[Qemu-devel] TCG slowdown

[Qemu-devel] TCG slowdown

[Edgar E. Iglesias]

Lately, my TCG emulated machines have been running terribly slow.
Like a factor 3 or 4 slower, hardly usable.

I bisected it to this:

commit d5ab9713d2d4037fd56b0adddd26c8d4dc11cf09
Author: Jan Kiszka <jan.kiszka@siemens.com>
Date:   Tue Aug 2 16:10:21 2011 +0200

    Avoid allocating TCG resources in non-TCG mode
    
    Do not allocate TCG-only resources like the translation buffer when
    running over KVM or XEN. Saves a "few" bytes in the qemu address space
    and is also conceptually cleaner.
    
    Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
    Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>


It turns out that the init of the tb cache is now moved to before ram_size
gets it's default value. And because the default tb cache size, if no
-tb-size option is given, is based on the ram_size it is now zero..

Cheers

Re: [Qemu-devel] TCG slowdown

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1077 bytes --]

On 2011-08-20 13:51, Edgar E. Iglesias wrote:
> Lately, my TCG emulated machines have been running terribly slow.
> Like a factor 3 or 4 slower, hardly usable.
> 
> I bisected it to this:
> 
> commit d5ab9713d2d4037fd56b0adddd26c8d4dc11cf09
> Author: Jan Kiszka <jan.kiszka@siemens.com>
> Date:   Tue Aug 2 16:10:21 2011 +0200
> 
>     Avoid allocating TCG resources in non-TCG mode
>     
>     Do not allocate TCG-only resources like the translation buffer when
>     running over KVM or XEN. Saves a "few" bytes in the qemu address space
>     and is also conceptually cleaner.
>     
>     Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
>     Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
> 
> 
> It turns out that the init of the tb cache is now moved to before ram_size
> gets it's default value. And because the default tb cache size, if no
> -tb-size option is given, is based on the ram_size it is now zero..

Yes, already stumbled over that regression myself. See
http://thread.gmane.org/gmane.comp.emulators.qemu/113547

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: [Qemu-devel] TCG slowdown

[Edgar E. Iglesias]

On Sat, Aug 20, 2011 at 03:37:09PM -0700, Jan Kiszka wrote:
> On 2011-08-20 13:51, Edgar E. Iglesias wrote:
> > Lately, my TCG emulated machines have been running terribly slow.
> > Like a factor 3 or 4 slower, hardly usable.
> > 
> > I bisected it to this:
> > 
> > commit d5ab9713d2d4037fd56b0adddd26c8d4dc11cf09
> > Author: Jan Kiszka <jan.kiszka@siemens.com>
> > Date:   Tue Aug 2 16:10:21 2011 +0200
> > 
> >     Avoid allocating TCG resources in non-TCG mode
> >     
> >     Do not allocate TCG-only resources like the translation buffer when
> >     running over KVM or XEN. Saves a "few" bytes in the qemu address space
> >     and is also conceptually cleaner.
> >     
> >     Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> >     Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
> > 
> > 
> > It turns out that the init of the tb cache is now moved to before ram_size
> > gets it's default value. And because the default tb cache size, if no
> > -tb-size option is given, is based on the ram_size it is now zero..
> 
> Yes, already stumbled over that regression myself. See
> http://thread.gmane.org/gmane.comp.emulators.qemu/113547

OK, thanks. I've pushed your fix.

Cheers