Re: Allocating buffers for USB transfers (again)

[Sarah Sharp <sarah.a.sharp@linux.intel.com>]

On Thu, Aug 11, 2011 at 07:27:15PM +0200, Daniel Mack wrote:
> On Thu, Aug 11, 2011 at 6:45 PM, Sarah Sharp
> <sarah.a.sharp@linux.intel.com> wrote:
> > On Thu, Aug 11, 2011 at 02:57:41AM +0200, Daniel Mack wrote:
> >> On Thu, Aug 11, 2011 at 1:15 AM, Sarah Sharp
> >> <sarah.a.sharp@linux.intel.com> wrote:
> >> > On Wed, Aug 10, 2011 at 05:33:02PM +0200, Daniel Mack wrote:
> >> >> On 08/10/2011 04:32 PM, Alan Stern wrote:
> >> >> >Looking at the driver's current code, it appears that your patch
> >> >> >does not fix the bug properly.  Using discontiguous regions in the
> >> >> >transfer buffer is perfectly okay.  The real problem is later on,
> >> >> >where you do:
> >> >> >
> >> >> >if (send_it) { out->number_of_packets = FRAMES_PER_URB;
> >> >> >
> >> >> >This should be
> >> >> >
> >> >> >out->number_of_packets = outframe;
> >> >> >
> >> >> >The way it is now, the USB stack will try to use data from all the
> >> >> >frame descriptors, and the last few will be stale because the loop
> >> >> >doesn't set them.
> >> >>
> >> >> That's actually true, even though it doesn't seem to cause any trouble.
> >> >> I tested everything here of course, and the output URBs return back from
> >> >> the USB stack with their length fields zeroed out, which then
> >> >> causes the stack to send packets with zero-length fields at the end.
> >> >
> >> > Actually, it causes system hangs when the driver is loaded on a device
> >> > attached to a USB 3.0 port, as Alan Stern pointed out:
> >> >
> >> > https://bugzilla.kernel.org/show_bug.cgi?id=40702
> >>
> >> Yes, I've noticed this.
> >>
> >> > Please don't submit zero-length transfers.  The xHCI driver just isn't
> >> > able to handle it.  Arguably, it probably should have just rejected your
> >> > URB when it found a zero length buffer, so I'll probably be submitting a
> >> > patch to fix that.
> >>
> >> According to the spec, sending zero-length frames should be fine, no?
> >> Is there any particular reason why XCHI can't handle this while EHCI
> >> can? And does my patch fix the driver for XHCI?
> >
> > Ok, yes, you're correct that the xHCI spec allows the transfer length to
> > be set to zero.  In the case where the frame buffer is zero-length, is
> > the buffer pointer still valid?  It's not clear from the spec whether it
> > needs to be.
> 
> Well, the buffer pointer is set in the URB, not in its individual iso
> subframes which just denotes them via the offset field. So yes, it is
> valid in my case. But it doesn't matter anymore, as the code which
> does that is now gone :)

Do you mean it was removed with this patch:

http://git.kernel.org/?p=linux/kernel/git/tiwai/sound-2.6.git;a=commitdiff;h=15439b

Because according to Matej, he applied that patch, plus my patch to
reject zero-length buffers[1], and he saw debugging that indicated he
*did* see zero-length buffers.  Is there any chance your driver might
submit a zero-length buffer in the middle of the isochronous URB
transfer array?

Sarah Sharp

[1]
http://git.kernel.org/?p=linux/kernel/git/sarah/xhci.git;a=commitdiff;h=e70d79bda63050729fcef654167d22eb59380aa6