1 /* 2 * This file is part of the Chelsio T4 PCI-E SR-IOV Virtual Function Ethernet 3 * driver for Linux. 4 * 5 * Copyright (c) 2009-2010 Chelsio Communications, Inc. All rights reserved. 6 * 7 * This software is available to you under a choice of one of two 8 * licenses. You may choose to be licensed under the terms of the GNU 9 * General Public License (GPL) Version 2, available from the file 10 * COPYING in the main directory of this source tree, or the 11 * OpenIB.org BSD license below: 12 * 13 * Redistribution and use in source and binary forms, with or 14 * without modification, are permitted provided that the following 15 * conditions are met: 16 * 17 * - Redistributions of source code must retain the above 18 * copyright notice, this list of conditions and the following 19 * disclaimer. 20 * 21 * - Redistributions in binary form must reproduce the above 22 * copyright notice, this list of conditions and the following 23 * disclaimer in the documentation and/or other materials 24 * provided with the distribution. 25 * 26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 33 * SOFTWARE. 34 */ 35 36 /* 37 * This file should not be included directly. Include t4vf_common.h instead. 38 */ 39 40 #ifndef __CXGB4VF_ADAPTER_H__ 41 #define __CXGB4VF_ADAPTER_H__ 42 43 #include <linux/interrupt.h> 44 #include <linux/pci.h> 45 #include <linux/spinlock.h> 46 #include <linux/skbuff.h> 47 #include <linux/if_ether.h> 48 #include <linux/netdevice.h> 49 50 #include "../cxgb4/t4_hw.h" 51 52 /* 53 * Constants of the implementation. 54 */ 55 enum { 56 MAX_NPORTS = 1, /* max # of "ports" */ 57 MAX_PORT_QSETS = 8, /* max # of Queue Sets / "port" */ 58 MAX_ETH_QSETS = MAX_NPORTS*MAX_PORT_QSETS, 59 60 /* 61 * MSI-X interrupt index usage. 62 */ 63 MSIX_FW = 0, /* MSI-X index for firmware Q */ 64 MSIX_IQFLINT = 1, /* MSI-X index base for Ingress Qs */ 65 MSIX_EXTRAS = 1, 66 MSIX_ENTRIES = MAX_ETH_QSETS + MSIX_EXTRAS, 67 68 /* 69 * The maximum number of Ingress and Egress Queues is determined by 70 * the maximum number of "Queue Sets" which we support plus any 71 * ancillary queues. Each "Queue Set" requires one Ingress Queue 72 * for RX Packet Ingress Event notifications and two Egress Queues for 73 * a Free List and an Ethernet TX list. 74 */ 75 INGQ_EXTRAS = 2, /* firmware event queue and */ 76 /* forwarded interrupts */ 77 MAX_INGQ = MAX_ETH_QSETS+INGQ_EXTRAS, 78 MAX_EGRQ = MAX_ETH_QSETS*2, 79 }; 80 81 /* 82 * Forward structure definition references. 83 */ 84 struct adapter; 85 struct sge_eth_rxq; 86 struct sge_rspq; 87 88 /* 89 * Per-"port" information. This is really per-Virtual Interface information 90 * but the use of the "port" nomanclature makes it easier to go back and forth 91 * between the PF and VF drivers ... 92 */ 93 struct port_info { 94 struct adapter *adapter; /* our adapter */ 95 u16 viid; /* virtual interface ID */ 96 s16 xact_addr_filt; /* index of our MAC address filter */ 97 u16 rss_size; /* size of VI's RSS table slice */ 98 u8 pidx; /* index into adapter port[] */ 99 u8 port_id; /* physical port ID */ 100 u8 nqsets; /* # of "Queue Sets" */ 101 u8 first_qset; /* index of first "Queue Set" */ 102 struct link_config link_cfg; /* physical port configuration */ 103 }; 104 105 /* 106 * Scatter Gather Engine resources for the "adapter". Our ingress and egress 107 * queues are organized into "Queue Sets" with one ingress and one egress 108 * queue per Queue Set. These Queue Sets are aportionable between the "ports" 109 * (Virtual Interfaces). One extra ingress queue is used to receive 110 * asynchronous messages from the firmware. Note that the "Queue IDs" that we 111 * use here are really "Relative Queue IDs" which are returned as part of the 112 * firmware command to allocate queues. These queue IDs are relative to the 113 * absolute Queue ID base of the section of the Queue ID space allocated to 114 * the PF/VF. 115 */ 116 117 /* 118 * SGE free-list queue state. 119 */ 120 struct rx_sw_desc; 121 struct sge_fl { 122 unsigned int avail; /* # of available RX buffers */ 123 unsigned int pend_cred; /* new buffers since last FL DB ring */ 124 unsigned int cidx; /* consumer index */ 125 unsigned int pidx; /* producer index */ 126 unsigned long alloc_failed; /* # of buffer allocation failures */ 127 unsigned long large_alloc_failed; 128 unsigned long starving; /* # of times FL was found starving */ 129 130 /* 131 * Write-once/infrequently fields. 132 * ------------------------------- 133 */ 134 135 unsigned int cntxt_id; /* SGE relative QID for the free list */ 136 unsigned int abs_id; /* SGE absolute QID for the free list */ 137 unsigned int size; /* capacity of free list */ 138 struct rx_sw_desc *sdesc; /* address of SW RX descriptor ring */ 139 __be64 *desc; /* address of HW RX descriptor ring */ 140 dma_addr_t addr; /* PCI bus address of hardware ring */ 141 }; 142 143 /* 144 * An ingress packet gather list. 145 */ 146 struct pkt_gl { 147 struct page_frag frags[MAX_SKB_FRAGS]; 148 void *va; /* virtual address of first byte */ 149 unsigned int nfrags; /* # of fragments */ 150 unsigned int tot_len; /* total length of fragments */ 151 }; 152 153 typedef int (*rspq_handler_t)(struct sge_rspq *, const __be64 *, 154 const struct pkt_gl *); 155 156 /* 157 * State for an SGE Response Queue. 158 */ 159 struct sge_rspq { 160 struct napi_struct napi; /* NAPI scheduling control */ 161 const __be64 *cur_desc; /* current descriptor in queue */ 162 unsigned int cidx; /* consumer index */ 163 u8 gen; /* current generation bit */ 164 u8 next_intr_params; /* holdoff params for next interrupt */ 165 int offset; /* offset into current FL buffer */ 166 167 unsigned int unhandled_irqs; /* bogus interrupts */ 168 169 /* 170 * Write-once/infrequently fields. 171 * ------------------------------- 172 */ 173 174 u8 intr_params; /* interrupt holdoff parameters */ 175 u8 pktcnt_idx; /* interrupt packet threshold */ 176 u8 idx; /* queue index within its group */ 177 u16 cntxt_id; /* SGE rel QID for the response Q */ 178 u16 abs_id; /* SGE abs QID for the response Q */ 179 __be64 *desc; /* address of hardware response ring */ 180 dma_addr_t phys_addr; /* PCI bus address of ring */ 181 unsigned int iqe_len; /* entry size */ 182 unsigned int size; /* capcity of response Q */ 183 struct adapter *adapter; /* our adapter */ 184 struct net_device *netdev; /* associated net device */ 185 rspq_handler_t handler; /* the handler for this response Q */ 186 }; 187 188 /* 189 * Ethernet queue statistics 190 */ 191 struct sge_eth_stats { 192 unsigned long pkts; /* # of ethernet packets */ 193 unsigned long lro_pkts; /* # of LRO super packets */ 194 unsigned long lro_merged; /* # of wire packets merged by LRO */ 195 unsigned long rx_cso; /* # of Rx checksum offloads */ 196 unsigned long vlan_ex; /* # of Rx VLAN extractions */ 197 unsigned long rx_drops; /* # of packets dropped due to no mem */ 198 }; 199 200 /* 201 * State for an Ethernet Receive Queue. 202 */ 203 struct sge_eth_rxq { 204 struct sge_rspq rspq; /* Response Queue */ 205 struct sge_fl fl; /* Free List */ 206 struct sge_eth_stats stats; /* receive statistics */ 207 }; 208 209 /* 210 * SGE Transmit Queue state. This contains all of the resources associated 211 * with the hardware status of a TX Queue which is a circular ring of hardware 212 * TX Descriptors. For convenience, it also contains a pointer to a parallel 213 * "Software Descriptor" array but we don't know anything about it here other 214 * than its type name. 215 */ 216 struct tx_desc { 217 /* 218 * Egress Queues are measured in units of SGE_EQ_IDXSIZE by the 219 * hardware: Sizes, Producer and Consumer indices, etc. 220 */ 221 __be64 flit[SGE_EQ_IDXSIZE/sizeof(__be64)]; 222 }; 223 struct tx_sw_desc; 224 struct sge_txq { 225 unsigned int in_use; /* # of in-use TX descriptors */ 226 unsigned int size; /* # of descriptors */ 227 unsigned int cidx; /* SW consumer index */ 228 unsigned int pidx; /* producer index */ 229 unsigned long stops; /* # of times queue has been stopped */ 230 unsigned long restarts; /* # of queue restarts */ 231 232 /* 233 * Write-once/infrequently fields. 234 * ------------------------------- 235 */ 236 237 unsigned int cntxt_id; /* SGE relative QID for the TX Q */ 238 unsigned int abs_id; /* SGE absolute QID for the TX Q */ 239 struct tx_desc *desc; /* address of HW TX descriptor ring */ 240 struct tx_sw_desc *sdesc; /* address of SW TX descriptor ring */ 241 struct sge_qstat *stat; /* queue status entry */ 242 dma_addr_t phys_addr; /* PCI bus address of hardware ring */ 243 }; 244 245 /* 246 * State for an Ethernet Transmit Queue. 247 */ 248 struct sge_eth_txq { 249 struct sge_txq q; /* SGE TX Queue */ 250 struct netdev_queue *txq; /* associated netdev TX queue */ 251 unsigned long tso; /* # of TSO requests */ 252 unsigned long tx_cso; /* # of TX checksum offloads */ 253 unsigned long vlan_ins; /* # of TX VLAN insertions */ 254 unsigned long mapping_err; /* # of I/O MMU packet mapping errors */ 255 }; 256 257 /* 258 * The complete set of Scatter/Gather Engine resources. 259 */ 260 struct sge { 261 /* 262 * Our "Queue Sets" ... 263 */ 264 struct sge_eth_txq ethtxq[MAX_ETH_QSETS]; 265 struct sge_eth_rxq ethrxq[MAX_ETH_QSETS]; 266 267 /* 268 * Extra ingress queues for asynchronous firmware events and 269 * forwarded interrupts (when in MSI mode). 270 */ 271 struct sge_rspq fw_evtq ____cacheline_aligned_in_smp; 272 273 struct sge_rspq intrq ____cacheline_aligned_in_smp; 274 spinlock_t intrq_lock; 275 276 /* 277 * State for managing "starving Free Lists" -- Free Lists which have 278 * fallen below a certain threshold of buffers available to the 279 * hardware and attempts to refill them up to that threshold have 280 * failed. We have a regular "slow tick" timer process which will 281 * make periodic attempts to refill these starving Free Lists ... 282 */ 283 DECLARE_BITMAP(starving_fl, MAX_EGRQ); 284 struct timer_list rx_timer; 285 286 /* 287 * State for cleaning up completed TX descriptors. 288 */ 289 struct timer_list tx_timer; 290 291 /* 292 * Write-once/infrequently fields. 293 * ------------------------------- 294 */ 295 296 u16 max_ethqsets; /* # of available Ethernet queue sets */ 297 u16 ethqsets; /* # of active Ethernet queue sets */ 298 u16 ethtxq_rover; /* Tx queue to clean up next */ 299 u16 timer_val[SGE_NTIMERS]; /* interrupt holdoff timer array */ 300 u8 counter_val[SGE_NCOUNTERS]; /* interrupt RX threshold array */ 301 302 /* 303 * Reverse maps from Absolute Queue IDs to associated queue pointers. 304 * The absolute Queue IDs are in a compact range which start at a 305 * [potentially large] Base Queue ID. We perform the reverse map by 306 * first converting the Absolute Queue ID into a Relative Queue ID by 307 * subtracting off the Base Queue ID and then use a Relative Queue ID 308 * indexed table to get the pointer to the corresponding software 309 * queue structure. 310 */ 311 unsigned int egr_base; 312 unsigned int ingr_base; 313 void *egr_map[MAX_EGRQ]; 314 struct sge_rspq *ingr_map[MAX_INGQ]; 315 }; 316 317 /* 318 * Utility macros to convert Absolute- to Relative-Queue indices and Egress- 319 * and Ingress-Queues. The EQ_MAP() and IQ_MAP() macros which provide 320 * pointers to Ingress- and Egress-Queues can be used as both L- and R-values 321 */ 322 #define EQ_IDX(s, abs_id) ((unsigned int)((abs_id) - (s)->egr_base)) 323 #define IQ_IDX(s, abs_id) ((unsigned int)((abs_id) - (s)->ingr_base)) 324 325 #define EQ_MAP(s, abs_id) ((s)->egr_map[EQ_IDX(s, abs_id)]) 326 #define IQ_MAP(s, abs_id) ((s)->ingr_map[IQ_IDX(s, abs_id)]) 327 328 /* 329 * Macro to iterate across Queue Sets ("rxq" is a historic misnomer). 330 */ 331 #define for_each_ethrxq(sge, iter) \ 332 for (iter = 0; iter < (sge)->ethqsets; iter++) 333 334 /* 335 * Per-"adapter" (Virtual Function) information. 336 */ 337 struct adapter { 338 /* PCI resources */ 339 void __iomem *regs; 340 struct pci_dev *pdev; 341 struct device *pdev_dev; 342 343 /* "adapter" resources */ 344 unsigned long registered_device_map; 345 unsigned long open_device_map; 346 unsigned long flags; 347 struct adapter_params params; 348 349 /* queue and interrupt resources */ 350 struct { 351 unsigned short vec; 352 char desc[22]; 353 } msix_info[MSIX_ENTRIES]; 354 struct sge sge; 355 356 /* Linux network device resources */ 357 struct net_device *port[MAX_NPORTS]; 358 const char *name; 359 unsigned int msg_enable; 360 361 /* debugfs resources */ 362 struct dentry *debugfs_root; 363 364 /* various locks */ 365 spinlock_t stats_lock; 366 }; 367 368 enum { /* adapter flags */ 369 FULL_INIT_DONE = (1UL << 0), 370 USING_MSI = (1UL << 1), 371 USING_MSIX = (1UL << 2), 372 QUEUES_BOUND = (1UL << 3), 373 }; 374 375 /* 376 * The following register read/write routine definitions are required by 377 * the common code. 378 */ 379 380 /** 381 * t4_read_reg - read a HW register 382 * @adapter: the adapter 383 * @reg_addr: the register address 384 * 385 * Returns the 32-bit value of the given HW register. 386 */ 387 static inline u32 t4_read_reg(struct adapter *adapter, u32 reg_addr) 388 { 389 return readl(adapter->regs + reg_addr); 390 } 391 392 /** 393 * t4_write_reg - write a HW register 394 * @adapter: the adapter 395 * @reg_addr: the register address 396 * @val: the value to write 397 * 398 * Write a 32-bit value into the given HW register. 399 */ 400 static inline void t4_write_reg(struct adapter *adapter, u32 reg_addr, u32 val) 401 { 402 writel(val, adapter->regs + reg_addr); 403 } 404 405 #ifndef readq 406 static inline u64 readq(const volatile void __iomem *addr) 407 { 408 return readl(addr) + ((u64)readl(addr + 4) << 32); 409 } 410 411 static inline void writeq(u64 val, volatile void __iomem *addr) 412 { 413 writel(val, addr); 414 writel(val >> 32, addr + 4); 415 } 416 #endif 417 418 /** 419 * t4_read_reg64 - read a 64-bit HW register 420 * @adapter: the adapter 421 * @reg_addr: the register address 422 * 423 * Returns the 64-bit value of the given HW register. 424 */ 425 static inline u64 t4_read_reg64(struct adapter *adapter, u32 reg_addr) 426 { 427 return readq(adapter->regs + reg_addr); 428 } 429 430 /** 431 * t4_write_reg64 - write a 64-bit HW register 432 * @adapter: the adapter 433 * @reg_addr: the register address 434 * @val: the value to write 435 * 436 * Write a 64-bit value into the given HW register. 437 */ 438 static inline void t4_write_reg64(struct adapter *adapter, u32 reg_addr, 439 u64 val) 440 { 441 writeq(val, adapter->regs + reg_addr); 442 } 443 444 /** 445 * port_name - return the string name of a port 446 * @adapter: the adapter 447 * @pidx: the port index 448 * 449 * Return the string name of the selected port. 450 */ 451 static inline const char *port_name(struct adapter *adapter, int pidx) 452 { 453 return adapter->port[pidx]->name; 454 } 455 456 /** 457 * t4_os_set_hw_addr - store a port's MAC address in SW 458 * @adapter: the adapter 459 * @pidx: the port index 460 * @hw_addr: the Ethernet address 461 * 462 * Store the Ethernet address of the given port in SW. Called by the common 463 * code when it retrieves a port's Ethernet address from EEPROM. 464 */ 465 static inline void t4_os_set_hw_addr(struct adapter *adapter, int pidx, 466 u8 hw_addr[]) 467 { 468 memcpy(adapter->port[pidx]->dev_addr, hw_addr, ETH_ALEN); 469 memcpy(adapter->port[pidx]->perm_addr, hw_addr, ETH_ALEN); 470 } 471 472 /** 473 * netdev2pinfo - return the port_info structure associated with a net_device 474 * @dev: the netdev 475 * 476 * Return the struct port_info associated with a net_device 477 */ 478 static inline struct port_info *netdev2pinfo(const struct net_device *dev) 479 { 480 return netdev_priv(dev); 481 } 482 483 /** 484 * adap2pinfo - return the port_info of a port 485 * @adap: the adapter 486 * @pidx: the port index 487 * 488 * Return the port_info structure for the adapter. 489 */ 490 static inline struct port_info *adap2pinfo(struct adapter *adapter, int pidx) 491 { 492 return netdev_priv(adapter->port[pidx]); 493 } 494 495 /** 496 * netdev2adap - return the adapter structure associated with a net_device 497 * @dev: the netdev 498 * 499 * Return the struct adapter associated with a net_device 500 */ 501 static inline struct adapter *netdev2adap(const struct net_device *dev) 502 { 503 return netdev2pinfo(dev)->adapter; 504 } 505 506 /* 507 * OS "Callback" function declarations. These are functions that the OS code 508 * is "contracted" to provide for the common code. 509 */ 510 void t4vf_os_link_changed(struct adapter *, int, int); 511 512 /* 513 * SGE function prototype declarations. 514 */ 515 int t4vf_sge_alloc_rxq(struct adapter *, struct sge_rspq *, bool, 516 struct net_device *, int, 517 struct sge_fl *, rspq_handler_t); 518 int t4vf_sge_alloc_eth_txq(struct adapter *, struct sge_eth_txq *, 519 struct net_device *, struct netdev_queue *, 520 unsigned int); 521 void t4vf_free_sge_resources(struct adapter *); 522 523 int t4vf_eth_xmit(struct sk_buff *, struct net_device *); 524 int t4vf_ethrx_handler(struct sge_rspq *, const __be64 *, 525 const struct pkt_gl *); 526 527 irq_handler_t t4vf_intr_handler(struct adapter *); 528 irqreturn_t t4vf_sge_intr_msix(int, void *); 529 530 int t4vf_sge_init(struct adapter *); 531 void t4vf_sge_start(struct adapter *); 532 void t4vf_sge_stop(struct adapter *); 533 534 #endif /* __CXGB4VF_ADAPTER_H__ */ 535