netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "Kok, Auke" <auke-jan.h.kok@intel.com>
To: "Garzik, Jeff" <jgarzik@pobox.com>
Cc: netdev@vger.kernel.org, "Brandeburg,
	Jesse" <jesse.brandeburg@intel.com>,
	"Kok, Auke" <auke-jan.h.kok@intel.com>,
	"Kok, Auke" <auke@foo-projects.org>,
	"Ronciak, John" <john.ronciak@intel.com>
Subject: [PATCH 20/22] e1000: Fix poor quad port performance with NAPI
Date: Fri, 08 Dec 2006 15:03:29 -0800	[thread overview]
Message-ID: <20061208230329.30109.85896.stgit@gitlost.site> (raw)
In-Reply-To: <20061208230209.30109.49157.stgit@gitlost.site>


This fix attempts to solve a customer reported issue with NAPI enabled
e1000 having bad performance when transmitting simultaneously on
four ports.  The issue comes down to an interaction between NAPI
hardware interrupt balancing, and the driver rescheduling poll on
the same processor.

We try to fix by allowing the driver to re-enable interrupts sooner
instead of polling one more time, when there was recently all the
work completed in cleanup.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
---

 drivers/net/e1000/e1000_main.c |   25 ++++++++++++++++---------
 1 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/drivers/net/e1000/e1000_main.c b/drivers/net/e1000/e1000_main.c
index 1a2b052..458aa38 100644
--- a/drivers/net/e1000/e1000_main.c
+++ b/drivers/net/e1000/e1000_main.c
@@ -3826,7 +3826,7 @@ e1000_intr_msi(int irq, void *data)
 
 	for (i = 0; i < E1000_MAX_INTR; i++)
 		if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
-		   !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
+		   e1000_clean_tx_irq(adapter, adapter->tx_ring)))
 			break;
 
 	if (likely(adapter->itr_setting & 3))
@@ -3979,7 +3979,7 @@ e1000_clean(struct net_device *poll_dev,
 	poll_dev->quota -= work_done;
 
 	/* If no Tx and not enough Rx work done, exit the polling mode */
-	if ((!tx_cleaned && (work_done == 0)) ||
+	if ((tx_cleaned && (work_done < work_to_do)) ||
 	   !netif_running(poll_dev)) {
 quit_polling:
 		if (likely(adapter->itr_setting & 3))
@@ -4009,7 +4009,7 @@ e1000_clean_tx_irq(struct e1000_adapter 
 #ifdef CONFIG_E1000_NAPI
 	unsigned int count = 0;
 #endif
-	boolean_t cleaned = FALSE;
+	boolean_t cleaned = TRUE;
 	unsigned int total_tx_bytes=0, total_tx_packets=0;
 
 	i = tx_ring->next_to_clean;
@@ -4025,13 +4025,17 @@ e1000_clean_tx_irq(struct e1000_adapter 
 			if (cleaned) {
 				struct sk_buff *skb = buffer_info->skb;
 #ifdef NETIF_F_TSO
-				unsigned int segs = skb_shinfo(skb)->gso_segs;
-				if (segs)
-					total_tx_packets += segs;
-#endif
-
+				unsigned int segs, bytecount;
+				segs = skb_shinfo(skb)->gso_segs ?: 1;
+				/* multiply data chunks by size of headers */
+				bytecount = ((segs - 1) * skb_headlen(skb)) +
+				            skb->len;
+				total_tx_packets += segs;
+				total_tx_bytes += bytecount;
+#else
 				total_tx_packets++;
 				total_tx_bytes += skb->len;
+#endif
 			}
 			e1000_unmap_and_free_tx_resource(adapter, buffer_info);
 			tx_desc->upper.data = 0;
@@ -4044,7 +4048,10 @@ e1000_clean_tx_irq(struct e1000_adapter 
 #ifdef CONFIG_E1000_NAPI
 #define E1000_TX_WEIGHT 64
 		/* weight of a sort for tx, to avoid endless transmit cleanup */
-		if (count++ == E1000_TX_WEIGHT) break;
+		if (count++ == E1000_TX_WEIGHT) {
+			cleaned = FALSE;
+			break;
+		}
 #endif
 	}
 



---
Auke Kok <auke-jan.h.kok@intel.com>

  parent reply	other threads:[~2006-12-08 22:54 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2006-12-08 23:02 [PATCH 00/22] ixgb, e1000 fixes Kok, Auke
2006-12-08 23:02 ` [PATCH 01/22] e1000: Fix early TSO completion Kok, Auke
2006-12-11 14:36   ` Jeff Garzik
2006-12-08 23:02 ` [PATCH 02/22] ixgb: Maybe stop TX if not enough free descriptors Kok, Auke
2006-12-11 14:37   ` Jeff Garzik
2006-12-08 23:02 ` [PATCH 03/22] ixgb: Write RA register high word first, increment version Kok, Auke
2006-12-11 14:37   ` Jeff Garzik
2006-12-08 23:02 ` [PATCH 04/22] e1000: display FC link properties Kok, Auke
2006-12-08 23:02 ` [PATCH 05/22] e1000: dynamic itr code fixes Kok, Auke
2006-12-08 23:02 ` [PATCH 06/22] e1000: minor tso ipv6 cleanups Kok, Auke
2006-12-11 14:42   ` Jeff Garzik
2006-12-08 23:02 ` [PATCH 07/22] e1000: omit stats for broken counter in 82543 Kok, Auke
2006-12-11 14:42   ` Jeff Garzik
2006-12-08 23:03 ` [PATCH 08/22] e1000: Fix all manageability workarounds Kok, Auke
2006-12-11 14:41   ` Jeff Garzik
2006-12-08 23:03 ` [PATCH 09/22] e1000: Fix PBA allocation calculations Kok, Auke
2006-12-08 23:03 ` [PATCH 10/22] e1000: Make copybreak parameter changeable by user Kok, Auke
2006-12-08 23:03 ` [PATCH 11/22] e1000: Fire a link even interrupt instead of a watchdog at init Kok, Auke
2006-12-11 22:48   ` Shaw Vrana
2006-12-08 23:03 ` [PATCH 12/22] e1000: Fix powerdown link loss when forced to gig speed Kok, Auke
2006-12-11 14:46   ` Jeff Garzik
2006-12-08 23:03 ` [PATCH 13/22] e1000: disable TSO when debugging slab Kok, Auke
2006-12-11 14:47   ` Jeff Garzik
2006-12-15  2:42     ` Herbert Xu
2006-12-08 23:03 ` [PATCH 14/22] e1000: always turn on receives on esb2 Kok, Auke
2006-12-11 14:47   ` Jeff Garzik
2006-12-08 23:03 ` [PATCH 15/22] e1000: simplify msi specific interrupt handler Kok, Auke
2006-12-11 14:48   ` Jeff Garzik
2006-12-08 23:03 ` [PATCH 16/22] e1000: Rewrite parm handling code to warn when forcing autoneg to gig only Kok, Auke
2006-12-08 23:03 ` [PATCH 17/22] e1000: set max frame size before enabling new link settings Kok, Auke
2006-12-08 23:03 ` [PATCH 18/22] e1000: For older adapters, we know that the pci bus type is pci Kok, Auke
2006-12-08 23:03 ` [PATCH 19/22] e1000: Don't set tipg (fiber) timings for pci-e adapters to fix fiber output Kok, Auke
2006-12-08 23:03 ` Kok, Auke [this message]
2006-12-11 14:49   ` [PATCH 20/22] e1000: Fix poor quad port performance with NAPI Jeff Garzik
2006-12-08 23:03 ` [PATCH 21/22] e1000: Version increment 7.3.20-k2 and minor cleanups Kok, Auke
2006-12-11 14:50   ` Jeff Garzik
2006-12-08 23:03 ` [PATCH 22/22] # This series applies on GIT commit 0ae851352a87db3f829511816a2da227860bf585 Kok, Auke

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=20061208230329.30109.85896.stgit@gitlost.site \
    --to=auke-jan.h.kok@intel.com \
    --cc=auke@foo-projects.org \
    --cc=jesse.brandeburg@intel.com \
    --cc=jgarzik@pobox.com \
    --cc=john.ronciak@intel.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).