Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
accel/amdxdna: Fix runtime suspend deadlock when there is pending job
The runtime suspend callback drains the running job workqueue before
suspending the device. If a job is still executing and calls
pm_runtime_resume_and_get(), it can deadlock with the runtime suspend
path.
Fix this by moving pm_runtime_resume_and_get() from the job execution
routine to the job submission routine, ensuring the device is resumed
before the job is queued and avoiding the deadlock during runtime
suspend.
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/20260310180058.336348-1-lizhi.hou@amd.com
+12
-12
+2
-12
drivers/accel/amdxdna/aie2_ctx.c
+2
-12
drivers/accel/amdxdna/aie2_ctx.c
···
165
165
166
166
trace_xdna_job(&job->base, job->hwctx->name, "signaled fence", job->seq);
167
167
168
-
amdxdna_pm_suspend_put(job->hwctx->client->xdna);
169
168
job->hwctx->priv->completed++;
170
169
dma_fence_signal(fence);
171
170
···
289
290
struct dma_fence *fence;
290
291
int ret;
291
292
292
-
ret = amdxdna_pm_resume_get(hwctx->client->xdna);
293
-
if (ret)
293
+
if (!hwctx->priv->mbox_chann)
294
294
return NULL;
295
295
296
-
if (!hwctx->priv->mbox_chann) {
297
-
amdxdna_pm_suspend_put(hwctx->client->xdna);
298
-
return NULL;
299
-
}
300
-
301
-
if (!mmget_not_zero(job->mm)) {
302
-
amdxdna_pm_suspend_put(hwctx->client->xdna);
296
+
if (!mmget_not_zero(job->mm))
303
297
return ERR_PTR(-ESRCH);
304
-
}
305
298
306
299
kref_get(&job->refcnt);
307
300
fence = dma_fence_get(job->fence);
···
324
333
325
334
out:
326
335
if (ret) {
327
-
amdxdna_pm_suspend_put(hwctx->client->xdna);
328
336
dma_fence_put(job->fence);
329
337
aie2_job_put(job);
330
338
mmput(job->mm);
+10
drivers/accel/amdxdna/amdxdna_ctx.c
+10
drivers/accel/amdxdna/amdxdna_ctx.c
···
17
17
#include "amdxdna_ctx.h"
18
18
#include "amdxdna_gem.h"
19
19
#include "amdxdna_pci_drv.h"
20
+
#include "amdxdna_pm.h"
20
21
21
22
#define MAX_HWCTX_ID 255
22
23
#define MAX_ARG_COUNT 4095
···
446
445
void amdxdna_sched_job_cleanup(struct amdxdna_sched_job *job)
447
446
{
448
447
trace_amdxdna_debug_point(job->hwctx->name, job->seq, "job release");
448
+
amdxdna_pm_suspend_put(job->hwctx->client->xdna);
449
449
amdxdna_arg_bos_put(job);
450
450
amdxdna_gem_put_obj(job->cmd_bo);
451
451
dma_fence_put(job->fence);
···
482
480
if (ret) {
483
481
XDNA_ERR(xdna, "Argument BOs lookup failed, ret %d", ret);
484
482
goto cmd_put;
483
+
}
484
+
485
+
ret = amdxdna_pm_resume_get(xdna);
486
+
if (ret) {
487
+
XDNA_ERR(xdna, "Resume failed, ret %d", ret);
488
+
goto put_bos;
485
489
}
486
490
487
491
idx = srcu_read_lock(&client->hwctx_srcu);
···
530
522
dma_fence_put(job->fence);
531
523
unlock_srcu:
532
524
srcu_read_unlock(&client->hwctx_srcu, idx);
525
+
amdxdna_pm_suspend_put(xdna);
526
+
put_bos:
533
527
amdxdna_arg_bos_put(job);
534
528
cmd_put:
535
529
amdxdna_gem_put_obj(job->cmd_bo);