2 * Copyright (c) 2016-2018 Netflix, Inc.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #ifndef __tcp_hpts_h__
29 #define __tcp_hpts_h__
31 /* Number of useconds in a hpts tick */
32 #define HPTS_TICKS_PER_SLOT 10
33 #define HPTS_MS_TO_SLOTS(x) ((x * 100) + 1)
34 #define HPTS_USEC_TO_SLOTS(x) ((x+9) /10)
35 #define HPTS_USEC_IN_SEC 1000000
36 #define HPTS_MSEC_IN_SEC 1000
37 #define HPTS_USEC_IN_MSEC 1000
40 uint32_t p_hpts_active; /* bbr->flex7 x */
41 uint32_t p_nxt_slot; /* bbr->flex1 x */
42 uint32_t p_cur_slot; /* bbr->flex2 x */
43 uint32_t p_prev_slot; /* bbr->delivered */
44 uint32_t p_runningslot; /* bbr->inflight */
45 uint32_t slot_req; /* bbr->flex3 x */
46 uint32_t inp_hptsslot; /* bbr->flex4 x */
47 uint32_t slot_remaining; /* bbr->flex5 x */
48 uint32_t have_slept; /* bbr->epoch x */
49 uint32_t hpts_sleep_time; /* bbr->applimited x */
50 uint32_t yet_to_sleep; /* bbr->lt_epoch x */
51 uint32_t need_new_to; /* bbr->flex6 x */
52 uint32_t wheel_slot; /* bbr->bw_inuse x */
53 uint32_t maxslots; /* bbr->delRate x */
54 uint32_t wheel_cts; /* bbr->rttProp x */
55 int32_t co_ret; /* bbr->pkts_out x */
56 uint32_t p_curtick; /* upper bbr->cur_del_rate */
57 uint32_t p_lasttick; /* lower bbr->cur_del_rate */
58 uint8_t p_on_min_sleep; /* bbr->flex8 x */
61 /* Magic flags to tell whats cooking on the pacing wheel */
62 #define PACE_TMR_DELACK 0x01 /* Delayed ack timer running */
63 #define PACE_TMR_RACK 0x02 /* RACK timer running */
64 #define PACE_TMR_TLP 0x04 /* TLP timer running */
65 #define PACE_TMR_RXT 0x08 /* Retransmit timer running */
66 #define PACE_TMR_PERSIT 0x10 /* Persists timer running */
67 #define PACE_TMR_KEEP 0x20 /* Keep alive timer running */
68 #define PACE_PKT_OUTPUT 0x40 /* Output Packets being paced */
69 #define PACE_TMR_MASK (PACE_TMR_KEEP|PACE_TMR_PERSIT|PACE_TMR_RXT|PACE_TMR_TLP|PACE_TMR_RACK|PACE_TMR_DELACK)
71 #define DEFAULT_CONNECTION_THESHOLD 100
74 * When using the hpts, a TCP stack must make sure
75 * that once a INP_DROPPED flag is applied to a INP
76 * that it does not expect tcp_output() to ever be
77 * called by the hpts. The hpts will *not* call
78 * any output (or input) functions on a TCB that
79 * is in the DROPPED state.
81 * This implies final ACK's and RST's that might
82 * be sent when a TCB is still around must be
83 * sent from a routine like tcp_respond().
85 #define LOWEST_SLEEP_ALLOWED 50
86 #define DEFAULT_MIN_SLEEP 250 /* How many usec's is default for hpts sleep
87 * this determines min granularity of the
88 * hpts. If 1, granularity is 10useconds at
89 * the cost of more CPU (context switching).
90 * Note do not set this to 0.
92 #define DYNAMIC_MIN_SLEEP DEFAULT_MIN_SLEEP
93 #define DYNAMIC_MAX_SLEEP 100000 /* 100ms */
94 /* No of connections when wee start aligning to the cpu from syscalls */
95 #define OLDEST_THRESHOLD 1200
96 /* Thresholds for raising/lowering sleep */
97 #define TICKS_INDICATE_MORE_SLEEP 100 /* This would be 1ms */
98 #define TICKS_INDICATE_LESS_SLEEP 1000 /* This would indicate 10ms */
101 * Dynamic adjustment of sleeping times is done in "new" mode
102 * where we are depending on syscall returns and lro returns
103 * to push hpts forward mainly and the timer is only a backstop.
105 * When we are in the "new" mode i.e. conn_cnt > conn_cnt_thresh
106 * then we do a dynamic adjustment on the time we sleep.
107 * Our threshold is if the lateness of the first client served (in ticks) is
108 * greater than or equal too ticks_indicate_more_sleep (10ms
109 * or 10000 ticks). If we were that late, the actual sleep time
110 * is adjusted down by 50%. If the ticks_ran is less than
111 * ticks_indicate_more_sleep (100 ticks or 1000usecs).
117 void tcp_hpts_remove(struct inpcb *);
118 bool tcp_in_hpts(struct inpcb *);
121 * To insert a TCB on the hpts you *must* be holding the
122 * INP_WLOCK(). The hpts insert code will then acqurire
123 * the hpts's lock and insert the TCB on the requested
124 * slot possibly waking up the hpts if you are requesting
125 * a time earlier than what the hpts is sleeping to (if
126 * the hpts is sleeping). You may check the inp->inp_in_hpts
127 * flag without the hpts lock. The hpts is the only one
128 * that will clear this flag holding only the hpts lock. This
129 * means that in your tcp_output() routine when you test for
130 * it to be 1 (so you wont call output) it may be transitioning
131 * to 0 (by the hpts). That will be fine since that will just
132 * mean an extra call to tcp_output that most likely will find
133 * the call you executed (when the mis-match occured) will have
134 * put the TCB back on the hpts and it will return. If your
135 * call did not add it back to the hpts then you will either
136 * over-send or the cwnd will block you from sending more.
138 * Note you should also be holding the INP_WLOCK() when you
139 * call the remove from the hpts as well. Thoug usually
140 * you are either doing this from a timer, where you need
141 * that INP_WLOCK() or from destroying your TCB where again
142 * you should already have the INP_WLOCK().
144 uint32_t tcp_hpts_insert_diag(struct inpcb *inp, uint32_t slot, int32_t line,
145 struct hpts_diag *diag);
146 #define tcp_hpts_insert(inp, slot) \
147 tcp_hpts_insert_diag((inp), (slot), __LINE__, NULL)
149 void __tcp_set_hpts(struct inpcb *inp, int32_t line);
150 #define tcp_set_hpts(a) __tcp_set_hpts(a, __LINE__)
152 void tcp_set_inp_to_drop(struct inpcb *inp, uint16_t reason);
154 void tcp_run_hpts(void);
156 uint16_t hpts_random_cpu(struct inpcb *inp);
158 extern int32_t tcp_min_hptsi_time;
163 * The following functions should also be available
164 * to userspace as well.
166 static __inline uint32_t
167 tcp_tv_to_hptstick(const struct timeval *sv)
169 return ((sv->tv_sec * 100000) + (sv->tv_usec / HPTS_TICKS_PER_SLOT));
172 static __inline uint32_t
173 tcp_tv_to_usectick(const struct timeval *sv)
175 return ((uint32_t) ((sv->tv_sec * HPTS_USEC_IN_SEC) + sv->tv_usec));
178 static __inline uint32_t
179 tcp_tv_to_mssectick(const struct timeval *sv)
181 return ((uint32_t) ((sv->tv_sec * HPTS_MSEC_IN_SEC) + (sv->tv_usec/HPTS_USEC_IN_MSEC)));
184 static __inline uint64_t
185 tcp_tv_to_lusectick(const struct timeval *sv)
187 return ((uint64_t)((sv->tv_sec * HPTS_USEC_IN_SEC) + sv->tv_usec));
191 static __inline uint32_t
192 tcp_gethptstick(struct timeval *sv)
199 return (tcp_tv_to_hptstick(sv));
202 static __inline uint32_t
203 tcp_get_usecs(struct timeval *tv)
210 return (tcp_tv_to_usectick(tv));
214 #endif /* __tcp_hpts_h__ */