[dpdk-dev] [PATCH v2] common/cnxk: support BPHY telemetry

[Ferruh Yigit]

On 11/2/2021 11:11 AM, Jerin Jacob wrote:
> On Tue, Nov 2, 2021 at 3:58 PM Ferruh Yigit <ferruh.yigit at intel.com> wrote:
>>
>> On 11/2/2021 4:32 AM, Jerin Jacob wrote:
>>> On Mon, Nov 1, 2021 at 8:32 PM Ferruh Yigit <ferruh.yigit at intel.com> wrote:
>>>>
>>>> On 10/22/2021 12:56 PM, Tomasz Duszynski wrote:
>>>>> Add initial support for baseband telemetry.
>>>>>
>>>>> Signed-off-by: Tomasz Duszynski<tduszynski at marvell.com>
>>>>> ---
>>>>> v2:
>>>>> - make bphy telemetry available only on platforms supporting baseband
>>>>> - use platform types where possible
>>>>> - remove unused header
>>>>>
>>>>>     drivers/common/cnxk/cnxk_telemetry_bphy.c | 52 +++++++++++++++++++++++
>>>>
>>>> Since the telemetry support is for 'raw/cnxk_bphy', why it is implemented
>>>> in common code, instead of driver?
>>>
>>> It can be raw/cnxk_bphy, thought of keeping it in common due to
>>> 1) To reuse it for another common code consumer
>>
>> Is it reusable, since the code is to get telemetry data from raw device?
> 
> Yes. I meant, common code is reused on another library that has
> telemetry support too.
> 
>>
>>> 2) roc_bphy_sso_pf_func_get() and roc_bphy_npa_pf_func_get() manged by
>>> common code. aka there is no reverse dependency on the raw driver framework
>>> in common code.
>>
>> If telemetry code moved to raw driver, dependency will be from driver to
>> common code, which is correct dependency order I think.
>>
>> It just looks wrong to have rawdev related telemetry function in the common
>> code, but I may be missing details, trying to understand.
> 
> Currently, we are doing like this:
> - implementing HW specific telemetry endpoints in driver/common/cnxk
> - ethdev/mempool/eventdev/cryptodev driver-specific telemetry endpoint
> in driver/.../cnxk.
> The criteria for deciding what is the second category is, if it uses
> any ethdev/mempool/eventdev/cryptodev/rawdev
> symbols or if telemetry-related raw dev driver implementation details.
> i.e common code does not have
> any raw driver dependency.
> 

Got it, so I am proceeding as it is.