svn commit: r219561 - in head/sys: cddl/dev/systrace modules/dtrace modules/dtrace/dtraceall modules/dtrace/systrace_freebsd32 modules/dtrace/systrace_linux32

[John Baldwin]

On Sunday, March 13, 2011 3:10:52 am Andriy Gapon wrote:
> on 13/03/2011 08:35 Artem Belevich said the following:
> > On Sat, Mar 12, 2011 at 12:14 PM, Alexander Leidinger
> > <Alexander at leidinger.net> wrote:
> >> But this looks like it will be named linux32 in any case. In the short
> >> term I would prefer:
> >> ---snip---
> >> #if defined(__amd64__)
> >> #define MODNAME "linux32"
> >> #elif defined(__i386__)
> >> #define MODNAME "linux"
> >> #endif
> >> ---snip---
> > 
> > Makes sense. It's what's done with freebsd syscalls -- 'freebsd' for
> > native syscalls, freebsd32 for 32-bit compat.
> 
> BTW, in my opinion, it might not make a lot of sense.
> That is, we can have native and compat FreeBSD system calls, but Linux system
> calls are always compat and never native, whether the emulation is for the same
> arch or not.  Explicit suffix makes things clearer.  But that's just my opinion.

However, if we were to go that route, you would need to have 'linux-i386',
'linux-x86-64', 'linux-alpha', etc.  Recall that our Alpha port had Linux compat
as well, so for it 'linux.ko' was a 64-bit ABI.  I think it makes sense to assume
that 'linux' maps to Linux compat for whatever the native platform of the running
kernel is and to then use suffixes for other platform modes (such as 32-bit compat
mode in amd64).

-- 
John Baldwin