[PATCH V3] accel/amdxdna: Prevent PM resume deadlock in hwctx_sync_debug_bo()
From: Lizhi Hou <lizhi.hou@amd.com>
Date: 2026-06-16 21:24:38
Also in:
dri-devel
Subsystem:
amd xdna driver, drm compute accelerators drivers and framework, the rest · Maintainers:
Min Ma, Lizhi Hou, Oded Gabbay, Linus Torvalds
amdxdna_hwctx_sync_debug_bo() invokes the hardware hwctx_sync_debug_bo()
callback while holding xdna->dev_lock.
The callback may call amdxdna_cmd_submit(), which in turn calls
amdxdna_pm_resume_get(). If the device is suspended,
amdxdna_pm_resume_get() may synchronously execute amdxdna_pm_resume(),
which also acquires xdna->dev_lock, resulting in a deadlock.
Avoid the deadlock by calling amdxdna_pm_resume_get() before holding
xdna->dev_lock in both amdxdna_hwctx_sync_debug_bo() and
amdxdna_drm_config_hwctx_ioctl()
Fixes: 7ea046838021 ("accel/amdxdna: Support firmware debug buffer")
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
---
V2:
Fix jumping forward over the guard(mutex) declaration.
V3:
Fix the same in amdxdna_drm_config_hwctx_ioctl().
drivers/accel/amdxdna/amdxdna_ctx.c | 26 ++++++++++++++++++++++----
1 file changed, 22 insertions(+), 4 deletions(-)
diff --git a/drivers/accel/amdxdna/amdxdna_ctx.c b/drivers/accel/amdxdna/amdxdna_ctx.c
index ff6c3e8e5a15..2ab4443e7094 100644
--- a/drivers/accel/amdxdna/amdxdna_ctx.c
+++ b/drivers/accel/amdxdna/amdxdna_ctx.c@@ -356,16 +356,25 @@ int amdxdna_drm_config_hwctx_ioctl(struct drm_device *dev, void *data, struct dr return -EINVAL; } - guard(mutex)(&xdna->dev_lock); + ret = amdxdna_pm_resume_get(xdna); + if (ret) { + XDNA_ERR(xdna, "Resume failed, ret %d", ret); + goto free_buf; + } + + mutex_lock(&xdna->dev_lock); hwctx = xa_load(&client->hwctx_xa, args->handle); if (!hwctx) { XDNA_DBG(xdna, "PID %d failed to get hwctx %d", client->pid, args->handle); ret = -EINVAL; - goto free_buf; + goto unlock; } ret = xdna->dev_info->ops->hwctx_config(hwctx, args->param_type, val, buf, buf_size); +unlock: + mutex_unlock(&xdna->dev_lock); + amdxdna_pm_suspend_put(xdna); free_buf: kfree(buf); return ret;
@@ -386,16 +395,25 @@ int amdxdna_hwctx_sync_debug_bo(struct amdxdna_client *client, u32 debug_bo_hdl) if (!gobj) return -EINVAL; + ret = amdxdna_pm_resume_get(xdna); + if (ret) { + XDNA_ERR(xdna, "Resume failed, ret %d", ret); + goto put_obj; + } + abo = to_xdna_obj(gobj); - guard(mutex)(&xdna->dev_lock); + mutex_lock(&xdna->dev_lock); hwctx = xa_load(&client->hwctx_xa, abo->assigned_hwctx); if (!hwctx) { ret = -EINVAL; - goto put_obj; + goto unlock; } ret = xdna->dev_info->ops->hwctx_sync_debug_bo(hwctx, debug_bo_hdl); +unlock: + mutex_unlock(&xdna->dev_lock); + amdxdna_pm_suspend_put(xdna); put_obj: drm_gem_object_put(gobj); return ret;
--
2.34.1