netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Sasha Levin <sashal@kernel.org>
To: linux-kernel@vger.kernel.org, stable@vger.kernel.org
Cc: Edward Cree <ecree@solarflare.com>,
	Jakub Kicinski <kuba@kernel.org>, Sasha Levin <sashal@kernel.org>,
	netdev@vger.kernel.org
Subject: [PATCH AUTOSEL 5.9 054/147] sfc: add and use efx_tx_send_pending in tx.c
Date: Mon, 26 Oct 2020 19:47:32 -0400	[thread overview]
Message-ID: <20201026234905.1022767-54-sashal@kernel.org> (raw)
In-Reply-To: <20201026234905.1022767-1-sashal@kernel.org>

From: Edward Cree <ecree@solarflare.com>

[ Upstream commit 1c0544d24927e4fad04f858216b8ea767a3bd123 ]

Instead of using efx_tx_queue_partner(), which relies on the assumption
 that tx_queues_per_channel is 2, efx_tx_send_pending() iterates over
 txqs with efx_for_each_channel_tx_queue().
We unconditionally set tx_queue->xmit_pending (renamed from
 xmit_more_available), then condition on xmit_more for the call to
 efx_tx_send_pending(), which will clear xmit_pending.  Thus, after an
 xmit_more TX, the doorbell is un-rung and xmit_pending is true.

Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/net/ethernet/sfc/ef10.c       |  2 +-
 drivers/net/ethernet/sfc/ef100_tx.c   | 14 +++----
 drivers/net/ethernet/sfc/farch.c      |  2 +-
 drivers/net/ethernet/sfc/net_driver.h |  4 +-
 drivers/net/ethernet/sfc/tx.c         | 59 ++++++++++++++-------------
 drivers/net/ethernet/sfc/tx_common.c  |  4 +-
 6 files changed, 44 insertions(+), 41 deletions(-)

diff --git a/drivers/net/ethernet/sfc/ef10.c b/drivers/net/ethernet/sfc/ef10.c
index 4b0b2cf026a52..ad05710883f85 100644
--- a/drivers/net/ethernet/sfc/ef10.c
+++ b/drivers/net/ethernet/sfc/ef10.c
@@ -2367,7 +2367,7 @@ static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
 	unsigned int write_ptr;
 	efx_qword_t *txd;
 
-	tx_queue->xmit_more_available = false;
+	tx_queue->xmit_pending = false;
 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
 		return;
 
diff --git a/drivers/net/ethernet/sfc/ef100_tx.c b/drivers/net/ethernet/sfc/ef100_tx.c
index a09546e434085..8d478c5e720e1 100644
--- a/drivers/net/ethernet/sfc/ef100_tx.c
+++ b/drivers/net/ethernet/sfc/ef100_tx.c
@@ -131,7 +131,7 @@ void ef100_notify_tx_desc(struct efx_tx_queue *tx_queue)
 	efx_writed_page(tx_queue->efx, &reg,
 			ER_GZ_TX_RING_DOORBELL, tx_queue->queue);
 	tx_queue->notify_count = tx_queue->write_count;
-	tx_queue->xmit_more_available = false;
+	tx_queue->xmit_pending = false;
 }
 
 static void ef100_tx_push_buffers(struct efx_tx_queue *tx_queue)
@@ -373,14 +373,14 @@ int ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 	}
 
 	if (__netdev_tx_sent_queue(tx_queue->core_txq, skb->len, xmit_more))
-		tx_queue->xmit_more_available = false; /* push doorbell */
+		tx_queue->xmit_pending = false; /* push doorbell */
 	else if (tx_queue->write_count - tx_queue->notify_count > 255)
 		/* Ensure we never push more than 256 packets at once */
-		tx_queue->xmit_more_available = false; /* push */
+		tx_queue->xmit_pending = false; /* push */
 	else
-		tx_queue->xmit_more_available = true; /* don't push yet */
+		tx_queue->xmit_pending = true; /* don't push yet */
 
-	if (!tx_queue->xmit_more_available)
+	if (!tx_queue->xmit_pending)
 		ef100_tx_push_buffers(tx_queue);
 
 	if (segments) {
@@ -400,9 +400,9 @@ int ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb)
 	/* If we're not expecting another transmit and we had something to push
 	 * on this queue then we need to push here to get the previous packets
 	 * out.  We only enter this branch from before the 'Update BQL' section
-	 * above, so xmit_more_available still refers to the old state.
+	 * above, so xmit_pending still refers to the old state.
 	 */
-	if (tx_queue->xmit_more_available && !xmit_more)
+	if (tx_queue->xmit_pending && !xmit_more)
 		ef100_tx_push_buffers(tx_queue);
 	return rc;
 }
diff --git a/drivers/net/ethernet/sfc/farch.c b/drivers/net/ethernet/sfc/farch.c
index 4002f9a3ae909..0cc7a37b63445 100644
--- a/drivers/net/ethernet/sfc/farch.c
+++ b/drivers/net/ethernet/sfc/farch.c
@@ -320,7 +320,7 @@ void efx_farch_tx_write(struct efx_tx_queue *tx_queue)
 	unsigned write_ptr;
 	unsigned old_write_count = tx_queue->write_count;
 
-	tx_queue->xmit_more_available = false;
+	tx_queue->xmit_pending = false;
 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
 		return;
 
diff --git a/drivers/net/ethernet/sfc/net_driver.h b/drivers/net/ethernet/sfc/net_driver.h
index 062462a138475..fb7290bbb135c 100644
--- a/drivers/net/ethernet/sfc/net_driver.h
+++ b/drivers/net/ethernet/sfc/net_driver.h
@@ -244,7 +244,7 @@ struct efx_tx_buffer {
  * @tso_fallbacks: Number of times TSO fallback used
  * @pushes: Number of times the TX push feature has been used
  * @pio_packets: Number of times the TX PIO feature has been used
- * @xmit_more_available: Are any packets waiting to be pushed to the NIC
+ * @xmit_pending: Are any packets waiting to be pushed to the NIC
  * @cb_packets: Number of times the TX copybreak feature has been used
  * @notify_count: Count of notified descriptors to the NIC
  * @empty_read_count: If the completion path has seen the queue as empty
@@ -292,7 +292,7 @@ struct efx_tx_queue {
 	unsigned int tso_fallbacks;
 	unsigned int pushes;
 	unsigned int pio_packets;
-	bool xmit_more_available;
+	bool xmit_pending;
 	unsigned int cb_packets;
 	unsigned int notify_count;
 	/* Statistics to supplement MAC stats */
diff --git a/drivers/net/ethernet/sfc/tx.c b/drivers/net/ethernet/sfc/tx.c
index 727201d5eb24e..c502d226371a9 100644
--- a/drivers/net/ethernet/sfc/tx.c
+++ b/drivers/net/ethernet/sfc/tx.c
@@ -268,6 +268,19 @@ static int efx_enqueue_skb_pio(struct efx_tx_queue *tx_queue,
 }
 #endif /* EFX_USE_PIO */
 
+/* Send any pending traffic for a channel. xmit_more is shared across all
+ * queues for a channel, so we must check all of them.
+ */
+static void efx_tx_send_pending(struct efx_channel *channel)
+{
+	struct efx_tx_queue *q;
+
+	efx_for_each_channel_tx_queue(q, channel) {
+		if (q->xmit_pending)
+			efx_nic_push_buffers(q);
+	}
+}
+
 /*
  * Add a socket buffer to a TX queue
  *
@@ -336,21 +349,11 @@ netdev_tx_t __efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb
 
 	efx_tx_maybe_stop_queue(tx_queue);
 
-	/* Pass off to hardware */
-	if (__netdev_tx_sent_queue(tx_queue->core_txq, skb_len, xmit_more)) {
-		struct efx_tx_queue *txq2 = efx_tx_queue_partner(tx_queue);
-
-		/* There could be packets left on the partner queue if
-		 * xmit_more was set. If we do not push those they
-		 * could be left for a long time and cause a netdev watchdog.
-		 */
-		if (txq2->xmit_more_available)
-			efx_nic_push_buffers(txq2);
+	tx_queue->xmit_pending = true;
 
-		efx_nic_push_buffers(tx_queue);
-	} else {
-		tx_queue->xmit_more_available = xmit_more;
-	}
+	/* Pass off to hardware */
+	if (__netdev_tx_sent_queue(tx_queue->core_txq, skb_len, xmit_more))
+		efx_tx_send_pending(tx_queue->channel);
 
 	if (segments) {
 		tx_queue->tso_bursts++;
@@ -371,14 +374,8 @@ netdev_tx_t __efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb
 	 * on this queue or a partner queue then we need to push here to get the
 	 * previous packets out.
 	 */
-	if (!xmit_more) {
-		struct efx_tx_queue *txq2 = efx_tx_queue_partner(tx_queue);
-
-		if (txq2->xmit_more_available)
-			efx_nic_push_buffers(txq2);
-
-		efx_nic_push_buffers(tx_queue);
-	}
+	if (!xmit_more)
+		efx_tx_send_pending(tx_queue->channel);
 
 	return NETDEV_TX_OK;
 }
@@ -489,18 +486,24 @@ netdev_tx_t efx_hard_start_xmit(struct sk_buff *skb,
 
 	EFX_WARN_ON_PARANOID(!netif_device_present(net_dev));
 
-	/* PTP "event" packet */
-	if (unlikely(efx_xmit_with_hwtstamp(skb)) &&
-	    unlikely(efx_ptp_is_ptp_tx(efx, skb))) {
-		return efx_ptp_tx(efx, skb);
-	}
-
 	index = skb_get_queue_mapping(skb);
 	type = skb->ip_summed == CHECKSUM_PARTIAL ? EFX_TXQ_TYPE_OFFLOAD : 0;
 	if (index >= efx->n_tx_channels) {
 		index -= efx->n_tx_channels;
 		type |= EFX_TXQ_TYPE_HIGHPRI;
 	}
+
+	/* PTP "event" packet */
+	if (unlikely(efx_xmit_with_hwtstamp(skb)) &&
+	    unlikely(efx_ptp_is_ptp_tx(efx, skb))) {
+		/* There may be existing transmits on the channel that are
+		 * waiting for this packet to trigger the doorbell write.
+		 * We need to send the packets at this point.
+		 */
+		efx_tx_send_pending(efx_get_tx_channel(efx, index));
+		return efx_ptp_tx(efx, skb);
+	}
+
 	tx_queue = efx_get_tx_queue(efx, index, type);
 
 	return __efx_enqueue_skb(tx_queue, skb);
diff --git a/drivers/net/ethernet/sfc/tx_common.c b/drivers/net/ethernet/sfc/tx_common.c
index 793e234819a8c..187d5c379a377 100644
--- a/drivers/net/ethernet/sfc/tx_common.c
+++ b/drivers/net/ethernet/sfc/tx_common.c
@@ -78,7 +78,7 @@ void efx_init_tx_queue(struct efx_tx_queue *tx_queue)
 	tx_queue->read_count = 0;
 	tx_queue->old_read_count = 0;
 	tx_queue->empty_read_count = 0 | EFX_EMPTY_COUNT_VALID;
-	tx_queue->xmit_more_available = false;
+	tx_queue->xmit_pending = false;
 	tx_queue->timestamping = (efx_ptp_use_mac_tx_timestamps(efx) &&
 				  tx_queue->channel == efx_ptp_channel(efx));
 	tx_queue->completed_timestamp_major = 0;
@@ -116,7 +116,7 @@ void efx_fini_tx_queue(struct efx_tx_queue *tx_queue)
 
 		++tx_queue->read_count;
 	}
-	tx_queue->xmit_more_available = false;
+	tx_queue->xmit_pending = false;
 	netdev_tx_reset_queue(tx_queue->core_txq);
 }
 
-- 
2.25.1


  parent reply	other threads:[~2020-10-26 23:50 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <20201026234905.1022767-1-sashal@kernel.org>
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 030/147] ath10k: fix retry packets update in station dump Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 035/147] ath10k: start recovery process when payload length exceeds max htc length for sdio Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 036/147] ath10k: fix VHT NSS calculation when STBC is enabled Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 053/147] brcmfmac: increase F2 watermark for BCM4329 Sasha Levin
2020-10-26 23:47 ` Sasha Levin [this message]
2020-10-27 13:24   ` [PATCH AUTOSEL 5.9 054/147] sfc: add and use efx_tx_send_pending in tx.c Edward Cree
2020-11-01 23:52     ` Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 060/147] bpf: Permit map_ptr arithmetic with opcode add and offset 0 Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 066/147] nfc: s3fwrn5: Add missing CRYPTO_HASH dependency Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 067/147] selftests/bpf: Define string const as global for test_sysctl_prog.c Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 069/147] samples/bpf: Fix possible deadlock in xdpsock Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 077/147] mac80211: add missing queue/hash initialization to 802.3 xmit Sasha Levin
2020-10-26 23:47 ` [PATCH AUTOSEL 5.9 079/147] SUNRPC: Mitigate cond_resched() in xprt_transmit() Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 082/147] can: flexcan: disable clocks during stop mode Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 088/147] octeontx2-af: fix LD CUSTOM LTYPE aliasing Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 090/147] brcmfmac: Fix warning message after dongle setup failed Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 091/147] ath11k: Use GFP_ATOMIC instead of GFP_KERNEL in ath11k_dp_htt_get_ppdu_desc Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 092/147] ath11k: fix warning caused by lockdep_assert_held Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 093/147] ath11k: change to disable softirqs for ath11k_regd_update to solve deadlock Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 094/147] drivers/net/wan/hdlc_fr: Correctly handle special skb->protocol values Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 107/147] bnxt_en: Log unknown link speed appropriately Sasha Levin
2020-10-26 23:48 ` [PATCH AUTOSEL 5.9 115/147] net: 9p: initialize sun_server.sun_path to have addr's value only when addr is valid Sasha Levin

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=20201026234905.1022767-54-sashal@kernel.org \
    --to=sashal@kernel.org \
    --cc=ecree@solarflare.com \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=stable@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).