[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAK7LNATFPWRaQb876p4=5K0H-Uc7KCLsVNnYVdxSypzmmfSv9A@mail.gmail.com>
Date: Tue, 21 Nov 2023 11:20:50 +0900
From: Masahiro Yamada <masahiroy@...nel.org>
To: Andi Kleen <ak@...ux.intel.com>
Cc: linux-kbuild@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] kbuild: Add inline-account tool
On Tue, Nov 21, 2023 at 5:08 AM Andi Kleen <ak@...ux.intel.com> wrote:
>
> >
> > > + sym_sizes = [syms[x + 1] - syms[x] for x, _ in enumerate(syms[:-1])]
> > > + sym_total = sum(sym_sizes)
> >
> >
> > This is equivalent to
> >
> > sym_total = syms[-1] - syms[0]
> >
> > isn't it?
>
> No it's not when there are gaps in the symbols, which always happens in
> the kernel. The previous version used the range but that led to inbalances.
I do not understand because there is nothing to handle gaps
in your code.
get_syms() appends only 'pc' into the list.
Then,
sym_sizes = [syms[x + 1] - syms[x] for x, _ in enumerate(syms[:-1])]
computes each symbol size by:
(address of the next symbol) - (address of the current symbol).
So, your code is equivalent to:
sym_total = syms[-1] - syms[0]
> > def get_ranges(syms, threads):
> >
> > ranges = []
> > prev = syms[0]
> > i = 1
> >
> > while threads > 0:
> >
> > boundary = prev + 1 + (syms[-1] - prev - 1) // threads
> >
> > while syms[i] < boundary:
> > i += 1
>
>
> Note the boundary is not the address, but the sum of symbol sizes
>
>
> -Andi
--
Best Regards
Masahiro Yamada
Powered by blists - more mailing lists