From: Leon Romanovsky <leon@kernel.org>
To: Bob Pearson <rpearsonhpe@gmail.com>
Cc: jgg@nvidia.com, zyjzyj2000@gmail.com, linux-rdma@vger.kernel.org,
jhack@hpe.com
Subject: Re: [PATCH for-next 2/2] RDMA/rxe: Enable rcu locking of indexed objects
Date: Sun, 23 Jul 2023 18:00:50 +0300 [thread overview]
Message-ID: <20230723150050.GA60079@unreal> (raw)
In-Reply-To: <4450c401-2a02-d66d-5615-22f65e291a04@gmail.com>
On Wed, Jul 19, 2023 at 11:43:30AM -0500, Bob Pearson wrote:
> On 7/19/23 02:49, Leon Romanovsky wrote:
> > On Tue, Jul 18, 2023 at 12:59:44PM -0500, Bob Pearson wrote:
> >> Make rcu_read locking of critical sections with the indexed
> >> verbs objects be protected from early freeing of those objects.
> >> The AH, QP, MR and MW objects are looked up from their indices
> >> contained in received packets or WQEs during I/O processing.
> >> Make these objects be freed using kfree_rcu to avoid races.
> >
> > Sorry, how use of RCU avoid races?
> >
> > Thanks
>
> The races are between destroy/deallocate/dereg verbs API calls and packets arriving or completing send
> or deferred processing of wqes. Packets and wqes contain indices/keys/numbers that refer to objects.
> The rxe driver maintains xarrays for each type of object that allow to lookup the address of the object
> from its index and then take a reference to protect the pointer. The destroy verbs defer completion
> until the reference count falls to zero and then removes the entry in the xarray. These operations
> need to be atomic. One alternative is to use spinlocks to protect them but that places a load on
> performance under heavy load which is typically dominated by the lookup function since objects tend
> to have a long lifetime. rcu readlocks are a better alternative but depend on the deferred destruction
> of the objects used in the rcu critical section.
You rarely can replace locks with RCU without careful design changes.
Thanks
next prev parent reply other threads:[~2023-07-23 15:00 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-07-18 17:59 [PATCH for-next 0/2] Enable rcu locking of verbs objects Bob Pearson
2023-07-18 17:59 ` [PATCH for-next 1/2] RDMA/core: Support drivers use of rcu locking Bob Pearson
2023-08-09 19:10 ` Jason Gunthorpe
2023-07-18 17:59 ` [PATCH for-next 2/2] RDMA/rxe: Enable rcu locking of indexed objects Bob Pearson
2023-07-19 5:38 ` Zhu Yanjun
2023-07-19 14:34 ` Bob Pearson
2023-07-19 7:49 ` Leon Romanovsky
2023-07-19 16:43 ` Bob Pearson
2023-07-23 15:00 ` Leon Romanovsky [this message]
2023-08-09 19:17 ` Jason Gunthorpe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230723150050.GA60079@unreal \
--to=leon@kernel.org \
--cc=jgg@nvidia.com \
--cc=jhack@hpe.com \
--cc=linux-rdma@vger.kernel.org \
--cc=rpearsonhpe@gmail.com \
--cc=zyjzyj2000@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).