1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_NET_QUEUES_H 3 #define _LINUX_NET_QUEUES_H 4 5 #include <linux/netdevice.h> 6 7 /** 8 * struct netdev_config - queue-related configuration for a netdev 9 * @hds_thresh: HDS Threshold value. 10 * @hds_config: HDS value from userspace. 11 */ 12 struct netdev_config { 13 u32 hds_thresh; 14 u8 hds_config; 15 }; 16 17 /* See the netdev.yaml spec for definition of each statistic */ 18 struct netdev_queue_stats_rx { 19 u64 bytes; 20 u64 packets; 21 u64 alloc_fail; 22 23 u64 hw_drops; 24 u64 hw_drop_overruns; 25 26 u64 csum_complete; 27 u64 csum_unnecessary; 28 u64 csum_none; 29 u64 csum_bad; 30 31 u64 hw_gro_packets; 32 u64 hw_gro_bytes; 33 u64 hw_gro_wire_packets; 34 u64 hw_gro_wire_bytes; 35 36 u64 hw_drop_ratelimits; 37 }; 38 39 struct netdev_queue_stats_tx { 40 u64 bytes; 41 u64 packets; 42 43 u64 hw_drops; 44 u64 hw_drop_errors; 45 46 u64 csum_none; 47 u64 needs_csum; 48 49 u64 hw_gso_packets; 50 u64 hw_gso_bytes; 51 u64 hw_gso_wire_packets; 52 u64 hw_gso_wire_bytes; 53 54 u64 hw_drop_ratelimits; 55 56 u64 stop; 57 u64 wake; 58 }; 59 60 /** 61 * struct netdev_stat_ops - netdev ops for fine grained stats 62 * @get_queue_stats_rx: get stats for a given Rx queue 63 * @get_queue_stats_tx: get stats for a given Tx queue 64 * @get_base_stats: get base stats (not belonging to any live instance) 65 * 66 * Query stats for a given object. The values of the statistics are undefined 67 * on entry (specifically they are *not* zero-initialized). Drivers should 68 * assign values only to the statistics they collect. Statistics which are not 69 * collected must be left undefined. 70 * 71 * Queue objects are not necessarily persistent, and only currently active 72 * queues are queried by the per-queue callbacks. This means that per-queue 73 * statistics will not generally add up to the total number of events for 74 * the device. The @get_base_stats callback allows filling in the delta 75 * between events for currently live queues and overall device history. 76 * @get_base_stats can also be used to report any miscellaneous packets 77 * transferred outside of the main set of queues used by the networking stack. 78 * When the statistics for the entire device are queried, first @get_base_stats 79 * is issued to collect the delta, and then a series of per-queue callbacks. 80 * Only statistics which are set in @get_base_stats will be reported 81 * at the device level, meaning that unlike in queue callbacks, setting 82 * a statistic to zero in @get_base_stats is a legitimate thing to do. 83 * This is because @get_base_stats has a second function of designating which 84 * statistics are in fact correct for the entire device (e.g. when history 85 * for some of the events is not maintained, and reliable "total" cannot 86 * be provided). 87 * 88 * Ops are called under the instance lock if netdev_need_ops_lock() 89 * returns true, otherwise under rtnl_lock. 90 * Device drivers can assume that when collecting total device stats, 91 * the @get_base_stats and subsequent per-queue calls are performed 92 * "atomically" (without releasing the relevant lock). 93 * 94 * Device drivers are encouraged to reset the per-queue statistics when 95 * number of queues change. This is because the primary use case for 96 * per-queue statistics is currently to detect traffic imbalance. 97 */ 98 struct netdev_stat_ops { 99 void (*get_queue_stats_rx)(struct net_device *dev, int idx, 100 struct netdev_queue_stats_rx *stats); 101 void (*get_queue_stats_tx)(struct net_device *dev, int idx, 102 struct netdev_queue_stats_tx *stats); 103 void (*get_base_stats)(struct net_device *dev, 104 struct netdev_queue_stats_rx *rx, 105 struct netdev_queue_stats_tx *tx); 106 }; 107 108 void netdev_stat_queue_sum(struct net_device *netdev, 109 int rx_start, int rx_end, 110 struct netdev_queue_stats_rx *rx_sum, 111 int tx_start, int tx_end, 112 struct netdev_queue_stats_tx *tx_sum); 113 114 /** 115 * struct netdev_queue_mgmt_ops - netdev ops for queue management 116 * 117 * @ndo_queue_mem_size: Size of the struct that describes a queue's memory. 118 * 119 * @ndo_queue_mem_alloc: Allocate memory for an RX queue at the specified index. 120 * The new memory is written at the specified address. 121 * 122 * @ndo_queue_mem_free: Free memory from an RX queue. 123 * 124 * @ndo_queue_start: Start an RX queue with the specified memory and at the 125 * specified index. 126 * 127 * @ndo_queue_stop: Stop the RX queue at the specified index. The stopped 128 * queue's memory is written at the specified address. 129 * 130 * Note that @ndo_queue_mem_alloc and @ndo_queue_mem_free may be called while 131 * the interface is closed. @ndo_queue_start and @ndo_queue_stop will only 132 * be called for an interface which is open. 133 */ 134 struct netdev_queue_mgmt_ops { 135 size_t ndo_queue_mem_size; 136 int (*ndo_queue_mem_alloc)(struct net_device *dev, 137 void *per_queue_mem, 138 int idx); 139 void (*ndo_queue_mem_free)(struct net_device *dev, 140 void *per_queue_mem); 141 int (*ndo_queue_start)(struct net_device *dev, 142 void *per_queue_mem, 143 int idx); 144 int (*ndo_queue_stop)(struct net_device *dev, 145 void *per_queue_mem, 146 int idx); 147 }; 148 149 /** 150 * DOC: Lockless queue stopping / waking helpers. 151 * 152 * The netif_txq_maybe_stop() and __netif_txq_completed_wake() 153 * macros are designed to safely implement stopping 154 * and waking netdev queues without full lock protection. 155 * 156 * We assume that there can be no concurrent stop attempts and no concurrent 157 * wake attempts. The try-stop should happen from the xmit handler, 158 * while wake up should be triggered from NAPI poll context. 159 * The two may run concurrently (single producer, single consumer). 160 * 161 * The try-stop side is expected to run from the xmit handler and therefore 162 * it does not reschedule Tx (netif_tx_start_queue() instead of 163 * netif_tx_wake_queue()). Uses of the ``stop`` macros outside of the xmit 164 * handler may lead to xmit queue being enabled but not run. 165 * The waking side does not have similar context restrictions. 166 * 167 * The macros guarantee that rings will not remain stopped if there's 168 * space available, but they do *not* prevent false wake ups when 169 * the ring is full! Drivers should check for ring full at the start 170 * for the xmit handler. 171 * 172 * All descriptor ring indexes (and other relevant shared state) must 173 * be updated before invoking the macros. 174 */ 175 176 #define netif_txq_try_stop(txq, get_desc, start_thrs) \ 177 ({ \ 178 int _res; \ 179 \ 180 netif_tx_stop_queue(txq); \ 181 /* Producer index and stop bit must be visible \ 182 * to consumer before we recheck. \ 183 * Pairs with a barrier in __netif_txq_completed_wake(). \ 184 */ \ 185 smp_mb__after_atomic(); \ 186 \ 187 /* We need to check again in a case another \ 188 * CPU has just made room available. \ 189 */ \ 190 _res = 0; \ 191 if (unlikely(get_desc >= start_thrs)) { \ 192 netif_tx_start_queue(txq); \ 193 _res = -1; \ 194 } \ 195 _res; \ 196 }) \ 197 198 /** 199 * netif_txq_maybe_stop() - locklessly stop a Tx queue, if needed 200 * @txq: struct netdev_queue to stop/start 201 * @get_desc: get current number of free descriptors (see requirements below!) 202 * @stop_thrs: minimal number of available descriptors for queue to be left 203 * enabled 204 * @start_thrs: minimal number of descriptors to re-enable the queue, can be 205 * equal to @stop_thrs or higher to avoid frequent waking 206 * 207 * All arguments may be evaluated multiple times, beware of side effects. 208 * @get_desc must be a formula or a function call, it must always 209 * return up-to-date information when evaluated! 210 * Expected to be used from ndo_start_xmit, see the comment on top of the file. 211 * 212 * Returns: 213 * 0 if the queue was stopped 214 * 1 if the queue was left enabled 215 * -1 if the queue was re-enabled (raced with waking) 216 */ 217 #define netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs) \ 218 ({ \ 219 int _res; \ 220 \ 221 _res = 1; \ 222 if (unlikely(get_desc < stop_thrs)) \ 223 _res = netif_txq_try_stop(txq, get_desc, start_thrs); \ 224 _res; \ 225 }) \ 226 227 /* Variant of netdev_tx_completed_queue() which guarantees smp_mb() if 228 * @bytes != 0, regardless of kernel config. 229 */ 230 static inline void 231 netdev_txq_completed_mb(struct netdev_queue *dev_queue, 232 unsigned int pkts, unsigned int bytes) 233 { 234 if (IS_ENABLED(CONFIG_BQL)) 235 netdev_tx_completed_queue(dev_queue, pkts, bytes); 236 else if (bytes) 237 smp_mb(); 238 } 239 240 /** 241 * __netif_txq_completed_wake() - locklessly wake a Tx queue, if needed 242 * @txq: struct netdev_queue to stop/start 243 * @pkts: number of packets completed 244 * @bytes: number of bytes completed 245 * @get_desc: get current number of free descriptors (see requirements below!) 246 * @start_thrs: minimal number of descriptors to re-enable the queue 247 * @down_cond: down condition, predicate indicating that the queue should 248 * not be woken up even if descriptors are available 249 * 250 * All arguments may be evaluated multiple times. 251 * @get_desc must be a formula or a function call, it must always 252 * return up-to-date information when evaluated! 253 * Reports completed pkts/bytes to BQL. 254 * 255 * Returns: 256 * 0 if the queue was woken up 257 * 1 if the queue was already enabled (or disabled but @down_cond is true) 258 * -1 if the queue was left unchanged (@start_thrs not reached) 259 */ 260 #define __netif_txq_completed_wake(txq, pkts, bytes, \ 261 get_desc, start_thrs, down_cond) \ 262 ({ \ 263 int _res; \ 264 \ 265 /* Report to BQL and piggy back on its barrier. \ 266 * Barrier makes sure that anybody stopping the queue \ 267 * after this point sees the new consumer index. \ 268 * Pairs with barrier in netif_txq_try_stop(). \ 269 */ \ 270 netdev_txq_completed_mb(txq, pkts, bytes); \ 271 \ 272 _res = -1; \ 273 if (pkts && likely(get_desc >= start_thrs)) { \ 274 _res = 1; \ 275 if (unlikely(netif_tx_queue_stopped(txq)) && \ 276 !(down_cond)) { \ 277 netif_tx_wake_queue(txq); \ 278 _res = 0; \ 279 } \ 280 } \ 281 _res; \ 282 }) 283 284 #define netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs) \ 285 __netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs, false) 286 287 /* subqueue variants follow */ 288 289 #define netif_subqueue_try_stop(dev, idx, get_desc, start_thrs) \ 290 ({ \ 291 struct netdev_queue *_txq; \ 292 \ 293 _txq = netdev_get_tx_queue(dev, idx); \ 294 netif_txq_try_stop(_txq, get_desc, start_thrs); \ 295 }) 296 297 #define netif_subqueue_maybe_stop(dev, idx, get_desc, stop_thrs, start_thrs) \ 298 ({ \ 299 struct netdev_queue *_txq; \ 300 \ 301 _txq = netdev_get_tx_queue(dev, idx); \ 302 netif_txq_maybe_stop(_txq, get_desc, stop_thrs, start_thrs); \ 303 }) 304 305 #define netif_subqueue_completed_wake(dev, idx, pkts, bytes, \ 306 get_desc, start_thrs) \ 307 ({ \ 308 struct netdev_queue *_txq; \ 309 \ 310 _txq = netdev_get_tx_queue(dev, idx); \ 311 netif_txq_completed_wake(_txq, pkts, bytes, \ 312 get_desc, start_thrs); \ 313 }) 314 315 #endif 316