[01:01] *** yang has quit IRC (Ping timeout: 276 seconds)
[16:22] *** prabhakarlad has left 
[21:25] <ndufresne> javier__: good news, got the decoder to work, thansk for the patch, note that in 1.8 we added ability to negotaited the output format, as some decoder can output multiple formats
[21:26] <ndufresne> javier__: it fails now with mfc decoder because when we S_FMT(CAPTURE) before we do STREAMON(CAPTURE), the driver wrongly returns EBUSY
[21:27] <ndufresne> right now it fails and stop in gstreamer, I'll workaround this by making this a warning instead, but I'll probably propose you a fix, as this is wrong doing of the driver, EBUS is reserved to when we are streaming and/or when buffers have been allocated, in other case, it should just fill the format structure with what is supported
[21:30] <javier__> ndufresne: you mean't s/output format/capture format right?
[21:30] <javier__> since I see that format for the output device is being negotiated now
[21:30] <javier__> *meant
[21:32] <ndufresne> yeah S_FMT(CAPTURE)
[21:32] <ndufresne> Which is the decoder output right ;-P
[21:32] <javier__> ndufresne: indeed
[21:32] <javier__> I always get confused that the output format is the input to the decoder and the decoder output is the capture
[21:33] <ndufresne> same
[21:33] <ndufresne> I re-read myself after you asked
[21:34] <ndufresne> but yeah, in V4L2 it's rare you are allowed to return an error when you look at the spec attentively, most of the time you should be modifying the structure to what you support instead
[21:34] <ndufresne> javier__: interestingly, the decoder on Exynos 5 is supposed to let you pick an output format, but the driver requires you to pick that format before you start the OUTPUT side apparently
[21:34] <ndufresne> anyway, next week
[21:35] <ndufresne> have a great weekend !
[21:35] <ndufresne> I'm glad mainline kernel work without much work
[21:35] <javier__> ndufresne: thanks for testing!
[21:36] <javier__> ndufresne: I'll resend Ayaka's patch then and cc you
[21:36] <ndufresne> javier__: note that I could get the m2m fimc module to load though, need to figure-out, on Exynos 4412, it's rather important due to the fact the decoder output tiles
[21:36] <ndufresne> sorry "I could not"
[21:36] <ndufresne> could be a regression, or a DT issue, not sure
[21:36] <javier__> ndufresne: there isn't a fimc module on Exynos5 so I can't test it... that's what indeed an issue to test the mfc test app
[21:36] <javier__> since only works on exynos4
[21:37] <ndufresne> yeah, Kamil test was for Ex4 indeed
[21:37] <javier__> ndufresne: so you only needed "[media] s5p-mfc: don't close instance after free OUTPUT buffers" on top of latest mainline then
[21:37] <javier__> and tested just with CMA but with IOMMU disabled
[21:37] <ndufresne> I also wrote a patch "ARM: dts: exynos4412-odroidu3: enable MFC device"
[21:38] <javier__> ndufresne: right, it would be great if you post that one
[21:38] <ndufresne> javier__: the one in your branch is creating the wrong names for the reserved cma, and then it's reverted in another patch
[21:38] <ndufresne> I wonder if I could just move that code into the parent DT, and clear that up from all the other
[21:38] <ndufresne> all 4412 have the mfc right ?
[21:39] <javier__> ndufresne: yeah, Marek's series revert those but I wonder if that could be done since it will break DT backward compatibility
[21:39] <ndufresne> at the same time, I wonder if some vendor need not to do that for their use case
[21:39] <ndufresne> or if you setup the HW differently, it's simple/small enough to be put in each board dt tbh
[21:40] <ndufresne> and with iommu, it would be removed of course
[21:40] <ndufresne> without using it, I just turned that on, and it just crashed at start
[21:41] <ndufresne> anyway, will post that next week
[21:41] <javier__> ndufresne: Ok, so as you guessed IOMMU in Exynos4 is not working properly
[21:41] <javier__> ndufresne: sure, no rush. Thanks!
[21:41] <ndufresne> and fix the regression in gst
[21:42] <ndufresne> javier__: will it ever work in fact ? I never seen anyone claiming he managed to get that working, outside Exynos 5
[21:42] <ndufresne> and dev have considerably slow down on Ex 4
[21:43] <javier__> ndufresne: yeah, probably never did
[21:44] <ndufresne> so yeah, enabling that DT settings make sense
[21:44] <ndufresne> is the iommu stuff in you branch out-compilable ?
[21:45] <javier__> ndufresne: I believe so, I didn't test it with IOMMU disabled since I used a big video
[21:46] <javier__> ndufresne: what's your CMA conf btw?
[21:47] <javier__> I will give a try to video decoding later with only CMA and no IOMMU
[21:48] <ndufresne> javier__: for the now the same as tiny4412, http://paste.fedoraproject.org/363531/46257126/
[21:48] <ndufresne> but I remember having different settings to achieve at least two time 1080p, you can't have 4 instance 1080p with that memory size, that I am pretty sure
[21:48] <javier__> ndufresne: no, I meant the CONFIG_CMA_*
[21:48] <javier__> or is just the default in exynos_defconfig?
[21:49] <ndufresne> sorry http://paste.fedoraproject.org/363532/71357146
[21:49] <ndufresne> (it's the odroid default)
[21:50] <javier__> right, just checked in arch/arm/configs/exynos_defconfig
[21:50] <javier__> ndufresne: so for now I'll wait until my s5p-mfc module removal fixes and Ayaka's land and then I'll try to re-send Marek's IOMMU patches
[21:50] <ndufresne> ok, got to go
[21:50] <javier__> ndufresne: Ok, have a nice weekend!
[21:52] *** awalls1 has left