Re: [PATCH] RFC: xen/x86: Enable --gc-sections

[Jan Beulich <jbeulich@xxxxxxxx>]

On 09.12.2025 22:55, Jason Andryuk wrote:
> On 2025-12-08 03:46, Jan Beulich wrote:
>> On 05.12.2025 23:40, Andrew Cooper wrote:
>>> On 05/12/2025 10:28 pm, Jason Andryuk wrote:
>>>> When linking to create xen-syms, add --gc-sections to garbage collect
>>>> unused stuff.  Relies on CONFIG_CC_SPLIT_SECTIONS
>>>>
>>>> We need to add KEEP() to the linker script in assorted places to retain
>>>> appropriate data - especially the arrays created therein.
>>>>
>>>> Something is off though.  In a test where memory_add() is unreachable,
>>>> it is still included.  I'm not sure, but I am wondering if it's the
>>>> alternatives somehow keeping a reference to it.
>>>
>>> Yes, .altinstructions contains relocations against the origin patch
>>> site, which will cause it to appear to be referenced. The same will be
>>> happening with a bunch of other sections.
>>
>> We will need to derive helper section names from base section ones. See
>> e.g. HAVE_AS_SECTNAME_SUBST as introduced by "common: honor
>> CONFIG_CC_SPLIT_SECTIONS also for assembly functions", controlling the
>> use of the --sectname-subst asssembler option, to in turn be able to use
>> %S in section names (available from gas 2.26 onwards).
> 
> I tried to add your patch and change ALTERNATIVE to:
> ".pushsection .altinstructions.%S, \"a\", @progbits\n"
> but it fails to build.  One example:
> ./include/xen/compiler.h:62:21: error: invalid 'asm': operand number 
> missing after %-letter
>     62 | # define asm_inline asm __inline
>        |                     ^~~
> ./arch/x86/include/asm/pdx.h:13:5: note: in expansion of macro ‘asm_inline’
>     13 |     asm_inline goto (                               \
>        |     ^~~~~~~~~~
> ./arch/x86/include/asm/pdx.h:22:5: note: in expansion of macro 
> ‘PDX_ASM_GOTO’
>     22 |     PDX_ASM_GOTO(skip);
>        |     ^~~~~~~~~~~~
> 
> ".pushsection .altinstructions.%%S, \"a\", @progbits\n" ends up creating 
> section ".altinstructions.%S" which doesn't helpful.
> 
> Is %S expected to work with inline asm, or only standalone?

Both, as long as --sectname-subst is passed to the assembler. My patch adds
that option to AFLAGS only, whereas for inline assembly it would need adding
to CFLAGS. Did you perhaps overlook that?

Jan