On Fri, 29 Sep 2006, Laurent Pinchart wrote:
> > > It
> > > seems the bug is never triggered when using Microsoft Windows, so I
> > > suspect that Linux uses a faster scheduler. Is there a way to hack the
> > > USB scheduler to make it split control transfers across USB frames ? We
> > > would like to test if the hardware bug would then be still triggered.
> >
> > Which host controller driver are you using? uhci-hcd always splits
> > control transfers like this (each transaction in a different frame) for
> > unconfigured devices. If you want to hack uhci-hcd so that it splits up
> > control transfers all the time, I can send you a patch. As far as I know,
> > it's impossible to make OHCI hardware do the same thing.
>
> I'm using an EHCI controller. Does it matter if the USB 1.1 controller is an
> OHCI or UHCI ?
Is the Logitech webcam a full-speed device or does it use high-speed? If
it is full-speed, is it behind a high-speed hub? These things will affect
which controller gets used. If it is behind a high-speed hub, does it
work better when you connect it directly to the computer?
For most purposes it doesn't matter whether a full-speed controller is
OHCI or UHCI. OHCI does more of the work in hardware, so it puts less
load on the host computer. By the same token, its scheduling is less
flexible since the host has less direct control.
> It would be nice if you could send me the patch.
See below.
> > BTW, Windows does not split up control transfers for devices after they
> > are configured. Not in general, anyway -- the Logitech device might be an
> > exception.
>
> I'm not aware of anything specific being done by Logitech in respect to the
> way Windows schedules USB control transfers.
>
> The problem seems quite weird to me. On one hand, this is clearly a hardware
> bug. When the control transfer is sent "too fast", the device misses an
> interrupt, and either times out or STALL. On the other hand, the problem
> doesn't occur when using Microsoft Windows.
Does this happen while the camera is in normal operation or only during
the initial enumeration? If it happens during operation, why do you need
to send control transfers to the camera while it is sending you data?
> Logitech is working on a firmware workaround for their next release, but
> devices already on the marker will still suffer from the bug. I was hoping to
> understand why Windows does not trigger the bug, and try to find a software
> fix for Linux.
You probably won't get too far trying to understand Windows' behavior by
asking questions on a Linux mailing list. :-)
Before anything can be fixed we need more details about the problem, like
when it occurs, what is being transferred, maybe even USB bus traces if
you can get them.
How did you determine the source of the problem? Did the engineers at
Logitech figure it out for you? If so, have they made a similar analysis
of what happens with Windows to see why it works better?
Alan Stern
Index: usb-2.6/drivers/usb/host/uhci-q.c
===================================================================
--- usb-2.6.orig/drivers/usb/host/uhci-q.c
+++ usb-2.6/drivers/usb/host/uhci-q.c
@@ -667,6 +667,7 @@ static int uhci_submit_control(struct uh
qh->dummy_td->status |= __constant_cpu_to_le32(TD_CTRL_ACTIVE);
qh->dummy_td = td;
+#if 0
/* Low-speed transfers get a different queue, and won't hog the bus.
* Also, some devices enumerate better without FSBR; the easiest way
* to do that is to put URBs on the low-speed queue while the device
@@ -678,6 +679,11 @@ static int uhci_submit_control(struct uh
qh->skel = uhci->skel_fs_control_qh;
uhci_add_fsbr(uhci, urb);
}
+#else
+ /* As a test, put all URBs on the low-speed queue so they only get
+ * one transaction per frame. */
+ qh->skel = uhci->skel_ls_control_qh;
+#endif
urb->actual_length = -8; /* Account for the SETUP packet */
return 0;
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
