From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
To: Ferruh Yigit <ferruh.yigit@amd.com>,
Konstantin Ananyev <konstantin.v.ananyev@yandex.ru>,
"dev@dpdk.org" <dev@dpdk.org>
Cc: "hujiayu.hu@foxmail.com" <hujiayu.hu@foxmail.com>,
"roretzla@linux.microsoft.com" <roretzla@linux.microsoft.com>,
"bruce.richardson@intel.com" <bruce.richardson@intel.com>,
"anatoly.burakov@intel.com" <anatoly.burakov@intel.com>,
"vladimir.medvedkin@intel.com" <vladimir.medvedkin@intel.com>
Subject: RE: [RFC 4/4] net/ice: remove use of VLAs
Date: Thu, 13 Jun 2024 10:32:13 +0000 [thread overview]
Message-ID: <773311c7507d4bad987d90bbf7ea41b8@huawei.com> (raw)
In-Reply-To: <9b196355-578a-48f3-926b-38cadff28052@amd.com>
>
> On 5/23/2024 5:26 PM, Konstantin Ananyev wrote:
> > From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
> >
> > ../drivers/net/ice/ice_rxtx.c:1871:29: warning: variable length array used [-Wvla]
> >
> > Here VLA is used as a temp array for mbufs that will be used as a split
> > RX data buffers.
> > As at any given time only one thread can do RX from particular queue,
> > at rx_queue_setup() we can allocate extra space for that array, and then
> > safely use it at RX fast-path.
> >
>
> Is there a reason to allocate extra space in sw_ring and used some part
> of it for split buffer, instead of allocating a new buffer for it?
Less allocations - less points to fail, less checks to do.
Again, having it close to sw_ring is probably a good thing too,
possibly less pressure on MMU, etc. - even though I don't think it is really critical.
But yes, it could be a separate rte_zmalloc(), even though
I don't see good reason for that.
>
> > Signed-off-by: Konstantin Ananyev <konstantin.ananyev@huawei.com>
> > ---
> > drivers/net/ice/ice_rxtx.c | 18 ++++++++++++------
> > drivers/net/ice/ice_rxtx.h | 2 ++
> > 2 files changed, 14 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
> > index 95a2db3432..6395a3b50a 100644
> > --- a/drivers/net/ice/ice_rxtx.c
> > +++ b/drivers/net/ice/ice_rxtx.c
> > @@ -1171,7 +1171,7 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
> > struct ice_vsi *vsi = pf->main_vsi;
> > struct ice_rx_queue *rxq;
> > const struct rte_memzone *rz;
> > - uint32_t ring_size;
> > + uint32_t ring_size, tlen;
> > uint16_t len;
> > int use_def_burst_func = 1;
> > uint64_t offloads;
> > @@ -1279,9 +1279,14 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
> > /* always reserve more for bulk alloc */
> > len = (uint16_t)(nb_desc + ICE_RX_MAX_BURST);
> >
> > + /* allocate extra entries for SW split buffer */
> > + tlen = ((rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) != 0) ?
> > + rxq->rx_free_thresh : 0;
> > + tlen += len;
> > +
> > /* Allocate the software ring. */
> > rxq->sw_ring = rte_zmalloc_socket(NULL,
> > - sizeof(struct ice_rx_entry) * len,
> > + sizeof(struct ice_rx_entry) * tlen,
> > RTE_CACHE_LINE_SIZE,
> > socket_id);
> > if (!rxq->sw_ring) {
> > @@ -1290,6 +1295,8 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
> > return -ENOMEM;
> > }
> >
> > + rxq->sw_split_buf = (tlen == len) ? NULL : rxq->sw_ring + len;
> > +
> > ice_reset_rx_queue(rxq);
> > rxq->q_set = true;
> > dev->data->rx_queues[queue_idx] = rxq;
> > @@ -1868,7 +1875,6 @@ ice_rx_alloc_bufs(struct ice_rx_queue *rxq)
> > uint64_t dma_addr;
> > int diag, diag_pay;
> > uint64_t pay_addr;
> > - struct rte_mbuf *mbufs_pay[rxq->rx_free_thresh];
> >
> > /* Allocate buffers in bulk */
> > alloc_idx = (uint16_t)(rxq->rx_free_trigger -
> > @@ -1883,7 +1889,7 @@ ice_rx_alloc_bufs(struct ice_rx_queue *rxq)
> >
> > if (rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) {
> > diag_pay = rte_mempool_get_bulk(rxq->rxseg[1].mp,
> > - (void *)mbufs_pay, rxq->rx_free_thresh);
> > + (void *)rxq->sw_split_buf, rxq->rx_free_thresh);
> >
>
> Are we allowed to call 'rte_mempool_get_bulk()' with NULL object_table,
Nope.
> as 'rxq->sw_split_buf' can be NULL?
> Perhaps can allocate 'rxq->sw_split_buf' even buffer split offload is
> not enabled?
No, if (rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) !=0, then
rxq->sw_split_buf should not be NULL.
If it is, then there is a bug in my changes, though right now I don't see
how it can happen: as in ice_rx_queue_setup() we always allocate space
rxq->sw_split_buf when RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT is set.
next prev parent reply other threads:[~2024-06-13 10:32 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-05-23 16:26 [RFC 0/4] remove use of VLA Konstantin Ananyev
2024-05-23 16:26 ` [RFC 1/4] gro: fix overwrite unprocessed packets Konstantin Ananyev
2024-06-12 0:48 ` Ferruh Yigit
2024-05-23 16:26 ` [RFC 2/4] gro: remove use of VLAs Konstantin Ananyev
2024-06-12 0:48 ` Ferruh Yigit
2024-06-13 10:20 ` Konstantin Ananyev
2024-06-14 15:11 ` Ferruh Yigit
2024-06-28 12:57 ` Konstantin Ananyev
2024-05-23 16:26 ` [RFC 3/4] net/ixgbe: " Konstantin Ananyev
2024-06-12 1:00 ` Ferruh Yigit
2024-05-23 16:26 ` [RFC 4/4] net/ice: " Konstantin Ananyev
2024-06-12 1:12 ` Ferruh Yigit
2024-06-13 10:32 ` Konstantin Ananyev [this message]
2024-06-14 15:31 ` Ferruh Yigit
2024-06-12 1:14 ` [RFC 0/4] remove use of VLA Ferruh Yigit
2024-06-13 10:43 ` Konstantin Ananyev
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=773311c7507d4bad987d90bbf7ea41b8@huawei.com \
--to=konstantin.ananyev@huawei.com \
--cc=anatoly.burakov@intel.com \
--cc=bruce.richardson@intel.com \
--cc=dev@dpdk.org \
--cc=ferruh.yigit@amd.com \
--cc=hujiayu.hu@foxmail.com \
--cc=konstantin.v.ananyev@yandex.ru \
--cc=roretzla@linux.microsoft.com \
--cc=vladimir.medvedkin@intel.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).