2
0
mirror of https://github.com/openvswitch/ovs synced 2025-08-22 01:51:26 +00:00
ovs/lib/dpif-netdev-extract-study.c
Kumar Amber 72dd22a0df dpif-netdev: Add study function to select the best mfex function
The study function runs all the available implementations
of miniflow_extract and makes a choice whose hitmask has
maximum hits and sets the mfex to that function.

Study can be run at runtime using the following command:

$ ovs-appctl dpif-netdev/miniflow-parser-set study

Signed-off-by: Kumar Amber <kumar.amber@intel.com>
Co-authored-by: Harry van Haaren <harry.van.haaren@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Eelco Chaudron <echaudro@redhat.com>
Acked-by: Flavio Leitner <fbl@sysclose.org>
Signed-off-by: Ian Stokes <ian.stokes@intel.com>
2021-07-16 10:29:52 +01:00

136 lines
4.9 KiB
C

/*
* Copyright (c) 2021 Intel.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at:
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <config.h>
#include <errno.h>
#include <stdint.h>
#include <string.h>
#include "dpif-netdev-private-thread.h"
#include "openvswitch/vlog.h"
#include "ovs-thread.h"
VLOG_DEFINE_THIS_MODULE(dpif_mfex_extract_study);
static atomic_uint32_t mfex_study_pkts_count = 0;
/* Struct to hold miniflow study stats. */
struct study_stats {
uint32_t pkt_count;
uint32_t impl_hitcount[MFEX_IMPL_MAX];
};
/* Define per thread data to hold the study stats. */
DEFINE_PER_THREAD_MALLOCED_DATA(struct study_stats *, study_stats);
/* Allocate per thread PMD pointer space for study_stats. */
static inline struct study_stats *
mfex_study_get_study_stats_ptr(void)
{
struct study_stats *stats = study_stats_get();
if (OVS_UNLIKELY(!stats)) {
stats = xzalloc(sizeof *stats);
study_stats_set_unsafe(stats);
}
return stats;
}
uint32_t
mfex_study_traffic(struct dp_packet_batch *packets,
struct netdev_flow_key *keys,
uint32_t keys_size, odp_port_t in_port,
struct dp_netdev_pmd_thread *pmd_handle)
{
uint32_t hitmask = 0;
uint32_t mask = 0;
struct dp_netdev_pmd_thread *pmd = pmd_handle;
struct dpif_miniflow_extract_impl *miniflow_funcs;
struct study_stats *stats = mfex_study_get_study_stats_ptr();
miniflow_funcs = dpif_mfex_impl_info_get();
/* Run traffic optimized miniflow_extract to collect the hitmask
* to be compared after certain packets have been hit to choose
* the best miniflow_extract version for that traffic.
*/
for (int i = MFEX_IMPL_START_IDX; i < MFEX_IMPL_MAX; i++) {
if (!miniflow_funcs[i].available) {
continue;
}
hitmask = miniflow_funcs[i].extract_func(packets, keys, keys_size,
in_port, pmd_handle);
stats->impl_hitcount[i] += count_1bits(hitmask);
/* If traffic is not classified then we dont overwrite the keys
* array in minfiflow implementations so its safe to create a
* mask for all those packets whose miniflow have been created.
*/
mask |= hitmask;
}
stats->pkt_count += dp_packet_batch_size(packets);
/* Choose the best implementation after a minimum packets have been
* processed.
*/
if (stats->pkt_count >= MFEX_MAX_PKT_COUNT) {
uint32_t best_func_index = MFEX_IMPL_START_IDX;
uint32_t max_hits = 0;
for (int i = MFEX_IMPL_START_IDX; i < MFEX_IMPL_MAX; i++) {
if (stats->impl_hitcount[i] > max_hits) {
max_hits = stats->impl_hitcount[i];
best_func_index = i;
}
}
/* If 50% of the packets hit, enable the function. */
if (max_hits >= (mfex_study_pkts_count / 2)) {
miniflow_extract_func mf_func =
miniflow_funcs[best_func_index].extract_func;
atomic_uintptr_t *pmd_func = (void *)&pmd->miniflow_extract_opt;
atomic_store_relaxed(pmd_func, (uintptr_t) mf_func);
VLOG_INFO("MFEX study chose impl %s: (hits %u/%u pkts)",
miniflow_funcs[best_func_index].name, max_hits,
stats->pkt_count);
} else {
/* Set the implementation to null for default miniflow. */
miniflow_extract_func mf_func =
miniflow_funcs[MFEX_IMPL_SCALAR].extract_func;
atomic_uintptr_t *pmd_func = (void *)&pmd->miniflow_extract_opt;
atomic_store_relaxed(pmd_func, (uintptr_t) mf_func);
VLOG_INFO("Not enough packets matched (%u/%u), disabling"
" optimized MFEX.", max_hits, stats->pkt_count);
}
/* In debug mode show stats for all the counters. */
if (VLOG_IS_DBG_ENABLED()) {
for (int i = MFEX_IMPL_START_IDX; i < MFEX_IMPL_MAX; i++) {
VLOG_DBG("MFEX study results for implementation %s:"
" (hits %u/%u pkts)", miniflow_funcs[i].name,
stats->impl_hitcount[i], stats->pkt_count);
}
}
/* Reset stats so that study function can be called again
* for next traffic type and optimal function ptr can be
* chosen.
*/
memset(stats, 0, sizeof(struct study_stats));
}
return mask;
}