Re: [PATCH 00/10] in-kernel APIC v3 (kernel side)

["Gregory Haskins" <ghaskins-Et1tbQHTxzrQT0dZR+AlfA@public.gmane.org>]

>>> On Sun, May 13, 2007 at  8:38 AM, in message <464706BF.6000808-atKUWr5tajBWk0Htik3J/w@public.gmane.org>,
Avi Kivity <avi-atKUWr5tajBWk0Htik3J/w@public.gmane.org> wrote: 
> Gregory Haskins wrote:
>> The load- average on my system is about 1 while XP is idling.  qemu seems to 
> be mostly at "0%" but will bounce up to 1% on occasion.  Here is the output 
> of "top - b - p <qemu- pid>" over a few seconds:
>>
>> top -  09:17:45 up 16:58,  3 users,  load average: 1.02, 0.86, 0.42
>> Tasks:   1 total,   0 running,   1 sleeping,   0 stopped,   0 zombie
>> Cpu(s):  0.1%us,  0.2%sy,  0.0%ni, 99.5%id,  0.1%wa,  0.0%hi,  0.1%si,  
> 0.0%st
>> Mem:   3994704k total,  2018980k used,  1975724k free,    70996k buffers
>> Swap:  2104472k total,        0k used,  2104472k free,  1148284k cached
>>
>>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND         
>    
>> 10359 ghaskins  15   0  598m  83m  75m S    0  2.1   1:37.87 qemu- system- x86   
>  
>>   
> 
> A good test is to let Windows boot and idle itself, then compare the 
> process cpu time under the TIME+ column with model- 0 and model- 1.

Will do.  One that that I notice that I can't explain yet is as follows:

When I boot windows + level-1, the point at which windows is running in 16 bit real-mode in the very beginning (before the splashscreen comes up), we seem to take a very large number of exits for instruction emulation.  This ends up being a little storm of activity for about 1 second or so.  I am not really sure why this happens with the new code and not with the old.  It doesnt seem to hurt anything other than extra CPU used.  But its weird nonetheless. 

> 
> Since the vast majority of exits in the scenario are hitting the tpr, 
> I'd be unsurprised if the time if 50% lower or so.

Yeah, the new code essentially converts all those TPR exits to lightweight.  Nothing more, nothing less.  I could be crazy, but my perception is the GUI is *much* more responsive because of it, however.  Windows draw very fast and it actually seems usable.  Whereas trunk+ACPI always feels sluggish.  I don't know of a good benchmark to run to see if there really is an improvement, however.





-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/