Re: [Qemu-ppc] [PATCH] spapr_cpu_core: instantiate CPUs separately

[Greg Kurz]

On Tue, 7 Nov 2017 06:04:55 +1100
David Gibson <address@hidden> wrote:

> On Mon, Nov 06, 2017 at 04:03:07PM +0100, Greg Kurz wrote:
> > On Tue, 17 Oct 2017 17:16:09 +1100
> > David Gibson <address@hidden> wrote:
> >   
> > > On Mon, Oct 16, 2017 at 10:26:38AM +0200, Igor Mammedov wrote:  
> > > > On Sat, 14 Oct 2017 20:33:37 +1100
> > > > David Gibson <address@hidden> wrote:
> > > >     
> > > > > On Fri, Oct 13, 2017 at 01:31:44PM +0200, Greg Kurz wrote:    
> > > > > > The current code assumes that only the CPU core object holds a
> > > > > > reference on each individual CPU object, and happily frees their
> > > > > > allocated memory when the core is unrealized. This is dangerous
> > > > > > as some other code can legitimely keep a pointer to a CPU if it
> > > > > > calls object_ref(), but it would end up with a dangling pointer.
> > > > > > 
> > > > > > Let's allocate all CPUs with object_new() and let QOM frees them
> > > > > > when their reference count reaches zero. This greatly simplify the
> > > > > > code as we don't have to fiddle with the instance size anymore.
> > > > > > 
> > > > > > Signed-off-by: Greg Kurz <address@hidden>      
> > > > > 
> > > > > So, I'm pretty sure my first drafts of the core stuff did things this
> > > > > waym and it got nacked, for QOM lifetime reasons that I never really
> > > > > understood.    
> > > > From what I remember, Andreas would like to see composite CPU object
> > > > allocated in one go and then its children initialized with 
> > > > object_initialize()
> > > > so that no more allocation were needed.    
> > > 
> > > Ah, ok.
> > >   
> > > > That potentially would benefit hotplug, since we could gracefully
> > > > fail object creation early if there is not enough memory.    
> > > 
> > > Yeah, it sounds nice, but I don't see how we can do it.  In order to
> > > do that the core object has to have enough space for all the threads,
> > > which means we need both the size of each thread object and the number
> > > of them.  The size we have (and will be easier to handle after Igor's
> > > cleanups).  The number, we don't.
> > >   
> > > > But the way it's implemented currently doesn't really match that initial
> > > > goal as array for threads is dynamically allocated later
> > > > and then we need to dance around it with pointer arithmetic.
> > > > 
> > > > BTW: almost any allocation failure in qemu currently
> > > > is fatal so whether we fail on array alloc or on individual
> > > > object_new() won't make any difference.
> > > > 
> > > > I'd rather see this clean up merged as it simplifies code
> > > > in these case.    
> > > 
> > > Ok, works for me.
> > >   
> > 
> > Since we're in soft freeze already, I guess this won't go to 2.11. Maybe 
> > it's
> > time to create ppc-for-2.12 and apply it there ?  
> 
> Yeah, sounds like a plan.
> 

Friendly reminder: can you push this to ppc-for-2.12 so that it doesn't fall
through the cracks ? :)

Cheers,

--
Greg

pgphlW7atRe0k.pgp
Description: OpenPGP digital signature