[ofw] [Patch 14/62] Reference implementation of NDv2

[Fab Tillier]

Distinguish between CQ catastrophic error and CQ overflow.  The hardware knows which of the two happened, propagate the information via the CQ notification, rather than losing it in a trace message.

Signed-off-by: Fab Tillier <ftillier at microsoft.com>

diff -dwup3 -x *svn* -r c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\hw\mlx4\kernel\bus\ib\cq.c .\hw\mlx4\kernel\bus\ib\cq.c
--- c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\hw\mlx4\kernel\bus\ib\cq.c	Thu Aug 02 16:47:19 2012
+++ .\hw\mlx4\kernel\bus\ib\cq.c	Sat Jun 23 14:15:09 2012
@@ -34,7 +34,7 @@
 #include "mlx4_ib.h"
 #include "cq.h"
 #include "qp.h"
-#include "user.h"
+#include "mx_abi.h"
 
 #if defined(EVENT_TRACING)
 #ifdef offsetof
@@ -58,7 +58,13 @@ static void mlx4_ib_cq_event(struct mlx4
 	ib_event_rec_t event;
 	struct ib_cq *ibcq;
 
-	if (type != MLX4_EVENT_TYPE_CQ_ERROR) {
+    if (type == MLX4_EVENT_TYPE_CQ_ERROR) {
+        event.type = IB_AE_CQ_ERROR;
+    }
+    else if (type == MLX4_EVENT_TYPE_CQ_OVERFLOW) {
+        event.type = IB_AE_CQ_OVERFLOW;
+    }
+    else {
 		MLX4_PRINT(TRACE_LEVEL_WARNING, MLX4_DBG_DRV,( "mlx4_ib: Unexpected event type %d "
 		       "on CQ %06x\n", type, cq->cqn));
 		return;
@@ -66,7 +72,6 @@ static void mlx4_ib_cq_event(struct mlx4
 
 	ibcq = &to_mibcq(cq)->ibcq;
 	if (ibcq->event_handler) {
-		event.type = (ib_async_event_t)IB_EVENT_CQ_ERR;
 		event.context = ibcq->cq_context;
 		event.vendor_specific = type;
 		ibcq->event_handler(&event);
diff -dwup3 -x *svn* -r c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\hw\mlx4\kernel\bus\inc\device.h .\hw\mlx4\kernel\bus\inc\device.h
--- c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\hw\mlx4\kernel\bus\inc\device.h	Thu Mar 29 00:15:28 2012
+++ .\hw\mlx4\kernel\bus\inc\device.h	Thu Jul 26 15:31:14 2012
@@ -111,6 +111,7 @@ enum mlx4_event {
 	MLX4_EVENT_TYPE_LOCAL_CATAS_ERROR  = 0x08,
 	MLX4_EVENT_TYPE_PORT_CHANGE 	   = 0x09,
 	MLX4_EVENT_TYPE_CMD 			   = 0x0a,
+	MLX4_EVENT_TYPE_CQ_OVERFLOW	   = 0x0c,
 	MLX4_EVENT_TYPE_ECC_DETECT		   = 0x0e,
 	MLX4_EVENT_TYPE_EQ_OVERFLOW 	   = 0x0f,
 	MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR = 0x10,
@@ -120,11 +121,11 @@ enum mlx4_event {
 	MLX4_EVENT_TYPE_SRQ_LIMIT		   = 0x14,
 	MLX4_EVENT_TYPE_COMM_CHANNEL	   = 0x18,
 	MLX4_EVENT_TYPE_VEP_UPDATE		   = 0x19,
+	MLX4_EVENT_TYPE_OP_REQUIRED	   = 0x1a,
 	MLX4_EVENT_TYPE_MAC_UPDATE		   = 0x20,
 	MLX4_EVENT_TYPE_PPF_REMOVE		   = 0xf0,
 	MLX4_EVENT_TYPE_SQP_UPDATE	   = 0xfe,
-	MLX4_EVENT_TYPE_NONE		   = 0xff,
-	MLX4_EVENT_TYPE_OP_REQUIRED	   = 0x1a
+	MLX4_EVENT_TYPE_NONE		   = 0xff
 };
 
 enum {
diff -dwup3 -x *svn* -r c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\hw\mlx4\kernel\bus\net\eq.c .\hw\mlx4\kernel\bus\net\eq.c
--- c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\hw\mlx4\kernel\bus\net\eq.c	Thu Mar 29 00:15:28 2012
+++ .\hw\mlx4\kernel\bus\net\eq.c	Thu Jul 26 15:31:14 2012
@@ -521,18 +521,24 @@ static int mlx4_eq_int(struct mlx4_dev *
 			break;
 
 		case MLX4_EVENT_TYPE_CQ_ERROR:
-			MLX4_PRINT(TRACE_LEVEL_WARNING, MLX4_DBG_DRV,( "%s: CQ %s on CQN %06x\n",
+            if (eqe->event.cq_err.syndrome == 1) {
+				MLX4_PRINT(TRACE_LEVEL_WARNING, MLX4_DBG_DRV,( "%s: CQ overrun on CQN %06x\n",
+					  dev->pdev->name, 
+					  be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff));
+                eqe->type = MLX4_EVENT_TYPE_CQ_OVERFLOW;
+            } else {
+				MLX4_PRINT(TRACE_LEVEL_WARNING, MLX4_DBG_DRV,( "%s: CQ access violation on CQN %06x\n",
 				dev->pdev->name, 
-				  eqe->event.cq_err.syndrome == 1 ?
-				  "overrun" : "access violation",
 				  be32_to_cpu(eqe->event.cq_err.cqn) & 0xffffff));
+            }
 			if (mlx4_is_master(dev)) {
 				/* TODO: forward only to slave owning the CQ */
 				ret = mlx4_get_slave_from_resource_id(dev, RES_CQ, eqe->event.cq_err.cqn, &slave);
 				if (!ret)
 					mlx4_slave_event(dev, slave, eqe);
-			} else
+			} else {
 				mlx4_cq_event(dev, eq, be32_to_cpu(eqe->event.cq_err.cqn), eqe->type);
+			}
 			break;
 
 		case MLX4_EVENT_TYPE_EQ_OVERFLOW:
diff -dwup3 -x *svn* -r c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\inc\iba\ib_types.h .\inc\iba\ib_types.h
--- c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\inc\iba\ib_types.h	Thu Aug 02 13:08:45 2012
+++ .\inc\iba\ib_types.h	Thu Jul 26 15:31:14 2012
@@ -9622,6 +9622,7 @@ typedef enum _ib_async_event_t {
 	IB_AE_SM_CHANGE,
 	IB_AE_GID_CHANGE,
 	IB_AE_RESET_4_RMV,
+    IB_AE_CQ_OVERFLOW,
 	IB_AE_UNKNOWN		/* ALWAYS LAST ENUM VALUE */
 
 }	ib_async_event_t;
diff -dwup3 -x *svn* -r c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\core\al\ib_statustext.c .\core\al\ib_statustext.c
--- c:\dev\openib\ofw\gen1\branches\mlx4_30\trunk\core\al\ib_statustext.c	Thu May 31 11:22:16 2012
+++ .\core\al\ib_statustext.c	Wed May 23 18:26:47 2012
@@ -151,6 +151,7 @@ static const char* const __ib_async_even
 	"IB_AE_SM_CHANGE",
 	"IB_AE_GID_CHANGE",
 	"IB_AE_RESET_4_RMV",
+    "IB_AE_CQ_OVERFLOW",
 	"IB_AE_UNKNOWN"
 };
 
@@ -229,6 +230,7 @@ ib_get_wc_type_str(
 		else
 			return __ib_wc_send_type_str[wc_type];
 }
+
 
 static const char* const __ib_wr_type_str[] =
 {
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ndv2.14.patch
Type: application/octet-stream
Size: 4910 bytes
Desc: ndv2.14.patch
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20130221/ce02e32b/attachment.obj>