1 /* SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) */ 2 /* Copyright (C) 2015-2018 Netronome Systems, Inc. */ 3 4 /* 5 * nfp_net.h 6 * Declarations for Netronome network device driver. 7 * Authors: Jakub Kicinski <jakub.kicinski@netronome.com> 8 * Jason McMullan <jason.mcmullan@netronome.com> 9 * Rolf Neugebauer <rolf.neugebauer@netronome.com> 10 */ 11 12 #ifndef _NFP_NET_H_ 13 #define _NFP_NET_H_ 14 15 #include <linux/atomic.h> 16 #include <linux/interrupt.h> 17 #include <linux/list.h> 18 #include <linux/netdevice.h> 19 #include <linux/pci.h> 20 #include <linux/dim.h> 21 #include <linux/io-64-nonatomic-hi-lo.h> 22 #include <linux/semaphore.h> 23 #include <linux/workqueue.h> 24 #include <net/xdp.h> 25 26 #include "nfp_net_ctrl.h" 27 28 #define nn_pr(nn, lvl, fmt, args...) \ 29 ({ \ 30 struct nfp_net *__nn = (nn); \ 31 \ 32 if (__nn->dp.netdev) \ 33 netdev_printk(lvl, __nn->dp.netdev, fmt, ## args); \ 34 else \ 35 dev_printk(lvl, __nn->dp.dev, "ctrl: " fmt, ## args); \ 36 }) 37 38 #define nn_err(nn, fmt, args...) nn_pr(nn, KERN_ERR, fmt, ## args) 39 #define nn_warn(nn, fmt, args...) nn_pr(nn, KERN_WARNING, fmt, ## args) 40 #define nn_info(nn, fmt, args...) nn_pr(nn, KERN_INFO, fmt, ## args) 41 #define nn_dbg(nn, fmt, args...) nn_pr(nn, KERN_DEBUG, fmt, ## args) 42 43 #define nn_dp_warn(dp, fmt, args...) \ 44 ({ \ 45 struct nfp_net_dp *__dp = (dp); \ 46 \ 47 if (unlikely(net_ratelimit())) { \ 48 if (__dp->netdev) \ 49 netdev_warn(__dp->netdev, fmt, ## args); \ 50 else \ 51 dev_warn(__dp->dev, fmt, ## args); \ 52 } \ 53 }) 54 55 /* Max time to wait for NFP to respond on updates (in seconds) */ 56 #define NFP_NET_POLL_TIMEOUT 5 57 58 /* Interval for reading offloaded filter stats */ 59 #define NFP_NET_STAT_POLL_IVL msecs_to_jiffies(100) 60 61 /* Bar allocation */ 62 #define NFP_NET_CTRL_BAR 0 63 #define NFP_NET_Q0_BAR 2 64 #define NFP_NET_Q1_BAR 4 /* OBSOLETE */ 65 66 /* Max bits in DMA address */ 67 #define NFP_NET_MAX_DMA_BITS 40 68 69 /* Default size for MTU and freelist buffer sizes */ 70 #define NFP_NET_DEFAULT_MTU 1500U 71 72 /* Maximum number of bytes prepended to a packet */ 73 #define NFP_NET_MAX_PREPEND 64 74 75 /* Interrupt definitions */ 76 #define NFP_NET_NON_Q_VECTORS 2 77 #define NFP_NET_IRQ_LSC_IDX 0 78 #define NFP_NET_IRQ_EXN_IDX 1 79 #define NFP_NET_MIN_VNIC_IRQS (NFP_NET_NON_Q_VECTORS + 1) 80 81 /* Queue/Ring definitions */ 82 #define NFP_NET_MAX_TX_RINGS 64 /* Max. # of Tx rings per device */ 83 #define NFP_NET_MAX_RX_RINGS 64 /* Max. # of Rx rings per device */ 84 #define NFP_NET_MAX_R_VECS (NFP_NET_MAX_TX_RINGS > NFP_NET_MAX_RX_RINGS ? \ 85 NFP_NET_MAX_TX_RINGS : NFP_NET_MAX_RX_RINGS) 86 #define NFP_NET_MAX_IRQS (NFP_NET_NON_Q_VECTORS + NFP_NET_MAX_R_VECS) 87 88 #define NFP_NET_MIN_TX_DESCS 256 /* Min. # of Tx descs per ring */ 89 #define NFP_NET_MIN_RX_DESCS 256 /* Min. # of Rx descs per ring */ 90 #define NFP_NET_MAX_TX_DESCS (256 * 1024) /* Max. # of Tx descs per ring */ 91 #define NFP_NET_MAX_RX_DESCS (256 * 1024) /* Max. # of Rx descs per ring */ 92 93 #define NFP_NET_TX_DESCS_DEFAULT 4096 /* Default # of Tx descs per ring */ 94 #define NFP_NET_RX_DESCS_DEFAULT 4096 /* Default # of Rx descs per ring */ 95 96 #define NFP_NET_FL_BATCH 16 /* Add freelist in this Batch size */ 97 #define NFP_NET_XDP_MAX_COMPLETE 2048 /* XDP bufs to reclaim in NAPI poll */ 98 99 /* Offload definitions */ 100 #define NFP_NET_N_VXLAN_PORTS (NFP_NET_CFG_VXLAN_SZ / sizeof(__be16)) 101 102 #define NFP_NET_RX_BUF_HEADROOM (NET_SKB_PAD + NET_IP_ALIGN) 103 #define NFP_NET_RX_BUF_NON_DATA (NFP_NET_RX_BUF_HEADROOM + \ 104 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 105 106 /* Forward declarations */ 107 struct nfp_cpp; 108 struct nfp_eth_table_port; 109 struct nfp_net; 110 struct nfp_net_r_vector; 111 struct nfp_port; 112 struct xsk_buff_pool; 113 114 /* Convenience macro for wrapping descriptor index on ring size */ 115 #define D_IDX(ring, idx) ((idx) & ((ring)->cnt - 1)) 116 117 /* Convenience macro for writing dma address into RX/TX descriptors */ 118 #define nfp_desc_set_dma_addr(desc, dma_addr) \ 119 do { \ 120 __typeof(desc) __d = (desc); \ 121 dma_addr_t __addr = (dma_addr); \ 122 \ 123 __d->dma_addr_lo = cpu_to_le32(lower_32_bits(__addr)); \ 124 __d->dma_addr_hi = upper_32_bits(__addr) & 0xff; \ 125 } while (0) 126 127 /* TX descriptor format */ 128 129 #define PCIE_DESC_TX_EOP BIT(7) 130 #define PCIE_DESC_TX_OFFSET_MASK GENMASK(6, 0) 131 #define PCIE_DESC_TX_MSS_MASK GENMASK(13, 0) 132 133 /* Flags in the host TX descriptor */ 134 #define PCIE_DESC_TX_CSUM BIT(7) 135 #define PCIE_DESC_TX_IP4_CSUM BIT(6) 136 #define PCIE_DESC_TX_TCP_CSUM BIT(5) 137 #define PCIE_DESC_TX_UDP_CSUM BIT(4) 138 #define PCIE_DESC_TX_VLAN BIT(3) 139 #define PCIE_DESC_TX_LSO BIT(2) 140 #define PCIE_DESC_TX_ENCAP BIT(1) 141 #define PCIE_DESC_TX_O_IP4_CSUM BIT(0) 142 143 struct nfp_net_tx_desc { 144 union { 145 struct { 146 u8 dma_addr_hi; /* High bits of host buf address */ 147 __le16 dma_len; /* Length to DMA for this desc */ 148 u8 offset_eop; /* Offset in buf where pkt starts + 149 * highest bit is eop flag. 150 */ 151 __le32 dma_addr_lo; /* Low 32bit of host buf addr */ 152 153 __le16 mss; /* MSS to be used for LSO */ 154 u8 lso_hdrlen; /* LSO, TCP payload offset */ 155 u8 flags; /* TX Flags, see @PCIE_DESC_TX_* */ 156 union { 157 struct { 158 u8 l3_offset; /* L3 header offset */ 159 u8 l4_offset; /* L4 header offset */ 160 }; 161 __le16 vlan; /* VLAN tag to add if indicated */ 162 }; 163 __le16 data_len; /* Length of frame + meta data */ 164 } __packed; 165 __le32 vals[4]; 166 __le64 vals8[2]; 167 }; 168 }; 169 170 /** 171 * struct nfp_net_tx_buf - software TX buffer descriptor 172 * @skb: normal ring, sk_buff associated with this buffer 173 * @frag: XDP ring, page frag associated with this buffer 174 * @xdp: XSK buffer pool handle (for AF_XDP) 175 * @dma_addr: DMA mapping address of the buffer 176 * @fidx: Fragment index (-1 for the head and [0..nr_frags-1] for frags) 177 * @pkt_cnt: Number of packets to be produced out of the skb associated 178 * with this buffer (valid only on the head's buffer). 179 * Will be 1 for all non-TSO packets. 180 * @is_xsk_tx: Flag if buffer is a RX buffer after a XDP_TX action and not a 181 * buffer from the TX queue (for AF_XDP). 182 * @real_len: Number of bytes which to be produced out of the skb (valid only 183 * on the head's buffer). Equal to skb->len for non-TSO packets. 184 */ 185 struct nfp_net_tx_buf { 186 union { 187 struct sk_buff *skb; 188 void *frag; 189 struct xdp_buff *xdp; 190 }; 191 dma_addr_t dma_addr; 192 union { 193 struct { 194 short int fidx; 195 u16 pkt_cnt; 196 }; 197 struct { 198 bool is_xsk_tx; 199 }; 200 }; 201 u32 real_len; 202 }; 203 204 /** 205 * struct nfp_net_tx_ring - TX ring structure 206 * @r_vec: Back pointer to ring vector structure 207 * @idx: Ring index from Linux's perspective 208 * @qcidx: Queue Controller Peripheral (QCP) queue index for the TX queue 209 * @qcp_q: Pointer to base of the QCP TX queue 210 * @cnt: Size of the queue in number of descriptors 211 * @wr_p: TX ring write pointer (free running) 212 * @rd_p: TX ring read pointer (free running) 213 * @qcp_rd_p: Local copy of QCP TX queue read pointer 214 * @wr_ptr_add: Accumulated number of buffers to add to QCP write pointer 215 * (used for .xmit_more delayed kick) 216 * @txbufs: Array of transmitted TX buffers, to free on transmit 217 * @txds: Virtual address of TX ring in host memory 218 * @dma: DMA address of the TX ring 219 * @size: Size, in bytes, of the TX ring (needed to free) 220 * @is_xdp: Is this a XDP TX ring? 221 */ 222 struct nfp_net_tx_ring { 223 struct nfp_net_r_vector *r_vec; 224 225 u32 idx; 226 int qcidx; 227 u8 __iomem *qcp_q; 228 229 u32 cnt; 230 u32 wr_p; 231 u32 rd_p; 232 u32 qcp_rd_p; 233 234 u32 wr_ptr_add; 235 236 struct nfp_net_tx_buf *txbufs; 237 struct nfp_net_tx_desc *txds; 238 239 dma_addr_t dma; 240 size_t size; 241 bool is_xdp; 242 } ____cacheline_aligned; 243 244 /* RX and freelist descriptor format */ 245 246 #define PCIE_DESC_RX_DD BIT(7) 247 #define PCIE_DESC_RX_META_LEN_MASK GENMASK(6, 0) 248 249 /* Flags in the RX descriptor */ 250 #define PCIE_DESC_RX_RSS cpu_to_le16(BIT(15)) 251 #define PCIE_DESC_RX_I_IP4_CSUM cpu_to_le16(BIT(14)) 252 #define PCIE_DESC_RX_I_IP4_CSUM_OK cpu_to_le16(BIT(13)) 253 #define PCIE_DESC_RX_I_TCP_CSUM cpu_to_le16(BIT(12)) 254 #define PCIE_DESC_RX_I_TCP_CSUM_OK cpu_to_le16(BIT(11)) 255 #define PCIE_DESC_RX_I_UDP_CSUM cpu_to_le16(BIT(10)) 256 #define PCIE_DESC_RX_I_UDP_CSUM_OK cpu_to_le16(BIT(9)) 257 #define PCIE_DESC_RX_DECRYPTED cpu_to_le16(BIT(8)) 258 #define PCIE_DESC_RX_EOP cpu_to_le16(BIT(7)) 259 #define PCIE_DESC_RX_IP4_CSUM cpu_to_le16(BIT(6)) 260 #define PCIE_DESC_RX_IP4_CSUM_OK cpu_to_le16(BIT(5)) 261 #define PCIE_DESC_RX_TCP_CSUM cpu_to_le16(BIT(4)) 262 #define PCIE_DESC_RX_TCP_CSUM_OK cpu_to_le16(BIT(3)) 263 #define PCIE_DESC_RX_UDP_CSUM cpu_to_le16(BIT(2)) 264 #define PCIE_DESC_RX_UDP_CSUM_OK cpu_to_le16(BIT(1)) 265 #define PCIE_DESC_RX_VLAN cpu_to_le16(BIT(0)) 266 267 #define PCIE_DESC_RX_CSUM_ALL (PCIE_DESC_RX_IP4_CSUM | \ 268 PCIE_DESC_RX_TCP_CSUM | \ 269 PCIE_DESC_RX_UDP_CSUM | \ 270 PCIE_DESC_RX_I_IP4_CSUM | \ 271 PCIE_DESC_RX_I_TCP_CSUM | \ 272 PCIE_DESC_RX_I_UDP_CSUM) 273 #define PCIE_DESC_RX_CSUM_OK_SHIFT 1 274 #define __PCIE_DESC_RX_CSUM_ALL le16_to_cpu(PCIE_DESC_RX_CSUM_ALL) 275 #define __PCIE_DESC_RX_CSUM_ALL_OK (__PCIE_DESC_RX_CSUM_ALL >> \ 276 PCIE_DESC_RX_CSUM_OK_SHIFT) 277 278 struct nfp_net_rx_desc { 279 union { 280 struct { 281 u8 dma_addr_hi; /* High bits of the buf address */ 282 __le16 reserved; /* Must be zero */ 283 u8 meta_len_dd; /* Must be zero */ 284 285 __le32 dma_addr_lo; /* Low bits of the buffer address */ 286 } __packed fld; 287 288 struct { 289 __le16 data_len; /* Length of the frame + meta data */ 290 u8 reserved; 291 u8 meta_len_dd; /* Length of meta data prepended + 292 * descriptor done flag. 293 */ 294 295 __le16 flags; /* RX flags. See @PCIE_DESC_RX_* */ 296 __le16 vlan; /* VLAN if stripped */ 297 } __packed rxd; 298 299 __le32 vals[2]; 300 }; 301 }; 302 303 #define NFP_NET_META_FIELD_MASK GENMASK(NFP_NET_META_FIELD_SIZE - 1, 0) 304 305 struct nfp_meta_parsed { 306 u8 hash_type; 307 u8 csum_type; 308 u32 hash; 309 u32 mark; 310 u32 portid; 311 __wsum csum; 312 }; 313 314 struct nfp_net_rx_hash { 315 __be32 hash_type; 316 __be32 hash; 317 }; 318 319 /** 320 * struct nfp_net_rx_buf - software RX buffer descriptor 321 * @frag: page fragment buffer 322 * @dma_addr: DMA mapping address of the buffer 323 */ 324 struct nfp_net_rx_buf { 325 void *frag; 326 dma_addr_t dma_addr; 327 }; 328 329 /** 330 * struct nfp_net_xsk_rx_buf - software RX XSK buffer descriptor 331 * @dma_addr: DMA mapping address of the buffer 332 * @xdp: XSK buffer pool handle (for AF_XDP) 333 */ 334 struct nfp_net_xsk_rx_buf { 335 dma_addr_t dma_addr; 336 struct xdp_buff *xdp; 337 }; 338 339 /** 340 * struct nfp_net_rx_ring - RX ring structure 341 * @r_vec: Back pointer to ring vector structure 342 * @cnt: Size of the queue in number of descriptors 343 * @wr_p: FL/RX ring write pointer (free running) 344 * @rd_p: FL/RX ring read pointer (free running) 345 * @idx: Ring index from Linux's perspective 346 * @fl_qcidx: Queue Controller Peripheral (QCP) queue index for the freelist 347 * @qcp_fl: Pointer to base of the QCP freelist queue 348 * @rxbufs: Array of transmitted FL/RX buffers 349 * @xsk_rxbufs: Array of transmitted FL/RX buffers (for AF_XDP) 350 * @rxds: Virtual address of FL/RX ring in host memory 351 * @xdp_rxq: RX-ring info avail for XDP 352 * @dma: DMA address of the FL/RX ring 353 * @size: Size, in bytes, of the FL/RX ring (needed to free) 354 */ 355 struct nfp_net_rx_ring { 356 struct nfp_net_r_vector *r_vec; 357 358 u32 cnt; 359 u32 wr_p; 360 u32 rd_p; 361 362 u32 idx; 363 364 int fl_qcidx; 365 u8 __iomem *qcp_fl; 366 367 struct nfp_net_rx_buf *rxbufs; 368 struct nfp_net_xsk_rx_buf *xsk_rxbufs; 369 struct nfp_net_rx_desc *rxds; 370 371 struct xdp_rxq_info xdp_rxq; 372 373 dma_addr_t dma; 374 size_t size; 375 } ____cacheline_aligned; 376 377 /** 378 * struct nfp_net_r_vector - Per ring interrupt vector configuration 379 * @nfp_net: Backpointer to nfp_net structure 380 * @napi: NAPI structure for this ring vec 381 * @tasklet: ctrl vNIC, tasklet for servicing the r_vec 382 * @queue: ctrl vNIC, send queue 383 * @lock: ctrl vNIC, r_vec lock protects @queue 384 * @tx_ring: Pointer to TX ring 385 * @rx_ring: Pointer to RX ring 386 * @xdp_ring: Pointer to an extra TX ring for XDP 387 * @xsk_pool: XSK buffer pool active on vector queue pair (for AF_XDP) 388 * @irq_entry: MSI-X table entry (use for talking to the device) 389 * @event_ctr: Number of interrupt 390 * @rx_dim: Dynamic interrupt moderation structure for RX 391 * @tx_dim: Dynamic interrupt moderation structure for TX 392 * @rx_sync: Seqlock for atomic updates of RX stats 393 * @rx_pkts: Number of received packets 394 * @rx_bytes: Number of received bytes 395 * @rx_drops: Number of packets dropped on RX due to lack of resources 396 * @hw_csum_rx_ok: Counter of packets where the HW checksum was OK 397 * @hw_csum_rx_inner_ok: Counter of packets where the inner HW checksum was OK 398 * @hw_csum_rx_complete: Counter of packets with CHECKSUM_COMPLETE reported 399 * @hw_csum_rx_error: Counter of packets with bad checksums 400 * @hw_tls_rx: Number of packets with TLS decrypted by hardware 401 * @tx_sync: Seqlock for atomic updates of TX stats 402 * @tx_pkts: Number of Transmitted packets 403 * @tx_bytes: Number of Transmitted bytes 404 * @hw_csum_tx: Counter of packets with TX checksum offload requested 405 * @hw_csum_tx_inner: Counter of inner TX checksum offload requests 406 * @tx_gather: Counter of packets with Gather DMA 407 * @tx_lso: Counter of LSO packets sent 408 * @hw_tls_tx: Counter of TLS packets sent with crypto offloaded to HW 409 * @tls_tx_fallback: Counter of TLS packets sent which had to be encrypted 410 * by the fallback path because packets came out of order 411 * @tls_tx_no_fallback: Counter of TLS packets not sent because the fallback 412 * path could not encrypt them 413 * @tx_errors: How many TX errors were encountered 414 * @tx_busy: How often was TX busy (no space)? 415 * @rx_replace_buf_alloc_fail: Counter of RX buffer allocation failures 416 * @irq_vector: Interrupt vector number (use for talking to the OS) 417 * @handler: Interrupt handler for this ring vector 418 * @name: Name of the interrupt vector 419 * @affinity_mask: SMP affinity mask for this vector 420 * 421 * This structure ties RX and TX rings to interrupt vectors and a NAPI 422 * context. This currently only supports one RX and TX ring per 423 * interrupt vector but might be extended in the future to allow 424 * association of multiple rings per vector. 425 */ 426 struct nfp_net_r_vector { 427 struct nfp_net *nfp_net; 428 union { 429 struct napi_struct napi; 430 struct { 431 struct tasklet_struct tasklet; 432 struct sk_buff_head queue; 433 spinlock_t lock; 434 }; 435 }; 436 437 struct nfp_net_tx_ring *tx_ring; 438 struct nfp_net_rx_ring *rx_ring; 439 440 u16 irq_entry; 441 442 u16 event_ctr; 443 struct dim rx_dim; 444 struct dim tx_dim; 445 446 struct u64_stats_sync rx_sync; 447 u64 rx_pkts; 448 u64 rx_bytes; 449 u64 rx_drops; 450 u64 hw_csum_rx_ok; 451 u64 hw_csum_rx_inner_ok; 452 u64 hw_csum_rx_complete; 453 u64 hw_tls_rx; 454 455 u64 hw_csum_rx_error; 456 u64 rx_replace_buf_alloc_fail; 457 458 struct nfp_net_tx_ring *xdp_ring; 459 struct xsk_buff_pool *xsk_pool; 460 461 struct u64_stats_sync tx_sync; 462 u64 tx_pkts; 463 u64 tx_bytes; 464 465 u64 ____cacheline_aligned_in_smp hw_csum_tx; 466 u64 hw_csum_tx_inner; 467 u64 tx_gather; 468 u64 tx_lso; 469 u64 hw_tls_tx; 470 471 u64 tls_tx_fallback; 472 u64 tls_tx_no_fallback; 473 u64 tx_errors; 474 u64 tx_busy; 475 476 /* Cold data follows */ 477 478 u32 irq_vector; 479 irq_handler_t handler; 480 char name[IFNAMSIZ + 8]; 481 cpumask_t affinity_mask; 482 } ____cacheline_aligned; 483 484 /* Firmware version as it is written in the 32bit value in the BAR */ 485 struct nfp_net_fw_version { 486 u8 minor; 487 u8 major; 488 u8 class; 489 u8 resv; 490 } __packed; 491 492 static inline bool nfp_net_fw_ver_eq(struct nfp_net_fw_version *fw_ver, 493 u8 resv, u8 class, u8 major, u8 minor) 494 { 495 return fw_ver->resv == resv && 496 fw_ver->class == class && 497 fw_ver->major == major && 498 fw_ver->minor == minor; 499 } 500 501 struct nfp_stat_pair { 502 u64 pkts; 503 u64 bytes; 504 }; 505 506 /** 507 * struct nfp_net_dp - NFP network device datapath data structure 508 * @dev: Backpointer to struct device 509 * @netdev: Backpointer to net_device structure 510 * @is_vf: Is the driver attached to a VF? 511 * @chained_metadata_format: Firemware will use new metadata format 512 * @ktls_tx: Is kTLS TX enabled? 513 * @rx_dma_dir: Mapping direction for RX buffers 514 * @rx_dma_off: Offset at which DMA packets (for XDP headroom) 515 * @rx_offset: Offset in the RX buffers where packet data starts 516 * @ctrl: Local copy of the control register/word. 517 * @fl_bufsz: Currently configured size of the freelist buffers 518 * @xdp_prog: Installed XDP program 519 * @tx_rings: Array of pre-allocated TX ring structures 520 * @rx_rings: Array of pre-allocated RX ring structures 521 * @ctrl_bar: Pointer to mapped control BAR 522 * 523 * @txd_cnt: Size of the TX ring in number of descriptors 524 * @rxd_cnt: Size of the RX ring in number of descriptors 525 * @num_r_vecs: Number of used ring vectors 526 * @num_tx_rings: Currently configured number of TX rings 527 * @num_stack_tx_rings: Number of TX rings used by the stack (not XDP) 528 * @num_rx_rings: Currently configured number of RX rings 529 * @mtu: Device MTU 530 * @xsk_pools: XSK buffer pools, @max_r_vecs in size (for AF_XDP). 531 */ 532 struct nfp_net_dp { 533 struct device *dev; 534 struct net_device *netdev; 535 536 u8 is_vf:1; 537 u8 chained_metadata_format:1; 538 u8 ktls_tx:1; 539 540 u8 rx_dma_dir; 541 u8 rx_offset; 542 543 u32 rx_dma_off; 544 545 u32 ctrl; 546 u32 fl_bufsz; 547 548 struct bpf_prog *xdp_prog; 549 550 struct nfp_net_tx_ring *tx_rings; 551 struct nfp_net_rx_ring *rx_rings; 552 553 u8 __iomem *ctrl_bar; 554 555 /* Cold data follows */ 556 557 unsigned int txd_cnt; 558 unsigned int rxd_cnt; 559 560 unsigned int num_r_vecs; 561 562 unsigned int num_tx_rings; 563 unsigned int num_stack_tx_rings; 564 unsigned int num_rx_rings; 565 566 unsigned int mtu; 567 568 struct xsk_buff_pool **xsk_pools; 569 }; 570 571 /** 572 * struct nfp_net - NFP network device structure 573 * @dp: Datapath structure 574 * @id: vNIC id within the PF (0 for VFs) 575 * @fw_ver: Firmware version 576 * @cap: Capabilities advertised by the Firmware 577 * @max_mtu: Maximum support MTU advertised by the Firmware 578 * @rss_hfunc: RSS selected hash function 579 * @rss_cfg: RSS configuration 580 * @rss_key: RSS secret key 581 * @rss_itbl: RSS indirection table 582 * @xdp: Information about the driver XDP program 583 * @xdp_hw: Information about the HW XDP program 584 * @max_r_vecs: Number of allocated interrupt vectors for RX/TX 585 * @max_tx_rings: Maximum number of TX rings supported by the Firmware 586 * @max_rx_rings: Maximum number of RX rings supported by the Firmware 587 * @stride_rx: Queue controller RX queue spacing 588 * @stride_tx: Queue controller TX queue spacing 589 * @r_vecs: Pre-allocated array of ring vectors 590 * @irq_entries: Pre-allocated array of MSI-X entries 591 * @lsc_handler: Handler for Link State Change interrupt 592 * @lsc_name: Name for Link State Change interrupt 593 * @exn_handler: Handler for Exception interrupt 594 * @exn_name: Name for Exception interrupt 595 * @shared_handler: Handler for shared interrupts 596 * @shared_name: Name for shared interrupt 597 * @reconfig_lock: Protects @reconfig_posted, @reconfig_timer_active, 598 * @reconfig_sync_present and HW reconfiguration request 599 * regs/machinery from async requests (sync must take 600 * @bar_lock) 601 * @reconfig_posted: Pending reconfig bits coming from async sources 602 * @reconfig_timer_active: Timer for reading reconfiguration results is pending 603 * @reconfig_sync_present: Some thread is performing synchronous reconfig 604 * @reconfig_timer: Timer for async reading of reconfig results 605 * @reconfig_in_progress_update: Update FW is processing now (debug only) 606 * @bar_lock: vNIC config BAR access lock, protects: update, 607 * mailbox area, crypto TLV 608 * @link_up: Is the link up? 609 * @link_status_lock: Protects @link_* and ensures atomicity with BAR reading 610 * @rx_coalesce_adapt_on: Is RX interrupt moderation adaptive? 611 * @tx_coalesce_adapt_on: Is TX interrupt moderation adaptive? 612 * @rx_coalesce_usecs: RX interrupt moderation usecs delay parameter 613 * @rx_coalesce_max_frames: RX interrupt moderation frame count parameter 614 * @tx_coalesce_usecs: TX interrupt moderation usecs delay parameter 615 * @tx_coalesce_max_frames: TX interrupt moderation frame count parameter 616 * @qcp_cfg: Pointer to QCP queue used for configuration notification 617 * @tx_bar: Pointer to mapped TX queues 618 * @rx_bar: Pointer to mapped FL/RX queues 619 * @tlv_caps: Parsed TLV capabilities 620 * @ktls_tx_conn_cnt: Number of offloaded kTLS TX connections 621 * @ktls_rx_conn_cnt: Number of offloaded kTLS RX connections 622 * @ktls_conn_id_gen: Trivial generator for kTLS connection ids (for TX) 623 * @ktls_no_space: Counter of firmware rejecting kTLS connection due to 624 * lack of space 625 * @ktls_rx_resync_req: Counter of TLS RX resync requested 626 * @ktls_rx_resync_ign: Counter of TLS RX resync requests ignored 627 * @ktls_rx_resync_sent: Counter of TLS RX resync completed 628 * @mbox_cmsg: Common Control Message via vNIC mailbox state 629 * @mbox_cmsg.queue: CCM mbox queue of pending messages 630 * @mbox_cmsg.wq: CCM mbox wait queue of waiting processes 631 * @mbox_cmsg.workq: CCM mbox work queue for @wait_work and @runq_work 632 * @mbox_cmsg.wait_work: CCM mbox posted msg reconfig wait work 633 * @mbox_cmsg.runq_work: CCM mbox posted msg queue runner work 634 * @mbox_cmsg.tag: CCM mbox message tag allocator 635 * @debugfs_dir: Device directory in debugfs 636 * @vnic_list: Entry on device vNIC list 637 * @pdev: Backpointer to PCI device 638 * @app: APP handle if available 639 * @vnic_no_name: For non-port PF vNIC make ndo_get_phys_port_name return 640 * -EOPNOTSUPP to keep backwards compatibility (set by app) 641 * @port: Pointer to nfp_port structure if vNIC is a port 642 * @app_priv: APP private data for this vNIC 643 */ 644 struct nfp_net { 645 struct nfp_net_dp dp; 646 647 struct nfp_net_fw_version fw_ver; 648 649 u32 id; 650 651 u32 cap; 652 u32 max_mtu; 653 654 u8 rss_hfunc; 655 u32 rss_cfg; 656 u8 rss_key[NFP_NET_CFG_RSS_KEY_SZ]; 657 u8 rss_itbl[NFP_NET_CFG_RSS_ITBL_SZ]; 658 659 struct xdp_attachment_info xdp; 660 struct xdp_attachment_info xdp_hw; 661 662 unsigned int max_tx_rings; 663 unsigned int max_rx_rings; 664 665 int stride_tx; 666 int stride_rx; 667 668 unsigned int max_r_vecs; 669 struct nfp_net_r_vector r_vecs[NFP_NET_MAX_R_VECS]; 670 struct msix_entry irq_entries[NFP_NET_MAX_IRQS]; 671 672 irq_handler_t lsc_handler; 673 char lsc_name[IFNAMSIZ + 8]; 674 675 irq_handler_t exn_handler; 676 char exn_name[IFNAMSIZ + 8]; 677 678 irq_handler_t shared_handler; 679 char shared_name[IFNAMSIZ + 8]; 680 681 bool link_up; 682 spinlock_t link_status_lock; 683 684 spinlock_t reconfig_lock; 685 u32 reconfig_posted; 686 bool reconfig_timer_active; 687 bool reconfig_sync_present; 688 struct timer_list reconfig_timer; 689 u32 reconfig_in_progress_update; 690 691 struct semaphore bar_lock; 692 693 bool rx_coalesce_adapt_on; 694 bool tx_coalesce_adapt_on; 695 u32 rx_coalesce_usecs; 696 u32 rx_coalesce_max_frames; 697 u32 tx_coalesce_usecs; 698 u32 tx_coalesce_max_frames; 699 700 u8 __iomem *qcp_cfg; 701 702 u8 __iomem *tx_bar; 703 u8 __iomem *rx_bar; 704 705 struct nfp_net_tlv_caps tlv_caps; 706 707 unsigned int ktls_tx_conn_cnt; 708 unsigned int ktls_rx_conn_cnt; 709 710 atomic64_t ktls_conn_id_gen; 711 712 atomic_t ktls_no_space; 713 atomic_t ktls_rx_resync_req; 714 atomic_t ktls_rx_resync_ign; 715 atomic_t ktls_rx_resync_sent; 716 717 struct { 718 struct sk_buff_head queue; 719 wait_queue_head_t wq; 720 struct workqueue_struct *workq; 721 struct work_struct wait_work; 722 struct work_struct runq_work; 723 u16 tag; 724 } mbox_cmsg; 725 726 struct dentry *debugfs_dir; 727 728 struct list_head vnic_list; 729 730 struct pci_dev *pdev; 731 struct nfp_app *app; 732 733 bool vnic_no_name; 734 735 struct nfp_port *port; 736 737 void *app_priv; 738 }; 739 740 /* Functions to read/write from/to a BAR 741 * Performs any endian conversion necessary. 742 */ 743 static inline u16 nn_readb(struct nfp_net *nn, int off) 744 { 745 return readb(nn->dp.ctrl_bar + off); 746 } 747 748 static inline void nn_writeb(struct nfp_net *nn, int off, u8 val) 749 { 750 writeb(val, nn->dp.ctrl_bar + off); 751 } 752 753 static inline u16 nn_readw(struct nfp_net *nn, int off) 754 { 755 return readw(nn->dp.ctrl_bar + off); 756 } 757 758 static inline void nn_writew(struct nfp_net *nn, int off, u16 val) 759 { 760 writew(val, nn->dp.ctrl_bar + off); 761 } 762 763 static inline u32 nn_readl(struct nfp_net *nn, int off) 764 { 765 return readl(nn->dp.ctrl_bar + off); 766 } 767 768 static inline void nn_writel(struct nfp_net *nn, int off, u32 val) 769 { 770 writel(val, nn->dp.ctrl_bar + off); 771 } 772 773 static inline u64 nn_readq(struct nfp_net *nn, int off) 774 { 775 return readq(nn->dp.ctrl_bar + off); 776 } 777 778 static inline void nn_writeq(struct nfp_net *nn, int off, u64 val) 779 { 780 writeq(val, nn->dp.ctrl_bar + off); 781 } 782 783 /* Flush posted PCI writes by reading something without side effects */ 784 static inline void nn_pci_flush(struct nfp_net *nn) 785 { 786 nn_readl(nn, NFP_NET_CFG_VERSION); 787 } 788 789 /* Queue Controller Peripheral access functions and definitions. 790 * 791 * Some of the BARs of the NFP are mapped to portions of the Queue 792 * Controller Peripheral (QCP) address space on the NFP. A QCP queue 793 * has a read and a write pointer (as well as a size and flags, 794 * indicating overflow etc). The QCP offers a number of different 795 * operation on queue pointers, but here we only offer function to 796 * either add to a pointer or to read the pointer value. 797 */ 798 #define NFP_QCP_QUEUE_ADDR_SZ 0x800 799 #define NFP_QCP_QUEUE_AREA_SZ 0x80000 800 #define NFP_QCP_QUEUE_OFF(_x) ((_x) * NFP_QCP_QUEUE_ADDR_SZ) 801 #define NFP_QCP_QUEUE_ADD_RPTR 0x0000 802 #define NFP_QCP_QUEUE_ADD_WPTR 0x0004 803 #define NFP_QCP_QUEUE_STS_LO 0x0008 804 #define NFP_QCP_QUEUE_STS_LO_READPTR_mask 0x3ffff 805 #define NFP_QCP_QUEUE_STS_HI 0x000c 806 #define NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask 0x3ffff 807 808 /* The offset of a QCP queues in the PCIe Target */ 809 #define NFP_PCIE_QUEUE(_q) (0x80000 + (NFP_QCP_QUEUE_ADDR_SZ * ((_q) & 0xff))) 810 811 /* nfp_qcp_ptr - Read or Write Pointer of a queue */ 812 enum nfp_qcp_ptr { 813 NFP_QCP_READ_PTR = 0, 814 NFP_QCP_WRITE_PTR 815 }; 816 817 /* There appear to be an *undocumented* upper limit on the value which 818 * one can add to a queue and that value is either 0x3f or 0x7f. We 819 * go with 0x3f as a conservative measure. 820 */ 821 #define NFP_QCP_MAX_ADD 0x3f 822 823 static inline void _nfp_qcp_ptr_add(u8 __iomem *q, 824 enum nfp_qcp_ptr ptr, u32 val) 825 { 826 u32 off; 827 828 if (ptr == NFP_QCP_READ_PTR) 829 off = NFP_QCP_QUEUE_ADD_RPTR; 830 else 831 off = NFP_QCP_QUEUE_ADD_WPTR; 832 833 while (val > NFP_QCP_MAX_ADD) { 834 writel(NFP_QCP_MAX_ADD, q + off); 835 val -= NFP_QCP_MAX_ADD; 836 } 837 838 writel(val, q + off); 839 } 840 841 /** 842 * nfp_qcp_rd_ptr_add() - Add the value to the read pointer of a queue 843 * 844 * @q: Base address for queue structure 845 * @val: Value to add to the queue pointer 846 * 847 * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed. 848 */ 849 static inline void nfp_qcp_rd_ptr_add(u8 __iomem *q, u32 val) 850 { 851 _nfp_qcp_ptr_add(q, NFP_QCP_READ_PTR, val); 852 } 853 854 /** 855 * nfp_qcp_wr_ptr_add() - Add the value to the write pointer of a queue 856 * 857 * @q: Base address for queue structure 858 * @val: Value to add to the queue pointer 859 * 860 * If @val is greater than @NFP_QCP_MAX_ADD multiple writes are performed. 861 */ 862 static inline void nfp_qcp_wr_ptr_add(u8 __iomem *q, u32 val) 863 { 864 _nfp_qcp_ptr_add(q, NFP_QCP_WRITE_PTR, val); 865 } 866 867 static inline u32 _nfp_qcp_read(u8 __iomem *q, enum nfp_qcp_ptr ptr) 868 { 869 u32 off; 870 u32 val; 871 872 if (ptr == NFP_QCP_READ_PTR) 873 off = NFP_QCP_QUEUE_STS_LO; 874 else 875 off = NFP_QCP_QUEUE_STS_HI; 876 877 val = readl(q + off); 878 879 if (ptr == NFP_QCP_READ_PTR) 880 return val & NFP_QCP_QUEUE_STS_LO_READPTR_mask; 881 else 882 return val & NFP_QCP_QUEUE_STS_HI_WRITEPTR_mask; 883 } 884 885 /** 886 * nfp_qcp_rd_ptr_read() - Read the current read pointer value for a queue 887 * @q: Base address for queue structure 888 * 889 * Return: Value read. 890 */ 891 static inline u32 nfp_qcp_rd_ptr_read(u8 __iomem *q) 892 { 893 return _nfp_qcp_read(q, NFP_QCP_READ_PTR); 894 } 895 896 /** 897 * nfp_qcp_wr_ptr_read() - Read the current write pointer value for a queue 898 * @q: Base address for queue structure 899 * 900 * Return: Value read. 901 */ 902 static inline u32 nfp_qcp_wr_ptr_read(u8 __iomem *q) 903 { 904 return _nfp_qcp_read(q, NFP_QCP_WRITE_PTR); 905 } 906 907 static inline bool nfp_net_is_data_vnic(struct nfp_net *nn) 908 { 909 WARN_ON_ONCE(!nn->dp.netdev && nn->port); 910 return !!nn->dp.netdev; 911 } 912 913 static inline bool nfp_net_running(struct nfp_net *nn) 914 { 915 return nn->dp.ctrl & NFP_NET_CFG_CTRL_ENABLE; 916 } 917 918 static inline const char *nfp_net_name(struct nfp_net *nn) 919 { 920 return nn->dp.netdev ? nn->dp.netdev->name : "ctrl"; 921 } 922 923 static inline void nfp_ctrl_lock(struct nfp_net *nn) 924 __acquires(&nn->r_vecs[0].lock) 925 { 926 spin_lock_bh(&nn->r_vecs[0].lock); 927 } 928 929 static inline void nfp_ctrl_unlock(struct nfp_net *nn) 930 __releases(&nn->r_vecs[0].lock) 931 { 932 spin_unlock_bh(&nn->r_vecs[0].lock); 933 } 934 935 static inline void nn_ctrl_bar_lock(struct nfp_net *nn) 936 { 937 down(&nn->bar_lock); 938 } 939 940 static inline bool nn_ctrl_bar_trylock(struct nfp_net *nn) 941 { 942 return !down_trylock(&nn->bar_lock); 943 } 944 945 static inline void nn_ctrl_bar_unlock(struct nfp_net *nn) 946 { 947 up(&nn->bar_lock); 948 } 949 950 /* Globals */ 951 extern const char nfp_driver_version[]; 952 953 extern const struct net_device_ops nfp_net_netdev_ops; 954 955 static inline bool nfp_netdev_is_nfp_net(struct net_device *netdev) 956 { 957 return netdev->netdev_ops == &nfp_net_netdev_ops; 958 } 959 960 static inline int nfp_net_coalesce_para_check(u32 usecs, u32 pkts) 961 { 962 if ((usecs >= ((1 << 16) - 1)) || (pkts >= ((1 << 16) - 1))) 963 return -EINVAL; 964 965 return 0; 966 } 967 968 /* Prototypes */ 969 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver, 970 void __iomem *ctrl_bar); 971 972 struct nfp_net * 973 nfp_net_alloc(struct pci_dev *pdev, void __iomem *ctrl_bar, bool needs_netdev, 974 unsigned int max_tx_rings, unsigned int max_rx_rings); 975 void nfp_net_free(struct nfp_net *nn); 976 977 int nfp_net_init(struct nfp_net *nn); 978 void nfp_net_clean(struct nfp_net *nn); 979 980 int nfp_ctrl_open(struct nfp_net *nn); 981 void nfp_ctrl_close(struct nfp_net *nn); 982 983 void nfp_net_set_ethtool_ops(struct net_device *netdev); 984 void nfp_net_info(struct nfp_net *nn); 985 int __nfp_net_reconfig(struct nfp_net *nn, u32 update); 986 int nfp_net_reconfig(struct nfp_net *nn, u32 update); 987 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn); 988 void nfp_net_rss_write_itbl(struct nfp_net *nn); 989 void nfp_net_rss_write_key(struct nfp_net *nn); 990 void nfp_net_coalesce_write_cfg(struct nfp_net *nn); 991 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size); 992 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd); 993 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd); 994 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 update); 995 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn); 996 997 void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr); 998 unsigned int 999 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries, 1000 unsigned int min_irqs, unsigned int want_irqs); 1001 void nfp_net_irqs_disable(struct pci_dev *pdev); 1002 void 1003 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries, 1004 unsigned int n); 1005 1006 void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring); 1007 void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget); 1008 1009 bool 1010 nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta, 1011 void *data, void *pkt, unsigned int pkt_len, int meta_len); 1012 1013 void nfp_net_rx_csum(const struct nfp_net_dp *dp, 1014 struct nfp_net_r_vector *r_vec, 1015 const struct nfp_net_rx_desc *rxd, 1016 const struct nfp_meta_parsed *meta, 1017 struct sk_buff *skb); 1018 1019 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn); 1020 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *new, 1021 struct netlink_ext_ack *extack); 1022 1023 #ifdef CONFIG_NFP_DEBUG 1024 void nfp_net_debugfs_create(void); 1025 void nfp_net_debugfs_destroy(void); 1026 struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev); 1027 void nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir); 1028 void nfp_net_debugfs_dir_clean(struct dentry **dir); 1029 #else 1030 static inline void nfp_net_debugfs_create(void) 1031 { 1032 } 1033 1034 static inline void nfp_net_debugfs_destroy(void) 1035 { 1036 } 1037 1038 static inline struct dentry *nfp_net_debugfs_device_add(struct pci_dev *pdev) 1039 { 1040 return NULL; 1041 } 1042 1043 static inline void 1044 nfp_net_debugfs_vnic_add(struct nfp_net *nn, struct dentry *ddir) 1045 { 1046 } 1047 1048 static inline void nfp_net_debugfs_dir_clean(struct dentry **dir) 1049 { 1050 } 1051 #endif /* CONFIG_NFP_DEBUG */ 1052 1053 #endif /* _NFP_NET_H_ */ 1054