Re: [PATCH] btrfs: Remove noinline attribute from wait_for_commit

[David Sterba <dsterba@suse.cz>]

On Tue, Nov 24, 2020 at 05:05:32PM +0100, David Sterba wrote:
> On Tue, Nov 24, 2020 at 05:42:57PM +0200, Nikolay Borisov wrote:
> > On 24.11.20 г. 17:39 ч., David Sterba wrote:
> > > On Tue, Nov 24, 2020 at 04:45:02PM +0200, Nikolay Borisov wrote:
> > >> The function is a plain wrapper that noinline makes no sense.
> > > 
> > > Or it is a way to let the function name appear in a stack trace, which
> > > could be useful for debugging or analyzing system state.
> > > 
> > 
> > Well, this information could generally be derived by having the debug
> > info  either in crash or one of the "beautify" scripts. In any case I
> > won't insist.
> 
> The way it's used is 'cat /proc/*/stack', without the need of debugging
> kernels and not in a post-mortem analysis with crash.

Actually, there's noinline_for_stack annotation that could make a bit
more understandable, but reading the comment it's more about conserving
stack space, not making the function visible in stack trace.

There are functions like free_reloc_roots, memcmp_node_keys or
cleanup_bitmap_list that do a trivial operations, no chance to block or
wait. So these are the obvious cases where we don't want it, there are
still many more for long functions that would need closer inspection.