Re: strange call stack when running uvc-gadget application

[Peter Chen <peter.chen@nxp.com>]

On 20-11-03 11:50:17, Sriharsha Allenki wrote:
> Hi Peter,
> 
> On 11/2/2020 3:20 PM, Peter Chen wrote:
> > Hi all,
> >
> > When running run uvc-gadget application at HS using dwc3 gadget at Linux
> > v5.10-rc1, the video stream will be stopped after 1-2 minutes running. The
> > trace log like below, I wonder how _raw_spin_lock_irqsave calls __switch_to?
> > Any hints? Thanks.
> >
> > usb_test# [ 4757.322728] configfs-gadget gadget: uvc: VS request completed with status -18.
> > [ 4757.329971] configfs-gadget gadget: uvc: VS request completed with status -18.
> > UVC: Possible USB shutdown requested from Host, seen during VIDIOC_DQBUF
> >
> > usb_test# [ 4812.376465] check_interval: 37 callbacks suppressed
> > [ 4825.307665] configfs-gadget gadget: uvc: VS request completed with status -18.
> > [ 4825.314912] configfs-gadget gadget: uvc: VS request completed with status -18.
> > UVC: Possible USB shutdown requested from Host, seen during VIDIOC_DQBUF
> > [ 4826.231392] check_interval: 3 callbacks suppressed
> > select timeout
> > [ 4827.336088] dwc3 38100000.dwc3: request 0000000080ebefd3 was not queued to ep2in
> > [ 4827.343547] dwc3 38100000.dwc3: request 00000000b578605c was not queued to ep2in
> > [ 4827.350989] dwc3 38100000.dwc3: request 00000000c6d191cd was not queued to ep2in
> > [ 4827.358422] dwc3 38100000.dwc3: request 0000000085205409 was not queued to ep2in
> > UVC: Stopping video stream.
> >
> > [ 4848.381718] rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
> > [ 4848.387824] rcu:     3-...0: (1 GPs behind) idle=162/1/0x4000000000000000 softirq=10819/10819 fqs=2356
> > [ 4848.396956]  (detected by 2, t=5252 jiffies, g=20129, q=3770)
> > [ 4848.396959] Task dump for CPU 3:
> > [ 4848.405925] task:uvc-gadget_wlhe state:R  running task     stack:    0 pid:  674 ppid:   636 flags:0x00000202
> > [ 4848.415842] Call trace:
> > [ 4848.418294]  __switch_to+0xc0/0x170
> > [ 4848.421785]  _raw_spin_lock_irqsave+0x84/0xb0
> > [ 4848.426143]  composite_disconnect+0x28/0x78
> > [ 4848.430327]  configfs_composite_disconnect+0x68/0x70
> > [ 4848.435290]  usb_gadget_disconnect+0x10c/0x128
> > [ 4848.439733]  usb_gadget_deactivate+0xd4/0x108
> > [ 4848.444089]  usb_function_deactivate+0x6c/0x80
> > [ 4848.448534]  uvc_function_disconnect+0x20/0x58
> > [ 4848.452976]  uvc_v4l2_release+0x30/0x88
> > [ 4848.456812]  v4l2_release+0xbc/0xf0
> > [ 4848.460301]  __fput+0x7c/0x230
> > [ 4848.463353]  ____fput+0x14/0x20
> > [ 4848.466495]  task_work_run+0x88/0x140
> > [ 4848.470157]  do_notify_resume+0x240/0x6f0
> > [ 4848.474166]  work_pending+0x8/0x200
> The reason for this seems to be that the usb_gadget_deactivate is being called with
> spinlock held from the usb_function_deactivate and the same lock is being used
> in the composite_disconnect (&cdev->lock).
> 
> This should be able to resolve it.
> 
> diff --git a/drivers/usb/gadget/composite.c b/drivers/usb/gadget/composite.c
> index 05b176c82cc5..5fced737e4ef 100644
> --- a/drivers/usb/gadget/composite.c
> +++ b/drivers/usb/gadget/composite.c
> @@ -392,8 +392,11 @@ int usb_function_deactivate(struct usb_function *function)
> 
>         spin_lock_irqsave(&cdev->lock, flags);
> 
> -       if (cdev->deactivations == 0)
> +       if (cdev->deactivations == 0) {
> +               spin_unlock_irqrestore(&cdev->lock, flags);
>                 status = usb_gadget_deactivate(cdev->gadget);
> +               spin_lock_irqsave(&cdev->lock, flags);
> +       }
>         if (status == 0)
>                 cdev->deactivations++;
> 
> @@ -424,8 +427,11 @@ int usb_function_activate(struct usb_function *function)
>                 status = -EINVAL;
>         else {
>                 cdev->deactivations--;
> -               if (cdev->deactivations == 0)
> +               if (cdev->deactivations == 0) {
> +                       spin_unlock_irqrestore(&cdev->lock, flags);
>                         status = usb_gadget_activate(cdev->gadget);
> +                       spin_lock_irqsave(&cdev->lock, flags);
> +               }
>         }
> 
>         spin_unlock_irqrestore(&cdev->lock, flags);
> 

Thanks, Sriharsha. It fixed the kernel dump after video stream has
stopped, I did not check the whole trace carefully, and not found this
spin recursion issue. You could add my Tested-by for it. Meanwhile,
this issue was reported before, and unlock at usb_function_activate
could also fix the possible sleep at atomic context issue for dwc3.

https://lore.kernel.org/linux-usb/20191115070122.GF30608@b29397-desktop/T/

-- 

Thanks,
Peter Chen