Hi,
> -----Original Message-----
> From: Rongwei Liu <rongw...@nvidia.com>
> Sent: Wednesday, July 21, 2021 10:06 AM
> To: Matan Azrad <ma...@nvidia.com>; Slava Ovsiienko
> <viachesl...@nvidia.com>; Ori Kam <or...@nvidia.com>; NBU-Contact-
> Thomas Monjalon <tho...@monjalon.net>; Wisam Monther
> <wis...@nvidia.com>
> Cc: dev@dpdk.org; Raslan Darawsheh <rasl...@nvidia.com>; Haifei Luo
> <haif...@nvidia.com>; Jiawei(Jonny) Wang <jiaw...@nvidia.com>
> Subject: [PATCH v2 1/3] app/flow-perf: support meter policy API
>
> Add option "policy-mtr" to indicate if meter creation will include policy or
> not.
> Meter creation will keep same without it.
>
> With "policy-mtr", policy is introduced. API create_meter_policy is to create
> a
> policy. API create_meter_rule will use it to create meter.
>
> Add option "policy-g_actions" to specify meter policy green color actions.
> W/o this, policy creation will fail since there is no default one.
>
> Signed-off-by: Haifei Luo <haif...@nvidia.com>
> Signed-off-by: Jiawei Wang <jiaw...@nvidia.com>
> Signed-off-by: Rongwei Liu <rongw...@nvidia.com>
> ---
> app/test-flow-perf/main.c | 121 ++++++++++++++++++++++++++++++--
> -
> doc/guides/tools/flow-perf.rst | 6 ++
> 2 files changed, 119 insertions(+), 8 deletions(-)
>
> diff --git a/app/test-flow-perf/main.c b/app/test-flow-perf/main.c index
> 9be8edc31d..e0d94f943a 100644
> --- a/app/test-flow-perf/main.c
> +++ b/app/test-flow-perf/main.c
> @@ -37,6 +37,7 @@
> #include <rte_mtr.h>
>
> #include "config.h"
> +#include "actions_gen.h"
> #include "flow_gen.h"
>
> #define MAX_BATCHES_COUNT 100
> @@ -49,10 +50,12 @@ static uint8_t flow_group;
>
> static uint64_t encap_data;
> static uint64_t decap_data;
> +static uint64_t g_actions;
>
> static uint64_t flow_items[MAX_ITEMS_NUM]; static uint64_t
> flow_actions[MAX_ACTIONS_NUM]; static uint64_t
> flow_attrs[MAX_ATTRS_NUM];
> +static uint32_t g_policy_id[MAX_PORTS];
> static uint8_t items_idx, actions_idx, attrs_idx;
>
> static uint64_t ports_mask;
> @@ -62,6 +65,7 @@ static bool delete_flag; static bool
> dump_socket_mem_flag; static bool enable_fwd; static bool unique_data;
> +static bool policy_mtr;
>
> static struct rte_mempool *mbuf_mp;
> static uint32_t nb_lcores;
> @@ -69,6 +73,7 @@ static uint32_t rules_count; static uint32_t rules_batch;
> static uint32_t hairpin_queues_num; /* total hairpin q number - default: 0 */
> static uint32_t nb_lcores;
> +static uint64_t meter_cir;
i see it's defined but not really used, i think all the meter_cir should be
squashed w/ next commit
>
> #define MAX_PKT_BURST 32
> #define LCORE_MODE_PKT 1
> @@ -134,6 +139,8 @@ usage(char *progname)
> printf(" --portmask=N: hexadecimal bitmask of ports used\n");
> printf(" --unique-data: flag to set using unique data for all"
> " actions that support data, such as header modify and encap
> actions\n");
> + printf(" --policy-mtr: To create meter with policy\n");
> + printf(" --policy-g_actions: To set meter policy green color
> +actions\n");
>
> printf("To set flow attributes:\n");
> printf(" --ingress: set ingress attribute in flows\n"); @@ -573,6
> +580,9 @@ args_parse(int argc, char **argv)
> { "unique-data", 0, 0, 0 },
> { "portmask", 1, 0, 0 },
> { "cores", 1, 0, 0 },
> + { "policy-mtr", 0, 0, 0 },
> + { "policy-g_actions", 1, 0, 0 },
i prefer to have those in the end with the actions
> + { "meter-profile-alg", 1, 0, 0 },
i don't see any use or parse for meter-profile-alg
> /* Attributes */
> { "ingress", 0, 0, 0 },
> { "egress", 0, 0, 0 },
> @@ -802,6 +812,32 @@ args_parse(int argc, char **argv)
> RTE_MAX_LCORE);
> }
> }
> + if (strcmp(lgopts[opt_idx].name, "policy-mtr") == 0)
do we really need this parameter,
for std meter user can use --meter
for the new one, something like: --policy-meter=[G_ACTIONS]
I'm trying to reduce the dependencies for any new parameters
> + policy_mtr = true;
> + if (strcmp(lgopts[opt_idx].name,
> + "policy-g_actions") == 0) {
> + token = strtok(optarg, ",");
> + while (token != NULL) {
> + for (i = 0;
> + i < RTE_DIM(flow_options); i++) {
> + if (strcmp(optarg,
> + flow_options[i].str) == 0) {
> + g_actions |=
> + flow_options[i].mask;
> + break;
> + }
> + }
> + /* Reached last item with no match
> */
> + if (i == (RTE_DIM(flow_options) - 1)) {
> + fprintf(stderr,
> + "Invalid g_actions "
> + "item: %s\n", token);
> + usage(argv[0]);
> + rte_exit(EXIT_SUCCESS,
> "Invalid g_actions item\n");
> + }
> + token = strtok(NULL, ",");
> + }
> + }
> break;
> default:
> usage(argv[0]);
> @@ -912,6 +948,62 @@ has_meter(void)
> return 0;
> }
>
> +static void
> +create_meter_policy(void)
> +{
> + struct rte_mtr_error error;
> + uint32_t policy_id;
> + int ret, i, port_id;
> + struct rte_mtr_meter_policy_params policy;
> + struct rte_flow_action r_actions[2];
> + uint16_t nr_ports;
> + struct rte_flow_action actions[MAX_ACTIONS_NUM];
> + uint64_t flow_actions[MAX_ACTIONS_NUM];
> + int lcore_counter = 0;
> + int lcore_id = rte_lcore_id();
not sure that you are using the correct mapping of multiple cores.
In short the user can pass: cores=90,91,92,93,94,95
The app usually map all cores to new internal ids from 0 - N.
While this solution will break the user freedom of choosing any core
> +
> + memset(actions, 0, sizeof(actions));
struct rte_flow_action actions[MAX_ACTIONS_NUM] = {0, 0};
> + memset(flow_actions, 0, sizeof(flow_actions));
uint64_t flow_actions[MAX_ACTIONS_NUM] = { 0 };
> + memset(&policy, 0, sizeof(policy));
in rte_mtr_meter_policy_params it have const struct rte_flow_action,
and since it's const i really don't like the idea of setting it to 0 by memset.
it's one time init and that's it
> + RTE_LCORE_FOREACH(i) {
> + /* If core not needed return. */
> + if (lcore_id == i) {
> + if (lcore_counter >= (int) mc_pool.cores_count)
> + return;
> + break;
> + }
> + lcore_counter++;
> + }
> + lcore_id = lcore_counter;
> +
> + if (lcore_id >= (int) mc_pool.cores_count)
> + return;
no need for such logic, just take the core id as parameter, all this logic
already done
> +
> + flow_actions[0] = g_actions;
> + fill_actions(actions, flow_actions, 0, 0, 0,
> + 0, 0, lcore_id, unique_data);
Will this create one meter for all flows or you are doing new meter for each
flow?
can you please elaborate?
What is the exact goal here?
> +
> + r_actions[0].type = RTE_FLOW_ACTION_TYPE_DROP;
> + r_actions[0].conf = NULL;
> + r_actions[1].type = RTE_FLOW_ACTION_TYPE_END;
> + r_actions[1].conf = NULL;
> +
> + policy.actions[RTE_COLOR_GREEN] = &actions[0];
> + policy.actions[RTE_COLOR_YELLOW] = NULL;
> + policy.actions[RTE_COLOR_RED] = &r_actions[0];
> +
> + nr_ports = rte_eth_dev_count_avail();
> + for (port_id = 0; port_id < nr_ports; port_id++) {
> + policy_id = port_id + 10;
> + ret = rte_mtr_meter_policy_add(port_id, policy_id,
> + &policy, &error);
> + if (ret)
> + printf("meter policy add failed "
> + "port_id %d\n", port_id);
> + g_policy_id[port_id] = policy_id;
> + }
> +}
> +
> static void
> create_meter_rule(int port_id, uint32_t counter) { @@ -928,7 +1020,14 @@
> create_meter_rule(int port_id, uint32_t counter)
>
> /*create meter*/
> params.meter_profile_id = default_prof_id;
> - ret = rte_mtr_create(port_id, counter, ¶ms, 1, &error);
> +
> + if (!policy_mtr) {
> + ret = rte_mtr_create(port_id, counter, ¶ms, 1, &error);
> + } else {
> + params.meter_policy_id = g_policy_id[port_id];
> + ret = rte_mtr_create(port_id, counter, ¶ms, 0, &error);
> + }
> +
> if (ret != 0) {
> printf("Port %u create meter idx(%d) error(%d) message:
> %s\n",
> port_id, counter, error.type,
> @@ -942,11 +1041,16 @@ destroy_meter_rule(int port_id, uint32_t counter)
> {
> struct rte_mtr_error error;
>
> - if (rte_mtr_destroy(port_id, counter, &error)) {
> - printf("Port %u destroy meter(%d) error(%d) message:
> %s\n",
> + if (policy_mtr) {
> + if (rte_mtr_meter_policy_delete(port_id, counter+1,
> &error))
> + printf("error delete policy %d\n", counter+1);
> + } else {
> + if (rte_mtr_destroy(port_id, counter, &error)) {
> + printf("Port %u destroy meter(%d) error(%d)
> message: %s\n",
> port_id, counter, error.type,
> error.message ? error.message : "(no stated
> reason)");
> - rte_exit(EXIT_FAILURE, "Error in deleting meter rule\n");
> + rte_exit(EXIT_FAILURE, "Error in deleting meter
> rule");
> + }
> }
> }
>
> @@ -1051,12 +1155,10 @@ create_meter_profile(void)
> /* If port outside portmask */
> if (!((ports_mask >> port_id) & 0x1))
> continue;
> -
> mp.alg = RTE_MTR_SRTCM_RFC2697;
> - mp.srtcm_rfc2697.cir = METER_CIR;
> - mp.srtcm_rfc2697.cbs = METER_CIR / 8;
> + mp.srtcm_rfc2697.cir = meter_cir;
> + mp.srtcm_rfc2697.cbs = meter_cir / 8;
> mp.srtcm_rfc2697.ebs = 0;
> -
> ret = rte_mtr_meter_profile_add
> (port_id, DEFAULT_METER_PROF_ID, &mp, &error);
> if (ret != 0) {
> @@ -1875,6 +1977,7 @@ main(int argc, char **argv)
> dump_socket_mem_flag = false;
> flow_group = DEFAULT_GROUP;
> unique_data = false;
> + meter_cir = METER_CIR;
>
> signal(SIGINT, signal_handler);
> signal(SIGTERM, signal_handler);
> @@ -1894,6 +1997,8 @@ main(int argc, char **argv)
>
> if (has_meter())
> create_meter_profile();
> + if (policy_mtr)
> + create_meter_policy();
> rte_eal_mp_remote_launch(run_rte_flow_handler_cores, NULL,
> CALL_MAIN);
>
> if (enable_fwd) {
> diff --git a/doc/guides/tools/flow-perf.rst b/doc/guides/tools/flow-perf.rst
> index 280bf7e0e0..90b6934537 100644
> --- a/doc/guides/tools/flow-perf.rst
> +++ b/doc/guides/tools/flow-perf.rst
> @@ -354,3 +354,9 @@ Actions:
> * ``--meter``
> Add meter action to all flows actions.
> Currently, 1 meter profile -> N meter rules -> N rte flows.
> +
> +* ``--policy-mtr``
> + Add policy-mtr to create meter with policy.
> +
> +* ``--policy-g_actions``
> + Add policy-g_actions to specify policy green color actions.
> --
> 2.27.0
