Re: [PATCH v3] purgatory: fix disabling debug info

From: Alyssa Ross
Date: Fri Mar 31 2023 - 16:27:32 EST


On Sat, Apr 01, 2023 at 12:42:13AM +0900, Masahiro Yamada wrote:
> On Fri, Mar 31, 2023 at 7:29 AM Nathan Chancellor <nathan@xxxxxxxxxx> wrote:
> >
> > On Thu, Mar 30, 2023 at 06:22:24PM +0000, Alyssa Ross wrote:
> > > Since 32ef9e5054ec, -Wa,-gdwarf-2 is no longer used in KBUILD_AFLAGS.
> > > Instead, it includes -g, the appropriate -gdwarf-* flag, and also the
> > > -Wa versions of both of those if building with Clang and GNU as. As a
> > > result, debug info was being generated for the purgatory objects, even
> > > though the intention was that it not be.
> > >
> > > Fixes: 32ef9e5054ec ("Makefile.debug: re-enable debug info for .S files")
> > > Signed-off-by: Alyssa Ross <hi@xxxxxxxxx>
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Acked-by: Nick Desaulniers <ndesaulniers@xxxxxxxxxx>
> >
> > This is definitely more future proof.
> >
> > Reviewed-by: Nathan Chancellor <nathan@xxxxxxxxxx>
> > Tested-by: Nathan Chancellor <nathan@xxxxxxxxxx>
>
>
>
> I prefer v3 since it is cleaner, but unfortunately
> it does not work for Clang+GAS.
>
>
> With v3 applied, I still see the debug info.
>
>
>
> $ make LLVM=1 LLVM_IAS=0 arch/x86/purgatory/setup-x86_64.o
> UPD include/config/kernel.release
> UPD include/generated/utsrelease.h
> CALL scripts/checksyscalls.sh
> DESCEND objtool
> INSTALL libsubcmd_headers
> AS arch/x86/purgatory/setup-x86_64.o
> $ readelf -S arch/x86/purgatory/setup-x86_64.o
> There are 18 section headers, starting at offset 0x14d8:
>
> Section Headers:
> [Nr] Name Type Address Offset
> Size EntSize Flags Link Info Align
> [ 0] NULL 0000000000000000 00000000
> 0000000000000000 0000000000000000 0 0 0
> [ 1] .text PROGBITS 0000000000000000 00000040
> 0000000000000027 0000000000000000 AX 0 0 16
> [ 2] .rela.text RELA 0000000000000000 000012f8
> 0000000000000060 0000000000000018 I 15 1 8
> [ 3] .data PROGBITS 0000000000000000 00000067
> 0000000000000000 0000000000000000 WA 0 0 1
> [ 4] .bss NOBITS 0000000000000000 00001000
> 0000000000001000 0000000000000000 WA 0 0 4096
> [ 5] .rodata PROGBITS 0000000000000000 00001000
> 0000000000000020 0000000000000000 A 0 0 16
> [ 6] .rela.rodata RELA 0000000000000000 00001358
> 0000000000000018 0000000000000018 I 15 5 8
> [ 7] .debug_line PROGBITS 0000000000000000 00001020
> 000000000000005f 0000000000000000 0 0 1
> [ 8] .rela.debug_line RELA 0000000000000000 00001370
> 0000000000000018 0000000000000018 I 15 7 8
> [ 9] .debug_info PROGBITS 0000000000000000 0000107f
> 0000000000000027 0000000000000000 0 0 1
> [10] .rela.debug_info RELA 0000000000000000 00001388
> 0000000000000090 0000000000000018 I 15 9 8
> [11] .debug_abbrev PROGBITS 0000000000000000 000010a6
> 0000000000000014 0000000000000000 0 0 1
> [12] .debug_aranges PROGBITS 0000000000000000 000010c0
> 0000000000000030 0000000000000000 0 0 16
> [13] .rela.debug_[...] RELA 0000000000000000 00001418
> 0000000000000030 0000000000000018 I 15 12 8
> [14] .debug_str PROGBITS 0000000000000000 000010f0
> 0000000000000054 0000000000000001 MS 0 0 1
> [15] .symtab SYMTAB 0000000000000000 00001148
> 0000000000000168 0000000000000018 16 12 8
> [16] .strtab STRTAB 0000000000000000 000012b0
> 0000000000000041 0000000000000000 0 0 1
> [17] .shstrtab STRTAB 0000000000000000 00001448
> 000000000000008d 0000000000000000 0 0 1
> Key to Flags:
> W (write), A (alloc), X (execute), M (merge), S (strings), I (info),
> L (link order), O (extra OS processing required), G (group), T (TLS),
> C (compressed), x (unknown), o (OS specific), E (exclude),
> D (mbind), l (large), p (processor specific)
>
>
>
>
>
>
> With -g0 given, GCC stops passing -g -gdwarf-4 down to GAS.
>
>
> Clang does not do anything about -g0 for the external assembler.

You're right. Thank you for your thoughtful testing — I forgot to check
LLVM for v3. I thought maybe adding -Wa,-g0 would be enough, but it
turns out that GAS doesn't support that. If -g has been specified,
there doesn't seem to be any way to disable debug info again later in
the command line.

So we probably can't do better than v2 while LLVM_IAS=0 is supported.
The only other option I see is (untested):

asflags-y += -g0
asflags-remove-y += -Wa,-g -Wa,-gdwarf-4 -Wa,-gdwarf-5

But I don't like that option, because it means there are two completely
different ways of doing it depending on the compiler setup, and it makes
it even less likely anybody would remember to update asflags-remove-y
when DWARF 6 comes around or whatever.

> I was thinking of dropping LLVM_IAS=0 support.
> When we decide to give up -fno-integrated-as,
> we can clean up the code in various places.
>
>
> Anyway, v3 does not work in the current situation.
>
>
> V2 works for all usecases.
>
>
>
>
> --
> Best Regards
> Masahiro Yamada

Attachment: signature.asc
Description: PGP signature