> #0 dapl_llist_remove_entry (head=0x636960, entry=0x7ffff0004bf8) at
> dapl/common/dapl_llist.c:272
> #1 0x00007ffff799fb09 in dapl_sp_remove_cr (sp_ptr=0x6368c0,
> cr_ptr=0x7ffff0004be0) at dapl/common/dapl_sp_util.c:229
> #2 0x00007ffff7998148 in dapli_connection_request (ib_cm_handle=<value
> optimized out>, sp_ptr=0x6368c0, prd_ptr=<value optimized out>,
> private_data_size=<value optimized out>, evd_ptr=0x633fb0) at
> dapl/common/dapl_cr_callback.c:424
>
> ...
>
> Now, it seems that some time back, a new release of dapl (dapl-2.0.34-
> 1.src.rpm) was
> introduced in OFED-1.5.4. So, I am just wondering if this is a
> regression in the new
> release of dapl?
> Or if anyone is aware of this issue and what could possibly lead to
> this
> dapltest-server segfault then, it would be helpful if someone can shed
> some light.
>
You should have seen a message like "WARNING: overflow event on EVD".
It appears that the default dapltest server allocates too small of a CR EVD for
many client test configurations. When it hits the overflow queue case, the CR
callback incorrectly frees the CR before it is removed from SP list. In your
case, I am guessing that another CR came in on another thread and this memory
was reallocated with flink ptr reinitialized.
Please try the following patches.
---------
Common: CR EVD overflow causes segfault.
The CR is freed up incorrectly before unlinking with SP.
Signed-off-by: Arlin Davis <arlin.r.da...@intel.com>
diff --git a/dapl/common/dapl_cr_callback.c b/dapl/common/dapl_cr_callback.c
index 3997b38..c58444b 100644
--- a/dapl/common/dapl_cr_callback.c
+++ b/dapl/common/dapl_cr_callback.c
@@ -414,7 +414,6 @@ dapli_connection_request(IN dp_ib_cm_handle_t ib_cm_handle,
(DAT_CR_HANDLE) cr_ptr);
if (dat_status != DAT_SUCCESS) {
- dapls_cr_free(cr_ptr);
(void)dapls_ib_reject_connection(ib_cm_handle,
DAT_CONNECTION_EVENT_BROKEN,
0, NULL);
@@ -423,6 +422,7 @@ dapli_connection_request(IN dp_ib_cm_handle_t ib_cm_handle,
dapl_os_lock(&sp_ptr->header.lock);
dapl_sp_remove_cr(sp_ptr, cr_ptr);
dapl_os_unlock(&sp_ptr->header.lock);
+ dapls_cr_free(cr_ptr);
return DAT_INSUFFICIENT_RESOURCES;
}
----------
dapltest: server CR EVD is too small for multi-client configurations.
Increase default size from 8 to 32.
Signed-off-by: Arlin Davis <arlin.r.da...@intel.com>
diff --git a/test/dapltest/test/dapl_server.c b/test/dapltest/test/dapl_server.c
index 443425c..92e0d21 100644
--- a/test/dapltest/test/dapl_server.c
+++ b/test/dapltest/test/dapl_server.c
@@ -34,7 +34,7 @@
#undef DFLT_QLEN
#endif
-#define DFLT_QLEN 8 /* default event queue length */
+#define DFLT_QLEN 32 /* default event queue length */
int send_control_data(DT_Tdep_Print_Head * phead,
unsigned char *buffp,
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
