Skip to contentSkip to conversation

migration: SMRAM dirty bitmap not fetched from kvm-kmod and not send to destination

Qemu-devel Herongguang (Stephen) 2 participants 14 messages Sep 14, 2016 – Sep 26, 2016

Subsystems (3)
x86 KVM KVM x86
lore
Share & tools
lore

Herongguang (Stephen)

author
#0 lore raw

Hi,
We found a problem that when a redhat 6 VM reboots (in grub countdown UI), migrating this VM will result in VM’s memory difference between source and destination side. The difference always resides in GPA 0xA0000~0xC0000, i.e. SMRAM area.

Occasionally this result in VM instruction emulation error in destination side.

After some digging, I think this is because in migration code, in migration_bitmap_sync(), only memory slots in address space address_space_memory’s dirty bitmap fetched from kvm-kmod, while SMRAM memory slot, in address space smram_address_space’s dirty bitmap not fetched from kvm-kmod, thus modifications in SMRAM in source side are not sent to destination side.

I tried following patch, and this phenomenon does not happen anymore. Do you think this patch is OK or do you have better idea? Thanks.

migration/ram.c +3 -0
index a3d70c4..1cc4360 100644
@@ -607,6 +607,8 @@ static void migration_bitmap_sync_init(void)
     iterations_prev = 0;
 }
 
+extern AddressSpace smram_address_space;
+
 static void migration_bitmap_sync(void)
 {
     RAMBlock *block;

@@ -627,6 +629,7 @@ static void migration_bitmap_sync(void)
 
     trace_migration_bitmap_sync_start();
     address_space_sync_dirty_bitmap(&address_space_memory);
+    address_space_sync_dirty_bitmap(&smram_address_space);
 
     qemu_mutex_lock(&migration_bitmap_mutex);
     rcu_read_lock();
target-i386/kvm.c +1 -1
index d1a25c5..b98fe22 100644
@@ -1111,7 +1111,7 @@ static int kvm_get_supported_msrs(KVMState *s)
 
 static Notifier smram_machine_done;
 static KVMMemoryListener smram_listener;
-static AddressSpace smram_address_space;
+AddressSpace smram_address_space;
 static MemoryRegion smram_as_root;
 static MemoryRegion smram_as_mem;
maintainer
#1 lore raw
Quoted 16 lines · Herongguang (Stephen)

Hi,
We found a problem that when a redhat 6 VM reboots (in grub countdown UI), migrating this VM will result in VM’s memory difference between source and destination side. The difference always resides in GPA 0xA0000~0xC0000, i.e. SMRAM area.

Occasionally this result in VM instruction emulation error in destination side.

After some digging, I think this is because in migration code, in migration_bitmap_sync(), only memory slots in address space address_space_memory’s dirty bitmap fetched from kvm-kmod, while SMRAM memory slot, in address space smram_address_space’s dirty bitmap not fetched from kvm-kmod, thus modifications in SMRAM in source side are not sent to destination side.

Quoted 2 lines

I tried following patch, and this phenomenon does not happen anymore. Do you think this patch is OK or do you have better idea? Thanks.

Nice caatch!

I think the right solution here is to sync all RAM memory regions instead of the address spaces. You can do that by putting a notifier in MemoryRegion; register the notifier in all the RAM creation functions (basically after every mr->ram=true or mr->rom_device=true), and unregister it in memory_region_destructor_ram.

Thanks,

Paolo

Quoted 36 lines
migration/ram.c +3 -0
index a3d70c4..1cc4360 100644
@@ -607,6 +607,8 @@ static void migration_bitmap_sync_init(void)
     iterations_prev = 0;
 }
 
+extern AddressSpace smram_address_space;
+
 static void migration_bitmap_sync(void)
 {
     RAMBlock *block;

@@ -627,6 +629,7 @@ static void migration_bitmap_sync(void)
 
     trace_migration_bitmap_sync_start();
     address_space_sync_dirty_bitmap(&address_space_memory);
+    address_space_sync_dirty_bitmap(&smram_address_space);
 
     qemu_mutex_lock(&migration_bitmap_mutex);
     rcu_read_lock();
target-i386/kvm.c +1 -1
index d1a25c5..b98fe22 100644
@@ -1111,7 +1111,7 @@ static int kvm_get_supported_msrs(KVMState *s)
 
 static Notifier smram_machine_done;
 static KVMMemoryListener smram_listener;
-static AddressSpace smram_address_space;
+AddressSpace smram_address_space;
 static MemoryRegion smram_as_root;
 static MemoryRegion smram_as_mem;
author
#2 lore raw
Quoted 30 lines · Paolo Bonzini
Quoted 16 lines · Herongguang (Stephen)

Hi,
We found a problem that when a redhat 6 VM reboots (in grub countdown UI), migrating this VM will result in VM’s memory difference between source and destination side. The difference always resides in GPA 0xA0000~0xC0000, i.e. SMRAM area.

Occasionally this result in VM instruction emulation error in destination side.

After some digging, I think this is because in migration code, in migration_bitmap_sync(), only memory slots in address space address_space_memory’s dirty bitmap fetched from kvm-kmod, while SMRAM memory slot, in address space smram_address_space’s dirty bitmap not fetched from kvm-kmod, thus modifications in SMRAM in source side are not sent to destination side.

Quoted 2 lines

I tried following patch, and this phenomenon does not happen anymore. Do you think this patch is OK or do you have better idea? Thanks.

Nice caatch!

I think the right solution here is to sync all RAM memory regions instead of the address spaces. You can do that by putting a notifier in MemoryRegion; register the notifier in all the RAM creation functions (basically after every mr->ram=true or mr->rom_device=true), and unregister it in memory_region_destructor_ram.

Quoted 4 lines

Thanks,

Paolo

I have some concern:
1. For example, vhost does not know about as_id, I wonder if guests in SMM can operate disk or ether card, as in that case vhost would not logging dirty pages correctly, without knowing as_id.

2. If a memory region is disabled/enabled/disabled frequently, since disabled memory regions would be removed from memory slots in kvm-kmod, dirty pages would be discarded in kvm-kmod and qemu when disabled, thus missing. Is my assumption correct?

3. I agree your opinion that the right solution is to get dirty-page information for all memory region from kvm-kmod. But I found it’s somewhat hard to implement since kvm_log_sync() expects a MemoryRegionSection* parameter. Do you have good idea?

As to all the ram memory regions, I think they are all in the ram_list.blocks, so there is no need to create a notifier, is this correct?

author
#3 lore raw
Quoted 38 lines · Herongguang (Stephen)
Quoted 30 lines · Paolo Bonzini
Quoted 16 lines · Herongguang (Stephen)

Hi,
We found a problem that when a redhat 6 VM reboots (in grub countdown UI), migrating this VM will result in VM’s memory difference between source and destination side. The difference always resides in GPA 0xA0000~0xC0000, i.e. SMRAM area.

Occasionally this result in VM instruction emulation error in destination side.

After some digging, I think this is because in migration code, in migration_bitmap_sync(), only memory slots in address space address_space_memory’s dirty bitmap fetched from kvm-kmod, while SMRAM memory slot, in address space smram_address_space’s dirty bitmap not fetched from kvm-kmod, thus modifications in SMRAM in source side are not sent to destination side.

Quoted 2 lines

I tried following patch, and this phenomenon does not happen anymore. Do you think this patch is OK or do you have better idea? Thanks.

Nice caatch!

I think the right solution here is to sync all RAM memory regions instead of the address spaces. You can do that by putting a notifier in MemoryRegion; register the notifier in all the RAM creation functions (basically after every mr->ram=true or mr->rom_device=true), and unregister it in memory_region_destructor_ram.

Quoted 4 lines

Thanks,

Paolo

Quoted 7 lines

I have some concern:
1. For example, vhost does not know about as_id, I wonder if guests in SMM can operate disk or ether card, as in that case vhost would not logging dirty pages correctly, without knowing as_id.

2. If a memory region is disabled/enabled/disabled frequently, since disabled memory regions would be removed from memory slots in kvm-kmod, dirty pages would be discarded in kvm-kmod and qemu when disabled, thus missing. Is my assumption correct?

After reviewing code, I think question 2 does not exist as qemu will sync dirty page before removing memory slots in kvm_set_phys_mem.

Quoted 7 lines

3. I agree your opinion that the right solution is to get dirty-page information for all memory region from kvm-kmod. But I found it’s somewhat hard to implement since kvm_log_sync() expects a MemoryRegionSection* parameter. Do you have good idea?

As to all the ram memory regions, I think they are all in the ram_list.blocks, so there is no need to create a notifier, is this correct?

maintainer
#4 lore raw
In reply to Herongguang (Stephen)
Quoted 5 lines · Herongguang (Stephen)

I have some concern:
1. For example, vhost does not know about as_id, I wonder if guests in SMM can operate disk or ether card, as in
that case vhost would not logging dirty pages correctly, without knowing as_id.

In the end memory is logged by ram_addr_t, not by address space. So if vhost_sync_dirty_bitmap is called on the right region everything works.

Guests in SMM can operate on storage devices, but storage devices cannot write to 0xA0000-0xBFFFF so that's safe.

Quoted 5 lines

2. If a memory region is disabled/enabled/disabled frequently, since disabled memory regions would be removed
from memory slots in kvm-kmod, dirty pages would be discarded in kvm-kmod and qemu when disabled, thus missing.
Is my assumption correct?

As you found, this is handled by kvm_set_phys_mem:

        if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
            kvm_physical_sync_dirty_bitmap(kml, section);
        }
Quoted 5 lines

3. I agree your opinion that the right solution is to get dirty-page information for all memory region from
kvm-kmod. But I found it’s somewhat hard to implement since kvm_log_sync() expects a MemoryRegionSection*
parameter. Do you have good idea?

You're right, memory_region_sync_dirty_bitmap handles this but it's inefficient.

So your patch is correct, but it breaks for !x86 as you probably know. We need to look at the address spaces which have a listener that implements log_sync.

Quoted 3 lines

As to all the ram memory regions, I think they are all in the ram_list.blocks, so there is no need to create
a notifier, is this correct?

Yes, makes sense.

The patch here is a bit of a hack but is efficient. It looks at each address space just once and calls log_sync just once per FlatRange.

Paolo

include/exec/memory.h +3 -4
index 3e4d416..23b086d 100644
@@ -1154,12 +1154,11 @@ MemoryRegionSection memory_region_find(MemoryRegion *mr,
                                        hwaddr addr, uint64_t size);
 
 /**
- * address_space_sync_dirty_bitmap: synchronize the dirty log for all memory
+ * memory_global_dirty_log_sync: synchronize the dirty log for all memory
  *
- * Synchronizes the dirty page log for an entire address space.
- * @as: the address space that contains the memory being synchronized
+ * Synchronizes the dirty page log for all address spaces.
  */
-void address_space_sync_dirty_bitmap(AddressSpace *as);
+void memory_global_dirty_log_sync(void);
 
 /**
  * memory_region_transaction_begin: Start a transaction.
memory.c +32 -14

69 lines not on this page · raw message

migration/ram.c +1 -1
index a6e1c63..e33ff5e 100644
@@ -626,7 +626,7 @@ static void migration_bitmap_sync(void)
     }
 
     trace_migration_bitmap_sync_start();
-    address_space_sync_dirty_bitmap(&address_space_memory);
+    memory_global_dirty_log_sync();
 
     qemu_mutex_lock(&migration_bitmap_mutex);
     rcu_read_lock();
author Reviewed-by
#5 lore raw
Quoted 9 lines · Paolo Bonzini
Quoted 5 lines · Herongguang (Stephen)

I have some concern:
1. For example, vhost does not know about as_id, I wonder if guests in SMM can operate disk or ether card, as in
that case vhost would not logging dirty pages correctly, without knowing as_id.

Quoted 5 lines

In the end memory is logged by ram_addr_t, not by address space. So if vhost_sync_dirty_bitmap is called on the right region everything works.

Guests in SMM can operate on storage devices, but storage devices cannot write to 0xA0000-0xBFFFF so that's safe.

You are right, as vhost does not have a smram address space listener. So is this a qemu's flaw that virtual devices can not write to SMRAM?

Quoted 140 lines
Quoted 6 lines

2. If a memory region is disabled/enabled/disabled frequently, since disabled memory regions would be removed
from memory slots in kvm-kmod, dirty pages would be discarded in kvm-kmod and qemu when disabled, thus missing.
Is my assumption correct?

As you found, this is handled by kvm_set_phys_mem:

         if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
             kvm_physical_sync_dirty_bitmap(kml, section);
         }
Quoted 5 lines

3. I agree your opinion that the right solution is to get dirty-page information for all memory region from
kvm-kmod. But I found it’s somewhat hard to implement since kvm_log_sync() expects a MemoryRegionSection*
parameter. Do you have good idea?

You're right, memory_region_sync_dirty_bitmap handles this but it's inefficient.

So your patch is correct, but it breaks for !x86 as you probably know. We need to look at the address spaces which have a listener that implements log_sync.

Quoted 3 lines

As to all the ram memory regions, I think they are all in the ram_list.blocks, so there is no need to create
a notifier, is this correct?

Yes, makes sense.

The patch here is a bit of a hack but is efficient. It looks at each address space just once and calls log_sync just once per FlatRange.

Paolo

include/exec/memory.h +3 -4
index 3e4d416..23b086d 100644
@@ -1154,12 +1154,11 @@ MemoryRegionSection memory_region_find(MemoryRegion *mr,
                                         hwaddr addr, uint64_t size);
 
  /**
- * address_space_sync_dirty_bitmap: synchronize the dirty log for all memory
+ * memory_global_dirty_log_sync: synchronize the dirty log for all memory
   *
- * Synchronizes the dirty page log for an entire address space.
- * @as: the address space that contains the memory being synchronized
+ * Synchronizes the dirty page log for all address spaces.
   */
-void address_space_sync_dirty_bitmap(AddressSpace *as);
+void memory_global_dirty_log_sync(void);
 
  /**
   * memory_region_transaction_begin: Start a transaction.
memory.c +32 -14
index 1a1baf5..6cac674 100644
@@ -158,15 +158,11 @@ static bool memory_listener_match(MemoryListener *listener,
 
  /* No need to ref/unref .mr, the FlatRange keeps it alive.  */
  #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...)  \
-    MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) {       \
-        .mr = (fr)->mr,                                                 \
-        .address_space = (as),                                          \
-        .offset_within_region = (fr)->offset_in_region,                 \
-        .size = (fr)->addr.size,                                        \
-        .offset_within_address_space = int128_get64((fr)->addr.start),  \
-        .readonly = (fr)->readonly,                                     \
-              }), ##_args)
-
+    do {                                                                \
+        MemoryRegionSection mrs = section_from_flat_range(fr, as);      \
+        MEMORY_LISTENER_CALL(callback, dir, &mrs, ##_args);             \
+    } while(0)
+
  struct CoalescedMemoryRange {
      AddrRange addr;
      QTAILQ_ENTRY(CoalescedMemoryRange) link;

@@ -245,6 +241,19 @@ typedef struct AddressSpaceOps AddressSpaceOps;
  #define FOR_EACH_FLAT_RANGE(var, view)          \
      for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 
+static inline MemoryRegionSection
+section_from_flat_range(FlatRange *fr, AddressSpace *as)
+{
+    return (MemoryRegionSection) {
+        .mr = fr->mr,
+        .address_space = as,
+        .offset_within_region = fr->offset_in_region,
+        .size = fr->addr.size,
+        .offset_within_address_space = int128_get64(fr->addr.start),
+        .readonly = fr->readonly,
+    };
+}
+
  static bool flatrange_equal(FlatRange *a, FlatRange *b)
  {
      return a->mr == b->mr

@@ -2124,16 +2133,25 @@ bool memory_region_present(MemoryRegion *container, hwaddr addr)
      return mr && mr != container;
  }
 
-void address_space_sync_dirty_bitmap(AddressSpace *as)
+void memory_global_dirty_log_sync(void)
  {
+    MemoryListener *listener;
+    AddressSpace *as;
      FlatView *view;
      FlatRange *fr;
 
-    view = address_space_get_flatview(as);
-    FOR_EACH_FLAT_RANGE(fr, view) {
-        MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
+    QTAILQ_FOREACH(listener, &memory_listeners, link) {
+        if (!listener->log_sync || !listener->address_space_filter) {
+            continue;
+        }
+        as = listener->address_space_filter;
+        view = address_space_get_flatview(as);
+        FOR_EACH_FLAT_RANGE(fr, view) {
+            MemoryRegionSection mrs = section_from_flat_range(fr, as);
+            listener->log_sync(listener, &mrs);
+        }
+        flatview_unref(view);
      }
-    flatview_unref(view);
  }
 
  void memory_global_dirty_log_start(void)
migration/ram.c +0 -0
index a6e1c63..e33ff5e 100644

migration/ram.c static void migration_bitmap_sync(void) · line 632

Quoted 8 lines
migration/ram.c +1 -1
@@ -626,7 +626,7 @@ static void migration_bitmap_sync(void)
      }
 
      trace_migration_bitmap_sync_start();
-    address_space_sync_dirty_bitmap(&address_space_memory);
+    memory_global_dirty_log_sync();
 
      qemu_mutex_lock(&migration_bitmap_mutex);
      rcu_read_lock();

Fine, your patch is simpler than I thought and functions right.

Reviewed-by He Rongguang <[email protected]>
Quoted 2 lines

.

maintainer
#6 lore raw
Quoted 12 lines · Herongguang (Stephen)
Quoted 9 lines · Paolo Bonzini
Quoted 5 lines · Herongguang (Stephen)

I have some concern:
1. For example, vhost does not know about as_id, I wonder if guests in SMM can operate disk or ether card, as in
that case vhost would not logging dirty pages correctly, without knowing as_id.

Quoted 7 lines
Quoted 5 lines

In the end memory is logged by ram_addr_t, not by address space. So if vhost_sync_dirty_bitmap is called on the right region everything works.

Guests in SMM can operate on storage devices, but storage devices cannot write to 0xA0000-0xBFFFF so that's safe.

You are right, as vhost does not have a smram address space listener. So is this a qemu's flaw that virtual devices can not write to SMRAM?

No, it's how it works in real hardware. However, it is a (minor) bug that vhost doesn't have the equivalent of kvm_set_phys_mem's call to kvm_physical_sync_dirty_bitmap.

Quoted 1 line

Fine, your patch is simpler than I thought and functions right.

Great!

Quoted 1 line
Reviewed-by He Rongguang <[email protected]>

Also Tested-by?

Thanks,

Paolo

Quoted 4 lines
Quoted 3 lines

.

author
#7 lore raw
Quoted 27 lines · Paolo Bonzini
Quoted 12 lines · Herongguang (Stephen)
Quoted 9 lines · Paolo Bonzini
Quoted 5 lines · Herongguang (Stephen)

I have some concern:
1. For example, vhost does not know about as_id, I wonder if guests in SMM can operate disk or ether card, as in
that case vhost would not logging dirty pages correctly, without knowing as_id.

Quoted 7 lines
Quoted 5 lines

In the end memory is logged by ram_addr_t, not by address space. So if vhost_sync_dirty_bitmap is called on the right region everything works.

Guests in SMM can operate on storage devices, but storage devices cannot write to 0xA0000-0xBFFFF so that's safe.

You are right, as vhost does not have a smram address space listener. So is this a qemu's flaw that virtual devices can not write to SMRAM?

No, it's how it works in real hardware. However, it is a (minor) bug that vhost doesn't have the equivalent of kvm_set_phys_mem's call to kvm_physical_sync_dirty_bitmap.

Quoted 7 lines
Quoted 1 line

Fine, your patch is simpler than I thought and functions right.

Great!

Quoted 1 line
Reviewed-by He Rongguang <[email protected]>

Also Tested-by?

Yes, I will test this patch in these days.

Quoted 11 lines

Thanks,

Paolo

Quoted 4 lines
Quoted 3 lines

.

.

author Tested-by
#8 lore raw
In reply to Paolo Bonzini
Quoted 154 lines · Paolo Bonzini
Quoted 5 lines · Herongguang (Stephen)

I have some concern:
1. For example, vhost does not know about as_id, I wonder if guests in SMM can operate disk or ether card, as in
that case vhost would not logging dirty pages correctly, without knowing as_id.

In the end memory is logged by ram_addr_t, not by address space. So if vhost_sync_dirty_bitmap is called on the right region everything works.

Guests in SMM can operate on storage devices, but storage devices cannot write to 0xA0000-0xBFFFF so that's safe.

Quoted 5 lines

2. If a memory region is disabled/enabled/disabled frequently, since disabled memory regions would be removed
from memory slots in kvm-kmod, dirty pages would be discarded in kvm-kmod and qemu when disabled, thus missing.
Is my assumption correct?

As you found, this is handled by kvm_set_phys_mem:

         if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
             kvm_physical_sync_dirty_bitmap(kml, section);
         }
Quoted 5 lines

3. I agree your opinion that the right solution is to get dirty-page information for all memory region from
kvm-kmod. But I found it’s somewhat hard to implement since kvm_log_sync() expects a MemoryRegionSection*
parameter. Do you have good idea?

You're right, memory_region_sync_dirty_bitmap handles this but it's inefficient.

So your patch is correct, but it breaks for !x86 as you probably know. We need to look at the address spaces which have a listener that implements log_sync.

Quoted 3 lines

As to all the ram memory regions, I think they are all in the ram_list.blocks, so there is no need to create
a notifier, is this correct?

Yes, makes sense.

The patch here is a bit of a hack but is efficient. It looks at each address space just once and calls log_sync just once per FlatRange.

Paolo

include/exec/memory.h +3 -4
index 3e4d416..23b086d 100644
@@ -1154,12 +1154,11 @@ MemoryRegionSection memory_region_find(MemoryRegion *mr,
                                         hwaddr addr, uint64_t size);
 
  /**
- * address_space_sync_dirty_bitmap: synchronize the dirty log for all memory
+ * memory_global_dirty_log_sync: synchronize the dirty log for all memory
   *
- * Synchronizes the dirty page log for an entire address space.
- * @as: the address space that contains the memory being synchronized
+ * Synchronizes the dirty page log for all address spaces.
   */
-void address_space_sync_dirty_bitmap(AddressSpace *as);
+void memory_global_dirty_log_sync(void);
 
  /**
   * memory_region_transaction_begin: Start a transaction.
memory.c +32 -14
index 1a1baf5..6cac674 100644
@@ -158,15 +158,11 @@ static bool memory_listener_match(MemoryListener *listener,
 
  /* No need to ref/unref .mr, the FlatRange keeps it alive.  */
  #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...)  \
-    MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) {       \
-        .mr = (fr)->mr,                                                 \
-        .address_space = (as),                                          \
-        .offset_within_region = (fr)->offset_in_region,                 \
-        .size = (fr)->addr.size,                                        \
-        .offset_within_address_space = int128_get64((fr)->addr.start),  \
-        .readonly = (fr)->readonly,                                     \
-              }), ##_args)
-
+    do {                                                                \
+        MemoryRegionSection mrs = section_from_flat_range(fr, as);      \
+        MEMORY_LISTENER_CALL(callback, dir, &mrs, ##_args);             \
+    } while(0)
+
  struct CoalescedMemoryRange {
      AddrRange addr;
      QTAILQ_ENTRY(CoalescedMemoryRange) link;

@@ -245,6 +241,19 @@ typedef struct AddressSpaceOps AddressSpaceOps;
  #define FOR_EACH_FLAT_RANGE(var, view)          \
      for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 
+static inline MemoryRegionSection
+section_from_flat_range(FlatRange *fr, AddressSpace *as)
+{
+    return (MemoryRegionSection) {
+        .mr = fr->mr,
+        .address_space = as,
+        .offset_within_region = fr->offset_in_region,
+        .size = fr->addr.size,
+        .offset_within_address_space = int128_get64(fr->addr.start),
+        .readonly = fr->readonly,
+    };
+}
+
  static bool flatrange_equal(FlatRange *a, FlatRange *b)
  {
      return a->mr == b->mr

@@ -2124,16 +2133,25 @@ bool memory_region_present(MemoryRegion *container, hwaddr addr)
      return mr && mr != container;
  }
 
-void address_space_sync_dirty_bitmap(AddressSpace *as)
+void memory_global_dirty_log_sync(void)
  {
+    MemoryListener *listener;
+    AddressSpace *as;
      FlatView *view;
      FlatRange *fr;
 
-    view = address_space_get_flatview(as);
-    FOR_EACH_FLAT_RANGE(fr, view) {
-        MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
+    QTAILQ_FOREACH(listener, &memory_listeners, link) {
+        if (!listener->log_sync || !listener->address_space_filter) {
+            continue;
+        }
+        as = listener->address_space_filter;
+        view = address_space_get_flatview(as);
+        FOR_EACH_FLAT_RANGE(fr, view) {
+            MemoryRegionSection mrs = section_from_flat_range(fr, as);
+            listener->log_sync(listener, &mrs);
+        }
+        flatview_unref(view);
      }
-    flatview_unref(view);
  }
 
  void memory_global_dirty_log_start(void)
migration/ram.c +0 -0
index a6e1c63..e33ff5e 100644

migration/ram.c static void migration_bitmap_sync(void) · line 632

Quoted 8 lines
migration/ram.c +1 -1
@@ -626,7 +626,7 @@ static void migration_bitmap_sync(void)
      }
 
      trace_migration_bitmap_sync_start();
-    address_space_sync_dirty_bitmap(&address_space_memory);
+    memory_global_dirty_log_sync();
 
      qemu_mutex_lock(&migration_bitmap_mutex);
      rcu_read_lock();
Tested-by He Rongguang <[email protected]>
Quoted 2 lines

.

author
#10 lore raw

[RFC/PATCH 2] kvm: x86: handle KVM_SET_VCPU_EVENTS/KVM_VCPUEVENT_VALID_SMM properly

In reply to Herongguang (Stephen)

After making memory consistent between source and destination (https://lists.gnu.org/archive/html/qemu-devel/2016-09/msg03069.html), there can still reproduce instruction emulation failure in destination side if migration when VM’s in grub stage:

[2016-09-15 06:29:24] monitor_qapi_event_emit:478 {"timestamp": {"seconds": 1473892164, "microseconds": 99652}, "event": "RESUME"} KVM internal error. Suberror: 1
emulation failure
EAX=000000b5 EBX=00008fc6 ECX=00005678 EDX=00000000
ESI=000f254b EDI=00000000 EBP=0000f958 ESP=000ee958
EIP=00008000 EFL=00010002 [-------] CPL=0 II=0 A20=1 SMM=0 HLT=0
ES =0000 00000000 ffffffff 00809300
CS =a000 000a0000 ffffffff 00809300
SS =0000 00000000 ffffffff 00809300
DS =0000 00000000 ffffffff 00809300
FS =0000 00000000 ffffffff 00809300
GS =0000 00000000 ffffffff 00809300
LDT=0000 00000000 0000ffff 00008200
TR =0000 00000000 0000ffff 00008b00
GDT= 000f71a0 00000037
IDT= 00000000 00000000
CR0=00000010 CR2=00000000 CR3=00000000 CR4=00000000
DR0=0000000000000000 DR1=0000000000000000 DR2=0000000000000000 DR3=0000000000000000
DR6=00000000ffff0ff0 DR7=0000000000000400
EFER=0000000000000000
Code=ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff <ff> ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff [2016-09-15 06:29:24] virtio_set_status:712 virtio-blk device status is 7 that means DRIVER OK
[2016-09-15 06:29:24] monitor_qapi_event_emit:478 {"timestamp": {"seconds": 1473892164, "microseconds": 106738}, "event": "STOP"}

Note CS+EIP=0xA8000, which is within SMRAM, however SMM=0 is confusing! So I found that in kvm_put_vcpu_events, events.flags(KVM_VCPUEVENT_VALID_SMM) is overwritten by setting events.flags to 0, which is obvious mistaken.

So it comes to patch 3. After patch 3, however it results kvm-kmod crash:

[69328.761479] kvm [51353]: vcpu7, guest rIP: 0xfd31c unhandled wrmsr: 0xd1 data 0
[69337.406083] BUG: unable to handle kernel NULL pointer dereference at           (null)
[69337.414193] IP: [<ffffffffa056a59d>] gfn_to_rmap+0xcd/0xe0 [kvm]
[69337.420357] PGD 0
[69337.422514] Oops: 0000 [#1] SMP
[69337.425783] Modules linked in: rpcsec_gss_krb5 nfsv4 dns_resolver nfs fscache nfsd auth_rpcgss nfs_acl lockd grace sunrpc xt_CHECKSUM iptable_mangle ipt_MASQUERADE nf_nat_masquerade_ipv4 iptable_nat xt_conntrack ipt_REJECT nf_reject_ipv4 bridge stp llc ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter openvswitch nf_conntrack_ipv6 nf_nat_ipv6 nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_defrag_ipv6 nf_nat nf_conntrack libcrc32c ipmi_devintf ipmi_si ipmi_msghandler intel_rapl sb_edac edac_core x86_pkg_temp_thermal intel_powerclamp coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel aesni_intel lrw kvm_intel gf128mul iTCO_wdt kvm glue_helper iTCO_vendor_support ablk_helper cryptd pcspkr sg mei_me mei ioatdma shpchp lpc_ich wmi mfd_core i2c_i801 vhost_net tun vhos
t macvtap macvlan ip_tables ext4 jbd2 mbcache sr_mod sd_mod cdrom isci libsas igb ahci libahci scsi_transport_sas crc32c_intel libata ptp pps_core serio_raw i2c_algo_bit megaraid_sas i2c_co!
re dca d
m_mod vfio_pci irqbypass vfio_virqfd vfio_iommu_type1 vfio
[69337.524020] CPU: 3 PID: 51375 Comm: CPU 0/KVM Not tainted 4.7.0 #1
[69337.530315] Hardware name: Huawei Technologies Co., Ltd. Tecal RH2288 V2-8S/BC11SRSB1, BIOS RMIBV365 09/06/2013
[69337.540644] task: ffff88104bcc8000 ti: ffff881055650000 task.ti: ffff881055650000
[69337.548382] RIP: 0010:[<ffffffffa056a59d>]  [<ffffffffa056a59d>] gfn_to_rmap+0xcd/0xe0 [kvm]
[69337.557105] RSP: 0018:ffff881055653a88  EFLAGS: 00010202
[69337.562541] RAX: ffffc9000b3cf428 RBX: ffff88085624a538 RCX: 000000000000000c
[69337.569788] RDX: ffff88083a6e2bd0 RSI: 00000000000000a7 RDI: 00000000000000a0
[69337.577036] RBP: ffff881055653a88 R08: 000000000000000c R09: ffffc9000b3cf008
[69337.584283] R10: ffffc9000b3cf000 R11: 0000000000000037 R12: ffff8810569d0000
[69337.591530] R13: 0000000000000000 R14: 00000000000000a7 R15: ffff8808d624a538
[69337.598777] FS:  00007faeb900d700(0000) GS:ffff88085e0c0000(0000) knlGS:0000000000000000
[69337.607116] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[69337.612978] CR2: 0000000000000000 CR3: 000000104c203000 CR4: 00000000000426e0
[69337.620225] Stack:
[69337.622369]  ffff881055653af0 ffffffffa056b28d ffffffff00000000 0000000000000001
[69337.630059]  0000000000000001 00007fadb50a7000 0100000000000000 ffff88083a6e2bd0
[69337.637751]  ffff8810569d0000 0000000000000001 000ffffffffff000 ffff880000000000
[69337.645442] Call Trace:
[69337.648029]  [<ffffffffa056b28d>] mmu_set_spte+0x19d/0x280 [kvm]
[69337.654166]  [<ffffffffa056f56b>] __direct_map.part.122+0x19b/0x210 [kvm]
[69337.661078]  [<ffffffffa056fa29>] tdp_page_fault+0x239/0x260 [kvm]
[69337.667385]  [<ffffffffa0568bf0>] kvm_mmu_page_fault+0x60/0xf0 [kvm]
[69337.673856]  [<ffffffffa03f8656>] handle_ept_violation+0x96/0x190 [kvm_intel]
[69337.681106]  [<ffffffffa03ff740>] vmx_handle_exit+0x1f0/0xc30 [kvm_intel]
[69337.688019]  [<ffffffffa055d9ce>] vcpu_enter_guest+0x90e/0x1190 [kvm]
[69337.694585]  [<ffffffffa056444d>] kvm_arch_vcpu_ioctl_run+0xcd/0x3f0 [kvm]
[69337.701581]  [<ffffffffa0549fb5>] kvm_vcpu_ioctl+0x295/0x610 [kvm]
[69337.707880]  [<ffffffff8110b6cc>] ? do_futex+0x11c/0x530
[69337.713314]  [<ffffffff8122e436>] do_vfs_ioctl+0xa6/0x5c0
[69337.718833]  [<ffffffff81003276>] ? do_audit_syscall_entry+0x66/0x70
[69337.725303]  [<ffffffff810037cf>] ? syscall_trace_enter_phase1+0x11f/0x140
[69337.732291]  [<ffffffff8122e9c9>] SyS_ioctl+0x79/0x90
[69337.737470]  [<ffffffff81003b12>] do_syscall_64+0x62/0x110
[69337.743085]  [<ffffffff816c88a1>] entry_SYSCALL64_slow_path+0x25/0x25
[69337.749641] Code: 8b 38 0f b6 52 28 5d 83 e2 0f 83 ea 01 8d 0c d2 48 63 d2 48 8b 44 d0 18 48 d3 ee 48 d3 ef 48 29 fe 48 8d 04 f0 c3 8d 48 01 eb 80 <48> 8b 3c 25 00 00 00 00 31 c0 eb cb 0f 1f 80 00 00 00 00 66 66
[69337.769707] RIP  [<ffffffffa056a59d>] gfn_to_rmap+0xcd/0xe0 [kvm]
[69337.775938]  RSP <ffff881055653a88>
[69337.779552] CR2: 0000000000000000
[69337.783467] ---[ end trace 5350f10b8de91e83 ]---
[69337.843957] Kernel panic - not syncing: Fatal exception
[69337.849430] Kernel Offset: disabled

I found that when kernel crashes, in tdp_page_fault ->__direct_map ->mmu_set_spte ->rmap_add->gfn_to_rmap-> __gfn_to_memslot &__gfn_to_rmap, in __gfn_to_memslot returned NULL, and __gfn_to_rmap uses this NULL pointer (slot) without checking, which is bad.

After some investigation, I found when kernel crashes, gfn is 0xA7, in SMRAM region again. And at this time, kvm_memslots_for_spte_role is false, however actually (vcpu->arch.hflags & HF_SMM_MASK) is true!

So I think there is some lacking in kvm-kmod’s kvm_vcpu_ioctl_x86_set_vcpu_events that handles KVM_VCPUEVENT_VALID_SMM. I tried following patch, it seems works fine.

Do you think this patch is appropriate or not enough? Thanks.

arch/x86/kvm/x86.c +3 -1
index 19f9f9e..f39e839 100644
@@ -3013,8 +3013,10 @@ static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
                vcpu->arch.apic->sipi_vector = events->sipi_vector;
 
        if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
-               if (events->smi.smm)
+               if (events->smi.smm) {
                        vcpu->arch.hflags |= HF_SMM_MASK;
+                       kvm_mmu_reset_context(vcpu);
+               }
                else
                        vcpu->arch.hflags &= ~HF_SMM_MASK;
                vcpu->arch.smi_pending = events->smi.pending;
maintainer
#11 lore raw

Re: [RFC/PATCH 2] kvm: x86: handle KVM_SET_VCPU_EVENTS/KVM_VCPUEVENT_VALID_SMM properly

Quoted 5 lines · Herongguang (Stephen)

After making memory consistent between source and destination
(https://lists.gnu.org/archive/html/qemu-devel/2016-09/msg03069.html),
there can
still reproduce instruction emulation failure in destination side if migration when VM’s in grub stage:

Hi! Did you follow up on that patch, by the way?

Quoted 6 lines

So I think there is some lacking in kvm-kmod’s kvm_vcpu_ioctl_x86_set_vcpu_events that handles KVM_VCPUEVENT_VALID_SMM. I tried following patch,
it seems works fine.

Do you think this patch is appropriate or not enough? Thanks.

Yes. I would just call kvm_mmu_reset_context unconditionally at the end of kvm_vcpu_iocyl_x86_set_x86_vcpu_events. Please send this patch as non-RFC.

Patch 3 is also okay, please send it separately.

Thanks,

Paolo

Quoted 19 lines
arch/x86/kvm/x86.c +3 -1
index 19f9f9e..f39e839 100644
@@ -3013,8 +3013,10 @@ static int
 kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
                vcpu->arch.apic->sipi_vector = events->sipi_vector;
 
        if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
-               if (events->smi.smm)
+               if (events->smi.smm) {
                        vcpu->arch.hflags |= HF_SMM_MASK;
+                       kvm_mmu_reset_context(vcpu);
+               }
                else
                        vcpu->arch.hflags &= ~HF_SMM_MASK;

                vcpu->arch.smi_pending = events->smi.pending;
author
#12 lore raw

Re: [RFC/PATCH 2] kvm: x86: handle KVM_SET_VCPU_EVENTS/KVM_VCPUEVENT_VALID_SMM properly

Quoted 10 lines · Paolo Bonzini
Quoted 5 lines · Herongguang (Stephen)

After making memory consistent between source and destination
(https://lists.gnu.org/archive/html/qemu-devel/2016-09/msg03069.html),
there can
still reproduce instruction emulation failure in destination side if migration when VM’s in grub stage:

Hi! Did you follow up on that patch, by the way?

Yes, I have some concern, see that post.

Quoted 6 lines
Quoted 6 lines

So I think there is some lacking in kvm-kmod’s kvm_vcpu_ioctl_x86_set_vcpu_events that handles KVM_VCPUEVENT_VALID_SMM. I tried following patch,
it seems works fine.

Quoted 7 lines
Quoted 1 line

Do you think this patch is appropriate or not enough? Thanks.

Yes. I would just call kvm_mmu_reset_context unconditionally at the end of kvm_vcpu_iocyl_x86_set_x86_vcpu_events. Please send this patch as non-RFC.

Patch 3 is also okay, please send it separately.

Ok, I will test and post it tomorrow, thanks!

Quoted 27 lines

Thanks,

Paolo

Quoted 16 lines
arch/x86/kvm/x86.c +3 -1
index 19f9f9e..f39e839 100644
@@ -3013,8 +3013,10 @@ static int
 kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
                 vcpu->arch.apic->sipi_vector = events->sipi_vector;
 
         if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
-               if (events->smi.smm)
+               if (events->smi.smm) {
                         vcpu->arch.hflags |= HF_SMM_MASK;
+                       kvm_mmu_reset_context(vcpu);
+               }
                 else
                         vcpu->arch.hflags &= ~HF_SMM_MASK;
Quoted 3 lines
                 vcpu->arch.smi_pending = events->smi.pending;

.

author
#13 lore raw

[Qemu-devel] [RFC/PATCH 3] kvm: fix events.flags (KVM_VCPUEVENT_VALID_SMM) overwritten by 0

In reply to Herongguang (Stephen)

Fix events.flags (KVM_VCPUEVENT_VALID_SMM) overwritten by 0.

Signed-off-by He Rongguang <[email protected]>

---
Note without patch 2, this would result in kvm-kmod crash, as described in patch 2

---
 target-i386/kvm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target-i386/kvm.c +1 -1
index d1a25c5..7db33d2 100644
@@ -2456,6 +2456,7 @@ static int kvm_put_vcpu_events(X86CPU *cpu, int level)
 
     events.sipi_vector = env->sipi_vector;
 
+    events.flags = 0;
     if (has_msr_smbase) {
         events.smi.smm = !!(env->hflags & HF_SMM_MASK);
         events.smi.smm_inside_nmi = !!(env->hflags2 & HF2_SMM_INSIDE_NMI_MASK);

@@ -2474,7 +2475,6 @@ static int kvm_put_vcpu_events(X86CPU *cpu, int level)
         events.flags |= KVM_VCPUEVENT_VALID_SMM;
     }
 
-    events.flags = 0;
     if (level >= KVM_PUT_RESET_STATE) {
         events.flags |=
             KVM_VCPUEVENT_VALID_NMI_PENDING | KVM_VCPUEVENT_VALID_SIPI_VECTOR;

--
1.8.3.4