kvm-userspace: VGA/VESA framebuffer broken

kvm-userspace: VGA/VESA framebuffer broken

[Jan Kiszka]

Hi,

latest kvm-userspace git 6892f63c18a526c7b54bbde2f59287787eabe1f8
appears to have a bug /wrt VGA/VESA modes. I just fired up one of my
Linux test kernels which runs a framebuffer console in mode 0x317, but
the display just contains garbage. Reverting to
82daa70a1d5bcad3a93150ffc5afbcb9e77361fb makes the problem disappear again.

I also tried latest qemu with -enable-kvm against the same kernel
modules, and the result is somehow better in that there is some output
on the screen -- but it is horribly slow.

Glauber, Avi, is this a problem of latest qemu upstream changes to the
vga emulation or a merge issue?

Jan

-- 
Siemens AG, Corporate Technology, CT SE 26
Corporate Competence Center Embedded Linux

Re: kvm-userspace: VGA/VESA framebuffer broken

[Glauber Costa]

On Fri, Dec 5, 2008 at 8:04 AM, Jan Kiszka <jan.kiszka@siemens.com> wrote:
> Hi,
>
> latest kvm-userspace git 6892f63c18a526c7b54bbde2f59287787eabe1f8
> appears to have a bug /wrt VGA/VESA modes. I just fired up one of my
> Linux test kernels which runs a framebuffer console in mode 0x317, but
> the display just contains garbage. Reverting to
> 82daa70a1d5bcad3a93150ffc5afbcb9e77361fb makes the problem disappear again.
>
> I also tried latest qemu with -enable-kvm against the same kernel
> modules, and the result is somehow better in that there is some output
> on the screen -- but it is horribly slow.
>
> Glauber, Avi, is this a problem of latest qemu upstream changes to the
> vga emulation or a merge issue?

There's a bit of each here.

There are some bits that differ because kvm upstream still does things
a little bit different, and some
problems arose. I spoke to avi yesterday, and he intends to release
fixes today (I think, he can confirm that)

Also, in the process of getting the vga optimization cleaned up for
qemu, I found a bug in kvm.ko that would
prevent it from working. If you can, please try the latest git kernel
to see if it fixes for you.

As much as I know, there's the intent to release compatibility code
for people running older kernels for this bug fix specifically


-- 
Glauber  Costa.
"Free as in Freedom"
http://glommer.net

"The less confident you are, the more serious you have to act."

Re: kvm-userspace: VGA/VESA framebuffer broken

[Avi Kivity]

Jan Kiszka wrote:
> Hi,
>
> latest kvm-userspace git 6892f63c18a526c7b54bbde2f59287787eabe1f8
> appears to have a bug /wrt VGA/VESA modes. I just fired up one of my
> Linux test kernels which runs a framebuffer console in mode 0x317, but
> the display just contains garbage. Reverting to
> 82daa70a1d5bcad3a93150ffc5afbcb9e77361fb makes the problem disappear again.
>
> I also tried latest qemu with -enable-kvm against the same kernel
> modules, and the result is somehow better in that there is some output
> on the screen -- but it is horribly slow.
>
> Glauber, Avi, is this a problem of latest qemu upstream changes to the
> vga emulation or a merge issue?
>   

It's caused by the recent merge; qemu upstream now has vga dirty bit 
tracking, done in a different way from kvm-userspace.git.

I've got it fixed here, just need to test a bit more.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: kvm-userspace: VGA/VESA framebuffer broken

[Jan Kiszka]

Avi Kivity wrote:
> Jan Kiszka wrote:
>> Hi,
>>
>> latest kvm-userspace git 6892f63c18a526c7b54bbde2f59287787eabe1f8
>> appears to have a bug /wrt VGA/VESA modes. I just fired up one of my
>> Linux test kernels which runs a framebuffer console in mode 0x317, but
>> the display just contains garbage. Reverting to
>> 82daa70a1d5bcad3a93150ffc5afbcb9e77361fb makes the problem disappear
>> again.
>>
>> I also tried latest qemu with -enable-kvm against the same kernel
>> modules, and the result is somehow better in that there is some output
>> on the screen -- but it is horribly slow.
>>
>> Glauber, Avi, is this a problem of latest qemu upstream changes to the
>> vga emulation or a merge issue?
>>   
> 
> It's caused by the recent merge; qemu upstream now has vga dirty bit
> tracking, done in a different way from kvm-userspace.git.
> 
> I've got it fixed here, just need to test a bit more.

That's good. If you want me to test as well, just throw something over.

Thanks,
Jan

PS: I was on latest kernel.git, Glauber.

-- 
Siemens AG, Corporate Technology, CT SE 26
Corporate Competence Center Embedded Linux

Re: kvm-userspace: VGA/VESA framebuffer broken

[Avi Kivity]

Jan Kiszka wrote:
> That's good. If you want me to test as well, just throw something over.
>
>   

It's pushed as tag kvm-80rc3 on kernel.org (will push as master once it 
passes the regression tests).  Thanks.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: kvm-userspace: VGA/VESA framebuffer broken

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 435 bytes --]

Avi Kivity wrote:
> Jan Kiszka wrote:
>> That's good. If you want me to test as well, just throw something over.
>>
>>   
> 
> It's pushed as tag kvm-80rc3 on kernel.org (will push as master once it
> passes the regression tests).  Thanks.

Garbage is gone, but update does not happen automatically. I have to
switch to the monitor or some other SDL console in back in order to see
the latest framebuffer updates.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]

Re: kvm-userspace: VGA/VESA framebuffer broken

[Avi Kivity]

Jan Kiszka wrote:
> Garbage is gone, but update does not happen automatically. I have to
> switch to the monitor or some other SDL console in back in order to see
> the latest framebuffer updates.
>   

That was indeed a symptom during the brokenness.  Which guest are you 
running?  also, please supply commit hashes you're working with so I can 
check I didn't push the wrong thing.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: kvm-userspace: VGA/VESA framebuffer broken

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 752 bytes --]

Avi Kivity wrote:
> Jan Kiszka wrote:
>> Garbage is gone, but update does not happen automatically. I have to
>> switch to the monitor or some other SDL console in back in order to see
>> the latest framebuffer updates.
>>   
> 
> That was indeed a symptom during the brokenness.  Which guest are you
> running?

Various home-brewed or openSuse kernels from 2.6.23 to 27.

>  also, please supply commit hashes you're working with so I can
> check I didn't push the wrong thing.

58e235dc112a8d0cc309af0f2e526087235ab633 (kernel) and
38fb92b29d43f8503fd78a30158f93d40cd88de2 (userspace).

Jan

PS: fc94d163d1e3424199166cf50449e03447400579 (fold vcpu_info into
CPUState) broke -kvm-no-irqchip. Currently trying to understand why.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]

Re: kvm-userspace: VGA/VESA framebuffer broken

[Avi Kivity]

Jan Kiszka wrote:
> Avi Kivity wrote:
>   
>> Jan Kiszka wrote:
>>     
>>> Garbage is gone, but update does not happen automatically. I have to
>>> switch to the monitor or some other SDL console in back in order to see
>>> the latest framebuffer updates.
>>>   
>>>       
>> That was indeed a symptom during the brokenness.  Which guest are you
>> running?
>>     
>
> Various home-brewed or openSuse kernels from 2.6.23 to 27.
>   

I reproduced this on opensuse 11, and applied a fix.  Unfortunately 
kvm-80 is broken wrt this bug.  Will release kvm-81 sooner than usual.

-- 
error compiling committee.c: too many arguments to function

Re: kvm-userspace: VGA/VESA framebuffer broken

[Michael Tokarev]

Avi Kivity wrote:
> Jan Kiszka wrote:
>> Avi Kivity wrote:
>>  
>>> Jan Kiszka wrote:
>>>    
>>>> Garbage is gone, but update does not happen automatically. I have to
>>>> switch to the monitor or some other SDL console in back in order to see
>>>> the latest framebuffer updates.
[]
> I reproduced this on opensuse 11, and applied a fix.  Unfortunately
> kvm-80 is broken wrt this bug.  Will release kvm-81 sooner than usual.

Oh, it's not just framebuffer, and not just opensuse. It also
is broken on X window and windows(xp) -- see my reply to the
release announcement.

Is this fix available as a patchch somewhere, for me to try please? :)

Thanks!

/mjt

Re: kvm-userspace: VGA/VESA framebuffer broken

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1126 bytes --]

Michael Tokarev wrote:
> Avi Kivity wrote:
>> Jan Kiszka wrote:
>>> Avi Kivity wrote:
>>>  
>>>> Jan Kiszka wrote:
>>>>    
>>>>> Garbage is gone, but update does not happen automatically. I have to
>>>>> switch to the monitor or some other SDL console in back in order to see
>>>>> the latest framebuffer updates.
> []
>> I reproduced this on opensuse 11, and applied a fix.  Unfortunately
>> kvm-80 is broken wrt this bug.  Will release kvm-81 sooner than usual.
> 
> Oh, it's not just framebuffer, and not just opensuse. It also
> is broken on X window and windows(xp) -- see my reply to the
> release announcement.
> 
> Is this fix available as a patchch somewhere, for me to try please? :)

You should either pull the latest git versions from kernel.org
(git://git.kernel.org/pub/scm/virt/kvm/kvm-userspace.git and
git://git.kernel.org/pub/scm/linux/kernel/git/avi/kvm.git) or give this
patch against kvm-userspace a try:

http://git.kernel.org/?p=virt/kvm/kvm-userspace.git;a=commitdiff;h=8eae225cf8cd82316fcc78569aeb1adbbc077cb8

I did the former, and I can confirm it works again.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]