/srv/irclogs.linaro.org/2011/10/10/#linaro-mm-sig.txt

benjiGibiris: ok07:23
ibirissumits: benjiG: can we agree on a blueprint/ or work items for the work benjiG will be doing?07:23
ibirissumits: benjiG: eg look at https://blueprints.launchpad.net/linaro-mm-sig/+spec/engr-mm-bo-sharing-2011.1007:24
benjiGmaybe my test could fit in "Implement basic test case set for generic dma_buf_XXX APIs to test buffer-sharing mechanism" work item07:25
ibirisbenjiG: ack - this is what I had in mind07:25
ibirisbenjiG: pls confirm this with sumits :-)07:26
ibirisbenjiG: sumits: also remember the release date for this month - 20Oct (that's next thursday)07:26
ibirissorry next week's thursday07:26
benjiGok07:27
ibirisbenjiG: for the tree that JesseBarker uses for the UMM release, the date is somewhat flexible but try to agree with sumits and get the work done by that date as a first target - JesseBarker will confirm if there is possibly more time07:28
benjiGibiris: sumits: JesseBarker: I will send you my current dev so you can evaluate it and said if it could be merge in UMM tree07:38
sumitsHi ibiris, benjiG, JesseBarker08:15
sumitsbenjiG: I would review the patch you sent;08:16
sumitshowever, we would need to keep the userspace somewhere else. I would create a base repository in git.linaro.org and keep it there.08:17
sumitsbase repo for dma-buf-test userspace side. kernel space we can keep in tree from JesseBarker08:17
benjiGsumits: yes you can split the test like that08:18
sumitsalso, yes, I am ok about the workitem that you mentioned.08:18
ibirissumits: afaik JesseBarker had acknowledged that his tree could contain also userspace componentd eg xorg tree and any other relevant user-space bits (libdrm, dri2proto, and xserver).08:19
ibirissumits: even though now it contains only the kernel08:19
sumitsibiris: ok, I will check with him; I think he won't be able to have one git with both kernel and userspace components. He would need separate git trees for each.08:20
ibirissumits: benjiG: actually I thought this was meant to be like that (containing both the kernel code and code which enables tests) in 1 place08:21
sumitsibiris: hmmm.. 1 place can still be 'set of git trees from JesseBarker'08:22
sumitsrather than, '1 git tree containing everything'?08:23
sumitsbut we should confirm it. anyways till then, I will make the test case available on my git.linaro.org, so JesseBarker can pull it if needed.08:23
sumitsthis will also give benjiG and me a base to start building upon.08:24
benjiGsumits: it is ok for me08:25
ibirisbenjiG: sumits: ack - also please update the blueprint to put benjiG's name against a work item to reflect this work08:25
sumitsibiris: ack. doing it now.08:26
sumitsibiris: the assignment should be with the irc nick, or the launchpad name?08:28
ibirissumits: launchpad name :-)08:28
ibirissumits: https://launchpad.net/~benjamin-gaignard08:29
benjiGsumits: you don't have the patch in my mail ?08:43
benjiGsumits: you don't have the patch in my mail ?08:54
sumitsbenjiG: no, I cant see any patch attached?08:56
sumitsibiris: do you see the patch as attachment?08:56
benjiGsumits: lets me re-try ...08:56
sumitssure08:57
benjiGmail send (again)08:57
benjiGit is a .patch file ....08:59
benjiGit is seems it comes from gmail ....09:01
sumits:( I guess something wrong with the linaro mail server? could you please send it to sumit.semwal@ti.com?09:01
ibirissumits: benjiG - use https://pastebin.linaro.org/ instead09:01
ibiristhen pass on the pastebin link ;-)09:02
sumitsbenjiG: that should be good too :)09:02
benjiGhttps://pastebin.linaro.org/271/09:04
CosmicPenguinOkay, so playing with CMA17:20
CosmicPenguindma_alloc_contiguous is successful, returns a page17:20
CosmicPenguinthen NULL pointer derefernce in memzero from _dma_clear_buffer17:21
CosmicPenguinThat would imply that the CMA reserved region ended outside of kernel mapped space, right?17:21
robclarkif page_address() returns NULL, then it does sound a bit like highmem..17:22
robclarkerr, nonmem17:23
robclark(or at least no kernel mapping)17:23
robclarkare you using the API to declare a contiguous zone, or just using the default?17:24
CosmicPenguinJust using the default - this is very early during boot which makes me think that the coherent_init is the one that is failing17:24
CosmicPenguinI'm debugging with early_printk17:24
robclarkhmm17:24
robclarkthat stuff just worked for me..  but I guess add some printks to print out the pfn of the pages..17:26
CosmicPenguinYeah.  I need to learn a lot more about how this stuff actually works17:27
robclarkhmm, I wonder what happens if you make coherent_pool_size bigger than the default CMA pool size17:28
robclarkI guess in theory __alloc_from_contiguous() would just fail..17:29
CosmicPenguinYep, the memory is ending up in the HighMem zone19:11
robclarkyup, that would do it19:15
CosmicPenguinSet the limit on dma_contiguous_reserve to arm_lowmem_limit and it seemed to fix things19:52
CosmicPenguinDoes everybody else use CONFIG_ZONE_DMA?19:53
CosmicPenguinMaybe I hit a untested path19:53
robclarknope, apparently I am not using that20:02
CosmicPenguinMoving on to the next bug - for my first sizable allocation all the pages aren't isolated20:28
robclarkisolated as in there is still an aliased alternate mapping?20:30
robclarkoh, I guess you mean MIGRATE_ISOLATE stuff..20:31
robclarkeek20:31
CosmicPenguinIt wasn't a completely clean merge into our tree, so I'm still not sure if any problems I have are merge related or code related21:02
CosmicPenguinor platform related21:02
CosmicPenguinHmm.. the offending pages are marked as MIGRATE_CMA21:48
robclarksounds like maybe some of the merge got missed.. and something isn't handling MIGRATE_CMA case..21:58

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!