Re: [PATCH V2 2/3] dmaselftest: add memcpy selftest support functions

[Sinan Kaya <okaya@codeaurora.org>]

On 11/3/2015 11:08 AM, Vinod Koul wrote:
> On Tue, Nov 03, 2015 at 10:22:25AM +0200, Andy Shevchenko wrote:
>> On Tue, Nov 3, 2015 at 9:44 AM, Dan Williams <dan.j.williams@intel.com> wrote:
>>> On Mon, Nov 2, 2015 at 10:30 PM, Vinod Koul <vinod.koul@intel.com> wrote:
>>>> On Mon, Nov 02, 2015 at 11:18:37PM -0500, Sinan Kaya wrote:
>>>>> On 11/2/2015 11:15 PM, Vinod Koul wrote:
>>>>>> On Mon, Nov 02, 2015 at 01:07:38AM -0500, Sinan Kaya wrote:
>>
>>
>>>>> This one; on the other hand, is selftest to verify hardware is
>>>>> working as expected during power up.
>>
>> I prefer to have such nice case by run time parameter (let's say
>> common to all DMA Engine drivers)
>>
>>>> We can have common code which is used for dmatest as well as selftest. I do
>>>> not want to see same code duplicated..
>>
>> First thought was to merge this to dmatest, however, some DMA
>> controllers doesn't have a memcpy capability.
>
> The tests should be based on capablity of memcpy
>
>> How would we test them?

Here is what I proposed.

- a common file that gets compiled into a module that wants to use 
self-test with a public API. It can be called from driver's probe routine.
- the test is independent of my implementation. It uses dmaengine API 
and should be portable to most drivers.
- there *are* still drivers in the kernel that has selftest code 
embedded inside them. I followed the design pattern from other drivers 
thinking this must have been a good idea and it paid off for me.

As far as I understand, there is interest in doing more than this and 
reusing the dmatest code for code duplication.

Facts:
- Dmatest can be actually configured to run during boot.
- Nobody besides the dma driver developer uses dmatest. This leaves 
holes for regressions that are really hard to debug due to interaction 
with the rest of the system.
- Dmatest doesn't exist in most distribution kernels.

If we want to do something else, I need clear directions. I can remove 
the self test code completely from my driver. But, I need an equivalent 
functionality.

 >
 > That part is tricky, you need to do so thru clients, spi/audio/serial etc
 >

My selftest code actually attaches to all slave devices and issues a 
memcpy command and then detaches from the slave devices.

-- 
Sinan Kaya
Qualcomm Technologies, Inc. on behalf of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a 
Linux Foundation Collaborative Project