Re: kern/57320: ATF test case kernel/t_magic_symlinks:machine_arch fails on RPI02W/evbarm only

[mlelstv%serpens.de@localhost (Michael van Elst)]

The following reply was made to PR kern/57320; it has been noted by GNATS.

From: mlelstv%serpens.de@localhost (Michael van Elst)
To: gnats-bugs%netbsd.org@localhost
Cc: 
Subject: Re: kern/57320: ATF test case kernel/t_magic_symlinks:machine_arch fails on RPI02W/evbarm only
Date: Wed, 12 Apr 2023 08:45:54 -0000 (UTC)

 martin%duskware.de@localhost (Martin Husemann) writes:
 
 > I am confused. uname -p and -m print hw.machine ("evbarm") and
 > hw.machine_arch ("earmv6hf") respectively. Those are the values that
 > the magic symlinks should use too (and I think they do).
 > 
 > Where does the ABI name ("earm") come into play?
 
 vfs uses the compile time constant MACHINE_ARCH == "earm". It might
 better use the variable machine_arch, but that wouldn't change
 anything.
 
 MACHINE_ARCH is set in arm/param.h, when not overriden by
 build.sh, the kernel only uses earm, arm, earmeb or armeb.
 
 Userland gets more refined values depending on the compiler
 target.
 
 
 sysctl uses (P)->p_md.md_march, which is initially MACHINE_ARCH
 but then gets overridden by ep_machine_arch which is the
 ELF_NOTE_MARCH_NAME of the executable currently running.
 
 Displaying notes found in: .note.netbsd.march
   Owner                Data size        Description
   NetBSD                0x00000009      MARCH <earmv6hf>
 
 
 Changing vfs to use PROC_MACHINE_ARCH(p) might help, but
 PROC_MACHINE_ARCH is a MD thing (arm + mips). sysctl defaults
 to use machine_arch if the arch doesn't define the macro.