Re: [PATCH 0/2] s5p-mfc fix for using reserved memory

[pankaj.dubey]

Hi Marek,

On Tuesday 20 December 2016 05:28 PM, Marek Szyprowski wrote:
> Hi Pankaj
> 
> 
> On 2016-12-16 12:48, Pankaj Dubey wrote:
>> It has been observed on ARM64 based Exynos SoC, if IOMMU is not enabled
>> and we try to use reserved memory for MFC, reqbufs fails with below
>> mentioned error
>> ---------------------------------------------------------------------------
>>
>> V4L2 Codec decoding example application
>> Kamil Debski <k.debski@xxxxxxxxxxx>
>> Copyright 2012 Samsung Electronics Co., Ltd.
>>
>> Opening MFC.
>> (mfc.c:mfc_open:58): MFC Info (/dev/video0): driver="s5p-mfc" \
>> bus_info="platform:12c30000.mfc0" card="s5p-mfc-dec" fd=0x4[
>> 42.339165] Remapping memory failed, error: -6
>>
>> MFC Open Success.
>> (main.c:main:711): Successfully opened all necessary files and devices
>> (mfc.c:mfc_dec_setup_output:103): Setup MFC decoding OUTPUT buffer \
>> size=4194304 (requested=4194304)
>> (mfc.c:mfc_dec_setup_output:120): Number of MFC OUTPUT buffers is 2 \
>> (requested 2)
>>
>> [App] Out buf phy : 0x00000000, virt : 0xffffffff
>> Output Length is = 0x300000
>> Error (mfc.c:mfc_dec_setup_output:145): Failed to MMAP MFC OUTPUT buffer
>> -------------------------------------------------------------------------
>> This is because the device requesting for memory is mfc0.left not the
>> parent mfc0.
>> Hence setting of alloc_devs need to be done only if IOMMU is enabled
>> and in that case both the left and right device is treated as mfc0 only.
>> Also we need to populate vb2_queue's dev pointer with mfc dev pointer.
> 
> I also got this issue but Your solution is imho not the proper approach.
> Too much hacking in the driver, while the issue is in the core. ARM64
> requires
> to call arch_setup_dma_ops() for each device that will be used for
> dma-mapping.
> So the issue is in drivers/of/of_reserved_mem.c - in
> of_reserved_mem_device_init_by_idx() function, which should ensure that
> arch_setup_dma_ops() is called also for the virtual devices for reserved
> memory.
> 
> s5p-mfc driver however still requires some patching for
> dma-mapping/iommu glue
> used on ARM64 architecture.
> 

Thanks for review and suggestion.
Our initial investigation pointed out issue due to change in vb2_queue
dev field by commit 2548fee63d9e ("[media] media/platform: convert
drivers to use the new vb2_queue dev field"), where only for mfc q->dev
pointer was not set.

But I think you are right, and of_reserved_mem_device_init_by_idx should
take care of setting dma_ops for devices which calls this function. So I
will be submitting new change soon to fix this issue after testing it.

Thanks,
Pankaj Dubey

>> Smitha T Murthy (2):
>>    media: s5p-mfc: convert drivers to use the new vb2_queue dev field
>>    media: s5p-mfc: fix MMAP of mfc buffer during reqbufs
>>
>>   drivers/media/platform/s5p-mfc/s5p_mfc.c     |  2 ++
>>   drivers/media/platform/s5p-mfc/s5p_mfc_dec.c | 17 ++++++++++-------
>>   drivers/media/platform/s5p-mfc/s5p_mfc_enc.c | 18 +++++++++++-------
>>   3 files changed, 23 insertions(+), 14 deletions(-)
>>
> 
> Best regards