[EXT] Re: [PATCH 2/2] block: remove the per-bio/request write hint.
Jens Axboe
axboe at kernel.dk
Mon Mar 14 12:58:25 PDT 2022
On 3/14/22 1:40 AM, Avi Shchislowski wrote:
>>>
>>> Hi Jens,
>>>
>>> The "upstream first" policy applies to the Android kernel (see also
>>> https://arstechnica.com/gadgets/2021/09/android-to-take-an-upstream-
>> first-development-model-for-the-linux-kernel/).
>>> If anyone requests inclusion in the Android kernel tree of a patch
>>> that is not upstream, that request is rejected unless a very strong
>>> reason can be provided why it should be included in the Android kernel
>>> only instead of being sent upstream. It is not clear to me why the
>>> patch Bean mentioned is not upstream nor in the upstream Android
>>> kernel tree.
>>>
>>> From a UFS vendor I received the feedback that the F2FS write hint
>>> information helps to reduce write amplification significantly. If the
>>> write hint information is retained in the upstream kernel I can help
>>> with making sure that the UFS patch mentioned above is integrated in
>>> the upstream Linux kernel.
>>
>> I'm really not that interested at this point, and I don't want to gate removal or
>> inclusion of code on some potential future event that may never happen.
>>
>> That doesn't mean that the work and process can't continue on the Android
>> front, the only difference is what the baseline kernel looks like for the
>> submitted patchset.
>>
>> Hence I do think we should go ahead as planned, and then we'll just revisit
>> the topic if/when some actual code comes up.
>>
> We also supports Samsung & Micron approach and sorry to see that this
> functionality has been removed.
This isn't some setup to solicit votes on who supports what. If the code
isn't upstream, it by definition doesn't exist to the kernel. No amount
of "we're also interested in this" changes that.
What I wrote earlier still applies - whoever is interested in supporting
lifetime hints should submit that code upstream. The existing patchset
to clean this up doesn't change that process AT ALL. As mentioned, the
only difference is what the baseline looks like in terms of what the
patchset is based on.
--
Jens Axboe
More information about the Linux-nvme
mailing list