Skip to contentSkip to conversation

Beyond per-CPU atomics and rseq syscall: thoroughly abusing mm with the do_on_cpu syscall

Ksummit-discuss Mathieu Desnoyers 1 participant 1 message Jul 8, 2019

lore
Share & tools
lore
#0 lore raw

The Restartable Sequences system call [1,2,3,4] introduced in Linux 4.18 has limitations which can be solved by introducing a bytecode interpreter running in inter-processor interrupt context which accesses user-space data.

However, there are a handful of reasons why accessing user-space data from an IPI handler concurrently with user-space execution is a bad idea, starting with page fault handling. This discussion aims at gathering feedback from the mm community on the do_on_cpu system call aiming to solve this gracefully.

[1] "PerCpu Atomics" http://www.linuxplumbersconf.org/2013/ocw/system/presentations/1695/original/LPC%20-%20PerCpu%20Atomics.pdf
[2] "Restartable sequences" https://lwn.net/Articles/650333/
[3] "Restartable sequences restarted" https://lwn.net/Articles/697979/
[4] "Restartable sequences and ops vectors" https://lwn.net/Articles/737662/

Thanks,

Mathieu

Signature · 3 lines
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com