x86: convert mm_context_t semaphore to a mutex
[linux-2.6-block.git] / include / linux / pkt_sched.h
CommitLineData
1da177e4
LT
1#ifndef __LINUX_PKT_SCHED_H
2#define __LINUX_PKT_SCHED_H
3
4/* Logical priority bands not depending on specific packet scheduler.
5 Every scheduler will map them to real traffic classes, if it has
6 no more precise mechanism to classify packets.
7
8 These numbers have no special meaning, though their coincidence
9 with obsolete IPv6 values is not occasional :-). New IPv6 drafts
10 preferred full anarchy inspired by diffserv group.
11
12 Note: TC_PRIO_BESTEFFORT does not mean that it is the most unhappy
13 class, actually, as rule it will be handled with more care than
14 filler or even bulk.
15 */
16
17#define TC_PRIO_BESTEFFORT 0
18#define TC_PRIO_FILLER 1
19#define TC_PRIO_BULK 2
20#define TC_PRIO_INTERACTIVE_BULK 4
21#define TC_PRIO_INTERACTIVE 6
22#define TC_PRIO_CONTROL 7
23
24#define TC_PRIO_MAX 15
25
26/* Generic queue statistics, available for all the elements.
27 Particular schedulers may have also their private records.
28 */
29
30struct tc_stats
31{
32 __u64 bytes; /* NUmber of enqueues bytes */
33 __u32 packets; /* Number of enqueued packets */
34 __u32 drops; /* Packets dropped because of lack of resources */
35 __u32 overlimits; /* Number of throttle events when this
36 * flow goes out of allocated bandwidth */
37 __u32 bps; /* Current flow byte rate */
38 __u32 pps; /* Current flow packet rate */
39 __u32 qlen;
40 __u32 backlog;
41};
42
43struct tc_estimator
44{
45 signed char interval;
46 unsigned char ewma_log;
47};
48
49/* "Handles"
50 ---------
51
52 All the traffic control objects have 32bit identifiers, or "handles".
53
54 They can be considered as opaque numbers from user API viewpoint,
55 but actually they always consist of two fields: major and
56 minor numbers, which are interpreted by kernel specially,
57 that may be used by applications, though not recommended.
58
59 F.e. qdisc handles always have minor number equal to zero,
60 classes (or flows) have major equal to parent qdisc major, and
61 minor uniquely identifying class inside qdisc.
62
63 Macros to manipulate handles:
64 */
65
66#define TC_H_MAJ_MASK (0xFFFF0000U)
67#define TC_H_MIN_MASK (0x0000FFFFU)
68#define TC_H_MAJ(h) ((h)&TC_H_MAJ_MASK)
69#define TC_H_MIN(h) ((h)&TC_H_MIN_MASK)
70#define TC_H_MAKE(maj,min) (((maj)&TC_H_MAJ_MASK)|((min)&TC_H_MIN_MASK))
71
72#define TC_H_UNSPEC (0U)
73#define TC_H_ROOT (0xFFFFFFFFU)
74#define TC_H_INGRESS (0xFFFFFFF1U)
75
76struct tc_ratespec
77{
78 unsigned char cell_log;
79 unsigned char __reserved;
e08b0998
JDB
80 unsigned short overhead;
81 short cell_align;
1da177e4
LT
82 unsigned short mpu;
83 __u32 rate;
84};
85
86/* FIFO section */
87
88struct tc_fifo_qopt
89{
90 __u32 limit; /* Queue length: bytes for bfifo, packets for pfifo */
91};
92
93/* PRIO section */
94
95#define TCQ_PRIO_BANDS 16
bdc450a0 96#define TCQ_MIN_PRIO_BANDS 2
1da177e4
LT
97
98struct tc_prio_qopt
99{
100 int bands; /* Number of bands */
101 __u8 priomap[TC_PRIO_MAX+1]; /* Map: logical priority -> PRIO band */
102};
103
d62733c8
PWJ
104enum
105{
106 TCA_PRIO_UNSPEC,
107 TCA_PRIO_MQ,
108 __TCA_PRIO_MAX
109};
110
111#define TCA_PRIO_MAX (__TCA_PRIO_MAX - 1)
112
1da177e4
LT
113/* TBF section */
114
115struct tc_tbf_qopt
116{
117 struct tc_ratespec rate;
118 struct tc_ratespec peakrate;
119 __u32 limit;
120 __u32 buffer;
121 __u32 mtu;
122};
123
124enum
125{
126 TCA_TBF_UNSPEC,
127 TCA_TBF_PARMS,
128 TCA_TBF_RTAB,
129 TCA_TBF_PTAB,
130 __TCA_TBF_MAX,
131};
132
133#define TCA_TBF_MAX (__TCA_TBF_MAX - 1)
134
135
136/* TEQL section */
137
138/* TEQL does not require any parameters */
139
140/* SFQ section */
141
142struct tc_sfq_qopt
143{
144 unsigned quantum; /* Bytes per round allocated to flow */
145 int perturb_period; /* Period of hash perturbation */
146 __u32 limit; /* Maximal packets in queue */
147 unsigned divisor; /* Hash divisor */
148 unsigned flows; /* Maximal number of flows */
149};
150
151/*
152 * NOTE: limit, divisor and flows are hardwired to code at the moment.
153 *
154 * limit=flows=128, divisor=1024;
155 *
156 * The only reason for this is efficiency, it is possible
157 * to change these parameters in compile time.
158 */
159
160/* RED section */
161
162enum
163{
164 TCA_RED_UNSPEC,
165 TCA_RED_PARMS,
166 TCA_RED_STAB,
167 __TCA_RED_MAX,
168};
169
170#define TCA_RED_MAX (__TCA_RED_MAX - 1)
171
172struct tc_red_qopt
173{
174 __u32 limit; /* HARD maximal queue length (bytes) */
175 __u32 qth_min; /* Min average length threshold (bytes) */
176 __u32 qth_max; /* Max average length threshold (bytes) */
177 unsigned char Wlog; /* log(W) */
178 unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */
179 unsigned char Scell_log; /* cell size for idle damping */
180 unsigned char flags;
181#define TC_RED_ECN 1
bdc450a0 182#define TC_RED_HARDDROP 2
1da177e4
LT
183};
184
185struct tc_red_xstats
186{
187 __u32 early; /* Early drops */
188 __u32 pdrop; /* Drops due to queue limits */
189 __u32 other; /* Drops due to drop() calls */
190 __u32 marked; /* Marked packets */
191};
192
193/* GRED section */
194
195#define MAX_DPs 16
196
197enum
198{
199 TCA_GRED_UNSPEC,
200 TCA_GRED_PARMS,
201 TCA_GRED_STAB,
202 TCA_GRED_DPS,
203 __TCA_GRED_MAX,
204};
205
206#define TCA_GRED_MAX (__TCA_GRED_MAX - 1)
207
1da177e4
LT
208struct tc_gred_qopt
209{
1e4dfaf9
TG
210 __u32 limit; /* HARD maximal queue length (bytes) */
211 __u32 qth_min; /* Min average length threshold (bytes) */
212 __u32 qth_max; /* Max average length threshold (bytes) */
213 __u32 DP; /* upto 2^32 DPs */
214 __u32 backlog;
215 __u32 qave;
216 __u32 forced;
217 __u32 early;
218 __u32 other;
219 __u32 pdrop;
220 __u8 Wlog; /* log(W) */
221 __u8 Plog; /* log(P_max/(qth_max-qth_min)) */
222 __u8 Scell_log; /* cell size for idle damping */
223 __u8 prio; /* prio of this VQ */
224 __u32 packets;
225 __u32 bytesin;
1da177e4 226};
1e4dfaf9 227
1da177e4
LT
228/* gred setup */
229struct tc_gred_sopt
230{
1e4dfaf9
TG
231 __u32 DPs;
232 __u32 def_DP;
233 __u8 grio;
b38c7eef
TG
234 __u8 flags;
235 __u16 pad1;
1da177e4
LT
236};
237
238/* HTB section */
239#define TC_HTB_NUMPRIO 8
240#define TC_HTB_MAXDEPTH 8
241#define TC_HTB_PROTOVER 3 /* the same as HTB and TC's major */
242
243struct tc_htb_opt
244{
245 struct tc_ratespec rate;
246 struct tc_ratespec ceil;
247 __u32 buffer;
248 __u32 cbuffer;
249 __u32 quantum;
250 __u32 level; /* out only */
251 __u32 prio;
252};
253struct tc_htb_glob
254{
255 __u32 version; /* to match HTB/TC */
256 __u32 rate2quantum; /* bps->quantum divisor */
257 __u32 defcls; /* default class number */
258 __u32 debug; /* debug flags */
259
260 /* stats */
261 __u32 direct_pkts; /* count of non shapped packets */
262};
263enum
264{
265 TCA_HTB_UNSPEC,
266 TCA_HTB_PARMS,
267 TCA_HTB_INIT,
268 TCA_HTB_CTAB,
269 TCA_HTB_RTAB,
270 __TCA_HTB_MAX,
271};
272
273#define TCA_HTB_MAX (__TCA_HTB_MAX - 1)
274
275struct tc_htb_xstats
276{
277 __u32 lends;
278 __u32 borrows;
279 __u32 giants; /* too big packets (rate will not be accurate) */
280 __u32 tokens;
281 __u32 ctokens;
282};
283
284/* HFSC section */
285
286struct tc_hfsc_qopt
287{
288 __u16 defcls; /* default class */
289};
290
291struct tc_service_curve
292{
293 __u32 m1; /* slope of the first segment in bps */
294 __u32 d; /* x-projection of the first segment in us */
295 __u32 m2; /* slope of the second segment in bps */
296};
297
298struct tc_hfsc_stats
299{
300 __u64 work; /* total work done */
301 __u64 rtwork; /* work done by real-time criteria */
302 __u32 period; /* current period */
303 __u32 level; /* class level in hierarchy */
304};
305
306enum
307{
308 TCA_HFSC_UNSPEC,
309 TCA_HFSC_RSC,
310 TCA_HFSC_FSC,
311 TCA_HFSC_USC,
312 __TCA_HFSC_MAX,
313};
314
315#define TCA_HFSC_MAX (__TCA_HFSC_MAX - 1)
316
317
318/* CBQ section */
319
320#define TC_CBQ_MAXPRIO 8
321#define TC_CBQ_MAXLEVEL 8
322#define TC_CBQ_DEF_EWMA 5
323
324struct tc_cbq_lssopt
325{
326 unsigned char change;
327 unsigned char flags;
328#define TCF_CBQ_LSS_BOUNDED 1
329#define TCF_CBQ_LSS_ISOLATED 2
330 unsigned char ewma_log;
331 unsigned char level;
332#define TCF_CBQ_LSS_FLAGS 1
333#define TCF_CBQ_LSS_EWMA 2
334#define TCF_CBQ_LSS_MAXIDLE 4
335#define TCF_CBQ_LSS_MINIDLE 8
336#define TCF_CBQ_LSS_OFFTIME 0x10
337#define TCF_CBQ_LSS_AVPKT 0x20
338 __u32 maxidle;
339 __u32 minidle;
340 __u32 offtime;
341 __u32 avpkt;
342};
343
344struct tc_cbq_wrropt
345{
346 unsigned char flags;
347 unsigned char priority;
348 unsigned char cpriority;
349 unsigned char __reserved;
350 __u32 allot;
351 __u32 weight;
352};
353
354struct tc_cbq_ovl
355{
356 unsigned char strategy;
357#define TC_CBQ_OVL_CLASSIC 0
358#define TC_CBQ_OVL_DELAY 1
359#define TC_CBQ_OVL_LOWPRIO 2
360#define TC_CBQ_OVL_DROP 3
361#define TC_CBQ_OVL_RCLASSIC 4
362 unsigned char priority2;
8a47077a 363 __u16 pad;
1da177e4
LT
364 __u32 penalty;
365};
366
367struct tc_cbq_police
368{
369 unsigned char police;
370 unsigned char __res1;
371 unsigned short __res2;
372};
373
374struct tc_cbq_fopt
375{
376 __u32 split;
377 __u32 defmap;
378 __u32 defchange;
379};
380
381struct tc_cbq_xstats
382{
383 __u32 borrows;
384 __u32 overactions;
385 __s32 avgidle;
386 __s32 undertime;
387};
388
389enum
390{
391 TCA_CBQ_UNSPEC,
392 TCA_CBQ_LSSOPT,
393 TCA_CBQ_WRROPT,
394 TCA_CBQ_FOPT,
395 TCA_CBQ_OVL_STRATEGY,
396 TCA_CBQ_RATE,
397 TCA_CBQ_RTAB,
398 TCA_CBQ_POLICE,
399 __TCA_CBQ_MAX,
400};
401
402#define TCA_CBQ_MAX (__TCA_CBQ_MAX - 1)
403
404/* dsmark section */
405
406enum {
407 TCA_DSMARK_UNSPEC,
408 TCA_DSMARK_INDICES,
409 TCA_DSMARK_DEFAULT_INDEX,
410 TCA_DSMARK_SET_TC_INDEX,
411 TCA_DSMARK_MASK,
412 TCA_DSMARK_VALUE,
413 __TCA_DSMARK_MAX,
414};
415
416#define TCA_DSMARK_MAX (__TCA_DSMARK_MAX - 1)
417
418/* ATM section */
419
420enum {
421 TCA_ATM_UNSPEC,
422 TCA_ATM_FD, /* file/socket descriptor */
423 TCA_ATM_PTR, /* pointer to descriptor - later */
424 TCA_ATM_HDR, /* LL header */
425 TCA_ATM_EXCESS, /* excess traffic class (0 for CLP) */
426 TCA_ATM_ADDR, /* PVC address (for output only) */
427 TCA_ATM_STATE, /* VC state (ATM_VS_*; for output only) */
428 __TCA_ATM_MAX,
429};
430
431#define TCA_ATM_MAX (__TCA_ATM_MAX - 1)
432
433/* Network emulator */
434
435enum
436{
437 TCA_NETEM_UNSPEC,
438 TCA_NETEM_CORR,
439 TCA_NETEM_DELAY_DIST,
0dca51d3 440 TCA_NETEM_REORDER,
c865e5d9 441 TCA_NETEM_CORRUPT,
1da177e4
LT
442 __TCA_NETEM_MAX,
443};
444
445#define TCA_NETEM_MAX (__TCA_NETEM_MAX - 1)
446
447struct tc_netem_qopt
448{
449 __u32 latency; /* added delay (us) */
450 __u32 limit; /* fifo limit (packets) */
451 __u32 loss; /* random packet loss (0=none ~0=100%) */
0dca51d3 452 __u32 gap; /* re-ordering gap (0 for none) */
1da177e4
LT
453 __u32 duplicate; /* random packet dup (0=none ~0=100%) */
454 __u32 jitter; /* random jitter in latency (us) */
455};
456
457struct tc_netem_corr
458{
459 __u32 delay_corr; /* delay correlation */
460 __u32 loss_corr; /* packet loss correlation */
461 __u32 dup_corr; /* duplicate correlation */
462};
463
0dca51d3
SH
464struct tc_netem_reorder
465{
466 __u32 probability;
467 __u32 correlation;
468};
469
c865e5d9
SH
470struct tc_netem_corrupt
471{
472 __u32 probability;
473 __u32 correlation;
474};
475
1da177e4
LT
476#define NETEM_DIST_SCALE 8192
477
478#endif