freebsd-dev/sys/netinet/tcp_hpts.h
Randall Stewart d7955cc0ff tcp: HPTS performance enhancements
HPTS drives both rack and bbr, and yet there have been many complaints
about performance. This bit of work restructures hpts to help reduce CPU
overhead. It does this by now instead of relying on the timer/callout to
drive it instead use user return from a system call as well as lro flushes
to drive hpts. The timer becomes a backstop that dynamically adjusts
based on how "late" we are.

Reviewed by: tuexen, glebius
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D31083
2021-07-07 07:22:35 -04:00

317 lines
11 KiB
C

/*-
* 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.
*
* $FreeBSD$
*/
#ifndef __tcp_hpts_h__
#define __tcp_hpts_h__
/*
* The hpts uses a 102400 wheel. The wheel
* defines the time in 10 usec increments (102400 x 10).
* This gives a range of 10usec - 1024ms to place
* an entry within. If the user requests more than
* 1.024 second, a remaineder is attached and the hpts
* when seeing the remainder will re-insert the
* inpcb forward in time from where it is until
* the remainder is zero.
*/
#define NUM_OF_HPTSI_SLOTS 102400
TAILQ_HEAD(hptsh, inpcb);
/* Number of useconds in a hpts tick */
#define HPTS_TICKS_PER_SLOT 10
#define HPTS_MS_TO_SLOTS(x) ((x * 100) + 1)
#define HPTS_USEC_TO_SLOTS(x) ((x+9) /10)
#define HPTS_USEC_IN_SEC 1000000
#define HPTS_MSEC_IN_SEC 1000
#define HPTS_USEC_IN_MSEC 1000
struct hpts_diag {
uint32_t p_hpts_active; /* bbr->flex7 x */
uint32_t p_nxt_slot; /* bbr->flex1 x */
uint32_t p_cur_slot; /* bbr->flex2 x */
uint32_t p_prev_slot; /* bbr->delivered */
uint32_t p_runningslot; /* bbr->inflight */
uint32_t slot_req; /* bbr->flex3 x */
uint32_t inp_hptsslot; /* bbr->flex4 x */
uint32_t slot_remaining; /* bbr->flex5 x */
uint32_t have_slept; /* bbr->epoch x */
uint32_t hpts_sleep_time; /* bbr->applimited x */
uint32_t yet_to_sleep; /* bbr->lt_epoch x */
uint32_t need_new_to; /* bbr->flex6 x */
uint32_t wheel_slot; /* bbr->bw_inuse x */
uint32_t maxslots; /* bbr->delRate x */
uint32_t wheel_cts; /* bbr->rttProp x */
int32_t co_ret; /* bbr->pkts_out x */
uint32_t p_curtick; /* upper bbr->cur_del_rate */
uint32_t p_lasttick; /* lower bbr->cur_del_rate */
uint8_t p_on_min_sleep; /* bbr->flex8 x */
};
/* Magic flags to tell whats cooking on the pacing wheel */
#define PACE_TMR_DELACK 0x01 /* Delayed ack timer running */
#define PACE_TMR_RACK 0x02 /* RACK timer running */
#define PACE_TMR_TLP 0x04 /* TLP timer running */
#define PACE_TMR_RXT 0x08 /* Retransmit timer running */
#define PACE_TMR_PERSIT 0x10 /* Persists timer running */
#define PACE_TMR_KEEP 0x20 /* Keep alive timer running */
#define PACE_PKT_OUTPUT 0x40 /* Output Packets being paced */
#define PACE_TMR_MASK (PACE_TMR_KEEP|PACE_TMR_PERSIT|PACE_TMR_RXT|PACE_TMR_TLP|PACE_TMR_RACK|PACE_TMR_DELACK)
#define DEFAULT_CONNECTION_THESHOLD 100
#ifdef _KERNEL
/* Each hpts has its own p_mtx which is used for locking */
struct tcp_hpts_entry {
/* Cache line 0x00 */
struct mtx p_mtx; /* Mutex for hpts */
struct timeval p_mysleep; /* Our min sleep time */
uint64_t syscall_cnt;
uint64_t sleeping; /* What the actual sleep was (if sleeping) */
uint16_t p_hpts_active; /* Flag that says hpts is awake */
uint8_t p_wheel_complete; /* have we completed the wheel arc walk? */
uint32_t p_curtick; /* Tick in 10 us the hpts is going to */
uint32_t p_runningslot; /* Current tick we are at if we are running */
uint32_t p_prev_slot; /* Previous slot we were on */
uint32_t p_cur_slot; /* Current slot in wheel hpts is draining */
uint32_t p_nxt_slot; /* The next slot outside the current range of
* slots that the hpts is running on. */
int32_t p_on_queue_cnt; /* Count on queue in this hpts */
uint32_t p_lasttick; /* Last tick before the current one */
uint8_t p_direct_wake :1, /* boolean */
p_on_min_sleep:1, /* boolean */
p_hpts_wake_scheduled:1, /* boolean */
p_avail:5;
uint8_t p_fill[3]; /* Fill to 32 bits */
/* Cache line 0x40 */
void *p_inp;
struct hptsh p_input; /* For the tcp-input runner */
/* Hptsi wheel */
struct hptsh *p_hptss;
int32_t p_on_inqueue_cnt; /* Count on input queue in this hpts */
uint32_t p_hpts_sleep_time; /* Current sleep interval having a max
* of 255ms */
uint32_t overidden_sleep; /* what was overrided by min-sleep for logging */
uint32_t saved_lasttick; /* for logging */
uint32_t saved_curtick; /* for logging */
uint32_t saved_curslot; /* for logging */
uint32_t saved_prev_slot; /* for logging */
uint32_t p_delayed_by; /* How much were we delayed by */
/* Cache line 0x80 */
struct sysctl_ctx_list hpts_ctx;
struct sysctl_oid *hpts_root;
struct intr_event *ie;
void *ie_cookie;
uint16_t p_num; /* The hpts number one per cpu */
uint16_t p_cpu; /* The hpts CPU */
/* There is extra space in here */
/* Cache line 0x100 */
struct callout co __aligned(CACHE_LINE_SIZE);
} __aligned(CACHE_LINE_SIZE);
struct tcp_hptsi {
struct proc *rp_proc; /* Process structure for hpts */
struct tcp_hpts_entry **rp_ent; /* Array of hptss */
uint32_t *cts_last_ran;
uint32_t rp_num_hptss; /* Number of hpts threads */
};
#endif
#define HPTS_REMOVE_INPUT 0x01
#define HPTS_REMOVE_OUTPUT 0x02
#define HPTS_REMOVE_ALL (HPTS_REMOVE_INPUT | HPTS_REMOVE_OUTPUT)
/*
* When using the hpts, a TCP stack must make sure
* that once a INP_DROPPED flag is applied to a INP
* that it does not expect tcp_output() to ever be
* called by the hpts. The hpts will *not* call
* any output (or input) functions on a TCB that
* is in the DROPPED state.
*
* This implies final ACK's and RST's that might
* be sent when a TCB is still around must be
* sent from a routine like tcp_respond().
*/
#define LOWEST_SLEEP_ALLOWED 50
#define DEFAULT_MIN_SLEEP 250 /* How many usec's is default for hpts sleep
* this determines min granularity of the
* hpts. If 1, granularity is 10useconds at
* the cost of more CPU (context switching).
* Note do not set this to 0.
*/
#define DYNAMIC_MIN_SLEEP DEFAULT_MIN_SLEEP
#define DYNAMIC_MAX_SLEEP 100000 /* 100ms */
/* No of connections when wee start aligning to the cpu from syscalls */
#define OLDEST_THRESHOLD 1200
/* Thresholds for raising/lowering sleep */
#define TICKS_INDICATE_MORE_SLEEP 100 /* This would be 1ms */
#define TICKS_INDICATE_LESS_SLEEP 1000 /* This would indicate 10ms */
/**
*
* Dynamic adjustment of sleeping times is done in "new" mode
* where we are depending on syscall returns and lro returns
* to push hpts forward mainly and the timer is only a backstop.
*
* When we are in the "new" mode i.e. conn_cnt > conn_cnt_thresh
* then we do a dynamic adjustment on the time we sleep.
* Our threshold is if the lateness of the first client served (in ticks) is
* greater than or equal too ticks_indicate_more_sleep (10ms
* or 10000 ticks). If we were that late, the actual sleep time
* is adjusted down by 50%. If the ticks_ran is less than
* ticks_indicate_more_sleep (100 ticks or 1000usecs).
*
*/
#ifdef _KERNEL
#define HPTS_MTX_ASSERT(hpts) mtx_assert(&(hpts)->p_mtx, MA_OWNED)
struct tcp_hpts_entry *tcp_hpts_lock(struct inpcb *inp);
struct tcp_hpts_entry *tcp_input_lock(struct inpcb *inp);
int __tcp_queue_to_hpts_immediate(struct inpcb *inp, int32_t line);
#define tcp_queue_to_hpts_immediate(a)__tcp_queue_to_hpts_immediate(a, __LINE__)
struct tcp_hpts_entry *tcp_cur_hpts(struct inpcb *inp);
#define tcp_hpts_remove(a, b) __tcp_hpts_remove(a, b, __LINE__)
void __tcp_hpts_remove(struct inpcb *inp, int32_t flags, int32_t line);
/*
* To insert a TCB on the hpts you *must* be holding the
* INP_WLOCK(). The hpts insert code will then acqurire
* the hpts's lock and insert the TCB on the requested
* slot possibly waking up the hpts if you are requesting
* a time earlier than what the hpts is sleeping to (if
* the hpts is sleeping). You may check the inp->inp_in_hpts
* flag without the hpts lock. The hpts is the only one
* that will clear this flag holding only the hpts lock. This
* means that in your tcp_output() routine when you test for
* it to be 1 (so you wont call output) it may be transitioning
* to 0 (by the hpts). That will be fine since that will just
* mean an extra call to tcp_output that most likely will find
* the call you executed (when the mis-match occured) will have
* put the TCB back on the hpts and it will return. If your
* call did not add it back to the hpts then you will either
* over-send or the cwnd will block you from sending more.
*
* Note you should also be holding the INP_WLOCK() when you
* call the remove from the hpts as well. Thoug usually
* you are either doing this from a timer, where you need
* that INP_WLOCK() or from destroying your TCB where again
* you should already have the INP_WLOCK().
*/
uint32_t __tcp_hpts_insert(struct inpcb *inp, uint32_t slot, int32_t line);
#define tcp_hpts_insert(a, b) __tcp_hpts_insert(a, b, __LINE__)
uint32_t
tcp_hpts_insert_diag(struct inpcb *inp, uint32_t slot, int32_t line, struct hpts_diag *diag);
int
__tcp_queue_to_input_locked(struct inpcb *inp, struct tcp_hpts_entry *hpts, int32_t line);
#define tcp_queue_to_input_locked(a, b) __tcp_queue_to_input_locked(a, b, __LINE__);
int
__tcp_queue_to_input(struct inpcb *inp, int32_t line);
#define tcp_queue_to_input(a) __tcp_queue_to_input(a, __LINE__)
uint16_t tcp_hpts_delayedby(struct inpcb *inp);
void __tcp_set_hpts(struct inpcb *inp, int32_t line);
#define tcp_set_hpts(a) __tcp_set_hpts(a, __LINE__)
void __tcp_set_inp_to_drop(struct inpcb *inp, uint16_t reason, int32_t line);
#define tcp_set_inp_to_drop(a, b) __tcp_set_inp_to_drop(a, b, __LINE__)
void tcp_run_hpts(void);
uint16_t hpts_random_cpu(struct inpcb *inp);
extern int32_t tcp_min_hptsi_time;
#endif /* _KERNEL */
/*
* The following functions should also be available
* to userspace as well.
*/
static __inline uint32_t
tcp_tv_to_hptstick(const struct timeval *sv)
{
return ((sv->tv_sec * 100000) + (sv->tv_usec / HPTS_TICKS_PER_SLOT));
}
static __inline uint32_t
tcp_tv_to_usectick(const struct timeval *sv)
{
return ((uint32_t) ((sv->tv_sec * HPTS_USEC_IN_SEC) + sv->tv_usec));
}
static __inline uint32_t
tcp_tv_to_mssectick(const struct timeval *sv)
{
return ((uint32_t) ((sv->tv_sec * HPTS_MSEC_IN_SEC) + (sv->tv_usec/HPTS_USEC_IN_MSEC)));
}
static __inline uint64_t
tcp_tv_to_lusectick(const struct timeval *sv)
{
return ((uint64_t)((sv->tv_sec * HPTS_USEC_IN_SEC) + sv->tv_usec));
}
#ifdef _KERNEL
static __inline void
tcp_hpts_unlock(struct tcp_hpts_entry *hpts)
{
mtx_unlock(&hpts->p_mtx);
}
static __inline uint32_t
tcp_gethptstick(struct timeval *sv)
{
struct timeval tv;
if (sv == NULL)
sv = &tv;
microuptime(sv);
return (tcp_tv_to_hptstick(sv));
}
static __inline uint32_t
tcp_get_usecs(struct timeval *tv)
{
struct timeval tvd;
if (tv == NULL)
tv = &tvd;
microuptime(tv);
return (tcp_tv_to_usectick(tv));
}
#endif /* _KERNEL */
#endif /* __tcp_hpts_h__ */