qemu discuss - Feb 2007 - Re: qemu-0.9.0 with initial tap interface

> >% gdiff  ../kqemu-osol-1.3.0pre9/kqemu-solaris.c . 
> >54c54
> >< static uint_t kqemu_busy = 0;
> >---
> >  
> >
> >>static uint32_t kqemu_busy = 0;
> >>    
> >>
> >73c73
> >< static uint_t upages_locked = 0;
> >---
> >  
> >
> >>static uint32_t upages_locked = 0;
> >>    
> >>
> >102c102
> ><               atomic_inc_uint(&upages_locked);
> >---
> >  
> >
> >>              atomic_add_32(&upages_locked, 1);
> >>    
> >>
> >141c141
> ><               atomic_inc_uint(&upages_locked);
> >---
> >  
> >
> >>              atomic_add_32(&upages_locked, 1);
> >>    
> >>
> >158c158
> ><       atomic_dec_uint(&upages_locked);
> >---
> >  
> >
> >>      atomic_add_32(&upages_locked, -1);
> >>    
> >>
> >626c626
> ><               atomic_inc_uint(&kqemu_busy);
> >---
> >  
> >
> >>              atomic_add_32(&kqemu_busy, 1);
> >>    
> >>
> >642c642
> ><               atomic_dec_uint(&kqemu_busy);
> >---
> >  
> >
> >>              atomic_add_32(&kqemu_busy, -1);
> >>    
> >>
> And it all works?  Interesting.  ...
> 
> So the next question is, do you think this will work on Solaris 9/X86?

The 32-bit kqemu binary compiled on snv_55 loads on a S9 x86 kernel;
but panics the kernel as soon as qemu tries to use the /dev/kqemu
device.

AFAICT, the problem is that the kthread_t structure has changed between
s9 and s10/s11, so that the statement...

struct kqemu_user_page * CDECL
kqemu_lock_user_page(unsigned long *ppage_index, unsigned long user_addr)
{
        proc_t *p = curproc;


sets ''p'' to a junk proc_t*.  



% grep curproc /usr/include/sys/thread.h 
#define curproc         (ttoproc(curthread))    /* current process pointer */


Compiling on S9 for S9 might fix it, but I''ve not yet tested that...

I wrote:
> The 32-bit kqemu binary compiled on snv_55 loads on a S9 x86 kernel;
> but panics the kernel as soon as qemu tries to use the /dev/kqemu
> device.
> 
> AFAICT, the problem is that the kthread_t structure has changed between
> s9 and s10/s11, so that the statement...
> 
> struct kqemu_user_page * CDECL
> kqemu_lock_user_page(unsigned long *ppage_index,
> unsigned long user_addr)
> {
>         proc_t *p = curproc;
>
> sets ''p'' to a junk proc_t*.  
> 
> 
> Compiling on S9 for S9 might fix it, but I''ve not yet tested
that...
Trying to compile kqemu-solaris.c on S9 revealed more problems:
the kernel''s "struct page" has changed between S9 and
S10/Nevada,
too.  The physical page number member p_pagenum in page_t is
missing.

Changes like this (kthread_t, page_t) could also affect S10 and Nevada;
whenever kernel data structures like kthread_t, page_t changes
(patch installed, bfu''ed or upgraded to a newer OpenSolaris kernel),
the binary kqemu module could break and crash the kernel. In such a case,
the fix would be to recompile kqemu from sources and reinstall it.
--
This message posted from opensolaris.org