Currently there's a KCOV remote coverage collection section in
__usb_hcd_giveback_urb(). Initially that section was added based on the
assumption that usb_hcd_giveback_urb() can only be called in interrupt
context as indicated by a comment before it. This is what happens when
syzkaller is fuzzing the USB stack via the dummy_hcd driver.
As it turns out, it's actually valid to call usb_hcd_giveback_urb() in task
context, provided that the caller turned off the interrupts; USB/IP does
exactly that. This can lead to a nested KCOV remote coverage collection
sections both trying to collect coverage in task context. This isn't
supported by KCOV, and leads to a WARNING.
Change __usb_hcd_giveback_urb() to only call kcov_remote_*() callbacks
when it's being executed in a softirq. As the result, the coverage from
USB/IP related usb_hcd_giveback_urb() calls won't be collected, but the
WARNING is fixed.
A potential future improvement would be to support nested remote coverage
collection sections, but this patch doesn't address that.
Signed-off-by: Andrey Konovalov <[email protected]>
Acked-by: Marco Elver <[email protected]>
---
Changes v3->v4:
- Don't make any kcov changes, do a softirq context check in usb code
instead.
---
drivers/usb/core/hcd.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/drivers/usb/core/hcd.c b/drivers/usb/core/hcd.c
index a33b849e8beb..2f6a39e09dc6 100644
--- a/drivers/usb/core/hcd.c
+++ b/drivers/usb/core/hcd.c
@@ -1646,9 +1646,16 @@ static void __usb_hcd_giveback_urb(struct urb *urb)
/* pass ownership to the completion handler */
urb->status = status;
- kcov_remote_start_usb((u64)urb->dev->bus->busnum);
+ /*
+ * This function can be called in task context inside another remote
+ * coverage collection section, but KCOV doesn't support that kind of
+ * recursion yet. Only collect coverage in softirq context for now.
+ */
+ if (in_serving_softirq())
+ kcov_remote_start_usb((u64)urb->dev->bus->busnum);
urb->complete(urb);
- kcov_remote_stop();
+ if (in_serving_softirq())
+ kcov_remote_stop();
usb_anchor_resume_wakeups(anchor);
atomic_dec(&urb->use_count);
--
2.29.0.rc1.297.gfa9743e501-goog
On Fri, Oct 16, 2020 at 3:57 PM Andrey Konovalov <[email protected]> wrote:
>
> Currently there's a KCOV remote coverage collection section in
> __usb_hcd_giveback_urb(). Initially that section was added based on the
> assumption that usb_hcd_giveback_urb() can only be called in interrupt
> context as indicated by a comment before it. This is what happens when
> syzkaller is fuzzing the USB stack via the dummy_hcd driver.
>
> As it turns out, it's actually valid to call usb_hcd_giveback_urb() in task
> context, provided that the caller turned off the interrupts; USB/IP does
> exactly that. This can lead to a nested KCOV remote coverage collection
> sections both trying to collect coverage in task context. This isn't
> supported by KCOV, and leads to a WARNING.
>
> Change __usb_hcd_giveback_urb() to only call kcov_remote_*() callbacks
> when it's being executed in a softirq. As the result, the coverage from
> USB/IP related usb_hcd_giveback_urb() calls won't be collected, but the
> WARNING is fixed.
>
> A potential future improvement would be to support nested remote coverage
> collection sections, but this patch doesn't address that.
>
> Signed-off-by: Andrey Konovalov <[email protected]>
> Acked-by: Marco Elver <[email protected]>
> ---
>
> Changes v3->v4:
> - Don't make any kcov changes, do a softirq context check in usb code
> instead.
>
> ---
> drivers/usb/core/hcd.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/usb/core/hcd.c b/drivers/usb/core/hcd.c
> index a33b849e8beb..2f6a39e09dc6 100644
> --- a/drivers/usb/core/hcd.c
> +++ b/drivers/usb/core/hcd.c
> @@ -1646,9 +1646,16 @@ static void __usb_hcd_giveback_urb(struct urb *urb)
>
> /* pass ownership to the completion handler */
> urb->status = status;
> - kcov_remote_start_usb((u64)urb->dev->bus->busnum);
> + /*
> + * This function can be called in task context inside another remote
> + * coverage collection section, but KCOV doesn't support that kind of
> + * recursion yet. Only collect coverage in softirq context for now.
> + */
> + if (in_serving_softirq())
> + kcov_remote_start_usb((u64)urb->dev->bus->busnum);
> urb->complete(urb);
> - kcov_remote_stop();
> + if (in_serving_softirq())
> + kcov_remote_stop();
Reviewed-by: Dmitry Vyukov <[email protected]>
Looks simpler :)
> usb_anchor_resume_wakeups(anchor);
> atomic_dec(&urb->use_count);
> --
> 2.29.0.rc1.297.gfa9743e501-goog
>
On 2020-10-16 15:57:45 [+0200], Andrey Konovalov wrote:
> --- a/drivers/usb/core/hcd.c
> +++ b/drivers/usb/core/hcd.c
> @@ -1646,9 +1646,16 @@ static void __usb_hcd_giveback_urb(struct urb *urb)
>
> /* pass ownership to the completion handler */
> urb->status = status;
> - kcov_remote_start_usb((u64)urb->dev->bus->busnum);
> + /*
> + * This function can be called in task context inside another remote
> + * coverage collection section, but KCOV doesn't support that kind of
> + * recursion yet. Only collect coverage in softirq context for now.
> + */
> + if (in_serving_softirq())
Could this in_serving_softirq() usage be replaced, please?
> + kcov_remote_start_usb((u64)urb->dev->bus->busnum);
> urb->complete(urb);
> - kcov_remote_stop();
> + if (in_serving_softirq())
> + kcov_remote_stop();
>
> usb_anchor_resume_wakeups(anchor);
> atomic_dec(&urb->use_count);
Sebastian
On Fri, Nov 13, 2020 at 1:30 PM Sebastian Andrzej Siewior
<[email protected]> wrote:
>
> On 2020-10-16 15:57:45 [+0200], Andrey Konovalov wrote:
> > --- a/drivers/usb/core/hcd.c
> > +++ b/drivers/usb/core/hcd.c
> > @@ -1646,9 +1646,16 @@ static void __usb_hcd_giveback_urb(struct urb *urb)
> >
> > /* pass ownership to the completion handler */
> > urb->status = status;
> > - kcov_remote_start_usb((u64)urb->dev->bus->busnum);
> > + /*
> > + * This function can be called in task context inside another remote
> > + * coverage collection section, but KCOV doesn't support that kind of
> > + * recursion yet. Only collect coverage in softirq context for now.
> > + */
> > + if (in_serving_softirq())
>
> Could this in_serving_softirq() usage be replaced, please?
Hi Sebastian,
Replaced with what and why?
Thanks!
On 2020-11-13 13:51:19 [+0100], Andrey Konovalov wrote:
> Hi Sebastian,
Hi Andrey,
> Replaced with what and why?
Linus requested in
https://lkml.kernel.org/r/CAHk-=wht7kAeyR5xEW2ORj7m0hibVxZ3t+2ie8vNHLQfdbN2_g@mail.gmail.com/
that drivers should not change their behaviour on context magic like
in_atomic(), in_interrupt() and so on.
The USB bits were posted in
https://lkml.kernel.org/r/[email protected]
and merged (which is probably the same time as this patch).
I haven't look what this code should do or does but there are HCDs for
which this is never true like the UHCI/OHCI controller for instance.
> Thanks!
Sebastian
On Fri, Nov 13, 2020 at 2:28 PM Sebastian Andrzej Siewior
<[email protected]> wrote:
>
> On 2020-11-13 13:51:19 [+0100], Andrey Konovalov wrote:
> > Hi Sebastian,
>
> Hi Andrey,
>
> > Replaced with what and why?
>
> Linus requested in
> https://lkml.kernel.org/r/CAHk-=wht7kAeyR5xEW2ORj7m0hibVxZ3t+2ie8vNHLQfdbN2_g@mail.gmail.com/
>
> that drivers should not change their behaviour on context magic like
> in_atomic(), in_interrupt() and so on.
> The USB bits were posted in
> https://lkml.kernel.org/r/[email protected]
>
> and merged (which is probably the same time as this patch).
>
> I haven't look what this code should do or does but there are HCDs for
> which this is never true like the UHCI/OHCI controller for instance.
We could go back to adding softirq-specific kcov callbacks. Perhaps
with a simpler implementation than what we had before to only cover
this case. Something like kcov_remote_start_usb_softirq() and
kcov_remote_stop_softirq() that do the softirq check internally.
Greg, what would you prefer?
On Fri, Nov 13, 2020 at 02:42:44PM +0100, Andrey Konovalov wrote:
> On Fri, Nov 13, 2020 at 2:28 PM Sebastian Andrzej Siewior
> <[email protected]> wrote:
> >
> > On 2020-11-13 13:51:19 [+0100], Andrey Konovalov wrote:
> > > Hi Sebastian,
> >
> > Hi Andrey,
> >
> > > Replaced with what and why?
> >
> > Linus requested in
> > https://lkml.kernel.org/r/CAHk-=wht7kAeyR5xEW2ORj7m0hibVxZ3t+2ie8vNHLQfdbN2_g@mail.gmail.com/
> >
> > that drivers should not change their behaviour on context magic like
> > in_atomic(), in_interrupt() and so on.
> > The USB bits were posted in
> > https://lkml.kernel.org/r/[email protected]
> >
> > and merged (which is probably the same time as this patch).
> >
> > I haven't look what this code should do or does but there are HCDs for
> > which this is never true like the UHCI/OHCI controller for instance.
>
> We could go back to adding softirq-specific kcov callbacks. Perhaps
> with a simpler implementation than what we had before to only cover
> this case. Something like kcov_remote_start_usb_softirq() and
> kcov_remote_stop_softirq() that do the softirq check internally.
>
> Greg, what would you prefer?
I really have no idea, sorry.
On Fri, 13 Nov 2020 at 14:42, Andrey Konovalov <[email protected]> wrote:
> On Fri, Nov 13, 2020 at 2:28 PM Sebastian Andrzej Siewior
> <[email protected]> wrote:
> >
> > On 2020-11-13 13:51:19 [+0100], Andrey Konovalov wrote:
> > > Hi Sebastian,
> >
> > Hi Andrey,
> >
> > > Replaced with what and why?
> >
> > Linus requested in
> > https://lkml.kernel.org/r/CAHk-=wht7kAeyR5xEW2ORj7m0hibVxZ3t+2ie8vNHLQfdbN2_g@mail.gmail.com/
> >
> > that drivers should not change their behaviour on context magic like
> > in_atomic(), in_interrupt() and so on.
> > The USB bits were posted in
> > https://lkml.kernel.org/r/[email protected]
Arguably this patch is *not* changing "driver behaviour", it's only
changing how and when KCOV collects coverage, which is not related to
how the driver behaves.
> > and merged (which is probably the same time as this patch).
> >
> > I haven't look what this code should do or does but there are HCDs for
> > which this is never true like the UHCI/OHCI controller for instance.
>
> We could go back to adding softirq-specific kcov callbacks. Perhaps
> with a simpler implementation than what we had before to only cover
> this case. Something like kcov_remote_start_usb_softirq() and
> kcov_remote_stop_softirq() that do the softirq check internally.
Is this a matter of simply banning such functions entirely without
understanding their use? Because that sounds wrong. But if it is, we
probably have to just add some static inline functions in
include/linux/kcov.h that simply does the check.
Thanks,
-- Marco
On Fri, Nov 13, 2020 at 4:47 PM Marco Elver <[email protected]> wrote:
>
> On Fri, 13 Nov 2020 at 14:42, Andrey Konovalov <[email protected]> wrote:
> > On Fri, Nov 13, 2020 at 2:28 PM Sebastian Andrzej Siewior
> > <[email protected]> wrote:
> > >
> > > On 2020-11-13 13:51:19 [+0100], Andrey Konovalov wrote:
> > > > Hi Sebastian,
> > >
> > > Hi Andrey,
> > >
> > > > Replaced with what and why?
> > >
> > > Linus requested in
> > > https://lkml.kernel.org/r/CAHk-=wht7kAeyR5xEW2ORj7m0hibVxZ3t+2ie8vNHLQfdbN2_g@mail.gmail.com/
> > >
> > > that drivers should not change their behaviour on context magic like
> > > in_atomic(), in_interrupt() and so on.
> > > The USB bits were posted in
> > > https://lkml.kernel.org/r/[email protected]
>
> Arguably this patch is *not* changing "driver behaviour", it's only
> changing how and when KCOV collects coverage, which is not related to
> how the driver behaves.
>
> > > and merged (which is probably the same time as this patch).
> > >
> > > I haven't look what this code should do or does but there are HCDs for
> > > which this is never true like the UHCI/OHCI controller for instance.
> >
> > We could go back to adding softirq-specific kcov callbacks. Perhaps
> > with a simpler implementation than what we had before to only cover
> > this case. Something like kcov_remote_start_usb_softirq() and
> > kcov_remote_stop_softirq() that do the softirq check internally.
>
> Is this a matter of simply banning such functions entirely without
> understanding their use? Because that sounds wrong. But if it is, we
> probably have to just add some static inline functions in
> include/linux/kcov.h that simply does the check.
Yeah, this seems like a solution that will satisfy everyone. Will mail
a new version shortly.