrapper macros xe_devcoredump and xe_devcoredump_gt in the header file.

More below.

Quoted 36 lines
drivers/gpu/drm/xe/xe_devcoredump.c +28 -0
+{
+	struct xe_device *xe = gt_to_xe(gt);
+	struct xe_devcoredump *coredump = &xe->devcoredump;
+	va_list varg;
+
+	mutex_lock(&coredump->lock);
+
+	if (coredump->captured) {
+		drm_dbg(&xe->drm, "Multiple hangs are occurring, but only the first snapshot was taken\n");
+		mutex_unlock(&coredump->lock);
+		return;
+	}
+
+	coredump->captured = true;
+
+	va_start(varg, fmt);
+	coredump->snapshot.reason = kvasprintf(GFP_ATOMIC, fmt, varg);
+	va_end(varg);
+
+	devcoredump_snapshot_gt(coredump, gt);
+
+	drm_info(&xe->drm, "Xe device coredump has been created\n");
+	drm_info(&xe->drm, "Check your /sys/class/drm/card%d/device/devcoredump/data\n",
+		 xe->drm.primary->index);
+
+	mutex_unlock(&coredump->lock);
+}
+
 static void xe_driver_devcoredump_fini(void *arg)
 {
 	struct drm_device *drm = arg;
drivers/gpu/drm/xe/xe_devcoredump.h +0 -0
index 5391a80a4d1..f071bd11f24 100644

drivers/gpu/drm/xe/xe_devcoredump.h line 19

Quoted 9 lines
drivers/gpu/drm/xe/xe_devcoredump.h +2 -0
@@ -11,10 +11,12 @@
 struct drm_printer;
 struct xe_device;
 struct xe_exec_queue;
+struct xe_gt;
 struct xe_sched_job;
 
 #ifdef CONFIG_DEV_COREDUMP
 void xe_devcoredump(struct xe_exec_queue *q, struct xe_sched_job *job, const char *fmt, ...);
+void xe_devcoredump_gt(struct xe_gt *gt, const char *fmt, ...);

This what I'm suggesting for a header...

void __xe_devcoredump(struct xe_gt *gt, struct xe_exec_queue *q,

                     struct xe_sched_job *job, const char *fmt, ...);
#define xe_devcoredump(_q, _job, _fmt, ...) \
       __xe_devcoredump((_q)->gt, _q, _job, _fmt, ##__VA_ARGS__)
#define xe_devcoredump_gt(_gt, _fmt, ...) \
       __xe_devcoredump(_gt, NULL, NULL, _fmt, ##__VA_ARGS__)

I think we need wrapper macros rather than inline wrappers because of how .../##__VA_ARGS__ work.

Matt

Quoted 76 lines
drivers/gpu/drm/xe/xe_devcoredump.h +4 -0
 int xe_devcoredump_init(struct xe_device *xe);
 #else
 static inline void xe_devcoredump(struct xe_exec_queue *q,
@@ -23,6 +25,10 @@ static inline void xe_devcoredump(struct xe_exec_queue *q,
 {
 }
 
+static inline void xe_devcoredump_gt(struct xe_gt *gt, const char *fmt, ...)
+{
+}
+
 static inline int xe_devcoredump_init(struct xe_device *xe)
 {
 	return 0;
drivers/gpu/drm/xe/xe_tlb_inval.c +20 -0
index bbd21d39306..833fb92cd3e 100644
@@ -5,6 +5,7 @@
 
 #include <drm/drm_managed.h>
 
+#include "xe_devcoredump.h"
 #include "xe_device_types.h"
 #include "xe_force_wake.h"
 #include "xe_gt_stats.h"
@@ -29,6 +30,12 @@
 
 #define FENCE_STACK_BIT		DMA_FENCE_FLAG_USER_BITS
 
+/* The frontend is only ever embedded in a GT */
+static struct xe_gt *tlb_inval_to_gt(struct xe_tlb_inval *tlb_inval)
+{
+	return container_of(tlb_inval, struct xe_gt, tlb_inval);
+}
+
 static void xe_tlb_inval_fence_fini(struct xe_tlb_inval_fence *fence)
 {
 	if (WARN_ON_ONCE(!fence->tlb_inval))
@@ -73,6 +80,7 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 	struct xe_device *xe = tlb_inval->xe;
 	struct xe_tlb_inval_fence *fence, *next;
 	long timeout_delay = tlb_inval->ops->timeout_delay(tlb_inval);
+	int timedout_seqno = 0;
 
 	tlb_inval->ops->flush(tlb_inval);
 
@@ -90,6 +98,8 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 			"TLB invalidation fence timeout, seqno=%d recv=%d",
 			fence->seqno, tlb_inval->seqno_recv);
 
+		timedout_seqno = fence->seqno;
+
 		fence->base.error = -ETIME;
 		xe_tlb_inval_fence_signal(fence);
 	}
@@ -97,6 +107,16 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 		queue_delayed_work(tlb_inval->timeout_wq, &tlb_inval->fence_tdr,
 				   timeout_delay);
 	spin_unlock_irq(&tlb_inval->pending_lock);
+
+	/*
+	 * Capture the GuC log and CT state so the firmware side of the hang
+	 * can be inspected; there is no queue or job to blame here. Must be
+	 * outside pending_lock as the capture takes sleeping locks.
+	 */
+	if (timedout_seqno)
+		xe_devcoredump_gt(tlb_inval_to_gt(tlb_inval),
+				  "TLB invalidation fence timeout, seqno=%d recv=%d",
+				  timedout_seqno, tlb_inval->seqno_recv);
 }
 
 /**
Signature · 1 line
2.55.0
#4 lore raw

RFC PATCH 2/3 drm/xe: Log when a timed out TLB invalidation ack finally arrives

In reply to Tales A. Mendonça

When a TLB invalidation fence times out we log the timeout, but if the ack for that seqno later shows up there is no record of it, making it impossible to tell from logs whether the ack was lost forever or merely (very) late.

Track the most recent timed out seqno and log how late its ack arrives, relative to both the original request and the moment the fence was signaled with -ETIME.

On ARL with GuC 70.53.0 this shows the acks are never lost: they consistently arrive ~2.3s after the request, tens of milliseconds after the TDR has already signaled the fence:

TLB invalidation fence timeout, seqno=10992 recv=10991
TLB invalidation late ack: seqno=10992 recv=10992, request-to-ack=2314ms, timeout-to-ack=45ms

---
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 16 ++++++++++++++++
 drivers/gpu/drm/xe/xe_tlb_inval_types.h | 17 +++++++++++++++++
 2 files changed, 33 insertions(+)
drivers/gpu/drm/xe/xe_tlb_inval.c +16 -0
index 833fb92cd3e..9dd04d5bc4c 100644
@@ -99,6 +99,9 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 			fence->seqno, tlb_inval->seqno_recv);
 
 		timedout_seqno = fence->seqno;
+		tlb_inval->timedout_seqno = fence->seqno;
+		tlb_inval->timedout_inval_time = fence->inval_time;
+		tlb_inval->timedout_time = ktime_get();
 
 		fence->base.error = -ETIME;
 		xe_tlb_inval_fence_signal(fence);
@@ -227,6 +230,7 @@ void xe_tlb_inval_reset(struct xe_tlb_inval *tlb_inval)
 	else
 		pending_seqno = tlb_inval->seqno - 1;
 	WRITE_ONCE(tlb_inval->seqno_recv, pending_seqno);
+	tlb_inval->timedout_seqno = 0;
 
 	list_for_each_entry_safe(fence, next,
 				 &tlb_inval->pending_fences, link)
@@ -424,6 +428,18 @@ void xe_tlb_inval_done_handler(struct xe_tlb_inval *tlb_inval, int seqno)
 
 	WRITE_ONCE(tlb_inval->seqno_recv, seqno);
 
+	if (tlb_inval->timedout_seqno &&
+	    xe_tlb_inval_seqno_past(tlb_inval, tlb_inval->timedout_seqno)) {
+		ktime_t now = ktime_get();
+
+		drm_warn(&xe->drm,
+			 "TLB invalidation late ack: seqno=%d recv=%d, request-to-ack=%lldms, timeout-to-ack=%lldms",
+			 tlb_inval->timedout_seqno, seqno,
+			 ktime_ms_delta(now, tlb_inval->timedout_inval_time),
+			 ktime_ms_delta(now, tlb_inval->timedout_time));
+		tlb_inval->timedout_seqno = 0;
+	}
+
 	list_for_each_entry_safe(fence, next,
 				 &tlb_inval->pending_fences, link) {
 		trace_xe_tlb_inval_fence_recv(xe, fence);
drivers/gpu/drm/xe/xe_tlb_inval_types.h +17 -0
index 3d1797d186f..38288966254 100644
@@ -102,6 +102,23 @@ struct xe_tlb_inval {
 	 * @pending_lock: protects @pending_fences and updating @seqno_recv.
 	 */
 	spinlock_t pending_lock;
+	/**
+	 * @timedout_seqno: seqno of the most recent timed out TLB
+	 * invalidation, 0 if none. Used to measure how late the ack for a
+	 * timed out invalidation actually arrives. Protected by
+	 * @pending_lock.
+	 */
+	int timedout_seqno;
+	/**
+	 * @timedout_inval_time: request time of @timedout_seqno. Protected by
+	 * @pending_lock.
+	 */
+	ktime_t timedout_inval_time;
+	/**
+	 * @timedout_time: time @timedout_seqno was signaled with -ETIME.
+	 * Protected by @pending_lock.
+	 */
+	ktime_t timedout_time;
 	/**
 	 * @fence_tdr: schedules a delayed call to xe_tlb_fence_timeout after
 	 * the timeout interval is over.
Signature · 1 line
2.55.0
maintainer
#5 lore raw

Re: [RFC PATCH 2/3] drm/xe: Log when a timed out TLB invalidation ack finally arrives

Quoted 28 lines · Tales A. Mendonça

When a TLB invalidation fence times out we log the timeout, but if the ack for that seqno later shows up there is no record of it, making it impossible to tell from logs whether the ack was lost forever or merely (very) late.

Track the most recent timed out seqno and log how late its ack arrives, relative to both the original request and the moment the fence was signaled with -ETIME.

On ARL with GuC 70.53.0 this shows the acks are never lost: they consistently arrive ~2.3s after the request, tens of milliseconds after the TDR has already signaled the fence:

TLB invalidation fence timeout, seqno=10992 recv=10991
TLB invalidation late ack: seqno=10992 recv=10992, request-to-ack=2314ms, timeout-to-ack=45ms

---
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 16 ++++++++++++++++
 drivers/gpu/drm/xe/xe_tlb_inval_types.h | 17 +++++++++++++++++
 2 files changed, 33 insertions(+)
drivers/gpu/drm/xe/xe_tlb_inval.c +0 -0
index 833fb92cd3e..9dd04d5bc4c 100644

drivers/gpu/drm/xe/xe_tlb_inval.c static void xe_tlb_inval_fence_timeout(struct work_struct *work) · line 101

Quoted 3 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +0 -0
@@ -99,6 +99,9 @@ static void xe_tlb_inval_fence_timeout(struct work_struct *work)
 			fence->seqno, tlb_inval->seqno_recv);
 
 		timedout_seqno = fence->seqno;

Should this be:

if (!tlb_inval->timedout_seqno) {
	tlb_inval->timedout_seqno = fence->seqno;
	tlb_inval->timedout_inval_time = fence->inval_time;
	tlb_inval->timedout_time = ktime_get();
}

To record the very first timeout seqno? I think this makes more sense.

Quoted 15 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +4 -0
+		tlb_inval->timedout_seqno = fence->seqno;
+		tlb_inval->timedout_inval_time = fence->inval_time;
+		tlb_inval->timedout_time = ktime_get();
 
 		fence->base.error = -ETIME;
 		xe_tlb_inval_fence_signal(fence);
@@ -227,6 +230,7 @@ void xe_tlb_inval_reset(struct xe_tlb_inval *tlb_inval)
 	else
 		pending_seqno = tlb_inval->seqno - 1;
 	WRITE_ONCE(tlb_inval->seqno_recv, pending_seqno);
+	tlb_inval->timedout_seqno = 0;
 
 	list_for_each_entry_safe(fence, next,
 				 &tlb_inval->pending_fences, link)

drivers/gpu/drm/xe/xe_tlb_inval.c void xe_tlb_inval_done_handler(struct xe_tlb_inval *tlb_inval, int seqno) · line 435

Quoted 8 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +5 -0
@@ -424,6 +428,18 @@ void xe_tlb_inval_done_handler(struct xe_tlb_inval *tlb_inval, int seqno)
 
 	WRITE_ONCE(tlb_inval->seqno_recv, seqno);
 
+	if (tlb_inval->timedout_seqno &&
+	    xe_tlb_inval_seqno_past(tlb_inval, tlb_inval->timedout_seqno)) {
+		ktime_t now = ktime_get();
+
+		drm_warn(&xe->drm,

I think xe_warn is the preference here.

Matt

Quoted 41 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +7 -0
+			 "TLB invalidation late ack: seqno=%d recv=%d, request-to-ack=%lldms, timeout-to-ack=%lldms",
+			 tlb_inval->timedout_seqno, seqno,
+			 ktime_ms_delta(now, tlb_inval->timedout_inval_time),
+			 ktime_ms_delta(now, tlb_inval->timedout_time));
+		tlb_inval->timedout_seqno = 0;
+	}
+
 	list_for_each_entry_safe(fence, next,
 				 &tlb_inval->pending_fences, link) {
 		trace_xe_tlb_inval_fence_recv(xe, fence);
drivers/gpu/drm/xe/xe_tlb_inval_types.h +17 -0
index 3d1797d186f..38288966254 100644
@@ -102,6 +102,23 @@ struct xe_tlb_inval {
 	 * @pending_lock: protects @pending_fences and updating @seqno_recv.
 	 */
 	spinlock_t pending_lock;
+	/**
+	 * @timedout_seqno: seqno of the most recent timed out TLB
+	 * invalidation, 0 if none. Used to measure how late the ack for a
+	 * timed out invalidation actually arrives. Protected by
+	 * @pending_lock.
+	 */
+	int timedout_seqno;
+	/**
+	 * @timedout_inval_time: request time of @timedout_seqno. Protected by
+	 * @pending_lock.
+	 */
+	ktime_t timedout_inval_time;
+	/**
+	 * @timedout_time: time @timedout_seqno was signaled with -ETIME.
+	 * Protected by @pending_lock.
+	 */
+	ktime_t timedout_time;
 	/**
 	 * @fence_tdr: schedules a delayed call to xe_tlb_fence_timeout after
 	 * the timeout interval is over.
Signature · 1 line
2.55.0
#6 lore raw

RFC PATCH 3/3 drm/xe: Kick GuC while TLB invalidation acks are overdue

In reply to Tales A. Mendonça

On ARL with GuC 70.53.0, TLB invalidation acks intermittently stall:
the H2G request is consumed from the CTB immediately, but the G2H ack only arrives ~2.3s later, tens of milliseconds after the fence timeout has fired. During the stall the GPU keeps rendering; only work blocked on the invalidation (e.g. Wayland compositors performing buffer unmaps) hitches for the full 2.3s. Observed on three machines so far (7d51, 7dd1, plus an Arc Pro 130T report), see Link.

Experiments ruled out the obvious suspects:

* GT C6 parking: holding forcewake across the whole GT (C6 residency pinned at 0ms) still produced 9 timeouts in a row.
* Lost interrupt/CT processing on the host: the G2H CTB is empty at timeout time; the ack genuinely has not been sent by the firmware.

What does help is poking the GuC while the ack is overdue. Add a delayed work that fires XE_TLB_INVAL_KICK_DELAY_MS after an invalidation is issued and, while any ack is pending, re-reads the GuC status register, flushes the CT fast-path and rings the GuC doorbell (xe_guc_notify()), re-arming itself until the ack arrives; the existing TDR still bounds the total wait.

Instrumented results from two ARL machines over several days:

* Without kicks: every stall lasts the full ~2.3s and is reported as a fence timeout (-ETIME), ~1/hour on a desktop workload.
* With kicks: the majority of stalls resolve 15-276ms after one of the kicks, e.g.:

     TLB invalidation ack after kick: seqno=36405 recv=36405, request-to-ack=528ms, last-kick-to-ack=15ms, kicks=2

* A minority of severe episodes ignore 8-9 consecutive doorbells and still run to the timeout (worst observed: ack 5996ms after request, 3.7s after the last kick), suggesting the firmware is internally blocked for the whole window rather than missing a wake event.

This is a workaround, not a fix - the root cause looks like a GuC firmware issue - but it turns a guaranteed 2.3s stall into a sub-500ms hiccup for most occurrences, and the "ack after kick" log documents the firmware behavior for further debugging.

---
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 95 +++++++++++++++++++++++--
 drivers/gpu/drm/xe/xe_tlb_inval_types.h | 25 +++++++
 2 files changed, 116 insertions(+), 4 deletions(-)
drivers/gpu/drm/xe/xe_tlb_inval.c +91 -4
index 9dd04d5bc4c..16c32d5669f 100644
@@ -10,7 +10,9 @@
 #include "xe_force_wake.h"
 #include "xe_gt_stats.h"
 #include "xe_gt_types.h"
+#include "xe_guc.h"
 #include "xe_guc_ct.h"
+#include "xe_guc_pc.h"
 #include "xe_guc_tlb_inval.h"
 #include "xe_mmio.h"
 #include "xe_pm.h"
@@ -30,6 +32,15 @@
 
 #define FENCE_STACK_BIT		DMA_FENCE_FLAG_USER_BITS
 
+/*
+ * Delay before poking the GuC about a pending invalidation that has not been
+ * acked yet. Acks normally arrive in microseconds; when the GuC stalls they
+ * only show up seconds later, after the timeout has already fired.
+ */
+#define XE_TLB_INVAL_KICK_DELAY_MS	250
+
+static void xe_tlb_inval_kick(struct work_struct *work);
+
 /* The frontend is only ever embedded in a GT */
 static struct xe_gt *tlb_inval_to_gt(struct xe_tlb_inval *tlb_inval)
 {
@@ -54,8 +65,10 @@ xe_tlb_inval_fence_signal(struct xe_tlb_inval_fence *fence)
 	lockdep_assert_held(&fence->tlb_inval->pending_lock);
 
 	list_del(&fence->link);
-	if (list_empty(&tlb_inval->pending_fences))
+	if (list_empty(&tlb_inval->pending_fences)) {
 		cancel_delayed_work(&tlb_inval->fence_tdr);
+		cancel_delayed_work(&tlb_inval->kick_work);
+	}
 	trace_xe_tlb_inval_fence_signal(fence->tlb_inval->xe, fence);
 	xe_tlb_inval_fence_fini(fence);
 	dma_fence_signal(&fence->base);
@@ -168,6 +181,7 @@ int xe_gt_tlb_inval_init_early(struct xe_gt *gt)
 	spin_lock_init(&tlb_inval->pending_lock);
 	spin_lock_init(&tlb_inval->lock);
 	INIT_DELAYED_WORK(&tlb_inval->fence_tdr, xe_tlb_inval_fence_timeout);
+	INIT_DELAYED_WORK(&tlb_inval->kick_work, xe_tlb_inval_kick);
 
 	err = drmm_mutex_init(&xe->drm, &tlb_inval->seqno_lock);
 	if (err)
@@ -218,6 +232,7 @@ void xe_tlb_inval_reset(struct xe_tlb_inval *tlb_inval)
 	mutex_lock(&tlb_inval->seqno_lock);
 	spin_lock_irq(&tlb_inval->pending_lock);
 	cancel_delayed_work(&tlb_inval->fence_tdr);
+	cancel_delayed_work(&tlb_inval->kick_work);
 	/*
 	 * We might have various kworkers waiting for TLB flushes to complete
 	 * which are not tracked with an explicit TLB fence, however at this
@@ -231,6 +246,8 @@ void xe_tlb_inval_reset(struct xe_tlb_inval *tlb_inval)
 		pending_seqno = tlb_inval->seqno - 1;
 	WRITE_ONCE(tlb_inval->seqno_recv, pending_seqno);
 	tlb_inval->timedout_seqno = 0;
+	tlb_inval->kicked_seqno = 0;
+	tlb_inval->kicked_count = 0;
 
 	list_for_each_entry_safe(fence, next,
 				 &tlb_inval->pending_fences, link)
@@ -268,6 +285,55 @@ static bool xe_tlb_inval_seqno_past(struct xe_tlb_inval *tlb_inval, int seqno)
 	return seqno_recv >= seqno;
 }
 
+static void xe_tlb_inval_kick(struct work_struct *work)
+{
+	struct xe_tlb_inval *tlb_inval = container_of(work, struct xe_tlb_inval,
+						      kick_work.work);
+	struct xe_gt *gt = tlb_inval_to_gt(tlb_inval);
+	struct xe_tlb_inval_fence *fence;
+	ktime_t inval_time = 0;
+	int seqno = 0;
+
+	spin_lock_irq(&tlb_inval->pending_lock);
+	fence = list_first_entry_or_null(&tlb_inval->pending_fences,
+					 struct xe_tlb_inval_fence, link);
+	if (fence) {
+		seqno = fence->seqno;
+		inval_time = fence->inval_time;
+	}
+	spin_unlock_irq(&tlb_inval->pending_lock);
+
+	if (!seqno)
+		return;
+
+	/*
+	 * Poke the GuC: read its status register, flush the CT fast-path and
+	 * ring the doorbell. On ARL with GuC 70.53.0 the ack for a pending
+	 * invalidation sometimes only arrives seconds after the request even
+	 * though the H2G was consumed immediately; a doorbell ring while the
+	 * ack is overdue usually unsticks it within ~250ms (see Link in the
+	 * commit message). Keep kicking every interval until the ack shows
+	 * up; the TDR bounds how long this can go on.
+	 */
+	if (gt->gtidle.idle_residency)
+		xe_guc_pc_c_status(&gt->uc.guc.pc);
+	tlb_inval->ops->flush(tlb_inval);
+	xe_guc_notify(&gt->uc.guc);
+
+	spin_lock_irq(&tlb_inval->pending_lock);
+	if (!xe_tlb_inval_seqno_past(tlb_inval, seqno)) {
+		if (tlb_inval->kicked_seqno != seqno)
+			tlb_inval->kicked_count = 0;
+		tlb_inval->kicked_seqno = seqno;
+		tlb_inval->kicked_inval_time = inval_time;
+		tlb_inval->kicked_time = ktime_get();
+		tlb_inval->kicked_count++;
+		queue_delayed_work(tlb_inval->timeout_wq, &tlb_inval->kick_work,
+				   msecs_to_jiffies(XE_TLB_INVAL_KICK_DELAY_MS));
+	}
+	spin_unlock_irq(&tlb_inval->pending_lock);
+}
+
 static void xe_tlb_inval_fence_prep(struct xe_tlb_inval_fence *fence)
 {
 	struct xe_tlb_inval *tlb_inval = fence->tlb_inval;
@@ -279,9 +345,12 @@ static void xe_tlb_inval_fence_prep(struct xe_tlb_inval_fence *fence)
 	fence->inval_time = ktime_get();
 	list_add_tail(&fence->link, &tlb_inval->pending_fences);
 
-	if (list_is_singular(&tlb_inval->pending_fences))
+	if (list_is_singular(&tlb_inval->pending_fences)) {
 		queue_delayed_work(tlb_inval->timeout_wq, &tlb_inval->fence_tdr,
 				   tlb_inval->ops->timeout_delay(tlb_inval));
+		queue_delayed_work(tlb_inval->timeout_wq, &tlb_inval->kick_work,
+				   msecs_to_jiffies(XE_TLB_INVAL_KICK_DELAY_MS));
+	}
 	spin_unlock_irq(&tlb_inval->pending_lock);
 
 	tlb_inval->seqno = (tlb_inval->seqno + 1) %
@@ -440,6 +509,20 @@ void xe_tlb_inval_done_handler(struct xe_tlb_inval *tlb_inval, int seqno)
 		tlb_inval->timedout_seqno = 0;
 	}
 
+	if (tlb_inval->kicked_seqno &&
+	    xe_tlb_inval_seqno_past(tlb_inval, tlb_inval->kicked_seqno)) {
+		ktime_t now = ktime_get();
+
+		drm_warn(&xe->drm,
+			 "TLB invalidation ack after kick: seqno=%d recv=%d, request-to-ack=%lldms, last-kick-to-ack=%lldms, kicks=%d",
+			 tlb_inval->kicked_seqno, seqno,
+			 ktime_ms_delta(now, tlb_inval->kicked_inval_time),
+			 ktime_ms_delta(now, tlb_inval->kicked_time),
+			 tlb_inval->kicked_count);
+		tlb_inval->kicked_seqno = 0;
+		tlb_inval->kicked_count = 0;
+	}
+
 	list_for_each_entry_safe(fence, next,
 				 &tlb_inval->pending_fences, link) {
 		trace_xe_tlb_inval_fence_recv(xe, fence);
@@ -450,12 +533,16 @@ void xe_tlb_inval_done_handler(struct xe_tlb_inval *tlb_inval, int seqno)
 		xe_tlb_inval_fence_signal(fence);
 	}
 
-	if (!list_empty(&tlb_inval->pending_fences))
+	if (!list_empty(&tlb_inval->pending_fences)) {
 		mod_delayed_work(tlb_inval->timeout_wq,
 				 &tlb_inval->fence_tdr,
 				 tlb_inval->ops->timeout_delay(tlb_inval));
-	else
+		mod_delayed_work(tlb_inval->timeout_wq, &tlb_inval->kick_work,
+				 msecs_to_jiffies(XE_TLB_INVAL_KICK_DELAY_MS));
+	} else {
 		cancel_delayed_work(&tlb_inval->fence_tdr);
+		cancel_delayed_work(&tlb_inval->kick_work);
+	}
 
 	spin_unlock_irqrestore(&tlb_inval->pending_lock, flags);
 }
drivers/gpu/drm/xe/xe_tlb_inval_types.h +25 -0
index 38288966254..f8ae540dc4b 100644
@@ -124,6 +124,31 @@ struct xe_tlb_inval {
 	 * the timeout interval is over.
 	 */
 	struct delayed_work fence_tdr;
+	/**
+	 * @kick_work: pokes the GuC while an invalidation ack is overdue,
+	 * bounding ack stalls on GuC firmware that misses CT notifications.
+	 */
+	struct delayed_work kick_work;
+	/**
+	 * @kicked_seqno: seqno the last kick was issued for, 0 if none.
+	 * Protected by @pending_lock.
+	 */
+	int kicked_seqno;
+	/**
+	 * @kicked_inval_time: request time of @kicked_seqno. Protected by
+	 * @pending_lock.
+	 */
+	ktime_t kicked_inval_time;
+	/**
+	 * @kicked_time: time the last kick for @kicked_seqno ran. Protected
+	 * by @pending_lock.
+	 */
+	ktime_t kicked_time;
+	/**
+	 * @kicked_count: number of kicks issued for @kicked_seqno. Protected
+	 * by @pending_lock.
+	 */
+	int kicked_count;
 	/** @job_wq: schedules TLB invalidation jobs */
 	struct workqueue_struct *job_wq;
 	/** @tlb_inval.lock: protects TLB invalidation fences */
Signature · 1 line
2.55.0
#7 lore raw

Re: [RFC PATCH 3/3] drm/xe: Kick GuC while TLB invalidation acks are overdue

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] Unprotected MMIO access in `xe_tlb_inval_kick` can cause a crash if the device suspends…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Unprotected MMIO access in xe_tlb_inval_kick can cause a crash if the device suspends concurrently. --

--- Patch [3]: [RFC PATCH 3/3] drm/xe: Kick GuC while TLB invalidation acks are overdue --- commit c7a516f0e42fcf29db699b7dce1432a3b42378cd
Author: Tales A. Mendonça <[email protected]>

drm/xe: Kick GuC while TLB invalidation acks are overdue

This commit adds a delayed work to periodically poke the GuC when a TLB invalidation ack is delayed. This provides a workaround for a firmware stall issue by re-reading the GuC status, flushing the CT fast-path, and ringing the GuC doorbell until the ack arrives.

Quoted 4 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +0 -0
index 9dd04d5bc4cd2..16c32d5669f36 100644

[ ... ]

Quoted 28 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +24 -0
@@ -268,6 +285,55 @@ static bool xe_tlb_inval_seqno_past(struct xe_tlb_inval *tlb_inval, int seqno)
 	return seqno_recv >= seqno;
 }
 
+static void xe_tlb_inval_kick(struct work_struct *work)
+{
+	struct xe_tlb_inval *tlb_inval = container_of(work, struct xe_tlb_inval,
+						      kick_work.work);
+	struct xe_gt *gt = tlb_inval_to_gt(tlb_inval);
+	struct xe_tlb_inval_fence *fence;
+	ktime_t inval_time = 0;
+	int seqno = 0;
+
+	spin_lock_irq(&tlb_inval->pending_lock);
+	fence = list_first_entry_or_null(&tlb_inval->pending_fences,
+					 struct xe_tlb_inval_fence, link);
+	if (fence) {
+		seqno = fence->seqno;
+		inval_time = fence->inval_time;
+	}
+	spin_unlock_irq(&tlb_inval->pending_lock);
+
+	if (!seqno)
+		return;
+
+	/*
+	 * Poke the GuC: read its status register, flush the CT fast-path and
+	 * ring the doorbell. On ARL with GuC 70.53.0 the ack for a pending

drivers/gpu/drm/xe/xe_tlb_inval.c static bool xe_tlb_inval_seqno_past(struct xe_tlb_inval *tlb_inval, int seqno) · line 321

Quoted 10 lines
drivers/gpu/drm/xe/xe_tlb_inval.c +10 -0
+	 * invalidation sometimes only arrives seconds after the request even
+	 * though the H2G was consumed immediately; a doorbell ring while the
+	 * ack is overdue usually unsticks it within ~250ms (see Link in the
+	 * commit message). Keep kicking every interval until the ack shows
+	 * up; the TDR bounds how long this can go on.
+	 */
+	if (gt->gtidle.idle_residency)
+		xe_guc_pc_c_status(&gt->uc.guc.pc);
+	tlb_inval->ops->flush(tlb_inval);
+	xe_guc_notify(&gt->uc.guc);

[Severity: High]
Could this lead to an unprotected MMIO access if the device suspends concurrently?

When pending_lock is dropped above, the device is kept awake by the runtime PM reference tied to the pending fence. If the GuC ack arrives concurrently on another CPU, xe_tlb_inval_done_handler() will acquire the lock, signal the fence, and call xe_pm_runtime_put() via xe_tlb_inval_fence_fini().

Since cancel_delayed_work() does not block if the worker is already running, is it possible for the last PM reference to be dropped before we reach these MMIO accesses in xe_guc_pc_c_status() and xe_guc_notify()?

If the device suspends, accessing these registers could trigger a PCIe timeout or master abort and crash the system.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
#8 lore raw

✗ LGCI.VerificationFailed: failure for drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL

In reply to Tales A. Mendonça
== Series Details ==

== Series Details ==

Series: drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL
URL : https://patchwork.freedesktop.org/series/171539/
State : failure

== Summary ==

Series author address '[email protected]' is not on the allowlist, which prevents CI from being automatically triggered. If you want CI to run for this series, ask Patchwork project owners to click 'retest' on the series in Patchwork. Exception occurred during validation, bailing out!
Build URL: http://intel-gfx-ci-public.igk.intel.com:8080/job/xe_pw_trigger/1237486/ (on master)

#9 lore raw
In reply to Tales A. Mendonça

Patchwork reports my address is not on the CI allowlist, so CI was not triggered for this series:

Series author address '[email protected]' is not on the allowlist,
which prevents CI from being automatically triggered.

Could one of the project owners click 'retest' on the series (and/or
add me to the allowlist)? Series URL:

    https://patchwork.freedesktop.org/series/171539/

Thanks!
Tales

Em seg., 3 de ago. de 2026 às 23:14, Tales A. Mendonça <[email protected]> escreveu:

Quoted 58 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor buffer unmaps etc.) hitches for the full window.

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no exec queue or job to blame - leaves a devcoredump with the GuC log and CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into a sub-500ms hiccup for the majority of occurrences; a minority of severe episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
 drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 131 +++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
 4 files changed, 243 insertions(+), 4 deletions(-)

--
2.55.0

Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
maintainer
#10 lore raw
Quoted 11 lines · Tales A. Mendonça

Patchwork reports my address is not on the CI allowlist, so CI was not triggered for this series:

Series author address '[email protected]' is not on the allowlist,
which prevents CI from being automatically triggered.

Could one of the project owners click 'retest' on the series (and/or
add me to the allowlist)? Series URL:

    https://patchwork.freedesktop.org/series/171539/

We'd have to resend this ourselves. I can do this, but I've requested for you to be on our allow list as well. I'll ping here once that goes through.

Matt

Quoted 72 lines

Thanks!
Tales

Em seg., 3 de ago. de 2026 às 23:14, Tales A. Mendonça <[email protected]> escreveu:

Quoted 58 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor buffer unmaps etc.) hitches for the full window.

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no exec queue or job to blame - leaves a devcoredump with the GuC log and CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into a sub-500ms hiccup for the majority of occurrences; a minority of severe episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
 drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 131 +++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
 4 files changed, 243 insertions(+), 4 deletions(-)

--
2.55.0

Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
maintainer
#11 lore raw
Quoted 10 lines · Matthew Brost
Quoted 9 lines · Tales A. Mendonça

Patchwork reports my address is not on the CI allowlist, so CI was not triggered for this series:

Series author address '[email protected]' is not on the allowlist,
which prevents CI from being automatically triggered.

Could one of the project owners click 'retest' on the series (and/or
add me to the allowlist)? Series URL:

Quoted 7 lines
Quoted 2 lines
    https://patchwork.freedesktop.org/series/171539/

We'd have to resend this ourselves. I can do this, but I've requested for you to be on our allow list as well. I'll ping here once that goes through.

You are approved on our CI for future patches.

Also I came across this issue in the i915 for MTL [1] which seems to indicate the same issue (ARL and MTL are very close and share same GuC firmware), the suggestion there is turn off rc6. Unfortunately Xe doesn't have a knob to do this but according to [1] you can turn off rc6 in the BIOS. Might be worth a try.

Matt

[1] https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/14469

Quoted 74 lines

Matt

Quoted 72 lines

Thanks!
Tales

Em seg., 3 de ago. de 2026 às 23:14, Tales A. Mendonça <[email protected]> escreveu:

Quoted 58 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor buffer unmaps etc.) hitches for the full window.

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no exec queue or job to blame - leaves a devcoredump with the GuC log and CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into a sub-500ms hiccup for the majority of occurrences; a minority of severe episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
 drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 131 +++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
 4 files changed, 243 insertions(+), 4 deletions(-)

--
2.55.0

Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
#12 lore raw
Quoted 1 line

You've been approved in our CI for future revs.

Thanks!

Quoted 5 lines

Also I came across this issue on i915 for MTL [1] which appears to indicate the same issue (ARL and MTL are very close and share the same GuC firmware), the suggestion is to disable rc6. Unfortunately Xe doesn't have a knob for this, but per [1] you can turn off rc6 in the BIOS. Maybe worth trying.

Good find - that does look like the same ~2.3s signature, and it being visible on i915/MTL too is an important data point (Daniele asked exactly that in the other subthread).

On the rc6 angle: I ran an A/B experiment earlier that should be equivalent to disabling rc6 at the GT level - holding forcewake across the whole GT for hours (C6 residency pinned at 0ms for the entire window, verified) - and still hit 9 timeouts in a row, with the same ~2.3s request-to-ack. So at least keeping the GT out of RC6 does not avoid the stall here. If the BIOS suggestion covers more than GT RC6 (e.g. package C-states), that would be a different experiment - my consumer ASUS BIOS does not expose an rc6 knob, but I can look for C-state options if you think it is worth isolating.

Next on my side: switching to the GuC 70.72.1 build Daniele posted and reporting back, plus GuC logs from severe stalls are now attached to the gitlab issue.

Thanks,
Tales

Em qua., 5 de ago. de 2026 às 17:25, Matthew Brost <[email protected]> escreveu:

Quoted 106 lines · Matthew Brost
Quoted 10 lines · Tales A. Mendonça
Quoted 9 lines

Patchwork reports my address is not on the CI allowlist, so CI was not triggered for this series:

Series author address '[email protected]' is not on the allowlist,
which prevents CI from being automatically triggered.

Could one of the project owners click 'retest' on the series (and/or
add me to the allowlist)? Series URL:

Quoted 7 lines
Quoted 2 lines
    https://patchwork.freedesktop.org/series/171539/

We'd have to resend this ourselves. I can do this, but I've requested for you to be on our allow list as well. I'll ping here once that goes through.

You are approved on our CI for future patches.

Also I came across this issue in the i915 for MTL [1] which seems to indicate the same issue (ARL and MTL are very close and share same GuC firmware), the suggestion there is turn off rc6. Unfortunately Xe doesn't have a knob to do this but according to [1] you can turn off rc6 in the BIOS. Might be worth a try.

Matt

[1] https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/14469

Quoted 74 lines

Matt

Quoted 72 lines

Thanks!
Tales

Em seg., 3 de ago. de 2026 às 23:14, Tales A. Mendonça <[email protected]> escreveu:

Quoted 49 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor buffer unmaps etc.) hitches for the full window.

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no exec queue or job to blame - leaves a devcoredump with the GuC log and CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into a sub-500ms hiccup for the majority of occurrences; a minority of severe episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

Quoted 9 lines
 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
 drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
 drivers/gpu/drm/xe/xe_tlb_inval.c       | 131 +++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
 4 files changed, 243 insertions(+), 4 deletions(-)

--
2.55.0

--
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org

Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
#13 lore raw
Quoted 11 lines · Tales A. Mendonça
Quoted 1 line

You've been approved in our CI for future revs.

Thanks!

Quoted 5 lines

Also I came across this issue on i915 for MTL [1] which appears to indicate the same issue (ARL and MTL are very close and share the same GuC firmware), the suggestion is to disable rc6. Unfortunately Xe doesn't have a knob for this, but per [1] you can turn off rc6 in the BIOS. Maybe worth trying.

Good find - that does look like the same ~2.3s signature, and it being visible on i915/MTL too is an important data point (Daniele asked exactly that in the other subthread).

I am not sure if 14469 is the same issue. In that one it seems like the GuC just stops processing incoming messages (i.e., the H2G CTB has unprocessed data in it), while AFAIU in your case the GuC is still processing new commands and it is just being slow. It might be different manifestations of the same underlying issue, but it might also be completely separate bugs.

Quoted 10 lines

On the rc6 angle: I ran an A/B experiment earlier that should be equivalent to disabling rc6 at the GT level - holding forcewake across the whole GT for hours (C6 residency pinned at 0ms for the entire window, verified) - and still hit 9 timeouts in a row, with the same ~2.3s request-to-ack. So at least keeping the GT out of RC6 does not avoid the stall here. If the BIOS suggestion covers more than GT RC6 (e.g. package C-states), that would be a different experiment - my consumer ASUS BIOS does not expose an rc6 knob, but I can look for C-state options if you think it is worth isolating.

The fact that keeping the GT awake didn't help also indicates that this likely isn't the same 14469.

Daniele

Quoted 113 lines

Next on my side: switching to the GuC 70.72.1 build Daniele posted and reporting back, plus GuC logs from severe stalls are now attached to the gitlab issue.

Thanks,
Tales

Em qua., 5 de ago. de 2026 às 17:25, Matthew Brost <[email protected]> escreveu:

Quoted 101 lines · Matthew Brost
Quoted 7 lines · Tales A. Mendonça
Quoted 6 lines

Patchwork reports my address is not on the CI allowlist, so CI was not triggered for this series:

Series author address '[email protected]' is not on the allowlist,
which prevents CI from being automatically triggered.

Quoted 9 lines
Quoted 5 lines

Could one of the project owners click 'retest' on the series (and/or
add me to the allowlist)? Series URL:

     https://patchwork.freedesktop.org/series/171539/

We'd have to resend this ourselves. I can do this, but I've requested for you to be on our allow list as well. I'll ping here once that goes through.

You are approved on our CI for future patches.

Also I came across this issue in the i915 for MTL [1] which seems to indicate the same issue (ARL and MTL are very close and share same GuC firmware), the suggestion there is turn off rc6. Unfortunately Xe doesn't have a knob to do this but according to [1] you can turn off rc6 in the BIOS. Might be worth a try.

Matt

[1] https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/14469

Quoted 72 lines

Matt

Quoted 70 lines

Thanks!
Tales

Em seg., 3 de ago. de 2026 às 23:14, Tales A. Mendonça <[email protected]> escreveu:

Quoted 48 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

   https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor buffer unmaps etc.) hitches for the full window.

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no exec queue or job to blame - leaves a devcoredump with the GuC log and CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into a sub-500ms hiccup for the majority of occurrences; a minority of severe episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

Quoted 9 lines
  drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
  drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
  drivers/gpu/drm/xe/xe_tlb_inval.c       | 131 +++++++++++++++++++++++-
  drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
  4 files changed, 243 insertions(+), 4 deletions(-)

--
2.55.0

--
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org

#14 lore raw
In reply to Tales A. Mendonça
Quoted 7 lines · Tales A. Mendonça

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

  https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Quoted 9 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

Quoted 6 lines

exec queue or job to blame - leaves a devcoredump with the GuC log and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Quoted 7 lines

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into

I'm a little worried we're just papering over something here that needs to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on registers might be prematurely waking things up which is something we'd want to happen in GuC, not the KMD.

Thanks,
Stuart

Quoted 28 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
  drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++ drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++ drivers/gpu/drm/xe/xe_tlb_inval.c       | 131
+++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++

 4 files changed, 243 insertions(+), 4 deletions(-)
maintainer
#15 lore raw
Quoted 18 lines · Stuart
Quoted 7 lines · Tales A. Mendonça

Hi,

This series is a follow-up to the TLB invalidation ack stall I have been debugging on ARL, tracked in:

  https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Quoted 9 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and 7dd1 machines here, plus an independent Arc Pro 130T report on the issue above), TLB invalidation acks intermittently stall for ~2.3s. The H2G request is consumed from the CTB immediately and the G2H CTB is empty the whole time - the firmware simply does not send the ack until much later. The fence timeout fires at 2.25s and the ack lands tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 7 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

[1] https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 5 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job-based, so it cannot be used for hangs that are not associated with a job.

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g., TLB invalidation timeouts like those addressed in this series, or more generally any GuC protocol hang), makes sense to me.

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 14 lines
Quoted 6 lines

exec queue or job to blame - leaves a devcoredump with the GuC log and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to the issue above).

Quoted 7 lines

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than lost.

Patch 3 is the RFC part: a delayed work that pokes the GuC (status register read, CT flush, doorbell ring) every 250ms while an ack is overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 6 lines

I'm a little worried we're just papering over something here that needs to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on registers might be prematurely waking things up which is something we'd want to happen in GuC, not the KMD.

In general, I agree with this. We should avoid papering over the issue and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Matt

Quoted 32 lines

Thanks,
Stuart

Quoted 28 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC firmware being internally blocked for the whole window. Full data on the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly I would like the firmware side investigated, since no host-side poke can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
  drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives drm/xe: Kick GuC while TLB invalidation acks are overdue

 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++ drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++ drivers/gpu/drm/xe/xe_tlb_inval.c       | 131
+++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++

 4 files changed, 243 insertions(+), 4 deletions(-)
#16 lore raw
Quoted 38 lines · Matthew Brost
Quoted 25 lines · Stuart
Quoted 8 lines · Tales A. Mendonça

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

  https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 15 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 6 lines
Quoted 3 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 34 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 20 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Thanks,
Stuart

Quoted 41 lines

Matt

Quoted 38 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
  drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally arrives
  drm/xe: Kick GuC while TLB invalidation acks are overdue

 drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++ drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++ drivers/gpu/drm/xe/xe_tlb_inval.c       | 131
+++++++++++++++++++++++-
 drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++

 4 files changed, 243 insertions(+), 4 deletions(-)
#17 lore raw
Quoted 107 lines · Stuart
Quoted 36 lines · Matthew Brost
Quoted 24 lines · Stuart
Quoted 8 lines · Tales A. Mendonça

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

  https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915. Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 43 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
  drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
  drm/xe: Kick GuC while TLB invalidation acks are overdue

  drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
  drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
  drivers/gpu/drm/xe/xe_tlb_inval.c       | 131
+++++++++++++++++++++++-
  drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
  4 files changed, 243 insertions(+), 4 deletions(-)
#18 lore raw
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 175 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

   https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 43 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

  drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
  drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
  drivers/gpu/drm/xe/xe_tlb_inval.c       | 131
+++++++++++++++++++++++-
  drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
  4 files changed, 243 insertions(+), 4 deletions(-)
Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
#19 lore raw
Quoted 6 lines · Tales A. Mendonça
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Quoted 6 lines

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

Quoted 6 lines

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

Just a bit of a terminology update here, to make sure we're on the same page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell. I'm making this clarification because the GuC supports a separate per-context notification mechanism that is referred to as doorbell and which we currently do not implement in neither i915 nor Xe.

When receiving the H2G interrupt, the only thing that the GuC does is look into the CTB and process anything in there; however, you've said that the contents of the H2G CTB are processed immediately, so the follow up interrupt should result in the GuC just bailing out and doing nothing because there is no data to process. It feels like when the issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

I have pushed the latest GuC FW for MTL here in case you want to give it a go: https://gitlab.com/dceraolo/drm-firmware/-/blob/f783b931555be057dafc2400b7eb4d445c953fec/i915/mtl_guc_70.72.1.bin . You can override the GuC firmware used by the driver via the xe.guc_firmware_path modparam; the path is relative to /lib/firmware/ and the firmware needs to be in initramfs for the driver to find it at boot. Note that we haven't tested this image on MTL, so it might have unexpected results.

Also, would you be able to capture the GuC logs when the issue occurs? The default guc log size is relatively small, so you'd have to capture right when the issue happens. However, you can make them bigger by building the kernel with CONFIG_DRM_XE_DEBUG or by simply modifying the xe_guc_log.h file to pick the bigger size by default. If you go with the latter, please also set xe.guc_log_level=3 on the command line (this is automatically added by the kconfig).

Thanks,
Daniele

Quoted 181 lines

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 171 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

    https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 42 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

   drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
   drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
   drivers/gpu/drm/xe/xe_tlb_inval.c       | 131
+++++++++++++++++++++++-
   drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
   4 files changed, 243 insertions(+), 4 deletions(-)
#20 lore raw
Quoted 2 lines

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Two data points on that, pulling in different directions:

* My second ARL machine (7dd1) has been running i915 for 7 days now, with the same GuC 70.53.0, and there is not a single TLB invalidation timeout or GuC error in its logs. Caveat: i915's TLB invalidation timeout is longer than xe's 2.25s, so short stalls could be silent there.

* However, Matt just pointed at i915 MTL issue 14469, which looks like the same problem on i915 - so it may not be xe-specific after all.

If it helps I can also boot i915 on the primary machine (7d51), where I can compare against days of xe statistics on identical workloads.

Quoted 3 lines

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell.

Thanks for the correction - I will use H2G interrupt from here on and fix the terminology in v2.

Quoted 3 lines

It feels like when the
issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

That fits a pattern I can now see clearly with more data. Since enabling the bigger GuC logs (~1.5 days, 35 stalls): 12 stalls were unstuck by one of the H2G interrupts within 0.3-1.5s, but 23 ignored 8-9 consecutive interrupts and ran to the end. And in those severe cases the request-to-ack time is nearly constant: 2.28-2.34s, every single time. It does not look like congestion - it looks like a fixed internal timeout expiring somewhere and releasing things.

Related: an A/B experiment I ran earlier (holding forcewake across the whole GT, C6 residency pinned at 0ms for the whole window) still hit 9 timeouts in a row, so GT-level RC6 avoidance alone does not prevent it.

Quoted 1 line

Also, would you be able to capture the GuC logs when the issue occurs?

Done. I rebuilt with the debug-sized log buffers (8M event data / 1M crash dump / 1M state capture) and xe.guc_log_level=3, and the series' patch 1 (devcoredump on TLB invalidation timeout) captures the GuC log at the exact moment the timeout fires. I attached three devcoredumps (4-8.7MB each, containing the full GuC log around severe stalls that ignored 8-9 H2G interrupts) to the gitlab issue:

https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

I have more captures if useful (9 so far).

Quoted 2 lines

I have pushed the latest GuC FW for MTL here in case you want to give it a go:

Downloaded and staged - I will switch to 70.72.1 via xe.guc_firmware_path today and report back with a few days of data (this machine currently reproduces 20-35 stalls/day under my normal workload, so the signal should be quick).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 20:50, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 244 lines · Tales A. Mendonça
Quoted 6 lines
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Quoted 6 lines

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

Quoted 6 lines

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell. I'm making this clarification because the GuC supports a separate per-context notification mechanism that is referred to as doorbell and which we currently do not implement in neither i915 nor Xe.

When receiving the H2G interrupt, the only thing that the GuC does is look into the CTB and process anything in there; however, you've said that the contents of the H2G CTB are processed immediately, so the follow up interrupt should result in the GuC just bailing out and doing nothing because there is no data to process. It feels like when the issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

I have pushed the latest GuC FW for MTL here in case you want to give it a go:
https://gitlab.com/dceraolo/drm-firmware/-/blob/f783b931555be057dafc2400b7eb4d445c953fec/i915/mtl_guc_70.72.1.bin
. You can override the GuC firmware used by the driver via the xe.guc_firmware_path modparam; the path is relative to /lib/firmware/ and the firmware needs to be in initramfs for the driver to find it at boot. Note that we haven't tested this image on MTL, so it might have unexpected results.

Also, would you be able to capture the GuC logs when the issue occurs? The default guc log size is relatively small, so you'd have to capture right when the issue happens. However, you can make them bigger by building the kernel with CONFIG_DRM_XE_DEBUG or by simply modifying the xe_guc_log.h file to pick the bigger size by default. If you go with the latter, please also set xe.guc_log_level=3 on the command line (this is automatically added by the kconfig).

Thanks,
Daniele

Quoted 181 lines

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 171 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

    https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 42 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

   drivers/gpu/drm/xe/xe_devcoredump.c     |  68 ++++++++++++
   drivers/gpu/drm/xe/xe_devcoredump.h     |   6 ++
   drivers/gpu/drm/xe/xe_tlb_inval.c       | 131
+++++++++++++++++++++++-
   drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
   4 files changed, 243 insertions(+), 4 deletions(-)
Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
#21 lore raw
Quoted 45 lines · Tales A. Mendonça
Quoted 2 lines

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Two data points on that, pulling in different directions:

* My second ARL machine (7dd1) has been running i915 for 7 days now, with the same GuC 70.53.0, and there is not a single TLB invalidation timeout or GuC error in its logs. Caveat: i915's TLB invalidation timeout is longer than xe's 2.25s, so short stalls could be silent there.

* However, Matt just pointed at i915 MTL issue 14469, which looks like the same problem on i915 - so it may not be xe-specific after all.

If it helps I can also boot i915 on the primary machine (7d51), where I can compare against days of xe statistics on identical workloads.

Quoted 3 lines

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell.

Thanks for the correction - I will use H2G interrupt from here on and fix the terminology in v2.

Quoted 3 lines

It feels like when the
issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

That fits a pattern I can now see clearly with more data. Since enabling the bigger GuC logs (~1.5 days, 35 stalls): 12 stalls were unstuck by one of the H2G interrupts within 0.3-1.5s, but 23 ignored 8-9 consecutive interrupts and ran to the end. And in those severe cases the request-to-ack time is nearly constant: 2.28-2.34s, every single time. It does not look like congestion - it looks like a fixed internal timeout expiring somewhere and releasing things.

Related: an A/B experiment I ran earlier (holding forcewake across the whole GT, C6 residency pinned at 0ms for the whole window) still hit 9 timeouts in a row, so GT-level RC6 avoidance alone does not prevent it.

Quoted 1 line

Also, would you be able to capture the GuC logs when the issue occurs?

Done. I rebuilt with the debug-sized log buffers (8M event data / 1M crash dump / 1M state capture) and xe.guc_log_level=3, and the series' patch 1 (devcoredump on TLB invalidation timeout) captures the GuC log at the exact moment the timeout fires. I attached three devcoredumps (4-8.7MB each, containing the full GuC log around severe stalls that ignored 8-9 H2G interrupts) to the gitlab issue:

Quoted 3 lines
   https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

I have more captures if useful (9 so far).

Do you happen to have the matching dmesg for those? I decoded the logs but there are thousands of invalidation calls in them; I looked at the last few but from the GuC POV they've all been handled quickly. Dmesg error should log exactly which invalidations were delayed from the driver POV so I can just look at what happens around those.
Also, I have noticed that the failures seems to all be on GT1. On MTL, there is a caching bug on GT1 and we do not implement the WA for that in Xe (Wa_22016122933). Not sure if this is the actual root cause, but the fact that it only happens on GT1 makes me suspicious (i.e., it is possible that the GuC is replying in time but CPU doesn't see the reply because the cache-line is not correctly updated, or vice versa). Not sure if there is an easy way to implement this in Xe to test, that's not really my field of expertise; maybe adding XE_BO_FLAG_FORCE_WC or XE_BO_FLAG_NEEDS_UC to the CTB allocation could work as a quick hack? But I'd like either Matt or Thomas to confirm.

Daniele

Quoted 250 lines
Quoted 3 lines

I have pushed the latest GuC FW for MTL here in case you want to give it a go:

Downloaded and staged - I will switch to 70.72.1 via xe.guc_firmware_path today and report back with a few days of data (this machine currently reproduces 20-35 stalls/day under my normal workload, so the signal should be quick).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 20:50, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 236 lines · Tales A. Mendonça
Quoted 6 lines
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Quoted 11 lines

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell. I'm making this clarification because the GuC supports a separate per-context notification mechanism that is referred to as doorbell and which we currently do not implement in neither i915 nor Xe.

When receiving the H2G interrupt, the only thing that the GuC does is look into the CTB and process anything in there; however, you've said that the contents of the H2G CTB are processed immediately, so the follow up interrupt should result in the GuC just bailing out and doing nothing because there is no data to process. It feels like when the issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

I have pushed the latest GuC FW for MTL here in case you want to give it a go:
https://gitlab.com/dceraolo/drm-firmware/-/blob/f783b931555be057dafc2400b7eb4d445c953fec/i915/mtl_guc_70.72.1.bin
. You can override the GuC firmware used by the driver via the xe.guc_firmware_path modparam; the path is relative to /lib/firmware/ and the firmware needs to be in initramfs for the driver to find it at boot. Note that we haven't tested this image on MTL, so it might have unexpected results.

Also, would you be able to capture the GuC logs when the issue occurs? The default guc log size is relatively small, so you'd have to capture right when the issue happens. However, you can make them bigger by building the kernel with CONFIG_DRM_XE_DEBUG or by simply modifying the xe_guc_log.h file to pick the bigger size by default. If you go with the latter, please also set xe.guc_log_level=3 on the command line (this is automatically added by the kconfig).

Thanks,
Daniele

Quoted 178 lines

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 170 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

     https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 42 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

drivers/gpu/drm/xe/xe_devcoredump.c | 68 ++++++++++++
drivers/gpu/drm/xe/xe_devcoredump.h | 6 ++
drivers/gpu/drm/xe/xe_tlb_inval.c | 131
+++++++++++++++++++++++-

    drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
    4 files changed, 243 insertions(+), 4 deletions(-)
#22 lore raw

Results from a full day on GuC 70.72.1 under my normal workload:

59 stall events / 20 fence timeouts in ~20.6h, worst request-to-ack=2346ms, severe episodes still ignoring 8-9 consecutive H2G interrupts.

Statistically identical to 70.53.0 - the firmware version changes nothing, which matches your decode (GuC handling everything in time). I attached to the gitlab issue the full dmesg of the day plus a fresh devcoredump taken on 70.72.1 (same debug-sized GuC log buffers), in case you want to confirm the decode on the new firmware too.

That leaves the GT1 cache-line theory (Wa_22016122933) as the main suspect. I have a kernel built with XE_BO_FLAG_NEEDS_UC added to both CTB allocations as you suggested (blunt version: both directions, all GTs - just to test causality) and I am booting it tonight, with the firmware reverted to stock 70.53.0 so only one variable changes. Results tomorrow.

If the stalls disappear, I would be happy to help turn this into a proper Wa_22016122933 implementation for xe, mirroring what i915 does (scoped to the media GT on the affected platforms) - with guidance from Matt/Thomas on the preferred shape.

Thanks,
Tales

Em qui., 6 de ago. de 2026 às 18:13, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 322 lines · Tales A. Mendonça
Quoted 45 lines
Quoted 2 lines

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Two data points on that, pulling in different directions:

* My second ARL machine (7dd1) has been running i915 for 7 days now, with the same GuC 70.53.0, and there is not a single TLB invalidation timeout or GuC error in its logs. Caveat: i915's TLB invalidation timeout is longer than xe's 2.25s, so short stalls could be silent there.

* However, Matt just pointed at i915 MTL issue 14469, which looks like the same problem on i915 - so it may not be xe-specific after all.

If it helps I can also boot i915 on the primary machine (7d51), where I can compare against days of xe statistics on identical workloads.

Quoted 3 lines

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell.

Thanks for the correction - I will use H2G interrupt from here on and fix the terminology in v2.

Quoted 3 lines

It feels like when the
issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

That fits a pattern I can now see clearly with more data. Since enabling the bigger GuC logs (~1.5 days, 35 stalls): 12 stalls were unstuck by one of the H2G interrupts within 0.3-1.5s, but 23 ignored 8-9 consecutive interrupts and ran to the end. And in those severe cases the request-to-ack time is nearly constant: 2.28-2.34s, every single time. It does not look like congestion - it looks like a fixed internal timeout expiring somewhere and releasing things.

Related: an A/B experiment I ran earlier (holding forcewake across the whole GT, C6 residency pinned at 0ms for the whole window) still hit 9 timeouts in a row, so GT-level RC6 avoidance alone does not prevent it.

Quoted 1 line

Also, would you be able to capture the GuC logs when the issue occurs?

Done. I rebuilt with the debug-sized log buffers (8M event data / 1M crash dump / 1M state capture) and xe.guc_log_level=3, and the series' patch 1 (devcoredump on TLB invalidation timeout) captures the GuC log at the exact moment the timeout fires. I attached three devcoredumps (4-8.7MB each, containing the full GuC log around severe stalls that ignored 8-9 H2G interrupts) to the gitlab issue:

Quoted 3 lines
   https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

I have more captures if useful (9 so far).

Do you happen to have the matching dmesg for those? I decoded the logs but there are thousands of invalidation calls in them; I looked at the last few but from the GuC POV they've all been handled quickly. Dmesg error should log exactly which invalidations were delayed from the driver POV so I can just look at what happens around those. Also, I have noticed that the failures seems to all be on GT1. On MTL, there is a caching bug on GT1 and we do not implement the WA for that in Xe (Wa_22016122933). Not sure if this is the actual root cause, but the fact that it only happens on GT1 makes me suspicious (i.e., it is possible that the GuC is replying in time but CPU doesn't see the reply because the cache-line is not correctly updated, or vice versa). Not sure if there is an easy way to implement this in Xe to test, that's not really my field of expertise; maybe adding XE_BO_FLAG_FORCE_WC or XE_BO_FLAG_NEEDS_UC to the CTB allocation could work as a quick hack? But I'd like either Matt or Thomas to confirm.

Daniele

Quoted 250 lines
Quoted 3 lines

I have pushed the latest GuC FW for MTL here in case you want to give it a go:

Downloaded and staged - I will switch to 70.72.1 via xe.guc_firmware_path today and report back with a few days of data (this machine currently reproduces 20-35 stalls/day under my normal workload, so the signal should be quick).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 20:50, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 236 lines · Tales A. Mendonça
Quoted 6 lines
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Quoted 11 lines

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell. I'm making this clarification because the GuC supports a separate per-context notification mechanism that is referred to as doorbell and which we currently do not implement in neither i915 nor Xe.

When receiving the H2G interrupt, the only thing that the GuC does is look into the CTB and process anything in there; however, you've said that the contents of the H2G CTB are processed immediately, so the follow up interrupt should result in the GuC just bailing out and doing nothing because there is no data to process. It feels like when the issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

I have pushed the latest GuC FW for MTL here in case you want to give it a go:
https://gitlab.com/dceraolo/drm-firmware/-/blob/f783b931555be057dafc2400b7eb4d445c953fec/i915/mtl_guc_70.72.1.bin
. You can override the GuC firmware used by the driver via the xe.guc_firmware_path modparam; the path is relative to /lib/firmware/ and the firmware needs to be in initramfs for the driver to find it at boot. Note that we haven't tested this image on MTL, so it might have unexpected results.

Also, would you be able to capture the GuC logs when the issue occurs? The default guc log size is relatively small, so you'd have to capture right when the issue happens. However, you can make them bigger by building the kernel with CONFIG_DRM_XE_DEBUG or by simply modifying the xe_guc_log.h file to pick the bigger size by default. If you go with the latter, please also set xe.guc_log_level=3 on the command line (this is automatically added by the kconfig).

Thanks,
Daniele

Quoted 178 lines

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 170 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

     https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 42 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

drivers/gpu/drm/xe/xe_devcoredump.c | 68 ++++++++++++
drivers/gpu/drm/xe/xe_devcoredump.h | 6 ++
drivers/gpu/drm/xe/xe_tlb_inval.c | 131
+++++++++++++++++++++++-

    drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
    4 files changed, 243 insertions(+), 4 deletions(-)
Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
#23 lore raw
In reply to Daniele Ceraolo Spurio

Results: the GT1 cache-line theory is confirmed.

With XE_BO_FLAG_NEEDS_UC added to both CTB allocations (blunt version, all GTs), firmware back to stock 70.53.0, same kernel otherwise:

~6h of my normal workload (sustained load 4-6), ~480k TLB invalidations per GT processed (per gt stats), ZERO stall events, ZERO fence timeouts.

For comparison, yesterday on the same workload (cached CTBs): 14 stalls in the first ~1h45 of use, 59 stalls / 20 timeouts over the day. Every previous kernel reproduced within the first hour, every day, for weeks.

So the GuC was replying in time all along - the CPU was reading a stale cache line of the G2H CTB, and the near-constant 2.3s "stall" was just how long the line took to get evicted naturally. That explains the whole signature: H2G direction fine, G2H CTB "empty" at timeout, H2G interrupt kicks unsticking only some episodes, firmware version irrelevant.

I am happy to send a proper Wa_22016122933 implementation for xe. Two questions on the preferred shape before I do:

1. Scoping: mirror i915 (media GT only on the affected platforms), or is there a reason to cover both GTs on MTL/ARL? My blunt test cannot distinguish - Daniele's decode showed failures only on GT1, but I applied UC everywhere.

2. Coverage: is the CTB buffer enough, or should the descriptor and/or the GuC log buffer get the same treatment (wherever the CPU polls GuC-written memory)?

I will keep running the UC kernel and report the full-day numbers, but after weeks of daily reproduction, six hours at zero with ~960k invalidations processed is already a very strong signal.

Thanks,
Tales

Em qui., 6 de ago. de 2026 às 18:13, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 322 lines · Tales A. Mendonça
Quoted 45 lines
Quoted 2 lines

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Two data points on that, pulling in different directions:

* My second ARL machine (7dd1) has been running i915 for 7 days now, with the same GuC 70.53.0, and there is not a single TLB invalidation timeout or GuC error in its logs. Caveat: i915's TLB invalidation timeout is longer than xe's 2.25s, so short stalls could be silent there.

* However, Matt just pointed at i915 MTL issue 14469, which looks like the same problem on i915 - so it may not be xe-specific after all.

If it helps I can also boot i915 on the primary machine (7d51), where I can compare against days of xe statistics on identical workloads.

Quoted 3 lines

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell.

Thanks for the correction - I will use H2G interrupt from here on and fix the terminology in v2.

Quoted 3 lines

It feels like when the
issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

That fits a pattern I can now see clearly with more data. Since enabling the bigger GuC logs (~1.5 days, 35 stalls): 12 stalls were unstuck by one of the H2G interrupts within 0.3-1.5s, but 23 ignored 8-9 consecutive interrupts and ran to the end. And in those severe cases the request-to-ack time is nearly constant: 2.28-2.34s, every single time. It does not look like congestion - it looks like a fixed internal timeout expiring somewhere and releasing things.

Related: an A/B experiment I ran earlier (holding forcewake across the whole GT, C6 residency pinned at 0ms for the whole window) still hit 9 timeouts in a row, so GT-level RC6 avoidance alone does not prevent it.

Quoted 1 line

Also, would you be able to capture the GuC logs when the issue occurs?

Done. I rebuilt with the debug-sized log buffers (8M event data / 1M crash dump / 1M state capture) and xe.guc_log_level=3, and the series' patch 1 (devcoredump on TLB invalidation timeout) captures the GuC log at the exact moment the timeout fires. I attached three devcoredumps (4-8.7MB each, containing the full GuC log around severe stalls that ignored 8-9 H2G interrupts) to the gitlab issue:

Quoted 3 lines
   https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

I have more captures if useful (9 so far).

Do you happen to have the matching dmesg for those? I decoded the logs but there are thousands of invalidation calls in them; I looked at the last few but from the GuC POV they've all been handled quickly. Dmesg error should log exactly which invalidations were delayed from the driver POV so I can just look at what happens around those. Also, I have noticed that the failures seems to all be on GT1. On MTL, there is a caching bug on GT1 and we do not implement the WA for that in Xe (Wa_22016122933). Not sure if this is the actual root cause, but the fact that it only happens on GT1 makes me suspicious (i.e., it is possible that the GuC is replying in time but CPU doesn't see the reply because the cache-line is not correctly updated, or vice versa). Not sure if there is an easy way to implement this in Xe to test, that's not really my field of expertise; maybe adding XE_BO_FLAG_FORCE_WC or XE_BO_FLAG_NEEDS_UC to the CTB allocation could work as a quick hack? But I'd like either Matt or Thomas to confirm.

Daniele

Quoted 250 lines
Quoted 3 lines

I have pushed the latest GuC FW for MTL here in case you want to give it a go:

Downloaded and staged - I will switch to 70.72.1 via xe.guc_firmware_path today and report back with a few days of data (this machine currently reproduces 20-35 stalls/day under my normal workload, so the signal should be quick).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 20:50, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 236 lines · Tales A. Mendonça
Quoted 6 lines
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Quoted 11 lines

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell. I'm making this clarification because the GuC supports a separate per-context notification mechanism that is referred to as doorbell and which we currently do not implement in neither i915 nor Xe.

When receiving the H2G interrupt, the only thing that the GuC does is look into the CTB and process anything in there; however, you've said that the contents of the H2G CTB are processed immediately, so the follow up interrupt should result in the GuC just bailing out and doing nothing because there is no data to process. It feels like when the issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

I have pushed the latest GuC FW for MTL here in case you want to give it a go:
https://gitlab.com/dceraolo/drm-firmware/-/blob/f783b931555be057dafc2400b7eb4d445c953fec/i915/mtl_guc_70.72.1.bin
. You can override the GuC firmware used by the driver via the xe.guc_firmware_path modparam; the path is relative to /lib/firmware/ and the firmware needs to be in initramfs for the driver to find it at boot. Note that we haven't tested this image on MTL, so it might have unexpected results.

Also, would you be able to capture the GuC logs when the issue occurs? The default guc log size is relatively small, so you'd have to capture right when the issue happens. However, you can make them bigger by building the kernel with CONFIG_DRM_XE_DEBUG or by simply modifying the xe_guc_log.h file to pick the bigger size by default. If you go with the latter, please also set xe.guc_log_level=3 on the command line (this is automatically added by the kconfig).

Thanks,
Daniele

Quoted 178 lines

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 170 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

     https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 42 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

drivers/gpu/drm/xe/xe_devcoredump.c | 68 ++++++++++++
drivers/gpu/drm/xe/xe_devcoredump.h | 6 ++
drivers/gpu/drm/xe/xe_tlb_inval.c | 131
+++++++++++++++++++++++-

    drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
    4 files changed, 243 insertions(+), 4 deletions(-)
Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
maintainer
#24 lore raw
Quoted 8 lines · Tales A. Mendonça

Results: the GT1 cache-line theory is confirmed.

With XE_BO_FLAG_NEEDS_UC added to both CTB allocations (blunt version, all GTs), firmware back to stock 70.53.0, same kernel otherwise:

~6h of my normal workload (sustained load 4-6), ~480k TLB invalidations per GT processed (per gt stats), ZERO stall events, ZERO fence timeouts.

Awesome.

Quoted 12 lines

For comparison, yesterday on the same workload (cached CTBs): 14 stalls in the first ~1h45 of use, 59 stalls / 20 timeouts over the day. Every previous kernel reproduced within the first hour, every day, for weeks.

So the GuC was replying in time all along - the CPU was reading a stale cache line of the G2H CTB, and the near-constant 2.3s "stall" was just how long the line took to get evicted naturally. That explains the whole signature: H2G direction fine, G2H CTB "empty" at timeout, H2G interrupt kicks unsticking only some episodes, firmware version irrelevant.

Quoted 8 lines

I am happy to send a proper Wa_22016122933 implementation for xe. Two questions on the preferred shape before I do:

1. Scoping: mirror i915 (media GT only on the affected platforms), or is there a reason to cover both GTs on MTL/ARL? My blunt test cannot distinguish - Daniele's decode showed failures only on GT1, but I applied UC everywhere.

I'd scope this to mirror the i915 W/A.

Quoted 3 lines

2. Coverage: is the CTB buffer enough, or should the descriptor and/or the GuC log buffer get the same treatment (wherever the CPU polls GuC-written memory)?

I don't know enough about the W/A offhand so say, but again I'd just do whatever we did on the i915 in Xe.

Matt

Quoted 340 lines

I will keep running the UC kernel and report the full-day numbers, but after weeks of daily reproduction, six hours at zero with ~960k invalidations processed is already a very strong signal.

Thanks,
Tales

Em qui., 6 de ago. de 2026 às 18:13, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 322 lines · Tales A. Mendonça
Quoted 45 lines
Quoted 2 lines

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Two data points on that, pulling in different directions:

* My second ARL machine (7dd1) has been running i915 for 7 days now, with the same GuC 70.53.0, and there is not a single TLB invalidation timeout or GuC error in its logs. Caveat: i915's TLB invalidation timeout is longer than xe's 2.25s, so short stalls could be silent there.

* However, Matt just pointed at i915 MTL issue 14469, which looks like the same problem on i915 - so it may not be xe-specific after all.

If it helps I can also boot i915 on the primary machine (7d51), where I can compare against days of xe statistics on identical workloads.

Quoted 3 lines

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell.

Thanks for the correction - I will use H2G interrupt from here on and fix the terminology in v2.

Quoted 3 lines

It feels like when the
issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

That fits a pattern I can now see clearly with more data. Since enabling the bigger GuC logs (~1.5 days, 35 stalls): 12 stalls were unstuck by one of the H2G interrupts within 0.3-1.5s, but 23 ignored 8-9 consecutive interrupts and ran to the end. And in those severe cases the request-to-ack time is nearly constant: 2.28-2.34s, every single time. It does not look like congestion - it looks like a fixed internal timeout expiring somewhere and releasing things.

Related: an A/B experiment I ran earlier (holding forcewake across the whole GT, C6 residency pinned at 0ms for the whole window) still hit 9 timeouts in a row, so GT-level RC6 avoidance alone does not prevent it.

Quoted 1 line

Also, would you be able to capture the GuC logs when the issue occurs?

Done. I rebuilt with the debug-sized log buffers (8M event data / 1M crash dump / 1M state capture) and xe.guc_log_level=3, and the series' patch 1 (devcoredump on TLB invalidation timeout) captures the GuC log at the exact moment the timeout fires. I attached three devcoredumps (4-8.7MB each, containing the full GuC log around severe stalls that ignored 8-9 H2G interrupts) to the gitlab issue:

Quoted 3 lines
   https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

I have more captures if useful (9 so far).

Do you happen to have the matching dmesg for those? I decoded the logs but there are thousands of invalidation calls in them; I looked at the last few but from the GuC POV they've all been handled quickly. Dmesg error should log exactly which invalidations were delayed from the driver POV so I can just look at what happens around those. Also, I have noticed that the failures seems to all be on GT1. On MTL, there is a caching bug on GT1 and we do not implement the WA for that in Xe (Wa_22016122933). Not sure if this is the actual root cause, but the fact that it only happens on GT1 makes me suspicious (i.e., it is possible that the GuC is replying in time but CPU doesn't see the reply because the cache-line is not correctly updated, or vice versa). Not sure if there is an easy way to implement this in Xe to test, that's not really my field of expertise; maybe adding XE_BO_FLAG_FORCE_WC or XE_BO_FLAG_NEEDS_UC to the CTB allocation could work as a quick hack? But I'd like either Matt or Thomas to confirm.

Daniele

Quoted 250 lines
Quoted 3 lines

I have pushed the latest GuC FW for MTL here in case you want to give it a go:

Downloaded and staged - I will switch to 70.72.1 via xe.guc_firmware_path today and report back with a few days of data (this machine currently reproduces 20-35 stalls/day under my normal workload, so the signal should be quick).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 20:50, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 236 lines · Tales A. Mendonça
Quoted 6 lines
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Quoted 11 lines

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell. I'm making this clarification because the GuC supports a separate per-context notification mechanism that is referred to as doorbell and which we currently do not implement in neither i915 nor Xe.

When receiving the H2G interrupt, the only thing that the GuC does is look into the CTB and process anything in there; however, you've said that the contents of the H2G CTB are processed immediately, so the follow up interrupt should result in the GuC just bailing out and doing nothing because there is no data to process. It feels like when the issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

I have pushed the latest GuC FW for MTL here in case you want to give it a go:
https://gitlab.com/dceraolo/drm-firmware/-/blob/f783b931555be057dafc2400b7eb4d445c953fec/i915/mtl_guc_70.72.1.bin
. You can override the GuC firmware used by the driver via the xe.guc_firmware_path modparam; the path is relative to /lib/firmware/ and the firmware needs to be in initramfs for the driver to find it at boot. Note that we haven't tested this image on MTL, so it might have unexpected results.

Also, would you be able to capture the GuC logs when the issue occurs? The default guc log size is relatively small, so you'd have to capture right when the issue happens. However, you can make them bigger by building the kernel with CONFIG_DRM_XE_DEBUG or by simply modifying the xe_guc_log.h file to pick the bigger size by default. If you go with the latter, please also set xe.guc_log_level=3 on the command line (this is automatically added by the kconfig).

Thanks,
Daniele

Quoted 178 lines

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 170 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

     https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 42 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

drivers/gpu/drm/xe/xe_devcoredump.c | 68 ++++++++++++
drivers/gpu/drm/xe/xe_devcoredump.h | 6 ++
drivers/gpu/drm/xe/xe_tlb_inval.c | 131
+++++++++++++++++++++++-

    drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
    4 files changed, 243 insertions(+), 4 deletions(-)
Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
#25 lore raw
Quoted 4 lines

I'd scope this to mirror the i915 W/A.

I don't know enough about the W/A offhand so say, but again I'd just do whatever we did on the i915 in Xe.

Done - v2 is out with the Wa_22016122933 implementation mirroring i915:
scoped via an OOB rule to the media GT on media version 13.00, with coverage modeled on intel_guc_allocate_vma() (CTBs, GuC log, ADS and SLPC shared data):

https://lore.kernel.org/intel-xe/[email protected]/

Validated on both ARL machines here for several days - details in the cover letter.

Thanks,
Tales

Em ter., 11 de ago. de 2026 às 21:52, Matthew Brost <[email protected]> escreveu:

Quoted 385 lines · Tales A. Mendonça
Quoted 8 lines

Results: the GT1 cache-line theory is confirmed.

With XE_BO_FLAG_NEEDS_UC added to both CTB allocations (blunt version, all GTs), firmware back to stock 70.53.0, same kernel otherwise:

~6h of my normal workload (sustained load 4-6), ~480k TLB invalidations per GT processed (per gt stats), ZERO stall events, ZERO fence timeouts.

Awesome.

Quoted 12 lines

For comparison, yesterday on the same workload (cached CTBs): 14 stalls in the first ~1h45 of use, 59 stalls / 20 timeouts over the day. Every previous kernel reproduced within the first hour, every day, for weeks.

So the GuC was replying in time all along - the CPU was reading a stale cache line of the G2H CTB, and the near-constant 2.3s "stall" was just how long the line took to get evicted naturally. That explains the whole signature: H2G direction fine, G2H CTB "empty" at timeout, H2G interrupt kicks unsticking only some episodes, firmware version irrelevant.

Quoted 8 lines

I am happy to send a proper Wa_22016122933 implementation for xe. Two questions on the preferred shape before I do:

1. Scoping: mirror i915 (media GT only on the affected platforms), or is there a reason to cover both GTs on MTL/ARL? My blunt test cannot distinguish - Daniele's decode showed failures only on GT1, but I applied UC everywhere.

I'd scope this to mirror the i915 W/A.

Quoted 3 lines

2. Coverage: is the CTB buffer enough, or should the descriptor and/or the GuC log buffer get the same treatment (wherever the CPU polls GuC-written memory)?

I don't know enough about the W/A offhand so say, but again I'd just do whatever we did on the i915 in Xe.

Matt

Quoted 340 lines

I will keep running the UC kernel and report the full-day numbers, but after weeks of daily reproduction, six hours at zero with ~960k invalidations processed is already a very strong signal.

Thanks,
Tales

Em qui., 6 de ago. de 2026 às 18:13, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 69 lines · Tales A. Mendonça
Quoted 45 lines
Quoted 2 lines

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Two data points on that, pulling in different directions:

* My second ARL machine (7dd1) has been running i915 for 7 days now, with the same GuC 70.53.0, and there is not a single TLB invalidation timeout or GuC error in its logs. Caveat: i915's TLB invalidation timeout is longer than xe's 2.25s, so short stalls could be silent there.

* However, Matt just pointed at i915 MTL issue 14469, which looks like the same problem on i915 - so it may not be xe-specific after all.

If it helps I can also boot i915 on the primary machine (7d51), where I can compare against days of xe statistics on identical workloads.

Quoted 3 lines

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell.

Thanks for the correction - I will use H2G interrupt from here on and fix the terminology in v2.

Quoted 3 lines

It feels like when the
issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

That fits a pattern I can now see clearly with more data. Since enabling the bigger GuC logs (~1.5 days, 35 stalls): 12 stalls were unstuck by one of the H2G interrupts within 0.3-1.5s, but 23 ignored 8-9 consecutive interrupts and ran to the end. And in those severe cases the request-to-ack time is nearly constant: 2.28-2.34s, every single time. It does not look like congestion - it looks like a fixed internal timeout expiring somewhere and releasing things.

Related: an A/B experiment I ran earlier (holding forcewake across the whole GT, C6 residency pinned at 0ms for the whole window) still hit 9 timeouts in a row, so GT-level RC6 avoidance alone does not prevent it.

Quoted 1 line

Also, would you be able to capture the GuC logs when the issue occurs?

Done. I rebuilt with the debug-sized log buffers (8M event data / 1M crash dump / 1M state capture) and xe.guc_log_level=3, and the series' patch 1 (devcoredump on TLB invalidation timeout) captures the GuC log at the exact moment the timeout fires. I attached three devcoredumps (4-8.7MB each, containing the full GuC log around severe stalls that ignored 8-9 H2G interrupts) to the gitlab issue:

Quoted 3 lines
   https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678

I have more captures if useful (9 so far).

Do you happen to have the matching dmesg for those? I decoded the logs but there are thousands of invalidation calls in them; I looked at the last few but from the GuC POV they've all been handled quickly. Dmesg error should log exactly which invalidations were delayed from the driver POV so I can just look at what happens around those. Also, I have noticed that the failures seems to all be on GT1. On MTL, there is a caching bug on GT1 and we do not implement the WA for that in Xe (Wa_22016122933). Not sure if this is the actual root cause, but the fact that it only happens on GT1 makes me suspicious (i.e., it is possible that the GuC is replying in time but CPU doesn't see the reply because the cache-line is not correctly updated, or vice versa). Not sure if there is an easy way to implement this in Xe to test, that's not really my field of expertise; maybe adding XE_BO_FLAG_FORCE_WC or XE_BO_FLAG_NEEDS_UC to the CTB allocation could work as a quick hack? But I'd like either Matt or Thomas to confirm.

Quoted 253 lines

Daniele

Quoted 250 lines
Quoted 3 lines

I have pushed the latest GuC FW for MTL here in case you want to give it a go:

Downloaded and staged - I will switch to 70.72.1 via xe.guc_firmware_path today and report back with a few days of data (this machine currently reproduces 20-35 stalls/day under my normal workload, so the signal should be quick).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 20:50, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 236 lines · Tales A. Mendonça
Quoted 6 lines
Quoted 1 line

Are we seeing this on i915?

I have not tested i915 on the affected machines yet - the instrumentation that measured the stalls (late-ack logging, kick results) is xe-only, so I have no comparable i915 data. I can boot one of the ARL machines with i915 for a few days and watch for TLB invalidation timeouts there, if that data helps.

That would definitely help, because if the issue does not happen on i915 it likely means that we're missing a WA or something like that in Xe.

Quoted 11 lines

More generally: both machines here reproduce reliably (~1 stall/hour on a desktop workload, much more under memory pressure), so I am happy to test anything on them - including any GuC build the firmware team would like data on.

On Stuart's masking concern: fully agreed, that is why patch 3 is marked RFC. Patches 1-2 are pure diagnostics and stand on their own; I am fine holding patch 3 until the firmware side has been looked at. The data point it adds is that a doorbell ring unblocks the ack in the majority of episodes, while the severe ones ignore 8-9 consecutive rings - hopefully that narrows where to look inside the GuC.

Just a bit of a terminology update here, to make sure we're on the same
page: we usually refer to the notification you're sending to the GuC as an H2G interrupt and not a doorbell. I'm making this clarification because the GuC supports a separate per-context notification mechanism that is referred to as doorbell and which we currently do not implement in neither i915 nor Xe.

When receiving the H2G interrupt, the only thing that the GuC does is look into the CTB and process anything in there; however, you've said that the contents of the H2G CTB are processed immediately, so the follow up interrupt should result in the GuC just bailing out and doing nothing because there is no data to process. It feels like when the issue occurs something is stuck in HW rather than GuC FW and triggering the interrupt causes the HW to get unstuck.

I have pushed the latest GuC FW for MTL here in case you want to give it a go:
https://gitlab.com/dceraolo/drm-firmware/-/blob/f783b931555be057dafc2400b7eb4d445c953fec/i915/mtl_guc_70.72.1.bin
. You can override the GuC firmware used by the driver via the xe.guc_firmware_path modparam; the path is relative to /lib/firmware/ and the firmware needs to be in initramfs for the driver to find it at boot. Note that we haven't tested this image on MTL, so it might have unexpected results.

Also, would you be able to capture the GuC logs when the issue occurs? The default guc log size is relatively small, so you'd have to capture right when the issue happens. However, you can make them bigger by building the kernel with CONFIG_DRM_XE_DEBUG or by simply modifying the xe_guc_log.h file to pick the bigger size by default. If you go with the latter, please also set xe.guc_log_level=3 on the command line (this is automatically added by the kconfig).

Thanks,
Daniele

Quoted 178 lines

I will send a v2 addressing Matt's review comments (the __xe_devcoredump unification and the fixes on patch 2).

Thanks,
Tales

Em ter., 4 de ago. de 2026 às 19:08, Daniele Ceraolo Spurio <[email protected]> escreveu:

Quoted 170 lines · Stuart
Quoted 107 lines · Matthew Brost
Quoted 36 lines · Stuart
Quoted 24 lines · Tales A. Mendonça
Quoted 8 lines

Hi,

This series is a follow-up to the TLB invalidation ack stall I have
been debugging on ARL, tracked in:

     https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8678
Quoted 15 lines

Summary of the issue: with GuC 70.53.0 on ARL (reproduced on 7d51 and
7dd1 machines here, plus an independent Arc Pro 130T report on the
issue above), TLB invalidation acks intermittently stall for ~2.3s.
The H2G request is consumed from the CTB immediately and the G2H CTB
is empty the whole time - the firmware simply does not send the ack
until much later. The fence timeout fires at 2.25s and the ack lands
tens of ms after it. Userspace blocked on the invalidation (compositor
buffer unmaps etc.) hitches for the full window.

Quoted 8 lines

Firstly, thanks for the patch!

I haven't looked in to all the details of the sighting you were debugging, but we have had similar issues that were fixed in a later
GuC version. I think around 70.60.0? It might be worth trying on something later than that to see if that helps... (+Daniele)

I think this would require an AR on our end to make a new firmware version available.

Quoted 7 lines

The upstream repo only has 70.53.0 available for ARL [1] (iirc, ARL aliases to MTL for firmware). (+Julia too).

Presumably, the GuC changelogs should indicate whether an issue related
this has been fixed. If so, we need to update all GuC versions across both i915 and Xe.

Right... I guess I'd still like to see if we can test this in GuC (or get confirmation we can't for some reason) before committing something. My worry is we will prevent bug reports like this by working around it and miss critical bugs that need to be fixed in the right component.

Quoted 11 lines

[1]
https://gitlab.com/kernel-firmware/linux-firmware/-/blob/main/i915/mtl_guc_70.bin?ref_type=heads

Quoted 4 lines
Quoted 2 lines

Patch 1 adds xe_devcoredump_gt() so this kind of hang - which has no

Is there a reason we don't just re-use the main xe_devcoredump()?

This is my suggestion: the main devcoredump infrastructure is job- based,
so it cannot be used for hangs that are not associated with a job.

Quoted 6 lines

In my opinion, this is a gap on our end. Introducing something like xe_devcoredump_gt(), which can be used for non-job-based hangs (e.g.,
TLB invalidation timeouts like those addressed in this series, or more
generally any GuC protocol hang), makes sense to me.

Ok makes sense. We can do that here. It would be nice to have a more inclusive implementation that lets us call this from anywhere so we aren't duplicating things around for different use cases. But not a blocker here.

Quoted 31 lines

I haven't looked at the patch yet, but at a high level, adding xe_devcoredump_gt() seems like a reasonable approach.

Quoted 19 lines
Quoted 13 lines

exec queue or job to blame - leaves a devcoredump with the GuC log
and
CT state behind (Matt suggested capturing devcoredumps when we discussed the issue; devcoredumps from both machines are attached to
the issue above).

Patch 2 logs when the ack for a timed out invalidation finally arrives. This is what established that the acks are late rather than
lost.

Quoted 6 lines

Patch 3 is the RFC part: a delayed work that pokes the GuC (status
register read, CT flush, doorbell ring) every 250ms while an ack is
overdue. On my machines this converts the guaranteed 2.3s stall into

Quoted 9 lines

I'm a little worried we're just papering over something here that needs
to be addressed in GuC, particularly around GT going to sleep or something around the time we're expecting a response, so the pings on
registers might be prematurely waking things up which is something we'd
want to happen in GuC, not the KMD.

Quoted 5 lines

In general, I agree with this. We should avoid papering over the issue
and instead fix it properly in the GuC. That said, this workaround provides a pretty strong data point, since it appears to get the TLB invalidation unstuck.

Quoted 8 lines

So if we hit this issue I guess we're already going to have some performance degredation and the workaround makes that better. I need to look at the implementation, but we could be potentially introducing performance penalties in other areas doing these pings.

Again, I'd like to see if we can fix this in the right place before implementing a workaround for it. Hopefully Daniele or Julia can give some direction there.

Are we seeing this on i915 at all? Given that Xe does not officially support MTL/ARL and is missing several critical WAs for those platforms, the approach so far has been to only update the GuC FW if it is required for i915.
Looking at the GuC release notes, there have been a couple of TLB-related fixes after 70.53, but they're both marked as only affecting PVC and Xe2+ platforms, so no fixes seem to be available for ARL (or at least they're not listed in the release notes).

Daniele

Quoted 42 lines

Thanks,
Stuart

Quoted 39 lines

Matt

Quoted 37 lines

Thanks,
Stuart

Quoted 34 lines

a
sub-500ms hiccup for the majority of occurrences; a minority of severe
episodes ignore 8-9 consecutive doorbells, which points at the GuC
firmware being internally blocked for the whole window. Full data on
the issue. I am happy to rework the approach (different delay, tying it to the G2H handler, dropping the status read, etc.) - mainly
I would like the firmware side investigated, since no host-side poke
can fix the severe cases.

Based on drm-tip. Tested for several days on both ARL machines under
desktop and VM-heavy workloads.

Thanks,
Tales

Tales A. Mendonça (3):
drm/xe: Capture devcoredump on TLB invalidation timeout drm/xe: Log when a timed out TLB invalidation ack finally
arrives
drm/xe: Kick GuC while TLB invalidation acks are overdue

drivers/gpu/drm/xe/xe_devcoredump.c | 68 ++++++++++++
drivers/gpu/drm/xe/xe_devcoredump.h | 6 ++
drivers/gpu/drm/xe/xe_tlb_inval.c | 131
+++++++++++++++++++++++-

    drivers/gpu/drm/xe/xe_tlb_inval_types.h |  42 ++++++++
    4 files changed, 243 insertions(+), 4 deletions(-)

--
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org

Signature · 5 lines
Com os cumprimentos,

Tales A. Mendonça
talesam.org
communitybig.org
#26 lore raw

✗ CI.checkpatch: warning for drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)

In reply to Tales A. Mendonça
== Series Details ==

== Series Details ==

Series: drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)
URL : https://patchwork.freedesktop.org/series/171539/
State : warning

== Summary ==

+ KERNEL=/kernel
+ git clone https://gitlab.freedesktop.org/drm/maintainer-tools mt Cloning into 'mt'...
warning: redirecting to https://gitlab.freedesktop.org/drm/maintainer-tools.git/
+ git -C mt rev-list -n1 origin/master 061140b9bc586ae7f40abc1249c97e1cc72d1b9d
+ cd /kernel
+ git config --global --add safe.directory /kernel
+ git log -n1
commit 34edb36f8f11bba53549e0a81459f317a6fac286

Author: Tales A. Mendonça <[email protected]>
Date:   Mon Aug 3 23:14:41 2026 -0300

drm/xe: Kick GuC while TLB invalidation acks are overdue

On ARL with GuC 70.53.0, TLB invalidation acks intermittently stall:
the H2G request is consumed from the CTB immediately, but the G2H ack only arrives ~2.3s later, tens of milliseconds after the fence timeout has fired. During the stall the GPU keeps rendering; only work blocked on the invalidation (e.g. Wayland compositors performing buffer unmaps) hitches for the full 2.3s. Observed on three machines so far (7d51, 7dd1, plus an Arc Pro 130T report), see Link.

Experiments ruled out the obvious suspects:

* GT C6 parking: holding forcewake across the whole GT (C6 residency pinned at 0ms) still produced 9 timeouts in a row.
* Lost interrupt/CT processing on the host: the G2H CTB is empty at timeout time; the ack genuinely has not been sent by the firmware.

What does help is poking the GuC while the ack is overdue. Add a delayed work that fires XE_TLB_INVAL_KICK_DELAY_MS after an invalidation is issued and, while any ack is pending, re-reads the GuC status register, flushes the CT fast-path and rings the GuC doorbell (xe_guc_notify()), re-arming itself until the ack arrives; the existing TDR still bounds the total wait.

Instrumented results from two ARL machines over several days:

* Without kicks: every stall lasts the full ~2.3s and is reported as a fence timeout (-ETIME), ~1/hour on a desktop workload.
* With kicks: the majority of stalls resolve 15-276ms after one of the kicks, e.g.:

TLB invalidation ack after kick: seqno=36405 recv=36405, request-to-ack=528ms, last-kick-to-ack=15ms, kicks=2

* A minority of severe episodes ignore 8-9 consecutive doorbells and still run to the timeout (worst observed: ack 5996ms after request, 3.7s after the last kick), suggesting the firmware is internally blocked for the whole window rather than missing a wake event.

This is a workaround, not a fix - the root cause looks like a GuC firmware issue - but it turns a guaranteed 2.3s stall into a sub-500ms hiccup for most occurrences, and the "ack after kick" log documents the firmware behavior for further debugging.

+ /mt/dim checkpatch 4c13a67db118ee0019231dbc1362bcb8115ac660 drm-intel
cc6ee7dbdd9a drm/xe: Capture devcoredump on TLB invalidation timeout
d5341199c3eb drm/xe: Log when a timed out TLB invalidation ack finally arrives
-:24: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?) #24:
TLB invalidation late ack: seqno=10992 recv=10992, request-to-ack=2314ms, timeout-to-ack=45ms

total: 0 errors, 1 warnings, 0 checks, 57 lines checked
34edb36f8f11 drm/xe: Kick GuC while TLB invalidation acks are overdue
-:38: WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?) #38:

     TLB invalidation ack after kick: seqno=36405 recv=36405, request-to-ack=528ms, last-kick-to-ack=15ms, kicks=2

total: 0 errors, 1 warnings, 0 checks, 194 lines checked

#27 lore raw

✓ CI.KUnit: success for drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)

In reply to Tales A. Mendonça
== Series Details ==

== Series Details ==

Series: drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)
URL : https://patchwork.freedesktop.org/series/171539/
State : success

== Summary ==

+ trap cleanup EXIT
+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/xe/.kunitconfig [12:32:46] Configuring KUnit Kernel ...
Generating .config ...
Populating config with:

$ make ARCH=um O=.kunit olddefconfig
[12:32:51] Building KUnit Kernel ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig
Building with:
$ make all compile_commands.json scripts_gdb ARCH=um O=.kunit --jobs=48
[12:33:22] Starting KUnit Kernel (1/1)...
[12:33:22] ============================================================
Running tests with:
$ .kunit/linux kunit.enable=1 mem=1G console=tty kunit_shutdown=halt
[12:33:22] ================== guc_buf (11 subtests) ===================
[12:33:22] [PASSED] test_smallest
[12:33:22] [PASSED] test_largest
[12:33:22] [PASSED] test_granular
[12:33:22] [PASSED] test_unique
[12:33:22] [PASSED] test_overlap
[12:33:22] [PASSED] test_reusable
[12:33:22] [PASSED] test_too_big
[12:33:22] [PASSED] test_flush
[12:33:22] [PASSED] test_lookup
[12:33:22] [PASSED] test_data
[12:33:22] [PASSED] test_class
[12:33:22] ===================== [PASSED] guc_buf =====================
[12:33:22] =================== guc_dbm (7 subtests) ===================
[12:33:22] [PASSED] test_empty
[12:33:22] [PASSED] test_default
[12:33:22] ======================== test_size  ========================
[12:33:22] [PASSED] 4
[12:33:23] [PASSED] 8
[12:33:23] [PASSED] 32
[12:33:23] [PASSED] 256
[12:33:23] ==================== [PASSED] test_size ====================
[12:33:23] ======================= test_reuse  ========================
[12:33:23] [PASSED] 4
[12:33:23] [PASSED] 8
[12:33:23] [PASSED] 32
[12:33:23] [PASSED] 256
[12:33:23] =================== [PASSED] test_reuse ====================
[12:33:23] =================== test_range_overlap  ====================
[12:33:23] [PASSED] 4
[12:33:23] [PASSED] 8
[12:33:23] [PASSED] 32
[12:33:23] [PASSED] 256
[12:33:23] =============== [PASSED] test_range_overlap ================
[12:33:23] =================== test_range_compact  ====================
[12:33:23] [PASSED] 4
[12:33:23] [PASSED] 8
[12:33:23] [PASSED] 32
[12:33:23] [PASSED] 256
[12:33:23] =============== [PASSED] test_range_compact ================
[12:33:23] ==================== test_range_spare  =====================
[12:33:23] [PASSED] 4
[12:33:23] [PASSED] 8
[12:33:23] [PASSED] 32
[12:33:23] [PASSED] 256
[12:33:23] ================ [PASSED] test_range_spare =================
[12:33:23] ===================== [PASSED] guc_dbm =====================
[12:33:23] =================== guc_idm (6 subtests) ===================
[12:33:23] [PASSED] bad_init
[12:33:23] [PASSED] no_init
[12:33:23] [PASSED] init_fini
[12:33:23] [PASSED] check_used
[12:33:23] [PASSED] check_quota
[12:33:23] [PASSED] check_all
[12:33:23] ===================== [PASSED] guc_idm =====================
[12:33:23] =============== guc_klv_helpers (9 subtests) ===============
[12:33:23] [PASSED] test_count
[12:33:23] [PASSED] test_encode_u32
[12:33:23] [PASSED] test_encode_u64
[12:33:23] [PASSED] test_encode_string
[12:33:23] [PASSED] test_encode_object_raw
[12:33:23] [PASSED] test_encode_object_klv
[12:33:23] [PASSED] test_encode_object_nested
[12:33:23] [PASSED] test_encode_object_basic
[12:33:23] [PASSED] test_print
[12:33:23] ================= [PASSED] guc_klv_helpers =================
[12:33:23] ================== no_relay (3 subtests) ===================
[12:33:23] [PASSED] xe_drops_guc2pf_if_not_ready
[12:33:23] [PASSED] xe_drops_guc2vf_if_not_ready
[12:33:23] [PASSED] xe_rejects_send_if_not_ready
[12:33:23] ==================== [PASSED] no_relay =====================
[12:33:23] ================== pf_relay (14 subtests) ==================
[12:33:23] [PASSED] pf_rejects_guc2pf_too_short
[12:33:23] [PASSED] pf_rejects_guc2pf_too_long
[12:33:23] [PASSED] pf_rejects_guc2pf_no_payload
[12:33:23] [PASSED] pf_fails_no_payload
[12:33:23] [PASSED] pf_fails_bad_origin
[12:33:23] [PASSED] pf_fails_bad_type
[12:33:23] [PASSED] pf_txn_reports_error
[12:33:23] [PASSED] pf_txn_sends_pf2guc
[12:33:23] [PASSED] pf_sends_pf2guc
[12:33:23] [SKIPPED] pf_loopback_nop (requires CONFIG_DRM_XE_DEBUG_SRIOV)
[12:33:23] [SKIPPED] pf_loopback_echo (requires CONFIG_DRM_XE_DEBUG_SRIOV)
[12:33:23] [SKIPPED] pf_loopback_fail (requires CONFIG_DRM_XE_DEBUG_SRIOV)
[12:33:23] [SKIPPED] pf_loopback_busy (requires CONFIG_DRM_XE_DEBUG_SRIOV)
[12:33:23] [SKIPPED] pf_loopback_retry (requires CONFIG_DRM_XE_DEBUG_SRIOV)
[12:33:23] ==================== [PASSED] pf_relay =====================
[12:33:23] ================== vf_relay (3 subtests) ===================
[12:33:23] [PASSED] vf_rejects_guc2vf_too_short
[12:33:23] [PASSED] vf_rejects_guc2vf_too_long
[12:33:23] [PASSED] vf_rejects_guc2vf_no_payload
[12:33:23] ==================== [PASSED] vf_relay =====================
[12:33:23] ================ pf_gt_config (9 subtests) =================
[12:33:23] [PASSED] fair_contexts_1vf
[12:33:23] [PASSED] fair_doorbells_1vf
[12:33:23] [PASSED] fair_ggtt_1vf
[12:33:23] ====================== fair_vram_1vf  ======================
[12:33:23] [PASSED] 3.50 GiB
[12:33:23] [PASSED] 11.5 GiB
[12:33:23] [PASSED] 15.5 GiB
[12:33:23] [PASSED] 31.5 GiB
[12:33:23] [PASSED] 63.5 GiB
[12:33:23] [PASSED] 1.91 GiB
[12:33:23] ================== [PASSED] fair_vram_1vf ==================
[12:33:23] ================ fair_vram_1vf_admin_only  =================
[12:33:23] [PASSED] 3.50 GiB
[12:33:23] [PASSED] 11.5 GiB
[12:33:23] [PASSED] 15.5 GiB
[12:33:23] [PASSED] 31.5 GiB
[12:33:23] [PASSED] 63.5 GiB
[12:33:23] [PASSED] 1.91 GiB
[12:33:23] ============ [PASSED] fair_vram_1vf_admin_only =============
[12:33:23] ====================== fair_contexts  ======================
[12:33:23] [PASSED] 1 VF
[12:33:23] [PASSED] 2 VFs
[12:33:23] [PASSED] 3 VFs
[12:33:23] [PASSED] 4 VFs
[12:33:23] [PASSED] 5 VFs
[12:33:23] [PASSED] 6 VFs
[12:33:23] [PASSED] 7 VFs
[12:33:23] [PASSED] 8 VFs
[12:33:23] [PASSED] 9 VFs
[12:33:23] [PASSED] 10 VFs
[12:33:23] [PASSED] 11 VFs
[12:33:23] [PASSED] 12 VFs
[12:33:23] [PASSED] 13 VFs
[12:33:23] [PASSED] 14 VFs
[12:33:23] [PASSED] 15 VFs
[12:33:23] [PASSED] 16 VFs
[12:33:23] [PASSED] 17 VFs
[12:33:23] [PASSED] 18 VFs
[12:33:23] [PASSED] 19 VFs
[12:33:23] [PASSED] 20 VFs
[12:33:23] [PASSED] 21 VFs
[12:33:23] [PASSED] 22 VFs
[12:33:23] [PASSED] 23 VFs
[12:33:23] [PASSED] 24 VFs
[12:33:23] [PASSED] 25 VFs
[12:33:23] [PASSED] 26 VFs
[12:33:23] [PASSED] 27 VFs
[12:33:23] [PASSED] 28 VFs
[12:33:23] [PASSED] 29 VFs
[12:33:23] [PASSED] 30 VFs
[12:33:23] [PASSED] 31 VFs
[12:33:23] [PASSED] 32 VFs
[12:33:23] [PASSED] 33 VFs
[12:33:23] [PASSED] 34 VFs
[12:33:23] [PASSED] 35 VFs
[12:33:23] [PASSED] 36 VFs
[12:33:23] [PASSED] 37 VFs
[12:33:23] [PASSED] 38 VFs
[12:33:23] [PASSED] 39 VFs
[12:33:23] [PASSED] 40 VFs
[12:33:23] [PASSED] 41 VFs
[12:33:23] [PASSED] 42 VFs
[12:33:23] [PASSED] 43 VFs
[12:33:23] [PASSED] 44 VFs
[12:33:23] [PASSED] 45 VFs
[12:33:23] [PASSED] 46 VFs
[12:33:23] [PASSED] 47 VFs
[12:33:23] [PASSED] 48 VFs
[12:33:23] [PASSED] 49 VFs
[12:33:23] [PASSED] 50 VFs
[12:33:23] [PASSED] 51 VFs
[12:33:23] [PASSED] 52 VFs
[12:33:23] [PASSED] 53 VFs
[12:33:23] [PASSED] 54 VFs
[12:33:23] [PASSED] 55 VFs
[12:33:23] [PASSED] 56 VFs
[12:33:23] [PASSED] 57 VFs
[12:33:23] [PASSED] 58 VFs
[12:33:23] [PASSED] 59 VFs
[12:33:23] [PASSED] 60 VFs
[12:33:23] [PASSED] 61 VFs
[12:33:23] [PASSED] 62 VFs
[12:33:23] [PASSED] 63 VFs
[12:33:23] ================== [PASSED] fair_contexts ==================
[12:33:23] ===================== fair_doorbells  ======================
[12:33:23] [PASSED] 1 VF
[12:33:23] [PASSED] 2 VFs
[12:33:23] [PASSED] 3 VFs
[12:33:23] [PASSED] 4 VFs
[12:33:23] [PASSED] 5 VFs
[12:33:23] [PASSED] 6 VFs
[12:33:23] [PASSED] 7 VFs
[12:33:23] [PASSED] 8 VFs
[12:33:23] [PASSED] 9 VFs
[12:33:23] [PASSED] 10 VFs
[12:33:23] [PASSED] 11 VFs
[12:33:23] [PASSED] 12 VFs
[12:33:23] [PASSED] 13 VFs
[12:33:23] [PASSED] 14 VFs
[12:33:23] [PASSED] 15 VFs
[12:33:23] [PASSED] 16 VFs
[12:33:23] [PASSED] 17 VFs
[12:33:23] [PASSED] 18 VFs
[12:33:23] [PASSED] 19 VFs
[12:33:23] [PASSED] 20 VFs
[12:33:23] [PASSED] 21 VFs
[12:33:23] [PASSED] 22 VFs
[12:33:23] [PASSED] 23 VFs
[12:33:23] [PASSED] 24 VFs
[12:33:23] [PASSED] 25 VFs
[12:33:23] [PASSED] 26 VFs
[12:33:23] [PASSED] 27 VFs
[12:33:23] [PASSED] 28 VFs
[12:33:23] [PASSED] 29 VFs
[12:33:23] [PASSED] 30 VFs
[12:33:23] [PASSED] 31 VFs
[12:33:23] [PASSED] 32 VFs
[12:33:23] [PASSED] 33 VFs
[12:33:23] [PASSED] 34 VFs
[12:33:23] [PASSED] 35 VFs
[12:33:23] [PASSED] 36 VFs
[12:33:23] [PASSED] 37 VFs
[12:33:23] [PASSED] 38 VFs
[12:33:23] [PASSED] 39 VFs
[12:33:23] [PASSED] 40 VFs
[12:33:23] [PASSED] 41 VFs
[12:33:23] [PASSED] 42 VFs
[12:33:23] [PASSED] 43 VFs
[12:33:23] [PASSED] 44 VFs
[12:33:23] [PASSED] 45 VFs
[12:33:23] [PASSED] 46 VFs
[12:33:23] [PASSED] 47 VFs
[12:33:23] [PASSED] 48 VFs
[12:33:23] [PASSED] 49 VFs
[12:33:23] [PASSED] 50 VFs
[12:33:23] [PASSED] 51 VFs
[12:33:23] [PASSED] 52 VFs
[12:33:23] [PASSED] 53 VFs
[12:33:23] [PASSED] 54 VFs
[12:33:23] [PASSED] 55 VFs
[12:33:23] [PASSED] 56 VFs
[12:33:23] [PASSED] 57 VFs
[12:33:23] [PASSED] 58 VFs
[12:33:23] [PASSED] 59 VFs
[12:33:23] [PASSED] 60 VFs
[12:33:23] [PASSED] 61 VFs
[12:33:23] [PASSED] 62 VFs
[12:33:23] [PASSED] 63 VFs
[12:33:23] ================= [PASSED] fair_doorbells ==================
[12:33:23] ======================== fair_ggtt  ========================
[12:33:23] [PASSED] 1 VF
[12:33:23] [PASSED] 2 VFs
[12:33:23] [PASSED] 3 VFs
[12:33:23] [PASSED] 4 VFs
[12:33:23] [PASSED] 5 VFs
[12:33:23] [PASSED] 6 VFs
[12:33:23] [PASSED] 7 VFs
[12:33:23] [PASSED] 8 VFs
[12:33:23] [PASSED] 9 VFs
[12:33:23] [PASSED] 10 VFs
[12:33:23] [PASSED] 11 VFs
[12:33:23] [PASSED] 12 VFs
[12:33:23] [PASSED] 13 VFs
[12:33:23] [PASSED] 14 VFs
[12:33:23] [PASSED] 15 VFs
[12:33:23] [PASSED] 16 VFs
[12:33:23] [PASSED] 17 VFs
[12:33:23] [PASSED] 18 VFs
[12:33:23] [PASSED] 19 VFs
[12:33:23] [PASSED] 20 VFs
[12:33:23] [PASSED] 21 VFs
[12:33:23] [PASSED] 22 VFs
[12:33:23] [PASSED] 23 VFs
[12:33:23] [PASSED] 24 VFs
[12:33:23] [PASSED] 25 VFs
[12:33:23] [PASSED] 26 VFs
[12:33:23] [PASSED] 27 VFs
[12:33:23] [PASSED] 28 VFs
[12:33:23] [PASSED] 29 VFs
[12:33:23] [PASSED] 30 VFs
[12:33:23] [PASSED] 31 VFs
[12:33:23] [PASSED] 32 VFs
[12:33:23] [PASSED] 33 VFs
[12:33:23] [PASSED] 34 VFs
[12:33:23] [PASSED] 35 VFs
[12:33:23] [PASSED] 36 VFs
[12:33:23] [PASSED] 37 VFs
[12:33:23] [PASSED] 38 VFs
[12:33:23] [PASSED] 39 VFs
[12:33:23] [PASSED] 40 VFs
[12:33:23] [PASSED] 41 VFs
[12:33:23] [PASSED] 42 VFs
[12:33:23] [PASSED] 43 VFs
[12:33:23] [PASSED] 44 VFs
[12:33:23] [PASSED] 45 VFs
[12:33:23] [PASSED] 46 VFs
[12:33:23] [PASSED] 47 VFs
[12:33:23] [PASSED] 48 VFs
[12:33:23] [PASSED] 49 VFs
[12:33:23] [PASSED] 50 VFs
[12:33:23] [PASSED] 51 VFs
[12:33:23] [PASSED] 52 VFs
[12:33:23] [PASSED] 53 VFs
[12:33:23] [PASSED] 54 VFs
[12:33:23] [PASSED] 55 VFs
[12:33:23] [PASSED] 56 VFs
[12:33:23] [PASSED] 57 VFs
[12:33:23] [PASSED] 58 VFs
[12:33:23] [PASSED] 59 VFs
[12:33:23] [PASSED] 60 VFs
[12:33:23] [PASSED] 61 VFs
[12:33:23] [PASSED] 62 VFs
[12:33:23] [PASSED] 63 VFs
[12:33:23] ==================== [PASSED] fair_ggtt ====================
[12:33:23] ======================== fair_vram  ========================
[12:33:23] [PASSED] 1 VF
[12:33:23] [PASSED] 2 VFs
[12:33:23] [PASSED] 3 VFs
[12:33:23] [PASSED] 4 VFs
[12:33:23] [PASSED] 5 VFs
[12:33:23] [PASSED] 6 VFs
[12:33:23] [PASSED] 7 VFs
[12:33:23] [PASSED] 8 VFs
[12:33:23] [PASSED] 9 VFs
[12:33:23] [PASSED] 10 VFs
[12:33:23] [PASSED] 11 VFs
[12:33:23] [PASSED] 12 VFs
[12:33:23] [PASSED] 13 VFs
[12:33:23] [PASSED] 14 VFs
[12:33:23] [PASSED] 15 VFs
[12:33:23] [PASSED] 16 VFs
[12:33:23] [PASSED] 17 VFs
[12:33:23] [PASSED] 18 VFs
[12:33:23] [PASSED] 19 VFs
[12:33:23] [PASSED] 20 VFs
[12:33:23] [PASSED] 21 VFs
[12:33:23] [PASSED] 22 VFs
[12:33:23] [PASSED] 23 VFs
[12:33:23] [PASSED] 24 VFs
[12:33:23] [PASSED] 25 VFs
[12:33:23] [PASSED] 26 VFs
[12:33:23] [PASSED] 27 VFs
[12:33:23] [PASSED] 28 VFs
[12:33:23] [PASSED] 29 VFs
[12:33:23] [PASSED] 30 VFs
[12:33:23] [PASSED] 31 VFs
[12:33:23] [PASSED] 32 VFs
[12:33:23] [PASSED] 33 VFs
[12:33:23] [PASSED] 34 VFs
[12:33:23] [PASSED] 35 VFs
[12:33:23] [PASSED] 36 VFs
[12:33:23] [PASSED] 37 VFs
[12:33:23] [PASSED] 38 VFs
[12:33:23] [PASSED] 39 VFs
[12:33:23] [PASSED] 40 VFs
[12:33:23] [PASSED] 41 VFs
[12:33:23] [PASSED] 42 VFs
[12:33:23] [PASSED] 43 VFs
[12:33:23] [PASSED] 44 VFs
[12:33:23] [PASSED] 45 VFs
[12:33:23] [PASSED] 46 VFs
[12:33:23] [PASSED] 47 VFs
[12:33:23] [PASSED] 48 VFs
[12:33:23] [PASSED] 49 VFs
[12:33:23] [PASSED] 50 VFs
[12:33:23] [PASSED] 51 VFs
[12:33:23] [PASSED] 52 VFs
[12:33:23] [PASSED] 53 VFs
[12:33:23] [PASSED] 54 VFs
[12:33:23] [PASSED] 55 VFs
[12:33:23] [PASSED] 56 VFs
[12:33:23] [PASSED] 57 VFs
[12:33:23] [PASSED] 58 VFs
[12:33:23] [PASSED] 59 VFs
[12:33:23] [PASSED] 60 VFs
[12:33:23] [PASSED] 61 VFs
[12:33:23] [PASSED] 62 VFs
[12:33:23] [PASSED] 63 VFs
[12:33:23] ==================== [PASSED] fair_vram ====================
[12:33:23] ================== [PASSED] pf_gt_config ===================
[12:33:23] ===================== lmtt (1 subtest) =====================
[12:33:23] ======================== test_ops  =========================
[12:33:23] [PASSED] 2-level
[12:33:23] [PASSED] multi-level
[12:33:23] ==================== [PASSED] test_ops =====================
[12:33:23] ====================== [PASSED] lmtt =======================
[12:33:23] ================= sriov_packet (1 subtest) =================
[12:33:23] [PASSED] test_descriptor_init
[12:33:23] ================== [PASSED] sriov_packet ===================
[12:33:23] ================= pf_service (11 subtests) =================
[12:33:23] [PASSED] pf_negotiate_any
[12:33:23] [PASSED] pf_negotiate_base_match
[12:33:23] [PASSED] pf_negotiate_base_newer
[12:33:23] [PASSED] pf_negotiate_base_next
[12:33:23] [SKIPPED] pf_negotiate_base_older (no older minor)
[12:33:23] [PASSED] pf_negotiate_base_prev
[12:33:23] [PASSED] pf_negotiate_latest_match
[12:33:23] [PASSED] pf_negotiate_latest_newer
[12:33:23] [PASSED] pf_negotiate_latest_next
[12:33:23] [SKIPPED] pf_negotiate_latest_older (no older minor)
[12:33:23] [SKIPPED] pf_negotiate_latest_prev (no prev major)
[12:33:23] =================== [PASSED] pf_service ====================
[12:33:23] ================= xe_guc_g2g (2 subtests) ==================
[12:33:23] ============== xe_live_guc_g2g_kunit_default  ==============
[12:33:23] ========= [SKIPPED] xe_live_guc_g2g_kunit_default ==========
[12:33:23] ============== xe_live_guc_g2g_kunit_allmem  ===============
[12:33:23] ========== [SKIPPED] xe_live_guc_g2g_kunit_allmem ==========
[12:33:23] =================== [SKIPPED] xe_guc_g2g ===================
[12:33:23] =================== xe_mocs (2 subtests) ===================
[12:33:23] ================ xe_live_mocs_kernel_kunit  ================
[12:33:23] =========== [SKIPPED] xe_live_mocs_kernel_kunit ============
[12:33:23] ================ xe_live_mocs_reset_kunit  =================
[12:33:23] ============ [SKIPPED] xe_live_mocs_reset_kunit ============
[12:33:23] ==================== [SKIPPED] xe_mocs =====================
[12:33:23] ================= xe_migrate (2 subtests) ==================
[12:33:23] ================= xe_migrate_sanity_kunit  =================
[12:33:23] ============ [SKIPPED] xe_migrate_sanity_kunit =============
[12:33:23] ================== xe_validate_ccs_kunit  ==================
[12:33:23] ============= [SKIPPED] xe_validate_ccs_kunit ==============
[12:33:23] =================== [SKIPPED] xe_migrate ===================
[12:33:23] ================== xe_dma_buf (1 subtest) ==================
[12:33:23] ==================== xe_dma_buf_kunit  =====================
[12:33:23] ================ [SKIPPED] xe_dma_buf_kunit ================
[12:33:23] =================== [SKIPPED] xe_dma_buf ===================
[12:33:23] ================= xe_bo_shrink (1 subtest) =================
[12:33:23] =================== xe_bo_shrink_kunit  ====================
[12:33:23] =============== [SKIPPED] xe_bo_shrink_kunit ===============
[12:33:23] ================== [SKIPPED] xe_bo_shrink ==================
[12:33:23] ==================== xe_bo (2 subtests) ====================
[12:33:23] ================== xe_ccs_migrate_kunit  ===================
[12:33:23] ============== [SKIPPED] xe_ccs_migrate_kunit ==============
[12:33:23] ==================== xe_bo_evict_kunit  ====================
[12:33:23] =============== [SKIPPED] xe_bo_evict_kunit ================
[12:33:23] ===================== [SKIPPED] xe_bo ======================
[12:33:23] ==================== args (13 subtests) ====================
[12:33:23] [PASSED] count_args_test
[12:33:23] [PASSED] call_args_example
[12:33:23] [PASSED] call_args_test
[12:33:23] [PASSED] drop_first_arg_example
[12:33:23] [PASSED] drop_first_arg_test
[12:33:23] [PASSED] first_arg_example
[12:33:23] [PASSED] first_arg_test
[12:33:23] [PASSED] last_arg_example
[12:33:23] [PASSED] last_arg_test
[12:33:23] [PASSED] pick_arg_example
[12:33:23] [PASSED] if_args_example
[12:33:23] [PASSED] if_args_test
[12:33:23] [PASSED] sep_comma_example
[12:33:23] ====================== [PASSED] args =======================
[12:33:23] =================== xe_pci (3 subtests) ====================
[12:33:23] ==================== check_graphics_ip  ====================
[12:33:23] [PASSED] 12.00 Xe_LP
[12:33:23] [PASSED] 12.10 Xe_LP+
[12:33:23] [PASSED] 12.55 Xe_HPG
[12:33:23] [PASSED] 12.60 Xe_HPC
[12:33:23] [PASSED] 12.70 Xe_LPG
[12:33:23] [PASSED] 12.71 Xe_LPG
[12:33:23] [PASSED] 12.74 Xe_LPG+
[12:33:23] [PASSED] 20.01 Xe2_HPG
[12:33:23] [PASSED] 20.02 Xe2_HPG
[12:33:23] [PASSED] 20.04 Xe2_LPG
[12:33:23] [PASSED] 30.00 Xe3_LPG
[12:33:23] [PASSED] 30.01 Xe3_LPG
[12:33:23] [PASSED] 30.03 Xe3_LPG
[12:33:23] [PASSED] 30.04 Xe3_LPG
[12:33:23] [PASSED] 30.05 Xe3_LPG
[12:33:23] [PASSED] 35.10 Xe3p_LPG
[12:33:23] [PASSED] 35.11 Xe3p_XPC
[12:33:23] ================ [PASSED] check_graphics_ip ================
[12:33:23] ===================== check_media_ip  ======================
[12:33:23] [PASSED] 12.00 Xe_M
[12:33:23] [PASSED] 12.55 Xe_HPM
[12:33:23] [PASSED] 13.00 Xe_LPM+
[12:33:23] [PASSED] 13.01 Xe2_HPM
[12:33:23] [PASSED] 20.00 Xe2_LPM
[12:33:23] [PASSED] 30.00 Xe3_LPM
[12:33:23] [PASSED] 30.02 Xe3_LPM
[12:33:23] [PASSED] 35.00 Xe3p_LPM
[12:33:23] [PASSED] 35.03 Xe3p_HPM
[12:33:23] ================= [PASSED] check_media_ip ==================
[12:33:23] =================== check_platform_desc  ===================
[12:33:23] [PASSED] 0x9A60 (TIGERLAKE)
[12:33:23] [PASSED] 0x9A68 (TIGERLAKE)
[12:33:23] [PASSED] 0x9A70 (TIGERLAKE)
[12:33:23] [PASSED] 0x9A40 (TIGERLAKE)
[12:33:23] [PASSED] 0x9A49 (TIGERLAKE)
[12:33:23] [PASSED] 0x9A59 (TIGERLAKE)
[12:33:23] [PASSED] 0x9A78 (TIGERLAKE)
[12:33:23] [PASSED] 0x9AC0 (TIGERLAKE)
[12:33:23] [PASSED] 0x9AC9 (TIGERLAKE)
[12:33:23] [PASSED] 0x9AD9 (TIGERLAKE)
[12:33:23] [PASSED] 0x9AF8 (TIGERLAKE)
[12:33:23] [PASSED] 0x4C80 (ROCKETLAKE)
[12:33:23] [PASSED] 0x4C8A (ROCKETLAKE)
[12:33:23] [PASSED] 0x4C8B (ROCKETLAKE)
[12:33:23] [PASSED] 0x4C8C (ROCKETLAKE)
[12:33:23] [PASSED] 0x4C90 (ROCKETLAKE)
[12:33:23] [PASSED] 0x4C9A (ROCKETLAKE)
[12:33:23] [PASSED] 0x4680 (ALDERLAKE_S)
[12:33:23] [PASSED] 0x4682 (ALDERLAKE_S)
[12:33:23] [PASSED] 0x4688 (ALDERLAKE_S)
[12:33:23] [PASSED] 0x468A (ALDERLAKE_S)
[12:33:23] [PASSED] 0x468B (ALDERLAKE_S)
[12:33:23] [PASSED] 0x4690 (ALDERLAKE_S)
[12:33:23] [PASSED] 0x4692 (ALDERLAKE_S)
[12:33:23] [PASSED] 0x4693 (ALDERLAKE_S)
[12:33:23] [PASSED] 0x46A0 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46A1 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46A2 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46A3 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46A6 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46A8 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46AA (ALDERLAKE_P)
[12:33:23] [PASSED] 0x462A (ALDERLAKE_P)
[12:33:23] [PASSED] 0x4626 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x4628 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46B0 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46B1 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46B2 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46B3 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46C0 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46C1 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46C2 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46C3 (ALDERLAKE_P)
[12:33:23] [PASSED] 0x46D0 (ALDERLAKE_N)
[12:33:23] [PASSED] 0x46D1 (ALDERLAKE_N)
[12:33:23] [PASSED] 0x46D2 (ALDERLAKE_N)
[12:33:23] [PASSED] 0x46D3 (ALDERLAKE_N)
[12:33:23] [PASSED] 0x46D4 (ALDERLAKE_N)
[12:33:23] [PASSED] 0xA721 (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7A1 (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7A9 (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7AC (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7AD (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA720 (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7A0 (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7A8 (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7AA (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA7AB (ALDERLAKE_P)
[12:33:23] [PASSED] 0xA780 (ALDERLAKE_S)
[12:33:23] [PASSED] 0xA781 (ALDERLAKE_S)
[12:33:23] [PASSED] 0xA782 (ALDERLAKE_S)
[12:33:23] [PASSED] 0xA783 (ALDERLAKE_S)
[12:33:23] [PASSED] 0xA788 (ALDERLAKE_S)
[12:33:23] [PASSED] 0xA789 (ALDERLAKE_S)
[12:33:23] [PASSED] 0xA78A (ALDERLAKE_S)
[12:33:23] [PASSED] 0xA78B (ALDERLAKE_S)
[12:33:23] [PASSED] 0x4905 (DG1)
[12:33:23] [PASSED] 0x4906 (DG1)
[12:33:23] [PASSED] 0x4907 (DG1)
[12:33:23] [PASSED] 0x4908 (DG1)
[12:33:23] [PASSED] 0x4909 (DG1)
[12:33:23] [PASSED] 0x56C0 (DG2)
[12:33:23] [PASSED] 0x56C2 (DG2)
[12:33:23] [PASSED] 0x56C1 (DG2)
[12:33:23] [PASSED] 0x7D51 (METEORLAKE)
[12:33:23] [PASSED] 0x7DD1 (METEORLAKE)
[12:33:23] [PASSED] 0x7D41 (METEORLAKE)
[12:33:23] [PASSED] 0x7D67 (METEORLAKE)
[12:33:23] [PASSED] 0xB640 (METEORLAKE)
[12:33:23] [PASSED] 0x56A0 (DG2)
[12:33:23] [PASSED] 0x56A1 (DG2)
[12:33:23] [PASSED] 0x56A2 (DG2)
[12:33:23] [PASSED] 0x56BE (DG2)
[12:33:23] [PASSED] 0x56BF (DG2)
[12:33:23] [PASSED] 0x5690 (DG2)
[12:33:23] [PASSED] 0x5691 (DG2)
[12:33:23] [PASSED] 0x5692 (DG2)
[12:33:23] [PASSED] 0x56A5 (DG2)
[12:33:23] [PASSED] 0x56A6 (DG2)
[12:33:23] [PASSED] 0x56B0 (DG2)
[12:33:23] [PASSED] 0x56B1 (DG2)
[12:33:23] [PASSED] 0x56BA (DG2)
[12:33:23] [PASSED] 0x56BB (DG2)
[12:33:23] [PASSED] 0x56BC (DG2)
[12:33:23] [PASSED] 0x56BD (DG2)
[12:33:23] [PASSED] 0x5693 (DG2)
[12:33:23] [PASSED] 0x5694 (DG2)
[12:33:23] [PASSED] 0x5695 (DG2)
[12:33:23] [PASSED] 0x56A3 (DG2)
[12:33:23] [PASSED] 0x56A4 (DG2)
[12:33:23] [PASSED] 0x56B2 (DG2)
[12:33:23] [PASSED] 0x56B3 (DG2)
[12:33:23] [PASSED] 0x5696 (DG2)
[12:33:23] [PASSED] 0x5697 (DG2)
[12:33:23] [PASSED] 0xB69 (PVC)
[12:33:23] [PASSED] 0xB6E (PVC)
[12:33:23] [PASSED] 0xBD4 (PVC)
[12:33:23] [PASSED] 0xBD5 (PVC)
[12:33:23] [PASSED] 0xBD6 (PVC)
[12:33:23] [PASSED] 0xBD7 (PVC)
[12:33:23] [PASSED] 0xBD8 (PVC)
[12:33:23] [PASSED] 0xBD9 (PVC)
[12:33:23] [PASSED] 0xBDA (PVC)
[12:33:23] [PASSED] 0xBDB (PVC)
[12:33:23] [PASSED] 0xBE0 (PVC)
[12:33:23] [PASSED] 0xBE1 (PVC)
[12:33:23] [PASSED] 0xBE5 (PVC)
[12:33:23] [PASSED] 0x7D40 (METEORLAKE)
[12:33:23] [PASSED] 0x7D45 (METEORLAKE)
[12:33:23] [PASSED] 0x7D55 (METEORLAKE)
[12:33:23] [PASSED] 0x7D60 (METEORLAKE)
[12:33:23] [PASSED] 0x7DD5 (METEORLAKE)
[12:33:23] [PASSED] 0x6420 (LUNARLAKE)
[12:33:23] [PASSED] 0x64A0 (LUNARLAKE)
[12:33:23] [PASSED] 0x64B0 (LUNARLAKE)
[12:33:23] [PASSED] 0xE202 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE209 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE20B (BATTLEMAGE)
[12:33:23] [PASSED] 0xE20C (BATTLEMAGE)
[12:33:23] [PASSED] 0xE20D (BATTLEMAGE)
[12:33:23] [PASSED] 0xE210 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE211 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE212 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE216 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE220 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE221 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE222 (BATTLEMAGE)
[12:33:23] [PASSED] 0xE223 (BATTLEMAGE)
[12:33:23] [PASSED] 0xB080 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB081 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB082 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB083 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB084 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB085 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB086 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB087 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB08F (PANTHERLAKE)
[12:33:23] [PASSED] 0xB090 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB0A0 (PANTHERLAKE)
[12:33:23] [PASSED] 0xB0B0 (PANTHERLAKE)
[12:33:23] [PASSED] 0xFD80 (PANTHERLAKE)
[12:33:23] [PASSED] 0xFD81 (PANTHERLAKE)
[12:33:23] [PASSED] 0xD740 (NOVALAKE_S)
[12:33:23] [PASSED] 0xD741 (NOVALAKE_S)
[12:33:23] [PASSED] 0xD742 (NOVALAKE_S)
[12:33:23] [PASSED] 0xD743 (NOVALAKE_S)
[12:33:23] [PASSED] 0xD745 (NOVALAKE_S)
[12:33:23] [PASSED] 0xD74A (NOVALAKE_S)
[12:33:23] [PASSED] 0xD74B (NOVALAKE_S)
[12:33:23] [PASSED] 0x674C (CRESCENTISLAND)
[12:33:23] [PASSED] 0x674D (CRESCENTISLAND)
[12:33:23] [PASSED] 0x674E (CRESCENTISLAND)
[12:33:23] [PASSED] 0x674F (CRESCENTISLAND)
[12:33:23] [PASSED] 0x6750 (CRESCENTISLAND)
[12:33:23] [PASSED] 0xD750 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD751 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD752 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD753 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD754 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD755 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD756 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD757 (NOVALAKE_P)
[12:33:23] [PASSED] 0xD75F (NOVALAKE_P)
[12:33:23] =============== [PASSED] check_platform_desc ===============
[12:33:23] ===================== [PASSED] xe_pci ======================
[12:33:23] ============= xe_rtp_tables_test (5 subtests) ==============
[12:33:23] ================== xe_rtp_table_gt_test  ===================
[12:33:23] [PASSED] gt_was/14011060649
[12:33:23] [PASSED] gt_was/14011059788
[12:33:23] [PASSED] gt_was/14015795083
[12:33:23] [PASSED] gt_was/16021867713
[12:33:23] [PASSED] gt_was/14019449301
[12:33:23] [PASSED] gt_was/16028005424
[12:33:23] [PASSED] gt_was/14026578760
[12:33:23] [PASSED] gt_was/1409420604
[12:33:23] [PASSED] gt_was/1408615072
[12:33:23] [PASSED] gt_was/22010523718
[12:33:23] [PASSED] gt_was/14011006942
[12:33:23] [PASSED] gt_was/14014830051
[12:33:23] [PASSED] gt_was/18018781329
[12:33:23] [PASSED] gt_was/1509235366
[12:33:23] [PASSED] gt_was/18018781329
[12:33:23] [PASSED] gt_was/16016694945
[12:33:23] [PASSED] gt_was/14018575942
[12:33:23] [PASSED] gt_was/22016670082
[12:33:23] [PASSED] gt_was/22016670082
[12:33:23] [PASSED] gt_was/14017421178
[12:33:23] [PASSED] gt_was/16025250150
[12:33:23] [PASSED] gt_was/14021871409
[12:33:23] [PASSED] gt_was/16021865536
[12:33:23] [PASSED] gt_was/14021486841
[12:33:23] [PASSED] gt_was/14025160223
[12:33:23] [PASSED] gt_was/14026144927, 16029437861, 14026127056
[12:33:23] [PASSED] gt_was/14025635424
[12:33:23] [PASSED] gt_was/16028005424
[12:33:23] ============== [PASSED] xe_rtp_table_gt_test ===============
[12:33:23] ================== xe_rtp_table_gt_test  ===================
[12:33:23] [PASSED] gt_tunings/Tuning: Blend Fill Caching Optimization Disable
[12:33:23] [PASSED] gt_tunings/Tuning: 32B Access Enable
[12:33:23] [PASSED] gt_tunings/Tuning: L3 cache
[12:33:23] [PASSED] gt_tunings/Tuning: L3 cache - media
[12:33:23] [PASSED] gt_tunings/Tuning: Compression Overfetch
[12:33:23] [PASSED] gt_tunings/Tuning: Compression Overfetch - media
[12:33:23] [PASSED] gt_tunings/Tuning: Enable compressible partial write overfetch in L3
[12:33:23] [PASSED] gt_tunings/Tuning: Enable compressible partial write overfetch in L3 - media
[12:33:23] [PASSED] gt_tunings/Tuning: L2 Overfetch Compressible Only
[12:33:23] [PASSED] gt_tunings/Tuning: L2 Overfetch Compressible Only - media
[12:33:23] [PASSED] gt_tunings/Tuning: Stateless compression control
[12:33:23] [PASSED] gt_tunings/Tuning: Stateless compression control - media
[12:33:23] [PASSED] gt_tunings/Tuning: L3 RW flush all Cache
[12:33:23] [PASSED] gt_tunings/Tuning: L3 RW flush all cache - media
[12:33:23] [PASSED] gt_tunings/Tuning: Set STLB Bank Hash Mode to 4KB
[12:33:23] ============== [PASSED] xe_rtp_table_gt_test ===============
[12:33:23] ================== xe_rtp_table_oob_test  ==================
[12:33:23] [PASSED] oob_was/1607983814
[12:33:23] [PASSED] oob_was/16010904313
[12:33:23] [PASSED] oob_was/18022495364
[12:33:23] [PASSED] oob_was/22012773006
[12:33:23] [PASSED] oob_was/14014475959
[12:33:23] [PASSED] oob_was/22011391025
[12:33:23] [PASSED] oob_was/22012727170
[12:33:23] [PASSED] oob_was/22012727685
[12:33:23] [PASSED] oob_was/22016596838
[12:33:23] [PASSED] oob_was/18020744125
[12:33:23] [PASSED] oob_was/1409600907
[12:33:23] [PASSED] oob_was/22014953428
[12:33:23] [PASSED] oob_was/16017236439
[12:33:23] [PASSED] oob_was/14019821291
[12:33:23] [PASSED] oob_was/14015076503
[12:33:23] [PASSED] oob_was/14018913170
[12:33:23] [PASSED] oob_was/14018094691
[12:33:23] [PASSED] oob_was/18024947630
[12:33:23] [PASSED] oob_was/16022287689
[12:33:23] [PASSED] oob_was/13011645652
[12:33:23] [PASSED] oob_was/14022293748
[12:33:23] [PASSED] oob_was/22019794406
[12:33:23] [PASSED] oob_was/22019338487
[12:33:23] [PASSED] oob_was/16023588340
[12:33:23] [PASSED] oob_was/14019789679
[12:33:23] [PASSED] oob_was/14022866841
[12:33:23] [PASSED] oob_was/16021333562
[12:33:23] [PASSED] oob_was/14016712196
[12:33:23] [PASSED] oob_was/14015568240
[12:33:23] [PASSED] oob_was/18013179988
[12:33:23] [PASSED] oob_was/1508761755
[12:33:23] [PASSED] oob_was/16023105232
[12:33:23] [PASSED] oob_was/16026508708
[12:33:23] [PASSED] oob_was/14020001231
[12:33:23] [PASSED] oob_was/16023683509
[12:33:23] [PASSED] oob_was/14025515070
[12:33:23] [PASSED] oob_was/15015404425_disable
[12:33:23] [PASSED] oob_was/16026007364
[12:33:23] [PASSED] oob_was/14020316580
[12:33:23] [PASSED] oob_was/14025883347
[12:33:23] [PASSED] oob_was/16029380221
[12:33:23] [PASSED] oob_was/22022079272
[12:33:23] [PASSED] oob_was/16029897822
[12:33:23] [PASSED] oob_was/14027054324
[12:33:23] ============== [PASSED] xe_rtp_table_oob_test ==============
[12:33:23] ================ xe_rtp_table_dev_oob_test  ================
[12:33:23] [PASSED] device_oob_was/22010954014
[12:33:23] [PASSED] device_oob_was/15015404425
[12:33:23] [PASSED] device_oob_was/22019338487_display
[12:33:23] [PASSED] device_oob_was/14022085890
[12:33:23] [PASSED] device_oob_was/14026539277
[12:33:23] [PASSED] device_oob_was/14026633728
[12:33:23] [PASSED] device_oob_was/14026746987
[12:33:23] [PASSED] device_oob_was/14026779378
[12:33:23] ============ [PASSED] xe_rtp_table_dev_oob_test ============
[12:33:23] ========== xe_rtp_table_missing_upper_bound_test  ==========
[12:33:23] [PASSED] register_whitelist/WaAllowPMDepthAndInvocationCountAccessFromUMD, 1408556865
[12:33:23] [PASSED] register_whitelist/1508744258, 14012131227, 1808121037
[12:33:23] [PASSED] register_whitelist/1806527549
[12:33:23] [PASSED] register_whitelist/allow_read_ctx_timestamp
[12:33:23] [PASSED] register_whitelist/allow_read_queue_timestamp
[12:33:23] [PASSED] register_whitelist/16014440446
[12:33:23] [PASSED] register_whitelist/16017236439
[12:33:23] [PASSED] register_whitelist/16020183090
[12:33:23] [PASSED] register_whitelist/14024997852
[12:33:23] [PASSED] register_whitelist/14024997852
[12:33:23] ====== [PASSED] xe_rtp_table_missing_upper_bound_test ======
[12:33:23] =============== [PASSED] xe_rtp_tables_test ================
[12:33:23] =================== xe_rtp (3 subtests) ====================
[12:33:23] =================== xe_rtp_rules_tests  ====================
[12:33:23] [PASSED] no
[12:33:23] [PASSED] yes
[12:33:23] [PASSED] no-and-no
[12:33:23] [PASSED] no-and-yes
[12:33:23] [PASSED] yes-and-no
[12:33:23] [PASSED] yes-and-yes
[12:33:23] [PASSED] no-or-no
[12:33:23] [PASSED] no-or-yes
[12:33:23] [PASSED] yes-or-no
[12:33:23] [PASSED] yes-or-yes
[12:33:23] [PASSED] no-yes-or-yes-no
[12:33:23] [PASSED] no-yes-or-yes-yes
[12:33:23] [PASSED] yes-yes-or-no-yes
[12:33:23] [PASSED] yes-yes-or-yes-yes
[12:33:23] [PASSED] no-no-or-yes-or-no
[12:33:23] [PASSED] or
[12:33:23] [PASSED] or-yes
[12:33:23] [PASSED] or-no
[12:33:23] [PASSED] yes-or
[12:33:23] [PASSED] no-or
[12:33:23] [PASSED] no-or-or-yes
[12:33:23] [PASSED] yes-or-or-no
[12:33:23] [PASSED] no-or-or-no
[12:33:23] [PASSED] missing-context-engine-class
[12:33:23] [PASSED] missing-context-engine-class-or-yes
[12:33:23] [PASSED] missing-context-engine-class-or-or-yes
[12:33:23] =============== [PASSED] xe_rtp_rules_tests ================
[12:33:23] =============== xe_rtp_process_to_sr_tests  ================
[12:33:23] [PASSED] coalesce-same-reg
[12:33:23] [PASSED] coalesce-same-reg-literal-and-func
[12:33:23] [PASSED] no-match-no-add
[12:33:23] [PASSED] two-regs-two-entries
[12:33:23] [PASSED] clr-one-set-other
[12:33:23] [PASSED] set-field
[12:33:23] [PASSED] conflict-duplicate
[12:33:23] [PASSED] conflict-not-disjoint
[12:33:23] [PASSED] conflict-not-disjoint-literal-and-func
[12:33:23] [PASSED] conflict-reg-type
[12:33:23] [PASSED] bad-mcr-reg-forced-to-regular
[12:33:23] [PASSED] bad-regular-reg-forced-to-mcr
[12:33:23] =========== [PASSED] xe_rtp_process_to_sr_tests ============
[12:33:23] ================== xe_rtp_process_tests  ===================
[12:33:23] [PASSED] active1
[12:33:23] [PASSED] active2
[12:33:23] [PASSED] active-inactive
[12:33:23] [PASSED] inactive-active
[12:33:23] [PASSED] inactive-active-inactive
[12:33:23] [PASSED] inactive-inactive-inactive
[12:33:23] ============== [PASSED] xe_rtp_process_tests ===============
[12:33:23] ===================== [PASSED] xe_rtp ======================
[12:33:23] ==================== xe_wa (1 subtest) =====================
[12:33:23] ======================== xe_wa_gt  =========================
[12:33:23] [PASSED] TIGERLAKE B0
[12:33:23] [PASSED] DG1 A0
[12:33:23] [PASSED] DG1 B0
[12:33:23] [PASSED] ALDERLAKE_S A0
[12:33:23] [PASSED] ALDERLAKE_S B0
[12:33:23] [PASSED] ALDERLAKE_S C0
[12:33:23] [PASSED] ALDERLAKE_S D0
[12:33:23] [PASSED] ALDERLAKE_P A0
[12:33:23] [PASSED] ALDERLAKE_P B0
[12:33:23] [PASSED] ALDERLAKE_P C0
[12:33:23] [PASSED] ALDERLAKE_S RPLS D0
[12:33:23] [PASSED] ALDERLAKE_P RPLU E0
[12:33:23] [PASSED] DG2 G10 C0
[12:33:23] [PASSED] DG2 G11 B1
[12:33:23] [PASSED] DG2 G12 A1
[12:33:23] [PASSED] METEORLAKE 12.70(Xe_LPG) A0 13.00(Xe_LPM+) A0
[12:33:23] [PASSED] METEORLAKE 12.71(Xe_LPG) A0 13.00(Xe_LPM+) A0
[12:33:23] [PASSED] METEORLAKE 12.74(Xe_LPG+) A0 13.00(Xe_LPM+) A0
[12:33:23] [PASSED] LUNARLAKE 20.04(Xe2_LPG) A0 20.00(Xe2_LPM) A0
[12:33:23] [PASSED] LUNARLAKE 20.04(Xe2_LPG) B0 20.00(Xe2_LPM) A0
[12:33:23] [PASSED] BATTLEMAGE 20.01(Xe2_HPG) A0 13.01(Xe2_HPM) A1
[12:33:23] [PASSED] PANTHERLAKE 30.00(Xe3_LPG) A0 30.00(Xe3_LPM) A0
[12:33:23] ==================== [PASSED] xe_wa_gt =====================
[12:33:23] ====================== [PASSED] xe_wa ======================
[12:33:23] ============================================================
[12:33:23] Testing complete. Ran 742 tests: passed: 724, skipped: 18
[12:33:23] Elapsed time: 36.657s total, 4.413s configuring, 31.578s building, 0.624s running

+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/tests/.kunitconfig [12:33:23] Configuring KUnit Kernel ...
Regenerating .config ...
Populating config with:

$ make ARCH=um O=.kunit olddefconfig
[12:33:25] Building KUnit Kernel ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig
Building with:
$ make all compile_commands.json scripts_gdb ARCH=um O=.kunit --jobs=48
[12:33:49] Starting KUnit Kernel (1/1)...
[12:33:49] ============================================================
Running tests with:
$ .kunit/linux kunit.enable=1 mem=1G console=tty kunit_shutdown=halt
[12:33:50] ============ drm_test_pick_cmdline (2 subtests) ============
[12:33:50] [PASSED] drm_test_pick_cmdline_res_1920_1080_60
[12:33:50] =============== drm_test_pick_cmdline_named  ===============
[12:33:50] [PASSED] NTSC
[12:33:50] [PASSED] NTSC-J
[12:33:50] [PASSED] PAL
[12:33:50] [PASSED] PAL-M
[12:33:50] =========== [PASSED] drm_test_pick_cmdline_named ===========
[12:33:50] ============== [PASSED] drm_test_pick_cmdline ==============
[12:33:50] == drm_test_atomic_get_connector_for_encoder (1 subtest) ===
[12:33:50] [PASSED] drm_test_drm_atomic_get_connector_for_encoder
[12:33:50] ==== [PASSED] drm_test_atomic_get_connector_for_encoder ====
[12:33:50] =========== drm_validate_clone_mode (2 subtests) ===========
[12:33:50] ============== drm_test_check_in_clone_mode  ===============
[12:33:50] [PASSED] in_clone_mode
[12:33:50] [PASSED] not_in_clone_mode
[12:33:50] ========== [PASSED] drm_test_check_in_clone_mode ===========
[12:33:50] =============== drm_test_check_valid_clones  ===============
[12:33:50] [PASSED] not_in_clone_mode
[12:33:50] [PASSED] valid_clone
[12:33:50] [PASSED] invalid_clone
[12:33:50] =========== [PASSED] drm_test_check_valid_clones ===========
[12:33:50] ============= [PASSED] drm_validate_clone_mode =============
[12:33:50] ============= drm_validate_modeset (1 subtest) =============
[12:33:50] [PASSED] drm_test_check_connector_changed_modeset
[12:33:50] ============== [PASSED] drm_validate_modeset ===============
[12:33:50] ====== drm_test_bridge_get_current_state (1 subtest) =======
[12:33:50] [PASSED] drm_test_drm_bridge_get_current_state_atomic
[12:33:50] ======== [PASSED] drm_test_bridge_get_current_state ========
[12:33:50] ====== drm_test_bridge_helper_reset_crtc (3 subtests) ======
[12:33:50] [PASSED] drm_test_drm_bridge_helper_reset_crtc_atomic
[12:33:50] [PASSED] drm_test_drm_bridge_helper_reset_crtc_atomic_disabled
[12:33:50] [PASSED] drm_test_drm_bridge_helper_hdmi_output_bus_fmts
[12:33:50] ======== [PASSED] drm_test_bridge_helper_reset_crtc ========
[12:33:50] ============== drm_bridge_alloc (2 subtests) ===============
[12:33:50] [PASSED] drm_test_drm_bridge_alloc_basic
[12:33:50] [PASSED] drm_test_drm_bridge_alloc_get_put
[12:33:50] ================ [PASSED] drm_bridge_alloc =================
[12:33:50] ============= drm_bridge_bus_fmt (5 subtests) ==============
[12:33:50] [PASSED] drm_test_bridge_rgb_yuv_rgb
[12:33:50] [PASSED] drm_test_bridge_must_convert_to_yuv444
[12:33:50] [PASSED] drm_test_bridge_hdmi_auto_rgb
[12:33:50] [PASSED] drm_test_bridge_auto_first
[12:33:50] [PASSED] drm_test_bridge_rgb_yuv_no_path
[12:33:50] =============== [PASSED] drm_bridge_bus_fmt ================
[12:33:50] ============= drm_cmdline_parser (40 subtests) =============
[12:33:50] [PASSED] drm_test_cmdline_force_d_only
[12:33:50] [PASSED] drm_test_cmdline_force_D_only_dvi
[12:33:50] [PASSED] drm_test_cmdline_force_D_only_hdmi
[12:33:50] [PASSED] drm_test_cmdline_force_D_only_not_digital
[12:33:50] [PASSED] drm_test_cmdline_force_e_only
[12:33:50] [PASSED] drm_test_cmdline_res
[12:33:50] [PASSED] drm_test_cmdline_res_vesa
[12:33:50] [PASSED] drm_test_cmdline_res_vesa_rblank
[12:33:50] [PASSED] drm_test_cmdline_res_rblank
[12:33:50] [PASSED] drm_test_cmdline_res_bpp
[12:33:50] [PASSED] drm_test_cmdline_res_refresh
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh_interlaced
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh_margins
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh_force_off
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh_force_on
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh_force_on_analog
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh_force_on_digital
[12:33:50] [PASSED] drm_test_cmdline_res_bpp_refresh_interlaced_margins_force_on
[12:33:50] [PASSED] drm_test_cmdline_res_margins_force_on
[12:33:50] [PASSED] drm_test_cmdline_res_vesa_margins
[12:33:50] [PASSED] drm_test_cmdline_name
[12:33:50] [PASSED] drm_test_cmdline_name_bpp
[12:33:50] [PASSED] drm_test_cmdline_name_option
[12:33:50] [PASSED] drm_test_cmdline_name_bpp_option
[12:33:50] [PASSED] drm_test_cmdline_rotate_0
[12:33:50] [PASSED] drm_test_cmdline_rotate_90
[12:33:50] [PASSED] drm_test_cmdline_rotate_180
[12:33:50] [PASSED] drm_test_cmdline_rotate_270
[12:33:50] [PASSED] drm_test_cmdline_hmirror
[12:33:50] [PASSED] drm_test_cmdline_vmirror
[12:33:50] [PASSED] drm_test_cmdline_margin_options
[12:33:50] [PASSED] drm_test_cmdline_multiple_options
[12:33:50] [PASSED] drm_test_cmdline_bpp_extra_and_option
[12:33:50] [PASSED] drm_test_cmdline_extra_and_option
[12:33:50] [PASSED] drm_test_cmdline_freestanding_options
[12:33:50] [PASSED] drm_test_cmdline_freestanding_force_e_and_options
[12:33:50] [PASSED] drm_test_cmdline_panel_orientation
[12:33:50] ================ drm_test_cmdline_invalid  =================
[12:33:50] [PASSED] margin_only
[12:33:50] [PASSED] interlace_only
[12:33:50] [PASSED] res_missing_x
[12:33:50] [PASSED] res_missing_y
[12:33:50] [PASSED] res_bad_y
[12:33:50] [PASSED] res_missing_y_bpp
[12:33:50] [PASSED] res_bad_bpp
[12:33:50] [PASSED] res_bad_refresh
[12:33:50] [PASSED] res_bpp_refresh_force_on_off
[12:33:50] [PASSED] res_invalid_mode
[12:33:50] [PASSED] res_bpp_wrong_place_mode
[12:33:50] [PASSED] name_bpp_refresh
[12:33:50] [PASSED] name_refresh
[12:33:50] [PASSED] name_refresh_wrong_mode
[12:33:50] [PASSED] name_refresh_invalid_mode
[12:33:50] [PASSED] rotate_multiple
[12:33:50] [PASSED] rotate_invalid_val
[12:33:50] [PASSED] rotate_truncated
[12:33:50] [PASSED] invalid_option
[12:33:50] [PASSED] invalid_tv_option
[12:33:50] [PASSED] truncated_tv_option
[12:33:50] ============ [PASSED] drm_test_cmdline_invalid =============
[12:33:50] =============== drm_test_cmdline_tv_options  ===============
[12:33:50] [PASSED] NTSC
[12:33:50] [PASSED] NTSC_443
[12:33:50] [PASSED] NTSC_J
[12:33:50] [PASSED] PAL
[12:33:50] [PASSED] PAL_M
[12:33:50] [PASSED] PAL_N
[12:33:50] [PASSED] SECAM
[12:33:50] [PASSED] MONO_525
[12:33:50] [PASSED] MONO_625
[12:33:50] =========== [PASSED] drm_test_cmdline_tv_options ===========
[12:33:50] =============== [PASSED] drm_cmdline_parser ================
[12:33:50] ========== drmm_connector_hdmi_init (20 subtests) ==========
[12:33:50] [PASSED] drm_test_connector_hdmi_init_valid
[12:33:50] [PASSED] drm_test_connector_hdmi_init_bpc_8
[12:33:50] [PASSED] drm_test_connector_hdmi_init_bpc_10
[12:33:50] [PASSED] drm_test_connector_hdmi_init_bpc_12
[12:33:50] [PASSED] drm_test_connector_hdmi_init_bpc_invalid
[12:33:50] [PASSED] drm_test_connector_hdmi_init_bpc_null
[12:33:50] [PASSED] drm_test_connector_hdmi_init_formats_empty
[12:33:50] [PASSED] drm_test_connector_hdmi_init_formats_no_rgb
[12:33:50] === drm_test_connector_hdmi_init_formats_yuv420_allowed  ===
[12:33:50] [PASSED] supported_formats=0x9 yuv420_allowed=1
[12:33:50] [PASSED] supported_formats=0x9 yuv420_allowed=0
[12:33:50] [PASSED] supported_formats=0x5 yuv420_allowed=1
[12:33:50] [PASSED] supported_formats=0x5 yuv420_allowed=0
[12:33:50] === [PASSED] drm_test_connector_hdmi_init_formats_yuv420_allowed ===
[12:33:50] [PASSED] drm_test_connector_hdmi_init_null_ddc
[12:33:50] [PASSED] drm_test_connector_hdmi_init_null_product
[12:33:50] [PASSED] drm_test_connector_hdmi_init_null_vendor
[12:33:50] [PASSED] drm_test_connector_hdmi_init_product_length_exact
[12:33:50] [PASSED] drm_test_connector_hdmi_init_product_length_too_long
[12:33:50] [PASSED] drm_test_connector_hdmi_init_product_valid
[12:33:50] [PASSED] drm_test_connector_hdmi_init_vendor_length_exact
[12:33:50] [PASSED] drm_test_connector_hdmi_init_vendor_length_too_long
[12:33:50] [PASSED] drm_test_connector_hdmi_init_vendor_valid
[12:33:50] ========= drm_test_connector_hdmi_init_type_valid  =========
[12:33:50] [PASSED] HDMI-A
[12:33:50] [PASSED] HDMI-B
[12:33:50] ===== [PASSED] drm_test_connector_hdmi_init_type_valid =====
[12:33:50] ======== drm_test_connector_hdmi_init_type_invalid  ========
[12:33:50] [PASSED] Unknown
[12:33:50] [PASSED] VGA
[12:33:50] [PASSED] DVI-I
[12:33:50] [PASSED] DVI-D
[12:33:50] [PASSED] DVI-A
[12:33:50] [PASSED] Composite
[12:33:50] [PASSED] SVIDEO
[12:33:50] [PASSED] LVDS
[12:33:50] [PASSED] Component
[12:33:50] [PASSED] DIN
[12:33:50] [PASSED] DP
[12:33:50] [PASSED] TV
[12:33:50] [PASSED] eDP
[12:33:50] [PASSED] Virtual
[12:33:50] [PASSED] DSI
[12:33:50] [PASSED] DPI
[12:33:50] [PASSED] Writeback
[12:33:50] [PASSED] SPI
[12:33:50] [PASSED] USB
[12:33:50] ==== [PASSED] drm_test_connector_hdmi_init_type_invalid ====
[12:33:50] ============ [PASSED] drmm_connector_hdmi_init =============
[12:33:50] ============= drmm_connector_init (3 subtests) =============
[12:33:50] [PASSED] drm_test_drmm_connector_init
[12:33:50] [PASSED] drm_test_drmm_connector_init_null_ddc
[12:33:50] ========= drm_test_drmm_connector_init_type_valid  =========
[12:33:50] [PASSED] Unknown
[12:33:50] [PASSED] VGA
[12:33:50] [PASSED] DVI-I
[12:33:50] [PASSED] DVI-D
[12:33:50] [PASSED] DVI-A
[12:33:50] [PASSED] Composite
[12:33:50] [PASSED] SVIDEO
[12:33:50] [PASSED] LVDS
[12:33:50] [PASSED] Component
[12:33:50] [PASSED] DIN
[12:33:50] [PASSED] DP
[12:33:50] [PASSED] HDMI-A
[12:33:50] [PASSED] HDMI-B
[12:33:50] [PASSED] TV
[12:33:50] [PASSED] eDP
[12:33:50] [PASSED] Virtual
[12:33:50] [PASSED] DSI
[12:33:50] [PASSED] DPI
[12:33:50] [PASSED] Writeback
[12:33:50] [PASSED] SPI
[12:33:50] [PASSED] USB
[12:33:50] ===== [PASSED] drm_test_drmm_connector_init_type_valid =====
[12:33:50] =============== [PASSED] drmm_connector_init ===============
[12:33:50] ========= drm_connector_dynamic_init (6 subtests) ==========
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_init
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_init_null_ddc
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_init_not_added
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_init_properties
[12:33:50] ===== drm_test_drm_connector_dynamic_init_type_valid  ======
[12:33:50] [PASSED] Unknown
[12:33:50] [PASSED] VGA
[12:33:50] [PASSED] DVI-I
[12:33:50] [PASSED] DVI-D
[12:33:50] [PASSED] DVI-A
[12:33:50] [PASSED] Composite
[12:33:50] [PASSED] SVIDEO
[12:33:50] [PASSED] LVDS
[12:33:50] [PASSED] Component
[12:33:50] [PASSED] DIN
[12:33:50] [PASSED] DP
[12:33:50] [PASSED] HDMI-A
[12:33:50] [PASSED] HDMI-B
[12:33:50] [PASSED] TV
[12:33:50] [PASSED] eDP
[12:33:50] [PASSED] Virtual
[12:33:50] [PASSED] DSI
[12:33:50] [PASSED] DPI
[12:33:50] [PASSED] Writeback
[12:33:50] [PASSED] SPI
[12:33:50] [PASSED] USB
[12:33:50] = [PASSED] drm_test_drm_connector_dynamic_init_type_valid ==
[12:33:50] ======== drm_test_drm_connector_dynamic_init_name  =========
[12:33:50] [PASSED] Unknown
[12:33:50] [PASSED] VGA
[12:33:50] [PASSED] DVI-I
[12:33:50] [PASSED] DVI-D
[12:33:50] [PASSED] DVI-A
[12:33:50] [PASSED] Composite
[12:33:50] [PASSED] SVIDEO
[12:33:50] [PASSED] LVDS
[12:33:50] [PASSED] Component
[12:33:50] [PASSED] DIN
[12:33:50] [PASSED] DP
[12:33:50] [PASSED] HDMI-A
[12:33:50] [PASSED] HDMI-B
[12:33:50] [PASSED] TV
[12:33:50] [PASSED] eDP
[12:33:50] [PASSED] Virtual
[12:33:50] [PASSED] DSI
[12:33:50] [PASSED] DPI
[12:33:50] [PASSED] Writeback
[12:33:50] [PASSED] SPI
[12:33:50] [PASSED] USB
[12:33:50] ==== [PASSED] drm_test_drm_connector_dynamic_init_name =====
[12:33:50] =========== [PASSED] drm_connector_dynamic_init ============
[12:33:50] ==== drm_connector_dynamic_register_early (4 subtests) =====
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_early_on_list
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_early_defer
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_early_no_init
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_early_no_mode_object
[12:33:50] ====== [PASSED] drm_connector_dynamic_register_early =======
[12:33:50] ======= drm_connector_dynamic_register (7 subtests) ========
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_on_list
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_no_defer
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_no_init
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_mode_object
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_sysfs
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_sysfs_name
[12:33:50] [PASSED] drm_test_drm_connector_dynamic_register_debugfs
[12:33:50] ========= [PASSED] drm_connector_dynamic_register ==========
[12:33:50] = drm_connector_attach_broadcast_rgb_property (2 subtests) =
[12:33:50] [PASSED] drm_test_drm_connector_attach_broadcast_rgb_property
[12:33:50] [PASSED] drm_test_drm_connector_attach_broadcast_rgb_property_hdmi_connector
[12:33:50] === [PASSED] drm_connector_attach_broadcast_rgb_property ===
[12:33:50] ========== drm_get_tv_mode_from_name (2 subtests) ==========
[12:33:50] ========== drm_test_get_tv_mode_from_name_valid  ===========
[12:33:50] [PASSED] NTSC
[12:33:50] [PASSED] NTSC-443
[12:33:50] [PASSED] NTSC-J
[12:33:50] [PASSED] PAL
[12:33:50] [PASSED] PAL-M
[12:33:50] [PASSED] PAL-N
[12:33:50] [PASSED] SECAM
[12:33:50] [PASSED] Mono
[12:33:50] ====== [PASSED] drm_test_get_tv_mode_from_name_valid =======
[12:33:50] [PASSED] drm_test_get_tv_mode_from_name_truncated
[12:33:50] ============ [PASSED] drm_get_tv_mode_from_name ============
[12:33:50] = drm_test_connector_hdmi_compute_mode_clock (12 subtests) =
[12:33:50] [PASSED] drm_test_drm_hdmi_compute_mode_clock_rgb
[12:33:50] [PASSED] drm_test_drm_hdmi_compute_mode_clock_rgb_10bpc
[12:33:50] [PASSED] drm_test_drm_hdmi_compute_mode_clock_rgb_10bpc_vic_1
[12:33:50] [PASSED] drm_test_drm_hdmi_compute_mode_clock_rgb_12bpc
[12:33:50] [PASSED] drm_test_drm_hdmi_compute_mode_clock_rgb_12bpc_vic_1
[12:33:50] [PASSED] drm_test_drm_hdmi_compute_mode_clock_rgb_double
[12:33:50] = drm_test_connector_hdmi_compute_mode_clock_yuv420_valid  =
[12:33:50] [PASSED] VIC 96
[12:33:50] [PASSED] VIC 97
[12:33:50] [PASSED] VIC 101
[12:33:50] [PASSED] VIC 102
[12:33:50] [PASSED] VIC 106
[12:33:50] [PASSED] VIC 107
[12:33:50] === [PASSED] drm_test_connector_hdmi_compute_mode_clock_yuv420_valid ===
[12:33:50] [PASSED] drm_test_connector_hdmi_compute_mode_clock_yuv420_10_bpc
[12:33:50] [PASSED] drm_test_connector_hdmi_compute_mode_clock_yuv420_12_bpc
[12:33:50] [PASSED] drm_test_connector_hdmi_compute_mode_clock_yuv422_8_bpc
[12:33:50] [PASSED] drm_test_connector_hdmi_compute_mode_clock_yuv422_10_bpc
[12:33:50] [PASSED] drm_test_connector_hdmi_compute_mode_clock_yuv422_12_bpc
[12:33:50] === [PASSED] drm_test_connector_hdmi_compute_mode_clock ====
[12:33:50] == drm_hdmi_connector_get_broadcast_rgb_name (2 subtests) ==
[12:33:50] === drm_test_drm_hdmi_connector_get_broadcast_rgb_name  ====
[12:33:50] [PASSED] Automatic
[12:33:50] [PASSED] Full
[12:33:50] [PASSED] Limited 16:235
[12:33:50] === [PASSED] drm_test_drm_hdmi_connector_get_broadcast_rgb_name ===
[12:33:50] [PASSED] drm_test_drm_hdmi_connector_get_broadcast_rgb_name_invalid
[12:33:50] ==== [PASSED] drm_hdmi_connector_get_broadcast_rgb_name ====
[12:33:50] == drm_hdmi_connector_get_output_format_name (2 subtests) ==
[12:33:50] === drm_test_drm_hdmi_connector_get_output_format_name  ====
[12:33:50] [PASSED] RGB
[12:33:50] [PASSED] YUV 4:2:0
[12:33:50] [PASSED] YUV 4:2:2
[12:33:50] [PASSED] YUV 4:4:4
[12:33:50] === [PASSED] drm_test_drm_hdmi_connector_get_output_format_name ===
[12:33:50] [PASSED] drm_test_drm_hdmi_connector_get_output_format_name_invalid
[12:33:50] ==== [PASSED] drm_hdmi_connector_get_output_format_name ====
[12:33:50] ============= drm_damage_helper (21 subtests) ==============
[12:33:50] [PASSED] drm_test_damage_iter_no_damage
[12:33:50] [PASSED] drm_test_damage_iter_no_damage_fractional_src
[12:33:50] [PASSED] drm_test_damage_iter_no_damage_src_moved
[12:33:50] [PASSED] drm_test_damage_iter_no_damage_fractional_src_moved
[12:33:50] [PASSED] drm_test_damage_iter_no_damage_not_visible
[12:33:50] [PASSED] drm_test_damage_iter_no_damage_no_crtc
[12:33:50] [PASSED] drm_test_damage_iter_no_damage_no_fb
[12:33:50] [PASSED] drm_test_damage_iter_simple_damage
[12:33:50] [PASSED] drm_test_damage_iter_single_damage
[12:33:50] [PASSED] drm_test_damage_iter_single_damage_intersect_src
[12:33:50] [PASSED] drm_test_damage_iter_single_damage_outside_src
[12:33:50] [PASSED] drm_test_damage_iter_single_damage_fractional_src
[12:33:50] [PASSED] drm_test_damage_iter_single_damage_intersect_fractional_src
[12:33:50] [PASSED] drm_test_damage_iter_single_damage_outside_fractional_src
[12:33:50] [PASSED] drm_test_damage_iter_single_damage_src_moved
[12:33:50] [PASSED] drm_test_damage_iter_single_damage_fractional_src_moved
[12:33:50] [PASSED] drm_test_damage_iter_damage
[12:33:50] [PASSED] drm_test_damage_iter_damage_one_intersect
[12:33:50] [PASSED] drm_test_damage_iter_damage_one_outside
[12:33:50] [PASSED] drm_test_damage_iter_damage_src_moved
[12:33:50] [PASSED] drm_test_damage_iter_damage_not_visible
[12:33:50] ================ [PASSED] drm_damage_helper ================
[12:33:50] ============== drm_dp_mst_helper (3 subtests) ==============
[12:33:50] ============== drm_test_dp_mst_calc_pbn_mode  ==============
[12:33:50] [PASSED] Clock 154000 BPP 30 DSC disabled
[12:33:50] [PASSED] Clock 234000 BPP 30 DSC disabled
[12:33:50] [PASSED] Clock 297000 BPP 24 DSC disabled
[12:33:50] [PASSED] Clock 332880 BPP 24 DSC enabled
[12:33:50] [PASSED] Clock 324540 BPP 24 DSC enabled
[12:33:50] ========== [PASSED] drm_test_dp_mst_calc_pbn_mode ==========
[12:33:50] ============== drm_test_dp_mst_calc_pbn_div  ===============
[12:33:50] [PASSED] Link rate 2000000 lane count 4
[12:33:50] [PASSED] Link rate 2000000 lane count 2
[12:33:50] [PASSED] Link rate 2000000 lane count 1
[12:33:50] [PASSED] Link rate 1350000 lane count 4
[12:33:50] [PASSED] Link rate 1350000 lane count 2
[12:33:50] [PASSED] Link rate 1350000 lane count 1
[12:33:50] [PASSED] Link rate 1000000 lane count 4
[12:33:50] [PASSED] Link rate 1000000 lane count 2
[12:33:50] [PASSED] Link rate 1000000 lane count 1
[12:33:50] [PASSED] Link rate 810000 lane count 4
[12:33:50] [PASSED] Link rate 810000 lane count 2
[12:33:50] [PASSED] Link rate 810000 lane count 1
[12:33:50] [PASSED] Link rate 540000 lane count 4
[12:33:50] [PASSED] Link rate 540000 lane count 2
[12:33:50] [PASSED] Link rate 540000 lane count 1
[12:33:50] [PASSED] Link rate 270000 lane count 4
[12:33:50] [PASSED] Link rate 270000 lane count 2
[12:33:50] [PASSED] Link rate 270000 lane count 1
[12:33:50] [PASSED] Link rate 162000 lane count 4
[12:33:50] [PASSED] Link rate 162000 lane count 2
[12:33:50] [PASSED] Link rate 162000 lane count 1
[12:33:50] ========== [PASSED] drm_test_dp_mst_calc_pbn_div ===========
[12:33:50] ========= drm_test_dp_mst_sideband_msg_req_decode  =========
[12:33:50] [PASSED] DP_ENUM_PATH_RESOURCES with port number
[12:33:50] [PASSED] DP_POWER_UP_PHY with port number
[12:33:50] [PASSED] DP_POWER_DOWN_PHY with port number
[12:33:50] [PASSED] DP_ALLOCATE_PAYLOAD with SDP stream sinks
[12:33:50] [PASSED] DP_ALLOCATE_PAYLOAD with port number
[12:33:50] [PASSED] DP_ALLOCATE_PAYLOAD with VCPI
[12:33:50] [PASSED] DP_ALLOCATE_PAYLOAD with PBN
[12:33:50] [PASSED] DP_QUERY_PAYLOAD with port number
[12:33:50] [PASSED] DP_QUERY_PAYLOAD with VCPI
[12:33:50] [PASSED] DP_REMOTE_DPCD_READ with port number
[12:33:50] [PASSED] DP_REMOTE_DPCD_READ with DPCD address
[12:33:50] [PASSED] DP_REMOTE_DPCD_READ with max number of bytes
[12:33:50] [PASSED] DP_REMOTE_DPCD_WRITE with port number
[12:33:50] [PASSED] DP_REMOTE_DPCD_WRITE with DPCD address
[12:33:50] [PASSED] DP_REMOTE_DPCD_WRITE with data array
[12:33:50] [PASSED] DP_REMOTE_I2C_READ with port number
[12:33:50] [PASSED] DP_REMOTE_I2C_READ with I2C device ID
[12:33:50] [PASSED] DP_REMOTE_I2C_READ with transactions array
[12:33:50] [PASSED] DP_REMOTE_I2C_WRITE with port number
[12:33:50] [PASSED] DP_REMOTE_I2C_WRITE with I2C device ID
[12:33:50] [PASSED] DP_REMOTE_I2C_WRITE with data array
[12:33:50] [PASSED] DP_QUERY_STREAM_ENC_STATUS with stream ID
[12:33:50] [PASSED] DP_QUERY_STREAM_ENC_STATUS with client ID
[12:33:50] [PASSED] DP_QUERY_STREAM_ENC_STATUS with stream event
[12:33:50] [PASSED] DP_QUERY_STREAM_ENC_STATUS with valid stream event
[12:33:50] [PASSED] DP_QUERY_STREAM_ENC_STATUS with stream behavior
[12:33:50] [PASSED] DP_QUERY_STREAM_ENC_STATUS with a valid stream behavior
[12:33:50] ===== [PASSED] drm_test_dp_mst_sideband_msg_req_decode =====
[12:33:50] ================ [PASSED] drm_dp_mst_helper ================
[12:33:50] ================== drm_exec (7 subtests) ===================
[12:33:50] [PASSED] sanitycheck
[12:33:50] [PASSED] test_lock
[12:33:50] [PASSED] test_lock_unlock
[12:33:50] [PASSED] test_duplicates
[12:33:50] [PASSED] test_prepare
[12:33:50] [PASSED] test_prepare_array
[12:33:50] [PASSED] test_multiple_loops
[12:33:50] ==================== [PASSED] drm_exec =====================
[12:33:50] =========== drm_format_helper_test (17 subtests) ===========
[12:33:50] ============== drm_test_fb_xrgb8888_to_gray8  ==============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ========== [PASSED] drm_test_fb_xrgb8888_to_gray8 ==========
[12:33:50] ============= drm_test_fb_xrgb8888_to_rgb332  ==============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ========= [PASSED] drm_test_fb_xrgb8888_to_rgb332 ==========
[12:33:50] ============= drm_test_fb_xrgb8888_to_rgb565  ==============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ========= [PASSED] drm_test_fb_xrgb8888_to_rgb565 ==========
[12:33:50] ============ drm_test_fb_xrgb8888_to_xrgb1555  =============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======== [PASSED] drm_test_fb_xrgb8888_to_xrgb1555 =========
[12:33:50] ============ drm_test_fb_xrgb8888_to_argb1555  =============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======== [PASSED] drm_test_fb_xrgb8888_to_argb1555 =========
[12:33:50] ============ drm_test_fb_xrgb8888_to_rgba5551  =============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======== [PASSED] drm_test_fb_xrgb8888_to_rgba5551 =========
[12:33:50] ============= drm_test_fb_xrgb8888_to_rgb888  ==============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ========= [PASSED] drm_test_fb_xrgb8888_to_rgb888 ==========
[12:33:50] ============= drm_test_fb_xrgb8888_to_bgr888  ==============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ========= [PASSED] drm_test_fb_xrgb8888_to_bgr888 ==========
[12:33:50] ============ drm_test_fb_xrgb8888_to_argb8888  =============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======== [PASSED] drm_test_fb_xrgb8888_to_argb8888 =========
[12:33:50] =========== drm_test_fb_xrgb8888_to_xrgb2101010  ===========
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======= [PASSED] drm_test_fb_xrgb8888_to_xrgb2101010 =======
[12:33:50] =========== drm_test_fb_xrgb8888_to_argb2101010  ===========
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======= [PASSED] drm_test_fb_xrgb8888_to_argb2101010 =======
[12:33:50] ============== drm_test_fb_xrgb8888_to_mono  ===============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ========== [PASSED] drm_test_fb_xrgb8888_to_mono ===========
[12:33:50] ==================== drm_test_fb_swab  =====================
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ================ [PASSED] drm_test_fb_swab =================
[12:33:50] ============ drm_test_fb_xrgb8888_to_xbgr8888  =============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======== [PASSED] drm_test_fb_xrgb8888_to_xbgr8888 =========
[12:33:50] ============ drm_test_fb_xrgb8888_to_abgr8888  =============
[12:33:50] [PASSED] single_pixel_source_buffer
[12:33:50] [PASSED] single_pixel_clip_rectangle
[12:33:50] [PASSED] well_known_colors
[12:33:50] [PASSED] destination_pitch
[12:33:50] ======== [PASSED] drm_test_fb_xrgb8888_to_abgr8888 =========
[12:33:50] ================= drm_test_fb_clip_offset  =================
[12:33:50] [PASSED] pass through
[12:33:50] [PASSED] horizontal offset
[12:33:50] [PASSED] vertical offset
[12:33:50] [PASSED] horizontal and vertical offset
[12:33:50] [PASSED] horizontal offset (custom pitch)
[12:33:50] [PASSED] vertical offset (custom pitch)
[12:33:50] [PASSED] horizontal and vertical offset (custom pitch)
[12:33:50] ============= [PASSED] drm_test_fb_clip_offset =============
[12:33:50] =================== drm_test_fb_memcpy  ====================
[12:33:50] [PASSED] single_pixel_source_buffer: XR24 little-endian (0x34325258)
[12:33:50] [PASSED] single_pixel_source_buffer: XRA8 little-endian (0x38415258)
[12:33:50] [PASSED] single_pixel_source_buffer: YU24 little-endian (0x34325559)
[12:33:50] [PASSED] single_pixel_clip_rectangle: XB24 little-endian (0x34324258)
[12:33:50] [PASSED] single_pixel_clip_rectangle: XRA8 little-endian (0x38415258)
[12:33:50] [PASSED] single_pixel_clip_rectangle: YU24 little-endian (0x34325559)
[12:33:50] [PASSED] well_known_colors: XB24 little-endian (0x34324258)
[12:33:50] [PASSED] well_known_colors: XRA8 little-endian (0x38415258)
[12:33:50] [PASSED] well_known_colors: YU24 little-endian (0x34325559)
[12:33:50] [PASSED] destination_pitch: XB24 little-endian (0x34324258)
[12:33:50] [PASSED] destination_pitch: XRA8 little-endian (0x38415258)
[12:33:50] [PASSED] destination_pitch: YU24 little-endian (0x34325559)
[12:33:50] =============== [PASSED] drm_test_fb_memcpy ================
[12:33:50] ============= [PASSED] drm_format_helper_test ==============
[12:33:50] ================= drm_format (18 subtests) =================
[12:33:50] [PASSED] drm_test_format_block_width_invalid
[12:33:50] [PASSED] drm_test_format_block_width_one_plane
[12:33:50] [PASSED] drm_test_format_block_width_two_plane
[12:33:50] [PASSED] drm_test_format_block_width_three_plane
[12:33:50] [PASSED] drm_test_format_block_width_tiled
[12:33:50] [PASSED] drm_test_format_block_height_invalid
[12:33:50] [PASSED] drm_test_format_block_height_one_plane
[12:33:50] [PASSED] drm_test_format_block_height_two_plane
[12:33:50] [PASSED] drm_test_format_block_height_three_plane
[12:33:50] [PASSED] drm_test_format_block_height_tiled
[12:33:50] [PASSED] drm_test_format_min_pitch_invalid
[12:33:50] [PASSED] drm_test_format_min_pitch_one_plane_8bpp
[12:33:50] [PASSED] drm_test_format_min_pitch_one_plane_16bpp
[12:33:50] [PASSED] drm_test_format_min_pitch_one_plane_24bpp
[12:33:50] [PASSED] drm_test_format_min_pitch_one_plane_32bpp
[12:33:50] [PASSED] drm_test_format_min_pitch_two_plane
[12:33:50] [PASSED] drm_test_format_min_pitch_three_plane_8bpp
[12:33:50] [PASSED] drm_test_format_min_pitch_tiled
[12:33:50] =================== [PASSED] drm_format ====================
[12:33:50] ============== drm_framebuffer (10 subtests) ===============
[12:33:50] ========== drm_test_framebuffer_check_src_coords  ==========
[12:33:50] [PASSED] Success: source fits into fb
[12:33:50] [PASSED] Fail: overflowing fb with x-axis coordinate
[12:33:50] [PASSED] Fail: overflowing fb with y-axis coordinate
[12:33:50] [PASSED] Fail: overflowing fb with source width
[12:33:50] [PASSED] Fail: overflowing fb with source height
[12:33:50] ====== [PASSED] drm_test_framebuffer_check_src_coords ======
[12:33:50] [PASSED] drm_test_framebuffer_cleanup
[12:33:50] =============== drm_test_framebuffer_create  ===============
[12:33:50] [PASSED] ABGR8888 normal sizes
[12:33:50] [PASSED] ABGR8888 max sizes
[12:33:50] [PASSED] ABGR8888 pitch greater than min required
[12:33:50] [PASSED] ABGR8888 pitch less than min required
[12:33:50] [PASSED] ABGR8888 Invalid width
[12:33:50] [PASSED] ABGR8888 Invalid buffer handle
[12:33:50] [PASSED] No pixel format
[12:33:50] [PASSED] ABGR8888 Width 0
[12:33:50] [PASSED] ABGR8888 Height 0
[12:33:50] [PASSED] ABGR8888 Out of bound height * pitch combination
[12:33:50] [PASSED] ABGR8888 Large buffer offset
[12:33:50] [PASSED] ABGR8888 Buffer offset for inexistent plane
[12:33:50] [PASSED] ABGR8888 Invalid flag
[12:33:50] [PASSED] ABGR8888 Set DRM_MODE_FB_MODIFIERS without modifiers
[12:33:50] [PASSED] ABGR8888 Valid buffer modifier
[12:33:50] [PASSED] ABGR8888 Invalid buffer modifier(DRM_FORMAT_MOD_SAMSUNG_64_32_TILE)
[12:33:50] [PASSED] ABGR8888 Extra pitches without DRM_MODE_FB_MODIFIERS
[12:33:50] [PASSED] ABGR8888 Extra pitches with DRM_MODE_FB_MODIFIERS
[12:33:50] [PASSED] NV12 Normal sizes
[12:33:50] [PASSED] NV12 Max sizes
[12:33:50] [PASSED] NV12 Invalid pitch
[12:33:50] [PASSED] NV12 Invalid modifier/missing DRM_MODE_FB_MODIFIERS flag
[12:33:50] [PASSED] NV12 different  modifier per-plane
[12:33:50] [PASSED] NV12 with DRM_FORMAT_MOD_SAMSUNG_64_32_TILE
[12:33:50] [PASSED] NV12 Valid modifiers without DRM_MODE_FB_MODIFIERS
[12:33:50] [PASSED] NV12 Modifier for inexistent plane
[12:33:50] [PASSED] NV12 Handle for inexistent plane
[12:33:50] [PASSED] NV12 Handle for inexistent plane without DRM_MODE_FB_MODIFIERS
[12:33:50] [PASSED] YVU420 DRM_MODE_FB_MODIFIERS set without modifier
[12:33:50] [PASSED] YVU420 Normal sizes
[12:33:50] [PASSED] YVU420 Max sizes
[12:33:50] [PASSED] YVU420 Invalid pitch
[12:33:50] [PASSED] YVU420 Different pitches
[12:33:50] [PASSED] YVU420 Different buffer offsets/pitches
[12:33:50] [PASSED] YVU420 Modifier set just for plane 0, without DRM_MODE_FB_MODIFIERS
[12:33:50] [PASSED] YVU420 Modifier set just for planes 0, 1, without DRM_MODE_FB_MODIFIERS
[12:33:50] [PASSED] YVU420 Modifier set just for plane 0, 1, with DRM_MODE_FB_MODIFIERS
[12:33:50] [PASSED] YVU420 Valid modifier
[12:33:50] [PASSED] YVU420 Different modifiers per plane
[12:33:50] [PASSED] YVU420 Modifier for inexistent plane
[12:33:50] [PASSED] YUV420_10BIT Invalid modifier(DRM_FORMAT_MOD_LINEAR)
[12:33:50] [PASSED] X0L2 Normal sizes
[12:33:50] [PASSED] X0L2 Max sizes
[12:33:50] [PASSED] X0L2 Invalid pitch
[12:33:50] [PASSED] X0L2 Pitch greater than minimum required
[12:33:50] [PASSED] X0L2 Handle for inexistent plane
[12:33:50] [PASSED] X0L2 Offset for inexistent plane, without DRM_MODE_FB_MODIFIERS set
[12:33:50] [PASSED] X0L2 Modifier without DRM_MODE_FB_MODIFIERS set
[12:33:50] [PASSED] X0L2 Valid modifier
[12:33:50] [PASSED] X0L2 Modifier for inexistent plane
[12:33:50] =========== [PASSED] drm_test_framebuffer_create ===========
[12:33:50] [PASSED] drm_test_framebuffer_free
[12:33:50] [PASSED] drm_test_framebuffer_init
[12:33:50] [PASSED] drm_test_framebuffer_init_bad_format
[12:33:50] [PASSED] drm_test_framebuffer_init_dev_mismatch
[12:33:50] [PASSED] drm_test_framebuffer_lookup
[12:33:50] [PASSED] drm_test_framebuffer_lookup_inexistent
[12:33:50] [PASSED] drm_test_framebuffer_modifiers_not_supported
[12:33:50] ================= [PASSED] drm_framebuffer =================
[12:33:50] ================ drm_gem_shmem (8 subtests) ================
[12:33:50] [PASSED] drm_gem_shmem_test_obj_create
[12:33:50] [PASSED] drm_gem_shmem_test_obj_create_private
[12:33:50] [PASSED] drm_gem_shmem_test_pin_pages
[12:33:50] [PASSED] drm_gem_shmem_test_vmap
[12:33:50] [PASSED] drm_gem_shmem_test_get_sg_table
[12:33:50] [PASSED] drm_gem_shmem_test_get_pages_sgt
[12:33:50] [PASSED] drm_gem_shmem_test_madvise
[12:33:50] [PASSED] drm_gem_shmem_test_purge
[12:33:50] ================== [PASSED] drm_gem_shmem ==================
[12:33:50] === drm_atomic_helper_connector_hdmi_check (29 subtests) ===
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_auto_cea_mode
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_auto_cea_mode_vic_1
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_full_cea_mode
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_full_cea_mode_vic_1
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_limited_cea_mode
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_limited_cea_mode_vic_1
[12:33:50] ====== drm_test_check_broadcast_rgb_cea_mode_yuv420  =======
[12:33:50] [PASSED] Automatic
[12:33:50] [PASSED] Full
[12:33:50] [PASSED] Limited 16:235
[12:33:50] == [PASSED] drm_test_check_broadcast_rgb_cea_mode_yuv420 ===
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_crtc_mode_changed
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_crtc_mode_not_changed
[12:33:50] [PASSED] drm_test_check_disable_connector
[12:33:50] [PASSED] drm_test_check_hdmi_funcs_reject_rate
[12:33:50] [PASSED] drm_test_check_max_tmds_rate_bpc_fallback_rgb
[12:33:50] [PASSED] drm_test_check_max_tmds_rate_bpc_fallback_yuv420
[12:33:50] [PASSED] drm_test_check_max_tmds_rate_bpc_fallback_ignore_yuv422
[12:33:50] [PASSED] drm_test_check_max_tmds_rate_bpc_fallback_ignore_yuv420
[12:33:50] [PASSED] drm_test_check_driver_unsupported_fallback_yuv420
[12:33:50] [PASSED] drm_test_check_output_bpc_crtc_mode_changed
[12:33:50] [PASSED] drm_test_check_output_bpc_crtc_mode_not_changed
[12:33:50] [PASSED] drm_test_check_output_bpc_dvi
[12:33:50] [PASSED] drm_test_check_output_bpc_format_vic_1
[12:33:50] [PASSED] drm_test_check_output_bpc_format_display_8bpc_only
[12:33:50] [PASSED] drm_test_check_output_bpc_format_display_rgb_only
[12:33:50] [PASSED] drm_test_check_output_bpc_format_driver_8bpc_only
[12:33:50] [PASSED] drm_test_check_output_bpc_format_driver_rgb_only
[12:33:50] [PASSED] drm_test_check_tmds_char_rate_rgb_8bpc
[12:33:50] [PASSED] drm_test_check_tmds_char_rate_rgb_10bpc
[12:33:50] [PASSED] drm_test_check_tmds_char_rate_rgb_12bpc
[12:33:50] ============ drm_test_check_hdmi_color_format  =============
[12:33:50] [PASSED] AUTO -> RGB
[12:33:50] [PASSED] YCBCR422 -> YUV422
[12:33:50] [PASSED] YCBCR420 -> YUV420
[12:33:50] [PASSED] YCBCR444 -> YUV444
[12:33:50] [PASSED] RGB -> RGB
[12:33:50] ======== [PASSED] drm_test_check_hdmi_color_format =========
[12:33:50] ======== drm_test_check_hdmi_color_format_420_only  ========
[12:33:50] [PASSED] RGB should fail
[12:33:50] [PASSED] YUV444 should fail
[12:33:50] [PASSED] YUV422 should fail
[12:33:50] [PASSED] YUV420 should work
[12:33:50] ==== [PASSED] drm_test_check_hdmi_color_format_420_only ====
[12:33:50] ===== [PASSED] drm_atomic_helper_connector_hdmi_check ======
[12:33:50] === drm_atomic_helper_connector_hdmi_reset (6 subtests) ====
[12:33:50] [PASSED] drm_test_check_broadcast_rgb_value
[12:33:50] [PASSED] drm_test_check_bpc_8_value
[12:33:50] [PASSED] drm_test_check_bpc_10_value
[12:33:50] [PASSED] drm_test_check_bpc_12_value
[12:33:50] [PASSED] drm_test_check_format_value
[12:33:50] [PASSED] drm_test_check_tmds_char_value
[12:33:50] ===== [PASSED] drm_atomic_helper_connector_hdmi_reset ======
[12:33:50] = drm_atomic_helper_connector_hdmi_mode_valid (7 subtests) =
[12:33:50] [PASSED] drm_test_check_mode_valid
[12:33:50] [PASSED] drm_test_check_mode_valid_reject
[12:33:50] [PASSED] drm_test_check_mode_valid_reject_rate
[12:33:50] [PASSED] drm_test_check_mode_valid_reject_max_clock
[12:33:50] [PASSED] drm_test_check_mode_valid_yuv420_only_max_clock
[12:33:50] [PASSED] drm_test_check_mode_valid_reject_yuv420_only_connector
[12:33:50] [PASSED] drm_test_check_mode_valid_accept_yuv420_also_connector_rgb
[12:33:50] === [PASSED] drm_atomic_helper_connector_hdmi_mode_valid ===
[12:33:50] = drm_atomic_helper_connector_hdmi_infoframes (5 subtests) =
[12:33:50] [PASSED] drm_test_check_infoframes
[12:33:50] [PASSED] drm_test_check_reject_avi_infoframe
[12:33:50] [PASSED] drm_test_check_reject_hdr_infoframe_bpc_8
[12:33:50] [PASSED] drm_test_check_reject_hdr_infoframe_bpc_10
[12:33:50] [PASSED] drm_test_check_reject_audio_infoframe
[12:33:50] === [PASSED] drm_atomic_helper_connector_hdmi_infoframes ===
[12:33:50] ================= drm_managed (2 subtests) =================
[12:33:50] [PASSED] drm_test_managed_release_action
[12:33:50] [PASSED] drm_test_managed_run_action
[12:33:50] =================== [PASSED] drm_managed ===================
[12:33:50] =================== drm_mm (6 subtests) ====================
[12:33:50] [PASSED] drm_test_mm_init
[12:33:50] [PASSED] drm_test_mm_debug
[12:33:50] [PASSED] drm_test_mm_align32
[12:33:50] [PASSED] drm_test_mm_align64
[12:33:50] [PASSED] drm_test_mm_lowest
[12:33:50] [PASSED] drm_test_mm_highest
[12:33:50] ===================== [PASSED] drm_mm ======================
[12:33:50] ============= drm_modes_analog_tv (5 subtests) =============
[12:33:50] [PASSED] drm_test_modes_analog_tv_mono_576i
[12:33:50] [PASSED] drm_test_modes_analog_tv_ntsc_480i
[12:33:50] [PASSED] drm_test_modes_analog_tv_ntsc_480i_inlined
[12:33:50] [PASSED] drm_test_modes_analog_tv_pal_576i
[12:33:50] [PASSED] drm_test_modes_analog_tv_pal_576i_inlined
[12:33:50] =============== [PASSED] drm_modes_analog_tv ===============
[12:33:50] ============== drm_plane_helper (2 subtests) ===============
[12:33:50] =============== drm_test_check_plane_state  ================
[12:33:50] [PASSED] clipping_simple
[12:33:50] [PASSED] clipping_rotate_reflect
[12:33:50] [PASSED] positioning_simple
[12:33:50] [PASSED] upscaling
[12:33:50] [PASSED] downscaling
[12:33:50] [PASSED] rounding1
[12:33:50] [PASSED] rounding2
[12:33:50] [PASSED] rounding3
[12:33:50] [PASSED] rounding4
[12:33:50] =========== [PASSED] drm_test_check_plane_state ============
[12:33:50] =========== drm_test_check_invalid_plane_state  ============
[12:33:50] [PASSED] positioning_invalid
[12:33:50] [PASSED] upscaling_invalid
[12:33:50] [PASSED] downscaling_invalid
[12:33:50] ======= [PASSED] drm_test_check_invalid_plane_state ========
[12:33:50] ================ [PASSED] drm_plane_helper =================
[12:33:50] ====== drm_connector_helper_tv_get_modes (1 subtest) =======
[12:33:50] ====== drm_test_connector_helper_tv_get_modes_check  =======
[12:33:50] [PASSED] None
[12:33:50] [PASSED] PAL
[12:33:50] [PASSED] NTSC
[12:33:50] [PASSED] Both, NTSC Default
[12:33:50] [PASSED] Both, PAL Default
[12:33:50] [PASSED] Both, NTSC Default, with PAL on command-line
[12:33:50] [PASSED] Both, PAL Default, with NTSC on command-line
[12:33:50] == [PASSED] drm_test_connector_helper_tv_get_modes_check ===
[12:33:50] ======== [PASSED] drm_connector_helper_tv_get_modes ========
[12:33:50] ================== drm_rect (9 subtests) ===================
[12:33:50] [PASSED] drm_test_rect_clip_scaled_div_by_zero
[12:33:50] [PASSED] drm_test_rect_clip_scaled_not_clipped
[12:33:50] [PASSED] drm_test_rect_clip_scaled_clipped
[12:33:50] [PASSED] drm_test_rect_clip_scaled_signed_vs_unsigned
[12:33:50] ================= drm_test_rect_intersect  =================
[12:33:50] [PASSED] top-left x bottom-right: 2x2+1+1 x 2x2+0+0
[12:33:50] [PASSED] top-right x bottom-left: 2x2+0+0 x 2x2+1-1
[12:33:50] [PASSED] bottom-left x top-right: 2x2+1-1 x 2x2+0+0
[12:33:50] [PASSED] bottom-right x top-left: 2x2+0+0 x 2x2+1+1
[12:33:50] [PASSED] right x left: 2x1+0+0 x 3x1+1+0
[12:33:50] [PASSED] left x right: 3x1+1+0 x 2x1+0+0
[12:33:50] [PASSED] up x bottom: 1x2+0+0 x 1x3+0-1
[12:33:50] [PASSED] bottom x up: 1x3+0-1 x 1x2+0+0
[12:33:50] [PASSED] touching corner: 1x1+0+0 x 2x2+1+1
[12:33:50] [PASSED] touching side: 1x1+0+0 x 1x1+1+0
[12:33:50] [PASSED] equal rects: 2x2+0+0 x 2x2+0+0
[12:33:50] [PASSED] inside another: 2x2+0+0 x 1x1+1+1
[12:33:50] [PASSED] far away: 1x1+0+0 x 1x1+3+6
[12:33:50] [PASSED] points intersecting: 0x0+5+10 x 0x0+5+10
[12:33:50] [PASSED] points not intersecting: 0x0+0+0 x 0x0+5+10
[12:33:50] ============= [PASSED] drm_test_rect_intersect =============
[12:33:50] ================ drm_test_rect_calc_hscale  ================
[12:33:50] [PASSED] normal use
[12:33:50] [PASSED] out of max range
[12:33:50] [PASSED] out of min range
[12:33:50] [PASSED] zero dst
[12:33:50] [PASSED] negative src
[12:33:50] [PASSED] negative dst
[12:33:50] ============ [PASSED] drm_test_rect_calc_hscale ============
[12:33:50] ================ drm_test_rect_calc_vscale  ================
[12:33:50] [PASSED] normal use
[12:33:50] [PASSED] out of max range
[12:33:50] [PASSED] out of min range
[12:33:50] [PASSED] zero dst
[12:33:50] [PASSED] negative src
[12:33:50] [PASSED] negative dst
[12:33:50] ============ [PASSED] drm_test_rect_calc_vscale ============
[12:33:50] ================== drm_test_rect_rotate  ===================
[12:33:50] [PASSED] reflect-x
[12:33:50] [PASSED] reflect-y
[12:33:50] [PASSED] rotate-0
[12:33:50] [PASSED] rotate-90
[12:33:50] [PASSED] rotate-180
[12:33:50] [PASSED] rotate-270
[12:33:50] ============== [PASSED] drm_test_rect_rotate ===============
[12:33:50] ================ drm_test_rect_rotate_inv  =================
[12:33:50] [PASSED] reflect-x
[12:33:50] [PASSED] reflect-y
[12:33:50] [PASSED] rotate-0
[12:33:50] [PASSED] rotate-90
[12:33:50] [PASSED] rotate-180
[12:33:50] [PASSED] rotate-270
[12:33:50] ============ [PASSED] drm_test_rect_rotate_inv =============
[12:33:50] ==================== [PASSED] drm_rect =====================
[12:33:50] ============ drm_sysfb_modeset_test (1 subtest) ============
[12:33:50] ============ drm_test_sysfb_build_fourcc_list  =============
[12:33:50] [PASSED] no native formats
[12:33:50] [PASSED] XRGB8888 as native format
[12:33:50] [PASSED] remove duplicates
[12:33:50] [PASSED] convert alpha formats
[12:33:50] [PASSED] random formats
[12:33:50] ======== [PASSED] drm_test_sysfb_build_fourcc_list =========
[12:33:50] ============= [PASSED] drm_sysfb_modeset_test ==============
[12:33:50] ================== drm_fixp (2 subtests) ===================
[12:33:50] [PASSED] drm_test_int2fixp
[12:33:50] [PASSED] drm_test_sm2fixp
[12:33:50] ==================== [PASSED] drm_fixp =====================
[12:33:50] ============================================================
[12:33:50] Testing complete. Ran 637 tests: passed: 637
[12:33:50] Elapsed time: 26.605s total, 1.817s configuring, 24.568s building, 0.203s running

+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/ttm/tests/.kunitconfig [12:33:50] Configuring KUnit Kernel ...
Regenerating .config ...
Populating config with:

$ make ARCH=um O=.kunit olddefconfig
[12:33:52] Building KUnit Kernel ...
Populating config with:
$ make ARCH=um O=.kunit olddefconfig
Building with:
$ make all compile_commands.json scripts_gdb ARCH=um O=.kunit --jobs=48
[12:34:02] Starting KUnit Kernel (1/1)...
[12:34:02] ============================================================
Running tests with:
$ .kunit/linux kunit.enable=1 mem=1G console=tty kunit_shutdown=halt
[12:34:02] ================= ttm_device (5 subtests) ==================
[12:34:02] [PASSED] ttm_device_init_basic
[12:34:02] [PASSED] ttm_device_init_multiple
[12:34:02] [PASSED] ttm_device_fini_basic
[12:34:02] [PASSED] ttm_device_init_no_vma_man
[12:34:02] ================== ttm_device_init_pools  ==================
[12:34:02] [PASSED] No DMA allocations, no DMA32 required
[12:34:02] [PASSED] DMA allocations, DMA32 required
[12:34:02] [PASSED] No DMA allocations, DMA32 required
[12:34:02] [PASSED] DMA allocations, no DMA32 required
[12:34:02] ============== [PASSED] ttm_device_init_pools ==============
[12:34:02] =================== [PASSED] ttm_device ====================
[12:34:02] ================== ttm_pool (8 subtests) ===================
[12:34:02] ================== ttm_pool_alloc_basic  ===================
[12:34:02] [PASSED] One page
[12:34:02] [PASSED] More than one page
[12:34:02] [PASSED] Above the allocation limit
[12:34:02] [PASSED] One page, with coherent DMA mappings enabled
[12:34:02] [PASSED] Above the allocation limit, with coherent DMA mappings enabled
[12:34:02] ============== [PASSED] ttm_pool_alloc_basic ===============
[12:34:02] ============== ttm_pool_alloc_basic_dma_addr  ==============
[12:34:02] [PASSED] One page
[12:34:02] [PASSED] More than one page
[12:34:02] [PASSED] Above the allocation limit
[12:34:02] [PASSED] One page, with coherent DMA mappings enabled
[12:34:02] [PASSED] Above the allocation limit, with coherent DMA mappings enabled
[12:34:02] ========== [PASSED] ttm_pool_alloc_basic_dma_addr ==========
[12:34:02] [PASSED] ttm_pool_alloc_order_caching_match
[12:34:02] [PASSED] ttm_pool_alloc_caching_mismatch
[12:34:02] [PASSED] ttm_pool_alloc_order_mismatch
[12:34:02] [PASSED] ttm_pool_free_dma_alloc
[12:34:02] [PASSED] ttm_pool_free_no_dma_alloc
[12:34:02] [PASSED] ttm_pool_fini_basic
[12:34:02] ==================== [PASSED] ttm_pool =====================
[12:34:02] ================ ttm_resource (8 subtests) =================
[12:34:02] ================= ttm_resource_init_basic  =================
[12:34:02] [PASSED] Init resource in TTM_PL_SYSTEM
[12:34:02] [PASSED] Init resource in TTM_PL_VRAM
[12:34:02] [PASSED] Init resource in a private placement
[12:34:02] [PASSED] Init resource in TTM_PL_SYSTEM, set placement flags
[12:34:02] ============= [PASSED] ttm_resource_init_basic =============
[12:34:02] [PASSED] ttm_resource_init_pinned
[12:34:02] [PASSED] ttm_resource_fini_basic
[12:34:02] [PASSED] ttm_resource_manager_init_basic
[12:34:02] [PASSED] ttm_resource_manager_usage_basic
[12:34:02] [PASSED] ttm_resource_manager_set_used_basic
[12:34:02] [PASSED] ttm_sys_man_alloc_basic
[12:34:02] [PASSED] ttm_sys_man_free_basic
[12:34:02] ================== [PASSED] ttm_resource ===================
[12:34:02] =================== ttm_tt (15 subtests) ===================
[12:34:02] ==================== ttm_tt_init_basic  ====================
[12:34:02] [PASSED] Page-aligned size
[12:34:02] [PASSED] Extra pages requested
[12:34:02] ================ [PASSED] ttm_tt_init_basic ================
[12:34:02] [PASSED] ttm_tt_init_misaligned
[12:34:02] [PASSED] ttm_tt_fini_basic
[12:34:02] [PASSED] ttm_tt_fini_sg
[12:34:02] [PASSED] ttm_tt_fini_shmem
[12:34:02] [PASSED] ttm_tt_create_basic
[12:34:02] [PASSED] ttm_tt_create_invalid_bo_type
[12:34:02] [PASSED] ttm_tt_create_ttm_exists
[12:34:02] [PASSED] ttm_tt_create_failed
[12:34:02] [PASSED] ttm_tt_destroy_basic
[12:34:02] [PASSED] ttm_tt_populate_null_ttm
[12:34:02] [PASSED] ttm_tt_populate_populated_ttm
[12:34:02] [PASSED] ttm_tt_unpopulate_basic
[12:34:02] [PASSED] ttm_tt_unpopulate_empty_ttm
[12:34:02] [PASSED] ttm_tt_swapin_basic
[12:34:02] ===================== [PASSED] ttm_tt ======================
[12:34:02] =================== ttm_bo (14 subtests) ===================
[12:34:02] =========== ttm_bo_reserve_optimistic_no_ticket  ===========
[12:34:02] [PASSED] Cannot be interrupted and sleeps
[12:34:02] [PASSED] Cannot be interrupted, locks straight away
[12:34:02] [PASSED] Can be interrupted, sleeps
[12:34:02] ======= [PASSED] ttm_bo_reserve_optimistic_no_ticket =======
[12:34:02] [PASSED] ttm_bo_reserve_locked_no_sleep
[12:34:02] [PASSED] ttm_bo_reserve_no_wait_ticket
[12:34:02] [PASSED] ttm_bo_reserve_double_resv
[12:34:02] [PASSED] ttm_bo_reserve_interrupted
[12:34:02] [PASSED] ttm_bo_reserve_deadlock
[12:34:02] [PASSED] ttm_bo_unreserve_basic
[12:34:02] [PASSED] ttm_bo_unreserve_pinned
[12:34:02] [PASSED] ttm_bo_unreserve_bulk
[12:34:02] [PASSED] ttm_bo_fini_basic
[12:34:02] [PASSED] ttm_bo_fini_shared_resv
[12:34:02] [PASSED] ttm_bo_pin_basic
[12:34:02] [PASSED] ttm_bo_pin_unpin_resource
[12:34:02] [PASSED] ttm_bo_multiple_pin_one_unpin
[12:34:02] ===================== [PASSED] ttm_bo ======================
[12:34:02] ============== ttm_bo_validate (22 subtests) ===============
[12:34:02] ============== ttm_bo_init_reserved_sys_man  ===============
[12:34:02] [PASSED] Buffer object for userspace
[12:34:02] [PASSED] Kernel buffer object
[12:34:02] [PASSED] Shared buffer object
[12:34:02] ========== [PASSED] ttm_bo_init_reserved_sys_man ===========
[12:34:02] ============== ttm_bo_init_reserved_mock_man  ==============
[12:34:02] [PASSED] Buffer object for userspace
[12:34:02] [PASSED] Kernel buffer object
[12:34:02] [PASSED] Shared buffer object
[12:34:02] ========== [PASSED] ttm_bo_init_reserved_mock_man ==========
[12:34:02] [PASSED] ttm_bo_init_reserved_resv
[12:34:02] ================== ttm_bo_validate_basic  ==================
[12:34:02] [PASSED] Buffer object for userspace
[12:34:02] [PASSED] Kernel buffer object
[12:34:02] [PASSED] Shared buffer object
[12:34:02] ============== [PASSED] ttm_bo_validate_basic ==============
[12:34:02] [PASSED] ttm_bo_validate_invalid_placement
[12:34:02] ============= ttm_bo_validate_same_placement  ==============
[12:34:02] [PASSED] System manager
[12:34:02] [PASSED] VRAM manager
[12:34:02] ========= [PASSED] ttm_bo_validate_same_placement ==========
[12:34:02] [PASSED] ttm_bo_validate_failed_alloc
[12:34:02] [PASSED] ttm_bo_validate_pinned
[12:34:02] [PASSED] ttm_bo_validate_busy_placement
[12:34:02] ================ ttm_bo_validate_multihop  =================
[12:34:02] [PASSED] Buffer object for userspace
[12:34:02] [PASSED] Kernel buffer object
[12:34:02] [PASSED] Shared buffer object
[12:34:02] ============ [PASSED] ttm_bo_validate_multihop =============
[12:34:02] ========== ttm_bo_validate_no_placement_signaled  ==========
[12:34:02] [PASSED] Buffer object in system domain, no page vector
[12:34:02] [PASSED] Buffer object in system domain with an existing page vector
[12:34:02] ====== [PASSED] ttm_bo_validate_no_placement_signaled ======
[12:34:02] ======== ttm_bo_validate_no_placement_not_signaled  ========
[12:34:02] [PASSED] Buffer object for userspace
[12:34:02] [PASSED] Kernel buffer object
[12:34:02] [PASSED] Shared buffer object
[12:34:02] ==== [PASSED] ttm_bo_validate_no_placement_not_signaled ====
[12:34:02] [PASSED] ttm_bo_validate_move_fence_signaled
[12:34:02] ========= ttm_bo_validate_move_fence_not_signaled  =========
[12:34:02] [PASSED] Waits for GPU
[12:34:02] [PASSED] Tries to lock straight away
[12:34:02] ===== [PASSED] ttm_bo_validate_move_fence_not_signaled =====
[12:34:02] [PASSED] ttm_bo_validate_swapout
[12:34:02] [PASSED] ttm_bo_validate_happy_evict
[12:34:02] [PASSED] ttm_bo_validate_all_pinned_evict
[12:34:02] [PASSED] ttm_bo_validate_allowed_only_evict
[12:34:02] [PASSED] ttm_bo_validate_deleted_evict
[12:34:02] [PASSED] ttm_bo_validate_busy_domain_evict
[12:34:02] [PASSED] ttm_bo_validate_evict_gutting
[12:34:02] [PASSED] ttm_bo_validate_recrusive_evict
[12:34:02] ================= [PASSED] ttm_bo_validate =================
[12:34:02] ============================================================
[12:34:02] Testing complete. Ran 102 tests: passed: 102
[12:34:02] Elapsed time: 11.969s total, 1.791s configuring, 9.963s building, 0.184s running

+ cleanup
++ stat -c %u:%g /kernel
+ chown -R 1003:1003 /kernel

#28 lore raw

✗ Xe.CI.BAT: failure for drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)

In reply to Tales A. Mendonça
== Series Details ==

== Series Details ==

Series: drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)
URL : https://patchwork.freedesktop.org/series/171539/
State : failure

== Summary ==

CI Bug Log - changes from xe-5545-4c13a67db118ee0019231dbc1362bcb8115ac660_BAT -> xe-pw-171539v2_BAT
====================================================

Summary
-------

**WARNING**

Minor unknown changes coming with xe-pw-171539v2_BAT need to be verified manually.

If you think the reported changes have nothing to do with the changes
introduced in xe-pw-171539v2_BAT, please notify your bug team ([email protected]) to allow them to document this new failure mode, which will reduce false positives in CI.

Participating hosts (14 -> 15)
------------------------------

Additional (1): bat-bmg-2

Possible new issues
-------------------

Here are the unknown changes that may have been introduced in xe-pw-171539v2_BAT:

### IGT changes ###

#### Warnings ####

* igt@kms_frontbuffer_tracking@basic:
- bat-nvls-2: [SKIP][1] ([Intel XE#7779]) -> [SKIP][2]
[1]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-5545-4c13a67db118ee0019231dbc1362bcb8115ac660/bat-nvls-2/igt@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-nvls-2/igt@[email protected]

Known issues
------------

Here are the changes found in xe-pw-171539v2_BAT that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@fbdev@write:
- bat-bmg-2: NOTRUN -> [SKIP][3] ([Intel XE#2134]) +4 other tests skip
[3]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@kms_addfb_basic@addfb25-y-tiled-small-legacy:
- bat-bmg-2: NOTRUN -> [SKIP][4] ([Intel XE#2233])
[4]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@kms_cursor_legacy@basic-flip-after-cursor-legacy:
- bat-bmg-2: NOTRUN -> [SKIP][5] ([Intel XE#2489] / [Intel XE#3419]) +13 other tests skip
[5]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@kms_flip@basic-flip-vs-modeset:
- bat-bmg-2: NOTRUN -> [SKIP][6] ([Intel XE#2482]) +3 other tests skip
[6]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@kms_frontbuffer_tracking@basic:
- bat-bmg-2: NOTRUN -> [SKIP][7] ([Intel XE#2434] / [Intel XE#2548] / [Intel XE#6314])
[7]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@kms_psr@psr-sprite-plane-onoff:
- bat-bmg-2: NOTRUN -> [SKIP][8] ([Intel XE#2234] / [Intel XE#2850]) +2 other tests skip
[8]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@xe_exec_multi_queue@priority:
- bat-bmg-2: NOTRUN -> [SKIP][9] ([Intel XE#8364]) +13 other tests skip
[9]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@xe_live_ktest@xe_bo@xe_ccs_migrate_kunit:
- bat-bmg-2: NOTRUN -> [SKIP][10] ([Intel XE#2229])
[10]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@xe_live_ktest@xe_bo@xe_ccs_migrate_kunit.html

* igt@xe_pat@pat-index-xehpc:
- bat-bmg-2: NOTRUN -> [SKIP][11] ([Intel XE#1420] / [Intel XE#7590])
[11]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@xe_pat@pat-index-xelp:
- bat-bmg-2: NOTRUN -> [SKIP][12] ([Intel XE#2245] / [Intel XE#7590])
[12]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

* igt@xe_pat@pat-index-xelpg:
- bat-bmg-2: NOTRUN -> [SKIP][13] ([Intel XE#2236] / [Intel XE#7590])
[13]: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/bat-bmg-2/igt@[email protected]

[Intel XE#1420]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/1420
[Intel XE#2134]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2134
[Intel XE#2229]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2229
[Intel XE#2233]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2233
[Intel XE#2234]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2234
[Intel XE#2236]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2236
[Intel XE#2245]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2245
[Intel XE#2434]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2434
[Intel XE#2482]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2482
[Intel XE#2489]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2489
[Intel XE#2548]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2548
[Intel XE#2850]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/2850
[Intel XE#3419]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/3419
[Intel XE#6314]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/6314
[Intel XE#7590]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/7590
[Intel XE#7779]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/7779
[Intel XE#8364]: https://gitlab.freedesktop.org/drm/xe/kernel/issues/8364

Build changes
-------------

* Linux: xe-5545-4c13a67db118ee0019231dbc1362bcb8115ac660 -> xe-pw-171539v2

IGT_9038: 9038
xe-5545-4c13a67db118ee0019231dbc1362bcb8115ac660: 4c13a67db118ee0019231dbc1362bcb8115ac660
xe-pw-171539v2: 171539v2

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/index.html

#29 lore raw

✗ Xe.CI.FULL: failure for drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)

In reply to Tales A. Mendonça
== Series Details ==

== Series Details ==

Series: drm/xe: diagnostics and workaround for GuC TLB invalidation ack stalls on ARL (rev2)
URL : https://patchwork.freedesktop.org/series/171539/
State : failure

== Summary ==

ERROR: The runconfig 'xe-5545-4c13a67db118ee0019231dbc1362bcb8115ac660_FULL' does not exist in the database

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-171539v2/index.html