Re: [PATCH v2 1/2] mm: memcg: print out cgroup name in the memcg tracepoints

[Dmitry Rokosov]

Shakeel,

On Thu, Nov 23, 2023 at 11:45:10AM +0300, Dmitry Rokosov wrote:
> On Thu, Nov 23, 2023 at 08:15:47AM +0000, Shakeel Butt wrote:
> > On Thu, Nov 23, 2023 at 11:03:34AM +0300, Dmitry Rokosov wrote:
> > [...]
> > > > > +		cgroup_name(memcg->css.cgroup,
> > > > > +			__entry->name,
> > > > > +			sizeof(__entry->name));
> > > > 
> > > > Any reason not to use cgroup_ino? cgroup_name may conflict and be
> > > > ambiguous.
> > > 
> > > I actually didn't consider it, as the cgroup name serves as a clear tag
> > > for filtering the appropriate cgroup in the entire trace file. However,
> > > you are correct that there might be conflicts with cgroup names.
> > > Therefore, it might be better to display both tags: ino and name. What
> > > do you think on this?
> > > 
> > 
> > I can see putting cgroup name can avoid pre or post processing, so
> > putting both are fine. Though keep in mind that cgroup_name acquires a
> > lock which may impact the applications running on the system.
> 
> Are you talking about kernfs_rename_lock? Yes, it's acquired each
> time... Unfortunatelly, I don't know a way to save cgroup_name one time
> somehow...

I delved deeper and realized that kernfs_rename_lock is a read-write
lock, but it's a global one. While it's true that we only enable
tracepoints during specific periods of the host's lifetime, the trace
system is still a fast way to debug things. So, you're absolutely right,
we shouldn't slow down the system unnecessarily.

Therefore, today I will prepare a new version with only the cgroup ino.
Thank you for pointing that out to me!

-- 
Thank you,
Dmitry