Re: [PATCH 7/7] block, drivers: make lots of attribute_group globals const

[Max Kellermann]

On Mon, Oct 9, 2023 at 7:30 PM Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Oct 09, 2023 at 06:57:40PM +0200, Max Kellermann wrote:
> > This moves those variables to the ".rodata" section which reduces the
> > kernel size a bit and protects the variables by putting them on
> > read-only pages at runtime.
>
> The kernel size should still be the same overall, you are just moving
> pointers from one section to another, right?
>
> If not, what are the numbers, show them please.

Before this patch series:

   text       data        bss        dec        hex    filename
10645342    6836166    12476420    29957928    1c91f28    vmlinux

After:

   text       data        bss        dec        hex    filename
10645342    6835398    12476420    29957160    1c91c28    vmlinux

(Config attached if you wish to reproduce. I'm using stock GCC from
Debian Bookworm.)

It's only a few hundred bytes, and I havn't really dug deep enough to
explain to you why this is - my goal was (semantic) const-correctness
and .rodata protection, but those numbers are a side effect that
exists.


> But step back, are you SURE you can make these attribute group pointers
> const?  They are modified by some subsystems by adding or removing items
> from the lists, so why does the core need to change for that?

Some few do modify them, but they do not modify those pointers; they
allocate a new one (the 3 I sbumitted a patch for) or they modify
global variables (arch/pmu) which I'm not going to make "const". But
most globals can indeed be made "const".

Attachment: .config
Description: Binary data