net/mlx5: E-Switch, Add SR-IOV (FDB) support
authorSaeed Mahameed <saeedm@mellanox.com>
Tue, 1 Dec 2015 16:03:20 +0000 (18:03 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 3 Dec 2015 17:08:46 +0000 (12:08 -0500)
Enabling E-Switch SRIOV for nvfs+1 vports.

Create E-Switch FDB for L2 UC/MC mac steering between VFs/PF and
external vport (Uplink).

FDB contains forwarding rules such as:
UC MAC0 -> vport0(PF).
UC MAC1 -> vport1.
UC MAC2 -> vport2.
MC MACX -> vport0, vport2, Uplink.
MC MACY -> vport1, Uplink.

For unmatched traffic FDB has the following default rules:
Unmached Traffic (src vport != Uplink) -> Uplink.
Unmached Traffic (src vport == Uplink) -> vport0(PF).

FDB rules population:
Each NIC vport (VF) will notify E-Switch manager of its UC/MC vport
context changes via modify vport context command, which will be
translated to an event that will be handled by E-Switch manager (PF)
which will update FDB table accordingly.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx5/core/eswitch.c
drivers/net/ethernet/mellanox/mlx5/core/eswitch.h
drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h
drivers/net/ethernet/mellanox/mlx5/core/sriov.c
include/linux/mlx5/device.h
include/linux/mlx5/flow_table.h
include/linux/mlx5/mlx5_ifc.h

index 1f2f804bde3e79cdf1f161433b8413598612da72..6fa6f013d2c5f89d6b5cadd04484aa2636e59c23 100644 (file)
 #include <linux/mlx5/driver.h>
 #include <linux/mlx5/mlx5_ifc.h>
 #include <linux/mlx5/vport.h>
+#include <linux/mlx5/flow_table.h>
 #include "mlx5_core.h"
 #include "eswitch.h"
 
+#define UPLINK_VPORT 0xFFFF
+
 #define MLX5_DEBUG_ESWITCH_MASK BIT(3)
 
 #define esw_info(dev, format, ...)                             \
@@ -54,18 +57,26 @@ enum {
        MLX5_ACTION_DEL  = 2,
 };
 
-/* HW UC L2 table hash node */
-struct mlx5_uc_l2addr {
+/* E-Switch UC L2 table hash node */
+struct esw_uc_addr {
        struct l2addr_node node;
-       u8                 action;
        u32                table_index;
        u32                vport;
 };
 
-/* Vport UC L2 table hash node */
-struct mlx5_vport_addr {
-       struct l2addr_node node;
-       u8                 action;
+/* E-Switch MC FDB table hash node */
+struct esw_mc_addr { /* SRIOV only */
+       struct l2addr_node     node;
+       struct mlx5_flow_rule *uplink_rule; /* Forward to uplink rule */
+       u32                    refcnt;
+};
+
+/* Vport UC/MC hash node */
+struct vport_addr {
+       struct l2addr_node     node;
+       u8                     action;
+       u32                    vport;
+       struct mlx5_flow_rule *flow_rule; /* SRIOV only */
 };
 
 enum {
@@ -73,7 +84,11 @@ enum {
        MC_ADDR_CHANGE = BIT(1),
 };
 
-static int arm_vport_context_events_cmd(struct mlx5_core_dev *dev, int vport,
+/* Vport context events */
+#define SRIOV_VPORT_EVENTS (UC_ADDR_CHANGE | \
+                           MC_ADDR_CHANGE)
+
+static int arm_vport_context_events_cmd(struct mlx5_core_dev *dev, u16 vport,
                                        u32 events_mask)
 {
        int in[MLX5_ST_SZ_DW(modify_nic_vport_context_in)];
@@ -196,97 +211,492 @@ static void del_l2_table_entry(struct mlx5_core_dev *dev, u32 index)
        free_l2_table_index(l2_table, index);
 }
 
-/* SW E-Switch L2 Table management */
-static int l2_table_addr_add(struct mlx5_eswitch *esw,
-                            u8 mac[ETH_ALEN], u32 vport)
+/* E-Switch FDB flow steering */
+struct dest_node {
+       struct list_head list;
+       struct mlx5_flow_destination dest;
+};
+
+static int _mlx5_flow_rule_apply(struct mlx5_flow_rule *fr)
 {
-       struct hlist_head *hash;
-       struct mlx5_uc_l2addr *addr;
+       bool was_valid = fr->valid;
+       struct dest_node *dest_n;
+       u32 dest_list_size = 0;
+       void *in_match_value;
+       u32 *flow_context;
+       u32 flow_index;
        int err;
+       int i;
+
+       if (list_empty(&fr->dest_list)) {
+               if (fr->valid)
+                       mlx5_del_flow_table_entry(fr->ft, fr->fi);
+               fr->valid = false;
+               return 0;
+       }
+
+       list_for_each_entry(dest_n, &fr->dest_list, list)
+               dest_list_size++;
 
-       hash = esw->l2_table.l2_hash;
-       addr = l2addr_hash_find(hash, mac, struct mlx5_uc_l2addr);
-       if (addr) {
+       flow_context = mlx5_vzalloc(MLX5_ST_SZ_BYTES(flow_context) +
+                                   MLX5_ST_SZ_BYTES(dest_format_struct) *
+                                   dest_list_size);
+       if (!flow_context)
+               return -ENOMEM;
+
+       MLX5_SET(flow_context, flow_context, flow_tag, fr->flow_tag);
+       MLX5_SET(flow_context, flow_context, action, fr->action);
+       MLX5_SET(flow_context, flow_context, destination_list_size,
+                dest_list_size);
+
+       i = 0;
+       list_for_each_entry(dest_n, &fr->dest_list, list) {
+               void *dest_addr = MLX5_ADDR_OF(flow_context, flow_context,
+                                              destination[i++]);
+
+               MLX5_SET(dest_format_struct, dest_addr, destination_type,
+                        dest_n->dest.type);
+               MLX5_SET(dest_format_struct, dest_addr, destination_id,
+                        dest_n->dest.vport_num);
+       }
+
+       in_match_value = MLX5_ADDR_OF(flow_context, flow_context, match_value);
+       memcpy(in_match_value, fr->match_value, MLX5_ST_SZ_BYTES(fte_match_param));
+
+       err = mlx5_add_flow_table_entry(fr->ft, fr->match_criteria_enable,
+                                       fr->match_criteria, flow_context,
+                                       &flow_index);
+       if (!err) {
+               if (was_valid)
+                       mlx5_del_flow_table_entry(fr->ft, fr->fi);
+               fr->fi = flow_index;
+               fr->valid = true;
+       }
+       kfree(flow_context);
+       return err;
+}
+
+static int mlx5_flow_rule_add_dest(struct mlx5_flow_rule *fr,
+                                  struct mlx5_flow_destination *new_dest)
+{
+       struct dest_node *dest_n;
+       int err;
+
+       dest_n = kzalloc(sizeof(*dest_n), GFP_KERNEL);
+       if (!dest_n)
+               return -ENOMEM;
+
+       memcpy(&dest_n->dest, new_dest, sizeof(dest_n->dest));
+       mutex_lock(&fr->mutex);
+       list_add(&dest_n->list, &fr->dest_list);
+       err = _mlx5_flow_rule_apply(fr);
+       if (err) {
+               list_del(&dest_n->list);
+               kfree(dest_n);
+       }
+       mutex_unlock(&fr->mutex);
+       return err;
+}
+
+static int mlx5_flow_rule_del_dest(struct mlx5_flow_rule *fr,
+                                  struct mlx5_flow_destination *dest)
+{
+       struct dest_node *dest_n;
+       struct dest_node *n;
+       int err;
+
+       mutex_lock(&fr->mutex);
+       list_for_each_entry_safe(dest_n, n, &fr->dest_list, list) {
+               if (dest->vport_num == dest_n->dest.vport_num)
+                       goto found;
+       }
+       mutex_unlock(&fr->mutex);
+       return -ENOENT;
+
+found:
+       list_del(&dest_n->list);
+       err = _mlx5_flow_rule_apply(fr);
+       mutex_unlock(&fr->mutex);
+       kfree(dest_n);
+
+       return err;
+}
+
+static struct mlx5_flow_rule *find_fr(struct mlx5_eswitch *esw,
+                                     u8 match_criteria_enable,
+                                     u32 *match_value)
+{
+       struct hlist_head *hash = esw->mc_table;
+       struct esw_mc_addr *esw_mc;
+       u8 *dmac_v;
+
+       dmac_v = MLX5_ADDR_OF(fte_match_param, match_value,
+                             outer_headers.dmac_47_16);
+
+       /* UNICAST FULL MATCH */
+       if (!is_multicast_ether_addr(dmac_v))
+               return NULL;
+
+       /* MULTICAST FULL MATCH */
+       esw_mc = l2addr_hash_find(hash, dmac_v, struct esw_mc_addr);
+
+       return esw_mc ? esw_mc->uplink_rule : NULL;
+}
+
+static struct mlx5_flow_rule *alloc_fr(void *ft,
+                                      u8 match_criteria_enable,
+                                      u32 *match_criteria,
+                                      u32 *match_value,
+                                      u32 action,
+                                      u32 flow_tag)
+{
+       struct mlx5_flow_rule *fr = kzalloc(sizeof(*fr), GFP_KERNEL);
+
+       if (!fr)
+               return NULL;
+
+       fr->match_criteria = kzalloc(MLX5_ST_SZ_BYTES(fte_match_param), GFP_KERNEL);
+       fr->match_value = kzalloc(MLX5_ST_SZ_BYTES(fte_match_param), GFP_KERNEL);
+       if (!fr->match_criteria || !fr->match_value) {
+               kfree(fr->match_criteria);
+               kfree(fr->match_value);
+               kfree(fr);
+               return NULL;
+       }
+
+       memcpy(fr->match_criteria, match_criteria, MLX5_ST_SZ_BYTES(fte_match_param));
+       memcpy(fr->match_value, match_value, MLX5_ST_SZ_BYTES(fte_match_param));
+       fr->match_criteria_enable = match_criteria_enable;
+       fr->flow_tag = flow_tag;
+       fr->action = action;
+
+       mutex_init(&fr->mutex);
+       INIT_LIST_HEAD(&fr->dest_list);
+       atomic_set(&fr->refcount, 0);
+       fr->ft = ft;
+       return fr;
+}
+
+static void deref_fr(struct mlx5_flow_rule *fr)
+{
+       if (!atomic_dec_and_test(&fr->refcount))
+               return;
+
+       kfree(fr->match_criteria);
+       kfree(fr->match_value);
+       kfree(fr);
+}
+
+static struct mlx5_flow_rule *
+mlx5_add_flow_rule(struct mlx5_eswitch *esw,
+                  u8 match_criteria_enable,
+                  u32 *match_criteria,
+                  u32 *match_value,
+                  u32 action,
+                  u32 flow_tag,
+                  struct mlx5_flow_destination *dest)
+{
+       struct mlx5_flow_rule *fr;
+       int err;
+
+       fr = find_fr(esw, match_criteria_enable, match_value);
+       fr = fr ? fr : alloc_fr(esw->fdb_table.fdb, match_criteria_enable, match_criteria,
+                               match_value, action, flow_tag);
+       if (!fr)
+               return NULL;
+
+       atomic_inc(&fr->refcount);
+
+       err = mlx5_flow_rule_add_dest(fr, dest);
+       if (err) {
+               deref_fr(fr);
+               return NULL;
+       }
+
+       return fr;
+}
+
+static void mlx5_del_flow_rule(struct mlx5_flow_rule *fr, u32 vport)
+{
+       struct mlx5_flow_destination dest;
+
+       dest.vport_num = vport;
+       mlx5_flow_rule_del_dest(fr, &dest);
+       deref_fr(fr);
+}
+
+/* E-Switch FDB */
+static struct mlx5_flow_rule *
+esw_fdb_set_vport_rule(struct mlx5_eswitch *esw, u8 mac[ETH_ALEN], u32 vport)
+{
+       int match_header = MLX5_MATCH_OUTER_HEADERS;
+       struct mlx5_flow_destination dest;
+       struct mlx5_flow_rule *flow_rule = NULL;
+       u32 *match_v;
+       u32 *match_c;
+       u8 *dmac_v;
+       u8 *dmac_c;
+
+       match_v = kzalloc(MLX5_ST_SZ_BYTES(fte_match_param), GFP_KERNEL);
+       match_c = kzalloc(MLX5_ST_SZ_BYTES(fte_match_param), GFP_KERNEL);
+       if (!match_v || !match_c) {
+               pr_warn("FDB: Failed to alloc match parameters\n");
+               goto out;
+       }
+       dmac_v = MLX5_ADDR_OF(fte_match_param, match_v,
+                             outer_headers.dmac_47_16);
+       dmac_c = MLX5_ADDR_OF(fte_match_param, match_c,
+                             outer_headers.dmac_47_16);
+
+       ether_addr_copy(dmac_v, mac);
+       /* Match criteria mask */
+       memset(dmac_c, 0xff, 6);
+
+       dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
+       dest.vport_num = vport;
+
+       esw_debug(esw->dev,
+                 "\tFDB add rule dmac_v(%pM) dmac_c(%pM) -> vport(%d)\n",
+                 dmac_v, dmac_c, vport);
+       flow_rule =
+               mlx5_add_flow_rule(esw,
+                                  match_header,
+                                  match_c,
+                                  match_v,
+                                  MLX5_FLOW_CONTEXT_ACTION_FWD_DEST,
+                                  0, &dest);
+       if (IS_ERR_OR_NULL(flow_rule)) {
+               pr_warn(
+                       "FDB: Failed to add flow rule: dmac_v(%pM) dmac_c(%pM) -> vport(%d), err(%ld)\n",
+                        dmac_v, dmac_c, vport, PTR_ERR(flow_rule));
+               flow_rule = NULL;
+       }
+out:
+       kfree(match_v);
+       kfree(match_c);
+       return flow_rule;
+}
+
+static int esw_create_fdb_table(struct mlx5_eswitch *esw, int nvports)
+{
+       struct mlx5_core_dev *dev = esw->dev;
+       struct mlx5_flow_table_group g;
+       struct mlx5_flow_table *fdb;
+       u8 *dmac;
+
+       esw_debug(dev, "Create FDB log_max_size(%d)\n",
+                 MLX5_CAP_ESW_FLOWTABLE_FDB(dev, log_max_ft_size));
+
+       memset(&g, 0, sizeof(g));
+       /* UC MC Full match rules*/
+       g.log_sz = MLX5_CAP_ESW_FLOWTABLE_FDB(dev, log_max_ft_size);
+       g.match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
+       dmac = MLX5_ADDR_OF(fte_match_param, g.match_criteria,
+                           outer_headers.dmac_47_16);
+       /* Match criteria mask */
+       memset(dmac, 0xff, 6);
+
+       fdb = mlx5_create_flow_table(dev, 0,
+                                    MLX5_FLOW_TABLE_TYPE_ESWITCH,
+                                    1, &g);
+       if (fdb)
+               esw_debug(dev, "ESW: FDB Table created fdb->id %d\n", mlx5_get_flow_table_id(fdb));
+       else
+               esw_warn(dev, "ESW: Failed to create FDB Table\n");
+
+       esw->fdb_table.fdb = fdb;
+       return fdb ? 0 : -ENOMEM;
+}
+
+static void esw_destroy_fdb_table(struct mlx5_eswitch *esw)
+{
+       if (!esw->fdb_table.fdb)
+               return;
+
+       esw_debug(esw->dev, "Destroy FDB Table fdb(%d)\n",
+                 mlx5_get_flow_table_id(esw->fdb_table.fdb));
+       mlx5_destroy_flow_table(esw->fdb_table.fdb);
+       esw->fdb_table.fdb = NULL;
+}
+
+/* E-Switch vport UC/MC lists management */
+typedef int (*vport_addr_action)(struct mlx5_eswitch *esw,
+                                struct vport_addr *vaddr);
+
+static int esw_add_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
+{
+       struct hlist_head *hash = esw->l2_table.l2_hash;
+       struct esw_uc_addr *esw_uc;
+       u8 *mac = vaddr->node.addr;
+       u32 vport = vaddr->vport;
+       int err;
+
+       esw_uc = l2addr_hash_find(hash, mac, struct esw_uc_addr);
+       if (esw_uc) {
                esw_warn(esw->dev,
                         "Failed to set L2 mac(%pM) for vport(%d), mac is already in use by vport(%d)\n",
-                        mac, vport, addr->vport);
+                        mac, vport, esw_uc->vport);
                return -EEXIST;
        }
 
-       addr = l2addr_hash_add(hash, mac, struct mlx5_uc_l2addr,
-                              GFP_KERNEL);
-       if (!addr)
+       esw_uc = l2addr_hash_add(hash, mac, struct esw_uc_addr, GFP_KERNEL);
+       if (!esw_uc)
                return -ENOMEM;
+       esw_uc->vport = vport;
 
-       addr->vport = vport;
-       addr->action = MLX5_ACTION_NONE;
-       err = set_l2_table_entry(esw->dev, mac, 0, 0, &addr->table_index);
+       err = set_l2_table_entry(esw->dev, mac, 0, 0, &esw_uc->table_index);
        if (err)
-               l2addr_hash_del(addr);
-       else
-               esw_debug(esw->dev, "\tADDED L2 MAC: vport[%d] %pM index:%d\n",
-                         vport, addr->node.addr, addr->table_index);
+               goto abort;
+
+       if (esw->fdb_table.fdb) /* SRIOV is enabled: Forward UC MAC to vport */
+               vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
+
+       esw_debug(esw->dev, "\tADDED UC MAC: vport[%d] %pM index:%d fr(%p)\n",
+                 vport, mac, esw_uc->table_index, vaddr->flow_rule);
+       return err;
+abort:
+       l2addr_hash_del(esw_uc);
        return err;
 }
 
-static int l2_table_addr_del(struct mlx5_eswitch *esw,
-                            u8 mac[ETH_ALEN], u32 vport)
+static int esw_del_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
 {
-       struct hlist_head *hash;
-       struct mlx5_uc_l2addr *addr;
+       struct hlist_head *hash = esw->l2_table.l2_hash;
+       struct esw_uc_addr *esw_uc;
+       u8 *mac = vaddr->node.addr;
+       u32 vport = vaddr->vport;
+
+       esw_uc = l2addr_hash_find(hash, mac, struct esw_uc_addr);
+       if (!esw_uc || esw_uc->vport != vport) {
+               esw_debug(esw->dev,
+                         "MAC(%pM) doesn't belong to vport (%d)\n",
+                         mac, vport);
+               return -EINVAL;
+       }
+       esw_debug(esw->dev, "\tDELETE UC MAC: vport[%d] %pM index:%d fr(%p)\n",
+                 vport, mac, esw_uc->table_index, vaddr->flow_rule);
+
+       del_l2_table_entry(esw->dev, esw_uc->table_index);
+
+       if (vaddr->flow_rule)
+               mlx5_del_flow_rule(vaddr->flow_rule, vport);
+       vaddr->flow_rule = NULL;
+
+       l2addr_hash_del(esw_uc);
+       return 0;
+}
+
+static int esw_add_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
+{
+       struct hlist_head *hash = esw->mc_table;
+       struct esw_mc_addr *esw_mc;
+       u8 *mac = vaddr->node.addr;
+       u32 vport = vaddr->vport;
+
+       if (!esw->fdb_table.fdb)
+               return 0;
+
+       esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
+       if (esw_mc)
+               goto add;
+
+       esw_mc = l2addr_hash_add(hash, mac, struct esw_mc_addr, GFP_KERNEL);
+       if (!esw_mc)
+               return -ENOMEM;
+
+       esw_mc->uplink_rule = /* Forward MC MAC to Uplink */
+               esw_fdb_set_vport_rule(esw, mac, UPLINK_VPORT);
+add:
+       esw_mc->refcnt++;
+       /* Forward MC MAC to vport */
+       vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
+       esw_debug(esw->dev,
+                 "\tADDED MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
+                 vport, mac, vaddr->flow_rule,
+                 esw_mc->refcnt, esw_mc->uplink_rule);
+       return 0;
+}
+
+static int esw_del_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
+{
+       struct hlist_head *hash = esw->mc_table;
+       struct esw_mc_addr *esw_mc;
+       u8 *mac = vaddr->node.addr;
+       u32 vport = vaddr->vport;
 
-       hash = esw->l2_table.l2_hash;
-       addr = l2addr_hash_find(hash, mac, struct mlx5_uc_l2addr);
-       if (!addr || addr->vport != vport) {
-               esw_warn(esw->dev, "MAC(%pM) doesn't belong to vport (%d)\n",
+       if (!esw->fdb_table.fdb)
+               return 0;
+
+       esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
+       if (!esw_mc) {
+               esw_warn(esw->dev,
+                        "Failed to find eswitch MC addr for MAC(%pM) vport(%d)",
                         mac, vport);
                return -EINVAL;
        }
+       esw_debug(esw->dev,
+                 "\tDELETE MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
+                 vport, mac, vaddr->flow_rule, esw_mc->refcnt,
+                 esw_mc->uplink_rule);
 
-       esw_debug(esw->dev, "\tDELETE L2 MAC: vport[%d] %pM index:%d\n",
-                 vport, addr->node.addr, addr->table_index);
-       del_l2_table_entry(esw->dev, addr->table_index);
-       l2addr_hash_del(addr);
+       if (vaddr->flow_rule)
+               mlx5_del_flow_rule(vaddr->flow_rule, vport);
+       vaddr->flow_rule = NULL;
+
+       if (--esw_mc->refcnt)
+               return 0;
+
+       if (esw_mc->uplink_rule)
+               mlx5_del_flow_rule(esw_mc->uplink_rule, UPLINK_VPORT);
+
+       l2addr_hash_del(esw_mc);
        return 0;
 }
 
-/* E-Switch vport uc list management */
-
-/* Apply vport uc list to HW l2 table */
-static void esw_apply_vport_uc_list(struct mlx5_eswitch *esw,
-                                   u32 vport_num)
+/* Apply vport UC/MC list to HW l2 table and FDB table */
+static void esw_apply_vport_addr_list(struct mlx5_eswitch *esw,
+                                     u32 vport_num, int list_type)
 {
        struct mlx5_vport *vport = &esw->vports[vport_num];
-       struct mlx5_vport_addr *addr;
+       bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
+       vport_addr_action vport_addr_add;
+       vport_addr_action vport_addr_del;
+       struct vport_addr *addr;
        struct l2addr_node *node;
        struct hlist_head *hash;
        struct hlist_node *tmp;
        int hi;
 
-       hash = vport->uc_list;
+       vport_addr_add = is_uc ? esw_add_uc_addr :
+                                esw_add_mc_addr;
+       vport_addr_del = is_uc ? esw_del_uc_addr :
+                                esw_del_mc_addr;
+
+       hash = is_uc ? vport->uc_list : vport->mc_list;
        for_each_l2hash_node(node, tmp, hash, hi) {
-               addr = container_of(node, struct mlx5_vport_addr, node);
+               addr = container_of(node, struct vport_addr, node);
                switch (addr->action) {
                case MLX5_ACTION_ADD:
-                       l2_table_addr_add(esw, addr->node.addr, vport_num);
+                       vport_addr_add(esw, addr);
                        addr->action = MLX5_ACTION_NONE;
                        break;
                case MLX5_ACTION_DEL:
-                       l2_table_addr_del(esw, addr->node.addr, vport_num);
+                       vport_addr_del(esw, addr);
                        l2addr_hash_del(addr);
                        break;
                }
        }
 }
 
-/* Sync vport uc list from vport context */
-static void esw_update_vport_uc_list(struct mlx5_eswitch *esw,
-                                    u32 vport_num)
+/* Sync vport UC/MC list from vport context */
+static void esw_update_vport_addr_list(struct mlx5_eswitch *esw,
+                                      u32 vport_num, int list_type)
 {
        struct mlx5_vport *vport = &esw->vports[vport_num];
-       struct mlx5_vport_addr *addr;
-       struct l2addr_node *node;
+       bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
        u8 (*mac_list)[ETH_ALEN];
+       struct l2addr_node *node;
+       struct vport_addr *addr;
        struct hlist_head *hash;
        struct hlist_node *tmp;
        int size;
@@ -294,40 +704,41 @@ static void esw_update_vport_uc_list(struct mlx5_eswitch *esw,
        int hi;
        int i;
 
-       size = MLX5_MAX_UC_PER_VPORT(esw->dev);
+       size = is_uc ? MLX5_MAX_UC_PER_VPORT(esw->dev) :
+                      MLX5_MAX_MC_PER_VPORT(esw->dev);
 
        mac_list = kcalloc(size, ETH_ALEN, GFP_KERNEL);
        if (!mac_list)
                return;
 
-       hash = vport->uc_list;
+       hash = is_uc ? vport->uc_list : vport->mc_list;
 
        for_each_l2hash_node(node, tmp, hash, hi) {
-               addr = container_of(node, struct mlx5_vport_addr, node);
+               addr = container_of(node, struct vport_addr, node);
                addr->action = MLX5_ACTION_DEL;
        }
 
-       err = mlx5_query_nic_vport_mac_list(esw->dev, vport_num,
-                                           MLX5_NVPRT_LIST_TYPE_UC,
+       err = mlx5_query_nic_vport_mac_list(esw->dev, vport_num, list_type,
                                            mac_list, &size);
        if (err)
                return;
-       esw_debug(esw->dev, "vport[%d] context update UC list size (%d)\n",
-                 vport_num, size);
+       esw_debug(esw->dev, "vport[%d] context update %s list size (%d)\n",
+                 vport_num, is_uc ? "UC" : "MC", size);
 
        for (i = 0; i < size; i++) {
-               if (!is_valid_ether_addr(mac_list[i]))
+               if (is_uc && !is_valid_ether_addr(mac_list[i]))
+                       continue;
+
+               if (!is_uc && !is_multicast_ether_addr(mac_list[i]))
                        continue;
 
-               addr = l2addr_hash_find(hash, mac_list[i],
-                                       struct mlx5_vport_addr);
+               addr = l2addr_hash_find(hash, mac_list[i], struct vport_addr);
                if (addr) {
                        addr->action = MLX5_ACTION_NONE;
                        continue;
                }
 
-               addr = l2addr_hash_add(hash, mac_list[i],
-                                      struct mlx5_vport_addr,
+               addr = l2addr_hash_add(hash, mac_list[i], struct vport_addr,
                                       GFP_KERNEL);
                if (!addr) {
                        esw_warn(esw->dev,
@@ -335,6 +746,7 @@ static void esw_update_vport_uc_list(struct mlx5_eswitch *esw,
                                 mac_list[i], vport_num);
                        continue;
                }
+               addr->vport = vport_num;
                addr->action = MLX5_ACTION_ADD;
        }
        kfree(mac_list);
@@ -345,32 +757,80 @@ static void esw_vport_change_handler(struct work_struct *work)
        struct mlx5_vport *vport =
                container_of(work, struct mlx5_vport, vport_change_handler);
        struct mlx5_core_dev *dev = vport->dev;
+       struct mlx5_eswitch *esw = dev->priv.eswitch;
        u8 mac[ETH_ALEN];
 
        mlx5_query_nic_vport_mac_address(dev, vport->vport, mac);
+       esw_debug(dev, "vport[%d] Context Changed: perm mac: %pM\n",
+                 vport->vport, mac);
+
+       if (vport->enabled_events & UC_ADDR_CHANGE) {
+               esw_update_vport_addr_list(esw, vport->vport,
+                                          MLX5_NVPRT_LIST_TYPE_UC);
+               esw_apply_vport_addr_list(esw, vport->vport,
+                                         MLX5_NVPRT_LIST_TYPE_UC);
+       }
 
-       if (!is_valid_ether_addr(mac))
-               goto out;
-
-       esw_update_vport_uc_list(dev->priv.eswitch, vport->vport);
-       esw_apply_vport_uc_list(dev->priv.eswitch, vport->vport);
+       if (vport->enabled_events & MC_ADDR_CHANGE) {
+               esw_update_vport_addr_list(esw, vport->vport,
+                                          MLX5_NVPRT_LIST_TYPE_MC);
+               esw_apply_vport_addr_list(esw, vport->vport,
+                                         MLX5_NVPRT_LIST_TYPE_MC);
+       }
 
-out:
+       esw_debug(esw->dev, "vport[%d] Context Changed: Done\n", vport->vport);
        if (vport->enabled)
                arm_vport_context_events_cmd(dev, vport->vport,
-                                            UC_ADDR_CHANGE);
+                                            vport->enabled_events);
 }
 
-static void esw_enable_vport(struct mlx5_eswitch *esw, int vport_num)
+static void esw_enable_vport(struct mlx5_eswitch *esw, int vport_num,
+                            int enable_events)
 {
        struct mlx5_vport *vport = &esw->vports[vport_num];
        unsigned long flags;
 
+       WARN_ON(vport->enabled);
+
+       esw_debug(esw->dev, "Enabling VPORT(%d)\n", vport_num);
+       mlx5_modify_vport_admin_state(esw->dev,
+                                     MLX5_QUERY_VPORT_STATE_IN_OP_MOD_ESW_VPORT,
+                                     vport_num,
+                                     MLX5_ESW_VPORT_ADMIN_STATE_AUTO);
+
+       /* Sync with current vport context */
+       vport->enabled_events = enable_events;
+       esw_vport_change_handler(&vport->vport_change_handler);
+
        spin_lock_irqsave(&vport->lock, flags);
        vport->enabled = true;
        spin_unlock_irqrestore(&vport->lock, flags);
 
-       arm_vport_context_events_cmd(esw->dev, vport_num, UC_ADDR_CHANGE);
+       arm_vport_context_events_cmd(esw->dev, vport_num, enable_events);
+
+       esw->enabled_vports++;
+       esw_debug(esw->dev, "Enabled VPORT(%d)\n", vport_num);
+}
+
+static void esw_cleanup_vport(struct mlx5_eswitch *esw, u16 vport_num)
+{
+       struct mlx5_vport *vport = &esw->vports[vport_num];
+       struct l2addr_node *node;
+       struct vport_addr *addr;
+       struct hlist_node *tmp;
+       int hi;
+
+       for_each_l2hash_node(node, tmp, vport->uc_list, hi) {
+               addr = container_of(node, struct vport_addr, node);
+               addr->action = MLX5_ACTION_DEL;
+       }
+       esw_apply_vport_addr_list(esw, vport_num, MLX5_NVPRT_LIST_TYPE_UC);
+
+       for_each_l2hash_node(node, tmp, vport->mc_list, hi) {
+               addr = container_of(node, struct vport_addr, node);
+               addr->action = MLX5_ACTION_DEL;
+       }
+       esw_apply_vport_addr_list(esw, vport_num, MLX5_NVPRT_LIST_TYPE_MC);
 }
 
 static void esw_disable_vport(struct mlx5_eswitch *esw, int vport_num)
@@ -381,19 +841,82 @@ static void esw_disable_vport(struct mlx5_eswitch *esw, int vport_num)
        if (!vport->enabled)
                return;
 
+       esw_debug(esw->dev, "Disabling vport(%d)\n", vport_num);
        /* Mark this vport as disabled to discard new events */
        spin_lock_irqsave(&vport->lock, flags);
        vport->enabled = false;
+       vport->enabled_events = 0;
        spin_unlock_irqrestore(&vport->lock, flags);
 
+       mlx5_modify_vport_admin_state(esw->dev,
+                                     MLX5_QUERY_VPORT_STATE_IN_OP_MOD_ESW_VPORT,
+                                     vport_num,
+                                     MLX5_ESW_VPORT_ADMIN_STATE_DOWN);
        /* Wait for current already scheduled events to complete */
        flush_workqueue(esw->work_queue);
-
        /* Disable events from this vport */
        arm_vport_context_events_cmd(esw->dev, vport->vport, 0);
+       /* We don't assume VFs will cleanup after themselves */
+       esw_cleanup_vport(esw, vport_num);
+       esw->enabled_vports--;
 }
 
 /* Public E-Switch API */
+int mlx5_eswitch_enable_sriov(struct mlx5_eswitch *esw, int nvfs)
+{
+       int err;
+       int i;
+
+       if (!esw || !MLX5_CAP_GEN(esw->dev, vport_group_manager) ||
+           MLX5_CAP_GEN(esw->dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
+               return 0;
+
+       if (!MLX5_CAP_GEN(esw->dev, eswitch_flow_table) ||
+           !MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, ft_support)) {
+               esw_warn(esw->dev, "E-Switch FDB is not supported, aborting ...\n");
+               return -ENOTSUPP;
+       }
+
+       esw_info(esw->dev, "E-Switch enable SRIOV: nvfs(%d)\n", nvfs);
+
+       esw_disable_vport(esw, 0);
+
+       err = esw_create_fdb_table(esw, nvfs + 1);
+       if (err)
+               goto abort;
+
+       for (i = 0; i <= nvfs; i++)
+               esw_enable_vport(esw, i, SRIOV_VPORT_EVENTS);
+
+       esw_info(esw->dev, "SRIOV enabled: active vports(%d)\n",
+                esw->enabled_vports);
+       return 0;
+
+abort:
+       esw_enable_vport(esw, 0, UC_ADDR_CHANGE);
+       return err;
+}
+
+void mlx5_eswitch_disable_sriov(struct mlx5_eswitch *esw)
+{
+       int i;
+
+       if (!esw || !MLX5_CAP_GEN(esw->dev, vport_group_manager) ||
+           MLX5_CAP_GEN(esw->dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
+               return;
+
+       esw_info(esw->dev, "disable SRIOV: active vports(%d)\n",
+                esw->enabled_vports);
+
+       for (i = 0; i < esw->total_vports; i++)
+               esw_disable_vport(esw, i);
+
+       esw_destroy_fdb_table(esw);
+
+       /* VPORT 0 (PF) must be enabled back with non-sriov configuration */
+       esw_enable_vport(esw, 0, UC_ADDR_CHANGE);
+}
+
 int mlx5_eswitch_init(struct mlx5_core_dev *dev)
 {
        int l2_table_size = 1 << MLX5_CAP_GEN(dev, log_max_l2_table);
@@ -439,7 +962,6 @@ int mlx5_eswitch_init(struct mlx5_core_dev *dev)
                goto abort;
        }
 
-       esw->total_vports = total_vports;
        for (vport_num = 0; vport_num < total_vports; vport_num++) {
                struct mlx5_vport *vport = &esw->vports[vport_num];
 
@@ -450,9 +972,11 @@ int mlx5_eswitch_init(struct mlx5_core_dev *dev)
                spin_lock_init(&vport->lock);
        }
 
-       dev->priv.eswitch = esw;
+       esw->total_vports = total_vports;
+       esw->enabled_vports = 0;
 
-       esw_enable_vport(esw, 0);
+       dev->priv.eswitch = esw;
+       esw_enable_vport(esw, 0, UC_ADDR_CHANGE);
        /* VF Vports will be enabled when SRIOV is enabled */
        return 0;
 abort:
index 0c41f26578243e7ccab6f45d0e2d41d772347498..f222e336f34f215df426f9b702ba2b6a84d092ad 100644 (file)
@@ -86,10 +86,25 @@ struct l2addr_node {
        kfree(ptr);                                         \
 })
 
+struct mlx5_flow_rule {
+       void             *ft;
+       u32              fi;
+       u8               match_criteria_enable;
+       u32              *match_criteria;
+       u32              *match_value;
+       u32              action;
+       u32              flow_tag;
+       bool             valid;
+       atomic_t         refcount;
+       struct mutex     mutex; /* protect flow rule updates */
+       struct list_head dest_list;
+};
+
 struct mlx5_vport {
        struct mlx5_core_dev    *dev;
        int                     vport;
        struct hlist_head       uc_list[MLX5_L2_ADDR_HASH_SIZE];
+       struct hlist_head       mc_list[MLX5_L2_ADDR_HASH_SIZE];
        struct work_struct      vport_change_handler;
 
        /* This spinlock protects access to vport data, between
@@ -98,6 +113,7 @@ struct mlx5_vport {
         */
        spinlock_t              lock; /* vport events sync */
        bool                    enabled;
+       u16                     enabled_events;
 };
 
 struct mlx5_l2_table {
@@ -106,17 +122,26 @@ struct mlx5_l2_table {
        unsigned long        *bitmap;
 };
 
+struct mlx5_eswitch_fdb {
+       void *fdb;
+};
+
 struct mlx5_eswitch {
        struct mlx5_core_dev    *dev;
        struct mlx5_l2_table    l2_table;
+       struct mlx5_eswitch_fdb fdb_table;
+       struct hlist_head       mc_table[MLX5_L2_ADDR_HASH_SIZE];
        struct workqueue_struct *work_queue;
        struct mlx5_vport       *vports;
        int                     total_vports;
+       int                     enabled_vports;
 };
 
 /* E-Switch API */
 int mlx5_eswitch_init(struct mlx5_core_dev *dev);
 void mlx5_eswitch_cleanup(struct mlx5_eswitch *esw);
 void mlx5_eswitch_vport_event(struct mlx5_eswitch *esw, struct mlx5_eqe *eqe);
+int mlx5_eswitch_enable_sriov(struct mlx5_eswitch *esw, int nvfs);
+void mlx5_eswitch_disable_sriov(struct mlx5_eswitch *esw);
 
 #endif /* __MLX5_ESWITCH_H__ */
index 1649d5cf9e29a0e16df259f313efefe1ae691a48..bee7da822dfe208eb9a9dd8095d76bf48de6962f 100644 (file)
@@ -36,6 +36,7 @@
 #include <linux/types.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
+#include <linux/if_link.h>
 
 #define DRIVER_NAME "mlx5_core"
 #define DRIVER_VERSION "3.0-1"
index 19a43240e35920d952f54c9f89ff7dac5d93c79b..7b24386794f994c3bf3a9aed60182f306b92ffc2 100644 (file)
@@ -33,6 +33,9 @@
 #include <linux/pci.h>
 #include <linux/mlx5/driver.h>
 #include "mlx5_core.h"
+#ifdef CONFIG_MLX5_CORE_EN
+#include "eswitch.h"
+#endif
 
 static void enable_vfs(struct mlx5_core_dev *dev, int num_vfs)
 {
@@ -144,13 +147,15 @@ int mlx5_core_sriov_configure(struct pci_dev *pdev, int num_vfs)
        mlx5_core_cleanup_vfs(dev);
 
        if (!num_vfs) {
+#ifdef CONFIG_MLX5_CORE_EN
+               mlx5_eswitch_disable_sriov(dev->priv.eswitch);
+#endif
                kfree(sriov->vfs_ctx);
                sriov->vfs_ctx = NULL;
                if (!pci_vfs_assigned(pdev))
                        pci_disable_sriov(pdev);
                else
                        pr_info("unloading PF driver while leaving orphan VFs\n");
-
                return 0;
        }
 
@@ -161,6 +166,9 @@ int mlx5_core_sriov_configure(struct pci_dev *pdev, int num_vfs)
        }
 
        mlx5_core_init_vfs(dev, num_vfs);
+#ifdef CONFIG_MLX5_CORE_EN
+       mlx5_eswitch_enable_sriov(dev->priv.eswitch, num_vfs);
+#endif
 
        return num_vfs;
 }
@@ -199,6 +207,10 @@ int mlx5_sriov_init(struct mlx5_core_dev *dev)
        sriov->enabled_vfs = cur_vfs;
 
        mlx5_core_init_vfs(dev, cur_vfs);
+#ifdef CONFIG_MLX5_CORE_EN
+       if (cur_vfs)
+               mlx5_eswitch_enable_sriov(dev->priv.eswitch, cur_vfs);
+#endif
 
        enable_vfs(dev, cur_vfs);
 
index bce9caed1eed8046dea2904553a4287f15974547..88eb4490a8b327f5e296f4d3eb351ad693485406 100644 (file)
@@ -1074,6 +1074,12 @@ enum {
        VPORT_STATE_UP                  = 0x1,
 };
 
+enum {
+       MLX5_ESW_VPORT_ADMIN_STATE_DOWN  = 0x0,
+       MLX5_ESW_VPORT_ADMIN_STATE_UP    = 0x1,
+       MLX5_ESW_VPORT_ADMIN_STATE_AUTO  = 0x2,
+};
+
 enum {
        MLX5_L3_PROT_TYPE_IPV4          = 0,
        MLX5_L3_PROT_TYPE_IPV6          = 1,
index 5f922c6d4fc237d8a2ec6f62336cba4d50eaf96d..0f2a15cf33173de6ca17173c25a1af9e949e19ed 100644 (file)
@@ -41,6 +41,15 @@ struct mlx5_flow_table_group {
        u32     match_criteria[MLX5_ST_SZ_DW(fte_match_param)];
 };
 
+struct mlx5_flow_destination {
+       enum mlx5_flow_destination_type type;
+       union {
+               u32                     tir_num;
+               void                    *ft;
+               u32                     vport_num;
+       };
+};
+
 void *mlx5_create_flow_table(struct mlx5_core_dev *dev, u8 level, u8 table_type,
                             u16 num_groups,
                             struct mlx5_flow_table_group *group);
index ae7c08adba4a0f981dbb9b8389e8f9e8b403e0fd..a81b008738fd270fc68c022eea877ddd067146e4 100644 (file)
@@ -827,9 +827,10 @@ struct mlx5_ifc_cmd_hca_cap_bits {
        u8         reserved_69[0x220];
 };
 
-enum {
-       MLX5_DEST_FORMAT_STRUCT_DESTINATION_TYPE_FLOW_TABLE_  = 0x1,
-       MLX5_DEST_FORMAT_STRUCT_DESTINATION_TYPE_TIR          = 0x2,
+enum mlx5_flow_destination_type {
+       MLX5_FLOW_DESTINATION_TYPE_VPORT        = 0x0,
+       MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE   = 0x1,
+       MLX5_FLOW_DESTINATION_TYPE_TIR          = 0x2,
 };
 
 struct mlx5_ifc_dest_format_struct_bits {