Linux v2.6.17-rc2
[linux-2.6/next.git] / net / sched / sch_hfsc.c
blob91132f6871d76ee42fa4e9dc53fcf643b32b779d
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/config.h>
54 #include <linux/module.h>
55 #include <linux/types.h>
56 #include <linux/errno.h>
57 #include <linux/jiffies.h>
58 #include <linux/compiler.h>
59 #include <linux/spinlock.h>
60 #include <linux/skbuff.h>
61 #include <linux/string.h>
62 #include <linux/slab.h>
63 #include <linux/timer.h>
64 #include <linux/list.h>
65 #include <linux/rbtree.h>
66 #include <linux/init.h>
67 #include <linux/netdevice.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/pkt_sched.h>
70 #include <net/pkt_sched.h>
71 #include <net/pkt_cls.h>
72 #include <asm/system.h>
73 #include <asm/div64.h>
75 #define HFSC_DEBUG 1
78 * kernel internal service curve representation:
79 * coordinates are given by 64 bit unsigned integers.
80 * x-axis: unit is clock count.
81 * y-axis: unit is byte.
83 * The service curve parameters are converted to the internal
84 * representation. The slope values are scaled to avoid overflow.
85 * the inverse slope values as well as the y-projection of the 1st
86 * segment are kept in order to to avoid 64-bit divide operations
87 * that are expensive on 32-bit architectures.
90 struct internal_sc
92 u64 sm1; /* scaled slope of the 1st segment */
93 u64 ism1; /* scaled inverse-slope of the 1st segment */
94 u64 dx; /* the x-projection of the 1st segment */
95 u64 dy; /* the y-projection of the 1st segment */
96 u64 sm2; /* scaled slope of the 2nd segment */
97 u64 ism2; /* scaled inverse-slope of the 2nd segment */
100 /* runtime service curve */
101 struct runtime_sc
103 u64 x; /* current starting position on x-axis */
104 u64 y; /* current starting position on y-axis */
105 u64 sm1; /* scaled slope of the 1st segment */
106 u64 ism1; /* scaled inverse-slope of the 1st segment */
107 u64 dx; /* the x-projection of the 1st segment */
108 u64 dy; /* the y-projection of the 1st segment */
109 u64 sm2; /* scaled slope of the 2nd segment */
110 u64 ism2; /* scaled inverse-slope of the 2nd segment */
113 enum hfsc_class_flags
115 HFSC_RSC = 0x1,
116 HFSC_FSC = 0x2,
117 HFSC_USC = 0x4
120 struct hfsc_class
122 u32 classid; /* class id */
123 unsigned int refcnt; /* usage count */
125 struct gnet_stats_basic bstats;
126 struct gnet_stats_queue qstats;
127 struct gnet_stats_rate_est rate_est;
128 spinlock_t *stats_lock;
129 unsigned int level; /* class level in hierarchy */
130 struct tcf_proto *filter_list; /* filter list */
131 unsigned int filter_cnt; /* filter count */
133 struct hfsc_sched *sched; /* scheduler data */
134 struct hfsc_class *cl_parent; /* parent class */
135 struct list_head siblings; /* sibling classes */
136 struct list_head children; /* child classes */
137 struct Qdisc *qdisc; /* leaf qdisc */
139 struct rb_node el_node; /* qdisc's eligible tree member */
140 struct rb_root vt_tree; /* active children sorted by cl_vt */
141 struct rb_node vt_node; /* parent's vt_tree member */
142 struct rb_root cf_tree; /* active children sorted by cl_f */
143 struct rb_node cf_node; /* parent's cf_heap member */
144 struct list_head hlist; /* hash list member */
145 struct list_head dlist; /* drop list member */
147 u64 cl_total; /* total work in bytes */
148 u64 cl_cumul; /* cumulative work in bytes done by
149 real-time criteria */
151 u64 cl_d; /* deadline*/
152 u64 cl_e; /* eligible time */
153 u64 cl_vt; /* virtual time */
154 u64 cl_f; /* time when this class will fit for
155 link-sharing, max(myf, cfmin) */
156 u64 cl_myf; /* my fit-time (calculated from this
157 class's own upperlimit curve) */
158 u64 cl_myfadj; /* my fit-time adjustment (to cancel
159 history dependence) */
160 u64 cl_cfmin; /* earliest children's fit-time (used
161 with cl_myf to obtain cl_f) */
162 u64 cl_cvtmin; /* minimal virtual time among the
163 children fit for link-sharing
164 (monotonic within a period) */
165 u64 cl_vtadj; /* intra-period cumulative vt
166 adjustment */
167 u64 cl_vtoff; /* inter-period cumulative vt offset */
168 u64 cl_cvtmax; /* max child's vt in the last period */
169 u64 cl_cvtoff; /* cumulative cvtmax of all periods */
170 u64 cl_pcvtoff; /* parent's cvtoff at initalization
171 time */
173 struct internal_sc cl_rsc; /* internal real-time service curve */
174 struct internal_sc cl_fsc; /* internal fair service curve */
175 struct internal_sc cl_usc; /* internal upperlimit service curve */
176 struct runtime_sc cl_deadline; /* deadline curve */
177 struct runtime_sc cl_eligible; /* eligible curve */
178 struct runtime_sc cl_virtual; /* virtual curve */
179 struct runtime_sc cl_ulimit; /* upperlimit curve */
181 unsigned long cl_flags; /* which curves are valid */
182 unsigned long cl_vtperiod; /* vt period sequence number */
183 unsigned long cl_parentperiod;/* parent's vt period sequence number*/
184 unsigned long cl_nactive; /* number of active children */
187 #define HFSC_HSIZE 16
189 struct hfsc_sched
191 u16 defcls; /* default class id */
192 struct hfsc_class root; /* root class */
193 struct list_head clhash[HFSC_HSIZE]; /* class hash */
194 struct rb_root eligible; /* eligible tree */
195 struct list_head droplist; /* active leaf class list (for
196 dropping) */
197 struct sk_buff_head requeue; /* requeued packet */
198 struct timer_list wd_timer; /* watchdog timer */
202 * macros
204 #ifdef CONFIG_NET_SCH_CLK_GETTIMEOFDAY
205 #include <linux/time.h>
206 #undef PSCHED_GET_TIME
207 #define PSCHED_GET_TIME(stamp) \
208 do { \
209 struct timeval tv; \
210 do_gettimeofday(&tv); \
211 (stamp) = 1ULL * USEC_PER_SEC * tv.tv_sec + tv.tv_usec; \
212 } while (0)
213 #endif
215 #if HFSC_DEBUG
216 #define ASSERT(cond) \
217 do { \
218 if (unlikely(!(cond))) \
219 printk("assertion %s failed at %s:%i (%s)\n", \
220 #cond, __FILE__, __LINE__, __FUNCTION__); \
221 } while (0)
222 #else
223 #define ASSERT(cond)
224 #endif /* HFSC_DEBUG */
226 #define HT_INFINITY 0xffffffffffffffffULL /* infinite time value */
230 * eligible tree holds backlogged classes being sorted by their eligible times.
231 * there is one eligible tree per hfsc instance.
234 static void
235 eltree_insert(struct hfsc_class *cl)
237 struct rb_node **p = &cl->sched->eligible.rb_node;
238 struct rb_node *parent = NULL;
239 struct hfsc_class *cl1;
241 while (*p != NULL) {
242 parent = *p;
243 cl1 = rb_entry(parent, struct hfsc_class, el_node);
244 if (cl->cl_e >= cl1->cl_e)
245 p = &parent->rb_right;
246 else
247 p = &parent->rb_left;
249 rb_link_node(&cl->el_node, parent, p);
250 rb_insert_color(&cl->el_node, &cl->sched->eligible);
253 static inline void
254 eltree_remove(struct hfsc_class *cl)
256 rb_erase(&cl->el_node, &cl->sched->eligible);
259 static inline void
260 eltree_update(struct hfsc_class *cl)
262 eltree_remove(cl);
263 eltree_insert(cl);
266 /* find the class with the minimum deadline among the eligible classes */
267 static inline struct hfsc_class *
268 eltree_get_mindl(struct hfsc_sched *q, u64 cur_time)
270 struct hfsc_class *p, *cl = NULL;
271 struct rb_node *n;
273 for (n = rb_first(&q->eligible); n != NULL; n = rb_next(n)) {
274 p = rb_entry(n, struct hfsc_class, el_node);
275 if (p->cl_e > cur_time)
276 break;
277 if (cl == NULL || p->cl_d < cl->cl_d)
278 cl = p;
280 return cl;
283 /* find the class with minimum eligible time among the eligible classes */
284 static inline struct hfsc_class *
285 eltree_get_minel(struct hfsc_sched *q)
287 struct rb_node *n;
289 n = rb_first(&q->eligible);
290 if (n == NULL)
291 return NULL;
292 return rb_entry(n, struct hfsc_class, el_node);
296 * vttree holds holds backlogged child classes being sorted by their virtual
297 * time. each intermediate class has one vttree.
299 static void
300 vttree_insert(struct hfsc_class *cl)
302 struct rb_node **p = &cl->cl_parent->vt_tree.rb_node;
303 struct rb_node *parent = NULL;
304 struct hfsc_class *cl1;
306 while (*p != NULL) {
307 parent = *p;
308 cl1 = rb_entry(parent, struct hfsc_class, vt_node);
309 if (cl->cl_vt >= cl1->cl_vt)
310 p = &parent->rb_right;
311 else
312 p = &parent->rb_left;
314 rb_link_node(&cl->vt_node, parent, p);
315 rb_insert_color(&cl->vt_node, &cl->cl_parent->vt_tree);
318 static inline void
319 vttree_remove(struct hfsc_class *cl)
321 rb_erase(&cl->vt_node, &cl->cl_parent->vt_tree);
324 static inline void
325 vttree_update(struct hfsc_class *cl)
327 vttree_remove(cl);
328 vttree_insert(cl);
331 static inline struct hfsc_class *
332 vttree_firstfit(struct hfsc_class *cl, u64 cur_time)
334 struct hfsc_class *p;
335 struct rb_node *n;
337 for (n = rb_first(&cl->vt_tree); n != NULL; n = rb_next(n)) {
338 p = rb_entry(n, struct hfsc_class, vt_node);
339 if (p->cl_f <= cur_time)
340 return p;
342 return NULL;
346 * get the leaf class with the minimum vt in the hierarchy
348 static struct hfsc_class *
349 vttree_get_minvt(struct hfsc_class *cl, u64 cur_time)
351 /* if root-class's cfmin is bigger than cur_time nothing to do */
352 if (cl->cl_cfmin > cur_time)
353 return NULL;
355 while (cl->level > 0) {
356 cl = vttree_firstfit(cl, cur_time);
357 if (cl == NULL)
358 return NULL;
360 * update parent's cl_cvtmin.
362 if (cl->cl_parent->cl_cvtmin < cl->cl_vt)
363 cl->cl_parent->cl_cvtmin = cl->cl_vt;
365 return cl;
368 static void
369 cftree_insert(struct hfsc_class *cl)
371 struct rb_node **p = &cl->cl_parent->cf_tree.rb_node;
372 struct rb_node *parent = NULL;
373 struct hfsc_class *cl1;
375 while (*p != NULL) {
376 parent = *p;
377 cl1 = rb_entry(parent, struct hfsc_class, cf_node);
378 if (cl->cl_f >= cl1->cl_f)
379 p = &parent->rb_right;
380 else
381 p = &parent->rb_left;
383 rb_link_node(&cl->cf_node, parent, p);
384 rb_insert_color(&cl->cf_node, &cl->cl_parent->cf_tree);
387 static inline void
388 cftree_remove(struct hfsc_class *cl)
390 rb_erase(&cl->cf_node, &cl->cl_parent->cf_tree);
393 static inline void
394 cftree_update(struct hfsc_class *cl)
396 cftree_remove(cl);
397 cftree_insert(cl);
401 * service curve support functions
403 * external service curve parameters
404 * m: bps
405 * d: us
406 * internal service curve parameters
407 * sm: (bytes/psched_us) << SM_SHIFT
408 * ism: (psched_us/byte) << ISM_SHIFT
409 * dx: psched_us
411 * Clock source resolution (CONFIG_NET_SCH_CLK_*)
412 * JIFFIES: for 48<=HZ<=1534 resolution is between 0.63us and 1.27us.
413 * CPU: resolution is between 0.5us and 1us.
414 * GETTIMEOFDAY: resolution is exactly 1us.
416 * sm and ism are scaled in order to keep effective digits.
417 * SM_SHIFT and ISM_SHIFT are selected to keep at least 4 effective
418 * digits in decimal using the following table.
420 * Note: We can afford the additional accuracy (altq hfsc keeps at most
421 * 3 effective digits) thanks to the fact that linux clock is bounded
422 * much more tightly.
424 * bits/sec 100Kbps 1Mbps 10Mbps 100Mbps 1Gbps
425 * ------------+-------------------------------------------------------
426 * bytes/0.5us 6.25e-3 62.5e-3 625e-3 6250e-e 62500e-3
427 * bytes/us 12.5e-3 125e-3 1250e-3 12500e-3 125000e-3
428 * bytes/1.27us 15.875e-3 158.75e-3 1587.5e-3 15875e-3 158750e-3
430 * 0.5us/byte 160 16 1.6 0.16 0.016
431 * us/byte 80 8 0.8 0.08 0.008
432 * 1.27us/byte 63 6.3 0.63 0.063 0.0063
434 #define SM_SHIFT 20
435 #define ISM_SHIFT 18
437 #define SM_MASK ((1ULL << SM_SHIFT) - 1)
438 #define ISM_MASK ((1ULL << ISM_SHIFT) - 1)
440 static inline u64
441 seg_x2y(u64 x, u64 sm)
443 u64 y;
446 * compute
447 * y = x * sm >> SM_SHIFT
448 * but divide it for the upper and lower bits to avoid overflow
450 y = (x >> SM_SHIFT) * sm + (((x & SM_MASK) * sm) >> SM_SHIFT);
451 return y;
454 static inline u64
455 seg_y2x(u64 y, u64 ism)
457 u64 x;
459 if (y == 0)
460 x = 0;
461 else if (ism == HT_INFINITY)
462 x = HT_INFINITY;
463 else {
464 x = (y >> ISM_SHIFT) * ism
465 + (((y & ISM_MASK) * ism) >> ISM_SHIFT);
467 return x;
470 /* Convert m (bps) into sm (bytes/psched us) */
471 static u64
472 m2sm(u32 m)
474 u64 sm;
476 sm = ((u64)m << SM_SHIFT);
477 sm += PSCHED_JIFFIE2US(HZ) - 1;
478 do_div(sm, PSCHED_JIFFIE2US(HZ));
479 return sm;
482 /* convert m (bps) into ism (psched us/byte) */
483 static u64
484 m2ism(u32 m)
486 u64 ism;
488 if (m == 0)
489 ism = HT_INFINITY;
490 else {
491 ism = ((u64)PSCHED_JIFFIE2US(HZ) << ISM_SHIFT);
492 ism += m - 1;
493 do_div(ism, m);
495 return ism;
498 /* convert d (us) into dx (psched us) */
499 static u64
500 d2dx(u32 d)
502 u64 dx;
504 dx = ((u64)d * PSCHED_JIFFIE2US(HZ));
505 dx += USEC_PER_SEC - 1;
506 do_div(dx, USEC_PER_SEC);
507 return dx;
510 /* convert sm (bytes/psched us) into m (bps) */
511 static u32
512 sm2m(u64 sm)
514 u64 m;
516 m = (sm * PSCHED_JIFFIE2US(HZ)) >> SM_SHIFT;
517 return (u32)m;
520 /* convert dx (psched us) into d (us) */
521 static u32
522 dx2d(u64 dx)
524 u64 d;
526 d = dx * USEC_PER_SEC;
527 do_div(d, PSCHED_JIFFIE2US(HZ));
528 return (u32)d;
531 static void
532 sc2isc(struct tc_service_curve *sc, struct internal_sc *isc)
534 isc->sm1 = m2sm(sc->m1);
535 isc->ism1 = m2ism(sc->m1);
536 isc->dx = d2dx(sc->d);
537 isc->dy = seg_x2y(isc->dx, isc->sm1);
538 isc->sm2 = m2sm(sc->m2);
539 isc->ism2 = m2ism(sc->m2);
543 * initialize the runtime service curve with the given internal
544 * service curve starting at (x, y).
546 static void
547 rtsc_init(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
549 rtsc->x = x;
550 rtsc->y = y;
551 rtsc->sm1 = isc->sm1;
552 rtsc->ism1 = isc->ism1;
553 rtsc->dx = isc->dx;
554 rtsc->dy = isc->dy;
555 rtsc->sm2 = isc->sm2;
556 rtsc->ism2 = isc->ism2;
560 * calculate the y-projection of the runtime service curve by the
561 * given x-projection value
563 static u64
564 rtsc_y2x(struct runtime_sc *rtsc, u64 y)
566 u64 x;
568 if (y < rtsc->y)
569 x = rtsc->x;
570 else if (y <= rtsc->y + rtsc->dy) {
571 /* x belongs to the 1st segment */
572 if (rtsc->dy == 0)
573 x = rtsc->x + rtsc->dx;
574 else
575 x = rtsc->x + seg_y2x(y - rtsc->y, rtsc->ism1);
576 } else {
577 /* x belongs to the 2nd segment */
578 x = rtsc->x + rtsc->dx
579 + seg_y2x(y - rtsc->y - rtsc->dy, rtsc->ism2);
581 return x;
584 static u64
585 rtsc_x2y(struct runtime_sc *rtsc, u64 x)
587 u64 y;
589 if (x <= rtsc->x)
590 y = rtsc->y;
591 else if (x <= rtsc->x + rtsc->dx)
592 /* y belongs to the 1st segment */
593 y = rtsc->y + seg_x2y(x - rtsc->x, rtsc->sm1);
594 else
595 /* y belongs to the 2nd segment */
596 y = rtsc->y + rtsc->dy
597 + seg_x2y(x - rtsc->x - rtsc->dx, rtsc->sm2);
598 return y;
602 * update the runtime service curve by taking the minimum of the current
603 * runtime service curve and the service curve starting at (x, y).
605 static void
606 rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
608 u64 y1, y2, dx, dy;
609 u32 dsm;
611 if (isc->sm1 <= isc->sm2) {
612 /* service curve is convex */
613 y1 = rtsc_x2y(rtsc, x);
614 if (y1 < y)
615 /* the current rtsc is smaller */
616 return;
617 rtsc->x = x;
618 rtsc->y = y;
619 return;
623 * service curve is concave
624 * compute the two y values of the current rtsc
625 * y1: at x
626 * y2: at (x + dx)
628 y1 = rtsc_x2y(rtsc, x);
629 if (y1 <= y) {
630 /* rtsc is below isc, no change to rtsc */
631 return;
634 y2 = rtsc_x2y(rtsc, x + isc->dx);
635 if (y2 >= y + isc->dy) {
636 /* rtsc is above isc, replace rtsc by isc */
637 rtsc->x = x;
638 rtsc->y = y;
639 rtsc->dx = isc->dx;
640 rtsc->dy = isc->dy;
641 return;
645 * the two curves intersect
646 * compute the offsets (dx, dy) using the reverse
647 * function of seg_x2y()
648 * seg_x2y(dx, sm1) == seg_x2y(dx, sm2) + (y1 - y)
650 dx = (y1 - y) << SM_SHIFT;
651 dsm = isc->sm1 - isc->sm2;
652 do_div(dx, dsm);
654 * check if (x, y1) belongs to the 1st segment of rtsc.
655 * if so, add the offset.
657 if (rtsc->x + rtsc->dx > x)
658 dx += rtsc->x + rtsc->dx - x;
659 dy = seg_x2y(dx, isc->sm1);
661 rtsc->x = x;
662 rtsc->y = y;
663 rtsc->dx = dx;
664 rtsc->dy = dy;
665 return;
668 static void
669 init_ed(struct hfsc_class *cl, unsigned int next_len)
671 u64 cur_time;
673 PSCHED_GET_TIME(cur_time);
675 /* update the deadline curve */
676 rtsc_min(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
679 * update the eligible curve.
680 * for concave, it is equal to the deadline curve.
681 * for convex, it is a linear curve with slope m2.
683 cl->cl_eligible = cl->cl_deadline;
684 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
685 cl->cl_eligible.dx = 0;
686 cl->cl_eligible.dy = 0;
689 /* compute e and d */
690 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
691 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
693 eltree_insert(cl);
696 static void
697 update_ed(struct hfsc_class *cl, unsigned int next_len)
699 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
700 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
702 eltree_update(cl);
705 static inline void
706 update_d(struct hfsc_class *cl, unsigned int next_len)
708 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
711 static inline void
712 update_cfmin(struct hfsc_class *cl)
714 struct rb_node *n = rb_first(&cl->cf_tree);
715 struct hfsc_class *p;
717 if (n == NULL) {
718 cl->cl_cfmin = 0;
719 return;
721 p = rb_entry(n, struct hfsc_class, cf_node);
722 cl->cl_cfmin = p->cl_f;
725 static void
726 init_vf(struct hfsc_class *cl, unsigned int len)
728 struct hfsc_class *max_cl;
729 struct rb_node *n;
730 u64 vt, f, cur_time;
731 int go_active;
733 cur_time = 0;
734 go_active = 1;
735 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
736 if (go_active && cl->cl_nactive++ == 0)
737 go_active = 1;
738 else
739 go_active = 0;
741 if (go_active) {
742 n = rb_last(&cl->cl_parent->vt_tree);
743 if (n != NULL) {
744 max_cl = rb_entry(n, struct hfsc_class,vt_node);
746 * set vt to the average of the min and max
747 * classes. if the parent's period didn't
748 * change, don't decrease vt of the class.
750 vt = max_cl->cl_vt;
751 if (cl->cl_parent->cl_cvtmin != 0)
752 vt = (cl->cl_parent->cl_cvtmin + vt)/2;
754 if (cl->cl_parent->cl_vtperiod !=
755 cl->cl_parentperiod || vt > cl->cl_vt)
756 cl->cl_vt = vt;
757 } else {
759 * first child for a new parent backlog period.
760 * add parent's cvtmax to cvtoff to make a new
761 * vt (vtoff + vt) larger than the vt in the
762 * last period for all children.
764 vt = cl->cl_parent->cl_cvtmax;
765 cl->cl_parent->cl_cvtoff += vt;
766 cl->cl_parent->cl_cvtmax = 0;
767 cl->cl_parent->cl_cvtmin = 0;
768 cl->cl_vt = 0;
771 cl->cl_vtoff = cl->cl_parent->cl_cvtoff -
772 cl->cl_pcvtoff;
774 /* update the virtual curve */
775 vt = cl->cl_vt + cl->cl_vtoff;
776 rtsc_min(&cl->cl_virtual, &cl->cl_fsc, vt,
777 cl->cl_total);
778 if (cl->cl_virtual.x == vt) {
779 cl->cl_virtual.x -= cl->cl_vtoff;
780 cl->cl_vtoff = 0;
782 cl->cl_vtadj = 0;
784 cl->cl_vtperiod++; /* increment vt period */
785 cl->cl_parentperiod = cl->cl_parent->cl_vtperiod;
786 if (cl->cl_parent->cl_nactive == 0)
787 cl->cl_parentperiod++;
788 cl->cl_f = 0;
790 vttree_insert(cl);
791 cftree_insert(cl);
793 if (cl->cl_flags & HFSC_USC) {
794 /* class has upper limit curve */
795 if (cur_time == 0)
796 PSCHED_GET_TIME(cur_time);
798 /* update the ulimit curve */
799 rtsc_min(&cl->cl_ulimit, &cl->cl_usc, cur_time,
800 cl->cl_total);
801 /* compute myf */
802 cl->cl_myf = rtsc_y2x(&cl->cl_ulimit,
803 cl->cl_total);
804 cl->cl_myfadj = 0;
808 f = max(cl->cl_myf, cl->cl_cfmin);
809 if (f != cl->cl_f) {
810 cl->cl_f = f;
811 cftree_update(cl);
812 update_cfmin(cl->cl_parent);
817 static void
818 update_vf(struct hfsc_class *cl, unsigned int len, u64 cur_time)
820 u64 f; /* , myf_bound, delta; */
821 int go_passive = 0;
823 if (cl->qdisc->q.qlen == 0 && cl->cl_flags & HFSC_FSC)
824 go_passive = 1;
826 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
827 cl->cl_total += len;
829 if (!(cl->cl_flags & HFSC_FSC) || cl->cl_nactive == 0)
830 continue;
832 if (go_passive && --cl->cl_nactive == 0)
833 go_passive = 1;
834 else
835 go_passive = 0;
837 if (go_passive) {
838 /* no more active child, going passive */
840 /* update cvtmax of the parent class */
841 if (cl->cl_vt > cl->cl_parent->cl_cvtmax)
842 cl->cl_parent->cl_cvtmax = cl->cl_vt;
844 /* remove this class from the vt tree */
845 vttree_remove(cl);
847 cftree_remove(cl);
848 update_cfmin(cl->cl_parent);
850 continue;
854 * update vt and f
856 cl->cl_vt = rtsc_y2x(&cl->cl_virtual, cl->cl_total)
857 - cl->cl_vtoff + cl->cl_vtadj;
860 * if vt of the class is smaller than cvtmin,
861 * the class was skipped in the past due to non-fit.
862 * if so, we need to adjust vtadj.
864 if (cl->cl_vt < cl->cl_parent->cl_cvtmin) {
865 cl->cl_vtadj += cl->cl_parent->cl_cvtmin - cl->cl_vt;
866 cl->cl_vt = cl->cl_parent->cl_cvtmin;
869 /* update the vt tree */
870 vttree_update(cl);
872 if (cl->cl_flags & HFSC_USC) {
873 cl->cl_myf = cl->cl_myfadj + rtsc_y2x(&cl->cl_ulimit,
874 cl->cl_total);
875 #if 0
877 * This code causes classes to stay way under their
878 * limit when multiple classes are used at gigabit
879 * speed. needs investigation. -kaber
882 * if myf lags behind by more than one clock tick
883 * from the current time, adjust myfadj to prevent
884 * a rate-limited class from going greedy.
885 * in a steady state under rate-limiting, myf
886 * fluctuates within one clock tick.
888 myf_bound = cur_time - PSCHED_JIFFIE2US(1);
889 if (cl->cl_myf < myf_bound) {
890 delta = cur_time - cl->cl_myf;
891 cl->cl_myfadj += delta;
892 cl->cl_myf += delta;
894 #endif
897 f = max(cl->cl_myf, cl->cl_cfmin);
898 if (f != cl->cl_f) {
899 cl->cl_f = f;
900 cftree_update(cl);
901 update_cfmin(cl->cl_parent);
906 static void
907 set_active(struct hfsc_class *cl, unsigned int len)
909 if (cl->cl_flags & HFSC_RSC)
910 init_ed(cl, len);
911 if (cl->cl_flags & HFSC_FSC)
912 init_vf(cl, len);
914 list_add_tail(&cl->dlist, &cl->sched->droplist);
917 static void
918 set_passive(struct hfsc_class *cl)
920 if (cl->cl_flags & HFSC_RSC)
921 eltree_remove(cl);
923 list_del(&cl->dlist);
926 * vttree is now handled in update_vf() so that update_vf(cl, 0, 0)
927 * needs to be called explicitly to remove a class from vttree.
932 * hack to get length of first packet in queue.
934 static unsigned int
935 qdisc_peek_len(struct Qdisc *sch)
937 struct sk_buff *skb;
938 unsigned int len;
940 skb = sch->dequeue(sch);
941 if (skb == NULL) {
942 if (net_ratelimit())
943 printk("qdisc_peek_len: non work-conserving qdisc ?\n");
944 return 0;
946 len = skb->len;
947 if (unlikely(sch->ops->requeue(skb, sch) != NET_XMIT_SUCCESS)) {
948 if (net_ratelimit())
949 printk("qdisc_peek_len: failed to requeue\n");
950 return 0;
952 return len;
955 static void
956 hfsc_purge_queue(struct Qdisc *sch, struct hfsc_class *cl)
958 unsigned int len = cl->qdisc->q.qlen;
960 qdisc_reset(cl->qdisc);
961 if (len > 0) {
962 update_vf(cl, 0, 0);
963 set_passive(cl);
964 sch->q.qlen -= len;
968 static void
969 hfsc_adjust_levels(struct hfsc_class *cl)
971 struct hfsc_class *p;
972 unsigned int level;
974 do {
975 level = 0;
976 list_for_each_entry(p, &cl->children, siblings) {
977 if (p->level > level)
978 level = p->level;
980 cl->level = level + 1;
981 } while ((cl = cl->cl_parent) != NULL);
984 static inline unsigned int
985 hfsc_hash(u32 h)
987 h ^= h >> 8;
988 h ^= h >> 4;
990 return h & (HFSC_HSIZE - 1);
993 static inline struct hfsc_class *
994 hfsc_find_class(u32 classid, struct Qdisc *sch)
996 struct hfsc_sched *q = qdisc_priv(sch);
997 struct hfsc_class *cl;
999 list_for_each_entry(cl, &q->clhash[hfsc_hash(classid)], hlist) {
1000 if (cl->classid == classid)
1001 return cl;
1003 return NULL;
1006 static void
1007 hfsc_change_rsc(struct hfsc_class *cl, struct tc_service_curve *rsc,
1008 u64 cur_time)
1010 sc2isc(rsc, &cl->cl_rsc);
1011 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
1012 cl->cl_eligible = cl->cl_deadline;
1013 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
1014 cl->cl_eligible.dx = 0;
1015 cl->cl_eligible.dy = 0;
1017 cl->cl_flags |= HFSC_RSC;
1020 static void
1021 hfsc_change_fsc(struct hfsc_class *cl, struct tc_service_curve *fsc)
1023 sc2isc(fsc, &cl->cl_fsc);
1024 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, cl->cl_vt, cl->cl_total);
1025 cl->cl_flags |= HFSC_FSC;
1028 static void
1029 hfsc_change_usc(struct hfsc_class *cl, struct tc_service_curve *usc,
1030 u64 cur_time)
1032 sc2isc(usc, &cl->cl_usc);
1033 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, cur_time, cl->cl_total);
1034 cl->cl_flags |= HFSC_USC;
1037 static int
1038 hfsc_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
1039 struct rtattr **tca, unsigned long *arg)
1041 struct hfsc_sched *q = qdisc_priv(sch);
1042 struct hfsc_class *cl = (struct hfsc_class *)*arg;
1043 struct hfsc_class *parent = NULL;
1044 struct rtattr *opt = tca[TCA_OPTIONS-1];
1045 struct rtattr *tb[TCA_HFSC_MAX];
1046 struct tc_service_curve *rsc = NULL, *fsc = NULL, *usc = NULL;
1047 u64 cur_time;
1049 if (opt == NULL || rtattr_parse_nested(tb, TCA_HFSC_MAX, opt))
1050 return -EINVAL;
1052 if (tb[TCA_HFSC_RSC-1]) {
1053 if (RTA_PAYLOAD(tb[TCA_HFSC_RSC-1]) < sizeof(*rsc))
1054 return -EINVAL;
1055 rsc = RTA_DATA(tb[TCA_HFSC_RSC-1]);
1056 if (rsc->m1 == 0 && rsc->m2 == 0)
1057 rsc = NULL;
1060 if (tb[TCA_HFSC_FSC-1]) {
1061 if (RTA_PAYLOAD(tb[TCA_HFSC_FSC-1]) < sizeof(*fsc))
1062 return -EINVAL;
1063 fsc = RTA_DATA(tb[TCA_HFSC_FSC-1]);
1064 if (fsc->m1 == 0 && fsc->m2 == 0)
1065 fsc = NULL;
1068 if (tb[TCA_HFSC_USC-1]) {
1069 if (RTA_PAYLOAD(tb[TCA_HFSC_USC-1]) < sizeof(*usc))
1070 return -EINVAL;
1071 usc = RTA_DATA(tb[TCA_HFSC_USC-1]);
1072 if (usc->m1 == 0 && usc->m2 == 0)
1073 usc = NULL;
1076 if (cl != NULL) {
1077 if (parentid) {
1078 if (cl->cl_parent && cl->cl_parent->classid != parentid)
1079 return -EINVAL;
1080 if (cl->cl_parent == NULL && parentid != TC_H_ROOT)
1081 return -EINVAL;
1083 PSCHED_GET_TIME(cur_time);
1085 sch_tree_lock(sch);
1086 if (rsc != NULL)
1087 hfsc_change_rsc(cl, rsc, cur_time);
1088 if (fsc != NULL)
1089 hfsc_change_fsc(cl, fsc);
1090 if (usc != NULL)
1091 hfsc_change_usc(cl, usc, cur_time);
1093 if (cl->qdisc->q.qlen != 0) {
1094 if (cl->cl_flags & HFSC_RSC)
1095 update_ed(cl, qdisc_peek_len(cl->qdisc));
1096 if (cl->cl_flags & HFSC_FSC)
1097 update_vf(cl, 0, cur_time);
1099 sch_tree_unlock(sch);
1101 #ifdef CONFIG_NET_ESTIMATOR
1102 if (tca[TCA_RATE-1])
1103 gen_replace_estimator(&cl->bstats, &cl->rate_est,
1104 cl->stats_lock, tca[TCA_RATE-1]);
1105 #endif
1106 return 0;
1109 if (parentid == TC_H_ROOT)
1110 return -EEXIST;
1112 parent = &q->root;
1113 if (parentid) {
1114 parent = hfsc_find_class(parentid, sch);
1115 if (parent == NULL)
1116 return -ENOENT;
1119 if (classid == 0 || TC_H_MAJ(classid ^ sch->handle) != 0)
1120 return -EINVAL;
1121 if (hfsc_find_class(classid, sch))
1122 return -EEXIST;
1124 if (rsc == NULL && fsc == NULL)
1125 return -EINVAL;
1127 cl = kmalloc(sizeof(struct hfsc_class), GFP_KERNEL);
1128 if (cl == NULL)
1129 return -ENOBUFS;
1130 memset(cl, 0, sizeof(struct hfsc_class));
1132 if (rsc != NULL)
1133 hfsc_change_rsc(cl, rsc, 0);
1134 if (fsc != NULL)
1135 hfsc_change_fsc(cl, fsc);
1136 if (usc != NULL)
1137 hfsc_change_usc(cl, usc, 0);
1139 cl->refcnt = 1;
1140 cl->classid = classid;
1141 cl->sched = q;
1142 cl->cl_parent = parent;
1143 cl->qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
1144 if (cl->qdisc == NULL)
1145 cl->qdisc = &noop_qdisc;
1146 cl->stats_lock = &sch->dev->queue_lock;
1147 INIT_LIST_HEAD(&cl->children);
1148 cl->vt_tree = RB_ROOT;
1149 cl->cf_tree = RB_ROOT;
1151 sch_tree_lock(sch);
1152 list_add_tail(&cl->hlist, &q->clhash[hfsc_hash(classid)]);
1153 list_add_tail(&cl->siblings, &parent->children);
1154 if (parent->level == 0)
1155 hfsc_purge_queue(sch, parent);
1156 hfsc_adjust_levels(parent);
1157 cl->cl_pcvtoff = parent->cl_cvtoff;
1158 sch_tree_unlock(sch);
1160 #ifdef CONFIG_NET_ESTIMATOR
1161 if (tca[TCA_RATE-1])
1162 gen_new_estimator(&cl->bstats, &cl->rate_est,
1163 cl->stats_lock, tca[TCA_RATE-1]);
1164 #endif
1165 *arg = (unsigned long)cl;
1166 return 0;
1169 static void
1170 hfsc_destroy_filters(struct tcf_proto **fl)
1172 struct tcf_proto *tp;
1174 while ((tp = *fl) != NULL) {
1175 *fl = tp->next;
1176 tcf_destroy(tp);
1180 static void
1181 hfsc_destroy_class(struct Qdisc *sch, struct hfsc_class *cl)
1183 struct hfsc_sched *q = qdisc_priv(sch);
1185 hfsc_destroy_filters(&cl->filter_list);
1186 qdisc_destroy(cl->qdisc);
1187 #ifdef CONFIG_NET_ESTIMATOR
1188 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1189 #endif
1190 if (cl != &q->root)
1191 kfree(cl);
1194 static int
1195 hfsc_delete_class(struct Qdisc *sch, unsigned long arg)
1197 struct hfsc_sched *q = qdisc_priv(sch);
1198 struct hfsc_class *cl = (struct hfsc_class *)arg;
1200 if (cl->level > 0 || cl->filter_cnt > 0 || cl == &q->root)
1201 return -EBUSY;
1203 sch_tree_lock(sch);
1205 list_del(&cl->hlist);
1206 list_del(&cl->siblings);
1207 hfsc_adjust_levels(cl->cl_parent);
1208 hfsc_purge_queue(sch, cl);
1209 if (--cl->refcnt == 0)
1210 hfsc_destroy_class(sch, cl);
1212 sch_tree_unlock(sch);
1213 return 0;
1216 static struct hfsc_class *
1217 hfsc_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
1219 struct hfsc_sched *q = qdisc_priv(sch);
1220 struct hfsc_class *cl;
1221 struct tcf_result res;
1222 struct tcf_proto *tcf;
1223 int result;
1225 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0 &&
1226 (cl = hfsc_find_class(skb->priority, sch)) != NULL)
1227 if (cl->level == 0)
1228 return cl;
1230 *qerr = NET_XMIT_BYPASS;
1231 tcf = q->root.filter_list;
1232 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
1233 #ifdef CONFIG_NET_CLS_ACT
1234 switch (result) {
1235 case TC_ACT_QUEUED:
1236 case TC_ACT_STOLEN:
1237 *qerr = NET_XMIT_SUCCESS;
1238 case TC_ACT_SHOT:
1239 return NULL;
1241 #elif defined(CONFIG_NET_CLS_POLICE)
1242 if (result == TC_POLICE_SHOT)
1243 return NULL;
1244 #endif
1245 if ((cl = (struct hfsc_class *)res.class) == NULL) {
1246 if ((cl = hfsc_find_class(res.classid, sch)) == NULL)
1247 break; /* filter selected invalid classid */
1250 if (cl->level == 0)
1251 return cl; /* hit leaf class */
1253 /* apply inner filter chain */
1254 tcf = cl->filter_list;
1257 /* classification failed, try default class */
1258 cl = hfsc_find_class(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
1259 if (cl == NULL || cl->level > 0)
1260 return NULL;
1262 return cl;
1265 static int
1266 hfsc_graft_class(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1267 struct Qdisc **old)
1269 struct hfsc_class *cl = (struct hfsc_class *)arg;
1271 if (cl == NULL)
1272 return -ENOENT;
1273 if (cl->level > 0)
1274 return -EINVAL;
1275 if (new == NULL) {
1276 new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
1277 if (new == NULL)
1278 new = &noop_qdisc;
1281 sch_tree_lock(sch);
1282 hfsc_purge_queue(sch, cl);
1283 *old = xchg(&cl->qdisc, new);
1284 sch_tree_unlock(sch);
1285 return 0;
1288 static struct Qdisc *
1289 hfsc_class_leaf(struct Qdisc *sch, unsigned long arg)
1291 struct hfsc_class *cl = (struct hfsc_class *)arg;
1293 if (cl != NULL && cl->level == 0)
1294 return cl->qdisc;
1296 return NULL;
1299 static unsigned long
1300 hfsc_get_class(struct Qdisc *sch, u32 classid)
1302 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1304 if (cl != NULL)
1305 cl->refcnt++;
1307 return (unsigned long)cl;
1310 static void
1311 hfsc_put_class(struct Qdisc *sch, unsigned long arg)
1313 struct hfsc_class *cl = (struct hfsc_class *)arg;
1315 if (--cl->refcnt == 0)
1316 hfsc_destroy_class(sch, cl);
1319 static unsigned long
1320 hfsc_bind_tcf(struct Qdisc *sch, unsigned long parent, u32 classid)
1322 struct hfsc_class *p = (struct hfsc_class *)parent;
1323 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1325 if (cl != NULL) {
1326 if (p != NULL && p->level <= cl->level)
1327 return 0;
1328 cl->filter_cnt++;
1331 return (unsigned long)cl;
1334 static void
1335 hfsc_unbind_tcf(struct Qdisc *sch, unsigned long arg)
1337 struct hfsc_class *cl = (struct hfsc_class *)arg;
1339 cl->filter_cnt--;
1342 static struct tcf_proto **
1343 hfsc_tcf_chain(struct Qdisc *sch, unsigned long arg)
1345 struct hfsc_sched *q = qdisc_priv(sch);
1346 struct hfsc_class *cl = (struct hfsc_class *)arg;
1348 if (cl == NULL)
1349 cl = &q->root;
1351 return &cl->filter_list;
1354 static int
1355 hfsc_dump_sc(struct sk_buff *skb, int attr, struct internal_sc *sc)
1357 struct tc_service_curve tsc;
1359 tsc.m1 = sm2m(sc->sm1);
1360 tsc.d = dx2d(sc->dx);
1361 tsc.m2 = sm2m(sc->sm2);
1362 RTA_PUT(skb, attr, sizeof(tsc), &tsc);
1364 return skb->len;
1366 rtattr_failure:
1367 return -1;
1370 static inline int
1371 hfsc_dump_curves(struct sk_buff *skb, struct hfsc_class *cl)
1373 if ((cl->cl_flags & HFSC_RSC) &&
1374 (hfsc_dump_sc(skb, TCA_HFSC_RSC, &cl->cl_rsc) < 0))
1375 goto rtattr_failure;
1377 if ((cl->cl_flags & HFSC_FSC) &&
1378 (hfsc_dump_sc(skb, TCA_HFSC_FSC, &cl->cl_fsc) < 0))
1379 goto rtattr_failure;
1381 if ((cl->cl_flags & HFSC_USC) &&
1382 (hfsc_dump_sc(skb, TCA_HFSC_USC, &cl->cl_usc) < 0))
1383 goto rtattr_failure;
1385 return skb->len;
1387 rtattr_failure:
1388 return -1;
1391 static int
1392 hfsc_dump_class(struct Qdisc *sch, unsigned long arg, struct sk_buff *skb,
1393 struct tcmsg *tcm)
1395 struct hfsc_class *cl = (struct hfsc_class *)arg;
1396 unsigned char *b = skb->tail;
1397 struct rtattr *rta = (struct rtattr *)b;
1399 tcm->tcm_parent = cl->cl_parent ? cl->cl_parent->classid : TC_H_ROOT;
1400 tcm->tcm_handle = cl->classid;
1401 if (cl->level == 0)
1402 tcm->tcm_info = cl->qdisc->handle;
1404 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1405 if (hfsc_dump_curves(skb, cl) < 0)
1406 goto rtattr_failure;
1407 rta->rta_len = skb->tail - b;
1408 return skb->len;
1410 rtattr_failure:
1411 skb_trim(skb, b - skb->data);
1412 return -1;
1415 static int
1416 hfsc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
1417 struct gnet_dump *d)
1419 struct hfsc_class *cl = (struct hfsc_class *)arg;
1420 struct tc_hfsc_stats xstats;
1422 cl->qstats.qlen = cl->qdisc->q.qlen;
1423 xstats.level = cl->level;
1424 xstats.period = cl->cl_vtperiod;
1425 xstats.work = cl->cl_total;
1426 xstats.rtwork = cl->cl_cumul;
1428 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1429 #ifdef CONFIG_NET_ESTIMATOR
1430 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1431 #endif
1432 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1433 return -1;
1435 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
1440 static void
1441 hfsc_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1443 struct hfsc_sched *q = qdisc_priv(sch);
1444 struct hfsc_class *cl;
1445 unsigned int i;
1447 if (arg->stop)
1448 return;
1450 for (i = 0; i < HFSC_HSIZE; i++) {
1451 list_for_each_entry(cl, &q->clhash[i], hlist) {
1452 if (arg->count < arg->skip) {
1453 arg->count++;
1454 continue;
1456 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1457 arg->stop = 1;
1458 return;
1460 arg->count++;
1465 static void
1466 hfsc_watchdog(unsigned long arg)
1468 struct Qdisc *sch = (struct Qdisc *)arg;
1470 sch->flags &= ~TCQ_F_THROTTLED;
1471 netif_schedule(sch->dev);
1474 static void
1475 hfsc_schedule_watchdog(struct Qdisc *sch, u64 cur_time)
1477 struct hfsc_sched *q = qdisc_priv(sch);
1478 struct hfsc_class *cl;
1479 u64 next_time = 0;
1480 long delay;
1482 if ((cl = eltree_get_minel(q)) != NULL)
1483 next_time = cl->cl_e;
1484 if (q->root.cl_cfmin != 0) {
1485 if (next_time == 0 || next_time > q->root.cl_cfmin)
1486 next_time = q->root.cl_cfmin;
1488 ASSERT(next_time != 0);
1489 delay = next_time - cur_time;
1490 delay = PSCHED_US2JIFFIE(delay);
1492 sch->flags |= TCQ_F_THROTTLED;
1493 mod_timer(&q->wd_timer, jiffies + delay);
1496 static int
1497 hfsc_init_qdisc(struct Qdisc *sch, struct rtattr *opt)
1499 struct hfsc_sched *q = qdisc_priv(sch);
1500 struct tc_hfsc_qopt *qopt;
1501 unsigned int i;
1503 if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1504 return -EINVAL;
1505 qopt = RTA_DATA(opt);
1507 sch->stats_lock = &sch->dev->queue_lock;
1509 q->defcls = qopt->defcls;
1510 for (i = 0; i < HFSC_HSIZE; i++)
1511 INIT_LIST_HEAD(&q->clhash[i]);
1512 q->eligible = RB_ROOT;
1513 INIT_LIST_HEAD(&q->droplist);
1514 skb_queue_head_init(&q->requeue);
1516 q->root.refcnt = 1;
1517 q->root.classid = sch->handle;
1518 q->root.sched = q;
1519 q->root.qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
1520 if (q->root.qdisc == NULL)
1521 q->root.qdisc = &noop_qdisc;
1522 q->root.stats_lock = &sch->dev->queue_lock;
1523 INIT_LIST_HEAD(&q->root.children);
1524 q->root.vt_tree = RB_ROOT;
1525 q->root.cf_tree = RB_ROOT;
1527 list_add(&q->root.hlist, &q->clhash[hfsc_hash(q->root.classid)]);
1529 init_timer(&q->wd_timer);
1530 q->wd_timer.function = hfsc_watchdog;
1531 q->wd_timer.data = (unsigned long)sch;
1533 return 0;
1536 static int
1537 hfsc_change_qdisc(struct Qdisc *sch, struct rtattr *opt)
1539 struct hfsc_sched *q = qdisc_priv(sch);
1540 struct tc_hfsc_qopt *qopt;
1542 if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1543 return -EINVAL;
1544 qopt = RTA_DATA(opt);
1546 sch_tree_lock(sch);
1547 q->defcls = qopt->defcls;
1548 sch_tree_unlock(sch);
1550 return 0;
1553 static void
1554 hfsc_reset_class(struct hfsc_class *cl)
1556 cl->cl_total = 0;
1557 cl->cl_cumul = 0;
1558 cl->cl_d = 0;
1559 cl->cl_e = 0;
1560 cl->cl_vt = 0;
1561 cl->cl_vtadj = 0;
1562 cl->cl_vtoff = 0;
1563 cl->cl_cvtmin = 0;
1564 cl->cl_cvtmax = 0;
1565 cl->cl_cvtoff = 0;
1566 cl->cl_pcvtoff = 0;
1567 cl->cl_vtperiod = 0;
1568 cl->cl_parentperiod = 0;
1569 cl->cl_f = 0;
1570 cl->cl_myf = 0;
1571 cl->cl_myfadj = 0;
1572 cl->cl_cfmin = 0;
1573 cl->cl_nactive = 0;
1575 cl->vt_tree = RB_ROOT;
1576 cl->cf_tree = RB_ROOT;
1577 qdisc_reset(cl->qdisc);
1579 if (cl->cl_flags & HFSC_RSC)
1580 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, 0, 0);
1581 if (cl->cl_flags & HFSC_FSC)
1582 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, 0, 0);
1583 if (cl->cl_flags & HFSC_USC)
1584 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, 0, 0);
1587 static void
1588 hfsc_reset_qdisc(struct Qdisc *sch)
1590 struct hfsc_sched *q = qdisc_priv(sch);
1591 struct hfsc_class *cl;
1592 unsigned int i;
1594 for (i = 0; i < HFSC_HSIZE; i++) {
1595 list_for_each_entry(cl, &q->clhash[i], hlist)
1596 hfsc_reset_class(cl);
1598 __skb_queue_purge(&q->requeue);
1599 q->eligible = RB_ROOT;
1600 INIT_LIST_HEAD(&q->droplist);
1601 del_timer(&q->wd_timer);
1602 sch->flags &= ~TCQ_F_THROTTLED;
1603 sch->q.qlen = 0;
1606 static void
1607 hfsc_destroy_qdisc(struct Qdisc *sch)
1609 struct hfsc_sched *q = qdisc_priv(sch);
1610 struct hfsc_class *cl, *next;
1611 unsigned int i;
1613 for (i = 0; i < HFSC_HSIZE; i++) {
1614 list_for_each_entry_safe(cl, next, &q->clhash[i], hlist)
1615 hfsc_destroy_class(sch, cl);
1617 __skb_queue_purge(&q->requeue);
1618 del_timer(&q->wd_timer);
1621 static int
1622 hfsc_dump_qdisc(struct Qdisc *sch, struct sk_buff *skb)
1624 struct hfsc_sched *q = qdisc_priv(sch);
1625 unsigned char *b = skb->tail;
1626 struct tc_hfsc_qopt qopt;
1628 qopt.defcls = q->defcls;
1629 RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
1630 return skb->len;
1632 rtattr_failure:
1633 skb_trim(skb, b - skb->data);
1634 return -1;
1637 static int
1638 hfsc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1640 struct hfsc_class *cl;
1641 unsigned int len;
1642 int err;
1644 cl = hfsc_classify(skb, sch, &err);
1645 if (cl == NULL) {
1646 if (err == NET_XMIT_BYPASS)
1647 sch->qstats.drops++;
1648 kfree_skb(skb);
1649 return err;
1652 len = skb->len;
1653 err = cl->qdisc->enqueue(skb, cl->qdisc);
1654 if (unlikely(err != NET_XMIT_SUCCESS)) {
1655 cl->qstats.drops++;
1656 sch->qstats.drops++;
1657 return err;
1660 if (cl->qdisc->q.qlen == 1)
1661 set_active(cl, len);
1663 cl->bstats.packets++;
1664 cl->bstats.bytes += len;
1665 sch->bstats.packets++;
1666 sch->bstats.bytes += len;
1667 sch->q.qlen++;
1669 return NET_XMIT_SUCCESS;
1672 static struct sk_buff *
1673 hfsc_dequeue(struct Qdisc *sch)
1675 struct hfsc_sched *q = qdisc_priv(sch);
1676 struct hfsc_class *cl;
1677 struct sk_buff *skb;
1678 u64 cur_time;
1679 unsigned int next_len;
1680 int realtime = 0;
1682 if (sch->q.qlen == 0)
1683 return NULL;
1684 if ((skb = __skb_dequeue(&q->requeue)))
1685 goto out;
1687 PSCHED_GET_TIME(cur_time);
1690 * if there are eligible classes, use real-time criteria.
1691 * find the class with the minimum deadline among
1692 * the eligible classes.
1694 if ((cl = eltree_get_mindl(q, cur_time)) != NULL) {
1695 realtime = 1;
1696 } else {
1698 * use link-sharing criteria
1699 * get the class with the minimum vt in the hierarchy
1701 cl = vttree_get_minvt(&q->root, cur_time);
1702 if (cl == NULL) {
1703 sch->qstats.overlimits++;
1704 hfsc_schedule_watchdog(sch, cur_time);
1705 return NULL;
1709 skb = cl->qdisc->dequeue(cl->qdisc);
1710 if (skb == NULL) {
1711 if (net_ratelimit())
1712 printk("HFSC: Non-work-conserving qdisc ?\n");
1713 return NULL;
1716 update_vf(cl, skb->len, cur_time);
1717 if (realtime)
1718 cl->cl_cumul += skb->len;
1720 if (cl->qdisc->q.qlen != 0) {
1721 if (cl->cl_flags & HFSC_RSC) {
1722 /* update ed */
1723 next_len = qdisc_peek_len(cl->qdisc);
1724 if (realtime)
1725 update_ed(cl, next_len);
1726 else
1727 update_d(cl, next_len);
1729 } else {
1730 /* the class becomes passive */
1731 set_passive(cl);
1734 out:
1735 sch->flags &= ~TCQ_F_THROTTLED;
1736 sch->q.qlen--;
1738 return skb;
1741 static int
1742 hfsc_requeue(struct sk_buff *skb, struct Qdisc *sch)
1744 struct hfsc_sched *q = qdisc_priv(sch);
1746 __skb_queue_head(&q->requeue, skb);
1747 sch->q.qlen++;
1748 sch->qstats.requeues++;
1749 return NET_XMIT_SUCCESS;
1752 static unsigned int
1753 hfsc_drop(struct Qdisc *sch)
1755 struct hfsc_sched *q = qdisc_priv(sch);
1756 struct hfsc_class *cl;
1757 unsigned int len;
1759 list_for_each_entry(cl, &q->droplist, dlist) {
1760 if (cl->qdisc->ops->drop != NULL &&
1761 (len = cl->qdisc->ops->drop(cl->qdisc)) > 0) {
1762 if (cl->qdisc->q.qlen == 0) {
1763 update_vf(cl, 0, 0);
1764 set_passive(cl);
1765 } else {
1766 list_move_tail(&cl->dlist, &q->droplist);
1768 cl->qstats.drops++;
1769 sch->qstats.drops++;
1770 sch->q.qlen--;
1771 return len;
1774 return 0;
1777 static struct Qdisc_class_ops hfsc_class_ops = {
1778 .change = hfsc_change_class,
1779 .delete = hfsc_delete_class,
1780 .graft = hfsc_graft_class,
1781 .leaf = hfsc_class_leaf,
1782 .get = hfsc_get_class,
1783 .put = hfsc_put_class,
1784 .bind_tcf = hfsc_bind_tcf,
1785 .unbind_tcf = hfsc_unbind_tcf,
1786 .tcf_chain = hfsc_tcf_chain,
1787 .dump = hfsc_dump_class,
1788 .dump_stats = hfsc_dump_class_stats,
1789 .walk = hfsc_walk
1792 static struct Qdisc_ops hfsc_qdisc_ops = {
1793 .id = "hfsc",
1794 .init = hfsc_init_qdisc,
1795 .change = hfsc_change_qdisc,
1796 .reset = hfsc_reset_qdisc,
1797 .destroy = hfsc_destroy_qdisc,
1798 .dump = hfsc_dump_qdisc,
1799 .enqueue = hfsc_enqueue,
1800 .dequeue = hfsc_dequeue,
1801 .requeue = hfsc_requeue,
1802 .drop = hfsc_drop,
1803 .cl_ops = &hfsc_class_ops,
1804 .priv_size = sizeof(struct hfsc_sched),
1805 .owner = THIS_MODULE
1808 static int __init
1809 hfsc_init(void)
1811 return register_qdisc(&hfsc_qdisc_ops);
1814 static void __exit
1815 hfsc_cleanup(void)
1817 unregister_qdisc(&hfsc_qdisc_ops);
1820 MODULE_LICENSE("GPL");
1821 module_init(hfsc_init);
1822 module_exit(hfsc_cleanup);