[Cbe-oss-dev] [PATCH 0:3] spusched: assorted spu scheduler fixes

[Kazunori Asayama]

Gerhard Stenzel wrote:
> cbe-oss-dev-bounces+stenzel=de.ibm.com at ozlabs.org wrote on 04/25/2008
> 11:43:21 AM:
> 
>> Jeremy Kerr wrote:
>>>> I couldn't apply the patch 1 to the master branch of Jeremy's tree.
>>>> Do I need other additional patches?
>>> I have an experimental tree which Luke is using as a base for these
>>> patches; I'll post a complete series soon.
>> I see.
>>
>> Will the problem with MFC_CNTL[Ds] and MFC_CNTL[Q]
>> (http://ozlabs.org/pipermail/cbe-oss-dev/2008-April/004605.html)
>> be fixed by it?
>>
>> FYI: the problem with MFC_CNTL[Q] can be easily reproduced by a test
>> in the libspe2 testsuite. Try:
>>
>>   make check-test_dma.elf TEST_SPES=7 (PS3)
>>   make check-test_dma.elf TEST_SPES=17 (Cell Blade)
>>
>> in the directory libspe2/tests/libspe2.mfc/.
> 
> it seems like it is fixed:

Did you apply the patch following?

  http://ozlabs.org/pipermail/cbe-oss-dev/2007-August/003139.html

It looks like that, without this patch, the SPUFS can restore invalid 
DMA queue if all outstanding DMAs are completed between save_mfc_cntl
and save_spu_status. Actually, without it, check-test_dma.elf on 
PS3 failed with the current Geoff's ps3-linux tree.

Or does Jeremy's experimental tree have any alternative fix?

> 
> [root at station libspe2.mfc]# make check-test_dma.elf TEST_SPES=17
> make -C /root/libspe2/tests/common
> ...
> test_dma.elf: RUNNING
> test_dma.elf: PASS

-- 
(ASAYAMA Kazunori
  (asayama at sm.sony.co.jp))
t