Re: 5.19-rc1 x86 build failure

[Andrew Cooper]

On 07/06/2022 13:19, Joe Damato wrote:
> Greetings:
>
> My apologies if this is the incorrect place to report this, but I got a
> build error when trying to compile the net-next 5.19-rc1 tree.
>
> git bisect says that commit a1e2c031ec394 ("x86/mm: Simplify
> RESERVE_BRK()") is responsible for the build issue I am hitting.
>
> I am performing this build on an x86_64 system with GNU C11 (Ubuntu
> 5.4.0-6ubuntu1~16.04.12) version 5.4.0 20160609 (x86_64-linux-gnu).
>
> The assembler outputs a cryptic error message:
>
> /tmp/ccnGOKZ5.s: Assembler messages:
> /tmp/ccnGOKZ5.s:98: Error: missing ')'
> /tmp/ccnGOKZ5.s:98: Error: missing ')'
> /tmp/ccnGOKZ5.s:98: Error: missing ')'
> /tmp/ccnGOKZ5.s:98: Error: junk at end of line, first unrecognized
> character is `U'
> /tmp/ccnGOKZ5.s:99: Error: missing ')'
> /tmp/ccnGOKZ5.s:99: Error: missing ')'
> /tmp/ccnGOKZ5.s:99: Error: missing ')'
> /tmp/ccnGOKZ5.s:99: Error: junk at end of line, first unrecognized
> character is `U'
>
> I've asked GCC to generate the assembly and output so I can see more
> specifically where this issue is (via "-fverbose-asm -Wa,-adhln=output"):
>
>   96                            .pushsection .brk_reservation,"aw",@nobits
>   97                            .brk.early_pgt_alloc:
>   98 ???? 00000000              .skip ((2 * 3) * ((1UL) << 12))
> ****  Error: missing ')'
> ****  Error: missing ')'
> ****  Error: missing ')'
> ****  Error: junk at end of line, first unrecognized character is `U'
>   98      0000
>  100                            .popsection
>
> This comes from arch/x86/mm/init.c, which has the following code:
>
> RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE);
>
> wherein INIT_PGT_BUF_SIZE (via PAGE_SIZE) has a "1UL" which makes the
> assembler unhappy.
>
> I don't really know what the correct way to fix this is; it seems that the
> macro _AC should handle this if ASSEMBLY is defined, IIUC, but that does
> not seem to be the case at this point in init.c.
>
> Perhaps I am doing something incorrect during the build process causing
> this to happen?

The problem is that _AC() is evaluated in C context (so gains the UL/ULL
suffix), and the C'd string is fed directly into the assembler (where
older binutils doesn't tolerate the suffix).

Short of having a _PAGE_SIZE which is an explicitly non-AC()'d constant,
I'm not sure what to suggest.  Ideally, you'd want to temporarily define
__ASSEMBLY__ around the expansion of __stringify(), but I don't think
that's possible as RESERVE_BRK() is a macro itself.

~Andrew