summaryrefslogtreecommitdiffstats
path: root/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
diff options
context:
space:
mode:
authorMaxime Ripard <maxime@cerno.tech>2021-05-11 13:35:52 +0200
committerMaxime Ripard <maxime@cerno.tech>2021-05-11 13:35:52 +0200
commitc55b44c9386f3ee1b08752638559f19deaf6040d (patch)
treec843a21f45180387fcd9eb2625cc9d1f166a3156 /drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
parentMAINTAINERS: Update my e-mail (diff)
parentLinux 5.13-rc1 (diff)
downloadlinux-c55b44c9386f3ee1b08752638559f19deaf6040d.tar.xz
linux-c55b44c9386f3ee1b08752638559f19deaf6040d.zip
Merge drm/drm-fixes into drm-misc-fixes
Start this new release drm-misc-fixes branch Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Diffstat (limited to 'drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c')
-rw-r--r--drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c682
1 files changed, 596 insertions, 86 deletions
diff --git a/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c b/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
index eda99d82766a..41259c0004d1 100644
--- a/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
+++ b/drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
@@ -113,6 +113,10 @@ struct mlxsw_sp_rif_ops {
void (*fdb_del)(struct mlxsw_sp_rif *rif, const char *mac);
};
+struct mlxsw_sp_router_ops {
+ int (*init)(struct mlxsw_sp *mlxsw_sp);
+};
+
static struct mlxsw_sp_rif *
mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
const struct net_device *dev);
@@ -2662,6 +2666,10 @@ static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
goto out;
}
+ if (neigh_entry->connected && entry_connected &&
+ !memcmp(neigh_entry->ha, ha, ETH_ALEN))
+ goto out;
+
memcpy(neigh_entry->ha, ha, ETH_ALEN);
mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected,
@@ -2842,6 +2850,15 @@ enum mlxsw_sp_nexthop_type {
MLXSW_SP_NEXTHOP_TYPE_IPIP,
};
+enum mlxsw_sp_nexthop_action {
+ /* Nexthop forwards packets to an egress RIF */
+ MLXSW_SP_NEXTHOP_ACTION_FORWARD,
+ /* Nexthop discards packets */
+ MLXSW_SP_NEXTHOP_ACTION_DISCARD,
+ /* Nexthop traps packets */
+ MLXSW_SP_NEXTHOP_ACTION_TRAP,
+};
+
struct mlxsw_sp_nexthop_key {
struct fib_nh *fib_nh;
};
@@ -2862,16 +2879,16 @@ struct mlxsw_sp_nexthop {
int norm_nh_weight;
int num_adj_entries;
struct mlxsw_sp_rif *rif;
- u8 should_offload:1, /* set indicates this neigh is connected and
- * should be put to KVD linear area of this group.
+ u8 should_offload:1, /* set indicates this nexthop should be written
+ * to the adjacency table.
*/
- offloaded:1, /* set in case the neigh is actually put into
- * KVD linear area of this group.
+ offloaded:1, /* set indicates this nexthop was written to the
+ * adjacency table.
*/
- update:1, /* set indicates that MAC of this neigh should be
- * updated in HW
+ update:1; /* set indicates this nexthop should be updated in the
+ * adjacency table (f.e., its MAC changed).
*/
- discard:1; /* nexthop is programmed to discard packets */
+ enum mlxsw_sp_nexthop_action action;
enum mlxsw_sp_nexthop_type type;
union {
struct mlxsw_sp_neigh_entry *neigh_entry;
@@ -2894,7 +2911,9 @@ struct mlxsw_sp_nexthop_group_info {
u16 count;
int sum_norm_weight;
u8 adj_index_valid:1,
- gateway:1; /* routes using the group use a gateway */
+ gateway:1, /* routes using the group use a gateway */
+ is_resilient:1;
+ struct list_head list; /* member in nh_res_grp_list */
struct mlxsw_sp_nexthop nexthops[0];
#define nh_rif nexthops[0].rif
};
@@ -2979,14 +2998,15 @@ struct mlxsw_sp_nexthop *mlxsw_sp_nexthop_next(struct mlxsw_sp_router *router,
return list_next_entry(nh, router_list_node);
}
-bool mlxsw_sp_nexthop_offload(struct mlxsw_sp_nexthop *nh)
+bool mlxsw_sp_nexthop_is_forward(const struct mlxsw_sp_nexthop *nh)
{
- return nh->offloaded;
+ return nh->offloaded && nh->action == MLXSW_SP_NEXTHOP_ACTION_FORWARD;
}
unsigned char *mlxsw_sp_nexthop_ha(struct mlxsw_sp_nexthop *nh)
{
- if (!nh->offloaded)
+ if (nh->type != MLXSW_SP_NEXTHOP_TYPE_ETH ||
+ !mlxsw_sp_nexthop_is_forward(nh))
return NULL;
return nh->neigh_entry->ha;
}
@@ -3036,11 +3056,6 @@ bool mlxsw_sp_nexthop_group_has_ipip(struct mlxsw_sp_nexthop *nh)
return false;
}
-bool mlxsw_sp_nexthop_is_discard(const struct mlxsw_sp_nexthop *nh)
-{
- return nh->discard;
-}
-
static const struct rhashtable_params mlxsw_sp_nexthop_group_vr_ht_params = {
.key_offset = offsetof(struct mlxsw_sp_nexthop_group_vr_entry, key),
.head_offset = offsetof(struct mlxsw_sp_nexthop_group_vr_entry, ht_node),
@@ -3403,20 +3418,38 @@ err_mass_update_vr:
return err;
}
-static int __mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
- struct mlxsw_sp_nexthop *nh)
+static int __mlxsw_sp_nexthop_eth_update(struct mlxsw_sp *mlxsw_sp,
+ u32 adj_index,
+ struct mlxsw_sp_nexthop *nh,
+ bool force, char *ratr_pl)
{
struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
- char ratr_pl[MLXSW_REG_RATR_LEN];
+ enum mlxsw_reg_ratr_op op;
+ u16 rif_index;
- mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
- true, MLXSW_REG_RATR_TYPE_ETHERNET,
- adj_index, nh->rif->rif_index);
- if (nh->discard)
+ rif_index = nh->rif ? nh->rif->rif_index :
+ mlxsw_sp->router->lb_rif_index;
+ op = force ? MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY :
+ MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY_ON_ACTIVITY;
+ mlxsw_reg_ratr_pack(ratr_pl, op, true, MLXSW_REG_RATR_TYPE_ETHERNET,
+ adj_index, rif_index);
+ switch (nh->action) {
+ case MLXSW_SP_NEXTHOP_ACTION_FORWARD:
+ mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
+ break;
+ case MLXSW_SP_NEXTHOP_ACTION_DISCARD:
mlxsw_reg_ratr_trap_action_set(ratr_pl,
MLXSW_REG_RATR_TRAP_ACTION_DISCARD_ERRORS);
- else
- mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
+ break;
+ case MLXSW_SP_NEXTHOP_ACTION_TRAP:
+ mlxsw_reg_ratr_trap_action_set(ratr_pl,
+ MLXSW_REG_RATR_TRAP_ACTION_TRAP);
+ mlxsw_reg_ratr_trap_id_set(ratr_pl, MLXSW_TRAP_ID_RTR_EGRESS0);
+ break;
+ default:
+ WARN_ON_ONCE(1);
+ return -EINVAL;
+ }
if (nh->counter_valid)
mlxsw_reg_ratr_counter_pack(ratr_pl, nh->counter_index, true);
else
@@ -3425,15 +3458,17 @@ static int __mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
}
-int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
- struct mlxsw_sp_nexthop *nh)
+int mlxsw_sp_nexthop_eth_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
+ struct mlxsw_sp_nexthop *nh, bool force,
+ char *ratr_pl)
{
int i;
for (i = 0; i < nh->num_adj_entries; i++) {
int err;
- err = __mlxsw_sp_nexthop_update(mlxsw_sp, adj_index + i, nh);
+ err = __mlxsw_sp_nexthop_eth_update(mlxsw_sp, adj_index + i,
+ nh, force, ratr_pl);
if (err)
return err;
}
@@ -3443,17 +3478,20 @@ int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
static int __mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
u32 adj_index,
- struct mlxsw_sp_nexthop *nh)
+ struct mlxsw_sp_nexthop *nh,
+ bool force, char *ratr_pl)
{
const struct mlxsw_sp_ipip_ops *ipip_ops;
ipip_ops = mlxsw_sp->router->ipip_ops_arr[nh->ipip_entry->ipipt];
- return ipip_ops->nexthop_update(mlxsw_sp, adj_index, nh->ipip_entry);
+ return ipip_ops->nexthop_update(mlxsw_sp, adj_index, nh->ipip_entry,
+ force, ratr_pl);
}
static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
u32 adj_index,
- struct mlxsw_sp_nexthop *nh)
+ struct mlxsw_sp_nexthop *nh, bool force,
+ char *ratr_pl)
{
int i;
@@ -3461,7 +3499,7 @@ static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
int err;
err = __mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index + i,
- nh);
+ nh, force, ratr_pl);
if (err)
return err;
}
@@ -3469,11 +3507,29 @@ static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
return 0;
}
+static int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
+ struct mlxsw_sp_nexthop *nh, bool force,
+ char *ratr_pl)
+{
+ /* When action is discard or trap, the nexthop must be
+ * programmed as an Ethernet nexthop.
+ */
+ if (nh->type == MLXSW_SP_NEXTHOP_TYPE_ETH ||
+ nh->action == MLXSW_SP_NEXTHOP_ACTION_DISCARD ||
+ nh->action == MLXSW_SP_NEXTHOP_ACTION_TRAP)
+ return mlxsw_sp_nexthop_eth_update(mlxsw_sp, adj_index, nh,
+ force, ratr_pl);
+ else
+ return mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index, nh,
+ force, ratr_pl);
+}
+
static int
mlxsw_sp_nexthop_group_update(struct mlxsw_sp *mlxsw_sp,
struct mlxsw_sp_nexthop_group_info *nhgi,
bool reallocate)
{
+ char ratr_pl[MLXSW_REG_RATR_LEN];
u32 adj_index = nhgi->adj_index; /* base */
struct mlxsw_sp_nexthop *nh;
int i;
@@ -3489,16 +3545,8 @@ mlxsw_sp_nexthop_group_update(struct mlxsw_sp *mlxsw_sp,
if (nh->update || reallocate) {
int err = 0;
- switch (nh->type) {
- case MLXSW_SP_NEXTHOP_TYPE_ETH:
- err = mlxsw_sp_nexthop_update
- (mlxsw_sp, adj_index, nh);
- break;
- case MLXSW_SP_NEXTHOP_TYPE_IPIP:
- err = mlxsw_sp_nexthop_ipip_update
- (mlxsw_sp, adj_index, nh);
- break;
- }
+ err = mlxsw_sp_nexthop_update(mlxsw_sp, adj_index, nh,
+ true, ratr_pl);
if (err)
return err;
nh->update = 0;
@@ -3524,34 +3572,69 @@ mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
return 0;
}
-static void mlxsw_sp_adj_grp_size_round_up(u16 *p_adj_grp_size)
+struct mlxsw_sp_adj_grp_size_range {
+ u16 start; /* Inclusive */
+ u16 end; /* Inclusive */
+};
+
+/* Ordered by range start value */
+static const struct mlxsw_sp_adj_grp_size_range
+mlxsw_sp1_adj_grp_size_ranges[] = {
+ { .start = 1, .end = 64 },
+ { .start = 512, .end = 512 },
+ { .start = 1024, .end = 1024 },
+ { .start = 2048, .end = 2048 },
+ { .start = 4096, .end = 4096 },
+};
+
+/* Ordered by range start value */
+static const struct mlxsw_sp_adj_grp_size_range
+mlxsw_sp2_adj_grp_size_ranges[] = {
+ { .start = 1, .end = 128 },
+ { .start = 256, .end = 256 },
+ { .start = 512, .end = 512 },
+ { .start = 1024, .end = 1024 },
+ { .start = 2048, .end = 2048 },
+ { .start = 4096, .end = 4096 },
+};
+
+static void mlxsw_sp_adj_grp_size_round_up(const struct mlxsw_sp *mlxsw_sp,
+ u16 *p_adj_grp_size)
{
- /* Valid sizes for an adjacency group are:
- * 1-64, 512, 1024, 2048 and 4096.
- */
- if (*p_adj_grp_size <= 64)
- return;
- else if (*p_adj_grp_size <= 512)
- *p_adj_grp_size = 512;
- else if (*p_adj_grp_size <= 1024)
- *p_adj_grp_size = 1024;
- else if (*p_adj_grp_size <= 2048)
- *p_adj_grp_size = 2048;
- else
- *p_adj_grp_size = 4096;
+ int i;
+
+ for (i = 0; i < mlxsw_sp->router->adj_grp_size_ranges_count; i++) {
+ const struct mlxsw_sp_adj_grp_size_range *size_range;
+
+ size_range = &mlxsw_sp->router->adj_grp_size_ranges[i];
+
+ if (*p_adj_grp_size >= size_range->start &&
+ *p_adj_grp_size <= size_range->end)
+ return;
+
+ if (*p_adj_grp_size <= size_range->end) {
+ *p_adj_grp_size = size_range->end;
+ return;
+ }
+ }
}
-static void mlxsw_sp_adj_grp_size_round_down(u16 *p_adj_grp_size,
+static void mlxsw_sp_adj_grp_size_round_down(const struct mlxsw_sp *mlxsw_sp,
+ u16 *p_adj_grp_size,
unsigned int alloc_size)
{
- if (alloc_size >= 4096)
- *p_adj_grp_size = 4096;
- else if (alloc_size >= 2048)
- *p_adj_grp_size = 2048;
- else if (alloc_size >= 1024)
- *p_adj_grp_size = 1024;
- else if (alloc_size >= 512)
- *p_adj_grp_size = 512;
+ int i;
+
+ for (i = mlxsw_sp->router->adj_grp_size_ranges_count - 1; i >= 0; i--) {
+ const struct mlxsw_sp_adj_grp_size_range *size_range;
+
+ size_range = &mlxsw_sp->router->adj_grp_size_ranges[i];
+
+ if (alloc_size >= size_range->end) {
+ *p_adj_grp_size = size_range->end;
+ return;
+ }
+ }
}
static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
@@ -3563,7 +3646,7 @@ static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
/* Round up the requested group size to the next size supported
* by the device and make sure the request can be satisfied.
*/
- mlxsw_sp_adj_grp_size_round_up(p_adj_grp_size);
+ mlxsw_sp_adj_grp_size_round_up(mlxsw_sp, p_adj_grp_size);
err = mlxsw_sp_kvdl_alloc_count_query(mlxsw_sp,
MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
*p_adj_grp_size, &alloc_size);
@@ -3573,7 +3656,7 @@ static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
* entries than requested. Try to use as much of them as
* possible.
*/
- mlxsw_sp_adj_grp_size_round_down(p_adj_grp_size, alloc_size);
+ mlxsw_sp_adj_grp_size_round_down(mlxsw_sp, p_adj_grp_size, alloc_size);
return 0;
}
@@ -3681,9 +3764,29 @@ mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
}
static void
+mlxsw_sp_nexthop_bucket_offload_refresh(struct mlxsw_sp *mlxsw_sp,
+ const struct mlxsw_sp_nexthop *nh,
+ u16 bucket_index)
+{
+ struct mlxsw_sp_nexthop_group *nh_grp = nh->nhgi->nh_grp;
+ bool offload = false, trap = false;
+
+ if (nh->offloaded) {
+ if (nh->action == MLXSW_SP_NEXTHOP_ACTION_TRAP)
+ trap = true;
+ else
+ offload = true;
+ }
+ nexthop_bucket_set_hw_flags(mlxsw_sp_net(mlxsw_sp), nh_grp->obj.id,
+ bucket_index, offload, trap);
+}
+
+static void
mlxsw_sp_nexthop_obj_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
struct mlxsw_sp_nexthop_group *nh_grp)
{
+ int i;
+
/* Do not update the flags if the nexthop group is being destroyed
* since:
* 1. The nexthop objects is being deleted, in which case the flags are
@@ -3697,6 +3800,18 @@ mlxsw_sp_nexthop_obj_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
nexthop_set_hw_flags(mlxsw_sp_net(mlxsw_sp), nh_grp->obj.id,
nh_grp->nhgi->adj_index_valid, false);
+
+ /* Update flags of individual nexthop buckets in case of a resilient
+ * nexthop group.
+ */
+ if (!nh_grp->nhgi->is_resilient)
+ return;
+
+ for (i = 0; i < nh_grp->nhgi->count; i++) {
+ struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
+
+ mlxsw_sp_nexthop_bucket_offload_refresh(mlxsw_sp, nh, i);
+ }
}
static void
@@ -3750,6 +3865,10 @@ mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
goto set_trap;
}
+ /* Flags of individual nexthop buckets might need to be
+ * updated.
+ */
+ mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
return 0;
}
mlxsw_sp_nexthop_group_normalize(nhgi);
@@ -3832,10 +3951,15 @@ set_trap:
static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
bool removing)
{
- if (!removing)
+ if (!removing) {
+ nh->action = MLXSW_SP_NEXTHOP_ACTION_FORWARD;
nh->should_offload = 1;
- else
+ } else if (nh->nhgi->is_resilient) {
+ nh->action = MLXSW_SP_NEXTHOP_ACTION_TRAP;
+ nh->should_offload = 1;
+ } else {
nh->should_offload = 0;
+ }
nh->update = 1;
}
@@ -4250,6 +4374,85 @@ static void mlxsw_sp_nexthop_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
}
}
+static void
+mlxsw_sp_nh_grp_activity_get(struct mlxsw_sp *mlxsw_sp,
+ const struct mlxsw_sp_nexthop_group *nh_grp,
+ unsigned long *activity)
+{
+ char *ratrad_pl;
+ int i, err;
+
+ ratrad_pl = kmalloc(MLXSW_REG_RATRAD_LEN, GFP_KERNEL);
+ if (!ratrad_pl)
+ return;
+
+ mlxsw_reg_ratrad_pack(ratrad_pl, nh_grp->nhgi->adj_index,
+ nh_grp->nhgi->count);
+ err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ratrad), ratrad_pl);
+ if (err)
+ goto out;
+
+ for (i = 0; i < nh_grp->nhgi->count; i++) {
+ if (!mlxsw_reg_ratrad_activity_vector_get(ratrad_pl, i))
+ continue;
+ bitmap_set(activity, i, 1);
+ }
+
+out:
+ kfree(ratrad_pl);
+}
+
+#define MLXSW_SP_NH_GRP_ACTIVITY_UPDATE_INTERVAL 1000 /* ms */
+
+static void
+mlxsw_sp_nh_grp_activity_update(struct mlxsw_sp *mlxsw_sp,
+ const struct mlxsw_sp_nexthop_group *nh_grp)
+{
+ unsigned long *activity;
+
+ activity = bitmap_zalloc(nh_grp->nhgi->count, GFP_KERNEL);
+ if (!activity)
+ return;
+
+ mlxsw_sp_nh_grp_activity_get(mlxsw_sp, nh_grp, activity);
+ nexthop_res_grp_activity_update(mlxsw_sp_net(mlxsw_sp), nh_grp->obj.id,
+ nh_grp->nhgi->count, activity);
+
+ bitmap_free(activity);
+}
+
+static void
+mlxsw_sp_nh_grp_activity_work_schedule(struct mlxsw_sp *mlxsw_sp)
+{
+ unsigned int interval = MLXSW_SP_NH_GRP_ACTIVITY_UPDATE_INTERVAL;
+
+ mlxsw_core_schedule_dw(&mlxsw_sp->router->nh_grp_activity_dw,
+ msecs_to_jiffies(interval));
+}
+
+static void mlxsw_sp_nh_grp_activity_work(struct work_struct *work)
+{
+ struct mlxsw_sp_nexthop_group_info *nhgi;
+ struct mlxsw_sp_router *router;
+ bool reschedule = false;
+
+ router = container_of(work, struct mlxsw_sp_router,
+ nh_grp_activity_dw.work);
+
+ mutex_lock(&router->lock);
+
+ list_for_each_entry(nhgi, &router->nh_res_grp_list, list) {
+ mlxsw_sp_nh_grp_activity_update(router->mlxsw_sp, nhgi->nh_grp);
+ reschedule = true;
+ }
+
+ mutex_unlock(&router->lock);
+
+ if (!reschedule)
+ return;
+ mlxsw_sp_nh_grp_activity_work_schedule(router->mlxsw_sp);
+}
+
static int
mlxsw_sp_nexthop_obj_single_validate(struct mlxsw_sp *mlxsw_sp,
const struct nh_notifier_single_info *nh,
@@ -4268,6 +4471,29 @@ mlxsw_sp_nexthop_obj_single_validate(struct mlxsw_sp *mlxsw_sp,
}
static int
+mlxsw_sp_nexthop_obj_group_entry_validate(struct mlxsw_sp *mlxsw_sp,
+ const struct nh_notifier_single_info *nh,
+ struct netlink_ext_ack *extack)
+{
+ int err;
+
+ err = mlxsw_sp_nexthop_obj_single_validate(mlxsw_sp, nh, extack);
+ if (err)
+ return err;
+
+ /* Device only nexthops with an IPIP device are programmed as
+ * encapsulating adjacency entries.
+ */
+ if (!nh->gw_family && !nh->is_reject &&
+ !mlxsw_sp_netdev_ipip_type(mlxsw_sp, nh->dev, NULL)) {
+ NL_SET_ERR_MSG_MOD(extack, "Nexthop group entry does not have a gateway");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int
mlxsw_sp_nexthop_obj_group_validate(struct mlxsw_sp *mlxsw_sp,
const struct nh_notifier_grp_info *nh_grp,
struct netlink_ext_ack *extack)
@@ -4284,21 +4510,83 @@ mlxsw_sp_nexthop_obj_group_validate(struct mlxsw_sp *mlxsw_sp,
int err;
nh = &nh_grp->nh_entries[i].nh;
- err = mlxsw_sp_nexthop_obj_single_validate(mlxsw_sp, nh,
- extack);
+ err = mlxsw_sp_nexthop_obj_group_entry_validate(mlxsw_sp, nh,
+ extack);
if (err)
return err;
+ }
- /* Device only nexthops with an IPIP device are programmed as
- * encapsulating adjacency entries.
- */
- if (!nh->gw_family && !nh->is_reject &&
- !mlxsw_sp_netdev_ipip_type(mlxsw_sp, nh->dev, NULL)) {
- NL_SET_ERR_MSG_MOD(extack, "Nexthop group entry does not have a gateway");
- return -EINVAL;
+ return 0;
+}
+
+static int
+mlxsw_sp_nexthop_obj_res_group_size_validate(struct mlxsw_sp *mlxsw_sp,
+ const struct nh_notifier_res_table_info *nh_res_table,
+ struct netlink_ext_ack *extack)
+{
+ unsigned int alloc_size;
+ bool valid_size = false;
+ int err, i;
+
+ if (nh_res_table->num_nh_buckets < 32) {
+ NL_SET_ERR_MSG_MOD(extack, "Minimum number of buckets is 32");
+ return -EINVAL;
+ }
+
+ for (i = 0; i < mlxsw_sp->router->adj_grp_size_ranges_count; i++) {
+ const struct mlxsw_sp_adj_grp_size_range *size_range;
+
+ size_range = &mlxsw_sp->router->adj_grp_size_ranges[i];
+
+ if (nh_res_table->num_nh_buckets >= size_range->start &&
+ nh_res_table->num_nh_buckets <= size_range->end) {
+ valid_size = true;
+ break;
}
}
+ if (!valid_size) {
+ NL_SET_ERR_MSG_MOD(extack, "Invalid number of buckets");
+ return -EINVAL;
+ }
+
+ err = mlxsw_sp_kvdl_alloc_count_query(mlxsw_sp,
+ MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
+ nh_res_table->num_nh_buckets,
+ &alloc_size);
+ if (err || nh_res_table->num_nh_buckets != alloc_size) {
+ NL_SET_ERR_MSG_MOD(extack, "Number of buckets does not fit allocation size of any KVDL partition");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int
+mlxsw_sp_nexthop_obj_res_group_validate(struct mlxsw_sp *mlxsw_sp,
+ const struct nh_notifier_res_table_info *nh_res_table,
+ struct netlink_ext_ack *extack)
+{
+ int err;
+ u16 i;
+
+ err = mlxsw_sp_nexthop_obj_res_group_size_validate(mlxsw_sp,
+ nh_res_table,
+ extack);
+ if (err)
+ return err;
+
+ for (i = 0; i < nh_res_table->num_nh_buckets; i++) {
+ const struct nh_notifier_single_info *nh;
+ int err;
+
+ nh = &nh_res_table->nhs[i];
+ err = mlxsw_sp_nexthop_obj_group_entry_validate(mlxsw_sp, nh,
+ extack);
+ if (err)
+ return err;
+ }
+
return 0;
}
@@ -4306,7 +4594,11 @@ static int mlxsw_sp_nexthop_obj_validate(struct mlxsw_sp *mlxsw_sp,
unsigned long event,
struct nh_notifier_info *info)
{
- if (event != NEXTHOP_EVENT_REPLACE)
+ struct nh_notifier_single_info *nh;
+
+ if (event != NEXTHOP_EVENT_REPLACE &&
+ event != NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE &&
+ event != NEXTHOP_EVENT_BUCKET_REPLACE)
return 0;
switch (info->type) {
@@ -4317,6 +4609,14 @@ static int mlxsw_sp_nexthop_obj_validate(struct mlxsw_sp *mlxsw_sp,
return mlxsw_sp_nexthop_obj_group_validate(mlxsw_sp,
info->nh_grp,
info->extack);
+ case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
+ return mlxsw_sp_nexthop_obj_res_group_validate(mlxsw_sp,
+ info->nh_res_table,
+ info->extack);
+ case NH_NOTIFIER_INFO_TYPE_RES_BUCKET:
+ nh = &info->nh_res_bucket->new_nh;
+ return mlxsw_sp_nexthop_obj_group_entry_validate(mlxsw_sp, nh,
+ info->extack);
default:
NL_SET_ERR_MSG_MOD(info->extack, "Unsupported nexthop type");
return -EOPNOTSUPP;
@@ -4334,6 +4634,7 @@ static bool mlxsw_sp_nexthop_obj_is_gateway(struct mlxsw_sp *mlxsw_sp,
return info->nh->gw_family || info->nh->is_reject ||
mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL);
case NH_NOTIFIER_INFO_TYPE_GRP:
+ case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
/* Already validated earlier. */
return true;
default:
@@ -4346,7 +4647,7 @@ static void mlxsw_sp_nexthop_obj_blackhole_init(struct mlxsw_sp *mlxsw_sp,
{
u16 lb_rif_index = mlxsw_sp->router->lb_rif_index;
- nh->discard = 1;
+ nh->action = MLXSW_SP_NEXTHOP_ACTION_DISCARD;
nh->should_offload = 1;
/* While nexthops that discard packets do not forward packets
* via an egress RIF, they still need to be programmed using a
@@ -4398,6 +4699,15 @@ mlxsw_sp_nexthop_obj_init(struct mlxsw_sp *mlxsw_sp,
if (nh_obj->is_reject)
mlxsw_sp_nexthop_obj_blackhole_init(mlxsw_sp, nh);
+ /* In a resilient nexthop group, all the nexthops must be written to
+ * the adjacency table. Even if they do not have a valid neighbour or
+ * RIF.
+ */
+ if (nh_grp->nhgi->is_resilient && !nh->should_offload) {
+ nh->action = MLXSW_SP_NEXTHOP_ACTION_TRAP;
+ nh->should_offload = 1;
+ }
+
return 0;
err_type_init:
@@ -4409,11 +4719,12 @@ err_type_init:
static void mlxsw_sp_nexthop_obj_fini(struct mlxsw_sp *mlxsw_sp,
struct mlxsw_sp_nexthop *nh)
{
- if (nh->discard)
+ if (nh->action == MLXSW_SP_NEXTHOP_ACTION_DISCARD)
mlxsw_sp_nexthop_obj_blackhole_fini(mlxsw_sp, nh);
mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
list_del(&nh->router_list_node);
mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
+ nh->should_offload = 0;
}
static int
@@ -4423,6 +4734,7 @@ mlxsw_sp_nexthop_obj_group_info_init(struct mlxsw_sp *mlxsw_sp,
{
struct mlxsw_sp_nexthop_group_info *nhgi;
struct mlxsw_sp_nexthop *nh;
+ bool is_resilient = false;
unsigned int nhs;
int err, i;
@@ -4433,6 +4745,10 @@ mlxsw_sp_nexthop_obj_group_info_init(struct mlxsw_sp *mlxsw_sp,
case NH_NOTIFIER_INFO_TYPE_GRP:
nhs = info->nh_grp->num_nh;
break;
+ case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
+ nhs = info->nh_res_table->num_nh_buckets;
+ is_resilient = true;
+ break;
default:
return -EINVAL;
}
@@ -4443,6 +4759,7 @@ mlxsw_sp_nexthop_obj_group_info_init(struct mlxsw_sp *mlxsw_sp,
nh_grp->nhgi = nhgi;
nhgi->nh_grp = nh_grp;
nhgi->gateway = mlxsw_sp_nexthop_obj_is_gateway(mlxsw_sp, info);
+ nhgi->is_resilient = is_resilient;
nhgi->count = nhs;
for (i = 0; i < nhgi->count; i++) {
struct nh_notifier_single_info *nh_obj;
@@ -4458,6 +4775,10 @@ mlxsw_sp_nexthop_obj_group_info_init(struct mlxsw_sp *mlxsw_sp,
nh_obj = &info->nh_grp->nh_entries[i].nh;
weight = info->nh_grp->nh_entries[i].weight;
break;
+ case NH_NOTIFIER_INFO_TYPE_RES_TABLE:
+ nh_obj = &info->nh_res_table->nhs[i];
+ weight = 1;
+ break;
default:
err = -EINVAL;
goto err_nexthop_obj_init;
@@ -4473,6 +4794,15 @@ mlxsw_sp_nexthop_obj_group_info_init(struct mlxsw_sp *mlxsw_sp,
goto err_group_refresh;
}
+ /* Add resilient nexthop groups to a list so that the activity of their
+ * nexthop buckets will be periodically queried and cleared.
+ */
+ if (nhgi->is_resilient) {
+ if (list_empty(&mlxsw_sp->router->nh_res_grp_list))
+ mlxsw_sp_nh_grp_activity_work_schedule(mlxsw_sp);
+ list_add(&nhgi->list, &mlxsw_sp->router->nh_res_grp_list);
+ }
+
return 0;
err_group_refresh:
@@ -4491,8 +4821,15 @@ mlxsw_sp_nexthop_obj_group_info_fini(struct mlxsw_sp *mlxsw_sp,
struct mlxsw_sp_nexthop_group *nh_grp)
{
struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
+ struct mlxsw_sp_router *router = mlxsw_sp->router;
int i;
+ if (nhgi->is_resilient) {
+ list_del(&nhgi->list);
+ if (list_empty(&mlxsw_sp->router->nh_res_grp_list))
+ cancel_delayed_work(&router->nh_grp_activity_dw);
+ }
+
for (i = nhgi->count - 1; i >= 0; i--) {
struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
@@ -4685,6 +5022,136 @@ static void mlxsw_sp_nexthop_obj_del(struct mlxsw_sp *mlxsw_sp,
mlxsw_sp_nexthop_obj_group_destroy(mlxsw_sp, nh_grp);
}
+static int mlxsw_sp_nexthop_obj_bucket_query(struct mlxsw_sp *mlxsw_sp,
+ u32 adj_index, char *ratr_pl)
+{
+ MLXSW_REG_ZERO(ratr, ratr_pl);
+ mlxsw_reg_ratr_op_set(ratr_pl, MLXSW_REG_RATR_OP_QUERY_READ);
+ mlxsw_reg_ratr_adjacency_index_low_set(ratr_pl, adj_index);
+ mlxsw_reg_ratr_adjacency_index_high_set(ratr_pl, adj_index >> 16);
+
+ return mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
+}
+
+static int mlxsw_sp_nexthop_obj_bucket_compare(char *ratr_pl, char *ratr_pl_new)
+{
+ /* Clear the opcode and activity on both the old and new payload as
+ * they are irrelevant for the comparison.
+ */
+ mlxsw_reg_ratr_op_set(ratr_pl, MLXSW_REG_RATR_OP_QUERY_READ);
+ mlxsw_reg_ratr_a_set(ratr_pl, 0);
+ mlxsw_reg_ratr_op_set(ratr_pl_new, MLXSW_REG_RATR_OP_QUERY_READ);
+ mlxsw_reg_ratr_a_set(ratr_pl_new, 0);
+
+ /* If the contents of the adjacency entry are consistent with the
+ * replacement request, then replacement was successful.
+ */
+ if (!memcmp(ratr_pl, ratr_pl_new, MLXSW_REG_RATR_LEN))
+ return 0;
+
+ return -EINVAL;
+}
+
+static int
+mlxsw_sp_nexthop_obj_bucket_adj_update(struct mlxsw_sp *mlxsw_sp,
+ struct mlxsw_sp_nexthop *nh,
+ struct nh_notifier_info *info)
+{
+ u16 bucket_index = info->nh_res_bucket->bucket_index;
+ struct netlink_ext_ack *extack = info->extack;
+ bool force = info->nh_res_bucket->force;
+ char ratr_pl_new[MLXSW_REG_RATR_LEN];
+ char ratr_pl[MLXSW_REG_RATR_LEN];
+ u32 adj_index;
+ int err;
+
+ /* No point in trying an atomic replacement if the idle timer interval
+ * is smaller than the interval in which we query and clear activity.
+ */
+ if (!force && info->nh_res_bucket->idle_timer_ms <
+ MLXSW_SP_NH_GRP_ACTIVITY_UPDATE_INTERVAL)
+ force = true;
+
+ adj_index = nh->nhgi->adj_index + bucket_index;
+ err = mlxsw_sp_nexthop_update(mlxsw_sp, adj_index, nh, force, ratr_pl);
+ if (err) {
+ NL_SET_ERR_MSG_MOD(extack, "Failed to overwrite nexthop bucket");
+ return err;
+ }
+
+ if (!force) {
+ err = mlxsw_sp_nexthop_obj_bucket_query(mlxsw_sp, adj_index,
+ ratr_pl_new);
+ if (err) {
+ NL_SET_ERR_MSG_MOD(extack, "Failed to query nexthop bucket state after replacement. State might be inconsistent");
+ return err;
+ }
+
+ err = mlxsw_sp_nexthop_obj_bucket_compare(ratr_pl, ratr_pl_new);
+ if (err) {
+ NL_SET_ERR_MSG_MOD(extack, "Nexthop bucket was not replaced because it was active during replacement");
+ return err;
+ }
+ }
+
+ nh->update = 0;
+ nh->offloaded = 1;
+ mlxsw_sp_nexthop_bucket_offload_refresh(mlxsw_sp, nh, bucket_index);
+
+ return 0;
+}
+
+static int mlxsw_sp_nexthop_obj_bucket_replace(struct mlxsw_sp *mlxsw_sp,
+ struct nh_notifier_info *info)
+{
+ u16 bucket_index = info->nh_res_bucket->bucket_index;
+ struct netlink_ext_ack *extack = info->extack;
+ struct mlxsw_sp_nexthop_group_info *nhgi;
+ struct nh_notifier_single_info *nh_obj;
+ struct mlxsw_sp_nexthop_group *nh_grp;
+ struct mlxsw_sp_nexthop *nh;
+ int err;
+
+ nh_grp = mlxsw_sp_nexthop_obj_group_lookup(mlxsw_sp, info->id);
+ if (!nh_grp) {
+ NL_SET_ERR_MSG_MOD(extack, "Nexthop group was not found");
+ return -EINVAL;
+ }
+
+ nhgi = nh_grp->nhgi;
+
+ if (bucket_index >= nhgi->count) {
+ NL_SET_ERR_MSG_MOD(extack, "Nexthop bucket index out of range");
+ return -EINVAL;
+ }
+
+ nh = &nhgi->nexthops[bucket_index];
+ mlxsw_sp_nexthop_obj_fini(mlxsw_sp, nh);
+
+ nh_obj = &info->nh_res_bucket->new_nh;
+ err = mlxsw_sp_nexthop_obj_init(mlxsw_sp, nh_grp, nh, nh_obj, 1);
+ if (err) {
+ NL_SET_ERR_MSG_MOD(extack, "Failed to initialize nexthop object for nexthop bucket replacement");
+ goto err_nexthop_obj_init;
+ }
+
+ err = mlxsw_sp_nexthop_obj_bucket_adj_update(mlxsw_sp, nh, info);
+ if (err)
+ goto err_nexthop_obj_bucket_adj_update;
+
+ return 0;
+
+err_nexthop_obj_bucket_adj_update:
+ mlxsw_sp_nexthop_obj_fini(mlxsw_sp, nh);
+err_nexthop_obj_init:
+ nh_obj = &info->nh_res_bucket->old_nh;
+ mlxsw_sp_nexthop_obj_init(mlxsw_sp, nh_grp, nh, nh_obj, 1);
+ /* The old adjacency entry was not overwritten */
+ nh->update = 0;
+ nh->offloaded = 1;
+ return err;
+}
+
static int mlxsw_sp_nexthop_obj_event(struct notifier_block *nb,
unsigned long event, void *ptr)
{
@@ -4699,8 +5166,6 @@ static int mlxsw_sp_nexthop_obj_event(struct notifier_block *nb,
mutex_lock(&router->lock);
- ASSERT_RTNL();
-
switch (event) {
case NEXTHOP_EVENT_REPLACE:
err = mlxsw_sp_nexthop_obj_new(router->mlxsw_sp, info);
@@ -4708,6 +5173,10 @@ static int mlxsw_sp_nexthop_obj_event(struct notifier_block *nb,
case NEXTHOP_EVENT_DEL:
mlxsw_sp_nexthop_obj_del(router->mlxsw_sp, info);
break;
+ case NEXTHOP_EVENT_BUCKET_REPLACE:
+ err = mlxsw_sp_nexthop_obj_bucket_replace(router->mlxsw_sp,
+ info);
+ break;
default:
break;
}
@@ -7667,7 +8136,7 @@ mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
int i, err;
type = mlxsw_sp_dev_rif_type(mlxsw_sp, params->dev);
- ops = mlxsw_sp->rif_ops_arr[type];
+ ops = mlxsw_sp->router->rif_ops_arr[type];
vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id ? : RT_TABLE_MAIN, extack);
if (IS_ERR(vr))
@@ -8865,7 +9334,7 @@ static const struct mlxsw_sp_rif_ops mlxsw_sp1_rif_ipip_lb_ops = {
.deconfigure = mlxsw_sp1_rif_ipip_lb_deconfigure,
};
-const struct mlxsw_sp_rif_ops *mlxsw_sp1_rif_ops_arr[] = {
+static const struct mlxsw_sp_rif_ops *mlxsw_sp1_rif_ops_arr[] = {
[MLXSW_SP_RIF_TYPE_SUBPORT] = &mlxsw_sp_rif_subport_ops,
[MLXSW_SP_RIF_TYPE_VLAN] = &mlxsw_sp_rif_vlan_emu_ops,
[MLXSW_SP_RIF_TYPE_FID] = &mlxsw_sp_rif_fid_ops,
@@ -9050,7 +9519,7 @@ static const struct mlxsw_sp_rif_ops mlxsw_sp2_rif_ipip_lb_ops = {
.deconfigure = mlxsw_sp2_rif_ipip_lb_deconfigure,
};
-const struct mlxsw_sp_rif_ops *mlxsw_sp2_rif_ops_arr[] = {
+static const struct mlxsw_sp_rif_ops *mlxsw_sp2_rif_ops_arr[] = {
[MLXSW_SP_RIF_TYPE_SUBPORT] = &mlxsw_sp_rif_subport_ops,
[MLXSW_SP_RIF_TYPE_VLAN] = &mlxsw_sp_rif_vlan_emu_ops,
[MLXSW_SP_RIF_TYPE_FID] = &mlxsw_sp_rif_fid_ops,
@@ -9302,6 +9771,36 @@ static void mlxsw_sp_lb_rif_fini(struct mlxsw_sp *mlxsw_sp)
mlxsw_sp_router_ul_rif_put(mlxsw_sp, mlxsw_sp->router->lb_rif_index);
}
+static int mlxsw_sp1_router_init(struct mlxsw_sp *mlxsw_sp)
+{
+ size_t size_ranges_count = ARRAY_SIZE(mlxsw_sp1_adj_grp_size_ranges);
+
+ mlxsw_sp->router->rif_ops_arr = mlxsw_sp1_rif_ops_arr;
+ mlxsw_sp->router->adj_grp_size_ranges = mlxsw_sp1_adj_grp_size_ranges;
+ mlxsw_sp->router->adj_grp_size_ranges_count = size_ranges_count;
+
+ return 0;
+}
+
+const struct mlxsw_sp_router_ops mlxsw_sp1_router_ops = {
+ .init = mlxsw_sp1_router_init,
+};
+
+static int mlxsw_sp2_router_init(struct mlxsw_sp *mlxsw_sp)
+{
+ size_t size_ranges_count = ARRAY_SIZE(mlxsw_sp2_adj_grp_size_ranges);
+
+ mlxsw_sp->router->rif_ops_arr = mlxsw_sp2_rif_ops_arr;
+ mlxsw_sp->router->adj_grp_size_ranges = mlxsw_sp2_adj_grp_size_ranges;
+ mlxsw_sp->router->adj_grp_size_ranges_count = size_ranges_count;
+
+ return 0;
+}
+
+const struct mlxsw_sp_router_ops mlxsw_sp2_router_ops = {
+ .init = mlxsw_sp2_router_init,
+};
+
int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp,
struct netlink_ext_ack *extack)
{
@@ -9315,6 +9814,10 @@ int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp,
mlxsw_sp->router = router;
router->mlxsw_sp = mlxsw_sp;
+ err = mlxsw_sp->router_ops->init(mlxsw_sp);
+ if (err)
+ goto err_router_ops_init;
+
err = mlxsw_sp_router_xm_init(mlxsw_sp);
if (err)
goto err_xm_init;
@@ -9328,6 +9831,10 @@ int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp,
if (err)
goto err_ll_op_ctx_init;
+ INIT_LIST_HEAD(&mlxsw_sp->router->nh_res_grp_list);
+ INIT_DELAYED_WORK(&mlxsw_sp->router->nh_grp_activity_dw,
+ mlxsw_sp_nh_grp_activity_work);
+
INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_neighs_list);
err = __mlxsw_sp_router_init(mlxsw_sp);
if (err)
@@ -9451,10 +9958,12 @@ err_ipips_init:
err_rifs_init:
__mlxsw_sp_router_fini(mlxsw_sp);
err_router_init:
+ cancel_delayed_work_sync(&mlxsw_sp->router->nh_grp_activity_dw);
mlxsw_sp_router_ll_op_ctx_fini(router);
err_ll_op_ctx_init:
mlxsw_sp_router_xm_fini(mlxsw_sp);
err_xm_init:
+err_router_ops_init:
mutex_destroy(&mlxsw_sp->router->lock);
kfree(mlxsw_sp->router);
return err;
@@ -9481,6 +9990,7 @@ void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
mlxsw_sp_ipips_fini(mlxsw_sp);
mlxsw_sp_rifs_fini(mlxsw_sp);
__mlxsw_sp_router_fini(mlxsw_sp);
+ cancel_delayed_work_sync(&mlxsw_sp->router->nh_grp_activity_dw);
mlxsw_sp_router_ll_op_ctx_fini(mlxsw_sp->router);
mlxsw_sp_router_xm_fini(mlxsw_sp);
mutex_destroy(&mlxsw_sp->router->lock);