Xorg 7.0-rc1 and EXA (radeon 9200)

Pierre Ossman drzeus-list at drzeus.cx
Mon Oct 31 02:16:32 PST 2005


Carsten Haitzler (The Rasterman) wrote:
> On Mon, 31 Oct 2005 10:30:53 +0100 Pierre Ossman <drzeus-list at drzeus.cx>
> babbled:
>
>   
>> Carsten Haitzler (The Rasterman) wrote:
>>     
>>> On Mon, 31 Oct 2005 08:33:55 +0100 Pierre Ossman <drzeus-list at drzeus.cx>
>>> babbled:
>>>
>>>   
>>>       
>>>> It would seem that something leaks memory though. If I keep mapping and
>>>> unmapping windows I gradually see the pixmap memory rising. When I hit
>>>> the size of my graphics card (64 MB) things deteriorate very quickly.
>>>> Who's the most likely culprit? X or xcompmgr? The latter seems more
>>>> likely since the pixmaps are freed once I kill xcompmgr.
>>>>     
>>>>         
>>> it could also just be vram fragmentation - but i woudl look at xcompmgr
>>> first. but it could be xcomposite itself too. have u tried xrestop to see
>>> who is allocating all the pixmaps? (or who xresources thinks they belong
>>> to) ?
>>>
>>>   
>>>       
>> xrestop just says unknown:
>>
>> 4600000 0 1 0 49 850 42342K 19K 42362K ? <unknown>
>>     
>
> ok - then this is likely xcomosite based allocation. possibly xcompmgr. w'ed
> need to modify xcompmgr to make a window with a name/clas or a title for
> xrestop to display useful info if it is xcompmgr (no need to show the window -
> simply create it and set properties).
>
>   

Done. It is xcompmgr that is allocating the pixmaps:

4e00000 1 1 0 11 342 27225K 8K 27233K ? xcompmgr

>> But when it starts to go downhill xcompmgr start emitting errors:
>>
>> error 3 request 20 minor 0 serial 67186
>> error 3 request 2 minor 0 serial 67199
>> error 3 request 20 minor 0 serial 67200
>> error 3 request 20 minor 0 serial 68205
>> error 3 request 15 minor 0 serial 68206
>> error 3 request 2 minor 0 serial 68207
>> error 3 request 20 minor 0 serial 68208
>> error 3 request 2 minor 0 serial 68221
>> error 3 request 20 minor 0 serial 68222
>>     
>
> interesting... hmmm. it could be xcomposite, a leak in it or x, or xcompmgr or
> maybe somewhere else, but triggered by use of xcomposite + xcompmgr - not sure
> though. xcomposite + xdamage + xrender and friends is still "immature" and
> needs testing and settling and fixing in some ways :)
>
>   

I'm no X hacker so I'm afraid I can't be much help there. I am willing
to test any patches sent my way though. :)

Rgds
Pierre




More information about the xorg mailing list