Re: [PATCH 2/2][4.15?] x86: fix build when NR_CPUS == 1

[Jan Beulich <jbeulich@suse.com>]

On 01.03.2021 15:01, Roger Pau Monné wrote:
> On Mon, Mar 01, 2021 at 09:31:07AM +0100, Jan Beulich wrote:
>> In this case the compiler is recognizing that no valid array indexes
>> remain (in x2apic_cluster()'s access to per_cpu(cpu_2_logical_apicid,
>> ...)), but oddly enough isn't really consistent about the checking it
>> does (see the code comment).
> 
> I assume this is because of the underlying per_cpu access to
> __per_cpu_offset using cpu as the index, in which case wouldn't it be
> better to place the BUG_ON there?

Not sure, to be honest. It seemed more logical to me to place it
next to where the issue is.

> Also I wonder why the accesses the same function does to the per_cpu
> area before the modified chunk using this_cpu as index don't also
> trigger such warnings.

The compiler appears to be issuing the warning when it can prove
that no legitimate index can make it to a respective use. in this
case it means that is is

        if ( this_cpu == cpu )
            continue;

which makes it possible for the compiler to know that what gets
past this would be an out of bounds access, since for NR_CPUS=1
both this_cpu and cpu can only validly both be zero. (This also
plays into my choice of placement, because it is not
x2apic_cluster() on its own which has this issue.)

Jan