diff options
author | 2017-05-16 14:51:32 +0200 | |
---|---|---|
committer | 2017-05-16 14:51:32 +0200 | |
commit | fca143f059a0bddd7d47b8dc2df646a891b0eb0f (patch) | |
tree | 4bfeadc905c977e45e54a90c42330553b8942e4e /lib/librte_distributor/rte_distributor_v20.c | |
parent | ce3d555e43e3795b5d9507fcfc76b7a0a92fd0d6 (diff) |
Imported Upstream version 17.05
Diffstat (limited to 'lib/librte_distributor/rte_distributor_v20.c')
-rw-r--r-- | lib/librte_distributor/rte_distributor_v20.c | 427 |
1 files changed, 427 insertions, 0 deletions
diff --git a/lib/librte_distributor/rte_distributor_v20.c b/lib/librte_distributor/rte_distributor_v20.c new file mode 100644 index 00000000..bb6c5d70 --- /dev/null +++ b/lib/librte_distributor/rte_distributor_v20.c @@ -0,0 +1,427 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <stdio.h> +#include <sys/queue.h> +#include <string.h> +#include <rte_mbuf.h> +#include <rte_memory.h> +#include <rte_memzone.h> +#include <rte_errno.h> +#include <rte_compat.h> +#include <rte_string_fns.h> +#include <rte_eal_memconfig.h> +#include "rte_distributor_v20.h" +#include "rte_distributor_private.h" + +TAILQ_HEAD(rte_distributor_list, rte_distributor_v20); + +static struct rte_tailq_elem rte_distributor_tailq = { + .name = "RTE_DISTRIBUTOR", +}; +EAL_REGISTER_TAILQ(rte_distributor_tailq) + +/**** APIs called by workers ****/ + +void +rte_distributor_request_pkt_v20(struct rte_distributor_v20 *d, + unsigned worker_id, struct rte_mbuf *oldpkt) +{ + union rte_distributor_buffer_v20 *buf = &d->bufs[worker_id]; + int64_t req = (((int64_t)(uintptr_t)oldpkt) << RTE_DISTRIB_FLAG_BITS) + | RTE_DISTRIB_GET_BUF; + while (unlikely(buf->bufptr64 & RTE_DISTRIB_FLAGS_MASK)) + rte_pause(); + buf->bufptr64 = req; +} +VERSION_SYMBOL(rte_distributor_request_pkt, _v20, 2.0); + +struct rte_mbuf * +rte_distributor_poll_pkt_v20(struct rte_distributor_v20 *d, + unsigned worker_id) +{ + union rte_distributor_buffer_v20 *buf = &d->bufs[worker_id]; + if (buf->bufptr64 & RTE_DISTRIB_GET_BUF) + return NULL; + + /* since bufptr64 is signed, this should be an arithmetic shift */ + int64_t ret = buf->bufptr64 >> RTE_DISTRIB_FLAG_BITS; + return (struct rte_mbuf *)((uintptr_t)ret); +} +VERSION_SYMBOL(rte_distributor_poll_pkt, _v20, 2.0); + +struct rte_mbuf * +rte_distributor_get_pkt_v20(struct rte_distributor_v20 *d, + unsigned worker_id, struct rte_mbuf *oldpkt) +{ + struct rte_mbuf *ret; + rte_distributor_request_pkt_v20(d, worker_id, oldpkt); + while ((ret = rte_distributor_poll_pkt_v20(d, worker_id)) == NULL) + rte_pause(); + return ret; +} +VERSION_SYMBOL(rte_distributor_get_pkt, _v20, 2.0); + +int +rte_distributor_return_pkt_v20(struct rte_distributor_v20 *d, + unsigned worker_id, struct rte_mbuf *oldpkt) +{ + union rte_distributor_buffer_v20 *buf = &d->bufs[worker_id]; + uint64_t req = (((int64_t)(uintptr_t)oldpkt) << RTE_DISTRIB_FLAG_BITS) + | RTE_DISTRIB_RETURN_BUF; + buf->bufptr64 = req; + return 0; +} +VERSION_SYMBOL(rte_distributor_return_pkt, _v20, 2.0); + +/**** APIs called on distributor core ***/ + +/* as name suggests, adds a packet to the backlog for a particular worker */ +static int +add_to_backlog(struct rte_distributor_backlog *bl, int64_t item) +{ + if (bl->count == RTE_DISTRIB_BACKLOG_SIZE) + return -1; + + bl->pkts[(bl->start + bl->count++) & (RTE_DISTRIB_BACKLOG_MASK)] + = item; + return 0; +} + +/* takes the next packet for a worker off the backlog */ +static int64_t +backlog_pop(struct rte_distributor_backlog *bl) +{ + bl->count--; + return bl->pkts[bl->start++ & RTE_DISTRIB_BACKLOG_MASK]; +} + +/* stores a packet returned from a worker inside the returns array */ +static inline void +store_return(uintptr_t oldbuf, struct rte_distributor_v20 *d, + unsigned *ret_start, unsigned *ret_count) +{ + /* store returns in a circular buffer - code is branch-free */ + d->returns.mbufs[(*ret_start + *ret_count) & RTE_DISTRIB_RETURNS_MASK] + = (void *)oldbuf; + *ret_start += (*ret_count == RTE_DISTRIB_RETURNS_MASK) & !!(oldbuf); + *ret_count += (*ret_count != RTE_DISTRIB_RETURNS_MASK) & !!(oldbuf); +} + +static inline void +handle_worker_shutdown(struct rte_distributor_v20 *d, unsigned int wkr) +{ + d->in_flight_tags[wkr] = 0; + d->in_flight_bitmask &= ~(1UL << wkr); + d->bufs[wkr].bufptr64 = 0; + if (unlikely(d->backlog[wkr].count != 0)) { + /* On return of a packet, we need to move the + * queued packets for this core elsewhere. + * Easiest solution is to set things up for + * a recursive call. That will cause those + * packets to be queued up for the next free + * core, i.e. it will return as soon as a + * core becomes free to accept the first + * packet, as subsequent ones will be added to + * the backlog for that core. + */ + struct rte_mbuf *pkts[RTE_DISTRIB_BACKLOG_SIZE]; + unsigned i; + struct rte_distributor_backlog *bl = &d->backlog[wkr]; + + for (i = 0; i < bl->count; i++) { + unsigned idx = (bl->start + i) & + RTE_DISTRIB_BACKLOG_MASK; + pkts[i] = (void *)((uintptr_t)(bl->pkts[idx] >> + RTE_DISTRIB_FLAG_BITS)); + } + /* recursive call. + * Note that the tags were set before first level call + * to rte_distributor_process. + */ + rte_distributor_process_v20(d, pkts, i); + bl->count = bl->start = 0; + } +} + +/* this function is called when process() fn is called without any new + * packets. It goes through all the workers and clears any returned packets + * to do a partial flush. + */ +static int +process_returns(struct rte_distributor_v20 *d) +{ + unsigned wkr; + unsigned flushed = 0; + unsigned ret_start = d->returns.start, + ret_count = d->returns.count; + + for (wkr = 0; wkr < d->num_workers; wkr++) { + + const int64_t data = d->bufs[wkr].bufptr64; + uintptr_t oldbuf = 0; + + if (data & RTE_DISTRIB_GET_BUF) { + flushed++; + if (d->backlog[wkr].count) + d->bufs[wkr].bufptr64 = + backlog_pop(&d->backlog[wkr]); + else { + d->bufs[wkr].bufptr64 = RTE_DISTRIB_GET_BUF; + d->in_flight_tags[wkr] = 0; + d->in_flight_bitmask &= ~(1UL << wkr); + } + oldbuf = data >> RTE_DISTRIB_FLAG_BITS; + } else if (data & RTE_DISTRIB_RETURN_BUF) { + handle_worker_shutdown(d, wkr); + oldbuf = data >> RTE_DISTRIB_FLAG_BITS; + } + + store_return(oldbuf, d, &ret_start, &ret_count); + } + + d->returns.start = ret_start; + d->returns.count = ret_count; + + return flushed; +} + +/* process a set of packets to distribute them to workers */ +int +rte_distributor_process_v20(struct rte_distributor_v20 *d, + struct rte_mbuf **mbufs, unsigned num_mbufs) +{ + unsigned next_idx = 0; + unsigned wkr = 0; + struct rte_mbuf *next_mb = NULL; + int64_t next_value = 0; + uint32_t new_tag = 0; + unsigned ret_start = d->returns.start, + ret_count = d->returns.count; + + if (unlikely(num_mbufs == 0)) + return process_returns(d); + + while (next_idx < num_mbufs || next_mb != NULL) { + + int64_t data = d->bufs[wkr].bufptr64; + uintptr_t oldbuf = 0; + + if (!next_mb) { + next_mb = mbufs[next_idx++]; + next_value = (((int64_t)(uintptr_t)next_mb) + << RTE_DISTRIB_FLAG_BITS); + /* + * User is advocated to set tag vaue for each + * mbuf before calling rte_distributor_process. + * User defined tags are used to identify flows, + * or sessions. + */ + new_tag = next_mb->hash.usr; + + /* + * Note that if RTE_DISTRIB_MAX_WORKERS is larger than 64 + * then the size of match has to be expanded. + */ + uint64_t match = 0; + unsigned i; + /* + * to scan for a match use "xor" and "not" to get a 0/1 + * value, then use shifting to merge to single "match" + * variable, where a one-bit indicates a match for the + * worker given by the bit-position + */ + for (i = 0; i < d->num_workers; i++) + match |= (!(d->in_flight_tags[i] ^ new_tag) + << i); + + /* Only turned-on bits are considered as match */ + match &= d->in_flight_bitmask; + + if (match) { + next_mb = NULL; + unsigned worker = __builtin_ctzl(match); + if (add_to_backlog(&d->backlog[worker], + next_value) < 0) + next_idx--; + } + } + + if ((data & RTE_DISTRIB_GET_BUF) && + (d->backlog[wkr].count || next_mb)) { + + if (d->backlog[wkr].count) + d->bufs[wkr].bufptr64 = + backlog_pop(&d->backlog[wkr]); + + else { + d->bufs[wkr].bufptr64 = next_value; + d->in_flight_tags[wkr] = new_tag; + d->in_flight_bitmask |= (1UL << wkr); + next_mb = NULL; + } + oldbuf = data >> RTE_DISTRIB_FLAG_BITS; + } else if (data & RTE_DISTRIB_RETURN_BUF) { + handle_worker_shutdown(d, wkr); + oldbuf = data >> RTE_DISTRIB_FLAG_BITS; + } + + /* store returns in a circular buffer */ + store_return(oldbuf, d, &ret_start, &ret_count); + + if (++wkr == d->num_workers) + wkr = 0; + } + /* to finish, check all workers for backlog and schedule work for them + * if they are ready */ + for (wkr = 0; wkr < d->num_workers; wkr++) + if (d->backlog[wkr].count && + (d->bufs[wkr].bufptr64 & RTE_DISTRIB_GET_BUF)) { + + int64_t oldbuf = d->bufs[wkr].bufptr64 >> + RTE_DISTRIB_FLAG_BITS; + store_return(oldbuf, d, &ret_start, &ret_count); + + d->bufs[wkr].bufptr64 = backlog_pop(&d->backlog[wkr]); + } + + d->returns.start = ret_start; + d->returns.count = ret_count; + return num_mbufs; +} +VERSION_SYMBOL(rte_distributor_process, _v20, 2.0); + +/* return to the caller, packets returned from workers */ +int +rte_distributor_returned_pkts_v20(struct rte_distributor_v20 *d, + struct rte_mbuf **mbufs, unsigned max_mbufs) +{ + struct rte_distributor_returned_pkts *returns = &d->returns; + unsigned retval = (max_mbufs < returns->count) ? + max_mbufs : returns->count; + unsigned i; + + for (i = 0; i < retval; i++) { + unsigned idx = (returns->start + i) & RTE_DISTRIB_RETURNS_MASK; + mbufs[i] = returns->mbufs[idx]; + } + returns->start += i; + returns->count -= i; + + return retval; +} +VERSION_SYMBOL(rte_distributor_returned_pkts, _v20, 2.0); + +/* return the number of packets in-flight in a distributor, i.e. packets + * being workered on or queued up in a backlog. */ +static inline unsigned +total_outstanding(const struct rte_distributor_v20 *d) +{ + unsigned wkr, total_outstanding; + + total_outstanding = __builtin_popcountl(d->in_flight_bitmask); + + for (wkr = 0; wkr < d->num_workers; wkr++) + total_outstanding += d->backlog[wkr].count; + + return total_outstanding; +} + +/* flush the distributor, so that there are no outstanding packets in flight or + * queued up. */ +int +rte_distributor_flush_v20(struct rte_distributor_v20 *d) +{ + const unsigned flushed = total_outstanding(d); + + while (total_outstanding(d) > 0) + rte_distributor_process_v20(d, NULL, 0); + + return flushed; +} +VERSION_SYMBOL(rte_distributor_flush, _v20, 2.0); + +/* clears the internal returns array in the distributor */ +void +rte_distributor_clear_returns_v20(struct rte_distributor_v20 *d) +{ + d->returns.start = d->returns.count = 0; +#ifndef __OPTIMIZE__ + memset(d->returns.mbufs, 0, sizeof(d->returns.mbufs)); +#endif +} +VERSION_SYMBOL(rte_distributor_clear_returns, _v20, 2.0); + +/* creates a distributor instance */ +struct rte_distributor_v20 * +rte_distributor_create_v20(const char *name, + unsigned socket_id, + unsigned num_workers) +{ + struct rte_distributor_v20 *d; + struct rte_distributor_list *distributor_list; + char mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *mz; + + /* compilation-time checks */ + RTE_BUILD_BUG_ON((sizeof(*d) & RTE_CACHE_LINE_MASK) != 0); + RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0); + RTE_BUILD_BUG_ON(RTE_DISTRIB_MAX_WORKERS > + sizeof(d->in_flight_bitmask) * CHAR_BIT); + + if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) { + rte_errno = EINVAL; + return NULL; + } + + snprintf(mz_name, sizeof(mz_name), RTE_DISTRIB_PREFIX"%s", name); + mz = rte_memzone_reserve(mz_name, sizeof(*d), socket_id, NO_FLAGS); + if (mz == NULL) { + rte_errno = ENOMEM; + return NULL; + } + + d = mz->addr; + snprintf(d->name, sizeof(d->name), "%s", name); + d->num_workers = num_workers; + + distributor_list = RTE_TAILQ_CAST(rte_distributor_tailq.head, + rte_distributor_list); + + rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK); + TAILQ_INSERT_TAIL(distributor_list, d, next); + rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK); + + return d; +} +VERSION_SYMBOL(rte_distributor_create, _v20, 2.0); |