[v2] memory: Sanity checks memory transaction when releasing BQL

[PATCH v2 9/9] memory: Delay the transaction pop() until commit completed

Posted by Peter Xu 4 years, 6 months ago

This should be functionally the same as before, but this allows the
memory_region_transaction_depth to be non-zero during commit, which can help us
to do sanity check on misuses.

Since at it, fix an indentation issue on the bracket.

Signed-off-by: Peter Xu <peterx@redhat.com>
---
 softmmu/memory.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/softmmu/memory.c b/softmmu/memory.c
index 08327c22e2..cf7943c02c 100644
--- a/softmmu/memory.c
+++ b/softmmu/memory.c
@@ -1114,8 +1114,7 @@ void memory_region_transaction_commit(void)
     assert(memory_region_transaction_depth);
     assert(qemu_mutex_iothread_locked());
 
-    memory_region_transaction_pop();
-    if (!memory_region_transaction_depth) {
+    if (memory_region_transaction_depth == 1) {
         if (memory_region_update_pending) {
             flatviews_reset();
 
@@ -1134,7 +1133,14 @@ void memory_region_transaction_commit(void)
             }
             ioeventfd_update_pending = false;
         }
-   }
+    }
+
+    /*
+     * Pop the depth at last, so that memory_region_transaction_depth will
+     * still be non-zero during committing.  This can help us to do some sanity
+     * check within the process of committing.
+     */
+    memory_region_transaction_pop();
 }
 
 static void memory_region_destructor_none(MemoryRegion *mr)
-- 
2.31.1

Re: [PATCH v2 9/9] memory: Delay the transaction pop() until commit completed

Posted by David Hildenbrand 4 years, 6 months ago

On 23.07.21 21:34, Peter Xu wrote:
> This should be functionally the same as before, but this allows the
> memory_region_transaction_depth to be non-zero during commit, which can help us
> to do sanity check on misuses.
> 
> Since at it, fix an indentation issue on the bracket.
> 
> Signed-off-by: Peter Xu <peterx@redhat.com>
> ---
>   softmmu/memory.c | 12 +++++++++---
>   1 file changed, 9 insertions(+), 3 deletions(-)
> 
> diff --git a/softmmu/memory.c b/softmmu/memory.c
> index 08327c22e2..cf7943c02c 100644
> --- a/softmmu/memory.c
> +++ b/softmmu/memory.c
> @@ -1114,8 +1114,7 @@ void memory_region_transaction_commit(void)
>       assert(memory_region_transaction_depth);
>       assert(qemu_mutex_iothread_locked());
>   
> -    memory_region_transaction_pop();
> -    if (!memory_region_transaction_depth) {
> +    if (memory_region_transaction_depth == 1) {
>           if (memory_region_update_pending) {
>               flatviews_reset();
>   
> @@ -1134,7 +1133,14 @@ void memory_region_transaction_commit(void)
>               }
>               ioeventfd_update_pending = false;
>           }
> -   }
> +    }
> +
> +    /*
> +     * Pop the depth at last, so that memory_region_transaction_depth will
> +     * still be non-zero during committing.  This can help us to do some sanity
> +     * check within the process of committing.
> +     */
> +    memory_region_transaction_pop();
>   }
>   
>   static void memory_region_destructor_none(MemoryRegion *mr)
> 

Sounds sane to me!

-- 
Thanks,

David / dhildenb