[Qemu-devel] USB-OHCI / UHCI: Isochronous transfer not working on Linux host

[Qemu-devel] USB-OHCI / UHCI: Isochronous transfer not working on Linux host

[Thomas Bandelier]

[-- Attachment #1: Type: text/plain, Size: 1336 bytes --]

Hi all,

Following several tests we did with qemu trunk, Isochronous transfer seems
to be broken with the head of qemu.

We tried following Os (targets):
-WinXP (i386)
-Win2k (i386)
-Several linux kernel ( 2.6.26-1 (i386), 2.6.27 (ARM) )

On our host we are using Gentoo with 2.6.23 and 2.6.27 kernel.

Following USB devices were tested:
-USB 2.0 logitech webcams
-Terratec USB2.0 Cynergy hybrid XS FM tuner.

Both OHCI and UHCI configurations were tested in i386 target.
Only OHCI was tested in ARM target.

In all these configurations, it was *never* possible to initiate isochronous
USB transfer. It was possible to enumerate the devices but as soon as
isochronous packets were used, the procedure failed. It means for example
that no picture was retrieved from the webcams, and no audio data was
retrieved from FM tuner.

It is important to mention that all theses devices perfectly work on our
host.

After some search in this maling list archives, it seems that no work has
been made on these aspects since Arnon Gilboa sent some patches more than
one year ago. Am I wrong?

So It would be very useful for us to have a feedback from the community on
this point.
Has someone met the same problems? Has someone investigated this aspect?
Is someone regularly using this feature?

How can we investigate this problem?

Thanks

Thomas

[-- Attachment #2: Type: text/html, Size: 1498 bytes --]

[Qemu-devel] RE: USB-OHCI / UHCI: Isochronous transfer not working on Linux host

[Arnon Gilboa]

[-- Attachment #1: Type: text/plain, Size: 2224 bytes --]

Hello Thomas,
 
Since the last time I sent a patch, usb-linux.c & usb-uhci.c were
completly rewritten by 
Max Krasnyansky and merged by aliguori.
 
However, looking at my post below from 24 Sep 2007, you can see that
usb-uhci, 
which is USB 1.1 host controller emulation, was mostly tested on USB 1.1
devices. 
 
I am sure that many USB 2.0 and even several 1.1 devices (especially
isochronous) 
will NOT work with the emulation.
 
http://lists.gnu.org/archive/html/qemu-devel/2007-09/msg00409.html
 
Regards,
Arnon


________________________________

From: Thomas Bandelier [mailto:thomas.bandelier@gmail.com] 
Sent: Thursday, November 27, 2008 4:06 PM
To: qemu-devel@nongnu.org
Cc: lemagoup@gmail.com; Arnon Gilboa
Subject: USB-OHCI / UHCI: Isochronous transfer not working on Linux host


Hi all,

Following several tests we did with qemu trunk, Isochronous transfer
seems to be broken with the head of qemu.

We tried following Os (targets):
-WinXP (i386)
-Win2k (i386)
-Several linux kernel ( 2.6.26-1 (i386), 2.6.27 (ARM) )

On our host we are using Gentoo with 2.6.23 and 2.6.27 kernel.

Following USB devices were tested:
-USB 2.0 logitech webcams
-Terratec USB2.0 Cynergy hybrid XS FM tuner.

Both OHCI and UHCI configurations were tested in i386 target.
Only OHCI was tested in ARM target.

In all these configurations, it was never possible to initiate
isochronous USB transfer. It was possible to enumerate the devices but
as soon as isochronous packets were used, the procedure failed. It means
for example that no picture was retrieved from the webcams, and no audio
data was retrieved from FM tuner.

It is important to mention that all theses devices perfectly work on our
host.

After some search in this maling list archives, it seems that no work
has been made on these aspects since Arnon Gilboa sent some patches more
than one year ago. Am I wrong?

So It would be very useful for us to have a feedback from the community
on this point.
Has someone met the same problems? Has someone investigated this aspect?
Is someone regularly using this feature?

How can we investigate this problem?

Thanks

Thomas














[-- Attachment #2: Type: text/html, Size: 3904 bytes --]

[Qemu-devel] Re: USB-OHCI / UHCI: Isochronous transfer not working on Linux host

[Thomas Bandelier]

[-- Attachment #1: Type: text/plain, Size: 3013 bytes --]

Hi Arnon, Max, Pierre,

Arnon, Thanks for your feedback on this point.

As we need to manage several USB 2.0 devices in our environment, we are
going to work on this aspect. We are very motivated to make it work.

Do you (Max and/or Arnon) have any idea of what should be done / reworked in
order to support USB 2.0 devices and especially isochronous transfers in
QEMU?

We would prefer keeping OHCI implementation, but do you think we should
switch to UHCI? Or maybe EHCI?

If we can keep OHCI, what should we begin to do in order to debug current
QEMU OHCI implementation? Same question for the other controllers if we have
to switch.

Best Regards,

Thomas


On Fri, Nov 28, 2008 at 8:32 PM, Arnon Gilboa <arnon.gilboa@qumranet.com>wrote:

>  Hello Thomas,
>
> Since the last time I sent a patch, usb-linux.c & usb-uhci.c were completly
> rewritten by
> Max Krasnyansky and merged by aliguori.
>
> However, looking at my post below from 24 Sep 2007, you can see
> that usb-uhci,
> which is USB 1.1 host controller emulation, was mostly tested on USB 1.1
> devices.
>
> I am sure that many USB 2.0 and even several 1.1 devices (especially
> isochronous)
> will NOT work with the emulation.
>
> http://lists.gnu.org/archive/html/qemu-devel/2007-09/msg00409.html
>
> Regards,
> Arnon
>
>  ------------------------------
> *From:* Thomas Bandelier [mailto:thomas.bandelier@gmail.com]
> *Sent:* Thursday, November 27, 2008 4:06 PM
> *To:* qemu-devel@nongnu.org
> *Cc:* lemagoup@gmail.com; Arnon Gilboa
> *Subject:* USB-OHCI / UHCI: Isochronous transfer not working on Linux host
>
> Hi all,
>
> Following several tests we did with qemu trunk, Isochronous transfer seems
> to be broken with the head of qemu.
>
> We tried following Os (targets):
> -WinXP (i386)
> -Win2k (i386)
> -Several linux kernel ( 2.6.26-1 (i386), 2.6.27 (ARM) )
>
> On our host we are using Gentoo with 2.6.23 and 2.6.27 kernel.
>
> Following USB devices were tested:
> -USB 2.0 logitech webcams
> -Terratec USB2.0 Cynergy hybrid XS FM tuner.
>
> Both OHCI and UHCI configurations were tested in i386 target.
> Only OHCI was tested in ARM target.
>
> In all these configurations, it was *never* possible to initiate
> isochronous USB transfer. It was possible to enumerate the devices but as
> soon as isochronous packets were used, the procedure failed. It means for
> example that no picture was retrieved from the webcams, and no audio data
> was retrieved from FM tuner.
>
> It is important to mention that all theses devices perfectly work on our
> host.
>
> After some search in this maling list archives, it seems that no work has
> been made on these aspects since Arnon Gilboa sent some patches more than
> one year ago. Am I wrong?
>
> So It would be very useful for us to have a feedback from the community on
> this point.
> Has someone met the same problems? Has someone investigated this aspect?
> Is someone regularly using this feature?
>
> How can we investigate this problem?
>
> Thanks
>
> Thomas
>
>
>
>
>
>
>
>
>
>
>
>
>

[-- Attachment #2: Type: text/html, Size: 4834 bytes --]

[Qemu-devel] Re: USB-OHCI / UHCI: Isochronous transfer not working on Linux host

[Max Krasnyansky]

Sorry for delay in replying.
(CCing Mark Burkley who recently posted EHCI patches)

Thomas Bandelier wrote:
> Hi Arnon, Max, Pierre,
> 
> Arnon, Thanks for your feedback on this point.
> 
> As we need to manage several USB 2.0 devices in our environment, we are 
> going to work on this aspect. We are very motivated to make it work.
> 
> Do you (Max and/or Arnon) have any idea of what should be done / 
> reworked in order to support USB 2.0 devices and especially isochronous 
> transfers in QEMU?
Most USB 2.0 devices should work just fine with the current UHCI code. 
Including isochronous transfers. While developing new UHCI 
implementation I was testing it with several 2.0 devices like USB 
storage devices and MS Webcam. Webcam uses isoc transfers and worked 
fairly well.
However since UHCI does not do 2.0 rates USB bandwidth is much lower in 
the guest.

> We would prefer keeping OHCI implementation, but do you think we should 
> switch to UHCI? Or maybe EHCI?
> 
> If we can keep OHCI, what should we begin to do in order to debug 
> current QEMU OHCI implementation? Same question for the other 
> controllers if we have to switch.
Ideally we need to switch to EHCI in order to support high USB2.0 
speeds. Not too long ago Mark posted a EHCI implementation (I have not 
had much luck with it and XP yet though).
Otherwise if you're ok with lower BW then UHCI should work fine.

Max

Re: [Qemu-devel] USB-OHCI / UHCI: Isochronous transfer not working on Linux host

[baibai_tao]

I met the same problem when using usb-ohci.c with usb-linux.c and usb.c in
head version.
Comparing usb-ohci.c on head and in 0.9.1, there is little difference but
the other 2 files are totally different. 
It could work if I pickup  usb-linux.c and usb.c from 0.9.1.


Thomas Bandelier-2 wrote:
> 
> Hi all,
> 
> So It would be very useful for us to have a feedback from the community on
> this point.
> Has someone met the same problems? Has someone investigated this aspect?
> Is someone regularly using this feature?
> 
> How can we investigate this problem?
> 
> Thanks
> 
> Thomas
> 
> 

-- 
View this message in context: http://www.nabble.com/USB-OHCI---UHCI%3A-Isochronous-transfer-not-working-on-Linux-host-tp20719918p20914454.html
Sent from the QEMU - Dev mailing list archive at Nabble.com.