On Fri, Jun 7, 2019 at 12:27 AM Jan Glauber <jglau...@marvell.com> wrote:
>
> To clarify, with 224 threads & CPUs queued_spin_lock_slowpath is the top hit
> even without a retry limit in lockref. This could be unrelated to the lockref
> fallback, it looks like it's coming from the spinlock in:
> do_sys_open -> get_unused_fd_flags -> __alloc_fd
At some point I stop worrying about microbenchmarks just because it's
easy to hit some locking paths in them, without it necessarily being
relevant in real loads.
But I'll apply the lockref patch because I think the "limit cmpxchg
loop" is conceptually a valid model, and while I think the "hitting
the same dentry lockref over and over again" is likely also an
artifact of a microbenchmark, I could at least imagine that it happens
with some common dentries (root, cwd) in some situations.
Linus
