xref: /illumos-gate/usr/src/uts/common/sys/mac_soft_ring.h (revision 8a2b682e57a046b828f37bcde1776f131ef4629f)
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 /*
23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  * Copyright 2017 Joyent, Inc.
26  */
27 
28 #ifndef	_SYS_MAC_SOFT_RING_H
29 #define	_SYS_MAC_SOFT_RING_H
30 
31 #ifdef	__cplusplus
32 extern "C" {
33 #endif
34 
35 #include <sys/types.h>
36 #include <sys/cpuvar.h>
37 #include <sys/cpupart.h>
38 #include <sys/processor.h>
39 #include <sys/stream.h>
40 #include <sys/squeue.h>
41 #include <sys/dlpi.h>
42 #include <sys/mac_impl.h>
43 #include <sys/mac_stat.h>
44 
45 #define	S_RING_NAMELEN 64
46 
47 #define	MAX_SR_FANOUT	24
48 
49 extern boolean_t mac_soft_ring_enable;
50 extern boolean_t mac_latency_optimize;
51 
52 typedef struct mac_soft_ring_s mac_soft_ring_t;
53 typedef struct mac_soft_ring_set_s mac_soft_ring_set_t;
54 
55 typedef void (*mac_soft_ring_drain_func_t)(mac_soft_ring_t *);
56 typedef mac_tx_cookie_t (*mac_tx_func_t)(mac_soft_ring_set_t *, mblk_t *,
57     uintptr_t, uint16_t, mblk_t **);
58 
59 
60 /* Tx notify callback */
61 typedef struct mac_tx_notify_cb_s {
62 	mac_cb_t		mtnf_link;	/* Linked list of callbacks */
63 	mac_tx_notify_t		mtnf_fn;	/* The callback function */
64 	void			*mtnf_arg;	/* Callback function argument */
65 } mac_tx_notify_cb_t;
66 
67 struct mac_soft_ring_s {
68 	/* Keep the most used members 64bytes cache aligned */
69 	kmutex_t	s_ring_lock;	/* lock before using any member */
70 	uint16_t	s_ring_type;	/* processing model of the sq */
71 	uint16_t	s_ring_state;	/* state flags and message count */
72 	int		s_ring_count;	/* # of mblocks in mac_soft_ring */
73 	size_t		s_ring_size;	/* Size of data queued */
74 	mblk_t		*s_ring_first;	/* first mblk chain or NULL */
75 	mblk_t		*s_ring_last;	/* last mblk chain or NULL */
76 
77 	mac_direct_rx_t	s_ring_rx_func;
78 	void		*s_ring_rx_arg1;
79 	mac_resource_handle_t  s_ring_rx_arg2;
80 
81 	/*
82 	 * Threshold after which packets get dropped.
83 	 * Is always greater than s_ring_tx_hiwat
84 	 */
85 	int		s_ring_tx_max_q_cnt;
86 	/* # of mblocks after which to apply flow control */
87 	int		s_ring_tx_hiwat;
88 	/* # of mblocks after which to relieve flow control */
89 	int		s_ring_tx_lowat;
90 	boolean_t	s_ring_tx_woken_up;
91 	uint32_t	s_ring_hiwat_cnt;	/* times blocked for Tx descs */
92 
93 	void		*s_ring_tx_arg1;
94 	void		*s_ring_tx_arg2;
95 
96 	/* Tx notify callback */
97 	mac_cb_info_t	s_ring_notify_cb_info;		/* cb list info */
98 	mac_cb_t	*s_ring_notify_cb_list;		/* The cb list */
99 
100 	clock_t		s_ring_awaken;	/* time async thread was awakened */
101 
102 	kthread_t	*s_ring_run;	/* Current thread processing sq */
103 	processorid_t	s_ring_cpuid;	/* processor to bind to */
104 	processorid_t	s_ring_cpuid_save;	/* saved cpuid during offline */
105 	kcondvar_t	s_ring_async;	/* async thread blocks on */
106 	clock_t		s_ring_wait;	/* lbolts to wait after a fill() */
107 	timeout_id_t	s_ring_tid;	/* timer id of pending timeout() */
108 	kthread_t	*s_ring_worker;	/* kernel thread id */
109 	char		s_ring_name[S_RING_NAMELEN + 1];
110 	uint32_t	s_ring_total_inpkt;
111 	uint32_t	s_ring_total_rbytes;
112 	uint32_t	s_ring_drops;
113 	struct mac_client_impl_s *s_ring_mcip;
114 	kstat_t		*s_ring_ksp;
115 
116 	/* Teardown, poll disable control ops */
117 	kcondvar_t	s_ring_client_cv; /* Client wait for control op */
118 
119 	mac_soft_ring_set_t *s_ring_set;   /* The SRS this ring belongs to */
120 	mac_soft_ring_t	*s_ring_next;
121 	mac_soft_ring_t	*s_ring_prev;
122 	mac_soft_ring_drain_func_t s_ring_drain_func;
123 
124 	mac_tx_stats_t	s_st_stat;
125 };
126 
127 typedef void (*mac_srs_drain_proc_t)(mac_soft_ring_set_t *, uint_t);
128 
129 /* Transmit side Soft Ring Set */
130 typedef struct mac_srs_tx_s {
131 	/* Members for Tx size processing */
132 	uint32_t	st_mode;
133 	mac_tx_func_t	st_func;
134 	void		*st_arg1;
135 	void		*st_arg2;
136 	mac_group_t	*st_group;	/* TX group for share */
137 	boolean_t	st_woken_up;
138 
139 	/*
140 	 * st_max_q_cnt is the queue depth threshold to limit
141 	 * outstanding packets on the Tx SRS. Once the limit
142 	 * is reached, Tx SRS will drop packets until the
143 	 * limit goes below the threshold.
144 	 */
145 	uint32_t	st_max_q_cnt;	/* max. outstanding packets */
146 	/*
147 	 * st_hiwat is used Tx serializer and bandwidth mode.
148 	 * This is the queue depth threshold upto which
149 	 * packets will get buffered with no flow-control
150 	 * back pressure applied to the caller. Once this
151 	 * threshold is reached, back pressure will be
152 	 * applied to the caller of mac_tx() (mac_tx() starts
153 	 * returning a cookie to indicate a blocked SRS).
154 	 * st_hiwat should always be lesser than or equal to
155 	 * st_max_q_cnt.
156 	 */
157 	uint32_t	st_hiwat;	/* mblk cnt to apply flow control */
158 	uint32_t	st_lowat;	/* mblk cnt to relieve flow control */
159 	uint32_t	st_hiwat_cnt; /* times blocked for Tx descs */
160 	mac_tx_stats_t	st_stat;
161 	mac_capab_aggr_t	st_capab_aggr;
162 	/*
163 	 * st_soft_rings is used as an array to store aggr Tx soft
164 	 * rings. When aggr_find_tx_ring() returns a pseudo ring,
165 	 * the associated soft ring has to be found. st_soft_rings
166 	 * array stores the soft ring associated with a pseudo Tx
167 	 * ring and it can be accessed using the pseudo ring
168 	 * index (mr_index). Note that the ring index is unique
169 	 * for each ring in a group.
170 	 */
171 	mac_soft_ring_t **st_soft_rings;
172 } mac_srs_tx_t;
173 
174 /* Receive side Soft Ring Set */
175 typedef struct mac_srs_rx_s {
176 	/*
177 	 * Upcall Function for fanout, Rx processing etc. Perhaps
178 	 * the same 3 members below can be used for Tx
179 	 * processing, but looking around, mac_rx_func_t has
180 	 * proliferated too much into various files at different
181 	 * places. I am leaving the consolidation battle for
182 	 * another day.
183 	 */
184 	mac_direct_rx_t		sr_func;	/* srs_lock */
185 	void			*sr_arg1;	/* srs_lock */
186 	mac_resource_handle_t	sr_arg2;	/* srs_lock */
187 	mac_rx_func_t		sr_lower_proc;	/* Atomically changed */
188 	uint32_t		sr_poll_pkt_cnt;
189 	uint32_t		sr_poll_thres;
190 
191 	/* mblk cnt to apply flow control */
192 	uint32_t		sr_hiwat;
193 	/* mblk cnt to relieve flow control */
194 	uint32_t		sr_lowat;
195 	mac_rx_stats_t		sr_stat;
196 
197 	/* Times polling was enabled */
198 	uint32_t		sr_poll_on;
199 	/* Times polling was enabled by worker thread */
200 	uint32_t		sr_worker_poll_on;
201 	/* Times polling was disabled */
202 	uint32_t		sr_poll_off;
203 	/* Poll thread signalled count */
204 	uint32_t		sr_poll_thr_sig;
205 	/* Poll thread busy */
206 	uint32_t		sr_poll_thr_busy;
207 	/* SRS drains, stays in poll mode but doesn't poll */
208 	uint32_t		sr_poll_drain_no_poll;
209 	/*
210 	 * SRS has nothing to do and no packets in H/W but
211 	 * there is a backlog in softrings. SRS stays in
212 	 * poll mode but doesn't do polling.
213 	 */
214 	uint32_t		sr_poll_no_poll;
215 	/* Active polling restarted */
216 	uint32_t		sr_below_hiwat;
217 	/* Found packets in last poll so try and poll again */
218 	uint32_t		sr_poll_again;
219 	/*
220 	 * Packets in queue but poll thread not allowed to process so
221 	 * signal the worker thread.
222 	 */
223 	uint32_t		sr_poll_sig_worker;
224 	/*
225 	 * Poll thread has nothing to do and H/W has nothing so
226 	 * reenable the interrupts.
227 	 */
228 	uint32_t		sr_poll_intr_enable;
229 	/*
230 	 * Poll thread has nothing to do and worker thread was already
231 	 * running so it can decide to reenable interrupt or poll again.
232 	 */
233 	uint32_t		sr_poll_goto_sleep;
234 	/* Worker thread goes back to draining the queue */
235 	uint32_t		sr_drain_again;
236 	/* More Packets in queue so signal the poll thread to drain */
237 	uint32_t		sr_drain_poll_sig;
238 	/* More Packets in queue so signal the worker thread to drain */
239 	uint32_t		sr_drain_worker_sig;
240 	/* Poll thread is already running so worker has nothing to do */
241 	uint32_t		sr_drain_poll_running;
242 	/* We have packets already queued so keep polling */
243 	uint32_t		sr_drain_keep_polling;
244 	/* Drain is done and interrupts are reenabled */
245 	uint32_t		sr_drain_finish_intr;
246 	/* Polling thread needs to schedule worker wakeup */
247 	uint32_t		sr_poll_worker_wakeup;
248 } mac_srs_rx_t;
249 
250 /*
251  * mac_soft_ring_set_s:
252  * This is used both for Tx and Rx side. The srs_type identifies Rx or
253  * Tx type.
254  *
255  * Note that the structure is carefully crafted, with Rx elements coming
256  * first followed by Tx specific members. Future additions to this
257  * structure should follow the same guidelines.
258  *
259  * Rx-side notes:
260  * mac_rx_classify_flow_add() always creates a mac_soft_ring_set_t and fn_flow
261  * points to info from it (func = srs_lower_proc, arg = soft_ring_set). On
262  * interrupt path, srs_lower_proc does B/W adjustment and switch to polling mode
263  * (if poll capable) and feeds the packets to soft_ring_list via choosen
264  * fanout type (specified by srs_type). In poll mode, the poll thread which is
265  * also a pointer can pick up the packets and feed them to various
266  * soft_ring_list.
267  *
268  * The srs_type can either be protocol based or fanout based where fanout itelf
269  * can be various types
270  *
271  * The polling works by turning off interrupts as soon as a packets
272  * are queued on the soft ring set. Once the backlog is clear and poll
273  * thread return empty handed i.e. Rx ring doesn't have anything, the
274  * interrupt is turned back on. For this purpose we keep a separate
275  * srs_poll_pkt_cnt counter which tracks the packets queued between SRS
276  * and the soft rings as well. The counter is incremented when packets
277  * are queued and decremented when SRS processes them (in case it has
278  * no soft rings) or the soft ring process them. Its important that
279  * in case SRS has softrings, the decrement doesn't happen till the
280  * packet is processed by the soft rings since it takes very little time
281  * for SRS to queue packet from SRS to soft rings and it will keep
282  * bringing more packets in the system faster than soft rings can
283  * process them.
284  *
285  * Tx side notes:
286  * The srs structure acts as a serializer with a worker thread. The
287  * default behavior of srs though is to act as a pass-thru. The queues
288  * (srs_first, srs_last, srs_count) get used when Tx ring runs out of Tx
289  * descriptors or to enforce bandwidth limits.
290  *
291  * When multiple Tx rings are present, the SRS state will be set to
292  * SRS_FANOUT_OTH. Outgoing packets coming into mac_tx_srs_process()
293  * function will be fanned out to one of the Tx side soft rings based on
294  * a hint passed in mac_tx_srs_process(). Each soft ring, in turn, will
295  * be associated with a distinct h/w Tx ring.
296  */
297 
298 struct mac_soft_ring_set_s {
299 	/*
300 	 * Common elements, common to both Rx and Tx SRS type.
301 	 * The following block of fields are protected by srs_lock
302 	 */
303 	kmutex_t	srs_lock;
304 	uint32_t	srs_type;
305 	uint32_t	srs_state;	/* state flags */
306 	uint32_t	srs_count;
307 	mblk_t		*srs_first;	/* first mblk chain or NULL */
308 	mblk_t		*srs_last;	/* last mblk chain or NULL */
309 	kcondvar_t	srs_async;	/* cv for worker thread */
310 	kcondvar_t	srs_cv;		/* cv for poll thread */
311 	kcondvar_t	srs_quiesce_done_cv;	/* cv for removal */
312 	timeout_id_t	srs_tid;	/* timeout id for pending timeout */
313 
314 	/*
315 	 * List of soft rings & processing function.
316 	 * The following block is protected by Rx quiescence.
317 	 * i.e. they can be changed only after quiescing the SRS
318 	 * Protected by srs_lock.
319 	 */
320 	mac_soft_ring_t	*srs_soft_ring_head;
321 	mac_soft_ring_t	*srs_soft_ring_tail;
322 	int		srs_soft_ring_count;
323 	int		srs_soft_ring_quiesced_count;
324 	int		srs_soft_ring_condemned_count;
325 	mac_soft_ring_t	**srs_tcp_soft_rings;
326 	int		srs_tcp_ring_count;
327 	mac_soft_ring_t	**srs_udp_soft_rings;
328 	int		srs_udp_ring_count;
329 	mac_soft_ring_t	**srs_oth_soft_rings;
330 	int		srs_oth_ring_count;
331 	/*
332 	 * srs_tx_soft_rings is used by tx_srs in
333 	 * when operating in multi tx ring mode.
334 	 */
335 	mac_soft_ring_t	**srs_tx_soft_rings;
336 	int		srs_tx_ring_count;
337 
338 	/*
339 	 * Bandwidth control related members.
340 	 * They are common to both Rx- and Tx-side.
341 	 * Following protected by srs_lock
342 	 */
343 	mac_bw_ctl_t	*srs_bw;
344 	size_t		srs_size;	/* Size of packets queued in bytes */
345 	pri_t		srs_pri;
346 
347 	mac_soft_ring_set_t	*srs_next;	/* mac_srs_g_lock */
348 	mac_soft_ring_set_t	*srs_prev;	/* mac_srs_g_lock */
349 
350 	/* Attribute specific drain func (BW ctl vs non-BW ctl)	*/
351 	mac_srs_drain_proc_t	srs_drain_func;	/* Write once (WO) */
352 
353 	/*
354 	 * If the associated ring is exclusively used by a mac client, e.g.,
355 	 * an aggregation, this fields is used to keep a reference to the
356 	 * MAC client's pseudo ring.
357 	 */
358 	mac_resource_handle_t	srs_mrh;
359 	/*
360 	 * The following blocks are write once (WO) and valid for the life
361 	 * of the SRS
362 	 */
363 	struct mac_client_impl_s *srs_mcip;	/* back ptr to mac client */
364 	void			*srs_flent;	/* back ptr to flent */
365 	mac_ring_t		*srs_ring;	/*  Ring Descriptor */
366 
367 	/* Teardown, disable control ops */
368 	kcondvar_t	srs_client_cv;	/* Client wait for the control op */
369 
370 	kthread_t	*srs_worker;	/* WO, worker thread */
371 	kthread_t	*srs_poll_thr;	/* WO, poll thread */
372 
373 	uint_t		srs_ind;	/* Round Robin indx for picking up SR */
374 	processorid_t	srs_worker_cpuid;	/* processor to bind to */
375 	processorid_t	srs_worker_cpuid_save;	/* saved cpuid during offline */
376 	processorid_t	srs_poll_cpuid;		/* processor to bind to */
377 	processorid_t	srs_poll_cpuid_save;	/* saved cpuid during offline */
378 	uint_t		srs_fanout_state;
379 	mac_cpus_t	srs_cpu;
380 
381 	mac_srs_rx_t	srs_rx;
382 	mac_srs_tx_t	srs_tx;
383 	kstat_t		*srs_ksp;
384 };
385 
386 /*
387  * type flags - combination allowed to process and drain the queue
388  */
389 #define	ST_RING_WORKER_ONLY	0x0001	/* Worker thread only */
390 #define	ST_RING_ANY		0x0002	/* Any thread can process the queue */
391 #define	ST_RING_TCP		0x0004
392 #define	ST_RING_UDP		0x0008
393 #define	ST_RING_OTH		0x0010
394 
395 #define	ST_RING_BW_CTL		0x0020
396 #define	ST_RING_TX		0x0040
397 
398 /*
399  * State flags.
400  */
401 #define	S_RING_PROC		0x0001	/* being processed */
402 #define	S_RING_BOUND		0x0002	/* Worker thread is bound to a cpu */
403 #define	S_RING_BLOCK		0x0004	/* No Tx descs */
404 #define	S_RING_TX_HIWAT		0x0008	/* Tx high watermark reached */
405 
406 #define	S_RING_WAKEUP_CLIENT	0x0010	/* flow ctrl, client wakeup needed */
407 #define	S_RING_BLANK		0x0020	/* Has been put into polling mode */
408 #define	S_RING_CLIENT_WAIT	0x0040	/* Client waiting for control op */
409 
410 #define	S_RING_CONDEMNED	0x0100	/* Being torn down */
411 #define	S_RING_CONDEMNED_DONE	0x0200	/* Being torn down */
412 #define	S_RING_QUIESCE		0x0400	/* No traffic flow, transient flag */
413 #define	S_RING_QUIESCE_DONE	0x0800	/* No traffic flow, transient flag */
414 
415 #define	S_RING_RESTART		0x1000	/* Go back to normal traffic flow */
416 #define	S_RING_ENQUEUED		0x2000	/* Pkts enqueued in Tx soft ring */
417 
418 /*
419  * arguments for processors to bind to
420  */
421 #define	S_RING_BIND_NONE	-1
422 
423 /*
424  * defines for srs_type - identifies a link or a sub-flow
425  * and other static characteristics of a SRS like a tx
426  * srs, tcp only srs, etc.
427  */
428 #define	SRST_LINK		0x00000001
429 #define	SRST_FLOW		0x00000002
430 #define	SRST_NO_SOFT_RINGS	0x00000004
431 #define	SRST_TCP_ONLY		0x00000008
432 
433 #define	SRST_FANOUT_PROTO	0x00000010
434 #define	SRST_FANOUT_SRC_IP	0x00000020
435 #define	SRST_FANOUT_OTH		0x00000040
436 #define	SRST_DEFAULT_GRP	0x00000080
437 
438 #define	SRST_TX			0x00000100
439 #define	SRST_BW_CONTROL		0x00000200
440 #define	SRST_DIRECT_POLL	0x00000400
441 
442 #define	SRST_DLS_BYPASS		0x00001000
443 #define	SRST_CLIENT_POLL_ENABLED 0x00002000
444 
445 /*
446  * soft ring set flags. These bits are dynamic in nature and get
447  * applied to srs_state. They reflect the state of SRS at any
448  * point of time
449  */
450 #define	SRS_BLANK		0x00000001
451 #define	SRS_WORKER_BOUND	0x00000002
452 #define	SRS_POLL_BOUND		0x00000004
453 #define	SRS_POLLING_CAPAB	0x00000008
454 
455 #define	SRS_PROC		0x00000010
456 #define	SRS_GET_PKTS		0x00000020
457 #define	SRS_POLLING		0x00000040
458 #define	SRS_BW_ENFORCED		0x00000080
459 
460 #define	SRS_WORKER		0x00000100
461 #define	SRS_ENQUEUED		0x00000200
462 #define	SRS_ANY_PROCESS		0x00000400
463 #define	SRS_PROC_FAST		0x00000800
464 
465 #define	SRS_POLL_PROC		0x00001000
466 #define	SRS_TX_BLOCKED		0x00002000	/* out of Tx descs */
467 #define	SRS_TX_HIWAT		0x00004000	/* Tx count exceeds hiwat */
468 #define	SRS_TX_WAKEUP_CLIENT	0x00008000	/* Flow-ctl: wakeup client */
469 
470 #define	SRS_CLIENT_PROC		0x00010000
471 #define	SRS_CLIENT_WAIT		0x00020000
472 #define	SRS_QUIESCE		0x00040000
473 #define	SRS_QUIESCE_DONE	0x00080000
474 
475 #define	SRS_CONDEMNED		0x00100000
476 #define	SRS_CONDEMNED_DONE	0x00200000
477 #define	SRS_POLL_THR_QUIESCED	0x00400000
478 #define	SRS_RESTART		0x00800000
479 
480 #define	SRS_RESTART_DONE	0x01000000
481 #define	SRS_POLL_THR_RESTART	0x02000000
482 #define	SRS_IN_GLIST		0x04000000
483 #define	SRS_POLL_THR_EXITED	0x08000000
484 
485 #define	SRS_QUIESCE_PERM	0x10000000
486 #define	SRS_LATENCY_OPT		0x20000000
487 #define	SRS_SOFTRING_QUEUE	0x40000000
488 
489 #define	SRS_QUIESCED(srs)	(srs->srs_state & SRS_QUIESCE_DONE)
490 
491 /*
492  * If the SRS_QUIESCE_PERM flag is set, the SRS worker thread will not be
493  * able to be restarted.
494  */
495 #define	SRS_QUIESCED_PERMANENT(srs)	(srs->srs_state & SRS_QUIESCE_PERM)
496 
497 /*
498  * soft ring set (SRS) Tx modes
499  */
500 typedef enum {
501 	SRS_TX_DEFAULT = 0,
502 	SRS_TX_SERIALIZE,
503 	SRS_TX_FANOUT,
504 	SRS_TX_BW,
505 	SRS_TX_BW_FANOUT,
506 	SRS_TX_AGGR,
507 	SRS_TX_BW_AGGR
508 } mac_tx_srs_mode_t;
509 
510 /*
511  * SRS fanout states
512  */
513 typedef enum {
514 	SRS_FANOUT_UNINIT = 0,
515 	SRS_FANOUT_INIT,
516 	SRS_FANOUT_REINIT
517 } mac_srs_fanout_state_t;
518 
519 /*
520  * Structure for dls statistics
521  */
522 struct dls_kstats {
523 	kstat_named_t	dlss_soft_ring_pkt_drop;
524 };
525 
526 extern struct dls_kstats dls_kstat;
527 
528 #define	DLS_BUMP_STAT(x, y)	(dls_kstat.x.value.ui32 += y)
529 
530 /* Turn dynamic polling off */
531 #define	MAC_SRS_POLLING_OFF(mac_srs) {					\
532 	ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock));			\
533 	if (((mac_srs)->srs_state & (SRS_POLLING_CAPAB|SRS_POLLING)) == \
534 	    (SRS_POLLING_CAPAB|SRS_POLLING)) {				\
535 		(mac_srs)->srs_state &= ~SRS_POLLING;			\
536 		(void) mac_hwring_enable_intr((mac_ring_handle_t)	\
537 		    (mac_srs)->srs_ring);				\
538 		(mac_srs)->srs_rx.sr_poll_off++;			\
539 	}								\
540 }
541 
542 #define	MAC_COUNT_CHAIN(mac_srs, head, tail, cnt, sz)	{	\
543 	mblk_t		*tmp;					\
544 	boolean_t	bw_ctl = B_FALSE;			\
545 								\
546 	ASSERT((head) != NULL);					\
547 	cnt = 0;						\
548 	sz = 0;							\
549 	if ((mac_srs)->srs_type & SRST_BW_CONTROL)		\
550 		bw_ctl = B_TRUE;				\
551 	tmp = tail = (head);					\
552 	if ((head)->b_next == NULL) {				\
553 		cnt = 1;					\
554 		if (bw_ctl)					\
555 			sz += msgdsize(head);			\
556 	} else {						\
557 		while (tmp != NULL) {				\
558 			tail = tmp;				\
559 			cnt++;					\
560 			if (bw_ctl)				\
561 				sz += msgdsize(tmp);		\
562 			tmp = tmp->b_next;			\
563 		}						\
564 	}							\
565 }
566 
567 /*
568  * Decrement the cumulative packet count in SRS and its
569  * soft rings. If the srs_poll_pkt_cnt goes below lowat, then check
570  * if if the interface was left in a polling mode and no one
571  * is really processing the queue (to get the interface out
572  * of poll mode). If no one is processing the queue, then
573  * acquire the PROC and signal the poll thread to check the
574  * interface for packets and get the interface back to interrupt
575  * mode if nothing is found.
576  */
577 #define	MAC_UPDATE_SRS_COUNT_LOCKED(mac_srs, cnt) {		        \
578 	mac_srs_rx_t	*srs_rx = &(mac_srs)->srs_rx;			\
579 	ASSERT(MUTEX_HELD(&(mac_srs)->srs_lock));			\
580 									\
581 	srs_rx->sr_poll_pkt_cnt -= cnt;					\
582 	if ((srs_rx->sr_poll_pkt_cnt <= srs_rx->sr_poll_thres) &&	\
583 	    (((mac_srs)->srs_state &					\
584 	    (SRS_POLLING|SRS_PROC|SRS_GET_PKTS)) == SRS_POLLING))	\
585 	{								\
586 		(mac_srs)->srs_state |= (SRS_PROC|SRS_GET_PKTS);	\
587 		cv_signal(&(mac_srs)->srs_cv);				\
588 		srs_rx->sr_below_hiwat++;				\
589 	}								\
590 }
591 
592 /*
593  * The following two macros are used to update the inbound packet and byte.
594  * count. The packet and byte count reflect the packets and bytes that are
595  * taken out of the SRS's queue, i.e. indicating they are being delivered.
596  * The srs_count and srs_size are updated in different locations as the
597  * srs_size is also used to take into account any bandwidth limits. The
598  * srs_size is updated only when a soft ring, if any, sends a packet up,
599  * as opposed to updating it when the SRS sends a packet to the SR, i.e.
600  * the srs_size reflects the packets in the SRS and SRs. These
601  * macros decrement the srs_size and srs_count and also increment the
602  * ipackets and ibytes stats resp.
603  *
604  * xxx-venu These are done under srs_lock, for now we still update
605  * mci_stat_ibytes/mci_stat_ipackets atomically, need to check if
606  * just updating them would be accurate enough.
607  *
608  * If we are updating these for a sub-flow SRS, then we need to also
609  * updated it's MAC client bandwidth info, if the MAC client is also
610  * bandwidth regulated.
611  */
612 #define	MAC_UPDATE_SRS_SIZE_LOCKED(srs, sz) {				\
613 	if ((srs)->srs_type & SRST_BW_CONTROL) {			\
614 		mutex_enter(&(srs)->srs_bw->mac_bw_lock);		\
615 		(srs)->srs_bw->mac_bw_sz -= (sz);			\
616 		(srs)->srs_bw->mac_bw_used += (sz);			\
617 		mutex_exit(&(srs)->srs_bw->mac_bw_lock);		\
618 	}								\
619 }
620 
621 #define	MAC_TX_UPDATE_BW_INFO(srs, sz) {				\
622 	(srs)->srs_bw->mac_bw_sz -= (sz);				\
623 	(srs)->srs_bw->mac_bw_used += (sz);				\
624 }
625 
626 #define	MAC_TX_SOFT_RINGS(mac_srs) ((mac_srs)->srs_tx_ring_count >= 1)
627 
628 /* Soft ring flags for teardown */
629 #define	SRS_POLL_THR_OWNER	(SRS_PROC | SRS_POLLING | SRS_GET_PKTS)
630 #define	SRS_PAUSE		(SRS_CONDEMNED | SRS_QUIESCE)
631 #define	S_RING_PAUSE		(S_RING_CONDEMNED | S_RING_QUIESCE)
632 
633 /* Soft rings */
634 extern void mac_soft_ring_init(void);
635 extern void mac_soft_ring_finish(void);
636 extern void mac_fanout_setup(mac_client_impl_t *, flow_entry_t *,
637     mac_resource_props_t *, mac_direct_rx_t, void *, mac_resource_handle_t,
638     cpupart_t *);
639 
640 extern void mac_soft_ring_worker_wakeup(mac_soft_ring_t *);
641 extern void mac_soft_ring_blank(void *, time_t, uint_t, int);
642 extern mblk_t *mac_soft_ring_poll(mac_soft_ring_t *, size_t);
643 extern void mac_soft_ring_destroy(mac_soft_ring_t *);
644 extern void mac_soft_ring_dls_bypass(void *, mac_direct_rx_t, void *);
645 
646 /* Rx SRS */
647 extern mac_soft_ring_set_t *mac_srs_create(struct mac_client_impl_s *,
648     flow_entry_t *, uint32_t, mac_direct_rx_t, void *, mac_resource_handle_t,
649     mac_ring_t *);
650 extern void mac_srs_free(mac_soft_ring_set_t *);
651 extern void mac_srs_signal(mac_soft_ring_set_t *, uint_t);
652 extern cpu_t *mac_srs_bind(mac_soft_ring_set_t *, processorid_t);
653 extern void mac_rx_srs_retarget_intr(mac_soft_ring_set_t *, processorid_t);
654 extern void mac_tx_srs_retarget_intr(mac_soft_ring_set_t *);
655 
656 extern void mac_srs_quiesce_initiate(mac_soft_ring_set_t *);
657 extern void mac_srs_client_poll_enable(struct mac_client_impl_s *,
658     mac_soft_ring_set_t *);
659 extern void mac_srs_client_poll_disable(struct mac_client_impl_s *,
660     mac_soft_ring_set_t *);
661 extern void mac_srs_client_poll_quiesce(struct mac_client_impl_s *,
662     mac_soft_ring_set_t *);
663 extern void mac_srs_client_poll_restart(struct mac_client_impl_s *,
664     mac_soft_ring_set_t *);
665 extern void mac_rx_srs_quiesce(mac_soft_ring_set_t *, uint_t);
666 extern void mac_rx_srs_restart(mac_soft_ring_set_t *);
667 extern void mac_rx_srs_subflow_process(void *, mac_resource_handle_t, mblk_t *,
668     boolean_t);
669 extern void mac_tx_srs_quiesce(mac_soft_ring_set_t *, uint_t);
670 
671 /* Tx SRS, Tx softring */
672 extern void mac_tx_srs_wakeup(mac_soft_ring_set_t *, mac_ring_handle_t);
673 extern void mac_tx_srs_setup(struct mac_client_impl_s *, flow_entry_t *);
674 extern mac_tx_func_t mac_tx_get_func(uint32_t);
675 extern mblk_t *mac_tx_send(mac_client_handle_t, mac_ring_handle_t, mblk_t *,
676     mac_tx_stats_t *);
677 extern boolean_t mac_tx_srs_ring_present(mac_soft_ring_set_t *, mac_ring_t *);
678 extern mac_soft_ring_t *mac_tx_srs_get_soft_ring(mac_soft_ring_set_t *,
679     mac_ring_t *);
680 extern void mac_tx_srs_add_ring(mac_soft_ring_set_t *, mac_ring_t *);
681 extern void mac_tx_srs_del_ring(mac_soft_ring_set_t *, mac_ring_t *);
682 extern mac_tx_cookie_t mac_tx_srs_no_desc(mac_soft_ring_set_t *, mblk_t *,
683     uint16_t, mblk_t **);
684 
685 /* Subflow specific stuff */
686 extern int mac_srs_flow_create(struct mac_client_impl_s *, flow_entry_t *,
687     mac_resource_props_t *, int, int, mac_direct_rx_t);
688 extern void mac_srs_update_bwlimit(flow_entry_t *, mac_resource_props_t *);
689 extern void mac_srs_adjust_subflow_bwlimit(struct mac_client_impl_s *);
690 extern void mac_srs_update_drv(struct mac_client_impl_s *);
691 extern void mac_update_srs_priority(mac_soft_ring_set_t *, pri_t);
692 extern void mac_client_update_classifier(mac_client_impl_t *, boolean_t);
693 
694 extern int mac_soft_ring_intr_enable(void *);
695 extern boolean_t mac_soft_ring_intr_disable(void *);
696 extern mac_soft_ring_t *mac_soft_ring_create(int, clock_t, uint16_t,
697     pri_t, mac_client_impl_t *, mac_soft_ring_set_t *,
698     processorid_t, mac_direct_rx_t, void *, mac_resource_handle_t);
699 extern cpu_t *mac_soft_ring_bind(mac_soft_ring_t *, processorid_t);
700 	extern void mac_soft_ring_unbind(mac_soft_ring_t *);
701 extern void mac_soft_ring_free(mac_soft_ring_t *);
702 extern void mac_soft_ring_signal(mac_soft_ring_t *, uint_t);
703 extern void mac_rx_soft_ring_process(mac_client_impl_t *, mac_soft_ring_t *,
704     mblk_t *, mblk_t *, int, size_t);
705 extern mac_tx_cookie_t mac_tx_soft_ring_process(mac_soft_ring_t *,
706     mblk_t *, uint16_t, mblk_t **);
707 extern void mac_srs_worker_quiesce(mac_soft_ring_set_t *);
708 extern void mac_srs_worker_restart(mac_soft_ring_set_t *);
709 extern void mac_rx_attach_flow_srs(mac_impl_t *, flow_entry_t *,
710     mac_soft_ring_set_t *, mac_ring_t *, mac_classify_type_t);
711 
712 extern void mac_rx_srs_drain_bw(mac_soft_ring_set_t *, uint_t);
713 extern void mac_rx_srs_drain(mac_soft_ring_set_t *, uint_t);
714 extern void mac_rx_srs_process(void *, mac_resource_handle_t, mblk_t *,
715     boolean_t);
716 extern void mac_srs_worker(mac_soft_ring_set_t *);
717 extern void mac_rx_srs_poll_ring(mac_soft_ring_set_t *);
718 extern void mac_tx_srs_drain(mac_soft_ring_set_t *, uint_t);
719 
720 extern void mac_tx_srs_restart(mac_soft_ring_set_t *);
721 extern void mac_rx_srs_remove(mac_soft_ring_set_t *);
722 
723 #ifdef	__cplusplus
724 }
725 #endif
726 
727 #endif	/* _SYS_MAC_SOFT_RING_H */
728