diff mbox series

[net-next,08/15] net/mlx5: fs, add support for no append at software level

Message ID 20240911201757.1505453-9-saeed@kernel.org (mailing list archive)
State Accepted
Commit 1217e6989c99c1c4b76866ea395dda757c79deb4
Delegated to: Netdev Maintainers
Headers show
Series [net-next,01/15] net/mlx5: HWS, updated API functions comments to kernel doc | expand

Checks

Context Check Description
netdev/series_format success Pull request is its own cover letter
netdev/tree_selection success Clearly marked for net-next
netdev/ynl success Generated files up to date; no warnings/errors; no diff in generated;
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 16 this patch: 16
netdev/build_tools success No tools touched, skip
netdev/cc_maintainers success CCed 4 of 4 maintainers
netdev/build_clang success Errors and warnings before: 16 this patch: 16
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 17 this patch: 17
netdev/checkpatch warning WARNING: line length of 83 exceeds 80 columns WARNING: line length of 85 exceeds 80 columns WARNING: line length of 86 exceeds 80 columns WARNING: line length of 87 exceeds 80 columns WARNING: line length of 89 exceeds 80 columns WARNING: line length of 95 exceeds 80 columns
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0
netdev/contest success net-next-2024-09-12--00-00 (tests: 764)

Commit Message

Saeed Mahameed Sept. 11, 2024, 8:17 p.m. UTC
From: Mark Bloch <mbloch@nvidia.com>

Native capability for some steering engines lacks support for adding an
additional match with the same value to the same flow group. To accommodate
the NO APPEND flag in these scenarios, we include the new rule in the
existing flow table entry (fte) without immediate hardware commitment. When
a request is made to delete the corresponding hardware rule, we then commit
the pending rule to hardware.

Only one pending rule is supported because NO_APPEND is primarily used
during replacement operations. In this scenario, a rule is initially added.
When it needs replacement, the new rule is added with NO_APPEND set. Only
after the insertion of the new rule is the original rule deleted.

Signed-off-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/fs_cmd.c  |   2 +-
 .../net/ethernet/mellanox/mlx5/core/fs_core.c | 245 +++++++++++++++++-
 .../net/ethernet/mellanox/mlx5/core/fs_core.h |   7 +
 .../mellanox/mlx5/core/steering/fs_dr.c       |   4 +-
 4 files changed, 242 insertions(+), 16 deletions(-)

Comments

Keller, Jacob E Sept. 12, 2024, 7:10 p.m. UTC | #1
On 9/11/2024 1:17 PM, Saeed Mahameed wrote:
> From: Mark Bloch <mbloch@nvidia.com>
> 
> Native capability for some steering engines lacks support for adding an
> additional match with the same value to the same flow group. To accommodate
> the NO APPEND flag in these scenarios, we include the new rule in the
> existing flow table entry (fte) without immediate hardware commitment. When
> a request is made to delete the corresponding hardware rule, we then commit
> the pending rule to hardware.
> 
> Only one pending rule is supported because NO_APPEND is primarily used
> during replacement operations. In this scenario, a rule is initially added.
> When it needs replacement, the new rule is added with NO_APPEND set. Only
> after the insertion of the new rule is the original rule deleted.
> 
> Signed-off-by: Mark Bloch <mbloch@nvidia.com>
> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
> ---
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
diff mbox series

Patch

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c
index 1a14f8dd878f..676005854dad 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c
@@ -1071,7 +1071,7 @@  static int mlx5_cmd_create_match_definer(struct mlx5_flow_root_namespace *ns,
 static u32 mlx5_cmd_get_capabilities(struct mlx5_flow_root_namespace *ns,
 				     enum fs_flow_table_type ft_type)
 {
-	return 0;
+	return MLX5_FLOW_STEERING_CAP_DUPLICATE_MATCH;
 }
 
 static const struct mlx5_flow_cmds mlx5_flow_cmds = {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
index 57df2a6bcb42..8505d5e241e1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
@@ -613,6 +613,31 @@  static void modify_fte(struct fs_fte *fte)
 	fte->act_dests.modify_mask = 0;
 }
 
+static void del_sw_hw_dup_rule(struct fs_node *node)
+{
+	struct mlx5_flow_rule *rule;
+	struct fs_fte *fte;
+
+	fs_get_obj(rule, node);
+	fs_get_obj(fte, rule->node.parent);
+	trace_mlx5_fs_del_rule(rule);
+
+	if (is_fwd_next_action(rule->sw_action)) {
+		mutex_lock(&rule->dest_attr.ft->lock);
+		list_del(&rule->next_ft);
+		mutex_unlock(&rule->dest_attr.ft->lock);
+	}
+
+	/* If a pending rule is being deleted it means
+	 * this is a NO APPEND rule, so there are no partial deletions,
+	 * all the rules of the mlx5_flow_handle are going to be deleted
+	 * and the rules aren't shared with any other mlx5_flow_handle instance
+	 * so no need to do any bookkeeping like in del_sw_hw_rule().
+	 */
+
+	kfree(rule);
+}
+
 static void del_sw_hw_rule(struct fs_node *node)
 {
 	struct mlx5_flow_rule *rule;
@@ -658,12 +683,33 @@  static void del_sw_hw_rule(struct fs_node *node)
 	kfree(rule);
 }
 
+static void switch_to_pending_act_dests(struct fs_fte *fte)
+{
+	struct fs_node *iter;
+
+	memcpy(&fte->act_dests, &fte->dup->act_dests, sizeof(fte->act_dests));
+
+	list_bulk_move_tail(&fte->node.children,
+			    fte->dup->children.next,
+			    fte->dup->children.prev);
+
+	list_for_each_entry(iter, &fte->node.children, list)
+		iter->del_sw_func = del_sw_hw_rule;
+
+	/* Make sure the fte isn't deleted
+	 * as mlx5_del_flow_rules() decreases the refcount
+	 * of the fte to trigger deletion.
+	 */
+	tree_get_node(&fte->node);
+}
+
 static void del_hw_fte(struct fs_node *node)
 {
 	struct mlx5_flow_root_namespace *root;
 	struct mlx5_flow_table *ft;
 	struct mlx5_flow_group *fg;
 	struct mlx5_core_dev *dev;
+	bool pending_used = false;
 	struct fs_fte *fte;
 	int err;
 
@@ -675,16 +721,33 @@  static void del_hw_fte(struct fs_node *node)
 	WARN_ON(fte->act_dests.dests_size);
 	dev = get_dev(&ft->node);
 	root = find_root(&ft->node);
+
+	if (fte->dup && !list_empty(&fte->dup->children)) {
+		switch_to_pending_act_dests(fte);
+		pending_used = true;
+	} else {
+		/* Avoid double call to del_hw_fte */
+		node->del_hw_func = NULL;
+	}
+
 	if (node->active) {
-		err = root->cmds->delete_fte(root, ft, fte);
-		if (err)
-			mlx5_core_warn(dev,
-				       "flow steering can't delete fte in index %d of flow group id %d\n",
-				       fte->index, fg->id);
-		node->active = false;
+		if (pending_used) {
+			err = root->cmds->update_fte(root, ft, fg,
+						     fte->act_dests.modify_mask, fte);
+			if (err)
+				mlx5_core_warn(dev,
+					       "flow steering can't update to pending rule in index %d of flow group id %d\n",
+					       fte->index, fg->id);
+			fte->act_dests.modify_mask = 0;
+		} else {
+			err = root->cmds->delete_fte(root, ft, fte);
+			if (err)
+				mlx5_core_warn(dev,
+					       "flow steering can't delete fte in index %d of flow group id %d\n",
+					       fte->index, fg->id);
+			node->active = false;
+		}
 	}
-	/* Avoid double call to del_hw_fte */
-	fte->node.del_hw_func = NULL;
 }
 
 static void del_sw_fte(struct fs_node *node)
@@ -702,6 +765,7 @@  static void del_sw_fte(struct fs_node *node)
 				     rhash_fte);
 	WARN_ON(err);
 	ida_free(&fg->fte_allocator, fte->index - fg->start_index);
+	kvfree(fte->dup);
 	kmem_cache_free(steering->ftes_cache, fte);
 }
 
@@ -1105,18 +1169,45 @@  static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 	return err;
 }
 
+static bool rule_is_pending(struct fs_fte *fte, struct mlx5_flow_rule *rule)
+{
+	struct mlx5_flow_rule *tmp_rule;
+	struct fs_node *iter;
+
+	if (!fte->dup || list_empty(&fte->dup->children))
+		return false;
+
+	list_for_each_entry(iter, &fte->dup->children, list) {
+		tmp_rule = container_of(iter, struct mlx5_flow_rule, node);
+
+		if (tmp_rule == rule)
+			return true;
+	}
+
+	return false;
+}
+
 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 					 struct mlx5_flow_destination *dest)
 {
 	struct mlx5_flow_root_namespace *root;
+	struct fs_fte_action *act_dests;
 	struct mlx5_flow_table *ft;
 	struct mlx5_flow_group *fg;
+	bool pending = false;
 	struct fs_fte *fte;
 	int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 	int err = 0;
 
 	fs_get_obj(fte, rule->node.parent);
-	if (!(fte->act_dests.action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
+
+	pending = rule_is_pending(fte, rule);
+	if (pending)
+		act_dests = &fte->dup->act_dests;
+	else
+		act_dests = &fte->act_dests;
+
+	if (!(act_dests->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 		return -EINVAL;
 	down_write_ref_node(&fte->node, false);
 	fs_get_obj(fg, fte->node.parent);
@@ -1124,8 +1215,9 @@  static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 
 	memcpy(&rule->dest_attr, dest, sizeof(*dest));
 	root = find_root(&ft->node);
-	err = root->cmds->update_fte(root, ft, fg,
-				     modify_mask, fte);
+	if (!pending)
+		err = root->cmds->update_fte(root, ft, fg,
+					     modify_mask, fte);
 	up_write_ref_node(&fte->node, false);
 
 	return err;
@@ -1455,6 +1547,16 @@  static struct mlx5_flow_handle *alloc_handle(int num_rules)
 	return handle;
 }
 
+static void destroy_flow_handle_dup(struct mlx5_flow_handle *handle,
+				    int i)
+{
+	for (; --i >= 0;) {
+		list_del(&handle->rule[i]->node.list);
+		kfree(handle->rule[i]);
+	}
+	kfree(handle);
+}
+
 static void destroy_flow_handle(struct fs_fte *fte,
 				struct mlx5_flow_handle *handle,
 				struct mlx5_flow_destination *dest,
@@ -1470,6 +1572,61 @@  static void destroy_flow_handle(struct fs_fte *fte,
 	kfree(handle);
 }
 
+static struct mlx5_flow_handle *
+create_flow_handle_dup(struct list_head *children,
+		       struct mlx5_flow_destination *dest,
+		       int dest_num,
+		       struct fs_fte_action *act_dests)
+{
+	static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
+	static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
+	struct mlx5_flow_rule *rule = NULL;
+	struct mlx5_flow_handle *handle;
+	int i = 0;
+	int type;
+
+	handle = alloc_handle((dest_num) ? dest_num : 1);
+	if (!handle)
+		return NULL;
+
+	do {
+		rule = alloc_rule(dest + i);
+		if (!rule)
+			goto free_rules;
+
+		/* Add dest to dests list- we need flow tables to be in the
+		 * end of the list for forward to next prio rules.
+		 */
+		tree_init_node(&rule->node, NULL, del_sw_hw_dup_rule);
+		if (dest &&
+		    dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
+			list_add(&rule->node.list, children);
+		else
+			list_add_tail(&rule->node.list, children);
+
+		if (dest) {
+			act_dests->dests_size++;
+
+			if (is_fwd_dest_type(dest[i].type))
+				act_dests->fwd_dests++;
+
+			type = dest[i].type ==
+				MLX5_FLOW_DESTINATION_TYPE_COUNTER;
+			act_dests->modify_mask |= type ? count : dst;
+		}
+		handle->rule[i] = rule;
+	} while (++i < dest_num);
+
+	return handle;
+
+free_rules:
+	destroy_flow_handle_dup(handle, i);
+	act_dests->dests_size = 0;
+	act_dests->fwd_dests = 0;
+
+	return NULL;
+}
+
 static struct mlx5_flow_handle *
 create_flow_handle(struct fs_fte *fte,
 		   struct mlx5_flow_destination *dest,
@@ -1963,6 +2120,62 @@  lookup_fte_locked(struct mlx5_flow_group *g,
 	return fte_tmp;
 }
 
+/* Native capability lacks support for adding an additional match with the same value
+ * to the same flow group. To accommodate the NO APPEND flag in these scenarios,
+ * we include the new rule in the existing flow table entry (fte) without immediate
+ * hardware commitment. When a request is made to delete the corresponding hardware rule,
+ * we then commit the pending rule to hardware.
+ */
+static struct mlx5_flow_handle *
+add_rule_dup_match_fte(struct fs_fte *fte,
+		       const struct mlx5_flow_spec *spec,
+		       struct mlx5_flow_act *flow_act,
+		       struct mlx5_flow_destination *dest,
+		       int dest_num)
+{
+	struct mlx5_flow_handle *handle;
+	struct fs_fte_dup *dup;
+	int i = 0;
+
+	if (!fte->dup) {
+		dup = kvzalloc(sizeof(*dup), GFP_KERNEL);
+		if (!dup)
+			return ERR_PTR(-ENOMEM);
+		/* dup will be freed when the fte is freed
+		 * this way we don't allocate / free dup on every rule deletion
+		 * or creation
+		 */
+		INIT_LIST_HEAD(&dup->children);
+		fte->dup = dup;
+	}
+
+	if (!list_empty(&fte->dup->children)) {
+		mlx5_core_warn(get_dev(&fte->node),
+			       "Can have only a single duplicate rule\n");
+
+		return ERR_PTR(-EEXIST);
+	}
+
+	fte->dup->act_dests.action = *flow_act;
+	fte->dup->act_dests.flow_context = spec->flow_context;
+	fte->dup->act_dests.dests_size = 0;
+	fte->dup->act_dests.fwd_dests = 0;
+	fte->dup->act_dests.modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
+
+	handle = create_flow_handle_dup(&fte->dup->children,
+					dest, dest_num,
+					&fte->dup->act_dests);
+	if (!handle)
+		return ERR_PTR(-ENOMEM);
+
+	for (i = 0; i < handle->num_rules; i++) {
+		tree_add_node(&handle->rule[i]->node, &fte->node);
+		trace_mlx5_fs_add_rule(handle->rule[i]);
+	}
+
+	return handle;
+}
+
 static struct mlx5_flow_handle *
 try_add_to_existing_fg(struct mlx5_flow_table *ft,
 		       struct list_head *match_head,
@@ -1973,6 +2186,7 @@  try_add_to_existing_fg(struct mlx5_flow_table *ft,
 		       int ft_version)
 {
 	struct mlx5_flow_steering *steering = get_steering(&ft->node);
+	struct mlx5_flow_root_namespace *root = find_root(&ft->node);
 	struct mlx5_flow_group *g;
 	struct mlx5_flow_handle *rule;
 	struct match_list *iter;
@@ -1986,7 +2200,9 @@  try_add_to_existing_fg(struct mlx5_flow_table *ft,
 		return  ERR_PTR(-ENOMEM);
 
 search_again_locked:
-	if (flow_act->flags & FLOW_ACT_NO_APPEND)
+	if (flow_act->flags & FLOW_ACT_NO_APPEND &&
+	    (root->cmds->get_capabilities(root, root->table_type) &
+	     MLX5_FLOW_STEERING_CAP_DUPLICATE_MATCH))
 		goto skip_search;
 	version = matched_fgs_get_version(match_head);
 	/* Try to find an fte with identical match value and attempt update its
@@ -1999,7 +2215,10 @@  try_add_to_existing_fg(struct mlx5_flow_table *ft,
 		fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
 		if (!fte_tmp)
 			continue;
-		rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
+		if (flow_act->flags & FLOW_ACT_NO_APPEND)
+			rule = add_rule_dup_match_fte(fte_tmp, spec, flow_act, dest, dest_num);
+		else
+			rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
 		/* No error check needed here, because insert_fte() is not called */
 		up_write_ref_node(&fte_tmp->node, false);
 		tree_put_node(&fte_tmp->node, false);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.h b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.h
index 19ba63063d09..964937f17cf5 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.h
@@ -133,6 +133,7 @@  enum mlx5_flow_steering_capabilty {
 	MLX5_FLOW_STEERING_CAP_VLAN_PUSH_ON_RX = 1UL << 0,
 	MLX5_FLOW_STEERING_CAP_VLAN_POP_ON_TX = 1UL << 1,
 	MLX5_FLOW_STEERING_CAP_MATCH_RANGES = 1UL << 2,
+	MLX5_FLOW_STEERING_CAP_DUPLICATE_MATCH = 1UL << 3,
 };
 
 struct mlx5_flow_steering {
@@ -238,12 +239,18 @@  struct fs_fte_action {
 	struct mlx5_flow_act		action;
 };
 
+struct fs_fte_dup {
+	struct list_head children;
+	struct fs_fte_action act_dests;
+};
+
 /* Type of children is mlx5_flow_rule */
 struct fs_fte {
 	struct fs_node			node;
 	struct mlx5_fs_dr_rule		fs_dr_rule;
 	u32				val[MLX5_ST_SZ_DW_MATCH_PARAM];
 	struct fs_fte_action		act_dests;
+	struct fs_fte_dup		*dup;
 	u32				index;
 	enum fs_fte_status		status;
 	struct rhash_head		hash;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/fs_dr.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/fs_dr.c
index 15ef118eade0..833cb68c744f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/fs_dr.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/fs_dr.c
@@ -813,11 +813,11 @@  static int mlx5_cmd_dr_destroy_ns(struct mlx5_flow_root_namespace *ns)
 static u32 mlx5_cmd_dr_get_capabilities(struct mlx5_flow_root_namespace *ns,
 					enum fs_flow_table_type ft_type)
 {
-	u32 steering_caps = 0;
+	u32 steering_caps = MLX5_FLOW_STEERING_CAP_DUPLICATE_MATCH;
 
 	if (ft_type != FS_FT_FDB ||
 	    MLX5_CAP_GEN(ns->dev, steering_format_version) == MLX5_STEERING_FORMAT_CONNECTX_5)
-		return 0;
+		return steering_caps;
 
 	steering_caps |= MLX5_FLOW_STEERING_CAP_VLAN_PUSH_ON_RX;
 	steering_caps |= MLX5_FLOW_STEERING_CAP_VLAN_POP_ON_TX;