net/sched/sch_hfsc.c

   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>
  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/init.h>
  66 #include <linux/netdevice.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/pkt_sched.h>
  69 #include <net/pkt_sched.h>
  70 #include <net/pkt_cls.h>
  71 #include <asm/system.h>
  72 #include <asm/div64.h>
  73
  74 #define HFSC_DEBUG 1
  75
  76 /*
  77  * kernel internal service curve representation:
  78  *   coordinates are given by 64 bit unsigned integers.
  79  *   x-axis: unit is clock count.
  80  *   y-axis: unit is byte.
  81  *
  82  *   The service curve parameters are converted to the internal
  83  *   representation. The slope values are scaled to avoid overflow.
  84  *   the inverse slope values as well as the y-projection of the 1st
  85  *   segment are kept in order to to avoid 64-bit divide operations
  86  *   that are expensive on 32-bit architectures.
  87  */
  88
  89 struct internal_sc
  90 {
  91         u64     sm1;    /* scaled slope of the 1st segment */
  92         u64     ism1;   /* scaled inverse-slope of the 1st segment */
  93         u64     dx;     /* the x-projection of the 1st segment */
  94         u64     dy;     /* the y-projection of the 1st segment */
  95         u64     sm2;    /* scaled slope of the 2nd segment */
  96         u64     ism2;   /* scaled inverse-slope of the 2nd segment */
  97 };
  98
  99 /* runtime service curve */
 100 struct runtime_sc
 101 {
 102         u64     x;      /* current starting position on x-axis */
 103         u64     y;      /* current starting position on y-axis */
 104         u64     sm1;    /* scaled slope of the 1st segment */
 105         u64     ism1;   /* scaled inverse-slope of the 1st segment */
 106         u64     dx;     /* the x-projection of the 1st segment */
 107         u64     dy;     /* the y-projection of the 1st segment */
 108         u64     sm2;    /* scaled slope of the 2nd segment */
 109         u64     ism2;   /* scaled inverse-slope of the 2nd segment */
 110 };
 111
 112 enum hfsc_class_flags
 113 {
 114         HFSC_RSC = 0x1,
 115         HFSC_FSC = 0x2,
 116         HFSC_USC = 0x4
 117 };
 118
 119 struct hfsc_class
 120 {
 121         u32             classid;        /* class id */
 122         unsigned int    refcnt;         /* usage count */
 123
 124         struct tc_stats stats;          /* generic statistics */
 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 list_head actlist;       /* active children list */
 137         struct list_head alist;         /* active children list member */
 138         struct list_head ellist;        /* eligible list member */
 139         struct list_head hlist;         /* hash list member */
 140         struct list_head dlist;         /* drop list member */
 141
 142         u64     cl_total;               /* total work in bytes */
 143         u64     cl_cumul;               /* cumulative work in bytes done by
 144                                            real-time criteria */
 145
 146         u64     cl_d;                   /* deadline*/
 147         u64     cl_e;                   /* eligible time */
 148         u64     cl_vt;                  /* virtual time */
 149         u64     cl_f;                   /* time when this class will fit for
 150                                            link-sharing, max(myf, cfmin) */
 151         u64     cl_myf;                 /* my fit-time (calculated from this
 152                                            class's own upperlimit curve) */
 153         u64     cl_myfadj;              /* my fit-time adjustment (to cancel
 154                                            history dependence) */
 155         u64     cl_cfmin;               /* earliest children's fit-time (used
 156                                            with cl_myf to obtain cl_f) */
 157         u64     cl_cvtmin;              /* minimal virtual time among the
 158                                            children fit for link-sharing
 159                                            (monotonic within a period) */
 160         u64     cl_vtadj;               /* intra-period cumulative vt
 161                                            adjustment */
 162         u64     cl_vtoff;               /* inter-period cumulative vt offset */
 163         u64     cl_cvtmax;              /* max child's vt in the last period */
 164
 165         struct internal_sc cl_rsc;      /* internal real-time service curve */
 166         struct internal_sc cl_fsc;      /* internal fair service curve */
 167         struct internal_sc cl_usc;      /* internal upperlimit service curve */
 168         struct runtime_sc cl_deadline;  /* deadline curve */
 169         struct runtime_sc cl_eligible;  /* eligible curve */
 170         struct runtime_sc cl_virtual;   /* virtual curve */
 171         struct runtime_sc cl_ulimit;    /* upperlimit curve */
 172
 173         unsigned long   cl_flags;       /* which curves are valid */
 174         unsigned long   cl_vtperiod;    /* vt period sequence number */
 175         unsigned long   cl_parentperiod;/* parent's vt period sequence number*/
 176         unsigned long   cl_nactive;     /* number of active children */
 177 };
 178
 179 #define HFSC_HSIZE      16
 180
 181 struct hfsc_sched
 182 {
 183         u16     defcls;                         /* default class id */
 184         struct hfsc_class root;                 /* root class */
 185         struct list_head clhash[HFSC_HSIZE];    /* class hash */
 186         struct list_head eligible;              /* eligible list */
 187         struct list_head droplist;              /* active leaf class list (for
 188                                                    dropping) */
 189         struct sk_buff_head requeue;            /* requeued packet */
 190         struct timer_list wd_timer;             /* watchdog timer */
 191 };
 192
 193 /*
 194  * macros
 195  */
 196 #ifdef CONFIG_NET_SCH_CLK_GETTIMEOFDAY
 197 #include <linux/time.h>
 198 #undef PSCHED_GET_TIME
 199 #define PSCHED_GET_TIME(stamp)                                          \
 200 do {                                                                    \
 201         struct timeval tv;                                              \
 202         do_gettimeofday(&tv);                                           \
 203         (stamp) = 1000000ULL * tv.tv_sec + tv.tv_usec;                  \
 204 } while (0)
 205 #endif
 206
 207 #if HFSC_DEBUG
 208 #define ASSERT(cond)                                                    \
 209 do {                                                                    \
 210         if (unlikely(!(cond)))                                          \
 211                 printk("assertion %s failed at %s:%i (%s)\n",           \
 212                        #cond, __FILE__, __LINE__, __FUNCTION__);        \
 213 } while (0)
 214 #else
 215 #define ASSERT(cond)
 216 #endif /* HFSC_DEBUG */
 217
 218 #define HT_INFINITY     0xffffffffffffffffULL   /* infinite time value */
 219
 220
 221 /*
 222  * eligible list holds backlogged classes being sorted by their eligible times.
 223  * there is one eligible list per hfsc instance.
 224  */
 225
 226 static void
 227 ellist_insert(struct hfsc_class *cl)
 228 {
 229         struct list_head *head = &cl->sched->eligible;
 230         struct hfsc_class *p;
 231
 232         /* check the last entry first */
 233         if (list_empty(head) ||
 234             ((p = list_entry(head->prev, struct hfsc_class, ellist)) &&
 235              p->cl_e <= cl->cl_e)) {
 236                 list_add_tail(&cl->ellist, head);
 237                 return;
 238         }
 239
 240         list_for_each_entry(p, head, ellist) {
 241                 if (cl->cl_e < p->cl_e) {
 242                         /* insert cl before p */
 243                         list_add_tail(&cl->ellist, &p->ellist);
 244                         return;
 245                 }
 246         }
 247         ASSERT(0); /* should not reach here */
 248 }
 249
 250 static inline void
 251 ellist_remove(struct hfsc_class *cl)
 252 {
 253         list_del(&cl->ellist);
 254 }
 255
 256 static void
 257 ellist_update(struct hfsc_class *cl)
 258 {
 259         struct list_head *head = &cl->sched->eligible;
 260         struct hfsc_class *p, *last;
 261
 262         /*
 263          * the eligible time of a class increases monotonically.
 264          * if the next entry has a larger eligible time, nothing to do.
 265          */
 266         if (cl->ellist.next == head ||
 267             ((p = list_entry(cl->ellist.next, struct hfsc_class, ellist)) &&
 268              cl->cl_e <= p->cl_e))
 269                 return;
 270
 271         /* check the last entry */
 272         last = list_entry(head->prev, struct hfsc_class, ellist);
 273         if (last->cl_e <= cl->cl_e) {
 274                 list_move_tail(&cl->ellist, head);
 275                 return;
 276         }
 277
 278         /*
 279          * the new position must be between the next entry
 280          * and the last entry
 281          */
 282         list_for_each_entry_continue(p, head, ellist) {
 283                 if (cl->cl_e < p->cl_e) {
 284                         list_move_tail(&cl->ellist, &p->ellist);
 285                         return;
 286                 }
 287         }
 288         ASSERT(0); /* should not reach here */
 289 }
 290
 291 /* find the class with the minimum deadline among the eligible classes */
 292 static inline struct hfsc_class *
 293 ellist_get_mindl(struct list_head *head, u64 cur_time)
 294 {
 295         struct hfsc_class *p, *cl = NULL;
 296
 297         list_for_each_entry(p, head, ellist) {
 298                 if (p->cl_e > cur_time)
 299                         break;
 300                 if (cl == NULL || p->cl_d < cl->cl_d)
 301                         cl = p;
 302         }
 303         return cl;
 304 }
 305
 306 /* find the class with minimum eligible time among the eligible classes */
 307 static inline struct hfsc_class *
 308 ellist_get_minel(struct list_head *head)
 309 {
 310         if (list_empty(head))
 311                 return NULL;
 312         return list_entry(head->next, struct hfsc_class, ellist);
 313 }
 314
 315 /*
 316  * active children list holds backlogged child classes being sorted
 317  * by their virtual time. each intermediate class has one active
 318  * children list.
 319  */
 320 static void
 321 actlist_insert(struct hfsc_class *cl)
 322 {
 323         struct list_head *head = &cl->cl_parent->actlist;
 324         struct hfsc_class *p;
 325
 326         /* check the last entry first */
 327         if (list_empty(head) ||
 328             ((p = list_entry(head->prev, struct hfsc_class, alist)) &&
 329              p->cl_vt <= cl->cl_vt)) {
 330                 list_add_tail(&cl->alist, head);
 331                 return;
 332         }
 333
 334         list_for_each_entry(p, head, alist) {
 335                 if (cl->cl_vt < p->cl_vt) {
 336                         /* insert cl before p */
 337                         list_add_tail(&cl->alist, &p->alist);
 338                         return;
 339                 }
 340         }
 341         ASSERT(0); /* should not reach here */
 342 }
 343
 344 static inline void
 345 actlist_remove(struct hfsc_class *cl)
 346 {
 347         list_del(&cl->alist);
 348 }
 349
 350 static void
 351 actlist_update(struct hfsc_class *cl)
 352 {
 353         struct list_head *head = &cl->cl_parent->actlist;
 354         struct hfsc_class *p, *last;
 355
 356         /*
 357          * the virtual time of a class increases monotonically.
 358          * if the next entry has a larger virtual time, nothing to do.
 359          */
 360         if (cl->alist.next == head ||
 361             ((p = list_entry(cl->alist.next, struct hfsc_class, alist)) &&
 362              cl->cl_vt <= p->cl_vt))
 363                 return;
 364
 365         /* check the last entry */
 366         last = list_entry(head->prev, struct hfsc_class, alist);
 367         if (last->cl_vt <= cl->cl_vt) {
 368                 list_move_tail(&cl->alist, head);
 369                 return;
 370         }
 371
 372         /*
 373          * the new position must be between the next entry
 374          * and the last entry
 375          */
 376         list_for_each_entry_continue(p, head, alist) {
 377                 if (cl->cl_vt < p->cl_vt) {
 378                         list_move_tail(&cl->alist, &p->alist);
 379                         return;
 380                 }
 381         }
 382         ASSERT(0); /* should not reach here */
 383 }
 384
 385 static inline struct hfsc_class *
 386 actlist_firstfit(struct hfsc_class *cl, u64 cur_time)
 387 {
 388         struct hfsc_class *p;
 389
 390         list_for_each_entry(p, &cl->actlist, alist) {
 391                 if (p->cl_f <= cur_time) {
 392                         return p;
 393                 }
 394         }
 395         return NULL;
 396 }
 397
 398 /*
 399  * get the leaf class with the minimum vt in the hierarchy
 400  */
 401 static struct hfsc_class *
 402 actlist_get_minvt(struct hfsc_class *cl, u64 cur_time)
 403 {
 404         /* if root-class's cfmin is bigger than cur_time nothing to do */
 405         if (cl->cl_cfmin > cur_time)
 406                 return NULL;
 407
 408         while (cl->level > 0) {
 409                 cl = actlist_firstfit(cl, cur_time);
 410                 if (cl == NULL)
 411                         return NULL;
 412                 /*
 413                  * update parent's cl_cvtmin.
 414                  */
 415                 if (cl->cl_parent->cl_cvtmin < cl->cl_vt)
 416                         cl->cl_parent->cl_cvtmin = cl->cl_vt;
 417         }
 418         return cl;
 419 }
 420
 421 /*
 422  * service curve support functions
 423  *
 424  *  external service curve parameters
 425  *      m: bps
 426  *      d: us
 427  *  internal service curve parameters
 428  *      sm: (bytes/psched_us) << SM_SHIFT
 429  *      ism: (psched_us/byte) << ISM_SHIFT
 430  *      dx: psched_us
 431  *
 432  * Clock source resolution (CONFIG_NET_SCH_CLK_*)
 433  *  JIFFIES: for 48<=HZ<=1534 resolution is between 0.63us and 1.27us.
 434  *  CPU: resolution is between 0.5us and 1us.
 435  *  GETTIMEOFDAY: resolution is exactly 1us.
 436  *
 437  * sm and ism are scaled in order to keep effective digits.
 438  * SM_SHIFT and ISM_SHIFT are selected to keep at least 4 effective
 439  * digits in decimal using the following table.
 440  *
 441  * Note: We can afford the additional accuracy (altq hfsc keeps at most
 442  * 3 effective digits) thanks to the fact that linux clock is bounded
 443  * much more tightly.
 444  *
 445  *  bits/sec      100Kbps     1Mbps     10Mbps     100Mbps    1Gbps
 446  *  ------------+-------------------------------------------------------
 447  *  bytes/0.5us   6.25e-3    62.5e-3    625e-3     6250e-e    62500e-3
 448  *  bytes/us      12.5e-3    125e-3     1250e-3    12500e-3   125000e-3
 449  *  bytes/1.27us  15.875e-3  158.75e-3  1587.5e-3  15875e-3   158750e-3
 450  *
 451  *  0.5us/byte    160        16         1.6        0.16       0.016
 452  *  us/byte       80         8          0.8        0.08       0.008
 453  *  1.27us/byte   63         6.3        0.63       0.063      0.0063
 454  */
 455 #define SM_SHIFT        20
 456 #define ISM_SHIFT       18
 457
 458 #define SM_MASK         ((1ULL << SM_SHIFT) - 1)
 459 #define ISM_MASK        ((1ULL << ISM_SHIFT) - 1)
 460
 461 static inline u64
 462 seg_x2y(u64 x, u64 sm)
 463 {
 464         u64 y;
 465
 466         /*
 467          * compute
 468          *      y = x * sm >> SM_SHIFT
 469          * but divide it for the upper and lower bits to avoid overflow
 470          */
 471         y = (x >> SM_SHIFT) * sm + (((x & SM_MASK) * sm) >> SM_SHIFT);
 472         return y;
 473 }
 474
 475 static inline u64
 476 seg_y2x(u64 y, u64 ism)
 477 {
 478         u64 x;
 479
 480         if (y == 0)
 481                 x = 0;
 482         else if (ism == HT_INFINITY)
 483                 x = HT_INFINITY;
 484         else {
 485                 x = (y >> ISM_SHIFT) * ism
 486                     + (((y & ISM_MASK) * ism) >> ISM_SHIFT);
 487         }
 488         return x;
 489 }
 490
 491 /* Convert m (bps) into sm (bytes/psched us) */
 492 static u64
 493 m2sm(u32 m)
 494 {
 495         u64 sm;
 496
 497         sm = ((u64)m << SM_SHIFT);
 498         sm += PSCHED_JIFFIE2US(HZ) - 1;
 499         do_div(sm, PSCHED_JIFFIE2US(HZ));
 500         return sm;
 501 }
 502
 503 /* convert m (bps) into ism (psched us/byte) */
 504 static u64
 505 m2ism(u32 m)
 506 {
 507         u64 ism;
 508
 509         if (m == 0)
 510                 ism = HT_INFINITY;
 511         else {
 512                 ism = ((u64)PSCHED_JIFFIE2US(HZ) << ISM_SHIFT);
 513                 ism += m - 1;
 514                 do_div(ism, m);
 515         }
 516         return ism;
 517 }
 518
 519 /* convert d (us) into dx (psched us) */
 520 static u64
 521 d2dx(u32 d)
 522 {
 523         u64 dx;
 524
 525         dx = ((u64)d * PSCHED_JIFFIE2US(HZ));
 526         dx += 1000000 - 1;
 527         do_div(dx, 1000000);
 528         return dx;
 529 }
 530
 531 /* convert sm (bytes/psched us) into m (bps) */
 532 static u32
 533 sm2m(u64 sm)
 534 {
 535         u64 m;
 536
 537         m = (sm * PSCHED_JIFFIE2US(HZ)) >> SM_SHIFT;
 538         return (u32)m;
 539 }
 540
 541 /* convert dx (psched us) into d (us) */
 542 static u32
 543 dx2d(u64 dx)
 544 {
 545         u64 d;
 546
 547         d = dx * 1000000;
 548         do_div(d, PSCHED_JIFFIE2US(HZ));
 549         return (u32)d;
 550 }
 551
 552 static void
 553 sc2isc(struct tc_service_curve *sc, struct internal_sc *isc)
 554 {
 555         isc->sm1  = m2sm(sc->m1);
 556         isc->ism1 = m2ism(sc->m1);
 557         isc->dx   = d2dx(sc->d);
 558         isc->dy   = seg_x2y(isc->dx, isc->sm1);
 559         isc->sm2  = m2sm(sc->m2);
 560         isc->ism2 = m2ism(sc->m2);
 561 }
 562
 563 /*
 564  * initialize the runtime service curve with the given internal
 565  * service curve starting at (x, y).
 566  */
 567 static void
 568 rtsc_init(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
 569 {
 570         rtsc->x    = x;
 571         rtsc->y    = y;
 572         rtsc->sm1  = isc->sm1;
 573         rtsc->ism1 = isc->ism1;
 574         rtsc->dx   = isc->dx;
 575         rtsc->dy   = isc->dy;
 576         rtsc->sm2  = isc->sm2;
 577         rtsc->ism2 = isc->ism2;
 578 }
 579
 580 /*
 581  * calculate the y-projection of the runtime service curve by the
 582  * given x-projection value
 583  */
 584 static u64
 585 rtsc_y2x(struct runtime_sc *rtsc, u64 y)
 586 {
 587         u64 x;
 588
 589         if (y < rtsc->y)
 590                 x = rtsc->x;
 591         else if (y <= rtsc->y + rtsc->dy) {
 592                 /* x belongs to the 1st segment */
 593                 if (rtsc->dy == 0)
 594                         x = rtsc->x + rtsc->dx;
 595                 else
 596                         x = rtsc->x + seg_y2x(y - rtsc->y, rtsc->ism1);
 597         } else {
 598                 /* x belongs to the 2nd segment */
 599                 x = rtsc->x + rtsc->dx
 600                     + seg_y2x(y - rtsc->y - rtsc->dy, rtsc->ism2);
 601         }
 602         return x;
 603 }
 604
 605 static u64
 606 rtsc_x2y(struct runtime_sc *rtsc, u64 x)
 607 {
 608         u64 y;
 609
 610         if (x <= rtsc->x)
 611                 y = rtsc->y;
 612         else if (x <= rtsc->x + rtsc->dx)
 613                 /* y belongs to the 1st segment */
 614                 y = rtsc->y + seg_x2y(x - rtsc->x, rtsc->sm1);
 615         else
 616                 /* y belongs to the 2nd segment */
 617                 y = rtsc->y + rtsc->dy
 618                     + seg_x2y(x - rtsc->x - rtsc->dx, rtsc->sm2);
 619         return y;
 620 }
 621
 622 /*
 623  * update the runtime service curve by taking the minimum of the current
 624  * runtime service curve and the service curve starting at (x, y).
 625  */
 626 static void
 627 rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u64 x, u64 y)
 628 {
 629         u64 y1, y2, dx, dy;
 630         u32 dsm;
 631
 632         if (isc->sm1 <= isc->sm2) {
 633                 /* service curve is convex */
 634                 y1 = rtsc_x2y(rtsc, x);
 635                 if (y1 < y)
 636                         /* the current rtsc is smaller */
 637                         return;
 638                 rtsc->x = x;
 639                 rtsc->y = y;
 640                 return;
 641         }
 642
 643         /*
 644          * service curve is concave
 645          * compute the two y values of the current rtsc
 646          *      y1: at x
 647          *      y2: at (x + dx)
 648          */
 649         y1 = rtsc_x2y(rtsc, x);
 650         if (y1 <= y) {
 651                 /* rtsc is below isc, no change to rtsc */
 652                 return;
 653         }
 654
 655         y2 = rtsc_x2y(rtsc, x + isc->dx);
 656         if (y2 >= y + isc->dy) {
 657                 /* rtsc is above isc, replace rtsc by isc */
 658                 rtsc->x = x;
 659                 rtsc->y = y;
 660                 rtsc->dx = isc->dx;
 661                 rtsc->dy = isc->dy;
 662                 return;
 663         }
 664
 665         /*
 666          * the two curves intersect
 667          * compute the offsets (dx, dy) using the reverse
 668          * function of seg_x2y()
 669          *      seg_x2y(dx, sm1) == seg_x2y(dx, sm2) + (y1 - y)
 670          */
 671         dx = (y1 - y) << SM_SHIFT;
 672         dsm = isc->sm1 - isc->sm2;
 673         do_div(dx, dsm);
 674         /*
 675          * check if (x, y1) belongs to the 1st segment of rtsc.
 676          * if so, add the offset.
 677          */
 678         if (rtsc->x + rtsc->dx > x)
 679                 dx += rtsc->x + rtsc->dx - x;
 680         dy = seg_x2y(dx, isc->sm1);
 681
 682         rtsc->x = x;
 683         rtsc->y = y;
 684         rtsc->dx = dx;
 685         rtsc->dy = dy;
 686         return;
 687 }
 688
 689 static void
 690 init_ed(struct hfsc_class *cl, unsigned int next_len)
 691 {
 692         u64 cur_time;
 693
 694         PSCHED_GET_TIME(cur_time);
 695
 696         /* update the deadline curve */
 697         rtsc_min(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
 698
 699         /*
 700          * update the eligible curve.
 701          * for concave, it is equal to the deadline curve.
 702          * for convex, it is a linear curve with slope m2.
 703          */
 704         cl->cl_eligible = cl->cl_deadline;
 705         if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
 706                 cl->cl_eligible.dx = 0;
 707                 cl->cl_eligible.dy = 0;
 708         }
 709
 710         /* compute e and d */
 711         cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
 712         cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
 713
 714         ellist_insert(cl);
 715 }
 716
 717 static void
 718 update_ed(struct hfsc_class *cl, unsigned int next_len)
 719 {
 720         cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul);
 721         cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
 722
 723         ellist_update(cl);
 724 }
 725
 726 static inline void
 727 update_d(struct hfsc_class *cl, unsigned int next_len)
 728 {
 729         cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len);
 730 }
 731
 732 static void
 733 update_cfmin(struct hfsc_class *cl)
 734 {
 735         struct hfsc_class *p;
 736         u64 cfmin;
 737
 738         if (list_empty(&cl->actlist)) {
 739                 cl->cl_cfmin = 0;
 740                 return;
 741         }
 742         cfmin = HT_INFINITY;
 743         list_for_each_entry(p, &cl->actlist, alist) {
 744                 if (p->cl_f == 0) {
 745                         cl->cl_cfmin = 0;
 746                         return;
 747                 }
 748                 if (p->cl_f < cfmin)
 749                         cfmin = p->cl_f;
 750         }
 751         cl->cl_cfmin = cfmin;
 752 }
 753
 754 static void
 755 init_vf(struct hfsc_class *cl, unsigned int len)
 756 {
 757         struct hfsc_class *max_cl, *p;
 758         u64 vt, f, cur_time;
 759         int go_active;
 760
 761         cur_time = 0;
 762         go_active = 1;
 763         for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
 764                 if (go_active && cl->cl_nactive++ == 0)
 765                         go_active = 1;
 766                 else
 767                         go_active = 0;
 768
 769                 if (go_active) {
 770                         if (!list_empty(&cl->cl_parent->actlist)) {
 771                                 max_cl = list_entry(cl->cl_parent->actlist.prev,
 772                                                     struct hfsc_class, alist);
 773                                 /*
 774                                  * set vt to the average of the min and max
 775                                  * classes.  if the parent's period didn't
 776                                  * change, don't decrease vt of the class.
 777                                  */
 778                                 vt = max_cl->cl_vt;
 779                                 if (cl->cl_parent->cl_cvtmin != 0)
 780                                         vt = (cl->cl_parent->cl_cvtmin + vt)/2;
 781
 782                                 if (cl->cl_parent->cl_vtperiod !=
 783                                     cl->cl_parentperiod || vt > cl->cl_vt)
 784                                         cl->cl_vt = vt;
 785                         } else {
 786                                 /*
 787                                  * first child for a new parent backlog period.
 788                                  * add parent's cvtmax to vtoff of children
 789                                  * to make a new vt (vtoff + vt) larger than
 790                                  * the vt in the last period for all children.
 791                                  */
 792                                 vt = cl->cl_parent->cl_cvtmax;
 793                                 list_for_each_entry(p, &cl->cl_parent->children,
 794                                                                        siblings)
 795                                         p->cl_vtoff += vt;
 796                                 cl->cl_vt = 0;
 797                                 cl->cl_parent->cl_cvtmax = 0;
 798                                 cl->cl_parent->cl_cvtmin = 0;
 799                         }
 800
 801                         /* update the virtual curve */
 802                         vt = cl->cl_vt + cl->cl_vtoff;
 803                         rtsc_min(&cl->cl_virtual, &cl->cl_fsc, vt,
 804                                                       cl->cl_total);
 805                         if (cl->cl_virtual.x == vt) {
 806                                 cl->cl_virtual.x -= cl->cl_vtoff;
 807                                 cl->cl_vtoff = 0;
 808                         }
 809                         cl->cl_vtadj = 0;
 810
 811                         cl->cl_vtperiod++;  /* increment vt period */
 812                         cl->cl_parentperiod = cl->cl_parent->cl_vtperiod;
 813                         if (cl->cl_parent->cl_nactive == 0)
 814                                 cl->cl_parentperiod++;
 815                         cl->cl_f = 0;
 816
 817                         actlist_insert(cl);
 818
 819                         if (cl->cl_flags & HFSC_USC) {
 820                                 /* class has upper limit curve */
 821                                 if (cur_time == 0)
 822                                         PSCHED_GET_TIME(cur_time);
 823
 824                                 /* update the ulimit curve */
 825                                 rtsc_min(&cl->cl_ulimit, &cl->cl_usc, cur_time,
 826                                          cl->cl_total);
 827                                 /* compute myf */
 828                                 cl->cl_myf = rtsc_y2x(&cl->cl_ulimit,
 829                                                       cl->cl_total);
 830                                 cl->cl_myfadj = 0;
 831                         }
 832                 }
 833
 834                 f = max(cl->cl_myf, cl->cl_cfmin);
 835                 if (f != cl->cl_f) {
 836                         cl->cl_f = f;
 837                         update_cfmin(cl->cl_parent);
 838                 }
 839         }
 840 }
 841
 842 static void
 843 update_vf(struct hfsc_class *cl, unsigned int len, u64 cur_time)
 844 {
 845         u64 f; /* , myf_bound, delta; */
 846         int go_passive = 0;
 847
 848         if (cl->qdisc->q.qlen == 0 && cl->cl_flags & HFSC_FSC)
 849                 go_passive = 1;
 850
 851         for (; cl->cl_parent != NULL; cl = cl->cl_parent) {
 852                 cl->cl_total += len;
 853
 854                 if (!(cl->cl_flags & HFSC_FSC) || cl->cl_nactive == 0)
 855                         continue;
 856
 857                 if (go_passive && --cl->cl_nactive == 0)
 858                         go_passive = 1;
 859                 else
 860                         go_passive = 0;
 861
 862                 if (go_passive) {
 863                         /* no more active child, going passive */
 864
 865                         /* update cvtmax of the parent class */
 866                         if (cl->cl_vt > cl->cl_parent->cl_cvtmax)
 867                                 cl->cl_parent->cl_cvtmax = cl->cl_vt;
 868
 869                         /* remove this class from the vt list */
 870                         actlist_remove(cl);
 871
 872                         update_cfmin(cl->cl_parent);
 873
 874                         continue;
 875                 }
 876
 877                 /*
 878                  * update vt and f
 879                  */
 880                 cl->cl_vt = rtsc_y2x(&cl->cl_virtual, cl->cl_total)
 881                             - cl->cl_vtoff + cl->cl_vtadj;
 882
 883                 /*
 884                  * if vt of the class is smaller than cvtmin,
 885                  * the class was skipped in the past due to non-fit.
 886                  * if so, we need to adjust vtadj.
 887                  */
 888                 if (cl->cl_vt < cl->cl_parent->cl_cvtmin) {
 889                         cl->cl_vtadj += cl->cl_parent->cl_cvtmin - cl->cl_vt;
 890                         cl->cl_vt = cl->cl_parent->cl_cvtmin;
 891                 }
 892
 893                 /* update the vt list */
 894                 actlist_update(cl);
 895
 896                 if (cl->cl_flags & HFSC_USC) {
 897                         cl->cl_myf = cl->cl_myfadj + rtsc_y2x(&cl->cl_ulimit,
 898                                                               cl->cl_total);
 899 #if 0
 900                         /*
 901                          * This code causes classes to stay way under their
 902                          * limit when multiple classes are used at gigabit
 903                          * speed. needs investigation. -kaber
 904                          */
 905                         /*
 906                          * if myf lags behind by more than one clock tick
 907                          * from the current time, adjust myfadj to prevent
 908                          * a rate-limited class from going greedy.
 909                          * in a steady state under rate-limiting, myf
 910                          * fluctuates within one clock tick.
 911                          */
 912                         myf_bound = cur_time - PSCHED_JIFFIE2US(1);
 913                         if (cl->cl_myf < myf_bound) {
 914                                 delta = cur_time - cl->cl_myf;
 915                                 cl->cl_myfadj += delta;
 916                                 cl->cl_myf += delta;
 917                         }
 918 #endif
 919                 }
 920
 921                 f = max(cl->cl_myf, cl->cl_cfmin);
 922                 if (f != cl->cl_f) {
 923                         cl->cl_f = f;
 924                         update_cfmin(cl->cl_parent);
 925                 }
 926         }
 927 }
 928
 929 static void
 930 set_active(struct hfsc_class *cl, unsigned int len)
 931 {
 932         if (cl->cl_flags & HFSC_RSC)
 933                 init_ed(cl, len);
 934         if (cl->cl_flags & HFSC_FSC)
 935                 init_vf(cl, len);
 936
 937         list_add_tail(&cl->dlist, &cl->sched->droplist);
 938 }
 939
 940 static void
 941 set_passive(struct hfsc_class *cl)
 942 {
 943         if (cl->cl_flags & HFSC_RSC)
 944                 ellist_remove(cl);
 945
 946         list_del(&cl->dlist);
 947
 948         /*
 949          * actlist is now handled in update_vf() so that update_vf(cl, 0, 0)
 950          * needs to be called explicitly to remove a class from actlist
 951          */
 952 }
 953
 954 /*
 955  * hack to get length of first packet in queue.
 956  */
 957 static unsigned int
 958 qdisc_peek_len(struct Qdisc *sch)
 959 {
 960         struct sk_buff *skb;
 961         unsigned int len;
 962
 963         skb = sch->dequeue(sch);
 964         if (skb == NULL) {
 965                 if (net_ratelimit())
 966                         printk("qdisc_peek_len: non work-conserving qdisc ?\n");
 967                 return 0;
 968         }
 969         len = skb->len;
 970         if (unlikely(sch->ops->requeue(skb, sch) != NET_XMIT_SUCCESS)) {
 971                 if (net_ratelimit())
 972                         printk("qdisc_peek_len: failed to requeue\n");
 973                 return 0;
 974         }
 975         return len;
 976 }
 977
 978 static void
 979 hfsc_purge_queue(struct Qdisc *sch, struct hfsc_class *cl)
 980 {
 981         unsigned int len = cl->qdisc->q.qlen;
 982
 983         qdisc_reset(cl->qdisc);
 984         if (len > 0) {
 985                 update_vf(cl, 0, 0);
 986                 set_passive(cl);
 987                 sch->q.qlen -= len;
 988         }
 989 }
 990
 991 static void
 992 hfsc_adjust_levels(struct hfsc_class *cl)
 993 {
 994         struct hfsc_class *p;
 995         unsigned int level;
 996
 997         do {
 998                 level = 0;
 999                 list_for_each_entry(p, &cl->children, siblings) {
1000                         if (p->level > level)
1001                                 level = p->level;
1002                 }
1003                 cl->level = level + 1;
1004         } while ((cl = cl->cl_parent) != NULL);
1005 }
1006
1007 static inline unsigned int
1008 hfsc_hash(u32 h)
1009 {
1010         h ^= h >> 8;
1011         h ^= h >> 4;
1012
1013         return h & (HFSC_HSIZE - 1);
1014 }
1015
1016 static inline struct hfsc_class *
1017 hfsc_find_class(u32 classid, struct Qdisc *sch)
1018 {
1019         struct hfsc_sched *q = qdisc_priv(sch);
1020         struct hfsc_class *cl;
1021
1022         list_for_each_entry(cl, &q->clhash[hfsc_hash(classid)], hlist) {
1023                 if (cl->classid == classid)
1024                         return cl;
1025         }
1026         return NULL;
1027 }
1028
1029 static void
1030 hfsc_change_rsc(struct hfsc_class *cl, struct tc_service_curve *rsc,
1031                 u64 cur_time)
1032 {
1033         sc2isc(rsc, &cl->cl_rsc);
1034         rtsc_init(&cl->cl_deadline, &cl->cl_rsc, cur_time, cl->cl_cumul);
1035         cl->cl_eligible = cl->cl_deadline;
1036         if (cl->cl_rsc.sm1 <= cl->cl_rsc.sm2) {
1037                 cl->cl_eligible.dx = 0;
1038                 cl->cl_eligible.dy = 0;
1039         }
1040         cl->cl_flags |= HFSC_RSC;
1041 }
1042
1043 static void
1044 hfsc_change_fsc(struct hfsc_class *cl, struct tc_service_curve *fsc)
1045 {
1046         sc2isc(fsc, &cl->cl_fsc);
1047         rtsc_init(&cl->cl_virtual, &cl->cl_fsc, cl->cl_vt, cl->cl_total);
1048         cl->cl_flags |= HFSC_FSC;
1049 }
1050
1051 static void
1052 hfsc_change_usc(struct hfsc_class *cl, struct tc_service_curve *usc,
1053                 u64 cur_time)
1054 {
1055         sc2isc(usc, &cl->cl_usc);
1056         rtsc_init(&cl->cl_ulimit, &cl->cl_usc, cur_time, cl->cl_total);
1057         cl->cl_flags |= HFSC_USC;
1058 }
1059
1060 static int
1061 hfsc_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
1062                   struct rtattr **tca, unsigned long *arg)
1063 {
1064         struct hfsc_sched *q = qdisc_priv(sch);
1065         struct hfsc_class *cl = (struct hfsc_class *)*arg;
1066         struct hfsc_class *parent = NULL;
1067         struct rtattr *opt = tca[TCA_OPTIONS-1];
1068         struct rtattr *tb[TCA_HFSC_MAX];
1069         struct tc_service_curve *rsc = NULL, *fsc = NULL, *usc = NULL;
1070         u64 cur_time;
1071
1072         if (opt == NULL ||
1073             rtattr_parse(tb, TCA_HFSC_MAX, RTA_DATA(opt), RTA_PAYLOAD(opt)))
1074                 return -EINVAL;
1075
1076         if (tb[TCA_HFSC_RSC-1]) {
1077                 if (RTA_PAYLOAD(tb[TCA_HFSC_RSC-1]) < sizeof(*rsc))
1078                         return -EINVAL;
1079                 rsc = RTA_DATA(tb[TCA_HFSC_RSC-1]);
1080                 if (rsc->m1 == 0 && rsc->m2 == 0)
1081                         rsc = NULL;
1082         }
1083
1084         if (tb[TCA_HFSC_FSC-1]) {
1085                 if (RTA_PAYLOAD(tb[TCA_HFSC_FSC-1]) < sizeof(*fsc))
1086                         return -EINVAL;
1087                 fsc = RTA_DATA(tb[TCA_HFSC_FSC-1]);
1088                 if (fsc->m1 == 0 && fsc->m2 == 0)
1089                         fsc = NULL;
1090         }
1091
1092         if (tb[TCA_HFSC_USC-1]) {
1093                 if (RTA_PAYLOAD(tb[TCA_HFSC_USC-1]) < sizeof(*usc))
1094                         return -EINVAL;
1095                 usc = RTA_DATA(tb[TCA_HFSC_USC-1]);
1096                 if (usc->m1 == 0 && usc->m2 == 0)
1097                         usc = NULL;
1098         }
1099
1100         if (cl != NULL) {
1101                 if (parentid) {
1102                         if (cl->cl_parent && cl->cl_parent->classid != parentid)
1103                                 return -EINVAL;
1104                         if (cl->cl_parent == NULL && parentid != TC_H_ROOT)
1105                                 return -EINVAL;
1106                 }
1107                 PSCHED_GET_TIME(cur_time);
1108
1109                 sch_tree_lock(sch);
1110                 if (rsc != NULL)
1111                         hfsc_change_rsc(cl, rsc, cur_time);
1112                 if (fsc != NULL)
1113                         hfsc_change_fsc(cl, fsc);
1114                 if (usc != NULL)
1115                         hfsc_change_usc(cl, usc, cur_time);
1116
1117                 if (cl->qdisc->q.qlen != 0) {
1118                         if (cl->cl_flags & HFSC_RSC)
1119                                 update_ed(cl, qdisc_peek_len(cl->qdisc));
1120                         if (cl->cl_flags & HFSC_FSC)
1121                                 update_vf(cl, 0, cur_time);
1122                 }
1123                 sch_tree_unlock(sch);
1124
1125 #ifdef CONFIG_NET_ESTIMATOR
1126                 if (tca[TCA_RATE-1]) {
1127                         qdisc_kill_estimator(&cl->stats);
1128                         qdisc_new_estimator(&cl->stats, cl->stats_lock,
1129                                             tca[TCA_RATE-1]);
1130                 }
1131 #endif
1132                 return 0;
1133         }
1134
1135         if (parentid == TC_H_ROOT)
1136                 return -EEXIST;
1137
1138         parent = &q->root;
1139         if (parentid) {
1140                 parent = hfsc_find_class(parentid, sch);
1141                 if (parent == NULL)
1142                         return -ENOENT;
1143         }
1144
1145         if (classid == 0 || TC_H_MAJ(classid ^ sch->handle) != 0)
1146                 return -EINVAL;
1147         if (hfsc_find_class(classid, sch))
1148                 return -EEXIST;
1149
1150         if (rsc == NULL && fsc == NULL)
1151                 return -EINVAL;
1152
1153         cl = kmalloc(sizeof(struct hfsc_class), GFP_KERNEL);
1154         if (cl == NULL)
1155                 return -ENOBUFS;
1156         memset(cl, 0, sizeof(struct hfsc_class));
1157
1158         if (rsc != NULL)
1159                 hfsc_change_rsc(cl, rsc, 0);
1160         if (fsc != NULL)
1161                 hfsc_change_fsc(cl, fsc);
1162         if (usc != NULL)
1163                 hfsc_change_usc(cl, usc, 0);
1164
1165         cl->refcnt    = 1;
1166         cl->classid   = classid;
1167         cl->sched     = q;
1168         cl->cl_parent = parent;
1169         cl->qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
1170         if (cl->qdisc == NULL)
1171                 cl->qdisc = &noop_qdisc;
1172         cl->stats_lock = &sch->dev->queue_lock;
1173         INIT_LIST_HEAD(&cl->children);
1174         INIT_LIST_HEAD(&cl->actlist);
1175
1176         sch_tree_lock(sch);
1177         list_add_tail(&cl->hlist, &q->clhash[hfsc_hash(classid)]);
1178         list_add_tail(&cl->siblings, &parent->children);
1179         if (parent->level == 0)
1180                 hfsc_purge_queue(sch, parent);
1181         hfsc_adjust_levels(parent);
1182         sch_tree_unlock(sch);
1183
1184 #ifdef CONFIG_NET_ESTIMATOR
1185         if (tca[TCA_RATE-1])
1186                 qdisc_new_estimator(&cl->stats, cl->stats_lock,
1187                                     tca[TCA_RATE-1]);
1188 #endif
1189         *arg = (unsigned long)cl;
1190         return 0;
1191 }
1192
1193 static void
1194 hfsc_destroy_filters(struct tcf_proto **fl)
1195 {
1196         struct tcf_proto *tp;
1197
1198         while ((tp = *fl) != NULL) {
1199                 *fl = tp->next;
1200                 tcf_destroy(tp);
1201         }
1202 }
1203
1204 static void
1205 hfsc_destroy_class(struct Qdisc *sch, struct hfsc_class *cl)
1206 {
1207         struct hfsc_sched *q = qdisc_priv(sch);
1208
1209         hfsc_destroy_filters(&cl->filter_list);
1210         qdisc_destroy(cl->qdisc);
1211 #ifdef CONFIG_NET_ESTIMATOR
1212         qdisc_kill_estimator(&cl->stats);
1213 #endif
1214         if (cl != &q->root)
1215                 kfree(cl);
1216 }
1217
1218 static int
1219 hfsc_delete_class(struct Qdisc *sch, unsigned long arg)
1220 {
1221         struct hfsc_sched *q = qdisc_priv(sch);
1222         struct hfsc_class *cl = (struct hfsc_class *)arg;
1223
1224         if (cl->level > 0 || cl->filter_cnt > 0 || cl == &q->root)
1225                 return -EBUSY;
1226
1227         sch_tree_lock(sch);
1228
1229         list_del(&cl->hlist);
1230         list_del(&cl->siblings);
1231         hfsc_adjust_levels(cl->cl_parent);
1232         hfsc_purge_queue(sch, cl);
1233         if (--cl->refcnt == 0)
1234                 hfsc_destroy_class(sch, cl);
1235
1236         sch_tree_unlock(sch);
1237         return 0;
1238 }
1239
1240 static struct hfsc_class *
1241 hfsc_classify(struct sk_buff *skb, struct Qdisc *sch, int *qres)
1242 {
1243         struct hfsc_sched *q = qdisc_priv(sch);
1244         struct hfsc_class *cl;
1245         struct tcf_result res;
1246         struct tcf_proto *tcf;
1247         int result;
1248
1249         if (TC_H_MAJ(skb->priority ^ sch->handle) == 0 &&
1250             (cl = hfsc_find_class(skb->priority, sch)) != NULL)
1251                 if (cl->level == 0)
1252                         return cl;
1253
1254         tcf = q->root.filter_list;
1255         while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) {
1256 #ifdef CONFIG_NET_CLS_ACT
1257                 int terminal = 0;
1258                 switch (result) {
1259                 case TC_ACT_SHOT:
1260                         *qres = NET_XMIT_DROP;
1261                         terminal = 1;
1262                         break;
1263                 case TC_ACT_QUEUED:
1264                 case TC_ACT_STOLEN:
1265                         terminal = 1;
1266                         break;
1267                 case TC_ACT_RECLASSIFY:
1268                 case TC_ACT_OK:
1269                 case TC_ACT_UNSPEC:
1270                 default:
1271                 break;
1272                 }
1273
1274                 if (terminal) {
1275                         kfree_skb(skb);
1276                         return NULL;
1277                 }
1278 #else
1279 #ifdef CONFIG_NET_CLS_POLICE
1280                 if (result == TC_POLICE_SHOT)
1281                         return NULL;
1282 #endif
1283 #endif
1284                 if ((cl = (struct hfsc_class *)res.class) == NULL) {
1285                         if ((cl = hfsc_find_class(res.classid, sch)) == NULL)
1286                                 break; /* filter selected invalid classid */
1287                 }
1288
1289                 if (cl->level == 0)
1290                         return cl; /* hit leaf class */
1291
1292                 /* apply inner filter chain */
1293                 tcf = cl->filter_list;
1294         }
1295
1296         /* classification failed, try default class */
1297         cl = hfsc_find_class(TC_H_MAKE(TC_H_MAJ(sch->handle), q->defcls), sch);
1298         if (cl == NULL || cl->level > 0)
1299                 return NULL;
1300
1301         return cl;
1302 }
1303
1304 static int
1305 hfsc_graft_class(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
1306                  struct Qdisc **old)
1307 {
1308         struct hfsc_class *cl = (struct hfsc_class *)arg;
1309
1310         if (cl == NULL)
1311                 return -ENOENT;
1312         if (cl->level > 0)
1313                 return -EINVAL;
1314         if (new == NULL) {
1315                 new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
1316                 if (new == NULL)
1317                         new = &noop_qdisc;
1318         }
1319
1320         sch_tree_lock(sch);
1321         hfsc_purge_queue(sch, cl);
1322         *old = xchg(&cl->qdisc, new);
1323         sch_tree_unlock(sch);
1324         return 0;
1325 }
1326
1327 static struct Qdisc *
1328 hfsc_class_leaf(struct Qdisc *sch, unsigned long arg)
1329 {
1330         struct hfsc_class *cl = (struct hfsc_class *)arg;
1331
1332         if (cl != NULL && cl->level == 0)
1333                 return cl->qdisc;
1334
1335         return NULL;
1336 }
1337
1338 static unsigned long
1339 hfsc_get_class(struct Qdisc *sch, u32 classid)
1340 {
1341         struct hfsc_class *cl = hfsc_find_class(classid, sch);
1342
1343         if (cl != NULL)
1344                 cl->refcnt++;
1345
1346         return (unsigned long)cl;
1347 }
1348
1349 static void
1350 hfsc_put_class(struct Qdisc *sch, unsigned long arg)
1351 {
1352         struct hfsc_class *cl = (struct hfsc_class *)arg;
1353
1354         if (--cl->refcnt == 0)
1355                 hfsc_destroy_class(sch, cl);
1356 }
1357
1358 static unsigned long
1359 hfsc_bind_tcf(struct Qdisc *sch, unsigned long parent, u32 classid)
1360 {
1361         struct hfsc_class *p = (struct hfsc_class *)parent;
1362         struct hfsc_class *cl = hfsc_find_class(classid, sch);
1363
1364         if (cl != NULL) {
1365                 if (p != NULL && p->level <= cl->level)
1366                         return 0;
1367                 cl->filter_cnt++;
1368         }
1369
1370         return (unsigned long)cl;
1371 }
1372
1373 static void
1374 hfsc_unbind_tcf(struct Qdisc *sch, unsigned long arg)
1375 {
1376         struct hfsc_class *cl = (struct hfsc_class *)arg;
1377
1378         cl->filter_cnt--;
1379 }
1380
1381 static struct tcf_proto **
1382 hfsc_tcf_chain(struct Qdisc *sch, unsigned long arg)
1383 {
1384         struct hfsc_sched *q = qdisc_priv(sch);
1385         struct hfsc_class *cl = (struct hfsc_class *)arg;
1386
1387         if (cl == NULL)
1388                 cl = &q->root;
1389
1390         return &cl->filter_list;
1391 }
1392
1393 static int
1394 hfsc_dump_sc(struct sk_buff *skb, int attr, struct internal_sc *sc)
1395 {
1396         struct tc_service_curve tsc;
1397
1398         tsc.m1 = sm2m(sc->sm1);
1399         tsc.d  = dx2d(sc->dx);
1400         tsc.m2 = sm2m(sc->sm2);
1401         RTA_PUT(skb, attr, sizeof(tsc), &tsc);
1402
1403         return skb->len;
1404
1405  rtattr_failure:
1406         return -1;
1407 }
1408
1409 static inline int
1410 hfsc_dump_curves(struct sk_buff *skb, struct hfsc_class *cl)
1411 {
1412         if ((cl->cl_flags & HFSC_RSC) &&
1413             (hfsc_dump_sc(skb, TCA_HFSC_RSC, &cl->cl_rsc) < 0))
1414                 goto rtattr_failure;
1415
1416         if ((cl->cl_flags & HFSC_FSC) &&
1417             (hfsc_dump_sc(skb, TCA_HFSC_FSC, &cl->cl_fsc) < 0))
1418                 goto rtattr_failure;
1419
1420         if ((cl->cl_flags & HFSC_USC) &&
1421             (hfsc_dump_sc(skb, TCA_HFSC_USC, &cl->cl_usc) < 0))
1422                 goto rtattr_failure;
1423
1424         return skb->len;
1425
1426  rtattr_failure:
1427         return -1;
1428 }
1429
1430 static inline int
1431 hfsc_dump_stats(struct sk_buff *skb, struct hfsc_class *cl)
1432 {
1433         cl->stats.qlen = cl->qdisc->q.qlen;
1434         if (qdisc_copy_stats(skb, &cl->stats, cl->stats_lock) < 0)
1435                 goto rtattr_failure;
1436
1437         return skb->len;
1438
1439  rtattr_failure:
1440         return -1;
1441 }
1442
1443 static inline int
1444 hfsc_dump_xstats(struct sk_buff *skb, struct hfsc_class *cl)
1445 {
1446         struct tc_hfsc_stats xstats;
1447
1448         xstats.level  = cl->level;
1449         xstats.period = cl->cl_vtperiod;
1450         xstats.work   = cl->cl_total;
1451         xstats.rtwork = cl->cl_cumul;
1452         RTA_PUT(skb, TCA_XSTATS, sizeof(xstats), &xstats);
1453
1454         return skb->len;
1455
1456  rtattr_failure:
1457         return -1;
1458 }
1459
1460 static int
1461 hfsc_dump_class(struct Qdisc *sch, unsigned long arg, struct sk_buff *skb,
1462                 struct tcmsg *tcm)
1463 {
1464         struct hfsc_class *cl = (struct hfsc_class *)arg;
1465         unsigned char *b = skb->tail;
1466         struct rtattr *rta = (struct rtattr *)b;
1467
1468         tcm->tcm_parent = cl->cl_parent ? cl->cl_parent->classid : TC_H_ROOT;
1469         tcm->tcm_handle = cl->classid;
1470         if (cl->level == 0)
1471                 tcm->tcm_info = cl->qdisc->handle;
1472
1473         RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
1474         if (hfsc_dump_curves(skb, cl) < 0)
1475                 goto rtattr_failure;
1476         rta->rta_len = skb->tail - b;
1477
1478         if ((hfsc_dump_stats(skb, cl) < 0) ||
1479             (hfsc_dump_xstats(skb, cl) < 0))
1480                 goto rtattr_failure;
1481
1482         return skb->len;
1483
1484  rtattr_failure:
1485         skb_trim(skb, b - skb->data);
1486         return -1;
1487 }
1488
1489 static void
1490 hfsc_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1491 {
1492         struct hfsc_sched *q = qdisc_priv(sch);
1493         struct hfsc_class *cl;
1494         unsigned int i;
1495
1496         if (arg->stop)
1497                 return;
1498
1499         for (i = 0; i < HFSC_HSIZE; i++) {
1500                 list_for_each_entry(cl, &q->clhash[i], hlist) {
1501                         if (arg->count < arg->skip) {
1502                                 arg->count++;
1503                                 continue;
1504                         }
1505                         if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
1506                                 arg->stop = 1;
1507                                 return;
1508                         }
1509                         arg->count++;
1510                 }
1511         }
1512 }
1513
1514 static void
1515 hfsc_watchdog(unsigned long arg)
1516 {
1517         struct Qdisc *sch = (struct Qdisc *)arg;
1518
1519         sch->flags &= ~TCQ_F_THROTTLED;
1520         netif_schedule(sch->dev);
1521 }
1522
1523 static void
1524 hfsc_schedule_watchdog(struct Qdisc *sch, u64 cur_time)
1525 {
1526         struct hfsc_sched *q = qdisc_priv(sch);
1527         struct hfsc_class *cl;
1528         u64 next_time = 0;
1529         long delay;
1530
1531         if ((cl = ellist_get_minel(&q->eligible)) != NULL)
1532                 next_time = cl->cl_e;
1533         if (q->root.cl_cfmin != 0) {
1534                 if (next_time == 0 || next_time > q->root.cl_cfmin)
1535                         next_time = q->root.cl_cfmin;
1536         }
1537         ASSERT(next_time != 0);
1538         delay = next_time - cur_time;
1539         delay = PSCHED_US2JIFFIE(delay);
1540
1541         sch->flags |= TCQ_F_THROTTLED;
1542         mod_timer(&q->wd_timer, jiffies + delay);
1543 }
1544
1545 static int
1546 hfsc_init_qdisc(struct Qdisc *sch, struct rtattr *opt)
1547 {
1548         struct hfsc_sched *q = qdisc_priv(sch);
1549         struct tc_hfsc_qopt *qopt;
1550         unsigned int i;
1551
1552         if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1553                 return -EINVAL;
1554         qopt = RTA_DATA(opt);
1555
1556         memset(q, 0, sizeof(struct hfsc_sched));
1557         sch->stats_lock = &sch->dev->queue_lock;
1558
1559         q->defcls = qopt->defcls;
1560         for (i = 0; i < HFSC_HSIZE; i++)
1561                 INIT_LIST_HEAD(&q->clhash[i]);
1562         INIT_LIST_HEAD(&q->eligible);
1563         INIT_LIST_HEAD(&q->droplist);
1564         skb_queue_head_init(&q->requeue);
1565
1566         q->root.refcnt  = 1;
1567         q->root.classid = sch->handle;
1568         q->root.sched   = q;
1569         q->root.qdisc = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
1570         if (q->root.qdisc == NULL)
1571                 q->root.qdisc = &noop_qdisc;
1572         q->root.stats_lock = &sch->dev->queue_lock;
1573         INIT_LIST_HEAD(&q->root.children);
1574         INIT_LIST_HEAD(&q->root.actlist);
1575
1576         list_add(&q->root.hlist, &q->clhash[hfsc_hash(q->root.classid)]);
1577
1578         init_timer(&q->wd_timer);
1579         q->wd_timer.function = hfsc_watchdog;
1580         q->wd_timer.data = (unsigned long)sch;
1581
1582         return 0;
1583 }
1584
1585 static int
1586 hfsc_change_qdisc(struct Qdisc *sch, struct rtattr *opt)
1587 {
1588         struct hfsc_sched *q = qdisc_priv(sch);
1589         struct tc_hfsc_qopt *qopt;
1590
1591         if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(*qopt))
1592                 return -EINVAL;
1593         qopt = RTA_DATA(opt);
1594
1595         sch_tree_lock(sch);
1596         q->defcls = qopt->defcls;
1597         sch_tree_unlock(sch);
1598
1599         return 0;
1600 }
1601
1602 static void
1603 hfsc_reset_class(struct hfsc_class *cl)
1604 {
1605         cl->cl_total        = 0;
1606         cl->cl_cumul        = 0;
1607         cl->cl_d            = 0;
1608         cl->cl_e            = 0;
1609         cl->cl_vt           = 0;
1610         cl->cl_vtadj        = 0;
1611         cl->cl_vtoff        = 0;
1612         cl->cl_cvtmin       = 0;
1613         cl->cl_cvtmax       = 0;
1614         cl->cl_vtperiod     = 0;
1615         cl->cl_parentperiod = 0;
1616         cl->cl_f            = 0;
1617         cl->cl_myf          = 0;
1618         cl->cl_myfadj       = 0;
1619         cl->cl_cfmin        = 0;
1620         cl->cl_nactive      = 0;
1621         INIT_LIST_HEAD(&cl->actlist);
1622         qdisc_reset(cl->qdisc);
1623
1624         if (cl->cl_flags & HFSC_RSC)
1625                 rtsc_init(&cl->cl_deadline, &cl->cl_rsc, 0, 0);
1626         if (cl->cl_flags & HFSC_FSC)
1627                 rtsc_init(&cl->cl_virtual, &cl->cl_fsc, 0, 0);
1628         if (cl->cl_flags & HFSC_USC)
1629                 rtsc_init(&cl->cl_ulimit, &cl->cl_usc, 0, 0);
1630 }
1631
1632 static void
1633 hfsc_reset_qdisc(struct Qdisc *sch)
1634 {
1635         struct hfsc_sched *q = qdisc_priv(sch);
1636         struct hfsc_class *cl;
1637         unsigned int i;
1638
1639         for (i = 0; i < HFSC_HSIZE; i++) {
1640                 list_for_each_entry(cl, &q->clhash[i], hlist)
1641                         hfsc_reset_class(cl);
1642         }
1643         __skb_queue_purge(&q->requeue);
1644         INIT_LIST_HEAD(&q->eligible);
1645         INIT_LIST_HEAD(&q->droplist);
1646         del_timer(&q->wd_timer);
1647         sch->flags &= ~TCQ_F_THROTTLED;
1648         sch->q.qlen = 0;
1649 }
1650
1651 static void
1652 hfsc_destroy_qdisc(struct Qdisc *sch)
1653 {
1654         struct hfsc_sched *q = qdisc_priv(sch);
1655         struct hfsc_class *cl, *next;
1656         unsigned int i;
1657
1658         for (i = 0; i < HFSC_HSIZE; i++) {
1659                 list_for_each_entry_safe(cl, next, &q->clhash[i], hlist)
1660                         hfsc_destroy_class(sch, cl);
1661         }
1662         __skb_queue_purge(&q->requeue);
1663         del_timer(&q->wd_timer);
1664 }
1665
1666 static int
1667 hfsc_dump_qdisc(struct Qdisc *sch, struct sk_buff *skb)
1668 {
1669         struct hfsc_sched *q = qdisc_priv(sch);
1670         unsigned char *b = skb->tail;
1671         struct tc_hfsc_qopt qopt;
1672
1673         qopt.defcls = q->defcls;
1674         RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
1675
1676         sch->stats.qlen = sch->q.qlen;
1677         if (qdisc_copy_stats(skb, &sch->stats, sch->stats_lock) < 0)
1678                 goto rtattr_failure;
1679
1680         return skb->len;
1681
1682  rtattr_failure:
1683         skb_trim(skb, b - skb->data);
1684         return -1;
1685 }
1686
1687 static int
1688 hfsc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1689 {
1690         int ret = NET_XMIT_SUCCESS;
1691         struct hfsc_class *cl = hfsc_classify(skb, sch, &ret);
1692         unsigned int len = skb->len;
1693         int err;
1694
1695
1696 #ifdef CONFIG_NET_CLS_ACT
1697         if (cl == NULL) {
1698                 if (NET_XMIT_DROP == ret) {
1699                         sch->stats.drops++;
1700                 }
1701                 return ret;
1702         }
1703 #else
1704         if (cl == NULL) {
1705                 kfree_skb(skb);
1706                 sch->stats.drops++;
1707                 return NET_XMIT_DROP;
1708         }
1709 #endif
1710
1711         err = cl->qdisc->enqueue(skb, cl->qdisc);
1712         if (unlikely(err != NET_XMIT_SUCCESS)) {
1713                 cl->stats.drops++;
1714                 sch->stats.drops++;
1715                 return err;
1716         }
1717
1718         if (cl->qdisc->q.qlen == 1)
1719                 set_active(cl, len);
1720
1721         cl->stats.packets++;
1722         cl->stats.bytes += len;
1723         sch->stats.packets++;
1724         sch->stats.bytes += len;
1725         sch->q.qlen++;
1726
1727         return NET_XMIT_SUCCESS;
1728 }
1729
1730 static struct sk_buff *
1731 hfsc_dequeue(struct Qdisc *sch)
1732 {
1733         struct hfsc_sched *q = qdisc_priv(sch);
1734         struct hfsc_class *cl;
1735         struct sk_buff *skb;
1736         u64 cur_time;
1737         unsigned int next_len;
1738         int realtime = 0;
1739
1740         if (sch->q.qlen == 0)
1741                 return NULL;
1742         if ((skb = __skb_dequeue(&q->requeue)))
1743                 goto out;
1744
1745         PSCHED_GET_TIME(cur_time);
1746
1747         /*
1748          * if there are eligible classes, use real-time criteria.
1749          * find the class with the minimum deadline among
1750          * the eligible classes.
1751          */
1752         if ((cl = ellist_get_mindl(&q->eligible, cur_time)) != NULL) {
1753                 realtime = 1;
1754         } else {
1755                 /*
1756                  * use link-sharing criteria
1757                  * get the class with the minimum vt in the hierarchy
1758                  */
1759                 cl = actlist_get_minvt(&q->root, cur_time);
1760                 if (cl == NULL) {
1761                         sch->stats.overlimits++;
1762                         hfsc_schedule_watchdog(sch, cur_time);
1763                         return NULL;
1764                 }
1765         }
1766
1767         skb = cl->qdisc->dequeue(cl->qdisc);
1768         if (skb == NULL) {
1769                 if (net_ratelimit())
1770                         printk("HFSC: Non-work-conserving qdisc ?\n");
1771                 return NULL;
1772         }
1773
1774         update_vf(cl, skb->len, cur_time);
1775         if (realtime)
1776                 cl->cl_cumul += skb->len;
1777
1778         if (cl->qdisc->q.qlen != 0) {
1779                 if (cl->cl_flags & HFSC_RSC) {
1780                         /* update ed */
1781                         next_len = qdisc_peek_len(cl->qdisc);
1782                         if (realtime)
1783                                 update_ed(cl, next_len);
1784                         else
1785                                 update_d(cl, next_len);
1786                 }
1787         } else {
1788                 /* the class becomes passive */
1789                 set_passive(cl);
1790         }
1791
1792  out:
1793         sch->flags &= ~TCQ_F_THROTTLED;
1794         sch->q.qlen--;
1795
1796         return skb;
1797 }
1798
1799 static int
1800 hfsc_requeue(struct sk_buff *skb, struct Qdisc *sch)
1801 {
1802         struct hfsc_sched *q = qdisc_priv(sch);
1803
1804         __skb_queue_head(&q->requeue, skb);
1805         sch->q.qlen++;
1806         return NET_XMIT_SUCCESS;
1807 }
1808
1809 static unsigned int
1810 hfsc_drop(struct Qdisc *sch)
1811 {
1812         struct hfsc_sched *q = qdisc_priv(sch);
1813         struct hfsc_class *cl;
1814         unsigned int len;
1815
1816         list_for_each_entry(cl, &q->droplist, dlist) {
1817                 if (cl->qdisc->ops->drop != NULL &&
1818                     (len = cl->qdisc->ops->drop(cl->qdisc)) > 0) {
1819                         if (cl->qdisc->q.qlen == 0) {
1820                                 update_vf(cl, 0, 0);
1821                                 set_passive(cl);
1822                         } else {
1823                                 list_move_tail(&cl->dlist, &q->droplist);
1824                         }
1825                         cl->stats.drops++;
1826                         sch->stats.drops++;
1827                         sch->q.qlen--;
1828                         return len;
1829                 }
1830         }
1831         return 0;
1832 }
1833
1834 static struct Qdisc_class_ops hfsc_class_ops = {
1835         .change         = hfsc_change_class,
1836         .delete         = hfsc_delete_class,
1837         .graft          = hfsc_graft_class,
1838         .leaf           = hfsc_class_leaf,
1839         .get            = hfsc_get_class,
1840         .put            = hfsc_put_class,
1841         .bind_tcf       = hfsc_bind_tcf,
1842         .unbind_tcf     = hfsc_unbind_tcf,
1843         .tcf_chain      = hfsc_tcf_chain,
1844         .dump           = hfsc_dump_class,
1845         .walk           = hfsc_walk
1846 };
1847
1848 static struct Qdisc_ops hfsc_qdisc_ops = {
1849         .id             = "hfsc",
1850         .init           = hfsc_init_qdisc,
1851         .change         = hfsc_change_qdisc,
1852         .reset          = hfsc_reset_qdisc,
1853         .destroy        = hfsc_destroy_qdisc,
1854         .dump           = hfsc_dump_qdisc,
1855         .enqueue        = hfsc_enqueue,
1856         .dequeue        = hfsc_dequeue,
1857         .requeue        = hfsc_requeue,
1858         .drop           = hfsc_drop,
1859         .cl_ops         = &hfsc_class_ops,
1860         .priv_size      = sizeof(struct hfsc_sched),
1861         .owner          = THIS_MODULE
1862 };
1863
1864 static int __init
1865 hfsc_init(void)
1866 {
1867         return register_qdisc(&hfsc_qdisc_ops);
1868 }
1869
1870 static void __exit
1871 hfsc_cleanup(void)
1872 {
1873         unregister_qdisc(&hfsc_qdisc_ops);
1874 }
1875
1876 MODULE_LICENSE("GPL");
1877 module_init(hfsc_init);
1878 module_exit(hfsc_cleanup);