/*- * Copyright (c) 2016-2018 Netflix, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_rss.h" #include "opt_tcpdebug.h" /** * Some notes about usage. * * The tcp_hpts system is designed to provide a high precision timer * system for tcp. Its main purpose is to provide a mechanism for * pacing packets out onto the wire. It can be used in two ways * by a given TCP stack (and those two methods can be used simultaneously). * * First, and probably the main thing its used by Rack and BBR, it can * be used to call tcp_output() of a transport stack at some time in the future. * The normal way this is done is that tcp_output() of the stack schedules * itself to be called again by calling tcp_hpts_insert(tcpcb, slot). The * slot is the time from now that the stack wants to be called but it * must be converted to tcp_hpts's notion of slot. This is done with * one of the macros HPTS_MS_TO_SLOTS or HPTS_USEC_TO_SLOTS. So a typical * call from the tcp_output() routine might look like: * * tcp_hpts_insert(tp, HPTS_USEC_TO_SLOTS(550)); * * The above would schedule tcp_ouput() to be called in 550 useconds. * Note that if using this mechanism the stack will want to add near * its top a check to prevent unwanted calls (from user land or the * arrival of incoming ack's). So it would add something like: * * if (inp->inp_in_hpts) * return; * * to prevent output processing until the time alotted has gone by. * Of course this is a bare bones example and the stack will probably * have more consideration then just the above. * * Now the second function (actually two functions I guess :D) * the tcp_hpts system provides is the ability to either abort * a connection (later) or process input on a connection. * Why would you want to do this? To keep processor locality * and or not have to worry about untangling any recursive * locks. The input function now is hooked to the new LRO * system as well. * * In order to use the input redirection function the * tcp stack must define an input function for * tfb_do_queued_segments(). This function understands * how to dequeue a array of packets that were input and * knows how to call the correct processing routine. * * Locking in this is important as well so most likely the * stack will need to define the tfb_do_segment_nounlock() * splitting tfb_do_segment() into two parts. The main processing * part that does not unlock the INP and returns a value of 1 or 0. * It returns 0 if all is well and the lock was not released. It * returns 1 if we had to destroy the TCB (a reset received etc). * The remains of tfb_do_segment() then become just a simple call * to the tfb_do_segment_nounlock() function and check the return * code and possibly unlock. * * The stack must also set the flag on the INP that it supports this * feature i.e. INP_SUPPORTS_MBUFQ. The LRO code recoginizes * this flag as well and will queue packets when it is set. * There are other flags as well INP_MBUF_QUEUE_READY and * INP_DONT_SACK_QUEUE. The first flag tells the LRO code * that we are in the pacer for output so there is no * need to wake up the hpts system to get immediate * input. The second tells the LRO code that its okay * if a SACK arrives you can still defer input and let * the current hpts timer run (this is usually set when * a rack timer is up so we know SACK's are happening * on the connection already and don't want to wakeup yet). * * There is a common functions within the rack_bbr_common code * version i.e. ctf_do_queued_segments(). This function * knows how to take the input queue of packets from * tp->t_in_pkts and process them digging out * all the arguments, calling any bpf tap and * calling into tfb_do_segment_nounlock(). The common * function (ctf_do_queued_segments()) requires that * you have defined the tfb_do_segment_nounlock() as * described above. * * The second feature of the input side of hpts is the * dropping of a connection. This is due to the way that * locking may have occured on the INP_WLOCK. So if * a stack wants to drop a connection it calls: * * tcp_set_inp_to_drop(tp, ETIMEDOUT) * * To schedule the tcp_hpts system to call * * tcp_drop(tp, drop_reason) * * at a future point. This is quite handy to prevent locking * issues when dropping connections. * */ #include #include #include #include #include #include #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RSS #include #include #endif #define TCPSTATES /* for logging */ #include #include #include #include #include /* required for icmp_var.h */ #include /* for ICMP_BANDLIM */ #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef tcpdebug #include #endif /* tcpdebug */ #ifdef tcp_offload #include #endif MALLOC_DEFINE(M_TCPHPTS, "tcp_hpts", "TCP hpts"); #ifdef RSS static int tcp_bind_threads = 1; #else static int tcp_bind_threads = 2; #endif TUNABLE_INT("net.inet.tcp.bind_hptss", &tcp_bind_threads); static struct tcp_hptsi tcp_pace; static int hpts_does_tp_logging = 0; static void tcp_wakehpts(struct tcp_hpts_entry *p); static void tcp_wakeinput(struct tcp_hpts_entry *p); static void tcp_input_data(struct tcp_hpts_entry *hpts, struct timeval *tv); static void tcp_hptsi(struct tcp_hpts_entry *hpts); static void tcp_hpts_thread(void *ctx); static void tcp_init_hptsi(void *st); int32_t tcp_min_hptsi_time = DEFAULT_MIN_SLEEP; static int32_t tcp_hpts_callout_skip_swi = 0; SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hpts, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "TCP Hpts controls"); #define timersub(tvp, uvp, vvp) \ do { \ (vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec; \ (vvp)->tv_usec = (tvp)->tv_usec - (uvp)->tv_usec; \ if ((vvp)->tv_usec < 0) { \ (vvp)->tv_sec--; \ (vvp)->tv_usec += 1000000; \ } \ } while (0) static int32_t tcp_hpts_precision = 120; struct hpts_domain_info { int count; int cpu[MAXCPU]; }; struct hpts_domain_info hpts_domains[MAXMEMDOM]; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, precision, CTLFLAG_RW, &tcp_hpts_precision, 120, "Value for PRE() precision of callout"); counter_u64_t hpts_hopelessly_behind; SYSCTL_COUNTER_U64(_net_inet_tcp_hpts, OID_AUTO, hopeless, CTLFLAG_RD, &hpts_hopelessly_behind, "Number of times hpts could not catch up and was behind hopelessly"); counter_u64_t hpts_loops; SYSCTL_COUNTER_U64(_net_inet_tcp_hpts, OID_AUTO, loops, CTLFLAG_RD, &hpts_loops, "Number of times hpts had to loop to catch up"); counter_u64_t back_tosleep; SYSCTL_COUNTER_U64(_net_inet_tcp_hpts, OID_AUTO, no_tcbsfound, CTLFLAG_RD, &back_tosleep, "Number of times hpts found no tcbs"); counter_u64_t combined_wheel_wrap; SYSCTL_COUNTER_U64(_net_inet_tcp_hpts, OID_AUTO, comb_wheel_wrap, CTLFLAG_RD, &combined_wheel_wrap, "Number of times the wheel lagged enough to have an insert see wrap"); counter_u64_t wheel_wrap; SYSCTL_COUNTER_U64(_net_inet_tcp_hpts, OID_AUTO, wheel_wrap, CTLFLAG_RD, &wheel_wrap, "Number of times the wheel lagged enough to have an insert see wrap"); static int32_t out_ts_percision = 0; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, out_tspercision, CTLFLAG_RW, &out_ts_percision, 0, "Do we use a percise timestamp for every output cts"); SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, logging, CTLFLAG_RW, &hpts_does_tp_logging, 0, "Do we add to any tp that has logging on pacer logs"); static int32_t max_pacer_loops = 10; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, loopmax, CTLFLAG_RW, &max_pacer_loops, 10, "What is the maximum number of times the pacer will loop trying to catch up"); #define HPTS_MAX_SLEEP_ALLOWED (NUM_OF_HPTSI_SLOTS/2) static uint32_t hpts_sleep_max = HPTS_MAX_SLEEP_ALLOWED; static int sysctl_net_inet_tcp_hpts_max_sleep(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = hpts_sleep_max; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if ((new < (NUM_OF_HPTSI_SLOTS / 4)) || (new > HPTS_MAX_SLEEP_ALLOWED)) error = EINVAL; else hpts_sleep_max = new; } return (error); } SYSCTL_PROC(_net_inet_tcp_hpts, OID_AUTO, maxsleep, CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &hpts_sleep_max, 0, &sysctl_net_inet_tcp_hpts_max_sleep, "IU", "Maximum time hpts will sleep"); SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, minsleep, CTLFLAG_RW, &tcp_min_hptsi_time, 0, "The minimum time the hpts must sleep before processing more slots"); SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, skip_swi, CTLFLAG_RW, &tcp_hpts_callout_skip_swi, 0, "Do we have the callout call directly to the hpts?"); static void tcp_hpts_log(struct tcp_hpts_entry *hpts, struct tcpcb *tp, struct timeval *tv, int ticks_to_run, int idx) { union tcp_log_stackspecific log; memset(&log.u_bbr, 0, sizeof(log.u_bbr)); log.u_bbr.flex1 = hpts->p_nxt_slot; log.u_bbr.flex2 = hpts->p_cur_slot; log.u_bbr.flex3 = hpts->p_prev_slot; log.u_bbr.flex4 = idx; log.u_bbr.flex5 = hpts->p_curtick; log.u_bbr.flex6 = hpts->p_on_queue_cnt; log.u_bbr.use_lt_bw = 1; log.u_bbr.inflight = ticks_to_run; log.u_bbr.applimited = hpts->overidden_sleep; log.u_bbr.delivered = hpts->saved_curtick; log.u_bbr.timeStamp = tcp_tv_to_usectick(tv); log.u_bbr.epoch = hpts->saved_curslot; log.u_bbr.lt_epoch = hpts->saved_prev_slot; log.u_bbr.pkts_out = hpts->p_delayed_by; log.u_bbr.lost = hpts->p_hpts_sleep_time; log.u_bbr.cur_del_rate = hpts->p_runningtick; TCP_LOG_EVENTP(tp, NULL, &tp->t_inpcb->inp_socket->so_rcv, &tp->t_inpcb->inp_socket->so_snd, BBR_LOG_HPTSDIAG, 0, 0, &log, false, tv); } static void hpts_timeout_swi(void *arg) { struct tcp_hpts_entry *hpts; hpts = (struct tcp_hpts_entry *)arg; swi_sched(hpts->ie_cookie, 0); } static void hpts_timeout_dir(void *arg) { tcp_hpts_thread(arg); } static inline void hpts_sane_pace_remove(struct tcp_hpts_entry *hpts, struct inpcb *inp, struct hptsh *head, int clear) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_hpts_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if (inp->inp_in_hpts == 0) { /* We are not on the hpts? */ panic("%s: hpts:%p inp:%p not on the hpts?", __FUNCTION__, hpts, inp); } #endif TAILQ_REMOVE(head, inp, inp_hpts); hpts->p_on_queue_cnt--; if (hpts->p_on_queue_cnt < 0) { /* Count should not go negative .. */ #ifdef INVARIANTS panic("Hpts goes negative inp:%p hpts:%p", inp, hpts); #endif hpts->p_on_queue_cnt = 0; } if (clear) { inp->inp_hpts_request = 0; inp->inp_in_hpts = 0; } } static inline void hpts_sane_pace_insert(struct tcp_hpts_entry *hpts, struct inpcb *inp, struct hptsh *head, int line, int noref) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_hpts_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if ((noref == 0) && (inp->inp_in_hpts == 1)) { /* We are already on the hpts? */ panic("%s: hpts:%p inp:%p already on the hpts?", __FUNCTION__, hpts, inp); } #endif TAILQ_INSERT_TAIL(head, inp, inp_hpts); inp->inp_in_hpts = 1; hpts->p_on_queue_cnt++; if (noref == 0) { in_pcbref(inp); } } static inline void hpts_sane_input_remove(struct tcp_hpts_entry *hpts, struct inpcb *inp, int clear) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_input_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if (inp->inp_in_input == 0) { /* We are not on the input hpts? */ panic("%s: hpts:%p inp:%p not on the input hpts?", __FUNCTION__, hpts, inp); } #endif TAILQ_REMOVE(&hpts->p_input, inp, inp_input); hpts->p_on_inqueue_cnt--; if (hpts->p_on_inqueue_cnt < 0) { #ifdef INVARIANTS panic("Hpts in goes negative inp:%p hpts:%p", inp, hpts); #endif hpts->p_on_inqueue_cnt = 0; } #ifdef INVARIANTS if (TAILQ_EMPTY(&hpts->p_input) && (hpts->p_on_inqueue_cnt != 0)) { /* We should not be empty with a queue count */ panic("%s hpts:%p in_hpts input empty but cnt:%d", __FUNCTION__, hpts, hpts->p_on_inqueue_cnt); } #endif if (clear) inp->inp_in_input = 0; } static inline void hpts_sane_input_insert(struct tcp_hpts_entry *hpts, struct inpcb *inp, int line) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_input_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if (inp->inp_in_input == 1) { /* We are already on the input hpts? */ panic("%s: hpts:%p inp:%p already on the input hpts?", __FUNCTION__, hpts, inp); } #endif TAILQ_INSERT_TAIL(&hpts->p_input, inp, inp_input); inp->inp_in_input = 1; hpts->p_on_inqueue_cnt++; in_pcbref(inp); } static void tcp_wakehpts(struct tcp_hpts_entry *hpts) { HPTS_MTX_ASSERT(hpts); if (hpts->p_hpts_wake_scheduled == 0) { hpts->p_hpts_wake_scheduled = 1; swi_sched(hpts->ie_cookie, 0); } } static void tcp_wakeinput(struct tcp_hpts_entry *hpts) { HPTS_MTX_ASSERT(hpts); if (hpts->p_hpts_wake_scheduled == 0) { hpts->p_hpts_wake_scheduled = 1; swi_sched(hpts->ie_cookie, 0); } } struct tcp_hpts_entry * tcp_cur_hpts(struct inpcb *inp) { int32_t hpts_num; struct tcp_hpts_entry *hpts; hpts_num = inp->inp_hpts_cpu; hpts = tcp_pace.rp_ent[hpts_num]; return (hpts); } struct tcp_hpts_entry * tcp_hpts_lock(struct inpcb *inp) { struct tcp_hpts_entry *hpts; int32_t hpts_num; again: hpts_num = inp->inp_hpts_cpu; hpts = tcp_pace.rp_ent[hpts_num]; #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif mtx_lock(&hpts->p_mtx); if (hpts_num != inp->inp_hpts_cpu) { mtx_unlock(&hpts->p_mtx); goto again; } return (hpts); } struct tcp_hpts_entry * tcp_input_lock(struct inpcb *inp) { struct tcp_hpts_entry *hpts; int32_t hpts_num; again: hpts_num = inp->inp_input_cpu; hpts = tcp_pace.rp_ent[hpts_num]; #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif mtx_lock(&hpts->p_mtx); if (hpts_num != inp->inp_input_cpu) { mtx_unlock(&hpts->p_mtx); goto again; } return (hpts); } static void tcp_remove_hpts_ref(struct inpcb *inp, struct tcp_hpts_entry *hpts, int line) { int32_t add_freed; if (inp->inp_flags2 & INP_FREED) { /* * Need to play a special trick so that in_pcbrele_wlocked * does not return 1 when it really should have returned 0. */ add_freed = 1; inp->inp_flags2 &= ~INP_FREED; } else { add_freed = 0; } #ifndef INP_REF_DEBUG if (in_pcbrele_wlocked(inp)) { /* * This should not happen. We have the inpcb referred to by * the main socket (why we are called) and the hpts. It * should always return 0. */ panic("inpcb:%p release ret 1", inp); } #else if (__in_pcbrele_wlocked(inp, line)) { /* * This should not happen. We have the inpcb referred to by * the main socket (why we are called) and the hpts. It * should always return 0. */ panic("inpcb:%p release ret 1", inp); } #endif if (add_freed) { inp->inp_flags2 |= INP_FREED; } } static void tcp_hpts_remove_locked_output(struct tcp_hpts_entry *hpts, struct inpcb *inp, int32_t flags, int32_t line) { if (inp->inp_in_hpts) { hpts_sane_pace_remove(hpts, inp, &hpts->p_hptss[inp->inp_hptsslot], 1); tcp_remove_hpts_ref(inp, hpts, line); } } static void tcp_hpts_remove_locked_input(struct tcp_hpts_entry *hpts, struct inpcb *inp, int32_t flags, int32_t line) { HPTS_MTX_ASSERT(hpts); if (inp->inp_in_input) { hpts_sane_input_remove(hpts, inp, 1); tcp_remove_hpts_ref(inp, hpts, line); } } /* * Called normally with the INP_LOCKED but it * does not matter, the hpts lock is the key * but the lock order allows us to hold the * INP lock and then get the hpts lock. * * Valid values in the flags are * HPTS_REMOVE_OUTPUT - remove from the output of the hpts. * HPTS_REMOVE_INPUT - remove from the input of the hpts. * Note that you can use one or both values together * and get two actions. */ void __tcp_hpts_remove(struct inpcb *inp, int32_t flags, int32_t line) { struct tcp_hpts_entry *hpts; INP_WLOCK_ASSERT(inp); if (flags & HPTS_REMOVE_OUTPUT) { hpts = tcp_hpts_lock(inp); tcp_hpts_remove_locked_output(hpts, inp, flags, line); mtx_unlock(&hpts->p_mtx); } if (flags & HPTS_REMOVE_INPUT) { hpts = tcp_input_lock(inp); tcp_hpts_remove_locked_input(hpts, inp, flags, line); mtx_unlock(&hpts->p_mtx); } } static inline int hpts_tick(uint32_t wheel_tick, uint32_t plus) { /* * Given a slot on the wheel, what slot * is that plus ticks out? */ KASSERT(wheel_tick < NUM_OF_HPTSI_SLOTS, ("Invalid tick %u not on wheel", wheel_tick)); return ((wheel_tick + plus) % NUM_OF_HPTSI_SLOTS); } static inline int tick_to_wheel(uint32_t cts_in_wticks) { /* * Given a timestamp in wheel ticks (10usec inc's) * map it to our limited space wheel. */ return (cts_in_wticks % NUM_OF_HPTSI_SLOTS); } static inline int hpts_ticks_diff(int prev_tick, int tick_now) { /* * Given two ticks that are someplace * on our wheel. How far are they apart? */ if (tick_now > prev_tick) return (tick_now - prev_tick); else if (tick_now == prev_tick) /* * Special case, same means we can go all of our * wheel less one slot. */ return (NUM_OF_HPTSI_SLOTS - 1); else return ((NUM_OF_HPTSI_SLOTS - prev_tick) + tick_now); } /* * Given a tick on the wheel that is the current time * mapped to the wheel (wheel_tick), what is the maximum * distance forward that can be obtained without * wrapping past either prev_tick or running_tick * depending on the htps state? Also if passed * a uint32_t *, fill it with the tick location. * * Note if you do not give this function the current * time (that you think it is) mapped to the wheel * then the results will not be what you expect and * could lead to invalid inserts. */ static inline int32_t max_ticks_available(struct tcp_hpts_entry *hpts, uint32_t wheel_tick, uint32_t *target_tick) { uint32_t dis_to_travel, end_tick, pacer_to_now, avail_on_wheel; if ((hpts->p_hpts_active == 1) && (hpts->p_wheel_complete == 0)) { end_tick = hpts->p_runningtick; /* Back up one tick */ if (end_tick == 0) end_tick = NUM_OF_HPTSI_SLOTS - 1; else end_tick--; if (target_tick) *target_tick = end_tick; } else { /* * For the case where we are * not active, or we have * completed the pass over * the wheel, we can use the * prev tick and subtract one from it. This puts us * as far out as possible on the wheel. */ end_tick = hpts->p_prev_slot; if (end_tick == 0) end_tick = NUM_OF_HPTSI_SLOTS - 1; else end_tick--; if (target_tick) *target_tick = end_tick; /* * Now we have close to the full wheel left minus the * time it has been since the pacer went to sleep. Note * that wheel_tick, passed in, should be the current time * from the perspective of the caller, mapped to the wheel. */ if (hpts->p_prev_slot != wheel_tick) dis_to_travel = hpts_ticks_diff(hpts->p_prev_slot, wheel_tick); else dis_to_travel = 1; /* * dis_to_travel in this case is the space from when the * pacer stopped (p_prev_slot) and where our wheel_tick * is now. To know how many slots we can put it in we * subtract from the wheel size. We would not want * to place something after p_prev_slot or it will * get ran too soon. */ return (NUM_OF_HPTSI_SLOTS - dis_to_travel); } /* * So how many slots are open between p_runningtick -> p_cur_slot * that is what is currently un-available for insertion. Special * case when we are at the last slot, this gets 1, so that * the answer to how many slots are available is all but 1. */ if (hpts->p_runningtick == hpts->p_cur_slot) dis_to_travel = 1; else dis_to_travel = hpts_ticks_diff(hpts->p_runningtick, hpts->p_cur_slot); /* * How long has the pacer been running? */ if (hpts->p_cur_slot != wheel_tick) { /* The pacer is a bit late */ pacer_to_now = hpts_ticks_diff(hpts->p_cur_slot, wheel_tick); } else { /* The pacer is right on time, now == pacers start time */ pacer_to_now = 0; } /* * To get the number left we can insert into we simply * subract the distance the pacer has to run from how * many slots there are. */ avail_on_wheel = NUM_OF_HPTSI_SLOTS - dis_to_travel; /* * Now how many of those we will eat due to the pacer's * time (p_cur_slot) of start being behind the * real time (wheel_tick)? */ if (avail_on_wheel <= pacer_to_now) { /* * Wheel wrap, we can't fit on the wheel, that * is unusual the system must be way overloaded! * Insert into the assured tick, and return special * "0". */ counter_u64_add(combined_wheel_wrap, 1); *target_tick = hpts->p_nxt_slot; return (0); } else { /* * We know how many slots are open * on the wheel (the reverse of what * is left to run. Take away the time * the pacer started to now (wheel_tick) * and that tells you how many slots are * open that can be inserted into that won't * be touched by the pacer until later. */ return (avail_on_wheel - pacer_to_now); } } static int tcp_queue_to_hpts_immediate_locked(struct inpcb *inp, struct tcp_hpts_entry *hpts, int32_t line, int32_t noref) { uint32_t need_wake = 0; HPTS_MTX_ASSERT(hpts); if (inp->inp_in_hpts == 0) { /* Ok we need to set it on the hpts in the current slot */ inp->inp_hpts_request = 0; if ((hpts->p_hpts_active == 0) || (hpts->p_wheel_complete)) { /* * A sleeping hpts we want in next slot to run * note that in this state p_prev_slot == p_cur_slot */ inp->inp_hptsslot = hpts_tick(hpts->p_prev_slot, 1); if ((hpts->p_on_min_sleep == 0) && (hpts->p_hpts_active == 0)) need_wake = 1; } else if ((void *)inp == hpts->p_inp) { /* * The hpts system is running and the caller * was awoken by the hpts system. * We can't allow you to go into the same slot we * are in (we don't want a loop :-D). */ inp->inp_hptsslot = hpts->p_nxt_slot; } else inp->inp_hptsslot = hpts->p_runningtick; hpts_sane_pace_insert(hpts, inp, &hpts->p_hptss[inp->inp_hptsslot], line, noref); if (need_wake) { /* * Activate the hpts if it is sleeping and its * timeout is not 1. */ hpts->p_direct_wake = 1; tcp_wakehpts(hpts); } } return (need_wake); } int __tcp_queue_to_hpts_immediate(struct inpcb *inp, int32_t line) { int32_t ret; struct tcp_hpts_entry *hpts; INP_WLOCK_ASSERT(inp); hpts = tcp_hpts_lock(inp); ret = tcp_queue_to_hpts_immediate_locked(inp, hpts, line, 0); mtx_unlock(&hpts->p_mtx); return (ret); } #ifdef INVARIANTS static void check_if_slot_would_be_wrong(struct tcp_hpts_entry *hpts, struct inpcb *inp, uint32_t inp_hptsslot, int line) { /* * Sanity checks for the pacer with invariants * on insert. */ if (inp_hptsslot >= NUM_OF_HPTSI_SLOTS) panic("hpts:%p inp:%p slot:%d > max", hpts, inp, inp_hptsslot); if ((hpts->p_hpts_active) && (hpts->p_wheel_complete == 0)) { /* * If the pacer is processing a arc * of the wheel, we need to make * sure we are not inserting within * that arc. */ int distance, yet_to_run; distance = hpts_ticks_diff(hpts->p_runningtick, inp_hptsslot); if (hpts->p_runningtick != hpts->p_cur_slot) yet_to_run = hpts_ticks_diff(hpts->p_runningtick, hpts->p_cur_slot); else yet_to_run = 0; /* processing last slot */ if (yet_to_run > distance) { panic("hpts:%p inp:%p slot:%d distance:%d yet_to_run:%d rs:%d cs:%d", hpts, inp, inp_hptsslot, distance, yet_to_run, hpts->p_runningtick, hpts->p_cur_slot); } } } #endif static void tcp_hpts_insert_locked(struct tcp_hpts_entry *hpts, struct inpcb *inp, uint32_t slot, int32_t line, struct hpts_diag *diag, struct timeval *tv) { uint32_t need_new_to = 0; uint32_t wheel_cts, last_tick; int32_t wheel_tick, maxticks; int8_t need_wakeup = 0; HPTS_MTX_ASSERT(hpts); if (diag) { memset(diag, 0, sizeof(struct hpts_diag)); diag->p_hpts_active = hpts->p_hpts_active; diag->p_prev_slot = hpts->p_prev_slot; diag->p_runningtick = hpts->p_runningtick; diag->p_nxt_slot = hpts->p_nxt_slot; diag->p_cur_slot = hpts->p_cur_slot; diag->p_curtick = hpts->p_curtick; diag->p_lasttick = hpts->p_lasttick; diag->slot_req = slot; diag->p_on_min_sleep = hpts->p_on_min_sleep; diag->hpts_sleep_time = hpts->p_hpts_sleep_time; } if (inp->inp_in_hpts == 0) { if (slot == 0) { /* Immediate */ tcp_queue_to_hpts_immediate_locked(inp, hpts, line, 0); return; } /* Get the current time relative to the wheel */ wheel_cts = tcp_tv_to_hptstick(tv); /* Map it onto the wheel */ wheel_tick = tick_to_wheel(wheel_cts); /* Now what's the max we can place it at? */ maxticks = max_ticks_available(hpts, wheel_tick, &last_tick); if (diag) { diag->wheel_tick = wheel_tick; diag->maxticks = maxticks; diag->wheel_cts = wheel_cts; } if (maxticks == 0) { /* The pacer is in a wheel wrap behind, yikes! */ if (slot > 1) { /* * Reduce by 1 to prevent a forever loop in * case something else is wrong. Note this * probably does not hurt because the pacer * if its true is so far behind we will be * > 1second late calling anyway. */ slot--; } inp->inp_hptsslot = last_tick; inp->inp_hpts_request = slot; } else if (maxticks >= slot) { /* It all fits on the wheel */ inp->inp_hpts_request = 0; inp->inp_hptsslot = hpts_tick(wheel_tick, slot); } else { /* It does not fit */ inp->inp_hpts_request = slot - maxticks; inp->inp_hptsslot = last_tick; } if (diag) { diag->slot_remaining = inp->inp_hpts_request; diag->inp_hptsslot = inp->inp_hptsslot; } #ifdef INVARIANTS check_if_slot_would_be_wrong(hpts, inp, inp->inp_hptsslot, line); #endif hpts_sane_pace_insert(hpts, inp, &hpts->p_hptss[inp->inp_hptsslot], line, 0); if ((hpts->p_hpts_active == 0) && (inp->inp_hpts_request == 0) && (hpts->p_on_min_sleep == 0)) { /* * The hpts is sleeping and not on a minimum * sleep time, we need to figure out where * it will wake up at and if we need to reschedule * its time-out. */ uint32_t have_slept, yet_to_sleep; /* Now do we need to restart the hpts's timer? */ have_slept = hpts_ticks_diff(hpts->p_prev_slot, wheel_tick); if (have_slept < hpts->p_hpts_sleep_time) yet_to_sleep = hpts->p_hpts_sleep_time - have_slept; else { /* We are over-due */ yet_to_sleep = 0; need_wakeup = 1; } if (diag) { diag->have_slept = have_slept; diag->yet_to_sleep = yet_to_sleep; } if (yet_to_sleep && (yet_to_sleep > slot)) { /* * We need to reschedule the hpts's time-out. */ hpts->p_hpts_sleep_time = slot; need_new_to = slot * HPTS_TICKS_PER_USEC; } } /* * Now how far is the hpts sleeping to? if active is 1, its * up and ticking we do nothing, otherwise we may need to * reschedule its callout if need_new_to is set from above. */ if (need_wakeup) { hpts->p_direct_wake = 1; tcp_wakehpts(hpts); if (diag) { diag->need_new_to = 0; diag->co_ret = 0xffff0000; } } else if (need_new_to) { int32_t co_ret; struct timeval tv; sbintime_t sb; tv.tv_sec = 0; tv.tv_usec = 0; while (need_new_to > HPTS_USEC_IN_SEC) { tv.tv_sec++; need_new_to -= HPTS_USEC_IN_SEC; } tv.tv_usec = need_new_to; sb = tvtosbt(tv); if (tcp_hpts_callout_skip_swi == 0) { co_ret = callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_swi, hpts, hpts->p_cpu, (C_DIRECT_EXEC | C_PREL(tcp_hpts_precision))); } else { co_ret = callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_dir, hpts, hpts->p_cpu, C_PREL(tcp_hpts_precision)); } if (diag) { diag->need_new_to = need_new_to; diag->co_ret = co_ret; } } } else { #ifdef INVARIANTS panic("Hpts:%p tp:%p already on hpts and add?", hpts, inp); #endif } } uint32_t tcp_hpts_insert_diag(struct inpcb *inp, uint32_t slot, int32_t line, struct hpts_diag *diag) { struct tcp_hpts_entry *hpts; uint32_t slot_on; struct timeval tv; /* * We now return the next-slot the hpts will be on, beyond its * current run (if up) or where it was when it stopped if it is * sleeping. */ INP_WLOCK_ASSERT(inp); hpts = tcp_hpts_lock(inp); microuptime(&tv); tcp_hpts_insert_locked(hpts, inp, slot, line, diag, &tv); slot_on = hpts->p_nxt_slot; mtx_unlock(&hpts->p_mtx); return (slot_on); } uint32_t __tcp_hpts_insert(struct inpcb *inp, uint32_t slot, int32_t line){ return (tcp_hpts_insert_diag(inp, slot, line, NULL)); } int __tcp_queue_to_input_locked(struct inpcb *inp, struct tcp_hpts_entry *hpts, int32_t line) { int32_t retval = 0; HPTS_MTX_ASSERT(hpts); if (inp->inp_in_input == 0) { /* Ok we need to set it on the hpts in the current slot */ hpts_sane_input_insert(hpts, inp, line); retval = 1; if (hpts->p_hpts_active == 0) { /* * Activate the hpts if it is sleeping. */ retval = 2; hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } } else if (hpts->p_hpts_active == 0) { retval = 4; hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } return (retval); } int32_t __tcp_queue_to_input(struct inpcb *inp, int line) { struct tcp_hpts_entry *hpts; int32_t ret; hpts = tcp_input_lock(inp); ret = __tcp_queue_to_input_locked(inp, hpts, line); mtx_unlock(&hpts->p_mtx); return (ret); } void __tcp_set_inp_to_drop(struct inpcb *inp, uint16_t reason, int32_t line) { struct tcp_hpts_entry *hpts; struct tcpcb *tp; tp = intotcpcb(inp); hpts = tcp_input_lock(tp->t_inpcb); if (inp->inp_in_input == 0) { /* Ok we need to set it on the hpts in the current slot */ hpts_sane_input_insert(hpts, inp, line); if (hpts->p_hpts_active == 0) { /* * Activate the hpts if it is sleeping. */ hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } } else if (hpts->p_hpts_active == 0) { hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } inp->inp_hpts_drop_reas = reason; mtx_unlock(&hpts->p_mtx); } static uint16_t hpts_random_cpu(struct inpcb *inp){ /* * No flow type set distribute the load randomly. */ uint16_t cpuid; uint32_t ran; /* * If one has been set use it i.e. we want both in and out on the * same hpts. */ if (inp->inp_input_cpu_set) { return (inp->inp_input_cpu); } else if (inp->inp_hpts_cpu_set) { return (inp->inp_hpts_cpu); } /* Nothing set use a random number */ ran = arc4random(); cpuid = (ran & 0xffff) % mp_ncpus; return (cpuid); } static uint16_t hpts_cpuid(struct inpcb *inp) { u_int cpuid; #if !defined(RSS) && defined(NUMA) struct hpts_domain_info *di; #endif /* * If one has been set use it i.e. we want both in and out on the * same hpts. */ if (inp->inp_input_cpu_set) { return (inp->inp_input_cpu); } else if (inp->inp_hpts_cpu_set) { return (inp->inp_hpts_cpu); } /* If one is set the other must be the same */ #ifdef RSS cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype); if (cpuid == NETISR_CPUID_NONE) return (hpts_random_cpu(inp)); else return (cpuid); #else /* * We don't have a flowid -> cpuid mapping, so cheat and just map * unknown cpuids to curcpu. Not the best, but apparently better * than defaulting to swi 0. */ if (inp->inp_flowtype == M_HASHTYPE_NONE) return (hpts_random_cpu(inp)); /* * Hash to a thread based on the flowid. If we are using numa, * then restrict the hash to the numa domain where the inp lives. */ #ifdef NUMA if (tcp_bind_threads == 2 && inp->inp_numa_domain != M_NODOM) { di = &hpts_domains[inp->inp_numa_domain]; cpuid = di->cpu[inp->inp_flowid % di->count]; } else #endif cpuid = inp->inp_flowid % mp_ncpus; return (cpuid); #endif } static void tcp_drop_in_pkts(struct tcpcb *tp) { struct mbuf *m, *n; m = tp->t_in_pkt; if (m) n = m->m_nextpkt; else n = NULL; tp->t_in_pkt = NULL; while (m) { m_freem(m); m = n; if (m) n = m->m_nextpkt; } } /* * Do NOT try to optimize the processing of inp's * by first pulling off all the inp's into a temporary * list (e.g. TAILQ_CONCAT). If you do that the subtle * interactions of switching CPU's will kill because of * problems in the linked list manipulation. Basically * you would switch cpu's with the hpts mutex locked * but then while you were processing one of the inp's * some other one that you switch will get a new * packet on the different CPU. It will insert it * on the new hpts's input list. Creating a temporary * link in the inp will not fix it either, since * the other hpts will be doing the same thing and * you will both end up using the temporary link. * * You will die in an ASSERT for tailq corruption if you * run INVARIANTS or you will die horribly without * INVARIANTS in some unknown way with a corrupt linked * list. */ static void tcp_input_data(struct tcp_hpts_entry *hpts, struct timeval *tv) { struct tcpcb *tp; struct inpcb *inp; uint16_t drop_reason; int16_t set_cpu; uint32_t did_prefetch = 0; int dropped; HPTS_MTX_ASSERT(hpts); NET_EPOCH_ASSERT(); while ((inp = TAILQ_FIRST(&hpts->p_input)) != NULL) { HPTS_MTX_ASSERT(hpts); hpts_sane_input_remove(hpts, inp, 0); if (inp->inp_input_cpu_set == 0) { set_cpu = 1; } else { set_cpu = 0; } hpts->p_inp = inp; drop_reason = inp->inp_hpts_drop_reas; inp->inp_in_input = 0; mtx_unlock(&hpts->p_mtx); INP_WLOCK(inp); #ifdef VIMAGE CURVNET_SET(inp->inp_vnet); #endif if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) || (inp->inp_flags2 & INP_FREED)) { out: hpts->p_inp = NULL; if (in_pcbrele_wlocked(inp) == 0) { INP_WUNLOCK(inp); } #ifdef VIMAGE CURVNET_RESTORE(); #endif mtx_lock(&hpts->p_mtx); continue; } tp = intotcpcb(inp); if ((tp == NULL) || (tp->t_inpcb == NULL)) { goto out; } if (drop_reason) { /* This tcb is being destroyed for drop_reason */ tcp_drop_in_pkts(tp); tp = tcp_drop(tp, drop_reason); if (tp == NULL) { INP_WLOCK(inp); } if (in_pcbrele_wlocked(inp) == 0) INP_WUNLOCK(inp); #ifdef VIMAGE CURVNET_RESTORE(); #endif mtx_lock(&hpts->p_mtx); continue; } if (set_cpu) { /* * Setup so the next time we will move to the right * CPU. This should be a rare event. It will * sometimes happens when we are the client side * (usually not the server). Somehow tcp_output() * gets called before the tcp_do_segment() sets the * intial state. This means the r_cpu and r_hpts_cpu * is 0. We get on the hpts, and then tcp_input() * gets called setting up the r_cpu to the correct * value. The hpts goes off and sees the mis-match. * We simply correct it here and the CPU will switch * to the new hpts nextime the tcb gets added to the * the hpts (not this time) :-) */ tcp_set_hpts(inp); } if (tp->t_fb_ptr != NULL) { kern_prefetch(tp->t_fb_ptr, &did_prefetch); did_prefetch = 1; } if ((inp->inp_flags2 & INP_SUPPORTS_MBUFQ) && tp->t_in_pkt) { if (inp->inp_in_input) tcp_hpts_remove(inp, HPTS_REMOVE_INPUT); dropped = (*tp->t_fb->tfb_do_queued_segments)(inp->inp_socket, tp, 0); if (dropped) { /* Re-acquire the wlock so we can release the reference */ INP_WLOCK(inp); } } else if (tp->t_in_pkt) { /* * We reach here only if we had a * stack that supported INP_SUPPORTS_MBUFQ * and then somehow switched to a stack that * does not. The packets are basically stranded * and would hang with the connection until * cleanup without this code. Its not the * best way but I know of no other way to * handle it since the stack needs functions * it does not have to handle queued packets. */ tcp_drop_in_pkts(tp); } if (in_pcbrele_wlocked(inp) == 0) INP_WUNLOCK(inp); INP_UNLOCK_ASSERT(inp); #ifdef VIMAGE CURVNET_RESTORE(); #endif mtx_lock(&hpts->p_mtx); hpts->p_inp = NULL; } } static void tcp_hptsi(struct tcp_hpts_entry *hpts) { struct tcpcb *tp; struct inpcb *inp = NULL, *ninp; struct timeval tv; int32_t ticks_to_run, i, error; int32_t paced_cnt = 0; int32_t loop_cnt = 0; int32_t did_prefetch = 0; int32_t prefetch_ninp = 0; int32_t prefetch_tp = 0; int32_t wrap_loop_cnt = 0; int16_t set_cpu; HPTS_MTX_ASSERT(hpts); NET_EPOCH_ASSERT(); /* record previous info for any logging */ hpts->saved_lasttick = hpts->p_lasttick; hpts->saved_curtick = hpts->p_curtick; hpts->saved_curslot = hpts->p_cur_slot; hpts->saved_prev_slot = hpts->p_prev_slot; hpts->p_lasttick = hpts->p_curtick; hpts->p_curtick = tcp_gethptstick(&tv); hpts->p_cur_slot = tick_to_wheel(hpts->p_curtick); if ((hpts->p_on_queue_cnt == 0) || (hpts->p_lasttick == hpts->p_curtick)) { /* * No time has yet passed, * or nothing to do. */ hpts->p_prev_slot = hpts->p_cur_slot; hpts->p_lasttick = hpts->p_curtick; goto no_run; } again: hpts->p_wheel_complete = 0; HPTS_MTX_ASSERT(hpts); ticks_to_run = hpts_ticks_diff(hpts->p_prev_slot, hpts->p_cur_slot); if (((hpts->p_curtick - hpts->p_lasttick) > ticks_to_run) && (hpts->p_on_queue_cnt != 0)) { /* * Wheel wrap is occuring, basically we * are behind and the distance between * run's has spread so much it has exceeded * the time on the wheel (1.024 seconds). This * is ugly and should NOT be happening. We * need to run the entire wheel. We last processed * p_prev_slot, so that needs to be the last slot * we run. The next slot after that should be our * reserved first slot for new, and then starts * the running postion. Now the problem is the * reserved "not to yet" place does not exist * and there may be inp's in there that need * running. We can merge those into the * first slot at the head. */ wrap_loop_cnt++; hpts->p_nxt_slot = hpts_tick(hpts->p_prev_slot, 1); hpts->p_runningtick = hpts_tick(hpts->p_prev_slot, 2); /* * Adjust p_cur_slot to be where we are starting from * hopefully we will catch up (fat chance if something * is broken this bad :( ) */ hpts->p_cur_slot = hpts->p_prev_slot; /* * The next slot has guys to run too, and that would * be where we would normally start, lets move them into * the next slot (p_prev_slot + 2) so that we will * run them, the extra 10usecs of late (by being * put behind) does not really matter in this situation. */ #ifdef INVARIANTS /* * To prevent a panic we need to update the inpslot to the * new location. This is safe since it takes both the * INP lock and the pacer mutex to change the inp_hptsslot. */ TAILQ_FOREACH(inp, &hpts->p_hptss[hpts->p_nxt_slot], inp_hpts) { inp->inp_hptsslot = hpts->p_runningtick; } #endif TAILQ_CONCAT(&hpts->p_hptss[hpts->p_runningtick], &hpts->p_hptss[hpts->p_nxt_slot], inp_hpts); ticks_to_run = NUM_OF_HPTSI_SLOTS - 1; counter_u64_add(wheel_wrap, 1); } else { /* * Nxt slot is always one after p_runningtick though * its not used usually unless we are doing wheel wrap. */ hpts->p_nxt_slot = hpts->p_prev_slot; hpts->p_runningtick = hpts_tick(hpts->p_prev_slot, 1); } #ifdef INVARIANTS if (TAILQ_EMPTY(&hpts->p_input) && (hpts->p_on_inqueue_cnt != 0)) { panic("tp:%p in_hpts input empty but cnt:%d", hpts, hpts->p_on_inqueue_cnt); } #endif HPTS_MTX_ASSERT(hpts); if (hpts->p_on_queue_cnt == 0) { goto no_one; } HPTS_MTX_ASSERT(hpts); for (i = 0; i < ticks_to_run; i++) { /* * Calculate our delay, if there are no extra ticks there * was not any (i.e. if ticks_to_run == 1, no delay). */ hpts->p_delayed_by = (ticks_to_run - (i + 1)) * HPTS_TICKS_PER_USEC; HPTS_MTX_ASSERT(hpts); while ((inp = TAILQ_FIRST(&hpts->p_hptss[hpts->p_runningtick])) != NULL) { /* For debugging */ hpts->p_inp = inp; paced_cnt++; #ifdef INVARIANTS if (hpts->p_runningtick != inp->inp_hptsslot) { panic("Hpts:%p inp:%p slot mis-aligned %u vs %u", hpts, inp, hpts->p_runningtick, inp->inp_hptsslot); } #endif /* Now pull it */ if (inp->inp_hpts_cpu_set == 0) { set_cpu = 1; } else { set_cpu = 0; } hpts_sane_pace_remove(hpts, inp, &hpts->p_hptss[hpts->p_runningtick], 0); if ((ninp = TAILQ_FIRST(&hpts->p_hptss[hpts->p_runningtick])) != NULL) { /* We prefetch the next inp if possible */ kern_prefetch(ninp, &prefetch_ninp); prefetch_ninp = 1; } if (inp->inp_hpts_request) { /* * This guy is deferred out further in time * then our wheel had available on it. * Push him back on the wheel or run it * depending. */ uint32_t maxticks, last_tick, remaining_slots; remaining_slots = ticks_to_run - (i + 1); if (inp->inp_hpts_request > remaining_slots) { /* * How far out can we go? */ maxticks = max_ticks_available(hpts, hpts->p_cur_slot, &last_tick); if (maxticks >= inp->inp_hpts_request) { /* we can place it finally to be processed */ inp->inp_hptsslot = hpts_tick(hpts->p_runningtick, inp->inp_hpts_request); inp->inp_hpts_request = 0; } else { /* Work off some more time */ inp->inp_hptsslot = last_tick; inp->inp_hpts_request-= maxticks; } hpts_sane_pace_insert(hpts, inp, &hpts->p_hptss[inp->inp_hptsslot], __LINE__, 1); hpts->p_inp = NULL; continue; } inp->inp_hpts_request = 0; /* Fall through we will so do it now */ } /* * We clear the hpts flag here after dealing with * remaining slots. This way anyone looking with the * TCB lock will see its on the hpts until just * before we unlock. */ inp->inp_in_hpts = 0; mtx_unlock(&hpts->p_mtx); INP_WLOCK(inp); if (in_pcbrele_wlocked(inp)) { mtx_lock(&hpts->p_mtx); hpts->p_inp = NULL; continue; } if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) || (inp->inp_flags2 & INP_FREED)) { out_now: #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif INP_WUNLOCK(inp); mtx_lock(&hpts->p_mtx); hpts->p_inp = NULL; continue; } tp = intotcpcb(inp); if ((tp == NULL) || (tp->t_inpcb == NULL)) { goto out_now; } if (set_cpu) { /* * Setup so the next time we will move to * the right CPU. This should be a rare * event. It will sometimes happens when we * are the client side (usually not the * server). Somehow tcp_output() gets called * before the tcp_do_segment() sets the * intial state. This means the r_cpu and * r_hpts_cpu is 0. We get on the hpts, and * then tcp_input() gets called setting up * the r_cpu to the correct value. The hpts * goes off and sees the mis-match. We * simply correct it here and the CPU will * switch to the new hpts nextime the tcb * gets added to the the hpts (not this one) * :-) */ tcp_set_hpts(inp); } #ifdef VIMAGE CURVNET_SET(inp->inp_vnet); #endif /* Lets do any logging that we might want to */ if (hpts_does_tp_logging && (tp->t_logstate != TCP_LOG_STATE_OFF)) { tcp_hpts_log(hpts, tp, &tv, ticks_to_run, i); } /* * There is a hole here, we get the refcnt on the * inp so it will still be preserved but to make * sure we can get the INP we need to hold the p_mtx * above while we pull out the tp/inp, as long as * fini gets the lock first we are assured of having * a sane INP we can lock and test. */ #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx before tcp-output:%d", hpts, __LINE__); } #endif if (tp->t_fb_ptr != NULL) { kern_prefetch(tp->t_fb_ptr, &did_prefetch); did_prefetch = 1; } if ((inp->inp_flags2 & INP_SUPPORTS_MBUFQ) && tp->t_in_pkt) { error = (*tp->t_fb->tfb_do_queued_segments)(inp->inp_socket, tp, 0); if (error) { /* The input killed the connection */ goto skip_pacing; } } inp->inp_hpts_calls = 1; error = tp->t_fb->tfb_tcp_output(tp); inp->inp_hpts_calls = 0; if (ninp && ninp->inp_ppcb) { /* * If we have a nxt inp, see if we can * prefetch its ppcb. Note this may seem * "risky" since we have no locks (other * than the previous inp) and there no * assurance that ninp was not pulled while * we were processing inp and freed. If this * occured it could mean that either: * * a) Its NULL (which is fine we won't go * here) b) Its valid (which is cool we * will prefetch it) c) The inp got * freed back to the slab which was * reallocated. Then the piece of memory was * re-used and something else (not an * address) is in inp_ppcb. If that occurs * we don't crash, but take a TLB shootdown * performance hit (same as if it was NULL * and we tried to pre-fetch it). * * Considering that the likelyhood of is * quite rare we will take a risk on doing * this. If performance drops after testing * we can always take this out. NB: the * kern_prefetch on amd64 actually has * protection against a bad address now via * the DMAP_() tests. This will prevent the * TLB hit, and instead if occurs just * cause us to load cache with a useless * address (to us). */ kern_prefetch(ninp->inp_ppcb, &prefetch_tp); prefetch_tp = 1; } INP_WUNLOCK(inp); skip_pacing: #ifdef VIMAGE CURVNET_RESTORE(); #endif INP_UNLOCK_ASSERT(inp); #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif mtx_lock(&hpts->p_mtx); hpts->p_inp = NULL; } HPTS_MTX_ASSERT(hpts); hpts->p_inp = NULL; hpts->p_runningtick++; if (hpts->p_runningtick >= NUM_OF_HPTSI_SLOTS) { hpts->p_runningtick = 0; } } no_one: HPTS_MTX_ASSERT(hpts); hpts->p_delayed_by = 0; /* * Check to see if we took an excess amount of time and need to run * more ticks (if we did not hit eno-bufs). */ #ifdef INVARIANTS if (TAILQ_EMPTY(&hpts->p_input) && (hpts->p_on_inqueue_cnt != 0)) { panic("tp:%p in_hpts input empty but cnt:%d", hpts, hpts->p_on_inqueue_cnt); } #endif hpts->p_prev_slot = hpts->p_cur_slot; hpts->p_lasttick = hpts->p_curtick; if (loop_cnt > max_pacer_loops) { /* * Something is serious slow we have * looped through processing the wheel * and by the time we cleared the * needs to run max_pacer_loops time * we still needed to run. That means * the system is hopelessly behind and * can never catch up :( * * We will just lie to this thread * and let it thing p_curtick is * correct. When it next awakens * it will find itself further behind. */ counter_u64_add(hpts_hopelessly_behind, 1); goto no_run; } hpts->p_curtick = tcp_gethptstick(&tv); hpts->p_cur_slot = tick_to_wheel(hpts->p_curtick); if ((wrap_loop_cnt < 2) && (hpts->p_lasttick != hpts->p_curtick)) { counter_u64_add(hpts_loops, 1); loop_cnt++; goto again; } no_run: /* * Set flag to tell that we are done for * any slot input that happens during * input. */ hpts->p_wheel_complete = 1; /* * Run any input that may be there not covered * in running data. */ if (!TAILQ_EMPTY(&hpts->p_input)) { tcp_input_data(hpts, &tv); /* * Now did we spend too long running * input and need to run more ticks? */ KASSERT(hpts->p_prev_slot == hpts->p_cur_slot, ("H:%p p_prev_slot:%u not equal to p_cur_slot:%u", hpts, hpts->p_prev_slot, hpts->p_cur_slot)); KASSERT(hpts->p_lasttick == hpts->p_curtick, ("H:%p p_lasttick:%u not equal to p_curtick:%u", hpts, hpts->p_lasttick, hpts->p_curtick)); hpts->p_curtick = tcp_gethptstick(&tv); if (hpts->p_lasttick != hpts->p_curtick) { counter_u64_add(hpts_loops, 1); hpts->p_cur_slot = tick_to_wheel(hpts->p_curtick); goto again; } } { uint32_t t = 0, i, fnd = 0; if ((hpts->p_on_queue_cnt) && (wrap_loop_cnt < 2)) { /* * Find next slot that is occupied and use that to * be the sleep time. */ for (i = 0, t = hpts_tick(hpts->p_cur_slot, 1); i < NUM_OF_HPTSI_SLOTS; i++) { if (TAILQ_EMPTY(&hpts->p_hptss[t]) == 0) { fnd = 1; break; } t = (t + 1) % NUM_OF_HPTSI_SLOTS; } if (fnd) { hpts->p_hpts_sleep_time = min((i + 1), hpts_sleep_max); } else { #ifdef INVARIANTS panic("Hpts:%p cnt:%d but none found", hpts, hpts->p_on_queue_cnt); #endif counter_u64_add(back_tosleep, 1); hpts->p_on_queue_cnt = 0; goto non_found; } } else if (wrap_loop_cnt >= 2) { /* Special case handling */ hpts->p_hpts_sleep_time = tcp_min_hptsi_time; } else { /* No one on the wheel sleep for all but 400 slots or sleep max */ non_found: hpts->p_hpts_sleep_time = hpts_sleep_max; } } } void __tcp_set_hpts(struct inpcb *inp, int32_t line) { struct tcp_hpts_entry *hpts; INP_WLOCK_ASSERT(inp); hpts = tcp_hpts_lock(inp); if ((inp->inp_in_hpts == 0) && (inp->inp_hpts_cpu_set == 0)) { inp->inp_hpts_cpu = hpts_cpuid(inp); inp->inp_hpts_cpu_set = 1; } mtx_unlock(&hpts->p_mtx); hpts = tcp_input_lock(inp); if ((inp->inp_input_cpu_set == 0) && (inp->inp_in_input == 0)) { inp->inp_input_cpu = hpts_cpuid(inp); inp->inp_input_cpu_set = 1; } mtx_unlock(&hpts->p_mtx); } uint16_t tcp_hpts_delayedby(struct inpcb *inp){ return (tcp_pace.rp_ent[inp->inp_hpts_cpu]->p_delayed_by); } static void tcp_hpts_thread(void *ctx) { struct tcp_hpts_entry *hpts; struct epoch_tracker et; struct timeval tv; sbintime_t sb; hpts = (struct tcp_hpts_entry *)ctx; mtx_lock(&hpts->p_mtx); if (hpts->p_direct_wake) { /* Signaled by input */ callout_stop(&hpts->co); } else { /* Timed out */ if (callout_pending(&hpts->co) || !callout_active(&hpts->co)) { mtx_unlock(&hpts->p_mtx); return; } callout_deactivate(&hpts->co); } hpts->p_hpts_wake_scheduled = 0; hpts->p_hpts_active = 1; NET_EPOCH_ENTER(et); tcp_hptsi(hpts); NET_EPOCH_EXIT(et); HPTS_MTX_ASSERT(hpts); tv.tv_sec = 0; tv.tv_usec = hpts->p_hpts_sleep_time * HPTS_TICKS_PER_USEC; if (tcp_min_hptsi_time && (tv.tv_usec < tcp_min_hptsi_time)) { hpts->overidden_sleep = tv.tv_usec; tv.tv_usec = tcp_min_hptsi_time; hpts->p_on_min_sleep = 1; } else { /* Clear the min sleep flag */ hpts->overidden_sleep = 0; hpts->p_on_min_sleep = 0; } hpts->p_hpts_active = 0; sb = tvtosbt(tv); if (tcp_hpts_callout_skip_swi == 0) { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_swi, hpts, hpts->p_cpu, (C_DIRECT_EXEC | C_PREL(tcp_hpts_precision))); } else { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_dir, hpts, hpts->p_cpu, C_PREL(tcp_hpts_precision)); } hpts->p_direct_wake = 0; mtx_unlock(&hpts->p_mtx); } #undef timersub static void tcp_init_hptsi(void *st) { int32_t i, j, error, bound = 0, created = 0; size_t sz, asz; struct timeval tv; sbintime_t sb; struct tcp_hpts_entry *hpts; struct pcpu *pc; cpuset_t cs; char unit[16]; uint32_t ncpus = mp_ncpus ? mp_ncpus : MAXCPU; int count, domain; tcp_pace.rp_proc = NULL; tcp_pace.rp_num_hptss = ncpus; hpts_hopelessly_behind = counter_u64_alloc(M_WAITOK); hpts_loops = counter_u64_alloc(M_WAITOK); back_tosleep = counter_u64_alloc(M_WAITOK); combined_wheel_wrap = counter_u64_alloc(M_WAITOK); wheel_wrap = counter_u64_alloc(M_WAITOK); sz = (tcp_pace.rp_num_hptss * sizeof(struct tcp_hpts_entry *)); tcp_pace.rp_ent = malloc(sz, M_TCPHPTS, M_WAITOK | M_ZERO); asz = sizeof(struct hptsh) * NUM_OF_HPTSI_SLOTS; for (i = 0; i < tcp_pace.rp_num_hptss; i++) { tcp_pace.rp_ent[i] = malloc(sizeof(struct tcp_hpts_entry), M_TCPHPTS, M_WAITOK | M_ZERO); tcp_pace.rp_ent[i]->p_hptss = malloc(asz, M_TCPHPTS, M_WAITOK); hpts = tcp_pace.rp_ent[i]; /* * Init all the hpts structures that are not specifically * zero'd by the allocations. Also lets attach them to the * appropriate sysctl block as well. */ mtx_init(&hpts->p_mtx, "tcp_hpts_lck", "hpts", MTX_DEF | MTX_DUPOK); TAILQ_INIT(&hpts->p_input); for (j = 0; j < NUM_OF_HPTSI_SLOTS; j++) { TAILQ_INIT(&hpts->p_hptss[j]); } sysctl_ctx_init(&hpts->hpts_ctx); sprintf(unit, "%d", i); hpts->hpts_root = SYSCTL_ADD_NODE(&hpts->hpts_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_hpts), OID_AUTO, unit, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); SYSCTL_ADD_INT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "in_qcnt", CTLFLAG_RD, &hpts->p_on_inqueue_cnt, 0, "Count TCB's awaiting input processing"); SYSCTL_ADD_INT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "out_qcnt", CTLFLAG_RD, &hpts->p_on_queue_cnt, 0, "Count TCB's awaiting output processing"); SYSCTL_ADD_U16(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "active", CTLFLAG_RD, &hpts->p_hpts_active, 0, "Is the hpts active"); SYSCTL_ADD_UINT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "curslot", CTLFLAG_RD, &hpts->p_cur_slot, 0, "What the current running pacers goal"); SYSCTL_ADD_UINT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "runtick", CTLFLAG_RD, &hpts->p_runningtick, 0, "What the running pacers current slot is"); SYSCTL_ADD_UINT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "curtick", CTLFLAG_RD, &hpts->p_curtick, 0, "What the running pacers last tick mapped to the wheel was"); hpts->p_hpts_sleep_time = hpts_sleep_max; hpts->p_num = i; hpts->p_curtick = tcp_gethptstick(&tv); hpts->p_prev_slot = hpts->p_cur_slot = tick_to_wheel(hpts->p_curtick); hpts->p_cpu = 0xffff; hpts->p_nxt_slot = hpts_tick(hpts->p_cur_slot, 1); callout_init(&hpts->co, 1); } /* Don't try to bind to NUMA domains if we don't have any */ if (vm_ndomains == 1 && tcp_bind_threads == 2) tcp_bind_threads = 0; /* * Now lets start ithreads to handle the hptss. */ CPU_FOREACH(i) { hpts = tcp_pace.rp_ent[i]; hpts->p_cpu = i; error = swi_add(&hpts->ie, "hpts", tcp_hpts_thread, (void *)hpts, SWI_NET, INTR_MPSAFE, &hpts->ie_cookie); if (error) { panic("Can't add hpts:%p i:%d err:%d", hpts, i, error); } created++; if (tcp_bind_threads == 1) { if (intr_event_bind(hpts->ie, i) == 0) bound++; } else if (tcp_bind_threads == 2) { pc = pcpu_find(i); domain = pc->pc_domain; CPU_COPY(&cpuset_domain[domain], &cs); if (intr_event_bind_ithread_cpuset(hpts->ie, &cs) == 0) { bound++; count = hpts_domains[domain].count; hpts_domains[domain].cpu[count] = i; hpts_domains[domain].count++; } } tv.tv_sec = 0; tv.tv_usec = hpts->p_hpts_sleep_time * HPTS_TICKS_PER_USEC; sb = tvtosbt(tv); if (tcp_hpts_callout_skip_swi == 0) { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_swi, hpts, hpts->p_cpu, (C_DIRECT_EXEC | C_PREL(tcp_hpts_precision))); } else { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_dir, hpts, hpts->p_cpu, C_PREL(tcp_hpts_precision)); } } /* * If we somehow have an empty domain, fall back to choosing * among all htps threads. */ for (i = 0; i < vm_ndomains; i++) { if (hpts_domains[i].count == 0) { tcp_bind_threads = 0; break; } } printf("TCP Hpts created %d swi interrupt threads and bound %d to %s\n", created, bound, tcp_bind_threads == 2 ? "NUMA domains" : "cpus"); } SYSINIT(tcphptsi, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, tcp_init_hptsi, NULL); MODULE_VERSION(tcphpts, 1);