[PATCH v2 20/45] net: Use get/put_online_cpus_atomic() to prevent CPU offline

[Srivatsa S. Bhat]

Once stop_machine() is gone from the CPU offline path, we won't be able
to depend on disabling preemption to prevent CPUs from going offline
from under us.

Use the get/put_online_cpus_atomic() APIs to prevent CPUs from going
offline, while invoking from atomic context.

Cc: "David S. Miller" <davem at davemloft.net>
Cc: Eric Dumazet <edumazet at google.com>
Cc: Alexander Duyck <alexander.h.duyck at intel.com>
Cc: Cong Wang <xiyou.wangcong at gmail.com>
Cc: Ben Hutchings <bhutchings at solarflare.com>
Cc: netdev at vger.kernel.org
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat at linux.vnet.ibm.com>
---

 net/core/dev.c |    9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index fc1e289..90519e9 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3141,7 +3141,7 @@ int netif_rx(struct sk_buff *skb)
 		struct rps_dev_flow voidflow, *rflow = &voidflow;
 		int cpu;
 
-		preempt_disable();
+		get_online_cpus_atomic();
 		rcu_read_lock();
 
 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
@@ -3151,7 +3151,7 @@ int netif_rx(struct sk_buff *skb)
 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
 
 		rcu_read_unlock();
-		preempt_enable();
+		put_online_cpus_atomic();
 	} else
 #endif
 	{
@@ -3570,6 +3570,7 @@ int netif_receive_skb(struct sk_buff *skb)
 		struct rps_dev_flow voidflow, *rflow = &voidflow;
 		int cpu, ret;
 
+		get_online_cpus_atomic();
 		rcu_read_lock();
 
 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
@@ -3577,9 +3578,11 @@ int netif_receive_skb(struct sk_buff *skb)
 		if (cpu >= 0) {
 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
 			rcu_read_unlock();
+			put_online_cpus_atomic();
 			return ret;
 		}
 		rcu_read_unlock();
+		put_online_cpus_atomic();
 	}
 #endif
 	return __netif_receive_skb(skb);
@@ -3957,6 +3960,7 @@ static void net_rps_action_and_irq_enable(struct softnet_data *sd)
 		local_irq_enable();
 
 		/* Send pending IPI's to kick RPS processing on remote cpus. */
+		get_online_cpus_atomic();
 		while (remsd) {
 			struct softnet_data *next = remsd->rps_ipi_next;
 
@@ -3965,6 +3969,7 @@ static void net_rps_action_and_irq_enable(struct softnet_data *sd)
 							   &remsd->csd, 0);
 			remsd = next;
 		}
+		put_online_cpus_atomic();
 	} else
 #endif
 		local_irq_enable();