net: introduce netdev_napi_exit()
Subsystems (2)
PATCH 1/1 net: introduce netdev_napi_exit()
After 1b23cdbd2bbc ("net: protect netdev->napi_list with netdev_lock()") it makes sense to iterate through dev->napi_list while holding the device lock.
Also call synchronize_net() at most one time.
--- net/core/dev.c | 23 +++++++++++++++++------ 1 file changed, 17 insertions(+), 6 deletions(-)
net/core/dev.c +17 -6
@@ -11568,6 +11568,22 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, }EXPORT_SYMBOL (alloc_netdev_mqs); +static void netdev_napi_exit (struct net_device *dev) +{ +if (!list_empty (&dev->napi_list)) { +struct napi_struct *p, *n; + +netdev_lock (dev); +list_for_each_entry_safe (p, n, &dev->napi_list, dev_list) +__netif_napi_del_locked (p); +netdev_unlock (dev); + +synchronize_net (); + } + +kvfree (dev->napi_config); +} +/** * free_netdev - free network device * @dev: device
@@ -11579,8 +11595,6 @@ EXPORT_SYMBOL(alloc_netdev_mqs);*/ void free_netdev (struct net_device *dev) { -struct napi_struct *p, *n; -might_sleep ();/* When called immediately after register_netdevice() failed the unwind
@@ -11602,10 +11616,7 @@ void free_netdev(struct net_device *dev)/* Flush device addresses */ dev_addr_flush (dev); -list_for_each_entry_safe (p, n, &dev->napi_list, dev_list) -netif_napi_del (p); - -kvfree (dev->napi_config); +netdev_napi_exit (dev);ref_tracker_dir_exit (&dev->refcnt_tracker);#ifdef CONFIG_PCPU_DEV_REFCNT
Signature · 1 line
2.48.0.rc2.279.g1de40edade-goog
Quoted 5 lines · On Fri, 17 Jan 2025 23:21:13 +0000 Eric Dumazet wrote:
After
1b23cdbd2bbc("net: protectnetdev->napi_listwithnetdev_lock()") it makes sense to iterate throughdev->napi_listwhile holding the device lock.Also call
synchronize_net()at most one time.
I suspected you may send this :)
I was wondering whether we have to call sync_rcu() at all, Joe moved unhashing the NAPI to napi_disable(). Assuming the driver is sane it will call napi_disable() at latest in ndo_uninit(), and there's already a synchronize_net() between ndo_uninit() and free.
But maybe drivers are not sane. We don't have
/* napi_disable() sets the SCHED bit */ WARN_ON(!test_bit(NAPI_STATE_SCHED, &val));
in netif_napi_del(). I think we should add it, after Joe's changes if driver doesn't disable the napi it will leave a stale pointer in the hash table.
In any case, your change makes sense:
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Quoted 9 lines · On Fri, 17 Jan 2025 23:21:13 +0000 you wrote:
After
1b23cdbd2bbc("net: protectnetdev->napi_listwithnetdev_lock()") it makes sense to iterate throughdev->napi_listwhile holding the device lock.Also call
synchronize_net()at most one time.Signed-off-by Eric Dumazet <[email protected]>[...]
Here is the summary with links:
- [net-next] net: introduce netdev_napi_exit()
https://git.kernel.org/netdev/net-next/c/185e5b869071
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html