Re: [PATCH v9 09/15] syscalls: Add an option for offsetting the common table

["André Almeida" <andrealmeid@igalia.com>]

Em 12/11/2026 11:07, Arnd Bergmann escreveu:
> On Thu, Nov 5, 2026, at 21:05, André Almeida wrote:
>> To better support architectures that doesn't share the same syscall
>> numbers, but have them in the same sequence, add an new option to offset
>> the common syscall table numbers by a value. This make it possible to reuse
>> the common syscall table for alpha.
>>
>> Signed-off-by: André Almeida <andrealmeid@igalia.com>
>> ---
>> Changes from v4:
>>   - I was reusing the logic from --offset to apply the --common-offset, but
>>   the logic is different: --offset does a string concat, and --common-offset
>>   does a addition to the number
> 
> I don't understand why this should be different here.
> 
>      Arnd

This how --offset is used, from arch/mips/kernel/syscalls/Makefile:

       cmd_syshdr = $(CONFIG_SHELL) $(syshdr) --offset __NR_Linux $< $@

The effect of this is that the string "__NR_Linux +" will be added to 
every syscall number in the unistd header, e.g.:

#define __NR_file_setattr (__NR_Linux + 469)
#define __NR_listns (__NR_Linux + 470)
#define __NR_rseq_slice_yield (__NR_Linux + 471)

...

Now, for --common-offset, from arch/alpha/kernel/syscalls/Makefile:

       cmd_systbl = $(CONFIG_SHELL) $(systbl) \
		   --common-tbl $(systbl_common) --common-offset 110 \
		   --abis $(abis) $< $@

That means that 110 will be added to the syscall number itself:

#define __NR_file_setattr 579
#define __NR_listns 580
#define __NR_rseq_slice_yield 581

So one option does a string concat, the other option really does a sum 
operation. If I had used --offset 110, the result would be

#define __NR_file_setattr (110 + 469)
#define __NR_listns (110 + 470)
#define __NR_rseq_slice_yield (110 + 471)

Thanks!
	André