Re: [PATCH v2 1/4] usb: renesas_usbhs: fix spinlock suspected in a gadget complete function

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

On Thu, Mar 12, 2015 at 05:40:56AM +0000, yoshihiro shimoda wrote:
> Hi,
> 
> > On Thu, Mar 12, 2015 at 04:33:41AM +0000, yoshihiro shimoda wrote:
> > > Hi Geert-san again,
> > >
> > > > Hi Geert-san,
> > > >
> > > > Thank you for the reply again!
> > > >
> > > > > Hi Shimoda-san,
> > > > >
> > > > > On Mon, Feb 16, 2015 at 2:52 AM, Yoshihiro Shimoda
> > > > > <yoshihiro.shimoda.uh@xxxxxxxxxxx> wrote:
> > > > > > According to the gadget.h, a "complete" function will always be called
> > > > > > with interrupts disabled. However, sometimes usbhsg_queue_pop() function
> > > > > > is called with interrupts enabled. So, this function should calls
> > > > > > local_irq_save() before this calls the usb_gadget_giveback_request().
> > > > > > Otherwise, there is possible to cause a spinlock suspected in a gadget
> > > > > > complete function.
> > > > >
> > > > > I still feel uneasy about adding the call to local_irq_save(), as the need for
> > > > > this is usually an indicator of another locking problem.
> > > >
> > > > I also think that I would like to avoid using local_irq_save().
> > > > But, I have no idea to resolve this issue for now.
> > >
> > > After I modified usb-dmac driver to use a tasklet instead of a kthread,
> > > this issue disappeared.
> > >
> > > My understanding is the followings:
> > > - According to the backtrace below, during usbhsf_dma_complete() was running,
> > >  a softirq happened. After ncm_tx_tasklet() was called, the issue happened.
> > > http://thread.gmane.org/gmane.linux.usb.general/122023/focus=43729
> > >
> > > - This means that usbhsf_dma_complete() ran on a kthread. So, ncm driver is able
> > >   to do the ncm_tx_tasklet().
> > >
> > > - After the current usb-dmac driver, since usbhsf_dma_complete() runs on a tasklet,
> > >   ncm driver is not able to do the ncm_tx_tasklet during usbhsf_dma_complete() was running.
> > >
> > >
> > > So, I would like to recall this patch and I will resubmit this patch series as v3.
> > 
> > try something like below:
> > 
> > diff --git a/drivers/usb/renesas_usbhs/mod_gadget.c b/drivers/usb/renesas_usbhs/mod_gadget.c
> > index e0384af77e56..e9d75d85be59 100644
> > --- a/drivers/usb/renesas_usbhs/mod_gadget.c
> > +++ b/drivers/usb/renesas_usbhs/mod_gadget.c
> > @@ -119,7 +119,7 @@ struct usbhsg_recip_handle {
> >  /*
> >   *		queue push/pop
> >   */
> > -static void usbhsg_queue_pop(struct usbhsg_uep *uep,
> > +static void __usbhsg_queue_pop(struct usbhsg_uep *uep,
> >  			     struct usbhsg_request *ureq,
> >  			     int status)
> >  {
> > @@ -133,6 +133,15 @@ static void usbhsg_queue_pop(struct usbhsg_uep *uep,
> >  	usb_gadget_giveback_request(&uep->ep, &ureq->req);
> >  }
> > 
> > +static void usbhsg_queue_pop(struct usbhsg_uep *uep,
> > +			     struct usbhsg_request *ureq,
> > +			     int status)
> > +{
> > +	usbhs_lock(priv, flags);
> > +	__usbhsg_queue_pop(uep, ureq, status);
> > +	usbhs_unlock(priv, flags);
> > +}
> > +
> >  static void usbhsg_queue_done(struct usbhs_priv *priv, struct usbhs_pkt *pkt)
> >  {
> >  	struct usbhs_pipe *pipe = pkt->pipe;
> > 
> > 
> > then, for cases where lock is already held you call __usbhsg_queue_pop()
> > and for all other cases, call usbhsg_queue_pop().
> 
> Thank you for the suggestion. However, we cannot use this
> usbhsg_queue_pop() because a gadget driver might call usb_ep_queue()
> in the "complete" function and this driver calls usbhs_lock() in the
> usb_ep_queue().

right, in that case just call __usbhs_queue_pop() directly.

> Perhaps my explanation was bad, but this issue was caused by the
> following conditions:
>  - I use the renesas_usbhs driver as udc.
>  - I use an old usb-dmac driver that the callback runs on a kthread.
>  - I use the ncm driver. In this environment, the ncm driver might
>	cause a spinlock suspected.
> 
> As an old solution, I fixed the renesas_usbhs driver by this patch.
> However, if I use a new usb-dmac driver that the callback runs on a
> tasklet, I don't need this patch. (This is a new solution.)

then differentiate based on some revision register or something like
that ?

-- 
balbi

Attachment: signature.asc
Description: Digital signature


[Index of Archives]     [Linux Media]     [Linux Input]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Old Linux USB Devel Archive]

  Powered by Linux