kvm: Some more trivial fixes for clear dirty log

Two issues I've noticed when I'm drafting the QEMU support of it. With these two patches applied the QEMU binary (with the clear dirty log supported [1]) can migrate correctly otherwise the migration can stall forever if with/after heavy memory workload.

Please have a look, thanks.

[1] https://github.com/xzpeter/qemu/tree/kvm-clear-dirty-log

Peter Xu (2):
kvm: Fix the bitmap range to copy during clear dirty
kvm: Fix loop of clear dirty with off-by-one

 virt/kvm/kvm_main.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

2.17.1

kvm_dirty_bitmap_bytes() will return the size of the dirty bitmap of the memslot rather than the size of bitmap passed over from the ioctl. Here for KVM_CLEAR_DIRTY_LOG we should only copy exactly the size of bitmap that covers kvm_clear_dirty_log.num_pages.

---
 virt/kvm/kvm_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index 53de2f946f9e..ad39c57de82d 100644
@@ -1251,7 +1251,7 @@ int kvm_clear_dirty_log_protect(struct kvm *kvm,
 	if (!dirty_bitmap)
 		return -ENOENT;
 
-	n = kvm_dirty_bitmap_bytes(memslot);
+	n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
 
 	if (log->first_page > memslot->npages ||
 	    log->num_pages > memslot->npages - log->first_page ||

2.17.1

Just imaging the case where num_pages < BITS_PER_LONG, then the loop will be skipped while it shouldn't.

---
 virt/kvm/kvm_main.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index ad39c57de82d..15601a19a87a 100644
@@ -1264,8 +1264,8 @@ int kvm_clear_dirty_log_protect(struct kvm *kvm,
 		return -EFAULT;
 
 	spin_lock(&kvm->mmu_lock);
-	for (offset = log->first_page,
-	     i = offset / BITS_PER_LONG, n = log->num_pages / BITS_PER_LONG; n--;
+	for (offset = log->first_page, i = offset / BITS_PER_LONG,
+		 n = (log->num_pages / BITS_PER_LONG + 1); n--;
 	     i++, offset += BITS_PER_LONG) {
 		unsigned long mask = *dirty_bitmap_buffer++;
 		atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];

2.17.1

Looks good, but this is a blocker for using this feature in userspace. I think we should change the capability name and number.

Paolo

Ok, let me add another patch for it. Also I probably have made a mistake too in patch 2 (it's not really off-by-one, but it should be a DIV_ROUND_UP I think...). I'll post a new version soon.

Thanks,

Peter Xu