[NET_SCHED]: Remove unnecessary stats_lock pointers
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sched / sch_hfsc.c
blob7130a2441b0ddd6c8c820d29a48975fa411c2a95
1 /*
2 * Copyright (c) 2003 Patrick McHardy, <kaber@trash.net>
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version 2
7 * of the License, or (at your option) any later version.
9 * 2003-10-17 - Ported from altq
12 * Copyright (c) 1997-1999 Carnegie Mellon University. All Rights Reserved.
14 * Permission to use, copy, modify, and distribute this software and
15 * its documentation is hereby granted (including for commercial or
16 * for-profit use), provided that both the copyright notice and this
17 * permission notice appear in all copies of the software, derivative
18 * works, or modified versions, and any portions thereof.
20 * THIS SOFTWARE IS EXPERIMENTAL AND IS KNOWN TO HAVE BUGS, SOME OF
21 * WHICH MAY HAVE SERIOUS CONSEQUENCES. CARNEGIE MELLON PROVIDES THIS
22 * SOFTWARE IN ITS ``AS IS'' CONDITION, AND ANY EXPRESS OR IMPLIED
23 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
25 * DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
28 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
29 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
30 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
32 * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
33 * DAMAGE.
35 * Carnegie Mellon encourages (but does not require) users of this
36 * software to return any improvements or extensions that they make,
37 * and to grant Carnegie Mellon the rights to redistribute these
38 * changes without encumbrance.
41 * H-FSC is described in Proceedings of SIGCOMM'97,
42 * "A Hierarchical Fair Service Curve Algorithm for Link-Sharing,
43 * Real-Time and Priority Service"
44 * by Ion Stoica, Hui Zhang, and T. S. Eugene Ng.
46 * Oleg Cherevko <olwi@aq.ml.com.ua> added the upperlimit for link-sharing.
47 * when a class has an upperlimit, the fit-time is computed from the
48 * upperlimit service curve. the link-sharing scheduler does not schedule
49 * a class whose fit-time exceeds the current time.
52 #include <linux/kernel.h>
53 #include <linux/module.h>
54 #include <linux/types.h>
55 #include <linux/errno.h>
56 #include <linux/jiffies.h>
57 #include <linux/compiler.h>
58 #include <linux/spinlock.h>
59 #include <linux/skbuff.h>
60 #include <linux/string.h>
61 #include <linux/slab.h>
62 #include <linux/list.h>
63 #include <linux/rbtree.h>
64 #include <linux/init.h>
65 #include <linux/netdevice.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/pkt_sched.h>
68 #include <net/netlink.h>
69 #include <net/pkt_sched.h>
70 #include <net/pkt_cls.h>
71 #include <asm/system.h>
72 #include <asm/div64.h>
75 * kernel internal service curve representation:
76 * coordinates are given by 64 bit unsigned integers.
77 * x-axis: unit is clock count.
78 * y-axis: unit is byte.
80 * The service curve parameters are converted to the internal
81 * representation. The slope values are scaled to avoid overflow.
82 * the inverse slope values as well as the y-projection of the 1st
83 * segment are kept in order to to avoid 64-bit divide operations
84 * that are expensive on 32-bit architectures.
87 struct internal_sc
89 u64 sm1; /* scaled slope of the 1st segment */
90 u64 ism1; /* scaled inverse-slope of the 1st segment */
91 u64 dx; /* the x-projection of the 1st segment */
92 u64 dy; /* the y-projection of the 1st segment */
93 u64 sm2; /* scaled slope of the 2nd segment */
94 u64 ism2; /* scaled inverse-slope of the 2nd segment */
97 /* runtime service curve */
98 struct runtime_sc
100 u64 x; /* current starting position on x-axis */
101 u64 y; /* current starting position on y-axis */
102 u64 sm1; /* scaled slope of the 1st segment */
103 u64 ism1; /* scaled inverse-slope of the 1st segment */
104 u64 dx; /* the x-projection of the 1st segment */
105 u64 dy; /* the y-projection of the 1st segment */
106 u64 sm2; /* scaled slope of the 2nd segment */
107 u64 ism2; /* scaled inverse-slope of the 2nd segment */
110 enum hfsc_class_flags
112 HFSC_RSC = 0x1,
113 HFSC_FSC = 0x2,
114 HFSC_USC = 0x4
117 struct hfsc_class
119 u32 classid; /* class id */
120 unsigned int refcnt; /* usage count */
122 struct gnet_stats_basic bstats;
123 struct gnet_stats_queue qstats;
124 struct gnet_stats_rate_est rate_est;
125 unsigned int level; /* class level in hierarchy */
126 struct tcf_proto *filter_list; /* filter list */
127 unsigned int filter_cnt; /* filter count */
129 struct hfsc_sched *sched; /* scheduler data */
130 struct hfsc_class *cl_parent; /* parent class */
131 struct list_head siblings; /* sibling classes */
132 struct list_head children; /* child classes */
133 struct Qdisc *qdisc; /* leaf qdisc */
135 struct rb_node el_node; /* qdisc's eligible tree member */
136 struct rb_root vt_tree; /* active children sorted by cl_vt */
137 struct rb_node vt_node; /* parent's vt_tree member */
138 struct rb_root cf_tree; /* active children sorted by cl_f */
139 struct rb_node cf_node; /* parent's cf_heap member */
140 struct list_head hlist; /* hash list member */
141 struct list_head dlist; /* drop list member */
143 u64 cl_total; /* total work in bytes */
144 u64 cl_cumul; /* cumulative work in bytes done by
145 real-time criteria */
147 u64 cl_d; /* deadline*/
148 u64 cl_e; /* eligible time */
149 u64 cl_vt; /* virtual time */
150 u64 cl_f; /* time when this class will fit for
151 link-sharing, max(myf, cfmin) */
152 u64 cl_myf; /* my fit-time (calculated from this
153 class's own upperlimit curve) */
154 u64 cl_myfadj; /* my fit-time adjustment (to cancel
155 history dependence) */
156 u64 cl_cfmin; /* earliest children's fit-time (used
157 with cl_myf to obtain cl_f) */
158 u64 cl_cvtmin; /* minimal virtual time among the
159 children fit for link-sharing
160 (monotonic within a period) */
161 u64 cl_vtadj; /* intra-period cumulative vt
162 adjustment */
163 u64 cl_vtoff; /* inter-period cumulative vt offset */
164 u64 cl_cvtmax; /* max child's vt in the last period */
165 u64 cl_cvtoff; /* cumulative cvtmax of all periods */
166 u64 cl_pcvtoff; /* parent's cvtoff at initalization
167 time */
169 struct internal_sc cl_rsc; /* internal real-time service curve */
170 struct internal_sc cl_fsc; /* internal fair service curve */
171 struct internal_sc cl_usc; /* internal upperlimit service curve */
172 struct runtime_sc cl_deadline; /* deadline curve */
173 struct runtime_sc cl_eligible; /* eligible curve */
174 struct runtime_sc cl_virtual; /* virtual curve */
175 struct runtime_sc cl_ulimit; /* upperlimit curve */
177 unsigned long cl_flags; /* which curves are valid */
178 unsigned long cl_vtperiod; /* vt period sequence number */
179 unsigned long cl_parentperiod;/* parent's vt period sequence number*/
180 unsigned long cl_nactive; /* number of active children */
183 #define HFSC_HSIZE 16
185 struct hfsc_sched
187 u16 defcls; /* default class id */
188 struct hfsc_class root; /* root class */
189 struct list_head clhash[HFSC_HSIZE]; /* class hash */
190 struct rb_root eligible; /* eligible tree */
191 struct list_head droplist; /* active leaf class list (for
192 dropping) */
193 struct sk_buff_head requeue; /* requeued packet */
194 struct qdisc_watchdog watchdog; /* watchdog timer */
197 #define HT_INFINITY 0xffffffffffffffffULL /* infinite time value */
201 * eligible tree holds backlogged classes being sorted by their eligible times.
202 * there is one eligible tree per hfsc instance.
205 static void
206 eltree_insert(struct hfsc_class *cl)
208 struct rb_node **p = &cl->sched->eligible.rb_node;
209 struct rb_node *parent = NULL;
210 struct hfsc_class *cl1;
212 while (*p != NULL) {
213 parent = *p;
214 cl1 = rb_entry(parent, struct hfsc_class, el_node);
215 if (cl->cl_e >= cl1->cl_e)
216 p = &parent->rb_right;
217 else
218 p = &parent->rb_left;
220 rb_link_node(&cl->el_node, parent, p);
221 rb_insert_color(&cl->el_node, &cl->sched->eligible);
224 static inline void
225 eltree_remove(struct hfsc_class *cl)
227 rb_erase(&cl->el_node, &cl->sched->eligible);
230 static inline void
231 eltree_update(struct hfsc_class *cl)
233 eltree_remove(cl);
234 eltree_insert(cl);
237 /* find the class with the minimum deadline among the eligible classes */
238 static inline struct hfsc_class *
239 eltree_get_mindl(struct hfsc_sched *q, u64 cur_time)
241 struct hfsc_class *p, *cl = NULL;
242 struct rb_node *n;
244 for (n = rb_first(&q->eligible); n != NULL; n = rb_next(n)) {
245 p = rb_entry(n, struct hfsc_class, el_node);
246 if (p->cl_e > cur_time)
247 break;
248 if (cl == NULL || p->cl_d < cl->cl_d)
249 cl = p;
251 return cl;
254 /* find the class with minimum eligible time among the eligible classes */
255 static inline struct hfsc_class *
256 eltree_get_minel(struct hfsc_sched *q)
258 struct rb_node *n;
260 n = rb_first(&q->eligible);
261 if (n == NULL)
262 return NULL;
263 return rb_entry(n, struct hfsc_class, el_node);
267 * vttree holds holds backlogged child classes being sorted by their virtual
268 * time. each intermediate class has one vttree.
270 static void
271 vttree_insert(struct hfsc_class *cl)
273 struct rb_node **p = &cl->cl_parent->vt_tree.rb_node;
274 struct rb_node *parent = NULL;
275 struct hfsc_class *cl1;
277 while (*p != NULL) {
278 parent = *p;
279 cl1 = rb_entry(parent, struct hfsc_class, vt_node);
280 if (cl->cl_vt >= cl1->cl_vt)
281 p = &parent->rb_right;
282 else
283 p = &parent->rb_left;
285 rb_link_node(&cl->vt_node, parent, p);
286 rb_insert_color(&cl->vt_node, &cl->cl_parent->vt_tree);
289 static inline void
290 vttree_remove(struct hfsc_class *cl)
292 rb_erase(&cl->vt_node, &cl->cl_parent->vt_tree);
295 static inline void
296 vttree_update(struct hfsc_class *cl)
298 vttree_remove(cl);
299 vttree_insert(cl);
302 static inline struct hfsc_class *
303 vttree_firstfit(struct hfsc_class *cl, u64 cur_time)
305 struct hfsc_class *p;
306 struct rb_node *n;
308 for (n = rb_first(&cl->vt_tree); n != NULL; n = rb_next(n)) {
309 p = rb_entry(n, struct hfsc_class, vt_node);
310 if (p->cl_f <= cur_time)
311 return p;
313 return NULL;
317 * get the leaf class with the minimum vt in the hierarchy
319 static struct hfsc_class *
320 vttree_get_minvt(struct hfsc_class *cl, u64 cur_time)
322 /* if root-class's cfmin is bigger than cur_time nothing to do */
323 if (cl->cl_cfmin > cur_time)
324 return NULL;
326 while (cl->level > 0) {
327 cl = vttree_firstfit(cl, cur_time);
328 if (cl == NULL)
329 return NULL;
331 * update parent's cl_cvtmin.
333 if (cl->cl_parent->cl_cvtmin < cl->cl_vt)
334 cl->cl_parent->cl_cvtmin = cl->cl_vt;
336 return cl;
339 static void
340 cftree_insert(struct hfsc_class *cl)
342 struct rb_node **p = &cl->cl_parent->cf_tree.rb_node;
343 struct rb_node *parent = NULL;
344 struct hfsc_class *cl1;
346 while (*p != NULL) {
347 parent = *p;
348 cl1 = rb_entry(parent, struct hfsc_class, cf_node);
349 if (cl->cl_f >= cl1->cl_f)
350 p = &parent->rb_right;
351 else
352 p = &parent->rb_left;
354 rb_link_node(&cl->cf_node, parent, p);
355 rb_insert_color(&cl->cf_node, &cl->cl_parent->cf_tree);
358 static inline void
359 cftree_remove(struct hfsc_class *cl)
361 rb_erase(&cl->cf_node, &cl->cl_parent->cf_tree);
364 static inline void
365 cftree_update(struct hfsc_class *cl)
367 cftree_remove(cl);
368 cftree_insert(cl);
372 * service curve support functions
374 * external service curve parameters
375 * m: bps
376 * d: us
377 * internal service curve parameters
378 * sm: (bytes/psched_us) << SM_SHIFT
379 * ism: (psched_us/byte) << ISM_SHIFT
380 * dx: psched_us
382 * The clock source resolution with ktime is 1.024us.
384 * sm and ism are scaled in order to keep effective digits.
385 * SM_SHIFT and ISM_SHIFT are selected to keep at least 4 effective
386 * digits in decimal using the following table.
388 * bits/sec 100Kbps 1Mbps 10Mbps 100Mbps 1Gbps
389 * ------------+-------------------------------------------------------
390 * bytes/1.024us 12.8e-3 128e-3 1280e-3 12800e-3 128000e-3
392 * 1.024us/byte 78.125 7.8125 0.78125 0.078125 0.0078125
394 #define SM_SHIFT 20
395 #define ISM_SHIFT 18
397 #define SM_MASK ((1ULL << SM_SHIFT) - 1)
398 #define ISM_MASK ((1ULL << ISM_SHIFT) - 1)
400 static inline u64
401 seg_x2y(u64 x, u64 sm)
403 u64 y;
406 * compute
407 * y = x * sm >> SM_SHIFT
408 * but divide it for the upper and lower bits to avoid overflow
410 y = (x >> SM_SHIFT) * sm + (((x & SM_MASK) * sm) >> SM_SHIFT);
411 return y;
414 static inline u64
415 seg_y2x(u64 y, u64 ism)
417 u64 x;
419 if (y == 0)
420 x = 0;
421 else if (ism == HT_INFINITY)
422 x = HT_INFINITY;
423 else {
424 x = (y >> ISM_SHIFT) * ism
425 + (((y & ISM_MASK) * ism) >> ISM_SHIFT);
427 return x;
430 /* Convert m (bps) into sm (bytes/psched us) */
431 static u64
432 m2sm(u32 m)
434 u64 sm;
436 sm = ((u64)m << SM_SHIFT);
437 sm += PSCHED_TICKS_PER_SEC - 1;
438 do_div(sm, PSCHED_TICKS_PER_SEC);
439 return sm;
442 /* convert m (bps) into ism (psched us/byte) */
443 static u64
444 m2ism(u32 m)
446 u64 ism;
448 if (m == 0)
449 ism = HT_INFINITY;
450 else {
451 ism = ((u64)PSCHED_TICKS_PER_SEC << ISM_SHIFT);
452 ism += m - 1;
453 do_div(ism, m);
455 return ism;
458 /* convert d (us) into dx (psched us) */
459 static u64
460 d2dx(u32 d)
462 u64 dx;
464 dx = ((u64)d * PSCHED_TICKS_PER_SEC);
465 dx += USEC_PER_SEC - 1;
466 do_div(dx, USEC_PER_SEC);
467 return dx;
470 /* convert sm (bytes/psched us) into m (bps) */
471 static u32
472 sm2m(u64 sm)
474 u64 m;
476 m = (sm * PSCHED_TICKS_PER_SEC) >> SM_SHIFT;
477 return (u32)m;
480 /* convert dx (psched us) into d (us) */
481 static u32
482 dx2d(u64 dx)
484 u64 d;
486 d = dx * USEC_PER_SEC;
487 do_div(d, PSCHED_TICKS_PER_SEC);
488 return (u32)d;
491 static void
492 sc2isc(struct tc_service_curve *sc, struct internal_sc *isc)
494 isc->sm1 = m2sm(sc->m1);
495 isc->ism1 = m2ism(sc->m1);
496 isc->dx = d2dx(sc->d);
497 isc->dy = seg_x2y(isc->dx, isc->sm1);
498 isc->sm2 = m2sm(sc->m2);
499 isc->ism2 = m2ism(sc->m2);
503 * initialize the runtime service curve with the given internal
504 * service curve starting at (x, y).
506 static void
507 rtsc_init(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
509 rtsc->x = x;
510 rtsc->y = y;
511 rtsc->sm1 = isc->sm1;
512 rtsc->ism1 = isc->ism1;
513 rtsc->dx = isc->dx;
514 rtsc->dy = isc->dy;
515 rtsc->sm2 = isc->sm2;
516 rtsc->ism2 = isc->ism2;
520 * calculate the y-projection of the runtime service curve by the
521 * given x-projection value
523 static u64
524 rtsc_y2x(struct runtime_sc *rtsc, u64 y)
526 u64 x;
528 if (y < rtsc->y)
529 x = rtsc->x;
530 else if (y <= rtsc->y + rtsc->dy) {
531 /* x belongs to the 1st segment */
532 if (rtsc->dy == 0)
533 x = rtsc->x + rtsc->dx;
534 else
535 x = rtsc->x + seg_y2x(y - rtsc->y, rtsc->ism1);
536 } else {
537 /* x belongs to the 2nd segment */
538 x = rtsc->x + rtsc->dx
539 + seg_y2x(y - rtsc->y - rtsc->dy, rtsc->ism2);
541 return x;
544 static u64
545 rtsc_x2y(struct runtime_sc *rtsc, u64 x)
547 u64 y;
549 if (x <= rtsc->x)
550 y = rtsc->y;
551 else if (x <= rtsc->x + rtsc->dx)
552 /* y belongs to the 1st segment */
553 y = rtsc->y + seg_x2y(x - rtsc->x, rtsc->sm1);
554 else
555 /* y belongs to the 2nd segment */
556 y = rtsc->y + rtsc->dy
557 + seg_x2y(x - rtsc->x - rtsc->dx, rtsc->sm2);
558 return y;
562 * update the runtime service curve by taking the minimum of the current
563 * runtime service curve and the service curve starting at (x, y).
565 static void
566 rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
568 u64 y1, y2, dx, dy;
569 u32 dsm;
571 if (isc->sm1 <= isc->sm2) {
572 /* service curve is convex */
573 y1 = rtsc_x2y(rtsc, x);
574 if (y1 < y)
575 /* the current rtsc is smaller */
576 return;
577 rtsc->x = x;
578 rtsc->y = y;
579 return;
583 * service curve is concave
584 * compute the two y values of the current rtsc
585 * y1: at x
586 * y2: at (x + dx)
588 y1 = rtsc_x2y(rtsc, x);
589 if (y1 <= y) {
590 /* rtsc is below isc, no change to rtsc */
591 return;
594 y2 = rtsc_x2y(rtsc, x + isc->dx);
595 if (y2 >= y + isc->dy) {
596 /* rtsc is above isc, replace rtsc by isc */
597 rtsc->x = x;
598 rtsc->y = y;
599 rtsc->dx = isc->dx;
600 rtsc->dy = isc->dy;
601 return;
605 * the two curves intersect
606 * compute the offsets (dx, dy) using the reverse
607 * function of seg_x2y()
608 * seg_x2y(dx, sm1) == seg_x2y(dx, sm2) + (y1 - y)
610 dx = (y1 - y) << SM_SHIFT;
611 dsm = isc->sm1 - isc->sm2;
612 do_div(dx, dsm);
614 * check if (x, y1) belongs to the 1st segment of rtsc.
615 * if so, add the offset.
617 if (rtsc->x + rtsc->dx > x)
618 dx += rtsc->x + rtsc->dx - x;
619 dy = seg_x2y(dx, isc->sm1);
621 rtsc->x = x;
622 rtsc->y = y;
623 rtsc->dx = dx;
624 rtsc->dy = dy;
625 return;
628 static void
629 init_ed(struct hfsc_class *cl, unsigned int next_len)
631 u64 cur_time = psched_get_time();
633 /* update the deadline curve */
634 rtsc_min(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
637 * update the eligible curve.
638 * for concave, it is equal to the deadline curve.
639 * for convex, it is a linear curve with slope m2.
641 cl->cl_eligible = cl->cl_deadline;
642 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
643 cl->cl_eligible.dx = 0;
644 cl->cl_eligible.dy = 0;
647 /* compute e and d */
648 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
649 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
651 eltree_insert(cl);
654 static void
655 update_ed(struct hfsc_class *cl, unsigned int next_len)
657 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
658 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
660 eltree_update(cl);
663 static inline void
664 update_d(struct hfsc_class *cl, unsigned int next_len)
666 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
669 static inline void
670 update_cfmin(struct hfsc_class *cl)
672 struct rb_node *n = rb_first(&cl->cf_tree);
673 struct hfsc_class *p;
675 if (n == NULL) {
676 cl->cl_cfmin = 0;
677 return;
679 p = rb_entry(n, struct hfsc_class, cf_node);
680 cl->cl_cfmin = p->cl_f;
683 static void
684 init_vf(struct hfsc_class *cl, unsigned int len)
686 struct hfsc_class *max_cl;
687 struct rb_node *n;
688 u64 vt, f, cur_time;
689 int go_active;
691 cur_time = 0;
692 go_active = 1;
693 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
694 if (go_active && cl->cl_nactive++ == 0)
695 go_active = 1;
696 else
697 go_active = 0;
699 if (go_active) {
700 n = rb_last(&cl->cl_parent->vt_tree);
701 if (n != NULL) {
702 max_cl = rb_entry(n, struct hfsc_class,vt_node);
704 * set vt to the average of the min and max
705 * classes. if the parent's period didn't
706 * change, don't decrease vt of the class.
708 vt = max_cl->cl_vt;
709 if (cl->cl_parent->cl_cvtmin != 0)
710 vt = (cl->cl_parent->cl_cvtmin + vt)/2;
712 if (cl->cl_parent->cl_vtperiod !=
713 cl->cl_parentperiod || vt > cl->cl_vt)
714 cl->cl_vt = vt;
715 } else {
717 * first child for a new parent backlog period.
718 * add parent's cvtmax to cvtoff to make a new
719 * vt (vtoff + vt) larger than the vt in the
720 * last period for all children.
722 vt = cl->cl_parent->cl_cvtmax;
723 cl->cl_parent->cl_cvtoff += vt;
724 cl->cl_parent->cl_cvtmax = 0;
725 cl->cl_parent->cl_cvtmin = 0;
726 cl->cl_vt = 0;
729 cl->cl_vtoff = cl->cl_parent->cl_cvtoff -
730 cl->cl_pcvtoff;
732 /* update the virtual curve */
733 vt = cl->cl_vt + cl->cl_vtoff;
734 rtsc_min(&cl->cl_virtual, &cl->cl_fsc, vt,
735 cl->cl_total);
736 if (cl->cl_virtual.x == vt) {
737 cl->cl_virtual.x -= cl->cl_vtoff;
738 cl->cl_vtoff = 0;
740 cl->cl_vtadj = 0;
742 cl->cl_vtperiod++; /* increment vt period */
743 cl->cl_parentperiod = cl->cl_parent->cl_vtperiod;
744 if (cl->cl_parent->cl_nactive == 0)
745 cl->cl_parentperiod++;
746 cl->cl_f = 0;
748 vttree_insert(cl);
749 cftree_insert(cl);
751 if (cl->cl_flags & HFSC_USC) {
752 /* class has upper limit curve */
753 if (cur_time == 0)
754 cur_time = psched_get_time();
756 /* update the ulimit curve */
757 rtsc_min(&cl->cl_ulimit, &cl->cl_usc, cur_time,
758 cl->cl_total);
759 /* compute myf */
760 cl->cl_myf = rtsc_y2x(&cl->cl_ulimit,
761 cl->cl_total);
762 cl->cl_myfadj = 0;
766 f = max(cl->cl_myf, cl->cl_cfmin);
767 if (f != cl->cl_f) {
768 cl->cl_f = f;
769 cftree_update(cl);
770 update_cfmin(cl->cl_parent);
775 static void
776 update_vf(struct hfsc_class *cl, unsigned int len, u64 cur_time)
778 u64 f; /* , myf_bound, delta; */
779 int go_passive = 0;
781 if (cl->qdisc->q.qlen == 0 && cl->cl_flags & HFSC_FSC)
782 go_passive = 1;
784 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
785 cl->cl_total += len;
787 if (!(cl->cl_flags & HFSC_FSC) || cl->cl_nactive == 0)
788 continue;
790 if (go_passive && --cl->cl_nactive == 0)
791 go_passive = 1;
792 else
793 go_passive = 0;
795 if (go_passive) {
796 /* no more active child, going passive */
798 /* update cvtmax of the parent class */
799 if (cl->cl_vt > cl->cl_parent->cl_cvtmax)
800 cl->cl_parent->cl_cvtmax = cl->cl_vt;
802 /* remove this class from the vt tree */
803 vttree_remove(cl);
805 cftree_remove(cl);
806 update_cfmin(cl->cl_parent);
808 continue;
812 * update vt and f
814 cl->cl_vt = rtsc_y2x(&cl->cl_virtual, cl->cl_total)
815 - cl->cl_vtoff + cl->cl_vtadj;
818 * if vt of the class is smaller than cvtmin,
819 * the class was skipped in the past due to non-fit.
820 * if so, we need to adjust vtadj.
822 if (cl->cl_vt < cl->cl_parent->cl_cvtmin) {
823 cl->cl_vtadj += cl->cl_parent->cl_cvtmin - cl->cl_vt;
824 cl->cl_vt = cl->cl_parent->cl_cvtmin;
827 /* update the vt tree */
828 vttree_update(cl);
830 if (cl->cl_flags & HFSC_USC) {
831 cl->cl_myf = cl->cl_myfadj + rtsc_y2x(&cl->cl_ulimit,
832 cl->cl_total);
833 #if 0
835 * This code causes classes to stay way under their
836 * limit when multiple classes are used at gigabit
837 * speed. needs investigation. -kaber
840 * if myf lags behind by more than one clock tick
841 * from the current time, adjust myfadj to prevent
842 * a rate-limited class from going greedy.
843 * in a steady state under rate-limiting, myf
844 * fluctuates within one clock tick.
846 myf_bound = cur_time - PSCHED_JIFFIE2US(1);
847 if (cl->cl_myf < myf_bound) {
848 delta = cur_time - cl->cl_myf;
849 cl->cl_myfadj += delta;
850 cl->cl_myf += delta;
852 #endif
855 f = max(cl->cl_myf, cl->cl_cfmin);
856 if (f != cl->cl_f) {
857 cl->cl_f = f;
858 cftree_update(cl);
859 update_cfmin(cl->cl_parent);
864 static void
865 set_active(struct hfsc_class *cl, unsigned int len)
867 if (cl->cl_flags & HFSC_RSC)
868 init_ed(cl, len);
869 if (cl->cl_flags & HFSC_FSC)
870 init_vf(cl, len);
872 list_add_tail(&cl->dlist, &cl->sched->droplist);
875 static void
876 set_passive(struct hfsc_class *cl)
878 if (cl->cl_flags & HFSC_RSC)
879 eltree_remove(cl);
881 list_del(&cl->dlist);
884 * vttree is now handled in update_vf() so that update_vf(cl, 0, 0)
885 * needs to be called explicitly to remove a class from vttree.
890 * hack to get length of first packet in queue.
892 static unsigned int
893 qdisc_peek_len(struct Qdisc *sch)
895 struct sk_buff *skb;
896 unsigned int len;
898 skb = sch->dequeue(sch);
899 if (skb == NULL) {
900 if (net_ratelimit())
901 printk("qdisc_peek_len: non work-conserving qdisc ?\n");
902 return 0;
904 len = skb->len;
905 if (unlikely(sch->ops->requeue(skb, sch) != NET_XMIT_SUCCESS)) {
906 if (net_ratelimit())
907 printk("qdisc_peek_len: failed to requeue\n");
908 qdisc_tree_decrease_qlen(sch, 1);
909 return 0;
911 return len;
914 static void
915 hfsc_purge_queue(struct Qdisc *sch, struct hfsc_class *cl)
917 unsigned int len = cl->qdisc->q.qlen;
919 qdisc_reset(cl->qdisc);
920 qdisc_tree_decrease_qlen(cl->qdisc, len);
923 static void
924 hfsc_adjust_levels(struct hfsc_class *cl)
926 struct hfsc_class *p;
927 unsigned int level;
929 do {
930 level = 0;
931 list_for_each_entry(p, &cl->children, siblings) {
932 if (p->level >= level)
933 level = p->level + 1;
935 cl->level = level;
936 } while ((cl = cl->cl_parent) != NULL);
939 static inline unsigned int
940 hfsc_hash(u32 h)
942 h ^= h >> 8;
943 h ^= h >> 4;
945 return h & (HFSC_HSIZE - 1);
948 static inline struct hfsc_class *
949 hfsc_find_class(u32 classid, struct Qdisc *sch)
951 struct hfsc_sched *q = qdisc_priv(sch);
952 struct hfsc_class *cl;
954 list_for_each_entry(cl, &q->clhash[hfsc_hash(classid)], hlist) {
955 if (cl->classid == classid)
956 return cl;
958 return NULL;
961 static void
962 hfsc_change_rsc(struct hfsc_class *cl, struct tc_service_curve *rsc,
963 u64 cur_time)
965 sc2isc(rsc, &cl->cl_rsc);
966 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
967 cl->cl_eligible = cl->cl_deadline;
968 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
969 cl->cl_eligible.dx = 0;
970 cl->cl_eligible.dy = 0;
972 cl->cl_flags |= HFSC_RSC;
975 static void
976 hfsc_change_fsc(struct hfsc_class *cl, struct tc_service_curve *fsc)
978 sc2isc(fsc, &cl->cl_fsc);
979 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, cl->cl_vt, cl->cl_total);
980 cl->cl_flags |= HFSC_FSC;
983 static void
984 hfsc_change_usc(struct hfsc_class *cl, struct tc_service_curve *usc,
985 u64 cur_time)
987 sc2isc(usc, &cl->cl_usc);
988 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, cur_time, cl->cl_total);
989 cl->cl_flags |= HFSC_USC;
992 static int
993 hfsc_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
994 struct rtattr **tca, unsigned long *arg)
996 struct hfsc_sched *q = qdisc_priv(sch);
997 struct hfsc_class *cl = (struct hfsc_class *)*arg;
998 struct hfsc_class *parent = NULL;
999 struct rtattr *opt = tca[TCA_OPTIONS-1];
1000 struct rtattr *tb[TCA_HFSC_MAX];
1001 struct tc_service_curve *rsc = NULL, *fsc = NULL, *usc = NULL;
1002 u64 cur_time;
1004 if (opt == NULL || rtattr_parse_nested(tb, TCA_HFSC_MAX, opt))
1005 return -EINVAL;
1007 if (tb[TCA_HFSC_RSC-1]) {
1008 if (RTA_PAYLOAD(tb[TCA_HFSC_RSC-1]) < sizeof(*rsc))
1009 return -EINVAL;
1010 rsc = RTA_DATA(tb[TCA_HFSC_RSC-1]);
1011 if (rsc->m1 == 0 && rsc->m2 == 0)
1012 rsc = NULL;
1015 if (tb[TCA_HFSC_FSC-1]) {
1016 if (RTA_PAYLOAD(tb[TCA_HFSC_FSC-1]) < sizeof(*fsc))
1017 return -EINVAL;
1018 fsc = RTA_DATA(tb[TCA_HFSC_FSC-1]);
1019 if (fsc->m1 == 0 && fsc->m2 == 0)
1020 fsc = NULL;
1023 if (tb[TCA_HFSC_USC-1]) {
1024 if (RTA_PAYLOAD(tb[TCA_HFSC_USC-1]) < sizeof(*usc))
1025 return -EINVAL;
1026 usc = RTA_DATA(tb[TCA_HFSC_USC-1]);
1027 if (usc->m1 == 0 && usc->m2 == 0)
1028 usc = NULL;
1031 if (cl != NULL) {
1032 if (parentid) {
1033 if (cl->cl_parent && cl->cl_parent->classid != parentid)
1034 return -EINVAL;
1035 if (cl->cl_parent == NULL && parentid != TC_H_ROOT)
1036 return -EINVAL;
1038 cur_time = psched_get_time();
1040 sch_tree_lock(sch);
1041 if (rsc != NULL)
1042 hfsc_change_rsc(cl, rsc, cur_time);
1043 if (fsc != NULL)
1044 hfsc_change_fsc(cl, fsc);
1045 if (usc != NULL)
1046 hfsc_change_usc(cl, usc, cur_time);
1048 if (cl->qdisc->q.qlen != 0) {
1049 if (cl->cl_flags & HFSC_RSC)
1050 update_ed(cl, qdisc_peek_len(cl->qdisc));
1051 if (cl->cl_flags & HFSC_FSC)
1052 update_vf(cl, 0, cur_time);
1054 sch_tree_unlock(sch);
1056 if (tca[TCA_RATE-1])
1057 gen_replace_estimator(&cl->bstats, &cl->rate_est,
1058 &sch->dev->queue_lock,
1059 tca[TCA_RATE-1]);
1060 return 0;
1063 if (parentid == TC_H_ROOT)
1064 return -EEXIST;
1066 parent = &q->root;
1067 if (parentid) {
1068 parent = hfsc_find_class(parentid, sch);
1069 if (parent == NULL)
1070 return -ENOENT;
1073 if (classid == 0 || TC_H_MAJ(classid ^ sch->handle) != 0)
1074 return -EINVAL;
1075 if (hfsc_find_class(classid, sch))
1076 return -EEXIST;
1078 if (rsc == NULL && fsc == NULL)
1079 return -EINVAL;
1081 cl = kzalloc(sizeof(struct hfsc_class), GFP_KERNEL);
1082 if (cl == NULL)
1083 return -ENOBUFS;
1085 if (rsc != NULL)
1086 hfsc_change_rsc(cl, rsc, 0);
1087 if (fsc != NULL)
1088 hfsc_change_fsc(cl, fsc);
1089 if (usc != NULL)
1090 hfsc_change_usc(cl, usc, 0);
1092 cl->refcnt = 1;
1093 cl->classid = classid;
1094 cl->sched = q;
1095 cl->cl_parent = parent;
1096 cl->qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid);
1097 if (cl->qdisc == NULL)
1098 cl->qdisc = &noop_qdisc;
1099 INIT_LIST_HEAD(&cl->children);
1100 cl->vt_tree = RB_ROOT;
1101 cl->cf_tree = RB_ROOT;
1103 sch_tree_lock(sch);
1104 list_add_tail(&cl->hlist, &q->clhash[hfsc_hash(classid)]);
1105 list_add_tail(&cl->siblings, &parent->children);
1106 if (parent->level == 0)
1107 hfsc_purge_queue(sch, parent);
1108 hfsc_adjust_levels(parent);
1109 cl->cl_pcvtoff = parent->cl_cvtoff;
1110 sch_tree_unlock(sch);
1112 if (tca[TCA_RATE-1])
1113 gen_new_estimator(&cl->bstats, &cl->rate_est,
1114 &sch->dev->queue_lock, tca[TCA_RATE-1]);
1115 *arg = (unsigned long)cl;
1116 return 0;
1119 static void
1120 hfsc_destroy_class(struct Qdisc *sch, struct hfsc_class *cl)
1122 struct hfsc_sched *q = qdisc_priv(sch);
1124 tcf_destroy_chain(cl->filter_list);
1125 qdisc_destroy(cl->qdisc);
1126 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1127 if (cl != &q->root)
1128 kfree(cl);
1131 static int
1132 hfsc_delete_class(struct Qdisc *sch, unsigned long arg)
1134 struct hfsc_sched *q = qdisc_priv(sch);
1135 struct hfsc_class *cl = (struct hfsc_class *)arg;
1137 if (cl->level > 0 || cl->filter_cnt > 0 || cl == &q->root)
1138 return -EBUSY;
1140 sch_tree_lock(sch);
1142 list_del(&cl->siblings);
1143 hfsc_adjust_levels(cl->cl_parent);
1145 hfsc_purge_queue(sch, cl);
1146 list_del(&cl->hlist);
1148 if (--cl->refcnt == 0)
1149 hfsc_destroy_class(sch, cl);
1151 sch_tree_unlock(sch);
1152 return 0;
1155 static struct hfsc_class *
1156 hfsc_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
1158 struct hfsc_sched *q = qdisc_priv(sch);
1159 struct hfsc_class *cl;
1160 struct tcf_result res;
1161 struct tcf_proto *tcf;
1162 int result;
1164 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0 &&
1165 (cl = hfsc_find_class(skb->priority, sch)) != NULL)
1166 if (cl->level == 0)
1167 return cl;
1169 *qerr = NET_XMIT_BYPASS;
1170 tcf = q->root.filter_list;
1171 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
1172 #ifdef CONFIG_NET_CLS_ACT
1173 switch (result) {
1174 case TC_ACT_QUEUED:
1175 case TC_ACT_STOLEN:
1176 *qerr = NET_XMIT_SUCCESS;
1177 case TC_ACT_SHOT:
1178 return NULL;
1180 #elif defined(CONFIG_NET_CLS_POLICE)
1181 if (result == TC_POLICE_SHOT)
1182 return NULL;
1183 #endif
1184 if ((cl = (struct hfsc_class *)res.class) == NULL) {
1185 if ((cl = hfsc_find_class(res.classid, sch)) == NULL)
1186 break; /* filter selected invalid classid */
1189 if (cl->level == 0)
1190 return cl; /* hit leaf class */
1192 /* apply inner filter chain */
1193 tcf = cl->filter_list;
1196 /* classification failed, try default class */
1197 cl = hfsc_find_class(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
1198 if (cl == NULL || cl->level > 0)
1199 return NULL;
1201 return cl;
1204 static int
1205 hfsc_graft_class(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1206 struct Qdisc **old)
1208 struct hfsc_class *cl = (struct hfsc_class *)arg;
1210 if (cl == NULL)
1211 return -ENOENT;
1212 if (cl->level > 0)
1213 return -EINVAL;
1214 if (new == NULL) {
1215 new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1216 cl->classid);
1217 if (new == NULL)
1218 new = &noop_qdisc;
1221 sch_tree_lock(sch);
1222 hfsc_purge_queue(sch, cl);
1223 *old = xchg(&cl->qdisc, new);
1224 sch_tree_unlock(sch);
1225 return 0;
1228 static struct Qdisc *
1229 hfsc_class_leaf(struct Qdisc *sch, unsigned long arg)
1231 struct hfsc_class *cl = (struct hfsc_class *)arg;
1233 if (cl != NULL && cl->level == 0)
1234 return cl->qdisc;
1236 return NULL;
1239 static void
1240 hfsc_qlen_notify(struct Qdisc *sch, unsigned long arg)
1242 struct hfsc_class *cl = (struct hfsc_class *)arg;
1244 if (cl->qdisc->q.qlen == 0) {
1245 update_vf(cl, 0, 0);
1246 set_passive(cl);
1250 static unsigned long
1251 hfsc_get_class(struct Qdisc *sch, u32 classid)
1253 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1255 if (cl != NULL)
1256 cl->refcnt++;
1258 return (unsigned long)cl;
1261 static void
1262 hfsc_put_class(struct Qdisc *sch, unsigned long arg)
1264 struct hfsc_class *cl = (struct hfsc_class *)arg;
1266 if (--cl->refcnt == 0)
1267 hfsc_destroy_class(sch, cl);
1270 static unsigned long
1271 hfsc_bind_tcf(struct Qdisc *sch, unsigned long parent, u32 classid)
1273 struct hfsc_class *p = (struct hfsc_class *)parent;
1274 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1276 if (cl != NULL) {
1277 if (p != NULL && p->level <= cl->level)
1278 return 0;
1279 cl->filter_cnt++;
1282 return (unsigned long)cl;
1285 static void
1286 hfsc_unbind_tcf(struct Qdisc *sch, unsigned long arg)
1288 struct hfsc_class *cl = (struct hfsc_class *)arg;
1290 cl->filter_cnt--;
1293 static struct tcf_proto **
1294 hfsc_tcf_chain(struct Qdisc *sch, unsigned long arg)
1296 struct hfsc_sched *q = qdisc_priv(sch);
1297 struct hfsc_class *cl = (struct hfsc_class *)arg;
1299 if (cl == NULL)
1300 cl = &q->root;
1302 return &cl->filter_list;
1305 static int
1306 hfsc_dump_sc(struct sk_buff *skb, int attr, struct internal_sc *sc)
1308 struct tc_service_curve tsc;
1310 tsc.m1 = sm2m(sc->sm1);
1311 tsc.d = dx2d(sc->dx);
1312 tsc.m2 = sm2m(sc->sm2);
1313 RTA_PUT(skb, attr, sizeof(tsc), &tsc);
1315 return skb->len;
1317 rtattr_failure:
1318 return -1;
1321 static inline int
1322 hfsc_dump_curves(struct sk_buff *skb, struct hfsc_class *cl)
1324 if ((cl->cl_flags & HFSC_RSC) &&
1325 (hfsc_dump_sc(skb, TCA_HFSC_RSC, &cl->cl_rsc) < 0))
1326 goto rtattr_failure;
1328 if ((cl->cl_flags & HFSC_FSC) &&
1329 (hfsc_dump_sc(skb, TCA_HFSC_FSC, &cl->cl_fsc) < 0))
1330 goto rtattr_failure;
1332 if ((cl->cl_flags & HFSC_USC) &&
1333 (hfsc_dump_sc(skb, TCA_HFSC_USC, &cl->cl_usc) < 0))
1334 goto rtattr_failure;
1336 return skb->len;
1338 rtattr_failure:
1339 return -1;
1342 static int
1343 hfsc_dump_class(struct Qdisc *sch, unsigned long arg, struct sk_buff *skb,
1344 struct tcmsg *tcm)
1346 struct hfsc_class *cl = (struct hfsc_class *)arg;
1347 unsigned char *b = skb_tail_pointer(skb);
1348 struct rtattr *rta = (struct rtattr *)b;
1350 tcm->tcm_parent = cl->cl_parent ? cl->cl_parent->classid : TC_H_ROOT;
1351 tcm->tcm_handle = cl->classid;
1352 if (cl->level == 0)
1353 tcm->tcm_info = cl->qdisc->handle;
1355 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1356 if (hfsc_dump_curves(skb, cl) < 0)
1357 goto rtattr_failure;
1358 rta->rta_len = skb_tail_pointer(skb) - b;
1359 return skb->len;
1361 rtattr_failure:
1362 nlmsg_trim(skb, b);
1363 return -1;
1366 static int
1367 hfsc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
1368 struct gnet_dump *d)
1370 struct hfsc_class *cl = (struct hfsc_class *)arg;
1371 struct tc_hfsc_stats xstats;
1373 cl->qstats.qlen = cl->qdisc->q.qlen;
1374 xstats.level = cl->level;
1375 xstats.period = cl->cl_vtperiod;
1376 xstats.work = cl->cl_total;
1377 xstats.rtwork = cl->cl_cumul;
1379 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1380 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1381 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1382 return -1;
1384 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
1389 static void
1390 hfsc_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1392 struct hfsc_sched *q = qdisc_priv(sch);
1393 struct hfsc_class *cl;
1394 unsigned int i;
1396 if (arg->stop)
1397 return;
1399 for (i = 0; i < HFSC_HSIZE; i++) {
1400 list_for_each_entry(cl, &q->clhash[i], hlist) {
1401 if (arg->count < arg->skip) {
1402 arg->count++;
1403 continue;
1405 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1406 arg->stop = 1;
1407 return;
1409 arg->count++;
1414 static void
1415 hfsc_schedule_watchdog(struct Qdisc *sch)
1417 struct hfsc_sched *q = qdisc_priv(sch);
1418 struct hfsc_class *cl;
1419 u64 next_time = 0;
1421 if ((cl = eltree_get_minel(q)) != NULL)
1422 next_time = cl->cl_e;
1423 if (q->root.cl_cfmin != 0) {
1424 if (next_time == 0 || next_time > q->root.cl_cfmin)
1425 next_time = q->root.cl_cfmin;
1427 WARN_ON(next_time == 0);
1428 qdisc_watchdog_schedule(&q->watchdog, next_time);
1431 static int
1432 hfsc_init_qdisc(struct Qdisc *sch, struct rtattr *opt)
1434 struct hfsc_sched *q = qdisc_priv(sch);
1435 struct tc_hfsc_qopt *qopt;
1436 unsigned int i;
1438 if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1439 return -EINVAL;
1440 qopt = RTA_DATA(opt);
1442 q->defcls = qopt->defcls;
1443 for (i = 0; i < HFSC_HSIZE; i++)
1444 INIT_LIST_HEAD(&q->clhash[i]);
1445 q->eligible = RB_ROOT;
1446 INIT_LIST_HEAD(&q->droplist);
1447 skb_queue_head_init(&q->requeue);
1449 q->root.refcnt = 1;
1450 q->root.classid = sch->handle;
1451 q->root.sched = q;
1452 q->root.qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1453 sch->handle);
1454 if (q->root.qdisc == NULL)
1455 q->root.qdisc = &noop_qdisc;
1456 INIT_LIST_HEAD(&q->root.children);
1457 q->root.vt_tree = RB_ROOT;
1458 q->root.cf_tree = RB_ROOT;
1460 list_add(&q->root.hlist, &q->clhash[hfsc_hash(q->root.classid)]);
1462 qdisc_watchdog_init(&q->watchdog, sch);
1464 return 0;
1467 static int
1468 hfsc_change_qdisc(struct Qdisc *sch, struct rtattr *opt)
1470 struct hfsc_sched *q = qdisc_priv(sch);
1471 struct tc_hfsc_qopt *qopt;
1473 if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1474 return -EINVAL;
1475 qopt = RTA_DATA(opt);
1477 sch_tree_lock(sch);
1478 q->defcls = qopt->defcls;
1479 sch_tree_unlock(sch);
1481 return 0;
1484 static void
1485 hfsc_reset_class(struct hfsc_class *cl)
1487 cl->cl_total = 0;
1488 cl->cl_cumul = 0;
1489 cl->cl_d = 0;
1490 cl->cl_e = 0;
1491 cl->cl_vt = 0;
1492 cl->cl_vtadj = 0;
1493 cl->cl_vtoff = 0;
1494 cl->cl_cvtmin = 0;
1495 cl->cl_cvtmax = 0;
1496 cl->cl_cvtoff = 0;
1497 cl->cl_pcvtoff = 0;
1498 cl->cl_vtperiod = 0;
1499 cl->cl_parentperiod = 0;
1500 cl->cl_f = 0;
1501 cl->cl_myf = 0;
1502 cl->cl_myfadj = 0;
1503 cl->cl_cfmin = 0;
1504 cl->cl_nactive = 0;
1506 cl->vt_tree = RB_ROOT;
1507 cl->cf_tree = RB_ROOT;
1508 qdisc_reset(cl->qdisc);
1510 if (cl->cl_flags & HFSC_RSC)
1511 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, 0, 0);
1512 if (cl->cl_flags & HFSC_FSC)
1513 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, 0, 0);
1514 if (cl->cl_flags & HFSC_USC)
1515 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, 0, 0);
1518 static void
1519 hfsc_reset_qdisc(struct Qdisc *sch)
1521 struct hfsc_sched *q = qdisc_priv(sch);
1522 struct hfsc_class *cl;
1523 unsigned int i;
1525 for (i = 0; i < HFSC_HSIZE; i++) {
1526 list_for_each_entry(cl, &q->clhash[i], hlist)
1527 hfsc_reset_class(cl);
1529 __skb_queue_purge(&q->requeue);
1530 q->eligible = RB_ROOT;
1531 INIT_LIST_HEAD(&q->droplist);
1532 qdisc_watchdog_cancel(&q->watchdog);
1533 sch->q.qlen = 0;
1536 static void
1537 hfsc_destroy_qdisc(struct Qdisc *sch)
1539 struct hfsc_sched *q = qdisc_priv(sch);
1540 struct hfsc_class *cl, *next;
1541 unsigned int i;
1543 for (i = 0; i < HFSC_HSIZE; i++) {
1544 list_for_each_entry_safe(cl, next, &q->clhash[i], hlist)
1545 hfsc_destroy_class(sch, cl);
1547 __skb_queue_purge(&q->requeue);
1548 qdisc_watchdog_cancel(&q->watchdog);
1551 static int
1552 hfsc_dump_qdisc(struct Qdisc *sch, struct sk_buff *skb)
1554 struct hfsc_sched *q = qdisc_priv(sch);
1555 unsigned char *b = skb_tail_pointer(skb);
1556 struct tc_hfsc_qopt qopt;
1558 qopt.defcls = q->defcls;
1559 RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
1560 return skb->len;
1562 rtattr_failure:
1563 nlmsg_trim(skb, b);
1564 return -1;
1567 static int
1568 hfsc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1570 struct hfsc_class *cl;
1571 unsigned int len;
1572 int err;
1574 cl = hfsc_classify(skb, sch, &err);
1575 if (cl == NULL) {
1576 if (err == NET_XMIT_BYPASS)
1577 sch->qstats.drops++;
1578 kfree_skb(skb);
1579 return err;
1582 len = skb->len;
1583 err = cl->qdisc->enqueue(skb, cl->qdisc);
1584 if (unlikely(err != NET_XMIT_SUCCESS)) {
1585 cl->qstats.drops++;
1586 sch->qstats.drops++;
1587 return err;
1590 if (cl->qdisc->q.qlen == 1)
1591 set_active(cl, len);
1593 cl->bstats.packets++;
1594 cl->bstats.bytes += len;
1595 sch->bstats.packets++;
1596 sch->bstats.bytes += len;
1597 sch->q.qlen++;
1599 return NET_XMIT_SUCCESS;
1602 static struct sk_buff *
1603 hfsc_dequeue(struct Qdisc *sch)
1605 struct hfsc_sched *q = qdisc_priv(sch);
1606 struct hfsc_class *cl;
1607 struct sk_buff *skb;
1608 u64 cur_time;
1609 unsigned int next_len;
1610 int realtime = 0;
1612 if (sch->q.qlen == 0)
1613 return NULL;
1614 if ((skb = __skb_dequeue(&q->requeue)))
1615 goto out;
1617 cur_time = psched_get_time();
1620 * if there are eligible classes, use real-time criteria.
1621 * find the class with the minimum deadline among
1622 * the eligible classes.
1624 if ((cl = eltree_get_mindl(q, cur_time)) != NULL) {
1625 realtime = 1;
1626 } else {
1628 * use link-sharing criteria
1629 * get the class with the minimum vt in the hierarchy
1631 cl = vttree_get_minvt(&q->root, cur_time);
1632 if (cl == NULL) {
1633 sch->qstats.overlimits++;
1634 hfsc_schedule_watchdog(sch);
1635 return NULL;
1639 skb = cl->qdisc->dequeue(cl->qdisc);
1640 if (skb == NULL) {
1641 if (net_ratelimit())
1642 printk("HFSC: Non-work-conserving qdisc ?\n");
1643 return NULL;
1646 update_vf(cl, skb->len, cur_time);
1647 if (realtime)
1648 cl->cl_cumul += skb->len;
1650 if (cl->qdisc->q.qlen != 0) {
1651 if (cl->cl_flags & HFSC_RSC) {
1652 /* update ed */
1653 next_len = qdisc_peek_len(cl->qdisc);
1654 if (realtime)
1655 update_ed(cl, next_len);
1656 else
1657 update_d(cl, next_len);
1659 } else {
1660 /* the class becomes passive */
1661 set_passive(cl);
1664 out:
1665 sch->flags &= ~TCQ_F_THROTTLED;
1666 sch->q.qlen--;
1668 return skb;
1671 static int
1672 hfsc_requeue(struct sk_buff *skb, struct Qdisc *sch)
1674 struct hfsc_sched *q = qdisc_priv(sch);
1676 __skb_queue_head(&q->requeue, skb);
1677 sch->q.qlen++;
1678 sch->qstats.requeues++;
1679 return NET_XMIT_SUCCESS;
1682 static unsigned int
1683 hfsc_drop(struct Qdisc *sch)
1685 struct hfsc_sched *q = qdisc_priv(sch);
1686 struct hfsc_class *cl;
1687 unsigned int len;
1689 list_for_each_entry(cl, &q->droplist, dlist) {
1690 if (cl->qdisc->ops->drop != NULL &&
1691 (len = cl->qdisc->ops->drop(cl->qdisc)) > 0) {
1692 if (cl->qdisc->q.qlen == 0) {
1693 update_vf(cl, 0, 0);
1694 set_passive(cl);
1695 } else {
1696 list_move_tail(&cl->dlist, &q->droplist);
1698 cl->qstats.drops++;
1699 sch->qstats.drops++;
1700 sch->q.qlen--;
1701 return len;
1704 return 0;
1707 static struct Qdisc_class_ops hfsc_class_ops = {
1708 .change = hfsc_change_class,
1709 .delete = hfsc_delete_class,
1710 .graft = hfsc_graft_class,
1711 .leaf = hfsc_class_leaf,
1712 .qlen_notify = hfsc_qlen_notify,
1713 .get = hfsc_get_class,
1714 .put = hfsc_put_class,
1715 .bind_tcf = hfsc_bind_tcf,
1716 .unbind_tcf = hfsc_unbind_tcf,
1717 .tcf_chain = hfsc_tcf_chain,
1718 .dump = hfsc_dump_class,
1719 .dump_stats = hfsc_dump_class_stats,
1720 .walk = hfsc_walk
1723 static struct Qdisc_ops hfsc_qdisc_ops = {
1724 .id = "hfsc",
1725 .init = hfsc_init_qdisc,
1726 .change = hfsc_change_qdisc,
1727 .reset = hfsc_reset_qdisc,
1728 .destroy = hfsc_destroy_qdisc,
1729 .dump = hfsc_dump_qdisc,
1730 .enqueue = hfsc_enqueue,
1731 .dequeue = hfsc_dequeue,
1732 .requeue = hfsc_requeue,
1733 .drop = hfsc_drop,
1734 .cl_ops = &hfsc_class_ops,
1735 .priv_size = sizeof(struct hfsc_sched),
1736 .owner = THIS_MODULE
1739 static int __init
1740 hfsc_init(void)
1742 return register_qdisc(&hfsc_qdisc_ops);
1745 static void __exit
1746 hfsc_cleanup(void)
1748 unregister_qdisc(&hfsc_qdisc_ops);
1751 MODULE_LICENSE("GPL");
1752 module_init(hfsc_init);
1753 module_exit(hfsc_cleanup);