[Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Erik Rull]

Hi all,

is there a summary existing that shows up the rough or actual differences 
between qemu --enable-kvm and qemu-kvm? I tested both versions with the 
same compile and start options, the CPU performance results are identical, 
only the bootup time of my guest system with qemu-kvm seemed to be a bit 
faster (not measured, it just feeled so).

Thanks.

Best regards,

Erik

Re: [Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Stefan Hajnoczi]

On Mon, May 21, 2012 at 5:54 PM, Erik Rull <webmaster@rdsoftware.de> wrote:
> is there a summary existing that shows up the rough or actual differences
> between qemu --enable-kvm and qemu-kvm? I tested both versions with the same
> compile and start options, the CPU performance results are identical, only
> the bootup time of my guest system with qemu-kvm seemed to be a bit faster
> (not measured, it just feeled so).

For production KVM instances I think it still makes sense to use
qemu-kvm packages from your distro or qemu-kvm upstream source.

Jan Kiszka has reduced the delta between qemu.git and qemu-kvm.git to
the point where I think the list of differences is rather small -
maybe PCI passthrough stuff, irqfd for vhost-net (which is now also
being upstreamed into qemu.git), and a few other things I don't know
of.

For development most patches should be against qemu.git unless they
have a dependency on qemu-kvm.git code.

Stefan

Re: [Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Jan Kiszka]

On 2012-05-22 07:04, Stefan Hajnoczi wrote:
> On Mon, May 21, 2012 at 5:54 PM, Erik Rull <webmaster@rdsoftware.de> wrote:
>> is there a summary existing that shows up the rough or actual differences
>> between qemu --enable-kvm and qemu-kvm? I tested both versions with the same
>> compile and start options, the CPU performance results are identical, only
>> the bootup time of my guest system with qemu-kvm seemed to be a bit faster
>> (not measured, it just feeled so).

Current upstream does not enable the in-kernel irqchip of KVM by
default. This should explain the difference in boot-up times. Try
"-machine accel=kvm,kernel_irqchip=on". But the default will be on, just
like in qemu-kvm, once [1] is merged.

> 
> For production KVM instances I think it still makes sense to use
> qemu-kvm packages from your distro or qemu-kvm upstream source.
> 
> Jan Kiszka has reduced the delta between qemu.git and qemu-kvm.git to
> the point where I think the list of differences is rather small -
> maybe PCI passthrough stuff, irqfd for vhost-net (which is now also
> being upstreamed into qemu.git), and a few other things I don't know
> of.

Right, the list of differences is dramatically shrinking. As stated in
[2], soon only PCI passthrough and legacy interface dependencies on
qemu-kvm will be the remaining reasons to use it. If we are lucky, PCI
passthrough will also make it into upstream for QEMU 1.2, we are working
on this.

> 
> For development most patches should be against qemu.git unless they
> have a dependency on qemu-kvm.git code.

Yes, unless you are working on the upstream merge itself, there is
practically no reason anymore to develop against qemu-kvm directly.

Jan

[1] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91171
[2] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91026

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Jan Kiszka]

On 2012-05-22 07:34, Jan Kiszka wrote:
> On 2012-05-22 07:04, Stefan Hajnoczi wrote:
>> On Mon, May 21, 2012 at 5:54 PM, Erik Rull <webmaster@rdsoftware.de> wrote:
>>> is there a summary existing that shows up the rough or actual differences
>>> between qemu --enable-kvm and qemu-kvm? I tested both versions with the same
>>> compile and start options, the CPU performance results are identical, only
>>> the bootup time of my guest system with qemu-kvm seemed to be a bit faster
>>> (not measured, it just feeled so).
> 
> Current upstream does not enable the in-kernel irqchip of KVM by
> default. This should explain the difference in boot-up times. Try
> "-machine accel=kvm,kernel_irqchip=on". But the default will be on, just
> like in qemu-kvm, once [1] is merged.
> 
>>
>> For production KVM instances I think it still makes sense to use
>> qemu-kvm packages from your distro or qemu-kvm upstream source.
>>
>> Jan Kiszka has reduced the delta between qemu.git and qemu-kvm.git to
>> the point where I think the list of differences is rather small -
>> maybe PCI passthrough stuff, irqfd for vhost-net (which is now also
>> being upstreamed into qemu.git), and a few other things I don't know
>> of.
> 
> Right, the list of differences is dramatically shrinking. As stated in
> [2], soon only PCI passthrough and legacy interface dependencies on
> qemu-kvm will be the remaining reasons to use it. If we are lucky, PCI
> passthrough will also make it into upstream for QEMU 1.2, we are working
> on this.
> 
>>
>> For development most patches should be against qemu.git unless they
>> have a dependency on qemu-kvm.git code.
> 
> Yes, unless you are working on the upstream merge itself, there is
> practically no reason anymore to develop against qemu-kvm directly.
> 
> Jan
> 
> [1] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91171
> [2] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91026
> 

I've added some more details on this to the QEMU wiki, see
http://wiki.qemu.org/KVM.

BTW, if someone could have a look at the VGA diffs and resolve them,
that would be great. Gerd, what's the state of switching the BIOS?

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Gerd Hoffmann]

  Hi,

> BTW, if someone could have a look at the VGA diffs and resolve them,
> that would be great.

There isn't much ...

--- a/hw/vga_int.h
+++ b/hw/vga_int.h
@@ -31,8 +31,8 @@

-#define VBE_DISPI_MAX_XRES              1600
-#define VBE_DISPI_MAX_YRES              1200
+#define VBE_DISPI_MAX_XRES              2560
+#define VBE_DISPI_MAX_YRES              1600

The vgabios (both lgpl'ed and seavgabios) filters the mode list by
available memory anyway, so there is no need to keep those low enougth
that the xmax * ymax * 32bpp fits into vga memory.  And when the vgamem
becomes configurable (see below) this will be moot anyway.

I think we should just make them large enougth that they are practically
unlimited.  Those are 16bit registers in virtual hardware.  Guess we
better leave the high bit clear to avoid possible issues with signed
integers.  So (1<<14) aka 16384 maybe?  Or 16000?

-#define VGA_RAM_SIZE (8192 * 1024)
+#define VGA_RAM_SIZE (16 * 1024 * 1024)

That one is more tricky as it breaks migration.  I guess qemu has to
stick to 8M by default for that reason.  We certainly can make the vga
memory size configurable via property though, so you can easily go for
16M or even more (or 1M to reduce the memory footprint of your guests
when using text mode only).

> Gerd, what's the state of switching the BIOS?

Postponed to 1.2.  Too risky for 1.1, also malc vetoed the switch
without seavgabios supporting the vesa protected mode interface which
still needs to be done.

cheers,
  Gerd

Re: [Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Erik Rull]

Jan Kiszka wrote:
> On 2012-05-22 07:34, Jan Kiszka wrote:
>> On 2012-05-22 07:04, Stefan Hajnoczi wrote:
>>> On Mon, May 21, 2012 at 5:54 PM, Erik Rull<webmaster@rdsoftware.de>  wrote:
>>>> is there a summary existing that shows up the rough or actual differences
>>>> between qemu --enable-kvm and qemu-kvm? I tested both versions with the same
>>>> compile and start options, the CPU performance results are identical, only
>>>> the bootup time of my guest system with qemu-kvm seemed to be a bit faster
>>>> (not measured, it just feeled so).
>>
>> Current upstream does not enable the in-kernel irqchip of KVM by
>> default. This should explain the difference in boot-up times. Try
>> "-machine accel=kvm,kernel_irqchip=on". But the default will be on, just
>> like in qemu-kvm, once [1] is merged.
>>
>>>
>>> For production KVM instances I think it still makes sense to use
>>> qemu-kvm packages from your distro or qemu-kvm upstream source.
>>>
>>> Jan Kiszka has reduced the delta between qemu.git and qemu-kvm.git to
>>> the point where I think the list of differences is rather small -
>>> maybe PCI passthrough stuff, irqfd for vhost-net (which is now also
>>> being upstreamed into qemu.git), and a few other things I don't know
>>> of.
>>
>> Right, the list of differences is dramatically shrinking. As stated in
>> [2], soon only PCI passthrough and legacy interface dependencies on
>> qemu-kvm will be the remaining reasons to use it. If we are lucky, PCI
>> passthrough will also make it into upstream for QEMU 1.2, we are working
>> on this.
>>
>>>
>>> For development most patches should be against qemu.git unless they
>>> have a dependency on qemu-kvm.git code.
>>
>> Yes, unless you are working on the upstream merge itself, there is
>> practically no reason anymore to develop against qemu-kvm directly.
>>
>> Jan
>>
>> [1] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91171
>> [2] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91026
>>
>
> I've added some more details on this to the QEMU wiki, see
> http://wiki.qemu.org/KVM.
>
> BTW, if someone could have a look at the VGA diffs and resolve them,
> that would be great. Gerd, what's the state of switching the BIOS?
>
> Jan
>

Hi all,

thanks a lot!

I don't use PCI device assignment - so the missing irqchip-default-option 
should be the biggest difference between these two versions, right?

Best regards,

Erik

Re: [Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Jan Kiszka]

On 2012-05-22 13:12, Erik Rull wrote:
> Jan Kiszka wrote:
>> On 2012-05-22 07:34, Jan Kiszka wrote:
>>> On 2012-05-22 07:04, Stefan Hajnoczi wrote:
>>>> On Mon, May 21, 2012 at 5:54 PM, Erik Rull<webmaster@rdsoftware.de>  wrote:
>>>>> is there a summary existing that shows up the rough or actual differences
>>>>> between qemu --enable-kvm and qemu-kvm? I tested both versions with the same
>>>>> compile and start options, the CPU performance results are identical, only
>>>>> the bootup time of my guest system with qemu-kvm seemed to be a bit faster
>>>>> (not measured, it just feeled so).
>>>
>>> Current upstream does not enable the in-kernel irqchip of KVM by
>>> default. This should explain the difference in boot-up times. Try
>>> "-machine accel=kvm,kernel_irqchip=on". But the default will be on, just
>>> like in qemu-kvm, once [1] is merged.
>>>
>>>>
>>>> For production KVM instances I think it still makes sense to use
>>>> qemu-kvm packages from your distro or qemu-kvm upstream source.
>>>>
>>>> Jan Kiszka has reduced the delta between qemu.git and qemu-kvm.git to
>>>> the point where I think the list of differences is rather small -
>>>> maybe PCI passthrough stuff, irqfd for vhost-net (which is now also
>>>> being upstreamed into qemu.git), and a few other things I don't know
>>>> of.
>>>
>>> Right, the list of differences is dramatically shrinking. As stated in
>>> [2], soon only PCI passthrough and legacy interface dependencies on
>>> qemu-kvm will be the remaining reasons to use it. If we are lucky, PCI
>>> passthrough will also make it into upstream for QEMU 1.2, we are working
>>> on this.
>>>
>>>>
>>>> For development most patches should be against qemu.git unless they
>>>> have a dependency on qemu-kvm.git code.
>>>
>>> Yes, unless you are working on the upstream merge itself, there is
>>> practically no reason anymore to develop against qemu-kvm directly.
>>>
>>> Jan
>>>
>>> [1] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91171
>>> [2] http://thread.gmane.org/gmane.comp.emulators.kvm.devel/91026
>>>
>>
>> I've added some more details on this to the QEMU wiki, see
>> http://wiki.qemu.org/KVM.
>>
>> BTW, if someone could have a look at the VGA diffs and resolve them,
>> that would be great. Gerd, what's the state of switching the BIOS?
>>
>> Jan
>>
> 
> Hi all,
> 
> thanks a lot!
> 
> I don't use PCI device assignment - so the missing irqchip-default-option 
> should be the biggest difference between these two versions, right?

And the lacking MSI support when irqchip is on. If that matters depends
on your workload.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

[Qemu-devel] Current differences between qemu --enable-kvm and qemu-kvm?

[Erik Rull]

Hi all,

is there a summary existing that shows up the rough or actual differences 
between qemu --enable-kvm and qemu-kvm? I tested both versions with the 
same compile and start options, the CPU performance results are identical, 
only the bootup time of my guest system with qemu-kvm seemed to be a bit 
faster (not measured, it just feeled so).

Thanks.

Best regards,

Erik