summaryrefslogtreecommitdiffstats
path: root/drivers/net/ethernet/freescale/dpaa2/dpaa2-switch-flower.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/net/ethernet/freescale/dpaa2/dpaa2-switch-flower.c')
-rw-r--r--drivers/net/ethernet/freescale/dpaa2/dpaa2-switch-flower.c885
1 files changed, 885 insertions, 0 deletions
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-switch-flower.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-switch-flower.c
new file mode 100644
index 000000000..16d3c3610
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-switch-flower.c
@@ -0,0 +1,885 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * DPAA2 Ethernet Switch flower support
+ *
+ * Copyright 2021 NXP
+ *
+ */
+
+#include "dpaa2-switch.h"
+
+static int dpaa2_switch_flower_parse_key(struct flow_cls_offload *cls,
+ struct dpsw_acl_key *acl_key)
+{
+ struct flow_rule *rule = flow_cls_offload_flow_rule(cls);
+ struct flow_dissector *dissector = rule->match.dissector;
+ struct netlink_ext_ack *extack = cls->common.extack;
+ struct dpsw_acl_fields *acl_h, *acl_m;
+
+ if (dissector->used_keys &
+ ~(BIT(FLOW_DISSECTOR_KEY_BASIC) |
+ BIT(FLOW_DISSECTOR_KEY_CONTROL) |
+ BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
+ BIT(FLOW_DISSECTOR_KEY_VLAN) |
+ BIT(FLOW_DISSECTOR_KEY_PORTS) |
+ BIT(FLOW_DISSECTOR_KEY_IP) |
+ BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
+ BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS))) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Unsupported keys used");
+ return -EOPNOTSUPP;
+ }
+
+ acl_h = &acl_key->match;
+ acl_m = &acl_key->mask;
+
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
+ struct flow_match_basic match;
+
+ flow_rule_match_basic(rule, &match);
+ acl_h->l3_protocol = match.key->ip_proto;
+ acl_h->l2_ether_type = be16_to_cpu(match.key->n_proto);
+ acl_m->l3_protocol = match.mask->ip_proto;
+ acl_m->l2_ether_type = be16_to_cpu(match.mask->n_proto);
+ }
+
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
+ struct flow_match_eth_addrs match;
+
+ flow_rule_match_eth_addrs(rule, &match);
+ ether_addr_copy(acl_h->l2_dest_mac, &match.key->dst[0]);
+ ether_addr_copy(acl_h->l2_source_mac, &match.key->src[0]);
+ ether_addr_copy(acl_m->l2_dest_mac, &match.mask->dst[0]);
+ ether_addr_copy(acl_m->l2_source_mac, &match.mask->src[0]);
+ }
+
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
+ struct flow_match_vlan match;
+
+ flow_rule_match_vlan(rule, &match);
+ acl_h->l2_vlan_id = match.key->vlan_id;
+ acl_h->l2_tpid = be16_to_cpu(match.key->vlan_tpid);
+ acl_h->l2_pcp_dei = match.key->vlan_priority << 1 |
+ match.key->vlan_dei;
+
+ acl_m->l2_vlan_id = match.mask->vlan_id;
+ acl_m->l2_tpid = be16_to_cpu(match.mask->vlan_tpid);
+ acl_m->l2_pcp_dei = match.mask->vlan_priority << 1 |
+ match.mask->vlan_dei;
+ }
+
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV4_ADDRS)) {
+ struct flow_match_ipv4_addrs match;
+
+ flow_rule_match_ipv4_addrs(rule, &match);
+ acl_h->l3_source_ip = be32_to_cpu(match.key->src);
+ acl_h->l3_dest_ip = be32_to_cpu(match.key->dst);
+ acl_m->l3_source_ip = be32_to_cpu(match.mask->src);
+ acl_m->l3_dest_ip = be32_to_cpu(match.mask->dst);
+ }
+
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
+ struct flow_match_ports match;
+
+ flow_rule_match_ports(rule, &match);
+ acl_h->l4_source_port = be16_to_cpu(match.key->src);
+ acl_h->l4_dest_port = be16_to_cpu(match.key->dst);
+ acl_m->l4_source_port = be16_to_cpu(match.mask->src);
+ acl_m->l4_dest_port = be16_to_cpu(match.mask->dst);
+ }
+
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
+ struct flow_match_ip match;
+
+ flow_rule_match_ip(rule, &match);
+ if (match.mask->ttl != 0) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Matching on TTL not supported");
+ return -EOPNOTSUPP;
+ }
+
+ if ((match.mask->tos & 0x3) != 0) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Matching on ECN not supported, only DSCP");
+ return -EOPNOTSUPP;
+ }
+
+ acl_h->l3_dscp = match.key->tos >> 2;
+ acl_m->l3_dscp = match.mask->tos >> 2;
+ }
+
+ return 0;
+}
+
+int dpaa2_switch_acl_entry_add(struct dpaa2_switch_filter_block *filter_block,
+ struct dpaa2_switch_acl_entry *entry)
+{
+ struct dpsw_acl_entry_cfg *acl_entry_cfg = &entry->cfg;
+ struct ethsw_core *ethsw = filter_block->ethsw;
+ struct dpsw_acl_key *acl_key = &entry->key;
+ struct device *dev = ethsw->dev;
+ u8 *cmd_buff;
+ int err;
+
+ cmd_buff = kzalloc(DPAA2_ETHSW_PORT_ACL_CMD_BUF_SIZE, GFP_KERNEL);
+ if (!cmd_buff)
+ return -ENOMEM;
+
+ dpsw_acl_prepare_entry_cfg(acl_key, cmd_buff);
+
+ acl_entry_cfg->key_iova = dma_map_single(dev, cmd_buff,
+ DPAA2_ETHSW_PORT_ACL_CMD_BUF_SIZE,
+ DMA_TO_DEVICE);
+ if (unlikely(dma_mapping_error(dev, acl_entry_cfg->key_iova))) {
+ dev_err(dev, "DMA mapping failed\n");
+ kfree(cmd_buff);
+ return -EFAULT;
+ }
+
+ err = dpsw_acl_add_entry(ethsw->mc_io, 0, ethsw->dpsw_handle,
+ filter_block->acl_id, acl_entry_cfg);
+
+ dma_unmap_single(dev, acl_entry_cfg->key_iova,
+ DPAA2_ETHSW_PORT_ACL_CMD_BUF_SIZE,
+ DMA_TO_DEVICE);
+ if (err) {
+ dev_err(dev, "dpsw_acl_add_entry() failed %d\n", err);
+ kfree(cmd_buff);
+ return err;
+ }
+
+ kfree(cmd_buff);
+
+ return 0;
+}
+
+static int
+dpaa2_switch_acl_entry_remove(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_acl_entry *entry)
+{
+ struct dpsw_acl_entry_cfg *acl_entry_cfg = &entry->cfg;
+ struct dpsw_acl_key *acl_key = &entry->key;
+ struct ethsw_core *ethsw = block->ethsw;
+ struct device *dev = ethsw->dev;
+ u8 *cmd_buff;
+ int err;
+
+ cmd_buff = kzalloc(DPAA2_ETHSW_PORT_ACL_CMD_BUF_SIZE, GFP_KERNEL);
+ if (!cmd_buff)
+ return -ENOMEM;
+
+ dpsw_acl_prepare_entry_cfg(acl_key, cmd_buff);
+
+ acl_entry_cfg->key_iova = dma_map_single(dev, cmd_buff,
+ DPAA2_ETHSW_PORT_ACL_CMD_BUF_SIZE,
+ DMA_TO_DEVICE);
+ if (unlikely(dma_mapping_error(dev, acl_entry_cfg->key_iova))) {
+ dev_err(dev, "DMA mapping failed\n");
+ kfree(cmd_buff);
+ return -EFAULT;
+ }
+
+ err = dpsw_acl_remove_entry(ethsw->mc_io, 0, ethsw->dpsw_handle,
+ block->acl_id, acl_entry_cfg);
+
+ dma_unmap_single(dev, acl_entry_cfg->key_iova,
+ DPAA2_ETHSW_PORT_ACL_CMD_BUF_SIZE, DMA_TO_DEVICE);
+ if (err) {
+ dev_err(dev, "dpsw_acl_remove_entry() failed %d\n", err);
+ kfree(cmd_buff);
+ return err;
+ }
+
+ kfree(cmd_buff);
+
+ return 0;
+}
+
+static int
+dpaa2_switch_acl_entry_add_to_list(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_acl_entry *entry)
+{
+ struct dpaa2_switch_acl_entry *tmp;
+ struct list_head *pos, *n;
+ int index = 0;
+
+ if (list_empty(&block->acl_entries)) {
+ list_add(&entry->list, &block->acl_entries);
+ return index;
+ }
+
+ list_for_each_safe(pos, n, &block->acl_entries) {
+ tmp = list_entry(pos, struct dpaa2_switch_acl_entry, list);
+ if (entry->prio < tmp->prio)
+ break;
+ index++;
+ }
+ list_add(&entry->list, pos->prev);
+ return index;
+}
+
+static struct dpaa2_switch_acl_entry*
+dpaa2_switch_acl_entry_get_by_index(struct dpaa2_switch_filter_block *block,
+ int index)
+{
+ struct dpaa2_switch_acl_entry *tmp;
+ int i = 0;
+
+ list_for_each_entry(tmp, &block->acl_entries, list) {
+ if (i == index)
+ return tmp;
+ ++i;
+ }
+
+ return NULL;
+}
+
+static int
+dpaa2_switch_acl_entry_set_precedence(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_acl_entry *entry,
+ int precedence)
+{
+ int err;
+
+ err = dpaa2_switch_acl_entry_remove(block, entry);
+ if (err)
+ return err;
+
+ entry->cfg.precedence = precedence;
+ return dpaa2_switch_acl_entry_add(block, entry);
+}
+
+static int
+dpaa2_switch_acl_tbl_add_entry(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_acl_entry *entry)
+{
+ struct dpaa2_switch_acl_entry *tmp;
+ int index, i, precedence, err;
+
+ /* Add the new ACL entry to the linked list and get its index */
+ index = dpaa2_switch_acl_entry_add_to_list(block, entry);
+
+ /* Move up in priority the ACL entries to make space
+ * for the new filter.
+ */
+ precedence = DPAA2_ETHSW_PORT_MAX_ACL_ENTRIES - block->num_acl_rules - 1;
+ for (i = 0; i < index; i++) {
+ tmp = dpaa2_switch_acl_entry_get_by_index(block, i);
+
+ err = dpaa2_switch_acl_entry_set_precedence(block, tmp,
+ precedence);
+ if (err)
+ return err;
+
+ precedence++;
+ }
+
+ /* Add the new entry to hardware */
+ entry->cfg.precedence = precedence;
+ err = dpaa2_switch_acl_entry_add(block, entry);
+ block->num_acl_rules++;
+
+ return err;
+}
+
+static struct dpaa2_switch_acl_entry *
+dpaa2_switch_acl_tbl_find_entry_by_cookie(struct dpaa2_switch_filter_block *block,
+ unsigned long cookie)
+{
+ struct dpaa2_switch_acl_entry *tmp, *n;
+
+ list_for_each_entry_safe(tmp, n, &block->acl_entries, list) {
+ if (tmp->cookie == cookie)
+ return tmp;
+ }
+ return NULL;
+}
+
+static int
+dpaa2_switch_acl_entry_get_index(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_acl_entry *entry)
+{
+ struct dpaa2_switch_acl_entry *tmp, *n;
+ int index = 0;
+
+ list_for_each_entry_safe(tmp, n, &block->acl_entries, list) {
+ if (tmp->cookie == entry->cookie)
+ return index;
+ index++;
+ }
+ return -ENOENT;
+}
+
+static struct dpaa2_switch_mirror_entry *
+dpaa2_switch_mirror_find_entry_by_cookie(struct dpaa2_switch_filter_block *block,
+ unsigned long cookie)
+{
+ struct dpaa2_switch_mirror_entry *tmp, *n;
+
+ list_for_each_entry_safe(tmp, n, &block->mirror_entries, list) {
+ if (tmp->cookie == cookie)
+ return tmp;
+ }
+ return NULL;
+}
+
+static int
+dpaa2_switch_acl_tbl_remove_entry(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_acl_entry *entry)
+{
+ struct dpaa2_switch_acl_entry *tmp;
+ int index, i, precedence, err;
+
+ index = dpaa2_switch_acl_entry_get_index(block, entry);
+
+ /* Remove from hardware the ACL entry */
+ err = dpaa2_switch_acl_entry_remove(block, entry);
+ if (err)
+ return err;
+
+ block->num_acl_rules--;
+
+ /* Remove it from the list also */
+ list_del(&entry->list);
+
+ /* Move down in priority the entries over the deleted one */
+ precedence = entry->cfg.precedence;
+ for (i = index - 1; i >= 0; i--) {
+ tmp = dpaa2_switch_acl_entry_get_by_index(block, i);
+ err = dpaa2_switch_acl_entry_set_precedence(block, tmp,
+ precedence);
+ if (err)
+ return err;
+
+ precedence--;
+ }
+
+ kfree(entry);
+
+ return 0;
+}
+
+static int dpaa2_switch_tc_parse_action_acl(struct ethsw_core *ethsw,
+ struct flow_action_entry *cls_act,
+ struct dpsw_acl_result *dpsw_act,
+ struct netlink_ext_ack *extack)
+{
+ int err = 0;
+
+ switch (cls_act->id) {
+ case FLOW_ACTION_TRAP:
+ dpsw_act->action = DPSW_ACL_ACTION_REDIRECT_TO_CTRL_IF;
+ break;
+ case FLOW_ACTION_REDIRECT:
+ if (!dpaa2_switch_port_dev_check(cls_act->dev)) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Destination not a DPAA2 switch port");
+ return -EOPNOTSUPP;
+ }
+
+ dpsw_act->if_id = dpaa2_switch_get_index(ethsw, cls_act->dev);
+ dpsw_act->action = DPSW_ACL_ACTION_REDIRECT;
+ break;
+ case FLOW_ACTION_DROP:
+ dpsw_act->action = DPSW_ACL_ACTION_DROP;
+ break;
+ default:
+ NL_SET_ERR_MSG_MOD(extack,
+ "Action not supported");
+ err = -EOPNOTSUPP;
+ goto out;
+ }
+
+out:
+ return err;
+}
+
+static int
+dpaa2_switch_block_add_mirror(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_mirror_entry *entry,
+ u16 to, struct netlink_ext_ack *extack)
+{
+ unsigned long block_ports = block->ports;
+ struct ethsw_core *ethsw = block->ethsw;
+ struct ethsw_port_priv *port_priv;
+ unsigned long ports_added = 0;
+ u16 vlan = entry->cfg.vlan_id;
+ bool mirror_port_enabled;
+ int err, port;
+
+ /* Setup the mirroring port */
+ mirror_port_enabled = (ethsw->mirror_port != ethsw->sw_attr.num_ifs);
+ if (!mirror_port_enabled) {
+ err = dpsw_set_reflection_if(ethsw->mc_io, 0,
+ ethsw->dpsw_handle, to);
+ if (err)
+ return err;
+ ethsw->mirror_port = to;
+ }
+
+ /* Setup the same egress mirroring configuration on all the switch
+ * ports that share the same filter block.
+ */
+ for_each_set_bit(port, &block_ports, ethsw->sw_attr.num_ifs) {
+ port_priv = ethsw->ports[port];
+
+ /* We cannot add a per VLAN mirroring rule if the VLAN in
+ * question is not installed on the switch port.
+ */
+ if (entry->cfg.filter == DPSW_REFLECTION_FILTER_INGRESS_VLAN &&
+ !(port_priv->vlans[vlan] & ETHSW_VLAN_MEMBER)) {
+ NL_SET_ERR_MSG(extack,
+ "VLAN must be installed on the switch port");
+ err = -EINVAL;
+ goto err_remove_filters;
+ }
+
+ err = dpsw_if_add_reflection(ethsw->mc_io, 0,
+ ethsw->dpsw_handle,
+ port, &entry->cfg);
+ if (err)
+ goto err_remove_filters;
+
+ ports_added |= BIT(port);
+ }
+
+ list_add(&entry->list, &block->mirror_entries);
+
+ return 0;
+
+err_remove_filters:
+ for_each_set_bit(port, &ports_added, ethsw->sw_attr.num_ifs) {
+ dpsw_if_remove_reflection(ethsw->mc_io, 0, ethsw->dpsw_handle,
+ port, &entry->cfg);
+ }
+
+ if (!mirror_port_enabled)
+ ethsw->mirror_port = ethsw->sw_attr.num_ifs;
+
+ return err;
+}
+
+static int
+dpaa2_switch_block_remove_mirror(struct dpaa2_switch_filter_block *block,
+ struct dpaa2_switch_mirror_entry *entry)
+{
+ struct dpsw_reflection_cfg *cfg = &entry->cfg;
+ unsigned long block_ports = block->ports;
+ struct ethsw_core *ethsw = block->ethsw;
+ int port;
+
+ /* Remove this mirroring configuration from all the ports belonging to
+ * the filter block.
+ */
+ for_each_set_bit(port, &block_ports, ethsw->sw_attr.num_ifs)
+ dpsw_if_remove_reflection(ethsw->mc_io, 0, ethsw->dpsw_handle,
+ port, cfg);
+
+ /* Also remove it from the list of mirror filters */
+ list_del(&entry->list);
+ kfree(entry);
+
+ /* If this was the last mirror filter, then unset the mirror port */
+ if (list_empty(&block->mirror_entries))
+ ethsw->mirror_port = ethsw->sw_attr.num_ifs;
+
+ return 0;
+}
+
+static int
+dpaa2_switch_cls_flower_replace_acl(struct dpaa2_switch_filter_block *block,
+ struct flow_cls_offload *cls)
+{
+ struct flow_rule *rule = flow_cls_offload_flow_rule(cls);
+ struct netlink_ext_ack *extack = cls->common.extack;
+ struct dpaa2_switch_acl_entry *acl_entry;
+ struct ethsw_core *ethsw = block->ethsw;
+ struct flow_action_entry *act;
+ int err;
+
+ if (dpaa2_switch_acl_tbl_is_full(block)) {
+ NL_SET_ERR_MSG(extack, "Maximum filter capacity reached");
+ return -ENOMEM;
+ }
+
+ acl_entry = kzalloc(sizeof(*acl_entry), GFP_KERNEL);
+ if (!acl_entry)
+ return -ENOMEM;
+
+ err = dpaa2_switch_flower_parse_key(cls, &acl_entry->key);
+ if (err)
+ goto free_acl_entry;
+
+ act = &rule->action.entries[0];
+ err = dpaa2_switch_tc_parse_action_acl(ethsw, act,
+ &acl_entry->cfg.result, extack);
+ if (err)
+ goto free_acl_entry;
+
+ acl_entry->prio = cls->common.prio;
+ acl_entry->cookie = cls->cookie;
+
+ err = dpaa2_switch_acl_tbl_add_entry(block, acl_entry);
+ if (err)
+ goto free_acl_entry;
+
+ return 0;
+
+free_acl_entry:
+ kfree(acl_entry);
+
+ return err;
+}
+
+static int dpaa2_switch_flower_parse_mirror_key(struct flow_cls_offload *cls,
+ u16 *vlan)
+{
+ struct flow_rule *rule = flow_cls_offload_flow_rule(cls);
+ struct flow_dissector *dissector = rule->match.dissector;
+ struct netlink_ext_ack *extack = cls->common.extack;
+ int ret = -EOPNOTSUPP;
+
+ if (dissector->used_keys &
+ ~(BIT(FLOW_DISSECTOR_KEY_BASIC) |
+ BIT(FLOW_DISSECTOR_KEY_CONTROL) |
+ BIT(FLOW_DISSECTOR_KEY_VLAN))) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Mirroring is supported only per VLAN");
+ return -EOPNOTSUPP;
+ }
+
+ if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
+ struct flow_match_vlan match;
+
+ flow_rule_match_vlan(rule, &match);
+
+ if (match.mask->vlan_priority != 0 ||
+ match.mask->vlan_dei != 0) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Only matching on VLAN ID supported");
+ return -EOPNOTSUPP;
+ }
+
+ if (match.mask->vlan_id != 0xFFF) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Masked matching not supported");
+ return -EOPNOTSUPP;
+ }
+
+ *vlan = (u16)match.key->vlan_id;
+ ret = 0;
+ }
+
+ return ret;
+}
+
+static int
+dpaa2_switch_cls_flower_replace_mirror(struct dpaa2_switch_filter_block *block,
+ struct flow_cls_offload *cls)
+{
+ struct netlink_ext_ack *extack = cls->common.extack;
+ struct dpaa2_switch_mirror_entry *mirror_entry;
+ struct ethsw_core *ethsw = block->ethsw;
+ struct dpaa2_switch_mirror_entry *tmp;
+ struct flow_action_entry *cls_act;
+ struct list_head *pos, *n;
+ bool mirror_port_enabled;
+ u16 if_id, vlan;
+ int err;
+
+ mirror_port_enabled = (ethsw->mirror_port != ethsw->sw_attr.num_ifs);
+ cls_act = &cls->rule->action.entries[0];
+
+ /* Offload rules only when the destination is a DPAA2 switch port */
+ if (!dpaa2_switch_port_dev_check(cls_act->dev)) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Destination not a DPAA2 switch port");
+ return -EOPNOTSUPP;
+ }
+ if_id = dpaa2_switch_get_index(ethsw, cls_act->dev);
+
+ /* We have a single mirror port but can configure egress mirroring on
+ * all the other switch ports. We need to allow mirroring rules only
+ * when the destination port is the same.
+ */
+ if (mirror_port_enabled && ethsw->mirror_port != if_id) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Multiple mirror ports not supported");
+ return -EBUSY;
+ }
+
+ /* Parse the key */
+ err = dpaa2_switch_flower_parse_mirror_key(cls, &vlan);
+ if (err)
+ return err;
+
+ /* Make sure that we don't already have a mirror rule with the same
+ * configuration.
+ */
+ list_for_each_safe(pos, n, &block->mirror_entries) {
+ tmp = list_entry(pos, struct dpaa2_switch_mirror_entry, list);
+
+ if (tmp->cfg.filter == DPSW_REFLECTION_FILTER_INGRESS_VLAN &&
+ tmp->cfg.vlan_id == vlan) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "VLAN mirror filter already installed");
+ return -EBUSY;
+ }
+ }
+
+ mirror_entry = kzalloc(sizeof(*mirror_entry), GFP_KERNEL);
+ if (!mirror_entry)
+ return -ENOMEM;
+
+ mirror_entry->cfg.filter = DPSW_REFLECTION_FILTER_INGRESS_VLAN;
+ mirror_entry->cfg.vlan_id = vlan;
+ mirror_entry->cookie = cls->cookie;
+
+ return dpaa2_switch_block_add_mirror(block, mirror_entry, if_id,
+ extack);
+}
+
+int dpaa2_switch_cls_flower_replace(struct dpaa2_switch_filter_block *block,
+ struct flow_cls_offload *cls)
+{
+ struct flow_rule *rule = flow_cls_offload_flow_rule(cls);
+ struct netlink_ext_ack *extack = cls->common.extack;
+ struct flow_action_entry *act;
+
+ if (!flow_offload_has_one_action(&rule->action)) {
+ NL_SET_ERR_MSG(extack, "Only singular actions are supported");
+ return -EOPNOTSUPP;
+ }
+
+ act = &rule->action.entries[0];
+ switch (act->id) {
+ case FLOW_ACTION_REDIRECT:
+ case FLOW_ACTION_TRAP:
+ case FLOW_ACTION_DROP:
+ return dpaa2_switch_cls_flower_replace_acl(block, cls);
+ case FLOW_ACTION_MIRRED:
+ return dpaa2_switch_cls_flower_replace_mirror(block, cls);
+ default:
+ NL_SET_ERR_MSG_MOD(extack, "Action not supported");
+ return -EOPNOTSUPP;
+ }
+}
+
+int dpaa2_switch_cls_flower_destroy(struct dpaa2_switch_filter_block *block,
+ struct flow_cls_offload *cls)
+{
+ struct dpaa2_switch_mirror_entry *mirror_entry;
+ struct dpaa2_switch_acl_entry *acl_entry;
+
+ /* If this filter is a an ACL one, remove it */
+ acl_entry = dpaa2_switch_acl_tbl_find_entry_by_cookie(block,
+ cls->cookie);
+ if (acl_entry)
+ return dpaa2_switch_acl_tbl_remove_entry(block, acl_entry);
+
+ /* If not, then it has to be a mirror */
+ mirror_entry = dpaa2_switch_mirror_find_entry_by_cookie(block,
+ cls->cookie);
+ if (mirror_entry)
+ return dpaa2_switch_block_remove_mirror(block,
+ mirror_entry);
+
+ return 0;
+}
+
+static int
+dpaa2_switch_cls_matchall_replace_acl(struct dpaa2_switch_filter_block *block,
+ struct tc_cls_matchall_offload *cls)
+{
+ struct netlink_ext_ack *extack = cls->common.extack;
+ struct ethsw_core *ethsw = block->ethsw;
+ struct dpaa2_switch_acl_entry *acl_entry;
+ struct flow_action_entry *act;
+ int err;
+
+ if (dpaa2_switch_acl_tbl_is_full(block)) {
+ NL_SET_ERR_MSG(extack, "Maximum filter capacity reached");
+ return -ENOMEM;
+ }
+
+ acl_entry = kzalloc(sizeof(*acl_entry), GFP_KERNEL);
+ if (!acl_entry)
+ return -ENOMEM;
+
+ act = &cls->rule->action.entries[0];
+ err = dpaa2_switch_tc_parse_action_acl(ethsw, act,
+ &acl_entry->cfg.result, extack);
+ if (err)
+ goto free_acl_entry;
+
+ acl_entry->prio = cls->common.prio;
+ acl_entry->cookie = cls->cookie;
+
+ err = dpaa2_switch_acl_tbl_add_entry(block, acl_entry);
+ if (err)
+ goto free_acl_entry;
+
+ return 0;
+
+free_acl_entry:
+ kfree(acl_entry);
+
+ return err;
+}
+
+static int
+dpaa2_switch_cls_matchall_replace_mirror(struct dpaa2_switch_filter_block *block,
+ struct tc_cls_matchall_offload *cls)
+{
+ struct netlink_ext_ack *extack = cls->common.extack;
+ struct dpaa2_switch_mirror_entry *mirror_entry;
+ struct ethsw_core *ethsw = block->ethsw;
+ struct dpaa2_switch_mirror_entry *tmp;
+ struct flow_action_entry *cls_act;
+ struct list_head *pos, *n;
+ bool mirror_port_enabled;
+ u16 if_id;
+
+ mirror_port_enabled = (ethsw->mirror_port != ethsw->sw_attr.num_ifs);
+ cls_act = &cls->rule->action.entries[0];
+
+ /* Offload rules only when the destination is a DPAA2 switch port */
+ if (!dpaa2_switch_port_dev_check(cls_act->dev)) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Destination not a DPAA2 switch port");
+ return -EOPNOTSUPP;
+ }
+ if_id = dpaa2_switch_get_index(ethsw, cls_act->dev);
+
+ /* We have a single mirror port but can configure egress mirroring on
+ * all the other switch ports. We need to allow mirroring rules only
+ * when the destination port is the same.
+ */
+ if (mirror_port_enabled && ethsw->mirror_port != if_id) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Multiple mirror ports not supported");
+ return -EBUSY;
+ }
+
+ /* Make sure that we don't already have a mirror rule with the same
+ * configuration. One matchall rule per block is the maximum.
+ */
+ list_for_each_safe(pos, n, &block->mirror_entries) {
+ tmp = list_entry(pos, struct dpaa2_switch_mirror_entry, list);
+
+ if (tmp->cfg.filter == DPSW_REFLECTION_FILTER_INGRESS_ALL) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Matchall mirror filter already installed");
+ return -EBUSY;
+ }
+ }
+
+ mirror_entry = kzalloc(sizeof(*mirror_entry), GFP_KERNEL);
+ if (!mirror_entry)
+ return -ENOMEM;
+
+ mirror_entry->cfg.filter = DPSW_REFLECTION_FILTER_INGRESS_ALL;
+ mirror_entry->cookie = cls->cookie;
+
+ return dpaa2_switch_block_add_mirror(block, mirror_entry, if_id,
+ extack);
+}
+
+int dpaa2_switch_cls_matchall_replace(struct dpaa2_switch_filter_block *block,
+ struct tc_cls_matchall_offload *cls)
+{
+ struct netlink_ext_ack *extack = cls->common.extack;
+ struct flow_action_entry *act;
+
+ if (!flow_offload_has_one_action(&cls->rule->action)) {
+ NL_SET_ERR_MSG(extack, "Only singular actions are supported");
+ return -EOPNOTSUPP;
+ }
+
+ act = &cls->rule->action.entries[0];
+ switch (act->id) {
+ case FLOW_ACTION_REDIRECT:
+ case FLOW_ACTION_TRAP:
+ case FLOW_ACTION_DROP:
+ return dpaa2_switch_cls_matchall_replace_acl(block, cls);
+ case FLOW_ACTION_MIRRED:
+ return dpaa2_switch_cls_matchall_replace_mirror(block, cls);
+ default:
+ NL_SET_ERR_MSG_MOD(extack, "Action not supported");
+ return -EOPNOTSUPP;
+ }
+}
+
+int dpaa2_switch_block_offload_mirror(struct dpaa2_switch_filter_block *block,
+ struct ethsw_port_priv *port_priv)
+{
+ struct ethsw_core *ethsw = port_priv->ethsw_data;
+ struct dpaa2_switch_mirror_entry *tmp;
+ int err;
+
+ list_for_each_entry(tmp, &block->mirror_entries, list) {
+ err = dpsw_if_add_reflection(ethsw->mc_io, 0,
+ ethsw->dpsw_handle,
+ port_priv->idx, &tmp->cfg);
+ if (err)
+ goto unwind_add;
+ }
+
+ return 0;
+
+unwind_add:
+ list_for_each_entry(tmp, &block->mirror_entries, list)
+ dpsw_if_remove_reflection(ethsw->mc_io, 0,
+ ethsw->dpsw_handle,
+ port_priv->idx, &tmp->cfg);
+
+ return err;
+}
+
+int dpaa2_switch_block_unoffload_mirror(struct dpaa2_switch_filter_block *block,
+ struct ethsw_port_priv *port_priv)
+{
+ struct ethsw_core *ethsw = port_priv->ethsw_data;
+ struct dpaa2_switch_mirror_entry *tmp;
+ int err;
+
+ list_for_each_entry(tmp, &block->mirror_entries, list) {
+ err = dpsw_if_remove_reflection(ethsw->mc_io, 0,
+ ethsw->dpsw_handle,
+ port_priv->idx, &tmp->cfg);
+ if (err)
+ goto unwind_remove;
+ }
+
+ return 0;
+
+unwind_remove:
+ list_for_each_entry(tmp, &block->mirror_entries, list)
+ dpsw_if_add_reflection(ethsw->mc_io, 0, ethsw->dpsw_handle,
+ port_priv->idx, &tmp->cfg);
+
+ return err;
+}
+
+int dpaa2_switch_cls_matchall_destroy(struct dpaa2_switch_filter_block *block,
+ struct tc_cls_matchall_offload *cls)
+{
+ struct dpaa2_switch_mirror_entry *mirror_entry;
+ struct dpaa2_switch_acl_entry *acl_entry;
+
+ /* If this filter is a an ACL one, remove it */
+ acl_entry = dpaa2_switch_acl_tbl_find_entry_by_cookie(block,
+ cls->cookie);
+ if (acl_entry)
+ return dpaa2_switch_acl_tbl_remove_entry(block,
+ acl_entry);
+
+ /* If not, then it has to be a mirror */
+ mirror_entry = dpaa2_switch_mirror_find_entry_by_cookie(block,
+ cls->cookie);
+ if (mirror_entry)
+ return dpaa2_switch_block_remove_mirror(block,
+ mirror_entry);
+
+ return 0;
+}