lkml.org 
[lkml]   [2023]   [Oct]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH RFC 1/2] RDMA/rxe: don't allow registering !PAGE_SIZE mr
From
在 2023/10/29 7:07, Bart Van Assche 写道:
> On 10/27/23 19:48, Zhu Yanjun wrote:
>> In this case, ULP with folio will not work well with current RXE after
>> this commit is applied.
>
> Why not? RDMA ULPs like the SRP initiator driver use ib_map_mr_sg(). The
> latter function calls rxe_map_mr_sg() if the RXE driver is used.
> rxe_map_mr_sg() calls ib_sg_to_pages(). ib_sg_to_pages() translates
> SG-entries that are larger than a virtual memory page into multiple
> entries with size mr->page_size.

It seems that we are not on the same page. I am thinking that this fix
should also work for the future folio. And your idea is based on the
current implementation.

Perhaps it is not a good time to discuss folio currently. Let me focus
on the current implementation.

In this commit, if the page size is not equal to PAGE_SIZE, it will pop
out warning then exit. So "rxe_info_mr" should be changed to rxe_warning_mr?

In fact, I like to remove the page size assignment in rxe, that is,
partly reverting the commit 325a7eb85199 ("RDMA/rxe: Cleanup page
variables in rxe_mr.c"). But it seems that Jason did not like this.
I no longer insist on this.

I am not sure whether Jason, Leon, Bob and you have better solution to
this problem or not.

If not, this commit can fix this problem if no better solution.

Reviewed-by: Zhu Yanjun <yanjun.zhu@linux.dev>

Thanks,
Zhu Yanjun

>
> Thanks,
>
> Bart.
>

\
 
 \ /
  Last update: 2023-10-29 04:24    [W:0.134 / U:0.056 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site