netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Ben Hutchings <bhutchings@solarflare.com>
To: Jeff Garzik <jgarzik@pobox.com>
Cc: netdev@vger.kernel.org, linux-net-drivers@solarflare.com
Subject: [PATCH 22/52] sfc: Move CPU counting for RSS into a separate function, efx_wanted_rx_queues()
Date: Mon, 1 Sep 2008 12:47:33 +0100	[thread overview]
Message-ID: <20080901114732.GP7908@solarflare.com> (raw)
In-Reply-To: <20080901111621.GT7908@solarflare.com>


Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
---
 drivers/net/sfc/efx.c |   60 ++++++++++++++++++++++++++++--------------------
 1 files changed, 35 insertions(+), 25 deletions(-)

diff --git a/drivers/net/sfc/efx.c b/drivers/net/sfc/efx.c
index 9f1ac3a..487df94 100644
--- a/drivers/net/sfc/efx.c
+++ b/drivers/net/sfc/efx.c
@@ -819,38 +819,48 @@ static void efx_fini_io(struct efx_nic *efx)
 	pci_disable_device(efx->pci_dev);
 }
 
-/* Probe the number and type of interrupts we are able to obtain. */
+/* Get number of RX queues wanted.  Return number of online CPU
+ * packages in the expectation that an IRQ balancer will spread
+ * interrupts across them. */
+static int efx_wanted_rx_queues(void)
+{
+	cpumask_t core_mask;
+	int count;
+	int cpu;
+
+	cpus_clear(core_mask);
+	count = 0;
+	for_each_online_cpu(cpu) {
+		if (!cpu_isset(cpu, core_mask)) {
+			++count;
+			cpus_or(core_mask, core_mask,
+				topology_core_siblings(cpu));
+		}
+	}
+
+	return count;
+}
+
+/* Probe the number and type of interrupts we are able to obtain, and
+ * the resulting numbers of channels and RX queues.
+ */
 static void efx_probe_interrupts(struct efx_nic *efx)
 {
-	int max_channel = efx->type->phys_addr_channels - 1;
-	struct msix_entry xentries[EFX_MAX_CHANNELS];
+	int max_channels =
+		min_t(int, efx->type->phys_addr_channels, EFX_MAX_CHANNELS);
 	int rc, i;
 
 	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
-		BUG_ON(!pci_find_capability(efx->pci_dev, PCI_CAP_ID_MSIX));
-
-		if (rss_cpus == 0) {
-			cpumask_t core_mask;
-			int cpu;
-
-			cpus_clear(core_mask);
-			efx->rss_queues = 0;
-			for_each_online_cpu(cpu) {
-				if (!cpu_isset(cpu, core_mask)) {
-					++efx->rss_queues;
-					cpus_or(core_mask, core_mask,
-						topology_core_siblings(cpu));
-				}
-			}
-		} else {
-			efx->rss_queues = rss_cpus;
-		}
+		struct msix_entry xentries[EFX_MAX_CHANNELS];
+		int wanted_ints;
 
-		efx->rss_queues = min(efx->rss_queues, max_channel + 1);
-		efx->rss_queues = min(efx->rss_queues, EFX_MAX_CHANNELS);
+		/* We want one RX queue and interrupt per CPU package
+		 * (or as specified by the rss_cpus module parameter).
+		 * We will need one channel per interrupt.
+		 */
+		wanted_ints = rss_cpus ? rss_cpus : efx_wanted_rx_queues();
+		efx->rss_queues = min(wanted_ints, max_channels);
 
-		/* Request maximum number of MSI interrupts, and fill out
-		 * the channel interrupt information the allowed allocation */
 		for (i = 0; i < efx->rss_queues; i++)
 			xentries[i].entry = i;
 		rc = pci_enable_msix(efx->pci_dev, xentries, efx->rss_queues);
-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.


  parent reply	other threads:[~2008-09-01 11:47 UTC|newest]

Thread overview: 55+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-09-01 11:16 [PATCH 00/52] sfc: Changes for 2.6.28 Ben Hutchings
2008-09-01 11:43 ` [PATCH 01/52] sfc: Replace net_dev->priv with netdev_priv(net_dev) Ben Hutchings
2008-09-01 11:43 ` [PATCH 02/52] sfc: Change first parameter type of {set,clear}_bit_le() to unsigned Ben Hutchings
2008-09-01 11:44 ` [PATCH 03/52] sfc: Remove unused field efx_channel::reset_work Ben Hutchings
2008-09-01 11:44 ` [PATCH 04/52] sfc: Use separate hardware TX queues to select checksum generation Ben Hutchings
2008-09-01 11:45 ` [PATCH 05/52] sfc: Avoid mangling error codes in efx_test_loopback() Ben Hutchings
2008-09-01 11:45 ` [PATCH 06/52] sfc: Reduce delays in SFE4001 initialisation Ben Hutchings
2008-09-01 11:46 ` [PATCH 07/52] sfc: Remove mistaken hardware workaround Ben Hutchings
2008-09-01 11:46 ` [PATCH 08/52] sfc: XMAC statistics fix-ups Ben Hutchings
2008-09-01 11:46 ` [PATCH 09/52] sfc: Remove inclusion of workarounds.h from efx.c Ben Hutchings
2008-09-01 11:46 ` [PATCH 10/52] sfc: Reverse the XOFF/XON pause frame control fifo thresholds Ben Hutchings
2008-09-01 11:46 ` [PATCH 11/52] sfc: Reduce log level for XGXS lane status Ben Hutchings
2008-09-01 11:46 ` [PATCH 12/52] sfc: Self-test reporting cleanup Ben Hutchings
2008-09-01 11:46 ` [PATCH 13/52] sfc: Speed up loopback self-test Ben Hutchings
2008-09-01 11:46 ` [PATCH 14/52] sfc: Don't leak PCI DMA maps in the TSO code when the queue fills up Ben Hutchings
2008-09-01 11:46 ` [PATCH 15/52] sfc: Use pci_map_single() to map the skb header when doing TSO Ben Hutchings
2008-09-01 11:46 ` [PATCH 16/52] sfc: Reduce the size of struct efx_tx_buffer Ben Hutchings
2008-09-01 11:46 ` [PATCH 17/52] sfc: Use explicit bool for boolean variables, parameters and return values Ben Hutchings
2008-09-01 11:46 ` [PATCH 18/52] sfc: Set net_device::vlan_features appropriately Ben Hutchings
2008-09-01 11:47 ` [PATCH 19/52] sfc: Cleaned up struct tso_state fields Ben Hutchings
2008-09-01 11:47 ` [PATCH 20/52] sfc: Removed forced inlining of long functions Ben Hutchings
2008-09-01 11:47 ` [PATCH 21/52] sfc: Export boot configuration in EEPROM through ethtool Ben Hutchings
2008-09-01 11:47 ` Ben Hutchings [this message]
2008-09-01 11:47 ` [PATCH 23/52] sfc: Remove efx_channel::has_interrupt Ben Hutchings
2008-09-01 11:47 ` [PATCH 24/52] sfc: Cleanup RX queue information Ben Hutchings
2008-09-01 11:47 ` [PATCH 25/52] sfc: Remove initialisation of RX_FILTER_CTL_REG.NUM_KER Ben Hutchings
2008-09-01 11:47 ` [PATCH 26/52] sfc: Make efx_for_each_channel_rx_queue() more efficient Ben Hutchings
2008-09-01 11:48 ` [PATCH 27/52] sfc: Remove efx_channel::evqnum field Ben Hutchings
2008-09-01 11:48 ` [PATCH 28/52] sfc: Cleanup RX event processing Ben Hutchings
2008-09-01 11:48 ` [PATCH 29/52] sfc: Implement get_sset_count, replacing get_stats_count and self_test_count Ben Hutchings
2008-09-01 11:48 ` [PATCH 30/52] sfc: Make PHY flash mode a device attribute, not a module parameter Ben Hutchings
2008-09-01 11:48 ` [PATCH 31/52] sfc: Do not call netif_{stop,wake}_queue() before register_netdev Ben Hutchings
2008-09-01 11:48 ` [PATCH 32/52] sfc: Enable TSO for 802.1q VLAN devices Ben Hutchings
2008-09-01 11:48 ` [PATCH 33/52] sfc: Remove efx_nic_dummy_op_int() as redundant with efx_port_dummy_op_int() Ben Hutchings
2008-09-01 11:48 ` [PATCH 34/52] sfc: Remove remnants of multi-port abstraction for MAC registers Ben Hutchings
2008-09-01 11:48 ` [PATCH 35/52] sfc: Remove some unreachable error paths Ben Hutchings
2008-09-01 11:48 ` [PATCH 36/52] sfc: Cleanup reset code Ben Hutchings
2008-09-01 11:48 ` [PATCH 37/52] sfc: Rework the bitfield header so that we can identify fields by bit number Ben Hutchings
2008-09-01 11:49 ` [PATCH 38/52] sfc: Extend self-tests Ben Hutchings
2008-09-01 11:49 ` [PATCH 39/52] sfc: Remove the STATE_RESETTING flag Ben Hutchings
2008-09-01 11:49 ` [PATCH 40/52] sfc: Rework efx_set_multicast_hash() Ben Hutchings
2008-09-03 13:54   ` Jeff Garzik
2008-09-01 11:49 ` [PATCH 42/52] sfc: Fix memory BAR release call on error path Ben Hutchings
2008-09-01 11:49 ` [PATCH 43/52] sfc: Remove workaround for old firmware bug Ben Hutchings
2008-09-01 11:49 ` [PATCH 44/52] sfc: Serialise tenxpress_special_reset() with statistics fetches Ben Hutchings
2008-09-01 11:49 ` [PATCH 45/52] sfc: Don't use EFX_OWORD_FIELD on an event (64-bit, quad-word) Ben Hutchings
2008-09-01 11:49 ` [PATCH 46/52] sfc: Make queue flushes more reliable Ben Hutchings
2008-09-01 11:49 ` [PATCH 47/52] sfc: Don't include net_driver.h from falcon_io.h Ben Hutchings
2008-09-01 11:50 ` [PATCH 48/52] sfc: Stop generating bogus events in tenxpress_check_hw() Ben Hutchings
2008-09-01 11:50 ` [PATCH 49/52] sfc: Insert read memory barrier after checking MAC statistics flag Ben Hutchings
2008-09-01 11:50 ` [PATCH 50/52] sfc: Disable interrupts after a fatal interrupt occurs until reset Ben Hutchings
2008-09-01 11:50 ` [PATCH 51/52] sfc: Remove obsolete comment about PCI modes Ben Hutchings
2008-09-01 11:50 ` [PATCH 52/52] sfc: Use CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS where appropriate Ben Hutchings
2008-09-03 14:07 ` [PATCH 41/52] sfc: Add check for memory allocation failure in falcon_probe_nic() Ben Hutchings
2008-09-13 19:29   ` Jeff Garzik

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=20080901114732.GP7908@solarflare.com \
    --to=bhutchings@solarflare.com \
    --cc=jgarzik@pobox.com \
    --cc=linux-net-drivers@solarflare.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).