Skip to contentSkip to conversation

KVM: Aggressively drop and reacquire mmu_lock during CLEAR_DIRTY_LOG

PATCH Superseded David Matlack 2 participants 4 messages Dec 5, 2023 – Apr 2, 2024 KVM

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 KVM: Aggressively drop and reacquire mmu_lock during CLEAR_DIRTY_LOG

Aggressively drop and reacquire mmu_lock during CLEAR_DIRTY_LOG to avoid blocking other threads (e.g. vCPUs taking page faults) for too long.

Specifically, change kvm_clear_dirty_log_protect() to acquire/release mmu_lock only when calling kvm_arch_mmu_enable_log_dirty_pt_masked(), rather than around the entire for loop. This ensures that KVM will only hold mmu_lock for the time it takes the architecture-specific code to process up to 64 pages, rather than holding mmu_lock for log->num_pages, which is controllable by userspace. This also avoids holding mmu_lock when processing parts of the dirty_bitmap that are zero (i.e. when there is nothing to clear).

Moving the acquire/release points for mmu_lock should be safe since dirty_bitmap_buffer is already protected by slots_lock, and dirty_bitmap is already accessed with atomic_long_fetch_andnot(). And at least on x86 holding mmu_lock doesn't even serialize access to the memslot dirty bitmap, as vCPUs can call mark_page_dirty_in_slot() without holding mmu_lock.

This change eliminates dips in guest performance during live migration in a 160 vCPU VM when userspace is issuing CLEAR ioctls (tested with 1GiB and 8GiB CLEARs). Userspace could issue finer-grained CLEARs, which would also reduce contention on mmu_lock, but doing so will increase the rate of remote TLB flushing. And there's really no reason to punt this problem to userspace since KVM can just drop and reacquire mmu_lock more frequently.

Cc Marc Zyngier <[email protected]>
Cc Oliver Upton <[email protected]>
Cc Tianrui Zhao <[email protected]>
Cc Bibo Mao <[email protected]>
Cc Huacai Chen <[email protected]>
Cc Michael Ellerman <[email protected]>
Cc Anup Patel <[email protected]>
Cc Christian Borntraeger <[email protected]>
Cc Janosch Frank <[email protected]>
Cc Claudio Imbrenda <[email protected]>
Cc Sean Christopherson <[email protected]>

---
NOTE: This patch was originally sent as part of another series [1].

[1] https://lore.kernel.org/kvm/[email protected]/

 virt/kvm/kvm_main.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
virt/kvm/kvm_main.c +2 -2
index 486800a7024b..afa61a2309d2 100644
@@ -2297,7 +2297,6 @@ static int kvm_clear_dirty_log_protect(struct kvm *kvm,
 	if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
 		return -EFAULT;
 
-	KVM_MMU_LOCK(kvm);
 	for (offset = log->first_page, i = offset / BITS_PER_LONG,
 		 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
 	     i++, offset += BITS_PER_LONG) {
@@ -2316,11 +2315,12 @@ static int kvm_clear_dirty_log_protect(struct kvm *kvm,
 		*/
 		if (mask) {
 			flush = true;
+			KVM_MMU_LOCK(kvm);
 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
 								offset, mask);
+			KVM_MMU_UNLOCK(kvm);
 		}
 	}
-	KVM_MMU_UNLOCK(kvm);
 
 	if (flush)
 		kvm_flush_remote_tlbs_memslot(kvm, memslot);

base-commit: 45b890f7689eb0aba454fc5831d2d79763781677

Signature · 1 line
2.43.0.rc2.451.g8631bc7472-goog
#1 lore raw
Quoted 3 lines · David Matlack

Aggressively drop and reacquire mmu_lock during CLEAR_DIRTY_LOG to avoid blocking other threads (e.g. vCPUs taking page faults) for too long.

Ping.

KVM architecture maintainers: Do you have any concerns about the correctness of this patch? I'm confident dropping the lock is correct on x86 and it should be on other architectures as well, but confirmation would be helpful.

Thanks.

#2 lore raw
Quoted 6 lines · David Matlack
Quoted 3 lines · David Matlack

Aggressively drop and reacquire mmu_lock during CLEAR_DIRTY_LOG to avoid blocking other threads (e.g. vCPUs taking page faults) for too long.

Quoted 8 lines

Ping.

KVM architecture maintainers: Do you have any concerns about the correctness of this patch? I'm confident dropping the lock is correct on x86 and it should be on other architectures as well, but confirmation would be helpful.

Thanks.

Ping again. This patch has been sitting since December 5th. Is there anything I can do to help get it merged?

#3 lore raw
Quoted 14 lines · David Matlack
Quoted 13 lines · David Matlack
Quoted 3 lines · David Matlack

Aggressively drop and reacquire mmu_lock during CLEAR_DIRTY_LOG to avoid blocking other threads (e.g. vCPUs taking page faults) for too long.

Ping.

KVM architecture maintainers: Do you have any concerns about the correctness of this patch? I'm confident dropping the lock is correct on x86 and it should be on other architectures as well, but confirmation would be helpful.

Quoted 4 lines
Quoted 1 line

Thanks.

Ping again. This patch has been sitting since December 5th. Is there anything I can do to help get it merged?

Please send a rebased version as a V2. With the number of MMU patches flying around without much coordination, it is hard to keep track.

Thanks,

        M.
Signature · 1 line
Jazz is not dead. It just smells funny...