DPDK patches and discussions
 help / color / mirror / Atom feed
From: Andrew Rybchenko <arybchenko@solarflare.com>
To: <dev@dpdk.org>
Cc: <stable@dpdk.org>
Subject: [dpdk-dev] [PATCH 02/14] net/sfc: do not hold management event queue lock while MCDI
Date: Sun, 24 Dec 2017 10:46:32 +0000	[thread overview]
Message-ID: <1514112404-13398-3-git-send-email-arybchenko@solarflare.com> (raw)
In-Reply-To: <1514112404-13398-1-git-send-email-arybchenko@solarflare.com>

MCDI execution may require MCDI proxy handling which involves
management event queue polling. So, it is a bad idea to hold
managment event queue lock when MCDI is executed.

Event queue creation and destruction are MCDI operations.

Fixes: 4650ed44c120 ("net/sfc: support MCDI proxy")
Cc: stable@dpdk.org

Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
---
 drivers/net/sfc/sfc.h      | 22 ++++++++++++++++++++++
 drivers/net/sfc/sfc_ev.c   | 23 ++++++++++++++---------
 drivers/net/sfc/sfc_intr.c |  5 +++--
 3 files changed, 39 insertions(+), 11 deletions(-)

diff --git a/drivers/net/sfc/sfc.h b/drivers/net/sfc/sfc.h
index b72eba0..ef980a4 100644
--- a/drivers/net/sfc/sfc.h
+++ b/drivers/net/sfc/sfc.h
@@ -212,7 +212,29 @@ struct sfc_adapter {
 	unsigned int			evq_count;
 
 	unsigned int			mgmt_evq_index;
+	/*
+	 * The lock is used to serialise management event queue polling
+	 * which can be done from different context. Also the lock
+	 * guarantees that mgmt_evq_running is preserved while the lock
+	 * is held. It is used to serialise polling and start/stop
+	 * operations.
+	 *
+	 * Locks which may be held when the lock is acquired:
+	 *  - adapter lock, when:
+	 *    - device start/stop to change mgmt_evq_running
+	 *    - any control operations in client side MCDI proxy handling to
+	 *	poll management event queue waiting for proxy response
+	 *  - MCDI lock, when:
+	 *    - any control operations in client side MCDI proxy handling to
+	 *	poll management event queue waiting for proxy response
+	 *
+	 * Locks which are acquired with the lock held:
+	 *  - nic_lock, when:
+	 *    - MC event processing on management event queue polling
+	 *	(e.g. MC REBOOT or BADASSERT events)
+	 */
 	rte_spinlock_t			mgmt_evq_lock;
+	bool				mgmt_evq_running;
 	struct sfc_evq			*mgmt_evq;
 
 	unsigned int			rxq_count;
diff --git a/drivers/net/sfc/sfc_ev.c b/drivers/net/sfc/sfc_ev.c
index b29eb2f..5fbebbf 100644
--- a/drivers/net/sfc/sfc_ev.c
+++ b/drivers/net/sfc/sfc_ev.c
@@ -565,10 +565,8 @@ void
 sfc_ev_mgmt_qpoll(struct sfc_adapter *sa)
 {
 	if (rte_spinlock_trylock(&sa->mgmt_evq_lock)) {
-		struct sfc_evq *mgmt_evq = sa->mgmt_evq;
-
-		if (mgmt_evq->init_state == SFC_EVQ_STARTED)
-			sfc_ev_qpoll(mgmt_evq);
+		if (sa->mgmt_evq_running)
+			sfc_ev_qpoll(sa->mgmt_evq);
 
 		rte_spinlock_unlock(&sa->mgmt_evq_lock);
 	}
@@ -734,20 +732,26 @@ sfc_ev_start(struct sfc_adapter *sa)
 		goto fail_ev_init;
 
 	/* Start management EVQ used for global events */
-	rte_spinlock_lock(&sa->mgmt_evq_lock);
 
+	/*
+	 * Management event queue start polls the queue, but it cannot
+	 * interfere with other polling contexts since mgmt_evq_running
+	 * is false yet.
+	 */
 	rc = sfc_ev_qstart(sa->mgmt_evq, sa->mgmt_evq_index);
 	if (rc != 0)
 		goto fail_mgmt_evq_start;
 
+	rte_spinlock_lock(&sa->mgmt_evq_lock);
+	sa->mgmt_evq_running = true;
+	rte_spinlock_unlock(&sa->mgmt_evq_lock);
+
 	if (sa->intr.lsc_intr) {
 		rc = sfc_ev_qprime(sa->mgmt_evq);
 		if (rc != 0)
 			goto fail_mgmt_evq_prime;
 	}
 
-	rte_spinlock_unlock(&sa->mgmt_evq_lock);
-
 	/*
 	 * Start management EVQ polling. If interrupts are disabled
 	 * (not used), it is required to process link status change
@@ -767,7 +771,6 @@ sfc_ev_start(struct sfc_adapter *sa)
 	sfc_ev_qstop(sa->mgmt_evq);
 
 fail_mgmt_evq_start:
-	rte_spinlock_unlock(&sa->mgmt_evq_lock);
 	efx_ev_fini(sa->nic);
 
 fail_ev_init:
@@ -783,9 +786,11 @@ sfc_ev_stop(struct sfc_adapter *sa)
 	sfc_ev_mgmt_periodic_qpoll_stop(sa);
 
 	rte_spinlock_lock(&sa->mgmt_evq_lock);
-	sfc_ev_qstop(sa->mgmt_evq);
+	sa->mgmt_evq_running = false;
 	rte_spinlock_unlock(&sa->mgmt_evq_lock);
 
+	sfc_ev_qstop(sa->mgmt_evq);
+
 	efx_ev_fini(sa->nic);
 }
 
diff --git a/drivers/net/sfc/sfc_intr.c b/drivers/net/sfc/sfc_intr.c
index ee59cb1..de65f8c 100644
--- a/drivers/net/sfc/sfc_intr.c
+++ b/drivers/net/sfc/sfc_intr.c
@@ -57,8 +57,9 @@ sfc_intr_handle_mgmt_evq(struct sfc_adapter *sa)
 
 	evq = sa->mgmt_evq;
 
-	if (evq->init_state != SFC_EVQ_STARTED) {
-		sfc_log_init(sa, "interrupt on stopped EVQ %u", evq->evq_index);
+	if (!sa->mgmt_evq_running) {
+		sfc_log_init(sa, "interrupt on not running management EVQ %u",
+			     evq->evq_index);
 	} else {
 		sfc_ev_qpoll(evq);
 
-- 
2.7.4

  parent reply	other threads:[~2017-12-24 10:47 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-12-24 10:46 [dpdk-dev] [PATCH 00/14] net/sfc: support NVGRE, VXLAN and GENEVE tunnels Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 01/14] net/sfc: fix label name to be consistent Andrew Rybchenko
2017-12-24 10:46 ` Andrew Rybchenko [this message]
2017-12-24 10:46 ` [dpdk-dev] [PATCH 03/14] net/sfc: handle MC reboot event Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 04/14] net/sfc: retry port start to handle MC reboot in the middle Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 05/14] net/sfc/base: control RxQ scatter using flag instead of type Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 06/14] net/sfc/base: add function to create packed stream RxQ Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 07/14] net/sfc/base: allow to request inner classes for Rx packets Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 08/14] net/sfc/base: add API to control UDP tunnel ports Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 09/14] net/sfc: support UDP tunnel ports configuration Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 10/14] net/sfc: fix incorrect bitwise ORing of L3/L4 packet types Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 11/14] net/sfc: support VXLAN and NVGRE packet types classification Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 12/14] net/sfc: correct Rx checksum offloads for tunnel packets Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 13/14] net/sfc: support inner checksum offload on transmit Andrew Rybchenko
2017-12-24 10:46 ` [dpdk-dev] [PATCH 14/14] doc: add net/sfc tunnels support to release features Andrew Rybchenko
2018-01-09 17:31 ` [dpdk-dev] [PATCH 00/14] net/sfc: support NVGRE, VXLAN and GENEVE tunnels 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=1514112404-13398-3-git-send-email-arybchenko@solarflare.com \
    --to=arybchenko@solarflare.com \
    --cc=dev@dpdk.org \
    --cc=stable@dpdk.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).