From: Jakub Kicinski <kuba@kernel.org>
To: Claudiu Manoil <claudiu.manoil@nxp.com>
Cc: "netdev@vger.kernel.org" <netdev@vger.kernel.org>,
"David S . Miller" <davem@davemloft.net>,
"james.jurack@ametek.com" <james.jurack@ametek.com>
Subject: Re: [PATCH net] gianfar: Account for Tx PTP timestamp in the skb headroom
Date: Thu, 22 Oct 2020 19:54:55 -0700 [thread overview]
Message-ID: <20201022195455.09939040@kicinski-fedora-PC1C0HJN.hsd1.ca.comcast.net> (raw)
In-Reply-To: <AM0PR04MB6754877DC2BBA688F2DC249A961D0@AM0PR04MB6754.eurprd04.prod.outlook.com>
On Thu, 22 Oct 2020 12:09:00 +0000 Claudiu Manoil wrote:
> >-----Original Message-----
> >From: Jakub Kicinski <kuba@kernel.org>
> >Sent: Wednesday, October 21, 2020 9:00 PM
> >To: Claudiu Manoil <claudiu.manoil@nxp.com>
> >Cc: netdev@vger.kernel.org; David S . Miller <davem@davemloft.net>;
> >james.jurack@ametek.com
> >Subject: Re: [PATCH net] gianfar: Account for Tx PTP timestamp in the skb
> >headroom
> >
> >On Tue, 20 Oct 2020 20:36:05 +0300 Claudiu Manoil wrote:
> [...]
> >>
> >> if (dev->features & NETIF_F_IP_CSUM ||
> >> priv->device_flags & FSL_GIANFAR_DEV_HAS_TIMER)
> >> - dev->needed_headroom = GMAC_FCB_LEN;
> >> + dev->needed_headroom = GMAC_FCB_LEN + GMAC_TXPAL_LEN;
> >>
> >> /* Initializing some of the rx/tx queue level parameters */
> >> for (i = 0; i < priv->num_tx_queues; i++) {
> >
> >Claudiu, I think this may be papering over the real issue.
> >needed_headroom is best effort, if you were seeing crashes
> >the missing checks for skb being the right geometry are still
> >out there, they just not get hit in the case needed_headroom
> >is respected.
> >
> >So updating needed_headroom is definitely fine, but the cause of
> >crashes has to be found first.
>
> I agree Jakub, but this is a simple (and old) ring based driver. So the
> question is what checks or operations may be missing or be wrong
> in the below sequence of operations made on the skb designated for
> timestamping?
> As hinted in the commit description, the code is not crashing when
> multiple TCP streams are transmitted alone (skb frags manipulation was
> omitted for brevity below, and this is a well exercised path known to work),
> but only crashes when multiple TCP stream are run concurrently
> with a PTP Tx packet flow taking the skb_realloc_headroom() path.
> I don't find other drivers doing skb_realloc_headroom() for PTP packets,
> so maybe is this an untested use case of skb usage?
>
> init:
> dev->needed_headroom = GMAC_FCB_LEN;
>
> start_xmit:
> netdev_tx_t gfar_start_xmit(struct sk_buff *skb, struct net_device *dev)
> {
> do_tstamp = (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
> priv->hwts_tx_en;
> fcb_len = GMAC_FCB_LEN; // can also be 0
> if (do_tstamp)
> fcb_len = GMAC_FCB_LEN + GMAC_TXPAL_LEN;
>
> if (skb_headroom(skb) < fcb_len) {
> skb_new = skb_realloc_headroom(skb, fcb_len);
> if (!skb_new) {
> dev_kfree_skb_any(skb);
> return NETDEV_TX_OK;
> }
> if (skb->sk)
> skb_set_owner_w(skb_new, skb->sk);
> dev_consume_skb_any(skb);
> skb = skb_new;
> }
Try replacing this if () with skb_cow_head(). The skbs you're getting
are probably cloned.
> [...]
> if (do_tstamp)
> skb_push(skb, GMAC_TXPAL_LEN);
> if (fcb_len)
> skb_push(skb, GMAC_FCB_LEN);
>
> // dma map and send the frame
> }
>
> Tx napi (xmit completion):
> gfar_clean_tx_ring()
> {
> do_tstamp = (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
> priv->hwts_tx_en;
> // dma unmap
> if (do_tstamp) {
> struct skb_shared_hwtstamps shhwtstamps;
>
> // read timestamp from skb->data and write it to shhwtstamps
> skb_pull(skb, GMAC_FCB_LEN + GMAC_TXPAL_LEN);
> skb_tstamp_tx(skb, &shhwtstamps);
> }
> dev_kfree_skb_any(skb);
> }
>
next prev parent reply other threads:[~2020-10-23 2:55 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-09-14 13:53 PROBLEM: Crash when timestamping outgoing PTP packets under heavy network load (ppc, gianfar) James Jurack
2020-10-20 17:36 ` [PATCH net] gianfar: Account for Tx PTP timestamp in the skb headroom Claudiu Manoil
2020-10-21 17:59 ` Jakub Kicinski
2020-10-22 12:09 ` Claudiu Manoil
2020-10-23 2:54 ` Jakub Kicinski [this message]
2020-10-23 11:37 ` Claudiu Manoil
2020-10-29 8:10 ` [PATCH net v2 1/2] gianfar: Replace skb_realloc_headroom with skb_cow_head for PTP Claudiu Manoil
2020-10-29 8:10 ` [PATCH net v2 2/2] gianfar: Account for Tx PTP timestamp in the skb headroom Claudiu Manoil
2020-10-30 16:41 ` [PATCH net v2 1/2] gianfar: Replace skb_realloc_headroom with skb_cow_head for PTP Jakub Kicinski
2020-11-03 16:13 ` Vladimir Oltean
2020-11-03 16:30 ` Jakub Kicinski
2020-11-03 17:18 ` Claudiu Manoil
2020-11-03 17:30 ` Vladimir Oltean
2020-11-03 17:36 ` Jakub Kicinski
2020-11-03 17:39 ` Jakub Kicinski
2020-11-03 17:51 ` Vladimir Oltean
2020-11-03 17:41 ` Claudiu Manoil
2020-11-03 17:49 ` Vladimir Oltean
2020-11-03 18:16 ` Eric Dumazet
2020-11-04 16:45 ` Claudiu Manoil
2020-11-03 18:16 ` Vladimir Oltean
2020-11-03 16:36 ` Julian Wiedmann
2020-11-03 17:08 ` Claudiu Manoil
2020-11-03 17:12 ` Jakub Kicinski
2020-11-03 17:24 ` Vladimir Oltean
2020-11-03 17:38 ` Vladimir Oltean
2020-11-03 17:18 ` Vladimir Oltean
2020-11-03 17:27 ` Vladimir Oltean
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=20201022195455.09939040@kicinski-fedora-PC1C0HJN.hsd1.ca.comcast.net \
--to=kuba@kernel.org \
--cc=claudiu.manoil@nxp.com \
--cc=davem@davemloft.net \
--cc=james.jurack@ametek.com \
--cc=netdev@vger.kernel.org \
/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).