[PATCH] drm/xe: Take PM ref SVM copy to SRAM
Matthew Brost
matthew.brost at intel.com
Wed Apr 16 18:43:54 UTC 2025
It is possible, however unlikely, for the CPU to access memory which is
in the GPU triggering a fault without a PM reference. Ensure a PM ref is
held when doing a SVM copy to SRAM.
Fixes the below splat found in local testing:
[ 1269.500163] ------------[ cut here ]------------
[ 1269.500167] xe 0000:03:00.0: [drm] Missing outer runtime PM protection
[ 1269.500184] WARNING: CPU: 8 PID: 38648 at drivers/gpu/drm/xe/xe_pm.c:664 xe_pm_runtime_get_noresume+0x86/0xb0 [xe]
[ 1269.500226] Modules linked in: xe drm_gpusvm drm_gpuvm drm_ttm_helper ttm drm_exec gpu_sched drm_suballoc_helper drm_buddy drm_kms_helper snd_hda_codec_realtek snd_hda_codec_generic snd_hda_scodec_component snd_hda_intel snd_intel_dspcfg snd_hda_codec x86_pkg_temp_thermal snd_hwdep coretemp snd_hda_core i2c_i801 i2c_mux snd_pcm wmi_bmof i2c_smbus mei_pxp mei_hdcp video wmi mei_me mei fuse igb e1000e i2c_algo_bit ptp ghash_clmulni_intel pps_core intel_lpss_pci
[ 1269.500257] CPU: 8 UID: 0 PID: 38648 Comm: xe_exec_system_ Tainted: G W 6.15.0-rc2-xe+ #158 PREEMPT(undef)
[ 1269.500260] Tainted: [W]=WARN
[ 1269.500261] Hardware name: Intel Corporation Raptor Lake Client Platform/RPL-S ADP-S DDR5 UDIMM CRB, BIOS RPLSFWI1.R00.3492.A00.2211291114 11/29/2022
[ 1269.500262] RIP: 0010:xe_pm_runtime_get_noresume+0x86/0xb0 [xe]
[ 1269.500293] Code: ee 31 c0 48 85 db 48 0f 44 f8 4c 8b 67 50 4d 85 e4 74 2e e8 6c 0b 9a e1 4c 89 e2 48 c7 c7 80 d5 4e a0 48 89 c6 e8 aa 51 11 e1 <0f> 0b eb c1 48 8b 47 08 f0 ff 80 f8 02 00 00 5b 41 5c c3 cc cc cc
[ 1269.500294] RSP: 0000:ffffc9000ed439c0 EFLAGS: 00010282
[ 1269.500297] RAX: 0000000000000000 RBX: ffff888113568000 RCX: 0000000000000000
[ 1269.500298] RDX: 0000000000000002 RSI: 0000000000000001 RDI: 00000000ffffffff
[ 1269.500299] RBP: ffff888111bdf600 R08: ffff88888d5fffe8 R09: 00000000fffdffff
[ 1269.500300] R10: ffff88888c800000 R11: ffff88888d300000 R12: ffff888103b3dd10
[ 1269.500301] R13: ffffc9000ed43a70 R14: ffff88813e5a52c0 R15: ffff88813e5a52c0
[ 1269.500302] FS: 00007f1e596a3940(0000) GS:ffff88890ac15000(0000) knlGS:0000000000000000
[ 1269.500304] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1269.500305] CR2: 0000563cf48c3ee8 CR3: 000000031055e001 CR4: 0000000000f70ef0
[ 1269.500306] PKRU: 55555554
[ 1269.500307] Call Trace:
[ 1269.500308] <TASK>
[ 1269.500310] xe_sched_job_create+0x159/0x330 [xe]
[ 1269.500342] xe_bb_create_migration_job+0x7c/0x510 [xe]
[ 1269.500359] ? rcu_is_watching+0x11/0x50
[ 1269.500363] ? __kmalloc_cache_noprof+0x255/0x330
[ 1269.500366] ? xelp_pte_encode_addr+0x34/0x1d0 [xe]
[ 1269.500394] xe_migrate_vram+0x2c5/0x620 [xe]
[ 1269.500558] ? __iommu_dma_map+0x99/0x170
[ 1269.500569] xe_svm_copy+0x486/0x620 [xe]
[ 1269.500613] drm_gpusvm_migrate_to_ram+0x290/0x330 [drm_gpusvm]
[ 1269.500624] do_swap_page+0xff7/0x2440
[ 1269.500633] ? __pfx_default_wake_function+0x10/0x10
[ 1269.500640] ? rcu_is_watching+0x11/0x50
[ 1269.500646] __handle_mm_fault+0x617/0x950
[ 1269.500658] handle_mm_fault+0xbf/0x250
[ 1269.500664] do_user_addr_fault+0x177/0x6a0
[ 1269.500672] exc_page_fault+0x63/0x1c0
[ 1269.500678] asm_exc_page_fault+0x26/0x30
[ 1269.500681] RIP: 0033:0x7f1e5b8b1b0f
[ 1269.500684] Code: 15 00 49 8d 0c 1a 49 39 d4 49 89 4c 24 60 0f 95 c2 48 29 d8 0f b6 d2 48 83 c8 01 48 c1 e2 02 48 09 da 48 83 ca 01 49 89 52 08 <48> 89 41 08 49 8d 4a 10 eb af 48 8d 0d 78 ea 12 00 ba 64 10 00 00
[ 1269.500687] RSP: 002b:00007fff7ccb74c0 EFLAGS: 00010206
[ 1269.500690] RAX: 0000000000521121 RBX: 0000000000001010 RCX: 0000563cf48c3ee0
[ 1269.500692] RDX: 0000000000001011 RSI: ffffffffffffff20 RDI: 0000000000000000
[ 1269.500695] RBP: 00007fff7ccb7540 R08: 0000000000000000 R09: 0000000000000001
[ 1269.500697] R10: 0000563cf48c2ed0 R11: 0000000000000206 R12: 00007f1e5ba11ac0
[ 1269.500699] R13: 0000000000001000 R14: 0000000000000000 R15: 00007f1e5ba11b20
[ 1269.500708] </TASK>
[ 1269.500710] irq event stamp: 176580299
[ 1269.500712] hardirqs last enabled at (176580305): [<ffffffff813447f6>] __up_console_sem+0x66/0x70
[ 1269.500716] hardirqs last disabled at (176580310): [<ffffffff813447db>] __up_console_sem+0x4b/0x70
[ 1269.500719] softirqs last enabled at (176580168): [<ffffffff812a637e>] __irq_exit_rcu+0xbe/0x110
[ 1269.500723] softirqs last disabled at (176579351): [<ffffffff812a637e>] __irq_exit_rcu+0xbe/0x110
[ 1269.500726] ---[ end trace 0000000000000000 ]---
Fixes: c5b3eb5a906c ("drm/xe: Add GPUSVM device memory copy vfunc functions")
Cc: stable at vger.kernel.org
Signed-off-by: Matthew Brost <matthew.brost at intel.com>
---
drivers/gpu/drm/xe/xe_svm.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c
index 92aebb6b0902..44960d47874e 100644
--- a/drivers/gpu/drm/xe/xe_svm.c
+++ b/drivers/gpu/drm/xe/xe_svm.c
@@ -8,6 +8,7 @@
#include "xe_gt_tlb_invalidation.h"
#include "xe_migrate.h"
#include "xe_module.h"
+#include "xe_pm.h"
#include "xe_pt.h"
#include "xe_svm.h"
#include "xe_ttm_vram_mgr.h"
@@ -378,7 +379,7 @@ static int xe_svm_copy(struct page **pages, dma_addr_t *dma_addr,
unsigned long npages, const enum xe_svm_copy_dir dir)
{
struct xe_vram_region *vr = NULL;
- struct xe_tile *tile;
+ struct xe_tile *tile = NULL;
struct dma_fence *fence = NULL;
unsigned long i;
#define XE_VRAM_ADDR_INVALID ~0x0ull
@@ -412,6 +413,8 @@ static int xe_svm_copy(struct page **pages, dma_addr_t *dma_addr,
if (!vr && spage) {
vr = page_to_vr(spage);
tile = vr_to_tile(vr);
+ if (dir == XE_SVM_COPY_TO_SRAM)
+ xe_pm_runtime_get(tile->xe);
}
XE_WARN_ON(spage && page_to_vr(spage) != vr);
@@ -510,6 +513,8 @@ static int xe_svm_copy(struct page **pages, dma_addr_t *dma_addr,
dma_fence_wait(fence, false);
dma_fence_put(fence);
}
+ if (tile && dir == XE_SVM_COPY_TO_SRAM)
+ xe_pm_runtime_put(tile->xe);
return err;
#undef XE_MIGRATE_CHUNK_SIZE
--
2.34.1
More information about the Intel-xe
mailing list