[RFCv1,4/8] kvm: Introduce secondary dirty bitmap

Commit Message

Gavin Shan Feb. 6, 2023, 11:20 a.m. UTC

When dirty ring is enabled on ARM64, the backup bitmap may be used
to track the dirty pages in no-running-vcpu situations. The original
bitmap is the primary one, used for the dirty ring buffer. We need
the secondary bitmap to collect the backup bitmap for ARM64.

No functional change intended.

Signed-off-by: Gavin Shan <[e-mail]>
---
 accel/kvm/kvm-all.c      | 50 ++++++++++++++++++++++++++++++----------
 include/sysemu/kvm_int.h |  1 +
 2 files changed, 39 insertions(+), 12 deletions(-)

Comments

Juan Quintela Feb. 8, 2023, 10:07 p.m. UTC | #1

Gavin Shan <[e-mail]> wrote:
> When dirty ring is enabled on ARM64, the backup bitmap may be used
> to track the dirty pages in no-running-vcpu situations. The original
> bitmap is the primary one, used for the dirty ring buffer. We need
> the secondary bitmap to collect the backup bitmap for ARM64.
>
> No functional change intended.
>
> Signed-off-by: Gavin Shan <[e-mail]>
> ---
>  accel/kvm/kvm-all.c      | 50 ++++++++++++++++++++++++++++++----------
>  include/sysemu/kvm_int.h |  1 +
>  2 files changed, 39 insertions(+), 12 deletions(-)
>
> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
> index 01a6a026af..1a93985574 100644
> --- a/accel/kvm/kvm-all.c
> +++ b/accel/kvm/kvm-all.c
> @@ -553,13 +553,29 @@ static void kvm_log_stop(MemoryListener *listener,
>      }
>  }
>  
> +static unsigned long *kvm_slot_dirty_bitmap(KVMSlot *slot, bool primary)
> +{
> +    if (primary) {
> +        return slot->dirty_bmap;
> +    }
> +
> +    return slot->dirty_bmap +
> +           slot->dirty_bmap_size / sizeof(slot->dirty_bmap[0]);
> +}


Why?
Just use two bitmaps and call it a day.

Later, Juan.

Gavin Shan Feb. 9, 2023, 9:42 a.m. UTC | #2

On 2/9/23 9:07 AM, Juan Quintela wrote:
> Gavin Shan <[e-mail]> wrote:
>> When dirty ring is enabled on ARM64, the backup bitmap may be used
>> to track the dirty pages in no-running-vcpu situations. The original
>> bitmap is the primary one, used for the dirty ring buffer. We need
>> the secondary bitmap to collect the backup bitmap for ARM64.
>>
>> No functional change intended.
>>
>> Signed-off-by: Gavin Shan <[e-mail]>
>> ---
>>   accel/kvm/kvm-all.c      | 50 ++++++++++++++++++++++++++++++----------
>>   include/sysemu/kvm_int.h |  1 +
>>   2 files changed, 39 insertions(+), 12 deletions(-)
>>
>> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
>> index 01a6a026af..1a93985574 100644
>> --- a/accel/kvm/kvm-all.c
>> +++ b/accel/kvm/kvm-all.c
>> @@ -553,13 +553,29 @@ static void kvm_log_stop(MemoryListener *listener,
>>       }
>>   }
>>   
>> +static unsigned long *kvm_slot_dirty_bitmap(KVMSlot *slot, bool primary)
>> +{
>> +    if (primary) {
>> +        return slot->dirty_bmap;
>> +    }
>> +
>> +    return slot->dirty_bmap +
>> +           slot->dirty_bmap_size / sizeof(slot->dirty_bmap[0]);
>> +}
> 
> 
> Why?
> Just use two bitmaps and call it a day.
> 

Thanks for your review, Juan. Right, I had wrong assumption that the original
(primary) bitmap can't be reused. It's why the secondary bitmap is introduced.
The intention is to use the original (primary) bitmap to cover the dirty-ring
buffers while the secondary bitmap is to cover the backup bitmap, resident in
host kernel.

I think the original (primary) bitmap can be reused in this case. After the
dirty-ring buffer is synchronized to the original bitmap, which is updated
to the dirty bits. It can be reused to cover the backup bitmap. I will remove
the secondary bitmap in next revision.

Thanks,
Gavin

Patch

diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
index 01a6a026af..1a93985574 100644
--- a/accel/kvm/kvm-all.c
+++ b/accel/kvm/kvm-all.c
@@ -553,13 +553,29 @@  static void kvm_log_stop(MemoryListener *listener,
     }
 }
 
+static unsigned long *kvm_slot_dirty_bitmap(KVMSlot *slot, bool primary)
+{
+    if (primary) {
+        return slot->dirty_bmap;
+    }
+
+    return slot->dirty_bmap +
+           slot->dirty_bmap_size / sizeof(slot->dirty_bmap[0]);
+}
+
 /* get kvm's dirty pages bitmap and update qemu's */
-static void kvm_slot_sync_dirty_pages(KVMSlot *slot)
+static void kvm_slot_sync_dirty_pages(KVMSlot *slot, bool primary)
 {
+    KVMState *s = kvm_state;
+    unsigned long *bmap = kvm_slot_dirty_bitmap(slot, primary);
     ram_addr_t start = slot->ram_start_offset;
     ram_addr_t pages = slot->memory_size / qemu_real_host_page_size();
 
-    cpu_physical_memory_set_dirty_lebitmap(slot->dirty_bmap, start, pages);
+    if (!s->kvm_dirty_ring_with_bitmap && !primary) {
+        return;
+    }
+
+    cpu_physical_memory_set_dirty_lebitmap(bmap, start, pages);
 }
 
 static void kvm_slot_reset_dirty_pages(KVMSlot *slot)
@@ -572,6 +588,9 @@  static void kvm_slot_reset_dirty_pages(KVMSlot *slot)
 /* Allocate the dirty bitmap for a slot  */
 static void kvm_slot_init_dirty_bitmap(KVMSlot *mem)
 {
+    KVMState *s = kvm_state;
+    hwaddr bitmap_size, alloc_size;
+
     if (!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) || mem->dirty_bmap) {
         return;
     }
@@ -593,9 +612,11 @@  static void kvm_slot_init_dirty_bitmap(KVMSlot *mem)
      * And mem->memory_size is aligned to it (otherwise this mem can't
      * be registered to KVM).
      */
-    hwaddr bitmap_size = ALIGN(mem->memory_size / qemu_real_host_page_size(),
-                                        /*HOST_LONG_BITS*/ 64) / 8;
-    mem->dirty_bmap = g_malloc0(bitmap_size);
+    bitmap_size = ALIGN(mem->memory_size / qemu_real_host_page_size(),
+                        /*HOST_LONG_BITS*/ 64) / 8;
+    alloc_size = s->kvm_dirty_ring_with_bitmap ? 2 * bitmap_size : bitmap_size;
+
+    mem->dirty_bmap = g_malloc0(alloc_size);
     mem->dirty_bmap_size = bitmap_size;
 }
 
@@ -603,12 +624,16 @@  static void kvm_slot_init_dirty_bitmap(KVMSlot *mem)
  * Sync dirty bitmap from kernel to KVMSlot.dirty_bmap, return true if
  * succeeded, false otherwise
  */
-static bool kvm_slot_get_dirty_log(KVMState *s, KVMSlot *slot)
+static bool kvm_slot_get_dirty_log(KVMState *s, KVMSlot *slot, bool primary)
 {
     struct kvm_dirty_log d = {};
     int ret;
 
-    d.dirty_bitmap = slot->dirty_bmap;
+    if (!s->kvm_dirty_ring_with_bitmap && !primary) {
+        return false;
+    }
+
+    d.dirty_bitmap = kvm_slot_dirty_bitmap(slot, primary);
     d.slot = slot->slot | (slot->as_id << 16);
     ret = kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d);
 
@@ -839,8 +864,8 @@  static void kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml,
             /* We don't have a slot if we want to trap every access. */
             return;
         }
-        if (kvm_slot_get_dirty_log(s, mem)) {
-            kvm_slot_sync_dirty_pages(mem);
+        if (kvm_slot_get_dirty_log(s, mem, true)) {
+            kvm_slot_sync_dirty_pages(mem, true);
         }
         start_addr += slot_size;
         size -= slot_size;
@@ -1353,9 +1378,9 @@  static void kvm_set_phys_mem(KVMMemoryListener *kml,
                 if (kvm_state->kvm_dirty_ring_size) {
                     kvm_dirty_ring_reap_locked(kvm_state, NULL);
                 } else {
-                    kvm_slot_get_dirty_log(kvm_state, mem);
+                    kvm_slot_get_dirty_log(kvm_state, mem, true);
                 }
-                kvm_slot_sync_dirty_pages(mem);
+                kvm_slot_sync_dirty_pages(mem, true);
             }
 
             /* unregister the slot */
@@ -1572,7 +1597,7 @@  static void kvm_log_sync_global(MemoryListener *l, bool last_stage)
     for (i = 0; i < s->nr_slots; i++) {
         mem = &kml->slots[i];
         if (mem->memory_size && mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
-            kvm_slot_sync_dirty_pages(mem);
+            kvm_slot_sync_dirty_pages(mem, true);
             /*
              * This is not needed by KVM_GET_DIRTY_LOG because the
              * ioctl will unconditionally overwrite the whole region.
@@ -3701,6 +3726,7 @@  static void kvm_accel_instance_init(Object *obj)
     s->kernel_irqchip_split = ON_OFF_AUTO_AUTO;
     /* KVM dirty ring is by default off */
     s->kvm_dirty_ring_size = 0;
+    s->kvm_dirty_ring_with_bitmap = false;
     s->notify_vmexit = NOTIFY_VMEXIT_OPTION_RUN;
     s->notify_window = 0;
 }
diff --git a/include/sysemu/kvm_int.h b/include/sysemu/kvm_int.h
index 60b520a13e..fdd5b1bde0 100644
--- a/include/sysemu/kvm_int.h
+++ b/include/sysemu/kvm_int.h
@@ -115,6 +115,7 @@  struct KVMState
     } *as;
     uint64_t kvm_dirty_ring_bytes;  /* Size of the per-vcpu dirty ring */
     uint32_t kvm_dirty_ring_size;   /* Number of dirty GFNs per ring */
+    bool kvm_dirty_ring_with_bitmap;
     struct KVMDirtyRingReaper reaper;
     NotifyVmexitOption notify_vmexit;
     uint32_t notify_window;