Re: [PATCH] man/man2/sigaction.2: Update si_code list with Linux v6.16

[Thiago Jung Bauermann <thiago.bauermann@linaro.org>]

Hello,

Carlos O'Donell <carlos@redhat.com> writes:

> On 8/28/25 5:07 PM, Thiago Jung Bauermann wrote:
>> Update with missing si_code values from Linux v6.16's
>> include/uapi/asm-generic/siginfo.h.
>
> The top-level header would be "#include <linux/signal.h>"
>
> This header isn't used by a glibc-based userspace.
>
> I've added it to the list of known conflicts:
> https://sourceware.org/glibc/wiki/Synchronizing_Headers
>
> Instead /usr/include/bits/siginfo-consts.h (glibc) provides
> constants that meet the language and platform requirements.
>
> While it is possible to use the Linux kernel's signal.h directly,
> that isn't recommended if you're trying to achieve any level of
> language or runtime conformance.
>
> You certainly cannot use "#include <signal.h>" and expect all that
> you have written below to work.

I used the kernel's siginfo.h file as a reference of what a process can
potentially see in the si_code field when it gets a signal.

> I am reviewing those entries which work and those which don't.
>
> If we want the other values to work then someone needs to do the
> harder work of either:

To be honest, my motivation to write this patch was just to fix the fact
that SEGV_CPERR was not mentioned. Then I noticed that other constants
were missing too, and thought it would be easy enough to add them as
well...

>  (a) Adding the constants to C libraries in a conformant way.
>
>  (b) Cleaning up the UAPI header to be conforming and work with
>      the existing C libraries to include it indirectly.
>
>  (c) Cleaning up both headers to allow dual inclusion with
>      additional constants showing up as needed.
>
> In summary:
>
>  - This patch contains 2 constants that don't work today in a glibc-based
>    userspace.
>
>  - The existing man page documents 1 constants that doesn't work today
>    with the standard #include <signal.h>.

As you suggested (thanks!), I sent a patch to the glibc mailing list
adding the si_codes it's missing to its siginfo-conts.h header: 

https://inbox.sourceware.org/libc-alpha/20250903024151.3030839-1-thiago.bauermann@linaro.org/

If that one is accepted, hopefully this patch can go in?

>> +.TP
>> +.BR TRAP_PERF " (since Linux 5.13)"
>> +Perf event with sigtrap=1.
>
> Not present in glibc.
>
> Suggest sending a patch to libc-alpha@sourceware.org to patch
> glibc/sysdeps/unix/sysv/linux/bits/siginfo-consts.h with the
> added constant.
>
> Not directly usable by glibc-based userspace.

Added in the glibc patch I just sent.

>> @@ -850,6 +883,20 @@ signal:
>>   Triggered by a
>>   .BR seccomp (2)
>>   filter rule.
>
> ^^^^ the existing discussed SYS_SECCOMP is not exposed either and thus
> is a documentation issue with this man page.

Added in the glibc patch I just sent.

>> +.TP
>> +.BR SYS_USER_DISPATCH  " (since Linux 5.11)"
>> +Syscall user dispatch triggered.
>
> Not present in glibc.
>
> Not directly usable by glibc-based userspace.

Added in the glibc patch I just sent.

> Suggest sending a patch to libc-alpha@sourceware.org to patch
> glibc/sysdeps/unix/sysv/linux/bits/siginfo-consts.h with the
> added constant along with SYS_SECCOMP which is missing.
>
> This along with the other changes would bring userspace in line with
> the kernel constants.
>
> Note: glibc accepts DCOd contributions.
> https://sourceware.org/glibc/wiki/Contribution%20checklist

Thank you for the suggestions. They were very helpful.

-- 
Thiago