diff options
-rw-r--r-- | net/sched/sch_htb.c | 91 |
1 files changed, 31 insertions, 60 deletions
diff --git a/net/sched/sch_htb.c b/net/sched/sch_htb.c index 3c3294d01041..4d84200f097b 100644 --- a/net/sched/sch_htb.c +++ b/net/sched/sch_htb.c | |||
@@ -128,7 +128,7 @@ struct htb_class { | |||
128 | } un; | 128 | } un; |
129 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ | 129 | struct rb_node node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ |
130 | struct rb_node pq_node; /* node for event queue */ | 130 | struct rb_node pq_node; /* node for event queue */ |
131 | unsigned long pq_key; /* the same type as jiffies global */ | 131 | psched_time_t pq_key; |
132 | 132 | ||
133 | int prio_activity; /* for which prios are we active */ | 133 | int prio_activity; /* for which prios are we active */ |
134 | enum htb_cmode cmode; /* current mode of the class */ | 134 | enum htb_cmode cmode; /* current mode of the class */ |
@@ -179,10 +179,7 @@ struct htb_sched { | |||
179 | struct rb_root wait_pq[TC_HTB_MAXDEPTH]; | 179 | struct rb_root wait_pq[TC_HTB_MAXDEPTH]; |
180 | 180 | ||
181 | /* time of nearest event per level (row) */ | 181 | /* time of nearest event per level (row) */ |
182 | unsigned long near_ev_cache[TC_HTB_MAXDEPTH]; | 182 | psched_time_t near_ev_cache[TC_HTB_MAXDEPTH]; |
183 | |||
184 | /* cached value of jiffies in dequeue */ | ||
185 | unsigned long jiffies; | ||
186 | 183 | ||
187 | /* whether we hit non-work conserving class during this dequeue; we use */ | 184 | /* whether we hit non-work conserving class during this dequeue; we use */ |
188 | int nwc_hit; /* this to disable mindelay complaint in dequeue */ | 185 | int nwc_hit; /* this to disable mindelay complaint in dequeue */ |
@@ -195,7 +192,7 @@ struct htb_sched { | |||
195 | 192 | ||
196 | int rate2quantum; /* quant = rate / rate2quantum */ | 193 | int rate2quantum; /* quant = rate / rate2quantum */ |
197 | psched_time_t now; /* cached dequeue time */ | 194 | psched_time_t now; /* cached dequeue time */ |
198 | struct timer_list timer; /* send delay timer */ | 195 | struct qdisc_watchdog watchdog; |
199 | #ifdef HTB_RATECM | 196 | #ifdef HTB_RATECM |
200 | struct timer_list rttim; /* rate computer timer */ | 197 | struct timer_list rttim; /* rate computer timer */ |
201 | int recmp_bucket; /* which hash bucket to recompute next */ | 198 | int recmp_bucket; /* which hash bucket to recompute next */ |
@@ -342,19 +339,19 @@ static void htb_add_to_wait_tree(struct htb_sched *q, | |||
342 | { | 339 | { |
343 | struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; | 340 | struct rb_node **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; |
344 | 341 | ||
345 | cl->pq_key = q->jiffies + PSCHED_US2JIFFIE(delay); | 342 | cl->pq_key = q->now + delay; |
346 | if (cl->pq_key == q->jiffies) | 343 | if (cl->pq_key == q->now) |
347 | cl->pq_key++; | 344 | cl->pq_key++; |
348 | 345 | ||
349 | /* update the nearest event cache */ | 346 | /* update the nearest event cache */ |
350 | if (time_after(q->near_ev_cache[cl->level], cl->pq_key)) | 347 | if (q->near_ev_cache[cl->level] > cl->pq_key) |
351 | q->near_ev_cache[cl->level] = cl->pq_key; | 348 | q->near_ev_cache[cl->level] = cl->pq_key; |
352 | 349 | ||
353 | while (*p) { | 350 | while (*p) { |
354 | struct htb_class *c; | 351 | struct htb_class *c; |
355 | parent = *p; | 352 | parent = *p; |
356 | c = rb_entry(parent, struct htb_class, pq_node); | 353 | c = rb_entry(parent, struct htb_class, pq_node); |
357 | if (time_after_eq(cl->pq_key, c->pq_key)) | 354 | if (cl->pq_key >= c->pq_key) |
358 | p = &parent->rb_right; | 355 | p = &parent->rb_right; |
359 | else | 356 | else |
360 | p = &parent->rb_left; | 357 | p = &parent->rb_left; |
@@ -679,14 +676,6 @@ static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch) | |||
679 | return NET_XMIT_SUCCESS; | 676 | return NET_XMIT_SUCCESS; |
680 | } | 677 | } |
681 | 678 | ||
682 | static void htb_timer(unsigned long arg) | ||
683 | { | ||
684 | struct Qdisc *sch = (struct Qdisc *)arg; | ||
685 | sch->flags &= ~TCQ_F_THROTTLED; | ||
686 | wmb(); | ||
687 | netif_schedule(sch->dev); | ||
688 | } | ||
689 | |||
690 | #ifdef HTB_RATECM | 679 | #ifdef HTB_RATECM |
691 | #define RT_GEN(D,R) R+=D-(R/HTB_EWMAC);D=0 | 680 | #define RT_GEN(D,R) R+=D-(R/HTB_EWMAC);D=0 |
692 | static void htb_rate_timer(unsigned long arg) | 681 | static void htb_rate_timer(unsigned long arg) |
@@ -778,11 +767,11 @@ static void htb_charge_class(struct htb_sched *q, struct htb_class *cl, | |||
778 | /** | 767 | /** |
779 | * htb_do_events - make mode changes to classes at the level | 768 | * htb_do_events - make mode changes to classes at the level |
780 | * | 769 | * |
781 | * Scans event queue for pending events and applies them. Returns jiffies to | 770 | * Scans event queue for pending events and applies them. Returns time of |
782 | * next pending event (0 for no event in pq). | 771 | * next pending event (0 for no event in pq). |
783 | * Note: Aplied are events whose have cl->pq_key <= jiffies. | 772 | * Note: Applied are events whose have cl->pq_key <= q->now. |
784 | */ | 773 | */ |
785 | static long htb_do_events(struct htb_sched *q, int level) | 774 | static psched_time_t htb_do_events(struct htb_sched *q, int level) |
786 | { | 775 | { |
787 | int i; | 776 | int i; |
788 | 777 | ||
@@ -795,9 +784,9 @@ static long htb_do_events(struct htb_sched *q, int level) | |||
795 | return 0; | 784 | return 0; |
796 | 785 | ||
797 | cl = rb_entry(p, struct htb_class, pq_node); | 786 | cl = rb_entry(p, struct htb_class, pq_node); |
798 | if (time_after(cl->pq_key, q->jiffies)) { | 787 | if (cl->pq_key > q->now) |
799 | return cl->pq_key - q->jiffies; | 788 | return cl->pq_key; |
800 | } | 789 | |
801 | htb_safe_rb_erase(p, q->wait_pq + level); | 790 | htb_safe_rb_erase(p, q->wait_pq + level); |
802 | diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer); | 791 | diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32) cl->mbuffer); |
803 | htb_change_class_mode(q, cl, &diff); | 792 | htb_change_class_mode(q, cl, &diff); |
@@ -806,7 +795,7 @@ static long htb_do_events(struct htb_sched *q, int level) | |||
806 | } | 795 | } |
807 | if (net_ratelimit()) | 796 | if (net_ratelimit()) |
808 | printk(KERN_WARNING "htb: too many events !\n"); | 797 | printk(KERN_WARNING "htb: too many events !\n"); |
809 | return HZ / 10; | 798 | return q->now + PSCHED_TICKS_PER_SEC / 10; |
810 | } | 799 | } |
811 | 800 | ||
812 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL | 801 | /* Returns class->node+prio from id-tree where classe's id is >= id. NULL |
@@ -958,30 +947,12 @@ next: | |||
958 | return skb; | 947 | return skb; |
959 | } | 948 | } |
960 | 949 | ||
961 | static void htb_delay_by(struct Qdisc *sch, long delay) | ||
962 | { | ||
963 | struct htb_sched *q = qdisc_priv(sch); | ||
964 | if (delay <= 0) | ||
965 | delay = 1; | ||
966 | if (unlikely(delay > 5 * HZ)) { | ||
967 | if (net_ratelimit()) | ||
968 | printk(KERN_INFO "HTB delay %ld > 5sec\n", delay); | ||
969 | delay = 5 * HZ; | ||
970 | } | ||
971 | /* why don't use jiffies here ? because expires can be in past */ | ||
972 | mod_timer(&q->timer, q->jiffies + delay); | ||
973 | sch->flags |= TCQ_F_THROTTLED; | ||
974 | sch->qstats.overlimits++; | ||
975 | } | ||
976 | |||
977 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) | 950 | static struct sk_buff *htb_dequeue(struct Qdisc *sch) |
978 | { | 951 | { |
979 | struct sk_buff *skb = NULL; | 952 | struct sk_buff *skb = NULL; |
980 | struct htb_sched *q = qdisc_priv(sch); | 953 | struct htb_sched *q = qdisc_priv(sch); |
981 | int level; | 954 | int level; |
982 | long min_delay; | 955 | psched_time_t next_event; |
983 | |||
984 | q->jiffies = jiffies; | ||
985 | 956 | ||
986 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ | 957 | /* try to dequeue direct packets as high prio (!) to minimize cpu work */ |
987 | skb = __skb_dequeue(&q->direct_queue); | 958 | skb = __skb_dequeue(&q->direct_queue); |
@@ -995,21 +966,23 @@ static struct sk_buff *htb_dequeue(struct Qdisc *sch) | |||
995 | goto fin; | 966 | goto fin; |
996 | PSCHED_GET_TIME(q->now); | 967 | PSCHED_GET_TIME(q->now); |
997 | 968 | ||
998 | min_delay = LONG_MAX; | 969 | next_event = q->now + 5 * PSCHED_TICKS_PER_SEC; |
999 | q->nwc_hit = 0; | 970 | q->nwc_hit = 0; |
1000 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { | 971 | for (level = 0; level < TC_HTB_MAXDEPTH; level++) { |
1001 | /* common case optimization - skip event handler quickly */ | 972 | /* common case optimization - skip event handler quickly */ |
1002 | int m; | 973 | int m; |
1003 | long delay; | 974 | psched_time_t event; |
1004 | if (time_after_eq(q->jiffies, q->near_ev_cache[level])) { | 975 | |
1005 | delay = htb_do_events(q, level); | 976 | if (q->now >= q->near_ev_cache[level]) { |
1006 | q->near_ev_cache[level] = | 977 | event = htb_do_events(q, level); |
1007 | q->jiffies + (delay ? delay : HZ); | 978 | q->near_ev_cache[level] = event ? event : |
979 | PSCHED_TICKS_PER_SEC; | ||
1008 | } else | 980 | } else |
1009 | delay = q->near_ev_cache[level] - q->jiffies; | 981 | event = q->near_ev_cache[level]; |
982 | |||
983 | if (event && next_event > event) | ||
984 | next_event = event; | ||
1010 | 985 | ||
1011 | if (delay && min_delay > delay) | ||
1012 | min_delay = delay; | ||
1013 | m = ~q->row_mask[level]; | 986 | m = ~q->row_mask[level]; |
1014 | while (m != (int)(-1)) { | 987 | while (m != (int)(-1)) { |
1015 | int prio = ffz(m); | 988 | int prio = ffz(m); |
@@ -1022,7 +995,8 @@ static struct sk_buff *htb_dequeue(struct Qdisc *sch) | |||
1022 | } | 995 | } |
1023 | } | 996 | } |
1024 | } | 997 | } |
1025 | htb_delay_by(sch, min_delay > 5 * HZ ? 5 * HZ : min_delay); | 998 | sch->qstats.overlimits++; |
999 | qdisc_watchdog_schedule(&q->watchdog, next_event); | ||
1026 | fin: | 1000 | fin: |
1027 | return skb; | 1001 | return skb; |
1028 | } | 1002 | } |
@@ -1075,8 +1049,7 @@ static void htb_reset(struct Qdisc *sch) | |||
1075 | 1049 | ||
1076 | } | 1050 | } |
1077 | } | 1051 | } |
1078 | sch->flags &= ~TCQ_F_THROTTLED; | 1052 | qdisc_watchdog_cancel(&q->watchdog); |
1079 | del_timer(&q->timer); | ||
1080 | __skb_queue_purge(&q->direct_queue); | 1053 | __skb_queue_purge(&q->direct_queue); |
1081 | sch->q.qlen = 0; | 1054 | sch->q.qlen = 0; |
1082 | memset(q->row, 0, sizeof(q->row)); | 1055 | memset(q->row, 0, sizeof(q->row)); |
@@ -1113,14 +1086,12 @@ static int htb_init(struct Qdisc *sch, struct rtattr *opt) | |||
1113 | for (i = 0; i < TC_HTB_NUMPRIO; i++) | 1086 | for (i = 0; i < TC_HTB_NUMPRIO; i++) |
1114 | INIT_LIST_HEAD(q->drops + i); | 1087 | INIT_LIST_HEAD(q->drops + i); |
1115 | 1088 | ||
1116 | init_timer(&q->timer); | 1089 | qdisc_watchdog_init(&q->watchdog, sch); |
1117 | skb_queue_head_init(&q->direct_queue); | 1090 | skb_queue_head_init(&q->direct_queue); |
1118 | 1091 | ||
1119 | q->direct_qlen = sch->dev->tx_queue_len; | 1092 | q->direct_qlen = sch->dev->tx_queue_len; |
1120 | if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ | 1093 | if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ |
1121 | q->direct_qlen = 2; | 1094 | q->direct_qlen = 2; |
1122 | q->timer.function = htb_timer; | ||
1123 | q->timer.data = (unsigned long)sch; | ||
1124 | 1095 | ||
1125 | #ifdef HTB_RATECM | 1096 | #ifdef HTB_RATECM |
1126 | init_timer(&q->rttim); | 1097 | init_timer(&q->rttim); |
@@ -1341,7 +1312,7 @@ static void htb_destroy(struct Qdisc *sch) | |||
1341 | { | 1312 | { |
1342 | struct htb_sched *q = qdisc_priv(sch); | 1313 | struct htb_sched *q = qdisc_priv(sch); |
1343 | 1314 | ||
1344 | del_timer_sync(&q->timer); | 1315 | qdisc_watchdog_cancel(&q->watchdog); |
1345 | #ifdef HTB_RATECM | 1316 | #ifdef HTB_RATECM |
1346 | del_timer_sync(&q->rttim); | 1317 | del_timer_sync(&q->rttim); |
1347 | #endif | 1318 | #endif |