| Message ID | 20241118-netpoll_rcu-v1-0-a1888dcb4a02@debian.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | netpoll: Use RCU primitives for npinfo pointer access | expand |
Hello: This series was applied to netdev/net.git (main) by Jakub Kicinski <kuba@kernel.org>: On Mon, 18 Nov 2024 03:15:16 -0800 you wrote: > The net_device->npinfo pointer is marked with __rcu, indicating it requires > proper RCU access primitives: > > struct net_device { > ... > struct netpoll_info __rcu *npinfo; > ... > }; > > [...] Here is the summary with links: - [net,1/2] netpoll: Use rcu_access_pointer() in __netpoll_setup https://git.kernel.org/netdev/net/c/c69c5e10adb9 - [net,2/2] netpoll: Use rcu_access_pointer() in netpoll_poll_lock https://git.kernel.org/netdev/net/c/a57d5a72f8de You are awesome, thank you!
On Tue, Nov 19, 2024 at 03:40:31AM +0000, patchwork-bot+netdevbpf@kernel.org wrote: > Hello: > > This series was applied to netdev/net.git (main) > by Jakub Kicinski <kuba@kernel.org>: > > On Mon, 18 Nov 2024 03:15:16 -0800 you wrote: > > The net_device->npinfo pointer is marked with __rcu, indicating it requires > > proper RCU access primitives: > > > > struct net_device { > > ... > > struct netpoll_info __rcu *npinfo; > > ... > > }; > > > > [...] > > Here is the summary with links: > - [net,1/2] netpoll: Use rcu_access_pointer() in __netpoll_setup > https://git.kernel.org/netdev/net/c/c69c5e10adb9 > - [net,2/2] netpoll: Use rcu_access_pointer() in netpoll_poll_lock > https://git.kernel.org/netdev/net/c/a57d5a72f8de These are not bug fixes. They should not be going through during a merge window, especially with such a short period of review. Thanks,
On Tue, 19 Nov 2024 11:53:49 +0800 Herbert Xu wrote: > > Here is the summary with links: > > - [net,1/2] netpoll: Use rcu_access_pointer() in __netpoll_setup > > https://git.kernel.org/netdev/net/c/c69c5e10adb9 > > - [net,2/2] netpoll: Use rcu_access_pointer() in netpoll_poll_lock > > https://git.kernel.org/netdev/net/c/a57d5a72f8de > > These are not bug fixes. They should not be going through during > a merge window, especially with such a short period of review. Sorry, I do agree with your assessment, especially on patch 1. But it's good to silence the false positive, so I applied and stripped the Fixes tag.
The net_device->npinfo pointer is marked with __rcu, indicating it requires proper RCU access primitives: struct net_device { ... struct netpoll_info __rcu *npinfo; ... }; Direct access to this pointer can lead to issues such as: - Compiler incorrectly caching/reusing stale pointer values - Missing memory ordering guarantees - Non-atomic pointer loads Replace direct NULL checks of npinfo with rcu_access_pointer(), which provides the necessary memory ordering guarantees without the overhead of a full RCU dereference, since we only need to verify if the pointer is NULL. In both cases, the RCU read lock is not held when the function is being called. I checked that by using lockdep_assert_in_rcu_read_lock(), and seeing the warning on both cases. Signed-off-by: Breno Leitao <leitao@debian.org> --- Breno Leitao (2): netpoll: Use rcu_access_pointer() in __netpoll_setup netpoll: Use rcu_access_pointer() in netpoll_poll_lock include/linux/netpoll.h | 2 +- net/core/netpoll.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) --- base-commit: 8ffade77b6337a8767fae9820d57d7a6413dd1a1 change-id: 20241115-netpoll_rcu-eb1296511b71 Best regards,