opensm: Reduce heap consumption by multicast routing tables (MFTs)

Heap memory consumption by the unicast and multicast routing tables can be reduced.

This patch is analagous to the previous patch doing this for the unicast routing tables (LFTs).

Using valgrind --tool=massif (for heap profiling), there are couple of place ->38.75% (11,206,656B) 0x43267E: osm_switch_new (osm_switch.c:134) ->12.89% (3,728,256B) 0x40F8C9: osm_mcast_tbl_init (osm_mcast_tbl.c:96)

osm_mcast_tbl_init (osm_mcast_tbl.c:96):
p_tbl->p_mask_tbl = malloc(p_tbl->num_entries *
(IB_MCAST_POSITION_MAX +
1) * IB_MCAST_MASK_SIZE / 8);

num_entries above is set based on the switch's MulticastFDBCap (indicated in it's SM class SwitchInfo attribute).

MFTs are only be increased in size and are never reduced in size. If a realloc for MFT fails, it is treated as a fatal error and OpenSM is exited.

Signed-off-by Hal Rosenstock <[email protected]>

---
Changes since v2:
MFT allocation during routing preparation rather than on table access Changes since v1:
MFT allocation based on actual MLID requests

opensm/include/opensm/osm_mcast_tbl.h +2 -1

index 710d199..85730e6 100644
@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved.
- * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
+ * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
  * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
  *
  * This software is available to you under a choice of one of two

@@ -46,6 +46,7 @@
 #include <iba/ib_types.h>
 #include <complib/cl_qmap.h>
 #include <opensm/osm_base.h>
+#include <opensm/osm_subnet.h>
 
 #ifdef __cplusplus
 #  define BEGIN_C_DECLS extern "C" {

opensm/include/opensm/osm_mcast_tbl.h +36 -4

 	uint16_t(*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
 } osm_mcast_tbl_t;
 /*

@@ -97,7 +99,7 @@ typedef struct osm_mcast_fwdbl {
 *	max_mlid_ho
 *		Maximum MLID value (host order).
 *
-*	pp_mask_tbl
+*	p_mask_tbl
 *		Pointer to a two dimensional array of port_masks for this switch.
 *		The first dimension is MLID, the second dimension is mask position.
 *		This pointer is null for switches that do not support multicast.

@@ -114,8 +116,8 @@ typedef struct osm_mcast_fwdbl {
 *
 * SYNOPSIS
 */
-ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
-				   IN uint8_t num_ports, IN uint16_t capacity);
+void osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl, IN uint8_t num_ports,
+			IN uint16_t capacity);
 /*
 * PARAMETERS
 *	num_ports

@@ -126,7 +128,7 @@ ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
 *		by this switch.
 *
 * RETURN VALUE
-*	IB_SUCCESS on success.
+*	None.
 *
 * NOTES
 *

@@ -158,6 +160,38 @@ void osm_mcast_tbl_delete(IN osm_mcast_tbl_t ** pp_tbl);
 * SEE ALSO
 *********/
 
+/****f* OpenSM: Forwarding Table/osm_mcast_tbl_realloc_mask_tbl
+* NAME
+*	osm_mcast_tbl_realloc_mask_tbl
+*
+* DESCRIPTION
+*	This function reallocates the port mask table if necessary.
+*
+* SYNOPSIS
+*/
+void
+osm_mcast_tbl_realloc_mask_tbl(IN osm_mcast_tbl_t * p_tbl,
+			       IN osm_subn_t * p_subn, IN uintn_t mlid_offset);
+/*
+* PARAMETERS
+*
+*	p_tbl
+*		[in] Pointer to the Multicast Forwarding Table object.
+*
+*	p_subn
+*		[in] Pointer to the subnet object.
+*
+*	mlid_offset
+*		[in] Offset of MLID being accessed.
+*
+* RETURN VALUE
+*	None
+*
+* NOTES
+*
+* SEE ALSO
+*/
+
 /****f* OpenSM: Forwarding Table/osm_mcast_tbl_destroy
 * NAME
 *	osm_mcast_tbl_destroy

opensm/opensm/osm_dump.c +2 -1

index 08b3156..c004b6c 100644
@@ -1,7 +1,7 @@
 /*
  * Copyright (c) 2009 Sun Microsystems, Inc. All rights reserved.
  * Copyright (c) 2004-2008 Voltaire, Inc. All rights reserved.
- * Copyright (c) 2002-2006 Mellanox Technologies LTD. All rights reserved.
+ * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
  * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
  *
  * This software is available to you under a choice of one of two

@@ -232,6 +232,7 @@ static void dump_ucast_routes(cl_map_item_t * item, FILE * file, void *cxt)
 static void dump_mcast_routes(cl_map_item_t * item, FILE * file, void *cxt)
 {
 	osm_switch_t *p_sw = (osm_switch_t *) item;
+	osm_opensm_t *p_osm = cxt;
 	osm_mcast_tbl_t *p_tbl;
 	int16_t mlid_ho = 0;
 	int16_t mlid_start_ho;

opensm/opensm/osm_dump.c +0 -0

 				mask_entry =
 				    cl_ntoh16((*p_tbl->
 					       p_mask_tbl)[mlid_ho][position]);

opensm/opensm/osm_mcast_mgr.c +3 -0

index 0ee689c..e6da7f6 100644
@@ -1043,6 +1043,34 @@ static int mcast_mgr_set_mftables(osm_sm_t * sm)
 	return ret;
 }
 
+static void alloc_mfts(osm_sm_t * sm)
+{
+	int i;

opensm/opensm/osm_mcast_mgr.c +3 -0

+
 	for (i = 0; i <= sm->p_subn->max_mcast_lid_ho - IB_LID_MCAST_START_HO;
 	     i++)
 		if (sm->p_subn->mgroups[i] || sm->mlids_req[i])

@@ -1101,6 +1131,8 @@ int osm_mcast_mgr_process_mgroups(osm_sm_t * sm)
 		goto exit;
 	}
 
+	alloc_mfts(sm);
+
 	for (i = 0; i <= sm->mlids_req_max; i++) {
 		if (!sm->mlids_req[i])
 			continue;

opensm/opensm/osm_mcast_tbl.c +7 -23

index d7c9529..11070b1 100644
@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2004-2006 Voltaire, Inc. All rights reserved.
- * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
+ * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
  * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
  * Copyright (c) 2009 HNR Consulting. All rights reserved.
  *

@@ -50,11 +50,13 @@
 #include <complib/cl_math.h>
 #include <iba/ib_types.h>
 #include <opensm/osm_mcast_tbl.h>
+#include <opensm/osm_log.h>
+#include <opensm/osm_opensm.h>
 
 /**********************************************************************
  **********************************************************************/
-ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
-				   IN uint8_t num_ports, IN uint16_t capacity)
+void osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl, IN uint8_t num_ports,
+			IN uint16_t capacity)
 {
 	CL_ASSERT(p_tbl);
 	CL_ASSERT(num_ports);

@@ -68,7 +70,7 @@ ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
 		   This switch apparently doesn't support multicast.
 		   Everything is initialized to zero already, so return.
 		 */
-		return IB_SUCCESS;
+		return;
 	}
 
 	p_tbl->num_entries = capacity;

@@ -82,25 +84,6 @@ ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
 					IB_MCAST_BLOCK_SIZE) - 1);
 
 	p_tbl->max_mlid_ho = (uint16_t) (IB_LID_MCAST_START_HO + capacity - 1);
-
-	/*
-	   The number of bytes needed in the mask table is:
-	   The (maximum bit mask 'position' + 1) times the
-	   number of bytes in each bit mask times the
-	   number of MLIDs supported by the table.
-
-	   We must always allocate the array with the maximum position
-	   since it is (and must be) defined that way the table structure
-	   in order to create a pointer to a two dimensional array.
-	 */
-	p_tbl->p_mask_tbl = calloc(p_tbl->num_entries,
-				   (IB_MCAST_POSITION_MAX +
-				    1) * IB_MCAST_MASK_SIZE / 8);
-
-	if (p_tbl->p_mask_tbl == NULL)
-		return IB_INSUFFICIENT_MEMORY;
-
-	return IB_SUCCESS;
 }
 
 /**********************************************************************

@@ -123,6 +106,7 @@ void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
 	CL_ASSERT(p_tbl);
 	CL_ASSERT(mlid_ho >= IB_LID_MCAST_START_HO);
 	CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
+	CL_ASSERT(mlid_ho - IB_LID_MCAST_START_HO < p_tbl->mft_size);
 	CL_ASSERT(p_tbl->p_mask_tbl);
 
 	mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;

opensm/opensm/osm_mcast_tbl.c +11 -0

+{
+	size_t mft_size, size;
+	uint16_t (*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
+
+	if (mlid_offset < p_tbl->mft_size)
+		return;
+
+	/*
+	   The number of bytes needed in the mask table is:
+	   The (maximum bit mask 'position' + 1) times the
+	   number of bytes in each bit mask times the

opensm/opensm/osm_mcast_tbl.c +7 -0

+	size = mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8;
+	p_mask_tbl = realloc(p_tbl->p_mask_tbl, size);
+	if (!p_mask_tbl)
+		goto error;
+	memset((uint8_t *)p_mask_tbl + p_tbl->mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8,
+	       0,
+	       size - p_tbl->mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8);

opensm/opensm/osm_mcast_tbl.c +6 -1

 		mask_offset = port_num / IB_MCAST_MASK_SIZE;
 		bit_mask = cl_ntoh16((uint16_t)
 				     (1 << (port_num % IB_MCAST_MASK_SIZE)));

@@ -181,6 +212,8 @@ boolean_t osm_mcast_tbl_is_any_port(IN const osm_mcast_tbl_t * p_tbl,
 		CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
 
 		mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
+		if (mlid_offset >= p_tbl->mft_size)
+			return FALSE;
 
 		for (position = 0; position <= p_tbl->max_position; position++)
 			result |= (*p_tbl->p_mask_tbl)[mlid_offset][position];

@@ -210,7 +243,7 @@ ib_api_status_t osm_mcast_tbl_set_block(IN osm_mcast_tbl_t * p_tbl,
 
 	mlid_start_ho = (uint16_t) (block_num * IB_MCAST_BLOCK_SIZE);
 
-	if (mlid_start_ho + IB_MCAST_BLOCK_SIZE - 1 > p_tbl->max_mlid_ho)
+	if (mlid_start_ho + IB_MCAST_BLOCK_SIZE - 1 > p_tbl->mft_size)
 		return IB_INVALID_PARAMETER;
 
 	for (i = 0; i < IB_MCAST_BLOCK_SIZE; i++)

@@ -234,6 +267,8 @@ void osm_mcast_tbl_clear_mlid(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho)
 
 	if (p_tbl->p_mask_tbl && (mlid_ho <= p_tbl->max_mlid_ho)) {
 		mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
+		if (mlid_offset >= p_tbl->mft_size)
+			return;
 		for (i = 0; i <= p_tbl->max_position; i++)
 			(*p_tbl->p_mask_tbl)[mlid_offset][i] = 0;
 	}

@@ -250,6 +285,7 @@ boolean_t osm_mcast_tbl_get_block(IN osm_mcast_tbl_t * p_tbl,
 
 	CL_ASSERT(p_tbl);
 	CL_ASSERT(p_block);
+	CL_ASSERT(block_num * IB_MCAST_BLOCK_SIZE <= p_tbl->mft_size);
 
 	if (block_num > p_tbl->max_block_in_use)
 		return FALSE;

opensm/opensm/osm_switch.c +2 -4

index ed0bc66..223791c 100644
@@ -136,9 +136,8 @@ osm_switch_t *osm_switch_new(IN osm_node_t * p_node,
 
 	memset(p_sw->p_prof, 0, sizeof(*p_sw->p_prof) * num_ports);
 
-	if (osm_mcast_tbl_init(&p_sw->mcast_tbl, osm_node_get_num_physp(p_node),
-			       cl_ntoh16(p_si->mcast_cap)))
-		goto err;
+	osm_mcast_tbl_init(&p_sw->mcast_tbl, osm_node_get_num_physp(p_node),
+			   cl_ntoh16(p_si->mcast_cap));
 
 	for (port_num = 0; port_num < num_ports; port_num++)
 		osm_port_prof_construct(&p_sw->p_prof[port_num]);

@@ -508,7 +507,6 @@ static int alloc_lft(IN osm_switch_t * p_sw, uint16_t lids)
 		p_sw->lft = new_lft;
 		p_sw->lft_size = lft_size;
 	}
-
 	return 0;
 }
 

Quoted 50 lines · Hal Rosenstock

Heap memory consumption by the unicast and multicast routing tables can be reduced.

This patch is analagous to the previous patch doing this for the unicast routing tables (LFTs).

Using valgrind --tool=massif (for heap profiling), there are couple of place ->38.75% (11,206,656B) 0x43267E: osm_switch_new (osm_switch.c:134) ->12.89% (3,728,256B) 0x40F8C9: osm_mcast_tbl_init (osm_mcast_tbl.c:96)

osm_mcast_tbl_init (osm_mcast_tbl.c:96):
        p_tbl->p_mask_tbl = malloc(p_tbl->num_entries * (IB_MCAST_POSITION_MAX +
                                    1) * IB_MCAST_MASK_SIZE / 8);

num_entries above is set based on the switch's MulticastFDBCap (indicated in it's SM class SwitchInfo attribute).

MFTs are only be increased in size and are never reduced in size. If a realloc for MFT fails, it is treated as a fatal error and OpenSM is exited.

Signed-off-by Hal Rosenstock <[email protected]>

---
Changes since v2:
MFT allocation during routing preparation rather than on table access Changes since v1:
MFT allocation based on actual MLID requests

opensm/include/opensm/osm_mcast_tbl.h +2 -1

index 710d199..85730e6 100644
@@ -1,6 +1,6 @@
  /*
   * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved.
- * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
+ * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
   *
   * This software is available to you under a choice of one of two

@@ -46,6 +46,7 @@
  #include <iba/ib_types.h>
  #include <complib/cl_qmap.h>
  #include <opensm/osm_base.h>
+#include <opensm/osm_subnet.h>
 
  #ifdef __cplusplus
  #  define BEGIN_C_DECLS extern "C" {

opensm/include/opensm/osm_mcast_tbl.h typedef struct osm_mcast_fwdbl { · line 78

Quoted 4 lines

opensm/include/opensm/osm_mcast_tbl.h +1 -0

@@ -74,6 +75,7 @@ typedef struct osm_mcast_fwdbl {
       int16_t max_block_in_use;
       uint16_t num_entries;
       uint16_t max_mlid_ho;
+     uint16_t mft_size;

Description for this newly introduced field would be helpful - I had to spend some time realizing about which units are used here.

Quoted 93 lines

opensm/include/opensm/osm_mcast_tbl.h +36 -4

       uint16_t(*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
  } osm_mcast_tbl_t;
  /*

@@ -97,7 +99,7 @@ typedef struct osm_mcast_fwdbl {
  *    max_mlid_ho
  *            Maximum MLID value (host order).
  *
-*    pp_mask_tbl
+*    p_mask_tbl
  *            Pointer to a two dimensional array of port_masks for this switch.
  *            The first dimension is MLID, the second dimension is mask position.
  *            This pointer is null for switches that do not support multicast.

@@ -114,8 +116,8 @@ typedef struct osm_mcast_fwdbl {
  *
  * SYNOPSIS
  */
-ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
-                                IN uint8_t num_ports, IN uint16_t capacity);
+void osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl, IN uint8_t num_ports,
+                     IN uint16_t capacity);
  /*
  * PARAMETERS
  *    num_ports

@@ -126,7 +128,7 @@ ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
  *            by this switch.
  *
  * RETURN VALUE
-*    IB_SUCCESS on success.
+*    None.
  *
  * NOTES
  *

@@ -158,6 +160,38 @@ void osm_mcast_tbl_delete(IN osm_mcast_tbl_t ** pp_tbl);
  * SEE ALSO
  *********/
 
+/****f* OpenSM: Forwarding Table/osm_mcast_tbl_realloc_mask_tbl
+* NAME
+*    osm_mcast_tbl_realloc_mask_tbl
+*
+* DESCRIPTION
+*    This function reallocates the port mask table if necessary.
+*
+* SYNOPSIS
+*/
+void
+osm_mcast_tbl_realloc_mask_tbl(IN osm_mcast_tbl_t * p_tbl,
+                            IN osm_subn_t * p_subn, IN uintn_t mlid_offset);
+/*
+* PARAMETERS
+*
+*    p_tbl
+*            [in] Pointer to the Multicast Forwarding Table object.
+*
+*    p_subn
+*            [in] Pointer to the subnet object.
+*
+*    mlid_offset
+*            [in] Offset of MLID being accessed.
+*
+* RETURN VALUE
+*    None
+*
+* NOTES
+*
+* SEE ALSO
+*/
+
  /****f* OpenSM: Forwarding Table/osm_mcast_tbl_destroy
  * NAME
  *    osm_mcast_tbl_destroy

opensm/opensm/osm_dump.c +2 -1

index 08b3156..c004b6c 100644
@@ -1,7 +1,7 @@
  /*
   * Copyright (c) 2009 Sun Microsystems, Inc. All rights reserved.
   * Copyright (c) 2004-2008 Voltaire, Inc. All rights reserved.
- * Copyright (c) 2002-2006 Mellanox Technologies LTD. All rights reserved.
+ * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
   *
   * This software is available to you under a choice of one of two

@@ -232,6 +232,7 @@ static void dump_ucast_routes(cl_map_item_t * item, FILE * file, void *cxt)
  static void dump_mcast_routes(cl_map_item_t * item, FILE * file, void *cxt)
  {
       osm_switch_t *p_sw = (osm_switch_t *) item;
+     osm_opensm_t *p_osm = cxt;
       osm_mcast_tbl_t *p_tbl;
       int16_t mlid_ho = 0;
       int16_t mlid_start_ho;

opensm/opensm/osm_dump.c static void dump_mcast_routes(cl_map_item_t * item, FILE * file, void *cxt) · line 267

Quoted 6 lines

opensm/opensm/osm_dump.c +3 -0

@@ -261,6 +262,9 @@ static void dump_mcast_routes(cl_map_item_t * item, FILE * file, void *cxt)
                       sprintf(mlid_hdr, "0x%04X :",
                               mlid_ho + IB_LID_MCAST_START_HO);
                       while (position <= p_tbl->max_position) {
+                             osm_mcast_tbl_realloc_mask_tbl(p_tbl,
+                                                            &p_osm->subn,
+                                                            mlid_ho);

Hmm, why to realloc here?

Quoted 14 lines

                              mask_entry =
                                  cl_ntoh16((*p_tbl-> p_mask_tbl)[mlid_ho][position]);

opensm/opensm/osm_mcast_mgr.c +3 -0

index 0ee689c..e6da7f6 100644
@@ -1043,6 +1043,34 @@ static int mcast_mgr_set_mftables(osm_sm_t * sm)
       return ret;
  }
 
+static void alloc_mfts(osm_sm_t * sm)
+{
+     int i;

opensm/opensm/osm_mcast_mgr.c static int mcast_mgr_set_mftables(osm_sm_t * sm) · line 1058

Quoted 10 lines

opensm/opensm/osm_mcast_mgr.c +10 -0

+     cl_map_item_t *item;
+     osm_switch_t *p_sw;
+     int max_mlid = 0;
+
+     for (i = sm->p_subn->max_mcast_lid_ho - IB_LID_MCAST_START_HO; i >= 0;
+          i--) {
+             if (!sm->p_subn->mgroups[i])
+                     continue;
+             max_mlid = i + IB_LID_MCAST_START_HO;
+             break;

Something simpler like:

               if (sm->p_subn->mgroups[i]) { max_mlid = i + IB_LID_MCAST_START_HO; break;
               }

would make more sense.

opensm/opensm/osm_mcast_mgr.c static int mcast_mgr_set_mftables(osm_sm_t * sm) · line 1070

Quoted 12 lines

opensm/opensm/osm_mcast_mgr.c +12 -0

+     }
+
+     if (max_mlid == 0)
+             return;
+
+     /* Now, walk switches and (re)allocate multicast tables */
+     for (item = cl_qmap_head(&sm->p_subn->sw_guid_tbl);
+          item != cl_qmap_end(&sm->p_subn->sw_guid_tbl);
+          item = cl_qmap_next(item)) {
+             p_sw = (osm_switch_t *)item;
+             osm_mcast_tbl_realloc_mask_tbl(&p_sw->mcast_tbl, sm->p_subn,
+                                            max_mlid - IB_LID_MCAST_START_HO);

I would suggest to have a return value here and...

opensm/opensm/osm_mcast_mgr.c int osm_mcast_mgr_process(osm_sm_t * sm) · line 1094

Quoted 11 lines

opensm/opensm/osm_mcast_mgr.c +4 -0

+     }
+}
+
  /**********************************************************************
   **********************************************************************/
  int osm_mcast_mgr_process(osm_sm_t * sm)

@@ -1063,6 +1091,8 @@ int osm_mcast_mgr_process(osm_sm_t * sm)
               goto exit;
       }
 
+     alloc_mfts(sm);

to verify allocation status and break multicast routing calculation when there is no room for MFTs.

Quoted 85 lines

opensm/opensm/osm_mcast_mgr.c +3 -0

+
       for (i = 0; i <= sm->p_subn->max_mcast_lid_ho - IB_LID_MCAST_START_HO;
            i++)
               if (sm->p_subn->mgroups[i] || sm->mlids_req[i])

@@ -1101,6 +1131,8 @@ int osm_mcast_mgr_process_mgroups(osm_sm_t * sm)
               goto exit;
       }
 
+     alloc_mfts(sm);
+
       for (i = 0; i <= sm->mlids_req_max; i++) {
               if (!sm->mlids_req[i])
                       continue;

opensm/opensm/osm_mcast_tbl.c +5 -3

index d7c9529..11070b1 100644
@@ -1,6 +1,6 @@
  /*
   * Copyright (c) 2004-2006 Voltaire, Inc. All rights reserved.
- * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
+ * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
   * Copyright (c) 2009 HNR Consulting. All rights reserved.
   *

@@ -50,11 +50,13 @@
  #include <complib/cl_math.h>
  #include <iba/ib_types.h>
  #include <opensm/osm_mcast_tbl.h>
+#include <opensm/osm_log.h>
+#include <opensm/osm_opensm.h>
 
  /**********************************************************************
   **********************************************************************/
-ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
-                                IN uint8_t num_ports, IN uint16_t capacity)
+void osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl, IN uint8_t num_ports,
+                     IN uint16_t capacity)
  {
       CL_ASSERT(p_tbl);
       CL_ASSERT(num_ports);

@@ -68,7 +70,7 @@ ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,

                 This switch apparently doesn't support multicast. Everything is initialized to zero already, so return. */
-             return IB_SUCCESS;
+             return;
      }

      p_tbl->num_entries = capacity;

opensm/opensm/osm_mcast_tbl.c +1 -19

@@ -82,25 +84,6 @@ ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
                                       IB_MCAST_BLOCK_SIZE) - 1);
 
       p_tbl->max_mlid_ho = (uint16_t) (IB_LID_MCAST_START_HO + capacity - 1);
-
-     /*
-        The number of bytes needed in the mask table is:
-        The (maximum bit mask 'position' + 1) times the
-        number of bytes in each bit mask times the
-        number of MLIDs supported by the table.
-
-        We must always allocate the array with the maximum position
-        since it is (and must be) defined that way the table structure
-        in order to create a pointer to a two dimensional array.
-      */
-     p_tbl->p_mask_tbl = calloc(p_tbl->num_entries,
-                                (IB_MCAST_POSITION_MAX +
-                                 1) * IB_MCAST_MASK_SIZE / 8);
-
-     if (p_tbl->p_mask_tbl == NULL)
-             return IB_INSUFFICIENT_MEMORY;
-
-     return IB_SUCCESS;
  }
 
  /**********************************************************************

@@ -123,6 +106,7 @@ void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
       CL_ASSERT(p_tbl);
       CL_ASSERT(mlid_ho >= IB_LID_MCAST_START_HO);
       CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
+     CL_ASSERT(mlid_ho - IB_LID_MCAST_START_HO < p_tbl->mft_size);
       CL_ASSERT(p_tbl->p_mask_tbl);
 
       mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;

opensm/opensm/osm_mcast_tbl.c void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho, · line 127

Quoted 6 lines

opensm/opensm/osm_mcast_tbl.c +3 -0

@@ -138,6 +122,51 @@ void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
 
  /**********************************************************************
   **********************************************************************/
+void
+osm_mcast_tbl_realloc_mask_tbl(IN osm_mcast_tbl_t * p_tbl,
+                            IN osm_subn_t * p_subn, IN uintn_t mlid_offset)

I would suggest to simplify the function name to something like 'osm_mcast_tbl_realloc()' - osm_mcast_tbl API doesn't have any information about a buffer internal structure.

Quoted 11 lines

opensm/opensm/osm_mcast_tbl.c +11 -0

+{
+     size_t mft_size, size;
+     uint16_t (*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
+
+     if (mlid_offset < p_tbl->mft_size)
+             return;
+
+     /*
+        The number of bytes needed in the mask table is:
+        The (maximum bit mask 'position' + 1) times the
+        number of bytes in each bit mask times the

opensm/opensm/osm_mcast_tbl.c void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho, · line 148

Quoted 10 lines

opensm/opensm/osm_mcast_tbl.c +10 -0

+        number of MLIDs supported by the table.
+
+        We must always allocate the array with the maximum position
+        since it is (and must be) defined that way the table structure
+        in order to create a pointer to a two dimensional array.
+      */
+     mft_size = (mlid_offset + IB_MCAST_BLOCK_SIZE) /
+                 IB_MCAST_BLOCK_SIZE * IB_MCAST_BLOCK_SIZE;
+     if (mft_size > (p_tbl->max_block + 1) * IB_MCAST_BLOCK_SIZE)
+             mft_size = (p_tbl->max_block + 1) * IB_MCAST_BLOCK_SIZE;

Hmm, wouldn't this:

       mft_size = (mlid_offset / IB_MCAST_BLOCK_SIZE + 1) * IB_MCAST_BLOCK_SIZE;

do the same as lines above?

Quoted 7 lines

opensm/opensm/osm_mcast_tbl.c +7 -0

+     size = mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8;
+     p_mask_tbl = realloc(p_tbl->p_mask_tbl, size);
+     if (!p_mask_tbl)
+             goto error;
+     memset((uint8_t *)p_mask_tbl + p_tbl->mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8,
+            0,
+            size - p_tbl->mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8);

opensm/opensm/osm_mcast_tbl.c void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho, · line 165

Quoted 10 lines

opensm/opensm/osm_mcast_tbl.c +10 -0

+     p_tbl->p_mask_tbl = p_mask_tbl;
+     p_tbl->mft_size = mft_size;
+     return;
+
+error:
+     OSM_LOG(&p_subn->p_osm->log, OSM_LOG_SYS,
+             "Reallocation of multicast mask table failed - exiting\n");
+     OSM_LOG(&p_subn->p_osm->log, OSM_LOG_ERROR, " ERR 6401: "
+             "Reallocation of multicast mask table failed - exiting\n");
+     exit(1);

Why to break whole OpenSM execution (print syslog errors, etc.)? Just return -1 to the caller.

Quoted 10 lines

opensm/opensm/osm_mcast_tbl.c +4 -0

+}
+
+/**********************************************************************
+ **********************************************************************/
  boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl,
                               IN uint16_t mlid_ho, IN uint8_t port_num)
  {

@@ -154,6 +183,8 @@ boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl,
               CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
 

opensm/opensm/osm_mcast_tbl.c boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl, · line 187

Quoted 3 lines

opensm/opensm/osm_mcast_tbl.c +2 -0

               mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
+             if (mlid_offset >= p_tbl->mft_size)
+                     return FALSE;

If you are introducing new field mft_size, I would also suggest to change the meaning of max_mlid_ho field to be max configured mlid for this table and not max capability mlid (which is almost duplicated by num_entries field).

I suppose that this can be done as separate patch.

Then all such and similar (many below) checks should be performed against this actually configured max mlid and not against table size. As we discussed already this prevents some bugs for 'max_mlid < table_size - 1' case.

Sasha

Quoted 62 lines

opensm/opensm/osm_mcast_tbl.c +6 -1

               mask_offset = port_num / IB_MCAST_MASK_SIZE;
               bit_mask = cl_ntoh16((uint16_t)
                                    (1 << (port_num % IB_MCAST_MASK_SIZE)));

@@ -181,6 +212,8 @@ boolean_t osm_mcast_tbl_is_any_port(IN const osm_mcast_tbl_t * p_tbl,
               CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
 
               mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
+             if (mlid_offset >= p_tbl->mft_size)
+                     return FALSE;
 
               for (position = 0; position <= p_tbl->max_position; position++)
                       result |= (*p_tbl->p_mask_tbl)[mlid_offset][position];

@@ -210,7 +243,7 @@ ib_api_status_t osm_mcast_tbl_set_block(IN osm_mcast_tbl_t * p_tbl,
 
       mlid_start_ho = (uint16_t) (block_num * IB_MCAST_BLOCK_SIZE);
 
-     if (mlid_start_ho + IB_MCAST_BLOCK_SIZE - 1 > p_tbl->max_mlid_ho)
+     if (mlid_start_ho + IB_MCAST_BLOCK_SIZE - 1 > p_tbl->mft_size)
               return IB_INVALID_PARAMETER;
 
       for (i = 0; i < IB_MCAST_BLOCK_SIZE; i++)

@@ -234,6 +267,8 @@ void osm_mcast_tbl_clear_mlid(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho)
 
       if (p_tbl->p_mask_tbl && (mlid_ho <= p_tbl->max_mlid_ho)) {
               mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
+             if (mlid_offset >= p_tbl->mft_size)
+                     return;
               for (i = 0; i <= p_tbl->max_position; i++)
                       (*p_tbl->p_mask_tbl)[mlid_offset][i] = 0;
       }

@@ -250,6 +285,7 @@ boolean_t osm_mcast_tbl_get_block(IN osm_mcast_tbl_t * p_tbl,
 
       CL_ASSERT(p_tbl);
       CL_ASSERT(p_block);
+     CL_ASSERT(block_num * IB_MCAST_BLOCK_SIZE <= p_tbl->mft_size);
 
       if (block_num > p_tbl->max_block_in_use)
               return FALSE;

opensm/opensm/osm_switch.c +2 -4

index ed0bc66..223791c 100644
@@ -136,9 +136,8 @@ osm_switch_t *osm_switch_new(IN osm_node_t * p_node,
 
       memset(p_sw->p_prof, 0, sizeof(*p_sw->p_prof) * num_ports);
 
-     if (osm_mcast_tbl_init(&p_sw->mcast_tbl, osm_node_get_num_physp(p_node),
-                            cl_ntoh16(p_si->mcast_cap)))
-             goto err;
+     osm_mcast_tbl_init(&p_sw->mcast_tbl, osm_node_get_num_physp(p_node),
+                        cl_ntoh16(p_si->mcast_cap));
 
       for (port_num = 0; port_num < num_ports; port_num++)
               osm_port_prof_construct(&p_sw->p_prof[port_num]);

@@ -508,7 +507,6 @@ static int alloc_lft(IN osm_switch_t * p_sw, uint16_t lids)
               p_sw->lft = new_lft;
               p_sw->lft_size = lft_size;
       }
-
       return 0;
  }
 

opensm/opensm/osm_switch.c int osm_switch_prepare_path_rebuild(IN osm_switch_t * p_sw, IN uint16_t max_lids · line 552

Quoted 8 lines

opensm/opensm/osm_switch.c +0 -1

@@ -549,7 +547,6 @@ int osm_switch_prepare_path_rebuild(IN osm_switch_t * p_sw, IN uint16_t max_lids
               p_sw->num_hops = max_lids + 1;
       }
       p_sw->max_lid_ho = max_lids;
-
       return 0;
  }
 

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at  http://vger.kernel.org/majordomo-info.html

Quoted 37 lines

opensm/opensm/osm_mcast_tbl.c void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho, · line 127

Quoted 7 lines

opensm/opensm/osm_mcast_tbl.c +3 -0

@@ -138,6 +122,51 @@ void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
 
  /**********************************************************************
   **********************************************************************/
+void
+osm_mcast_tbl_realloc_mask_tbl(IN osm_mcast_tbl_t * p_tbl,
+                            IN osm_subn_t * p_subn, IN uintn_t mlid_offset)

I would suggest to simplify the function name to something like 'osm_mcast_tbl_realloc()' - osm_mcast_tbl API doesn't have any information about a buffer internal structure.

Quoted 11 lines

opensm/opensm/osm_mcast_tbl.c +11 -0

+{
+     size_t mft_size, size;
+     uint16_t (*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
+
+     if (mlid_offset < p_tbl->mft_size)
+             return;
+
+     /*
+        The number of bytes needed in the mask table is:
+        The (maximum bit mask 'position' + 1) times the
+        number of bytes in each bit mask times the

opensm/opensm/osm_mcast_tbl.c void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho, · line 148

Quoted 10 lines

opensm/opensm/osm_mcast_tbl.c +10 -0

+        number of MLIDs supported by the table.
+
+        We must always allocate the array with the maximum position
+        since it is (and must be) defined that way the table structure
+        in order to create a pointer to a two dimensional array.
+      */
+     mft_size = (mlid_offset + IB_MCAST_BLOCK_SIZE) /
+                 IB_MCAST_BLOCK_SIZE * IB_MCAST_BLOCK_SIZE;
+     if (mft_size > (p_tbl->max_block + 1) * IB_MCAST_BLOCK_SIZE)
+             mft_size = (p_tbl->max_block + 1) * IB_MCAST_BLOCK_SIZE;

Hmm, wouldn't this:

Quoted 3 lines

       mft_size = (mlid_offset / IB_MCAST_BLOCK_SIZE + 1) * IB_MCAST_BLOCK_SIZE;

do the same as lines above?

What about the limit (max_block) check ?

Quoted 11 lines

opensm/opensm/osm_mcast_tbl.c boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl, · line 187

Quoted 6 lines

opensm/opensm/osm_mcast_tbl.c +2 -0

@@ -154,6 +183,8 @@ boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl,
               CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
 
               mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
+             if (mlid_offset >= p_tbl->mft_size)
+                     return FALSE;

If you are introducing new field mft_size, I would also suggest to change the meaning of max_mlid_ho field to be max configured mlid for this table and not max capability mlid (which is almost duplicated by num_entries field).

Yes, almost duplicated.

Quoted 6 lines

I suppose that this can be done as separate patch.

Then all such and similar (many below) checks should be performed against this actually configured max mlid and not against table size. As we discussed already this prevents some bugs for 'max_mlid < table_size - 1' case.

I'll look at this subsequent to this patch.

Quoted 39 lines · Hal Rosenstock

Quoted 39 lines

opensm/opensm/osm_mcast_tbl.c void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho, · line 127

Quoted 7 lines

opensm/opensm/osm_mcast_tbl.c +3 -0

@@ -138,6 +122,51 @@ void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
 
  /**********************************************************************
   **********************************************************************/
+void
+osm_mcast_tbl_realloc_mask_tbl(IN osm_mcast_tbl_t * p_tbl,
+                            IN osm_subn_t * p_subn, IN uintn_t mlid_offset)

I would suggest to simplify the function name to something like 'osm_mcast_tbl_realloc()' - osm_mcast_tbl API doesn't have any information about a buffer internal structure.

Quoted 11 lines

opensm/opensm/osm_mcast_tbl.c +11 -0

+{
+     size_t mft_size, size;
+     uint16_t (*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
+
+     if (mlid_offset < p_tbl->mft_size)
+             return;
+
+     /*
+        The number of bytes needed in the mask table is:
+        The (maximum bit mask 'position' + 1) times the
+        number of bytes in each bit mask times the

opensm/opensm/osm_mcast_tbl.c void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho, · line 148

Quoted 10 lines

opensm/opensm/osm_mcast_tbl.c +10 -0

+        number of MLIDs supported by the table.
+
+        We must always allocate the array with the maximum position
+        since it is (and must be) defined that way the table structure
+        in order to create a pointer to a two dimensional array.
+      */
+     mft_size = (mlid_offset + IB_MCAST_BLOCK_SIZE) /
+                 IB_MCAST_BLOCK_SIZE * IB_MCAST_BLOCK_SIZE;
+     if (mft_size > (p_tbl->max_block + 1) * IB_MCAST_BLOCK_SIZE)
+             mft_size = (p_tbl->max_block + 1) * IB_MCAST_BLOCK_SIZE;

Hmm, wouldn't this:

       mft_size = (mlid_offset / IB_MCAST_BLOCK_SIZE + 1) * IB_MCAST_BLOCK_SIZE;

Quoted 3 lines

Quoted 1 line

do the same as lines above?

What about the limit (max_block) check ?

If passed mlid_offset can exceed this then it is likely needed, but I'm not sure that this is the case - then we are asking for more mlids than a subnet is capable to handle and probably it is blocked somewhere already (but I didn't check really).

Quoted 22 lines

Quoted 4 lines

Quoted 4 lines

(quoted diff) +0 -0

@@ -154,6 +183,8 @@ boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl,
               CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
 

Quoted 8 lines

opensm/opensm/osm_mcast_tbl.c boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl, · line 187

Quoted 3 lines

opensm/opensm/osm_mcast_tbl.c +2 -0

               mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
+             if (mlid_offset >= p_tbl->mft_size)
+                     return FALSE;

If you are introducing new field mft_size, I would also suggest to change the meaning of max_mlid_ho field to be max configured mlid for this table and not max capability mlid (which is almost duplicated by num_entries field).

Yes, almost duplicated.

Quoted 6 lines

I suppose that this can be done as separate patch.

Then all such and similar (many below) checks should be performed against this actually configured max mlid and not against table size. As we discussed already this prevents some bugs for 'max_mlid < table_size - 1' case.

Quoted 1 line

I'll look at this subsequent to this patch.

I would suggest to make it first (guess that it is possible without relation to MFT reallocation), so we will need to change all those lines only once.

Sasha