Re: [PATCH v5 18/19] KVM: Dynamically size memslot array based on number of used slots

From: Peter Xu
Date: Fri Feb 07 2020 - 11:05:56 EST


On Fri, Feb 07, 2020 at 07:38:29AM -0800, Sean Christopherson wrote:
> On Thu, Feb 06, 2020 at 05:12:08PM -0500, Peter Xu wrote:
> > On Tue, Jan 21, 2020 at 02:31:56PM -0800, Sean Christopherson wrote:
> > > Now that the memslot logic doesn't assume memslots are always non-NULL,
> > > dynamically size the array of memslots instead of unconditionally
> > > allocating memory for the maximum number of memslots.
> > >
> > > Note, because a to-be-deleted memslot must first be invalidated, the
> > > array size cannot be immediately reduced when deleting a memslot.
> > > However, consecutive deletions will realize the memory savings, i.e.
> > > a second deletion will trim the entry.
> > >
> > > Tested-by: Christoffer Dall <christoffer.dall@xxxxxxx>
> > > Tested-by: Marc Zyngier <maz@xxxxxxxxxx>
> > > Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> > > ---
> > > include/linux/kvm_host.h | 2 +-
> > > virt/kvm/kvm_main.c | 31 ++++++++++++++++++++++++++++---
> > > 2 files changed, 29 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> > > index 60ddfdb69378..8bb6fb127387 100644
> > > --- a/include/linux/kvm_host.h
> > > +++ b/include/linux/kvm_host.h
> > > @@ -431,11 +431,11 @@ static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
> > > */
> > > struct kvm_memslots {
> > > u64 generation;
> > > - struct kvm_memory_slot memslots[KVM_MEM_SLOTS_NUM];
> > > /* The mapping table from slot id to the index in memslots[]. */
> > > short id_to_index[KVM_MEM_SLOTS_NUM];
> > > atomic_t lru_slot;
> > > int used_slots;
> > > + struct kvm_memory_slot memslots[];
> >
> > This patch is tested so I believe this works, however normally I need
> > to do similar thing with [0] otherwise gcc might complaint. Is there
> > any trick behind to make this work? Or is that because of different
> > gcc versions?
>
> array[] and array[0] have the same net affect, but array[] is given special
> treatment by gcc to provide extra sanity checks, e.g. requires the field to
> be the end of the struct. Last I checked, gcc also doesn't allow array[]
> in unions. There are probably other restrictions.
>
> But, it's precisely because of those restrictions that using array[] is
> preferred, as it provides extra protections, e.g. if someone moved memslots
> to the top of the struct it would fail to compile.

However...

xz-x1:tmp $ cat a.c
struct a {
int s[];
};

int main(void) { }
xz-x1:tmp $ make a
cc a.c -o a
a.c:2:9: error: flexible array member in a struct with no named members
2 | int s[];
| ^
make: *** [<builtin>: a] Error 1

My gcc version is 9.2.1 20190827 (Red Hat 9.2.1-1) (GCC).

>
> > > };
> > >
> > > struct kvm {
> > > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> > > index 9b614cf2ca20..ed392ce64e59 100644
> > > --- a/virt/kvm/kvm_main.c
> > > +++ b/virt/kvm/kvm_main.c
> > > @@ -565,7 +565,7 @@ static struct kvm_memslots *kvm_alloc_memslots(void)
> > > return NULL;
> > >
> > > for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
> > > - slots->id_to_index[i] = slots->memslots[i].id = -1;
> > > + slots->id_to_index[i] = -1;
> > >
> > > return slots;
> > > }
> > > @@ -1077,6 +1077,32 @@ static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
> > > return old_memslots;
> > > }
> > >
> > > +/*
> > > + * Note, at a minimum, the current number of used slots must be allocated, even
> > > + * when deleting a memslot, as we need a complete duplicate of the memslots for
> > > + * use when invalidating a memslot prior to deleting/moving the memslot.
> > > + */
> > > +static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
> > > + enum kvm_mr_change change)
> > > +{
> > > + struct kvm_memslots *slots;
> > > + size_t old_size, new_size;
> > > +
> > > + old_size = sizeof(struct kvm_memslots) +
> > > + (sizeof(struct kvm_memory_slot) * old->used_slots);
> > > +
> > > + if (change == KVM_MR_CREATE)
> > > + new_size = old_size + sizeof(struct kvm_memory_slot);
> > > + else
> > > + new_size = old_size;
> > > +
> > > + slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
> > > + if (likely(slots))
> > > + memcpy(slots, old, old_size);
> >
> > (Maybe directly copy into it?)
>
> I don't follow, are you saying do "*slots = *old"?
>
> @new_size and @old_size are not guaranteed to be the same. More
> specifically, slots->memslots and old->slots are now flexible arrays with
> potentially different sizes. Doing "*slots = *old" would only copy the
> standard members, a memcpy() would still be needed for @memlots.
>
> A more effecient implementation would be:
>
> slots = kvalloc(new_size, GFP_KERNEL_ACCOUNT);
> if (likely(slots)) {
> memcpy(slots, old, old_size);
> if (change == KVM_MR_CREATE)
> memset((void *)slots + old_size, 0, new_size - old_size);
> }
>
> to avoid unnecessarily zeroing out the entire thing. I opted for the
> simpler implementation as this is not performance critical code, for most
> cases @slots won't be all that large, and I wanted to be absolutely sure
> any mixup would hit zeroed memory and not uninitialized memory.

I made a silly mistake on reading "slots" as "old". Ignore my
comment, sorry! And please take my R-b for this patch too:

Reviewed-by: Peter Xu <peterx@xxxxxxxxxx>

--
Peter Xu