[net-next,1/2] net: Busy polling should ignore sender CPUs

Submitted by Alexander Duyck on March 20, 2017, 9:48 p.m.

Details

Message ID 20170320214848.13922.79201.stgit@localhost.localdomain
State Changes Requested
Delegated to: David Miller
Headers show

Commit Message

Alexander Duyck March 20, 2017, 9:48 p.m.
From: Alexander Duyck <alexander.h.duyck@intel.com>

This patch is a cleanup/fix for NAPI IDs following the changes that made it
so that sender_cpu and napi_id were doing a better job of sharing the same
location in the sk_buff.

One issue I found is that we weren't validating the napi_id as being valid
before we started trying to setup the busy polling.  This change corrects
that by using the MIN_NAPI_ID value that is now used in both allocating the
NAPI IDs, as well as validating them.

Fixes: 52bd2d62ce675 ("net: better skb->sender_cpu and skb->napi_id cohabitation")
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
---
 include/net/busy_poll.h |   11 +++++++++--
 net/core/dev.c          |    6 +++---
 2 files changed, 12 insertions(+), 5 deletions(-)

Comments

Eric Dumazet March 20, 2017, 10:16 p.m.
On Mon, 2017-03-20 at 14:48 -0700, Alexander Duyck wrote:
> From: Alexander Duyck <alexander.h.duyck@intel.com>
> 
> This patch is a cleanup/fix for NAPI IDs following the changes that made it
> so that sender_cpu and napi_id were doing a better job of sharing the same
> location in the sk_buff.
> 
> One issue I found is that we weren't validating the napi_id as being valid
> before we started trying to setup the busy polling.  This change corrects
> that by using the MIN_NAPI_ID value that is now used in both allocating the
> NAPI IDs, as well as validating them.
> 
> Fixes: 52bd2d62ce675 ("net: better skb->sender_cpu and skb->napi_id cohabitation")
> Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
> ---

This Fixes: tag seems not really needed here.

If really busy polling is attempted to a socket with a <wrong> napi id,
nothing bad happens. This fits the advisory model of busy polling...

Otherwise, your patch would be a candidate for net tree.

Also note that as soon as sk_can_busy_loop(sk) returns some status,
another cpu might already have changed sk->sk_napi_id to something else,
possibly with a <wrong> napi id again.

If your upcoming code depends on sk->sk_napi_id being verified, then
you need to read it once.

Patch hide | download patch | download mbox

diff --git a/include/net/busy_poll.h b/include/net/busy_poll.h
index c0452de83086..edf1310212a1 100644
--- a/include/net/busy_poll.h
+++ b/include/net/busy_poll.h
@@ -35,6 +35,12 @@ 
 extern unsigned int sysctl_net_busy_read __read_mostly;
 extern unsigned int sysctl_net_busy_poll __read_mostly;
 
+/*		0 - Reserved to indicate value not set
+ *     1..NR_CPUS - Reserved for sender_cpu
+ *  NR_CPUS+1..~0 - Region available for NAPI IDs
+ */
+#define MIN_NAPI_ID ((unsigned int)(NR_CPUS + 1))
+
 static inline bool net_busy_loop_on(void)
 {
 	return sysctl_net_busy_poll;
@@ -58,10 +64,11 @@  static inline unsigned long busy_loop_end_time(void)
 
 static inline bool sk_can_busy_loop(const struct sock *sk)
 {
-	return sk->sk_ll_usec && sk->sk_napi_id && !signal_pending(current);
+	return sk->sk_ll_usec &&
+	       (sk->sk_napi_id >= MIN_NAPI_ID) &&
+	       !signal_pending(current);
 }
 
-
 static inline bool busy_loop_timeout(unsigned long end_time)
 {
 	unsigned long now = busy_loop_us_clock();
diff --git a/net/core/dev.c b/net/core/dev.c
index 7869ae3837ca..5bbe30c08a5b 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -5143,10 +5143,10 @@  static void napi_hash_add(struct napi_struct *napi)
 
 	spin_lock(&napi_hash_lock);
 
-	/* 0..NR_CPUS+1 range is reserved for sender_cpu use */
+	/* 0..NR_CPUS range is reserved for sender_cpu use */
 	do {
-		if (unlikely(++napi_gen_id < NR_CPUS + 1))
-			napi_gen_id = NR_CPUS + 1;
+		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
+			napi_gen_id = MIN_NAPI_ID;
 	} while (napi_by_id(napi_gen_id));
 	napi->napi_id = napi_gen_id;