David Dawes
dawes****@XFree*****
2003年 8月 15日 (金) 21:23:21 JST
On Thu, Aug 14, 2003 at 10:23:53PM -0400, David Dawes wrote:
>On Thu, Aug 14, 2003 at 09:55:37PM -0400, David Dawes wrote:
>>On Thu, Aug 14, 2003 at 04:17:41PM -0700, Mark Vojkovich wrote:
>>>On Thu, 14 Aug 2003, David Dawes wrote:
>>>
>>>> >
>>>> > XAACacheTile() should have never been called if no slots existed.
>>>> >MaxCacheableTileWidth and MaxCacheableTileHeight should have been
>>>> >consulted at some point.
>>>>
>>>> OK, we need to see what those values are at this point. XAAPaintWindow(),
>>>> where the crash is apparently happening, does have a test for them, so
>>>> maybe there's getting set incorrectly somewhere.
>>>>
>>>> if(infoRec->UsingPixmapCache &&
>>>> infoRec->FillCacheBltRects && !NoCache &&
>>>> ((what == PW_BORDER) ||
>>>> (pPix->drawable.height != pWin->drawable.height) ||
>>>> (pPix->drawable.width != pWin->drawable.width)) &&
>>>> (pPix->drawable.height <= infoRec->MaxCacheableTileHeight) &&
>>>> (pPix->drawable.width <= infoRec->MaxCacheableTileWidth)) {
>>>>
>>>> XAACacheInfoPtr pCache =
>>>> (*infoRec->CacheTile)(infoRec->pScrn, pPix);
>>>> (*infoRec->FillCacheBltRects)(infoRec->pScrn, GXcopy, ~0,
>>>> nBox, pBox, xorg, yorg, pCache);
>>>> return;
>>>> }
>>>>
>>>> This check assumes that if there are no 256x256 slots there won't
>>>> be any 512x512 slots. I'm not 100% sure if it can ever happen that
>>>> there are 512x512 slots but no 256x256 slots. If the debugging
>>>> could be extended to print both infoRec->MaxCacheableTileHeight
>>>> and Num512, it should clarify if this is the problem or not.
>>>
>>> It's expected that if there are 512x512 slots there will be
>>>smaller slots. The logic that sets up the slots in XAAInitPixmapCache
>>>is supposed to ensure that.
>>
>>I played around a little with some unusual cases. If the memory is very
>>fragmented, I can find a case where there are 4 512x512 slots and no
>>256x256 slots. The trick here is to have enough boxes for 4 512 slots
>>and lots of 128 slots, but none for any 256 slots to get allocated in
>>the first round. If there are enough 128x128 slots, the target for
>>512x512 slots will remain at 4 and none of them get split into 256x256
>>slots.
>>
>>A case that gives this result is 4 512x512 boxes and 70 250x250 boxes:
>>
>>128: 32, 256: 0, 512: 4, partial: 288
>>max cacheable: 512x512
>
>Here's a patch that should catch this type of case. Let us know if this
>is enough to fix the crash.
>
>David
>Index: xaaPCache.c
>===================================================================
>RCS file: /home/x-cvs/xc/programs/Xserver/hw/xfree86/xaa/xaaPCache.c,v
>retrieving revision 1.30
>diff -u -r1.30 xaaPCache.c
>--- xaaPCache.c 25 Sep 2000 23:56:14 -0000 1.30
>+++ xaaPCache.c 15 Aug 2003 02:18:24 -0000
>@@ -763,6 +763,7 @@
> 1) Don't take up more than half the memory.
> 2) Don't bother if you can't get at least four.
> 3) Don't make more than MAX_512.
>+ 4) Don't have any of there are no 256x256s.
>
> 256x256 -
> 1) Don't take up more than a quarter of the memory enless there
>@@ -785,6 +786,13 @@
> if(!Target512) Target256 = ntotal >> 3;
> else Target256 = ntotal >> 4;
> if(Target256 < 4) Target256 = 0;
>+
>+ if(Num512 && !Num256) {
>+ while(Num512 && Num256 < Target256) {
>+ SubdivideList(&List512, &List256);
>+ Num256 += 4; Num512--;
>+ }
>+ }
>
> if(!Num512) { /* no room */
> } else if((Num512 < 4) || (!Target512)) {
It occurred to me this morning that 'if(Num512 && !Num256)' needs to
be 'if(Num512 && Num256 < 4)'. The case that was reported had Num256
initially 3, not 0.
David
--
David Dawes
Founder/committer/developer The XFree86 Project
www.XFree86.org/~dawes