net: allow to call netif_reset_xps_queues() under cpus_read_lock

The definition of static_key_slow_inc() has cpus_read_lock in place. In the
virtio_net driver, XPS queues are initialized after setting the queue:cpu
affinity in virtnet_set_affinity() which is already protected within
cpus_read_lock. Lockdep prints a warning when we are trying to acquire
cpus_read_lock when it is already held.

This patch adds an ability to call __netif_set_xps_queue under
cpus_read_lock().
Acked-by: Jason Wang <jasowang@redhat.com>

============================================
WARNING: possible recursive locking detected
4.18.0-rc3-next-20180703+ #1 Not tainted
--------------------------------------------
swapper/0/1 is trying to acquire lock:
00000000cf973d46 (cpu_hotplug_lock.rw_sem){++++}, at: static_key_slow_inc+0xe/0x20

but task is already holding lock:
00000000cf973d46 (cpu_hotplug_lock.rw_sem){++++}, at: init_vqs+0x513/0x5a0

other info that might help us debug this:
 Possible unsafe locking scenario:

       CPU0
       ----
  lock(cpu_hotplug_lock.rw_sem);
  lock(cpu_hotplug_lock.rw_sem);

 *** DEADLOCK ***

 May be due to missing lock nesting notation

3 locks held by swapper/0/1:
 #0: 00000000244bc7da (&dev->mutex){....}, at: __driver_attach+0x5a/0x110
 #1: 00000000cf973d46 (cpu_hotplug_lock.rw_sem){++++}, at: init_vqs+0x513/0x5a0
 #2: 000000005cd8463f (xps_map_mutex){+.+.}, at: __netif_set_xps_queue+0x8d/0xc60

v2: move cpus_read_lock() out of __netif_set_xps_queue()

Cc: "Nambiar, Amritha" <amritha.nambiar@intel.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Fixes: 8af2c06ff4b1 ("net-sysfs: Add interface for Rx queue(s) map per Tx queue")

Signed-off-by: Andrei Vagin <avagin@gmail.com>

Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 22 insertions, 6 deletions

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 62311dde6e71..39a7f4452587 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -1903,9 +1903,11 @@ static void virtnet_set_affinity(struct virtnet_info *vi)
 
         i = 0;
         for_each_online_cpu(cpu) {
+                const unsigned long *mask = cpumask_bits(cpumask_of(cpu));
+
                 virtqueue_set_affinity(vi->rq[i].vq, cpu);
                 virtqueue_set_affinity(vi->sq[i].vq, cpu);
-                netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
+                __netif_set_xps_queue(vi->dev, mask, i, false);
                 i++;
         }
 
diff --git a/net/core/dev.c b/net/core/dev.c
index f68122f0ab02..325fc5088370 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2176,6 +2176,7 @@ static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
         if (!static_key_false(&xps_needed))
                 return;
 
+        cpus_read_lock();
         mutex_lock(&xps_map_mutex);
 
         if (static_key_false(&xps_rxqs_needed)) {
@@ -2199,10 +2200,11 @@ static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
 
 out_no_maps:
         if (static_key_enabled(&xps_rxqs_needed))
-                static_key_slow_dec(&xps_rxqs_needed);
+                static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
 
-        static_key_slow_dec(&xps_needed);
+        static_key_slow_dec_cpuslocked(&xps_needed);
         mutex_unlock(&xps_map_mutex);
+        cpus_read_unlock();
 }
 
 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
@@ -2250,6 +2252,7 @@ static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
         return new_map;
 }
 
+/* Must be called under cpus_read_lock */
 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
                           u16 index, bool is_rxqs_map)
 {
@@ -2317,9 +2320,9 @@ int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
         if (!new_dev_maps)
                 goto out_no_new_maps;
 
-        static_key_slow_inc(&xps_needed);
+        static_key_slow_inc_cpuslocked(&xps_needed);
         if (is_rxqs_map)
-                static_key_slow_inc(&xps_rxqs_needed);
+                static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
 
         for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
              j < nr_ids;) {
@@ -2448,11 +2451,18 @@ error:
         kfree(new_dev_maps);
         return -ENOMEM;
 }
+EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
 
 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
                         u16 index)
 {
-        return __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
+        int ret;
+
+        cpus_read_lock();
+        ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
+        cpus_read_unlock();
+
+        return ret;
 }
 EXPORT_SYMBOL(netif_set_xps_queue);
 
diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c
index 0a95bcf64cdc..bd67c4d0fcfd 100644
--- a/net/core/net-sysfs.c
+++ b/net/core/net-sysfs.c
@@ -26,6 +26,7 @@
 #include <linux/pm_runtime.h>
 #include <linux/of.h>
 #include <linux/of_net.h>
+#include <linux/cpu.h>
 
 #include "net-sysfs.h"
 
@@ -1400,7 +1401,10 @@ static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
                 return err;
         }
 
+        cpus_read_lock();
         err = __netif_set_xps_queue(dev, mask, index, true);
+        cpus_read_unlock();
+
         kfree(mask);
         return err ? : len;
 }