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