[v2] xen: Add RING_COPY_RESPONSE()

Commit Message

Marek Marczykowski-Górecki Jan. 28, 2021, 3:50 p.m. UTC

Using RING_GET_RESPONSE() on a shared ring is easy to use incorrectly
(i.e., by not considering that the other end may alter the data in the
shared ring while it is being inspected). Safe usage of a response
generally requires taking a local copy.

Provide a RING_COPY_RESPONSE() macro to use instead of
RING_GET_RESPONSE() and an open-coded memcpy().  This takes care of
ensuring that the copy is done correctly regardless of any possible
compiler optimizations.

Use a volatile source to prevent the compiler from reordering or
omitting the copy.

This generalizes similar RING_COPY_REQUEST() macro added in 3f20b8def0.

Signed-off-by: Marek Marczykowski-Górecki <[e-mail]>
---
Changes in v2:
 - action -> type
 - drop leading underscores from macro parameters (keep it in
   RING_GET_RESPONSE for consistency with RING_GET_REQUEST)
---
 xen/include/public/io/ring.h | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)

Comments

Jürgen Groß Feb. 4, 2021, 11:10 a.m. UTC | #1

On 28.01.21 16:50, Marek Marczykowski-Górecki wrote:
> Using RING_GET_RESPONSE() on a shared ring is easy to use incorrectly
> (i.e., by not considering that the other end may alter the data in the
> shared ring while it is being inspected). Safe usage of a response
> generally requires taking a local copy.
> 
> Provide a RING_COPY_RESPONSE() macro to use instead of
> RING_GET_RESPONSE() and an open-coded memcpy().  This takes care of
> ensuring that the copy is done correctly regardless of any possible
> compiler optimizations.
> 
> Use a volatile source to prevent the compiler from reordering or
> omitting the copy.
> 
> This generalizes similar RING_COPY_REQUEST() macro added in 3f20b8def0.
> 
> Signed-off-by: Marek Marczykowski-Górecki <[e-mail]>

Reviewed-by: Juergen Gross <[e-mail]>


Juergen

Patch

diff --git a/xen/include/public/io/ring.h b/xen/include/public/io/ring.h
index d68615ae2f67..d2129ed183ab 100644
--- a/xen/include/public/io/ring.h
+++ b/xen/include/public/io/ring.h
@@ -231,22 +231,25 @@  typedef struct __name##_back_ring __name##_back_ring_t
 #define RING_GET_REQUEST(_r, _idx)                                      \
     (&((_r)->sring->ring[((_idx) & (RING_SIZE(_r) - 1))].req))
 
+#define RING_GET_RESPONSE(_r, _idx)                                     \
+    (&((_r)->sring->ring[((_idx) & (RING_SIZE(_r) - 1))].rsp))
+
 /*
- * Get a local copy of a request.
+ * Get a local copy of a request/response.
  *
- * Use this in preference to RING_GET_REQUEST() so all processing is
+ * Use this in preference to RING_GET_{REQUEST,RESPONSE}() so all processing is
  * done on a local copy that cannot be modified by the other end.
  *
  * Note that https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58145 may cause this
- * to be ineffective where _req is a struct which consists of only bitfields.
+ * to be ineffective where dest is a struct which consists of only bitfields.
  */
-#define RING_COPY_REQUEST(_r, _idx, _req) do {				\
-	/* Use volatile to force the copy into _req. */			\
-	*(_req) = *(volatile typeof(_req))RING_GET_REQUEST(_r, _idx);	\
+#define RING_COPY_(type, r, idx, dest) do {				\
+	/* Use volatile to force the copy into dest. */			\
+	*(dest) = *(volatile typeof(dest))RING_GET_##type(r, idx);	\
 } while (0)
 
-#define RING_GET_RESPONSE(_r, _idx)                                     \
-    (&((_r)->sring->ring[((_idx) & (RING_SIZE(_r) - 1))].rsp))
+#define RING_COPY_REQUEST(r, idx, req)  RING_COPY_(REQUEST, r, idx, req)
+#define RING_COPY_RESPONSE(r, idx, rsp) RING_COPY_(RESPONSE, r, idx, rsp)
 
 /* Loop termination condition: Would the specified index overflow the ring? */
 #define RING_REQUEST_CONS_OVERFLOW(_r, _cons)                           \