[ofw] [PATCH 2/4] Avoid the SM

Fab Tillier ftillier at windows.microsoft.com
Thu Aug 21 09:51:35 PDT 2008


This patch adds a new IOCTL to IPoIB for converting a destination Ethernet MAC address to a path record.  The path records is created using information from the broadcast group and previous work completions.

The limitations of this patch is that the rate is that of the broadcast group, which in a mixed rate fabric (SDR/DDR/QDR) will result in suboptimal data transfer rates.

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

diff -up -r -X trunk\docs\dontdiff.txt -I \$Id: old\inc\iba\ib_at_ioctl.h trunk\inc\iba\ib_at_ioctl.h
--- old\inc\iba\ib_at_ioctl.h   Tue Jul 08 10:15:55 2008
+++ trunk\inc\iba\ib_at_ioctl.h Thu Aug 21 09:50:26 2008
@@ -146,6 +146,24 @@ typedef struct _IOCTL_IBAT_IP_TO_PORT_OU
 } IOCTL_IBAT_IP_TO_PORT_OUT;


+/** This IRP is used to convert a remote MAC addresses to a remote GID */
+#define        IOCTL_IBAT_MAC_TO_PATH IOCTL_IBAT( 5 )
+
+typedef struct _IOCTL_IBAT_MAC_TO_PATH_IN
+{
+       ULONG                           Version;
+       UINT64                          PortGuid;
+       UCHAR                           DestMac[IBAT_MAC_LEN];
+
+} IOCTL_IBAT_MAC_TO_PATH_IN;
+
+typedef struct _IOCTL_IBAT_MAC_TO_PATH_OUT
+{
+       ib_path_rec_t           Path;
+
+} IOCTL_IBAT_MAC_TO_PATH_OUT;
+
+
 #define        IBAT_DEV_NAME   L"\\Device\\ibat"
 #define        IBAT_DOS_DEV_NAME L"\\DosDevices\\Global\\ibat"
 #define        IBAT_WIN32_NAME L"\\\\.\\ibat"
diff -up -r -X trunk\docs\dontdiff.txt -I \$Id: old\ulp\ipoib\kernel\ipoib_ibat.c trunk\ulp\ipoib\kernel\ipoib_ibat.c
--- old\ulp\ipoib\kernel\ipoib_ibat.c   Tue Jul 08 10:15:30 2008
+++ trunk\ulp\ipoib\kernel\ipoib_ibat.c Thu Aug 21 09:50:26 2008
@@ -329,6 +329,80 @@ __ibat_mac_to_gid(


 static NTSTATUS
+__ibat_mac_to_path(
+       IN                              IRP                                                     *pIrp,
+       IN                              IO_STACK_LOCATION                       *pIoStack )
+{
+       NTSTATUS                                        status = STATUS_INVALID_PARAMETER;
+       IOCTL_IBAT_MAC_TO_PATH_IN       *pIn;
+       IOCTL_IBAT_MAC_TO_PATH_OUT      *pOut;
+       KLOCK_QUEUE_HANDLE                      hdl;
+       cl_list_item_t                          *pItem;
+       ipoib_adapter_t                         *pAdapter;
+
+       IPOIB_ENTER(IPOIB_DBG_IOCTL);
+
+       if( pIoStack->Parameters.DeviceIoControl.InputBufferLength !=
+               sizeof(IOCTL_IBAT_MAC_TO_PATH_IN) )
+       {
+               IPOIB_PRINT_EXIT( TRACE_LEVEL_ERROR, IPOIB_DBG_ERROR,
+                       ("Invalid input buffer size.\n") );
+               return STATUS_INVALID_PARAMETER;
+       }
+
+       if( pIoStack->Parameters.DeviceIoControl.OutputBufferLength !=
+               sizeof(IOCTL_IBAT_MAC_TO_PATH_OUT) )
+       {
+               IPOIB_PRINT_EXIT( TRACE_LEVEL_ERROR, IPOIB_DBG_ERROR,
+                       ("Invalid output buffer size.\n") );
+               return STATUS_INVALID_PARAMETER;
+       }
+
+       pIn = pIrp->AssociatedIrp.SystemBuffer;
+       pOut = pIrp->AssociatedIrp.SystemBuffer;
+
+       if( pIn->Version != IBAT_IOCTL_VERSION )
+       {
+               IPOIB_PRINT_EXIT( TRACE_LEVEL_ERROR, IPOIB_DBG_ERROR,
+                       ("Invalid version.\n") );
+               return STATUS_INVALID_PARAMETER;
+       }
+
+       KeAcquireInStackQueuedSpinLock( &g_ipoib.lock, &hdl );
+
+       for( pItem = cl_qlist_head( &g_ipoib.adapter_list );
+               pItem != cl_qlist_end( &g_ipoib.adapter_list );
+               pItem = cl_qlist_next( pItem ) )
+       {
+               pAdapter = CONTAINING_RECORD( pItem, ipoib_adapter_t, entry );
+               if( pIn->PortGuid != pAdapter->guids.port_guid.guid )
+                       continue;
+
+               /* Found the port - lookup the MAC. */
+               cl_obj_lock( &pAdapter->obj );
+               if( pAdapter->p_port )
+               {
+                       status = ipoib_mac_to_path(
+                               pAdapter->p_port, *(mac_addr_t*)pIn->DestMac, &pOut->Path );
+
+                       if( NT_SUCCESS( status ) )
+                       {
+                               pIrp->IoStatus.Information =
+                                       sizeof(IOCTL_IBAT_MAC_TO_PATH_OUT);
+                       }
+               }
+               cl_obj_unlock( &pAdapter->obj );
+               break;
+       }
+
+       KeReleaseInStackQueuedSpinLock( &hdl );
+
+       IPOIB_EXIT( IPOIB_DBG_IOCTL );
+       return status;
+}
+
+
+static NTSTATUS
 __ibat_ip_to_port(
        IN                              IRP                                                     *pIrp,
        IN                              IO_STACK_LOCATION                       *pIoStack )
@@ -571,6 +645,12 @@ __ipoib_dispatch(
                IPOIB_PRINT(TRACE_LEVEL_INFORMATION, IPOIB_DBG_IOCTL,
                        ("IOCTL_IBAT_IP_TO_PORT received\n" ));
                status = __ibat_ip_to_port( pIrp, pIoStack );
+               break;
+
+       case IOCTL_IBAT_MAC_TO_PATH:
+               IPOIB_PRINT(TRACE_LEVEL_INFORMATION, IPOIB_DBG_IOCTL,
+                       ("IOCTL_IBAT_MAC_TO_PATH received\n" ));
+               status = __ibat_mac_to_path( pIrp, pIoStack );
                break;

        default:
diff -up -r -X trunk\docs\dontdiff.txt -I \$Id: old\ulp\ipoib\kernel\ipoib_port.c trunk\ulp\ipoib\kernel\ipoib_port.c
--- old\ulp\ipoib\kernel\ipoib_port.c   Thu Aug 21 09:49:49 2008
+++ trunk\ulp\ipoib\kernel\ipoib_port.c Thu Aug 21 09:50:26 2008
@@ -4569,6 +4569,86 @@ ipoib_mac_to_gid(
 }


+NTSTATUS
+ipoib_mac_to_path(
+       IN                              ipoib_port_t* const                     p_port,
+       IN              const   mac_addr_t                                      mac,
+               OUT                     ib_path_rec_t*                          p_path )
+{
+       ipoib_endpt_t*  p_endpt;
+       cl_map_item_t   *p_item;
+       uint64_t                key = 0;
+       uint8_t                 sl;
+       net32_t                 flow_lbl;
+       uint8_t                 hop_limit;
+
+       IPOIB_ENTER( IPOIB_DBG_ENDPT );
+
+       cl_memcpy( &key, &mac, sizeof(mac_addr_t) );
+
+       cl_obj_lock( &p_port->obj );
+
+       if( p_port->p_local_endpt == NULL )
+       {
+               cl_obj_unlock( &p_port->obj );
+               IPOIB_PRINT_EXIT( TRACE_LEVEL_ERROR, IPOIB_DBG_ERROR,
+                       ("No local endpoint.\n") );
+               return STATUS_INVALID_PARAMETER;
+       }
+
+       if( mac.addr[0] == 0 && mac.addr[1] == 0 && mac.addr[2] == 0 &&
+               mac.addr[3] == 0 && mac.addr[4] == 0 && mac.addr[5] == 0 )
+       {
+               p_endpt = p_port->p_local_endpt;
+       }
+       else
+       {
+               p_item = cl_qmap_get( &p_port->endpt_mgr.mac_endpts, key );
+               if( p_item == cl_qmap_end( &p_port->endpt_mgr.mac_endpts ) )
+               {
+                       cl_obj_unlock( &p_port->obj );
+                       IPOIB_PRINT_EXIT( TRACE_LEVEL_ERROR, IPOIB_DBG_ERROR,
+                               ("Failed endpoint lookup.\n") );
+                       return STATUS_INVALID_PARAMETER;
+               }
+
+               p_endpt = PARENT_STRUCT( p_item, ipoib_endpt_t, mac_item );
+       }
+
+       p_path->resv0 = 0;
+       p_path->dgid = p_endpt->dgid;
+       p_path->sgid = p_port->p_local_endpt->dgid;
+       p_path->dlid = p_endpt->dlid;
+       p_path->slid = p_port->p_local_endpt->dlid;
+
+       ib_member_get_sl_flow_hop(
+               p_port->ib_mgr.bcast_rec.sl_flow_hop,
+               &sl,
+               &flow_lbl,
+               &hop_limit
+               );
+       ib_path_rec_set_hop_flow_raw( p_path, hop_limit, flow_lbl, FALSE );
+
+       p_path->tclass = p_port->ib_mgr.bcast_rec.tclass;
+       p_path->num_path = 1;
+       p_path->pkey = p_port->ib_mgr.bcast_rec.pkey;
+       p_path->mtu = p_port->ib_mgr.bcast_rec.mtu;
+       p_path->rate = p_port->ib_mgr.bcast_rec.rate;
+       if( p_path->slid == p_path->dlid )
+               p_path->pkt_life = 0;
+       else
+               p_path->pkt_life = p_port->ib_mgr.bcast_rec.pkt_life;
+       p_path->preference = 0;
+       p_path->resv1 = 0;
+       p_path->resv2 = 0;
+
+       cl_obj_unlock( &p_port->obj );
+
+       IPOIB_EXIT( IPOIB_DBG_ENDPT );
+       return STATUS_SUCCESS;
+}
+
+
 static inline NDIS_STATUS
 __endpt_mgr_ref(
        IN                              ipoib_port_t* const                     p_port,
diff -up -r -X trunk\docs\dontdiff.txt -I \$Id: old\ulp\ipoib\kernel\ipoib_port.h trunk\ulp\ipoib\kernel\ipoib_port.h
--- old\ulp\ipoib\kernel\ipoib_port.h   Thu Aug 21 08:40:18 2008
+++ trunk\ulp\ipoib\kernel\ipoib_port.h Thu Aug 21 09:50:26 2008
@@ -613,6 +613,12 @@ ipoib_mac_to_gid(
        IN              const   mac_addr_t                                      mac,
                OUT                     ib_gid_t*                                       p_gid );

+NTSTATUS
+ipoib_mac_to_path(
+       IN                              ipoib_port_t* const                     p_port,
+       IN              const   mac_addr_t                                      mac,
+               OUT                     ib_path_rec_t*                          p_path );
+
 inline void ipoib_port_ref(
        IN                              ipoib_port_t *                          p_port,
        IN                              int                                             type);
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ipoib_mac_to_path.patch
Type: application/octet-stream
Size: 6793 bytes
Desc: ipoib_mac_to_path.patch
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20080821/f12db6c1/attachment.obj>


More information about the ofw mailing list