On 09/10/2018 11:21 PM, Laura Abbott wrote:
> On 09/10/2018 03:00 AM, Alexey Skidanov wrote:
>>
>>
>> On 09/10/2018 12:36 PM, Dan Carpenter wrote:
>>> On Sun, Sep 09, 2018 at 01:44:31AM +0300, Alexey Skidanov wrote:
>>>> The heap statistics have been removed and currently even basics
>>>> statistics
>>>> are missing.
>>>
>>> Remind me why did we remove them? What was the git hash?
>> 15c6098cfec5 ("staging: android: ion: Remove ion_handle and ion_client")
>> According to Laura, "The goal was to rip out the racy handle/buffer mess
>> and the associated debugfs there"
>
> Yes, the debugfs was removed with that commit because the underlying
> handle/client were being removed and the associated debugfs
> tended to race with tear down. There was also a question about what
> information should go in Ion vs. dma_buf since Ion is now basically
> a wrapper around dma_buf.
>
> I don't disagree that we want some kind of debugfs counters but
> I'd like to see some description of what problems we can solve
> with this that aren't also solved by the debugfs with dma_buf.
The dma_buf debugfs provides very detailed information about allocated
buffers. However the heap related statistics are missing. I would
suggest to start with at least heap watermark, in order to profile heap
memory usage.
In addition, it might be good to correlate dma_buf debugfs and ion
debugfs. For example, to get per heap buffers list.
>
> Thanks,
> Laura
Thanks,
Alexey
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
