mlx5_hw_q_job struct held a reference to temporary flow rule pattern
and contained temporary REPRESENTED_PORT and TAG items structs.
They are used whenever it is required to prepend a flow rule pattern,
provided by the application with one of such items.
If prepending is required, then flow rule pattern is copied over to
temporary buffer and a new item added internally in PMD.
Such constructed buffer is passed to the HWS layer when flow create
operation is being enqueued.
After operation is enqueued, temporary flow pattern can be safely
discarded, so there is no need to store it during
the whole lifecycle of mlx5_hw_q_job.
This patch removes all references to flow rule pattern and items stored
inside mlx5_hw_q_job and removes relevant allocations to reduce job
memory footprint.
Temporary pattern and items stored per job are replaced with stack
allocated ones, contained in mlx5_flow_hw_pattern_params struct.
Signed-off-by: Dariusz Sosnowski <dsosnowski@nvidia.com>
---
drivers/net/mlx5/mlx5.h | 17 ++++-------
drivers/net/mlx5/mlx5_flow.h | 10 +++++++
drivers/net/mlx5/mlx5_flow_hw.c | 51 ++++++++++++++-------------------
3 files changed, 37 insertions(+), 41 deletions(-)
@@ -401,17 +401,12 @@ struct mlx5_hw_q_job {
const void *action; /* Indirect action attached to the job. */
};
void *user_data; /* Job user data. */
- struct rte_flow_item *items;
- union {
- struct {
- /* User memory for query output */
- void *user;
- /* Data extracted from hardware */
- void *hw;
- } __rte_packed query;
- struct rte_flow_item_ethdev port_spec;
- struct rte_flow_item_tag tag_spec;
- } __rte_packed;
+ struct {
+ /* User memory for query output */
+ void *user;
+ /* Data extracted from hardware */
+ void *hw;
+ } query;
struct rte_flow_hw *upd_flow; /* Flow with updated values. */
};
@@ -1304,6 +1304,16 @@ struct mlx5_flow_hw_action_params {
uint8_t ipv6_push_data[MLX5_PUSH_MAX_LEN];
};
+/** Container for dynamically generated flow items used during flow rule creation. */
+struct mlx5_flow_hw_pattern_params {
+ /** Array of dynamically generated flow items. */
+ struct rte_flow_item items[MLX5_HW_MAX_ITEMS];
+ /** Temporary REPRESENTED_PORT item generated by PMD. */
+ struct rte_flow_item_ethdev port_spec;
+ /** Temporary TAG item generated by PMD. */
+ struct rte_flow_item_tag tag_spec;
+};
+
/* rte flow action translate to DR action struct. */
struct mlx5_action_construct_data {
LIST_ENTRY(mlx5_action_construct_data) next;
@@ -3253,44 +3253,44 @@ flow_hw_get_rule_items(struct rte_eth_dev *dev,
const struct rte_flow_template_table *table,
const struct rte_flow_item items[],
uint8_t pattern_template_index,
- struct mlx5_hw_q_job *job)
+ struct mlx5_flow_hw_pattern_params *pp)
{
struct rte_flow_pattern_template *pt = table->its[pattern_template_index];
/* Only one implicit item can be added to flow rule pattern. */
MLX5_ASSERT(!pt->implicit_port || !pt->implicit_tag);
- /* At least one item was allocated in job descriptor for items. */
+ /* At least one item was allocated in pattern params for items. */
MLX5_ASSERT(MLX5_HW_MAX_ITEMS >= 1);
if (pt->implicit_port) {
if (pt->orig_item_nb + 1 > MLX5_HW_MAX_ITEMS) {
rte_errno = ENOMEM;
return NULL;
}
- /* Set up represented port item in job descriptor. */
- job->port_spec = (struct rte_flow_item_ethdev){
+ /* Set up represented port item in pattern params. */
+ pp->port_spec = (struct rte_flow_item_ethdev){
.port_id = dev->data->port_id,
};
- job->items[0] = (struct rte_flow_item){
+ pp->items[0] = (struct rte_flow_item){
.type = RTE_FLOW_ITEM_TYPE_REPRESENTED_PORT,
- .spec = &job->port_spec,
+ .spec = &pp->port_spec,
};
- rte_memcpy(&job->items[1], items, sizeof(*items) * pt->orig_item_nb);
- return job->items;
+ rte_memcpy(&pp->items[1], items, sizeof(*items) * pt->orig_item_nb);
+ return pp->items;
} else if (pt->implicit_tag) {
if (pt->orig_item_nb + 1 > MLX5_HW_MAX_ITEMS) {
rte_errno = ENOMEM;
return NULL;
}
- /* Set up tag item in job descriptor. */
- job->tag_spec = (struct rte_flow_item_tag){
+ /* Set up tag item in pattern params. */
+ pp->tag_spec = (struct rte_flow_item_tag){
.data = flow_hw_tx_tag_regc_value(dev),
};
- job->items[0] = (struct rte_flow_item){
+ pp->items[0] = (struct rte_flow_item){
.type = (enum rte_flow_item_type)MLX5_RTE_FLOW_ITEM_TYPE_TAG,
- .spec = &job->tag_spec,
+ .spec = &pp->tag_spec,
};
- rte_memcpy(&job->items[1], items, sizeof(*items) * pt->orig_item_nb);
- return job->items;
+ rte_memcpy(&pp->items[1], items, sizeof(*items) * pt->orig_item_nb);
+ return pp->items;
} else {
return items;
}
@@ -3345,6 +3345,7 @@ flow_hw_async_flow_create(struct rte_eth_dev *dev,
};
struct mlx5dr_rule_action *rule_acts;
struct mlx5_flow_hw_action_params ap;
+ struct mlx5_flow_hw_pattern_params pp;
struct rte_flow_hw *flow = NULL;
struct mlx5_hw_q_job *job = NULL;
const struct rte_flow_item *rule_items;
@@ -3409,7 +3410,7 @@ flow_hw_async_flow_create(struct rte_eth_dev *dev,
goto error;
}
rule_items = flow_hw_get_rule_items(dev, table, items,
- pattern_template_index, job);
+ pattern_template_index, &pp);
if (!rule_items)
goto error;
if (likely(!rte_flow_template_table_resizable(dev->data->port_id, &table->cfg.attr))) {
@@ -9990,11 +9991,8 @@ flow_hw_configure(struct rte_eth_dev *dev,
goto err;
}
mem_size += (sizeof(struct mlx5_hw_q_job *) +
- sizeof(struct mlx5_hw_q_job) +
- sizeof(struct rte_flow_item) *
- MLX5_HW_MAX_ITEMS +
- sizeof(struct rte_flow_hw)) *
- _queue_attr[i]->size;
+ sizeof(struct mlx5_hw_q_job) +
+ sizeof(struct rte_flow_hw)) * _queue_attr[i]->size;
}
priv->hw_q = mlx5_malloc(MLX5_MEM_ZERO, mem_size,
64, SOCKET_ID_ANY);
@@ -10003,7 +10001,6 @@ flow_hw_configure(struct rte_eth_dev *dev,
goto err;
}
for (i = 0; i < nb_q_updated; i++) {
- struct rte_flow_item *items = NULL;
struct rte_flow_hw *upd_flow = NULL;
priv->hw_q[i].job_idx = _queue_attr[i]->size;
@@ -10016,12 +10013,8 @@ flow_hw_configure(struct rte_eth_dev *dev,
&job[_queue_attr[i - 1]->size - 1].upd_flow[1];
job = (struct mlx5_hw_q_job *)
&priv->hw_q[i].job[_queue_attr[i]->size];
- items = (struct rte_flow_item *)
- &job[_queue_attr[i]->size];
- upd_flow = (struct rte_flow_hw *)
- &items[_queue_attr[i]->size * MLX5_HW_MAX_ITEMS];
+ upd_flow = (struct rte_flow_hw *)&job[_queue_attr[i]->size];
for (j = 0; j < _queue_attr[i]->size; j++) {
- job[j].items = &items[j * MLX5_HW_MAX_ITEMS];
job[j].upd_flow = &upd_flow[j];
priv->hw_q[i].job[j] = &job[j];
}
@@ -12193,14 +12186,12 @@ flow_hw_calc_table_hash(struct rte_eth_dev *dev,
uint32_t *hash, struct rte_flow_error *error)
{
const struct rte_flow_item *items;
- /* Temp job to allow adding missing items */
- static struct rte_flow_item tmp_items[MLX5_HW_MAX_ITEMS];
- static struct mlx5_hw_q_job job = {.items = tmp_items};
+ struct mlx5_flow_hw_pattern_params pp;
int res;
items = flow_hw_get_rule_items(dev, table, pattern,
pattern_template_index,
- &job);
+ &pp);
res = mlx5dr_rule_hash_calculate(mlx5_table_matcher(table), items,
pattern_template_index,
MLX5DR_RULE_HASH_CALC_MODE_RAW,