DPDK patches and discussions
 help / color / mirror / Atom feed
From: Andrew Rybchenko <arybchenko@solarflare.com>
To: <dev@dpdk.org>
Cc: Mark Spender <mspender@solarflare.com>
Subject: [dpdk-dev] [PATCH 02/53] net/sfc/base: do not use Tx desc push with TSO descriptors
Date: Thu, 16 Nov 2017 08:03:50 +0000	[thread overview]
Message-ID: <1510819481-6809-3-git-send-email-arybchenko@solarflare.com> (raw)
In-Reply-To: <1510819481-6809-1-git-send-email-arybchenko@solarflare.com>

From: Mark Spender <mspender@solarflare.com>

Pushing TSO option descriptors is unsafe if pacer bypass is enabled,
so to make sure that doesn't happen never push TSO option descriptors.

Signed-off-by: Mark Spender <mspender@solarflare.com>
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
---
 drivers/net/sfc/base/ef10_tx.c | 56 ++++++++++++++++++++++++++++++++----------
 1 file changed, 43 insertions(+), 13 deletions(-)

diff --git a/drivers/net/sfc/base/ef10_tx.c b/drivers/net/sfc/base/ef10_tx.c
index 211d265..2666943 100644
--- a/drivers/net/sfc/base/ef10_tx.c
+++ b/drivers/net/sfc/base/ef10_tx.c
@@ -473,9 +473,9 @@ ef10_tx_qpost(
 }
 
 /*
- * This improves performance by pushing a TX descriptor at the same time as the
- * doorbell. The descriptor must be added to the TXQ, so that can be used if the
- * hardware decides not to use the pushed descriptor.
+ * This improves performance by, when possible, pushing a TX descriptor at the
+ * same time as the doorbell. The descriptor must be added to the TXQ, so that
+ * can be used if the hardware decides not to use the pushed descriptor.
  */
 			void
 ef10_tx_qpush(
@@ -495,16 +495,46 @@ ef10_tx_qpush(
 	offset = id * sizeof (efx_qword_t);
 
 	EFSYS_MEM_READQ(etp->et_esmp, offset, &desc);
-	EFX_POPULATE_OWORD_3(oword,
-	    ERF_DZ_TX_DESC_WPTR, wptr,
-	    ERF_DZ_TX_DESC_HWORD, EFX_QWORD_FIELD(desc, EFX_DWORD_1),
-	    ERF_DZ_TX_DESC_LWORD, EFX_QWORD_FIELD(desc, EFX_DWORD_0));
-
-	/* Guarantee ordering of memory (descriptors) and PIO (doorbell) */
-	EFX_DMA_SYNC_QUEUE_FOR_DEVICE(etp->et_esmp, etp->et_mask + 1, wptr, id);
-	EFSYS_PIO_WRITE_BARRIER();
-	EFX_BAR_TBL_DOORBELL_WRITEO(enp, ER_DZ_TX_DESC_UPD_REG, etp->et_index,
-				    &oword);
+
+	/*
+	 * Bug 65776: TSO option descriptors cannot be pushed if pacer bypass is
+	 * enabled on the event queue this transmit queue is attached to.
+	 *
+	 * To ensure the code is safe, it is easiest to simply test the type of
+	 * the descriptor to push, and only push it is if it not a TSO option
+	 * descriptor.
+	 */
+	if ((EFX_QWORD_FIELD(desc, ESF_DZ_TX_DESC_IS_OPT) != 1) ||
+	    (EFX_QWORD_FIELD(desc, ESF_DZ_TX_OPTION_TYPE) !=
+	    ESE_DZ_TX_OPTION_DESC_TSO)) {
+		/* Push the descriptor and update the wptr. */
+		EFX_POPULATE_OWORD_3(oword, ERF_DZ_TX_DESC_WPTR, wptr,
+		    ERF_DZ_TX_DESC_HWORD, EFX_QWORD_FIELD(desc, EFX_DWORD_1),
+		    ERF_DZ_TX_DESC_LWORD, EFX_QWORD_FIELD(desc, EFX_DWORD_0));
+
+		/* Ensure ordering of memory (descriptors) and PIO (doorbell) */
+		EFX_DMA_SYNC_QUEUE_FOR_DEVICE(etp->et_esmp, etp->et_mask + 1,
+					    wptr, id);
+		EFSYS_PIO_WRITE_BARRIER();
+		EFX_BAR_TBL_DOORBELL_WRITEO(enp, ER_DZ_TX_DESC_UPD_REG,
+					    etp->et_index, &oword);
+	} else {
+		efx_dword_t dword;
+
+		/*
+		 * Only update the wptr. This is signalled to the hardware by
+		 * only writing one DWORD of the doorbell register.
+		 */
+		EFX_POPULATE_OWORD_1(oword, ERF_DZ_TX_DESC_WPTR, wptr);
+		dword = oword.eo_dword[2];
+
+		/* Ensure ordering of memory (descriptors) and PIO (doorbell) */
+		EFX_DMA_SYNC_QUEUE_FOR_DEVICE(etp->et_esmp, etp->et_mask + 1,
+					    wptr, id);
+		EFSYS_PIO_WRITE_BARRIER();
+		EFX_BAR_TBL_WRITED2(enp, ER_DZ_TX_DESC_UPD_REG,
+				    etp->et_index, &dword, B_FALSE);
+	}
 }
 
 	__checkReturn	efx_rc_t
-- 
2.7.4

  parent reply	other threads:[~2017-11-16  8:04 UTC|newest]

Thread overview: 71+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-11-16  8:03 [dpdk-dev] [PATCH 00/53] net/sfc: base driver update Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 01/53] net/sfc/base: copy new header from firmwaresrc Andrew Rybchenko
2017-11-27 19:58   ` Ferruh Yigit
2017-11-29  9:49     ` Andrew Rybchenko
2017-11-27 19:58   ` Ferruh Yigit
2017-11-16  8:03 ` Andrew Rybchenko [this message]
2017-11-16  8:03 ` [dpdk-dev] [PATCH 03/53] net/sfc/base: fix result code in MCDI NVRAM update finish Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 04/53] net/sfc/base: simplify verify result handling Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 05/53] net/sfc/base: report verify result from RW finish callback Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 06/53] net/sfc/base: extend NVRAM RW finish to return verify result Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 07/53] net/sfc/base: rename firmware update verify result cap field Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 08/53] net/sfc/base: simplify NVRAM type to partition mappings Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 09/53] net/sfc/base: check NVRAM locking by partition ID Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 10/53] net/sfc/base: report correct partition write chunk size Andrew Rybchenko
2017-11-16  8:03 ` [dpdk-dev] [PATCH 11/53] net/sfc/base: fix check in NVRAM validate Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 12/53] net/sfc/base: precheck and verify flash writes Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 13/53] net/sfc/base: remove duplicate NVRAM asserts Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 14/53] net/sfc/base: quieten get version methods Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 15/53] net/sfc/base: fix PreFAST issues Andrew Rybchenko
2017-11-27 19:58   ` [dpdk-dev] [dpdk-stable] " Ferruh Yigit
2017-11-28 11:49     ` Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 16/53] net/sfc/base: add support for MUM/SUC firmware partitions Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 17/53] net/sfc/base: improve PS credits push function name Andrew Rybchenko
2017-11-27 19:59   ` Ferruh Yigit
2017-11-28  9:20     ` Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 18/53] net/sfc/base: improve RxQ label init prototype Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 19/53] net/sfc/base: add description of the PS packets layout Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 20/53] net/sfc/base: clarify meaning of Rx desc lbits in PS mode Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 21/53] net/sfc/base: optimize credits overflow check Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 22/53] net/sfc/base: provide simple access to RxQ state in EvQ Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 23/53] net/sfc/base: enforce packed stream fake buffer size Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 24/53] net/sfc/base: more accurately calculate number of PS credits Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 25/53] net/sfc/base: make MAC naming consistent with other modules Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 26/53] net/sfc/base: support inner checksum offload on transmit Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 27/53] net/sfc/base: use MCDIv2 for requests with too long response Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 28/53] net/sfc/base: ignore error in completion event on MCDIv2 HW Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 29/53] net/sfc/base: request info about outer frame in Rx events Andrew Rybchenko
2017-11-27 19:57   ` Ferruh Yigit
2017-11-28 10:48     ` Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 30/53] net/sfc/base: remove assertion on no longer used define Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 31/53] net/sfc/base: improve names for EVQ descriptor counts Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 32/53] net/sfc/base: improve names for RXQ " Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 33/53] net/sfc/base: improve names for TXQ " Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 34/53] net/sfc/base: fix build issue with PHY LED control enabled Andrew Rybchenko
2017-11-27 19:59   ` [dpdk-dev] [dpdk-stable] " Ferruh Yigit
2017-11-16  8:04 ` [dpdk-dev] [PATCH 35/53] net/sfc/base: move Siena-specific defs to right header Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 36/53] net/sfc/base: run mkconfig.py to add a new sensors Andrew Rybchenko
2017-11-27 19:58   ` Ferruh Yigit
2017-11-28 12:01     ` Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 37/53] net/sfc/base: add support for " Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 38/53] net/sfc/base: remove unused defined for WPTR alignment Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 39/53] net/sfc/base: cstyle fixes Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 40/53] net/sfc/base: allow to use PHY stats on Huntington/Medford Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 41/53] net/sfc/base: fix diagnostics support build without Siena Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 42/53] net/sfc/base: fix probes in licensing support Andrew Rybchenko
2017-11-27 19:58   ` [dpdk-dev] [dpdk-stable] " Ferruh Yigit
2017-11-28 10:17     ` Andrew Rybchenko
2017-11-28 21:38       ` Ferruh Yigit
2017-11-29  9:51         ` Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 43/53] net/sfc/base: fix warnings from VS2015 C compiler (C4310) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 44/53] net/sfc/base: fix warnings from VS2015 C compiler (C4244) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 45/53] net/sfc/base: fix warnings from VS2015 C compiler (C4245) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 46/53] net/sfc/base: fix warnings from VS2015 C compiler (C4100) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 47/53] net/sfc/base: fix warnings from VS2015 C compiler (C4189) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 48/53] net/sfc/base: fix warnings from VS2015 C compiler (C4057) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 49/53] net/sfc/base: fix warnings from VS2015 C compiler (C4214) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 50/53] net/sfc/base: remove obsolete check for pre-Siena hardware Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 51/53] net/sfc/base: expand on comment on number of queues field Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 52/53] net/sfc/base: fix PreFAST static analysis warning (C6001) Andrew Rybchenko
2017-11-16  8:04 ` [dpdk-dev] [PATCH 53/53] net/sfc/base: move BIU test code into Siena-specific file Andrew Rybchenko
2017-11-27 19:59 ` [dpdk-dev] [PATCH 00/53] net/sfc: base driver update Ferruh Yigit

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=1510819481-6809-3-git-send-email-arybchenko@solarflare.com \
    --to=arybchenko@solarflare.com \
    --cc=dev@dpdk.org \
    --cc=mspender@solarflare.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).