The kernel locking changes submitted today for Linux 6.17 contain a temporary change worth discussion for yielding a 10x speed-up of a particular function call and as part of that yielding less network egress downtime until a better solution is developed.
The locking changes sent out today for the Linux 6.17 merge window include a number of low-level code changes while catching my eye was beginning to use synchronize_rcu_expedited() within lockdep_unregister_key() for better performance.
Meta engineer and Debian developer Breno Leitao authored the one line of code change in switching from synchronize_rcu to synchronize_rcu_expedited. He explained on the patch:
“lockdep: Speed up lockdep_unregister_key() with expedited RCU synchronization
lockdep_unregister_key() is called from critical code paths, including sections where rtnl_lock() is held. For example, when replacing a qdisc in a network device, network egress traffic is disabled while __qdisc_destroy() is called for every network queue.
If lockdep is enabled, __qdisc_destroy() calls lockdep_unregister_key(), which gets blocked waiting for synchronize_rcu() to complete.
For example, a simple tc command to replace a qdisc could take 13 seconds:
# time /usr/sbin/tc qdisc replace dev eth0 root handle 0x1: mq
real 0m13.195s
user 0m0.001s
sys 0m2.746sDuring this time, network egress is completely frozen while waiting for RCU synchronization.
Use synchronize_rcu_expedited() instead to minimize the impact on critical operations like network connectivity changes.
This improves 10x the function call to tc, when replacing the qdisc for a network card.
# time /usr/sbin/tc qdisc replace dev eth0 root handle 0x1: mq
real 0m1.789s
user 0m0.000s
sys 0m1.613s”
The original issue stems from a problem discovered at Meta. This principally will help with the performance of the debug kernel builds.
The synchronize RCU expedited call has been known for years to be much faster albeit despised by developers and at the cost of system IPIs. This workaround for faster performance in lockdep_unregister_key is expected to be temporary with ultimately working to remove RCU usage and relying on hazptr in its place. See this LKML thread for more details on that discussion. But for Linux 6.17 this one-liner change to use the expedited call is part of the locking pull request.
