include/net/ll_poll.h

   1 /*
   2  * Low Latency Sockets
   3  * Copyright(c) 2013 Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along with
  15  * this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  17  *
  18  * Author: Eliezer Tamir
  19  *
  20  * Contact Information:
  21  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  22  */
  23
  24 #ifndef _LINUX_NET_LL_POLL_H
  25 #define _LINUX_NET_LL_POLL_H
  26
  27 #include <linux/netdevice.h>
  28 #include <net/ip.h>
  29
  30 #ifdef CONFIG_NET_LL_RX_POLL
  31
  32 struct napi_struct;
  33 extern unsigned int sysctl_net_ll_read __read_mostly;
  34 extern unsigned int sysctl_net_ll_poll __read_mostly;
  35
  36 /* return values from ndo_ll_poll */
  37 #define LL_FLUSH_FAILED         -1
  38 #define LL_FLUSH_BUSY           -2
  39
  40 static inline bool net_busy_loop_on(void)
  41 {
  42         return sysctl_net_ll_poll;
  43 }
  44
  45 /* a wrapper to make debug_smp_processor_id() happy
  46  * we can use sched_clock() because we don't care much about precision
  47  * we only care that the average is bounded
  48  */
  49 #ifdef CONFIG_DEBUG_PREEMPT
  50 static inline u64 busy_loop_sched_clock(void)
  51 {
  52         u64 rc;
  53
  54         preempt_disable_notrace();
  55         rc = sched_clock();
  56         preempt_enable_no_resched_notrace();
  57
  58         return rc;
  59 }
  60 #else /* CONFIG_DEBUG_PREEMPT */
  61 static inline u64 busy_loop_sched_clock(void)
  62 {
  63         return sched_clock();
  64 }
  65 #endif /* CONFIG_DEBUG_PREEMPT */
  66
  67 /* we don't mind a ~2.5% imprecision so <<10 instead of *1000
  68  * sk->sk_ll_usec is a u_int so this can't overflow
  69  */
  70 static inline u64 sk_busy_loop_end_time(struct sock *sk)
  71 {
  72         return (u64)ACCESS_ONCE(sk->sk_ll_usec) << 10;
  73 }
  74
  75 /* in poll/select we use the global sysctl_net_ll_poll value
  76  * only call sched_clock() if enabled
  77  */
  78 static inline u64 busy_loop_end_time(void)
  79 {
  80         return (u64)ACCESS_ONCE(sysctl_net_ll_poll) << 10;
  81 }
  82
  83 /* if flag is not set we don't need to know the time
  84  * so we want to avoid a potentially expensive sched_clock()
  85  */
  86 static inline u64 busy_loop_start_time(unsigned int flag)
  87 {
  88         return flag ? busy_loop_sched_clock() : 0;
  89 }
  90
  91 static inline bool sk_can_busy_loop(struct sock *sk)
  92 {
  93         return sk->sk_ll_usec && sk->sk_napi_id &&
  94                !need_resched() && !signal_pending(current);
  95 }
  96
  97 /* careful! time_in_range64 will evaluate now twice */
  98 static inline bool busy_loop_range(u64 start_time, u64 run_time)
  99 {
 100         u64 now = busy_loop_sched_clock();
 101
 102         return time_in_range64(now, start_time, start_time + run_time);
 103 }
 104
 105 /* when used in sock_poll() nonblock is known at compile time to be true
 106  * so the loop and end_time will be optimized out
 107  */
 108 static inline bool sk_busy_loop(struct sock *sk, int nonblock)
 109 {
 110         u64 start_time = busy_loop_start_time(!nonblock);
 111         u64 end_time = sk_busy_loop_end_time(sk);
 112         const struct net_device_ops *ops;
 113         struct napi_struct *napi;
 114         int rc = false;
 115
 116         /*
 117          * rcu read lock for napi hash
 118          * bh so we don't race with net_rx_action
 119          */
 120         rcu_read_lock_bh();
 121
 122         napi = napi_by_id(sk->sk_napi_id);
 123         if (!napi)
 124                 goto out;
 125
 126         ops = napi->dev->netdev_ops;
 127         if (!ops->ndo_ll_poll)
 128                 goto out;
 129
 130         do {
 131                 rc = ops->ndo_ll_poll(napi);
 132
 133                 if (rc == LL_FLUSH_FAILED)
 134                         break; /* permanent failure */
 135
 136                 if (rc > 0)
 137                         /* local bh are disabled so it is ok to use _BH */
 138                         NET_ADD_STATS_BH(sock_net(sk),
 139                                          LINUX_MIB_LOWLATENCYRXPACKETS, rc);
 140
 141         } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
 142                  busy_loop_range(start_time, end_time));
 143
 144         rc = !skb_queue_empty(&sk->sk_receive_queue);
 145 out:
 146         rcu_read_unlock_bh();
 147         return rc;
 148 }
 149
 150 /* used in the NIC receive handler to mark the skb */
 151 static inline void skb_mark_ll(struct sk_buff *skb, struct napi_struct *napi)
 152 {
 153         skb->napi_id = napi->napi_id;
 154 }
 155
 156 /* used in the protocol hanlder to propagate the napi_id to the socket */
 157 static inline void sk_mark_ll(struct sock *sk, struct sk_buff *skb)
 158 {
 159         sk->sk_napi_id = skb->napi_id;
 160 }
 161
 162 #else /* CONFIG_NET_LL_RX_POLL */
 163 static inline unsigned long net_busy_loop_on(void)
 164 {
 165         return 0;
 166 }
 167
 168 static inline u64 busy_loop_start_time(unsigned int flag)
 169 {
 170         return 0;
 171 }
 172
 173 static inline u64 busy_loop_end_time(void)
 174 {
 175         return 0;
 176 }
 177
 178 static inline bool sk_can_busy_loop(struct sock *sk)
 179 {
 180         return false;
 181 }
 182
 183 static inline bool sk_busy_poll(struct sock *sk, int nonblock)
 184 {
 185         return false;
 186 }
 187
 188 static inline void skb_mark_ll(struct sk_buff *skb, struct napi_struct *napi)
 189 {
 190 }
 191
 192 static inline void sk_mark_ll(struct sock *sk, struct sk_buff *skb)
 193 {
 194 }
 195
 196 static inline bool busy_loop_range(u64 start_time, u64 run_time)
 197 {
 198         return false;
 199 }
 200
 201 #endif /* CONFIG_NET_LL_RX_POLL */
 202 #endif /* _LINUX_NET_LL_POLL_H */