Re: [PATCH] cgroup/rstat: avoid disabling irqs for O(num_cpu)

[Johannes Weiner]

On Wed, Mar 19, 2025 at 06:35:02PM +0000, Yosry Ahmed wrote:
> On Wed, Mar 19, 2025 at 02:06:43PM -0400, Johannes Weiner wrote:
> > (btw, why do we not have any locking around the root stats in
> > cgroup_base_stat_cputime_show()? There isn't anything preventing a
> > reader from seeing all zeroes if another reader runs the memset() on
> > cgrp->bstat, is there? Or double times...)
> 
> (I think root_cgroup_cputime() operates on a stack allocated bstat, not
> cgrp->bstat)

That was the case until:

commit b824766504e49f3fdcbb8c722e70996a78c3636e
Author: Chen Ridong <chenridong@xxxxxxxxxx>
Date:   Thu Jul 4 14:01:19 2024 +0000

    cgroup/rstat: add force idle show helper

Now it's doing this:

void cgroup_base_stat_cputime_show(struct seq_file *seq)
{
	struct cgroup *cgrp = seq_css(seq)->cgroup;

	if (cgroup_parent(cgrp)) {
		...
	} else {
		/* cgrp->bstat of root is not actually used, reuse it */
		root_cgroup_cputime(&cgrp->bstat);
		usage = cgrp->bstat.cputime.sum_exec_runtime;
		utime = cgrp->bstat.cputime.utime;
		stime = cgrp->bstat.cputime.stime;
		ntime = cgrp->bstat.ntime;
	}
}

and then:

static void root_cgroup_cputime(struct cgroup_base_stat *bstat)
{
	struct task_cputime *cputime = &bstat->cputime;
	int i;

	memset(bstat, 0, sizeof(*bstat));

	/* various += on bstat and cputime members */
}

So all readers are mucking with the root cgroup's bstat.