Le 21/09/2023 à 12:24, Hans Verkuil a écrit :

On 21/09/2023 11:28, Benjamin Gaignard wrote:

quoted

Le 20/09/2023 à 16:56, Hans Verkuil a écrit :

quoted

On 20/09/2023 16:30, Benjamin Gaignard wrote:
<snip>

quoted

quoted

quoted

        num_buffers = min_t(unsigned int, num_buffers,
                    q->max_allowed_buffers - vb2_get_num_buffers(q));
    -    first_index = vb2_get_num_buffers(q);
+    first_index = bitmap_find_next_zero_area(q->bufs_map, q->max_allowed_buffers,
+                         0, num_buffers, 0);
          if (first_index >= q->max_allowed_buffers)
            return 0;

@@ -675,7 +678,13 @@ static void __vb2_queue_free(struct vb2_queue *q, unsigned int buffers)
      struct vb2_buffer *vb2_get_buffer(struct vb2_queue *q, unsigned int index)
    {
-    if (index < q->num_buffers)
+    if (!q->bufs_map || !q->bufs)
+        return NULL;

I don't think this can ever happen.

I got kernel crash without them.
I will keep them.

What is the backtrace? How can this happen? It feels wrong that this can be
called with a vb2_queue that apparently is not properly initialized.

I have this log when adding dump_stack() in vb2_get_buffer() if !q->bufs_bitmap:

[   18.924627] Call trace:
[   18.927090]  dump_backtrace+0x94/0xec
[   18.930787]  show_stack+0x18/0x24
[   18.934137]  dump_stack_lvl+0x48/0x60
[   18.937833]  dump_stack+0x18/0x24
[   18.941166]  __vb2_queue_cancel+0x23c/0x2f0
[   18.945365]  vb2_core_queue_release+0x24/0x6c
[   18.949740]  vb2_queue_release+0x10/0x1c
[   18.953677]  v4l2_m2m_ctx_release+0x20/0x40
[   18.957892]  hantro_release+0x20/0x54
[   18.961584]  v4l2_release+0x74/0xec
[   18.965110]  __fput+0xb4/0x274
[   18.968205]  __fput_sync+0x50/0x5c
[   18.971626]  __arm64_sys_close+0x38/0x7c
[   18.975562]  invoke_syscall+0x48/0x114
[   18.979329]  el0_svc_common.constprop.0+0xc0/0xe0
[   18.984068]  do_el0_svc+0x1c/0x28
[   18.987402]  el0_svc+0x40/0xe8
[   18.990470]  el0t_64_sync_handler+0x100/0x12c
[   18.994842]  el0t_64_sync+0x190/0x194

This happen at boot time when hantro driver is open and close without other actions.

Ah, now I see the problem. q->bufs and q->bufs_map are allocated in
vb2_core_create_bufs and vb2_core_reqbufs, but they should be allocated
in vb2_queue_init: that's the counterpart of vb2_core_queue_release.

With that change you shouldn't have to check for q->bufs/bufs_map anymore.

It is a better solution but even like this vb2_core_queue_release() is called
at least 2 times on the same vivid queue and without testing q->bufs_bitmap
makes kernel crash.

Regards,

	Hans

quoted

quoted

quoted

quoted

quoted

+
+    return (bitmap_weight(q->bufs_map, q->max_allowed_buffers) > 0);

How about:

      return vb2_get_num_buffers(q) > 0;

vb2_get_num_buffers is defined in videobuf2-core.c, I'm not sure that
an inline function could depend of a module function.

Not a problem. E.g. v4l2-ctrls.h is full of such static inlines.

Regards,

     Hans

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel