On Fri, May 30, 2025 at 12:31:35PM +0200, Vlastimil Babka wrote:

On 5/29/25 23:14, Johannes Weiner wrote:

quoted

On Thu, May 29, 2025 at 04:28:46PM +0100, Matthew Wilcox wrote:

quoted

Barry's problem is that we're all nervous about possibly regressing
performance on some unknown workloads.  Just try Barry's proposal, see
if anyone actually compains or if we're just afraid of our own shadows.

I actually explained why I think this is a terrible idea. But okay, I
tried the patch anyway.

This is 'git log' on a hot kernel repo after a large IO stream:

                                     VANILLA                      BARRY
Real time                 49.93 (    +0.00%)         60.36 (   +20.48%)
User time                 32.10 (    +0.00%)         32.09 (    -0.04%)
System time               14.41 (    +0.00%)         14.64 (    +1.50%)
pgmajfault              9227.00 (    +0.00%)      18390.00 (   +99.30%)
workingset_refault_file  184.00 (    +0.00%)    236899.00 (+127954.05%)

Clearly we can't generally ignore page cache hits just because the
mmaps() are intermittent.

The whole point is to cache across processes and their various
apertures into a common, long-lived filesystem space.

Barry knows something about the relationship between certain processes
and certain files that he could exploit with MADV_COLD-on-exit
semantics. But that's not something the kernel can safely assume. Not
without defeating the page cache for an entire class of file accesses.

I've just read the previous threads about Barry's proposal and if doing this
always isn't feasible, I'm wondering if memcg would be a better interface to
opt-in for this kind of behavior than both prctl or mctl. I think at least
conceptually it fits what memcg is doing? The question is if the
implementation would be feasible, and if android puts apps in separate memcgs...

CCing Tejun.

Cgroups has been trying to resist flag settings like these. The cgroup
tree is a nested hierarchical structure designed for dividing up
system resources. But flag properties don't have natural inheritance
rules. What does it mean if the parent group says one thing and the
child says another? Which one has precedence?

Hence the proposal to make it a per-process property that propagates
through fork() and exec(). This also enables the container usecase (by
setting the flag in the container launching process), without there
being any confusion what the *effective* setting for any given process
in the system is.