Problem with drm-radeon-testing commit - drm/ttm: Avoid conflicting reserve_memtype during ttm_tt_set_page_caching.

Problem with drm-radeon-testing commit - drm/ttm: Avoid conflicting reserve_memtype during ttm_tt_set_page_caching.

[Andy Furniss]

Card is an AGP RV670. libdrm,mesa,ddx and xorg are current gits.

After the recent batch of commits to drm-radeon-testing I have a perf 
problem.

It doesn't show immediately, but after running a timedemo on nexuiz 
followed by openarena, further timedemos on either will have very poor 
performance - 4 fps and gears may be as low as 100fps.

Nothing in dmesg during bisect (though I saw a few errors when on head - 
I think they are a separate issue) and free looks OK.

Once this has happened 2D performance is also reduced a bit - menues on 
seamonkey slower than before and when displaying a page with a second 
counter, sometimes all rendering will glitch briefly every second - can 
be seen just with key repeat in an xterm.

Git bisect came up with -

db78e27de7e29a6db6be7caf607cf803d84094aa is first bad commit
commit db78e27de7e29a6db6be7caf607cf803d84094aa
Author: Francisco Jerez <currojerez@riseup.net>
Date:   Tue Jan 12 18:49:43 2010 +0100

     drm/ttm: Avoid conflicting reserve_memtype during 
ttm_tt_set_page_caching.

     Fixes errors like:
     > reserve_ram_pages_type failed 0x15b7a000-0x15b7b000, track 0x8, 
req 0x10
     when a BO is moved between WC and UC areas.


------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
--

Re: Problem with drm-radeon-testing commit - drm/ttm: Avoid conflicting reserve_memtype during ttm_tt_set_page_caching.

[Andy Furniss]

Andy Furniss wrote:

> db78e27de7e29a6db6be7caf607cf803d84094aa is first bad commit

I have confirmed by running both todays and yesterdays heads with and 
without that commit reverted that it is the one causing the problem.

Another test I tried was with agpmode=-1. This does avoid the problem, 
but unfortunately it is useless as a workaround for me as game perf 
other than openarena has always been poor using pcie gart.

eg. Nexuiz timedemo demos/demo1

KMS agpmode=4

result 1910 frames 70.4830000 seconds 27.0987330 fps, one-second fps 
min/avg/max: 19 28 47

KMS agpmode=-1 (or UMS with either gart)

result 1910 frames 254.2660000 seconds 7.5118183 fps, one-second fps 
min/avg/max: 4 10 36

------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
--