projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[NETLINK]: Use nlmsg_trim() where appropriate
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / sched / sch_hfsc.c
CommitLineData
1da177e4
LT		 1/*
2 * Copyright (c) 2003 Patrick McHardy, <kaber@trash.net>
3 *
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.
8 *
9 * 2003-10-17 - Ported from altq
10 */
11/*
12 * Copyright (c) 1997-1999 Carnegie Mellon University. All Rights Reserved.
13 *
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.
19 *
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.
34 *
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.
39 */
40/*
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.
45 *
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.
50 */
51
52#include <linux/kernel.h>
1da177e4
LT		 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>
1da177e4
LT		 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>
dc5fc579 68#include <net/netlink.h>
1da177e4
LT		 69#include <net/pkt_sched.h>
70#include <net/pkt_cls.h>
71#include <asm/system.h>
72#include <asm/div64.h>
73
1da177e4
LT		 74/*
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.
79 *
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.
85 */
86
87struct internal_sc
88{
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 */
95};
96
97/* runtime service curve */
98struct runtime_sc
99{
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 */
108};
109
110enum hfsc_class_flags
111{
112 HFSC_RSC = 0x1,
113 HFSC_FSC = 0x2,
114 HFSC_USC = 0x4
115};
116
117struct hfsc_class
118{
119 u32 classid; /* class id */
120 unsigned int refcnt; /* usage count */
121
122 struct gnet_stats_basic bstats;
123 struct gnet_stats_queue qstats;
124 struct gnet_stats_rate_est rate_est;
125 spinlock_t *stats_lock;
126 unsigned int level; /* class level in hierarchy */
127 struct tcf_proto *filter_list; /* filter list */
128 unsigned int filter_cnt; /* filter count */
129
130 struct hfsc_sched *sched; /* scheduler data */
131 struct hfsc_class *cl_parent; /* parent class */
132 struct list_head siblings; /* sibling classes */
133 struct list_head children; /* child classes */
134 struct Qdisc *qdisc; /* leaf qdisc */
135
136 struct rb_node el_node; /* qdisc's eligible tree member */
137 struct rb_root vt_tree; /* active children sorted by cl_vt */
138 struct rb_node vt_node; /* parent's vt_tree member */
139 struct rb_root cf_tree; /* active children sorted by cl_f */
140 struct rb_node cf_node; /* parent's cf_heap member */
141 struct list_head hlist; /* hash list member */
142 struct list_head dlist; /* drop list member */
143
144 u64 cl_total; /* total work in bytes */
145 u64 cl_cumul; /* cumulative work in bytes done by
146 real-time criteria */
147
148 u64 cl_d; /* deadline*/
149 u64 cl_e; /* eligible time */
150 u64 cl_vt; /* virtual time */
151 u64 cl_f; /* time when this class will fit for
152 link-sharing, max(myf, cfmin) */
153 u64 cl_myf; /* my fit-time (calculated from this
154 class's own upperlimit curve) */
155 u64 cl_myfadj; /* my fit-time adjustment (to cancel
156 history dependence) */
157 u64 cl_cfmin; /* earliest children's fit-time (used
158 with cl_myf to obtain cl_f) */
159 u64 cl_cvtmin; /* minimal virtual time among the
160 children fit for link-sharing
161 (monotonic within a period) */
162 u64 cl_vtadj; /* intra-period cumulative vt
163 adjustment */
164 u64 cl_vtoff; /* inter-period cumulative vt offset */
165 u64 cl_cvtmax; /* max child's vt in the last period */
166 u64 cl_cvtoff; /* cumulative cvtmax of all periods */
167 u64 cl_pcvtoff; /* parent's cvtoff at initalization
168 time */
169
170 struct internal_sc cl_rsc; /* internal real-time service curve */
171 struct internal_sc cl_fsc; /* internal fair service curve */
172 struct internal_sc cl_usc; /* internal upperlimit service curve */
173 struct runtime_sc cl_deadline; /* deadline curve */
174 struct runtime_sc cl_eligible; /* eligible curve */
175 struct runtime_sc cl_virtual; /* virtual curve */
176 struct runtime_sc cl_ulimit; /* upperlimit curve */
177
178 unsigned long cl_flags; /* which curves are valid */
179 unsigned long cl_vtperiod; /* vt period sequence number */
180 unsigned long cl_parentperiod;/* parent's vt period sequence number*/
181 unsigned long cl_nactive; /* number of active children */
182};
183
184#define HFSC_HSIZE 16
185
186struct hfsc_sched
187{
188 u16 defcls; /* default class id */
189 struct hfsc_class root; /* root class */
190 struct list_head clhash[HFSC_HSIZE]; /* class hash */
191 struct rb_root eligible; /* eligible tree */
192 struct list_head droplist; /* active leaf class list (for
193 dropping) */
194 struct sk_buff_head requeue; /* requeued packet */
ed2b229a 195 struct qdisc_watchdog watchdog; /* watchdog timer */
1da177e4
LT		 196};
197
1da177e4
LT		 198#define HT_INFINITY 0xffffffffffffffffULL /* infinite time value */
199
200
201/*
202 * eligible tree holds backlogged classes being sorted by their eligible times.
203 * there is one eligible tree per hfsc instance.
204 */
205
206static void
207eltree_insert(struct hfsc_class *cl)
208{
209 struct rb_node **p = &cl->sched->eligible.rb_node;
210 struct rb_node *parent = NULL;
211 struct hfsc_class *cl1;
212
213 while (*p != NULL) {
214 parent = *p;
215 cl1 = rb_entry(parent, struct hfsc_class, el_node);
216 if (cl->cl_e >= cl1->cl_e)
217 p = &parent->rb_right;
218 else
219 p = &parent->rb_left;
220 }
221 rb_link_node(&cl->el_node, parent, p);
222 rb_insert_color(&cl->el_node, &cl->sched->eligible);
223}
224
225static inline void
226eltree_remove(struct hfsc_class *cl)
227{
228 rb_erase(&cl->el_node, &cl->sched->eligible);
229}
230
231static inline void
232eltree_update(struct hfsc_class *cl)
233{
234 eltree_remove(cl);
235 eltree_insert(cl);
236}
237
238/* find the class with the minimum deadline among the eligible classes */
239static inline struct hfsc_class *
240eltree_get_mindl(struct hfsc_sched *q, u64 cur_time)
241{
242 struct hfsc_class *p, *cl = NULL;
243 struct rb_node *n;
244
245 for (n = rb_first(&q->eligible); n != NULL; n = rb_next(n)) {
246 p = rb_entry(n, struct hfsc_class, el_node);
247 if (p->cl_e > cur_time)
248 break;
249 if (cl == NULL || p->cl_d < cl->cl_d)
250 cl = p;
251 }
252 return cl;
253}
254
255/* find the class with minimum eligible time among the eligible classes */
256static inline struct hfsc_class *
257eltree_get_minel(struct hfsc_sched *q)
258{
259 struct rb_node *n;
10297b99 260
1da177e4
LT		 261 n = rb_first(&q->eligible);
262 if (n == NULL)
263 return NULL;
264 return rb_entry(n, struct hfsc_class, el_node);
265}
266
267/*
268 * vttree holds holds backlogged child classes being sorted by their virtual
269 * time. each intermediate class has one vttree.
270 */
271static void
272vttree_insert(struct hfsc_class *cl)
273{
274 struct rb_node **p = &cl->cl_parent->vt_tree.rb_node;
275 struct rb_node *parent = NULL;
276 struct hfsc_class *cl1;
277
278 while (*p != NULL) {
279 parent = *p;
280 cl1 = rb_entry(parent, struct hfsc_class, vt_node);
281 if (cl->cl_vt >= cl1->cl_vt)
282 p = &parent->rb_right;
283 else
284 p = &parent->rb_left;
285 }
286 rb_link_node(&cl->vt_node, parent, p);
287 rb_insert_color(&cl->vt_node, &cl->cl_parent->vt_tree);
288}
289
290static inline void
291vttree_remove(struct hfsc_class *cl)
292{
293 rb_erase(&cl->vt_node, &cl->cl_parent->vt_tree);
294}
295
296static inline void
297vttree_update(struct hfsc_class *cl)
298{
299 vttree_remove(cl);
300 vttree_insert(cl);
301}
302
303static inline struct hfsc_class *
304vttree_firstfit(struct hfsc_class *cl, u64 cur_time)
305{
306 struct hfsc_class *p;
307 struct rb_node *n;
308
309 for (n = rb_first(&cl->vt_tree); n != NULL; n = rb_next(n)) {
310 p = rb_entry(n, struct hfsc_class, vt_node);
311 if (p->cl_f <= cur_time)
312 return p;
313 }
314 return NULL;
315}
316
317/*
318 * get the leaf class with the minimum vt in the hierarchy
319 */
320static struct hfsc_class *
321vttree_get_minvt(struct hfsc_class *cl, u64 cur_time)
322{
323 /* if root-class's cfmin is bigger than cur_time nothing to do */
324 if (cl->cl_cfmin > cur_time)
325 return NULL;
326
327 while (cl->level > 0) {
328 cl = vttree_firstfit(cl, cur_time);
329 if (cl == NULL)
330 return NULL;
331 /*
332 * update parent's cl_cvtmin.
333 */
334 if (cl->cl_parent->cl_cvtmin < cl->cl_vt)
335 cl->cl_parent->cl_cvtmin = cl->cl_vt;
336 }
337 return cl;
338}
339
340static void
341cftree_insert(struct hfsc_class *cl)
342{
343 struct rb_node **p = &cl->cl_parent->cf_tree.rb_node;
344 struct rb_node *parent = NULL;
345 struct hfsc_class *cl1;
346
347 while (*p != NULL) {
348 parent = *p;
349 cl1 = rb_entry(parent, struct hfsc_class, cf_node);
350 if (cl->cl_f >= cl1->cl_f)
351 p = &parent->rb_right;
352 else
353 p = &parent->rb_left;
354 }
355 rb_link_node(&cl->cf_node, parent, p);
356 rb_insert_color(&cl->cf_node, &cl->cl_parent->cf_tree);
357}
358
359static inline void
360cftree_remove(struct hfsc_class *cl)
361{
362 rb_erase(&cl->cf_node, &cl->cl_parent->cf_tree);
363}
364
365static inline void
366cftree_update(struct hfsc_class *cl)
367{
368 cftree_remove(cl);
369 cftree_insert(cl);
370}
371
372/*
373 * service curve support functions
374 *
375 * external service curve parameters
376 * m: bps
377 * d: us
378 * internal service curve parameters
379 * sm: (bytes/psched_us) << SM_SHIFT
380 * ism: (psched_us/byte) << ISM_SHIFT
381 * dx: psched_us
382 *
641b9e0e 383 * The clock source resolution with ktime is 1.024us.
1da177e4
LT		 384 *
385 * sm and ism are scaled in order to keep effective digits.
386 * SM_SHIFT and ISM_SHIFT are selected to keep at least 4 effective
387 * digits in decimal using the following table.
388 *
1da177e4
LT		 389 * bits/sec 100Kbps 1Mbps 10Mbps 100Mbps 1Gbps
390 * ------------+-------------------------------------------------------
641b9e0e 391 * bytes/1.024us 12.8e-3 128e-3 1280e-3 12800e-3 128000e-3
1da177e4 392 *
641b9e0e 393 * 1.024us/byte 78.125 7.8125 0.78125 0.078125 0.0078125
1da177e4
LT		 394 */
395#define SM_SHIFT 20
396#define ISM_SHIFT 18
397
398#define SM_MASK ((1ULL << SM_SHIFT) - 1)
399#define ISM_MASK ((1ULL << ISM_SHIFT) - 1)
400
401static inline u64
402seg_x2y(u64 x, u64 sm)
403{
404 u64 y;
405
406 /*
407 * compute
408 * y = x * sm >> SM_SHIFT
409 * but divide it for the upper and lower bits to avoid overflow
410 */
411 y = (x >> SM_SHIFT) * sm + (((x & SM_MASK) * sm) >> SM_SHIFT);
412 return y;
413}
414
415static inline u64
416seg_y2x(u64 y, u64 ism)
417{
418 u64 x;
419
420 if (y == 0)
421 x = 0;
422 else if (ism == HT_INFINITY)
423 x = HT_INFINITY;
424 else {
425 x = (y >> ISM_SHIFT) * ism
426 + (((y & ISM_MASK) * ism) >> ISM_SHIFT);
427 }
428 return x;
429}
430
431/* Convert m (bps) into sm (bytes/psched us) */
432static u64
433m2sm(u32 m)
434{
435 u64 sm;
436
437 sm = ((u64)m << SM_SHIFT);
00c04af9
PM		 438 sm += PSCHED_TICKS_PER_SEC - 1;
439 do_div(sm, PSCHED_TICKS_PER_SEC);
1da177e4
LT		 440 return sm;
441}
442
443/* convert m (bps) into ism (psched us/byte) */
444static u64
445m2ism(u32 m)
446{
447 u64 ism;
448
449 if (m == 0)
450 ism = HT_INFINITY;
451 else {
00c04af9 452 ism = ((u64)PSCHED_TICKS_PER_SEC << ISM_SHIFT);
1da177e4
LT		 453 ism += m - 1;
454 do_div(ism, m);
455 }
456 return ism;
457}
458
459/* convert d (us) into dx (psched us) */
460static u64
461d2dx(u32 d)
462{
463 u64 dx;
464
00c04af9 465 dx = ((u64)d * PSCHED_TICKS_PER_SEC);
538e43a4
PM		 466 dx += USEC_PER_SEC - 1;
467 do_div(dx, USEC_PER_SEC);
1da177e4
LT		 468 return dx;
469}
470
471/* convert sm (bytes/psched us) into m (bps) */
472static u32
473sm2m(u64 sm)
474{
475 u64 m;
476
00c04af9 477 m = (sm * PSCHED_TICKS_PER_SEC) >> SM_SHIFT;
1da177e4
LT		 478 return (u32)m;
479}
480
481/* convert dx (psched us) into d (us) */
482static u32
483dx2d(u64 dx)
484{
485 u64 d;
486
538e43a4 487 d = dx * USEC_PER_SEC;
00c04af9 488 do_div(d, PSCHED_TICKS_PER_SEC);
1da177e4
LT		 489 return (u32)d;
490}
491
492static void
493sc2isc(struct tc_service_curve *sc, struct internal_sc *isc)
494{
495 isc->sm1 = m2sm(sc->m1);
496 isc->ism1 = m2ism(sc->m1);
497 isc->dx = d2dx(sc->d);
498 isc->dy = seg_x2y(isc->dx, isc->sm1);
499 isc->sm2 = m2sm(sc->m2);
500 isc->ism2 = m2ism(sc->m2);
501}
502
503/*
504 * initialize the runtime service curve with the given internal
505 * service curve starting at (x, y).
506 */
507static void
508rtsc_init(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
509{
510 rtsc->x = x;
511 rtsc->y = y;
512 rtsc->sm1 = isc->sm1;
513 rtsc->ism1 = isc->ism1;
514 rtsc->dx = isc->dx;
515 rtsc->dy = isc->dy;
516 rtsc->sm2 = isc->sm2;
517 rtsc->ism2 = isc->ism2;
518}
519
520/*
521 * calculate the y-projection of the runtime service curve by the
522 * given x-projection value
523 */
524static u64
525rtsc_y2x(struct runtime_sc *rtsc, u64 y)
526{
527 u64 x;
528
529 if (y < rtsc->y)
530 x = rtsc->x;
531 else if (y <= rtsc->y + rtsc->dy) {
532 /* x belongs to the 1st segment */
533 if (rtsc->dy == 0)
534 x = rtsc->x + rtsc->dx;
535 else
536 x = rtsc->x + seg_y2x(y - rtsc->y, rtsc->ism1);
537 } else {
538 /* x belongs to the 2nd segment */
539 x = rtsc->x + rtsc->dx
540 + seg_y2x(y - rtsc->y - rtsc->dy, rtsc->ism2);
541 }
542 return x;
543}
544
545static u64
546rtsc_x2y(struct runtime_sc *rtsc, u64 x)
547{
548 u64 y;
549
550 if (x <= rtsc->x)
551 y = rtsc->y;
552 else if (x <= rtsc->x + rtsc->dx)
553 /* y belongs to the 1st segment */
554 y = rtsc->y + seg_x2y(x - rtsc->x, rtsc->sm1);
555 else
556 /* y belongs to the 2nd segment */
557 y = rtsc->y + rtsc->dy
558 + seg_x2y(x - rtsc->x - rtsc->dx, rtsc->sm2);
559 return y;
560}
561
562/*
563 * update the runtime service curve by taking the minimum of the current
564 * runtime service curve and the service curve starting at (x, y).
565 */
566static void
567rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
568{
569 u64 y1, y2, dx, dy;
570 u32 dsm;
571
572 if (isc->sm1 <= isc->sm2) {
573 /* service curve is convex */
574 y1 = rtsc_x2y(rtsc, x);
575 if (y1 < y)
576 /* the current rtsc is smaller */
577 return;
578 rtsc->x = x;
579 rtsc->y = y;
580 return;
581 }
582
583 /*
584 * service curve is concave
585 * compute the two y values of the current rtsc
586 * y1: at x
587 * y2: at (x + dx)
588 */
589 y1 = rtsc_x2y(rtsc, x);
590 if (y1 <= y) {
591 /* rtsc is below isc, no change to rtsc */
592 return;
593 }
594
595 y2 = rtsc_x2y(rtsc, x + isc->dx);
596 if (y2 >= y + isc->dy) {
597 /* rtsc is above isc, replace rtsc by isc */
598 rtsc->x = x;
599 rtsc->y = y;
600 rtsc->dx = isc->dx;
601 rtsc->dy = isc->dy;
602 return;
603 }
604
605 /*
606 * the two curves intersect
607 * compute the offsets (dx, dy) using the reverse
608 * function of seg_x2y()
609 * seg_x2y(dx, sm1) == seg_x2y(dx, sm2) + (y1 - y)
610 */
611 dx = (y1 - y) << SM_SHIFT;
612 dsm = isc->sm1 - isc->sm2;
613 do_div(dx, dsm);
614 /*
615 * check if (x, y1) belongs to the 1st segment of rtsc.
616 * if so, add the offset.
617 */
618 if (rtsc->x + rtsc->dx > x)
619 dx += rtsc->x + rtsc->dx - x;
620 dy = seg_x2y(dx, isc->sm1);
621
622 rtsc->x = x;
623 rtsc->y = y;
624 rtsc->dx = dx;
625 rtsc->dy = dy;
626 return;
627}
628
629static void
630init_ed(struct hfsc_class *cl, unsigned int next_len)
631{
632 u64 cur_time;
633
634 PSCHED_GET_TIME(cur_time);
635
636 /* update the deadline curve */
637 rtsc_min(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
638
639 /*
640 * update the eligible curve.
641 * for concave, it is equal to the deadline curve.
642 * for convex, it is a linear curve with slope m2.
643 */
644 cl->cl_eligible = cl->cl_deadline;
645 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
646 cl->cl_eligible.dx = 0;
647 cl->cl_eligible.dy = 0;
648 }
649
650 /* compute e and d */
651 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
652 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
653
654 eltree_insert(cl);
655}
656
657static void
658update_ed(struct hfsc_class *cl, unsigned int next_len)
659{
660 cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
661 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
662
663 eltree_update(cl);
664}
665
666static inline void
667update_d(struct hfsc_class *cl, unsigned int next_len)
668{
669 cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
670}
671
672static inline void
673update_cfmin(struct hfsc_class *cl)
674{
675 struct rb_node *n = rb_first(&cl->cf_tree);
676 struct hfsc_class *p;
677
678 if (n == NULL) {
679 cl->cl_cfmin = 0;
680 return;
681 }
682 p = rb_entry(n, struct hfsc_class, cf_node);
683 cl->cl_cfmin = p->cl_f;
684}
685
686static void
687init_vf(struct hfsc_class *cl, unsigned int len)
688{
689 struct hfsc_class *max_cl;
690 struct rb_node *n;
691 u64 vt, f, cur_time;
692 int go_active;
693
694 cur_time = 0;
695 go_active = 1;
696 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
697 if (go_active && cl->cl_nactive++ == 0)
698 go_active = 1;
699 else
700 go_active = 0;
701
702 if (go_active) {
703 n = rb_last(&cl->cl_parent->vt_tree);
704 if (n != NULL) {
705 max_cl = rb_entry(n, struct hfsc_class,vt_node);
706 /*
707 * set vt to the average of the min and max
708 * classes. if the parent's period didn't
709 * change, don't decrease vt of the class.
710 */
711 vt = max_cl->cl_vt;
712 if (cl->cl_parent->cl_cvtmin != 0)
713 vt = (cl->cl_parent->cl_cvtmin + vt)/2;
714
715 if (cl->cl_parent->cl_vtperiod !=
716 cl->cl_parentperiod || vt > cl->cl_vt)
717 cl->cl_vt = vt;
718 } else {
719 /*
720 * first child for a new parent backlog period.
721 * add parent's cvtmax to cvtoff to make a new
722 * vt (vtoff + vt) larger than the vt in the
723 * last period for all children.
724 */
725 vt = cl->cl_parent->cl_cvtmax;
726 cl->cl_parent->cl_cvtoff += vt;
727 cl->cl_parent->cl_cvtmax = 0;
728 cl->cl_parent->cl_cvtmin = 0;
729 cl->cl_vt = 0;
730 }
731
732 cl->cl_vtoff = cl->cl_parent->cl_cvtoff -
733 cl->cl_pcvtoff;
734
735 /* update the virtual curve */
736 vt = cl->cl_vt + cl->cl_vtoff;
737 rtsc_min(&cl->cl_virtual, &cl->cl_fsc, vt,
10297b99 738 cl->cl_total);
1da177e4
LT		 739 if (cl->cl_virtual.x == vt) {
740 cl->cl_virtual.x -= cl->cl_vtoff;
741 cl->cl_vtoff = 0;
742 }
743 cl->cl_vtadj = 0;
744
745 cl->cl_vtperiod++; /* increment vt period */
746 cl->cl_parentperiod = cl->cl_parent->cl_vtperiod;
747 if (cl->cl_parent->cl_nactive == 0)
748 cl->cl_parentperiod++;
749 cl->cl_f = 0;
750
751 vttree_insert(cl);
752 cftree_insert(cl);
753
754 if (cl->cl_flags & HFSC_USC) {
755 /* class has upper limit curve */
756 if (cur_time == 0)
757 PSCHED_GET_TIME(cur_time);
758
759 /* update the ulimit curve */
760 rtsc_min(&cl->cl_ulimit, &cl->cl_usc, cur_time,
10297b99 761 cl->cl_total);
1da177e4
LT		 762 /* compute myf */
763 cl->cl_myf = rtsc_y2x(&cl->cl_ulimit,
10297b99 764 cl->cl_total);
1da177e4
LT		 765 cl->cl_myfadj = 0;
766 }
767 }
768
769 f = max(cl->cl_myf, cl->cl_cfmin);
770 if (f != cl->cl_f) {
771 cl->cl_f = f;
772 cftree_update(cl);
773 update_cfmin(cl->cl_parent);
774 }
775 }
776}
777
778static void
779update_vf(struct hfsc_class *cl, unsigned int len, u64 cur_time)
780{
781 u64 f; /* , myf_bound, delta; */
782 int go_passive = 0;
783
784 if (cl->qdisc->q.qlen == 0 && cl->cl_flags & HFSC_FSC)
785 go_passive = 1;
786
787 for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
788 cl->cl_total += len;
789
790 if (!(cl->cl_flags & HFSC_FSC) || cl->cl_nactive == 0)
791 continue;
792
793 if (go_passive && --cl->cl_nactive == 0)
794 go_passive = 1;
795 else
796 go_passive = 0;
797
798 if (go_passive) {
799 /* no more active child, going passive */
800
801 /* update cvtmax of the parent class */
802 if (cl->cl_vt > cl->cl_parent->cl_cvtmax)
803 cl->cl_parent->cl_cvtmax = cl->cl_vt;
804
805 /* remove this class from the vt tree */
806 vttree_remove(cl);
807
808 cftree_remove(cl);
809 update_cfmin(cl->cl_parent);
810
811 continue;
812 }
813
814 /*
815 * update vt and f
816 */
817 cl->cl_vt = rtsc_y2x(&cl->cl_virtual, cl->cl_total)
10297b99 818 - cl->cl_vtoff + cl->cl_vtadj;
1da177e4
LT		 819
820 /*
821 * if vt of the class is smaller than cvtmin,
822 * the class was skipped in the past due to non-fit.
823 * if so, we need to adjust vtadj.
824 */
825 if (cl->cl_vt < cl->cl_parent->cl_cvtmin) {
826 cl->cl_vtadj += cl->cl_parent->cl_cvtmin - cl->cl_vt;
827 cl->cl_vt = cl->cl_parent->cl_cvtmin;
828 }
829
830 /* update the vt tree */
831 vttree_update(cl);
832
833 if (cl->cl_flags & HFSC_USC) {
834 cl->cl_myf = cl->cl_myfadj + rtsc_y2x(&cl->cl_ulimit,
10297b99 835 cl->cl_total);
1da177e4
LT		 836#if 0
837 /*
838 * This code causes classes to stay way under their
839 * limit when multiple classes are used at gigabit
840 * speed. needs investigation. -kaber
841 */
842 /*
843 * if myf lags behind by more than one clock tick
844 * from the current time, adjust myfadj to prevent
845 * a rate-limited class from going greedy.
846 * in a steady state under rate-limiting, myf
847 * fluctuates within one clock tick.
848 */
849 myf_bound = cur_time - PSCHED_JIFFIE2US(1);
850 if (cl->cl_myf < myf_bound) {
851 delta = cur_time - cl->cl_myf;
852 cl->cl_myfadj += delta;
853 cl->cl_myf += delta;
854 }
855#endif
856 }
857
858 f = max(cl->cl_myf, cl->cl_cfmin);
859 if (f != cl->cl_f) {
860 cl->cl_f = f;
861 cftree_update(cl);
862 update_cfmin(cl->cl_parent);
863 }
864 }
865}
866
867static void
868set_active(struct hfsc_class *cl, unsigned int len)
869{
870 if (cl->cl_flags & HFSC_RSC)
871 init_ed(cl, len);
872 if (cl->cl_flags & HFSC_FSC)
873 init_vf(cl, len);
874
875 list_add_tail(&cl->dlist, &cl->sched->droplist);
876}
877
878static void
879set_passive(struct hfsc_class *cl)
880{
881 if (cl->cl_flags & HFSC_RSC)
882 eltree_remove(cl);
883
884 list_del(&cl->dlist);
885
886 /*
887 * vttree is now handled in update_vf() so that update_vf(cl, 0, 0)
888 * needs to be called explicitly to remove a class from vttree.
889 */
890}
891
892/*
893 * hack to get length of first packet in queue.
894 */
895static unsigned int
896qdisc_peek_len(struct Qdisc *sch)
897{
898 struct sk_buff *skb;
899 unsigned int len;
900
901 skb = sch->dequeue(sch);
902 if (skb == NULL) {
903 if (net_ratelimit())
904 printk("qdisc_peek_len: non work-conserving qdisc ?\n");
905 return 0;
906 }
907 len = skb->len;
908 if (unlikely(sch->ops->requeue(skb, sch) != NET_XMIT_SUCCESS)) {
909 if (net_ratelimit())
910 printk("qdisc_peek_len: failed to requeue\n");
e488eafc 911 qdisc_tree_decrease_qlen(sch, 1);
1da177e4
LT		 912 return 0;
913 }
914 return len;
915}
916
917static void
918hfsc_purge_queue(struct Qdisc *sch, struct hfsc_class *cl)
919{
920 unsigned int len = cl->qdisc->q.qlen;
921
922 qdisc_reset(cl->qdisc);
f973b913 923 qdisc_tree_decrease_qlen(cl->qdisc, len);
1da177e4
LT		 924}
925
926static void
927hfsc_adjust_levels(struct hfsc_class *cl)
928{
929 struct hfsc_class *p;
930 unsigned int level;
931
932 do {
933 level = 0;
934 list_for_each_entry(p, &cl->children, siblings) {
210525d6
PM		 935 if (p->level >= level)
936 level = p->level + 1;
1da177e4 937 }
210525d6 938 cl->level = level;
1da177e4
LT		 939 } while ((cl = cl->cl_parent) != NULL);
940}
941
942static inline unsigned int
943hfsc_hash(u32 h)
944{
945 h ^= h >> 8;
946 h ^= h >> 4;
947
948 return h & (HFSC_HSIZE - 1);
949}
950
951static inline struct hfsc_class *
952hfsc_find_class(u32 classid, struct Qdisc *sch)
953{
954 struct hfsc_sched *q = qdisc_priv(sch);
955 struct hfsc_class *cl;
956
957 list_for_each_entry(cl, &q->clhash[hfsc_hash(classid)], hlist) {
958 if (cl->classid == classid)
959 return cl;
960 }
961 return NULL;
962}
963
964static void
965hfsc_change_rsc(struct hfsc_class *cl, struct tc_service_curve *rsc,
10297b99 966 u64 cur_time)
1da177e4
LT		 967{
968 sc2isc(rsc, &cl->cl_rsc);
969 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
970 cl->cl_eligible = cl->cl_deadline;
971 if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
972 cl->cl_eligible.dx = 0;
973 cl->cl_eligible.dy = 0;
974 }
975 cl->cl_flags |= HFSC_RSC;
976}
977
978static void
979hfsc_change_fsc(struct hfsc_class *cl, struct tc_service_curve *fsc)
980{
981 sc2isc(fsc, &cl->cl_fsc);
982 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, cl->cl_vt, cl->cl_total);
983 cl->cl_flags |= HFSC_FSC;
984}
985
986static void
987hfsc_change_usc(struct hfsc_class *cl, struct tc_service_curve *usc,
10297b99 988 u64 cur_time)
1da177e4
LT		 989{
990 sc2isc(usc, &cl->cl_usc);
991 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, cur_time, cl->cl_total);
992 cl->cl_flags |= HFSC_USC;
993}
994
995static int
996hfsc_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
10297b99 997 struct rtattr **tca, unsigned long *arg)
1da177e4
LT		 998{
999 struct hfsc_sched *q = qdisc_priv(sch);
1000 struct hfsc_class *cl = (struct hfsc_class *)*arg;
1001 struct hfsc_class *parent = NULL;
1002 struct rtattr *opt = tca[TCA_OPTIONS-1];
1003 struct rtattr *tb[TCA_HFSC_MAX];
1004 struct tc_service_curve *rsc = NULL, *fsc = NULL, *usc = NULL;
1005 u64 cur_time;
1006
1007 if (opt == NULL || rtattr_parse_nested(tb, TCA_HFSC_MAX, opt))
1008 return -EINVAL;
1009
1010 if (tb[TCA_HFSC_RSC-1]) {
1011 if (RTA_PAYLOAD(tb[TCA_HFSC_RSC-1]) < sizeof(*rsc))
1012 return -EINVAL;
1013 rsc = RTA_DATA(tb[TCA_HFSC_RSC-1]);
1014 if (rsc->m1 == 0 && rsc->m2 == 0)
1015 rsc = NULL;
1016 }
1017
1018 if (tb[TCA_HFSC_FSC-1]) {
1019 if (RTA_PAYLOAD(tb[TCA_HFSC_FSC-1]) < sizeof(*fsc))
1020 return -EINVAL;
1021 fsc = RTA_DATA(tb[TCA_HFSC_FSC-1]);
1022 if (fsc->m1 == 0 && fsc->m2 == 0)
1023 fsc = NULL;
1024 }
1025
1026 if (tb[TCA_HFSC_USC-1]) {
1027 if (RTA_PAYLOAD(tb[TCA_HFSC_USC-1]) < sizeof(*usc))
1028 return -EINVAL;
1029 usc = RTA_DATA(tb[TCA_HFSC_USC-1]);
1030 if (usc->m1 == 0 && usc->m2 == 0)
1031 usc = NULL;
1032 }
1033
1034 if (cl != NULL) {
1035 if (parentid) {
1036 if (cl->cl_parent && cl->cl_parent->classid != parentid)
1037 return -EINVAL;
1038 if (cl->cl_parent == NULL && parentid != TC_H_ROOT)
1039 return -EINVAL;
1040 }
1041 PSCHED_GET_TIME(cur_time);
1042
1043 sch_tree_lock(sch);
1044 if (rsc != NULL)
1045 hfsc_change_rsc(cl, rsc, cur_time);
1046 if (fsc != NULL)
1047 hfsc_change_fsc(cl, fsc);
1048 if (usc != NULL)
1049 hfsc_change_usc(cl, usc, cur_time);
1050
1051 if (cl->qdisc->q.qlen != 0) {
1052 if (cl->cl_flags & HFSC_RSC)
1053 update_ed(cl, qdisc_peek_len(cl->qdisc));
1054 if (cl->cl_flags & HFSC_FSC)
1055 update_vf(cl, 0, cur_time);
1056 }
1057 sch_tree_unlock(sch);
1058
1059#ifdef CONFIG_NET_ESTIMATOR
1060 if (tca[TCA_RATE-1])
1061 gen_replace_estimator(&cl->bstats, &cl->rate_est,
1062 cl->stats_lock, tca[TCA_RATE-1]);
1063#endif
1064 return 0;
1065 }
1066
1067 if (parentid == TC_H_ROOT)
1068 return -EEXIST;
1069
1070 parent = &q->root;
1071 if (parentid) {
1072 parent = hfsc_find_class(parentid, sch);
1073 if (parent == NULL)
1074 return -ENOENT;
1075 }
1076
1077 if (classid == 0 || TC_H_MAJ(classid ^ sch->handle) != 0)
1078 return -EINVAL;
1079 if (hfsc_find_class(classid, sch))
1080 return -EEXIST;
1081
1082 if (rsc == NULL && fsc == NULL)
1083 return -EINVAL;
1084
0da974f4 1085 cl = kzalloc(sizeof(struct hfsc_class), GFP_KERNEL);
1da177e4
LT		 1086 if (cl == NULL)
1087 return -ENOBUFS;
1da177e4
LT		 1088
1089 if (rsc != NULL)
1090 hfsc_change_rsc(cl, rsc, 0);
1091 if (fsc != NULL)
1092 hfsc_change_fsc(cl, fsc);
1093 if (usc != NULL)
1094 hfsc_change_usc(cl, usc, 0);
1095
1096 cl->refcnt = 1;
1097 cl->classid = classid;
1098 cl->sched = q;
1099 cl->cl_parent = parent;
9f9afec4 1100 cl->qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, classid);
1da177e4
LT		 1101 if (cl->qdisc == NULL)
1102 cl->qdisc = &noop_qdisc;
1103 cl->stats_lock = &sch->dev->queue_lock;
1104 INIT_LIST_HEAD(&cl->children);
1105 cl->vt_tree = RB_ROOT;
1106 cl->cf_tree = RB_ROOT;
1107
1108 sch_tree_lock(sch);
1109 list_add_tail(&cl->hlist, &q->clhash[hfsc_hash(classid)]);
1110 list_add_tail(&cl->siblings, &parent->children);
1111 if (parent->level == 0)
1112 hfsc_purge_queue(sch, parent);
1113 hfsc_adjust_levels(parent);
1114 cl->cl_pcvtoff = parent->cl_cvtoff;
1115 sch_tree_unlock(sch);
1116
1117#ifdef CONFIG_NET_ESTIMATOR
1118 if (tca[TCA_RATE-1])
1119 gen_new_estimator(&cl->bstats, &cl->rate_est,
1120 cl->stats_lock, tca[TCA_RATE-1]);
1121#endif
1122 *arg = (unsigned long)cl;
1123 return 0;
1124}
1125
1126static void
1127hfsc_destroy_filters(struct tcf_proto **fl)
1128{
1129 struct tcf_proto *tp;
1130
1131 while ((tp = *fl) != NULL) {
1132 *fl = tp->next;
1133 tcf_destroy(tp);
1134 }
1135}
1136
1137static void
1138hfsc_destroy_class(struct Qdisc *sch, struct hfsc_class *cl)
1139{
1140 struct hfsc_sched *q = qdisc_priv(sch);
1141
1142 hfsc_destroy_filters(&cl->filter_list);
1143 qdisc_destroy(cl->qdisc);
1144#ifdef CONFIG_NET_ESTIMATOR
1145 gen_kill_estimator(&cl->bstats, &cl->rate_est);
1146#endif
1147 if (cl != &q->root)
1148 kfree(cl);
1149}
1150
1151static int
1152hfsc_delete_class(struct Qdisc *sch, unsigned long arg)
1153{
1154 struct hfsc_sched *q = qdisc_priv(sch);
1155 struct hfsc_class *cl = (struct hfsc_class *)arg;
1156
1157 if (cl->level > 0 || cl->filter_cnt > 0 || cl == &q->root)
1158 return -EBUSY;
1159
1160 sch_tree_lock(sch);
1161
1da177e4
LT		 1162 list_del(&cl->siblings);
1163 hfsc_adjust_levels(cl->cl_parent);
c38c83cb 1164
1da177e4 1165 hfsc_purge_queue(sch, cl);
c38c83cb
PM		 1166 list_del(&cl->hlist);
1167
1da177e4
LT		 1168 if (--cl->refcnt == 0)
1169 hfsc_destroy_class(sch, cl);
1170
1171 sch_tree_unlock(sch);
1172 return 0;
1173}
1174
1175static struct hfsc_class *
1176hfsc_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
1177{
1178 struct hfsc_sched *q = qdisc_priv(sch);
1179 struct hfsc_class *cl;
1180 struct tcf_result res;
1181 struct tcf_proto *tcf;
1182 int result;
1183
1184 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0 &&
1185 (cl = hfsc_find_class(skb->priority, sch)) != NULL)
1186 if (cl->level == 0)
1187 return cl;
1188
29f1df6c 1189 *qerr = NET_XMIT_BYPASS;
1da177e4
LT		 1190 tcf = q->root.filter_list;
1191 while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
1192#ifdef CONFIG_NET_CLS_ACT
1193 switch (result) {
1194 case TC_ACT_QUEUED:
10297b99 1195 case TC_ACT_STOLEN:
1da177e4 1196 *qerr = NET_XMIT_SUCCESS;
10297b99 1197 case TC_ACT_SHOT:
1da177e4
LT		 1198 return NULL;
1199 }
1200#elif defined(CONFIG_NET_CLS_POLICE)
1201 if (result == TC_POLICE_SHOT)
1202 return NULL;
1203#endif
1204 if ((cl = (struct hfsc_class *)res.class) == NULL) {
1205 if ((cl = hfsc_find_class(res.classid, sch)) == NULL)
1206 break; /* filter selected invalid classid */
1207 }
1208
1209 if (cl->level == 0)
1210 return cl; /* hit leaf class */
1211
1212 /* apply inner filter chain */
1213 tcf = cl->filter_list;
1214 }
1215
1216 /* classification failed, try default class */
1217 cl = hfsc_find_class(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
1218 if (cl == NULL || cl->level > 0)
1219 return NULL;
1220
1221 return cl;
1222}
1223
1224static int
1225hfsc_graft_class(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
10297b99 1226 struct Qdisc **old)
1da177e4
LT		 1227{
1228 struct hfsc_class *cl = (struct hfsc_class *)arg;
1229
1230 if (cl == NULL)
1231 return -ENOENT;
1232 if (cl->level > 0)
1233 return -EINVAL;
1234 if (new == NULL) {
9f9afec4
PM		 1235 new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1236 cl->classid);
1da177e4
LT		 1237 if (new == NULL)
1238 new = &noop_qdisc;
1239 }
1240
1241 sch_tree_lock(sch);
1242 hfsc_purge_queue(sch, cl);
1243 *old = xchg(&cl->qdisc, new);
1244 sch_tree_unlock(sch);
1245 return 0;
1246}
1247
1248static struct Qdisc *
1249hfsc_class_leaf(struct Qdisc *sch, unsigned long arg)
1250{
1251 struct hfsc_class *cl = (struct hfsc_class *)arg;
1252
1253 if (cl != NULL && cl->level == 0)
1254 return cl->qdisc;
1255
1256 return NULL;
1257}
1258
f973b913
PM		 1259static void
1260hfsc_qlen_notify(struct Qdisc *sch, unsigned long arg)
1261{
1262 struct hfsc_class *cl = (struct hfsc_class *)arg;
1263
1264 if (cl->qdisc->q.qlen == 0) {
1265 update_vf(cl, 0, 0);
1266 set_passive(cl);
1267 }
1268}
1269
1da177e4
LT		 1270static unsigned long
1271hfsc_get_class(struct Qdisc *sch, u32 classid)
1272{
1273 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1274
1275 if (cl != NULL)
1276 cl->refcnt++;
1277
1278 return (unsigned long)cl;
1279}
1280
1281static void
1282hfsc_put_class(struct Qdisc *sch, unsigned long arg)
1283{
1284 struct hfsc_class *cl = (struct hfsc_class *)arg;
1285
1286 if (--cl->refcnt == 0)
1287 hfsc_destroy_class(sch, cl);
1288}
1289
1290static unsigned long
1291hfsc_bind_tcf(struct Qdisc *sch, unsigned long parent, u32 classid)
1292{
1293 struct hfsc_class *p = (struct hfsc_class *)parent;
1294 struct hfsc_class *cl = hfsc_find_class(classid, sch);
1295
1296 if (cl != NULL) {
1297 if (p != NULL && p->level <= cl->level)
1298 return 0;
1299 cl->filter_cnt++;
1300 }
1301
1302 return (unsigned long)cl;
1303}
1304
1305static void
1306hfsc_unbind_tcf(struct Qdisc *sch, unsigned long arg)
1307{
1308 struct hfsc_class *cl = (struct hfsc_class *)arg;
1309
1310 cl->filter_cnt--;
1311}
1312
1313static struct tcf_proto **
1314hfsc_tcf_chain(struct Qdisc *sch, unsigned long arg)
1315{
1316 struct hfsc_sched *q = qdisc_priv(sch);
1317 struct hfsc_class *cl = (struct hfsc_class *)arg;
1318
1319 if (cl == NULL)
1320 cl = &q->root;
1321
1322 return &cl->filter_list;
1323}
1324
1325static int
1326hfsc_dump_sc(struct sk_buff *skb, int attr, struct internal_sc *sc)
1327{
1328 struct tc_service_curve tsc;
1329
1330 tsc.m1 = sm2m(sc->sm1);
1331 tsc.d = dx2d(sc->dx);
1332 tsc.m2 = sm2m(sc->sm2);
1333 RTA_PUT(skb, attr, sizeof(tsc), &tsc);
1334
1335 return skb->len;
1336
1337 rtattr_failure:
1338 return -1;
1339}
1340
1341static inline int
1342hfsc_dump_curves(struct sk_buff *skb, struct hfsc_class *cl)
1343{
1344 if ((cl->cl_flags & HFSC_RSC) &&
1345 (hfsc_dump_sc(skb, TCA_HFSC_RSC, &cl->cl_rsc) < 0))
1346 goto rtattr_failure;
1347
1348 if ((cl->cl_flags & HFSC_FSC) &&
1349 (hfsc_dump_sc(skb, TCA_HFSC_FSC, &cl->cl_fsc) < 0))
1350 goto rtattr_failure;
1351
1352 if ((cl->cl_flags & HFSC_USC) &&
1353 (hfsc_dump_sc(skb, TCA_HFSC_USC, &cl->cl_usc) < 0))
1354 goto rtattr_failure;
1355
1356 return skb->len;
1357
1358 rtattr_failure:
1359 return -1;
1360}
1361
1362static int
1363hfsc_dump_class(struct Qdisc *sch, unsigned long arg, struct sk_buff *skb,
10297b99 1364 struct tcmsg *tcm)
1da177e4
LT		 1365{
1366 struct hfsc_class *cl = (struct hfsc_class *)arg;
27a884dc 1367 unsigned char *b = skb_tail_pointer(skb);
1da177e4
LT		 1368 struct rtattr *rta = (struct rtattr *)b;
1369
1370 tcm->tcm_parent = cl->cl_parent ? cl->cl_parent->classid : TC_H_ROOT;
1371 tcm->tcm_handle = cl->classid;
1372 if (cl->level == 0)
1373 tcm->tcm_info = cl->qdisc->handle;
1374
1375 RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1376 if (hfsc_dump_curves(skb, cl) < 0)
1377 goto rtattr_failure;
27a884dc 1378 rta->rta_len = skb_tail_pointer(skb) - b;
1da177e4
LT		 1379 return skb->len;
1380
1381 rtattr_failure:
dc5fc579 1382 nlmsg_trim(skb, b);
1da177e4
LT		 1383 return -1;
1384}
1385
1386static int
1387hfsc_dump_class_stats(struct Qdisc *sch, unsigned long arg,
1388 struct gnet_dump *d)
1389{
1390 struct hfsc_class *cl = (struct hfsc_class *)arg;
1391 struct tc_hfsc_stats xstats;
1392
1393 cl->qstats.qlen = cl->qdisc->q.qlen;
1394 xstats.level = cl->level;
1395 xstats.period = cl->cl_vtperiod;
1396 xstats.work = cl->cl_total;
1397 xstats.rtwork = cl->cl_cumul;
1398
1399 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
1400#ifdef CONFIG_NET_ESTIMATOR
1401 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
1402#endif
1403 gnet_stats_copy_queue(d, &cl->qstats) < 0)
1404 return -1;
1405
1406 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
1407}
1408
1409
1410
1411static void
1412hfsc_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1413{
1414 struct hfsc_sched *q = qdisc_priv(sch);
1415 struct hfsc_class *cl;
1416 unsigned int i;
1417
1418 if (arg->stop)
1419 return;
1420
1421 for (i = 0; i < HFSC_HSIZE; i++) {
1422 list_for_each_entry(cl, &q->clhash[i], hlist) {
1423 if (arg->count < arg->skip) {
1424 arg->count++;
1425 continue;
1426 }
1427 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1428 arg->stop = 1;
1429 return;
1430 }
1431 arg->count++;
1432 }
1433 }
1434}
1435
1436static void
ed2b229a 1437hfsc_schedule_watchdog(struct Qdisc *sch)
1da177e4
LT		 1438{
1439 struct hfsc_sched *q = qdisc_priv(sch);
1440 struct hfsc_class *cl;
1441 u64 next_time = 0;
1da177e4
LT		 1442
1443 if ((cl = eltree_get_minel(q)) != NULL)
1444 next_time = cl->cl_e;
1445 if (q->root.cl_cfmin != 0) {
1446 if (next_time == 0 || next_time > q->root.cl_cfmin)
1447 next_time = q->root.cl_cfmin;
1448 }
3d50f231 1449 WARN_ON(next_time == 0);
ed2b229a 1450 qdisc_watchdog_schedule(&q->watchdog, next_time);
1da177e4
LT		 1451}
1452
1453static int
1454hfsc_init_qdisc(struct Qdisc *sch, struct rtattr *opt)
1455{
1456 struct hfsc_sched *q = qdisc_priv(sch);
1457 struct tc_hfsc_qopt *qopt;
1458 unsigned int i;
1459
1460 if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1461 return -EINVAL;
1462 qopt = RTA_DATA(opt);
1463
1464 sch->stats_lock = &sch->dev->queue_lock;
1465
1466 q->defcls = qopt->defcls;
1467 for (i = 0; i < HFSC_HSIZE; i++)
1468 INIT_LIST_HEAD(&q->clhash[i]);
1469 q->eligible = RB_ROOT;
1470 INIT_LIST_HEAD(&q->droplist);
1471 skb_queue_head_init(&q->requeue);
1472
1473 q->root.refcnt = 1;
1474 q->root.classid = sch->handle;
1475 q->root.sched = q;
9f9afec4
PM		 1476 q->root.qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
1477 sch->handle);
1da177e4
LT		 1478 if (q->root.qdisc == NULL)
1479 q->root.qdisc = &noop_qdisc;
1480 q->root.stats_lock = &sch->dev->queue_lock;
1481 INIT_LIST_HEAD(&q->root.children);
1482 q->root.vt_tree = RB_ROOT;
1483 q->root.cf_tree = RB_ROOT;
1484
1485 list_add(&q->root.hlist, &q->clhash[hfsc_hash(q->root.classid)]);
1486
ed2b229a 1487 qdisc_watchdog_init(&q->watchdog, sch);
1da177e4
LT		 1488
1489 return 0;
1490}
1491
1492static int
1493hfsc_change_qdisc(struct Qdisc *sch, struct rtattr *opt)
1494{
1495 struct hfsc_sched *q = qdisc_priv(sch);
1496 struct tc_hfsc_qopt *qopt;
1497
1498 if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1499 return -EINVAL;
1500 qopt = RTA_DATA(opt);
1501
1502 sch_tree_lock(sch);
1503 q->defcls = qopt->defcls;
1504 sch_tree_unlock(sch);
1505
1506 return 0;
1507}
1508
1509static void
1510hfsc_reset_class(struct hfsc_class *cl)
1511{
1512 cl->cl_total = 0;
1513 cl->cl_cumul = 0;
1514 cl->cl_d = 0;
1515 cl->cl_e = 0;
1516 cl->cl_vt = 0;
1517 cl->cl_vtadj = 0;
1518 cl->cl_vtoff = 0;
1519 cl->cl_cvtmin = 0;
1520 cl->cl_cvtmax = 0;
1521 cl->cl_cvtoff = 0;
1522 cl->cl_pcvtoff = 0;
1523 cl->cl_vtperiod = 0;
1524 cl->cl_parentperiod = 0;
1525 cl->cl_f = 0;
1526 cl->cl_myf = 0;
1527 cl->cl_myfadj = 0;
1528 cl->cl_cfmin = 0;
1529 cl->cl_nactive = 0;
1530
1531 cl->vt_tree = RB_ROOT;
1532 cl->cf_tree = RB_ROOT;
1533 qdisc_reset(cl->qdisc);
1534
1535 if (cl->cl_flags & HFSC_RSC)
1536 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, 0, 0);
1537 if (cl->cl_flags & HFSC_FSC)
1538 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, 0, 0);
1539 if (cl->cl_flags & HFSC_USC)
1540 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, 0, 0);
1541}
1542
1543static void
1544hfsc_reset_qdisc(struct Qdisc *sch)
1545{
1546 struct hfsc_sched *q = qdisc_priv(sch);
1547 struct hfsc_class *cl;
1548 unsigned int i;
1549
1550 for (i = 0; i < HFSC_HSIZE; i++) {
1551 list_for_each_entry(cl, &q->clhash[i], hlist)
1552 hfsc_reset_class(cl);
1553 }
1554 __skb_queue_purge(&q->requeue);
1555 q->eligible = RB_ROOT;
1556 INIT_LIST_HEAD(&q->droplist);
ed2b229a 1557 qdisc_watchdog_cancel(&q->watchdog);
1da177e4
LT		 1558 sch->q.qlen = 0;
1559}
1560
1561static void
1562hfsc_destroy_qdisc(struct Qdisc *sch)
1563{
1564 struct hfsc_sched *q = qdisc_priv(sch);
1565 struct hfsc_class *cl, *next;
1566 unsigned int i;
1567
1568 for (i = 0; i < HFSC_HSIZE; i++) {
1569 list_for_each_entry_safe(cl, next, &q->clhash[i], hlist)
1570 hfsc_destroy_class(sch, cl);
1571 }
1572 __skb_queue_purge(&q->requeue);
ed2b229a 1573 qdisc_watchdog_cancel(&q->watchdog);
1da177e4
LT		 1574}
1575
1576static int
1577hfsc_dump_qdisc(struct Qdisc *sch, struct sk_buff *skb)
1578{
1579 struct hfsc_sched *q = qdisc_priv(sch);
27a884dc 1580 unsigned char *b = skb_tail_pointer(skb);
1da177e4
LT		 1581 struct tc_hfsc_qopt qopt;
1582
1583 qopt.defcls = q->defcls;
1584 RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
1585 return skb->len;
1586
1587 rtattr_failure:
dc5fc579 1588 nlmsg_trim(skb, b);
1da177e4
LT		 1589 return -1;
1590}
1591
1592static int
1593hfsc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1594{
1595 struct hfsc_class *cl;
1596 unsigned int len;
1597 int err;
1598
1599 cl = hfsc_classify(skb, sch, &err);
1600 if (cl == NULL) {
29f1df6c 1601 if (err == NET_XMIT_BYPASS)
1da177e4
LT		 1602 sch->qstats.drops++;
1603 kfree_skb(skb);
1604 return err;
1605 }
1606
1607 len = skb->len;
1608 err = cl->qdisc->enqueue(skb, cl->qdisc);
1609 if (unlikely(err != NET_XMIT_SUCCESS)) {
1610 cl->qstats.drops++;
1611 sch->qstats.drops++;
1612 return err;
1613 }
1614
1615 if (cl->qdisc->q.qlen == 1)
1616 set_active(cl, len);
1617
1618 cl->bstats.packets++;
1619 cl->bstats.bytes += len;
1620 sch->bstats.packets++;
1621 sch->bstats.bytes += len;
1622 sch->q.qlen++;
1623
1624 return NET_XMIT_SUCCESS;
1625}
1626
1627static struct sk_buff *
1628hfsc_dequeue(struct Qdisc *sch)
1629{
1630 struct hfsc_sched *q = qdisc_priv(sch);
1631 struct hfsc_class *cl;
1632 struct sk_buff *skb;
1633 u64 cur_time;
1634 unsigned int next_len;
1635 int realtime = 0;
1636
1637 if (sch->q.qlen == 0)
1638 return NULL;
1639 if ((skb = __skb_dequeue(&q->requeue)))
1640 goto out;
1641
1642 PSCHED_GET_TIME(cur_time);
1643
1644 /*
1645 * if there are eligible classes, use real-time criteria.
1646 * find the class with the minimum deadline among
1647 * the eligible classes.
1648 */
1649 if ((cl = eltree_get_mindl(q, cur_time)) != NULL) {
1650 realtime = 1;
1651 } else {
1652 /*
1653 * use link-sharing criteria
1654 * get the class with the minimum vt in the hierarchy
1655 */
1656 cl = vttree_get_minvt(&q->root, cur_time);
1657 if (cl == NULL) {
1658 sch->qstats.overlimits++;
ed2b229a 1659 hfsc_schedule_watchdog(sch);
1da177e4
LT		 1660 return NULL;
1661 }
1662 }
1663
1664 skb = cl->qdisc->dequeue(cl->qdisc);
1665 if (skb == NULL) {
1666 if (net_ratelimit())
1667 printk("HFSC: Non-work-conserving qdisc ?\n");
1668 return NULL;
1669 }
1670
1671 update_vf(cl, skb->len, cur_time);
1672 if (realtime)
1673 cl->cl_cumul += skb->len;
1674
1675 if (cl->qdisc->q.qlen != 0) {
1676 if (cl->cl_flags & HFSC_RSC) {
1677 /* update ed */
1678 next_len = qdisc_peek_len(cl->qdisc);
1679 if (realtime)
1680 update_ed(cl, next_len);
1681 else
1682 update_d(cl, next_len);
1683 }
1684 } else {
1685 /* the class becomes passive */
1686 set_passive(cl);
1687 }
1688
1689 out:
1690 sch->flags &= ~TCQ_F_THROTTLED;
1691 sch->q.qlen--;
1692
1693 return skb;
1694}
1695
1696static int
1697hfsc_requeue(struct sk_buff *skb, struct Qdisc *sch)
1698{
1699 struct hfsc_sched *q = qdisc_priv(sch);
1700
1701 __skb_queue_head(&q->requeue, skb);
1702 sch->q.qlen++;
1703 sch->qstats.requeues++;
1704 return NET_XMIT_SUCCESS;
1705}
1706
1707static unsigned int
1708hfsc_drop(struct Qdisc *sch)
1709{
1710 struct hfsc_sched *q = qdisc_priv(sch);
1711 struct hfsc_class *cl;
1712 unsigned int len;
1713
1714 list_for_each_entry(cl, &q->droplist, dlist) {
1715 if (cl->qdisc->ops->drop != NULL &&
1716 (len = cl->qdisc->ops->drop(cl->qdisc)) > 0) {
1717 if (cl->qdisc->q.qlen == 0) {
1718 update_vf(cl, 0, 0);
1719 set_passive(cl);
1720 } else {
1721 list_move_tail(&cl->dlist, &q->droplist);
1722 }
1723 cl->qstats.drops++;
1724 sch->qstats.drops++;
1725 sch->q.qlen--;
1726 return len;
1727 }
1728 }
1729 return 0;
1730}
1731
1732static struct Qdisc_class_ops hfsc_class_ops = {
1733 .change = hfsc_change_class,
1734 .delete = hfsc_delete_class,
1735 .graft = hfsc_graft_class,
1736 .leaf = hfsc_class_leaf,
f973b913 1737 .qlen_notify = hfsc_qlen_notify,
1da177e4
LT		 1738 .get = hfsc_get_class,
1739 .put = hfsc_put_class,
1740 .bind_tcf = hfsc_bind_tcf,
1741 .unbind_tcf = hfsc_unbind_tcf,
1742 .tcf_chain = hfsc_tcf_chain,
1743 .dump = hfsc_dump_class,
1744 .dump_stats = hfsc_dump_class_stats,
1745 .walk = hfsc_walk
1746};
1747
1748static struct Qdisc_ops hfsc_qdisc_ops = {
1749 .id = "hfsc",
1750 .init = hfsc_init_qdisc,
1751 .change = hfsc_change_qdisc,
1752 .reset = hfsc_reset_qdisc,
1753 .destroy = hfsc_destroy_qdisc,
1754 .dump = hfsc_dump_qdisc,
1755 .enqueue = hfsc_enqueue,
1756 .dequeue = hfsc_dequeue,
1757 .requeue = hfsc_requeue,
1758 .drop = hfsc_drop,
1759 .cl_ops = &hfsc_class_ops,
1760 .priv_size = sizeof(struct hfsc_sched),
1761 .owner = THIS_MODULE
1762};
1763
1764static int __init
1765hfsc_init(void)
1766{
1767 return register_qdisc(&hfsc_qdisc_ops);
1768}
1769
1770static void __exit
1771hfsc_cleanup(void)
1772{
1773 unregister_qdisc(&hfsc_qdisc_ops);
1774}
1775
1776MODULE_LICENSE("GPL");
1777module_init(hfsc_init);
1778module_exit(hfsc_cleanup);
kernel source for telekom puls (alcatel/mediatek)