Why doesn't Linux have a NORCU RCU option that just implements the RCU API as plain rwlocks?
Why does everybody love breaking the important real-world case (1-8 cores) for the sake of HPC wackiness? :-(
-
-
Lots of important real-world cases with a wide range of CPUs. You are doing mobile or some such?
-
Mobile and embedded. But even for reasonable desktop and server use locks are better.
-
But I don't think this is solvable in Linux. The code using the RCU api actually does the alloc/free operations...
-
...so there's no way a "NORCU" (as plain rwlock) could give you update-in-place without unwanted invasive changes.
-
That is correct. A NORCU has the same problem in user mode in the presence of signal handlers.
-
That's trivial to solve in kernel because disabling irqs takes at most a couple cycles. OTOH sigprocmask is syscall.
-
BTW any tips on debugging RCU stalls (not perm, just long, >20s) in idle task with dyntick? :-)
-
(1) https://www.kernel.org/doc/Documentation/RCU/stallwarn.txt … (2) but please email me dmesg + .config + version
- 5 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.