[Qemu-devel] Re: playing with qemu usermode emulation on FreeBSD...

[Juergen Lock]

On Mon, Oct 12, 2009 at 10:55:24PM +0300, Blue Swirl wrote:
> On Mon, Oct 12, 2009 at 1:18 AM, Juergen Lock <nox at jelal.kn-bremen.de> wrote:
> > On Thu, Oct 08, 2009 at 12:05:49AM +0200, Juergen Lock wrote:
> >> I recently noticed there are x86 bsd-user targets now (yeah I totally
> >> missed those commits...) and now got it working a tiny little bit:
> >> I can run
> >>       qemu-x86_64 -bsd freebsd /rescue/echo foo bar
> >> here on FreeBSD 8/amd64 and it echoes foo bar as expected, but
> >> segfaults afterwards. :)  (in pthread_setcancelstate() invoked from
> >> a guest write() syscall, in case anyone is wondering.)  Other things
> >> I tried either exit with errors or segfault as well, and i386 hosts
> >> probably still don't work at all yet.  (qemu-i386 here on amd64 does
> >> at least something, but probably needs lock_user() treatment for all
> >> kinds of syscalls, I only tried adding that for sysctl so far.)
> >>
> >>  Anyway, here is an emulators/qemu-devel git head snapshot port
> >> update with my current patches (files/patch-bsd-user), feel free to
> >> test/debug/improve:
> >>       http://people.freebsd.org/~nox/qemu/qemu-devel-20091007.patch
> >> (For the folks reading this on the qemu list:  I shall start doing
> >> `proper' patch submissions later, this is more for the FreeBSD folks
> >> and because I was asked to send what I have...)
> >
> > New version at the same place, which now runs FreeBSD/{i386,sparc64}
> > /rescue/echo on FreeBSD/amd64, the FreeBSD/amd64 target now segfaults
> > in pthread_setcancelstate() invoked from the final writev() tho.
> > Oh and I also uploaded the snapshot tarball so others can now actually
> > build the port too... :)  And I have switched to the cpu-exec.c patch
> > posted by Aleksej Saushev on the qemu list and added back amd64
> > code there.
> >
> >  Here is the bsd-user patch again:
> 
> Please add Signed-off-by: line and use 'diff -u' (or preferably git diff).
> 
Well I wasn't expecting this diff to be committed just yet anyway,
it's still more a wip version...

> > +    if (1 /* bsd_type == target_freebsd */)
> > +        regs->rdi = infop->start_stack;
> 
> Why the if and comment?
> 
> > +        if (1 /* bsd_type == target_freebsd */) {
> > +            regs->u_regs[8] = infop->start_stack;
> > +            regs->u_regs[11] = infop->start_stack;
> 
> Same here.
> 
 Because bsd_type isn't available at these places in the code but
probably should be checked, I still wanted to fix that.  (Maybe
make it global?)

> >         case 0x100:
> > +        /* FreeBSD uses 0x141 for syscalls too */
> > +        case 0x141:
> > +            if (bsd_type != target_freebsd)
> > +                goto badtrap;
> 
> You are now also trapping on case 0x100 if bsd_type != target_freebsd,
> which probably breaks other BSDs.
> 
 Right, thats broken, the 0x141 case should come before the 0x100
here of course.

> > +/* XXX this needs to be emulated on non-FreeBSD hosts... */
> > +static abi_long do_freebsd_sysctl(abi_ulong namep, int32_t namelen, abi_ulong oldp,
> > +                          abi_ulong oldlenp, abi_ulong newp, abi_ulong newlen)
> 
> What kind of call is this, is it possible to emulate on other BSDs? Is
> it important?

 Its used mostly for things that on linux is done by manipulating
/proc or /sys, like getting the kernel version, number of cpus,
pagesize, etc. - and there are also sysctls that can be written to,
like to enable ip forwarding or change sysV ipc settings.  Although
changes are usually restriced to root so `regular' executables rarely
do them and I'm not really handling those yet.  See here:
	http://www.freebsd.org/cgi/man.cgi?query=sysctl&apropos=0&sektion=3&manpath=FreeBSD+7.2-RELEASE&format=html

>  I'm just wondering if the cross-BSD emulation makes
> sense after all. It would make the emulator much simpler if we could
> assume that host_bsdness == target_bsdness.

 Yeah I was wondering about that too...

 Cheers,
	Juergen