Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
accel/amdxdna: Fix dead lock for suspend and resume
When an application issues a query IOCTL while auto suspend is running,
a deadlock can occur. The query path holds dev_lock and then calls
pm_runtime_resume_and_get(), which waits for the ongoing suspend to
complete. Meanwhile, the suspend callback attempts to acquire dev_lock
and blocks, resulting in a deadlock.
Fix this by releasing dev_lock before calling pm_runtime_resume_and_get()
and reacquiring it after the call completes. Also acquire dev_lock in the
resume callback to keep the locking consistent.
Fixes: 063db451832b ("accel/amdxdna: Enhance runtime power management")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260211204644.722758-1-lizhi.hou@amd.com
+26
-19
+2
-2
drivers/accel/amdxdna/aie2_ctx.c
+2
-2
drivers/accel/amdxdna/aie2_ctx.c
···
629
629
goto free_entity;
630
630
}
631
631
632
-
ret = amdxdna_pm_resume_get(xdna);
632
+
ret = amdxdna_pm_resume_get_locked(xdna);
633
633
if (ret)
634
634
goto free_col_list;
635
635
···
760
760
if (!hwctx->cus)
761
761
return -ENOMEM;
762
762
763
-
ret = amdxdna_pm_resume_get(xdna);
763
+
ret = amdxdna_pm_resume_get_locked(xdna);
764
764
if (ret)
765
765
goto free_cus;
766
766
+3
-4
drivers/accel/amdxdna/aie2_pci.c
+3
-4
drivers/accel/amdxdna/aie2_pci.c
···
451
451
{
452
452
struct amdxdna_client *client;
453
453
454
-
guard(mutex)(&xdna->dev_lock);
455
454
list_for_each_entry(client, &xdna->client_list, node)
456
455
aie2_hwctx_suspend(client);
457
456
···
950
951
if (!drm_dev_enter(&xdna->ddev, &idx))
951
952
return -ENODEV;
952
953
953
-
ret = amdxdna_pm_resume_get(xdna);
954
+
ret = amdxdna_pm_resume_get_locked(xdna);
954
955
if (ret)
955
956
goto dev_exit;
956
957
···
1043
1044
if (!drm_dev_enter(&xdna->ddev, &idx))
1044
1045
return -ENODEV;
1045
1046
1046
-
ret = amdxdna_pm_resume_get(xdna);
1047
+
ret = amdxdna_pm_resume_get_locked(xdna);
1047
1048
if (ret)
1048
1049
goto dev_exit;
1049
1050
···
1133
1134
if (!drm_dev_enter(&xdna->ddev, &idx))
1134
1135
return -ENODEV;
1135
1136
1136
-
ret = amdxdna_pm_resume_get(xdna);
1137
+
ret = amdxdna_pm_resume_get_locked(xdna);
1137
1138
if (ret)
1138
1139
goto dev_exit;
1139
1140
+1
-1
drivers/accel/amdxdna/aie2_pm.c
+1
-1
drivers/accel/amdxdna/aie2_pm.c
+7
-12
drivers/accel/amdxdna/amdxdna_ctx.c
+7
-12
drivers/accel/amdxdna/amdxdna_ctx.c
···
266
266
struct amdxdna_drm_config_hwctx *args = data;
267
267
struct amdxdna_dev *xdna = to_xdna_dev(dev);
268
268
struct amdxdna_hwctx *hwctx;
269
-
int ret, idx;
270
269
u32 buf_size;
271
270
void *buf;
271
+
int ret;
272
272
u64 val;
273
273
274
274
if (XDNA_MBZ_DBG(xdna, &args->pad, sizeof(args->pad)))
···
310
310
return -EINVAL;
311
311
}
312
312
313
-
mutex_lock(&xdna->dev_lock);
314
-
idx = srcu_read_lock(&client->hwctx_srcu);
313
+
guard(mutex)(&xdna->dev_lock);
315
314
hwctx = xa_load(&client->hwctx_xa, args->handle);
316
315
if (!hwctx) {
317
316
XDNA_DBG(xdna, "PID %d failed to get hwctx %d", client->pid, args->handle);
318
317
ret = -EINVAL;
319
-
goto unlock_srcu;
318
+
goto free_buf;
320
319
}
321
320
322
321
ret = xdna->dev_info->ops->hwctx_config(hwctx, args->param_type, val, buf, buf_size);
323
322
324
-
unlock_srcu:
325
-
srcu_read_unlock(&client->hwctx_srcu, idx);
326
-
mutex_unlock(&xdna->dev_lock);
323
+
free_buf:
327
324
kfree(buf);
328
325
return ret;
329
326
}
···
331
334
struct amdxdna_hwctx *hwctx;
332
335
struct amdxdna_gem_obj *abo;
333
336
struct drm_gem_object *gobj;
334
-
int ret, idx;
337
+
int ret;
335
338
336
339
if (!xdna->dev_info->ops->hwctx_sync_debug_bo)
337
340
return -EOPNOTSUPP;
···
342
345
343
346
abo = to_xdna_obj(gobj);
344
347
guard(mutex)(&xdna->dev_lock);
345
-
idx = srcu_read_lock(&client->hwctx_srcu);
346
348
hwctx = xa_load(&client->hwctx_xa, abo->assigned_hwctx);
347
349
if (!hwctx) {
348
350
ret = -EINVAL;
349
-
goto unlock_srcu;
351
+
goto put_obj;
350
352
}
351
353
352
354
ret = xdna->dev_info->ops->hwctx_sync_debug_bo(hwctx, debug_bo_hdl);
353
355
354
-
unlock_srcu:
355
-
srcu_read_unlock(&client->hwctx_srcu, idx);
356
+
put_obj:
356
357
drm_gem_object_put(gobj);
357
358
return ret;
358
359
}
+2
drivers/accel/amdxdna/amdxdna_pm.c
+2
drivers/accel/amdxdna/amdxdna_pm.c
···
16
16
struct amdxdna_dev *xdna = to_xdna_dev(dev_get_drvdata(dev));
17
17
int ret = -EOPNOTSUPP;
18
18
19
+
guard(mutex)(&xdna->dev_lock);
19
20
if (xdna->dev_info->ops->suspend)
20
21
ret = xdna->dev_info->ops->suspend(xdna);
21
22
···
29
28
struct amdxdna_dev *xdna = to_xdna_dev(dev_get_drvdata(dev));
30
29
int ret = -EOPNOTSUPP;
31
30
31
+
guard(mutex)(&xdna->dev_lock);
32
32
if (xdna->dev_info->ops->resume)
33
33
ret = xdna->dev_info->ops->resume(xdna);
34
34
+11
drivers/accel/amdxdna/amdxdna_pm.h
+11
drivers/accel/amdxdna/amdxdna_pm.h
···
15
15
void amdxdna_pm_init(struct amdxdna_dev *xdna);
16
16
void amdxdna_pm_fini(struct amdxdna_dev *xdna);
17
17
18
+
static inline int amdxdna_pm_resume_get_locked(struct amdxdna_dev *xdna)
19
+
{
20
+
int ret;
21
+
22
+
mutex_unlock(&xdna->dev_lock);
23
+
ret = amdxdna_pm_resume_get(xdna);
24
+
mutex_lock(&xdna->dev_lock);
25
+
26
+
return ret;
27
+
}
28
+
18
29
#endif /* _AMDXDNA_PM_H_ */