linux-kernel - Re: [PATCH] numa: mark __next_node() as __always

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Mon, 6 Dec 2021 11:43:47 -0800
From:   Nick Desaulniers <ndesaulniers@...gle.com>
To:     Alexander Lobakin <alexandr.lobakin@...el.com>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Arnd Bergmann <arnd@...db.de>,
        Nathan Chancellor <nathan@...nel.org>,
        Zhen Lei <thunder.leizhen@...wei.com>,
        linux-kernel@...r.kernel.org, llvm@...ts.linux.dev
Subject: Re: [PATCH] numa: mark __next_node() as __always_inline to fix
 section mismatch

On Mon, Dec 6, 2021 at 8:19 AM Alexander Lobakin
<alexandr.lobakin@...el.com> wrote:
>
> Clang (13) uninlines __next_node() which emits the following warning
> due to that this function is used in init code (amd_numa_init(),
> sched_init_numa() etc.):
>
> WARNING: modpost: vmlinux.o(.text+0x927ee): Section mismatch
> in reference from the function __next_node() to the variable
> .init.data:numa_nodes_parsed
> The function __next_node() references
> the variable __initdata numa_nodes_parsed.
> This is often because __next_node lacks a __initdata
> annotation or the annotation of numa_nodes_parsed is wrong.
>
> Mark __next_node() as __always_inline() so it won't get uninlined.
> bloat-o-meter over x86_64 binaries says this:
>
> scripts/bloat-o-meter -c vmlinux.baseline vmlinux
> add/remove: 1/1 grow/shrink: 2/7 up/down: 446/-2166 (-1720)
> Function                                     old     new   delta
> apply_wqattrs_cleanup                          -     410    +410
> amd_numa_init                                814     842     +28
> sched_init_numa                             1338    1346      +8
> find_next_bit                                 38      19     -19
> __next_node                                   45       -     -45
> apply_wqattrs_prepare                       1069     799    -270
> wq_nice_store                                688     414    -274
> wq_numa_store                                805     433    -372
> wq_cpumask_store                             789     402    -387
> apply_workqueue_attrs                        538     147    -391
> workqueue_set_unbound_cpumask                947     539    -408
> Total: Before=14422603, After=14420883, chg -0.01%
>
> So it's both win-win in terms of resolving section mismatch and
> saving some text size (-1.7 Kb is quite nice).
>
> Signed-off-by: Alexander Lobakin <alexandr.lobakin@...el.com>

Thanks for the patch.  See this thread:
https://github.com/ClangBuiltLinux/linux/issues/1302

There's a lot more instances of these based on config.  Something like
https://github.com/ClangBuiltLinux/linux/issues/1302#issuecomment-807260475
would be more appropriate for fixing all instances, but I think this
is more so an issue with the inline cost model in LLVM.

I need to finish off https://reviews.llvm.org/D111456, and request
that https://reviews.llvm.org/D111272 which landed in clang-14 get
backported to the 13.0.1 release which should also help.

> ---
>  include/linux/nodemask.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/include/linux/nodemask.h b/include/linux/nodemask.h
> index 567c3ddba2c4..55ba2c56f39b 100644
> --- a/include/linux/nodemask.h
> +++ b/include/linux/nodemask.h
> @@ -266,7 +266,7 @@ static inline int __first_node(const nodemask_t *srcp)
>  }
>
>  #define next_node(n, src) __next_node((n), &(src))
> -static inline int __next_node(int n, const nodemask_t *srcp)
> +static __always_inline int __next_node(int n, const nodemask_t *srcp)
>  {
>         return min_t(int,MAX_NUMNODES,find_next_bit(srcp->bits, MAX_NUMNODES, n+1));
>  }
> --
> 2.33.1
>


-- 
Thanks,
~Nick Desaulniers