memory: fix race between TCG and accesses to dirty bitmap

There is a race between TCG and accesses to the dirty log:

      vCPU thread                  reader thread
      -----------------------      -----------------------
      TLB check -> slow path
        notdirty_mem_write
          write to RAM
          set dirty flag
                                   clear dirty flag
      TLB check -> fast path
                                   read memory
        write to RAM

Fortunately, in order to fix it, no change is required to the
vCPU thread.  However, the reader thread must delay the read after
the vCPU thread has finished the write.  This can be approximated
conservatively by run_on_cpu, which waits for the end of the current
translation block.

A similar technique is used by KVM, which has to do a synchronous TLB
flush after doing a test-and-clear of the dirty-page flags.

Reported-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

1 files changed, 12 insertions, 0 deletions

diff --git a/include/exec/memory.h b/include/exec/memory.h
index d99eb25d2e..fddc2ff48a 100644
--- a/include/exec/memory.h
+++ b/include/exec/memory.h
@@ -425,6 +425,7 @@ struct MemoryListener {
     void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section);
     void (*log_global_start)(MemoryListener *listener);
     void (*log_global_stop)(MemoryListener *listener);
+    void (*log_global_after_sync)(MemoryListener *listener);
     void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section,
                         bool match_data, uint64_t data, EventNotifier *e);
     void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section,
@@ -1688,6 +1689,17 @@ MemoryRegionSection memory_region_find(MemoryRegion *mr,
 void memory_global_dirty_log_sync(void);
 
 /**
+ * memory_global_dirty_log_sync: synchronize the dirty log for all memory
+ *
+ * Synchronizes the vCPUs with a thread that is reading the dirty bitmap.
+ * This function must be called after the dirty log bitmap is cleared, and
+ * before dirty guest memory pages are read.  If you are using
+ * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes
+ * care of doing this.
+ */
+void memory_global_after_dirty_log_sync(void);
+
+/**
  * memory_region_transaction_begin: Start a transaction.
  *
  * During a transaction, changes will be accumulated and made visible