[Crash-utility] [PATCH v2] Improve the performance of symbol searching for kernel modules

[Tao Liu]

On Wed, Aug 25, 2021 at 1:10 PM HAGIO KAZUHITO(萩尾　一仁)
<k-hagio-ab at nec.com> wrote:
>
> -----Original Message-----
> > > (BTW, at a glance I'm not sure why the 2nd stage is needed...
> > > sp_hashed was already checked with STREQ in symname_hash_search if found,
> > > otherwise st->synmane_hash should include all kernel symbols..)
> > >
> > Hello Kazu,
> >
> > Yes, from my view I also think the 2nd stage is not necessary, I have
> > thought about removing
> > it as well, but I just don't want to go too far at one step. :-)
>
> OK, I see.
>
> > > One more, do you have any slow command that will be especially improved
> > > by the patch?  I'm ok with the sym command taking 3ms :-)
> > > I'd like to know why the patch should be applied or how much the worst
> > > command etc. will be improved, to see the importance of it.
> > >
> >
> > No, I don't have a specific slow command. Because the modified functions
> > symbol_search and symbol_exists are fundamental and widely used by
> > other crash functions, thus the benefit of performance improvement can
> > get accumulated. In addition, the time consumption for T3 shortened greatly,
> > from 2.4 to 0.017, maybe this can be one of the reasons.
>
> OK.
>
> As far as I've checked a large machine (with 192 CPUs), there are not many
> commands but several commands call symbol_search and symbol_exists returning
> "not found" many times:
>
>       symbol_search  symbol_exists
>        found  NULL    found  NULL
> startup   95    21       29    21
> ps -m   1951  1951        0     0
> irq     1850     0        0 13448
> irq -a  1850     0      884 10777
> irq -s  1850     0      884 10777
> timer    193   193        0     0
>
> Actually, these commands were obviously improved by the patch.
> So let's add some explanation like this to the commit log:
>

Thanks for the testing.

> " symbol_search and symbol_exists are fundamental and widely used by
> other crash functions, thus the benefit of performance improvement can
> get accumulated.  For example, "ps -m" and "irq" commands, which call
> the functions many times, will become faster with the patch."
>
> Anyway, I'll review the patch, please wait for a while.
>

Thank you very much for the patch review! I will send new versions of the patch
with the updated commit log.

Thanks,
Tao Liu

> Thanks,
> Kazu
>
>
> --
> Crash-utility mailing list
> Crash-utility at redhat.com
> https://listman.redhat.com/mailman/listinfo/crash-utility
>