public inbox for kvm@vger.kernel.org
 help / color / mirror / Atom feed
From: Konstantin Khlebnikov <khlebnikov@parallels.com>
To: Prasad Joshi <prasadjoshi124@gmail.com>
Cc: "kvm@vger.kernel.org" <kvm@vger.kernel.org>
Subject: Re: GPU Pass-through need help.
Date: Thu, 10 Feb 2011 09:45:57 +0300	[thread overview]
Message-ID: <4D5389A5.1040705@parallels.com> (raw)
In-Reply-To: <AANLkTi=yLoU=q9xUbxQYFkQ-iqYSqN1cP1EvQv-PiMb=@mail.gmail.com>

Prasad Joshi wrote:
> On Wed, Feb 9, 2011 at 4:33 PM, Konstantin Khlebnikov
> <khlebnikov@openvz.org <mailto:khlebnikov@openvz.org>> wrote:
>
>     Prasad Joshi wrote:
>
>         I was previously seeing a problem with the ROM BIOS, but I could
>         solve
>         it by passing a correct BIOS file in function assigned_initfn.
>         For now
>         I am using hardcoded file name in the code.
>
>
>     Did you recheck video bios inside guest? (for example in sysfs)
>
>
> For GPU cards on my machine, reading from the
> /sys/bus/pci/devices/<device>/rom always give error 'Invalid Argument'.

Yep, rom in sysfs must be enabled before via 'echo 1 > rom'
If rom was corrupted you will see warning in dmesg.

> I had to extract the BIOS using some utilities. The BIOS seems to be
> correct as I have been able to pass-through a old Nvidia (NVS 290) and
> old ATI (RV370 FireGL V3100) card to Ubuntu Guest OS.
>
>
>     If this is bootstrap vga, guest linux kernel will use bios copy
>     placed at 0xc0000.
>     But in my setup qemu can only pass first 32k of option-roms,
>     due to curruption at seabios shadow-copying stage.
>     (seems like because improper i440fx PAM registers emulation)
>
>
> I am not sure if it is related to the problem you are facing. The
> current SeaBIOS has very limited pci memory space from 0xf0000000  to
> 0xfec00000 (236MB). This includes everything, prefectable and
> non-prefecthable memory. The range is very small for VGA cards to work.
> I have modified the memory range on my machine for prefectable memory.

But if in your guest /sys/bus/pci/devices/<device>/boot_vga == 1
linux will use shadow copy from 0xc0000 instead of pci rom bar

>
> There is one bug in qemu-kvm code, which acidently detects the
> prefectable memory as non-prefectable. I had sent a patch to correct
> this behavior yesterday. Hope it gets accepted.

Yes, I saw him, IORESOURCE_PREFETCH was accidentally changed in linux kernel.
Maybe better to detect it via opening and mapping /resorce<id>_wc instad of /resource<id>,
it exist only for prefechable resources. Plus qemu will get write-combined mapping.

>
> Thanks and Regards,
> Prasad


      parent reply	other threads:[~2011-02-10  6:46 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-01-06 15:01 GPU Pass-through need help Prasad Joshi
2011-02-09 16:33 ` Konstantin Khlebnikov
2011-02-09 20:40   ` Prasad Joshi
     [not found]   ` <AANLkTi=yLoU=q9xUbxQYFkQ-iqYSqN1cP1EvQv-PiMb=@mail.gmail.com>
2011-02-10  6:45     ` Konstantin Khlebnikov [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=4D5389A5.1040705@parallels.com \
    --to=khlebnikov@parallels.com \
    --cc=kvm@vger.kernel.org \
    --cc=prasadjoshi124@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox