Re: PCI passthrough issue

[Jean Baptiste Favre <xen-devel@jbfavre.org>]

Hello Ian,

Le 01/02/2011 12:34, Ian Campbell a écrit :
> On Fri, 2011-01-28 at 15:47 +0000, Jean Baptiste Favre wrote:
> > Hello,
> > I made some more tests today, still with 2.6.37 32bits kernel from
> > Debian experimental, with various memory allocation value.
> >
> > For each test, I make ping on my gateway with various packet size:
> > ping -s15 10.0.0.1
> > ping -s85 10.0.0.1
> > ping -s86 10.0.0.1
> > ping -s150 10.0.0.1
> >
> > Results bellow:
> >
> > - less than 256mb: works
> > - between 256 and 512mb: ping greater than 85 bytes does not work
> > - more than 512mb: works
> >
> > I'm lost...
>
> Me too, this really is the most inexplicable set of symptoms...
>
> Does it work correctly with any other guest kernel, e.g. the
> xen/stable-2.6.32.x branch from xen.git or maybe one of the old-style
> Xen kernels?
I'm compiling 2.6.32 kernel from Jeremy's GIT repos to check that.

> The network device in use is one of the Intel NICs below? Any luck just
> passing through that one device without all the others?
>
> Previously you mentioned using a Marvell NIC, so I guess the failure is
> independent of the NIC type?
I made all tests with Marvell NIC (driver sky2).
I don't know if the same behaviour occurs with another NIC type. I'll
try with an Intel one (I've a dual port Intel NIC, so I could
passthrough only one port)

> Your userspace is still OpenWRT in these most recent tests? Is that some
> sort of busybox based thing? Can you try with e.g. a regular Debian
> guest userspace to rule out any funnyness from that end?
I made tests with both OpenWRT and Debian Squeeze.
Had problems to compile OpenWRT kernel in 64bits :)
I tested Debian Squeeze with 2.6.37 kernel from experimental (because of
Xen PCI Frontend integration. Not sure it has been backported into 2.6.32).
I will test with Jeremy's 2.6.32 kernel (see above).
As a conclusion, last results are from Debian Squeeze with 2.6.37 and
that's why I wrote on debian-kernel maillist :)

> If you restrict dom0 to >256MB but <512MB (using dom0_mem= on hypervisor
> command line) does the NIC work correctly in non-passedthrough form?
My Xen hypervisor commandline is as follow:
placeholder dom0_mem=256M dom0_max_vcpus=1 dom0_vcpus_pin loglvl=all
guest_loglvl=all com1=115200,8n1 console=com1

Everything works great without passthrough, but my dom0 is 64bits which
may explain that (I do have this strange behaviour only with 32bits
kernels).

I did not tried changing dom0_mem param.

> Similarly does the kernel running native with mem= cause the failure?
Not sure I understand what you mean here.
BTW, I'm preparing a set of automatic tests with different memory
values. That will be:
* loop for each mem value
- set memory in domU configfile
- starting domU with 128Mb memory
- rc.local will ping my gateway with different packet size, store result
in file
- halt domU
* end of loop
* Check results :-/

> Bit of a long shot but are you able to try a 4.0.2-rc hypervisor+tools
> and/or a 4.1.0-rc setup (not branched yet so still in xen-unstable.hg)?
I can eventually try it, but after my looong test list :)

> Is the 10.0.0.1 address you are testing against a VM on the same host or
> some sort of external entity?
It's my gateway (for the history, WRT54GL with OpenWRT).

Regards,
JB