Re: Callback not called on bulk out transfer with usbfs

[Austin Schuh <aschuh@xxxxxxxxxx>]

On Wed, Jun 13, 2012 at 7:20 AM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> On Tue, 12 Jun 2012, Austin Schuh wrote:
>
>> On Tue, Jun 12, 2012 at 12:40 PM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
>> > If you get rid of the line saying:
>> >
>> >                ehci->need_io_watchdog = 0;
>> >
>> > under the PCI_VENDOR_ID_INTEL case in
>> > drivers/usb/host/ehci-pci.c:ehci_pci_setup(), it may help.
>>
>> That seems to have fixed it.  Hmm...  I used to be able to run my code
>> anywhere from 3-30 times before it would hang.  It is at 470 cycles
>> right now and counting without an issue.
>>
>> Apart from building and deploying custom patched kernels everywhere
>> that I need to run my code, is there something that I can do to help
>> get this hack packaged up into something that could be submitted?  Or,
>> is there something else that I should look around for that could be
>> the root cause of the problem?
>
> My first guess would be bad hardware.  On the other hand, the fact that
> the same thing happened on two computers with different chipsets argues
> that it's not a hardware problem.
>
> A known software bug could also cause interrupts to get lost, although
> I have never observed this.  A patch (for the 3.4 kernel) to fix this
> bug is below; you could try it with ehci->need_io_watchdog set back to
> 0.

The fix below with ehci->need_io_watchdog set to 0 locks up.

> Are the two computers you tried both SMP systems?

Both systems are SMP.

desktop:
2.6.38.8 #11 SMP Tue Jun 12 18:47:20 PDT 2012 x86_64 GNU/Linux

laptop:
2.6.38-15-generic #60~lucid1-Ubuntu SMP Wed May 23 15:04:42 UTC 2012
x86_64 GNU/Linux

To add another data point, I ran the code on a machine with
USB Controller: nVidia Corporation MCP79 EHCI USB 2.0 Controller (rev b1)
and kernel
2.6.32-41-generic #89-Ubuntu SMP Fri Apr 27 22:18:56 UTC 2012 x86_64 GNU/Linux
and it worked.  I ran it for over 1000 cycles without issue.

> Alan Stern
>
>
>
> Index: v/drivers/usb/host/ehci.h
> ===================================================================
> --- v.orig/drivers/usb/host/ehci.h
> +++ v/drivers/usb/host/ehci.h
> @@ -83,7 +83,8 @@ struct ehci_hcd {                     /* one per controlle
>        struct ehci_qh          *dummy;         /* For AMD quirk use */
>        struct ehci_qh          *reclaim;
>        struct ehci_qh          *qh_scan_next;
> -       unsigned                scanning : 1;
> +       bool                    scanning:1;
> +       bool                    need_rescan:1;
>
>        /* periodic schedule support */
>  #define        DEFAULT_I_TDPS          1024            /* some HCs can do less */
> Index: v/drivers/usb/host/ehci-hcd.c
> ===================================================================
> --- v.orig/drivers/usb/host/ehci-hcd.c
> +++ v/drivers/usb/host/ehci-hcd.c
> @@ -535,13 +535,20 @@ static void ehci_work (struct ehci_hcd *
>         * it reports urb completions.  this flag guards against bogus
>         * attempts at re-entrant schedule scanning.
>         */
> -       if (ehci->scanning)
> +       if (ehci->scanning) {
> +               ehci->need_rescan = true;
>                return;
> -       ehci->scanning = 1;
> +       }
> +       ehci->scanning = true;
> +
> + rescan:
> +       ehci->need_rescan = false;
>        scan_async (ehci);
>        if (ehci->next_uframe != -1)
>                scan_periodic (ehci);
> -       ehci->scanning = 0;
> +       if (ehci->need_rescan)
> +               goto rescan;
> +       ehci->scanning = false;
>
>        /* the IO watchdog guards against hardware or driver bugs that
>         * misplace IRQs, and should let us run completely without IRQs.
>
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html