Re: modload(8) v.s. alias

[Rin Okuyama <rokuyama.rk%gmail.com@localhost>]

Hi,

Thank you for your kind explanation! I understand the situation.
The current behavior is clean and safe, and we should not try to
``improve'' it easily.

For the case of NFS on m68k, we can rename ffs to __ffssi2 for
__KERNEL; we do not provide ffs(9) as a kernel routine.

Thanks,
rin

On 2020/03/10 0:49, Paul Goyette wrote:
> Our in-kernel linker does not understand weak references.  It would
> be a lot of work to implement, and there are some issues that would
> need to be considered carefully.
>
> For example, if module A defines a weak alias, and module B is loaded
> and resolves the alias, what do you do if module C gets loaded and we
> discover a "strong" definition of the same symbol?  Do you keep the
> original resolution, or do you re-resolve it to the new definition?
>
>
> On Mon, 9 Mar 2020, Rin Okuyama wrote:
>
> > Hi,
> >
> > modload(8) fails if some depended functions are alias, at least
> > on m68k. For example:
> >
> >   # uname -a
> >   NetBSD  9.99.49 NetBSD 9.99.49 (GENERIC) #6: Mon Mar  9 22:53:07 JST 2020 rin@latipes:/build/src/sys/arch/sun2/compile/GENERIC sun2
> >   # modload nfs
> >   [xxx.xxx] kobj_checksyms, 998: [nfs]: linker error: symbol '__ffssi2' not found
> >   [xxx.xxx] WARNING: module error: unable to affix module 'nfs', error 8
> >   modload: nfs: Exec format error
> >   #
> >
> > Here, __ffssi4 is weak alias of ffs on m68k:
> >
> >   # nm /netbsd | grep __ffssi2
> >   0012327c W __ffssi2
> >   # nm /netbsd | grep 0012327c
> >   0012327c W __ffssi2
> >   0012327c T ffs
> >   #
> >
> > If this symbol is turned into a ``real'' object by attached patch,
> > modload(8) and loaded module work fine:
> >
> >   # nm /netbsd | grep __ffssi4
> >   0012327c T __ffssi2
> >   # modload nfs
> >   # mount_nfs server:/path /mnt
> >   # ls /mnt
> >   foo bar baz ...
> >   #
> >
> > Is this a bug or feature? How can I fix this problem?
> >
> > Thanks,
> > rin
> > ----
> > Index: common/lib/libc/arch/m68k/string/ffs.S
> > ===================================================================
> > RCS file: /cvsroot/src/common/lib/libc/arch/m68k/string/ffs.S,v
> > retrieving revision 1.7
> > diff -p -u -r1.7 ffs.S
> > --- common/lib/libc/arch/m68k/string/ffs.S    9 Mar 2020 13:36:10 -0000 1.7
> > +++ common/lib/libc/arch/m68k/string/ffs.S    9 Mar 2020 14:09:13 -0000
> > @@ -45,7 +45,11 @@
> > /* bit = ffs(value) */
> > +#ifdef _LIBC
> > WEAK_ALIAS(__ffssi2,ffs)
> > +#else /* KERNEL */
> > +#define ffs __ffssi2
> > +#endif
> > #if (!defined(__mc68010__) && !defined(__mcoldfire__)) || defined(__mcfisac__)
> >
> > !DSPAM:5e664fb8246343444541299!
>
> +--------------------+--------------------------+-----------------------+
> | Paul Goyette       | PGP Key fingerprint:     | E-mail addresses:     |
> | (Retired)          | FA29 0E3B 35AF E8AE 6651 | paul%whooppee.com@localhost     |
> | Software Developer | 0786 F758 55DE 53BA 7731 | pgoyette%netbsd.org@localhost   |
> +--------------------+--------------------------+-----------------------+