summaryrefslogtreecommitdiffstats
path: root/lib/librte_distributor/rte_distributor_v20.c
blob: 9566b53f2b39d4f0613fe7aea4819c4e9a1d3f74 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
/* SPDX-License-Identifier: BSD-3-Clause
 * Copyright(c) 2010-2014 Intel Corporation
 */

#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_pause.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 value 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 worked 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);