* [PATCH net-next 1/4] net/mlx5: HWS, rework the check if matcher size can be increased
2025-01-14 13:06 [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering Tariq Toukan
@ 2025-01-14 13:06 ` Tariq Toukan
2025-01-14 13:06 ` [PATCH net-next 2/4] net/mlx5e: CT: Add initial support for Hardware Steering Tariq Toukan
` (3 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Tariq Toukan @ 2025-01-14 13:06 UTC (permalink / raw)
To: David S. Miller, Jakub Kicinski, Paolo Abeni, Eric Dumazet,
Andrew Lunn
Cc: netdev, Saeed Mahameed, Gal Pressman, Leon Romanovsky, Mark Bloch,
Moshe Shemesh, Yevgeny Kliteynik, Tariq Toukan
From: Yevgeny Kliteynik <kliteyn@nvidia.com>
When checking if the matcher size can be increased, check both
match and action RTCs. Also, consider the increasing step - check
that it won't cause the new matcher size to become unsupported.
Additionally, since we're using '+ 1' for action RTC size yet
again, define it as macro and use in all the required places.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../mellanox/mlx5/core/steering/hws/bwc.c | 18 ++++++++++++++++--
.../mellanox/mlx5/core/steering/hws/matcher.c | 6 ++++--
.../mellanox/mlx5/core/steering/hws/matcher.h | 5 +++++
3 files changed, 25 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c
index a8d886e92144..3dbd4efa21a2 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c
@@ -468,8 +468,22 @@ hws_bwc_matcher_size_maxed_out(struct mlx5hws_bwc_matcher *bwc_matcher)
{
struct mlx5hws_cmd_query_caps *caps = bwc_matcher->matcher->tbl->ctx->caps;
- return bwc_matcher->size_log + MLX5HWS_MATCHER_ASSURED_MAIN_TBL_DEPTH >=
- caps->ste_alloc_log_max - 1;
+ /* check the match RTC size */
+ if ((bwc_matcher->size_log +
+ MLX5HWS_MATCHER_ASSURED_MAIN_TBL_DEPTH +
+ MLX5HWS_BWC_MATCHER_SIZE_LOG_STEP) >
+ (caps->ste_alloc_log_max - 1))
+ return true;
+
+ /* check the action RTC size */
+ if ((bwc_matcher->size_log +
+ MLX5HWS_BWC_MATCHER_SIZE_LOG_STEP +
+ ilog2(roundup_pow_of_two(bwc_matcher->matcher->action_ste.max_stes)) +
+ MLX5HWS_MATCHER_ACTION_RTC_UPDATE_MULT) >
+ (caps->ste_alloc_log_max - 1))
+ return true;
+
+ return false;
}
static bool
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.c
index 80157a29a076..b61864b32053 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.c
@@ -289,7 +289,8 @@ static int hws_matcher_create_rtc(struct mlx5hws_matcher *matcher,
* (2 to support writing new STEs for update rule))
*/
ste->order = ilog2(roundup_pow_of_two(action_ste->max_stes)) +
- attr->table.sz_row_log + 1;
+ attr->table.sz_row_log +
+ MLX5HWS_MATCHER_ACTION_RTC_UPDATE_MULT;
rtc_attr.log_size = ste->order;
rtc_attr.log_depth = 0;
rtc_attr.update_index_mode = MLX5_IFC_RTC_STE_UPDATE_MODE_BY_OFFSET;
@@ -561,7 +562,8 @@ static int hws_matcher_bind_at(struct mlx5hws_matcher *matcher)
pool_attr.flags = MLX5HWS_POOL_FLAGS_FOR_STE_ACTION_POOL;
/* Pool size is similar to action RTC size */
pool_attr.alloc_log_sz = ilog2(roundup_pow_of_two(action_ste->max_stes)) +
- matcher->attr.table.sz_row_log + 1;
+ matcher->attr.table.sz_row_log +
+ MLX5HWS_MATCHER_ACTION_RTC_UPDATE_MULT;
hws_matcher_set_pool_attr(&pool_attr, matcher);
action_ste->pool = mlx5hws_pool_create(ctx, &pool_attr);
if (!action_ste->pool) {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.h b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.h
index cff4ae854a79..020de70270c5 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/matcher.h
@@ -18,6 +18,11 @@
/* Required depth of the main large table */
#define MLX5HWS_MATCHER_ASSURED_MAIN_TBL_DEPTH 2
+/* Action RTC size multiplier that is required in order
+ * to support rule update for rules with action STEs.
+ */
+#define MLX5HWS_MATCHER_ACTION_RTC_UPDATE_MULT 1
+
enum mlx5hws_matcher_offset {
MLX5HWS_MATCHER_OFFSET_TAG_DW1 = 12,
MLX5HWS_MATCHER_OFFSET_TAG_DW0 = 13,
--
2.45.0
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH net-next 2/4] net/mlx5e: CT: Add initial support for Hardware Steering
2025-01-14 13:06 [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering Tariq Toukan
2025-01-14 13:06 ` [PATCH net-next 1/4] net/mlx5: HWS, rework the check if matcher size can be increased Tariq Toukan
@ 2025-01-14 13:06 ` Tariq Toukan
2025-01-14 13:06 ` [PATCH net-next 3/4] net/mlx5e: CT: Make mlx5_ct_fs_smfs_ct_validate_flow_rule reusable Tariq Toukan
` (2 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Tariq Toukan @ 2025-01-14 13:06 UTC (permalink / raw)
To: David S. Miller, Jakub Kicinski, Paolo Abeni, Eric Dumazet,
Andrew Lunn
Cc: netdev, Saeed Mahameed, Gal Pressman, Leon Romanovsky, Mark Bloch,
Moshe Shemesh, Yevgeny Kliteynik, Cosmin Ratiu, Jianbo Liu,
Tariq Toukan
From: Cosmin Ratiu <cratiu@nvidia.com>
Connection tracking can offload tuple matches to the NIC either via
firmware commands (when the steering mode is dmfs or offload support is
disabled due to eswitch being set to legacy) or via software-managed
flow steering (smfs).
This commit adds stub operations for a third mode, hardware-managed flow
steering. This is enabled when both CONFIG_MLX5_TC_CT and
CONFIG_MLX5_HW_STEERING are enabled.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Jianbo Liu <jianbol@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../net/ethernet/mellanox/mlx5/core/Makefile | 1 +
.../ethernet/mellanox/mlx5/core/en/tc/ct_fs.h | 10 ++++
.../mellanox/mlx5/core/en/tc/ct_fs_hmfs.c | 47 +++++++++++++++++++
.../ethernet/mellanox/mlx5/core/en/tc_ct.c | 17 +++++--
4 files changed, 71 insertions(+), 4 deletions(-)
create mode 100644 drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/Makefile b/drivers/net/ethernet/mellanox/mlx5/core/Makefile
index d9a8817bb33c..568bbe5f83f5 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/Makefile
+++ b/drivers/net/ethernet/mellanox/mlx5/core/Makefile
@@ -60,6 +60,7 @@ mlx5_core-$(CONFIG_MLX5_CLS_ACT) += en/tc/act/act.o en/tc/act/drop.o en/tc/a
ifneq ($(CONFIG_MLX5_TC_CT),)
mlx5_core-y += en/tc_ct.o en/tc/ct_fs_dmfs.o
mlx5_core-$(CONFIG_MLX5_SW_STEERING) += en/tc/ct_fs_smfs.o
+ mlx5_core-$(CONFIG_MLX5_HW_STEERING) += en/tc/ct_fs_hmfs.o
endif
mlx5_core-$(CONFIG_MLX5_TC_SAMPLE) += en/tc/sample.o
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs.h b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs.h
index 62b3f7ff5562..e5b30801314b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs.h
@@ -48,4 +48,14 @@ mlx5_ct_fs_smfs_ops_get(void)
}
#endif /* IS_ENABLED(CONFIG_MLX5_SW_STEERING) */
+#if IS_ENABLED(CONFIG_MLX5_HW_STEERING)
+struct mlx5_ct_fs_ops *mlx5_ct_fs_hmfs_ops_get(void);
+#else
+static inline struct mlx5_ct_fs_ops *
+mlx5_ct_fs_hmfs_ops_get(void)
+{
+ return NULL;
+}
+#endif /* IS_ENABLED(CONFIG_MLX5_SW_STEERING) */
+
#endif /* __MLX5_EN_TC_CT_FS_H__ */
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c
new file mode 100644
index 000000000000..be1a36d1d778
--- /dev/null
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/* Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. */
+
+#include "en_tc.h"
+#include "en/tc_ct.h"
+#include "en/tc/ct_fs.h"
+
+static int mlx5_ct_fs_hmfs_init(struct mlx5_ct_fs *fs, struct mlx5_flow_table *ct,
+ struct mlx5_flow_table *ct_nat, struct mlx5_flow_table *post_ct)
+{
+ return 0;
+}
+
+static void mlx5_ct_fs_hmfs_destroy(struct mlx5_ct_fs *fs)
+{
+}
+
+static struct mlx5_ct_fs_rule *
+mlx5_ct_fs_hmfs_ct_rule_add(struct mlx5_ct_fs *fs, struct mlx5_flow_spec *spec,
+ struct mlx5_flow_attr *attr, struct flow_rule *flow_rule)
+{
+ return ERR_PTR(-EOPNOTSUPP);
+}
+
+static void mlx5_ct_fs_hmfs_ct_rule_del(struct mlx5_ct_fs *fs, struct mlx5_ct_fs_rule *fs_rule)
+{
+}
+
+static int mlx5_ct_fs_hmfs_ct_rule_update(struct mlx5_ct_fs *fs, struct mlx5_ct_fs_rule *fs_rule,
+ struct mlx5_flow_spec *spec, struct mlx5_flow_attr *attr)
+{
+ return -EOPNOTSUPP;
+}
+
+static struct mlx5_ct_fs_ops hmfs_ops = {
+ .ct_rule_add = mlx5_ct_fs_hmfs_ct_rule_add,
+ .ct_rule_del = mlx5_ct_fs_hmfs_ct_rule_del,
+ .ct_rule_update = mlx5_ct_fs_hmfs_ct_rule_update,
+
+ .init = mlx5_ct_fs_hmfs_init,
+ .destroy = mlx5_ct_fs_hmfs_destroy,
+};
+
+struct mlx5_ct_fs_ops *mlx5_ct_fs_hmfs_ops_get(void)
+{
+ return &hmfs_ops;
+}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
index a84ebac2f011..fec008c540f3 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
@@ -2065,10 +2065,19 @@ mlx5_tc_ct_fs_init(struct mlx5_tc_ct_priv *ct_priv)
struct mlx5_ct_fs_ops *fs_ops = mlx5_ct_fs_dmfs_ops_get();
int err;
- if (ct_priv->ns_type == MLX5_FLOW_NAMESPACE_FDB &&
- ct_priv->dev->priv.steering->mode == MLX5_FLOW_STEERING_MODE_SMFS) {
- ct_dbg("Using SMFS ct flow steering provider");
- fs_ops = mlx5_ct_fs_smfs_ops_get();
+ if (ct_priv->ns_type == MLX5_FLOW_NAMESPACE_FDB) {
+ if (ct_priv->dev->priv.steering->mode == MLX5_FLOW_STEERING_MODE_HMFS) {
+ ct_dbg("Using HMFS ct flow steering provider");
+ fs_ops = mlx5_ct_fs_hmfs_ops_get();
+ } else if (ct_priv->dev->priv.steering->mode == MLX5_FLOW_STEERING_MODE_SMFS) {
+ ct_dbg("Using SMFS ct flow steering provider");
+ fs_ops = mlx5_ct_fs_smfs_ops_get();
+ }
+
+ if (!fs_ops) {
+ ct_dbg("Requested flow steering mode is not enabled.");
+ return -EOPNOTSUPP;
+ }
}
ct_priv->fs = kzalloc(sizeof(*ct_priv->fs) + fs_ops->priv_size, GFP_KERNEL);
--
2.45.0
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH net-next 3/4] net/mlx5e: CT: Make mlx5_ct_fs_smfs_ct_validate_flow_rule reusable
2025-01-14 13:06 [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering Tariq Toukan
2025-01-14 13:06 ` [PATCH net-next 1/4] net/mlx5: HWS, rework the check if matcher size can be increased Tariq Toukan
2025-01-14 13:06 ` [PATCH net-next 2/4] net/mlx5e: CT: Add initial support for Hardware Steering Tariq Toukan
@ 2025-01-14 13:06 ` Tariq Toukan
2025-01-14 13:06 ` [PATCH net-next 4/4] net/mlx5e: CT: Offload connections with hardware steering rules Tariq Toukan
2025-01-16 3:30 ` [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering patchwork-bot+netdevbpf
4 siblings, 0 replies; 6+ messages in thread
From: Tariq Toukan @ 2025-01-14 13:06 UTC (permalink / raw)
To: David S. Miller, Jakub Kicinski, Paolo Abeni, Eric Dumazet,
Andrew Lunn
Cc: netdev, Saeed Mahameed, Gal Pressman, Leon Romanovsky, Mark Bloch,
Moshe Shemesh, Yevgeny Kliteynik, Cosmin Ratiu, Jianbo Liu,
Tariq Toukan
From: Cosmin Ratiu <cratiu@nvidia.com>
This function checks whether a flow_rule has the right flow dissector
keys and masks used for a connection tracking flow offload. It is
currently used locally by the tc_ct smfs module, but is about to be used
from another place, so this commit moves it to a better place, renames
it to mlx5e_tc_ct_is_valid_flow_rule and drops the unused fs argument.
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Jianbo Liu <jianbol@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../mellanox/mlx5/core/en/tc/ct_fs_smfs.c | 75 +------------------
.../ethernet/mellanox/mlx5/core/en/tc_ct.c | 71 ++++++++++++++++++
.../ethernet/mellanox/mlx5/core/en/tc_ct.h | 10 +++
3 files changed, 82 insertions(+), 74 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_smfs.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_smfs.c
index 45737d039252..0c97c5899904 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_smfs.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_smfs.c
@@ -13,7 +13,6 @@
#define INIT_ERR_PREFIX "ct_fs_smfs init failed"
#define ct_dbg(fmt, args...)\
netdev_dbg(fs->netdev, "ct_fs_smfs debug: " fmt "\n", ##args)
-#define MLX5_CT_TCP_FLAGS_MASK cpu_to_be16(be32_to_cpu(TCP_FLAG_RST | TCP_FLAG_FIN) >> 16)
struct mlx5_ct_fs_smfs_matcher {
struct mlx5dr_matcher *dr_matcher;
@@ -220,78 +219,6 @@ mlx5_ct_fs_smfs_destroy(struct mlx5_ct_fs *fs)
mlx5_smfs_action_destroy(fs_smfs->fwd_action);
}
-static inline bool
-mlx5_tc_ct_valid_used_dissector_keys(const u64 used_keys)
-{
-#define DISS_BIT(name) BIT_ULL(FLOW_DISSECTOR_KEY_ ## name)
- const u64 basic_keys = DISS_BIT(BASIC) | DISS_BIT(CONTROL) |
- DISS_BIT(META);
- const u64 ipv4_tcp = basic_keys | DISS_BIT(IPV4_ADDRS) |
- DISS_BIT(PORTS) | DISS_BIT(TCP);
- const u64 ipv6_tcp = basic_keys | DISS_BIT(IPV6_ADDRS) |
- DISS_BIT(PORTS) | DISS_BIT(TCP);
- const u64 ipv4_udp = basic_keys | DISS_BIT(IPV4_ADDRS) |
- DISS_BIT(PORTS);
- const u64 ipv6_udp = basic_keys | DISS_BIT(IPV6_ADDRS) |
- DISS_BIT(PORTS);
- const u64 ipv4_gre = basic_keys | DISS_BIT(IPV4_ADDRS);
- const u64 ipv6_gre = basic_keys | DISS_BIT(IPV6_ADDRS);
-
- return (used_keys == ipv4_tcp || used_keys == ipv4_udp || used_keys == ipv6_tcp ||
- used_keys == ipv6_udp || used_keys == ipv4_gre || used_keys == ipv6_gre);
-}
-
-static bool
-mlx5_ct_fs_smfs_ct_validate_flow_rule(struct mlx5_ct_fs *fs, struct flow_rule *flow_rule)
-{
- struct flow_match_ipv4_addrs ipv4_addrs;
- struct flow_match_ipv6_addrs ipv6_addrs;
- struct flow_match_control control;
- struct flow_match_basic basic;
- struct flow_match_ports ports;
- struct flow_match_tcp tcp;
-
- if (!mlx5_tc_ct_valid_used_dissector_keys(flow_rule->match.dissector->used_keys)) {
- ct_dbg("rule uses unexpected dissectors (0x%016llx)",
- flow_rule->match.dissector->used_keys);
- return false;
- }
-
- flow_rule_match_basic(flow_rule, &basic);
- flow_rule_match_control(flow_rule, &control);
- flow_rule_match_ipv4_addrs(flow_rule, &ipv4_addrs);
- flow_rule_match_ipv6_addrs(flow_rule, &ipv6_addrs);
- if (basic.key->ip_proto != IPPROTO_GRE)
- flow_rule_match_ports(flow_rule, &ports);
- if (basic.key->ip_proto == IPPROTO_TCP)
- flow_rule_match_tcp(flow_rule, &tcp);
-
- if (basic.mask->n_proto != htons(0xFFFF) ||
- (basic.key->n_proto != htons(ETH_P_IP) && basic.key->n_proto != htons(ETH_P_IPV6)) ||
- basic.mask->ip_proto != 0xFF ||
- (basic.key->ip_proto != IPPROTO_UDP && basic.key->ip_proto != IPPROTO_TCP &&
- basic.key->ip_proto != IPPROTO_GRE)) {
- ct_dbg("rule uses unexpected basic match (n_proto 0x%04x/0x%04x, ip_proto 0x%02x/0x%02x)",
- ntohs(basic.key->n_proto), ntohs(basic.mask->n_proto),
- basic.key->ip_proto, basic.mask->ip_proto);
- return false;
- }
-
- if (basic.key->ip_proto != IPPROTO_GRE &&
- (ports.mask->src != htons(0xFFFF) || ports.mask->dst != htons(0xFFFF))) {
- ct_dbg("rule uses ports match (src 0x%04x, dst 0x%04x)",
- ports.mask->src, ports.mask->dst);
- return false;
- }
-
- if (basic.key->ip_proto == IPPROTO_TCP && tcp.mask->flags != MLX5_CT_TCP_FLAGS_MASK) {
- ct_dbg("rule uses unexpected tcp match (flags 0x%02x)", tcp.mask->flags);
- return false;
- }
-
- return true;
-}
-
static struct mlx5_ct_fs_rule *
mlx5_ct_fs_smfs_ct_rule_add(struct mlx5_ct_fs *fs, struct mlx5_flow_spec *spec,
struct mlx5_flow_attr *attr, struct flow_rule *flow_rule)
@@ -304,7 +231,7 @@ mlx5_ct_fs_smfs_ct_rule_add(struct mlx5_ct_fs *fs, struct mlx5_flow_spec *spec,
int num_actions = 0, err;
bool nat, tcp, ipv4, gre;
- if (!mlx5_ct_fs_smfs_ct_validate_flow_rule(fs, flow_rule))
+ if (!mlx5e_tc_ct_is_valid_flow_rule(fs->netdev, flow_rule))
return ERR_PTR(-EOPNOTSUPP);
smfs_rule = kzalloc(sizeof(*smfs_rule), GFP_KERNEL);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
index fec008c540f3..a065e8fafb1d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
@@ -2430,3 +2430,74 @@ mlx5e_tc_ct_restore_flow(struct mlx5_tc_ct_priv *ct_priv,
atomic_inc(&ct_priv->debugfs.stats.rx_dropped);
return false;
}
+
+static bool mlx5e_tc_ct_valid_used_dissector_keys(const u64 used_keys)
+{
+#define DISS_BIT(name) BIT_ULL(FLOW_DISSECTOR_KEY_ ## name)
+ const u64 basic_keys = DISS_BIT(BASIC) | DISS_BIT(CONTROL) |
+ DISS_BIT(META);
+ const u64 ipv4_tcp = basic_keys | DISS_BIT(IPV4_ADDRS) |
+ DISS_BIT(PORTS) | DISS_BIT(TCP);
+ const u64 ipv6_tcp = basic_keys | DISS_BIT(IPV6_ADDRS) |
+ DISS_BIT(PORTS) | DISS_BIT(TCP);
+ const u64 ipv4_udp = basic_keys | DISS_BIT(IPV4_ADDRS) |
+ DISS_BIT(PORTS);
+ const u64 ipv6_udp = basic_keys | DISS_BIT(IPV6_ADDRS) |
+ DISS_BIT(PORTS);
+ const u64 ipv4_gre = basic_keys | DISS_BIT(IPV4_ADDRS);
+ const u64 ipv6_gre = basic_keys | DISS_BIT(IPV6_ADDRS);
+
+ return (used_keys == ipv4_tcp || used_keys == ipv4_udp || used_keys == ipv6_tcp ||
+ used_keys == ipv6_udp || used_keys == ipv4_gre || used_keys == ipv6_gre);
+}
+
+bool mlx5e_tc_ct_is_valid_flow_rule(const struct net_device *dev, struct flow_rule *flow_rule)
+{
+ struct flow_match_ipv4_addrs ipv4_addrs;
+ struct flow_match_ipv6_addrs ipv6_addrs;
+ struct flow_match_control control;
+ struct flow_match_basic basic;
+ struct flow_match_ports ports;
+ struct flow_match_tcp tcp;
+
+ if (!mlx5e_tc_ct_valid_used_dissector_keys(flow_rule->match.dissector->used_keys)) {
+ netdev_dbg(dev, "ct_debug: rule uses unexpected dissectors (0x%016llx)",
+ flow_rule->match.dissector->used_keys);
+ return false;
+ }
+
+ flow_rule_match_basic(flow_rule, &basic);
+ flow_rule_match_control(flow_rule, &control);
+ flow_rule_match_ipv4_addrs(flow_rule, &ipv4_addrs);
+ flow_rule_match_ipv6_addrs(flow_rule, &ipv6_addrs);
+ if (basic.key->ip_proto != IPPROTO_GRE)
+ flow_rule_match_ports(flow_rule, &ports);
+ if (basic.key->ip_proto == IPPROTO_TCP)
+ flow_rule_match_tcp(flow_rule, &tcp);
+
+ if (basic.mask->n_proto != htons(0xFFFF) ||
+ (basic.key->n_proto != htons(ETH_P_IP) && basic.key->n_proto != htons(ETH_P_IPV6)) ||
+ basic.mask->ip_proto != 0xFF ||
+ (basic.key->ip_proto != IPPROTO_UDP && basic.key->ip_proto != IPPROTO_TCP &&
+ basic.key->ip_proto != IPPROTO_GRE)) {
+ netdev_dbg(dev, "ct_debug: rule uses unexpected basic match (n_proto 0x%04x/0x%04x, ip_proto 0x%02x/0x%02x)",
+ ntohs(basic.key->n_proto), ntohs(basic.mask->n_proto),
+ basic.key->ip_proto, basic.mask->ip_proto);
+ return false;
+ }
+
+ if (basic.key->ip_proto != IPPROTO_GRE &&
+ (ports.mask->src != htons(0xFFFF) || ports.mask->dst != htons(0xFFFF))) {
+ netdev_dbg(dev, "ct_debug: rule uses ports match (src 0x%04x, dst 0x%04x)",
+ ports.mask->src, ports.mask->dst);
+ return false;
+ }
+
+ if (basic.key->ip_proto == IPPROTO_TCP && tcp.mask->flags != MLX5_CT_TCP_FLAGS_MASK) {
+ netdev_dbg(dev, "ct_debug: rule uses unexpected tcp match (flags 0x%02x)",
+ tcp.mask->flags);
+ return false;
+ }
+
+ return true;
+}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.h b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.h
index b66c5f98067f..5e9dbdd4a5e9 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.h
@@ -128,6 +128,9 @@ bool
mlx5e_tc_ct_restore_flow(struct mlx5_tc_ct_priv *ct_priv,
struct sk_buff *skb, u8 zone_restore_id);
+#define MLX5_CT_TCP_FLAGS_MASK cpu_to_be16(be32_to_cpu(TCP_FLAG_RST | TCP_FLAG_FIN) >> 16)
+bool mlx5e_tc_ct_is_valid_flow_rule(const struct net_device *dev, struct flow_rule *flow_rule);
+
#else /* CONFIG_MLX5_TC_CT */
static inline struct mlx5_tc_ct_priv *
@@ -202,5 +205,12 @@ mlx5e_tc_ct_restore_flow(struct mlx5_tc_ct_priv *ct_priv,
return false;
}
+static inline bool
+mlx5e_tc_ct_is_valid_flow_rule(const struct net_device *dev,
+ struct flow_rule *flow_rule)
+{
+ return false;
+}
+
#endif /* !IS_ENABLED(CONFIG_MLX5_TC_CT) */
#endif /* __MLX5_EN_TC_CT_H__ */
--
2.45.0
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH net-next 4/4] net/mlx5e: CT: Offload connections with hardware steering rules
2025-01-14 13:06 [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering Tariq Toukan
` (2 preceding siblings ...)
2025-01-14 13:06 ` [PATCH net-next 3/4] net/mlx5e: CT: Make mlx5_ct_fs_smfs_ct_validate_flow_rule reusable Tariq Toukan
@ 2025-01-14 13:06 ` Tariq Toukan
2025-01-16 3:30 ` [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering patchwork-bot+netdevbpf
4 siblings, 0 replies; 6+ messages in thread
From: Tariq Toukan @ 2025-01-14 13:06 UTC (permalink / raw)
To: David S. Miller, Jakub Kicinski, Paolo Abeni, Eric Dumazet,
Andrew Lunn
Cc: netdev, Saeed Mahameed, Gal Pressman, Leon Romanovsky, Mark Bloch,
Moshe Shemesh, Yevgeny Kliteynik, Cosmin Ratiu, Jianbo Liu,
Tariq Toukan
From: Cosmin Ratiu <cratiu@nvidia.com>
This is modeled similar to how software steering works:
- a reference-counted matcher is maintained for each
combination of nat/no_nat x ipv4/ipv6 x tcp/udp/gre.
- adding a rule involves finding+referencing or creating a corresponding
matcher, then actually adding a rule.
- updating rules is implemented using the bwc_rule update API, which can
change a rule's actions without touching the match value.
By using a T-Rex traffic generator to initiate multi-million UDP flows
per second, a kernel running with these patches on the RX side was able
to offload ~600K flows per second, which is about ~7x larger than what
software steering could do on the same hardware (256-thread AMD EPYC,
512 GB RAM, ConnectX-7 b2b).
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Jianbo Liu <jianbol@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../mellanox/mlx5/core/en/tc/ct_fs_hmfs.c | 249 +++++++++++++++++-
1 file changed, 247 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c
index be1a36d1d778..a4263137fef5 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/ct_fs_hmfs.c
@@ -3,33 +3,276 @@
#include "en_tc.h"
#include "en/tc_ct.h"
+#include "en/tc_priv.h"
#include "en/tc/ct_fs.h"
+#include "fs_core.h"
+#include "steering/hws/fs_hws_pools.h"
+#include "steering/hws/mlx5hws.h"
+#include "steering/hws/table.h"
+
+struct mlx5_ct_fs_hmfs_matcher {
+ struct mlx5hws_bwc_matcher *hws_bwc_matcher;
+ refcount_t ref;
+};
+
+/* We need {ipv4, ipv6} x {tcp, udp, gre} matchers. */
+#define NUM_MATCHERS (2 * 3)
+
+struct mlx5_ct_fs_hmfs {
+ struct mlx5hws_table *ct_tbl;
+ struct mlx5hws_table *ct_nat_tbl;
+ struct mlx5_flow_table *ct_nat;
+ struct mlx5hws_action *fwd_action;
+ struct mlx5hws_action *last_action;
+ struct mlx5hws_context *ctx;
+ struct mutex lock; /* Guards matchers */
+ struct mlx5_ct_fs_hmfs_matcher matchers[NUM_MATCHERS];
+ struct mlx5_ct_fs_hmfs_matcher matchers_nat[NUM_MATCHERS];
+};
+
+struct mlx5_ct_fs_hmfs_rule {
+ struct mlx5_ct_fs_rule fs_rule;
+ struct mlx5hws_bwc_rule *hws_bwc_rule;
+ struct mlx5_ct_fs_hmfs_matcher *hmfs_matcher;
+ struct mlx5_fc *counter;
+};
+
+static u32 get_matcher_idx(bool ipv4, bool tcp, bool gre)
+{
+ return ipv4 * 3 + tcp * 2 + gre;
+}
static int mlx5_ct_fs_hmfs_init(struct mlx5_ct_fs *fs, struct mlx5_flow_table *ct,
struct mlx5_flow_table *ct_nat, struct mlx5_flow_table *post_ct)
{
+ u32 flags = MLX5HWS_ACTION_FLAG_HWS_FDB | MLX5HWS_ACTION_FLAG_SHARED;
+ struct mlx5hws_table *ct_tbl, *ct_nat_tbl, *post_ct_tbl;
+ struct mlx5_ct_fs_hmfs *fs_hmfs = mlx5_ct_fs_priv(fs);
+
+ ct_tbl = ct->fs_hws_table.hws_table;
+ ct_nat_tbl = ct_nat->fs_hws_table.hws_table;
+ post_ct_tbl = post_ct->fs_hws_table.hws_table;
+ fs_hmfs->ct_nat = ct_nat;
+
+ if (!ct_tbl || !ct_nat_tbl || !post_ct_tbl) {
+ netdev_warn(fs->netdev, "ct_fs_hmfs: failed to init, missing backing hws tables");
+ return -EOPNOTSUPP;
+ }
+
+ netdev_dbg(fs->netdev, "using hmfs steering");
+
+ fs_hmfs->ct_tbl = ct_tbl;
+ fs_hmfs->ct_nat_tbl = ct_nat_tbl;
+ fs_hmfs->ctx = ct_tbl->ctx;
+ mutex_init(&fs_hmfs->lock);
+
+ fs_hmfs->fwd_action = mlx5hws_action_create_dest_table(ct_tbl->ctx, post_ct_tbl, flags);
+ if (!fs_hmfs->fwd_action) {
+ netdev_warn(fs->netdev, "ct_fs_hmfs: failed to create fwd action\n");
+ return -EINVAL;
+ }
+ fs_hmfs->last_action = mlx5hws_action_create_last(ct_tbl->ctx, flags);
+ if (!fs_hmfs->last_action) {
+ netdev_warn(fs->netdev, "ct_fs_hmfs: failed to create last action\n");
+ mlx5hws_action_destroy(fs_hmfs->fwd_action);
+ return -EINVAL;
+ }
+
return 0;
}
static void mlx5_ct_fs_hmfs_destroy(struct mlx5_ct_fs *fs)
{
+ struct mlx5_ct_fs_hmfs *fs_hmfs = mlx5_ct_fs_priv(fs);
+
+ mlx5hws_action_destroy(fs_hmfs->last_action);
+ mlx5hws_action_destroy(fs_hmfs->fwd_action);
+}
+
+static struct mlx5hws_bwc_matcher *
+mlx5_ct_fs_hmfs_matcher_create(struct mlx5_ct_fs *fs, struct mlx5hws_table *tbl,
+ struct mlx5_flow_spec *spec, bool ipv4, bool tcp, bool gre)
+{
+ u8 match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS_2 | MLX5_MATCH_OUTER_HEADERS;
+ struct mlx5hws_match_parameters mask = {
+ .match_buf = spec->match_criteria,
+ .match_sz = sizeof(spec->match_criteria),
+ };
+ u32 priority = get_matcher_idx(ipv4, tcp, gre); /* Static priority based on params. */
+ struct mlx5hws_bwc_matcher *hws_bwc_matcher;
+
+ hws_bwc_matcher = mlx5hws_bwc_matcher_create(tbl, priority, match_criteria_enable, &mask);
+ if (!hws_bwc_matcher)
+ return ERR_PTR(-EINVAL);
+
+ return hws_bwc_matcher;
+}
+
+static struct mlx5_ct_fs_hmfs_matcher *
+mlx5_ct_fs_hmfs_matcher_get(struct mlx5_ct_fs *fs, struct mlx5_flow_spec *spec,
+ bool nat, bool ipv4, bool tcp, bool gre)
+{
+ struct mlx5_ct_fs_hmfs *fs_hmfs = mlx5_ct_fs_priv(fs);
+ u32 matcher_idx = get_matcher_idx(ipv4, tcp, gre);
+ struct mlx5_ct_fs_hmfs_matcher *hmfs_matcher;
+ struct mlx5hws_bwc_matcher *hws_bwc_matcher;
+ struct mlx5hws_table *tbl;
+
+ hmfs_matcher = nat ?
+ (fs_hmfs->matchers_nat + matcher_idx) :
+ (fs_hmfs->matchers + matcher_idx);
+
+ if (refcount_inc_not_zero(&hmfs_matcher->ref))
+ return hmfs_matcher;
+
+ mutex_lock(&fs_hmfs->lock);
+
+ /* Retry with lock, as the matcher might be already created by another cpu. */
+ if (refcount_inc_not_zero(&hmfs_matcher->ref))
+ goto out_unlock;
+
+ tbl = nat ? fs_hmfs->ct_nat_tbl : fs_hmfs->ct_tbl;
+
+ hws_bwc_matcher = mlx5_ct_fs_hmfs_matcher_create(fs, tbl, spec, ipv4, tcp, gre);
+ if (IS_ERR(hws_bwc_matcher)) {
+ netdev_warn(fs->netdev,
+ "ct_fs_hmfs: failed to create bwc matcher (nat %d, ipv4 %d, tcp %d, gre %d), err: %ld\n",
+ nat, ipv4, tcp, gre, PTR_ERR(hws_bwc_matcher));
+
+ hmfs_matcher = ERR_CAST(hws_bwc_matcher);
+ goto out_unlock;
+ }
+
+ hmfs_matcher->hws_bwc_matcher = hws_bwc_matcher;
+ refcount_set(&hmfs_matcher->ref, 1);
+
+out_unlock:
+ mutex_unlock(&fs_hmfs->lock);
+ return hmfs_matcher;
+}
+
+static void
+mlx5_ct_fs_hmfs_matcher_put(struct mlx5_ct_fs *fs, struct mlx5_ct_fs_hmfs_matcher *hmfs_matcher)
+{
+ struct mlx5_ct_fs_hmfs *fs_hmfs = mlx5_ct_fs_priv(fs);
+
+ if (!refcount_dec_and_mutex_lock(&hmfs_matcher->ref, &fs_hmfs->lock))
+ return;
+
+ mlx5hws_bwc_matcher_destroy(hmfs_matcher->hws_bwc_matcher);
+ mutex_unlock(&fs_hmfs->lock);
+}
+
+#define NUM_CT_HMFS_RULES 4
+
+static void mlx5_ct_fs_hmfs_fill_rule_actions(struct mlx5_ct_fs_hmfs *fs_hmfs,
+ struct mlx5_flow_attr *attr,
+ struct mlx5hws_rule_action *rule_actions)
+{
+ struct mlx5_fs_hws_action *mh_action = &attr->modify_hdr->fs_hws_action;
+
+ memset(rule_actions, 0, NUM_CT_HMFS_RULES * sizeof(*rule_actions));
+ rule_actions[0].action = mlx5_fc_get_hws_action(fs_hmfs->ctx, attr->counter);
+ /* Modify header is special, it may require extra arguments outside the action itself. */
+ if (mh_action->mh_data) {
+ rule_actions[1].modify_header.offset = mh_action->mh_data->offset;
+ rule_actions[1].modify_header.data = mh_action->mh_data->data;
+ }
+ rule_actions[1].action = mh_action->hws_action;
+ rule_actions[2].action = fs_hmfs->fwd_action;
+ rule_actions[3].action = fs_hmfs->last_action;
}
static struct mlx5_ct_fs_rule *
mlx5_ct_fs_hmfs_ct_rule_add(struct mlx5_ct_fs *fs, struct mlx5_flow_spec *spec,
struct mlx5_flow_attr *attr, struct flow_rule *flow_rule)
{
- return ERR_PTR(-EOPNOTSUPP);
+ struct mlx5hws_rule_action rule_actions[NUM_CT_HMFS_RULES];
+ struct mlx5_ct_fs_hmfs *fs_hmfs = mlx5_ct_fs_priv(fs);
+ struct mlx5hws_match_parameters match_params = {
+ .match_buf = spec->match_value,
+ .match_sz = ARRAY_SIZE(spec->match_value),
+ };
+ struct mlx5_ct_fs_hmfs_matcher *hmfs_matcher;
+ struct mlx5_ct_fs_hmfs_rule *hmfs_rule;
+ bool nat, tcp, ipv4, gre;
+ int err;
+
+ if (!mlx5e_tc_ct_is_valid_flow_rule(fs->netdev, flow_rule))
+ return ERR_PTR(-EOPNOTSUPP);
+
+ hmfs_rule = kzalloc(sizeof(*hmfs_rule), GFP_KERNEL);
+ if (!hmfs_rule)
+ return ERR_PTR(-ENOMEM);
+
+ nat = (attr->ft == fs_hmfs->ct_nat);
+ ipv4 = mlx5e_tc_get_ip_version(spec, true) == 4;
+ tcp = MLX5_GET(fte_match_param, spec->match_value,
+ outer_headers.ip_protocol) == IPPROTO_TCP;
+ gre = MLX5_GET(fte_match_param, spec->match_value,
+ outer_headers.ip_protocol) == IPPROTO_GRE;
+
+ hmfs_matcher = mlx5_ct_fs_hmfs_matcher_get(fs, spec, nat, ipv4, tcp, gre);
+ if (IS_ERR(hmfs_matcher)) {
+ err = PTR_ERR(hmfs_matcher);
+ goto err_free_rule;
+ }
+ hmfs_rule->hmfs_matcher = hmfs_matcher;
+
+ mlx5_ct_fs_hmfs_fill_rule_actions(fs_hmfs, attr, rule_actions);
+ hmfs_rule->counter = attr->counter;
+
+ hmfs_rule->hws_bwc_rule =
+ mlx5hws_bwc_rule_create(hmfs_matcher->hws_bwc_matcher, &match_params,
+ spec->flow_context.flow_source, rule_actions);
+ if (!hmfs_rule->hws_bwc_rule) {
+ err = -EINVAL;
+ goto err_put_matcher;
+ }
+
+ return &hmfs_rule->fs_rule;
+
+err_put_matcher:
+ mlx5_fc_put_hws_action(hmfs_rule->counter);
+ mlx5_ct_fs_hmfs_matcher_put(fs, hmfs_matcher);
+err_free_rule:
+ kfree(hmfs_rule);
+ return ERR_PTR(err);
}
static void mlx5_ct_fs_hmfs_ct_rule_del(struct mlx5_ct_fs *fs, struct mlx5_ct_fs_rule *fs_rule)
{
+ struct mlx5_ct_fs_hmfs_rule *hmfs_rule = container_of(fs_rule,
+ struct mlx5_ct_fs_hmfs_rule,
+ fs_rule);
+ mlx5hws_bwc_rule_destroy(hmfs_rule->hws_bwc_rule);
+ mlx5_fc_put_hws_action(hmfs_rule->counter);
+ mlx5_ct_fs_hmfs_matcher_put(fs, hmfs_rule->hmfs_matcher);
+ kfree(hmfs_rule);
}
static int mlx5_ct_fs_hmfs_ct_rule_update(struct mlx5_ct_fs *fs, struct mlx5_ct_fs_rule *fs_rule,
struct mlx5_flow_spec *spec, struct mlx5_flow_attr *attr)
{
- return -EOPNOTSUPP;
+ struct mlx5_ct_fs_hmfs_rule *hmfs_rule = container_of(fs_rule,
+ struct mlx5_ct_fs_hmfs_rule,
+ fs_rule);
+ struct mlx5hws_rule_action rule_actions[NUM_CT_HMFS_RULES];
+ struct mlx5_ct_fs_hmfs *fs_hmfs = mlx5_ct_fs_priv(fs);
+ int err;
+
+ mlx5_ct_fs_hmfs_fill_rule_actions(fs_hmfs, attr, rule_actions);
+
+ err = mlx5hws_bwc_rule_action_update(hmfs_rule->hws_bwc_rule, rule_actions);
+ if (err) {
+ mlx5_fc_put_hws_action(attr->counter);
+ return err;
+ }
+
+ mlx5_fc_put_hws_action(hmfs_rule->counter);
+ hmfs_rule->counter = attr->counter;
+
+ return 0;
}
static struct mlx5_ct_fs_ops hmfs_ops = {
@@ -39,6 +282,8 @@ static struct mlx5_ct_fs_ops hmfs_ops = {
.init = mlx5_ct_fs_hmfs_init,
.destroy = mlx5_ct_fs_hmfs_destroy,
+
+ .priv_size = sizeof(struct mlx5_ct_fs_hmfs),
};
struct mlx5_ct_fs_ops *mlx5_ct_fs_hmfs_ops_get(void)
--
2.45.0
^ permalink raw reply related [flat|nested] 6+ messages in thread* Re: [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering
2025-01-14 13:06 [PATCH net-next 0/4] net/mlx5e: CT: Add support for hardware steering Tariq Toukan
` (3 preceding siblings ...)
2025-01-14 13:06 ` [PATCH net-next 4/4] net/mlx5e: CT: Offload connections with hardware steering rules Tariq Toukan
@ 2025-01-16 3:30 ` patchwork-bot+netdevbpf
4 siblings, 0 replies; 6+ messages in thread
From: patchwork-bot+netdevbpf @ 2025-01-16 3:30 UTC (permalink / raw)
To: Tariq Toukan
Cc: davem, kuba, pabeni, edumazet, andrew+netdev, netdev, saeedm, gal,
leonro, mbloch, moshe, kliteyn
Hello:
This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Tue, 14 Jan 2025 15:06:42 +0200 you wrote:
> This series start with one more HWS patch by Yevgeny, followed by
> patches that add support for connection tracking in hardware steering
> mode. It consists of:
> - patch #2 hooks up the CT ops for the new mode in the right places.
> - patch #3 moves a function into a common file, so it can be reused.
> - patch #4 uses the HWS API to implement connection tracking.
>
> [...]
Here is the summary with links:
- [net-next,1/4] net/mlx5: HWS, rework the check if matcher size can be increased
https://git.kernel.org/netdev/net-next/c/af02dbfe3740
- [net-next,2/4] net/mlx5e: CT: Add initial support for Hardware Steering
https://git.kernel.org/netdev/net-next/c/34eea5b12a10
- [net-next,3/4] net/mlx5e: CT: Make mlx5_ct_fs_smfs_ct_validate_flow_rule reusable
https://git.kernel.org/netdev/net-next/c/554f9773fdee
- [net-next,4/4] net/mlx5e: CT: Offload connections with hardware steering rules
https://git.kernel.org/netdev/net-next/c/066d49c199a6
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply [flat|nested] 6+ messages in thread