[Qemu-devel] USB support

[Qemu-devel] USB support

[Fabrice Bellard]

Hi,

I just commited an initial USB support for QEMU. This USB layer will 
ultimately enable QEMU to use some host USB devices and to simulate USB 
devices.

WARNING: the current USB support is for USB hackers only - it is not 
meant to be fully usable yet.

The following features are implemented:

- PCI UHCI USB controller (I finally decided to implement UHCI because I 
know it better than OHCI and because Bochs has a similar driver. Of 
course it would still be very interesting to have an equivalent OHCI 
controller for non PC targets and an EHCI controller for USB 2.0 devices).

- Virtual USB 1.1 hub.

- Linux host USB redirector to use the USB 1.1 host devices which are 
not requested by the host OS (i.e. no host driver is loaded for them). 
It is *very* limited and buggy at the moment, but I was able (once !) to 
mount a disk-on-key flash device.

I plan to implement a USB mouse device as in Bochs just to have at least 
one "useful" virtual USB device to play with. Anyone is free to 
implement other devices by looking at the USB hub implementation in usb.c.

Fabrice.

Re: [Qemu-devel] USB support

[Volker Ruppert]

Hi,

> The following features are implemented:
>
> - PCI UHCI USB controller (I finally decided to implement UHCI because I
> know it better than OHCI and because Bochs has a similar driver. Of
> course it would still be very interesting to have an equivalent OHCI
> controller for non PC targets and an EHCI controller for USB 2.0 devices).

I compared the Qemu USB controller with the Bochs one using Ralph Brown's 
pcifg utility and found two different things.
- Bochs USB appears at function 2 of the PIIX3
- Bochs USB uses PIRQ line INTD
You can find both things in the PIIX3 documentation.

I tried the Qemu USB implementation with Win98 here. The hubs are detected 
correctly, but it makes Win98 hang on shutdown.

> - Linux host USB redirector to use the USB 1.1 host devices which are
> not requested by the host OS (i.e. no host driver is loaded for them).
> It is *very* limited and buggy at the moment, but I was able (once !) to
> mount a disk-on-key flash device.

I guess the host OS doesn't like modifing data on a mounted devices. It might 
be okay for input-only devices. I cannot try it here, since it requires root 
permissions.

--
Bye

Volker

Re: [Qemu-devel] USB support

[Fabrice Bellard]

Volker Ruppert wrote:
> Hi,
> 
> 
>>The following features are implemented:
>>
>>- PCI UHCI USB controller (I finally decided to implement UHCI because I
>>know it better than OHCI and because Bochs has a similar driver. Of
>>course it would still be very interesting to have an equivalent OHCI
>>controller for non PC targets and an EHCI controller for USB 2.0 devices).
> 
> 
> I compared the Qemu USB controller with the Bochs one using Ralph Brown's 
> pcifg utility and found two different things.
> - Bochs USB appears at function 2 of the PIIX3
> - Bochs USB uses PIRQ line INTD
> You can find both things in the PIIX3 documentation.

I'll try to update that.

> I tried the Qemu USB implementation with Win98 here. The hubs are detected 
> correctly, but it makes Win98 hang on shutdown.

I only tested with a Linux 2.4 guest OS. The USB mouse is working in X11 
and the USB hubs seem to work too.

>>- Linux host USB redirector to use the USB 1.1 host devices which are
>>not requested by the host OS (i.e. no host driver is loaded for them).
>>It is *very* limited and buggy at the moment, but I was able (once !) to
>>mount a disk-on-key flash device.
> 
> 
> I guess the host OS doesn't like modifing data on a mounted devices. It might 
> be okay for input-only devices. I cannot try it here, since it requires root 
> permissions.

I will add a documentation once it works better, but here are some 
information :

1) The host OS must not use the USB device. It means in particular that 
no host OS driver must be present for that device. The solution I am 
using is to rename the host kernel module "usb-storage.o" to 
"usb-storage.o.disabled" so that it is not loaded by Linux. Then QEMU 
can exclusively access to the corresponding host storage USB device. The 
same apply to every other type of USB devices.

2) In order not to launch QEMU as root, I changed the permissions in 
/proc/bus/usb : chown -R myuid /proc/bus/usb. I am sure it is possible 
to find a better solution !

3) Isosynchronous USB packets are not redirected yet, so host webcams 
have no chance to work.

Fabrice.

Re: [Qemu-devel] USB support

[Lonnie Mendez]

On Saturday 05 November 2005 11:10 am, Fabrice Bellard wrote: 
> 1) The host OS must not use the USB device. It means in particular that 
> no host OS driver must be present for that device. The solution I am 
> using is to rename the host kernel module "usb-storage.o" to 
> "usb-storage.o.disabled" so that it is not loaded by Linux. Then QEMU 
> can exclusively access to the corresponding host storage USB device. The 
> same apply to every other type of USB devices.

   usb devfs has a disconnect ioctl (USBDEVFS_DISCONNECT) which
allows you to disconnect a single device from a driver.

   A suggestion is to hook the interface up to qemu monitor.  Then you would 
be able to list usb devices and then claim only the ones you want to use.

Re: [Qemu-devel] USB support

[Fabrice Bellard]

Lonnie Mendez wrote:
> On Saturday 05 November 2005 11:10 am, Fabrice Bellard wrote: 
> 
>>1) The host OS must not use the USB device. It means in particular that 
>>no host OS driver must be present for that device. The solution I am 
>>using is to rename the host kernel module "usb-storage.o" to 
>>"usb-storage.o.disabled" so that it is not loaded by Linux. Then QEMU 
>>can exclusively access to the corresponding host storage USB device. The 
>>same apply to every other type of USB devices.
> 
> 
>    usb devfs has a disconnect ioctl (USBDEVFS_DISCONNECT) which
> allows you to disconnect a single device from a driver.
> 
>    A suggestion is to hook the interface up to qemu monitor.  Then you would 
> be able to list usb devices and then claim only the ones you want to use.

Good idea. I am adding something like that:

In the monitor:

'info usbhost' lists the available host USB devices

'usbhost_add 3.4' or 'usbhost_add 1234:5678' adds the USB host device on 
bus 3 at address 4 or the first one matching the vendor ID 0x1234 and 
product ID 0x5678.

The QEMU option '-usbhost x' can be used to do the same as usbhost_add.

Fabrice.

Re: [Qemu-devel] [patch] USB support

[Oliver Gerlich]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Fabrice Bellard schrieb:
> Lonnie Mendez wrote:
> 
>> On Saturday 05 November 2005 11:10 am, Fabrice Bellard wrote:
>>
>>> 1) The host OS must not use the USB device. It means in particular
>>> that no host OS driver must be present for that device. The solution
>>> I am using is to rename the host kernel module "usb-storage.o" to
>>> "usb-storage.o.disabled" so that it is not loaded by Linux. Then QEMU
>>> can exclusively access to the corresponding host storage USB device.
>>> The same apply to every other type of USB devices.
>>
>>
>>
>>    usb devfs has a disconnect ioctl (USBDEVFS_DISCONNECT) which
>> allows you to disconnect a single device from a driver.
>>
>>    A suggestion is to hook the interface up to qemu monitor.  Then you
>> would be able to list usb devices and then claim only the ones you
>> want to use.
> 
> 
> Good idea. I am adding something like that:
> 
> In the monitor:
> 
> 'info usbhost' lists the available host USB devices
> 
> 'usbhost_add 3.4' or 'usbhost_add 1234:5678' adds the USB host device on
> bus 3 at address 4 or the first one matching the vendor ID 0x1234 and
> product ID 0x5678.
> 
> The QEMU option '-usbhost x' can be used to do the same as usbhost_add.
> 
> Fabrice.
> 
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://lists.nongnu.org/mailman/listinfo/qemu-devel
> 
> 

When I use usb_add, I get:
USBDEVFS_DISCONNECT: Inappropriate ioctl for device

This is with snapshot from 2005-11-11, on Debian testing, kernel 2.6.13
(with 1000Hz timer).
The patch below fixes this (plainly taken from
http://www.mail-archive.com/linux-usb-devel@lists.sourceforge.net/msg35550.html
; is ifno = 0 correct?)


diff -Naur qemu-snapshot-2005-11-11_23/usb-linux.c
qemu-snapshot-2005-11-11_23-usb/usb-linux.c
- --- qemu-snapshot-2005-11-11_23/usb-linux.c     2005-11-06
17:13:29.000000000 +0
100
+++ qemu-snapshot-2005-11-11_23-usb/usb-linux.c 2005-11-13
01:10:19.000000000 +0
100
@@ -181,7 +181,10 @@

 #ifdef USBDEVFS_DISCONNECT
     /* earlier Linux 2.4 do not support that */
- -    ret = ioctl(fd, USBDEVFS_DISCONNECT);
+    struct usbdevfs_ioctl ctrl;
+    ctrl.ioctl_code = USBDEVFS_DISCONNECT;
+    ctrl.ifno = 0;
+    ret = ioctl(fd, USBDEVFS_IOCTL, &ctrl);
     if (ret < 0 && errno != ENODATA) {
         perror("USBDEVFS_DISCONNECT");
         goto fail;


With this patch, qemu recognized all USB devices I had available - which
is actually only one device, a joystick :)
Win2k showed it in the control panel with all available buttons and axes
and reacted to events correctly.

Though, a strange thing is that as long as the device is attached to the
guest, Windows gets very slow. The Windows task manager shows that the
guest CPU load is at 100%; but "top" on the guest shows that qemu uses
only 20% host CPU. This starts as soon as usb_add is executed, and stops
as soon as usb_del is executed.
Does qemu use sleep in the usb code, or does maybe Windows sent the CPU
a HLT ? I don't really understand that behaviour :)


Regards,
Oliver
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQFDdpAWTFOM6DcNJ6cRAjzgAJ4rJlhVBe8thmg3VCBRRs+iw7dCsgCeI8s/
vDSywn5WpfH4vkCNSfEpDvI=
=Djpx
-----END PGP SIGNATURE-----

Re: [Qemu-devel] [patch] USB support

[Fabrice Bellard]

> When I use usb_add, I get:
> USBDEVFS_DISCONNECT: Inappropriate ioctl for device
> 
> This is with snapshot from 2005-11-11, on Debian testing, kernel 2.6.13
> (with 1000Hz timer).
> The patch below fixes this (plainly taken from
> http://www.mail-archive.com/linux-usb-devel@lists.sourceforge.net/msg35550.html
> ; is ifno = 0 correct?)

Thank you. I was not able to test this ioctl on my PC, hence the problem...

> With this patch, qemu recognized all USB devices I had available - which
> is actually only one device, a joystick :)
> Win2k showed it in the control panel with all available buttons and axes
> and reacted to events correctly.
> 
> Though, a strange thing is that as long as the device is attached to the
> guest, Windows gets very slow. The Windows task manager shows that the
> guest CPU load is at 100%; but "top" on the guest shows that qemu uses
> only 20% host CPU. This starts as soon as usb_add is executed, and stops
> as soon as usb_del is executed.
> Does qemu use sleep in the usb code, or does maybe Windows sent the CPU
> a HLT ? I don't really understand that behaviour :)

As mentionned in the documentation, the USB host emulation is very slow 
in QEMU. The problem is that USB requests block the whole VM. I plan to 
implement non blocking USB I/Os soon.

Fabrice.

[Qemu-devel] Re: USB support

[Matthew Mastracci]

> 1) The host OS must not use the USB device. It means in particular that 
> no host OS driver must be present for that device. The solution I am 
> using is to rename the host kernel module "usb-storage.o" to 
> "usb-storage.o.disabled" so that it is not loaded by Linux. Then QEMU 
> can exclusively access to the corresponding host storage USB device. The 
> same apply to every other type of USB devices.

For those using Fedora Core (and possibly a number of other Linux 
distributions), you can add your module to /etc/hotplug/blacklist to 
ensure it won't load.  I've been using this trick to get the USB support 
in VMWare to work for some time.

Matt.

Re: [Qemu-devel] USB support

[Mark Williamson]

> I just commited an initial USB support for QEMU. This USB layer will
> ultimately enable QEMU to use some host USB devices and to simulate USB
> devices.

Coolness!

> - Linux host USB redirector to use the USB 1.1 host devices which are
> not requested by the host OS (i.e. no host driver is loaded for them).
> It is *very* limited and buggy at the moment, but I was able (once !) to
> mount a disk-on-key flash device.

Had you thought of adding a minimal kernel module to allow "stealing" of host 
drivers for use by QEmu?  When I looked into this, I had the impression that 
the 2.6 USB code ought to be able to support it, although I'm not sure how it 
would interact with the userspace driver layer in your case.

Cheers,
Mark

> I plan to implement a USB mouse device as in Bochs just to have at least
> one "useful" virtual USB device to play with. Anyone is free to
> implement other devices by looking at the USB hub implementation in usb.c.
>
> Fabrice.
>
>
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://lists.nongnu.org/mailman/listinfo/qemu-devel

Re: [Qemu-devel] USB support

[Paul Brook]

> > - Linux host USB redirector to use the USB 1.1 host devices which are
> > not requested by the host OS (i.e. no host driver is loaded for them).
> > It is *very* limited and buggy at the moment, but I was able (once !) to
> > mount a disk-on-key flash device.
>
> Had you thought of adding a minimal kernel module to allow "stealing" of
> host drivers for use by QEmu?  When I looked into this, I had the
> impression that the 2.6 USB code ought to be able to support it, although
> I'm not sure how it would interact with the userspace driver layer in your
> case.

Someone else already pointed out there's a usbdevfs ioctl for this.

Paul