Re: [Qemu-devel] [Patch] Support translating Guest physical address to Host virtual address.

[Max Asbock]

On Wed, 2010-01-27 at 15:39 -0600, Anthony Liguori wrote:
> On 01/26/2010 09:25 PM, Zheng, Jiajia wrote:
> > Add command p2v to translate Guest physical address to Host virtual address.
> >    
> 
> For what purpose?
> 
> > Signed-off-by: Max Asbock<address@hidden>
> >             Jiajia Zheng<address@hidden>
> > ---
> > diff --git a/monitor.c b/monitor.c
> > index b33b01f..83d9ac7 100644
> > --- a/monitor.c
> > +++ b/monitor.c
> > @@ -668,6 +668,11 @@ static void do_info_uuid(Monitor *mon, QObject 
> > **ret_data)
> >       *ret_data = qobject_from_jsonf("{ 'UUID': %s }", uuid);
> >   }
> >
> > +static void do_info_p2v(Monitor *mon)
> > +{
> > +    monitor_printf(mon, "p2v implemented\n");
> > +}
> >    
> 
> These should be implemented as QMP commands.
> 
> >   /* get the current CPU defined by the user */
> >   static int mon_set_cpu(int cpu_index)
> >   {
> > @@ -2283,6 +2288,14 @@ static void do_inject_mce(Monitor *mon, const QDict 
> > *qdict)
> >               break;
> >           }
> >   }
> > +static void do_p2v(Monitor *mon, const QDict *qdict)
> > +{
> > +    target_long size = 4096;
> > +    target_long addr = qdict_get_int(qdict, "addr");
> > +
> > +    monitor_printf(mon, "Guest physical address %p is mapped at host 
> > virtual address %p\n", (void *)addr, cpu_physical_memory_map( 
> > (target_phys_addr_t)addr, (target_phys_addr_t *)&size, 0));
> >    
> 
> This isn't quite right.  It assumes TARGET_PAGE_SIZE is 4k which is 
> certainly not always true.  It also assumes that 
> cpu_physical_memory_map() something that has some meaning which isn't 
> necessarily the case.  It could be a pointer to a bounce buffer.
> 
> Could you give an end-to-end description of how you expect this 
> mechanism to be used so we can work out a more appropriate set of 
> interfaces.  I assume this is MCE related.
> 

The purpose of this is to translate a guest physical address to a host
virtual address. 
This was indeed used for MCE testing. The p2v command provides one step
in a chain of translations from guest virtual to guest physical to host
virtual to host physical. Host physical is then used to inject a machine
check error. As a consequence the HPOISON code on the host and the MCE
injection code in qemu are exercised.
I was always assuming that this implementation perhaps isn't the most
optimal, but it simply worked for our test case.

What would an appropriate method be to get a host virtual address for
guest physical address that represents a page of RAM?

thanks,
Max