Home | History | Annotate | Line # | Download | only in altq
altq_rmclass.c revision 1.7
      1 /*	$NetBSD: altq_rmclass.c,v 1.7 2003/10/07 21:22:11 mycroft Exp $	*/
      2 /*	$KAME: altq_rmclass.c,v 1.9 2000/12/14 08:12:46 thorpej Exp $	*/
      3 
      4 /*
      5  * Copyright (c) 1991-1997 Regents of the University of California.
      6  * All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. All advertising materials mentioning features or use of this software
     17  *    must display the following acknowledgement:
     18  *      This product includes software developed by the Network Research
     19  *      Group at Lawrence Berkeley Laboratory.
     20  * 4. Neither the name of the University nor of the Laboratory may be used
     21  *    to endorse or promote products derived from this software without
     22  *    specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  *
     36  * LBL code modified by speer (at) eng.sun.com, May 1977.
     37  * For questions and/or comments, please send mail to cbq (at) ee.lbl.gov
     38  */
     39 
     40 #include <sys/cdefs.h>
     41 __KERNEL_RCSID(0, "$NetBSD: altq_rmclass.c,v 1.7 2003/10/07 21:22:11 mycroft Exp $");
     42 
     43 #ident "@(#)rm_class.c  1.48     97/12/05 SMI"
     44 
     45 #if defined(__FreeBSD__) || defined(__NetBSD__)
     46 #include "opt_altq.h"
     47 #if (__FreeBSD__ != 2)
     48 #include "opt_inet.h"
     49 #ifdef __FreeBSD__
     50 #include "opt_inet6.h"
     51 #endif
     52 #endif
     53 #endif /* __FreeBSD__ || __NetBSD__ */
     54 #ifdef ALTQ_CBQ	/* cbq is enabled by ALTQ_CBQ option in opt_altq.h */
     55 
     56 #include <sys/param.h>
     57 #include <sys/malloc.h>
     58 #include <sys/mbuf.h>
     59 #include <sys/socket.h>
     60 #include <sys/systm.h>
     61 #include <sys/errno.h>
     62 #include <sys/time.h>
     63 #include <sys/kernel.h>
     64 
     65 #include <net/if.h>
     66 #include <netinet/in.h>
     67 #include <netinet/in_systm.h>
     68 #include <netinet/ip.h>
     69 
     70 #include <altq/altq.h>
     71 #include <altq/altq_rmclass.h>
     72 #include <altq/altq_rmclass_debug.h>
     73 #include <altq/altq_red.h>
     74 #include <altq/altq_rio.h>
     75 
     76 /*
     77  * Local Macros
     78  */
     79 
     80 #define	reset_cutoff(ifd)	{ ifd->cutoff_ = RM_MAXDEPTH; }
     81 
     82 /*
     83  * Local routines.
     84  */
     85 
     86 static int	rmc_satisfied __P((struct rm_class *, struct timeval *));
     87 static void	rmc_wrr_set_weights __P((struct rm_ifdat *));
     88 static void	rmc_depth_compute __P((struct rm_class *));
     89 static void	rmc_depth_recompute __P((rm_class_t *));
     90 
     91 static mbuf_t	*_rmc_wrr_dequeue_next __P((struct rm_ifdat *, int));
     92 static mbuf_t	*_rmc_prr_dequeue_next __P((struct rm_ifdat *, int));
     93 
     94 static int	_rmc_addq __P((rm_class_t *, mbuf_t *));
     95 static void	_rmc_dropq __P((rm_class_t *));
     96 static mbuf_t	*_rmc_getq __P((rm_class_t *));
     97 static mbuf_t	*_rmc_pollq __P((rm_class_t *));
     98 
     99 static int	rmc_under_limit __P((struct rm_class *, struct timeval *));
    100 static void	rmc_tl_satisfied __P((struct rm_ifdat *, struct timeval *));
    101 static void	rmc_drop_action __P((struct rm_class *));
    102 static void	rmc_restart __P((struct rm_class *));
    103 static void	rmc_root_overlimit __P((struct rm_class *, struct rm_class *));
    104 
    105 #define	BORROW_OFFTIME
    106 /*
    107  * BORROW_OFFTIME (experimental):
    108  * borrow the offtime of the class borrowing from.
    109  * the reason is that when its own offtime is set, the class is unable
    110  * to borrow much, especially when cutoff is taking effect.
    111  * but when the borrowed class is overloaded (advidle is close to minidle),
    112  * use the borrowing class's offtime to avoid overload.
    113  */
    114 #define	ADJUST_CUTOFF
    115 /*
    116  * ADJUST_CUTOFF (experimental):
    117  * if no underlimit class is found due to cutoff, increase cutoff and
    118  * retry the scheduling loop.
    119  * also, don't invoke delay_actions while cutoff is taking effect,
    120  * since a sleeping class won't have a chance to be scheduled in the
    121  * next loop.
    122  *
    123  * now heuristics for setting the top-level variable (cutoff_) becomes:
    124  *	1. if a packet arrives for a not-overlimit class, set cutoff
    125  *	   to the depth of the class.
    126  *	2. if cutoff is i, and a packet arrives for an overlimit class
    127  *	   with an underlimit ancestor at a lower level than i (say j),
    128  *	   then set cutoff to j.
    129  *	3. at scheduling a packet, if there is no underlimit class
    130  *	   due to the current cutoff level, increase cutoff by 1 and
    131  *	   then try to schedule again.
    132  */
    133 
    134 /*
    135  * rm_class_t *
    136  * rmc_newclass(...) - Create a new resource management class at priority
    137  * 'pri' on the interface given by 'ifd'.
    138  *
    139  * nsecPerByte  is the data rate of the interface in nanoseconds/byte.
    140  *              E.g., 800 for a 10Mb/s ethernet.  If the class gets less
    141  *              than 100% of the bandwidth, this number should be the
    142  *              'effective' rate for the class.  Let f be the
    143  *              bandwidth fraction allocated to this class, and let
    144  *              nsPerByte be the data rate of the output link in
    145  *              nanoseconds/byte.  Then nsecPerByte is set to
    146  *              nsPerByte / f.  E.g., 1600 (= 800 / .5)
    147  *              for a class that gets 50% of an ethernet's bandwidth.
    148  *
    149  * action       the routine to call when the class is over limit.
    150  *
    151  * maxq         max allowable queue size for class (in packets).
    152  *
    153  * parent       parent class pointer.
    154  *
    155  * borrow       class to borrow from (should be either 'parent' or null).
    156  *
    157  * maxidle      max value allowed for class 'idle' time estimate (this
    158  *              parameter determines how large an initial burst of packets
    159  *              can be before overlimit action is invoked.
    160  *
    161  * offtime      how long 'delay' action will delay when class goes over
    162  *              limit (this parameter determines the steady-state burst
    163  *              size when a class is running over its limit).
    164  *
    165  * Maxidle and offtime have to be computed from the following:  If the
    166  * average packet size is s, the bandwidth fraction allocated to this
    167  * class is f, we want to allow b packet bursts, and the gain of the
    168  * averaging filter is g (= 1 - 2^(-RM_FILTER_GAIN)), then:
    169  *
    170  *   ptime = s * nsPerByte * (1 - f) / f
    171  *   maxidle = ptime * (1 - g^b) / g^b
    172  *   minidle = -ptime * (1 / (f - 1))
    173  *   offtime = ptime * (1 + 1/(1 - g) * (1 - g^(b - 1)) / g^(b - 1)
    174  *
    175  * Operationally, it's convenient to specify maxidle & offtime in units
    176  * independent of the link bandwidth so the maxidle & offtime passed to
    177  * this routine are the above values multiplied by 8*f/(1000*nsPerByte).
    178  * (The constant factor is a scale factor needed to make the parameters
    179  * integers.  This scaling also means that the 'unscaled' values of
    180  * maxidle*nsecPerByte/8 and offtime*nsecPerByte/8 will be in microseconds,
    181  * not nanoseconds.)  Also note that the 'idle' filter computation keeps
    182  * an estimate scaled upward by 2^RM_FILTER_GAIN so the passed value of
    183  * maxidle also must be scaled upward by this value.  Thus, the passed
    184  * values for maxidle and offtime can be computed as follows:
    185  *
    186  * maxidle = maxidle * 2^RM_FILTER_GAIN * 8 / (1000 * nsecPerByte)
    187  * offtime = offtime * 8 / (1000 * nsecPerByte)
    188  *
    189  * When USE_HRTIME is employed, then maxidle and offtime become:
    190  * 	maxidle = maxilde * (8.0 / nsecPerByte);
    191  * 	offtime = offtime * (8.0 / nsecPerByte);
    192  */
    193 
    194 struct rm_class *
    195 rmc_newclass(pri, ifd, nsecPerByte, action, maxq, parent, borrow,
    196 	     maxidle, minidle, offtime, pktsize, flags)
    197 	int		pri;
    198 	struct rm_ifdat	*ifd;
    199 	u_int		nsecPerByte;
    200 	void		(*action)(rm_class_t *, rm_class_t *);
    201 	int		maxq;
    202 	struct rm_class	*parent;
    203 	struct rm_class	*borrow;
    204 	u_int		maxidle;
    205 	int		minidle;
    206 	u_int		offtime;
    207 	int		pktsize;
    208 	int		flags;
    209 {
    210 	struct rm_class *cl;
    211 	struct rm_class *peer;
    212 	int s;
    213 
    214 	if (pri >= RM_MAXPRIO)
    215 		return (NULL);
    216 #ifndef ALTQ_RED
    217 	if (flags & RMCF_RED) {
    218 		printf("rmc_newclass: RED not configured for CBQ!\n");
    219 		return (NULL);
    220 	}
    221 #endif
    222 #ifndef ALTQ_RIO
    223 	if (flags & RMCF_RIO) {
    224 		printf("rmc_newclass: RIO not configured for CBQ!\n");
    225 		return (NULL);
    226 	}
    227 #endif
    228 
    229 	MALLOC(cl, struct rm_class *, sizeof(struct rm_class),
    230 	       M_DEVBUF, M_WAITOK);
    231 	if (cl == NULL)
    232 		return (NULL);
    233 	bzero(cl, sizeof(struct rm_class));
    234 	CALLOUT_INIT(&cl->callout_);
    235 	MALLOC(cl->q_, class_queue_t *, sizeof(class_queue_t),
    236 	       M_DEVBUF, M_WAITOK);
    237 	if (cl->q_ == NULL) {
    238 		FREE(cl, M_DEVBUF);
    239 		return (NULL);
    240 	}
    241 	bzero(cl->q_, sizeof(class_queue_t));
    242 
    243 	/*
    244 	 * Class initialization.
    245 	 */
    246 	cl->children_ = NULL;
    247 	cl->parent_ = parent;
    248 	cl->borrow_ = borrow;
    249 	cl->leaf_ = 1;
    250 	cl->ifdat_ = ifd;
    251 	cl->pri_ = pri;
    252 	cl->allotment_ = RM_NS_PER_SEC / nsecPerByte; /* Bytes per sec */
    253 	cl->depth_ = 0;
    254 	cl->qthresh_ = 0;
    255 	cl->ns_per_byte_ = nsecPerByte;
    256 
    257 	qlimit(cl->q_) = maxq;
    258 	qtype(cl->q_) = Q_DROPHEAD;
    259 	qlen(cl->q_) = 0;
    260 	cl->flags_ = flags;
    261 
    262 #if 1 /* minidle is also scaled in ALTQ */
    263 	cl->minidle_ = (minidle * (int)nsecPerByte) / 8;
    264 	if (cl->minidle_ > 0)
    265 		cl->minidle_ = 0;
    266 #else
    267 	cl->minidle_ = minidle;
    268 #endif
    269 	cl->maxidle_ = (maxidle * nsecPerByte) / 8;
    270 	if (cl->maxidle_ == 0)
    271 		cl->maxidle_ = 1;
    272 #if 1 /* offtime is also scaled in ALTQ */
    273 	cl->avgidle_ = cl->maxidle_;
    274 	cl->offtime_ = ((offtime * nsecPerByte) / 8) >> RM_FILTER_GAIN;
    275 	if (cl->offtime_ == 0)
    276 		cl->offtime_ = 1;
    277 #else
    278 	cl->avgidle_ = 0;
    279 	cl->offtime_ = (offtime * nsecPerByte) / 8;
    280 #endif
    281 	cl->overlimit = action;
    282 
    283 #ifdef ALTQ_RED
    284 	if (flags & (RMCF_RED|RMCF_RIO)) {
    285 		int red_flags, red_pkttime;
    286 
    287 		red_flags = 0;
    288 		if (flags & RMCF_ECN)
    289 			red_flags |= REDF_ECN;
    290 		if (flags & RMCF_FLOWVALVE)
    291 			red_flags |= REDF_FLOWVALVE;
    292 #ifdef ALTQ_RIO
    293 		if (flags & RMCF_CLEARDSCP)
    294 			red_flags |= RIOF_CLEARDSCP;
    295 #endif
    296 		red_pkttime = nsecPerByte * pktsize  / 1000;
    297 
    298 		if (flags & RMCF_RED) {
    299 			cl->red_ = red_alloc(0, 0, 0, 0,
    300 					     red_flags, red_pkttime);
    301 			if (cl->red_ != NULL)
    302 				qtype(cl->q_) = Q_RED;
    303 		}
    304 #ifdef ALTQ_RIO
    305 		else {
    306 			cl->red_ = (red_t *)rio_alloc(0, NULL,
    307 						      red_flags, red_pkttime);
    308 			if (cl->red_ != NULL)
    309 				qtype(cl->q_) = Q_RIO;
    310 		}
    311 #endif
    312 	}
    313 #endif /* ALTQ_RED */
    314 
    315 	/*
    316 	 * put the class into the class tree
    317 	 */
    318 	s = splnet();
    319 	if ((peer = ifd->active_[pri]) != NULL) {
    320 		/* find the last class at this pri */
    321 		cl->peer_ = peer;
    322 		while (peer->peer_ != ifd->active_[pri])
    323 			peer = peer->peer_;
    324 		peer->peer_ = cl;
    325 	} else {
    326 		ifd->active_[pri] = cl;
    327 		cl->peer_ = cl;
    328 	}
    329 
    330 	if (cl->parent_) {
    331 		cl->next_ = parent->children_;
    332 		parent->children_ = cl;
    333 		parent->leaf_ = 0;
    334 	}
    335 
    336 	/*
    337 	 * Compute the depth of this class and it's ancestors in the class
    338 	 * hierarchy.
    339 	 */
    340 	rmc_depth_compute(cl);
    341 
    342 	/*
    343 	 * If CBQ's WRR is enabled, then initailize the class WRR state.
    344 	 */
    345 	if (ifd->wrr_) {
    346 		ifd->num_[pri]++;
    347 		ifd->alloc_[pri] += cl->allotment_;
    348 		rmc_wrr_set_weights(ifd);
    349 	}
    350 	splx(s);
    351 	return (cl);
    352 }
    353 
    354 int
    355 rmc_modclass(cl, nsecPerByte, maxq, maxidle, minidle, offtime, pktsize)
    356 	struct rm_class *cl;
    357 	u_int		nsecPerByte;
    358 	int		maxq;
    359 	u_int		maxidle;
    360 	int		minidle;
    361 	u_int		offtime;
    362 	int		pktsize;
    363 {
    364 	struct rm_ifdat	*ifd;
    365 	u_int old_allotment;
    366 	int s;
    367 
    368 	ifd = cl->ifdat_;
    369 	old_allotment = cl->allotment_;
    370 
    371 	s = splnet();
    372 	cl->allotment_ = RM_NS_PER_SEC / nsecPerByte; /* Bytes per sec */
    373 	cl->qthresh_ = 0;
    374 	cl->ns_per_byte_ = nsecPerByte;
    375 
    376 	qlimit(cl->q_) = maxq;
    377 
    378 #if 1 /* minidle is also scaled in ALTQ */
    379 	cl->minidle_ = (minidle * nsecPerByte) / 8;
    380 	if (cl->minidle_ > 0)
    381 		cl->minidle_ = 0;
    382 #else
    383 	cl->minidle_ = minidle;
    384 #endif
    385 	cl->maxidle_ = (maxidle * nsecPerByte) / 8;
    386 	if (cl->maxidle_ == 0)
    387 		cl->maxidle_ = 1;
    388 #if 1 /* offtime is also scaled in ALTQ */
    389 	cl->avgidle_ = cl->maxidle_;
    390 	cl->offtime_ = ((offtime * nsecPerByte) / 8) >> RM_FILTER_GAIN;
    391 	if (cl->offtime_ == 0)
    392 		cl->offtime_ = 1;
    393 #else
    394 	cl->avgidle_ = 0;
    395 	cl->offtime_ = (offtime * nsecPerByte) / 8;
    396 #endif
    397 
    398 	/*
    399 	 * If CBQ's WRR is enabled, then initailize the class WRR state.
    400 	 */
    401 	if (ifd->wrr_) {
    402 		ifd->alloc_[cl->pri_] += cl->allotment_ - old_allotment;
    403 		rmc_wrr_set_weights(ifd);
    404 	}
    405 	splx(s);
    406 	return (0);
    407 }
    408 
    409 /*
    410  * static void
    411  * rmc_wrr_set_weights(struct rm_ifdat *ifdat) - This function computes
    412  * 	the appropriate run robin weights for the CBQ weighted round robin
    413  *	algorithm.
    414  *
    415  *	Returns: NONE
    416  */
    417 
    418 static void
    419 rmc_wrr_set_weights(ifd)
    420 	struct rm_ifdat *ifd;
    421 {
    422 	int		i;
    423 	struct rm_class	*cl, *clh;
    424 
    425 	for (i = 0; i < RM_MAXPRIO; i++) {
    426 		/*
    427 		 * This is inverted from that of the simulator to
    428 		 * maintain precision.
    429 		 */
    430 		if (ifd->num_[i] == 0)
    431 			ifd->M_[i] = 0;
    432 		else
    433 			ifd->M_[i] = ifd->alloc_[i] /
    434 				(ifd->num_[i] * ifd->maxpkt_);
    435 		/*
    436 		 * Compute the weigthed allotment for each class.
    437 		 * This takes the expensive div instruction out
    438 		 * of the main loop for the wrr scheduling path.
    439 		 * These only get recomputed when a class comes or
    440 		 * goes.
    441 		 */
    442 		if (ifd->active_[i] != NULL) {
    443 			clh = cl = ifd->active_[i];
    444 			do {
    445 				/* safe-guard for slow link or alloc_ == 0 */
    446 				if (ifd->M_[i] == 0)
    447 					cl->w_allotment_ = 0;
    448 				else
    449 					cl->w_allotment_ = cl->allotment_ /
    450 						ifd->M_[i];
    451 				cl = cl->peer_;
    452 			} while ((cl != NULL) && (cl != clh));
    453 		}
    454 	}
    455 }
    456 
    457 int
    458 rmc_get_weight(ifd, pri)
    459 	struct rm_ifdat *ifd;
    460 	int pri;
    461 {
    462 	if ((pri >= 0) && (pri < RM_MAXPRIO))
    463 		return (ifd->M_[pri]);
    464 	else
    465 		return (0);
    466 }
    467 
    468 /*
    469  * static void
    470  * rmc_depth_compute(struct rm_class *cl) - This function computes the
    471  * 	appropriate depth of class 'cl' and its ancestors.
    472  *
    473  *	Returns:	NONE
    474  */
    475 
    476 static void
    477 rmc_depth_compute(cl)
    478 	struct rm_class *cl;
    479 {
    480 	rm_class_t *t = cl, *p;
    481 
    482 	/*
    483 	 * Recompute the depth for the branch of the tree.
    484 	 */
    485 	while (t != NULL) {
    486 		p = t->parent_;
    487 		if (p && (t->depth_ >= p->depth_)) {
    488 			p->depth_ = t->depth_ + 1;
    489 			t = p;
    490 		} else
    491 			t = NULL;
    492 	}
    493 }
    494 
    495 /*
    496  * static void
    497  * rmc_depth_recompute(struct rm_class *cl) - This function re-computes
    498  *	the depth of the tree after a class has been deleted.
    499  *
    500  *	Returns: 	NONE
    501  */
    502 
    503 static void
    504 rmc_depth_recompute(rm_class_t *cl)
    505 {
    506 #if 1 /* ALTQ */
    507 	rm_class_t	*p, *t;
    508 
    509 	p = cl;
    510 	while (p != NULL) {
    511 		if ((t = p->children_) == NULL) {
    512 			p->depth_ = 0;
    513 		} else {
    514 			int cdepth = 0;
    515 
    516 			while (t != NULL) {
    517 				if (t->depth_ > cdepth)
    518 					cdepth = t->depth_;
    519 				t = t->next_;
    520 			}
    521 
    522 			if (p->depth_ == cdepth + 1)
    523 				/* no change to this parent */
    524 				return;
    525 
    526 			p->depth_ = cdepth + 1;
    527 		}
    528 
    529 		p = p->parent_;
    530 	}
    531 #else
    532 	rm_class_t	*t;
    533 
    534 	if (cl->depth_ >= 1) {
    535 		if (cl->children_ == NULL) {
    536 			cl->depth_ = 0;
    537 		} else if ((t = cl->children_) != NULL) {
    538 			while (t != NULL) {
    539 				if (t->children_ != NULL)
    540 					rmc_depth_recompute(t);
    541 				t = t->next_;
    542 			}
    543 		} else
    544 			rmc_depth_compute(cl);
    545 	}
    546 #endif
    547 }
    548 
    549 /*
    550  * void
    551  * rmc_delete_class(struct rm_ifdat *ifdat, struct rm_class *cl) - This
    552  *	function deletes a class from the link-sharing stucture and frees
    553  *	all resources associated with the class.
    554  *
    555  *	Returns: NONE
    556  */
    557 
    558 void
    559 rmc_delete_class(ifd, cl)
    560 	struct rm_ifdat *ifd;
    561 	struct rm_class *cl;
    562 {
    563 	struct rm_class	*p, *head, *previous;
    564 	int		s;
    565 
    566 	ASSERT(cl->children_ == NULL);
    567 
    568 	if (cl->sleeping_)
    569 		CALLOUT_STOP(&cl->callout_);
    570 
    571 	s = splnet();
    572 	/*
    573 	 * Free packets in the packet queue.
    574 	 * XXX - this may not be a desired behavior.  Packets should be
    575 	 * 		re-queued.
    576 	 */
    577 	rmc_dropall(cl);
    578 
    579 	/*
    580 	 * If the class has a parent, then remove the class from the
    581 	 * class from the parent's children chain.
    582 	 */
    583 	if (cl->parent_ != NULL) {
    584 		head = cl->parent_->children_;
    585 		p = previous = head;
    586 		if (head->next_ == NULL) {
    587 			ASSERT(head == cl);
    588 			cl->parent_->children_ = NULL;
    589 			cl->parent_->leaf_ = 1;
    590 		} else while (p != NULL) {
    591 			if (p == cl) {
    592 				if (cl == head)
    593 					cl->parent_->children_ = cl->next_;
    594 				else
    595 					previous->next_ = cl->next_;
    596 				cl->next_ = NULL;
    597 				p = NULL;
    598 			} else {
    599 				previous = p;
    600 				p = p->next_;
    601 			}
    602 		}
    603 	}
    604 
    605 	/*
    606 	 * Delete class from class priority peer list.
    607 	 */
    608 	if ((p = ifd->active_[cl->pri_]) != NULL) {
    609 		/*
    610 		 * If there is more than one member of this priority
    611 		 * level, then look for class(cl) in the priority level.
    612 		 */
    613 		if (p != p->peer_) {
    614 			while (p->peer_ != cl)
    615 				p = p->peer_;
    616 			p->peer_ = cl->peer_;
    617 
    618 			if (ifd->active_[cl->pri_] == cl)
    619 				ifd->active_[cl->pri_] = cl->peer_;
    620 		} else {
    621 			ASSERT(p == cl);
    622 			ifd->active_[cl->pri_] = NULL;
    623 		}
    624 	}
    625 
    626 	/*
    627 	 * Recompute the WRR weights.
    628 	 */
    629 	if (ifd->wrr_) {
    630 		ifd->alloc_[cl->pri_] -= cl->allotment_;
    631 		ifd->num_[cl->pri_]--;
    632 		rmc_wrr_set_weights(ifd);
    633 	}
    634 
    635 	/*
    636 	 * Re-compute the depth of the tree.
    637 	 */
    638 #if 1 /* ALTQ */
    639 	rmc_depth_recompute(cl->parent_);
    640 #else
    641 	rmc_depth_recompute(ifd->root_);
    642 #endif
    643 
    644 	splx(s);
    645 
    646 	/*
    647 	 * Free the class structure.
    648 	 */
    649 	if (cl->red_ != NULL) {
    650 #ifdef ALTQ_RIO
    651 		if (q_is_rio(cl->q_))
    652 			rio_destroy((rio_t *)cl->red_);
    653 #endif
    654 #ifdef ALTQ_RED
    655 		if (q_is_red(cl->q_))
    656 			red_destroy(cl->red_);
    657 #endif
    658 	}
    659 	FREE(cl->q_, M_DEVBUF);
    660 	FREE(cl, M_DEVBUF);
    661 }
    662 
    663 
    664 /*
    665  * void
    666  * rmc_init(...) - Initialize the resource management data structures
    667  *	associated with the output portion of interface 'ifp'.  'ifd' is
    668  *	where the structures will be built (for backwards compatibility, the
    669  *	structures aren't kept in the ifnet struct).  'nsecPerByte'
    670  *	gives the link speed (inverse of bandwidth) in nanoseconds/byte.
    671  *	'restart' is the driver-specific routine that the generic 'delay
    672  *	until under limit' action will call to restart output.  `maxq'
    673  *	is the queue size of the 'link' & 'default' classes.  'maxqueued'
    674  *	is the maximum number of packets that the resource management
    675  *	code will allow to be queued 'downstream' (this is typically 1).
    676  *
    677  *	Returns:	NONE
    678  */
    679 
    680 void
    681 rmc_init(ifq, ifd, nsecPerByte, restart, maxq, maxqueued, maxidle,
    682 	 minidle, offtime, flags)
    683 	struct ifaltq	*ifq;
    684 	struct rm_ifdat *ifd;
    685 	u_int	nsecPerByte;
    686 	void	(*restart)(struct ifaltq *);
    687 	int	maxq, maxqueued;
    688 	u_int	maxidle;
    689 	int	minidle;
    690 	u_int	offtime;
    691 	int	flags;
    692 {
    693 	int		i, mtu;
    694 
    695 	/*
    696 	 * Initialize the CBQ traciing/debug facility.
    697 	 */
    698 	CBQTRACEINIT();
    699 
    700 	bzero((char *)ifd, sizeof (*ifd));
    701 	mtu = ifq->altq_ifp->if_mtu;
    702 	ifd->ifq_ = ifq;
    703 	ifd->restart = restart;
    704 	ifd->maxqueued_ = maxqueued;
    705 	ifd->ns_per_byte_ = nsecPerByte;
    706 	ifd->maxpkt_ = mtu;
    707 	ifd->wrr_ = (flags & RMCF_WRR) ? 1 : 0;
    708 	ifd->efficient_ = (flags & RMCF_EFFICIENT) ? 1 : 0;
    709 #if 1
    710 	ifd->maxiftime_ = mtu * nsecPerByte / 1000 * 16;
    711 	if (mtu * nsecPerByte > 10 * 1000000)
    712 		ifd->maxiftime_ /= 4;
    713 #endif
    714 
    715 	reset_cutoff(ifd);
    716 	CBQTRACE(rmc_init, 'INIT', ifd->cutoff_);
    717 
    718 	/*
    719 	 * Initialize the CBQ's WRR state.
    720 	 */
    721 	for (i = 0; i < RM_MAXPRIO; i++) {
    722 		ifd->alloc_[i] = 0;
    723 		ifd->M_[i] = 0;
    724 		ifd->num_[i] = 0;
    725 		ifd->na_[i] = 0;
    726 		ifd->active_[i] = NULL;
    727 	}
    728 
    729 	/*
    730 	 * Initialize current packet state.
    731 	 */
    732 	ifd->qi_ = 0;
    733 	ifd->qo_ = 0;
    734 	for (i = 0; i < RM_MAXQUEUED; i++) {
    735 		ifd->class_[i] = NULL;
    736 		ifd->curlen_[i] = 0;
    737 		ifd->borrowed_[i] = NULL;
    738 	}
    739 
    740 	/*
    741 	 * Create the root class of the link-sharing structure.
    742 	 */
    743 	if ((ifd->root_ = rmc_newclass(0, ifd,
    744 				       nsecPerByte,
    745 				       rmc_root_overlimit, maxq, 0, 0,
    746 				       maxidle, minidle, offtime,
    747 				       0, 0)) == NULL) {
    748 		printf("rmc_init: root class not allocated\n");
    749 		return ;
    750 	}
    751 	ifd->root_->depth_ = 0;
    752 }
    753 
    754 /*
    755  * void
    756  * rmc_queue_packet(struct rm_class *cl, mbuf_t *m) - Add packet given by
    757  *	mbuf 'm' to queue for resource class 'cl'.  This routine is called
    758  *	by a driver's if_output routine.  This routine must be called with
    759  *	output packet completion interrupts locked out (to avoid racing with
    760  *	rmc_dequeue_next).
    761  *
    762  *	Returns:	0 on successful queueing
    763  *			-1 when packet drop occurs
    764  */
    765 int
    766 rmc_queue_packet(cl, m)
    767 	struct rm_class *cl;
    768 	mbuf_t *m;
    769 {
    770 	struct timeval	now;
    771 	struct rm_ifdat *ifd = cl->ifdat_;
    772 	int		cpri = cl->pri_;
    773 	int		is_empty = qempty(cl->q_);
    774 
    775 	RM_GETTIME(now);
    776 	if (ifd->cutoff_ > 0) {
    777 		if (TV_LT(&cl->undertime_, &now)) {
    778 			if (ifd->cutoff_ > cl->depth_)
    779 				ifd->cutoff_ = cl->depth_;
    780 			CBQTRACE(rmc_queue_packet, 'ffoc', cl->depth_);
    781 		}
    782 #if 1 /* ALTQ */
    783 		else {
    784 			/*
    785 			 * the class is overlimit. if the class has
    786 			 * underlimit ancestors, set cutoff to the lowest
    787 			 * depth among them.
    788 			 */
    789 			struct rm_class *borrow = cl->borrow_;
    790 
    791 			while (borrow != NULL &&
    792 			       borrow->depth_ < ifd->cutoff_) {
    793 				if (TV_LT(&borrow->undertime_, &now)) {
    794 					ifd->cutoff_ = borrow->depth_;
    795 					CBQTRACE(rmc_queue_packet, 'ffob', ifd->cutoff_);
    796 					break;
    797 				}
    798 				borrow = borrow->borrow_;
    799 			}
    800 		}
    801 #else /* !ALTQ */
    802 		else if ((ifd->cutoff_ > 1) && cl->borrow_) {
    803 			if (TV_LT(&cl->borrow_->undertime_, &now)) {
    804 				ifd->cutoff_ = cl->borrow_->depth_;
    805 				CBQTRACE(rmc_queue_packet, 'ffob',
    806 					 cl->borrow_->depth_);
    807 			}
    808 		}
    809 #endif /* !ALTQ */
    810 	}
    811 
    812 	if (_rmc_addq(cl, m) < 0)
    813 		/* failed */
    814 		return (-1);
    815 
    816 	if (is_empty) {
    817 		CBQTRACE(rmc_queue_packet, 'ytpe', cl->stats_.handle);
    818 		ifd->na_[cpri]++;
    819 	}
    820 
    821 	if (qlen(cl->q_) > qlimit(cl->q_)) {
    822 		/* note: qlimit can be set to 0 or 1 */
    823 		rmc_drop_action(cl);
    824 		return (-1);
    825 	}
    826 	return (0);
    827 }
    828 
    829 /*
    830  * void
    831  * rmc_tl_satisfied(struct rm_ifdat *ifd, struct timeval *now) - Check all
    832  *	classes to see if there are satified.
    833  */
    834 
    835 static void
    836 rmc_tl_satisfied(ifd, now)
    837 	struct rm_ifdat *ifd;
    838 	struct timeval *now;
    839 {
    840 	int	i;
    841 	rm_class_t	*p, *bp;
    842 
    843 	for (i = RM_MAXPRIO - 1; i >= 0; i--) {
    844 		if ((bp = ifd->active_[i]) != NULL) {
    845 			p = bp;
    846 			do {
    847 				if (!rmc_satisfied(p, now)) {
    848 					ifd->cutoff_ = p->depth_;
    849 					return;
    850 				}
    851 				p = p->peer_;
    852 			} while (p != bp);
    853 		}
    854 	}
    855 
    856 	reset_cutoff(ifd);
    857 }
    858 
    859 /*
    860  * rmc_satisfied - Return 1 of the class is satisfied.  O, otherwise.
    861  */
    862 
    863 static int
    864 rmc_satisfied(cl, now)
    865 	struct rm_class *cl;
    866 	struct timeval *now;
    867 {
    868 	rm_class_t 	*p;
    869 
    870 	if (cl == NULL)
    871 		return (1);
    872 	if (TV_LT(now, &cl->undertime_))
    873 		return (1);
    874 	if (cl->depth_ == 0) {
    875 		if (!cl->sleeping_ && (qlen(cl->q_) > cl->qthresh_))
    876 			return (0);
    877 		else
    878 			return (1);
    879 	}
    880 	if (cl->children_ != NULL) {
    881 		p = cl->children_;
    882 		while (p != NULL) {
    883 			if (!rmc_satisfied(p, now))
    884 				return (0);
    885 			p = p->next_;
    886 		}
    887 	}
    888 
    889 	return (1);
    890 }
    891 
    892 /*
    893  * Return 1 if class 'cl' is under limit or can borrow from a parent,
    894  * 0 if overlimit.  As a side-effect, this routine will invoke the
    895  * class overlimit action if the class if overlimit.
    896  */
    897 
    898 static int
    899 rmc_under_limit(cl, now)
    900 	struct rm_class *cl;
    901 	struct timeval *now;
    902 {
    903 	rm_class_t	*p = cl;
    904 	rm_class_t	*top;
    905 	struct rm_ifdat	*ifd = cl->ifdat_;
    906 
    907 	ifd->borrowed_[ifd->qi_] = NULL;
    908 	/*
    909 	 * If cl is the root class, then always return that it is
    910 	 * underlimit.  Otherwise, check to see if the class is underlimit.
    911 	 */
    912 	if (cl->parent_ == NULL)
    913 		return (1);
    914 
    915 	if (cl->sleeping_) {
    916 		if (TV_LT(now, &cl->undertime_))
    917 			return (0);
    918 
    919 		CALLOUT_STOP(&cl->callout_);
    920 		cl->sleeping_ = 0;
    921 		cl->undertime_.tv_sec = 0;
    922 		return (1);
    923 	}
    924 
    925 	top = NULL;
    926 	while (cl->undertime_.tv_sec && TV_LT(now, &cl->undertime_)) {
    927 		if (((cl = cl->borrow_) == NULL) ||
    928 		    (cl->depth_ > ifd->cutoff_)) {
    929 #ifdef ADJUST_CUTOFF
    930 			if (cl != NULL)
    931 				/* cutoff is taking effect, just
    932 				   return false without calling
    933 				   the delay action. */
    934 				return (0);
    935 #endif
    936 #ifdef BORROW_OFFTIME
    937 			/*
    938 			 * check if the class can borrow offtime too.
    939 			 * borrow offtime from the top of the borrow
    940 			 * chain if the top class is not overloaded.
    941 			 */
    942 			if (cl != NULL) {
    943 				/* cutoff is taking effect, use this class as top. */
    944 				top = cl;
    945 				CBQTRACE(rmc_under_limit, 'ffou', ifd->cutoff_);
    946 			}
    947 			if (top != NULL && top->avgidle_ == top->minidle_)
    948 				top = NULL;
    949 			p->overtime_ = *now;
    950 			(p->overlimit)(p, top);
    951 #else
    952 			p->overtime_ = *now;
    953 			(p->overlimit)(p, NULL);
    954 #endif
    955 			return (0);
    956 		}
    957 		top = cl;
    958 	}
    959 
    960 	if (cl != p)
    961 		ifd->borrowed_[ifd->qi_] = cl;
    962 	return (1);
    963 }
    964 
    965 /*
    966  * _rmc_wrr_dequeue_next() - This is scheduler for WRR as opposed to
    967  *	Packet-by-packet round robin.
    968  *
    969  * The heart of the weigthed round-robin scheduler, which decides which
    970  * class next gets to send a packet.  Highest priority first, then
    971  * weighted round-robin within priorites.
    972  *
    973  * Each able-to-send class gets to send until its byte allocation is
    974  * exhausted.  Thus, the active pointer is only changed after a class has
    975  * exhausted its allocation.
    976  *
    977  * If the scheduler finds no class that is underlimit or able to borrow,
    978  * then the first class found that had a nonzero queue and is allowed to
    979  * borrow gets to send.
    980  */
    981 
    982 static mbuf_t *
    983 _rmc_wrr_dequeue_next(ifd, op)
    984 	struct rm_ifdat *ifd;
    985 	int op;
    986 {
    987 	struct rm_class	*cl = NULL, *first = NULL;
    988 	u_int		deficit;
    989 	int		cpri;
    990 	mbuf_t		*m;
    991 	struct timeval	now;
    992 
    993 	RM_GETTIME(now);
    994 
    995 	/*
    996 	 * if the driver polls the top of the queue and then removes
    997 	 * the polled packet, we must return the same packet.
    998 	 */
    999 	if (op == ALTDQ_REMOVE && ifd->pollcache_) {
   1000 		cl = ifd->pollcache_;
   1001 		cpri = cl->pri_;
   1002 		if (ifd->efficient_) {
   1003 			/* check if this class is overlimit */
   1004 			if (cl->undertime_.tv_sec != 0 &&
   1005 			    rmc_under_limit(cl, &now) == 0)
   1006 				first = cl;
   1007 		}
   1008 		ifd->pollcache_ = NULL;
   1009 		goto _wrr_out;
   1010 	}
   1011 	else {
   1012 		/* mode == ALTDQ_POLL || pollcache == NULL */
   1013 		ifd->pollcache_ = NULL;
   1014 		ifd->borrowed_[ifd->qi_] = NULL;
   1015 	}
   1016 #ifdef ADJUST_CUTOFF
   1017  _again:
   1018 #endif
   1019 	for (cpri = RM_MAXPRIO - 1; cpri >= 0; cpri--) {
   1020 		if (ifd->na_[cpri] == 0)
   1021 			continue;
   1022 		deficit = 0;
   1023 		/*
   1024 		 * Loop through twice for a priority level, if some class
   1025 		 * was unable to send a packet the first round because
   1026 		 * of the weighted round-robin mechanism.
   1027 		 * During the second loop at this level, deficit==2.
   1028 		 * (This second loop is not needed if for every class,
   1029 		 * "M[cl->pri_])" times "cl->allotment" is greater than
   1030 		 * the byte size for the largest packet in the class.)
   1031 		 */
   1032  _wrr_loop:
   1033 		cl = ifd->active_[cpri];
   1034 		ASSERT(cl != NULL);
   1035 		do {
   1036 			if ((deficit < 2) && (cl->bytes_alloc_ <= 0))
   1037 				cl->bytes_alloc_ += cl->w_allotment_;
   1038 			if (!qempty(cl->q_)) {
   1039 				if ((cl->undertime_.tv_sec == 0) ||
   1040 				    rmc_under_limit(cl, &now)) {
   1041 					if (cl->bytes_alloc_ > 0 || deficit > 1)
   1042 						goto _wrr_out;
   1043 
   1044 					/* underlimit but no alloc */
   1045 					deficit = 1;
   1046 #if 1
   1047 					ifd->borrowed_[ifd->qi_] = NULL;
   1048 #endif
   1049 				}
   1050 				else if (first == NULL && cl->borrow_ != NULL)
   1051 					first = cl; /* borrowing candidate */
   1052 			}
   1053 
   1054 			cl->bytes_alloc_ = 0;
   1055 			cl = cl->peer_;
   1056 		} while (cl != ifd->active_[cpri]);
   1057 
   1058 		if (deficit == 1) {
   1059 			/* first loop found an underlimit class with deficit */
   1060 			/* Loop on same priority level, with new deficit.  */
   1061 			deficit = 2;
   1062 			goto _wrr_loop;
   1063 		}
   1064 	}
   1065 
   1066 #ifdef ADJUST_CUTOFF
   1067 	/*
   1068 	 * no underlimit class found.  if cutoff is taking effect,
   1069 	 * increase cutoff and try again.
   1070 	 */
   1071 	if (first != NULL && ifd->cutoff_ < ifd->root_->depth_) {
   1072 		ifd->cutoff_++;
   1073 		CBQTRACE(_rmc_wrr_dequeue_next, 'ojda', ifd->cutoff_);
   1074 		goto _again;
   1075 	}
   1076 #endif /* ADJUST_CUTOFF */
   1077 	/*
   1078 	 * If LINK_EFFICIENCY is turned on, then the first overlimit
   1079 	 * class we encounter will send a packet if all the classes
   1080 	 * of the link-sharing structure are overlimit.
   1081 	 */
   1082 	reset_cutoff(ifd);
   1083 	CBQTRACE(_rmc_wrr_dequeue_next, 'otsr', ifd->cutoff_);
   1084 
   1085 	if (!ifd->efficient_ || first == NULL)
   1086 		return (NULL);
   1087 
   1088 	cl = first;
   1089 	cpri = cl->pri_;
   1090 #if 0	/* too time-consuming for nothing */
   1091 	if (cl->sleeping_)
   1092 		CALLOUT_STOP(&cl->callout_);
   1093 	cl->sleeping_ = 0;
   1094 	cl->undertime_.tv_sec = 0;
   1095 #endif
   1096 	ifd->borrowed_[ifd->qi_] = cl->borrow_;
   1097 	ifd->cutoff_ = cl->borrow_->depth_;
   1098 
   1099 	/*
   1100 	 * Deque the packet and do the book keeping...
   1101 	 */
   1102  _wrr_out:
   1103 	if (op == ALTDQ_REMOVE) {
   1104 		m = _rmc_getq(cl);
   1105 		if (m == NULL)
   1106 			panic("_rmc_wrr_dequeue_next");
   1107 		if (qempty(cl->q_))
   1108 			ifd->na_[cpri]--;
   1109 
   1110 		/*
   1111 		 * Update class statistics and link data.
   1112 		 */
   1113 		if (cl->bytes_alloc_ > 0)
   1114 			cl->bytes_alloc_ -= m_pktlen(m);
   1115 
   1116 		if ((cl->bytes_alloc_ <= 0) || first == cl)
   1117 			ifd->active_[cl->pri_] = cl->peer_;
   1118 		else
   1119 			ifd->active_[cl->pri_] = cl;
   1120 
   1121 		ifd->class_[ifd->qi_] = cl;
   1122 		ifd->curlen_[ifd->qi_] = m_pktlen(m);
   1123 		ifd->now_[ifd->qi_] = now;
   1124 		ifd->qi_ = (ifd->qi_ + 1) % ifd->maxqueued_;
   1125 		ifd->queued_++;
   1126 	} else {
   1127 		/* mode == ALTDQ_PPOLL */
   1128 		m = _rmc_pollq(cl);
   1129 		ifd->pollcache_ = cl;
   1130 	}
   1131 	return (m);
   1132 }
   1133 
   1134 /*
   1135  * Dequeue & return next packet from the highest priority class that
   1136  * has a packet to send & has enough allocation to send it.  This
   1137  * routine is called by a driver whenever it needs a new packet to
   1138  * output.
   1139  */
   1140 static mbuf_t *
   1141 _rmc_prr_dequeue_next(ifd, op)
   1142 	struct rm_ifdat *ifd;
   1143 	int op;
   1144 {
   1145 	mbuf_t		*m;
   1146 	int		cpri;
   1147 	struct rm_class	*cl, *first = NULL;
   1148 	struct timeval	now;
   1149 
   1150 	RM_GETTIME(now);
   1151 
   1152 	/*
   1153 	 * if the driver polls the top of the queue and then removes
   1154 	 * the polled packet, we must return the same packet.
   1155 	 */
   1156 	if (op == ALTDQ_REMOVE && ifd->pollcache_) {
   1157 		cl = ifd->pollcache_;
   1158 		cpri = cl->pri_;
   1159 		ifd->pollcache_ = NULL;
   1160 		goto _prr_out;
   1161 	} else {
   1162 		/* mode == ALTDQ_POLL || pollcache == NULL */
   1163 		ifd->pollcache_ = NULL;
   1164 		ifd->borrowed_[ifd->qi_] = NULL;
   1165 	}
   1166 #ifdef ADJUST_CUTOFF
   1167  _again:
   1168 #endif
   1169 	for (cpri = RM_MAXPRIO - 1; cpri >= 0; cpri--) {
   1170 		if (ifd->na_[cpri] == 0)
   1171 			continue;
   1172 		cl = ifd->active_[cpri];
   1173 		ASSERT(cl != NULL);
   1174 		do {
   1175 			if (!qempty(cl->q_)) {
   1176 				if ((cl->undertime_.tv_sec == 0) ||
   1177 				    rmc_under_limit(cl, &now))
   1178 					goto _prr_out;
   1179 				if (first == NULL && cl->borrow_ != NULL)
   1180 					first = cl;
   1181 			}
   1182 			cl = cl->peer_;
   1183 		} while (cl != ifd->active_[cpri]);
   1184 	}
   1185 
   1186 #ifdef ADJUST_CUTOFF
   1187 	/*
   1188 	 * no underlimit class found.  if cutoff is taking effect, increase
   1189 	 * cutoff and try again.
   1190 	 */
   1191 	if (first != NULL && ifd->cutoff_ < ifd->root_->depth_) {
   1192 		ifd->cutoff_++;
   1193 		goto _again;
   1194 	}
   1195 #endif /* ADJUST_CUTOFF */
   1196 	/*
   1197 	 * If LINK_EFFICIENCY is turned on, then the first overlimit
   1198 	 * class we encounter will send a packet if all the classes
   1199 	 * of the link-sharing structure are overlimit.
   1200 	 */
   1201 	reset_cutoff(ifd);
   1202 	if (!ifd->efficient_ || first == NULL)
   1203 		return (NULL);
   1204 
   1205 	cl = first;
   1206 	cpri = cl->pri_;
   1207 #if 0	/* too time-consuming for nothing */
   1208 	if (cl->sleeping_)
   1209 		CALLOUT_STOP(&cl->callout_);
   1210 	cl->sleeping_ = 0;
   1211 	cl->undertime_.tv_sec = 0;
   1212 #endif
   1213 	ifd->borrowed_[ifd->qi_] = cl->borrow_;
   1214 	ifd->cutoff_ = cl->borrow_->depth_;
   1215 
   1216 	/*
   1217 	 * Deque the packet and do the book keeping...
   1218 	 */
   1219  _prr_out:
   1220 	if (op == ALTDQ_REMOVE) {
   1221 		m = _rmc_getq(cl);
   1222 		if (m == NULL)
   1223 			panic("_rmc_prr_dequeue_next");
   1224 		if (qempty(cl->q_))
   1225 			ifd->na_[cpri]--;
   1226 
   1227 		ifd->active_[cpri] = cl->peer_;
   1228 
   1229 		ifd->class_[ifd->qi_] = cl;
   1230 		ifd->curlen_[ifd->qi_] = m_pktlen(m);
   1231 		ifd->now_[ifd->qi_] = now;
   1232 		ifd->qi_ = (ifd->qi_ + 1) % ifd->maxqueued_;
   1233 		ifd->queued_++;
   1234 	} else {
   1235 		/* mode == ALTDQ_POLL */
   1236 		m = _rmc_pollq(cl);
   1237 		ifd->pollcache_ = cl;
   1238 	}
   1239 	return (m);
   1240 }
   1241 
   1242 /*
   1243  * mbuf_t *
   1244  * rmc_dequeue_next(struct rm_ifdat *ifd, struct timeval *now) - this function
   1245  *	is invoked by the packet driver to get the next packet to be
   1246  *	dequeued and output on the link.  If WRR is enabled, then the
   1247  *	WRR dequeue next routine will determine the next packet to sent.
   1248  *	Otherwise, packet-by-packet round robin is invoked.
   1249  *
   1250  *	Returns:	NULL, if a packet is not available or if all
   1251  *			classes are overlimit.
   1252  *
   1253  *			Otherwise, Pointer to the next packet.
   1254  */
   1255 
   1256 mbuf_t *
   1257 rmc_dequeue_next(ifd, mode)
   1258 	struct rm_ifdat *ifd;
   1259 	int mode;
   1260 {
   1261 	if (ifd->queued_ >= ifd->maxqueued_)
   1262 		return (NULL);
   1263 	else if (ifd->wrr_)
   1264 		return (_rmc_wrr_dequeue_next(ifd, mode));
   1265 	else
   1266 		return (_rmc_prr_dequeue_next(ifd, mode));
   1267 }
   1268 
   1269 /*
   1270  * Update the utilization estimate for the packet that just completed.
   1271  * The packet's class & the parent(s) of that class all get their
   1272  * estimators updated.  This routine is called by the driver's output-
   1273  * packet-completion interrupt service routine.
   1274  */
   1275 
   1276 /*
   1277  * a macro to approximate "divide by 1000" that gives 0.000999,
   1278  * if a value has enough effective digits.
   1279  * (on pentium, mul takes 9 cycles but div takes 46!)
   1280  */
   1281 #define	NSEC_TO_USEC(t)	(((t) >> 10) + ((t) >> 16) + ((t) >> 17))
   1282 void
   1283 rmc_update_class_util(ifd)
   1284 	struct rm_ifdat *ifd;
   1285 {
   1286 	int		idle, avgidle, pktlen;
   1287 	int 		pkt_time, tidle;
   1288 	rm_class_t	*cl, *borrowed;
   1289 	rm_class_t	*borrows;
   1290 	struct timeval	*nowp;
   1291 
   1292 	/*
   1293 	 * Get the most recent completed class.
   1294 	 */
   1295 	if ((cl = ifd->class_[ifd->qo_]) == NULL)
   1296 		return;
   1297 
   1298 	pktlen = ifd->curlen_[ifd->qo_];
   1299 	borrowed = ifd->borrowed_[ifd->qo_];
   1300 	borrows = borrowed;
   1301 
   1302 	PKTCNTR_ADD(&cl->stats_.xmit_cnt, pktlen);
   1303 
   1304 	/*
   1305 	 * Run estimator on class and it's ancesstors.
   1306 	 */
   1307 	/*
   1308 	 * rm_update_class_util is designed to be called when the
   1309 	 * transfer is completed from a xmit complete interrupt,
   1310 	 * but most drivers don't implement an upcall for that.
   1311 	 * so, just use estimated completion time.
   1312 	 * as a result, ifd->qi_ and ifd->qo_ are always synced.
   1313 	 */
   1314 	nowp = &ifd->now_[ifd->qo_];
   1315 	/* get pkt_time (for link) in usec */
   1316 #if 1  /* use approximation */
   1317 	pkt_time = ifd->curlen_[ifd->qo_] * ifd->ns_per_byte_;
   1318 	pkt_time = NSEC_TO_USEC(pkt_time);
   1319 #else
   1320 	pkt_time = ifd->curlen_[ifd->qo_] * ifd->ns_per_byte_ / 1000;
   1321 #endif
   1322 #if 1 /* ALTQ4PPP */
   1323 	if (TV_LT(nowp, &ifd->ifnow_)) {
   1324 		int iftime;
   1325 
   1326 		/*
   1327 		 * make sure the estimated completion time does not go
   1328 		 * too far.  it can happen when the link layer supports
   1329 		 * data compression or the interface speed is set to
   1330 		 * a much lower value.
   1331 		 */
   1332 		TV_DELTA(&ifd->ifnow_, nowp, iftime);
   1333 		if (iftime+pkt_time < ifd->maxiftime_) {
   1334 			TV_ADD_DELTA(&ifd->ifnow_, pkt_time, &ifd->ifnow_);
   1335 		} else {
   1336 			TV_ADD_DELTA(nowp, ifd->maxiftime_, &ifd->ifnow_);
   1337 		}
   1338 	} else {
   1339 		TV_ADD_DELTA(nowp, pkt_time, &ifd->ifnow_);
   1340 	}
   1341 #else
   1342 	if (TV_LT(nowp, &ifd->ifnow_)) {
   1343 		TV_ADD_DELTA(&ifd->ifnow_, pkt_time, &ifd->ifnow_);
   1344 	} else {
   1345 		TV_ADD_DELTA(nowp, pkt_time, &ifd->ifnow_);
   1346 	}
   1347 #endif
   1348 
   1349 	while (cl != NULL) {
   1350 		TV_DELTA(&ifd->ifnow_, &cl->last_, idle);
   1351 		if (idle >= 2000000)
   1352 			/*
   1353 			 * this class is idle enough, reset avgidle.
   1354 			 * (TV_DELTA returns 2000000 us when delta is large.)
   1355 			 */
   1356 			cl->avgidle_ = cl->maxidle_;
   1357 
   1358 		/* get pkt_time (for class) in usec */
   1359 #if 1  /* use approximation */
   1360 		pkt_time = pktlen * cl->ns_per_byte_;
   1361 		pkt_time = NSEC_TO_USEC(pkt_time);
   1362 #else
   1363 		pkt_time = pktlen * cl->ns_per_byte_ / 1000;
   1364 #endif
   1365 		idle -= pkt_time;
   1366 
   1367 		avgidle = cl->avgidle_;
   1368 		avgidle += idle - (avgidle >> RM_FILTER_GAIN);
   1369 		cl->avgidle_ = avgidle;
   1370 
   1371 		/* Are we overlimit ? */
   1372 		if (avgidle <= 0) {
   1373 			CBQTRACE(rmc_update_class_util, 'milo', cl->stats_.handle);
   1374 #if 1 /* ALTQ */
   1375 			/*
   1376 			 * need some lower bound for avgidle, otherwise
   1377 			 * a borrowing class gets unbounded penalty.
   1378 			 */
   1379 			if (avgidle < cl->minidle_)
   1380 				avgidle = cl->avgidle_ = cl->minidle_;
   1381 #endif
   1382 			/* set next idle to make avgidle 0 */
   1383 			tidle = pkt_time +
   1384 				(((1 - RM_POWER) * avgidle) >> RM_FILTER_GAIN);
   1385 			TV_ADD_DELTA(nowp, tidle, &cl->undertime_);
   1386 			++cl->stats_.over;
   1387 		} else {
   1388 			cl->avgidle_ =
   1389 			    (avgidle > cl->maxidle_) ? cl->maxidle_ : avgidle;
   1390 			cl->undertime_.tv_sec = 0;
   1391 			if (cl->sleeping_) {
   1392 				CALLOUT_STOP(&cl->callout_);
   1393 				cl->sleeping_ = 0;
   1394 			}
   1395 		}
   1396 
   1397 		if (borrows != NULL) {
   1398 			if (borrows != cl)
   1399 				++cl->stats_.borrows;
   1400 			else
   1401 				borrows = NULL;
   1402 		}
   1403 		cl->last_ = ifd->ifnow_;
   1404 		cl->last_pkttime_ = pkt_time;
   1405 
   1406 #if 1
   1407 		if (cl->parent_ == NULL) {
   1408 			/* take stats of root class */
   1409 			PKTCNTR_ADD(&cl->stats_.xmit_cnt, pktlen);
   1410 		}
   1411 #endif
   1412 
   1413 		cl = cl->parent_;
   1414 	}
   1415 
   1416 	/*
   1417 	 * Check to see if cutoff needs to set to a new level.
   1418 	 */
   1419 	cl = ifd->class_[ifd->qo_];
   1420 	if (borrowed && (ifd->cutoff_ >= borrowed->depth_)) {
   1421 #if 1 /* ALTQ */
   1422 		if ((qlen(cl->q_) <= 0) || TV_LT(nowp, &borrowed->undertime_)) {
   1423 			rmc_tl_satisfied(ifd, nowp);
   1424 			CBQTRACE(rmc_update_class_util, 'broe', ifd->cutoff_);
   1425 		} else {
   1426 			ifd->cutoff_ = borrowed->depth_;
   1427 			CBQTRACE(rmc_update_class_util, 'ffob', borrowed->depth_);
   1428 		}
   1429 #else /* !ALTQ */
   1430 		if ((qlen(cl->q_) <= 1) || TV_LT(&now, &borrowed->undertime_)) {
   1431 			reset_cutoff(ifd);
   1432 #ifdef notdef
   1433 			rmc_tl_satisfied(ifd, &now);
   1434 #endif
   1435 			CBQTRACE(rmc_update_class_util, 'broe', ifd->cutoff_);
   1436 		} else {
   1437 			ifd->cutoff_ = borrowed->depth_;
   1438 			CBQTRACE(rmc_update_class_util, 'ffob', borrowed->depth_);
   1439 		}
   1440 #endif /* !ALTQ */
   1441 	}
   1442 
   1443 	/*
   1444 	 * Release class slot
   1445 	 */
   1446 	ifd->borrowed_[ifd->qo_] = NULL;
   1447 	ifd->class_[ifd->qo_] = NULL;
   1448 	ifd->qo_ = (ifd->qo_ + 1) % ifd->maxqueued_;
   1449 	ifd->queued_--;
   1450 }
   1451 
   1452 /*
   1453  * void
   1454  * rmc_drop_action(struct rm_class *cl) - Generic (not protocol-specific)
   1455  *	over-limit action routines.  These get invoked by rmc_under_limit()
   1456  *	if a class with packets to send if over its bandwidth limit & can't
   1457  *	borrow from a parent class.
   1458  *
   1459  *	Returns: NONE
   1460  */
   1461 
   1462 static void
   1463 rmc_drop_action(cl)
   1464 	struct rm_class *cl;
   1465 {
   1466 	struct rm_ifdat	*ifd = cl->ifdat_;
   1467 
   1468 	ASSERT(qlen(cl->q_) > 0);
   1469 	_rmc_dropq(cl);
   1470 	if (qempty(cl->q_))
   1471 		ifd->na_[cl->pri_]--;
   1472 }
   1473 
   1474 void rmc_dropall(cl)
   1475     struct rm_class *cl;
   1476 {
   1477 	struct rm_ifdat *ifd = cl->ifdat_;
   1478 
   1479 	if (!qempty(cl->q_)) {
   1480 		_flushq(cl->q_);
   1481 
   1482 		ifd->na_[cl->pri_]--;
   1483 	}
   1484 }
   1485 
   1486 #if (__FreeBSD_version > 300000)
   1487 /* hzto() is removed from FreeBSD-3.0 */
   1488 static int hzto __P((struct timeval *));
   1489 
   1490 static int
   1491 hzto(tv)
   1492 	struct timeval *tv;
   1493 {
   1494 	struct timeval t2;
   1495 
   1496 	getmicrotime(&t2);
   1497 	t2.tv_sec = tv->tv_sec - t2.tv_sec;
   1498 	t2.tv_usec = tv->tv_usec - t2.tv_usec;
   1499 	return (tvtohz(&t2));
   1500 }
   1501 #endif /* __FreeBSD_version > 300000 */
   1502 
   1503 /*
   1504  * void
   1505  * rmc_delay_action(struct rm_class *cl) - This function is the generic CBQ
   1506  *	delay action routine.  It is invoked via rmc_under_limit when the
   1507  *	packet is discoverd to be overlimit.
   1508  *
   1509  *	If the delay action is result of borrow class being overlimit, then
   1510  *	delay for the offtime of the borrowing class that is overlimit.
   1511  *
   1512  *	Returns: NONE
   1513  */
   1514 
   1515 void
   1516 rmc_delay_action(cl, borrow)
   1517 	struct rm_class *cl, *borrow;
   1518 {
   1519 	int	delay, t, extradelay;
   1520 
   1521 	cl->stats_.overactions++;
   1522 	TV_DELTA(&cl->undertime_, &cl->overtime_, delay);
   1523 #ifndef BORROW_OFFTIME
   1524 	delay += cl->offtime_;
   1525 #endif
   1526 
   1527 	if (!cl->sleeping_) {
   1528 		CBQTRACE(rmc_delay_action, 'yled', cl->stats_.handle);
   1529 #ifdef BORROW_OFFTIME
   1530 		if (borrow != NULL)
   1531 			extradelay = borrow->offtime_;
   1532 		else
   1533 #endif
   1534 			extradelay = cl->offtime_;
   1535 
   1536 #ifdef ALTQ
   1537 		/*
   1538 		 * XXX recalculate suspend time:
   1539 		 * current undertime is (tidle + pkt_time) calculated
   1540 		 * from the last transmission.
   1541 		 *	tidle: time required to bring avgidle back to 0
   1542 		 *	pkt_time: target waiting time for this class
   1543 		 * we need to replace pkt_time by offtime
   1544 		 */
   1545 		extradelay -= cl->last_pkttime_;
   1546 #endif
   1547 		if (extradelay > 0) {
   1548 			TV_ADD_DELTA(&cl->undertime_, extradelay, &cl->undertime_);
   1549 			delay += extradelay;
   1550 		}
   1551 
   1552 		cl->sleeping_ = 1;
   1553 		cl->stats_.delays++;
   1554 
   1555 		/*
   1556 		 * Since packets are phased randomly with respect to the
   1557 		 * clock, 1 tick (the next clock tick) can be an arbitrarily
   1558 		 * short time so we have to wait for at least two ticks.
   1559 		 * NOTE:  If there's no other traffic, we need the timer as
   1560 		 * a 'backstop' to restart this class.
   1561 		 */
   1562 		if (delay > tick * 2) {
   1563 #ifdef __FreeBSD__
   1564 			/* FreeBSD rounds up the tick */
   1565 			t = hzto(&cl->undertime_);
   1566 #else
   1567 			/* other BSDs round down the tick */
   1568 			t = hzto(&cl->undertime_) + 1;
   1569 #endif
   1570 		} else
   1571 			t = 2;
   1572 		CALLOUT_RESET(&cl->callout_, t,
   1573 			      (timeout_t *)rmc_restart, (caddr_t)cl);
   1574 	}
   1575 }
   1576 
   1577 /*
   1578  * void
   1579  * rmc_restart() - is just a helper routine for rmc_delay_action -- it is
   1580  *	called by the system timer code & is responsible checking if the
   1581  *	class is still sleeping (it might have been restarted as a side
   1582  *	effect of the queue scan on a packet arrival) and, if so, restarting
   1583  *	output for the class.  Inspecting the class state & restarting output
   1584  *	require locking the class structure.  In general the driver is
   1585  *	responsible for locking but this is the only routine that is not
   1586  *	called directly or indirectly from the interface driver so it has
   1587  *	know about system locking conventions.  Under bsd, locking is done
   1588  *	by raising IPL to splnet so that's what's implemented here.  On a
   1589  *	different system this would probably need to be changed.
   1590  *
   1591  *	Returns:	NONE
   1592  */
   1593 
   1594 static void
   1595 rmc_restart(cl)
   1596 	struct rm_class *cl;
   1597 {
   1598 	struct rm_ifdat *ifd = cl->ifdat_;
   1599 	int s;
   1600 
   1601 	s = splnet();
   1602 	if (cl->sleeping_) {
   1603 		cl->sleeping_ = 0;
   1604 		cl->undertime_.tv_sec = 0;
   1605 
   1606 		if (ifd->queued_ < ifd->maxqueued_ && ifd->restart != NULL) {
   1607 			CBQTRACE(rmc_restart, 'trts', cl->stats_.handle);
   1608 			(ifd->restart)(ifd->ifq_);
   1609 		}
   1610 	}
   1611 	splx(s);
   1612 }
   1613 
   1614 /*
   1615  * void
   1616  * rmc_root_overlimit(struct rm_class *cl) - This the generic overlimit
   1617  *	handling routine for the root class of the link sharing structure.
   1618  *
   1619  *	Returns: NONE
   1620  */
   1621 
   1622 static void
   1623 rmc_root_overlimit(cl, borrow)
   1624 	struct rm_class *cl, *borrow;
   1625 {
   1626     panic("rmc_root_overlimit");
   1627 }
   1628 
   1629 /*
   1630  * Packet Queue handling routines.  Eventually, this is to localize the
   1631  *	effects on the code whether queues are red queues or droptail
   1632  *	queues.
   1633  */
   1634 
   1635 static int
   1636 _rmc_addq(cl, m)
   1637 	rm_class_t *cl;
   1638 	mbuf_t *m;
   1639 {
   1640 #ifdef ALTQ_RIO
   1641 	if (q_is_rio(cl->q_))
   1642 		return rio_addq((rio_t *)cl->red_, cl->q_, m, cl->pktattr_);
   1643 #endif
   1644 #ifdef ALTQ_RED
   1645 	if (q_is_red(cl->q_))
   1646 		return red_addq(cl->red_, cl->q_, m, cl->pktattr_);
   1647 #endif /* ALTQ_RED */
   1648 
   1649 	if (cl->flags_ & RMCF_CLEARDSCP)
   1650 		write_dsfield(m, cl->pktattr_, 0);
   1651 
   1652 	_addq(cl->q_, m);
   1653 	return (0);
   1654 }
   1655 
   1656 /* note: _rmc_dropq is not called for red */
   1657 static void
   1658 _rmc_dropq(cl)
   1659 	rm_class_t *cl;
   1660 {
   1661 	mbuf_t  *m;
   1662 
   1663 	if ((m = _getq(cl->q_)) != NULL)
   1664 		m_freem(m);
   1665 }
   1666 
   1667 static mbuf_t *
   1668 _rmc_getq(cl)
   1669 	rm_class_t *cl;
   1670 {
   1671 #ifdef ALTQ_RIO
   1672 	if (q_is_rio(cl->q_))
   1673 		return rio_getq((rio_t *)cl->red_, cl->q_);
   1674 #endif
   1675 #ifdef ALTQ_RED
   1676 	if (q_is_red(cl->q_))
   1677 		return red_getq(cl->red_, cl->q_);
   1678 #endif
   1679 	return _getq(cl->q_);
   1680 }
   1681 
   1682 static mbuf_t *
   1683 _rmc_pollq(cl)
   1684 	rm_class_t *cl;
   1685 {
   1686 	return qhead(cl->q_);
   1687 }
   1688 
   1689 #ifdef CBQ_TRACE
   1690 
   1691 struct cbqtrace		cbqtrace_buffer[NCBQTRACE+1];
   1692 struct cbqtrace		*cbqtrace_ptr = NULL;
   1693 int			cbqtrace_count;
   1694 
   1695 /*
   1696  * DDB hook to trace cbq events:
   1697  *  the last 1024 events are held in a circular buffer.
   1698  *  use "call cbqtrace_dump(N)" to display 20 events from Nth event.
   1699  */
   1700 void cbqtrace_dump(int);
   1701 static char *rmc_funcname(void *);
   1702 
   1703 static struct rmc_funcs {
   1704 	void *func;
   1705 	char *name;
   1706 } rmc_funcs[] =
   1707 {
   1708 	rmc_init, 		"rmc_init",
   1709 	rmc_queue_packet, 	"rmc_queue_packet",
   1710 	rmc_under_limit, 	"rmc_under_limit",
   1711 	rmc_update_class_util, 	"rmc_update_class_util",
   1712 	rmc_delay_action, 	"rmc_delay_action",
   1713 	rmc_restart, 		"rmc_restart",
   1714 	_rmc_wrr_dequeue_next, 	"_rmc_wrr_dequeue_next",
   1715 	NULL, 			NULL
   1716 };
   1717 
   1718 static char *rmc_funcname(func)
   1719 	void *func;
   1720 {
   1721 	struct rmc_funcs *fp;
   1722 
   1723 	for (fp = rmc_funcs; fp->func != NULL; fp++)
   1724 		if (fp->func == func)
   1725 			return (fp->name);
   1726 	return ("unknown");
   1727 }
   1728 
   1729 void cbqtrace_dump(counter)
   1730 	int counter;
   1731 {
   1732 	int i, *p;
   1733 	char *cp;
   1734 
   1735 	counter = counter % NCBQTRACE;
   1736 	p = (int *)&cbqtrace_buffer[counter];
   1737 
   1738 	for (i=0; i<20; i++) {
   1739 		printf("[0x%x] ", *p++);
   1740 		printf("%s: ", rmc_funcname((void *)*p++));
   1741 		cp = (char *)p++;
   1742 		printf("%c%c%c%c: ", cp[0], cp[1], cp[2], cp[3]);
   1743 		printf("%d\n",*p++);
   1744 
   1745 		if (p >= (int *)&cbqtrace_buffer[NCBQTRACE])
   1746 			p = (int *)cbqtrace_buffer;
   1747 	}
   1748 }
   1749 #endif /* CBQ_TRACE */
   1750 
   1751 #endif /* ALTQ_CBQ */
   1752 
   1753 #if defined(ALTQ_CBQ) || defined(ALTQ_RED) || defined(ALTQ_RIO) || defined(ALTQ_HFSC) || defined(ALTQ_PRIQ)
   1754 #if !defined(__GNUC__) || defined(ALTQ_DEBUG)
   1755 
   1756 void
   1757 _addq(q, m)
   1758 	class_queue_t *q;
   1759 	mbuf_t *m;
   1760 {
   1761         mbuf_t  *m0;
   1762 
   1763 	if ((m0 = qtail(q)) != NULL)
   1764 		m->m_nextpkt = m0->m_nextpkt;
   1765 	else
   1766 		m0 = m;
   1767 	m0->m_nextpkt = m;
   1768 	qtail(q) = m;
   1769 	qlen(q)++;
   1770 }
   1771 
   1772 mbuf_t *
   1773 _getq(q)
   1774 	class_queue_t *q;
   1775 {
   1776 	mbuf_t  *m, *m0;
   1777 
   1778 	if ((m = qtail(q)) == NULL)
   1779 		return (NULL);
   1780 	if ((m0 = m->m_nextpkt) != m)
   1781 		m->m_nextpkt = m0->m_nextpkt;
   1782 	else {
   1783 		ASSERT(qlen(q) == 1);
   1784 		qtail(q) = NULL;
   1785 	}
   1786 	qlen(q)--;
   1787 	m0->m_nextpkt = NULL;
   1788 	return (m0);
   1789 }
   1790 
   1791 /* drop a packet at the tail of the queue */
   1792 mbuf_t *
   1793 _getq_tail(q)
   1794 	class_queue_t *q;
   1795 {
   1796 	mbuf_t *m, *m0, *prev;
   1797 
   1798 	if ((m = m0 = qtail(q)) == NULL)
   1799 		return NULL;
   1800 	do {
   1801 		prev = m0;
   1802 		m0 = m0->m_nextpkt;
   1803 	} while (m0 != m);
   1804 	prev->m_nextpkt = m->m_nextpkt;
   1805 	if (prev == m)  {
   1806 		ASSERT(qlen(q) == 1);
   1807 		qtail(q) = NULL;
   1808 	} else
   1809 		qtail(q) = prev;
   1810 	qlen(q)--;
   1811 	m->m_nextpkt = NULL;
   1812 	return (m);
   1813 }
   1814 
   1815 /* randomly select a packet in the queue */
   1816 mbuf_t *
   1817 _getq_random(q)
   1818 	class_queue_t *q;
   1819 {
   1820 	struct mbuf *m;
   1821 	int i, n;
   1822 
   1823 	if ((m = qtail(q)) == NULL)
   1824 		return NULL;
   1825 	if (m->m_nextpkt == m) {
   1826 		ASSERT(qlen(q) == 1);
   1827 		qtail(q) = NULL;
   1828 	} else {
   1829 		struct mbuf *prev = NULL;
   1830 
   1831 		n = random() % qlen(q) + 1;
   1832 		for (i = 0; i < n; i++) {
   1833 			prev = m;
   1834 			m = m->m_nextpkt;
   1835 		}
   1836 		prev->m_nextpkt = m->m_nextpkt;
   1837 		if (m == qtail(q))
   1838 			qtail(q) = prev;
   1839 	}
   1840 	qlen(q)--;
   1841 	m->m_nextpkt = NULL;
   1842 	return (m);
   1843 }
   1844 
   1845 void
   1846 _removeq(q, m)
   1847 	class_queue_t *q;
   1848 	mbuf_t *m;
   1849 {
   1850 	mbuf_t *m0, *prev;
   1851 
   1852 	m0 = qtail(q);
   1853 	do {
   1854 		prev = m0;
   1855 		m0 = m0->m_nextpkt;
   1856 	} while (m0 != m);
   1857 	prev->m_nextpkt = m->m_nextpkt;
   1858 	if (prev == m)
   1859 		qtail(q) = NULL;
   1860 	else if (qtail(q) == m)
   1861 		qtail(q) = prev;
   1862 	qlen(q)--;
   1863 }
   1864 
   1865 void
   1866 _flushq(q)
   1867 	class_queue_t *q;
   1868 {
   1869 	mbuf_t *m;
   1870 
   1871 	while ((m = _getq(q)) != NULL)
   1872 		m_freem(m);
   1873 	ASSERT(qlen(q) == 0);
   1874 }
   1875 
   1876 #endif /* !__GNUC__ || ALTQ_DEBUG */
   1877 #endif /* ALTQ_CBQ || ALTQ_RED || ALTQ_RIO || ALTQ_HFSC || ALTQ_PRIQ */
   1878