Andrew,
On Tue, Mar 18, 2014 at 01:55:01PM +0400, Andrew Rybchenko wrote:
A> sfxge: limit software Tx queue size
A>
A> Previous implementation limits put queue size only (when Tx lock can't
A> be acquired),
A> but get queue may grow unboundedly which results in mbuf pools
A> exhaustion and
A> latency growth.
A>
A> Submitted-by: Andrew Rybchenko <andrew.rybche...@oktetlabs.ru>
A> Sponsored by: Solarflare Communications, Inc.
The interaction between sfxge_tx_qdpl_put() and sfxge_tx_packet_add()
is quite complex and I couldn't resist from suggesting you to
simplify the code.
Can you please look into attached patch?
- Inline sfxge_tx_qdpl_put() into sfxge_tx_packet_add().
- Simplify the 'locked' logic.
- Add your PATCH 1/6, the mbuf leak fix.
- Add your PATCH 2/6, the SFXGE_TX_DPL_GET_PKT_LIMIT_DEFAULT check.
--
Totus tuus, Glebius.
Index: sfxge_tx.c
===================================================================
--- sfxge_tx.c (revision 263297)
+++ sfxge_tx.c (working copy)
@@ -437,32 +437,39 @@ sfxge_tx_qdpl_service(struct sfxge_txq *txq)
}
/*
- * Put a packet on the deferred packet list.
- *
- * If we are called with the txq lock held, we put the packet on the "get
- * list", otherwise we atomically push it on the "put list". The swizzle
- * function takes care of ordering.
- *
- * The length of the put list is bounded by SFXGE_TX_MAX_DEFFERED. We
- * overload the csum_data field in the mbuf to keep track of this length
- * because there is no cheap alternative to avoid races.
+ * Called from if_transmit - will try to grab the txq lock and enqueue to the
+ * put list if it succeeds, otherwise will push onto the defer list.
*/
-static inline int
-sfxge_tx_qdpl_put(struct sfxge_txq *txq, struct mbuf *mbuf, int locked)
+int
+sfxge_tx_packet_add(struct sfxge_txq *txq, struct mbuf *m)
{
- struct sfxge_tx_dpl *stdp;
+ struct sfxge_tx_dpl *stdp = &txq->dpl;
- stdp = &txq->dpl;
+ KASSERT(m->m_nextpkt == NULL, ("m->m_nextpkt != NULL"));
- KASSERT(mbuf->m_nextpkt == NULL, ("mbuf->m_nextpkt != NULL"));
+ /*
+ * Try to grab the txq lock. If we are able to get the lock,
+ * the packet will be appended to the "get list" of the deferred
+ * packet list. Otherwise, it will be pushed on the "put list".
+ * The swizzle function takes care of ordering.
+ *
+ * The length of the get and put lists are bounded by
+ * SFXGE_TX_DPL_GET_PKT_LIMIT_DEFAULT and
+ * SFXGE_TX_DPL_PUT_PKT_LIMIT_DEFAULT respectively.
+ * We overload the csum_data field in the mbuf to keep track of this
+ * length because there is no cheap alternative to avoid races.
+ */
+ if (mtx_trylock(&txq->lock)) {
+ sfxge_tx_qdpl_swizzle(txq);
- if (locked) {
- mtx_assert(&txq->lock, MA_OWNED);
+ if (stdp->std_count >= SFXGE_TX_DPL_GET_PKT_LIMIT_DEFAULT) {
+ mtx_unlock(&txq->lock);
+ m_freem(m);
+ return (ENOBUFS);
+ }
- sfxge_tx_qdpl_swizzle(txq);
-
- *(stdp->std_getp) = mbuf;
- stdp->std_getp = &mbuf->m_nextpkt;
+ *(stdp->std_getp) = m;
+ stdp->std_getp = &m->m_nextpkt;
stdp->std_count++;
} else {
volatile uintptr_t *putp;
@@ -471,7 +478,7 @@ sfxge_tx_qdpl_service(struct sfxge_txq *txq)
unsigned old_len;
putp = &stdp->std_put;
- new = (uintptr_t)mbuf;
+ new = (uintptr_t)m;
do {
old = *putp;
@@ -480,64 +487,30 @@ sfxge_tx_qdpl_service(struct sfxge_txq *txq)
old_len = mp->m_pkthdr.csum_data;
} else
old_len = 0;
- if (old_len >= SFXGE_TX_MAX_DEFERRED)
- return ENOBUFS;
- mbuf->m_pkthdr.csum_data = old_len + 1;
- mbuf->m_nextpkt = (void *)old;
+ if (old_len >= SFXGE_TX_DPL_PUT_PKT_LIMIT_DEFAULT) {
+ m_freem(m);
+ return (ENOBUFS);
+ }
+ m->m_pkthdr.csum_data = old_len + 1;
+ m->m_nextpkt = (void *)old;
} while (atomic_cmpset_ptr(putp, old, new) == 0);
- }
- return (0);
-}
-
-/*
- * Called from if_transmit - will try to grab the txq lock and enqueue to the
- * put list if it succeeds, otherwise will push onto the defer list.
- */
-int
-sfxge_tx_packet_add(struct sfxge_txq *txq, struct mbuf *m)
-{
- int locked;
- int rc;
-
- /*
- * Try to grab the txq lock. If we are able to get the lock,
- * the packet will be appended to the "get list" of the deferred
- * packet list. Otherwise, it will be pushed on the "put list".
- */
- locked = mtx_trylock(&txq->lock);
-
- /*
- * Can only fail if we weren't able to get the lock.
- */
- if (sfxge_tx_qdpl_put(txq, m, locked) != 0) {
- KASSERT(!locked,
- ("sfxge_tx_qdpl_put() failed locked"));
- rc = ENOBUFS;
- goto fail;
+ /*
+ * Again try to grab the lock.
+ *
+ * If we are able to get the lock, we need to process the
+ * deferred packet list. If we are not able to get the lock,
+ * another thread is processing the list, and we return.
+ */
+ if (mtx_trylock(&txq->lock) == 0)
+ return (0);
}
- /*
- * Try to grab the lock again.
- *
- * If we are able to get the lock, we need to process the deferred
- * packet list. If we are not able to get the lock, another thread
- * is processing the list.
- */
- if (!locked)
- locked = mtx_trylock(&txq->lock);
+ /* Try to service the list. */
+ sfxge_tx_qdpl_service(txq);
+ /* Lock has been dropped. */
- if (locked) {
- /* Try to service the list. */
- sfxge_tx_qdpl_service(txq);
- /* Lock has been dropped. */
- }
-
return (0);
-
-fail:
- return (rc);
-
}
static void
Index: sfxge_tx.h
===================================================================
--- sfxge_tx.h (revision 263297)
+++ sfxge_tx.h (working copy)
@@ -75,7 +75,8 @@ struct sfxge_tx_mapping {
enum sfxge_tx_buf_flags flags;
};
-#define SFXGE_TX_MAX_DEFERRED 64
+#define SFXGE_TX_DPL_GET_PKT_LIMIT_DEFAULT 64
+#define SFXGE_TX_DPL_PUT_PKT_LIMIT_DEFAULT 64
/*
* Deferred packet list.
_______________________________________________
freebsd-net@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
