Re: [PATCH v5 12/19] KVM: Move memslot deletion to helper function

[Sean Christopherson]

On Thu, Feb 06, 2020 at 11:51:16AM -0500, Peter Xu wrote:
> On Thu, Feb 06, 2020 at 08:28:18AM -0800, Sean Christopherson wrote:
> > On Thu, Feb 06, 2020 at 11:14:15AM -0500, Peter Xu wrote:
> > > On Tue, Jan 21, 2020 at 02:31:50PM -0800, Sean Christopherson wrote:
> > > > Move memslot deletion into its own routine so that the success path for
> > > > other memslot updates does not need to use kvm_free_memslot(), i.e. can
> > > > explicitly destroy the dirty bitmap when necessary.  This paves the way
> > > > for dropping @dont from kvm_free_memslot(), i.e. all callers now pass
> > > > NULL for @dont.
> > > > 
> > > > Add a comment above the code to make a copy of the existing memslot
> > > > prior to deletion, it is not at all obvious that the pointer will become
> > > > stale during sorting and/or installation of new memslots.
> > > 
> > > Could you help explain a bit on this explicit comment?  I can follow
> > > up with the patch itself which looks all correct to me, but I failed
> > > to catch what this extra comment wants to emphasize...
> > 
> > It's tempting to write the code like this (I know, because I did it):
> > 
> > 	if (!mem->memory_size)
> > 		return kvm_delete_memslot(kvm, mem, slot, as_id);
> > 
> > 	new = *slot;
> > 
> > Where @slot is a pointer to the memslot to be deleted.  At first, second,
> > and third glances, this seems perfectly sane.
> > 
> > The issue is that slot was pulled from struct kvm_memslots.memslots, e.g.
> > 
> > 	slot = &slots->memslots[index];
> > 
> > Note that slots->memslots holds actual "struct kvm_memory_slot" objects,
> > not pointers to slots.  When update_memslots() sorts the slots, it swaps
> > the actual slot objects, not pointers.  I.e. after update_memslots(), even
> > though @slot points at the same address, it's could be pointing at a
> > different slot.  As a result kvm_free_memslot() in kvm_delete_memslot()
> > will free the dirty page info and arch-specific points for some random
> > slot, not the intended slot, and will set npages=0 for that random slot.
> 
> Ah I see, thanks.  Another alternative is we move the "old = *slot"
> copy into kvm_delete_memslot(), which could be even clearer imo.

The copy is also needed in __kvm_set_memory_region() for the MOVE case.

> However I'm not sure whether it's a good idea to drop the test-by for
> this.  Considering that comment change should not affect it, would you
> mind enrich the comment into something like this (or anything better)?
> 
> /*
>  * Make a full copy of the old memslot, the pointer will become stale
>  * when the memslots are re-sorted by update_memslots() in
>  * kvm_delete_memslot(), while to make the kvm_free_memslot() work as
>  * expected later on, we still need the cached memory slot.
>  */

As above, it's more subtle than just the kvm_delete_memslot() case.

	/*
	 * Make a full copy of the old memslot, the pointer will become stale
	 * when the memslots are re-sorted by update_memslots() when deleting
	 * or moving a memslot, and additional modifications to the old memslot
	 * need to be made after calling update_memslots().
	 */