[Xen-devel] [PATCH 2 of 4] x86/mm: Allow to not sleep on mem event ring

[Andres Lagar-Cavilla <andres@xxxxxxxxxxxxxxxx>]

 xen/arch/x86/mm/mem_event.c     |   5 +++--
 xen/include/asm-x86/mem_event.h |  30 +++++++++++++++++++++++++-----
 2 files changed, 28 insertions(+), 7 deletions(-)


Under extreme congestion conditions, generating a mem event may put the vcpu to
sleep on a wait queue if the ring is full. This is generally desirable, although
fairly convoluted to work with, since sleeping on a wait queue requires a
non-atomic context (i.e. no locks held).

Introduce an allow_sleep flag to make this optional. The API remains such that
all current callers set allow_sleep to true and thus will sleep if necessary.

The end-use is for cases in which loss of guest mem events is tolerable. One
such consumer to be added later is the unsharing code under ENOMEM conditions.

Signed-off-by: Andres Lagar-Cavilla <andres@xxxxxxxxxxxxxxxx>
Signed-off-by: Adin Scannell <adin@xxxxxxxxxxx>

diff -r 11fd4e0a1e1a -r 09746decbd28 xen/arch/x86/mm/mem_event.c
--- a/xen/arch/x86/mm/mem_event.c
+++ b/xen/arch/x86/mm/mem_event.c
@@ -441,9 +441,10 @@ bool_t mem_event_check_ring(struct mem_e
  *               0: a spot has been reserved
  *
  */
-int mem_event_claim_slot(struct domain *d, struct mem_event_domain *med)
+int __mem_event_claim_slot(struct domain *d, struct mem_event_domain *med,
+                            bool_t allow_sleep)
 {
-    if ( current->domain == d )
+    if ( (current->domain == d) && allow_sleep )
         return mem_event_wait_slot(med);
     else
         return mem_event_grab_slot(med, 1);
diff -r 11fd4e0a1e1a -r 09746decbd28 xen/include/asm-x86/mem_event.h
--- a/xen/include/asm-x86/mem_event.h
+++ b/xen/include/asm-x86/mem_event.h
@@ -28,11 +28,31 @@
 bool_t mem_event_check_ring(struct mem_event_domain *med);
 
 /* Returns 0 on success, -ENOSYS if there is no ring, -EBUSY if there is no
- * available space. For success or -EBUSY, the vCPU may be left blocked
- * temporarily to ensure that the ring does not lose future events.  In
- * general, you must follow a claim_slot() call with either put_request() or
- * cancel_slot(), both of which are guaranteed to succeed. */
-int mem_event_claim_slot(struct domain *d, struct mem_event_domain *med);
+ * available space and the caller is a foreign domain. If the guest itself
+ * is the caller, -EBUSY is avoided by sleeping on a wait queue to ensure
+ * that the ring does not lose future events. 
+ *
+ * However, the allow_sleep flag can be set to false in cases in which it is ok
+ * to lose future events, and thus -EBUSY can be returned to guest vcpus
+ * (handle with care!). 
+ *
+ * In general, you must follow a claim_slot() call with either put_request() or
+ * cancel_slot(), both of which are guaranteed to
+ * succeed. 
+ */
+int __mem_event_claim_slot(struct domain *d, struct mem_event_domain *med,
+                            bool_t allow_sleep);
+static inline int mem_event_claim_slot(struct domain *d, 
+                                        struct mem_event_domain *med)
+{
+    return __mem_event_claim_slot(d, med, 1);
+}
+
+static inline int mem_event_claim_slot_nosleep(struct domain *d,
+                                        struct mem_event_domain *med)
+{
+    return __mem_event_claim_slot(d, med, 0);
+}
 
 void mem_event_cancel_slot(struct domain *d, struct mem_event_domain *med);
 

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel