From: Honnappa Nagarahalli <Honnappa.Nagarahalli@arm.com>
To: "Pavan Nikhilesh Bhagavatula" <pbhagavatula@marvell.com>,
"Bruce Richardson" <bruce.richardson@intel.com>,
"Morten Brørup" <mb@smartsharesystems.com>
Cc: Shijith Thotton <sthotton@marvell.com>,
"dev@dpdk.org" <dev@dpdk.org>,
"jerinj@marvell.com" <jerinj@marvell.com>,
"olivier.matz@6wind.com" <olivier.matz@6wind.com>,
"stephen@networkplumber.org" <stephen@networkplumber.org>,
"thomas@monjalon.net" <thomas@monjalon.net>, nd <nd@arm.com>
Subject: RE: [EXT] Re: [PATCH v1 2/4] mbuf: add second dynamic field member for VA only build
Date: Tue, 30 Aug 2022 13:22:19 +0000 [thread overview]
Message-ID: <DBAPR08MB58144DC3F7E2C2E4FBBE8AEE98799@DBAPR08MB5814.eurprd08.prod.outlook.com> (raw)
In-Reply-To: <PH0PR18MB408665FCDF05C630C7184AE3DE799@PH0PR18MB4086.namprd18.prod.outlook.com>
<snip>
> >
> > ----------------------------------------------------------------------
> > On Mon, Aug 29, 2022 at 08:32:20PM +0200, Morten Brørup wrote:
> > >
> > > > From: Shijith Thotton [mailto:sthotton@marvell.com]
> > > > Sent: Monday, 29 August 2022 17.16
> > > >
> > > > mbuf physical address field is not used in builds which only uses VA.
> > > > It is used to expand the dynamic field area.
> > > >
> > > > Signed-off-by: Shijith Thotton <sthotton@marvell.com>
> > > > ---
> > > > lib/mbuf/rte_mbuf_core.h | 26 +++++++++++++++++---------
> > > > lib/mbuf/rte_mbuf_dyn.c | 2 ++
> > > > 2 files changed, 19 insertions(+), 9 deletions(-)
> > > >
> > > > diff --git a/lib/mbuf/rte_mbuf_core.h b/lib/mbuf/rte_mbuf_core.h
> > > > index 81cb07c2e4..98ce62fd6a 100644
> > > > --- a/lib/mbuf/rte_mbuf_core.h
> > > > +++ b/lib/mbuf/rte_mbuf_core.h
> > > > @@ -579,15 +579,23 @@ struct rte_mbuf {
> > > > RTE_MARKER cacheline0;
> > > >
> > > > void *buf_addr; /**< Virtual address of segment buffer.
> > > > */
> > > > - /**
> > > > - * Physical address of segment buffer.
> > > > - * This field is invalid if the build is configured to use only
> > > > - * virtual address as IOVA (i.e. RTE_IOVA_AS_VA is defined).
> > > > - * Force alignment to 8-bytes, so as to ensure we have the exact
> > > > - * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
> > > > - * working on vector drivers easier.
> > > > - */
> > > > - rte_iova_t buf_iova __rte_aligned(sizeof(rte_iova_t));
> > > > + RTE_STD_C11
> > > > + union {
> > > > + /**
> > > > + * Physical address of segment buffer.
> > > > + * This field is invalid if the build is configured to use
> > > > only
> > > > + * virtual address as IOVA (i.e. RTE_IOVA_AS_VA is
> > > > defined).
> > > > + * Force alignment to 8-bytes, so as to ensure we have the
> > > > exact
> > > > + * same mbuf cacheline0 layout for 32-bit and 64-bit. This
> > > > makes
> > > > + * working on vector drivers easier.
> > > > + */
> > > > + rte_iova_t buf_iova __rte_aligned(sizeof(rte_iova_t));
> > > > + /**
> > > > + * Reserved for dynamic field in builds where physical
> > > > address
> > > > + * field is invalid.
> > > > + */
> > > > + uint64_t dynfield2;
> > > > + };
> > > >
> > > > /* next 8 bytes are initialised on RX descriptor rearm */
> > > > RTE_MARKER64 rearm_data;
> > >
> > > I know that the intention here is to keep the rte_mbuf structure
> > > intact,
> > which will certainly improve the probability of getting this patch
> > series into DPDK.
> > >
> > > So, I will add a comment for the benefit of the other participants
> > > in the
> > discussion:
> > >
> > > With this patch, and in RTE_IOVA_AS_VA mode, it becomes possible to
> > move m->next into the first cache line, so rte_pktmbuf_prefree_seg()
> > does not have to touch the second cache line, thus potentially
> > improving performance by eliminating one cache miss per freed packet
> > segment. (I also recall someone mentioning that some PMDs set m->next
> > on RX... If that is the case, a cache miss per packet might also be
> > avoidable in those PMDs.)
> > >
> > > Obviously, moving m->next to the first cache line is not related to
> > > this patch
> > series, but would belong in a completely different patch.
> > >
> >
> > +1 to that, with the exception that if it is decided to move the next
> > pointer rather than use this as dynamic space, I think it *should* be
> > in this patch series, rather than mucking about with mbuf twice. :-)
>
> +1 When RTE_IOVA_AS_VA is set we can set mbuf->next as the dynamic field
> and move it to mbuf->buf_iova.
> mbuf->next write is one of the prominent hotspot in arm platforms.
+1 for reducing the cachelines that need to be touched
next prev parent reply other threads:[~2022-08-30 13:22 UTC|newest]
Thread overview: 88+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-06-30 16:25 [PATCH] mbuf: add mbuf physical address field to dynamic field Shijith Thotton
2022-06-30 16:45 ` Stephen Hemminger
2022-07-01 12:16 ` Shijith Thotton
2022-07-01 12:24 ` Shijith Thotton
2022-07-03 7:31 ` Morten Brørup
2022-07-04 14:00 ` Bruce Richardson
2022-08-03 15:34 ` [EXT] " Shijith Thotton
2022-08-29 15:16 ` [PATCH v1 0/4] mbuf dynamic field expansion Shijith Thotton
2022-09-07 13:43 ` [PATCH v2 0/5] " Shijith Thotton
2022-09-21 9:43 ` David Marchand
2022-09-21 14:01 ` [EXT] " Shijith Thotton
2022-09-21 13:56 ` [PATCH v3 " Shijith Thotton
2022-09-21 13:56 ` [PATCH v3 1/5] build: add meson option to configure IOVA mode as VA Shijith Thotton
2022-09-28 12:52 ` Olivier Matz
2022-09-29 5:48 ` [EXT] " Shijith Thotton
2022-09-21 13:56 ` [PATCH v3 2/5] mbuf: add second dynamic field member for VA only build Shijith Thotton
2022-09-28 7:24 ` Thomas Monjalon
2022-09-28 12:52 ` Olivier Matz
2022-09-28 19:33 ` Thomas Monjalon
2022-09-28 19:48 ` Stephen Hemminger
2022-09-29 6:13 ` [EXT] " Shijith Thotton
2022-09-28 12:52 ` Olivier Matz
2022-09-21 13:56 ` [PATCH v3 3/5] lib: move mbuf next pointer to first cache line Shijith Thotton
2022-09-21 14:07 ` Morten Brørup
2022-09-28 12:52 ` Olivier Matz
2022-09-29 6:14 ` [EXT] " Shijith Thotton
2022-09-21 13:56 ` [PATCH v3 4/5] drivers: mark Marvell cnxk PMDs work with IOVA as VA Shijith Thotton
2022-09-28 12:53 ` Olivier Matz
2022-09-29 6:19 ` [EXT] " Shijith Thotton
2022-09-29 7:44 ` Olivier Matz
2022-09-29 8:10 ` Shijith Thotton
2022-10-07 20:17 ` Olivier Matz
2022-10-07 20:22 ` [EXT] " Shijith Thotton
2022-09-21 13:56 ` [PATCH v3 5/5] drivers: mark software " Shijith Thotton
2022-09-28 5:41 ` [PATCH v3 0/5] mbuf dynamic field expansion Shijith Thotton
2022-09-28 12:52 ` Olivier Matz
2022-09-29 4:51 ` [EXT] " Shijith Thotton
2022-10-07 13:50 ` Thomas Monjalon
2022-10-07 19:35 ` [EXT] " Shijith Thotton
2022-10-07 19:30 ` [PATCH v4 0/7] " Shijith Thotton
2022-10-07 19:30 ` [PATCH v4 1/7] mbuf: add API to get and set mbuf physical address Shijith Thotton
2022-10-07 20:16 ` Olivier Matz
2022-10-07 20:20 ` [EXT] " Shijith Thotton
2022-10-07 19:30 ` [PATCH v4 2/7] test/dma: use API to get mbuf data " Shijith Thotton
2022-10-07 20:17 ` Olivier Matz
2022-10-07 19:30 ` [PATCH v4 3/7] build: add meson option to configure IOVA mode as PA Shijith Thotton
2022-10-07 19:30 ` [PATCH v4 4/7] mbuf: add second dynamic field member Shijith Thotton
2022-10-07 19:30 ` [PATCH v4 5/7] lib: move mbuf next pointer to first cache line Shijith Thotton
2022-10-07 19:30 ` [PATCH v4 6/7] drivers: mark cnxk PMDs work with IOVA as PA disabled Shijith Thotton
2022-10-07 19:30 ` [PATCH v4 7/7] drivers: mark software " Shijith Thotton
2022-10-07 20:19 ` [PATCH v4 0/7] mbuf dynamic field expansion Olivier Matz
2022-10-07 21:02 ` [PATCH v5 " Shijith Thotton
2022-10-07 21:02 ` [PATCH v5 1/7] mbuf: add API to get and set mbuf physical address Shijith Thotton
2022-10-07 21:20 ` Stephen Hemminger
2022-10-07 21:02 ` [PATCH v5 2/7] test/dma: use API to get mbuf data " Shijith Thotton
2022-10-07 21:02 ` [PATCH v5 3/7] build: add meson option to configure IOVA mode as PA Shijith Thotton
2022-10-07 21:02 ` [PATCH v5 4/7] mbuf: add second dynamic field member Shijith Thotton
2022-10-07 21:02 ` [PATCH v5 5/7] lib: move mbuf next pointer to first cache line Shijith Thotton
2022-10-07 21:22 ` Stephen Hemminger
2022-10-07 21:30 ` [EXT] " Shijith Thotton
2022-10-07 21:02 ` [PATCH v5 6/7] drivers: mark cnxk PMDs work with IOVA as PA disabled Shijith Thotton
2022-10-07 21:02 ` [PATCH v5 7/7] drivers: mark software " Shijith Thotton
2022-10-09 9:34 ` [PATCH v5 0/7] mbuf dynamic field expansion Thomas Monjalon
2022-09-07 13:43 ` [PATCH v2 1/5] build: add meson option to configure IOVA mode as VA Shijith Thotton
2022-09-07 15:31 ` Stephen Hemminger
2022-09-07 15:38 ` Bruce Richardson
2022-09-07 21:33 ` Morten Brørup
2022-09-07 13:43 ` [PATCH v2 2/5] mbuf: add second dynamic field member for VA only build Shijith Thotton
2022-09-07 13:43 ` [PATCH v2 3/5] lib: move mbuf next pointer to first cache line Shijith Thotton
2022-09-07 13:43 ` [PATCH v2 4/5] drivers: mark Marvell cnxk PMDs work with IOVA as VA Shijith Thotton
2022-09-07 13:43 ` [PATCH v2 5/5] drivers: mark software " Shijith Thotton
2022-08-29 15:16 ` [PATCH v1 1/4] build: add meson option to configure IOVA mode " Shijith Thotton
2022-08-29 18:18 ` Morten Brørup
2022-08-30 8:32 ` Bruce Richardson
2022-08-29 15:16 ` [PATCH v1 2/4] mbuf: add second dynamic field member for VA only build Shijith Thotton
2022-08-29 18:32 ` Morten Brørup
2022-08-30 8:35 ` Bruce Richardson
2022-08-30 8:41 ` [EXT] " Pavan Nikhilesh Bhagavatula
2022-08-30 13:22 ` Honnappa Nagarahalli [this message]
2022-09-07 13:55 ` Shijith Thotton
2022-08-29 15:16 ` [PATCH v1 3/4] drivers: mark Marvell cnxk PMDs work with IOVA as VA Shijith Thotton
2022-08-29 15:16 ` [PATCH v1 4/4] drivers: mark software " Shijith Thotton
2022-08-30 13:07 ` [PATCH] mbuf: add mbuf physical address field to dynamic field Ferruh Yigit
2022-09-12 13:19 ` [EXT] " Shijith Thotton
2022-06-30 16:55 ` Bruce Richardson
2022-07-01 9:48 ` Olivier Matz
2022-07-01 11:53 ` Slava Ovsiienko
2022-07-01 12:01 ` [EXT] " Shijith Thotton
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=DBAPR08MB58144DC3F7E2C2E4FBBE8AEE98799@DBAPR08MB5814.eurprd08.prod.outlook.com \
--to=honnappa.nagarahalli@arm.com \
--cc=bruce.richardson@intel.com \
--cc=dev@dpdk.org \
--cc=jerinj@marvell.com \
--cc=mb@smartsharesystems.com \
--cc=nd@arm.com \
--cc=olivier.matz@6wind.com \
--cc=pbhagavatula@marvell.com \
--cc=stephen@networkplumber.org \
--cc=sthotton@marvell.com \
--cc=thomas@monjalon.net \
/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).