On Fri, May 17, 2024 at 11:16:31PM +0200, Andrey Konovalov wrote:
> > Hello.
> >
> > syzbot is still reporting warning in kcov_remote_start() from
> > __usb_hcd_giveback_urb() path. I guess that commit 8fea0c8fda30
> > ("usb: core: hcd: Convert from tasklet to BH workqueue") broke
> > the in_serving_softirq() workaround explained in commit e89eed02a5f1
> > ("kcov, usb: hide in_serving_softirq checks in __usb_hcd_giveback_urb").
> >
> > How can we fix this workaround?
> So the problem here is that __usb_hcd_giveback_urb gets called from an
> interrupt that comes during a __usb_hcd_giveback_urb call from
> usb_giveback_urb_bh. And KCOV annotations don't expect that
> __usb_hcd_giveback_urb can be interrupted.
>
> We had a similar issue with USB/IP before, and AFAIU the resolution
> was that it's invalid to call usb_giveback_urb with interrupts enabled
> [1]. But I'm not sure if the same applies to usb_giveback_urb_bh.
>
> Alan, is this a bug in usb_giveback_urb_bh? Perhaps we need to disable
> interrupts before calling __usb_hcd_giveback_urb?
Well, the code which used to run in a tasklet now runs in a work queue,
so it's not surprising that some things can go wrong.
Perhaps the simplest approach is to disable interrupts in
__usb_hcd_giveback_urb() around the kcov_* calls. And change the
preceding comment, of course.
Alan Stern
> Thank you!
>
> [1] https://lore.kernel.org/linux-usb/20201006012333.GA399825@xxxxxxxxxxxxxxxxxxx/
