Re: [Qemu-devel] [PATCH 4/5] uhci: rewrite UHCI emulator, fully async operation with multiple outstanding transactions

[Juergen Lock]

On Thu, Oct 16, 2008 at 12:05:43AM +0200, andrzej zaborowski wrote:
> 2008/10/15 Max Krasnyansky <address@hidden>:
> > Hi Juergen,
> >
> > Juergen Lock wrote:
> >>
> >> This is whats got committed as r5050, right?  It causes a large slowdown
> >> with some linux guests, as posted here:
> >>        http://lists.gnu.org/archive/html/qemu-devel/2008-09/msg01201.html
> >> As mentioned the guest I tested was:
> >>        sidux-2008-03-ourea-pre1-kde-lite-i386-200809142136.iso
> >> It uses a Linux 2.6.26 kernel, a FreeBSD guest is not affected by the
> >> slowness, could this be related to dynticks?
> >>
> >>  Oh and on one box transfers over the emulated usb nic even sometimes
> >> stall, I have to ping the guest's ip for them to continue, and the ping
> >> then loses packets too.
> >
> > Can you please clarify your setup. As I understood, from the thread you
> > mentioned above, you're running Linux 2.6.26 guest on QEMU under FreeBSD
> > host,correct ?

Correct.
> 
> For the record on Linux host I observed the same issues.  Juergen
> mentioned that in an earlier mail.  The transfer rate is about 60kB/s
> reading from an emulated mass storage device using a raw image.  It
> was before about 1MB/s according to Juergen.  I wasn't using dynticks.
> 
 Also no dynticks in your guest?  The sidux iso we both tried does seem to
use dynticks at least...

 Anyway I have enabled debug output like Max suggested, booted a sidux guest
to runlevel 3 (which took quite a while...) and when it finally was up ran:

$ wget 10.0.2.2:631/help/ref-cupsd-conf.html
(gave <40 kB/s)
$ su
# rmmod uhci-hcd
# modprobe uhci-hcd
# dhclient eth0
# ^D
$ wget 10.0.2.2:631/help/ref-cupsd-conf.html
(again gave <40 kB/s)

 (the wget assumes cups running on the host, the file it gets is 85264
bytes here.)

 The resulting log is 1.6G, tho running it thru bzip2 twice shrinks it
down to 10M, I put them here:
        http://people.freebsd.org/~nox/qemu/usblog.bz2 (16M)
        http://people.freebsd.org/~nox/qemu/usblog.bz2.bz2 (10M)

 Oh and about whether I'm sure its the uhci changes that cause the slowdown,
all I can say is that I played with several svn revisions and r5049 and
earlier were way faster with this(!) guest than r5050 and later, as posted
here:
        http://lists.gnu.org/archive/html/qemu-devel/2008-09/msg01201.html

 And btw, I also since noticed that with this guest and a recent snapshot
(I haven't tried r5050) emulated scsi disks are slower (about 500 K/s here
with dd bs=64k) than in FreeBSD guests (more than 15 MB/s), and I remember
seeing the opposite, i.e. decent thruput with earlier linux guests and qemu
versions, and less (2 MB/s or so) with FreeBSD guests on that qemu version.

 HTH,
        Juergen