1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_NET_QUEUES_H 2 #ifndef _LINUX_NET_QUEUES_H
3 #define _LINUX_NET_QUEUES_H 3 #define _LINUX_NET_QUEUES_H
4 4
5 #include <linux/netdevice.h> 5 #include <linux/netdevice.h>
6 6
7 /* See the netdev.yaml spec for definition of 7 /* See the netdev.yaml spec for definition of each statistic */
8 struct netdev_queue_stats_rx { 8 struct netdev_queue_stats_rx {
9 u64 bytes; 9 u64 bytes;
10 u64 packets; 10 u64 packets;
11 u64 alloc_fail; 11 u64 alloc_fail;
12 <<
13 u64 hw_drops; <<
14 u64 hw_drop_overruns; <<
15 <<
16 u64 csum_unnecessary; <<
17 u64 csum_none; <<
18 u64 csum_bad; <<
19 <<
20 u64 hw_gro_packets; <<
21 u64 hw_gro_bytes; <<
22 u64 hw_gro_wire_packets; <<
23 u64 hw_gro_wire_bytes; <<
24 <<
25 u64 hw_drop_ratelimits; <<
26 }; 12 };
27 13
28 struct netdev_queue_stats_tx { 14 struct netdev_queue_stats_tx {
29 u64 bytes; 15 u64 bytes;
30 u64 packets; 16 u64 packets;
31 <<
32 u64 hw_drops; <<
33 u64 hw_drop_errors; <<
34 <<
35 u64 csum_none; <<
36 u64 needs_csum; <<
37 <<
38 u64 hw_gso_packets; <<
39 u64 hw_gso_bytes; <<
40 u64 hw_gso_wire_packets; <<
41 u64 hw_gso_wire_bytes; <<
42 <<
43 u64 hw_drop_ratelimits; <<
44 <<
45 u64 stop; <<
46 u64 wake; <<
47 }; 17 };
48 18
49 /** 19 /**
50 * struct netdev_stat_ops - netdev ops for fin 20 * struct netdev_stat_ops - netdev ops for fine grained stats
51 * @get_queue_stats_rx: get stats for a given 21 * @get_queue_stats_rx: get stats for a given Rx queue
52 * @get_queue_stats_tx: get stats for a given 22 * @get_queue_stats_tx: get stats for a given Tx queue
53 * @get_base_stats: get base stats (not be 23 * @get_base_stats: get base stats (not belonging to any live instance)
54 * 24 *
55 * Query stats for a given object. The values 25 * Query stats for a given object. The values of the statistics are undefined
56 * on entry (specifically they are *not* zero- 26 * on entry (specifically they are *not* zero-initialized). Drivers should
57 * assign values only to the statistics they c 27 * assign values only to the statistics they collect. Statistics which are not
58 * collected must be left undefined. 28 * collected must be left undefined.
59 * 29 *
60 * Queue objects are not necessarily persisten 30 * Queue objects are not necessarily persistent, and only currently active
61 * queues are queried by the per-queue callbac 31 * queues are queried by the per-queue callbacks. This means that per-queue
62 * statistics will not generally add up to the 32 * statistics will not generally add up to the total number of events for
63 * the device. The @get_base_stats callback al 33 * the device. The @get_base_stats callback allows filling in the delta
64 * between events for currently live queues an 34 * between events for currently live queues and overall device history.
65 * @get_base_stats can also be used to report <<
66 * transferred outside of the main set of queu <<
67 * When the statistics for the entire device a 35 * When the statistics for the entire device are queried, first @get_base_stats
68 * is issued to collect the delta, and then a 36 * is issued to collect the delta, and then a series of per-queue callbacks.
69 * Only statistics which are set in @get_base_ 37 * Only statistics which are set in @get_base_stats will be reported
70 * at the device level, meaning that unlike in 38 * at the device level, meaning that unlike in queue callbacks, setting
71 * a statistic to zero in @get_base_stats is a 39 * a statistic to zero in @get_base_stats is a legitimate thing to do.
72 * This is because @get_base_stats has a secon 40 * This is because @get_base_stats has a second function of designating which
73 * statistics are in fact correct for the enti 41 * statistics are in fact correct for the entire device (e.g. when history
74 * for some of the events is not maintained, a 42 * for some of the events is not maintained, and reliable "total" cannot
75 * be provided). 43 * be provided).
76 * 44 *
77 * Device drivers can assume that when collect 45 * Device drivers can assume that when collecting total device stats,
78 * the @get_base_stats and subsequent per-queu 46 * the @get_base_stats and subsequent per-queue calls are performed
79 * "atomically" (without releasing the rtnl_lo 47 * "atomically" (without releasing the rtnl_lock).
80 * 48 *
81 * Device drivers are encouraged to reset the 49 * Device drivers are encouraged to reset the per-queue statistics when
82 * number of queues change. This is because th 50 * number of queues change. This is because the primary use case for
83 * per-queue statistics is currently to detect 51 * per-queue statistics is currently to detect traffic imbalance.
84 */ 52 */
85 struct netdev_stat_ops { 53 struct netdev_stat_ops {
86 void (*get_queue_stats_rx)(struct net_ 54 void (*get_queue_stats_rx)(struct net_device *dev, int idx,
87 struct netd 55 struct netdev_queue_stats_rx *stats);
88 void (*get_queue_stats_tx)(struct net_ 56 void (*get_queue_stats_tx)(struct net_device *dev, int idx,
89 struct netd 57 struct netdev_queue_stats_tx *stats);
90 void (*get_base_stats)(struct net_devi 58 void (*get_base_stats)(struct net_device *dev,
91 struct netdev_q 59 struct netdev_queue_stats_rx *rx,
92 struct netdev_q 60 struct netdev_queue_stats_tx *tx);
93 }; <<
94 <<
95 /** <<
96 * struct netdev_queue_mgmt_ops - netdev ops f <<
97 * <<
98 * @ndo_queue_mem_size: Size of the struct tha <<
99 * <<
100 * @ndo_queue_mem_alloc: Allocate memory for a <<
101 * The new memory is wri <<
102 * <<
103 * @ndo_queue_mem_free: Free memory from an RX <<
104 * <<
105 * @ndo_queue_start: Start an RX queue with <<
106 * specified index. <<
107 * <<
108 * @ndo_queue_stop: Stop the RX queue at t <<
109 * queue's memory is writ <<
110 */ <<
111 struct netdev_queue_mgmt_ops { <<
112 size_t ndo_queue_mem_ <<
113 int (*ndo_queue_me <<
114 <<
115 <<
116 void (*ndo_queue_me <<
117 <<
118 int (*ndo_queue_st <<
119 <<
120 <<
121 int (*ndo_queue_st <<
122 <<
123 <<
124 }; 61 };
125 62
126 /** 63 /**
127 * DOC: Lockless queue stopping / waking helpe 64 * DOC: Lockless queue stopping / waking helpers.
128 * 65 *
129 * The netif_txq_maybe_stop() and __netif_txq_ 66 * The netif_txq_maybe_stop() and __netif_txq_completed_wake()
130 * macros are designed to safely implement sto 67 * macros are designed to safely implement stopping
131 * and waking netdev queues without full lock 68 * and waking netdev queues without full lock protection.
132 * 69 *
133 * We assume that there can be no concurrent s 70 * We assume that there can be no concurrent stop attempts and no concurrent
134 * wake attempts. The try-stop should happen f 71 * wake attempts. The try-stop should happen from the xmit handler,
135 * while wake up should be triggered from NAPI 72 * while wake up should be triggered from NAPI poll context.
136 * The two may run concurrently (single produc 73 * The two may run concurrently (single producer, single consumer).
137 * 74 *
138 * The try-stop side is expected to run from t 75 * The try-stop side is expected to run from the xmit handler and therefore
139 * it does not reschedule Tx (netif_tx_start_q 76 * it does not reschedule Tx (netif_tx_start_queue() instead of
140 * netif_tx_wake_queue()). Uses of the ``stop` 77 * netif_tx_wake_queue()). Uses of the ``stop`` macros outside of the xmit
141 * handler may lead to xmit queue being enable 78 * handler may lead to xmit queue being enabled but not run.
142 * The waking side does not have similar conte 79 * The waking side does not have similar context restrictions.
143 * 80 *
144 * The macros guarantee that rings will not re 81 * The macros guarantee that rings will not remain stopped if there's
145 * space available, but they do *not* prevent 82 * space available, but they do *not* prevent false wake ups when
146 * the ring is full! Drivers should check for 83 * the ring is full! Drivers should check for ring full at the start
147 * for the xmit handler. 84 * for the xmit handler.
148 * 85 *
149 * All descriptor ring indexes (and other rele 86 * All descriptor ring indexes (and other relevant shared state) must
150 * be updated before invoking the macros. 87 * be updated before invoking the macros.
151 */ 88 */
152 89
153 #define netif_txq_try_stop(txq, get_desc, star 90 #define netif_txq_try_stop(txq, get_desc, start_thrs) \
154 ({ 91 ({ \
155 int _res; 92 int _res; \
156 93 \
157 netif_tx_stop_queue(txq); 94 netif_tx_stop_queue(txq); \
158 /* Producer index and stop bit 95 /* Producer index and stop bit must be visible \
159 * to consumer before we reche 96 * to consumer before we recheck. \
160 * Pairs with a barrier in __n 97 * Pairs with a barrier in __netif_txq_completed_wake(). \
161 */ 98 */ \
162 smp_mb__after_atomic(); 99 smp_mb__after_atomic(); \
163 100 \
164 /* We need to check again in a 101 /* We need to check again in a case another \
165 * CPU has just made room avai 102 * CPU has just made room available. \
166 */ 103 */ \
167 _res = 0; 104 _res = 0; \
168 if (unlikely(get_desc >= start 105 if (unlikely(get_desc >= start_thrs)) { \
169 netif_tx_start_queue(t 106 netif_tx_start_queue(txq); \
170 _res = -1; 107 _res = -1; \
171 } 108 } \
172 _res; 109 _res; \
173 }) 110 }) \
174 111
175 /** 112 /**
176 * netif_txq_maybe_stop() - locklessly stop a 113 * netif_txq_maybe_stop() - locklessly stop a Tx queue, if needed
177 * @txq: struct netdev_queue to stop/st 114 * @txq: struct netdev_queue to stop/start
178 * @get_desc: get current number of free des 115 * @get_desc: get current number of free descriptors (see requirements below!)
179 * @stop_thrs: minimal number of available de 116 * @stop_thrs: minimal number of available descriptors for queue to be left
180 * enabled 117 * enabled
181 * @start_thrs: minimal number of descriptors 118 * @start_thrs: minimal number of descriptors to re-enable the queue, can be
182 * equal to @stop_thrs or higher 119 * equal to @stop_thrs or higher to avoid frequent waking
183 * 120 *
184 * All arguments may be evaluated multiple tim 121 * All arguments may be evaluated multiple times, beware of side effects.
185 * @get_desc must be a formula or a function c 122 * @get_desc must be a formula or a function call, it must always
186 * return up-to-date information when evaluate 123 * return up-to-date information when evaluated!
187 * Expected to be used from ndo_start_xmit, se 124 * Expected to be used from ndo_start_xmit, see the comment on top of the file.
188 * 125 *
189 * Returns: 126 * Returns:
190 * 0 if the queue was stopped 127 * 0 if the queue was stopped
191 * 1 if the queue was left enabled 128 * 1 if the queue was left enabled
192 * -1 if the queue was re-enabled (raced 129 * -1 if the queue was re-enabled (raced with waking)
193 */ 130 */
194 #define netif_txq_maybe_stop(txq, get_desc, st 131 #define netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs) \
195 ({ 132 ({ \
196 int _res; 133 int _res; \
197 134 \
198 _res = 1; 135 _res = 1; \
199 if (unlikely(get_desc < stop_t 136 if (unlikely(get_desc < stop_thrs)) \
200 _res = netif_txq_try_s 137 _res = netif_txq_try_stop(txq, get_desc, start_thrs); \
201 _res; 138 _res; \
202 }) 139 }) \
203 140
204 /* Variant of netdev_tx_completed_queue() whic 141 /* Variant of netdev_tx_completed_queue() which guarantees smp_mb() if
205 * @bytes != 0, regardless of kernel config. 142 * @bytes != 0, regardless of kernel config.
206 */ 143 */
207 static inline void 144 static inline void
208 netdev_txq_completed_mb(struct netdev_queue *d 145 netdev_txq_completed_mb(struct netdev_queue *dev_queue,
209 unsigned int pkts, uns 146 unsigned int pkts, unsigned int bytes)
210 { 147 {
211 if (IS_ENABLED(CONFIG_BQL)) 148 if (IS_ENABLED(CONFIG_BQL))
212 netdev_tx_completed_queue(dev_ 149 netdev_tx_completed_queue(dev_queue, pkts, bytes);
213 else if (bytes) 150 else if (bytes)
214 smp_mb(); 151 smp_mb();
215 } 152 }
216 153
217 /** 154 /**
218 * __netif_txq_completed_wake() - locklessly w 155 * __netif_txq_completed_wake() - locklessly wake a Tx queue, if needed
219 * @txq: struct netdev_queue to stop/st 156 * @txq: struct netdev_queue to stop/start
220 * @pkts: number of packets completed 157 * @pkts: number of packets completed
221 * @bytes: number of bytes completed 158 * @bytes: number of bytes completed
222 * @get_desc: get current number of free des 159 * @get_desc: get current number of free descriptors (see requirements below!)
223 * @start_thrs: minimal number of descriptors 160 * @start_thrs: minimal number of descriptors to re-enable the queue
224 * @down_cond: down condition, predicate indi 161 * @down_cond: down condition, predicate indicating that the queue should
225 * not be woken up even if descri 162 * not be woken up even if descriptors are available
226 * 163 *
227 * All arguments may be evaluated multiple tim 164 * All arguments may be evaluated multiple times.
228 * @get_desc must be a formula or a function c 165 * @get_desc must be a formula or a function call, it must always
229 * return up-to-date information when evaluate 166 * return up-to-date information when evaluated!
230 * Reports completed pkts/bytes to BQL. 167 * Reports completed pkts/bytes to BQL.
231 * 168 *
232 * Returns: 169 * Returns:
233 * 0 if the queue was woken up 170 * 0 if the queue was woken up
234 * 1 if the queue was already enabled (o 171 * 1 if the queue was already enabled (or disabled but @down_cond is true)
235 * -1 if the queue was left unchanged (@s 172 * -1 if the queue was left unchanged (@start_thrs not reached)
236 */ 173 */
237 #define __netif_txq_completed_wake(txq, pkts, 174 #define __netif_txq_completed_wake(txq, pkts, bytes, \
238 get_desc, s 175 get_desc, start_thrs, down_cond) \
239 ({ 176 ({ \
240 int _res; 177 int _res; \
241 178 \
242 /* Report to BQL and piggy bac 179 /* Report to BQL and piggy back on its barrier. \
243 * Barrier makes sure that any 180 * Barrier makes sure that anybody stopping the queue \
244 * after this point sees the n 181 * after this point sees the new consumer index. \
245 * Pairs with barrier in netif 182 * Pairs with barrier in netif_txq_try_stop(). \
246 */ 183 */ \
247 netdev_txq_completed_mb(txq, p 184 netdev_txq_completed_mb(txq, pkts, bytes); \
248 185 \
249 _res = -1; 186 _res = -1; \
250 if (pkts && likely(get_desc >= 187 if (pkts && likely(get_desc >= start_thrs)) { \
251 _res = 1; 188 _res = 1; \
252 if (unlikely(netif_tx_ 189 if (unlikely(netif_tx_queue_stopped(txq)) && \
253 !(down_cond)) { 190 !(down_cond)) { \
254 netif_tx_wake_ 191 netif_tx_wake_queue(txq); \
255 _res = 0; 192 _res = 0; \
256 } 193 } \
257 } 194 } \
258 _res; 195 _res; \
259 }) 196 })
260 197
261 #define netif_txq_completed_wake(txq, pkts, by 198 #define netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs) \
262 __netif_txq_completed_wake(txq, pkts, 199 __netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs, false)
263 200
264 /* subqueue variants follow */ 201 /* subqueue variants follow */
265 202
266 #define netif_subqueue_try_stop(dev, idx, get_ 203 #define netif_subqueue_try_stop(dev, idx, get_desc, start_thrs) \
267 ({ 204 ({ \
268 struct netdev_queue *txq; 205 struct netdev_queue *txq; \
269 206 \
270 txq = netdev_get_tx_queue(dev, 207 txq = netdev_get_tx_queue(dev, idx); \
271 netif_txq_try_stop(txq, get_de 208 netif_txq_try_stop(txq, get_desc, start_thrs); \
272 }) 209 })
273 210
274 #define netif_subqueue_maybe_stop(dev, idx, ge 211 #define netif_subqueue_maybe_stop(dev, idx, get_desc, stop_thrs, start_thrs) \
275 ({ 212 ({ \
276 struct netdev_queue *txq; 213 struct netdev_queue *txq; \
277 214 \
278 txq = netdev_get_tx_queue(dev, 215 txq = netdev_get_tx_queue(dev, idx); \
279 netif_txq_maybe_stop(txq, get_ 216 netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs); \
280 }) 217 })
281 218
282 #define netif_subqueue_completed_wake(dev, idx 219 #define netif_subqueue_completed_wake(dev, idx, pkts, bytes, \
283 get_desc 220 get_desc, start_thrs) \
284 ({ 221 ({ \
285 struct netdev_queue *txq; 222 struct netdev_queue *txq; \
286 223 \
287 txq = netdev_get_tx_queue(dev, 224 txq = netdev_get_tx_queue(dev, idx); \
288 netif_txq_completed_wake(txq, 225 netif_txq_completed_wake(txq, pkts, bytes, \
289 get_d 226 get_desc, start_thrs); \
290 }) 227 })
291 228
292 #endif 229 #endif
293 230
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.