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 * Device drivers can assume that when collecting total device stats, 89 * the @get_base_stats and subsequent per-queue calls are performed 90 * "atomically" (without releasing the rtnl_lock). 91 * 92 * Device drivers are encouraged to reset the per-queue statistics when 93 * number of queues change. This is because the primary use case for 94 * per-queue statistics is currently to detect traffic imbalance. 95 */ 96 struct netdev_stat_ops { 97 void (*get_queue_stats_rx)(struct net_device *dev, int idx, 98 struct netdev_queue_stats_rx *stats); 99 void (*get_queue_stats_tx)(struct net_device *dev, int idx, 100 struct netdev_queue_stats_tx *stats); 101 void (*get_base_stats)(struct net_device *dev, 102 struct netdev_queue_stats_rx *rx, 103 struct netdev_queue_stats_tx *tx); 104 }; 105 106 void netdev_stat_queue_sum(struct net_device *netdev, 107 int rx_start, int rx_end, 108 struct netdev_queue_stats_rx *rx_sum, 109 int tx_start, int tx_end, 110 struct netdev_queue_stats_tx *tx_sum); 111 112 /** 113 * struct netdev_queue_mgmt_ops - netdev ops for queue management 114 * 115 * @ndo_queue_mem_size: Size of the struct that describes a queue's memory. 116 * 117 * @ndo_queue_mem_alloc: Allocate memory for an RX queue at the specified index. 118 * The new memory is written at the specified address. 119 * 120 * @ndo_queue_mem_free: Free memory from an RX queue. 121 * 122 * @ndo_queue_start: Start an RX queue with the specified memory and at the 123 * specified index. 124 * 125 * @ndo_queue_stop: Stop the RX queue at the specified index. The stopped 126 * queue's memory is written at the specified address. 127 * 128 * Note that @ndo_queue_mem_alloc and @ndo_queue_mem_free may be called while 129 * the interface is closed. @ndo_queue_start and @ndo_queue_stop will only 130 * be called for an interface which is open. 131 */ 132 struct netdev_queue_mgmt_ops { 133 size_t ndo_queue_mem_size; 134 int (*ndo_queue_mem_alloc)(struct net_device *dev, 135 void *per_queue_mem, 136 int idx); 137 void (*ndo_queue_mem_free)(struct net_device *dev, 138 void *per_queue_mem); 139 int (*ndo_queue_start)(struct net_device *dev, 140 void *per_queue_mem, 141 int idx); 142 int (*ndo_queue_stop)(struct net_device *dev, 143 void *per_queue_mem, 144 int idx); 145 }; 146 147 /** 148 * DOC: Lockless queue stopping / waking helpers. 149 * 150 * The netif_txq_maybe_stop() and __netif_txq_completed_wake() 151 * macros are designed to safely implement stopping 152 * and waking netdev queues without full lock protection. 153 * 154 * We assume that there can be no concurrent stop attempts and no concurrent 155 * wake attempts. The try-stop should happen from the xmit handler, 156 * while wake up should be triggered from NAPI poll context. 157 * The two may run concurrently (single producer, single consumer). 158 * 159 * The try-stop side is expected to run from the xmit handler and therefore 160 * it does not reschedule Tx (netif_tx_start_queue() instead of 161 * netif_tx_wake_queue()). Uses of the ``stop`` macros outside of the xmit 162 * handler may lead to xmit queue being enabled but not run. 163 * The waking side does not have similar context restrictions. 164 * 165 * The macros guarantee that rings will not remain stopped if there's 166 * space available, but they do *not* prevent false wake ups when 167 * the ring is full! Drivers should check for ring full at the start 168 * for the xmit handler. 169 * 170 * All descriptor ring indexes (and other relevant shared state) must 171 * be updated before invoking the macros. 172 */ 173 174 #define netif_txq_try_stop(txq, get_desc, start_thrs) \ 175 ({ \ 176 int _res; \ 177 \ 178 netif_tx_stop_queue(txq); \ 179 /* Producer index and stop bit must be visible \ 180 * to consumer before we recheck. \ 181 * Pairs with a barrier in __netif_txq_completed_wake(). \ 182 */ \ 183 smp_mb__after_atomic(); \ 184 \ 185 /* We need to check again in a case another \ 186 * CPU has just made room available. \ 187 */ \ 188 _res = 0; \ 189 if (unlikely(get_desc >= start_thrs)) { \ 190 netif_tx_start_queue(txq); \ 191 _res = -1; \ 192 } \ 193 _res; \ 194 }) \ 195 196 /** 197 * netif_txq_maybe_stop() - locklessly stop a Tx queue, if needed 198 * @txq: struct netdev_queue to stop/start 199 * @get_desc: get current number of free descriptors (see requirements below!) 200 * @stop_thrs: minimal number of available descriptors for queue to be left 201 * enabled 202 * @start_thrs: minimal number of descriptors to re-enable the queue, can be 203 * equal to @stop_thrs or higher to avoid frequent waking 204 * 205 * All arguments may be evaluated multiple times, beware of side effects. 206 * @get_desc must be a formula or a function call, it must always 207 * return up-to-date information when evaluated! 208 * Expected to be used from ndo_start_xmit, see the comment on top of the file. 209 * 210 * Returns: 211 * 0 if the queue was stopped 212 * 1 if the queue was left enabled 213 * -1 if the queue was re-enabled (raced with waking) 214 */ 215 #define netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs) \ 216 ({ \ 217 int _res; \ 218 \ 219 _res = 1; \ 220 if (unlikely(get_desc < stop_thrs)) \ 221 _res = netif_txq_try_stop(txq, get_desc, start_thrs); \ 222 _res; \ 223 }) \ 224 225 /* Variant of netdev_tx_completed_queue() which guarantees smp_mb() if 226 * @bytes != 0, regardless of kernel config. 227 */ 228 static inline void 229 netdev_txq_completed_mb(struct netdev_queue *dev_queue, 230 unsigned int pkts, unsigned int bytes) 231 { 232 if (IS_ENABLED(CONFIG_BQL)) 233 netdev_tx_completed_queue(dev_queue, pkts, bytes); 234 else if (bytes) 235 smp_mb(); 236 } 237 238 /** 239 * __netif_txq_completed_wake() - locklessly wake a Tx queue, if needed 240 * @txq: struct netdev_queue to stop/start 241 * @pkts: number of packets completed 242 * @bytes: number of bytes completed 243 * @get_desc: get current number of free descriptors (see requirements below!) 244 * @start_thrs: minimal number of descriptors to re-enable the queue 245 * @down_cond: down condition, predicate indicating that the queue should 246 * not be woken up even if descriptors are available 247 * 248 * All arguments may be evaluated multiple times. 249 * @get_desc must be a formula or a function call, it must always 250 * return up-to-date information when evaluated! 251 * Reports completed pkts/bytes to BQL. 252 * 253 * Returns: 254 * 0 if the queue was woken up 255 * 1 if the queue was already enabled (or disabled but @down_cond is true) 256 * -1 if the queue was left unchanged (@start_thrs not reached) 257 */ 258 #define __netif_txq_completed_wake(txq, pkts, bytes, \ 259 get_desc, start_thrs, down_cond) \ 260 ({ \ 261 int _res; \ 262 \ 263 /* Report to BQL and piggy back on its barrier. \ 264 * Barrier makes sure that anybody stopping the queue \ 265 * after this point sees the new consumer index. \ 266 * Pairs with barrier in netif_txq_try_stop(). \ 267 */ \ 268 netdev_txq_completed_mb(txq, pkts, bytes); \ 269 \ 270 _res = -1; \ 271 if (pkts && likely(get_desc >= start_thrs)) { \ 272 _res = 1; \ 273 if (unlikely(netif_tx_queue_stopped(txq)) && \ 274 !(down_cond)) { \ 275 netif_tx_wake_queue(txq); \ 276 _res = 0; \ 277 } \ 278 } \ 279 _res; \ 280 }) 281 282 #define netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs) \ 283 __netif_txq_completed_wake(txq, pkts, bytes, get_desc, start_thrs, false) 284 285 /* subqueue variants follow */ 286 287 #define netif_subqueue_try_stop(dev, idx, get_desc, start_thrs) \ 288 ({ \ 289 struct netdev_queue *txq; \ 290 \ 291 txq = netdev_get_tx_queue(dev, idx); \ 292 netif_txq_try_stop(txq, get_desc, start_thrs); \ 293 }) 294 295 #define netif_subqueue_maybe_stop(dev, idx, get_desc, stop_thrs, start_thrs) \ 296 ({ \ 297 struct netdev_queue *txq; \ 298 \ 299 txq = netdev_get_tx_queue(dev, idx); \ 300 netif_txq_maybe_stop(txq, get_desc, stop_thrs, start_thrs); \ 301 }) 302 303 #define netif_subqueue_completed_wake(dev, idx, pkts, bytes, \ 304 get_desc, start_thrs) \ 305 ({ \ 306 struct netdev_queue *txq; \ 307 \ 308 txq = netdev_get_tx_queue(dev, idx); \ 309 netif_txq_completed_wake(txq, pkts, bytes, \ 310 get_desc, start_thrs); \ 311 }) 312 313 #endif 314