James,

This small patch fixes another issue with scale-out MPI testing. Please
review this patch since you understand the history around NO_EP_STATE
better then me. Apparently there is some NO_EP_STATE hack in the common
code that is provided to protect against race conditions but I not clear
on its usage model. This unknown state is causing the
dat_ep_disconnect() to incorrectly fail with invalid state instead of
processing in the proper DAT_EP_STATE_COMPLETION_PENDING state. 

Thanks,

-arlin 

Signed-off by: Arlin Davis <[EMAIL PROTECTED]>

Index: dapl/dapl/common/dapl_cr_callback.c
===================================================================
--- dapl/dapl/common/dapl_cr_callback.c (revision 5642)
+++ dapl/dapl/common/dapl_cr_callback.c (working copy)
@@ -532,18 +532,10 @@ dapli_get_sp_ep (
     {
        /* Remove the CR from the queue */
        dapl_sp_remove_cr (sp_ptr, cr_ptr);
-       /*
-        * Last event, time to clean up and dispose of the resource.
-        * Put the EP in a state that it can't get hurt, avoiding
-        * race conditions between now and when it gets disconnected,
-        * i.e. an app invoking ep_disconnect before the callback
-        * code above finishes.
-        */
-#define NO_EP_STATE    111             /* larger than any ep state */
+
        if (ep_ptr != NULL)
        {
            ep_ptr->cr_ptr = NULL;
-           ep_ptr->param.ep_state = NO_EP_STATE;
        }
 
        /*


Attachment: udapl_dconn.patch
Description: udapl_dconn.patch

_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to