* [PATCH 00/15] Improve rte_tm support in ICE driver
@ 2024-08-07 9:33 Bruce Richardson
2024-08-07 9:33 ` [PATCH 01/15] net/ice: add traffic management node query function Bruce Richardson
` (16 more replies)
0 siblings, 17 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
This patchset expands the capabilities of the traffic management
support in the ICE driver. It allows the driver to support different
sizes of topologies, and support >256 queues and more than 3 hierarchy
layers.
Bruce Richardson (15):
net/ice: add traffic management node query function
net/ice: detect stopping a flow-director queue twice
net/ice: improve Tx scheduler graph output
net/ice: add option to choose DDP package file
net/ice: add option to download scheduler topology
net/ice/base: allow init without TC class sched nodes
net/ice/base: set VSI index on newly created nodes
net/ice/base: read VSI layer info from VSI
net/ice/base: remove 255 limit on sched child nodes
net/ice/base: optimize subtree searches
net/ice/base: make functions non-static
net/ice/base: remove flag checks before topology upload
net/ice: limit the number of queues to sched capabilities
net/ice: enhance Tx scheduler hierarchy support
net/ice: add minimal capability reporting API
doc/guides/nics/ice.rst | 9 +
drivers/net/ice/base/ice_ddp.c | 51 +--
drivers/net/ice/base/ice_ddp.h | 4 +-
drivers/net/ice/base/ice_sched.c | 56 ++--
drivers/net/ice/base/ice_sched.h | 8 +
drivers/net/ice/base/ice_type.h | 3 +-
drivers/net/ice/ice_diagnose.c | 196 ++++-------
drivers/net/ice/ice_ethdev.c | 92 +++--
drivers/net/ice/ice_ethdev.h | 18 +-
drivers/net/ice/ice_rxtx.c | 15 +
drivers/net/ice/ice_tm.c | 558 +++++++++++++++----------------
11 files changed, 487 insertions(+), 523 deletions(-)
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 01/15] net/ice: add traffic management node query function
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
@ 2024-08-07 9:33 ` Bruce Richardson
2024-08-07 9:33 ` [PATCH 02/15] net/ice: detect stopping a flow-director queue twice Bruce Richardson
` (15 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Implement the new node querying function for the "ice" net driver.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_tm.c | 48 ++++++++++++++++++++++++++++++++++++++++
1 file changed, 48 insertions(+)
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 8a29a9e744..459446a6b0 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -17,6 +17,11 @@ static int ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
uint32_t weight, uint32_t level_id,
const struct rte_tm_node_params *params,
struct rte_tm_error *error);
+static int ice_node_query(const struct rte_eth_dev *dev, uint32_t node_id,
+ uint32_t *parent_node_id, uint32_t *priority,
+ uint32_t *weight, uint32_t *level_id,
+ struct rte_tm_node_params *params,
+ struct rte_tm_error *error);
static int ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
struct rte_tm_error *error);
static int ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
@@ -35,6 +40,7 @@ const struct rte_tm_ops ice_tm_ops = {
.node_add = ice_tm_node_add,
.node_delete = ice_tm_node_delete,
.node_type_get = ice_node_type_get,
+ .node_query = ice_node_query,
.hierarchy_commit = ice_hierarchy_commit,
};
@@ -219,6 +225,48 @@ ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
return 0;
}
+static int
+ice_node_query(const struct rte_eth_dev *dev, uint32_t node_id,
+ uint32_t *parent_node_id, uint32_t *priority,
+ uint32_t *weight, uint32_t *level_id,
+ struct rte_tm_node_params *params,
+ struct rte_tm_error *error)
+{
+ struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_tm_node *tm_node;
+
+ if (node_id == RTE_TM_NODE_ID_NULL) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "invalid node id";
+ return -EINVAL;
+ }
+
+ /* check if the node id exists */
+ tm_node = find_node(pf->tm_conf.root, node_id);
+ if (!tm_node) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "no such node";
+ return -EEXIST;
+ }
+
+ if (parent_node_id != NULL)
+ *parent_node_id = tm_node->parent->id;
+
+ if (priority != NULL)
+ *priority = tm_node->priority;
+
+ if (weight != NULL)
+ *weight = tm_node->weight;
+
+ if (level_id != NULL)
+ *level_id = tm_node->level;
+
+ if (params != NULL)
+ *params = tm_node->params;
+
+ return 0;
+}
+
static inline struct ice_tm_shaper_profile *
ice_shaper_profile_search(struct rte_eth_dev *dev,
uint32_t shaper_profile_id)
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 02/15] net/ice: detect stopping a flow-director queue twice
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-07 9:33 ` [PATCH 01/15] net/ice: add traffic management node query function Bruce Richardson
@ 2024-08-07 9:33 ` Bruce Richardson
2024-08-07 9:33 ` [PATCH 03/15] net/ice: improve Tx scheduler graph output Bruce Richardson
` (14 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
If the flow-director queue is stopped at some point during the running
of an application, the shutdown procedure for the port issues an error
as it tries to stop the queue a second time, and fails to do so. We can
eliminate this error by setting the tail-register pointer to NULL on
stop, and checking for that condition in subsequent stop calls. Since
the register pointer is set on start, any restarting of the queue will
allow a stop call to progress as normal.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_rxtx.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index f270498ed1..a150d28e73 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -1139,6 +1139,10 @@ ice_fdir_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
tx_queue_id);
return -EINVAL;
}
+ if (txq->qtx_tail == NULL) {
+ PMD_DRV_LOG(INFO, "TX queue %u not started\n", tx_queue_id);
+ return 0;
+ }
vsi = txq->vsi;
q_ids[0] = txq->reg_idx;
@@ -1153,6 +1157,7 @@ ice_fdir_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
}
txq->tx_rel_mbufs(txq);
+ txq->qtx_tail = NULL;
return 0;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 03/15] net/ice: improve Tx scheduler graph output
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-07 9:33 ` [PATCH 01/15] net/ice: add traffic management node query function Bruce Richardson
2024-08-07 9:33 ` [PATCH 02/15] net/ice: detect stopping a flow-director queue twice Bruce Richardson
@ 2024-08-07 9:33 ` Bruce Richardson
2024-08-07 9:33 ` [PATCH 04/15] net/ice: add option to choose DDP package file Bruce Richardson
` (13 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The function to dump the TX scheduler topology only adds to the chart
nodes connected to TX queues or for the flow director VSI. Change the
function to work recursively from the root node and thereby include all
scheduler nodes, whether in use or not, in the dump.
Also, improve the output of the Tx scheduler graphing function:
* Add VSI details to each node in graph
* When number of children is >16, skip middle nodes to reduce size of
the graph, otherwise dot output is unviewable for large hierarchies
* For VSIs other than zero, use dot's clustering method to put those
VSIs into subgraphs with borders
* For leaf nodes, display queue numbers for the any nodes asigned to
ethdev NIC Tx queues
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_diagnose.c | 196 ++++++++++++---------------------
1 file changed, 69 insertions(+), 127 deletions(-)
diff --git a/drivers/net/ice/ice_diagnose.c b/drivers/net/ice/ice_diagnose.c
index c357554707..623d84e37d 100644
--- a/drivers/net/ice/ice_diagnose.c
+++ b/drivers/net/ice/ice_diagnose.c
@@ -545,29 +545,15 @@ static void print_rl_profile(struct ice_aqc_rl_profile_elem *prof,
fprintf(stream, "\t\t\t\t\t</td>\n");
}
-static
-void print_elem_type(FILE *stream, u8 type)
+static const char *
+get_elem_type(u8 type)
{
- switch (type) {
- case 1:
- fprintf(stream, "root");
- break;
- case 2:
- fprintf(stream, "tc");
- break;
- case 3:
- fprintf(stream, "se_generic");
- break;
- case 4:
- fprintf(stream, "entry_point");
- break;
- case 5:
- fprintf(stream, "leaf");
- break;
- default:
- fprintf(stream, "%d", type);
- break;
- }
+ static const char * const ice_sched_node_types[] = {
+ "Undefined", "Root", "TC", "SE Generic", "SW Entry", "Leaf"
+ };
+ if (type < RTE_DIM(ice_sched_node_types))
+ return ice_sched_node_types[type];
+ return "*UNKNOWN*";
}
static
@@ -602,7 +588,9 @@ void print_priority_mode(FILE *stream, bool flag)
}
static
-void print_node(struct ice_aqc_txsched_elem_data *data,
+void print_node(struct ice_sched_node *node,
+ struct rte_eth_dev_data *ethdata,
+ struct ice_aqc_txsched_elem_data *data,
struct ice_aqc_rl_profile_elem *cir_prof,
struct ice_aqc_rl_profile_elem *eir_prof,
struct ice_aqc_rl_profile_elem *shared_prof,
@@ -613,17 +601,19 @@ void print_node(struct ice_aqc_txsched_elem_data *data,
fprintf(stream, "\t\t\t<table>\n");
- fprintf(stream, "\t\t\t\t<tr>\n");
- fprintf(stream, "\t\t\t\t\t<td> teid </td>\n");
- fprintf(stream, "\t\t\t\t\t<td> %d </td>\n", data->node_teid);
- fprintf(stream, "\t\t\t\t</tr>\n");
-
- fprintf(stream, "\t\t\t\t<tr>\n");
- fprintf(stream, "\t\t\t\t\t<td> type </td>\n");
- fprintf(stream, "\t\t\t\t\t<td>");
- print_elem_type(stream, data->data.elem_type);
- fprintf(stream, "</td>\n");
- fprintf(stream, "\t\t\t\t</tr>\n");
+ fprintf(stream, "\t\t\t\t<tr><td>teid</td><td>%d</td></tr>\n", data->node_teid);
+ fprintf(stream, "\t\t\t\t<tr><td>type</td><td>%s</td></tr>\n",
+ get_elem_type(data->data.elem_type));
+ fprintf(stream, "\t\t\t\t<tr><td>VSI</td><td>%u</td></tr>\n", node->vsi_handle);
+ if (data->data.elem_type == ICE_AQC_ELEM_TYPE_LEAF) {
+ for (uint16_t i = 0; i < ethdata->nb_tx_queues; i++) {
+ struct ice_tx_queue *q = ethdata->tx_queues[i];
+ if (q->q_teid == data->node_teid) {
+ fprintf(stream, "\t\t\t\t<tr><td>TXQ</td><td>%u</td></tr>\n", i);
+ break;
+ }
+ }
+ }
if (!detail)
goto brief;
@@ -705,8 +695,6 @@ void print_node(struct ice_aqc_txsched_elem_data *data,
fprintf(stream, "\t\tshape=plain\n");
fprintf(stream, "\t]\n");
- if (data->parent_teid != 0xFFFFFFFF)
- fprintf(stream, "\tNODE_%d -> NODE_%d\n", data->parent_teid, data->node_teid);
}
static
@@ -731,112 +719,92 @@ int query_rl_profile(struct ice_hw *hw,
return 0;
}
-static
-int query_node(struct ice_hw *hw, uint32_t child, uint32_t *parent,
- uint8_t level, bool detail, FILE *stream)
+static int
+query_node(struct ice_hw *hw, struct rte_eth_dev_data *ethdata,
+ struct ice_sched_node *node, bool detail, FILE *stream)
{
- struct ice_aqc_txsched_elem_data data;
+ struct ice_aqc_txsched_elem_data *data = &node->info;
struct ice_aqc_rl_profile_elem cir_prof;
struct ice_aqc_rl_profile_elem eir_prof;
struct ice_aqc_rl_profile_elem shared_prof;
struct ice_aqc_rl_profile_elem *cp = NULL;
struct ice_aqc_rl_profile_elem *ep = NULL;
struct ice_aqc_rl_profile_elem *sp = NULL;
- int status, ret;
-
- status = ice_sched_query_elem(hw, child, &data);
- if (status != ICE_SUCCESS) {
- if (level == hw->num_tx_sched_layers) {
- /* ignore the error when a queue has been stopped. */
- PMD_DRV_LOG(WARNING, "Failed to query queue node %d.", child);
- *parent = 0xffffffff;
- return 0;
- }
- PMD_DRV_LOG(ERR, "Failed to query scheduling node %d.", child);
- return -EINVAL;
- }
-
- *parent = data.parent_teid;
+ u8 level = node->tx_sched_layer;
+ int ret;
- if (data.data.cir_bw.bw_profile_idx != 0) {
- ret = query_rl_profile(hw, level, 0, data.data.cir_bw.bw_profile_idx, &cir_prof);
+ if (data->data.cir_bw.bw_profile_idx != 0) {
+ ret = query_rl_profile(hw, level, 0, data->data.cir_bw.bw_profile_idx, &cir_prof);
if (ret)
return ret;
cp = &cir_prof;
}
- if (data.data.eir_bw.bw_profile_idx != 0) {
- ret = query_rl_profile(hw, level, 1, data.data.eir_bw.bw_profile_idx, &eir_prof);
+ if (data->data.eir_bw.bw_profile_idx != 0) {
+ ret = query_rl_profile(hw, level, 1, data->data.eir_bw.bw_profile_idx, &eir_prof);
if (ret)
return ret;
ep = &eir_prof;
}
- if (data.data.srl_id != 0) {
- ret = query_rl_profile(hw, level, 2, data.data.srl_id, &shared_prof);
+ if (data->data.srl_id != 0) {
+ ret = query_rl_profile(hw, level, 2, data->data.srl_id, &shared_prof);
if (ret)
return ret;
sp = &shared_prof;
}
- print_node(&data, cp, ep, sp, detail, stream);
+ print_node(node, ethdata, data, cp, ep, sp, detail, stream);
return 0;
}
-static
-int query_nodes(struct ice_hw *hw,
- uint32_t *children, int child_num,
- uint32_t *parents, int *parent_num,
- uint8_t level, bool detail,
- FILE *stream)
+static int
+query_node_recursive(struct ice_hw *hw, struct rte_eth_dev_data *ethdata,
+ struct ice_sched_node *node, bool detail, FILE *stream)
{
- uint32_t parent;
- int i;
- int j;
-
- *parent_num = 0;
- for (i = 0; i < child_num; i++) {
- bool exist = false;
- int ret;
+ bool close = false;
+ if (node->parent != NULL && node->vsi_handle != node->parent->vsi_handle) {
+ fprintf(stream, "subgraph cluster_%u {\n", node->vsi_handle);
+ fprintf(stream, "\tlabel = \"VSI %u\";\n", node->vsi_handle);
+ close = true;
+ }
- ret = query_node(hw, children[i], &parent, level, detail, stream);
- if (ret)
- return -EINVAL;
+ int ret = query_node(hw, ethdata, node, detail, stream);
+ if (ret != 0)
+ return ret;
- for (j = 0; j < *parent_num; j++) {
- if (parents[j] == parent) {
- exist = true;
- break;
- }
+ for (uint16_t i = 0; i < node->num_children; i++) {
+ ret = query_node_recursive(hw, ethdata, node->children[i], detail, stream);
+ if (ret != 0)
+ return ret;
+ /* if we have a lot of nodes, skip a bunch in the middle */
+ if (node->num_children > 16 && i == 2) {
+ uint16_t inc = node->num_children - 5;
+ fprintf(stream, "\tn%d_children [label=\"... +%d child nodes ...\"];\n",
+ node->info.node_teid, inc);
+ fprintf(stream, "\tNODE_%d -> n%d_children;\n",
+ node->info.node_teid, node->info.node_teid);
+ i += inc;
}
-
- if (!exist && parent != 0xFFFFFFFF)
- parents[(*parent_num)++] = parent;
}
+ if (close)
+ fprintf(stream, "}\n");
+ if (node->info.parent_teid != 0xFFFFFFFF)
+ fprintf(stream, "\tNODE_%d -> NODE_%d\n",
+ node->info.parent_teid, node->info.node_teid);
return 0;
}
-int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
+int
+rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
{
struct rte_eth_dev *dev;
struct ice_hw *hw;
- struct ice_pf *pf;
- struct ice_q_ctx *q_ctx;
- uint16_t q_num;
- uint16_t i;
- struct ice_tx_queue *txq;
- uint32_t buf1[256];
- uint32_t buf2[256];
- uint32_t *children = buf1;
- uint32_t *parents = buf2;
- int child_num = 0;
- int parent_num = 0;
- uint8_t level;
RTE_ETH_VALID_PORTID_OR_ERR_RET(port, -ENODEV);
@@ -846,35 +814,9 @@ int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
dev = &rte_eth_devices[port];
hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- level = hw->num_tx_sched_layers;
-
- q_num = dev->data->nb_tx_queues;
-
- /* main vsi */
- for (i = 0; i < q_num; i++) {
- txq = dev->data->tx_queues[i];
- q_ctx = ice_get_lan_q_ctx(hw, txq->vsi->idx, 0, i);
- children[child_num++] = q_ctx->q_teid;
- }
-
- /* fdir vsi */
- q_ctx = ice_get_lan_q_ctx(hw, pf->fdir.fdir_vsi->idx, 0, 0);
- children[child_num++] = q_ctx->q_teid;
fprintf(stream, "digraph tx_sched {\n");
- while (child_num > 0) {
- int ret;
- ret = query_nodes(hw, children, child_num,
- parents, &parent_num,
- level, detail, stream);
- if (ret)
- return ret;
-
- children = parents;
- child_num = parent_num;
- level--;
- }
+ query_node_recursive(hw, dev->data, hw->port_info->root, detail, stream);
fprintf(stream, "}\n");
return 0;
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 04/15] net/ice: add option to choose DDP package file
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (2 preceding siblings ...)
2024-08-07 9:33 ` [PATCH 03/15] net/ice: improve Tx scheduler graph output Bruce Richardson
@ 2024-08-07 9:33 ` Bruce Richardson
2024-08-07 9:33 ` [PATCH 05/15] net/ice: add option to download scheduler topology Bruce Richardson
` (12 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The "Dynamic Device Personalization" package is loaded at initialization
time by the driver, but the specific package file loaded depends upon
what package file is found first by searching through a hard-coded list
of firmware paths. To enable greater control over the package loading,
we can add a device option to choose a specific DDP package file to
load.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
doc/guides/nics/ice.rst | 9 +++++++++
drivers/net/ice/ice_ethdev.c | 34 ++++++++++++++++++++++++++++++++++
drivers/net/ice/ice_ethdev.h | 1 +
3 files changed, 44 insertions(+)
diff --git a/doc/guides/nics/ice.rst b/doc/guides/nics/ice.rst
index ae975d19ad..58ccfbd1a5 100644
--- a/doc/guides/nics/ice.rst
+++ b/doc/guides/nics/ice.rst
@@ -108,6 +108,15 @@ Runtime Configuration
-a 80:00.0,default-mac-disable=1
+- ``DDP Package File``
+
+ Rather than have the driver search for the DDP package to load,
+ or to override what package is used,
+ the ``ddp_pkg_file`` option can be used to provide the path to a specific package file.
+ For example::
+
+ -a 80:00.0,ddp_pkg_file=/path/to/ice-version.pkg
+
- ``Protocol extraction for per queue``
Configure the RX queues to do protocol extraction into mbuf for protocol
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 304f959b7e..3e7ceda9ce 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -36,6 +36,7 @@
#define ICE_ONE_PPS_OUT_ARG "pps_out"
#define ICE_RX_LOW_LATENCY_ARG "rx_low_latency"
#define ICE_MBUF_CHECK_ARG "mbuf_check"
+#define ICE_DDP_FILENAME "ddp_pkg_file"
#define ICE_CYCLECOUNTER_MASK 0xffffffffffffffffULL
@@ -52,6 +53,7 @@ static const char * const ice_valid_args[] = {
ICE_RX_LOW_LATENCY_ARG,
ICE_DEFAULT_MAC_DISABLE,
ICE_MBUF_CHECK_ARG,
+ ICE_DDP_FILENAME,
NULL
};
@@ -692,6 +694,18 @@ handle_field_name_arg(__rte_unused const char *key, const char *value,
return 0;
}
+static int
+handle_ddp_filename_arg(__rte_unused const char *key, const char *value, void *name_args)
+{
+ const char **filename = name_args;
+ if (strlen(value) >= ICE_MAX_PKG_FILENAME_SIZE) {
+ PMD_DRV_LOG(ERR, "The DDP package filename is too long : '%s'", value);
+ return -1;
+ }
+ *filename = strdup(value);
+ return 0;
+}
+
static void
ice_check_proto_xtr_support(struct ice_hw *hw)
{
@@ -1882,6 +1896,16 @@ int ice_load_pkg(struct ice_adapter *adapter, bool use_dsn, uint64_t dsn)
size_t bufsz;
int err;
+ if (adapter->devargs.ddp_filename != NULL) {
+ strlcpy(pkg_file, adapter->devargs.ddp_filename, sizeof(pkg_file));
+ if (rte_firmware_read(pkg_file, &buf, &bufsz) == 0) {
+ goto load_fw;
+ } else {
+ PMD_INIT_LOG(ERR, "Cannot load DDP file: %s\n", pkg_file);
+ return -1;
+ }
+ }
+
if (!use_dsn)
goto no_dsn;
@@ -2216,6 +2240,13 @@ static int ice_parse_devargs(struct rte_eth_dev *dev)
ret = rte_kvargs_process(kvlist, ICE_RX_LOW_LATENCY_ARG,
&parse_bool, &ad->devargs.rx_low_latency);
+ if (ret)
+ goto bail;
+
+ ret = rte_kvargs_process(kvlist, ICE_DDP_FILENAME,
+ &handle_ddp_filename_arg, &ad->devargs.ddp_filename);
+ if (ret)
+ goto bail;
bail:
rte_kvargs_free(kvlist);
@@ -2689,6 +2720,8 @@ ice_dev_close(struct rte_eth_dev *dev)
ice_free_hw_tbls(hw);
rte_free(hw->port_info);
hw->port_info = NULL;
+ free((void *)(uintptr_t)ad->devargs.ddp_filename);
+ ad->devargs.ddp_filename = NULL;
ice_shutdown_all_ctrlq(hw, true);
rte_free(pf->proto_xtr);
pf->proto_xtr = NULL;
@@ -6981,6 +7014,7 @@ RTE_PMD_REGISTER_PARAM_STRING(net_ice,
ICE_PROTO_XTR_ARG "=[queue:]<vlan|ipv4|ipv6|ipv6_flow|tcp|ip_offset>"
ICE_SAFE_MODE_SUPPORT_ARG "=<0|1>"
ICE_DEFAULT_MAC_DISABLE "=<0|1>"
+ ICE_DDP_FILENAME "=</path/to/file>"
ICE_RX_LOW_LATENCY_ARG "=<0|1>");
RTE_LOG_REGISTER_SUFFIX(ice_logtype_init, init, NOTICE);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index 3ea9f37dc8..c211b5b9cc 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -568,6 +568,7 @@ struct ice_devargs {
/* Name of the field. */
char xtr_field_name[RTE_MBUF_DYN_NAMESIZE];
uint64_t mbuf_check;
+ const char *ddp_filename;
};
/**
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 05/15] net/ice: add option to download scheduler topology
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (3 preceding siblings ...)
2024-08-07 9:33 ` [PATCH 04/15] net/ice: add option to choose DDP package file Bruce Richardson
@ 2024-08-07 9:33 ` Bruce Richardson
2024-08-07 9:33 ` [PATCH 06/15] net/ice/base: allow init without TC class sched nodes Bruce Richardson
` (11 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The DDP package file being loaded at init time may contain an
alternative Tx Scheduler topology in it. Add driver option to load this
topology at init time.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_ddp.c | 18 +++++++++++++++---
drivers/net/ice/base/ice_ddp.h | 4 ++--
drivers/net/ice/ice_ethdev.c | 24 +++++++++++++++---------
drivers/net/ice/ice_ethdev.h | 1 +
4 files changed, 33 insertions(+), 14 deletions(-)
diff --git a/drivers/net/ice/base/ice_ddp.c b/drivers/net/ice/base/ice_ddp.c
index 24506dfaea..e6c42c5274 100644
--- a/drivers/net/ice/base/ice_ddp.c
+++ b/drivers/net/ice/base/ice_ddp.c
@@ -1326,7 +1326,7 @@ ice_fill_hw_ptype(struct ice_hw *hw)
* ice_copy_and_init_pkg() instead of directly calling ice_init_pkg() in this
* case.
*/
-enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
+enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len, bool load_sched)
{
bool already_loaded = false;
enum ice_ddp_state state;
@@ -1344,6 +1344,18 @@ enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
return state;
}
+ if (load_sched) {
+ enum ice_status res = ice_cfg_tx_topo(hw, buf, len);
+ if (res != ICE_SUCCESS) {
+ ice_debug(hw, ICE_DBG_INIT, "failed to apply sched topology (err: %d)\n",
+ res);
+ return ICE_DDP_PKG_ERR;
+ }
+ ice_debug(hw, ICE_DBG_INIT, "Topology download successful, reinitializing device\n");
+ ice_deinit_hw(hw);
+ ice_init_hw(hw);
+ }
+
/* initialize package info */
state = ice_init_pkg_info(hw, pkg);
if (state)
@@ -1416,7 +1428,7 @@ enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
* related routines.
*/
enum ice_ddp_state
-ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len)
+ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len, bool load_sched)
{
enum ice_ddp_state state;
u8 *buf_copy;
@@ -1426,7 +1438,7 @@ ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len)
buf_copy = (u8 *)ice_memdup(hw, buf, len, ICE_NONDMA_TO_NONDMA);
- state = ice_init_pkg(hw, buf_copy, len);
+ state = ice_init_pkg(hw, buf_copy, len, load_sched);
if (!ice_is_init_pkg_successful(state)) {
/* Free the copy, since we failed to initialize the package */
ice_free(hw, buf_copy);
diff --git a/drivers/net/ice/base/ice_ddp.h b/drivers/net/ice/base/ice_ddp.h
index 5761920207..2feba2e91d 100644
--- a/drivers/net/ice/base/ice_ddp.h
+++ b/drivers/net/ice/base/ice_ddp.h
@@ -451,9 +451,9 @@ ice_pkg_enum_entry(struct ice_seg *ice_seg, struct ice_pkg_enum *state,
void *
ice_pkg_enum_section(struct ice_seg *ice_seg, struct ice_pkg_enum *state,
u32 sect_type);
-enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buff, u32 len);
+enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buff, u32 len, bool load_sched);
enum ice_ddp_state
-ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len);
+ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len, bool load_sched);
bool ice_is_init_pkg_successful(enum ice_ddp_state state);
void ice_free_seg(struct ice_hw *hw);
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 3e7ceda9ce..0d2445a317 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -37,6 +37,7 @@
#define ICE_RX_LOW_LATENCY_ARG "rx_low_latency"
#define ICE_MBUF_CHECK_ARG "mbuf_check"
#define ICE_DDP_FILENAME "ddp_pkg_file"
+#define ICE_DDP_LOAD_SCHED "ddp_load_sched_topo"
#define ICE_CYCLECOUNTER_MASK 0xffffffffffffffffULL
@@ -54,6 +55,7 @@ static const char * const ice_valid_args[] = {
ICE_DEFAULT_MAC_DISABLE,
ICE_MBUF_CHECK_ARG,
ICE_DDP_FILENAME,
+ ICE_DDP_LOAD_SCHED,
NULL
};
@@ -1938,7 +1940,7 @@ int ice_load_pkg(struct ice_adapter *adapter, bool use_dsn, uint64_t dsn)
load_fw:
PMD_INIT_LOG(DEBUG, "DDP package name: %s", pkg_file);
- err = ice_copy_and_init_pkg(hw, buf, bufsz);
+ err = ice_copy_and_init_pkg(hw, buf, bufsz, adapter->devargs.ddp_load_sched);
if (!ice_is_init_pkg_successful(err)) {
PMD_INIT_LOG(ERR, "ice_copy_and_init_hw failed: %d\n", err);
free(buf);
@@ -1971,19 +1973,18 @@ static int
parse_bool(const char *key, const char *value, void *args)
{
int *i = (int *)args;
- char *end;
- int num;
- num = strtoul(value, &end, 10);
-
- if (num != 0 && num != 1) {
- PMD_DRV_LOG(WARNING, "invalid value:\"%s\" for key:\"%s\", "
- "value must be 0 or 1",
+ if (value == NULL || value[0] == '\0') {
+ PMD_DRV_LOG(WARNING, "key:\"%s\", requires a value, which must be 0 or 1", key);
+ return -1;
+ }
+ if (value[1] != '\0' || (value[0] != '0' && value[0] != '1')) {
+ PMD_DRV_LOG(WARNING, "invalid value:\"%s\" for key:\"%s\", value must be 0 or 1",
value, key);
return -1;
}
- *i = num;
+ *i = value[0] - '0';
return 0;
}
@@ -2248,6 +2249,10 @@ static int ice_parse_devargs(struct rte_eth_dev *dev)
if (ret)
goto bail;
+ ret = rte_kvargs_process(kvlist, ICE_DDP_LOAD_SCHED,
+ &parse_bool, &ad->devargs.ddp_load_sched);
+ if (ret)
+ goto bail;
bail:
rte_kvargs_free(kvlist);
return ret;
@@ -7014,6 +7019,7 @@ RTE_PMD_REGISTER_PARAM_STRING(net_ice,
ICE_PROTO_XTR_ARG "=[queue:]<vlan|ipv4|ipv6|ipv6_flow|tcp|ip_offset>"
ICE_SAFE_MODE_SUPPORT_ARG "=<0|1>"
ICE_DEFAULT_MAC_DISABLE "=<0|1>"
+ ICE_DDP_LOAD_SCHED "=<0|1>"
ICE_DDP_FILENAME "=</path/to/file>"
ICE_RX_LOW_LATENCY_ARG "=<0|1>");
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index c211b5b9cc..f31addb122 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -563,6 +563,7 @@ struct ice_devargs {
uint8_t proto_xtr[ICE_MAX_QUEUE_NUM];
uint8_t pin_idx;
uint8_t pps_out_ena;
+ int ddp_load_sched;
int xtr_field_offs;
uint8_t xtr_flag_offs[PROTO_XTR_MAX];
/* Name of the field. */
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 06/15] net/ice/base: allow init without TC class sched nodes
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (4 preceding siblings ...)
2024-08-07 9:33 ` [PATCH 05/15] net/ice: add option to download scheduler topology Bruce Richardson
@ 2024-08-07 9:33 ` Bruce Richardson
2024-08-07 9:33 ` [PATCH 07/15] net/ice/base: set VSI index on newly created nodes Bruce Richardson
` (10 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
If DCB support is disabled via DDP image, there will not be any traffic
class (TC) nodes in the scheduler tree immediately above the root level.
To allow the driver to work with this scenario, we allow use of the root
node as a dummy TC0 node in case where there are no TC nodes in the
tree. For use of any other TC other than 0 (used by default in the
driver), existing behaviour of returning NULL pointer is maintained.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 6 ++++++
drivers/net/ice/base/ice_type.h | 1 +
2 files changed, 7 insertions(+)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index 373c32a518..f75e5ae599 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -292,6 +292,10 @@ struct ice_sched_node *ice_sched_get_tc_node(struct ice_port_info *pi, u8 tc)
if (!pi || !pi->root)
return NULL;
+ /* if no TC nodes, use root as TC node 0 */
+ if (pi->has_tc == 0)
+ return tc == 0 ? pi->root : NULL;
+
for (i = 0; i < pi->root->num_children; i++)
if (pi->root->children[i]->tc_num == tc)
return pi->root->children[i];
@@ -1306,6 +1310,8 @@ int ice_sched_init_port(struct ice_port_info *pi)
ICE_AQC_ELEM_TYPE_ENTRY_POINT)
hw->sw_entry_point_layer = j;
+ if (buf[0].generic[j].data.elem_type == ICE_AQC_ELEM_TYPE_TC)
+ pi->has_tc = 1;
status = ice_sched_add_node(pi, j, &buf[i].generic[j], NULL);
if (status)
goto err_init_port;
diff --git a/drivers/net/ice/base/ice_type.h b/drivers/net/ice/base/ice_type.h
index 598a80155b..a70e4a8afa 100644
--- a/drivers/net/ice/base/ice_type.h
+++ b/drivers/net/ice/base/ice_type.h
@@ -1260,6 +1260,7 @@ struct ice_port_info {
struct ice_qos_cfg qos_cfg;
u8 is_vf:1;
u8 is_custom_tx_enabled:1;
+ u8 has_tc:1;
};
struct ice_switch_info {
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 07/15] net/ice/base: set VSI index on newly created nodes
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (5 preceding siblings ...)
2024-08-07 9:33 ` [PATCH 06/15] net/ice/base: allow init without TC class sched nodes Bruce Richardson
@ 2024-08-07 9:33 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 08/15] net/ice/base: read VSI layer info from VSI Bruce Richardson
` (9 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:33 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The ice_sched_node type has got a field for the vsi to which the node
belongs. This field was not getting set in "ice_sched_add_node", so add
a line configuring this field for each node from its parent node.
Similarly, when searching for a qgroup node, we can check for each node
that the VSI information is correct.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f75e5ae599..f6dc5ae173 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -200,6 +200,7 @@ ice_sched_add_node(struct ice_port_info *pi, u8 layer,
node->in_use = true;
node->parent = parent;
node->tx_sched_layer = layer;
+ node->vsi_handle = parent->vsi_handle;
parent->children[parent->num_children++] = node;
node->info = elem;
return 0;
@@ -1581,7 +1582,7 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
/* make sure the qgroup node is part of the VSI subtree */
if (ice_sched_find_node_in_subtree(pi->hw, vsi_node, qgrp_node))
if (qgrp_node->num_children < max_children &&
- qgrp_node->owner == owner)
+ qgrp_node->owner == owner && qgrp_node->vsi_handle == vsi_handle)
break;
qgrp_node = qgrp_node->sibling;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 08/15] net/ice/base: read VSI layer info from VSI
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (6 preceding siblings ...)
2024-08-07 9:33 ` [PATCH 07/15] net/ice/base: set VSI index on newly created nodes Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 09/15] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
` (8 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Rather than computing from the number of HW layers the layer of the VSI,
we can instead just read that info from the VSI node itself. This allows
the layer to be changed at runtime.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f6dc5ae173..e398984bf2 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -1559,7 +1559,6 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
u16 max_children;
qgrp_layer = ice_sched_get_qgrp_layer(pi->hw);
- vsi_layer = ice_sched_get_vsi_layer(pi->hw);
max_children = pi->hw->max_children[qgrp_layer];
vsi_ctx = ice_get_vsi_ctx(pi->hw, vsi_handle);
@@ -1569,6 +1568,7 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
/* validate invalid VSI ID */
if (!vsi_node)
return NULL;
+ vsi_layer = vsi_node->tx_sched_layer;
/* If the queue group and vsi layer are same then queues
* are all attached directly to VSI
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 09/15] net/ice/base: remove 255 limit on sched child nodes
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (7 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 08/15] net/ice/base: read VSI layer info from VSI Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 10/15] net/ice/base: optimize subtree searches Bruce Richardson
` (7 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The Tx scheduler in the ice driver can be configured to have large
numbers of child nodes at a given layer, but the driver code implicitly
limited the number of nodes to 255 by using a u8 datatype for the number
of children. Increase this to a 16-bit value throughout the code.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 25 ++++++++++++++-----------
drivers/net/ice/base/ice_type.h | 2 +-
2 files changed, 15 insertions(+), 12 deletions(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index e398984bf2..be13833e1e 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -289,7 +289,7 @@ ice_sched_get_first_node(struct ice_port_info *pi,
*/
struct ice_sched_node *ice_sched_get_tc_node(struct ice_port_info *pi, u8 tc)
{
- u8 i;
+ u16 i;
if (!pi || !pi->root)
return NULL;
@@ -316,7 +316,7 @@ void ice_free_sched_node(struct ice_port_info *pi, struct ice_sched_node *node)
{
struct ice_sched_node *parent;
struct ice_hw *hw = pi->hw;
- u8 i, j;
+ u16 i, j;
/* Free the children before freeing up the parent node
* The parent array is updated below and that shifts the nodes
@@ -1473,7 +1473,7 @@ bool
ice_sched_find_node_in_subtree(struct ice_hw *hw, struct ice_sched_node *base,
struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
for (i = 0; i < base->num_children; i++) {
struct ice_sched_node *child = base->children[i];
@@ -1510,7 +1510,7 @@ ice_sched_get_free_qgrp(struct ice_port_info *pi,
struct ice_sched_node *qgrp_node, u8 owner)
{
struct ice_sched_node *min_qgrp;
- u8 min_children;
+ u16 min_children;
if (!qgrp_node)
return qgrp_node;
@@ -2070,7 +2070,7 @@ static void ice_sched_rm_agg_vsi_info(struct ice_port_info *pi, u16 vsi_handle)
*/
static bool ice_sched_is_leaf_node_present(struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
for (i = 0; i < node->num_children; i++)
if (ice_sched_is_leaf_node_present(node->children[i]))
@@ -2105,7 +2105,7 @@ ice_sched_rm_vsi_cfg(struct ice_port_info *pi, u16 vsi_handle, u8 owner)
ice_for_each_traffic_class(i) {
struct ice_sched_node *vsi_node, *tc_node;
- u8 j = 0;
+ u16 j = 0;
tc_node = ice_sched_get_tc_node(pi, i);
if (!tc_node)
@@ -2173,7 +2173,7 @@ int ice_rm_vsi_lan_cfg(struct ice_port_info *pi, u16 vsi_handle)
*/
bool ice_sched_is_tree_balanced(struct ice_hw *hw, struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
/* start from the leaf node */
for (i = 0; i < node->num_children; i++)
@@ -2247,7 +2247,8 @@ ice_sched_get_free_vsi_parent(struct ice_hw *hw, struct ice_sched_node *node,
u16 *num_nodes)
{
u8 l = node->tx_sched_layer;
- u8 vsil, i;
+ u8 vsil;
+ u16 i;
vsil = ice_sched_get_vsi_layer(hw);
@@ -2289,7 +2290,7 @@ ice_sched_update_parent(struct ice_sched_node *new_parent,
struct ice_sched_node *node)
{
struct ice_sched_node *old_parent;
- u8 i, j;
+ u16 i, j;
old_parent = node->parent;
@@ -2389,7 +2390,8 @@ ice_sched_move_vsi_to_agg(struct ice_port_info *pi, u16 vsi_handle, u32 agg_id,
u16 num_nodes[ICE_AQC_TOPO_MAX_LEVEL_NUM] = { 0 };
u32 first_node_teid, vsi_teid;
u16 num_nodes_added;
- u8 aggl, vsil, i;
+ u8 aggl, vsil;
+ u16 i;
int status;
tc_node = ice_sched_get_tc_node(pi, tc);
@@ -2505,7 +2507,8 @@ ice_move_all_vsi_to_dflt_agg(struct ice_port_info *pi,
static bool
ice_sched_is_agg_inuse(struct ice_port_info *pi, struct ice_sched_node *node)
{
- u8 vsil, i;
+ u8 vsil;
+ u16 i;
vsil = ice_sched_get_vsi_layer(pi->hw);
if (node->tx_sched_layer < vsil - 1) {
diff --git a/drivers/net/ice/base/ice_type.h b/drivers/net/ice/base/ice_type.h
index a70e4a8afa..35f832eb9f 100644
--- a/drivers/net/ice/base/ice_type.h
+++ b/drivers/net/ice/base/ice_type.h
@@ -1030,9 +1030,9 @@ struct ice_sched_node {
struct ice_aqc_txsched_elem_data info;
u32 agg_id; /* aggregator group ID */
u16 vsi_handle;
+ u16 num_children;
u8 in_use; /* suspended or in use */
u8 tx_sched_layer; /* Logical Layer (1-9) */
- u8 num_children;
u8 tc_num;
u8 owner;
#define ICE_SCHED_NODE_OWNER_LAN 0
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 10/15] net/ice/base: optimize subtree searches
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (8 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 09/15] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 11/15] net/ice/base: make functions non-static Bruce Richardson
` (6 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
In a number of places throughout the driver code, we want to confirm
that a scheduler node is indeed a child of another node. Currently, this
is confirmed by searching down the tree from the base until the desired
node is hit, a search which may hit many irrelevant tree nodes when
recursing down wrong branches. By switching the direction of search, to
check upwards from the node to the parent, we can avoid any incorrect
paths, and so speed up processing.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 18 +++++-------------
1 file changed, 5 insertions(+), 13 deletions(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index be13833e1e..f7d5f8f415 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -1475,20 +1475,12 @@ ice_sched_find_node_in_subtree(struct ice_hw *hw, struct ice_sched_node *base,
{
u16 i;
- for (i = 0; i < base->num_children; i++) {
- struct ice_sched_node *child = base->children[i];
-
- if (node == child)
- return true;
-
- if (child->tx_sched_layer > node->tx_sched_layer)
- return false;
-
- /* this recursion is intentional, and wouldn't
- * go more than 8 calls
- */
- if (ice_sched_find_node_in_subtree(hw, child, node))
+ if (base == node)
+ return true;
+ while (node->tx_sched_layer != 0 && node->parent != NULL) {
+ if (node->parent == base)
return true;
+ node = node->parent;
}
return false;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 11/15] net/ice/base: make functions non-static
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (9 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 10/15] net/ice/base: optimize subtree searches Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 12/15] net/ice/base: remove flag checks before topology upload Bruce Richardson
` (5 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
We will need to allocate more lanq contexts after a scheduler rework, so
make that function non-static so accessible outside the file. For similar
reasons, make the function to add a Tx scheduler node non-static
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 2 +-
drivers/net/ice/base/ice_sched.h | 8 ++++++++
2 files changed, 9 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f7d5f8f415..d88b836c38 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -570,7 +570,7 @@ ice_sched_suspend_resume_elems(struct ice_hw *hw, u8 num_nodes, u32 *node_teids,
* @tc: TC number
* @new_numqs: number of queues
*/
-static int
+int
ice_alloc_lan_q_ctx(struct ice_hw *hw, u16 vsi_handle, u8 tc, u16 new_numqs)
{
struct ice_vsi_ctx *vsi_ctx;
diff --git a/drivers/net/ice/base/ice_sched.h b/drivers/net/ice/base/ice_sched.h
index 9f78516dfb..c7eb794963 100644
--- a/drivers/net/ice/base/ice_sched.h
+++ b/drivers/net/ice/base/ice_sched.h
@@ -270,4 +270,12 @@ int ice_sched_replay_q_bw(struct ice_port_info *pi, struct ice_q_ctx *q_ctx);
int
ice_sched_cfg_node_bw_alloc(struct ice_hw *hw, struct ice_sched_node *node,
enum ice_rl_type rl_type, u16 bw_alloc);
+
+int
+ice_sched_add_elems(struct ice_port_info *pi, struct ice_sched_node *tc_node,
+ struct ice_sched_node *parent, u8 layer, u16 num_nodes,
+ u16 *num_nodes_added, u32 *first_node_teid,
+ struct ice_sched_node **prealloc_nodes);
+int
+ice_alloc_lan_q_ctx(struct ice_hw *hw, u16 vsi_handle, u8 tc, u16 new_numqs);
#endif /* _ICE_SCHED_H_ */
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 12/15] net/ice/base: remove flag checks before topology upload
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (10 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 11/15] net/ice/base: make functions non-static Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 13/15] net/ice: limit the number of queues to sched capabilities Bruce Richardson
` (4 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
DPDK should support more than just 9-level or 5-level topologies, so
remove the checks for those particular settings.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_ddp.c | 33 ---------------------------------
1 file changed, 33 deletions(-)
diff --git a/drivers/net/ice/base/ice_ddp.c b/drivers/net/ice/base/ice_ddp.c
index e6c42c5274..744f015fe5 100644
--- a/drivers/net/ice/base/ice_ddp.c
+++ b/drivers/net/ice/base/ice_ddp.c
@@ -2373,38 +2373,6 @@ int ice_cfg_tx_topo(struct ice_hw *hw, u8 *buf, u32 len)
return status;
}
- /* Is default topology already applied ? */
- if (!(flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 9) {
- ice_debug(hw, ICE_DBG_INIT, "Loaded default topology\n");
- /* Already default topology is loaded */
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Is new topology already applied ? */
- if ((flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 5) {
- ice_debug(hw, ICE_DBG_INIT, "Loaded new topology\n");
- /* Already new topology is loaded */
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Is set topology issued already ? */
- if (flags & ICE_AQC_TX_TOPO_FLAGS_ISSUED) {
- ice_debug(hw, ICE_DBG_INIT, "Update tx topology was done by another PF\n");
- /* add a small delay before exiting */
- for (i = 0; i < 20; i++)
- ice_msec_delay(100, true);
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Change the topology from new to default (5 to 9) */
- if (!(flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 5) {
- ice_debug(hw, ICE_DBG_INIT, "Change topology from 5 to 9 layers\n");
- goto update_topo;
- }
-
pkg_hdr = (struct ice_pkg_hdr *)buf;
state = ice_verify_pkg(pkg_hdr, len);
if (state) {
@@ -2451,7 +2419,6 @@ int ice_cfg_tx_topo(struct ice_hw *hw, u8 *buf, u32 len)
/* Get the new topology buffer */
new_topo = ((u8 *)section) + offset;
-update_topo:
/* acquire global lock to make sure that set topology issued
* by one PF
*/
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 13/15] net/ice: limit the number of queues to sched capabilities
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (11 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 12/15] net/ice/base: remove flag checks before topology upload Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 14/15] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
` (3 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Rather than assuming that each VSI can hold up to 256 queue pairs,
or the reported device limit, query the available nodes in the scheduler
tree to check that we are not overflowing the limit for number of
child scheduling nodes at each level. Do this by multiplying
max_children for each level beyond the VSI and using that as an
additional cap on the number of queues.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.c | 25 ++++++++++++++++++++-----
1 file changed, 20 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 0d2445a317..ab3f88fd7d 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -913,7 +913,7 @@ ice_vsi_config_default_rss(struct ice_aqc_vsi_props *info)
}
static int
-ice_vsi_config_tc_queue_mapping(struct ice_vsi *vsi,
+ice_vsi_config_tc_queue_mapping(struct ice_hw *hw, struct ice_vsi *vsi,
struct ice_aqc_vsi_props *info,
uint8_t enabled_tcmap)
{
@@ -929,13 +929,28 @@ ice_vsi_config_tc_queue_mapping(struct ice_vsi *vsi,
}
/* vector 0 is reserved and 1 vector for ctrl vsi */
- if (vsi->adapter->hw.func_caps.common_cap.num_msix_vectors < 2)
+ if (vsi->adapter->hw.func_caps.common_cap.num_msix_vectors < 2) {
vsi->nb_qps = 0;
- else
+ } else {
vsi->nb_qps = RTE_MIN
((uint16_t)vsi->adapter->hw.func_caps.common_cap.num_msix_vectors - 2,
RTE_MIN(vsi->nb_qps, ICE_MAX_Q_PER_TC));
+ /* cap max QPs to what the HW reports as num-children for each layer.
+ * Multiply num_children for each layer from the entry_point layer to
+ * the qgroup, or second-last layer.
+ * Avoid any potential overflow by using uint32_t type and breaking loop
+ * once we have a number greater than the already configured max.
+ */
+ uint32_t max_sched_vsi_nodes = 1;
+ for (uint8_t i = hw->sw_entry_point_layer; i < hw->num_tx_sched_layers - 1; i++) {
+ max_sched_vsi_nodes *= hw->max_children[i];
+ if (max_sched_vsi_nodes >= vsi->nb_qps)
+ break;
+ }
+ vsi->nb_qps = RTE_MIN(vsi->nb_qps, max_sched_vsi_nodes);
+ }
+
/* nb_qps(hex) -> fls */
/* 0000 -> 0 */
/* 0001 -> 0 */
@@ -1707,7 +1722,7 @@ ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type)
rte_cpu_to_le_16(hw->func_caps.fd_fltr_best_effort);
/* Enable VLAN/UP trip */
- ret = ice_vsi_config_tc_queue_mapping(vsi,
+ ret = ice_vsi_config_tc_queue_mapping(hw, vsi,
&vsi_ctx.info,
ICE_DEFAULT_TCMAP);
if (ret) {
@@ -1731,7 +1746,7 @@ ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type)
vsi_ctx.info.fd_options = rte_cpu_to_le_16(cfg);
vsi_ctx.info.sw_id = hw->port_info->sw_id;
vsi_ctx.info.sw_flags2 = ICE_AQ_VSI_SW_FLAG_LAN_ENA;
- ret = ice_vsi_config_tc_queue_mapping(vsi,
+ ret = ice_vsi_config_tc_queue_mapping(hw, vsi,
&vsi_ctx.info,
ICE_DEFAULT_TCMAP);
if (ret) {
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 14/15] net/ice: enhance Tx scheduler hierarchy support
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (12 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 13/15] net/ice: limit the number of queues to sched capabilities Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:34 ` [PATCH 15/15] net/ice: add minimal capability reporting API Bruce Richardson
` (2 subsequent siblings)
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Increase the flexibility of the Tx scheduler hierarchy support in the
driver. If the HW/firmware allows it, allow creating up to 2k child
nodes per scheduler node. Also expand the number of supported layers to
the max available, rather than always just having 3 layers. One
restriction on this change is that the topology needs to be configured
and enabled before port queue setup, in many cases, and before port
start in all cases.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.c | 9 -
drivers/net/ice/ice_ethdev.h | 15 +-
drivers/net/ice/ice_rxtx.c | 10 +
drivers/net/ice/ice_tm.c | 495 ++++++++++++++---------------------
4 files changed, 212 insertions(+), 317 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index ab3f88fd7d..5a5967ff71 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -3832,7 +3832,6 @@ ice_dev_start(struct rte_eth_dev *dev)
int mask, ret;
uint8_t timer = hw->func_caps.ts_func_info.tmr_index_owned;
uint32_t pin_idx = ad->devargs.pin_idx;
- struct rte_tm_error tm_err;
ice_declare_bitmap(pmask, ICE_PROMISC_MAX);
ice_zero_bitmap(pmask, ICE_PROMISC_MAX);
@@ -3864,14 +3863,6 @@ ice_dev_start(struct rte_eth_dev *dev)
}
}
- if (pf->tm_conf.committed) {
- ret = ice_do_hierarchy_commit(dev, pf->tm_conf.clear_on_fail, &tm_err);
- if (ret) {
- PMD_DRV_LOG(ERR, "fail to commit Tx scheduler");
- goto rx_err;
- }
- }
-
ice_set_rx_function(dev);
ice_set_tx_function(dev);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index f31addb122..cb1a7e8e0d 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -479,14 +479,6 @@ struct ice_tm_node {
struct ice_sched_node *sched_node;
};
-/* node type of Traffic Manager */
-enum ice_tm_node_type {
- ICE_TM_NODE_TYPE_PORT,
- ICE_TM_NODE_TYPE_QGROUP,
- ICE_TM_NODE_TYPE_QUEUE,
- ICE_TM_NODE_TYPE_MAX,
-};
-
/* Struct to store all the Traffic Manager configuration. */
struct ice_tm_conf {
struct ice_shaper_profile_list shaper_profile_list;
@@ -690,9 +682,6 @@ int ice_rem_rss_cfg_wrap(struct ice_pf *pf, uint16_t vsi_id,
struct ice_rss_hash_cfg *cfg);
void ice_tm_conf_init(struct rte_eth_dev *dev);
void ice_tm_conf_uninit(struct rte_eth_dev *dev);
-int ice_do_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error);
extern const struct rte_tm_ops ice_tm_ops;
static inline int
@@ -750,4 +739,8 @@ int rte_pmd_ice_dump_switch(uint16_t port, uint8_t **buff, uint32_t *size);
__rte_experimental
int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream);
+
+int
+ice_tm_setup_txq_node(struct ice_pf *pf, struct ice_hw *hw, uint16_t qid, uint32_t node_teid);
+
#endif /* _ICE_ETHDEV_H_ */
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index a150d28e73..7a421bb364 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -747,6 +747,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
int err;
struct ice_vsi *vsi;
struct ice_hw *hw;
+ struct ice_pf *pf;
struct ice_aqc_add_tx_qgrp *txq_elem;
struct ice_tlan_ctx tx_ctx;
int buf_len;
@@ -777,6 +778,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
vsi = txq->vsi;
hw = ICE_VSI_TO_HW(vsi);
+ pf = ICE_VSI_TO_PF(vsi);
memset(&tx_ctx, 0, sizeof(tx_ctx));
txq_elem->num_txqs = 1;
@@ -812,6 +814,14 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
/* store the schedule node id */
txq->q_teid = txq_elem->txqs[0].q_teid;
+ /* move the queue to correct position in hierarchy, if explicit hierarchy configured */
+ if (pf->tm_conf.committed)
+ if (ice_tm_setup_txq_node(pf, hw, tx_queue_id, txq->q_teid) != 0) {
+ PMD_DRV_LOG(ERR, "Failed to set up txq traffic management node");
+ rte_free(txq_elem);
+ return -EIO;
+ }
+
dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
rte_free(txq_elem);
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 459446a6b0..a86943a5b2 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -1,17 +1,17 @@
/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(c) 2022 Intel Corporation
*/
+#include <rte_ethdev.h>
#include <rte_tm_driver.h>
#include "ice_ethdev.h"
#include "ice_rxtx.h"
-#define MAX_CHILDREN_PER_SCHED_NODE 8
-#define MAX_CHILDREN_PER_TM_NODE 256
+#define MAX_CHILDREN_PER_TM_NODE 2048
static int ice_hierarchy_commit(struct rte_eth_dev *dev,
int clear_on_fail,
- __rte_unused struct rte_tm_error *error);
+ struct rte_tm_error *error);
static int ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
uint32_t parent_node_id, uint32_t priority,
uint32_t weight, uint32_t level_id,
@@ -86,9 +86,10 @@ ice_tm_conf_uninit(struct rte_eth_dev *dev)
}
static int
-ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
+ice_node_param_check(uint32_t node_id,
uint32_t priority, uint32_t weight,
const struct rte_tm_node_params *params,
+ bool is_leaf,
struct rte_tm_error *error)
{
/* checked all the unsupported parameter */
@@ -123,7 +124,7 @@ ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
}
/* for non-leaf node */
- if (node_id >= pf->dev_data->nb_tx_queues) {
+ if (!is_leaf) {
if (params->nonleaf.wfq_weight_mode) {
error->type =
RTE_TM_ERROR_TYPE_NODE_PARAMS_WFQ_WEIGHT_MODE;
@@ -147,6 +148,11 @@ ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
}
/* for leaf node */
+ if (node_id >= RTE_MAX_QUEUES_PER_PORT) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "Node ID out of range for a leaf node.";
+ return -EINVAL;
+ }
if (params->leaf.cman) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS_CMAN;
error->message = "Congestion management not supported";
@@ -193,11 +199,18 @@ find_node(struct ice_tm_node *root, uint32_t id)
return NULL;
}
+static inline uint8_t
+ice_get_leaf_level(struct ice_hw *hw)
+{
+ return hw->num_tx_sched_layers - 1 - hw->port_info->has_tc;
+}
+
static int
ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
int *is_leaf, struct rte_tm_error *error)
{
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_tm_node *tm_node;
if (!is_leaf || !error)
@@ -217,7 +230,7 @@ ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
return -EINVAL;
}
- if (tm_node->level == ICE_TM_NODE_TYPE_QUEUE)
+ if (tm_node->level == ice_get_leaf_level(hw))
*is_leaf = true;
else
*is_leaf = false;
@@ -389,16 +402,28 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
struct rte_tm_error *error)
{
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_tm_shaper_profile *shaper_profile = NULL;
struct ice_tm_node *tm_node;
- struct ice_tm_node *parent_node;
+ struct ice_tm_node *parent_node = NULL;
int ret;
if (!params || !error)
return -EINVAL;
- ret = ice_node_param_check(pf, node_id, priority, weight,
- params, error);
+ if (parent_node_id != RTE_TM_NODE_ID_NULL) {
+ parent_node = find_node(pf->tm_conf.root, parent_node_id);
+ if (!parent_node) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
+ error->message = "parent not exist";
+ return -EINVAL;
+ }
+ }
+ if (level_id == RTE_TM_NODE_LEVEL_ID_ANY && parent_node != NULL)
+ level_id = parent_node->level + 1;
+
+ ret = ice_node_param_check(node_id, priority, weight,
+ params, level_id == ice_get_leaf_level(hw), error);
if (ret)
return ret;
@@ -424,9 +449,9 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
/* root node if not have a parent */
if (parent_node_id == RTE_TM_NODE_ID_NULL) {
/* check level */
- if (level_id != ICE_TM_NODE_TYPE_PORT) {
+ if (level_id != 0) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
- error->message = "Wrong level";
+ error->message = "Wrong level, root node (NULL parent) must be at level 0";
return -EINVAL;
}
@@ -445,7 +470,7 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
if (!tm_node)
return -ENOMEM;
tm_node->id = node_id;
- tm_node->level = ICE_TM_NODE_TYPE_PORT;
+ tm_node->level = 0;
tm_node->parent = NULL;
tm_node->reference_count = 0;
tm_node->shaper_profile = shaper_profile;
@@ -458,48 +483,21 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
}
/* check the parent node */
- parent_node = find_node(pf->tm_conf.root, parent_node_id);
- if (!parent_node) {
- error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
- error->message = "parent not exist";
- return -EINVAL;
- }
- if (parent_node->level != ICE_TM_NODE_TYPE_PORT &&
- parent_node->level != ICE_TM_NODE_TYPE_QGROUP) {
+ /* for n-level hierarchy, level n-1 is leaf, so last level with children is n-2 */
+ if ((int)parent_node->level > hw->num_tx_sched_layers - 2) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
error->message = "parent is not valid";
return -EINVAL;
}
/* check level */
- if (level_id != RTE_TM_NODE_LEVEL_ID_ANY &&
- level_id != parent_node->level + 1) {
+ if (level_id != parent_node->level + 1) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
error->message = "Wrong level";
return -EINVAL;
}
/* check the node number */
- if (parent_node->level == ICE_TM_NODE_TYPE_PORT) {
- /* check the queue group number */
- if (parent_node->reference_count >= pf->dev_data->nb_tx_queues) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too many queue groups";
- return -EINVAL;
- }
- } else {
- /* check the queue number */
- if (parent_node->reference_count >=
- MAX_CHILDREN_PER_SCHED_NODE) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too many queues";
- return -EINVAL;
- }
- if (node_id >= pf->dev_data->nb_tx_queues) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too large queue id";
- return -EINVAL;
- }
- }
+ /* TODO, check max children allowed and max nodes at this level */
tm_node = rte_zmalloc(NULL,
sizeof(struct ice_tm_node) +
@@ -518,13 +516,12 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
(void *)((uint8_t *)tm_node + sizeof(struct ice_tm_node));
tm_node->parent->children[tm_node->parent->reference_count] = tm_node;
- if (tm_node->priority != 0 && level_id != ICE_TM_NODE_TYPE_QUEUE &&
- level_id != ICE_TM_NODE_TYPE_QGROUP)
+ if (tm_node->priority != 0)
+ /* TODO fixme, some levels may support this perhaps? */
PMD_DRV_LOG(WARNING, "priority != 0 not supported in level %d",
level_id);
- if (tm_node->weight != 1 &&
- level_id != ICE_TM_NODE_TYPE_QUEUE && level_id != ICE_TM_NODE_TYPE_QGROUP)
+ if (tm_node->weight != 1 && level_id == 0)
PMD_DRV_LOG(WARNING, "weight != 1 not supported in level %d",
level_id);
@@ -569,7 +566,7 @@ ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
}
/* root node */
- if (tm_node->level == ICE_TM_NODE_TYPE_PORT) {
+ if (tm_node->level == 0) {
rte_free(tm_node);
pf->tm_conf.root = NULL;
return 0;
@@ -589,53 +586,6 @@ ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
return 0;
}
-static int ice_move_recfg_lan_txq(struct rte_eth_dev *dev,
- struct ice_sched_node *queue_sched_node,
- struct ice_sched_node *dst_node,
- uint16_t queue_id)
-{
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_aqc_move_txqs_data *buf;
- struct ice_sched_node *queue_parent_node;
- uint8_t txqs_moved;
- int ret = ICE_SUCCESS;
- uint16_t buf_size = ice_struct_size(buf, txqs, 1);
-
- buf = (struct ice_aqc_move_txqs_data *)ice_malloc(hw, sizeof(*buf));
- if (buf == NULL)
- return -ENOMEM;
-
- queue_parent_node = queue_sched_node->parent;
- buf->src_teid = queue_parent_node->info.node_teid;
- buf->dest_teid = dst_node->info.node_teid;
- buf->txqs[0].q_teid = queue_sched_node->info.node_teid;
- buf->txqs[0].txq_id = queue_id;
-
- ret = ice_aq_move_recfg_lan_txq(hw, 1, true, false, false, false, 50,
- NULL, buf, buf_size, &txqs_moved, NULL);
- if (ret || txqs_moved == 0) {
- PMD_DRV_LOG(ERR, "move lan queue %u failed", queue_id);
- rte_free(buf);
- return ICE_ERR_PARAM;
- }
-
- if (queue_parent_node->num_children > 0) {
- queue_parent_node->num_children--;
- queue_parent_node->children[queue_parent_node->num_children] = NULL;
- } else {
- PMD_DRV_LOG(ERR, "invalid children number %d for queue %u",
- queue_parent_node->num_children, queue_id);
- rte_free(buf);
- return ICE_ERR_PARAM;
- }
- dst_node->children[dst_node->num_children++] = queue_sched_node;
- queue_sched_node->parent = dst_node;
- ice_sched_query_elem(hw, queue_sched_node->info.node_teid, &queue_sched_node->info);
-
- rte_free(buf);
- return ret;
-}
-
static int ice_set_node_rate(struct ice_hw *hw,
struct ice_tm_node *tm_node,
struct ice_sched_node *sched_node)
@@ -723,240 +673,191 @@ static int ice_cfg_hw_node(struct ice_hw *hw,
return 0;
}
-static struct ice_sched_node *ice_get_vsi_node(struct ice_hw *hw)
+int
+ice_tm_setup_txq_node(struct ice_pf *pf, struct ice_hw *hw, uint16_t qid, uint32_t teid)
{
- struct ice_sched_node *node = hw->port_info->root;
- uint32_t vsi_layer = hw->num_tx_sched_layers - ICE_VSI_LAYER_OFFSET;
- uint32_t i;
+ struct ice_sched_node *hw_node = ice_sched_find_node_by_teid(hw->port_info->root, teid);
+ struct ice_tm_node *sw_node = find_node(pf->tm_conf.root, qid);
- for (i = 0; i < vsi_layer; i++)
- node = node->children[0];
-
- return node;
-}
-
-static int ice_reset_noleaf_nodes(struct rte_eth_dev *dev)
-{
- struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_sched_node *vsi_node = ice_get_vsi_node(hw);
- struct ice_tm_node *root = pf->tm_conf.root;
- uint32_t i;
- int ret;
-
- /* reset vsi_node */
- ret = ice_set_node_rate(hw, NULL, vsi_node);
- if (ret) {
- PMD_DRV_LOG(ERR, "reset vsi node failed");
- return ret;
- }
-
- if (root == NULL)
+ /* not configured in hierarchy */
+ if (sw_node == NULL)
return 0;
- for (i = 0; i < root->reference_count; i++) {
- struct ice_tm_node *tm_node = root->children[i];
+ sw_node->sched_node = hw_node;
- if (tm_node->sched_node == NULL)
- continue;
+ /* if the queue node has been put in the wrong place in hierarchy */
+ if (hw_node->parent != sw_node->parent->sched_node) {
+ struct ice_aqc_move_txqs_data *buf;
+ uint8_t txqs_moved = 0;
+ uint16_t buf_size = ice_struct_size(buf, txqs, 1);
+
+ buf = ice_malloc(hw, buf_size);
+ if (buf == NULL)
+ return -ENOMEM;
- ret = ice_cfg_hw_node(hw, NULL, tm_node->sched_node);
- if (ret) {
- PMD_DRV_LOG(ERR, "reset queue group node %u failed", tm_node->id);
- return ret;
+ struct ice_sched_node *parent = hw_node->parent;
+ struct ice_sched_node *new_parent = sw_node->parent->sched_node;
+ buf->src_teid = parent->info.node_teid;
+ buf->dest_teid = new_parent->info.node_teid;
+ buf->txqs[0].q_teid = hw_node->info.node_teid;
+ buf->txqs[0].txq_id = qid;
+
+ int ret = ice_aq_move_recfg_lan_txq(hw, 1, true, false, false, false, 50,
+ NULL, buf, buf_size, &txqs_moved, NULL);
+ if (ret || txqs_moved == 0) {
+ PMD_DRV_LOG(ERR, "move lan queue %u failed", qid);
+ ice_free(hw, buf);
+ return ICE_ERR_PARAM;
}
- tm_node->sched_node = NULL;
+
+ /* now update the ice_sched_nodes to match physical layout */
+ new_parent->children[new_parent->num_children++] = hw_node;
+ hw_node->parent = new_parent;
+ ice_sched_query_elem(hw, hw_node->info.node_teid, &hw_node->info);
+ for (uint16_t i = 0; i < parent->num_children; i++)
+ if (parent->children[i] == hw_node) {
+ /* to remove, just overwrite the old node slot with the last ptr */
+ parent->children[i] = parent->children[--parent->num_children];
+ break;
+ }
}
- return 0;
+ return ice_cfg_hw_node(hw, sw_node, hw_node);
}
-static int ice_remove_leaf_nodes(struct rte_eth_dev *dev)
+/* from a given node, recursively deletes all the nodes that belong to that vsi.
+ * Any nodes which can't be deleted because they have children belonging to a different
+ * VSI, are now also adjusted to belong to that VSI also
+ */
+static int
+free_sched_node_recursive(struct ice_port_info *pi, const struct ice_sched_node *root,
+ struct ice_sched_node *node, uint8_t vsi_id)
{
- int ret = 0;
- int i;
+ uint16_t i = 0;
- for (i = 0; i < dev->data->nb_tx_queues; i++) {
- ret = ice_tx_queue_stop(dev, i);
- if (ret) {
- PMD_DRV_LOG(ERR, "stop queue %u failed", i);
- break;
+ while (i < node->num_children) {
+ if (node->children[i]->vsi_handle != vsi_id) {
+ i++;
+ continue;
}
+ free_sched_node_recursive(pi, root, node->children[i], vsi_id);
}
- return ret;
-}
-
-static int ice_add_leaf_nodes(struct rte_eth_dev *dev)
-{
- int ret = 0;
- int i;
-
- for (i = 0; i < dev->data->nb_tx_queues; i++) {
- ret = ice_tx_queue_start(dev, i);
- if (ret) {
- PMD_DRV_LOG(ERR, "start queue %u failed", i);
- break;
- }
+ if (node != root) {
+ if (node->num_children == 0)
+ ice_free_sched_node(pi, node);
+ else
+ node->vsi_handle = node->children[0]->vsi_handle;
}
- return ret;
+ return 0;
}
-int ice_do_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error)
+static int
+create_sched_node_recursive(struct ice_port_info *pi, struct ice_tm_node *sw_node,
+ struct ice_sched_node *hw_root, uint16_t *created)
{
- struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_tm_node *root;
- struct ice_sched_node *vsi_node = NULL;
- struct ice_sched_node *queue_node;
- struct ice_tx_queue *txq;
- int ret_val = 0;
- uint32_t i;
- uint32_t idx_vsi_child;
- uint32_t idx_qg;
- uint32_t nb_vsi_child;
- uint32_t nb_qg;
- uint32_t qid;
- uint32_t q_teid;
-
- /* remove leaf nodes */
- ret_val = ice_remove_leaf_nodes(dev);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "reset no-leaf nodes failed");
- goto fail_clear;
- }
-
- /* reset no-leaf nodes. */
- ret_val = ice_reset_noleaf_nodes(dev);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "reset leaf nodes failed");
- goto add_leaf;
- }
-
- /* config vsi node */
- vsi_node = ice_get_vsi_node(hw);
- root = pf->tm_conf.root;
-
- ret_val = ice_set_node_rate(hw, root, vsi_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure vsi node %u bandwidth failed",
- root->id);
- goto add_leaf;
- }
-
- /* config queue group nodes */
- nb_vsi_child = vsi_node->num_children;
- nb_qg = vsi_node->children[0]->num_children;
-
- idx_vsi_child = 0;
- idx_qg = 0;
-
- if (root == NULL)
- goto commit;
-
- for (i = 0; i < root->reference_count; i++) {
- struct ice_tm_node *tm_node = root->children[i];
- struct ice_tm_node *tm_child_node;
- struct ice_sched_node *qgroup_sched_node =
- vsi_node->children[idx_vsi_child]->children[idx_qg];
- uint32_t j;
-
- ret_val = ice_cfg_hw_node(hw, tm_node, qgroup_sched_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure queue group node %u failed",
- tm_node->id);
- goto reset_leaf;
- }
-
- for (j = 0; j < tm_node->reference_count; j++) {
- tm_child_node = tm_node->children[j];
- qid = tm_child_node->id;
- ret_val = ice_tx_queue_start(dev, qid);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "start queue %u failed", qid);
- goto reset_leaf;
- }
- txq = dev->data->tx_queues[qid];
- q_teid = txq->q_teid;
- queue_node = ice_sched_get_node(hw->port_info, q_teid);
- if (queue_node == NULL) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "get queue %u node failed", qid);
- goto reset_leaf;
- }
- if (queue_node->info.parent_teid != qgroup_sched_node->info.node_teid) {
- ret_val = ice_move_recfg_lan_txq(dev, queue_node,
- qgroup_sched_node, qid);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "move queue %u failed", qid);
- goto reset_leaf;
- }
- }
- ret_val = ice_cfg_hw_node(hw, tm_child_node, queue_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure queue group node %u failed",
- tm_node->id);
- goto reset_leaf;
- }
- }
-
- idx_qg++;
- if (idx_qg >= nb_qg) {
- idx_qg = 0;
- idx_vsi_child++;
+ struct ice_sched_node *parent = sw_node->sched_node;
+ uint32_t teid;
+ uint16_t added;
+
+ /* first create all child nodes */
+ for (uint16_t i = 0; i < sw_node->reference_count; i++) {
+ struct ice_tm_node *tm_node = sw_node->children[i];
+ int res = ice_sched_add_elems(pi, hw_root,
+ parent, parent->tx_sched_layer + 1,
+ 1 /* num nodes */, &added, &teid,
+ NULL /* no pre-alloc */);
+ if (res != 0) {
+ PMD_DRV_LOG(ERR, "Error with ice_sched_add_elems, adding child node to teid %u\n",
+ parent->info.node_teid);
+ return -1;
}
- if (idx_vsi_child >= nb_vsi_child) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "too many queues");
- goto reset_leaf;
+ struct ice_sched_node *hw_node = ice_sched_find_node_by_teid(parent, teid);
+ if (ice_cfg_hw_node(pi->hw, tm_node, hw_node) != 0) {
+ PMD_DRV_LOG(ERR, "Error configuring node %u at layer %u",
+ teid, parent->tx_sched_layer + 1);
+ return -1;
}
+ tm_node->sched_node = hw_node;
+ created[hw_node->tx_sched_layer]++;
}
-commit:
- pf->tm_conf.committed = true;
- pf->tm_conf.clear_on_fail = clear_on_fail;
+ /* if we have just created the child nodes in the q-group, i.e. last non-leaf layer,
+ * then just return, rather than trying to create leaf nodes.
+ * That is done later at queue start.
+ */
+ if (sw_node->level + 2 == ice_get_leaf_level(pi->hw))
+ return 0;
- return ret_val;
+ for (uint16_t i = 0; i < sw_node->reference_count; i++) {
+ if (sw_node->children[i]->reference_count == 0)
+ continue;
-reset_leaf:
- ice_remove_leaf_nodes(dev);
-add_leaf:
- ice_add_leaf_nodes(dev);
- ice_reset_noleaf_nodes(dev);
-fail_clear:
- /* clear all the traffic manager configuration */
- if (clear_on_fail) {
- ice_tm_conf_uninit(dev);
- ice_tm_conf_init(dev);
+ if (create_sched_node_recursive(pi, sw_node->children[i], hw_root, created) < 0)
+ return -1;
}
- return ret_val;
+ return 0;
}
-static int ice_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error)
+static int
+apply_topology_updates(struct rte_eth_dev *dev __rte_unused)
{
+ return 0;
+}
+
+static int
+commit_new_hierarchy(struct rte_eth_dev *dev)
+{
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_port_info *pi = hw->port_info;
+ struct ice_tm_node *sw_root = pf->tm_conf.root;
+ struct ice_sched_node *new_vsi_root = (pi->has_tc) ? pi->root->children[0] : pi->root;
+ uint16_t nodes_created_per_level[10] = {0}; /* counted per hw level, not per logical */
+ uint8_t q_lvl = ice_get_leaf_level(hw);
+ uint8_t qg_lvl = q_lvl - 1;
+
+ /* check if we have a previously applied topology */
+ if (sw_root->sched_node != NULL)
+ return apply_topology_updates(dev);
+
+ free_sched_node_recursive(pi, new_vsi_root, new_vsi_root, new_vsi_root->vsi_handle);
+
+ sw_root->sched_node = new_vsi_root;
+ if (create_sched_node_recursive(pi, sw_root, new_vsi_root, nodes_created_per_level) < 0)
+ return -1;
+ for (uint16_t i = 0; i < RTE_DIM(nodes_created_per_level); i++)
+ PMD_DRV_LOG(DEBUG, "Created %u nodes at level %u\n",
+ nodes_created_per_level[i], i);
+ hw->vsi_ctx[pf->main_vsi->idx]->sched.vsi_node[0] = new_vsi_root;
+
+ pf->main_vsi->nb_qps =
+ RTE_MIN(nodes_created_per_level[qg_lvl] * hw->max_children[qg_lvl],
+ hw->layer_info[q_lvl].max_device_nodes);
+
+ pf->tm_conf.committed = true; /* set flag to be checks on queue start */
+
+ return ice_alloc_lan_q_ctx(hw, 0, 0, pf->main_vsi->nb_qps);
+}
- /* if device not started, simply set committed flag and return. */
- if (!dev->data->dev_started) {
- pf->tm_conf.committed = true;
- pf->tm_conf.clear_on_fail = clear_on_fail;
- return 0;
+static int
+ice_hierarchy_commit(struct rte_eth_dev *dev,
+ int clear_on_fail,
+ struct rte_tm_error *error)
+{
+ RTE_SET_USED(error);
+ /* TODO - commit should only be done to topology before start! */
+ if (dev->data->dev_started)
+ return -1;
+
+ uint64_t start = rte_rdtsc();
+ int ret = commit_new_hierarchy(dev);
+ if (ret < 0 && clear_on_fail) {
+ ice_tm_conf_uninit(dev);
+ ice_tm_conf_init(dev);
}
-
- return ice_do_hierarchy_commit(dev, clear_on_fail, error);
+ uint64_t time = rte_rdtsc() - start;
+ PMD_DRV_LOG(DEBUG, "Time to apply hierarchy = %.1f\n", (float)time / rte_get_timer_hz());
+ return ret;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH 15/15] net/ice: add minimal capability reporting API
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (13 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 14/15] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
@ 2024-08-07 9:34 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
16 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:34 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Incomplete but reports number of available layers
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.h | 1 +
drivers/net/ice/ice_tm.c | 17 +++++++++++++++++
2 files changed, 18 insertions(+)
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index cb1a7e8e0d..6bebc511e4 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -682,6 +682,7 @@ int ice_rem_rss_cfg_wrap(struct ice_pf *pf, uint16_t vsi_id,
struct ice_rss_hash_cfg *cfg);
void ice_tm_conf_init(struct rte_eth_dev *dev);
void ice_tm_conf_uninit(struct rte_eth_dev *dev);
+
extern const struct rte_tm_ops ice_tm_ops;
static inline int
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index a86943a5b2..d7def61756 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -33,8 +33,12 @@ static int ice_shaper_profile_add(struct rte_eth_dev *dev,
static int ice_shaper_profile_del(struct rte_eth_dev *dev,
uint32_t shaper_profile_id,
struct rte_tm_error *error);
+static int ice_tm_capabilities_get(struct rte_eth_dev *dev,
+ struct rte_tm_capabilities *cap,
+ struct rte_tm_error *error);
const struct rte_tm_ops ice_tm_ops = {
+ .capabilities_get = ice_tm_capabilities_get,
.shaper_profile_add = ice_shaper_profile_add,
.shaper_profile_delete = ice_shaper_profile_del,
.node_add = ice_tm_node_add,
@@ -861,3 +865,16 @@ ice_hierarchy_commit(struct rte_eth_dev *dev,
PMD_DRV_LOG(DEBUG, "Time to apply hierarchy = %.1f\n", (float)time / rte_get_timer_hz());
return ret;
}
+
+static int
+ice_tm_capabilities_get(struct rte_eth_dev *dev, struct rte_tm_capabilities *cap,
+ struct rte_tm_error *error)
+{
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+ *cap = (struct rte_tm_capabilities){
+ .n_levels_max = hw->num_tx_sched_layers - hw->port_info->has_tc,
+ };
+ if (error)
+ error->type = RTE_TM_ERROR_TYPE_NONE;
+ return 0;
+}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 00/15] Improve rte_tm support in ICE driver
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (14 preceding siblings ...)
2024-08-07 9:34 ` [PATCH 15/15] net/ice: add minimal capability reporting API Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 01/15] net/ice: add traffic management node query function Bruce Richardson
` (14 more replies)
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
16 siblings, 15 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
This patchset expands the capabilities of the traffic management
support in the ICE driver. It allows the driver to support different
sizes of topologies, and support >256 queues and more than 3 hierarchy
layers.
---
Depends-on: series-32719 ("improve rte_rm APIs")
v2:
* Correct typo in commit log of one patch
* Add missing depends-on tag to the cover letter
Bruce Richardson (15):
net/ice: add traffic management node query function
net/ice: detect stopping a flow-director queue twice
net/ice: improve Tx scheduler graph output
net/ice: add option to choose DDP package file
net/ice: add option to download scheduler topology
net/ice/base: allow init without TC class sched nodes
net/ice/base: set VSI index on newly created nodes
net/ice/base: read VSI layer info from VSI
net/ice/base: remove 255 limit on sched child nodes
net/ice/base: optimize subtree searches
net/ice/base: make functions non-static
net/ice/base: remove flag checks before topology upload
net/ice: limit the number of queues to sched capabilities
net/ice: enhance Tx scheduler hierarchy support
net/ice: add minimal capability reporting API
doc/guides/nics/ice.rst | 9 +
drivers/net/ice/base/ice_ddp.c | 51 +--
drivers/net/ice/base/ice_ddp.h | 4 +-
drivers/net/ice/base/ice_sched.c | 56 ++--
drivers/net/ice/base/ice_sched.h | 8 +
drivers/net/ice/base/ice_type.h | 3 +-
drivers/net/ice/ice_diagnose.c | 196 ++++-------
drivers/net/ice/ice_ethdev.c | 92 +++--
drivers/net/ice/ice_ethdev.h | 18 +-
drivers/net/ice/ice_rxtx.c | 15 +
drivers/net/ice/ice_tm.c | 558 +++++++++++++++----------------
11 files changed, 487 insertions(+), 523 deletions(-)
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 01/15] net/ice: add traffic management node query function
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 02/15] net/ice: detect stopping a flow-director queue twice Bruce Richardson
` (13 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Implement the new node querying function for the "ice" net driver.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_tm.c | 48 ++++++++++++++++++++++++++++++++++++++++
1 file changed, 48 insertions(+)
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 8a29a9e744..459446a6b0 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -17,6 +17,11 @@ static int ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
uint32_t weight, uint32_t level_id,
const struct rte_tm_node_params *params,
struct rte_tm_error *error);
+static int ice_node_query(const struct rte_eth_dev *dev, uint32_t node_id,
+ uint32_t *parent_node_id, uint32_t *priority,
+ uint32_t *weight, uint32_t *level_id,
+ struct rte_tm_node_params *params,
+ struct rte_tm_error *error);
static int ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
struct rte_tm_error *error);
static int ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
@@ -35,6 +40,7 @@ const struct rte_tm_ops ice_tm_ops = {
.node_add = ice_tm_node_add,
.node_delete = ice_tm_node_delete,
.node_type_get = ice_node_type_get,
+ .node_query = ice_node_query,
.hierarchy_commit = ice_hierarchy_commit,
};
@@ -219,6 +225,48 @@ ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
return 0;
}
+static int
+ice_node_query(const struct rte_eth_dev *dev, uint32_t node_id,
+ uint32_t *parent_node_id, uint32_t *priority,
+ uint32_t *weight, uint32_t *level_id,
+ struct rte_tm_node_params *params,
+ struct rte_tm_error *error)
+{
+ struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_tm_node *tm_node;
+
+ if (node_id == RTE_TM_NODE_ID_NULL) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "invalid node id";
+ return -EINVAL;
+ }
+
+ /* check if the node id exists */
+ tm_node = find_node(pf->tm_conf.root, node_id);
+ if (!tm_node) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "no such node";
+ return -EEXIST;
+ }
+
+ if (parent_node_id != NULL)
+ *parent_node_id = tm_node->parent->id;
+
+ if (priority != NULL)
+ *priority = tm_node->priority;
+
+ if (weight != NULL)
+ *weight = tm_node->weight;
+
+ if (level_id != NULL)
+ *level_id = tm_node->level;
+
+ if (params != NULL)
+ *params = tm_node->params;
+
+ return 0;
+}
+
static inline struct ice_tm_shaper_profile *
ice_shaper_profile_search(struct rte_eth_dev *dev,
uint32_t shaper_profile_id)
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 02/15] net/ice: detect stopping a flow-director queue twice
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 01/15] net/ice: add traffic management node query function Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 03/15] net/ice: improve Tx scheduler graph output Bruce Richardson
` (12 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
If the flow-director queue is stopped at some point during the running
of an application, the shutdown procedure for the port issues an error
as it tries to stop the queue a second time, and fails to do so. We can
eliminate this error by setting the tail-register pointer to NULL on
stop, and checking for that condition in subsequent stop calls. Since
the register pointer is set on start, any restarting of the queue will
allow a stop call to progress as normal.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_rxtx.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index f270498ed1..a150d28e73 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -1139,6 +1139,10 @@ ice_fdir_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
tx_queue_id);
return -EINVAL;
}
+ if (txq->qtx_tail == NULL) {
+ PMD_DRV_LOG(INFO, "TX queue %u not started\n", tx_queue_id);
+ return 0;
+ }
vsi = txq->vsi;
q_ids[0] = txq->reg_idx;
@@ -1153,6 +1157,7 @@ ice_fdir_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
}
txq->tx_rel_mbufs(txq);
+ txq->qtx_tail = NULL;
return 0;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 03/15] net/ice: improve Tx scheduler graph output
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 01/15] net/ice: add traffic management node query function Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 02/15] net/ice: detect stopping a flow-director queue twice Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 04/15] net/ice: add option to choose DDP package file Bruce Richardson
` (11 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The function to dump the TX scheduler topology only adds to the chart
nodes connected to TX queues or for the flow director VSI. Change the
function to work recursively from the root node and thereby include all
scheduler nodes, whether in use or not, in the dump.
Also, improve the output of the Tx scheduler graphing function:
* Add VSI details to each node in graph
* When number of children is >16, skip middle nodes to reduce size of
the graph, otherwise dot output is unviewable for large hierarchies
* For VSIs other than zero, use dot's clustering method to put those
VSIs into subgraphs with borders
* For leaf nodes, display queue numbers for the any nodes assigned to
ethdev NIC Tx queues
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_diagnose.c | 196 ++++++++++++---------------------
1 file changed, 69 insertions(+), 127 deletions(-)
diff --git a/drivers/net/ice/ice_diagnose.c b/drivers/net/ice/ice_diagnose.c
index c357554707..623d84e37d 100644
--- a/drivers/net/ice/ice_diagnose.c
+++ b/drivers/net/ice/ice_diagnose.c
@@ -545,29 +545,15 @@ static void print_rl_profile(struct ice_aqc_rl_profile_elem *prof,
fprintf(stream, "\t\t\t\t\t</td>\n");
}
-static
-void print_elem_type(FILE *stream, u8 type)
+static const char *
+get_elem_type(u8 type)
{
- switch (type) {
- case 1:
- fprintf(stream, "root");
- break;
- case 2:
- fprintf(stream, "tc");
- break;
- case 3:
- fprintf(stream, "se_generic");
- break;
- case 4:
- fprintf(stream, "entry_point");
- break;
- case 5:
- fprintf(stream, "leaf");
- break;
- default:
- fprintf(stream, "%d", type);
- break;
- }
+ static const char * const ice_sched_node_types[] = {
+ "Undefined", "Root", "TC", "SE Generic", "SW Entry", "Leaf"
+ };
+ if (type < RTE_DIM(ice_sched_node_types))
+ return ice_sched_node_types[type];
+ return "*UNKNOWN*";
}
static
@@ -602,7 +588,9 @@ void print_priority_mode(FILE *stream, bool flag)
}
static
-void print_node(struct ice_aqc_txsched_elem_data *data,
+void print_node(struct ice_sched_node *node,
+ struct rte_eth_dev_data *ethdata,
+ struct ice_aqc_txsched_elem_data *data,
struct ice_aqc_rl_profile_elem *cir_prof,
struct ice_aqc_rl_profile_elem *eir_prof,
struct ice_aqc_rl_profile_elem *shared_prof,
@@ -613,17 +601,19 @@ void print_node(struct ice_aqc_txsched_elem_data *data,
fprintf(stream, "\t\t\t<table>\n");
- fprintf(stream, "\t\t\t\t<tr>\n");
- fprintf(stream, "\t\t\t\t\t<td> teid </td>\n");
- fprintf(stream, "\t\t\t\t\t<td> %d </td>\n", data->node_teid);
- fprintf(stream, "\t\t\t\t</tr>\n");
-
- fprintf(stream, "\t\t\t\t<tr>\n");
- fprintf(stream, "\t\t\t\t\t<td> type </td>\n");
- fprintf(stream, "\t\t\t\t\t<td>");
- print_elem_type(stream, data->data.elem_type);
- fprintf(stream, "</td>\n");
- fprintf(stream, "\t\t\t\t</tr>\n");
+ fprintf(stream, "\t\t\t\t<tr><td>teid</td><td>%d</td></tr>\n", data->node_teid);
+ fprintf(stream, "\t\t\t\t<tr><td>type</td><td>%s</td></tr>\n",
+ get_elem_type(data->data.elem_type));
+ fprintf(stream, "\t\t\t\t<tr><td>VSI</td><td>%u</td></tr>\n", node->vsi_handle);
+ if (data->data.elem_type == ICE_AQC_ELEM_TYPE_LEAF) {
+ for (uint16_t i = 0; i < ethdata->nb_tx_queues; i++) {
+ struct ice_tx_queue *q = ethdata->tx_queues[i];
+ if (q->q_teid == data->node_teid) {
+ fprintf(stream, "\t\t\t\t<tr><td>TXQ</td><td>%u</td></tr>\n", i);
+ break;
+ }
+ }
+ }
if (!detail)
goto brief;
@@ -705,8 +695,6 @@ void print_node(struct ice_aqc_txsched_elem_data *data,
fprintf(stream, "\t\tshape=plain\n");
fprintf(stream, "\t]\n");
- if (data->parent_teid != 0xFFFFFFFF)
- fprintf(stream, "\tNODE_%d -> NODE_%d\n", data->parent_teid, data->node_teid);
}
static
@@ -731,112 +719,92 @@ int query_rl_profile(struct ice_hw *hw,
return 0;
}
-static
-int query_node(struct ice_hw *hw, uint32_t child, uint32_t *parent,
- uint8_t level, bool detail, FILE *stream)
+static int
+query_node(struct ice_hw *hw, struct rte_eth_dev_data *ethdata,
+ struct ice_sched_node *node, bool detail, FILE *stream)
{
- struct ice_aqc_txsched_elem_data data;
+ struct ice_aqc_txsched_elem_data *data = &node->info;
struct ice_aqc_rl_profile_elem cir_prof;
struct ice_aqc_rl_profile_elem eir_prof;
struct ice_aqc_rl_profile_elem shared_prof;
struct ice_aqc_rl_profile_elem *cp = NULL;
struct ice_aqc_rl_profile_elem *ep = NULL;
struct ice_aqc_rl_profile_elem *sp = NULL;
- int status, ret;
-
- status = ice_sched_query_elem(hw, child, &data);
- if (status != ICE_SUCCESS) {
- if (level == hw->num_tx_sched_layers) {
- /* ignore the error when a queue has been stopped. */
- PMD_DRV_LOG(WARNING, "Failed to query queue node %d.", child);
- *parent = 0xffffffff;
- return 0;
- }
- PMD_DRV_LOG(ERR, "Failed to query scheduling node %d.", child);
- return -EINVAL;
- }
-
- *parent = data.parent_teid;
+ u8 level = node->tx_sched_layer;
+ int ret;
- if (data.data.cir_bw.bw_profile_idx != 0) {
- ret = query_rl_profile(hw, level, 0, data.data.cir_bw.bw_profile_idx, &cir_prof);
+ if (data->data.cir_bw.bw_profile_idx != 0) {
+ ret = query_rl_profile(hw, level, 0, data->data.cir_bw.bw_profile_idx, &cir_prof);
if (ret)
return ret;
cp = &cir_prof;
}
- if (data.data.eir_bw.bw_profile_idx != 0) {
- ret = query_rl_profile(hw, level, 1, data.data.eir_bw.bw_profile_idx, &eir_prof);
+ if (data->data.eir_bw.bw_profile_idx != 0) {
+ ret = query_rl_profile(hw, level, 1, data->data.eir_bw.bw_profile_idx, &eir_prof);
if (ret)
return ret;
ep = &eir_prof;
}
- if (data.data.srl_id != 0) {
- ret = query_rl_profile(hw, level, 2, data.data.srl_id, &shared_prof);
+ if (data->data.srl_id != 0) {
+ ret = query_rl_profile(hw, level, 2, data->data.srl_id, &shared_prof);
if (ret)
return ret;
sp = &shared_prof;
}
- print_node(&data, cp, ep, sp, detail, stream);
+ print_node(node, ethdata, data, cp, ep, sp, detail, stream);
return 0;
}
-static
-int query_nodes(struct ice_hw *hw,
- uint32_t *children, int child_num,
- uint32_t *parents, int *parent_num,
- uint8_t level, bool detail,
- FILE *stream)
+static int
+query_node_recursive(struct ice_hw *hw, struct rte_eth_dev_data *ethdata,
+ struct ice_sched_node *node, bool detail, FILE *stream)
{
- uint32_t parent;
- int i;
- int j;
-
- *parent_num = 0;
- for (i = 0; i < child_num; i++) {
- bool exist = false;
- int ret;
+ bool close = false;
+ if (node->parent != NULL && node->vsi_handle != node->parent->vsi_handle) {
+ fprintf(stream, "subgraph cluster_%u {\n", node->vsi_handle);
+ fprintf(stream, "\tlabel = \"VSI %u\";\n", node->vsi_handle);
+ close = true;
+ }
- ret = query_node(hw, children[i], &parent, level, detail, stream);
- if (ret)
- return -EINVAL;
+ int ret = query_node(hw, ethdata, node, detail, stream);
+ if (ret != 0)
+ return ret;
- for (j = 0; j < *parent_num; j++) {
- if (parents[j] == parent) {
- exist = true;
- break;
- }
+ for (uint16_t i = 0; i < node->num_children; i++) {
+ ret = query_node_recursive(hw, ethdata, node->children[i], detail, stream);
+ if (ret != 0)
+ return ret;
+ /* if we have a lot of nodes, skip a bunch in the middle */
+ if (node->num_children > 16 && i == 2) {
+ uint16_t inc = node->num_children - 5;
+ fprintf(stream, "\tn%d_children [label=\"... +%d child nodes ...\"];\n",
+ node->info.node_teid, inc);
+ fprintf(stream, "\tNODE_%d -> n%d_children;\n",
+ node->info.node_teid, node->info.node_teid);
+ i += inc;
}
-
- if (!exist && parent != 0xFFFFFFFF)
- parents[(*parent_num)++] = parent;
}
+ if (close)
+ fprintf(stream, "}\n");
+ if (node->info.parent_teid != 0xFFFFFFFF)
+ fprintf(stream, "\tNODE_%d -> NODE_%d\n",
+ node->info.parent_teid, node->info.node_teid);
return 0;
}
-int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
+int
+rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
{
struct rte_eth_dev *dev;
struct ice_hw *hw;
- struct ice_pf *pf;
- struct ice_q_ctx *q_ctx;
- uint16_t q_num;
- uint16_t i;
- struct ice_tx_queue *txq;
- uint32_t buf1[256];
- uint32_t buf2[256];
- uint32_t *children = buf1;
- uint32_t *parents = buf2;
- int child_num = 0;
- int parent_num = 0;
- uint8_t level;
RTE_ETH_VALID_PORTID_OR_ERR_RET(port, -ENODEV);
@@ -846,35 +814,9 @@ int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
dev = &rte_eth_devices[port];
hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- level = hw->num_tx_sched_layers;
-
- q_num = dev->data->nb_tx_queues;
-
- /* main vsi */
- for (i = 0; i < q_num; i++) {
- txq = dev->data->tx_queues[i];
- q_ctx = ice_get_lan_q_ctx(hw, txq->vsi->idx, 0, i);
- children[child_num++] = q_ctx->q_teid;
- }
-
- /* fdir vsi */
- q_ctx = ice_get_lan_q_ctx(hw, pf->fdir.fdir_vsi->idx, 0, 0);
- children[child_num++] = q_ctx->q_teid;
fprintf(stream, "digraph tx_sched {\n");
- while (child_num > 0) {
- int ret;
- ret = query_nodes(hw, children, child_num,
- parents, &parent_num,
- level, detail, stream);
- if (ret)
- return ret;
-
- children = parents;
- child_num = parent_num;
- level--;
- }
+ query_node_recursive(hw, dev->data, hw->port_info->root, detail, stream);
fprintf(stream, "}\n");
return 0;
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 04/15] net/ice: add option to choose DDP package file
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (2 preceding siblings ...)
2024-08-07 9:46 ` [PATCH v2 03/15] net/ice: improve Tx scheduler graph output Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 05/15] net/ice: add option to download scheduler topology Bruce Richardson
` (10 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The "Dynamic Device Personalization" package is loaded at initialization
time by the driver, but the specific package file loaded depends upon
what package file is found first by searching through a hard-coded list
of firmware paths. To enable greater control over the package loading,
we can add a device option to choose a specific DDP package file to
load.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
doc/guides/nics/ice.rst | 9 +++++++++
drivers/net/ice/ice_ethdev.c | 34 ++++++++++++++++++++++++++++++++++
drivers/net/ice/ice_ethdev.h | 1 +
3 files changed, 44 insertions(+)
diff --git a/doc/guides/nics/ice.rst b/doc/guides/nics/ice.rst
index ae975d19ad..58ccfbd1a5 100644
--- a/doc/guides/nics/ice.rst
+++ b/doc/guides/nics/ice.rst
@@ -108,6 +108,15 @@ Runtime Configuration
-a 80:00.0,default-mac-disable=1
+- ``DDP Package File``
+
+ Rather than have the driver search for the DDP package to load,
+ or to override what package is used,
+ the ``ddp_pkg_file`` option can be used to provide the path to a specific package file.
+ For example::
+
+ -a 80:00.0,ddp_pkg_file=/path/to/ice-version.pkg
+
- ``Protocol extraction for per queue``
Configure the RX queues to do protocol extraction into mbuf for protocol
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 304f959b7e..3e7ceda9ce 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -36,6 +36,7 @@
#define ICE_ONE_PPS_OUT_ARG "pps_out"
#define ICE_RX_LOW_LATENCY_ARG "rx_low_latency"
#define ICE_MBUF_CHECK_ARG "mbuf_check"
+#define ICE_DDP_FILENAME "ddp_pkg_file"
#define ICE_CYCLECOUNTER_MASK 0xffffffffffffffffULL
@@ -52,6 +53,7 @@ static const char * const ice_valid_args[] = {
ICE_RX_LOW_LATENCY_ARG,
ICE_DEFAULT_MAC_DISABLE,
ICE_MBUF_CHECK_ARG,
+ ICE_DDP_FILENAME,
NULL
};
@@ -692,6 +694,18 @@ handle_field_name_arg(__rte_unused const char *key, const char *value,
return 0;
}
+static int
+handle_ddp_filename_arg(__rte_unused const char *key, const char *value, void *name_args)
+{
+ const char **filename = name_args;
+ if (strlen(value) >= ICE_MAX_PKG_FILENAME_SIZE) {
+ PMD_DRV_LOG(ERR, "The DDP package filename is too long : '%s'", value);
+ return -1;
+ }
+ *filename = strdup(value);
+ return 0;
+}
+
static void
ice_check_proto_xtr_support(struct ice_hw *hw)
{
@@ -1882,6 +1896,16 @@ int ice_load_pkg(struct ice_adapter *adapter, bool use_dsn, uint64_t dsn)
size_t bufsz;
int err;
+ if (adapter->devargs.ddp_filename != NULL) {
+ strlcpy(pkg_file, adapter->devargs.ddp_filename, sizeof(pkg_file));
+ if (rte_firmware_read(pkg_file, &buf, &bufsz) == 0) {
+ goto load_fw;
+ } else {
+ PMD_INIT_LOG(ERR, "Cannot load DDP file: %s\n", pkg_file);
+ return -1;
+ }
+ }
+
if (!use_dsn)
goto no_dsn;
@@ -2216,6 +2240,13 @@ static int ice_parse_devargs(struct rte_eth_dev *dev)
ret = rte_kvargs_process(kvlist, ICE_RX_LOW_LATENCY_ARG,
&parse_bool, &ad->devargs.rx_low_latency);
+ if (ret)
+ goto bail;
+
+ ret = rte_kvargs_process(kvlist, ICE_DDP_FILENAME,
+ &handle_ddp_filename_arg, &ad->devargs.ddp_filename);
+ if (ret)
+ goto bail;
bail:
rte_kvargs_free(kvlist);
@@ -2689,6 +2720,8 @@ ice_dev_close(struct rte_eth_dev *dev)
ice_free_hw_tbls(hw);
rte_free(hw->port_info);
hw->port_info = NULL;
+ free((void *)(uintptr_t)ad->devargs.ddp_filename);
+ ad->devargs.ddp_filename = NULL;
ice_shutdown_all_ctrlq(hw, true);
rte_free(pf->proto_xtr);
pf->proto_xtr = NULL;
@@ -6981,6 +7014,7 @@ RTE_PMD_REGISTER_PARAM_STRING(net_ice,
ICE_PROTO_XTR_ARG "=[queue:]<vlan|ipv4|ipv6|ipv6_flow|tcp|ip_offset>"
ICE_SAFE_MODE_SUPPORT_ARG "=<0|1>"
ICE_DEFAULT_MAC_DISABLE "=<0|1>"
+ ICE_DDP_FILENAME "=</path/to/file>"
ICE_RX_LOW_LATENCY_ARG "=<0|1>");
RTE_LOG_REGISTER_SUFFIX(ice_logtype_init, init, NOTICE);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index 3ea9f37dc8..c211b5b9cc 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -568,6 +568,7 @@ struct ice_devargs {
/* Name of the field. */
char xtr_field_name[RTE_MBUF_DYN_NAMESIZE];
uint64_t mbuf_check;
+ const char *ddp_filename;
};
/**
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 05/15] net/ice: add option to download scheduler topology
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (3 preceding siblings ...)
2024-08-07 9:46 ` [PATCH v2 04/15] net/ice: add option to choose DDP package file Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 06/15] net/ice/base: allow init without TC class sched nodes Bruce Richardson
` (9 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The DDP package file being loaded at init time may contain an
alternative Tx Scheduler topology in it. Add driver option to load this
topology at init time.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_ddp.c | 18 +++++++++++++++---
drivers/net/ice/base/ice_ddp.h | 4 ++--
drivers/net/ice/ice_ethdev.c | 24 +++++++++++++++---------
drivers/net/ice/ice_ethdev.h | 1 +
4 files changed, 33 insertions(+), 14 deletions(-)
diff --git a/drivers/net/ice/base/ice_ddp.c b/drivers/net/ice/base/ice_ddp.c
index 24506dfaea..e6c42c5274 100644
--- a/drivers/net/ice/base/ice_ddp.c
+++ b/drivers/net/ice/base/ice_ddp.c
@@ -1326,7 +1326,7 @@ ice_fill_hw_ptype(struct ice_hw *hw)
* ice_copy_and_init_pkg() instead of directly calling ice_init_pkg() in this
* case.
*/
-enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
+enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len, bool load_sched)
{
bool already_loaded = false;
enum ice_ddp_state state;
@@ -1344,6 +1344,18 @@ enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
return state;
}
+ if (load_sched) {
+ enum ice_status res = ice_cfg_tx_topo(hw, buf, len);
+ if (res != ICE_SUCCESS) {
+ ice_debug(hw, ICE_DBG_INIT, "failed to apply sched topology (err: %d)\n",
+ res);
+ return ICE_DDP_PKG_ERR;
+ }
+ ice_debug(hw, ICE_DBG_INIT, "Topology download successful, reinitializing device\n");
+ ice_deinit_hw(hw);
+ ice_init_hw(hw);
+ }
+
/* initialize package info */
state = ice_init_pkg_info(hw, pkg);
if (state)
@@ -1416,7 +1428,7 @@ enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
* related routines.
*/
enum ice_ddp_state
-ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len)
+ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len, bool load_sched)
{
enum ice_ddp_state state;
u8 *buf_copy;
@@ -1426,7 +1438,7 @@ ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len)
buf_copy = (u8 *)ice_memdup(hw, buf, len, ICE_NONDMA_TO_NONDMA);
- state = ice_init_pkg(hw, buf_copy, len);
+ state = ice_init_pkg(hw, buf_copy, len, load_sched);
if (!ice_is_init_pkg_successful(state)) {
/* Free the copy, since we failed to initialize the package */
ice_free(hw, buf_copy);
diff --git a/drivers/net/ice/base/ice_ddp.h b/drivers/net/ice/base/ice_ddp.h
index 5761920207..2feba2e91d 100644
--- a/drivers/net/ice/base/ice_ddp.h
+++ b/drivers/net/ice/base/ice_ddp.h
@@ -451,9 +451,9 @@ ice_pkg_enum_entry(struct ice_seg *ice_seg, struct ice_pkg_enum *state,
void *
ice_pkg_enum_section(struct ice_seg *ice_seg, struct ice_pkg_enum *state,
u32 sect_type);
-enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buff, u32 len);
+enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buff, u32 len, bool load_sched);
enum ice_ddp_state
-ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len);
+ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len, bool load_sched);
bool ice_is_init_pkg_successful(enum ice_ddp_state state);
void ice_free_seg(struct ice_hw *hw);
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 3e7ceda9ce..0d2445a317 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -37,6 +37,7 @@
#define ICE_RX_LOW_LATENCY_ARG "rx_low_latency"
#define ICE_MBUF_CHECK_ARG "mbuf_check"
#define ICE_DDP_FILENAME "ddp_pkg_file"
+#define ICE_DDP_LOAD_SCHED "ddp_load_sched_topo"
#define ICE_CYCLECOUNTER_MASK 0xffffffffffffffffULL
@@ -54,6 +55,7 @@ static const char * const ice_valid_args[] = {
ICE_DEFAULT_MAC_DISABLE,
ICE_MBUF_CHECK_ARG,
ICE_DDP_FILENAME,
+ ICE_DDP_LOAD_SCHED,
NULL
};
@@ -1938,7 +1940,7 @@ int ice_load_pkg(struct ice_adapter *adapter, bool use_dsn, uint64_t dsn)
load_fw:
PMD_INIT_LOG(DEBUG, "DDP package name: %s", pkg_file);
- err = ice_copy_and_init_pkg(hw, buf, bufsz);
+ err = ice_copy_and_init_pkg(hw, buf, bufsz, adapter->devargs.ddp_load_sched);
if (!ice_is_init_pkg_successful(err)) {
PMD_INIT_LOG(ERR, "ice_copy_and_init_hw failed: %d\n", err);
free(buf);
@@ -1971,19 +1973,18 @@ static int
parse_bool(const char *key, const char *value, void *args)
{
int *i = (int *)args;
- char *end;
- int num;
- num = strtoul(value, &end, 10);
-
- if (num != 0 && num != 1) {
- PMD_DRV_LOG(WARNING, "invalid value:\"%s\" for key:\"%s\", "
- "value must be 0 or 1",
+ if (value == NULL || value[0] == '\0') {
+ PMD_DRV_LOG(WARNING, "key:\"%s\", requires a value, which must be 0 or 1", key);
+ return -1;
+ }
+ if (value[1] != '\0' || (value[0] != '0' && value[0] != '1')) {
+ PMD_DRV_LOG(WARNING, "invalid value:\"%s\" for key:\"%s\", value must be 0 or 1",
value, key);
return -1;
}
- *i = num;
+ *i = value[0] - '0';
return 0;
}
@@ -2248,6 +2249,10 @@ static int ice_parse_devargs(struct rte_eth_dev *dev)
if (ret)
goto bail;
+ ret = rte_kvargs_process(kvlist, ICE_DDP_LOAD_SCHED,
+ &parse_bool, &ad->devargs.ddp_load_sched);
+ if (ret)
+ goto bail;
bail:
rte_kvargs_free(kvlist);
return ret;
@@ -7014,6 +7019,7 @@ RTE_PMD_REGISTER_PARAM_STRING(net_ice,
ICE_PROTO_XTR_ARG "=[queue:]<vlan|ipv4|ipv6|ipv6_flow|tcp|ip_offset>"
ICE_SAFE_MODE_SUPPORT_ARG "=<0|1>"
ICE_DEFAULT_MAC_DISABLE "=<0|1>"
+ ICE_DDP_LOAD_SCHED "=<0|1>"
ICE_DDP_FILENAME "=</path/to/file>"
ICE_RX_LOW_LATENCY_ARG "=<0|1>");
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index c211b5b9cc..f31addb122 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -563,6 +563,7 @@ struct ice_devargs {
uint8_t proto_xtr[ICE_MAX_QUEUE_NUM];
uint8_t pin_idx;
uint8_t pps_out_ena;
+ int ddp_load_sched;
int xtr_field_offs;
uint8_t xtr_flag_offs[PROTO_XTR_MAX];
/* Name of the field. */
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 06/15] net/ice/base: allow init without TC class sched nodes
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (4 preceding siblings ...)
2024-08-07 9:46 ` [PATCH v2 05/15] net/ice: add option to download scheduler topology Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 07/15] net/ice/base: set VSI index on newly created nodes Bruce Richardson
` (8 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
If DCB support is disabled via DDP image, there will not be any traffic
class (TC) nodes in the scheduler tree immediately above the root level.
To allow the driver to work with this scenario, we allow use of the root
node as a dummy TC0 node in case where there are no TC nodes in the
tree. For use of any other TC other than 0 (used by default in the
driver), existing behaviour of returning NULL pointer is maintained.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 6 ++++++
drivers/net/ice/base/ice_type.h | 1 +
2 files changed, 7 insertions(+)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index 373c32a518..f75e5ae599 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -292,6 +292,10 @@ struct ice_sched_node *ice_sched_get_tc_node(struct ice_port_info *pi, u8 tc)
if (!pi || !pi->root)
return NULL;
+ /* if no TC nodes, use root as TC node 0 */
+ if (pi->has_tc == 0)
+ return tc == 0 ? pi->root : NULL;
+
for (i = 0; i < pi->root->num_children; i++)
if (pi->root->children[i]->tc_num == tc)
return pi->root->children[i];
@@ -1306,6 +1310,8 @@ int ice_sched_init_port(struct ice_port_info *pi)
ICE_AQC_ELEM_TYPE_ENTRY_POINT)
hw->sw_entry_point_layer = j;
+ if (buf[0].generic[j].data.elem_type == ICE_AQC_ELEM_TYPE_TC)
+ pi->has_tc = 1;
status = ice_sched_add_node(pi, j, &buf[i].generic[j], NULL);
if (status)
goto err_init_port;
diff --git a/drivers/net/ice/base/ice_type.h b/drivers/net/ice/base/ice_type.h
index 598a80155b..a70e4a8afa 100644
--- a/drivers/net/ice/base/ice_type.h
+++ b/drivers/net/ice/base/ice_type.h
@@ -1260,6 +1260,7 @@ struct ice_port_info {
struct ice_qos_cfg qos_cfg;
u8 is_vf:1;
u8 is_custom_tx_enabled:1;
+ u8 has_tc:1;
};
struct ice_switch_info {
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 07/15] net/ice/base: set VSI index on newly created nodes
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (5 preceding siblings ...)
2024-08-07 9:46 ` [PATCH v2 06/15] net/ice/base: allow init without TC class sched nodes Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 08/15] net/ice/base: read VSI layer info from VSI Bruce Richardson
` (7 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The ice_sched_node type has got a field for the vsi to which the node
belongs. This field was not getting set in "ice_sched_add_node", so add
a line configuring this field for each node from its parent node.
Similarly, when searching for a qgroup node, we can check for each node
that the VSI information is correct.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f75e5ae599..f6dc5ae173 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -200,6 +200,7 @@ ice_sched_add_node(struct ice_port_info *pi, u8 layer,
node->in_use = true;
node->parent = parent;
node->tx_sched_layer = layer;
+ node->vsi_handle = parent->vsi_handle;
parent->children[parent->num_children++] = node;
node->info = elem;
return 0;
@@ -1581,7 +1582,7 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
/* make sure the qgroup node is part of the VSI subtree */
if (ice_sched_find_node_in_subtree(pi->hw, vsi_node, qgrp_node))
if (qgrp_node->num_children < max_children &&
- qgrp_node->owner == owner)
+ qgrp_node->owner == owner && qgrp_node->vsi_handle == vsi_handle)
break;
qgrp_node = qgrp_node->sibling;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 08/15] net/ice/base: read VSI layer info from VSI
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (6 preceding siblings ...)
2024-08-07 9:46 ` [PATCH v2 07/15] net/ice/base: set VSI index on newly created nodes Bruce Richardson
@ 2024-08-07 9:46 ` Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 09/15] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
` (6 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:46 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Rather than computing from the number of HW layers the layer of the VSI,
we can instead just read that info from the VSI node itself. This allows
the layer to be changed at runtime.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f6dc5ae173..e398984bf2 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -1559,7 +1559,6 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
u16 max_children;
qgrp_layer = ice_sched_get_qgrp_layer(pi->hw);
- vsi_layer = ice_sched_get_vsi_layer(pi->hw);
max_children = pi->hw->max_children[qgrp_layer];
vsi_ctx = ice_get_vsi_ctx(pi->hw, vsi_handle);
@@ -1569,6 +1568,7 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
/* validate invalid VSI ID */
if (!vsi_node)
return NULL;
+ vsi_layer = vsi_node->tx_sched_layer;
/* If the queue group and vsi layer are same then queues
* are all attached directly to VSI
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 09/15] net/ice/base: remove 255 limit on sched child nodes
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (7 preceding siblings ...)
2024-08-07 9:46 ` [PATCH v2 08/15] net/ice/base: read VSI layer info from VSI Bruce Richardson
@ 2024-08-07 9:47 ` Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 10/15] net/ice/base: optimize subtree searches Bruce Richardson
` (5 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:47 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The Tx scheduler in the ice driver can be configured to have large
numbers of child nodes at a given layer, but the driver code implicitly
limited the number of nodes to 255 by using a u8 datatype for the number
of children. Increase this to a 16-bit value throughout the code.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 25 ++++++++++++++-----------
drivers/net/ice/base/ice_type.h | 2 +-
2 files changed, 15 insertions(+), 12 deletions(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index e398984bf2..be13833e1e 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -289,7 +289,7 @@ ice_sched_get_first_node(struct ice_port_info *pi,
*/
struct ice_sched_node *ice_sched_get_tc_node(struct ice_port_info *pi, u8 tc)
{
- u8 i;
+ u16 i;
if (!pi || !pi->root)
return NULL;
@@ -316,7 +316,7 @@ void ice_free_sched_node(struct ice_port_info *pi, struct ice_sched_node *node)
{
struct ice_sched_node *parent;
struct ice_hw *hw = pi->hw;
- u8 i, j;
+ u16 i, j;
/* Free the children before freeing up the parent node
* The parent array is updated below and that shifts the nodes
@@ -1473,7 +1473,7 @@ bool
ice_sched_find_node_in_subtree(struct ice_hw *hw, struct ice_sched_node *base,
struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
for (i = 0; i < base->num_children; i++) {
struct ice_sched_node *child = base->children[i];
@@ -1510,7 +1510,7 @@ ice_sched_get_free_qgrp(struct ice_port_info *pi,
struct ice_sched_node *qgrp_node, u8 owner)
{
struct ice_sched_node *min_qgrp;
- u8 min_children;
+ u16 min_children;
if (!qgrp_node)
return qgrp_node;
@@ -2070,7 +2070,7 @@ static void ice_sched_rm_agg_vsi_info(struct ice_port_info *pi, u16 vsi_handle)
*/
static bool ice_sched_is_leaf_node_present(struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
for (i = 0; i < node->num_children; i++)
if (ice_sched_is_leaf_node_present(node->children[i]))
@@ -2105,7 +2105,7 @@ ice_sched_rm_vsi_cfg(struct ice_port_info *pi, u16 vsi_handle, u8 owner)
ice_for_each_traffic_class(i) {
struct ice_sched_node *vsi_node, *tc_node;
- u8 j = 0;
+ u16 j = 0;
tc_node = ice_sched_get_tc_node(pi, i);
if (!tc_node)
@@ -2173,7 +2173,7 @@ int ice_rm_vsi_lan_cfg(struct ice_port_info *pi, u16 vsi_handle)
*/
bool ice_sched_is_tree_balanced(struct ice_hw *hw, struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
/* start from the leaf node */
for (i = 0; i < node->num_children; i++)
@@ -2247,7 +2247,8 @@ ice_sched_get_free_vsi_parent(struct ice_hw *hw, struct ice_sched_node *node,
u16 *num_nodes)
{
u8 l = node->tx_sched_layer;
- u8 vsil, i;
+ u8 vsil;
+ u16 i;
vsil = ice_sched_get_vsi_layer(hw);
@@ -2289,7 +2290,7 @@ ice_sched_update_parent(struct ice_sched_node *new_parent,
struct ice_sched_node *node)
{
struct ice_sched_node *old_parent;
- u8 i, j;
+ u16 i, j;
old_parent = node->parent;
@@ -2389,7 +2390,8 @@ ice_sched_move_vsi_to_agg(struct ice_port_info *pi, u16 vsi_handle, u32 agg_id,
u16 num_nodes[ICE_AQC_TOPO_MAX_LEVEL_NUM] = { 0 };
u32 first_node_teid, vsi_teid;
u16 num_nodes_added;
- u8 aggl, vsil, i;
+ u8 aggl, vsil;
+ u16 i;
int status;
tc_node = ice_sched_get_tc_node(pi, tc);
@@ -2505,7 +2507,8 @@ ice_move_all_vsi_to_dflt_agg(struct ice_port_info *pi,
static bool
ice_sched_is_agg_inuse(struct ice_port_info *pi, struct ice_sched_node *node)
{
- u8 vsil, i;
+ u8 vsil;
+ u16 i;
vsil = ice_sched_get_vsi_layer(pi->hw);
if (node->tx_sched_layer < vsil - 1) {
diff --git a/drivers/net/ice/base/ice_type.h b/drivers/net/ice/base/ice_type.h
index a70e4a8afa..35f832eb9f 100644
--- a/drivers/net/ice/base/ice_type.h
+++ b/drivers/net/ice/base/ice_type.h
@@ -1030,9 +1030,9 @@ struct ice_sched_node {
struct ice_aqc_txsched_elem_data info;
u32 agg_id; /* aggregator group ID */
u16 vsi_handle;
+ u16 num_children;
u8 in_use; /* suspended or in use */
u8 tx_sched_layer; /* Logical Layer (1-9) */
- u8 num_children;
u8 tc_num;
u8 owner;
#define ICE_SCHED_NODE_OWNER_LAN 0
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 10/15] net/ice/base: optimize subtree searches
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (8 preceding siblings ...)
2024-08-07 9:47 ` [PATCH v2 09/15] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
@ 2024-08-07 9:47 ` Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 11/15] net/ice/base: make functions non-static Bruce Richardson
` (4 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:47 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
In a number of places throughout the driver code, we want to confirm
that a scheduler node is indeed a child of another node. Currently, this
is confirmed by searching down the tree from the base until the desired
node is hit, a search which may hit many irrelevant tree nodes when
recursing down wrong branches. By switching the direction of search, to
check upwards from the node to the parent, we can avoid any incorrect
paths, and so speed up processing.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 18 +++++-------------
1 file changed, 5 insertions(+), 13 deletions(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index be13833e1e..f7d5f8f415 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -1475,20 +1475,12 @@ ice_sched_find_node_in_subtree(struct ice_hw *hw, struct ice_sched_node *base,
{
u16 i;
- for (i = 0; i < base->num_children; i++) {
- struct ice_sched_node *child = base->children[i];
-
- if (node == child)
- return true;
-
- if (child->tx_sched_layer > node->tx_sched_layer)
- return false;
-
- /* this recursion is intentional, and wouldn't
- * go more than 8 calls
- */
- if (ice_sched_find_node_in_subtree(hw, child, node))
+ if (base == node)
+ return true;
+ while (node->tx_sched_layer != 0 && node->parent != NULL) {
+ if (node->parent == base)
return true;
+ node = node->parent;
}
return false;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 11/15] net/ice/base: make functions non-static
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (9 preceding siblings ...)
2024-08-07 9:47 ` [PATCH v2 10/15] net/ice/base: optimize subtree searches Bruce Richardson
@ 2024-08-07 9:47 ` Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 12/15] net/ice/base: remove flag checks before topology upload Bruce Richardson
` (3 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:47 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
We will need to allocate more lanq contexts after a scheduler rework, so
make that function non-static so accessible outside the file. For similar
reasons, make the function to add a Tx scheduler node non-static
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 2 +-
drivers/net/ice/base/ice_sched.h | 8 ++++++++
2 files changed, 9 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f7d5f8f415..d88b836c38 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -570,7 +570,7 @@ ice_sched_suspend_resume_elems(struct ice_hw *hw, u8 num_nodes, u32 *node_teids,
* @tc: TC number
* @new_numqs: number of queues
*/
-static int
+int
ice_alloc_lan_q_ctx(struct ice_hw *hw, u16 vsi_handle, u8 tc, u16 new_numqs)
{
struct ice_vsi_ctx *vsi_ctx;
diff --git a/drivers/net/ice/base/ice_sched.h b/drivers/net/ice/base/ice_sched.h
index 9f78516dfb..c7eb794963 100644
--- a/drivers/net/ice/base/ice_sched.h
+++ b/drivers/net/ice/base/ice_sched.h
@@ -270,4 +270,12 @@ int ice_sched_replay_q_bw(struct ice_port_info *pi, struct ice_q_ctx *q_ctx);
int
ice_sched_cfg_node_bw_alloc(struct ice_hw *hw, struct ice_sched_node *node,
enum ice_rl_type rl_type, u16 bw_alloc);
+
+int
+ice_sched_add_elems(struct ice_port_info *pi, struct ice_sched_node *tc_node,
+ struct ice_sched_node *parent, u8 layer, u16 num_nodes,
+ u16 *num_nodes_added, u32 *first_node_teid,
+ struct ice_sched_node **prealloc_nodes);
+int
+ice_alloc_lan_q_ctx(struct ice_hw *hw, u16 vsi_handle, u8 tc, u16 new_numqs);
#endif /* _ICE_SCHED_H_ */
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 12/15] net/ice/base: remove flag checks before topology upload
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (10 preceding siblings ...)
2024-08-07 9:47 ` [PATCH v2 11/15] net/ice/base: make functions non-static Bruce Richardson
@ 2024-08-07 9:47 ` Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 13/15] net/ice: limit the number of queues to sched capabilities Bruce Richardson
` (2 subsequent siblings)
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:47 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
DPDK should support more than just 9-level or 5-level topologies, so
remove the checks for those particular settings.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_ddp.c | 33 ---------------------------------
1 file changed, 33 deletions(-)
diff --git a/drivers/net/ice/base/ice_ddp.c b/drivers/net/ice/base/ice_ddp.c
index e6c42c5274..744f015fe5 100644
--- a/drivers/net/ice/base/ice_ddp.c
+++ b/drivers/net/ice/base/ice_ddp.c
@@ -2373,38 +2373,6 @@ int ice_cfg_tx_topo(struct ice_hw *hw, u8 *buf, u32 len)
return status;
}
- /* Is default topology already applied ? */
- if (!(flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 9) {
- ice_debug(hw, ICE_DBG_INIT, "Loaded default topology\n");
- /* Already default topology is loaded */
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Is new topology already applied ? */
- if ((flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 5) {
- ice_debug(hw, ICE_DBG_INIT, "Loaded new topology\n");
- /* Already new topology is loaded */
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Is set topology issued already ? */
- if (flags & ICE_AQC_TX_TOPO_FLAGS_ISSUED) {
- ice_debug(hw, ICE_DBG_INIT, "Update tx topology was done by another PF\n");
- /* add a small delay before exiting */
- for (i = 0; i < 20; i++)
- ice_msec_delay(100, true);
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Change the topology from new to default (5 to 9) */
- if (!(flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 5) {
- ice_debug(hw, ICE_DBG_INIT, "Change topology from 5 to 9 layers\n");
- goto update_topo;
- }
-
pkg_hdr = (struct ice_pkg_hdr *)buf;
state = ice_verify_pkg(pkg_hdr, len);
if (state) {
@@ -2451,7 +2419,6 @@ int ice_cfg_tx_topo(struct ice_hw *hw, u8 *buf, u32 len)
/* Get the new topology buffer */
new_topo = ((u8 *)section) + offset;
-update_topo:
/* acquire global lock to make sure that set topology issued
* by one PF
*/
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 13/15] net/ice: limit the number of queues to sched capabilities
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (11 preceding siblings ...)
2024-08-07 9:47 ` [PATCH v2 12/15] net/ice/base: remove flag checks before topology upload Bruce Richardson
@ 2024-08-07 9:47 ` Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 14/15] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 15/15] net/ice: add minimal capability reporting API Bruce Richardson
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:47 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Rather than assuming that each VSI can hold up to 256 queue pairs,
or the reported device limit, query the available nodes in the scheduler
tree to check that we are not overflowing the limit for number of
child scheduling nodes at each level. Do this by multiplying
max_children for each level beyond the VSI and using that as an
additional cap on the number of queues.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.c | 25 ++++++++++++++++++++-----
1 file changed, 20 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 0d2445a317..ab3f88fd7d 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -913,7 +913,7 @@ ice_vsi_config_default_rss(struct ice_aqc_vsi_props *info)
}
static int
-ice_vsi_config_tc_queue_mapping(struct ice_vsi *vsi,
+ice_vsi_config_tc_queue_mapping(struct ice_hw *hw, struct ice_vsi *vsi,
struct ice_aqc_vsi_props *info,
uint8_t enabled_tcmap)
{
@@ -929,13 +929,28 @@ ice_vsi_config_tc_queue_mapping(struct ice_vsi *vsi,
}
/* vector 0 is reserved and 1 vector for ctrl vsi */
- if (vsi->adapter->hw.func_caps.common_cap.num_msix_vectors < 2)
+ if (vsi->adapter->hw.func_caps.common_cap.num_msix_vectors < 2) {
vsi->nb_qps = 0;
- else
+ } else {
vsi->nb_qps = RTE_MIN
((uint16_t)vsi->adapter->hw.func_caps.common_cap.num_msix_vectors - 2,
RTE_MIN(vsi->nb_qps, ICE_MAX_Q_PER_TC));
+ /* cap max QPs to what the HW reports as num-children for each layer.
+ * Multiply num_children for each layer from the entry_point layer to
+ * the qgroup, or second-last layer.
+ * Avoid any potential overflow by using uint32_t type and breaking loop
+ * once we have a number greater than the already configured max.
+ */
+ uint32_t max_sched_vsi_nodes = 1;
+ for (uint8_t i = hw->sw_entry_point_layer; i < hw->num_tx_sched_layers - 1; i++) {
+ max_sched_vsi_nodes *= hw->max_children[i];
+ if (max_sched_vsi_nodes >= vsi->nb_qps)
+ break;
+ }
+ vsi->nb_qps = RTE_MIN(vsi->nb_qps, max_sched_vsi_nodes);
+ }
+
/* nb_qps(hex) -> fls */
/* 0000 -> 0 */
/* 0001 -> 0 */
@@ -1707,7 +1722,7 @@ ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type)
rte_cpu_to_le_16(hw->func_caps.fd_fltr_best_effort);
/* Enable VLAN/UP trip */
- ret = ice_vsi_config_tc_queue_mapping(vsi,
+ ret = ice_vsi_config_tc_queue_mapping(hw, vsi,
&vsi_ctx.info,
ICE_DEFAULT_TCMAP);
if (ret) {
@@ -1731,7 +1746,7 @@ ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type)
vsi_ctx.info.fd_options = rte_cpu_to_le_16(cfg);
vsi_ctx.info.sw_id = hw->port_info->sw_id;
vsi_ctx.info.sw_flags2 = ICE_AQ_VSI_SW_FLAG_LAN_ENA;
- ret = ice_vsi_config_tc_queue_mapping(vsi,
+ ret = ice_vsi_config_tc_queue_mapping(hw, vsi,
&vsi_ctx.info,
ICE_DEFAULT_TCMAP);
if (ret) {
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 14/15] net/ice: enhance Tx scheduler hierarchy support
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (12 preceding siblings ...)
2024-08-07 9:47 ` [PATCH v2 13/15] net/ice: limit the number of queues to sched capabilities Bruce Richardson
@ 2024-08-07 9:47 ` Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 15/15] net/ice: add minimal capability reporting API Bruce Richardson
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:47 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Increase the flexibility of the Tx scheduler hierarchy support in the
driver. If the HW/firmware allows it, allow creating up to 2k child
nodes per scheduler node. Also expand the number of supported layers to
the max available, rather than always just having 3 layers. One
restriction on this change is that the topology needs to be configured
and enabled before port queue setup, in many cases, and before port
start in all cases.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.c | 9 -
drivers/net/ice/ice_ethdev.h | 15 +-
drivers/net/ice/ice_rxtx.c | 10 +
drivers/net/ice/ice_tm.c | 495 ++++++++++++++---------------------
4 files changed, 212 insertions(+), 317 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index ab3f88fd7d..5a5967ff71 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -3832,7 +3832,6 @@ ice_dev_start(struct rte_eth_dev *dev)
int mask, ret;
uint8_t timer = hw->func_caps.ts_func_info.tmr_index_owned;
uint32_t pin_idx = ad->devargs.pin_idx;
- struct rte_tm_error tm_err;
ice_declare_bitmap(pmask, ICE_PROMISC_MAX);
ice_zero_bitmap(pmask, ICE_PROMISC_MAX);
@@ -3864,14 +3863,6 @@ ice_dev_start(struct rte_eth_dev *dev)
}
}
- if (pf->tm_conf.committed) {
- ret = ice_do_hierarchy_commit(dev, pf->tm_conf.clear_on_fail, &tm_err);
- if (ret) {
- PMD_DRV_LOG(ERR, "fail to commit Tx scheduler");
- goto rx_err;
- }
- }
-
ice_set_rx_function(dev);
ice_set_tx_function(dev);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index f31addb122..cb1a7e8e0d 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -479,14 +479,6 @@ struct ice_tm_node {
struct ice_sched_node *sched_node;
};
-/* node type of Traffic Manager */
-enum ice_tm_node_type {
- ICE_TM_NODE_TYPE_PORT,
- ICE_TM_NODE_TYPE_QGROUP,
- ICE_TM_NODE_TYPE_QUEUE,
- ICE_TM_NODE_TYPE_MAX,
-};
-
/* Struct to store all the Traffic Manager configuration. */
struct ice_tm_conf {
struct ice_shaper_profile_list shaper_profile_list;
@@ -690,9 +682,6 @@ int ice_rem_rss_cfg_wrap(struct ice_pf *pf, uint16_t vsi_id,
struct ice_rss_hash_cfg *cfg);
void ice_tm_conf_init(struct rte_eth_dev *dev);
void ice_tm_conf_uninit(struct rte_eth_dev *dev);
-int ice_do_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error);
extern const struct rte_tm_ops ice_tm_ops;
static inline int
@@ -750,4 +739,8 @@ int rte_pmd_ice_dump_switch(uint16_t port, uint8_t **buff, uint32_t *size);
__rte_experimental
int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream);
+
+int
+ice_tm_setup_txq_node(struct ice_pf *pf, struct ice_hw *hw, uint16_t qid, uint32_t node_teid);
+
#endif /* _ICE_ETHDEV_H_ */
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index a150d28e73..7a421bb364 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -747,6 +747,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
int err;
struct ice_vsi *vsi;
struct ice_hw *hw;
+ struct ice_pf *pf;
struct ice_aqc_add_tx_qgrp *txq_elem;
struct ice_tlan_ctx tx_ctx;
int buf_len;
@@ -777,6 +778,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
vsi = txq->vsi;
hw = ICE_VSI_TO_HW(vsi);
+ pf = ICE_VSI_TO_PF(vsi);
memset(&tx_ctx, 0, sizeof(tx_ctx));
txq_elem->num_txqs = 1;
@@ -812,6 +814,14 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
/* store the schedule node id */
txq->q_teid = txq_elem->txqs[0].q_teid;
+ /* move the queue to correct position in hierarchy, if explicit hierarchy configured */
+ if (pf->tm_conf.committed)
+ if (ice_tm_setup_txq_node(pf, hw, tx_queue_id, txq->q_teid) != 0) {
+ PMD_DRV_LOG(ERR, "Failed to set up txq traffic management node");
+ rte_free(txq_elem);
+ return -EIO;
+ }
+
dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
rte_free(txq_elem);
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 459446a6b0..a86943a5b2 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -1,17 +1,17 @@
/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(c) 2022 Intel Corporation
*/
+#include <rte_ethdev.h>
#include <rte_tm_driver.h>
#include "ice_ethdev.h"
#include "ice_rxtx.h"
-#define MAX_CHILDREN_PER_SCHED_NODE 8
-#define MAX_CHILDREN_PER_TM_NODE 256
+#define MAX_CHILDREN_PER_TM_NODE 2048
static int ice_hierarchy_commit(struct rte_eth_dev *dev,
int clear_on_fail,
- __rte_unused struct rte_tm_error *error);
+ struct rte_tm_error *error);
static int ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
uint32_t parent_node_id, uint32_t priority,
uint32_t weight, uint32_t level_id,
@@ -86,9 +86,10 @@ ice_tm_conf_uninit(struct rte_eth_dev *dev)
}
static int
-ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
+ice_node_param_check(uint32_t node_id,
uint32_t priority, uint32_t weight,
const struct rte_tm_node_params *params,
+ bool is_leaf,
struct rte_tm_error *error)
{
/* checked all the unsupported parameter */
@@ -123,7 +124,7 @@ ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
}
/* for non-leaf node */
- if (node_id >= pf->dev_data->nb_tx_queues) {
+ if (!is_leaf) {
if (params->nonleaf.wfq_weight_mode) {
error->type =
RTE_TM_ERROR_TYPE_NODE_PARAMS_WFQ_WEIGHT_MODE;
@@ -147,6 +148,11 @@ ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
}
/* for leaf node */
+ if (node_id >= RTE_MAX_QUEUES_PER_PORT) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "Node ID out of range for a leaf node.";
+ return -EINVAL;
+ }
if (params->leaf.cman) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS_CMAN;
error->message = "Congestion management not supported";
@@ -193,11 +199,18 @@ find_node(struct ice_tm_node *root, uint32_t id)
return NULL;
}
+static inline uint8_t
+ice_get_leaf_level(struct ice_hw *hw)
+{
+ return hw->num_tx_sched_layers - 1 - hw->port_info->has_tc;
+}
+
static int
ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
int *is_leaf, struct rte_tm_error *error)
{
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_tm_node *tm_node;
if (!is_leaf || !error)
@@ -217,7 +230,7 @@ ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
return -EINVAL;
}
- if (tm_node->level == ICE_TM_NODE_TYPE_QUEUE)
+ if (tm_node->level == ice_get_leaf_level(hw))
*is_leaf = true;
else
*is_leaf = false;
@@ -389,16 +402,28 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
struct rte_tm_error *error)
{
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_tm_shaper_profile *shaper_profile = NULL;
struct ice_tm_node *tm_node;
- struct ice_tm_node *parent_node;
+ struct ice_tm_node *parent_node = NULL;
int ret;
if (!params || !error)
return -EINVAL;
- ret = ice_node_param_check(pf, node_id, priority, weight,
- params, error);
+ if (parent_node_id != RTE_TM_NODE_ID_NULL) {
+ parent_node = find_node(pf->tm_conf.root, parent_node_id);
+ if (!parent_node) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
+ error->message = "parent not exist";
+ return -EINVAL;
+ }
+ }
+ if (level_id == RTE_TM_NODE_LEVEL_ID_ANY && parent_node != NULL)
+ level_id = parent_node->level + 1;
+
+ ret = ice_node_param_check(node_id, priority, weight,
+ params, level_id == ice_get_leaf_level(hw), error);
if (ret)
return ret;
@@ -424,9 +449,9 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
/* root node if not have a parent */
if (parent_node_id == RTE_TM_NODE_ID_NULL) {
/* check level */
- if (level_id != ICE_TM_NODE_TYPE_PORT) {
+ if (level_id != 0) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
- error->message = "Wrong level";
+ error->message = "Wrong level, root node (NULL parent) must be at level 0";
return -EINVAL;
}
@@ -445,7 +470,7 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
if (!tm_node)
return -ENOMEM;
tm_node->id = node_id;
- tm_node->level = ICE_TM_NODE_TYPE_PORT;
+ tm_node->level = 0;
tm_node->parent = NULL;
tm_node->reference_count = 0;
tm_node->shaper_profile = shaper_profile;
@@ -458,48 +483,21 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
}
/* check the parent node */
- parent_node = find_node(pf->tm_conf.root, parent_node_id);
- if (!parent_node) {
- error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
- error->message = "parent not exist";
- return -EINVAL;
- }
- if (parent_node->level != ICE_TM_NODE_TYPE_PORT &&
- parent_node->level != ICE_TM_NODE_TYPE_QGROUP) {
+ /* for n-level hierarchy, level n-1 is leaf, so last level with children is n-2 */
+ if ((int)parent_node->level > hw->num_tx_sched_layers - 2) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
error->message = "parent is not valid";
return -EINVAL;
}
/* check level */
- if (level_id != RTE_TM_NODE_LEVEL_ID_ANY &&
- level_id != parent_node->level + 1) {
+ if (level_id != parent_node->level + 1) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
error->message = "Wrong level";
return -EINVAL;
}
/* check the node number */
- if (parent_node->level == ICE_TM_NODE_TYPE_PORT) {
- /* check the queue group number */
- if (parent_node->reference_count >= pf->dev_data->nb_tx_queues) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too many queue groups";
- return -EINVAL;
- }
- } else {
- /* check the queue number */
- if (parent_node->reference_count >=
- MAX_CHILDREN_PER_SCHED_NODE) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too many queues";
- return -EINVAL;
- }
- if (node_id >= pf->dev_data->nb_tx_queues) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too large queue id";
- return -EINVAL;
- }
- }
+ /* TODO, check max children allowed and max nodes at this level */
tm_node = rte_zmalloc(NULL,
sizeof(struct ice_tm_node) +
@@ -518,13 +516,12 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
(void *)((uint8_t *)tm_node + sizeof(struct ice_tm_node));
tm_node->parent->children[tm_node->parent->reference_count] = tm_node;
- if (tm_node->priority != 0 && level_id != ICE_TM_NODE_TYPE_QUEUE &&
- level_id != ICE_TM_NODE_TYPE_QGROUP)
+ if (tm_node->priority != 0)
+ /* TODO fixme, some levels may support this perhaps? */
PMD_DRV_LOG(WARNING, "priority != 0 not supported in level %d",
level_id);
- if (tm_node->weight != 1 &&
- level_id != ICE_TM_NODE_TYPE_QUEUE && level_id != ICE_TM_NODE_TYPE_QGROUP)
+ if (tm_node->weight != 1 && level_id == 0)
PMD_DRV_LOG(WARNING, "weight != 1 not supported in level %d",
level_id);
@@ -569,7 +566,7 @@ ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
}
/* root node */
- if (tm_node->level == ICE_TM_NODE_TYPE_PORT) {
+ if (tm_node->level == 0) {
rte_free(tm_node);
pf->tm_conf.root = NULL;
return 0;
@@ -589,53 +586,6 @@ ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
return 0;
}
-static int ice_move_recfg_lan_txq(struct rte_eth_dev *dev,
- struct ice_sched_node *queue_sched_node,
- struct ice_sched_node *dst_node,
- uint16_t queue_id)
-{
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_aqc_move_txqs_data *buf;
- struct ice_sched_node *queue_parent_node;
- uint8_t txqs_moved;
- int ret = ICE_SUCCESS;
- uint16_t buf_size = ice_struct_size(buf, txqs, 1);
-
- buf = (struct ice_aqc_move_txqs_data *)ice_malloc(hw, sizeof(*buf));
- if (buf == NULL)
- return -ENOMEM;
-
- queue_parent_node = queue_sched_node->parent;
- buf->src_teid = queue_parent_node->info.node_teid;
- buf->dest_teid = dst_node->info.node_teid;
- buf->txqs[0].q_teid = queue_sched_node->info.node_teid;
- buf->txqs[0].txq_id = queue_id;
-
- ret = ice_aq_move_recfg_lan_txq(hw, 1, true, false, false, false, 50,
- NULL, buf, buf_size, &txqs_moved, NULL);
- if (ret || txqs_moved == 0) {
- PMD_DRV_LOG(ERR, "move lan queue %u failed", queue_id);
- rte_free(buf);
- return ICE_ERR_PARAM;
- }
-
- if (queue_parent_node->num_children > 0) {
- queue_parent_node->num_children--;
- queue_parent_node->children[queue_parent_node->num_children] = NULL;
- } else {
- PMD_DRV_LOG(ERR, "invalid children number %d for queue %u",
- queue_parent_node->num_children, queue_id);
- rte_free(buf);
- return ICE_ERR_PARAM;
- }
- dst_node->children[dst_node->num_children++] = queue_sched_node;
- queue_sched_node->parent = dst_node;
- ice_sched_query_elem(hw, queue_sched_node->info.node_teid, &queue_sched_node->info);
-
- rte_free(buf);
- return ret;
-}
-
static int ice_set_node_rate(struct ice_hw *hw,
struct ice_tm_node *tm_node,
struct ice_sched_node *sched_node)
@@ -723,240 +673,191 @@ static int ice_cfg_hw_node(struct ice_hw *hw,
return 0;
}
-static struct ice_sched_node *ice_get_vsi_node(struct ice_hw *hw)
+int
+ice_tm_setup_txq_node(struct ice_pf *pf, struct ice_hw *hw, uint16_t qid, uint32_t teid)
{
- struct ice_sched_node *node = hw->port_info->root;
- uint32_t vsi_layer = hw->num_tx_sched_layers - ICE_VSI_LAYER_OFFSET;
- uint32_t i;
+ struct ice_sched_node *hw_node = ice_sched_find_node_by_teid(hw->port_info->root, teid);
+ struct ice_tm_node *sw_node = find_node(pf->tm_conf.root, qid);
- for (i = 0; i < vsi_layer; i++)
- node = node->children[0];
-
- return node;
-}
-
-static int ice_reset_noleaf_nodes(struct rte_eth_dev *dev)
-{
- struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_sched_node *vsi_node = ice_get_vsi_node(hw);
- struct ice_tm_node *root = pf->tm_conf.root;
- uint32_t i;
- int ret;
-
- /* reset vsi_node */
- ret = ice_set_node_rate(hw, NULL, vsi_node);
- if (ret) {
- PMD_DRV_LOG(ERR, "reset vsi node failed");
- return ret;
- }
-
- if (root == NULL)
+ /* not configured in hierarchy */
+ if (sw_node == NULL)
return 0;
- for (i = 0; i < root->reference_count; i++) {
- struct ice_tm_node *tm_node = root->children[i];
+ sw_node->sched_node = hw_node;
- if (tm_node->sched_node == NULL)
- continue;
+ /* if the queue node has been put in the wrong place in hierarchy */
+ if (hw_node->parent != sw_node->parent->sched_node) {
+ struct ice_aqc_move_txqs_data *buf;
+ uint8_t txqs_moved = 0;
+ uint16_t buf_size = ice_struct_size(buf, txqs, 1);
+
+ buf = ice_malloc(hw, buf_size);
+ if (buf == NULL)
+ return -ENOMEM;
- ret = ice_cfg_hw_node(hw, NULL, tm_node->sched_node);
- if (ret) {
- PMD_DRV_LOG(ERR, "reset queue group node %u failed", tm_node->id);
- return ret;
+ struct ice_sched_node *parent = hw_node->parent;
+ struct ice_sched_node *new_parent = sw_node->parent->sched_node;
+ buf->src_teid = parent->info.node_teid;
+ buf->dest_teid = new_parent->info.node_teid;
+ buf->txqs[0].q_teid = hw_node->info.node_teid;
+ buf->txqs[0].txq_id = qid;
+
+ int ret = ice_aq_move_recfg_lan_txq(hw, 1, true, false, false, false, 50,
+ NULL, buf, buf_size, &txqs_moved, NULL);
+ if (ret || txqs_moved == 0) {
+ PMD_DRV_LOG(ERR, "move lan queue %u failed", qid);
+ ice_free(hw, buf);
+ return ICE_ERR_PARAM;
}
- tm_node->sched_node = NULL;
+
+ /* now update the ice_sched_nodes to match physical layout */
+ new_parent->children[new_parent->num_children++] = hw_node;
+ hw_node->parent = new_parent;
+ ice_sched_query_elem(hw, hw_node->info.node_teid, &hw_node->info);
+ for (uint16_t i = 0; i < parent->num_children; i++)
+ if (parent->children[i] == hw_node) {
+ /* to remove, just overwrite the old node slot with the last ptr */
+ parent->children[i] = parent->children[--parent->num_children];
+ break;
+ }
}
- return 0;
+ return ice_cfg_hw_node(hw, sw_node, hw_node);
}
-static int ice_remove_leaf_nodes(struct rte_eth_dev *dev)
+/* from a given node, recursively deletes all the nodes that belong to that vsi.
+ * Any nodes which can't be deleted because they have children belonging to a different
+ * VSI, are now also adjusted to belong to that VSI also
+ */
+static int
+free_sched_node_recursive(struct ice_port_info *pi, const struct ice_sched_node *root,
+ struct ice_sched_node *node, uint8_t vsi_id)
{
- int ret = 0;
- int i;
+ uint16_t i = 0;
- for (i = 0; i < dev->data->nb_tx_queues; i++) {
- ret = ice_tx_queue_stop(dev, i);
- if (ret) {
- PMD_DRV_LOG(ERR, "stop queue %u failed", i);
- break;
+ while (i < node->num_children) {
+ if (node->children[i]->vsi_handle != vsi_id) {
+ i++;
+ continue;
}
+ free_sched_node_recursive(pi, root, node->children[i], vsi_id);
}
- return ret;
-}
-
-static int ice_add_leaf_nodes(struct rte_eth_dev *dev)
-{
- int ret = 0;
- int i;
-
- for (i = 0; i < dev->data->nb_tx_queues; i++) {
- ret = ice_tx_queue_start(dev, i);
- if (ret) {
- PMD_DRV_LOG(ERR, "start queue %u failed", i);
- break;
- }
+ if (node != root) {
+ if (node->num_children == 0)
+ ice_free_sched_node(pi, node);
+ else
+ node->vsi_handle = node->children[0]->vsi_handle;
}
- return ret;
+ return 0;
}
-int ice_do_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error)
+static int
+create_sched_node_recursive(struct ice_port_info *pi, struct ice_tm_node *sw_node,
+ struct ice_sched_node *hw_root, uint16_t *created)
{
- struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_tm_node *root;
- struct ice_sched_node *vsi_node = NULL;
- struct ice_sched_node *queue_node;
- struct ice_tx_queue *txq;
- int ret_val = 0;
- uint32_t i;
- uint32_t idx_vsi_child;
- uint32_t idx_qg;
- uint32_t nb_vsi_child;
- uint32_t nb_qg;
- uint32_t qid;
- uint32_t q_teid;
-
- /* remove leaf nodes */
- ret_val = ice_remove_leaf_nodes(dev);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "reset no-leaf nodes failed");
- goto fail_clear;
- }
-
- /* reset no-leaf nodes. */
- ret_val = ice_reset_noleaf_nodes(dev);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "reset leaf nodes failed");
- goto add_leaf;
- }
-
- /* config vsi node */
- vsi_node = ice_get_vsi_node(hw);
- root = pf->tm_conf.root;
-
- ret_val = ice_set_node_rate(hw, root, vsi_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure vsi node %u bandwidth failed",
- root->id);
- goto add_leaf;
- }
-
- /* config queue group nodes */
- nb_vsi_child = vsi_node->num_children;
- nb_qg = vsi_node->children[0]->num_children;
-
- idx_vsi_child = 0;
- idx_qg = 0;
-
- if (root == NULL)
- goto commit;
-
- for (i = 0; i < root->reference_count; i++) {
- struct ice_tm_node *tm_node = root->children[i];
- struct ice_tm_node *tm_child_node;
- struct ice_sched_node *qgroup_sched_node =
- vsi_node->children[idx_vsi_child]->children[idx_qg];
- uint32_t j;
-
- ret_val = ice_cfg_hw_node(hw, tm_node, qgroup_sched_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure queue group node %u failed",
- tm_node->id);
- goto reset_leaf;
- }
-
- for (j = 0; j < tm_node->reference_count; j++) {
- tm_child_node = tm_node->children[j];
- qid = tm_child_node->id;
- ret_val = ice_tx_queue_start(dev, qid);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "start queue %u failed", qid);
- goto reset_leaf;
- }
- txq = dev->data->tx_queues[qid];
- q_teid = txq->q_teid;
- queue_node = ice_sched_get_node(hw->port_info, q_teid);
- if (queue_node == NULL) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "get queue %u node failed", qid);
- goto reset_leaf;
- }
- if (queue_node->info.parent_teid != qgroup_sched_node->info.node_teid) {
- ret_val = ice_move_recfg_lan_txq(dev, queue_node,
- qgroup_sched_node, qid);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "move queue %u failed", qid);
- goto reset_leaf;
- }
- }
- ret_val = ice_cfg_hw_node(hw, tm_child_node, queue_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure queue group node %u failed",
- tm_node->id);
- goto reset_leaf;
- }
- }
-
- idx_qg++;
- if (idx_qg >= nb_qg) {
- idx_qg = 0;
- idx_vsi_child++;
+ struct ice_sched_node *parent = sw_node->sched_node;
+ uint32_t teid;
+ uint16_t added;
+
+ /* first create all child nodes */
+ for (uint16_t i = 0; i < sw_node->reference_count; i++) {
+ struct ice_tm_node *tm_node = sw_node->children[i];
+ int res = ice_sched_add_elems(pi, hw_root,
+ parent, parent->tx_sched_layer + 1,
+ 1 /* num nodes */, &added, &teid,
+ NULL /* no pre-alloc */);
+ if (res != 0) {
+ PMD_DRV_LOG(ERR, "Error with ice_sched_add_elems, adding child node to teid %u\n",
+ parent->info.node_teid);
+ return -1;
}
- if (idx_vsi_child >= nb_vsi_child) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "too many queues");
- goto reset_leaf;
+ struct ice_sched_node *hw_node = ice_sched_find_node_by_teid(parent, teid);
+ if (ice_cfg_hw_node(pi->hw, tm_node, hw_node) != 0) {
+ PMD_DRV_LOG(ERR, "Error configuring node %u at layer %u",
+ teid, parent->tx_sched_layer + 1);
+ return -1;
}
+ tm_node->sched_node = hw_node;
+ created[hw_node->tx_sched_layer]++;
}
-commit:
- pf->tm_conf.committed = true;
- pf->tm_conf.clear_on_fail = clear_on_fail;
+ /* if we have just created the child nodes in the q-group, i.e. last non-leaf layer,
+ * then just return, rather than trying to create leaf nodes.
+ * That is done later at queue start.
+ */
+ if (sw_node->level + 2 == ice_get_leaf_level(pi->hw))
+ return 0;
- return ret_val;
+ for (uint16_t i = 0; i < sw_node->reference_count; i++) {
+ if (sw_node->children[i]->reference_count == 0)
+ continue;
-reset_leaf:
- ice_remove_leaf_nodes(dev);
-add_leaf:
- ice_add_leaf_nodes(dev);
- ice_reset_noleaf_nodes(dev);
-fail_clear:
- /* clear all the traffic manager configuration */
- if (clear_on_fail) {
- ice_tm_conf_uninit(dev);
- ice_tm_conf_init(dev);
+ if (create_sched_node_recursive(pi, sw_node->children[i], hw_root, created) < 0)
+ return -1;
}
- return ret_val;
+ return 0;
}
-static int ice_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error)
+static int
+apply_topology_updates(struct rte_eth_dev *dev __rte_unused)
{
+ return 0;
+}
+
+static int
+commit_new_hierarchy(struct rte_eth_dev *dev)
+{
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_port_info *pi = hw->port_info;
+ struct ice_tm_node *sw_root = pf->tm_conf.root;
+ struct ice_sched_node *new_vsi_root = (pi->has_tc) ? pi->root->children[0] : pi->root;
+ uint16_t nodes_created_per_level[10] = {0}; /* counted per hw level, not per logical */
+ uint8_t q_lvl = ice_get_leaf_level(hw);
+ uint8_t qg_lvl = q_lvl - 1;
+
+ /* check if we have a previously applied topology */
+ if (sw_root->sched_node != NULL)
+ return apply_topology_updates(dev);
+
+ free_sched_node_recursive(pi, new_vsi_root, new_vsi_root, new_vsi_root->vsi_handle);
+
+ sw_root->sched_node = new_vsi_root;
+ if (create_sched_node_recursive(pi, sw_root, new_vsi_root, nodes_created_per_level) < 0)
+ return -1;
+ for (uint16_t i = 0; i < RTE_DIM(nodes_created_per_level); i++)
+ PMD_DRV_LOG(DEBUG, "Created %u nodes at level %u\n",
+ nodes_created_per_level[i], i);
+ hw->vsi_ctx[pf->main_vsi->idx]->sched.vsi_node[0] = new_vsi_root;
+
+ pf->main_vsi->nb_qps =
+ RTE_MIN(nodes_created_per_level[qg_lvl] * hw->max_children[qg_lvl],
+ hw->layer_info[q_lvl].max_device_nodes);
+
+ pf->tm_conf.committed = true; /* set flag to be checks on queue start */
+
+ return ice_alloc_lan_q_ctx(hw, 0, 0, pf->main_vsi->nb_qps);
+}
- /* if device not started, simply set committed flag and return. */
- if (!dev->data->dev_started) {
- pf->tm_conf.committed = true;
- pf->tm_conf.clear_on_fail = clear_on_fail;
- return 0;
+static int
+ice_hierarchy_commit(struct rte_eth_dev *dev,
+ int clear_on_fail,
+ struct rte_tm_error *error)
+{
+ RTE_SET_USED(error);
+ /* TODO - commit should only be done to topology before start! */
+ if (dev->data->dev_started)
+ return -1;
+
+ uint64_t start = rte_rdtsc();
+ int ret = commit_new_hierarchy(dev);
+ if (ret < 0 && clear_on_fail) {
+ ice_tm_conf_uninit(dev);
+ ice_tm_conf_init(dev);
}
-
- return ice_do_hierarchy_commit(dev, clear_on_fail, error);
+ uint64_t time = rte_rdtsc() - start;
+ PMD_DRV_LOG(DEBUG, "Time to apply hierarchy = %.1f\n", (float)time / rte_get_timer_hz());
+ return ret;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v2 15/15] net/ice: add minimal capability reporting API
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (13 preceding siblings ...)
2024-08-07 9:47 ` [PATCH v2 14/15] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
@ 2024-08-07 9:47 ` Bruce Richardson
14 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-07 9:47 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Incomplete but reports number of available layers
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.h | 1 +
drivers/net/ice/ice_tm.c | 17 +++++++++++++++++
2 files changed, 18 insertions(+)
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index cb1a7e8e0d..6bebc511e4 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -682,6 +682,7 @@ int ice_rem_rss_cfg_wrap(struct ice_pf *pf, uint16_t vsi_id,
struct ice_rss_hash_cfg *cfg);
void ice_tm_conf_init(struct rte_eth_dev *dev);
void ice_tm_conf_uninit(struct rte_eth_dev *dev);
+
extern const struct rte_tm_ops ice_tm_ops;
static inline int
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index a86943a5b2..d7def61756 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -33,8 +33,12 @@ static int ice_shaper_profile_add(struct rte_eth_dev *dev,
static int ice_shaper_profile_del(struct rte_eth_dev *dev,
uint32_t shaper_profile_id,
struct rte_tm_error *error);
+static int ice_tm_capabilities_get(struct rte_eth_dev *dev,
+ struct rte_tm_capabilities *cap,
+ struct rte_tm_error *error);
const struct rte_tm_ops ice_tm_ops = {
+ .capabilities_get = ice_tm_capabilities_get,
.shaper_profile_add = ice_shaper_profile_add,
.shaper_profile_delete = ice_shaper_profile_del,
.node_add = ice_tm_node_add,
@@ -861,3 +865,16 @@ ice_hierarchy_commit(struct rte_eth_dev *dev,
PMD_DRV_LOG(DEBUG, "Time to apply hierarchy = %.1f\n", (float)time / rte_get_timer_hz());
return ret;
}
+
+static int
+ice_tm_capabilities_get(struct rte_eth_dev *dev, struct rte_tm_capabilities *cap,
+ struct rte_tm_error *error)
+{
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+ *cap = (struct rte_tm_capabilities){
+ .n_levels_max = hw->num_tx_sched_layers - hw->port_info->has_tc,
+ };
+ if (error)
+ error->type = RTE_TM_ERROR_TYPE_NONE;
+ return 0;
+}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 00/16] Improve rte_tm support in ICE driver
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
` (15 preceding siblings ...)
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
@ 2024-08-12 15:27 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 01/16] net/ice: add traffic management node query function Bruce Richardson
` (15 more replies)
16 siblings, 16 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:27 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
This patchset expands the capabilities of the traffic management
support in the ICE driver. It allows the driver to support different
sizes of topologies, and support >256 queues and more than 3 hierarchy
layers.
---
Depends-on: series-32719 ("improve rte_rm APIs")
v3:
* remove/implement some code TODOs
* add patch 16 to set.
v2:
* Correct typo in commit log of one patch
* Add missing depends-on tag to the cover letter
Bruce Richardson (16):
net/ice: add traffic management node query function
net/ice: detect stopping a flow-director queue twice
net/ice: improve Tx scheduler graph output
net/ice: add option to choose DDP package file
net/ice: add option to download scheduler topology
net/ice/base: allow init without TC class sched nodes
net/ice/base: set VSI index on newly created nodes
net/ice/base: read VSI layer info from VSI
net/ice/base: remove 255 limit on sched child nodes
net/ice/base: optimize subtree searches
net/ice/base: make functions non-static
net/ice/base: remove flag checks before topology upload
net/ice: limit the number of queues to sched capabilities
net/ice: enhance Tx scheduler hierarchy support
net/ice: add minimal capability reporting API
net/ice: do early check on node level when adding
doc/guides/nics/ice.rst | 9 +
drivers/net/ice/base/ice_ddp.c | 51 +--
drivers/net/ice/base/ice_ddp.h | 4 +-
drivers/net/ice/base/ice_sched.c | 56 +--
drivers/net/ice/base/ice_sched.h | 8 +
drivers/net/ice/base/ice_type.h | 3 +-
drivers/net/ice/ice_diagnose.c | 196 ++++-------
drivers/net/ice/ice_ethdev.c | 92 +++--
drivers/net/ice/ice_ethdev.h | 18 +-
drivers/net/ice/ice_rxtx.c | 15 +
drivers/net/ice/ice_tm.c | 574 +++++++++++++++----------------
11 files changed, 497 insertions(+), 529 deletions(-)
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 01/16] net/ice: add traffic management node query function
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 02/16] net/ice: detect stopping a flow-director queue twice Bruce Richardson
` (14 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Implement the new node querying function for the "ice" net driver.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_tm.c | 48 ++++++++++++++++++++++++++++++++++++++++
1 file changed, 48 insertions(+)
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 8a29a9e744..459446a6b0 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -17,6 +17,11 @@ static int ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
uint32_t weight, uint32_t level_id,
const struct rte_tm_node_params *params,
struct rte_tm_error *error);
+static int ice_node_query(const struct rte_eth_dev *dev, uint32_t node_id,
+ uint32_t *parent_node_id, uint32_t *priority,
+ uint32_t *weight, uint32_t *level_id,
+ struct rte_tm_node_params *params,
+ struct rte_tm_error *error);
static int ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
struct rte_tm_error *error);
static int ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
@@ -35,6 +40,7 @@ const struct rte_tm_ops ice_tm_ops = {
.node_add = ice_tm_node_add,
.node_delete = ice_tm_node_delete,
.node_type_get = ice_node_type_get,
+ .node_query = ice_node_query,
.hierarchy_commit = ice_hierarchy_commit,
};
@@ -219,6 +225,48 @@ ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
return 0;
}
+static int
+ice_node_query(const struct rte_eth_dev *dev, uint32_t node_id,
+ uint32_t *parent_node_id, uint32_t *priority,
+ uint32_t *weight, uint32_t *level_id,
+ struct rte_tm_node_params *params,
+ struct rte_tm_error *error)
+{
+ struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_tm_node *tm_node;
+
+ if (node_id == RTE_TM_NODE_ID_NULL) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "invalid node id";
+ return -EINVAL;
+ }
+
+ /* check if the node id exists */
+ tm_node = find_node(pf->tm_conf.root, node_id);
+ if (!tm_node) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "no such node";
+ return -EEXIST;
+ }
+
+ if (parent_node_id != NULL)
+ *parent_node_id = tm_node->parent->id;
+
+ if (priority != NULL)
+ *priority = tm_node->priority;
+
+ if (weight != NULL)
+ *weight = tm_node->weight;
+
+ if (level_id != NULL)
+ *level_id = tm_node->level;
+
+ if (params != NULL)
+ *params = tm_node->params;
+
+ return 0;
+}
+
static inline struct ice_tm_shaper_profile *
ice_shaper_profile_search(struct rte_eth_dev *dev,
uint32_t shaper_profile_id)
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 02/16] net/ice: detect stopping a flow-director queue twice
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 01/16] net/ice: add traffic management node query function Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 03/16] net/ice: improve Tx scheduler graph output Bruce Richardson
` (13 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
If the flow-director queue is stopped at some point during the running
of an application, the shutdown procedure for the port issues an error
as it tries to stop the queue a second time, and fails to do so. We can
eliminate this error by setting the tail-register pointer to NULL on
stop, and checking for that condition in subsequent stop calls. Since
the register pointer is set on start, any restarting of the queue will
allow a stop call to progress as normal.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_rxtx.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index f270498ed1..a150d28e73 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -1139,6 +1139,10 @@ ice_fdir_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
tx_queue_id);
return -EINVAL;
}
+ if (txq->qtx_tail == NULL) {
+ PMD_DRV_LOG(INFO, "TX queue %u not started\n", tx_queue_id);
+ return 0;
+ }
vsi = txq->vsi;
q_ids[0] = txq->reg_idx;
@@ -1153,6 +1157,7 @@ ice_fdir_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
}
txq->tx_rel_mbufs(txq);
+ txq->qtx_tail = NULL;
return 0;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 03/16] net/ice: improve Tx scheduler graph output
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 01/16] net/ice: add traffic management node query function Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 02/16] net/ice: detect stopping a flow-director queue twice Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 04/16] net/ice: add option to choose DDP package file Bruce Richardson
` (12 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The function to dump the TX scheduler topology only adds to the chart
nodes connected to TX queues or for the flow director VSI. Change the
function to work recursively from the root node and thereby include all
scheduler nodes, whether in use or not, in the dump.
Also, improve the output of the Tx scheduler graphing function:
* Add VSI details to each node in graph
* When number of children is >16, skip middle nodes to reduce size of
the graph, otherwise dot output is unviewable for large hierarchies
* For VSIs other than zero, use dot's clustering method to put those
VSIs into subgraphs with borders
* For leaf nodes, display queue numbers for the any nodes assigned to
ethdev NIC Tx queues
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_diagnose.c | 196 ++++++++++++---------------------
1 file changed, 69 insertions(+), 127 deletions(-)
diff --git a/drivers/net/ice/ice_diagnose.c b/drivers/net/ice/ice_diagnose.c
index c357554707..623d84e37d 100644
--- a/drivers/net/ice/ice_diagnose.c
+++ b/drivers/net/ice/ice_diagnose.c
@@ -545,29 +545,15 @@ static void print_rl_profile(struct ice_aqc_rl_profile_elem *prof,
fprintf(stream, "\t\t\t\t\t</td>\n");
}
-static
-void print_elem_type(FILE *stream, u8 type)
+static const char *
+get_elem_type(u8 type)
{
- switch (type) {
- case 1:
- fprintf(stream, "root");
- break;
- case 2:
- fprintf(stream, "tc");
- break;
- case 3:
- fprintf(stream, "se_generic");
- break;
- case 4:
- fprintf(stream, "entry_point");
- break;
- case 5:
- fprintf(stream, "leaf");
- break;
- default:
- fprintf(stream, "%d", type);
- break;
- }
+ static const char * const ice_sched_node_types[] = {
+ "Undefined", "Root", "TC", "SE Generic", "SW Entry", "Leaf"
+ };
+ if (type < RTE_DIM(ice_sched_node_types))
+ return ice_sched_node_types[type];
+ return "*UNKNOWN*";
}
static
@@ -602,7 +588,9 @@ void print_priority_mode(FILE *stream, bool flag)
}
static
-void print_node(struct ice_aqc_txsched_elem_data *data,
+void print_node(struct ice_sched_node *node,
+ struct rte_eth_dev_data *ethdata,
+ struct ice_aqc_txsched_elem_data *data,
struct ice_aqc_rl_profile_elem *cir_prof,
struct ice_aqc_rl_profile_elem *eir_prof,
struct ice_aqc_rl_profile_elem *shared_prof,
@@ -613,17 +601,19 @@ void print_node(struct ice_aqc_txsched_elem_data *data,
fprintf(stream, "\t\t\t<table>\n");
- fprintf(stream, "\t\t\t\t<tr>\n");
- fprintf(stream, "\t\t\t\t\t<td> teid </td>\n");
- fprintf(stream, "\t\t\t\t\t<td> %d </td>\n", data->node_teid);
- fprintf(stream, "\t\t\t\t</tr>\n");
-
- fprintf(stream, "\t\t\t\t<tr>\n");
- fprintf(stream, "\t\t\t\t\t<td> type </td>\n");
- fprintf(stream, "\t\t\t\t\t<td>");
- print_elem_type(stream, data->data.elem_type);
- fprintf(stream, "</td>\n");
- fprintf(stream, "\t\t\t\t</tr>\n");
+ fprintf(stream, "\t\t\t\t<tr><td>teid</td><td>%d</td></tr>\n", data->node_teid);
+ fprintf(stream, "\t\t\t\t<tr><td>type</td><td>%s</td></tr>\n",
+ get_elem_type(data->data.elem_type));
+ fprintf(stream, "\t\t\t\t<tr><td>VSI</td><td>%u</td></tr>\n", node->vsi_handle);
+ if (data->data.elem_type == ICE_AQC_ELEM_TYPE_LEAF) {
+ for (uint16_t i = 0; i < ethdata->nb_tx_queues; i++) {
+ struct ice_tx_queue *q = ethdata->tx_queues[i];
+ if (q->q_teid == data->node_teid) {
+ fprintf(stream, "\t\t\t\t<tr><td>TXQ</td><td>%u</td></tr>\n", i);
+ break;
+ }
+ }
+ }
if (!detail)
goto brief;
@@ -705,8 +695,6 @@ void print_node(struct ice_aqc_txsched_elem_data *data,
fprintf(stream, "\t\tshape=plain\n");
fprintf(stream, "\t]\n");
- if (data->parent_teid != 0xFFFFFFFF)
- fprintf(stream, "\tNODE_%d -> NODE_%d\n", data->parent_teid, data->node_teid);
}
static
@@ -731,112 +719,92 @@ int query_rl_profile(struct ice_hw *hw,
return 0;
}
-static
-int query_node(struct ice_hw *hw, uint32_t child, uint32_t *parent,
- uint8_t level, bool detail, FILE *stream)
+static int
+query_node(struct ice_hw *hw, struct rte_eth_dev_data *ethdata,
+ struct ice_sched_node *node, bool detail, FILE *stream)
{
- struct ice_aqc_txsched_elem_data data;
+ struct ice_aqc_txsched_elem_data *data = &node->info;
struct ice_aqc_rl_profile_elem cir_prof;
struct ice_aqc_rl_profile_elem eir_prof;
struct ice_aqc_rl_profile_elem shared_prof;
struct ice_aqc_rl_profile_elem *cp = NULL;
struct ice_aqc_rl_profile_elem *ep = NULL;
struct ice_aqc_rl_profile_elem *sp = NULL;
- int status, ret;
-
- status = ice_sched_query_elem(hw, child, &data);
- if (status != ICE_SUCCESS) {
- if (level == hw->num_tx_sched_layers) {
- /* ignore the error when a queue has been stopped. */
- PMD_DRV_LOG(WARNING, "Failed to query queue node %d.", child);
- *parent = 0xffffffff;
- return 0;
- }
- PMD_DRV_LOG(ERR, "Failed to query scheduling node %d.", child);
- return -EINVAL;
- }
-
- *parent = data.parent_teid;
+ u8 level = node->tx_sched_layer;
+ int ret;
- if (data.data.cir_bw.bw_profile_idx != 0) {
- ret = query_rl_profile(hw, level, 0, data.data.cir_bw.bw_profile_idx, &cir_prof);
+ if (data->data.cir_bw.bw_profile_idx != 0) {
+ ret = query_rl_profile(hw, level, 0, data->data.cir_bw.bw_profile_idx, &cir_prof);
if (ret)
return ret;
cp = &cir_prof;
}
- if (data.data.eir_bw.bw_profile_idx != 0) {
- ret = query_rl_profile(hw, level, 1, data.data.eir_bw.bw_profile_idx, &eir_prof);
+ if (data->data.eir_bw.bw_profile_idx != 0) {
+ ret = query_rl_profile(hw, level, 1, data->data.eir_bw.bw_profile_idx, &eir_prof);
if (ret)
return ret;
ep = &eir_prof;
}
- if (data.data.srl_id != 0) {
- ret = query_rl_profile(hw, level, 2, data.data.srl_id, &shared_prof);
+ if (data->data.srl_id != 0) {
+ ret = query_rl_profile(hw, level, 2, data->data.srl_id, &shared_prof);
if (ret)
return ret;
sp = &shared_prof;
}
- print_node(&data, cp, ep, sp, detail, stream);
+ print_node(node, ethdata, data, cp, ep, sp, detail, stream);
return 0;
}
-static
-int query_nodes(struct ice_hw *hw,
- uint32_t *children, int child_num,
- uint32_t *parents, int *parent_num,
- uint8_t level, bool detail,
- FILE *stream)
+static int
+query_node_recursive(struct ice_hw *hw, struct rte_eth_dev_data *ethdata,
+ struct ice_sched_node *node, bool detail, FILE *stream)
{
- uint32_t parent;
- int i;
- int j;
-
- *parent_num = 0;
- for (i = 0; i < child_num; i++) {
- bool exist = false;
- int ret;
+ bool close = false;
+ if (node->parent != NULL && node->vsi_handle != node->parent->vsi_handle) {
+ fprintf(stream, "subgraph cluster_%u {\n", node->vsi_handle);
+ fprintf(stream, "\tlabel = \"VSI %u\";\n", node->vsi_handle);
+ close = true;
+ }
- ret = query_node(hw, children[i], &parent, level, detail, stream);
- if (ret)
- return -EINVAL;
+ int ret = query_node(hw, ethdata, node, detail, stream);
+ if (ret != 0)
+ return ret;
- for (j = 0; j < *parent_num; j++) {
- if (parents[j] == parent) {
- exist = true;
- break;
- }
+ for (uint16_t i = 0; i < node->num_children; i++) {
+ ret = query_node_recursive(hw, ethdata, node->children[i], detail, stream);
+ if (ret != 0)
+ return ret;
+ /* if we have a lot of nodes, skip a bunch in the middle */
+ if (node->num_children > 16 && i == 2) {
+ uint16_t inc = node->num_children - 5;
+ fprintf(stream, "\tn%d_children [label=\"... +%d child nodes ...\"];\n",
+ node->info.node_teid, inc);
+ fprintf(stream, "\tNODE_%d -> n%d_children;\n",
+ node->info.node_teid, node->info.node_teid);
+ i += inc;
}
-
- if (!exist && parent != 0xFFFFFFFF)
- parents[(*parent_num)++] = parent;
}
+ if (close)
+ fprintf(stream, "}\n");
+ if (node->info.parent_teid != 0xFFFFFFFF)
+ fprintf(stream, "\tNODE_%d -> NODE_%d\n",
+ node->info.parent_teid, node->info.node_teid);
return 0;
}
-int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
+int
+rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
{
struct rte_eth_dev *dev;
struct ice_hw *hw;
- struct ice_pf *pf;
- struct ice_q_ctx *q_ctx;
- uint16_t q_num;
- uint16_t i;
- struct ice_tx_queue *txq;
- uint32_t buf1[256];
- uint32_t buf2[256];
- uint32_t *children = buf1;
- uint32_t *parents = buf2;
- int child_num = 0;
- int parent_num = 0;
- uint8_t level;
RTE_ETH_VALID_PORTID_OR_ERR_RET(port, -ENODEV);
@@ -846,35 +814,9 @@ int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream)
dev = &rte_eth_devices[port];
hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- level = hw->num_tx_sched_layers;
-
- q_num = dev->data->nb_tx_queues;
-
- /* main vsi */
- for (i = 0; i < q_num; i++) {
- txq = dev->data->tx_queues[i];
- q_ctx = ice_get_lan_q_ctx(hw, txq->vsi->idx, 0, i);
- children[child_num++] = q_ctx->q_teid;
- }
-
- /* fdir vsi */
- q_ctx = ice_get_lan_q_ctx(hw, pf->fdir.fdir_vsi->idx, 0, 0);
- children[child_num++] = q_ctx->q_teid;
fprintf(stream, "digraph tx_sched {\n");
- while (child_num > 0) {
- int ret;
- ret = query_nodes(hw, children, child_num,
- parents, &parent_num,
- level, detail, stream);
- if (ret)
- return ret;
-
- children = parents;
- child_num = parent_num;
- level--;
- }
+ query_node_recursive(hw, dev->data, hw->port_info->root, detail, stream);
fprintf(stream, "}\n");
return 0;
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 04/16] net/ice: add option to choose DDP package file
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (2 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 03/16] net/ice: improve Tx scheduler graph output Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 05/16] net/ice: add option to download scheduler topology Bruce Richardson
` (11 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The "Dynamic Device Personalization" package is loaded at initialization
time by the driver, but the specific package file loaded depends upon
what package file is found first by searching through a hard-coded list
of firmware paths. To enable greater control over the package loading,
we can add a device option to choose a specific DDP package file to
load.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
doc/guides/nics/ice.rst | 9 +++++++++
drivers/net/ice/ice_ethdev.c | 34 ++++++++++++++++++++++++++++++++++
drivers/net/ice/ice_ethdev.h | 1 +
3 files changed, 44 insertions(+)
diff --git a/doc/guides/nics/ice.rst b/doc/guides/nics/ice.rst
index ae975d19ad..58ccfbd1a5 100644
--- a/doc/guides/nics/ice.rst
+++ b/doc/guides/nics/ice.rst
@@ -108,6 +108,15 @@ Runtime Configuration
-a 80:00.0,default-mac-disable=1
+- ``DDP Package File``
+
+ Rather than have the driver search for the DDP package to load,
+ or to override what package is used,
+ the ``ddp_pkg_file`` option can be used to provide the path to a specific package file.
+ For example::
+
+ -a 80:00.0,ddp_pkg_file=/path/to/ice-version.pkg
+
- ``Protocol extraction for per queue``
Configure the RX queues to do protocol extraction into mbuf for protocol
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 304f959b7e..3e7ceda9ce 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -36,6 +36,7 @@
#define ICE_ONE_PPS_OUT_ARG "pps_out"
#define ICE_RX_LOW_LATENCY_ARG "rx_low_latency"
#define ICE_MBUF_CHECK_ARG "mbuf_check"
+#define ICE_DDP_FILENAME "ddp_pkg_file"
#define ICE_CYCLECOUNTER_MASK 0xffffffffffffffffULL
@@ -52,6 +53,7 @@ static const char * const ice_valid_args[] = {
ICE_RX_LOW_LATENCY_ARG,
ICE_DEFAULT_MAC_DISABLE,
ICE_MBUF_CHECK_ARG,
+ ICE_DDP_FILENAME,
NULL
};
@@ -692,6 +694,18 @@ handle_field_name_arg(__rte_unused const char *key, const char *value,
return 0;
}
+static int
+handle_ddp_filename_arg(__rte_unused const char *key, const char *value, void *name_args)
+{
+ const char **filename = name_args;
+ if (strlen(value) >= ICE_MAX_PKG_FILENAME_SIZE) {
+ PMD_DRV_LOG(ERR, "The DDP package filename is too long : '%s'", value);
+ return -1;
+ }
+ *filename = strdup(value);
+ return 0;
+}
+
static void
ice_check_proto_xtr_support(struct ice_hw *hw)
{
@@ -1882,6 +1896,16 @@ int ice_load_pkg(struct ice_adapter *adapter, bool use_dsn, uint64_t dsn)
size_t bufsz;
int err;
+ if (adapter->devargs.ddp_filename != NULL) {
+ strlcpy(pkg_file, adapter->devargs.ddp_filename, sizeof(pkg_file));
+ if (rte_firmware_read(pkg_file, &buf, &bufsz) == 0) {
+ goto load_fw;
+ } else {
+ PMD_INIT_LOG(ERR, "Cannot load DDP file: %s\n", pkg_file);
+ return -1;
+ }
+ }
+
if (!use_dsn)
goto no_dsn;
@@ -2216,6 +2240,13 @@ static int ice_parse_devargs(struct rte_eth_dev *dev)
ret = rte_kvargs_process(kvlist, ICE_RX_LOW_LATENCY_ARG,
&parse_bool, &ad->devargs.rx_low_latency);
+ if (ret)
+ goto bail;
+
+ ret = rte_kvargs_process(kvlist, ICE_DDP_FILENAME,
+ &handle_ddp_filename_arg, &ad->devargs.ddp_filename);
+ if (ret)
+ goto bail;
bail:
rte_kvargs_free(kvlist);
@@ -2689,6 +2720,8 @@ ice_dev_close(struct rte_eth_dev *dev)
ice_free_hw_tbls(hw);
rte_free(hw->port_info);
hw->port_info = NULL;
+ free((void *)(uintptr_t)ad->devargs.ddp_filename);
+ ad->devargs.ddp_filename = NULL;
ice_shutdown_all_ctrlq(hw, true);
rte_free(pf->proto_xtr);
pf->proto_xtr = NULL;
@@ -6981,6 +7014,7 @@ RTE_PMD_REGISTER_PARAM_STRING(net_ice,
ICE_PROTO_XTR_ARG "=[queue:]<vlan|ipv4|ipv6|ipv6_flow|tcp|ip_offset>"
ICE_SAFE_MODE_SUPPORT_ARG "=<0|1>"
ICE_DEFAULT_MAC_DISABLE "=<0|1>"
+ ICE_DDP_FILENAME "=</path/to/file>"
ICE_RX_LOW_LATENCY_ARG "=<0|1>");
RTE_LOG_REGISTER_SUFFIX(ice_logtype_init, init, NOTICE);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index 3ea9f37dc8..c211b5b9cc 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -568,6 +568,7 @@ struct ice_devargs {
/* Name of the field. */
char xtr_field_name[RTE_MBUF_DYN_NAMESIZE];
uint64_t mbuf_check;
+ const char *ddp_filename;
};
/**
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 05/16] net/ice: add option to download scheduler topology
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (3 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 04/16] net/ice: add option to choose DDP package file Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 06/16] net/ice/base: allow init without TC class sched nodes Bruce Richardson
` (10 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The DDP package file being loaded at init time may contain an
alternative Tx Scheduler topology in it. Add driver option to load this
topology at init time.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_ddp.c | 18 +++++++++++++++---
drivers/net/ice/base/ice_ddp.h | 4 ++--
drivers/net/ice/ice_ethdev.c | 24 +++++++++++++++---------
drivers/net/ice/ice_ethdev.h | 1 +
4 files changed, 33 insertions(+), 14 deletions(-)
diff --git a/drivers/net/ice/base/ice_ddp.c b/drivers/net/ice/base/ice_ddp.c
index 24506dfaea..e6c42c5274 100644
--- a/drivers/net/ice/base/ice_ddp.c
+++ b/drivers/net/ice/base/ice_ddp.c
@@ -1326,7 +1326,7 @@ ice_fill_hw_ptype(struct ice_hw *hw)
* ice_copy_and_init_pkg() instead of directly calling ice_init_pkg() in this
* case.
*/
-enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
+enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len, bool load_sched)
{
bool already_loaded = false;
enum ice_ddp_state state;
@@ -1344,6 +1344,18 @@ enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
return state;
}
+ if (load_sched) {
+ enum ice_status res = ice_cfg_tx_topo(hw, buf, len);
+ if (res != ICE_SUCCESS) {
+ ice_debug(hw, ICE_DBG_INIT, "failed to apply sched topology (err: %d)\n",
+ res);
+ return ICE_DDP_PKG_ERR;
+ }
+ ice_debug(hw, ICE_DBG_INIT, "Topology download successful, reinitializing device\n");
+ ice_deinit_hw(hw);
+ ice_init_hw(hw);
+ }
+
/* initialize package info */
state = ice_init_pkg_info(hw, pkg);
if (state)
@@ -1416,7 +1428,7 @@ enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buf, u32 len)
* related routines.
*/
enum ice_ddp_state
-ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len)
+ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len, bool load_sched)
{
enum ice_ddp_state state;
u8 *buf_copy;
@@ -1426,7 +1438,7 @@ ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len)
buf_copy = (u8 *)ice_memdup(hw, buf, len, ICE_NONDMA_TO_NONDMA);
- state = ice_init_pkg(hw, buf_copy, len);
+ state = ice_init_pkg(hw, buf_copy, len, load_sched);
if (!ice_is_init_pkg_successful(state)) {
/* Free the copy, since we failed to initialize the package */
ice_free(hw, buf_copy);
diff --git a/drivers/net/ice/base/ice_ddp.h b/drivers/net/ice/base/ice_ddp.h
index 5761920207..2feba2e91d 100644
--- a/drivers/net/ice/base/ice_ddp.h
+++ b/drivers/net/ice/base/ice_ddp.h
@@ -451,9 +451,9 @@ ice_pkg_enum_entry(struct ice_seg *ice_seg, struct ice_pkg_enum *state,
void *
ice_pkg_enum_section(struct ice_seg *ice_seg, struct ice_pkg_enum *state,
u32 sect_type);
-enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buff, u32 len);
+enum ice_ddp_state ice_init_pkg(struct ice_hw *hw, u8 *buff, u32 len, bool load_sched);
enum ice_ddp_state
-ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len);
+ice_copy_and_init_pkg(struct ice_hw *hw, const u8 *buf, u32 len, bool load_sched);
bool ice_is_init_pkg_successful(enum ice_ddp_state state);
void ice_free_seg(struct ice_hw *hw);
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 3e7ceda9ce..0d2445a317 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -37,6 +37,7 @@
#define ICE_RX_LOW_LATENCY_ARG "rx_low_latency"
#define ICE_MBUF_CHECK_ARG "mbuf_check"
#define ICE_DDP_FILENAME "ddp_pkg_file"
+#define ICE_DDP_LOAD_SCHED "ddp_load_sched_topo"
#define ICE_CYCLECOUNTER_MASK 0xffffffffffffffffULL
@@ -54,6 +55,7 @@ static const char * const ice_valid_args[] = {
ICE_DEFAULT_MAC_DISABLE,
ICE_MBUF_CHECK_ARG,
ICE_DDP_FILENAME,
+ ICE_DDP_LOAD_SCHED,
NULL
};
@@ -1938,7 +1940,7 @@ int ice_load_pkg(struct ice_adapter *adapter, bool use_dsn, uint64_t dsn)
load_fw:
PMD_INIT_LOG(DEBUG, "DDP package name: %s", pkg_file);
- err = ice_copy_and_init_pkg(hw, buf, bufsz);
+ err = ice_copy_and_init_pkg(hw, buf, bufsz, adapter->devargs.ddp_load_sched);
if (!ice_is_init_pkg_successful(err)) {
PMD_INIT_LOG(ERR, "ice_copy_and_init_hw failed: %d\n", err);
free(buf);
@@ -1971,19 +1973,18 @@ static int
parse_bool(const char *key, const char *value, void *args)
{
int *i = (int *)args;
- char *end;
- int num;
- num = strtoul(value, &end, 10);
-
- if (num != 0 && num != 1) {
- PMD_DRV_LOG(WARNING, "invalid value:\"%s\" for key:\"%s\", "
- "value must be 0 or 1",
+ if (value == NULL || value[0] == '\0') {
+ PMD_DRV_LOG(WARNING, "key:\"%s\", requires a value, which must be 0 or 1", key);
+ return -1;
+ }
+ if (value[1] != '\0' || (value[0] != '0' && value[0] != '1')) {
+ PMD_DRV_LOG(WARNING, "invalid value:\"%s\" for key:\"%s\", value must be 0 or 1",
value, key);
return -1;
}
- *i = num;
+ *i = value[0] - '0';
return 0;
}
@@ -2248,6 +2249,10 @@ static int ice_parse_devargs(struct rte_eth_dev *dev)
if (ret)
goto bail;
+ ret = rte_kvargs_process(kvlist, ICE_DDP_LOAD_SCHED,
+ &parse_bool, &ad->devargs.ddp_load_sched);
+ if (ret)
+ goto bail;
bail:
rte_kvargs_free(kvlist);
return ret;
@@ -7014,6 +7019,7 @@ RTE_PMD_REGISTER_PARAM_STRING(net_ice,
ICE_PROTO_XTR_ARG "=[queue:]<vlan|ipv4|ipv6|ipv6_flow|tcp|ip_offset>"
ICE_SAFE_MODE_SUPPORT_ARG "=<0|1>"
ICE_DEFAULT_MAC_DISABLE "=<0|1>"
+ ICE_DDP_LOAD_SCHED "=<0|1>"
ICE_DDP_FILENAME "=</path/to/file>"
ICE_RX_LOW_LATENCY_ARG "=<0|1>");
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index c211b5b9cc..f31addb122 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -563,6 +563,7 @@ struct ice_devargs {
uint8_t proto_xtr[ICE_MAX_QUEUE_NUM];
uint8_t pin_idx;
uint8_t pps_out_ena;
+ int ddp_load_sched;
int xtr_field_offs;
uint8_t xtr_flag_offs[PROTO_XTR_MAX];
/* Name of the field. */
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 06/16] net/ice/base: allow init without TC class sched nodes
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (4 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 05/16] net/ice: add option to download scheduler topology Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 07/16] net/ice/base: set VSI index on newly created nodes Bruce Richardson
` (9 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
If DCB support is disabled via DDP image, there will not be any traffic
class (TC) nodes in the scheduler tree immediately above the root level.
To allow the driver to work with this scenario, we allow use of the root
node as a dummy TC0 node in case where there are no TC nodes in the
tree. For use of any other TC other than 0 (used by default in the
driver), existing behaviour of returning NULL pointer is maintained.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 6 ++++++
drivers/net/ice/base/ice_type.h | 1 +
2 files changed, 7 insertions(+)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index 373c32a518..f75e5ae599 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -292,6 +292,10 @@ struct ice_sched_node *ice_sched_get_tc_node(struct ice_port_info *pi, u8 tc)
if (!pi || !pi->root)
return NULL;
+ /* if no TC nodes, use root as TC node 0 */
+ if (pi->has_tc == 0)
+ return tc == 0 ? pi->root : NULL;
+
for (i = 0; i < pi->root->num_children; i++)
if (pi->root->children[i]->tc_num == tc)
return pi->root->children[i];
@@ -1306,6 +1310,8 @@ int ice_sched_init_port(struct ice_port_info *pi)
ICE_AQC_ELEM_TYPE_ENTRY_POINT)
hw->sw_entry_point_layer = j;
+ if (buf[0].generic[j].data.elem_type == ICE_AQC_ELEM_TYPE_TC)
+ pi->has_tc = 1;
status = ice_sched_add_node(pi, j, &buf[i].generic[j], NULL);
if (status)
goto err_init_port;
diff --git a/drivers/net/ice/base/ice_type.h b/drivers/net/ice/base/ice_type.h
index 598a80155b..a70e4a8afa 100644
--- a/drivers/net/ice/base/ice_type.h
+++ b/drivers/net/ice/base/ice_type.h
@@ -1260,6 +1260,7 @@ struct ice_port_info {
struct ice_qos_cfg qos_cfg;
u8 is_vf:1;
u8 is_custom_tx_enabled:1;
+ u8 has_tc:1;
};
struct ice_switch_info {
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 07/16] net/ice/base: set VSI index on newly created nodes
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (5 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 06/16] net/ice/base: allow init without TC class sched nodes Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 08/16] net/ice/base: read VSI layer info from VSI Bruce Richardson
` (8 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The ice_sched_node type has got a field for the vsi to which the node
belongs. This field was not getting set in "ice_sched_add_node", so add
a line configuring this field for each node from its parent node.
Similarly, when searching for a qgroup node, we can check for each node
that the VSI information is correct.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f75e5ae599..f6dc5ae173 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -200,6 +200,7 @@ ice_sched_add_node(struct ice_port_info *pi, u8 layer,
node->in_use = true;
node->parent = parent;
node->tx_sched_layer = layer;
+ node->vsi_handle = parent->vsi_handle;
parent->children[parent->num_children++] = node;
node->info = elem;
return 0;
@@ -1581,7 +1582,7 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
/* make sure the qgroup node is part of the VSI subtree */
if (ice_sched_find_node_in_subtree(pi->hw, vsi_node, qgrp_node))
if (qgrp_node->num_children < max_children &&
- qgrp_node->owner == owner)
+ qgrp_node->owner == owner && qgrp_node->vsi_handle == vsi_handle)
break;
qgrp_node = qgrp_node->sibling;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 08/16] net/ice/base: read VSI layer info from VSI
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (6 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 07/16] net/ice/base: set VSI index on newly created nodes Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 09/16] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
` (7 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Rather than computing from the number of HW layers the layer of the VSI,
we can instead just read that info from the VSI node itself. This allows
the layer to be changed at runtime.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f6dc5ae173..e398984bf2 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -1559,7 +1559,6 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
u16 max_children;
qgrp_layer = ice_sched_get_qgrp_layer(pi->hw);
- vsi_layer = ice_sched_get_vsi_layer(pi->hw);
max_children = pi->hw->max_children[qgrp_layer];
vsi_ctx = ice_get_vsi_ctx(pi->hw, vsi_handle);
@@ -1569,6 +1568,7 @@ ice_sched_get_free_qparent(struct ice_port_info *pi, u16 vsi_handle, u8 tc,
/* validate invalid VSI ID */
if (!vsi_node)
return NULL;
+ vsi_layer = vsi_node->tx_sched_layer;
/* If the queue group and vsi layer are same then queues
* are all attached directly to VSI
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 09/16] net/ice/base: remove 255 limit on sched child nodes
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (7 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 08/16] net/ice/base: read VSI layer info from VSI Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 10/16] net/ice/base: optimize subtree searches Bruce Richardson
` (6 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
The Tx scheduler in the ice driver can be configured to have large
numbers of child nodes at a given layer, but the driver code implicitly
limited the number of nodes to 255 by using a u8 datatype for the number
of children. Increase this to a 16-bit value throughout the code.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 25 ++++++++++++++-----------
drivers/net/ice/base/ice_type.h | 2 +-
2 files changed, 15 insertions(+), 12 deletions(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index e398984bf2..be13833e1e 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -289,7 +289,7 @@ ice_sched_get_first_node(struct ice_port_info *pi,
*/
struct ice_sched_node *ice_sched_get_tc_node(struct ice_port_info *pi, u8 tc)
{
- u8 i;
+ u16 i;
if (!pi || !pi->root)
return NULL;
@@ -316,7 +316,7 @@ void ice_free_sched_node(struct ice_port_info *pi, struct ice_sched_node *node)
{
struct ice_sched_node *parent;
struct ice_hw *hw = pi->hw;
- u8 i, j;
+ u16 i, j;
/* Free the children before freeing up the parent node
* The parent array is updated below and that shifts the nodes
@@ -1473,7 +1473,7 @@ bool
ice_sched_find_node_in_subtree(struct ice_hw *hw, struct ice_sched_node *base,
struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
for (i = 0; i < base->num_children; i++) {
struct ice_sched_node *child = base->children[i];
@@ -1510,7 +1510,7 @@ ice_sched_get_free_qgrp(struct ice_port_info *pi,
struct ice_sched_node *qgrp_node, u8 owner)
{
struct ice_sched_node *min_qgrp;
- u8 min_children;
+ u16 min_children;
if (!qgrp_node)
return qgrp_node;
@@ -2070,7 +2070,7 @@ static void ice_sched_rm_agg_vsi_info(struct ice_port_info *pi, u16 vsi_handle)
*/
static bool ice_sched_is_leaf_node_present(struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
for (i = 0; i < node->num_children; i++)
if (ice_sched_is_leaf_node_present(node->children[i]))
@@ -2105,7 +2105,7 @@ ice_sched_rm_vsi_cfg(struct ice_port_info *pi, u16 vsi_handle, u8 owner)
ice_for_each_traffic_class(i) {
struct ice_sched_node *vsi_node, *tc_node;
- u8 j = 0;
+ u16 j = 0;
tc_node = ice_sched_get_tc_node(pi, i);
if (!tc_node)
@@ -2173,7 +2173,7 @@ int ice_rm_vsi_lan_cfg(struct ice_port_info *pi, u16 vsi_handle)
*/
bool ice_sched_is_tree_balanced(struct ice_hw *hw, struct ice_sched_node *node)
{
- u8 i;
+ u16 i;
/* start from the leaf node */
for (i = 0; i < node->num_children; i++)
@@ -2247,7 +2247,8 @@ ice_sched_get_free_vsi_parent(struct ice_hw *hw, struct ice_sched_node *node,
u16 *num_nodes)
{
u8 l = node->tx_sched_layer;
- u8 vsil, i;
+ u8 vsil;
+ u16 i;
vsil = ice_sched_get_vsi_layer(hw);
@@ -2289,7 +2290,7 @@ ice_sched_update_parent(struct ice_sched_node *new_parent,
struct ice_sched_node *node)
{
struct ice_sched_node *old_parent;
- u8 i, j;
+ u16 i, j;
old_parent = node->parent;
@@ -2389,7 +2390,8 @@ ice_sched_move_vsi_to_agg(struct ice_port_info *pi, u16 vsi_handle, u32 agg_id,
u16 num_nodes[ICE_AQC_TOPO_MAX_LEVEL_NUM] = { 0 };
u32 first_node_teid, vsi_teid;
u16 num_nodes_added;
- u8 aggl, vsil, i;
+ u8 aggl, vsil;
+ u16 i;
int status;
tc_node = ice_sched_get_tc_node(pi, tc);
@@ -2505,7 +2507,8 @@ ice_move_all_vsi_to_dflt_agg(struct ice_port_info *pi,
static bool
ice_sched_is_agg_inuse(struct ice_port_info *pi, struct ice_sched_node *node)
{
- u8 vsil, i;
+ u8 vsil;
+ u16 i;
vsil = ice_sched_get_vsi_layer(pi->hw);
if (node->tx_sched_layer < vsil - 1) {
diff --git a/drivers/net/ice/base/ice_type.h b/drivers/net/ice/base/ice_type.h
index a70e4a8afa..35f832eb9f 100644
--- a/drivers/net/ice/base/ice_type.h
+++ b/drivers/net/ice/base/ice_type.h
@@ -1030,9 +1030,9 @@ struct ice_sched_node {
struct ice_aqc_txsched_elem_data info;
u32 agg_id; /* aggregator group ID */
u16 vsi_handle;
+ u16 num_children;
u8 in_use; /* suspended or in use */
u8 tx_sched_layer; /* Logical Layer (1-9) */
- u8 num_children;
u8 tc_num;
u8 owner;
#define ICE_SCHED_NODE_OWNER_LAN 0
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 10/16] net/ice/base: optimize subtree searches
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (8 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 09/16] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 11/16] net/ice/base: make functions non-static Bruce Richardson
` (5 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
In a number of places throughout the driver code, we want to confirm
that a scheduler node is indeed a child of another node. Currently, this
is confirmed by searching down the tree from the base until the desired
node is hit, a search which may hit many irrelevant tree nodes when
recursing down wrong branches. By switching the direction of search, to
check upwards from the node to the parent, we can avoid any incorrect
paths, and so speed up processing.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 18 +++++-------------
1 file changed, 5 insertions(+), 13 deletions(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index be13833e1e..f7d5f8f415 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -1475,20 +1475,12 @@ ice_sched_find_node_in_subtree(struct ice_hw *hw, struct ice_sched_node *base,
{
u16 i;
- for (i = 0; i < base->num_children; i++) {
- struct ice_sched_node *child = base->children[i];
-
- if (node == child)
- return true;
-
- if (child->tx_sched_layer > node->tx_sched_layer)
- return false;
-
- /* this recursion is intentional, and wouldn't
- * go more than 8 calls
- */
- if (ice_sched_find_node_in_subtree(hw, child, node))
+ if (base == node)
+ return true;
+ while (node->tx_sched_layer != 0 && node->parent != NULL) {
+ if (node->parent == base)
return true;
+ node = node->parent;
}
return false;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 11/16] net/ice/base: make functions non-static
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (9 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 10/16] net/ice/base: optimize subtree searches Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 12/16] net/ice/base: remove flag checks before topology upload Bruce Richardson
` (4 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
We will need to allocate more lanq contexts after a scheduler rework, so
make that function non-static so accessible outside the file. For similar
reasons, make the function to add a Tx scheduler node non-static
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_sched.c | 2 +-
drivers/net/ice/base/ice_sched.h | 8 ++++++++
2 files changed, 9 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ice/base/ice_sched.c b/drivers/net/ice/base/ice_sched.c
index f7d5f8f415..d88b836c38 100644
--- a/drivers/net/ice/base/ice_sched.c
+++ b/drivers/net/ice/base/ice_sched.c
@@ -570,7 +570,7 @@ ice_sched_suspend_resume_elems(struct ice_hw *hw, u8 num_nodes, u32 *node_teids,
* @tc: TC number
* @new_numqs: number of queues
*/
-static int
+int
ice_alloc_lan_q_ctx(struct ice_hw *hw, u16 vsi_handle, u8 tc, u16 new_numqs)
{
struct ice_vsi_ctx *vsi_ctx;
diff --git a/drivers/net/ice/base/ice_sched.h b/drivers/net/ice/base/ice_sched.h
index 9f78516dfb..c7eb794963 100644
--- a/drivers/net/ice/base/ice_sched.h
+++ b/drivers/net/ice/base/ice_sched.h
@@ -270,4 +270,12 @@ int ice_sched_replay_q_bw(struct ice_port_info *pi, struct ice_q_ctx *q_ctx);
int
ice_sched_cfg_node_bw_alloc(struct ice_hw *hw, struct ice_sched_node *node,
enum ice_rl_type rl_type, u16 bw_alloc);
+
+int
+ice_sched_add_elems(struct ice_port_info *pi, struct ice_sched_node *tc_node,
+ struct ice_sched_node *parent, u8 layer, u16 num_nodes,
+ u16 *num_nodes_added, u32 *first_node_teid,
+ struct ice_sched_node **prealloc_nodes);
+int
+ice_alloc_lan_q_ctx(struct ice_hw *hw, u16 vsi_handle, u8 tc, u16 new_numqs);
#endif /* _ICE_SCHED_H_ */
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 12/16] net/ice/base: remove flag checks before topology upload
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (10 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 11/16] net/ice/base: make functions non-static Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 13/16] net/ice: limit the number of queues to sched capabilities Bruce Richardson
` (3 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
DPDK should support more than just 9-level or 5-level topologies, so
remove the checks for those particular settings.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/base/ice_ddp.c | 33 ---------------------------------
1 file changed, 33 deletions(-)
diff --git a/drivers/net/ice/base/ice_ddp.c b/drivers/net/ice/base/ice_ddp.c
index e6c42c5274..744f015fe5 100644
--- a/drivers/net/ice/base/ice_ddp.c
+++ b/drivers/net/ice/base/ice_ddp.c
@@ -2373,38 +2373,6 @@ int ice_cfg_tx_topo(struct ice_hw *hw, u8 *buf, u32 len)
return status;
}
- /* Is default topology already applied ? */
- if (!(flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 9) {
- ice_debug(hw, ICE_DBG_INIT, "Loaded default topology\n");
- /* Already default topology is loaded */
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Is new topology already applied ? */
- if ((flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 5) {
- ice_debug(hw, ICE_DBG_INIT, "Loaded new topology\n");
- /* Already new topology is loaded */
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Is set topology issued already ? */
- if (flags & ICE_AQC_TX_TOPO_FLAGS_ISSUED) {
- ice_debug(hw, ICE_DBG_INIT, "Update tx topology was done by another PF\n");
- /* add a small delay before exiting */
- for (i = 0; i < 20; i++)
- ice_msec_delay(100, true);
- return ICE_ERR_ALREADY_EXISTS;
- }
-
- /* Change the topology from new to default (5 to 9) */
- if (!(flags & ICE_AQC_TX_TOPO_FLAGS_LOAD_NEW) &&
- hw->num_tx_sched_layers == 5) {
- ice_debug(hw, ICE_DBG_INIT, "Change topology from 5 to 9 layers\n");
- goto update_topo;
- }
-
pkg_hdr = (struct ice_pkg_hdr *)buf;
state = ice_verify_pkg(pkg_hdr, len);
if (state) {
@@ -2451,7 +2419,6 @@ int ice_cfg_tx_topo(struct ice_hw *hw, u8 *buf, u32 len)
/* Get the new topology buffer */
new_topo = ((u8 *)section) + offset;
-update_topo:
/* acquire global lock to make sure that set topology issued
* by one PF
*/
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 13/16] net/ice: limit the number of queues to sched capabilities
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (11 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 12/16] net/ice/base: remove flag checks before topology upload Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 14/16] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
` (2 subsequent siblings)
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Rather than assuming that each VSI can hold up to 256 queue pairs,
or the reported device limit, query the available nodes in the scheduler
tree to check that we are not overflowing the limit for number of
child scheduling nodes at each level. Do this by multiplying
max_children for each level beyond the VSI and using that as an
additional cap on the number of queues.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.c | 25 ++++++++++++++++++++-----
1 file changed, 20 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 0d2445a317..ab3f88fd7d 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -913,7 +913,7 @@ ice_vsi_config_default_rss(struct ice_aqc_vsi_props *info)
}
static int
-ice_vsi_config_tc_queue_mapping(struct ice_vsi *vsi,
+ice_vsi_config_tc_queue_mapping(struct ice_hw *hw, struct ice_vsi *vsi,
struct ice_aqc_vsi_props *info,
uint8_t enabled_tcmap)
{
@@ -929,13 +929,28 @@ ice_vsi_config_tc_queue_mapping(struct ice_vsi *vsi,
}
/* vector 0 is reserved and 1 vector for ctrl vsi */
- if (vsi->adapter->hw.func_caps.common_cap.num_msix_vectors < 2)
+ if (vsi->adapter->hw.func_caps.common_cap.num_msix_vectors < 2) {
vsi->nb_qps = 0;
- else
+ } else {
vsi->nb_qps = RTE_MIN
((uint16_t)vsi->adapter->hw.func_caps.common_cap.num_msix_vectors - 2,
RTE_MIN(vsi->nb_qps, ICE_MAX_Q_PER_TC));
+ /* cap max QPs to what the HW reports as num-children for each layer.
+ * Multiply num_children for each layer from the entry_point layer to
+ * the qgroup, or second-last layer.
+ * Avoid any potential overflow by using uint32_t type and breaking loop
+ * once we have a number greater than the already configured max.
+ */
+ uint32_t max_sched_vsi_nodes = 1;
+ for (uint8_t i = hw->sw_entry_point_layer; i < hw->num_tx_sched_layers - 1; i++) {
+ max_sched_vsi_nodes *= hw->max_children[i];
+ if (max_sched_vsi_nodes >= vsi->nb_qps)
+ break;
+ }
+ vsi->nb_qps = RTE_MIN(vsi->nb_qps, max_sched_vsi_nodes);
+ }
+
/* nb_qps(hex) -> fls */
/* 0000 -> 0 */
/* 0001 -> 0 */
@@ -1707,7 +1722,7 @@ ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type)
rte_cpu_to_le_16(hw->func_caps.fd_fltr_best_effort);
/* Enable VLAN/UP trip */
- ret = ice_vsi_config_tc_queue_mapping(vsi,
+ ret = ice_vsi_config_tc_queue_mapping(hw, vsi,
&vsi_ctx.info,
ICE_DEFAULT_TCMAP);
if (ret) {
@@ -1731,7 +1746,7 @@ ice_setup_vsi(struct ice_pf *pf, enum ice_vsi_type type)
vsi_ctx.info.fd_options = rte_cpu_to_le_16(cfg);
vsi_ctx.info.sw_id = hw->port_info->sw_id;
vsi_ctx.info.sw_flags2 = ICE_AQ_VSI_SW_FLAG_LAN_ENA;
- ret = ice_vsi_config_tc_queue_mapping(vsi,
+ ret = ice_vsi_config_tc_queue_mapping(hw, vsi,
&vsi_ctx.info,
ICE_DEFAULT_TCMAP);
if (ret) {
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 14/16] net/ice: enhance Tx scheduler hierarchy support
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (12 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 13/16] net/ice: limit the number of queues to sched capabilities Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 15/16] net/ice: add minimal capability reporting API Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 16/16] net/ice: do early check on node level when adding Bruce Richardson
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Increase the flexibility of the Tx scheduler hierarchy support in the
driver. If the HW/firmware allows it, allow creating up to 2k child
nodes per scheduler node. Also expand the number of supported layers to
the max available, rather than always just having 3 layers. One
restriction on this change is that the topology needs to be configured
and enabled before port queue setup, in many cases, and before port
start in all cases.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.c | 9 -
drivers/net/ice/ice_ethdev.h | 15 +-
drivers/net/ice/ice_rxtx.c | 10 +
drivers/net/ice/ice_tm.c | 500 ++++++++++++++---------------------
4 files changed, 216 insertions(+), 318 deletions(-)
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index ab3f88fd7d..5a5967ff71 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -3832,7 +3832,6 @@ ice_dev_start(struct rte_eth_dev *dev)
int mask, ret;
uint8_t timer = hw->func_caps.ts_func_info.tmr_index_owned;
uint32_t pin_idx = ad->devargs.pin_idx;
- struct rte_tm_error tm_err;
ice_declare_bitmap(pmask, ICE_PROMISC_MAX);
ice_zero_bitmap(pmask, ICE_PROMISC_MAX);
@@ -3864,14 +3863,6 @@ ice_dev_start(struct rte_eth_dev *dev)
}
}
- if (pf->tm_conf.committed) {
- ret = ice_do_hierarchy_commit(dev, pf->tm_conf.clear_on_fail, &tm_err);
- if (ret) {
- PMD_DRV_LOG(ERR, "fail to commit Tx scheduler");
- goto rx_err;
- }
- }
-
ice_set_rx_function(dev);
ice_set_tx_function(dev);
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index f31addb122..cb1a7e8e0d 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -479,14 +479,6 @@ struct ice_tm_node {
struct ice_sched_node *sched_node;
};
-/* node type of Traffic Manager */
-enum ice_tm_node_type {
- ICE_TM_NODE_TYPE_PORT,
- ICE_TM_NODE_TYPE_QGROUP,
- ICE_TM_NODE_TYPE_QUEUE,
- ICE_TM_NODE_TYPE_MAX,
-};
-
/* Struct to store all the Traffic Manager configuration. */
struct ice_tm_conf {
struct ice_shaper_profile_list shaper_profile_list;
@@ -690,9 +682,6 @@ int ice_rem_rss_cfg_wrap(struct ice_pf *pf, uint16_t vsi_id,
struct ice_rss_hash_cfg *cfg);
void ice_tm_conf_init(struct rte_eth_dev *dev);
void ice_tm_conf_uninit(struct rte_eth_dev *dev);
-int ice_do_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error);
extern const struct rte_tm_ops ice_tm_ops;
static inline int
@@ -750,4 +739,8 @@ int rte_pmd_ice_dump_switch(uint16_t port, uint8_t **buff, uint32_t *size);
__rte_experimental
int rte_pmd_ice_dump_txsched(uint16_t port, bool detail, FILE *stream);
+
+int
+ice_tm_setup_txq_node(struct ice_pf *pf, struct ice_hw *hw, uint16_t qid, uint32_t node_teid);
+
#endif /* _ICE_ETHDEV_H_ */
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index a150d28e73..7a421bb364 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -747,6 +747,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
int err;
struct ice_vsi *vsi;
struct ice_hw *hw;
+ struct ice_pf *pf;
struct ice_aqc_add_tx_qgrp *txq_elem;
struct ice_tlan_ctx tx_ctx;
int buf_len;
@@ -777,6 +778,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
vsi = txq->vsi;
hw = ICE_VSI_TO_HW(vsi);
+ pf = ICE_VSI_TO_PF(vsi);
memset(&tx_ctx, 0, sizeof(tx_ctx));
txq_elem->num_txqs = 1;
@@ -812,6 +814,14 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
/* store the schedule node id */
txq->q_teid = txq_elem->txqs[0].q_teid;
+ /* move the queue to correct position in hierarchy, if explicit hierarchy configured */
+ if (pf->tm_conf.committed)
+ if (ice_tm_setup_txq_node(pf, hw, tx_queue_id, txq->q_teid) != 0) {
+ PMD_DRV_LOG(ERR, "Failed to set up txq traffic management node");
+ rte_free(txq_elem);
+ return -EIO;
+ }
+
dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
rte_free(txq_elem);
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 459446a6b0..80039c8aff 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -1,17 +1,17 @@
/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(c) 2022 Intel Corporation
*/
+#include <rte_ethdev.h>
#include <rte_tm_driver.h>
#include "ice_ethdev.h"
#include "ice_rxtx.h"
-#define MAX_CHILDREN_PER_SCHED_NODE 8
-#define MAX_CHILDREN_PER_TM_NODE 256
+#define MAX_CHILDREN_PER_TM_NODE 2048
static int ice_hierarchy_commit(struct rte_eth_dev *dev,
int clear_on_fail,
- __rte_unused struct rte_tm_error *error);
+ struct rte_tm_error *error);
static int ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
uint32_t parent_node_id, uint32_t priority,
uint32_t weight, uint32_t level_id,
@@ -86,9 +86,10 @@ ice_tm_conf_uninit(struct rte_eth_dev *dev)
}
static int
-ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
+ice_node_param_check(uint32_t node_id,
uint32_t priority, uint32_t weight,
const struct rte_tm_node_params *params,
+ bool is_leaf,
struct rte_tm_error *error)
{
/* checked all the unsupported parameter */
@@ -123,7 +124,7 @@ ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
}
/* for non-leaf node */
- if (node_id >= pf->dev_data->nb_tx_queues) {
+ if (!is_leaf) {
if (params->nonleaf.wfq_weight_mode) {
error->type =
RTE_TM_ERROR_TYPE_NODE_PARAMS_WFQ_WEIGHT_MODE;
@@ -147,6 +148,11 @@ ice_node_param_check(struct ice_pf *pf, uint32_t node_id,
}
/* for leaf node */
+ if (node_id >= RTE_MAX_QUEUES_PER_PORT) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_ID;
+ error->message = "Node ID out of range for a leaf node.";
+ return -EINVAL;
+ }
if (params->leaf.cman) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS_CMAN;
error->message = "Congestion management not supported";
@@ -193,11 +199,18 @@ find_node(struct ice_tm_node *root, uint32_t id)
return NULL;
}
+static inline uint8_t
+ice_get_leaf_level(struct ice_hw *hw)
+{
+ return hw->num_tx_sched_layers - 1 - hw->port_info->has_tc;
+}
+
static int
ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
int *is_leaf, struct rte_tm_error *error)
{
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_tm_node *tm_node;
if (!is_leaf || !error)
@@ -217,7 +230,7 @@ ice_node_type_get(struct rte_eth_dev *dev, uint32_t node_id,
return -EINVAL;
}
- if (tm_node->level == ICE_TM_NODE_TYPE_QUEUE)
+ if (tm_node->level == ice_get_leaf_level(hw))
*is_leaf = true;
else
*is_leaf = false;
@@ -389,16 +402,28 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
struct rte_tm_error *error)
{
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_tm_shaper_profile *shaper_profile = NULL;
struct ice_tm_node *tm_node;
- struct ice_tm_node *parent_node;
+ struct ice_tm_node *parent_node = NULL;
int ret;
if (!params || !error)
return -EINVAL;
- ret = ice_node_param_check(pf, node_id, priority, weight,
- params, error);
+ if (parent_node_id != RTE_TM_NODE_ID_NULL) {
+ parent_node = find_node(pf->tm_conf.root, parent_node_id);
+ if (!parent_node) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
+ error->message = "parent not exist";
+ return -EINVAL;
+ }
+ }
+ if (level_id == RTE_TM_NODE_LEVEL_ID_ANY && parent_node != NULL)
+ level_id = parent_node->level + 1;
+
+ ret = ice_node_param_check(node_id, priority, weight,
+ params, level_id == ice_get_leaf_level(hw), error);
if (ret)
return ret;
@@ -424,9 +449,9 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
/* root node if not have a parent */
if (parent_node_id == RTE_TM_NODE_ID_NULL) {
/* check level */
- if (level_id != ICE_TM_NODE_TYPE_PORT) {
+ if (level_id != 0) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
- error->message = "Wrong level";
+ error->message = "Wrong level, root node (NULL parent) must be at level 0";
return -EINVAL;
}
@@ -445,7 +470,7 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
if (!tm_node)
return -ENOMEM;
tm_node->id = node_id;
- tm_node->level = ICE_TM_NODE_TYPE_PORT;
+ tm_node->level = 0;
tm_node->parent = NULL;
tm_node->reference_count = 0;
tm_node->shaper_profile = shaper_profile;
@@ -458,52 +483,29 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
}
/* check the parent node */
- parent_node = find_node(pf->tm_conf.root, parent_node_id);
- if (!parent_node) {
- error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
- error->message = "parent not exist";
- return -EINVAL;
- }
- if (parent_node->level != ICE_TM_NODE_TYPE_PORT &&
- parent_node->level != ICE_TM_NODE_TYPE_QGROUP) {
+ /* for n-level hierarchy, level n-1 is leaf, so last level with children is n-2 */
+ if ((int)parent_node->level > hw->num_tx_sched_layers - 2) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARENT_NODE_ID;
error->message = "parent is not valid";
return -EINVAL;
}
/* check level */
- if (level_id != RTE_TM_NODE_LEVEL_ID_ANY &&
- level_id != parent_node->level + 1) {
+ if (level_id != parent_node->level + 1) {
error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
error->message = "Wrong level";
return -EINVAL;
}
- /* check the node number */
- if (parent_node->level == ICE_TM_NODE_TYPE_PORT) {
- /* check the queue group number */
- if (parent_node->reference_count >= pf->dev_data->nb_tx_queues) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too many queue groups";
- return -EINVAL;
- }
- } else {
- /* check the queue number */
- if (parent_node->reference_count >=
- MAX_CHILDREN_PER_SCHED_NODE) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too many queues";
- return -EINVAL;
- }
- if (node_id >= pf->dev_data->nb_tx_queues) {
- error->type = RTE_TM_ERROR_TYPE_NODE_ID;
- error->message = "too large queue id";
- return -EINVAL;
- }
+ /* check the max children allowed at this level */
+ if (parent_node->reference_count >= hw->max_children[parent_node->level]) {
+ error->type = RTE_TM_ERROR_TYPE_CAPABILITIES;
+ error->message = "insufficient number of child nodes supported";
+ return -EINVAL;
}
tm_node = rte_zmalloc(NULL,
sizeof(struct ice_tm_node) +
- sizeof(struct ice_tm_node *) * MAX_CHILDREN_PER_TM_NODE,
+ sizeof(struct ice_tm_node *) * hw->max_children[level_id],
0);
if (!tm_node)
return -ENOMEM;
@@ -518,13 +520,11 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
(void *)((uint8_t *)tm_node + sizeof(struct ice_tm_node));
tm_node->parent->children[tm_node->parent->reference_count] = tm_node;
- if (tm_node->priority != 0 && level_id != ICE_TM_NODE_TYPE_QUEUE &&
- level_id != ICE_TM_NODE_TYPE_QGROUP)
+ if (tm_node->priority != 0)
PMD_DRV_LOG(WARNING, "priority != 0 not supported in level %d",
level_id);
- if (tm_node->weight != 1 &&
- level_id != ICE_TM_NODE_TYPE_QUEUE && level_id != ICE_TM_NODE_TYPE_QGROUP)
+ if (tm_node->weight != 1 && level_id == 0)
PMD_DRV_LOG(WARNING, "weight != 1 not supported in level %d",
level_id);
@@ -569,7 +569,7 @@ ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
}
/* root node */
- if (tm_node->level == ICE_TM_NODE_TYPE_PORT) {
+ if (tm_node->level == 0) {
rte_free(tm_node);
pf->tm_conf.root = NULL;
return 0;
@@ -589,53 +589,6 @@ ice_tm_node_delete(struct rte_eth_dev *dev, uint32_t node_id,
return 0;
}
-static int ice_move_recfg_lan_txq(struct rte_eth_dev *dev,
- struct ice_sched_node *queue_sched_node,
- struct ice_sched_node *dst_node,
- uint16_t queue_id)
-{
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_aqc_move_txqs_data *buf;
- struct ice_sched_node *queue_parent_node;
- uint8_t txqs_moved;
- int ret = ICE_SUCCESS;
- uint16_t buf_size = ice_struct_size(buf, txqs, 1);
-
- buf = (struct ice_aqc_move_txqs_data *)ice_malloc(hw, sizeof(*buf));
- if (buf == NULL)
- return -ENOMEM;
-
- queue_parent_node = queue_sched_node->parent;
- buf->src_teid = queue_parent_node->info.node_teid;
- buf->dest_teid = dst_node->info.node_teid;
- buf->txqs[0].q_teid = queue_sched_node->info.node_teid;
- buf->txqs[0].txq_id = queue_id;
-
- ret = ice_aq_move_recfg_lan_txq(hw, 1, true, false, false, false, 50,
- NULL, buf, buf_size, &txqs_moved, NULL);
- if (ret || txqs_moved == 0) {
- PMD_DRV_LOG(ERR, "move lan queue %u failed", queue_id);
- rte_free(buf);
- return ICE_ERR_PARAM;
- }
-
- if (queue_parent_node->num_children > 0) {
- queue_parent_node->num_children--;
- queue_parent_node->children[queue_parent_node->num_children] = NULL;
- } else {
- PMD_DRV_LOG(ERR, "invalid children number %d for queue %u",
- queue_parent_node->num_children, queue_id);
- rte_free(buf);
- return ICE_ERR_PARAM;
- }
- dst_node->children[dst_node->num_children++] = queue_sched_node;
- queue_sched_node->parent = dst_node;
- ice_sched_query_elem(hw, queue_sched_node->info.node_teid, &queue_sched_node->info);
-
- rte_free(buf);
- return ret;
-}
-
static int ice_set_node_rate(struct ice_hw *hw,
struct ice_tm_node *tm_node,
struct ice_sched_node *sched_node)
@@ -723,240 +676,191 @@ static int ice_cfg_hw_node(struct ice_hw *hw,
return 0;
}
-static struct ice_sched_node *ice_get_vsi_node(struct ice_hw *hw)
+int
+ice_tm_setup_txq_node(struct ice_pf *pf, struct ice_hw *hw, uint16_t qid, uint32_t teid)
{
- struct ice_sched_node *node = hw->port_info->root;
- uint32_t vsi_layer = hw->num_tx_sched_layers - ICE_VSI_LAYER_OFFSET;
- uint32_t i;
+ struct ice_sched_node *hw_node = ice_sched_find_node_by_teid(hw->port_info->root, teid);
+ struct ice_tm_node *sw_node = find_node(pf->tm_conf.root, qid);
- for (i = 0; i < vsi_layer; i++)
- node = node->children[0];
-
- return node;
-}
-
-static int ice_reset_noleaf_nodes(struct rte_eth_dev *dev)
-{
- struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_sched_node *vsi_node = ice_get_vsi_node(hw);
- struct ice_tm_node *root = pf->tm_conf.root;
- uint32_t i;
- int ret;
-
- /* reset vsi_node */
- ret = ice_set_node_rate(hw, NULL, vsi_node);
- if (ret) {
- PMD_DRV_LOG(ERR, "reset vsi node failed");
- return ret;
- }
-
- if (root == NULL)
+ /* not configured in hierarchy */
+ if (sw_node == NULL)
return 0;
- for (i = 0; i < root->reference_count; i++) {
- struct ice_tm_node *tm_node = root->children[i];
+ sw_node->sched_node = hw_node;
- if (tm_node->sched_node == NULL)
- continue;
+ /* if the queue node has been put in the wrong place in hierarchy */
+ if (hw_node->parent != sw_node->parent->sched_node) {
+ struct ice_aqc_move_txqs_data *buf;
+ uint8_t txqs_moved = 0;
+ uint16_t buf_size = ice_struct_size(buf, txqs, 1);
+
+ buf = ice_malloc(hw, buf_size);
+ if (buf == NULL)
+ return -ENOMEM;
- ret = ice_cfg_hw_node(hw, NULL, tm_node->sched_node);
- if (ret) {
- PMD_DRV_LOG(ERR, "reset queue group node %u failed", tm_node->id);
- return ret;
+ struct ice_sched_node *parent = hw_node->parent;
+ struct ice_sched_node *new_parent = sw_node->parent->sched_node;
+ buf->src_teid = parent->info.node_teid;
+ buf->dest_teid = new_parent->info.node_teid;
+ buf->txqs[0].q_teid = hw_node->info.node_teid;
+ buf->txqs[0].txq_id = qid;
+
+ int ret = ice_aq_move_recfg_lan_txq(hw, 1, true, false, false, false, 50,
+ NULL, buf, buf_size, &txqs_moved, NULL);
+ if (ret || txqs_moved == 0) {
+ PMD_DRV_LOG(ERR, "move lan queue %u failed", qid);
+ ice_free(hw, buf);
+ return ICE_ERR_PARAM;
}
- tm_node->sched_node = NULL;
+
+ /* now update the ice_sched_nodes to match physical layout */
+ new_parent->children[new_parent->num_children++] = hw_node;
+ hw_node->parent = new_parent;
+ ice_sched_query_elem(hw, hw_node->info.node_teid, &hw_node->info);
+ for (uint16_t i = 0; i < parent->num_children; i++)
+ if (parent->children[i] == hw_node) {
+ /* to remove, just overwrite the old node slot with the last ptr */
+ parent->children[i] = parent->children[--parent->num_children];
+ break;
+ }
}
- return 0;
+ return ice_cfg_hw_node(hw, sw_node, hw_node);
}
-static int ice_remove_leaf_nodes(struct rte_eth_dev *dev)
+/* from a given node, recursively deletes all the nodes that belong to that vsi.
+ * Any nodes which can't be deleted because they have children belonging to a different
+ * VSI, are now also adjusted to belong to that VSI also
+ */
+static int
+free_sched_node_recursive(struct ice_port_info *pi, const struct ice_sched_node *root,
+ struct ice_sched_node *node, uint8_t vsi_id)
{
- int ret = 0;
- int i;
+ uint16_t i = 0;
- for (i = 0; i < dev->data->nb_tx_queues; i++) {
- ret = ice_tx_queue_stop(dev, i);
- if (ret) {
- PMD_DRV_LOG(ERR, "stop queue %u failed", i);
- break;
+ while (i < node->num_children) {
+ if (node->children[i]->vsi_handle != vsi_id) {
+ i++;
+ continue;
}
+ free_sched_node_recursive(pi, root, node->children[i], vsi_id);
}
- return ret;
-}
-
-static int ice_add_leaf_nodes(struct rte_eth_dev *dev)
-{
- int ret = 0;
- int i;
-
- for (i = 0; i < dev->data->nb_tx_queues; i++) {
- ret = ice_tx_queue_start(dev, i);
- if (ret) {
- PMD_DRV_LOG(ERR, "start queue %u failed", i);
- break;
- }
+ if (node != root) {
+ if (node->num_children == 0)
+ ice_free_sched_node(pi, node);
+ else
+ node->vsi_handle = node->children[0]->vsi_handle;
}
- return ret;
+ return 0;
}
-int ice_do_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error)
+static int
+create_sched_node_recursive(struct ice_port_info *pi, struct ice_tm_node *sw_node,
+ struct ice_sched_node *hw_root, uint16_t *created)
{
- struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
- struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
- struct ice_tm_node *root;
- struct ice_sched_node *vsi_node = NULL;
- struct ice_sched_node *queue_node;
- struct ice_tx_queue *txq;
- int ret_val = 0;
- uint32_t i;
- uint32_t idx_vsi_child;
- uint32_t idx_qg;
- uint32_t nb_vsi_child;
- uint32_t nb_qg;
- uint32_t qid;
- uint32_t q_teid;
-
- /* remove leaf nodes */
- ret_val = ice_remove_leaf_nodes(dev);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "reset no-leaf nodes failed");
- goto fail_clear;
- }
-
- /* reset no-leaf nodes. */
- ret_val = ice_reset_noleaf_nodes(dev);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "reset leaf nodes failed");
- goto add_leaf;
- }
-
- /* config vsi node */
- vsi_node = ice_get_vsi_node(hw);
- root = pf->tm_conf.root;
-
- ret_val = ice_set_node_rate(hw, root, vsi_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure vsi node %u bandwidth failed",
- root->id);
- goto add_leaf;
- }
-
- /* config queue group nodes */
- nb_vsi_child = vsi_node->num_children;
- nb_qg = vsi_node->children[0]->num_children;
-
- idx_vsi_child = 0;
- idx_qg = 0;
-
- if (root == NULL)
- goto commit;
-
- for (i = 0; i < root->reference_count; i++) {
- struct ice_tm_node *tm_node = root->children[i];
- struct ice_tm_node *tm_child_node;
- struct ice_sched_node *qgroup_sched_node =
- vsi_node->children[idx_vsi_child]->children[idx_qg];
- uint32_t j;
-
- ret_val = ice_cfg_hw_node(hw, tm_node, qgroup_sched_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure queue group node %u failed",
- tm_node->id);
- goto reset_leaf;
- }
-
- for (j = 0; j < tm_node->reference_count; j++) {
- tm_child_node = tm_node->children[j];
- qid = tm_child_node->id;
- ret_val = ice_tx_queue_start(dev, qid);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "start queue %u failed", qid);
- goto reset_leaf;
- }
- txq = dev->data->tx_queues[qid];
- q_teid = txq->q_teid;
- queue_node = ice_sched_get_node(hw->port_info, q_teid);
- if (queue_node == NULL) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "get queue %u node failed", qid);
- goto reset_leaf;
- }
- if (queue_node->info.parent_teid != qgroup_sched_node->info.node_teid) {
- ret_val = ice_move_recfg_lan_txq(dev, queue_node,
- qgroup_sched_node, qid);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "move queue %u failed", qid);
- goto reset_leaf;
- }
- }
- ret_val = ice_cfg_hw_node(hw, tm_child_node, queue_node);
- if (ret_val) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR,
- "configure queue group node %u failed",
- tm_node->id);
- goto reset_leaf;
- }
- }
-
- idx_qg++;
- if (idx_qg >= nb_qg) {
- idx_qg = 0;
- idx_vsi_child++;
+ struct ice_sched_node *parent = sw_node->sched_node;
+ uint32_t teid;
+ uint16_t added;
+
+ /* first create all child nodes */
+ for (uint16_t i = 0; i < sw_node->reference_count; i++) {
+ struct ice_tm_node *tm_node = sw_node->children[i];
+ int res = ice_sched_add_elems(pi, hw_root,
+ parent, parent->tx_sched_layer + 1,
+ 1 /* num nodes */, &added, &teid,
+ NULL /* no pre-alloc */);
+ if (res != 0) {
+ PMD_DRV_LOG(ERR, "Error with ice_sched_add_elems, adding child node to teid %u\n",
+ parent->info.node_teid);
+ return -1;
}
- if (idx_vsi_child >= nb_vsi_child) {
- error->type = RTE_TM_ERROR_TYPE_UNSPECIFIED;
- PMD_DRV_LOG(ERR, "too many queues");
- goto reset_leaf;
+ struct ice_sched_node *hw_node = ice_sched_find_node_by_teid(parent, teid);
+ if (ice_cfg_hw_node(pi->hw, tm_node, hw_node) != 0) {
+ PMD_DRV_LOG(ERR, "Error configuring node %u at layer %u",
+ teid, parent->tx_sched_layer + 1);
+ return -1;
}
+ tm_node->sched_node = hw_node;
+ created[hw_node->tx_sched_layer]++;
}
-commit:
- pf->tm_conf.committed = true;
- pf->tm_conf.clear_on_fail = clear_on_fail;
+ /* if we have just created the child nodes in the q-group, i.e. last non-leaf layer,
+ * then just return, rather than trying to create leaf nodes.
+ * That is done later at queue start.
+ */
+ if (sw_node->level + 2 == ice_get_leaf_level(pi->hw))
+ return 0;
- return ret_val;
+ for (uint16_t i = 0; i < sw_node->reference_count; i++) {
+ if (sw_node->children[i]->reference_count == 0)
+ continue;
-reset_leaf:
- ice_remove_leaf_nodes(dev);
-add_leaf:
- ice_add_leaf_nodes(dev);
- ice_reset_noleaf_nodes(dev);
-fail_clear:
- /* clear all the traffic manager configuration */
- if (clear_on_fail) {
- ice_tm_conf_uninit(dev);
- ice_tm_conf_init(dev);
+ if (create_sched_node_recursive(pi, sw_node->children[i], hw_root, created) < 0)
+ return -1;
}
- return ret_val;
+ return 0;
}
-static int ice_hierarchy_commit(struct rte_eth_dev *dev,
- int clear_on_fail,
- struct rte_tm_error *error)
+static int
+apply_topology_updates(struct rte_eth_dev *dev __rte_unused)
{
+ return 0;
+}
+
+static int
+commit_new_hierarchy(struct rte_eth_dev *dev)
+{
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+ struct ice_port_info *pi = hw->port_info;
+ struct ice_tm_node *sw_root = pf->tm_conf.root;
+ struct ice_sched_node *new_vsi_root = (pi->has_tc) ? pi->root->children[0] : pi->root;
+ uint16_t nodes_created_per_level[10] = {0}; /* counted per hw level, not per logical */
+ uint8_t q_lvl = ice_get_leaf_level(hw);
+ uint8_t qg_lvl = q_lvl - 1;
+
+ /* check if we have a previously applied topology */
+ if (sw_root->sched_node != NULL)
+ return apply_topology_updates(dev);
+
+ free_sched_node_recursive(pi, new_vsi_root, new_vsi_root, new_vsi_root->vsi_handle);
+
+ sw_root->sched_node = new_vsi_root;
+ if (create_sched_node_recursive(pi, sw_root, new_vsi_root, nodes_created_per_level) < 0)
+ return -1;
+ for (uint16_t i = 0; i < RTE_DIM(nodes_created_per_level); i++)
+ PMD_DRV_LOG(DEBUG, "Created %u nodes at level %u\n",
+ nodes_created_per_level[i], i);
+ hw->vsi_ctx[pf->main_vsi->idx]->sched.vsi_node[0] = new_vsi_root;
+
+ pf->main_vsi->nb_qps =
+ RTE_MIN(nodes_created_per_level[qg_lvl] * hw->max_children[qg_lvl],
+ hw->layer_info[q_lvl].max_device_nodes);
+
+ pf->tm_conf.committed = true; /* set flag to be checks on queue start */
+
+ return ice_alloc_lan_q_ctx(hw, 0, 0, pf->main_vsi->nb_qps);
+}
- /* if device not started, simply set committed flag and return. */
- if (!dev->data->dev_started) {
- pf->tm_conf.committed = true;
- pf->tm_conf.clear_on_fail = clear_on_fail;
- return 0;
+static int
+ice_hierarchy_commit(struct rte_eth_dev *dev,
+ int clear_on_fail,
+ struct rte_tm_error *error)
+{
+ RTE_SET_USED(error);
+ /* commit should only be done to topology before start! */
+ if (dev->data->dev_started)
+ return -1;
+
+ uint64_t start = rte_rdtsc();
+ int ret = commit_new_hierarchy(dev);
+ if (ret < 0 && clear_on_fail) {
+ ice_tm_conf_uninit(dev);
+ ice_tm_conf_init(dev);
}
-
- return ice_do_hierarchy_commit(dev, clear_on_fail, error);
+ uint64_t time = rte_rdtsc() - start;
+ PMD_DRV_LOG(DEBUG, "Time to apply hierarchy = %.1f\n", (float)time / rte_get_timer_hz());
+ return ret;
}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 15/16] net/ice: add minimal capability reporting API
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (13 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 14/16] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 16/16] net/ice: do early check on node level when adding Bruce Richardson
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
Incomplete but reports number of available layers
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_ethdev.h | 1 +
drivers/net/ice/ice_tm.c | 17 +++++++++++++++++
2 files changed, 18 insertions(+)
diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h
index cb1a7e8e0d..6bebc511e4 100644
--- a/drivers/net/ice/ice_ethdev.h
+++ b/drivers/net/ice/ice_ethdev.h
@@ -682,6 +682,7 @@ int ice_rem_rss_cfg_wrap(struct ice_pf *pf, uint16_t vsi_id,
struct ice_rss_hash_cfg *cfg);
void ice_tm_conf_init(struct rte_eth_dev *dev);
void ice_tm_conf_uninit(struct rte_eth_dev *dev);
+
extern const struct rte_tm_ops ice_tm_ops;
static inline int
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 80039c8aff..3dcd091c38 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -33,8 +33,12 @@ static int ice_shaper_profile_add(struct rte_eth_dev *dev,
static int ice_shaper_profile_del(struct rte_eth_dev *dev,
uint32_t shaper_profile_id,
struct rte_tm_error *error);
+static int ice_tm_capabilities_get(struct rte_eth_dev *dev,
+ struct rte_tm_capabilities *cap,
+ struct rte_tm_error *error);
const struct rte_tm_ops ice_tm_ops = {
+ .capabilities_get = ice_tm_capabilities_get,
.shaper_profile_add = ice_shaper_profile_add,
.shaper_profile_delete = ice_shaper_profile_del,
.node_add = ice_tm_node_add,
@@ -864,3 +868,16 @@ ice_hierarchy_commit(struct rte_eth_dev *dev,
PMD_DRV_LOG(DEBUG, "Time to apply hierarchy = %.1f\n", (float)time / rte_get_timer_hz());
return ret;
}
+
+static int
+ice_tm_capabilities_get(struct rte_eth_dev *dev, struct rte_tm_capabilities *cap,
+ struct rte_tm_error *error)
+{
+ struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+ *cap = (struct rte_tm_capabilities){
+ .n_levels_max = hw->num_tx_sched_layers - hw->port_info->has_tc,
+ };
+ if (error)
+ error->type = RTE_TM_ERROR_TYPE_NONE;
+ return 0;
+}
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
* [PATCH v3 16/16] net/ice: do early check on node level when adding
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
` (14 preceding siblings ...)
2024-08-12 15:28 ` [PATCH v3 15/16] net/ice: add minimal capability reporting API Bruce Richardson
@ 2024-08-12 15:28 ` Bruce Richardson
15 siblings, 0 replies; 49+ messages in thread
From: Bruce Richardson @ 2024-08-12 15:28 UTC (permalink / raw)
To: dev; +Cc: Bruce Richardson
When adding a new scheduler node, the parameters for leaf nodes and
non-leaf nodes are different, and which parameter checks are done is
determined by checking the node level i.e. if it's the lowest (leaf)
node level or not. However, if the node level itself is incorrectly
specified, the error messages got can be confusing since the user may
add a leaf node using e.g. the testpmd command to explicitly add a leaf
node, yet get error messages only relevant to non-leaf nodes due to an
incorrect level parameter.
We can avoid these confusing errors by doing a check that the level
matches "parent->level + 1" before doing a more detailed parameter
check.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
drivers/net/ice/ice_tm.c | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ice/ice_tm.c b/drivers/net/ice/ice_tm.c
index 3dcd091c38..e05ad8a8e7 100644
--- a/drivers/net/ice/ice_tm.c
+++ b/drivers/net/ice/ice_tm.c
@@ -426,6 +426,13 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
if (level_id == RTE_TM_NODE_LEVEL_ID_ANY && parent_node != NULL)
level_id = parent_node->level + 1;
+ /* check level */
+ if (parent_node != NULL && level_id != parent_node->level + 1) {
+ error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
+ error->message = "Wrong level";
+ return -EINVAL;
+ }
+
ret = ice_node_param_check(node_id, priority, weight,
params, level_id == ice_get_leaf_level(hw), error);
if (ret)
@@ -493,12 +500,6 @@ ice_tm_node_add(struct rte_eth_dev *dev, uint32_t node_id,
error->message = "parent is not valid";
return -EINVAL;
}
- /* check level */
- if (level_id != parent_node->level + 1) {
- error->type = RTE_TM_ERROR_TYPE_NODE_PARAMS;
- error->message = "Wrong level";
- return -EINVAL;
- }
/* check the max children allowed at this level */
if (parent_node->reference_count >= hw->max_children[parent_node->level]) {
--
2.43.0
^ permalink raw reply [flat|nested] 49+ messages in thread
end of thread, other threads:[~2024-08-12 15:30 UTC | newest]
Thread overview: 49+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-08-07 9:33 [PATCH 00/15] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-07 9:33 ` [PATCH 01/15] net/ice: add traffic management node query function Bruce Richardson
2024-08-07 9:33 ` [PATCH 02/15] net/ice: detect stopping a flow-director queue twice Bruce Richardson
2024-08-07 9:33 ` [PATCH 03/15] net/ice: improve Tx scheduler graph output Bruce Richardson
2024-08-07 9:33 ` [PATCH 04/15] net/ice: add option to choose DDP package file Bruce Richardson
2024-08-07 9:33 ` [PATCH 05/15] net/ice: add option to download scheduler topology Bruce Richardson
2024-08-07 9:33 ` [PATCH 06/15] net/ice/base: allow init without TC class sched nodes Bruce Richardson
2024-08-07 9:33 ` [PATCH 07/15] net/ice/base: set VSI index on newly created nodes Bruce Richardson
2024-08-07 9:34 ` [PATCH 08/15] net/ice/base: read VSI layer info from VSI Bruce Richardson
2024-08-07 9:34 ` [PATCH 09/15] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
2024-08-07 9:34 ` [PATCH 10/15] net/ice/base: optimize subtree searches Bruce Richardson
2024-08-07 9:34 ` [PATCH 11/15] net/ice/base: make functions non-static Bruce Richardson
2024-08-07 9:34 ` [PATCH 12/15] net/ice/base: remove flag checks before topology upload Bruce Richardson
2024-08-07 9:34 ` [PATCH 13/15] net/ice: limit the number of queues to sched capabilities Bruce Richardson
2024-08-07 9:34 ` [PATCH 14/15] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
2024-08-07 9:34 ` [PATCH 15/15] net/ice: add minimal capability reporting API Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 00/15] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 01/15] net/ice: add traffic management node query function Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 02/15] net/ice: detect stopping a flow-director queue twice Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 03/15] net/ice: improve Tx scheduler graph output Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 04/15] net/ice: add option to choose DDP package file Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 05/15] net/ice: add option to download scheduler topology Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 06/15] net/ice/base: allow init without TC class sched nodes Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 07/15] net/ice/base: set VSI index on newly created nodes Bruce Richardson
2024-08-07 9:46 ` [PATCH v2 08/15] net/ice/base: read VSI layer info from VSI Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 09/15] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 10/15] net/ice/base: optimize subtree searches Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 11/15] net/ice/base: make functions non-static Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 12/15] net/ice/base: remove flag checks before topology upload Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 13/15] net/ice: limit the number of queues to sched capabilities Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 14/15] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
2024-08-07 9:47 ` [PATCH v2 15/15] net/ice: add minimal capability reporting API Bruce Richardson
2024-08-12 15:27 ` [PATCH v3 00/16] Improve rte_tm support in ICE driver Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 01/16] net/ice: add traffic management node query function Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 02/16] net/ice: detect stopping a flow-director queue twice Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 03/16] net/ice: improve Tx scheduler graph output Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 04/16] net/ice: add option to choose DDP package file Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 05/16] net/ice: add option to download scheduler topology Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 06/16] net/ice/base: allow init without TC class sched nodes Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 07/16] net/ice/base: set VSI index on newly created nodes Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 08/16] net/ice/base: read VSI layer info from VSI Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 09/16] net/ice/base: remove 255 limit on sched child nodes Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 10/16] net/ice/base: optimize subtree searches Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 11/16] net/ice/base: make functions non-static Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 12/16] net/ice/base: remove flag checks before topology upload Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 13/16] net/ice: limit the number of queues to sched capabilities Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 14/16] net/ice: enhance Tx scheduler hierarchy support Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 15/16] net/ice: add minimal capability reporting API Bruce Richardson
2024-08-12 15:28 ` [PATCH v3 16/16] net/ice: do early check on node level when adding Bruce Richardson
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).