Re: [RFC PATCH net-next v5 00/11] net/smc: SMC intra-OS shortcut with loopback-ism

From: Wen Gu
Date: Thu Apr 11 2024 - 05:57:06 EST




On 2024/4/11 17:32, Wenjia Zhang wrote:


On 11.04.24 09:45, Wen Gu wrote:


On 2024/4/3 19:10, Gerd Bayer wrote:
On Wed, 2024-04-03 at 14:35 +0800, Wen Gu wrote:


On 2024/3/24 21:55, Wen Gu wrote:
This patch set acts as the second part of the new version of [1]
(The first
part can be referred from [2]), the updated things of this version
are listed
at the end.

Change log:

RFC v5->RFC v4:
- Patch #2: minor changes in description of config SMC_LO and
comments.
- Patch #10: minor changes in comments and
if(smc_ism_support_dmb_nocopy())
    check in smcd_cdc_msg_send().
- Patch #3: change smc_lo_generate_id() to smc_lo_generate_ids()
and SMC_LO_CHID
    to SMC_LO_RESERVED_CHID.
- Patch #5: memcpy while holding the ldev->dmb_ht_lock.
- Some expression changes in commit logs.


Hi, Jan. Do you have any comments on this version and should I post a
new patch series without 'RFC'? Thank you.

Hi Wen,

Jan has been out sick for a little while now, and Wenjia is expected
back from a longer vacation tomorrow. So if you could hold off until
begin of next week, Wenjia might have some more feedback.

In the meantime, I'm looking at your patchset...

Thank you, Gerd


Hi Gerd, is there any further information? I am wondering if I
should wait for more feedback from SMC maintainers. Thanks!


Hi Wenjia, when it's convenient for you, could you please confirm
if [1] and [2] need to be included in the next version? Thanks!

[1] https://lore.kernel.org/netdev/7291dd1b2d16fd9bbd90988ac5bcc3a46d17e3f4.camel@xxxxxxxxxxxxx/
[2] https://lore.kernel.org/netdev/60b4aec0b4bf4474d651b653c86c280dafc4518a.camel@xxxxxxxxxxxxx/


Hi Wen,

I'm just back, thank you for the patience!

Firstly I want to thank Gerd and Niklas for review and bringing up these points!

Here are some of my options on that:

To [1]:
I agree to document the ops as otional if it must not be supported. Since I don't really have any ideas, the classification souds reasonable to me. Going to the details, what about to take following options as mandatory:

* query_remote_gid()
* register_dmb()/unregister_dmb()
* move_data() : I do see the necessary here.
* get_local_gid()
* get_chid()
* get_dev()

To [2]:
I also agree to keep the ism-loopback at the very beginning of the List. That acting is also what I imaged previously. Thank you, gerd, again for testing it and find it out!

Thanks,
Wenjia

Hi Wenjia, welcome back! :)

OK, then I will take these in my next version. Thank you all!