On 7/18/24 17:49, David Finkel wrote:
I spent some time today attempting to implement this. Here's a branch on github that compiles, and I think is close to what you described, but is definitely still a WIP:
https://github.com/torvalds/linux/compare/master...dfinkel:linux:memcg2_memo...
Since there seems to be significant agreement that this approach is better long-term as a kernel interface, if that continues, I can factor out some of the changes so it supports both memory.peak and memory.swap.peak, fix the tests, and clean up any style issues tomorrow.
Also, If there are opinions on whether the cgroup_lock is a reasonable way of handling this synchronization, or if I should add a more appropriate spinlock or mutex onto either the pagecounter or the memcg, I'm all ears.
cgroup_lock() should only be used by the cgroup core code, though there are exceptions.
You may or may not need lock protection depending on what data you want to protect and if there is any chance that concurrent race may screw thing up. If lock protection is really needed, add your own lock to protect the data. Since your critical sections seem to be pretty short, a regular spinlock will be enough.
Cheers, Longman