AM35x dma_alloc_coherent failed

Jemish Patel37

Other Parts Discussed in Thread: AM3505

Hi All,

We are using AM3505 processor based custom board. We are using kernel version 2.6.37 on that.

We are facing very strange issue currently. On our board when we use 2 Telit modems together (module used by modem : cdc-acm.ko) in that case video driver is showing below error.

[ 63.569976] vpfe-capture vpfe-capture: adjusted width = 160, height = 128, bpp = 2, bytesperline = 320, sizeimage = 40960

[ 63.996337] vpfe-capture vpfe-capture: dma_alloc_coherent size 40960 failed

If we remove any one Telit modem then issue is not happening.

We have also done few more experiments based on TI e2e posts.

1. Increase MAX_ORDER to 12.

2. Increase CONSISTENT_DMA_SIZE to SZ_8M instead of SZ_4M.

We don't get success in both of above experiments error is still there.

Any pointers or suggestion for this issue will be appreciated. We have shipment of units in this month end , it would be better if someone can help for this critical issue.

Thanks,

Jemish

over 9 years ago

0 Biser Gatchev-XID over 9 years ago

TI__Guru**** 393215 points

Hi,

I will forward this to the software team.

0 Yordan Kovachev over 9 years ago

TI__Guru**** 161600 points

Hi Jemish,

Jemish Patel37 said:
1. Increase MAX_ORDER to 12.

Have you tried to redefine this in the defconfig file using:
#define CONFIG_FORCE_MAX_ZONEORDER 12 => this will set maximum of 8MB

#define CONFIG_FORCE_MAX_ZONEORDER 13 => this will set maximum of 16MB

Also have a look at the following wiki:

https://developer.ridgerun.com/wiki/index.php/Capturing_JPEG_images_at_full_(allowed)_resolution_on_DM36x_platform#Configuring_the_SDK_to_capture_2176x1944_still_images

It deals with similar error in DM36x devices: vpfe-capture vpfe-capture: dma_alloc_coherent size 8462336 failed so it should be useful to you.

Hope this helps.

Best Regards,
Yordan

0 Jemish Patel37 over 9 years ago in reply to Yordan Kovachev

Intellectual 490 points

Hi Yordan/Biser,

Thanks for your quick response!

Yordan my issue seems to be different. In my case dma_alloc_coherent is not failing because it requires large memory which is not by default available. If you have seen my error log "dma_alloc_coherent size 40960 failed" which is 40K only. The example you suggested required around 8MB memory together which will not be available with default order 11 which give us maximum 4MB. Such issues can be resolved by increasing MAX_ORDER but my case seems to be different.

Do you have any other suggestions or pointers for this issue?

Thanks,
Jemish

0 Jemish Patel37 over 9 years ago in reply to Jemish Patel37

Intellectual 490 points

Hi All,

Few more observations on this issue.

1. In our board System RAM is 256MB.
2. We figured out that even when system is completely running free memory is 131.68MB available.
3. This indicates that we have enough free memory.
4. But dma_alloc_coherent call is failing it indicates that required contiguous physical memory is not available. As per my understanding it needs to be allocated boot time or in driver init/probe function so that it can not fail run time.

Can anyone please suggest or give direction for this issue?

Thanks,
Jemish

0 Yordan Kovachev over 9 years ago in reply to Jemish Patel37

TI__Guru**** 161600 points

Have you checked what's been done in newer kernels, in that regard? For example kernel 3.2, which is still using a board file, instead of dts.

Best Regards,
Yordan

0 Jemish Patel37 over 9 years ago in reply to Yordan Kovachev

Intellectual 490 points

Hi Yordan,

Would you please tell me what should I check in board file specifically for this kind of issue?

Thanks,
Jemish

0 Yordan Kovachev over 9 years ago in reply to Jemish Patel37

TI__Guru**** 161600 points

Hi Jemish,

Sorry, maybe I wasn't clear, I referred to kernel 3.2, because it is the latest TI release that uses board files, so it is the closest one to the version you use 2.6.37.

You should check the dma related drivers. As this post explains the memory is taken from somewhere in the physical address space that's granted to the Linux kernel via the u-boot bootargs parameter mem=##.

So see how is your uboot environment set and does it match your hw. Maybe try to increase the vram.

Also see if you have vpfe memory buffers defined in your code & increase them.

Best Regards,

Yordan

0 Jemish Patel37 over 9 years ago in reply to Yordan Kovachev

Intellectual 490 points

Hi Biser,

Right we have managed to solve it with some work around so we can close this thread as of now.

Thanks,
Jemish

Processors

Processors forum

AM35x dma_alloc_coherent failed