Re: [Qemu-devel] sh : performance problem

[Lionel Landwerlin <lionel.landwerlin@openwide.fr>]

Le lundi 02 mars 2009 à 01:10 +0900, Shin-ichiro KAWASAKI a écrit :
> Lionel Landwerlin wrote:
> > Le dimanche 01 mars 2009 à 15:05 +0900, Shin-ichiro KAWASAKI a écrit :
> >> Hi, Lionel.
> >> Thank you for the response!
> >>
> >> Lionel Landwerlin wrote:
> >>> Le vendredi 27 février 2009 à 01:28 +0900, Shin-ichiro KAWASAKI a
> >>> écrit :
> >>>
> >>> Hi Shin-ichiro,
> >>>
> >>> I'm interested by your patch, but it does not apply on top of the r6215
> >>> (qemu svn) + qemu-sh patches
> >>> ( http://git.kernel.org/?p=virt/qemu/lethal/qemu-sh.git;a=summary) 
> >>>
> >>> Should I try on the last qemu svn ?
> >> Yes.  When I send patch to qemu-devel ml, I make diff against
> >> qemu main repository.  Could you try on it?
> > 
> > Yes, I will.
> > 
> > Do you know when the qemu-sh patches will be integrated to qemu ?
> 
> Nope.
> 
> Sometimes the main developers afford some time for sh kindly, then
> some of the patches in the qemu-sh staging repository are already
> integrated.  But some patches are left. I guess they are busy on
> other CPU archs.
> 
> Especially, it's not a good time to apply patches which is not so
> serious, because version cutting work is going now.
> 
> Regards,
> Shin-ichiro KAWASAKI
> 
> 

There is also one thing that is annoying me.

With the algorithm you proposed in your patch, you're not able anymore
to raise the MMU_ITLB_MULTIPLE and MMU_DTLB_MULTIPLE exceptions. This
makes the MMU emulation different from the real processor (even if
thoses cases aren't even handled in Linux and probably others OS).

For example you can probably not map the same physical area with
differents virtual page sizes.

-- 
Lionel Landwerlin <lionel.landwerlin@openwide.fr>