Skip to contentSkip to conversation

drm/xe: fixing endless PF range response

PATCH New [email protected] 3 participants 4 messages Oct 6

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover drm/xe: fixing endless PF range response

From: Fei Yang <[email protected]>

xe4_guc_ack_fault_range() reads vma_info, which is not updated by the SVM handler. This causes the ack fault to loop through the whole address space with 1G pages, and that is wrong and the loop takes a long while to complete.

Fei Yang (1):
drm/xe: fixing endless PF range response

 drivers/gpu/drm/xe/xe_pagefault.c | 7 ++++++-
 drivers/gpu/drm/xe/xe_pagefault.h | 2 ++
 2 files changed, 8 insertions(+), 1 deletion(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/1 drm/xe: fixing endless PF range response

From: Fei Yang <[email protected]>

[334.635636] PF range response: start=0x80200000  end=0xc0200000 pages=262144 (more ranges pending)
...
[519.065737] PF range response: start=0x5ca140200000 ...  (still going, 94888 messages logged)

SVM handler narrows the serviced range through functions xe_pagefault_set_[start|end]_addr, which updates only the consumer.[page|end]_addr. However a different field is used for "immediate ack" cache, xe4_guc_ack_fault_range() reads vma_info. This causes the ack fault to loop through the whole address space with 1G pages, and that is wrong and the loop takes a long while to complete.

VLK-99796

Assisted-by GitHub_Copilot:claude-opus-5
---
 drivers/gpu/drm/xe/xe_pagefault.c | 7 ++++++-
 drivers/gpu/drm/xe/xe_pagefault.h | 2 ++
 2 files changed, 8 insertions(+), 1 deletion(-)
drivers/gpu/drm/xe/xe_pagefault.c +6 -1
index 5247321c805ec..72dfe27025736 100644
@@ -694,6 +694,8 @@ static void xe_pagefault_queue_work(struct work_struct *w)
 		/* Last fault same address, ack immediately */
 		if (xe_pagefault_match(pf, cache_start, cache_end, cache_asid)) {
 			xe_gt_stats_incr(gt, XE_GT_STATS_ID_LAST_PAGEFAULT_COUNT, 1);
+			xe_pagefault_set_start_addr(pf, cache_start);
+			xe_pagefault_set_end_addr(pf, cache_end);
 			goto ack_fault;
 		}
 
@@ -751,9 +753,12 @@ static void xe_pagefault_queue_work(struct work_struct *w)
 				pf = xe_pagefault_queue_unchain_requeue(pf_queue, pf, gt);
 
 			/* Ensure resets are safe */
-			if (pf)
+			if (pf) {
 				pf->consumer.alloc_state =
 					XE_PAGEFAULT_ALLOC_STATE_ACTIVE;
+				xe_pagefault_set_start_addr(pf, cache_start);
+				xe_pagefault_set_end_addr(pf, cache_end);
+			}
 
 			spin_unlock_irq(&pf_queue->lock);
 		}
drivers/gpu/drm/xe/xe_pagefault.h +2 -0
index 799c984dee84f..5ce2f6ed10c56 100644
@@ -98,6 +98,7 @@ xe_pagefault_set_end_addr(struct xe_pagefault *pf, u64 end_addr)
 {
 	pf->consumer.end_addr &= ~XE_PAGEFAULT_END_ADDR_MASK;
 	pf->consumer.end_addr |= end_addr;
+	pf->consumer.vma_info.end_addr = end_addr;
 }
 
 /**
@@ -137,6 +138,7 @@ xe_pagefault_set_start_addr(struct xe_pagefault *pf, u64 start_addr)
 {
 	pf->consumer.page_addr &= XE_PAGEFAULT_ERROR_MASK;
 	pf->consumer.page_addr |= (start_addr & ~XE_PAGEFAULT_ERROR_MASK);
+	pf->consumer.vma_info.start_addr = start_addr;
 }
 
 /**
Signature · 1 line
2.43.0
#2 lore raw
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] drm/xe: build failure due to hallucinated vma_info field --

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] drm/xe: build failure due to hallucinated vma_info field
--

commit 634f5ddbb1931a170d6586933045107086b28191
Author: Fei Yang <[email protected]>
Subject: drm/xe: fixing endless PF range response

This commit modifies the page fault queue worker to immediately update the start and end addresses of a page fault. It also adds assignments to a new vma_info structure inside the page fault consumer.

drivers/gpu/drm/xe/xe_pagefault.h xe_pagefault_set_end_addr(struct xe_pagefault *pf, u64 end_addr) · line 101

Quoted 9 lines
drivers/gpu/drm/xe/xe_pagefault.h +1 -0
index 799c984dee84f..5ce2f6ed10c56 100644
@@ -98,6 +98,7 @@ xe_pagefault_set_end_addr(struct xe_pagefault *pf, u64 end_addr)
 {
 	pf->consumer.end_addr &= ~XE_PAGEFAULT_END_ADDR_MASK;
 	pf->consumer.end_addr |= end_addr;
+	pf->consumer.vma_info.end_addr = end_addr;

[Severity: Low]
Does the consumer structure actually have a vma_info member?

Looking at the definition in drivers/gpu/drm/xe/xe_pagefault_types.h:

struct xe_pagefault {
...

	struct {
		u64 page_addr;
		union {
			struct {
				u8 alloc_state;
...
				u32 id;
			};
			u64 end_addr;
		};
		struct xe_pagefault *next;
	} consumer;

...
};

It appears consumer lacks a vma_info field. Will this code compile, or will it result in a build error complaining about no member named vma_info?

[ ... ]

drivers/gpu/drm/xe/xe_pagefault.h xe_pagefault_set_start_addr(struct xe_pagefault *pf, u64 start_addr) · line 141

Quoted 5 lines
drivers/gpu/drm/xe/xe_pagefault.h +1 -0
@@ -137,6 +138,7 @@ xe_pagefault_set_start_addr(struct xe_pagefault *pf, u64 start_addr)
 {
 	pf->consumer.page_addr &= XE_PAGEFAULT_ERROR_MASK;
 	pf->consumer.page_addr |= (start_addr & ~XE_PAGEFAULT_ERROR_MASK);
+	pf->consumer.vma_info.start_addr = start_addr;

[Severity: Low]
Similarly, since vma_info doesn't exist in consumer, will this assignment also cause a compiler error?

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

✗ CI.KUnit: failure for drm/xe: fixing endless PF range response

In reply to [email protected]
== Series Details ==

== Series Details ==

Series: drm/xe: fixing endless PF range response
URL : https://patchwork.freedesktop.org/series/175636/
State : failure

== Summary ==

+ trap cleanup EXIT
+ kunitconfigs=('/kernel/drivers/gpu/tests/.kunitconfig' '/kernel/drivers/gpu/drm/xe/.kunitconfig' '/kernel/drivers/gpu/drm/tests/.kunitconfig' '/kernel/drivers/gpu/drm/ttm/tests/.kunitconfig' '/kernel/drivers/dma-buf/.kunitconfig')
+ for kcfg in "${kunitconfigs[@]}"
+ [[ ! -f /kernel/drivers/gpu/tests/.kunitconfig ]]
+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/tests/.kunitconfig
[21:35:52] Configuring KUnit Kernel ...
Generating .config ...
Populating config with:

$ make ARCH=um O=.kunit olddefconfig
[21:35:57] 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
[21:36:17] Starting KUnit Kernel (1/1)...
[21:36:17] ============================================================
Running tests with:
$ .kunit/linux kunit.enable=1 mem=1G console=tty kunit_shutdown=halt
[21:36:17] ================= gpu_buddy (14 subtests) ==================
[21:36:17] [PASSED] gpu_test_buddy_alloc_limit
[21:36:17] [PASSED] gpu_test_buddy_alloc_optimistic
[21:36:17] [PASSED] gpu_test_buddy_alloc_pessimistic
[21:36:17] [PASSED] gpu_test_buddy_alloc_pathological
[21:36:17] [PASSED] gpu_test_buddy_alloc_contiguous
[21:36:17] [PASSED] gpu_test_buddy_alloc_clear
[21:36:17] [PASSED] gpu_test_buddy_alloc_range
[21:36:17] [PASSED] gpu_test_buddy_alloc_range_bias
[21:36:18] [PASSED] gpu_test_buddy_fragmentation_performance
[21:36:19] [PASSED] gpu_test_buddy_dirty_tracker_performance
[21:36:19] [PASSED] gpu_test_buddy_alloc_exceeds_max_order
[21:36:19] [PASSED] gpu_test_buddy_offset_aligned_allocation
[21:36:19] [PASSED] gpu_test_buddy_subtree_offset_alignment_stress
[21:36:19] [PASSED] gpu_test_buddy_addr_to_block
[21:36:19] ==================== [PASSED] gpu_buddy ====================
[21:36:19] ============================================================
[21:36:19] Testing complete. Ran 14 tests: passed: 14
[21:36:19] Elapsed time: 26.641s total, 4.466s configuring, 20.307s building, 1.822s running

+ for kcfg in "${kunitconfigs[@]}"
+ [[ ! -f /kernel/drivers/gpu/drm/xe/.kunitconfig ]]
+ /kernel/tools/testing/kunit/kunit.py run --kunitconfig /kernel/drivers/gpu/drm/xe/.kunitconfig ERROR:root:In file included from ../drivers/gpu/drm/xe/xe_device.c:59:
../drivers/gpu/drm/xe/xe_pagefault.h: In function ‘xe_pagefault_set_end_addr’:
../drivers/gpu/drm/xe/xe_pagefault.h:101:21: error: ‘struct <anonymous>’ has no member named ‘vma_info’
101 | pf->consumer.vma_info.end_addr = end_addr;
| ^
../drivers/gpu/drm/xe/xe_pagefault.h: In function ‘xe_pagefault_set_start_addr’:
../drivers/gpu/drm/xe/xe_pagefault.h:141:21: error: ‘struct <anonymous>’ has no member named ‘vma_info’
141 | pf->consumer.vma_info.start_addr = start_addr;
| ^
make[7]: *** [../scripts/Makefile.build:290: drivers/gpu/drm/xe/xe_device.o] Error 1 make[7]: *** Waiting for unfinished jobs....
make[6]: *** [../scripts/Makefile.build:551: drivers/gpu/drm/xe] Error 2
make[6]: *** Waiting for unfinished jobs....
make[5]: *** [../scripts/Makefile.build:551: drivers/gpu/drm] Error 2
make[4]: *** [../scripts/Makefile.build:551: drivers/gpu] Error 2
make[4]: *** Waiting for unfinished jobs....
make[3]: *** [../scripts/Makefile.build:551: drivers] Error 2
make[3]: *** Waiting for unfinished jobs....
make[2]: *** [/kernel/Makefile:2229: .] Error 2
make[1]: *** [/kernel/Makefile:248: __sub-make] Error 2
make: *** [Makefile:248: __sub-make] Error 2

[21:36:19] Configuring KUnit Kernel ...
Regenerating .config ...
Populating config with:

$ make ARCH=um O=.kunit olddefconfig
[21:36:21] 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
+ cleanup
++ stat -c %u:%g /kernel
+ chown -R 1003:1003 /kernel