Re: [PATCH] arch/x86/setup.c: Ignore early boot parameters like no-real-mode

[Trammell Hudson <hudson@xxxxxxxx>]

On Wednesday, August 12, 2020 8:16 PM, Andrew Cooper 
<andrew.cooper3@xxxxxxxxxx> wrote:

> However, the use of LINE creates problems for livepatch builds, as
> it causes the binary diffing tools to believe these changed, based on a
> change earlier in the file.

Ah, I hadn't considered that.  Makes sense that the
deterministic __COUNTER__ would be better for many uses.
However...

One concern is that the __COUNTER__ is per compilation unit,
which I think would mean that every file would conflict by
creating __setup_str_ign_0 for the first one, __setup_str_ign_1
for the next, etc.  Unless they are static scoped or have a
variable-name-friendly unique prefix, they aren't
suitable for globals that share a namespace.

Another is that each evaluation increments it, so the macro
would need some tricks to avoid double-evaluation since it
both defines __setup_str_ign_XYZ and references it in the
__kparam structure.  This is in contrast to __LINE__,
which is constant in the macro and based on the line where
it was invoked so the double evaluation is not a problem.

> Instead of opencoding TEMP_NAME(), can we borrow Linux's __UNIQUE_ID()
> infrastructure?  COUNTER appears to have existed for ages, and
> exists in all of our supported compilers.

I'm definitely in favor of borrowing it from Linux, although
subject to those two caveats.

> If you want, I can sort that out as a prereq patch, and rebase this one
> on top?

That sounds fine to me. They really are two separate patches.

--
Trammell