Re: [PATCH 1/2] accel/ivpu: Fix deadlock in ivpu_ms_cleanup()

[Lizhi Hou]

On 3/25/25 04:43, Maciej Falkowski wrote:
From: Jacek Lawrynowicz <jacek.lawrynow...@linux.intel.com>

Fix deadlock in ivpu_ms_cleanup() by preventing runtime resume after
file_priv->ms_lock is acquired.

During a failure in runtime resume, a cold boot is executed, which
calls ivpu_ms_cleanup_all(). This function calls ivpu_ms_cleanup()
that acquires file_priv->ms_lock and causes the deadlock.

Fixes: cdfad4db7756 ("accel/ivpu: Add NPU profiling support")
Cc: <sta...@vger.kernel.org> # v6.11+
Signed-off-by: Jacek Lawrynowicz <jacek.lawrynow...@linux.intel.com>
Signed-off-by: Maciej Falkowski <maciej.falkow...@linux.intel.com>
---
  drivers/accel/ivpu/ivpu_ms.c | 6 ++++++
  1 file changed, 6 insertions(+)

diff --git a/drivers/accel/ivpu/ivpu_ms.c b/drivers/accel/ivpu/ivpu_ms.c
index ffe7b10f8a76..eb485cf15ad6 100644
--- a/drivers/accel/ivpu/ivpu_ms.c
+++ b/drivers/accel/ivpu/ivpu_ms.c
@@ -4,6 +4,7 @@
   */
  
  #include <drm/drm_file.h>
+#include <linux/pm_runtime.h>
  
  #include "ivpu_drv.h"
  #include "ivpu_gem.h"
@@ -281,6 +282,9 @@ int ivpu_ms_get_info_ioctl(struct drm_device *dev, void 
*data, struct drm_file *
  void ivpu_ms_cleanup(struct ivpu_file_priv *file_priv)
  {
        struct ivpu_ms_instance *ms, *tmp;
+       struct ivpu_device *vdev = file_priv->vdev;
+
+       pm_runtime_get_sync(vdev->drm.dev);

Could get_sync() be failed here? Maybe it is better to add warning for 
failure?


Lizhi

  
  	mutex_lock(&file_priv->ms_lock);
  
@@ -293,6 +297,8 @@ void ivpu_ms_cleanup(struct ivpu_file_priv *file_priv)
                free_instance(file_priv, ms);
  
  	mutex_unlock(&file_priv->ms_lock);
+
+       pm_runtime_put_autosuspend(vdev->drm.dev);
  }
  
  void ivpu_ms_cleanup_all(struct ivpu_device *vdev)