xref: /titanic_41/usr/src/uts/common/inet/squeue.c (revision 42cac157f878fbb7ae190eb0339c6932f3192b87)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * Squeues: General purpose serialization mechanism
28  * ------------------------------------------------
29  *
30  * Background:
31  * -----------
32  *
33  * This is a general purpose high-performance serialization mechanism
34  * currently used by TCP/IP. It is implement by means of a per CPU queue,
35  * a worker thread and a polling thread with are bound to the CPU
36  * associated with the squeue. The squeue is strictly FIFO for both read
37  * and write side and only one thread can process it at any given time.
38  * The design goal of squeue was to offer a very high degree of
39  * parallelization (on a per H/W execution pipeline basis) with at
40  * most one queuing.
41  *
42  * The modules needing protection typically calls SQUEUE_ENTER_ONE() or
43  * SQUEUE_ENTER() macro as soon as a thread enter the module
44  * from either direction. For each packet, the processing function
45  * and argument is stored in the mblk itself. When the packet is ready
46  * to be processed, the squeue retrieves the stored function and calls
47  * it with the supplied argument and the pointer to the packet itself.
48  * The called function can assume that no other thread is processing
49  * the squeue when it is executing.
50  *
51  * Squeue/connection binding:
52  * --------------------------
53  *
54  * TCP/IP uses an IP classifier in conjunction with squeue where specific
55  * connections are assigned to specific squeue (based on various policies),
56  * at the connection creation time. Once assigned, the connection to
57  * squeue mapping is never changed and all future packets for that
58  * connection are processed on that squeue. The connection ("conn") to
59  * squeue mapping is stored in "conn_t" member "conn_sqp".
60  *
61  * Since the processing of the connection cuts across multiple layers
62  * but still allows packets for different connnection to be processed on
63  * other CPU/squeues, squeues are also termed as "Vertical Perimeter" or
64  * "Per Connection Vertical Perimeter".
65  *
66  * Processing Model:
67  * -----------------
68  *
69  * Squeue doesn't necessary processes packets with its own worker thread.
70  * The callers can pick if they just want to queue the packet, process
71  * their packet if nothing is queued or drain and process. The first two
72  * modes are typically employed when the packet was generated while
73  * already doing the processing behind the squeue and last mode (drain
74  * and process) is typically employed when the thread is entering squeue
75  * for the first time. The squeue still imposes a finite time limit
76  * for which a external thread can do processing after which it switches
77  * processing to its own worker thread.
78  *
79  * Once created, squeues are never deleted. Hence squeue pointers are
80  * always valid. This means that functions outside the squeue can still
81  * refer safely to conn_sqp and their is no need for ref counts.
82  *
83  * Only a thread executing in the squeue can change the squeue of the
84  * connection. It does so by calling a squeue framework function to do this.
85  * After changing the squeue, the thread must leave the squeue. It must not
86  * continue to execute any code that needs squeue protection.
87  *
88  * The squeue framework, after entering the squeue, checks if the current
89  * squeue matches the conn_sqp. If the check fails, the packet is delivered
90  * to right squeue.
91  *
92  * Polling Model:
93  * --------------
94  *
95  * Squeues can control the rate of packet arrival into itself from the
96  * NIC or specific Rx ring within a NIC. As part of capability negotiation
97  * between IP and MAC layer, squeue are created for each TCP soft ring
98  * (or TCP Rx ring - to be implemented in future). As part of this
99  * negotiation, squeues get a cookie for underlying soft ring or Rx
100  * ring, a function to turn off incoming packets and a function to call
101  * to poll for packets. This helps schedule the receive side packet
102  * processing so that queue backlog doesn't build up and packet processing
103  * doesn't keep getting disturbed by high priority interrupts. As part
104  * of this mode, as soon as a backlog starts building, squeue turns off
105  * the interrupts and switches to poll mode. In poll mode, when poll
106  * thread goes down to retrieve packets, it retrieves them in the form of
107  * a chain which improves performance even more. As the squeue/softring
108  * system gets more packets, it gets more efficient by switching to
109  * polling more often and dealing with larger packet chains.
110  *
111  */
112 
113 #include <sys/types.h>
114 #include <sys/cmn_err.h>
115 #include <sys/debug.h>
116 #include <sys/kmem.h>
117 #include <sys/cpuvar.h>
118 #include <sys/condvar_impl.h>
119 #include <sys/systm.h>
120 #include <sys/callb.h>
121 #include <sys/sdt.h>
122 #include <sys/ddi.h>
123 #include <sys/sunddi.h>
124 
125 #include <inet/ipclassifier.h>
126 #include <inet/udp_impl.h>
127 
128 #include <sys/squeue_impl.h>
129 
130 static void squeue_fire(void *);
131 static void squeue_drain(squeue_t *, uint_t, hrtime_t);
132 static void squeue_worker(squeue_t *sqp);
133 static void squeue_polling_thread(squeue_t *sqp);
134 
135 kmem_cache_t *squeue_cache;
136 
137 #define	SQUEUE_MSEC_TO_NSEC 1000000
138 
139 int squeue_drain_ms = 20;
140 int squeue_workerwait_ms = 0;
141 
142 /* The values above converted to ticks or nano seconds */
143 static int squeue_drain_ns = 0;
144 static int squeue_workerwait_tick = 0;
145 
146 #define	MAX_BYTES_TO_PICKUP	150000
147 
148 #define	ENQUEUE_CHAIN(sqp, mp, tail, cnt) {			\
149 	/*							\
150 	 * Enqueue our mblk chain.				\
151 	 */							\
152 	ASSERT(MUTEX_HELD(&(sqp)->sq_lock));			\
153 								\
154 	if ((sqp)->sq_last != NULL)				\
155 		(sqp)->sq_last->b_next = (mp);			\
156 	else							\
157 		(sqp)->sq_first = (mp);				\
158 	(sqp)->sq_last = (tail);				\
159 	(sqp)->sq_count += (cnt);				\
160 	ASSERT((sqp)->sq_count > 0);				\
161 	DTRACE_PROBE4(squeue__enqueuechain, squeue_t *, sqp,	\
162 		mblk_t *, mp, mblk_t *, tail, int, cnt);	\
163 								\
164 }
165 
166 /*
167  * Blank the receive ring (in this case it is the soft ring). When
168  * blanked, the soft ring will not send any more packets up.
169  * Blanking may not succeed when there is a CPU already in the soft
170  * ring sending packets up. In that case, SQS_POLLING will not be
171  * set.
172  */
173 #define	SQS_POLLING_ON(sqp, sq_poll_capable, rx_ring) {		\
174 	ASSERT(MUTEX_HELD(&(sqp)->sq_lock));			\
175 	if (sq_poll_capable) {					\
176 		ASSERT(rx_ring != NULL);			\
177 		ASSERT(sqp->sq_state & SQS_POLL_CAPAB);		\
178 		if (!(sqp->sq_state & SQS_POLLING)) {		\
179 			if (rx_ring->rr_intr_disable(rx_ring->rr_intr_handle)) \
180 				sqp->sq_state |= SQS_POLLING;	\
181 		}						\
182 	}							\
183 }
184 
185 #define	SQS_POLLING_OFF(sqp, sq_poll_capable, rx_ring) {	\
186 	ASSERT(MUTEX_HELD(&(sqp)->sq_lock));			\
187 	if (sq_poll_capable) {					\
188 		ASSERT(rx_ring != NULL);			\
189 		ASSERT(sqp->sq_state & SQS_POLL_CAPAB);		\
190 		if (sqp->sq_state & SQS_POLLING) {		\
191 			sqp->sq_state &= ~SQS_POLLING;		\
192 			rx_ring->rr_intr_enable(rx_ring->rr_intr_handle); \
193 		}						\
194 	}							\
195 }
196 
197 /* Wakeup poll thread only if SQS_POLLING is set */
198 #define	SQS_POLL_RING(sqp) {			\
199 	ASSERT(MUTEX_HELD(&(sqp)->sq_lock));			\
200 	if (sqp->sq_state & SQS_POLLING) {			\
201 		ASSERT(sqp->sq_state & SQS_POLL_CAPAB);		\
202 		if (!(sqp->sq_state & SQS_GET_PKTS)) {		\
203 			sqp->sq_state |= SQS_GET_PKTS;		\
204 			cv_signal(&sqp->sq_poll_cv);		\
205 		}						\
206 	}							\
207 }
208 
209 #ifdef DEBUG
210 #define	SQUEUE_DBG_SET(sqp, mp, proc, connp, tag) {		\
211 	(sqp)->sq_curmp = (mp);					\
212 	(sqp)->sq_curproc = (proc);				\
213 	(sqp)->sq_connp = (connp);				\
214 	(mp)->b_tag = (sqp)->sq_tag = (tag);			\
215 }
216 
217 #define	SQUEUE_DBG_CLEAR(sqp)	{				\
218 	(sqp)->sq_curmp = NULL;					\
219 	(sqp)->sq_curproc = NULL;				\
220 	(sqp)->sq_connp = NULL;					\
221 }
222 #else
223 #define	SQUEUE_DBG_SET(sqp, mp, proc, connp, tag)
224 #define	SQUEUE_DBG_CLEAR(sqp)
225 #endif
226 
227 void
228 squeue_init(void)
229 {
230 	squeue_cache = kmem_cache_create("squeue_cache",
231 	    sizeof (squeue_t), 64, NULL, NULL, NULL, NULL, NULL, 0);
232 
233 	squeue_drain_ns = squeue_drain_ms * SQUEUE_MSEC_TO_NSEC;
234 	squeue_workerwait_tick = MSEC_TO_TICK_ROUNDUP(squeue_workerwait_ms);
235 }
236 
237 /* ARGSUSED */
238 squeue_t *
239 squeue_create(clock_t wait, pri_t pri)
240 {
241 	squeue_t *sqp = kmem_cache_alloc(squeue_cache, KM_SLEEP);
242 
243 	bzero(sqp, sizeof (squeue_t));
244 	sqp->sq_bind = PBIND_NONE;
245 	sqp->sq_priority = pri;
246 	sqp->sq_wait = MSEC_TO_TICK(wait);
247 	sqp->sq_worker = thread_create(NULL, 0, squeue_worker,
248 	    sqp, 0, &p0, TS_RUN, pri);
249 
250 	sqp->sq_poll_thr = thread_create(NULL, 0, squeue_polling_thread,
251 	    sqp, 0, &p0, TS_RUN, pri);
252 
253 	sqp->sq_enter = squeue_enter;
254 	sqp->sq_drain = squeue_drain;
255 
256 	return (sqp);
257 }
258 
259 /*
260  * Bind squeue worker thread to the specified CPU, given by CPU id.
261  * If the CPU id  value is -1, bind the worker thread to the value
262  * specified in sq_bind field. If a thread is already bound to a
263  * different CPU, unbind it from the old CPU and bind to the new one.
264  */
265 
266 void
267 squeue_bind(squeue_t *sqp, processorid_t bind)
268 {
269 	mutex_enter(&sqp->sq_lock);
270 	ASSERT(sqp->sq_bind != PBIND_NONE || bind != PBIND_NONE);
271 	ASSERT(MUTEX_HELD(&cpu_lock));
272 
273 	if (sqp->sq_state & SQS_BOUND) {
274 		if (sqp->sq_bind == bind) {
275 			mutex_exit(&sqp->sq_lock);
276 			return;
277 		}
278 		thread_affinity_clear(sqp->sq_worker);
279 	} else {
280 		sqp->sq_state |= SQS_BOUND;
281 	}
282 
283 	if (bind != PBIND_NONE)
284 		sqp->sq_bind = bind;
285 
286 	thread_affinity_set(sqp->sq_worker, sqp->sq_bind);
287 	mutex_exit(&sqp->sq_lock);
288 }
289 
290 void
291 squeue_unbind(squeue_t *sqp)
292 {
293 	mutex_enter(&sqp->sq_lock);
294 	if (!(sqp->sq_state & SQS_BOUND)) {
295 		mutex_exit(&sqp->sq_lock);
296 		return;
297 	}
298 
299 	sqp->sq_state &= ~SQS_BOUND;
300 	thread_affinity_clear(sqp->sq_worker);
301 	mutex_exit(&sqp->sq_lock);
302 }
303 
304 void
305 squeue_worker_wakeup(squeue_t *sqp)
306 {
307 	timeout_id_t tid = (sqp)->sq_tid;
308 
309 	ASSERT(MUTEX_HELD(&(sqp)->sq_lock));
310 
311 	if (sqp->sq_wait == 0) {
312 		ASSERT(tid == 0);
313 		ASSERT(!(sqp->sq_state & SQS_TMO_PROG));
314 		sqp->sq_awaken = ddi_get_lbolt();
315 		cv_signal(&sqp->sq_worker_cv);
316 		mutex_exit(&sqp->sq_lock);
317 		return;
318 	}
319 
320 	/*
321 	 * Queue isn't being processed, so take
322 	 * any post enqueue actions needed before leaving.
323 	 */
324 	if (tid != 0) {
325 		/*
326 		 * Waiting for an enter() to process mblk(s).
327 		 */
328 		clock_t now = ddi_get_lbolt();
329 		clock_t	waited = now - sqp->sq_awaken;
330 
331 		if (TICK_TO_MSEC(waited) >= sqp->sq_wait) {
332 			/*
333 			 * Times up and have a worker thread
334 			 * waiting for work, so schedule it.
335 			 */
336 			sqp->sq_tid = 0;
337 			sqp->sq_awaken = now;
338 			cv_signal(&sqp->sq_worker_cv);
339 			mutex_exit(&sqp->sq_lock);
340 			(void) untimeout(tid);
341 			return;
342 		}
343 		mutex_exit(&sqp->sq_lock);
344 		return;
345 	} else if (sqp->sq_state & SQS_TMO_PROG) {
346 		mutex_exit(&sqp->sq_lock);
347 		return;
348 	} else {
349 		clock_t	wait = sqp->sq_wait;
350 		/*
351 		 * Wait up to sqp->sq_wait ms for an
352 		 * enter() to process this queue. We
353 		 * don't want to contend on timeout locks
354 		 * with sq_lock held for performance reasons,
355 		 * so drop the sq_lock before calling timeout
356 		 * but we need to check if timeout is required
357 		 * after re acquiring the sq_lock. Once
358 		 * the sq_lock is dropped, someone else could
359 		 * have processed the packet or the timeout could
360 		 * have already fired.
361 		 */
362 		sqp->sq_state |= SQS_TMO_PROG;
363 		mutex_exit(&sqp->sq_lock);
364 		tid = timeout(squeue_fire, sqp, wait);
365 		mutex_enter(&sqp->sq_lock);
366 		/* Check again if we still need the timeout */
367 		if (((sqp->sq_state & (SQS_PROC|SQS_TMO_PROG)) ==
368 		    SQS_TMO_PROG) && (sqp->sq_tid == 0) &&
369 		    (sqp->sq_first != NULL)) {
370 				sqp->sq_state &= ~SQS_TMO_PROG;
371 				sqp->sq_tid = tid;
372 				mutex_exit(&sqp->sq_lock);
373 				return;
374 		} else {
375 			if (sqp->sq_state & SQS_TMO_PROG) {
376 				sqp->sq_state &= ~SQS_TMO_PROG;
377 				mutex_exit(&sqp->sq_lock);
378 				(void) untimeout(tid);
379 			} else {
380 				/*
381 				 * The timer fired before we could
382 				 * reacquire the sq_lock. squeue_fire
383 				 * removes the SQS_TMO_PROG flag
384 				 * and we don't need to	do anything
385 				 * else.
386 				 */
387 				mutex_exit(&sqp->sq_lock);
388 			}
389 		}
390 	}
391 
392 	ASSERT(MUTEX_NOT_HELD(&sqp->sq_lock));
393 }
394 
395 /*
396  * squeue_enter() - enter squeue sqp with mblk mp (which can be
397  * a chain), while tail points to the end and cnt in number of
398  * mblks in the chain.
399  *
400  * For a chain of single packet (i.e. mp == tail), go through the
401  * fast path if no one is processing the squeue and nothing is queued.
402  *
403  * The proc and arg for each mblk is already stored in the mblk in
404  * appropriate places.
405  *
406  * The process_flag specifies if we are allowed to process the mblk
407  * and drain in the entering thread context. If process_flag is
408  * SQ_FILL, then we just queue the mblk and return (after signaling
409  * the worker thread if no one else is processing the squeue).
410  *
411  * The ira argument can be used when the count is one.
412  * For a chain the caller needs to prepend any needed mblks from
413  * ip_recv_attr_to_mblk().
414  */
415 /* ARGSUSED */
416 void
417 squeue_enter(squeue_t *sqp, mblk_t *mp, mblk_t *tail, uint32_t cnt,
418     ip_recv_attr_t *ira, int process_flag, uint8_t tag)
419 {
420 	conn_t		*connp;
421 	sqproc_t	proc;
422 	hrtime_t	now;
423 
424 	ASSERT(sqp != NULL);
425 	ASSERT(mp != NULL);
426 	ASSERT(tail != NULL);
427 	ASSERT(cnt > 0);
428 	ASSERT(MUTEX_NOT_HELD(&sqp->sq_lock));
429 	ASSERT(ira == NULL || cnt == 1);
430 
431 	mutex_enter(&sqp->sq_lock);
432 
433 	/*
434 	 * Try to process the packet if SQ_FILL flag is not set and
435 	 * we are allowed to process the squeue. The SQ_NODRAIN is
436 	 * ignored if the packet chain consists of more than 1 packet.
437 	 */
438 	if (!(sqp->sq_state & SQS_PROC) && ((process_flag == SQ_PROCESS) ||
439 	    (process_flag == SQ_NODRAIN && sqp->sq_first == NULL))) {
440 		/*
441 		 * See if anything is already queued. If we are the
442 		 * first packet, do inline processing else queue the
443 		 * packet and do the drain.
444 		 */
445 		if (sqp->sq_first == NULL && cnt == 1) {
446 			/*
447 			 * Fast-path, ok to process and nothing queued.
448 			 */
449 			sqp->sq_state |= (SQS_PROC|SQS_FAST);
450 			sqp->sq_run = curthread;
451 			mutex_exit(&sqp->sq_lock);
452 
453 			/*
454 			 * We are the chain of 1 packet so
455 			 * go through this fast path.
456 			 */
457 			ASSERT(mp->b_prev != NULL);
458 			ASSERT(mp->b_queue != NULL);
459 			connp = (conn_t *)mp->b_prev;
460 			mp->b_prev = NULL;
461 			proc = (sqproc_t)mp->b_queue;
462 			mp->b_queue = NULL;
463 			ASSERT(proc != NULL && connp != NULL);
464 			ASSERT(mp->b_next == NULL);
465 
466 			/*
467 			 * Handle squeue switching. More details in the
468 			 * block comment at the top of the file
469 			 */
470 			if (connp->conn_sqp == sqp) {
471 				SQUEUE_DBG_SET(sqp, mp, proc, connp,
472 				    tag);
473 				connp->conn_on_sqp = B_TRUE;
474 				DTRACE_PROBE3(squeue__proc__start, squeue_t *,
475 				    sqp, mblk_t *, mp, conn_t *, connp);
476 				(*proc)(connp, mp, sqp, ira);
477 				DTRACE_PROBE2(squeue__proc__end, squeue_t *,
478 				    sqp, conn_t *, connp);
479 				connp->conn_on_sqp = B_FALSE;
480 				SQUEUE_DBG_CLEAR(sqp);
481 				CONN_DEC_REF(connp);
482 			} else {
483 				SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc,
484 				    connp, ira, SQ_FILL, SQTAG_SQUEUE_CHANGE);
485 			}
486 			ASSERT(MUTEX_NOT_HELD(&sqp->sq_lock));
487 			mutex_enter(&sqp->sq_lock);
488 			sqp->sq_state &= ~(SQS_PROC|SQS_FAST);
489 			sqp->sq_run = NULL;
490 			if (sqp->sq_first == NULL ||
491 			    process_flag == SQ_NODRAIN) {
492 				if (sqp->sq_first != NULL) {
493 					squeue_worker_wakeup(sqp);
494 					return;
495 				}
496 				/*
497 				 * We processed inline our packet and nothing
498 				 * new has arrived. We are done. In case any
499 				 * control actions are pending, wake up the
500 				 * worker.
501 				 */
502 				if (sqp->sq_state & SQS_WORKER_THR_CONTROL)
503 					cv_signal(&sqp->sq_worker_cv);
504 				mutex_exit(&sqp->sq_lock);
505 				return;
506 			}
507 		} else {
508 			if (ira != NULL) {
509 				mblk_t	*attrmp;
510 
511 				ASSERT(cnt == 1);
512 				attrmp = ip_recv_attr_to_mblk(ira);
513 				if (attrmp == NULL) {
514 					mutex_exit(&sqp->sq_lock);
515 					ip_drop_input("squeue: "
516 					    "ip_recv_attr_to_mblk",
517 					    mp, NULL);
518 					/* Caller already set b_prev/b_next */
519 					mp->b_prev = mp->b_next = NULL;
520 					freemsg(mp);
521 					return;
522 				}
523 				ASSERT(attrmp->b_cont == NULL);
524 				attrmp->b_cont = mp;
525 				/* Move connp and func to new */
526 				attrmp->b_queue = mp->b_queue;
527 				mp->b_queue = NULL;
528 				attrmp->b_prev = mp->b_prev;
529 				mp->b_prev = NULL;
530 
531 				ASSERT(mp == tail);
532 				tail = mp = attrmp;
533 			}
534 
535 			ENQUEUE_CHAIN(sqp, mp, tail, cnt);
536 #ifdef DEBUG
537 			mp->b_tag = tag;
538 #endif
539 		}
540 		/*
541 		 * We are here because either we couldn't do inline
542 		 * processing (because something was already queued),
543 		 * or we had a chain of more than one packet,
544 		 * or something else arrived after we were done with
545 		 * inline processing.
546 		 */
547 		ASSERT(MUTEX_HELD(&sqp->sq_lock));
548 		ASSERT(sqp->sq_first != NULL);
549 		now = gethrtime();
550 		sqp->sq_drain(sqp, SQS_ENTER, now + squeue_drain_ns);
551 
552 		/*
553 		 * If we didn't do a complete drain, the worker
554 		 * thread was already signalled by squeue_drain.
555 		 * In case any control actions are pending, wake
556 		 * up the worker.
557 		 */
558 		sqp->sq_run = NULL;
559 		if (sqp->sq_state & SQS_WORKER_THR_CONTROL)
560 			cv_signal(&sqp->sq_worker_cv);
561 		mutex_exit(&sqp->sq_lock);
562 		return;
563 	} else {
564 		/*
565 		 * We let a thread processing a squeue reenter only
566 		 * once. This helps the case of incoming connection
567 		 * where a SYN-ACK-ACK that triggers the conn_ind
568 		 * doesn't have to queue the packet if listener and
569 		 * eager are on the same squeue. Also helps the
570 		 * loopback connection where the two ends are bound
571 		 * to the same squeue (which is typical on single
572 		 * CPU machines).
573 		 *
574 		 * We let the thread reenter only once for the fear
575 		 * of stack getting blown with multiple traversal.
576 		 */
577 		connp = (conn_t *)mp->b_prev;
578 		if (!(sqp->sq_state & SQS_REENTER) &&
579 		    (process_flag != SQ_FILL) && (sqp->sq_first == NULL) &&
580 		    (sqp->sq_run == curthread) && (cnt == 1) &&
581 		    (connp->conn_on_sqp == B_FALSE)) {
582 			sqp->sq_state |= SQS_REENTER;
583 			mutex_exit(&sqp->sq_lock);
584 
585 			ASSERT(mp->b_prev != NULL);
586 			ASSERT(mp->b_queue != NULL);
587 
588 			mp->b_prev = NULL;
589 			proc = (sqproc_t)mp->b_queue;
590 			mp->b_queue = NULL;
591 
592 			/*
593 			 * Handle squeue switching. More details in the
594 			 * block comment at the top of the file
595 			 */
596 			if (connp->conn_sqp == sqp) {
597 				connp->conn_on_sqp = B_TRUE;
598 				DTRACE_PROBE3(squeue__proc__start, squeue_t *,
599 				    sqp, mblk_t *, mp, conn_t *, connp);
600 				(*proc)(connp, mp, sqp, ira);
601 				DTRACE_PROBE2(squeue__proc__end, squeue_t *,
602 				    sqp, conn_t *, connp);
603 				connp->conn_on_sqp = B_FALSE;
604 				CONN_DEC_REF(connp);
605 			} else {
606 				SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc,
607 				    connp, ira, SQ_FILL, SQTAG_SQUEUE_CHANGE);
608 			}
609 
610 			mutex_enter(&sqp->sq_lock);
611 			sqp->sq_state &= ~SQS_REENTER;
612 			mutex_exit(&sqp->sq_lock);
613 			return;
614 		}
615 
616 		/*
617 		 * Queue is already being processed or there is already
618 		 * one or more paquets on the queue. Enqueue the
619 		 * packet and wakeup the squeue worker thread if the
620 		 * squeue is not being processed.
621 		 */
622 #ifdef DEBUG
623 		mp->b_tag = tag;
624 #endif
625 		if (ira != NULL) {
626 			mblk_t	*attrmp;
627 
628 			ASSERT(cnt == 1);
629 			attrmp = ip_recv_attr_to_mblk(ira);
630 			if (attrmp == NULL) {
631 				mutex_exit(&sqp->sq_lock);
632 				ip_drop_input("squeue: ip_recv_attr_to_mblk",
633 				    mp, NULL);
634 				/* Caller already set b_prev/b_next */
635 				mp->b_prev = mp->b_next = NULL;
636 				freemsg(mp);
637 				return;
638 			}
639 			ASSERT(attrmp->b_cont == NULL);
640 			attrmp->b_cont = mp;
641 			/* Move connp and func to new */
642 			attrmp->b_queue = mp->b_queue;
643 			mp->b_queue = NULL;
644 			attrmp->b_prev = mp->b_prev;
645 			mp->b_prev = NULL;
646 
647 			ASSERT(mp == tail);
648 			tail = mp = attrmp;
649 		}
650 		ENQUEUE_CHAIN(sqp, mp, tail, cnt);
651 		if (!(sqp->sq_state & SQS_PROC)) {
652 			squeue_worker_wakeup(sqp);
653 			return;
654 		}
655 		/*
656 		 * In case any control actions are pending, wake
657 		 * up the worker.
658 		 */
659 		if (sqp->sq_state & SQS_WORKER_THR_CONTROL)
660 			cv_signal(&sqp->sq_worker_cv);
661 		mutex_exit(&sqp->sq_lock);
662 		return;
663 	}
664 }
665 
666 /*
667  * PRIVATE FUNCTIONS
668  */
669 
670 static void
671 squeue_fire(void *arg)
672 {
673 	squeue_t	*sqp = arg;
674 	uint_t		state;
675 
676 	mutex_enter(&sqp->sq_lock);
677 
678 	state = sqp->sq_state;
679 	if (sqp->sq_tid == 0 && !(state & SQS_TMO_PROG)) {
680 		mutex_exit(&sqp->sq_lock);
681 		return;
682 	}
683 
684 	sqp->sq_tid = 0;
685 	/*
686 	 * The timeout fired before we got a chance to set it.
687 	 * Process it anyway but remove the SQS_TMO_PROG so that
688 	 * the guy trying to set the timeout knows that it has
689 	 * already been processed.
690 	 */
691 	if (state & SQS_TMO_PROG)
692 		sqp->sq_state &= ~SQS_TMO_PROG;
693 
694 	if (!(state & SQS_PROC)) {
695 		sqp->sq_awaken = ddi_get_lbolt();
696 		cv_signal(&sqp->sq_worker_cv);
697 	}
698 	mutex_exit(&sqp->sq_lock);
699 }
700 
701 static void
702 squeue_drain(squeue_t *sqp, uint_t proc_type, hrtime_t expire)
703 {
704 	mblk_t		*mp;
705 	mblk_t 		*head;
706 	sqproc_t 	proc;
707 	conn_t		*connp;
708 	timeout_id_t 	tid;
709 	ill_rx_ring_t	*sq_rx_ring = sqp->sq_rx_ring;
710 	hrtime_t 	now;
711 	boolean_t	did_wakeup = B_FALSE;
712 	boolean_t	sq_poll_capable;
713 	ip_recv_attr_t	*ira, iras;
714 
715 	sq_poll_capable = (sqp->sq_state & SQS_POLL_CAPAB) != 0;
716 again:
717 	ASSERT(mutex_owned(&sqp->sq_lock));
718 	ASSERT(!(sqp->sq_state & (SQS_POLL_THR_QUIESCED |
719 	    SQS_POLL_QUIESCE_DONE)));
720 
721 	head = sqp->sq_first;
722 	sqp->sq_first = NULL;
723 	sqp->sq_last = NULL;
724 	sqp->sq_count = 0;
725 
726 	if ((tid = sqp->sq_tid) != 0)
727 		sqp->sq_tid = 0;
728 
729 	sqp->sq_state |= SQS_PROC | proc_type;
730 
731 	/*
732 	 * We have backlog built up. Switch to polling mode if the
733 	 * device underneath allows it. Need to do it so that
734 	 * more packets don't come in and disturb us (by contending
735 	 * for sq_lock or higher priority thread preempting us).
736 	 *
737 	 * The worker thread is allowed to do active polling while we
738 	 * just disable the interrupts for drain by non worker (kernel
739 	 * or userland) threads so they can peacefully process the
740 	 * packets during time allocated to them.
741 	 */
742 	SQS_POLLING_ON(sqp, sq_poll_capable, sq_rx_ring);
743 	mutex_exit(&sqp->sq_lock);
744 
745 	if (tid != 0)
746 		(void) untimeout(tid);
747 
748 	while ((mp = head) != NULL) {
749 
750 		head = mp->b_next;
751 		mp->b_next = NULL;
752 
753 		proc = (sqproc_t)mp->b_queue;
754 		mp->b_queue = NULL;
755 		connp = (conn_t *)mp->b_prev;
756 		mp->b_prev = NULL;
757 
758 		/* Is there an ip_recv_attr_t to handle? */
759 		if (ip_recv_attr_is_mblk(mp)) {
760 			mblk_t	*attrmp = mp;
761 
762 			ASSERT(attrmp->b_cont != NULL);
763 
764 			mp = attrmp->b_cont;
765 			attrmp->b_cont = NULL;
766 			ASSERT(mp->b_queue == NULL);
767 			ASSERT(mp->b_prev == NULL);
768 
769 			if (!ip_recv_attr_from_mblk(attrmp, &iras)) {
770 				/* The ill or ip_stack_t disappeared on us */
771 				ip_drop_input("ip_recv_attr_from_mblk",
772 				    mp, NULL);
773 				ira_cleanup(&iras, B_TRUE);
774 				CONN_DEC_REF(connp);
775 				continue;
776 			}
777 			ira = &iras;
778 		} else {
779 			ira = NULL;
780 		}
781 
782 
783 		/*
784 		 * Handle squeue switching. More details in the
785 		 * block comment at the top of the file
786 		 */
787 		if (connp->conn_sqp == sqp) {
788 			SQUEUE_DBG_SET(sqp, mp, proc, connp,
789 			    mp->b_tag);
790 			connp->conn_on_sqp = B_TRUE;
791 			DTRACE_PROBE3(squeue__proc__start, squeue_t *,
792 			    sqp, mblk_t *, mp, conn_t *, connp);
793 			(*proc)(connp, mp, sqp, ira);
794 			DTRACE_PROBE2(squeue__proc__end, squeue_t *,
795 			    sqp, conn_t *, connp);
796 			connp->conn_on_sqp = B_FALSE;
797 			CONN_DEC_REF(connp);
798 		} else {
799 			SQUEUE_ENTER_ONE(connp->conn_sqp, mp, proc, connp, ira,
800 			    SQ_FILL, SQTAG_SQUEUE_CHANGE);
801 		}
802 		if (ira != NULL)
803 			ira_cleanup(ira, B_TRUE);
804 	}
805 
806 	SQUEUE_DBG_CLEAR(sqp);
807 
808 	mutex_enter(&sqp->sq_lock);
809 
810 	/*
811 	 * Check if there is still work to do (either more arrived or timer
812 	 * expired). If we are the worker thread and we are polling capable,
813 	 * continue doing the work since no one else is around to do the
814 	 * work anyway (but signal the poll thread to retrieve some packets
815 	 * in the meanwhile). If we are not the worker thread, just
816 	 * signal the worker thread to take up the work if processing time
817 	 * has expired.
818 	 */
819 	if (sqp->sq_first != NULL) {
820 		/*
821 		 * Still more to process. If time quanta not expired, we
822 		 * should let the drain go on. The worker thread is allowed
823 		 * to drain as long as there is anything left.
824 		 */
825 		now = gethrtime();
826 		if ((now < expire) || (proc_type == SQS_WORKER)) {
827 			/*
828 			 * If time not expired or we are worker thread and
829 			 * this squeue is polling capable, continue to do
830 			 * the drain.
831 			 *
832 			 * We turn off interrupts for all userland threads
833 			 * doing drain but we do active polling only for
834 			 * worker thread.
835 			 *
836 			 * Calling SQS_POLL_RING() even in the case of
837 			 * SQS_POLLING_ON() not succeeding is ok as
838 			 * SQS_POLL_RING() will not wake up poll thread
839 			 * if SQS_POLLING bit is not set.
840 			 */
841 			if (proc_type == SQS_WORKER)
842 				SQS_POLL_RING(sqp);
843 			goto again;
844 		} else {
845 			did_wakeup = B_TRUE;
846 			sqp->sq_awaken = ddi_get_lbolt();
847 			cv_signal(&sqp->sq_worker_cv);
848 		}
849 	}
850 
851 	/*
852 	 * If the poll thread is already running, just return. The
853 	 * poll thread continues to hold the proc and will finish
854 	 * processing.
855 	 */
856 	if (sqp->sq_state & SQS_GET_PKTS) {
857 		ASSERT(!(sqp->sq_state & (SQS_POLL_THR_QUIESCED |
858 		    SQS_POLL_QUIESCE_DONE)));
859 		sqp->sq_state &= ~proc_type;
860 		return;
861 	}
862 
863 	/*
864 	 *
865 	 * If we are the worker thread and no work is left, send the poll
866 	 * thread down once more to see if something arrived. Otherwise,
867 	 * turn the interrupts back on and we are done.
868 	 */
869 	if ((proc_type == SQS_WORKER) && (sqp->sq_state & SQS_POLLING)) {
870 		/*
871 		 * Do one last check to see if anything arrived
872 		 * in the NIC. We leave the SQS_PROC set to ensure
873 		 * that poll thread keeps the PROC and can decide
874 		 * if it needs to turn polling off or continue
875 		 * processing.
876 		 *
877 		 * If we drop the SQS_PROC here and poll thread comes
878 		 * up empty handed, it can not safely turn polling off
879 		 * since someone else could have acquired the PROC
880 		 * and started draining. The previously running poll
881 		 * thread and the current thread doing drain would end
882 		 * up in a race for turning polling on/off and more
883 		 * complex code would be required to deal with it.
884 		 *
885 		 * Its lot simpler for drain to hand the SQS_PROC to
886 		 * poll thread (if running) and let poll thread finish
887 		 * without worrying about racing with any other thread.
888 		 */
889 		ASSERT(!(sqp->sq_state & (SQS_POLL_THR_QUIESCED |
890 		    SQS_POLL_QUIESCE_DONE)));
891 		SQS_POLL_RING(sqp);
892 		sqp->sq_state &= ~proc_type;
893 	} else {
894 		/*
895 		 * The squeue is either not capable of polling or the
896 		 * attempt to blank (i.e., turn SQS_POLLING_ON()) was
897 		 * unsuccessful or poll thread already finished
898 		 * processing and didn't find anything. Since there
899 		 * is nothing queued and we already turn polling on
900 		 * (for all threads doing drain), we should turn
901 		 * polling off and relinquish the PROC.
902 		 */
903 		ASSERT(!(sqp->sq_state & (SQS_POLL_THR_QUIESCED |
904 		    SQS_POLL_QUIESCE_DONE)));
905 		SQS_POLLING_OFF(sqp, sq_poll_capable, sq_rx_ring);
906 		sqp->sq_state &= ~(SQS_PROC | proc_type);
907 		if (!did_wakeup && sqp->sq_first != NULL) {
908 			squeue_worker_wakeup(sqp);
909 			mutex_enter(&sqp->sq_lock);
910 		}
911 		/*
912 		 * If we are not the worker and there is a pending quiesce
913 		 * event, wake up the worker
914 		 */
915 		if ((proc_type != SQS_WORKER) &&
916 		    (sqp->sq_state & SQS_WORKER_THR_CONTROL))
917 			cv_signal(&sqp->sq_worker_cv);
918 	}
919 }
920 
921 /*
922  * Quiesce, Restart, or Cleanup of the squeue poll thread.
923  *
924  * Quiesce and Restart: After an squeue poll thread has been quiesced, it does
925  * not attempt to poll the underlying soft ring any more. The quiesce is
926  * triggered by the mac layer when it wants to quiesce a soft ring. Typically
927  * control operations such as changing the fanout of a NIC or VNIC (dladm
928  * setlinkprop) need to quiesce data flow before changing the wiring.
929  * The operation is done by the mac layer, but it calls back into IP to
930  * quiesce the soft ring. After completing the operation (say increase or
931  * decrease of the fanout) the mac layer then calls back into IP to restart
932  * the quiesced soft ring.
933  *
934  * Cleanup: This is triggered when the squeue binding to a soft ring is
935  * removed permanently. Typically interface plumb and unplumb would trigger
936  * this. It can also be triggered from the mac layer when a soft ring is
937  * being deleted say as the result of a fanout reduction. Since squeues are
938  * never deleted, the cleanup marks the squeue as fit for recycling and
939  * moves it to the zeroth squeue set.
940  */
941 static void
942 squeue_poll_thr_control(squeue_t *sqp)
943 {
944 	if (sqp->sq_state & SQS_POLL_THR_RESTART) {
945 		/* Restart implies a previous quiesce */
946 		ASSERT(sqp->sq_state & SQS_POLL_THR_QUIESCED);
947 		sqp->sq_state &= ~(SQS_POLL_THR_QUIESCED |
948 		    SQS_POLL_THR_RESTART);
949 		sqp->sq_state |= SQS_POLL_CAPAB;
950 		cv_signal(&sqp->sq_worker_cv);
951 		return;
952 	}
953 
954 	if (sqp->sq_state & SQS_POLL_THR_QUIESCE) {
955 		sqp->sq_state |= SQS_POLL_THR_QUIESCED;
956 		sqp->sq_state &= ~SQS_POLL_THR_QUIESCE;
957 		cv_signal(&sqp->sq_worker_cv);
958 		return;
959 	}
960 }
961 
962 /*
963  * POLLING Notes
964  *
965  * With polling mode, we want to do as much processing as we possibly can
966  * in worker thread context. The sweet spot is worker thread keeps doing
967  * work all the time in polling mode and writers etc. keep dumping packets
968  * to worker thread. Occassionally, we send the poll thread (running at
969  * lower priority to NIC to get the chain of packets to feed to worker).
970  * Sending the poll thread down to NIC is dependant on 3 criterions
971  *
972  * 1) Its always driven from squeue_drain and only if worker thread is
973  *	doing the drain.
974  * 2) We clear the backlog once and more packets arrived in between.
975  *	Before starting drain again, send the poll thread down if
976  *	the drain is being done by worker thread.
977  * 3) Before exiting the squeue_drain, if the poll thread is not already
978  *	working and we are the worker thread, try to poll one more time.
979  *
980  * For latency sake, we do allow any thread calling squeue_enter
981  * to process its packet provided:
982  *
983  * 1) Nothing is queued
984  * 2) If more packets arrived in between, the non worker thread are allowed
985  *	to do the drain till their time quanta expired provided SQS_GET_PKTS
986  *	wasn't set in between.
987  *
988  * Avoiding deadlocks with interrupts
989  * ==================================
990  *
991  * One of the big problem is that we can't send poll_thr down while holding
992  * the sq_lock since the thread can block. So we drop the sq_lock before
993  * calling sq_get_pkts(). We keep holding the SQS_PROC as long as the
994  * poll thread is running so that no other thread can acquire the
995  * perimeter in between. If the squeue_drain gets done (no more work
996  * left), it leaves the SQS_PROC set if poll thread is running.
997  */
998 
999 /*
1000  * This is the squeue poll thread. In poll mode, it polls the underlying
1001  * TCP softring and feeds packets into the squeue. The worker thread then
1002  * drains the squeue. The poll thread also responds to control signals for
1003  * quiesceing, restarting, or cleanup of an squeue. These are driven by
1004  * control operations like plumb/unplumb or as a result of dynamic Rx ring
1005  * related operations that are driven from the mac layer.
1006  */
1007 static void
1008 squeue_polling_thread(squeue_t *sqp)
1009 {
1010 	kmutex_t *lock = &sqp->sq_lock;
1011 	kcondvar_t *async = &sqp->sq_poll_cv;
1012 	ip_mac_rx_t sq_get_pkts;
1013 	ip_accept_t ip_accept;
1014 	ill_rx_ring_t *sq_rx_ring;
1015 	ill_t *sq_ill;
1016 	mblk_t *head, *tail, *mp;
1017 	uint_t cnt;
1018 	void *sq_mac_handle;
1019 	callb_cpr_t cprinfo;
1020 	size_t bytes_to_pickup;
1021 	uint32_t ctl_state;
1022 
1023 	CALLB_CPR_INIT(&cprinfo, lock, callb_generic_cpr, "sq_poll");
1024 	mutex_enter(lock);
1025 
1026 	for (;;) {
1027 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1028 		cv_wait(async, lock);
1029 		CALLB_CPR_SAFE_END(&cprinfo, lock);
1030 
1031 		ctl_state = sqp->sq_state & (SQS_POLL_THR_CONTROL |
1032 		    SQS_POLL_THR_QUIESCED);
1033 		if (ctl_state != 0) {
1034 			/*
1035 			 * If the squeue is quiesced, then wait for a control
1036 			 * request. A quiesced squeue must not poll the
1037 			 * underlying soft ring.
1038 			 */
1039 			if (ctl_state == SQS_POLL_THR_QUIESCED)
1040 				continue;
1041 			/*
1042 			 * Act on control requests to quiesce, cleanup or
1043 			 * restart an squeue
1044 			 */
1045 			squeue_poll_thr_control(sqp);
1046 			continue;
1047 		}
1048 
1049 		if (!(sqp->sq_state & SQS_POLL_CAPAB))
1050 			continue;
1051 
1052 		ASSERT((sqp->sq_state &
1053 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS)) ==
1054 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS));
1055 
1056 poll_again:
1057 		sq_rx_ring = sqp->sq_rx_ring;
1058 		sq_get_pkts = sq_rx_ring->rr_rx;
1059 		sq_mac_handle = sq_rx_ring->rr_rx_handle;
1060 		ip_accept = sq_rx_ring->rr_ip_accept;
1061 		sq_ill = sq_rx_ring->rr_ill;
1062 		bytes_to_pickup = MAX_BYTES_TO_PICKUP;
1063 		mutex_exit(lock);
1064 		head = sq_get_pkts(sq_mac_handle, bytes_to_pickup);
1065 		mp = NULL;
1066 		if (head != NULL) {
1067 			/*
1068 			 * We got the packet chain from the mac layer. It
1069 			 * would be nice to be able to process it inline
1070 			 * for better performance but we need to give
1071 			 * IP a chance to look at this chain to ensure
1072 			 * that packets are really meant for this squeue
1073 			 * and do the IP processing.
1074 			 */
1075 			mp = ip_accept(sq_ill, sq_rx_ring, sqp, head,
1076 			    &tail, &cnt);
1077 		}
1078 		mutex_enter(lock);
1079 		if (mp != NULL) {
1080 			/*
1081 			 * The ip_accept function has already added an
1082 			 * ip_recv_attr_t mblk if that is needed.
1083 			 */
1084 			ENQUEUE_CHAIN(sqp, mp, tail, cnt);
1085 		}
1086 		ASSERT((sqp->sq_state &
1087 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS)) ==
1088 		    (SQS_PROC|SQS_POLLING|SQS_GET_PKTS));
1089 
1090 		if (sqp->sq_first != NULL && !(sqp->sq_state & SQS_WORKER)) {
1091 			/*
1092 			 * We have packets to process and worker thread
1093 			 * is not running.  Check to see if poll thread is
1094 			 * allowed to process. Let it do processing only if it
1095 			 * picked up some packets from the NIC otherwise
1096 			 * wakeup the worker thread.
1097 			 */
1098 			if (mp != NULL) {
1099 				hrtime_t  now;
1100 
1101 				now = gethrtime();
1102 				sqp->sq_run = curthread;
1103 				sqp->sq_drain(sqp, SQS_POLL_PROC, now +
1104 				    squeue_drain_ns);
1105 				sqp->sq_run = NULL;
1106 
1107 				if (sqp->sq_first == NULL)
1108 					goto poll_again;
1109 
1110 				/*
1111 				 * Couldn't do the entire drain because the
1112 				 * time limit expired, let the
1113 				 * worker thread take over.
1114 				 */
1115 			}
1116 
1117 			sqp->sq_awaken = ddi_get_lbolt();
1118 			/*
1119 			 * Put the SQS_PROC_HELD on so the worker
1120 			 * thread can distinguish where its called from. We
1121 			 * can remove the SQS_PROC flag here and turn off the
1122 			 * polling so that it wouldn't matter who gets the
1123 			 * processing but we get better performance this way
1124 			 * and save the cost of turn polling off and possibly
1125 			 * on again as soon as we start draining again.
1126 			 *
1127 			 * We can't remove the SQS_PROC flag without turning
1128 			 * polling off until we can guarantee that control
1129 			 * will return to squeue_drain immediately.
1130 			 */
1131 			sqp->sq_state |= SQS_PROC_HELD;
1132 			sqp->sq_state &= ~SQS_GET_PKTS;
1133 			cv_signal(&sqp->sq_worker_cv);
1134 		} else if (sqp->sq_first == NULL &&
1135 		    !(sqp->sq_state & SQS_WORKER)) {
1136 			/*
1137 			 * Nothing queued and worker thread not running.
1138 			 * Since we hold the proc, no other thread is
1139 			 * processing the squeue. This means that there
1140 			 * is no work to be done and nothing is queued
1141 			 * in squeue or in NIC. Turn polling off and go
1142 			 * back to interrupt mode.
1143 			 */
1144 			sqp->sq_state &= ~(SQS_PROC|SQS_GET_PKTS);
1145 			/* LINTED: constant in conditional context */
1146 			SQS_POLLING_OFF(sqp, B_TRUE, sq_rx_ring);
1147 
1148 			/*
1149 			 * If there is a pending control operation
1150 			 * wake up the worker, since it is currently
1151 			 * not running.
1152 			 */
1153 			if (sqp->sq_state & SQS_WORKER_THR_CONTROL)
1154 				cv_signal(&sqp->sq_worker_cv);
1155 		} else {
1156 			/*
1157 			 * Worker thread is already running. We don't need
1158 			 * to do anything. Indicate that poll thread is done.
1159 			 */
1160 			sqp->sq_state &= ~SQS_GET_PKTS;
1161 		}
1162 		if (sqp->sq_state & SQS_POLL_THR_CONTROL) {
1163 			/*
1164 			 * Act on control requests to quiesce, cleanup or
1165 			 * restart an squeue
1166 			 */
1167 			squeue_poll_thr_control(sqp);
1168 		}
1169 	}
1170 }
1171 
1172 /*
1173  * The squeue worker thread acts on any control requests to quiesce, cleanup
1174  * or restart an ill_rx_ring_t by calling this function. The worker thread
1175  * synchronizes with the squeue poll thread to complete the request and finally
1176  * wakes up the requestor when the request is completed.
1177  */
1178 static void
1179 squeue_worker_thr_control(squeue_t *sqp)
1180 {
1181 	ill_t	*ill;
1182 	ill_rx_ring_t	*rx_ring;
1183 
1184 	ASSERT(MUTEX_HELD(&sqp->sq_lock));
1185 
1186 	if (sqp->sq_state & SQS_POLL_RESTART) {
1187 		/* Restart implies a previous quiesce. */
1188 		ASSERT((sqp->sq_state & (SQS_PROC_HELD |
1189 		    SQS_POLL_QUIESCE_DONE | SQS_PROC | SQS_WORKER)) ==
1190 		    (SQS_POLL_QUIESCE_DONE | SQS_PROC | SQS_WORKER));
1191 		/*
1192 		 * Request the squeue poll thread to restart and wait till
1193 		 * it actually restarts.
1194 		 */
1195 		sqp->sq_state &= ~SQS_POLL_QUIESCE_DONE;
1196 		sqp->sq_state |= SQS_POLL_THR_RESTART;
1197 		cv_signal(&sqp->sq_poll_cv);
1198 		while (sqp->sq_state & SQS_POLL_THR_QUIESCED)
1199 			cv_wait(&sqp->sq_worker_cv, &sqp->sq_lock);
1200 		sqp->sq_state &= ~(SQS_POLL_RESTART | SQS_PROC |
1201 		    SQS_WORKER);
1202 		/*
1203 		 * Signal any waiter that is waiting for the restart
1204 		 * to complete
1205 		 */
1206 		sqp->sq_state |= SQS_POLL_RESTART_DONE;
1207 		cv_signal(&sqp->sq_ctrlop_done_cv);
1208 		return;
1209 	}
1210 
1211 	if (sqp->sq_state & SQS_PROC_HELD) {
1212 		/* The squeue poll thread handed control to us */
1213 		ASSERT(sqp->sq_state & SQS_PROC);
1214 	}
1215 
1216 	/*
1217 	 * Prevent any other thread from processing the squeue
1218 	 * until we finish the control actions by setting SQS_PROC.
1219 	 * But allow ourself to reenter by setting SQS_WORKER
1220 	 */
1221 	sqp->sq_state |= (SQS_PROC | SQS_WORKER);
1222 
1223 	/* Signal the squeue poll thread and wait for it to quiesce itself */
1224 	if (!(sqp->sq_state & SQS_POLL_THR_QUIESCED)) {
1225 		sqp->sq_state |= SQS_POLL_THR_QUIESCE;
1226 		cv_signal(&sqp->sq_poll_cv);
1227 		while (!(sqp->sq_state & SQS_POLL_THR_QUIESCED))
1228 			cv_wait(&sqp->sq_worker_cv, &sqp->sq_lock);
1229 	}
1230 
1231 	rx_ring = sqp->sq_rx_ring;
1232 	ill = rx_ring->rr_ill;
1233 	/*
1234 	 * The lock hierarchy is as follows.
1235 	 * cpu_lock -> ill_lock -> sqset_lock -> sq_lock
1236 	 */
1237 	mutex_exit(&sqp->sq_lock);
1238 	mutex_enter(&ill->ill_lock);
1239 	mutex_enter(&sqp->sq_lock);
1240 
1241 	SQS_POLLING_OFF(sqp, (sqp->sq_state & SQS_POLL_CAPAB) != 0,
1242 	    sqp->sq_rx_ring);
1243 	sqp->sq_state &= ~(SQS_POLL_CAPAB | SQS_GET_PKTS | SQS_PROC_HELD);
1244 	if (sqp->sq_state & SQS_POLL_CLEANUP) {
1245 		/*
1246 		 * Disassociate this squeue from its ill_rx_ring_t.
1247 		 * The rr_sqp, sq_rx_ring fields are protected by the
1248 		 * corresponding squeue, ill_lock* and sq_lock. Holding any
1249 		 * of them will ensure that the ring to squeue mapping does
1250 		 * not change.
1251 		 */
1252 		ASSERT(!(sqp->sq_state & SQS_DEFAULT));
1253 
1254 		sqp->sq_rx_ring = NULL;
1255 		rx_ring->rr_sqp = NULL;
1256 
1257 		sqp->sq_state &= ~(SQS_POLL_CLEANUP | SQS_POLL_THR_QUIESCED |
1258 		    SQS_POLL_QUIESCE_DONE);
1259 		sqp->sq_ill = NULL;
1260 
1261 		rx_ring->rr_rx_handle = NULL;
1262 		rx_ring->rr_intr_handle = NULL;
1263 		rx_ring->rr_intr_enable = NULL;
1264 		rx_ring->rr_intr_disable = NULL;
1265 		sqp->sq_state |= SQS_POLL_CLEANUP_DONE;
1266 	} else {
1267 		sqp->sq_state &= ~SQS_POLL_QUIESCE;
1268 		sqp->sq_state |= SQS_POLL_QUIESCE_DONE;
1269 	}
1270 	/*
1271 	 * Signal any waiter that is waiting for the quiesce or cleanup
1272 	 * to complete and also wait for it to actually see and reset the
1273 	 * SQS_POLL_CLEANUP_DONE.
1274 	 */
1275 	cv_signal(&sqp->sq_ctrlop_done_cv);
1276 	mutex_exit(&ill->ill_lock);
1277 	if (sqp->sq_state & SQS_POLL_CLEANUP_DONE) {
1278 		cv_wait(&sqp->sq_worker_cv, &sqp->sq_lock);
1279 		sqp->sq_state &= ~(SQS_PROC | SQS_WORKER);
1280 	}
1281 }
1282 
1283 static void
1284 squeue_worker(squeue_t *sqp)
1285 {
1286 	kmutex_t *lock = &sqp->sq_lock;
1287 	kcondvar_t *async = &sqp->sq_worker_cv;
1288 	callb_cpr_t cprinfo;
1289 	hrtime_t now;
1290 
1291 	CALLB_CPR_INIT(&cprinfo, lock, callb_generic_cpr, "sq_worker");
1292 	mutex_enter(lock);
1293 
1294 	for (;;) {
1295 		for (;;) {
1296 			/*
1297 			 * If the poll thread has handed control to us
1298 			 * we need to break out of the wait.
1299 			 */
1300 			if (sqp->sq_state & SQS_PROC_HELD)
1301 				break;
1302 
1303 			/*
1304 			 * If the squeue is not being processed and we either
1305 			 * have messages to drain or some thread has signaled
1306 			 * some control activity we need to break
1307 			 */
1308 			if (!(sqp->sq_state & SQS_PROC) &&
1309 			    ((sqp->sq_state & SQS_WORKER_THR_CONTROL) ||
1310 			    (sqp->sq_first != NULL)))
1311 				break;
1312 
1313 			/*
1314 			 * If we have started some control action, then check
1315 			 * for the SQS_WORKER flag (since we don't
1316 			 * release the squeue) to make sure we own the squeue
1317 			 * and break out
1318 			 */
1319 			if ((sqp->sq_state & SQS_WORKER_THR_CONTROL) &&
1320 			    (sqp->sq_state & SQS_WORKER))
1321 				break;
1322 
1323 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
1324 			cv_wait(async, lock);
1325 			CALLB_CPR_SAFE_END(&cprinfo, lock);
1326 		}
1327 		if (sqp->sq_state & SQS_WORKER_THR_CONTROL) {
1328 			squeue_worker_thr_control(sqp);
1329 			continue;
1330 		}
1331 		ASSERT(!(sqp->sq_state & (SQS_POLL_THR_QUIESCED |
1332 		    SQS_POLL_CLEANUP_DONE | SQS_POLL_QUIESCE_DONE |
1333 		    SQS_WORKER_THR_CONTROL | SQS_POLL_THR_CONTROL)));
1334 
1335 		if (sqp->sq_state & SQS_PROC_HELD)
1336 			sqp->sq_state &= ~SQS_PROC_HELD;
1337 
1338 		now = gethrtime();
1339 		sqp->sq_run = curthread;
1340 		sqp->sq_drain(sqp, SQS_WORKER, now +  squeue_drain_ns);
1341 		sqp->sq_run = NULL;
1342 	}
1343 }
1344 
1345 uintptr_t *
1346 squeue_getprivate(squeue_t *sqp, sqprivate_t p)
1347 {
1348 	ASSERT(p < SQPRIVATE_MAX);
1349 
1350 	return (&sqp->sq_private[p]);
1351 }
1352 
1353 /* ARGSUSED */
1354 void
1355 squeue_wakeup_conn(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy)
1356 {
1357 	conn_t *connp = (conn_t *)arg;
1358 	squeue_t *sqp = connp->conn_sqp;
1359 
1360 	/*
1361 	 * Mark the squeue as paused before waking up the thread stuck
1362 	 * in squeue_synch_enter().
1363 	 */
1364 	mutex_enter(&sqp->sq_lock);
1365 	sqp->sq_state |= SQS_PAUSE;
1366 
1367 	/*
1368 	 * Notify the thread that it's OK to proceed; that is done by
1369 	 * clearing the MSGWAITSYNC flag. The synch thread will free the mblk.
1370 	 */
1371 	ASSERT(mp->b_flag & MSGWAITSYNC);
1372 	mp->b_flag &= ~MSGWAITSYNC;
1373 	cv_broadcast(&connp->conn_sq_cv);
1374 
1375 	/*
1376 	 * We are doing something on behalf of another thread, so we have to
1377 	 * pause and wait until it finishes.
1378 	 */
1379 	while (sqp->sq_state & SQS_PAUSE) {
1380 		cv_wait(&sqp->sq_synch_cv, &sqp->sq_lock);
1381 	}
1382 	mutex_exit(&sqp->sq_lock);
1383 }
1384 
1385 int
1386 squeue_synch_enter(squeue_t *sqp, conn_t *connp, mblk_t *use_mp)
1387 {
1388 	mutex_enter(&sqp->sq_lock);
1389 	if (sqp->sq_first == NULL && !(sqp->sq_state & SQS_PROC)) {
1390 		/*
1391 		 * We are OK to proceed if the squeue is empty, and
1392 		 * no one owns the squeue.
1393 		 *
1394 		 * The caller won't own the squeue as this is called from the
1395 		 * application.
1396 		 */
1397 		ASSERT(sqp->sq_run == NULL);
1398 
1399 		sqp->sq_state |= SQS_PROC;
1400 		sqp->sq_run = curthread;
1401 		mutex_exit(&sqp->sq_lock);
1402 
1403 #if SQUEUE_DEBUG
1404 		sqp->sq_curmp = NULL;
1405 		sqp->sq_curproc = NULL;
1406 		sqp->sq_connp = connp;
1407 #endif
1408 		connp->conn_on_sqp = B_TRUE;
1409 		return (0);
1410 	} else {
1411 		mblk_t  *mp;
1412 
1413 		mp = (use_mp == NULL) ? allocb(0, BPRI_MED) : use_mp;
1414 		if (mp == NULL) {
1415 			mutex_exit(&sqp->sq_lock);
1416 			return (ENOMEM);
1417 		}
1418 
1419 		/*
1420 		 * We mark the mblk as awaiting synchronous squeue access
1421 		 * by setting the MSGWAITSYNC flag. Once squeue_wakeup_conn
1422 		 * fires, MSGWAITSYNC is cleared, at which point we know we
1423 		 * have exclusive access.
1424 		 */
1425 		mp->b_flag |= MSGWAITSYNC;
1426 
1427 		CONN_INC_REF(connp);
1428 		SET_SQUEUE(mp, squeue_wakeup_conn, connp);
1429 		ENQUEUE_CHAIN(sqp, mp, mp, 1);
1430 
1431 		ASSERT(sqp->sq_run != curthread);
1432 
1433 		/* Wait until the enqueued mblk get processed. */
1434 		while (mp->b_flag & MSGWAITSYNC)
1435 			cv_wait(&connp->conn_sq_cv, &sqp->sq_lock);
1436 		mutex_exit(&sqp->sq_lock);
1437 
1438 		if (use_mp == NULL)
1439 			freeb(mp);
1440 
1441 		return (0);
1442 	}
1443 }
1444 
1445 void
1446 squeue_synch_exit(squeue_t *sqp, conn_t *connp)
1447 {
1448 	mutex_enter(&sqp->sq_lock);
1449 	if (sqp->sq_run == curthread) {
1450 		ASSERT(sqp->sq_state & SQS_PROC);
1451 
1452 		sqp->sq_state &= ~SQS_PROC;
1453 		sqp->sq_run = NULL;
1454 		connp->conn_on_sqp = B_FALSE;
1455 
1456 		if (sqp->sq_first == NULL) {
1457 			mutex_exit(&sqp->sq_lock);
1458 		} else {
1459 			/*
1460 			 * If this was a normal thread, then it would
1461 			 * (most likely) continue processing the pending
1462 			 * requests. Since the just completed operation
1463 			 * was executed synchronously, the thread should
1464 			 * not be delayed. To compensate, wake up the
1465 			 * worker thread right away when there are outstanding
1466 			 * requests.
1467 			 */
1468 			sqp->sq_awaken = ddi_get_lbolt();
1469 			cv_signal(&sqp->sq_worker_cv);
1470 			mutex_exit(&sqp->sq_lock);
1471 		}
1472 	} else {
1473 		/*
1474 		 * The caller doesn't own the squeue, clear the SQS_PAUSE flag,
1475 		 * and wake up the squeue owner, such that owner can continue
1476 		 * processing.
1477 		 */
1478 		ASSERT(sqp->sq_state & SQS_PAUSE);
1479 		sqp->sq_state &= ~SQS_PAUSE;
1480 
1481 		/* There should be only one thread blocking on sq_synch_cv. */
1482 		cv_signal(&sqp->sq_synch_cv);
1483 		mutex_exit(&sqp->sq_lock);
1484 	}
1485 }
1486