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 /** 109 * struct netdev_queue_mgmt_ops - netdev ops for queue management 110 * 111 * @ndo_queue_mem_size: Size of the struct that describes a queue's memory. 112 * 113 * @ndo_queue_mem_alloc: Allocate memory for an RX queue at the specified index. 114 * The new memory is written at the specified address. 115 * 116 * @ndo_queue_mem_free: Free memory from an RX queue. 117 * 118 * @ndo_queue_start: Start an RX queue with the specified memory and at the 119 * specified index. 120 * 121 * @ndo_queue_stop: Stop the RX queue at the specified index. The stopped 122 * queue's memory is written at the specified address. 123 * 124 * Note that @ndo_queue_mem_alloc and @ndo_queue_mem_free may be called while 125 * the interface is closed. @ndo_queue_start and @ndo_queue_stop will only 126 * be called for an interface which is open. 127 */ 128 struct netdev_queue_mgmt_ops { 129 size_t ndo_queue_mem_size; 130 int (*ndo_queue_mem_alloc)(struct net_device *dev, 131 void *per_queue_mem, 132 int idx); 133 void (*ndo_queue_mem_free)(struct net_device *dev, 134 void *per_queue_mem); 135 int (*ndo_queue_start)(struct net_device *dev, 136 void *per_queue_mem, 137 int idx); 138 int (*ndo_queue_stop)(struct net_device *dev, 139 void *per_queue_mem, 140 int idx); 141 }; 142 143 /** 144 * DOC: Lockless queue stopping / waking helpers. 145 * 146 * The netif_txq_maybe_stop() and __netif_txq_completed_wake() 147 * macros are designed to safely implement stopping 148 * and waking netdev queues without full lock protection. 149 * 150 * We assume that there can be no concurrent stop attempts and no concurrent 151 * wake attempts. The try-stop should happen from the xmit handler, 152 * while wake up should be triggered from NAPI poll context. 153 * The two may run concurrently (single producer, single consumer). 154 * 155 * The try-stop side is expected to run from the xmit handler and therefore 156 * it does not reschedule Tx (netif_tx_start_queue() instead of 157 * netif_tx_wake_queue()). Uses of the ``stop`` macros outside of the xmit 158 * handler may lead to xmit queue being enabled but not run. 159 * The waking side does not have similar context restrictions. 160 * 161 * The macros guarantee that rings will not remain stopped if there's 162 * space available, but they do *not* prevent false wake ups when 163 * the ring is full! Drivers should check for ring full at the start 164 * for the xmit handler. 165 * 166 * All descriptor ring indexes (and other relevant shared state) must 167 * be updated before invoking the macros. 168 */ 169 170 #define netif_txq_try_stop(txq, get_desc, start_thrs) \ 171 ({ \ 172 int _res; \ 173 \ 174 netif_tx_stop_queue(txq); \ 175 /* Producer index and stop bit must be visible \ 176 * to consumer before we recheck. \ 177 * Pairs with a barrier in __netif_txq_completed_wake(). \ 178 */ \ 179 smp_mb__after_atomic(); \ 180 \ 181 /* We need to check again in a case another \ 182 * CPU has just made room available. \ 183 */ \ 184 _res = 0; \ 185 if (unlikely(get_desc >= start_thrs)) { \ 186 netif_tx_start_queue(txq); \ 187 _res = -1; \ 188 } \ 189 _res; \ 190 }) \ 191 192 /** 193 * netif_txq_maybe_stop() - locklessly stop a Tx queue, if needed 194 * @txq: struct netdev_queue to stop/start 195 * @get_desc: get current number of free descriptors (see requirements below!) 196 * @stop_thrs: minimal number of available descriptors for queue to be left 197 * enabled 198 * @start_thrs: minimal number of descriptors to re-enable the queue, can be 199 * equal to @stop_thrs or higher to avoid frequent waking 200 * 201 * All arguments may be evaluated multiple times, beware of side effects. 202 * @get_desc must be a formula or a function call, it must always 203 * return up-to-date information when evaluated! 204 * Expected to be used from ndo_start_xmit, see the comment on top of the file. 205 * 206 * Returns: 207 * 0 if the queue was stopped 208 * 1 if the queue was left enabled 209 * -1 if the queue was re-enabled (raced with waking) 210 */ 211 #define netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs) \ 212 ({ \ 213 int _res; \ 214 \ 215 _res = 1; \ 216 if (unlikely(get_desc < stop_thrs)) \ 217 _res = netif_txq_try_stop(txq, get_desc, start_thrs); \ 218 _res; \ 219 }) \ 220 221 /* Variant of netdev_tx_completed_queue() which guarantees smp_mb() if 222 * @bytes != 0, regardless of kernel config. 223 */ 224 static inline void 225 netdev_txq_completed_mb(struct netdev_queue *dev_queue, 226 unsigned int pkts, unsigned int bytes) 227 { 228 if (IS_ENABLED(CONFIG_BQL)) 229 netdev_tx_completed_queue(dev_queue, pkts, bytes); 230 else if (bytes) 231 smp_mb(); 232 } 233 234 /** 235 * __netif_txq_completed_wake() - locklessly wake a Tx queue, if needed 236 * @txq: struct netdev_queue to stop/start 237 * @pkts: number of packets completed 238 * @bytes: number of bytes completed 239 * @get_desc: get current number of free descriptors (see requirements below!) 240 * @start_thrs: minimal number of descriptors to re-enable the queue 241 * @down_cond: down condition, predicate indicating that the queue should 242 * not be woken up even if descriptors are available 243 * 244 * All arguments may be evaluated multiple times. 245 * @get_desc must be a formula or a function call, it must always 246 * return up-to-date information when evaluated! 247 * Reports completed pkts/bytes to BQL. 248 * 249 * Returns: 250 * 0 if the queue was woken up 251 * 1 if the queue was already enabled (or disabled but @down_cond is true) 252 * -1 if the queue was left unchanged (@start_thrs not reached) 253 */ 254 #define __netif_txq_completed_wake(txq, pkts, bytes, \ 255 get_desc, start_thrs, down_cond) \ 256 ({ \ 257 int _res; \ 258 \ 259 /* Report to BQL and piggy back on its barrier. \ 260 * Barrier makes sure that anybody stopping the queue \ 261 * after this point sees the new consumer index. \ 262 * Pairs with barrier in netif_txq_try_stop(). \ 263 */ \ 264 netdev_txq_completed_mb(txq, pkts, bytes); \ 265 \ 266 _res = -1; \ 267 if (pkts && likely(get_desc >= start_thrs)) { \ 268 _res = 1; \ 269 if (unlikely(netif_tx_queue_stopped(txq)) && \ 270 !(down_cond)) { \ 271 netif_tx_wake_queue(txq); \ 272 _res = 0; \ 273 } \ 274 } \ 275 _res; \ 276 }) 277 278 #define netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs) \ 279 __netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs, false) 280 281 /* subqueue variants follow */ 282 283 #define netif_subqueue_try_stop(dev, idx, get_desc, start_thrs) \ 284 ({ \ 285 struct netdev_queue *txq; \ 286 \ 287 txq = netdev_get_tx_queue(dev, idx); \ 288 netif_txq_try_stop(txq, get_desc, start_thrs); \ 289 }) 290 291 #define netif_subqueue_maybe_stop(dev, idx, get_desc, stop_thrs, start_thrs) \ 292 ({ \ 293 struct netdev_queue *txq; \ 294 \ 295 txq = netdev_get_tx_queue(dev, idx); \ 296 netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs); \ 297 }) 298 299 #define netif_subqueue_completed_wake(dev, idx, pkts, bytes, \ 300 get_desc, start_thrs) \ 301 ({ \ 302 struct netdev_queue *txq; \ 303 \ 304 txq = netdev_get_tx_queue(dev, idx); \ 305 netif_txq_completed_wake(txq, pkts, bytes, \ 306 get_desc, start_thrs); \ 307 }) 308 309 #endif 310