Re: Chromebook snow issues

Re: Chromebook snow issues

[Sylwester Nawrocki]

Adding linux-samsung-soc mailing list at Cc.

On 29/09/15 16:01, Mauro Carvalho Chehab wrote:
> Em Tue, 29 Sep 2015 15:37:51 +0200
> Sylwester Nawrocki <s.nawrocki@samsung.com> escreveu:
> 
>> Hi Mauro,
>>
>> On 29/09/15 14:43, Mauro Carvalho Chehab wrote:
>>> Hi Sylwester,
>>>
>>> I'm c/c Javier. Javier is working with us at the Open Souce Group and he
>>> has been working on making sure that Chromebooks are supported with vanilla
>>> upstream Kernels.
>>>
>>> That's said, while testing the latest upstream Kernel (4.3-rc3) on a
>>> Chromebook snow, I noticed this error message:
>>>
>>> [  211.116975] s5p_mfc_alloc_memdevs:1045: Failed to declare coherent memory for
>>>                MFC device
>>> [  211.117039] s5p-mfc: probe of 11000000.codec failed with error -12
>>>
>>> Do you have any idea about how to fix? Btw, do I need any firmware to test
>>> the MFC driver there?
>>
>> You can get v4.1 based kernel with MFC working (e.g. on Odroid XU3) here:
>>
>> https://USER@review.tizen.org/gerrit/p/platform/kernel/linux-exynos.git
>> ssh://USER@review.tizen.org:29418/platform/kernel/linux-exynos.git
>> branch: tizen
>>
>> I'm adding at Cc Marek who knows more details. AFAIU you may need patches
>> like these on top of current mainline kernel:
>>
>> https://review.tizen.org/gerrit/gitweb?p=platform/kernel/linux-exynos.git;a=commitdiff;h=dcbf130658cdd1731b8a219ecedd867af0b2670e
>>
>> https://review.tizen.org/gerrit/gitweb?p=platform/kernel/linux-exynos.git;a=commitdiff;h=b82fd3408fd58476da79996d4a3bcbbb72174961
>>
>> And Exynos IOMMU needs to be enabled in kernel config.
> 
> If this is a requirement, we should enforce it at s5p-mfc Kconfig entry,
> like:
> 
> 	depends on EXYNOS_IOMMU if !COMPILE_TEST

In general s5p-mfc doesn't depend on IOMMU, some SoC revisions don't
even have SYSMMU. Thus we can't add such constraint.

> Just enabled Exynos IOMMU here to see what would happen:
> 
> [   19.461839] ------------[ cut here ]------------
> [   19.464264] kernel BUG at net/core/dev.c:6513!
> [   19.466648] Internal error: Oops - BUG: 0 [#2] PREEMPT SMP ARM
> [   19.469039] Modules linked in: mwifiex_sdio mwifiex exynos_gsc(+) v4l2_mem2mem s5p_mfc(+) videobuf2_dma_contig videobuf2_memops videobuf2_core v4l2_common videodev media
> [   19.474158] CPU: 1 PID: 1390 Comm: kworker/1:2 Tainted: G      D         4.3.0-rc3-00492-g397ead8befa9 #10
> [   19.476753] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
> [   19.479326] Workqueue: events request_firmware_work_func
> [   19.481887] task: edb06d00 ti: ee18a000 task.ti: ee18a000
> [   19.484445] PC is at register_netdevice+0x23c/0x41c
> [   19.487013] LR is at register_netdevice+0x28/0x41c
> [   19.489572] pc : [<c047a184>]    lr : [<c0479f70>]    psr: 60010013
> [   19.489572] sp : ee18be00  ip : 0000000e  fp : ed4629a0
> [   19.494692] r10: bf102940  r9 : 00000000  r8 : ed462000
> [   19.497218] r7 : 00000002  r6 : ed462000  r5 : edbee000  r4 : ed738000
> [   19.499741] r3 : 00000000  r2 : 00000146  r1 : 000003e8  r0 : 00000001
> [   19.502253] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
> [   19.504768] Control: 10c5387d  Table: 6d5bc06a  DAC: 00000051
> [   19.507247] Process kworker/1:2 (pid: 1390, stack limit = 0xee18a210)
> [   19.509717] Stack: (0xee18be00 to 0xee18c000)
> [   19.512150] be00: 0000ffff 00000001 00000000 00000000 ed4629a0 ed738000 edbee000 ed462000
> [   19.514615] be20: 00000002 ed7392fc ed739000 bf0fab4c 00000004 00000001 edbef000 ffffffff
> [   19.517072] be40: c086697c edbee000 bf10b400 edbef000 bf12d784 00000000 00000000 ed9a4800
> [   19.519518] be60: ee7b58c0 bf0d95b0 00000000 00000000 00000000 00000000 f1694000 0006fd7c
> [   19.521959] be80: 00000000 ec118e00 ec118dc0 c02f711c f1694000 0006fd7c ed43a390 ec118d40
> [   19.524396] bea0: 00000003 ee18bee4 edbbf208 c0834674 ec118dc0 c08ba8bc 00000002 ec118d80
> [   19.526827] bec0: ed9a4800 ec118d80
> 
> Unfortunately, the serial console didn't got everything. It _seems_ that the
> bug happens during a mwifiex kthread, when it calls mwifiex_add_virtual_intf,
> but even at the device's screen the message is truncated.

Unfortunately this doesn't ring any bells here, we don't work with Chromebooks
in general and I'm not sure how to address this issue. Maybe someone subscribed
to the mailing list can help.

> I didn't apply those two patches you pointed from Tizen.org. Not sure if
> this is related or not.

It's possible the issue is caused by some other device (driver) that also
has a sysmmu assigned to it, like display controller, etc. I'm not sure.

>> Required firmware is already available the linux-firmware git repository,
>> the driver will select proper version if the firmware files (s5p-mfc*.fw)
>> are copied to /lib/firmware.
> 
> Ah, OK!
>>
>>> Thanks!
>>> Mauro

-- 
Regards,
Sylwester

Re: Chromebook snow issues

[Mauro Carvalho Chehab]

Em Tue, 29 Sep 2015 16:32:58 +0200
Sylwester Nawrocki <s.nawrocki@samsung.com> escreveu:

> Adding linux-samsung-soc mailing list at Cc.


Never mind. My mistake: I forgot to reinstall the modules (mwifiex and mfc
drivers were compiled as such).

It is working.

I guess the only remaining thing is to add the dependency of EXYNOS_IOMMU
to the MFC driver.

Regards,
Mauro

> 
> On 29/09/15 16:01, Mauro Carvalho Chehab wrote:
> > Em Tue, 29 Sep 2015 15:37:51 +0200
> > Sylwester Nawrocki <s.nawrocki@samsung.com> escreveu:
> > 
> >> Hi Mauro,
> >>
> >> On 29/09/15 14:43, Mauro Carvalho Chehab wrote:
> >>> Hi Sylwester,
> >>>
> >>> I'm c/c Javier. Javier is working with us at the Open Souce Group and he
> >>> has been working on making sure that Chromebooks are supported with vanilla
> >>> upstream Kernels.
> >>>
> >>> That's said, while testing the latest upstream Kernel (4.3-rc3) on a
> >>> Chromebook snow, I noticed this error message:
> >>>
> >>> [  211.116975] s5p_mfc_alloc_memdevs:1045: Failed to declare coherent memory for
> >>>                MFC device
> >>> [  211.117039] s5p-mfc: probe of 11000000.codec failed with error -12
> >>>
> >>> Do you have any idea about how to fix? Btw, do I need any firmware to test
> >>> the MFC driver there?
> >>
> >> You can get v4.1 based kernel with MFC working (e.g. on Odroid XU3) here:
> >>
> >> https://USER@review.tizen.org/gerrit/p/platform/kernel/linux-exynos.git
> >> ssh://USER@review.tizen.org:29418/platform/kernel/linux-exynos.git
> >> branch: tizen
> >>
> >> I'm adding at Cc Marek who knows more details. AFAIU you may need patches
> >> like these on top of current mainline kernel:
> >>
> >> https://review.tizen.org/gerrit/gitweb?p=platform/kernel/linux-exynos.git;a=commitdiff;h=dcbf130658cdd1731b8a219ecedd867af0b2670e
> >>
> >> https://review.tizen.org/gerrit/gitweb?p=platform/kernel/linux-exynos.git;a=commitdiff;h=b82fd3408fd58476da79996d4a3bcbbb72174961
> >>
> >> And Exynos IOMMU needs to be enabled in kernel config.
> > 
> > If this is a requirement, we should enforce it at s5p-mfc Kconfig entry,
> > like:
> > 
> > 	depends on EXYNOS_IOMMU if !COMPILE_TEST
> 
> In general s5p-mfc doesn't depend on IOMMU, some SoC revisions don't
> even have SYSMMU. Thus we can't add such constraint.
> 
> > Just enabled Exynos IOMMU here to see what would happen:
> > 
> > [   19.461839] ------------[ cut here ]------------
> > [   19.464264] kernel BUG at net/core/dev.c:6513!
> > [   19.466648] Internal error: Oops - BUG: 0 [#2] PREEMPT SMP ARM
> > [   19.469039] Modules linked in: mwifiex_sdio mwifiex exynos_gsc(+) v4l2_mem2mem s5p_mfc(+) videobuf2_dma_contig videobuf2_memops videobuf2_core v4l2_common videodev media
> > [   19.474158] CPU: 1 PID: 1390 Comm: kworker/1:2 Tainted: G      D         4.3.0-rc3-00492-g397ead8befa9 #10
> > [   19.476753] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
> > [   19.479326] Workqueue: events request_firmware_work_func
> > [   19.481887] task: edb06d00 ti: ee18a000 task.ti: ee18a000
> > [   19.484445] PC is at register_netdevice+0x23c/0x41c
> > [   19.487013] LR is at register_netdevice+0x28/0x41c
> > [   19.489572] pc : [<c047a184>]    lr : [<c0479f70>]    psr: 60010013
> > [   19.489572] sp : ee18be00  ip : 0000000e  fp : ed4629a0
> > [   19.494692] r10: bf102940  r9 : 00000000  r8 : ed462000
> > [   19.497218] r7 : 00000002  r6 : ed462000  r5 : edbee000  r4 : ed738000
> > [   19.499741] r3 : 00000000  r2 : 00000146  r1 : 000003e8  r0 : 00000001
> > [   19.502253] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
> > [   19.504768] Control: 10c5387d  Table: 6d5bc06a  DAC: 00000051
> > [   19.507247] Process kworker/1:2 (pid: 1390, stack limit = 0xee18a210)
> > [   19.509717] Stack: (0xee18be00 to 0xee18c000)
> > [   19.512150] be00: 0000ffff 00000001 00000000 00000000 ed4629a0 ed738000 edbee000 ed462000
> > [   19.514615] be20: 00000002 ed7392fc ed739000 bf0fab4c 00000004 00000001 edbef000 ffffffff
> > [   19.517072] be40: c086697c edbee000 bf10b400 edbef000 bf12d784 00000000 00000000 ed9a4800
> > [   19.519518] be60: ee7b58c0 bf0d95b0 00000000 00000000 00000000 00000000 f1694000 0006fd7c
> > [   19.521959] be80: 00000000 ec118e00 ec118dc0 c02f711c f1694000 0006fd7c ed43a390 ec118d40
> > [   19.524396] bea0: 00000003 ee18bee4 edbbf208 c0834674 ec118dc0 c08ba8bc 00000002 ec118d80
> > [   19.526827] bec0: ed9a4800 ec118d80
> > 
> > Unfortunately, the serial console didn't got everything. It _seems_ that the
> > bug happens during a mwifiex kthread, when it calls mwifiex_add_virtual_intf,
> > but even at the device's screen the message is truncated.
> 
> Unfortunately this doesn't ring any bells here, we don't work with Chromebooks
> in general and I'm not sure how to address this issue. Maybe someone subscribed
> to the mailing list can help.
> 
> > I didn't apply those two patches you pointed from Tizen.org. Not sure if
> > this is related or not.
> 
> It's possible the issue is caused by some other device (driver) that also
> has a sysmmu assigned to it, like display controller, etc. I'm not sure.
> 
> >> Required firmware is already available the linux-firmware git repository,
> >> the driver will select proper version if the firmware files (s5p-mfc*.fw)
> >> are copied to /lib/firmware.
> > 
> > Ah, OK!
> >>
> >>> Thanks!
> >>> Mauro
> 

Re: Chromebook snow issues

[Sylwester Nawrocki]

On 29/09/15 16:42, Mauro Carvalho Chehab wrote:
> Em Tue, 29 Sep 2015 16:32:58 +0200
> Sylwester Nawrocki <s.nawrocki@samsung.com> escreveu:
> 
>> > Adding linux-samsung-soc mailing list at Cc.
> 
> Never mind. My mistake: I forgot to reinstall the modules (mwifiex and mfc
> drivers were compiled as such).
> 
> It is working.
> 
> I guess the only remaining thing is to add the dependency of EXYNOS_IOMMU
> to the MFC driver.

Sounds good. The patches I pointed out are only a workaround, in general
MFC should also work without IOMMU. As the driver handles also SoCs without
a SYSMMU hardware support adding such a Kconfig entry would not be correct.

-- 
Regards,
Sylwester