[PATCH 0/8] optimize the firmware loading process

[Ferruh Yigit]

On 1/23/2024 11:27 AM, Chaoyong He wrote:
>>>> -----Original Message-----
>>>> From: Ferruh Yigit <ferruh.yigit at amd.com>
>>>> Sent: Monday, January 22, 2024 11:09 PM
>>>> To: Chaoyong He <chaoyong.he at corigine.com>; dev at dpdk.org
>>>> Cc: oss-drivers <oss-drivers at corigine.com>
>>>> Subject: Re: [PATCH 0/8] optimize the firmware loading process
>>>>
>>>> On 1/15/2024 2:54 AM, Chaoyong He wrote:
>>>>> This patch series aims to speedup the DPDK application start by
>>>>> optimize the firmware loading process in sereval places.
>>>>> We also simplify the port name in multiple PF firmware case to make
>>>>> the customer happy.
>>>>>
>>>>
>>>> <...>
>>>>
>>>>>   net/nfp: add the elf module
>>>>>   net/nfp: reload the firmware only when firmware changed
>>>>
>>>> Above commit adds elf parser capability and second one loads firmware
>>>> when build time is different.
>>>>
>>>> I can see this is an optimization effort, to understand FW status
>>>> before loading FW, but relying on build time seems fragile. Does it
>>>> help to add a new section to store version information and evaluate based
>> on this information?
>>>>
>>>
>>> We have a branch of firmware (several app type combined with
>>> NFD3/NFDk) with the same version information(monthly publish), so the
>> version information can't help us, because we can't distinguish among them.
>>>
>>> But the build time is different for every firmware, and that's the reason we
>> choose it.
>>>
>>
>> If version is same although FW itself is different, isn't this problem on its own?
>> Perhaps an additional field is required in version syntax.
> 
> Actually, it's just the role the build time which embed in the firmware plays, which is unique for every firmware, and which can't be modified once the firmware was published.
> 

I see it is already in the elf binary but relying on build time of a FW
to decide to load it or not looks a weak method to me, and fragile as
stated before.

> What you said also has its mean, but in practice we can't accept(at least can't do it immediately), which needs to discuss with firmware team.
> 

As it is an optimization I assume it is not urgent, so would you mind
discussing the issue with the FW team, perhaps it can lead to a better
solution, we can proceed after that.

Meanwhile I will continue with remaining patches, excluding these two
patches.

> If you insist that we should change the design, maybe we can just kick off two commits, and upstream the other commits?
> - net/nfp: add the elf module
> - net/nfp: reload the firmware only when firmware changed