1 /*- 2 * Copyright (c) 2016-2017 Alexander Motin <mav@FreeBSD.org> 3 * Copyright (C) 2013 Intel Corporation 4 * Copyright (C) 2015 EMC Corporation 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 /* 30 * The Non-Transparent Bridge (NTB) is a device that allows you to connect 31 * two or more systems using a PCI-e links, providing remote memory access. 32 * 33 * This module contains a driver for NTB hardware in Intel Xeon/Atom CPUs. 34 * 35 * NOTE: Much of the code in this module is shared with Linux. Any patches may 36 * be picked up and redistributed in Linux with a dual GPL/BSD license. 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include <sys/param.h> 43 #include <sys/kernel.h> 44 #include <sys/systm.h> 45 #include <sys/bus.h> 46 #include <sys/endian.h> 47 #include <sys/interrupt.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/module.h> 51 #include <sys/mutex.h> 52 #include <sys/pciio.h> 53 #include <sys/taskqueue.h> 54 #include <sys/tree.h> 55 #include <sys/queue.h> 56 #include <sys/rman.h> 57 #include <sys/sbuf.h> 58 #include <sys/sysctl.h> 59 #include <vm/vm.h> 60 #include <vm/pmap.h> 61 #include <machine/bus.h> 62 #include <machine/intr_machdep.h> 63 #include <machine/resource.h> 64 #include <dev/pci/pcireg.h> 65 #include <dev/pci/pcivar.h> 66 #include <dev/iommu/iommu.h> 67 68 #include "ntb_hw_intel.h" 69 #include "../ntb.h" 70 71 #define MAX_MSIX_INTERRUPTS \ 72 MAX(MAX(XEON_DB_COUNT, ATOM_DB_COUNT), XEON_GEN3_DB_COUNT) 73 74 #define NTB_HB_TIMEOUT 1 /* second */ 75 #define ATOM_LINK_RECOVERY_TIME 500 /* ms */ 76 #define BAR_HIGH_MASK (~((1ull << 12) - 1)) 77 78 #define NTB_MSIX_VER_GUARD 0xaabbccdd 79 #define NTB_MSIX_RECEIVED 0xe0f0e0f0 80 81 /* 82 * PCI constants could be somewhere more generic, but aren't defined/used in 83 * pci.c. 84 */ 85 #define PCI_MSIX_ENTRY_SIZE 16 86 #define PCI_MSIX_ENTRY_LOWER_ADDR 0 87 #define PCI_MSIX_ENTRY_UPPER_ADDR 4 88 #define PCI_MSIX_ENTRY_DATA 8 89 90 enum ntb_device_type { 91 NTB_XEON_GEN1, 92 NTB_XEON_GEN3, 93 NTB_ATOM 94 }; 95 96 /* ntb_conn_type are hardware numbers, cannot change. */ 97 enum ntb_conn_type { 98 NTB_CONN_TRANSPARENT = 0, 99 NTB_CONN_B2B = 1, 100 NTB_CONN_RP = 2, 101 }; 102 103 enum ntb_b2b_direction { 104 NTB_DEV_USD = 0, 105 NTB_DEV_DSD = 1, 106 }; 107 108 enum ntb_bar { 109 NTB_CONFIG_BAR = 0, 110 NTB_B2B_BAR_1, 111 NTB_B2B_BAR_2, 112 NTB_B2B_BAR_3, 113 NTB_MAX_BARS 114 }; 115 116 enum { 117 NTB_MSIX_GUARD = 0, 118 NTB_MSIX_DATA0, 119 NTB_MSIX_DATA1, 120 NTB_MSIX_DATA2, 121 NTB_MSIX_OFS0, 122 NTB_MSIX_OFS1, 123 NTB_MSIX_OFS2, 124 NTB_MSIX_DONE, 125 NTB_MAX_MSIX_SPAD 126 }; 127 128 /* Device features and workarounds */ 129 #define HAS_FEATURE(ntb, feature) \ 130 (((ntb)->features & (feature)) != 0) 131 132 struct ntb_hw_info { 133 uint32_t device_id; 134 const char *desc; 135 enum ntb_device_type type; 136 uint32_t features; 137 }; 138 139 struct ntb_pci_bar_info { 140 bus_space_tag_t pci_bus_tag; 141 bus_space_handle_t pci_bus_handle; 142 int pci_resource_id; 143 struct resource *pci_resource; 144 vm_paddr_t pbase; 145 caddr_t vbase; 146 vm_size_t size; 147 vm_memattr_t map_mode; 148 149 /* Configuration register offsets */ 150 uint32_t psz_off; 151 uint32_t ssz_off; 152 uint32_t pbarxlat_off; 153 }; 154 155 struct ntb_int_info { 156 struct resource *res; 157 int rid; 158 void *tag; 159 }; 160 161 struct ntb_vec { 162 struct ntb_softc *ntb; 163 uint32_t num; 164 unsigned masked; 165 }; 166 167 struct ntb_reg { 168 uint32_t ntb_ctl; 169 uint32_t lnk_sta; 170 uint8_t db_size; 171 unsigned mw_bar[NTB_MAX_BARS]; 172 }; 173 174 struct ntb_alt_reg { 175 uint32_t db_bell; 176 uint32_t db_mask; 177 uint32_t spad; 178 }; 179 180 struct ntb_xlat_reg { 181 uint32_t bar0_base; 182 uint32_t bar2_base; 183 uint32_t bar4_base; 184 uint32_t bar5_base; 185 186 uint32_t bar2_xlat; 187 uint32_t bar4_xlat; 188 uint32_t bar5_xlat; 189 190 uint32_t bar2_limit; 191 uint32_t bar4_limit; 192 uint32_t bar5_limit; 193 }; 194 195 struct ntb_b2b_addr { 196 uint64_t bar0_addr; 197 uint64_t bar2_addr64; 198 uint64_t bar4_addr64; 199 uint64_t bar4_addr32; 200 uint64_t bar5_addr32; 201 }; 202 203 struct ntb_msix_data { 204 uint32_t nmd_ofs; 205 uint32_t nmd_data; 206 }; 207 208 struct ntb_softc { 209 /* ntb.c context. Do not move! Must go first! */ 210 void *ntb_store; 211 212 device_t device; 213 enum ntb_device_type type; 214 uint32_t features; 215 216 struct ntb_pci_bar_info bar_info[NTB_MAX_BARS]; 217 struct ntb_int_info int_info[MAX_MSIX_INTERRUPTS]; 218 uint32_t allocated_interrupts; 219 220 struct ntb_msix_data peer_msix_data[XEON_NONLINK_DB_MSIX_BITS]; 221 struct ntb_msix_data msix_data[XEON_NONLINK_DB_MSIX_BITS]; 222 bool peer_msix_good; 223 bool peer_msix_done; 224 struct ntb_pci_bar_info *peer_lapic_bar; 225 struct callout peer_msix_work; 226 227 bus_dma_tag_t bar0_dma_tag; 228 bus_dmamap_t bar0_dma_map; 229 230 struct callout heartbeat_timer; 231 struct callout lr_timer; 232 233 struct ntb_vec *msix_vec; 234 235 uint32_t ppd; 236 enum ntb_conn_type conn_type; 237 enum ntb_b2b_direction dev_type; 238 239 /* Offset of peer bar0 in B2B BAR */ 240 uint64_t b2b_off; 241 /* Memory window used to access peer bar0 */ 242 #define B2B_MW_DISABLED UINT8_MAX 243 uint8_t b2b_mw_idx; 244 uint32_t msix_xlat; 245 uint8_t msix_mw_idx; 246 247 uint8_t mw_count; 248 uint8_t spad_count; 249 uint8_t db_count; 250 uint8_t db_vec_count; 251 uint8_t db_vec_shift; 252 253 /* Protects local db_mask. */ 254 #define DB_MASK_LOCK(sc) mtx_lock_spin(&(sc)->db_mask_lock) 255 #define DB_MASK_UNLOCK(sc) mtx_unlock_spin(&(sc)->db_mask_lock) 256 #define DB_MASK_ASSERT(sc,f) mtx_assert(&(sc)->db_mask_lock, (f)) 257 struct mtx db_mask_lock; 258 259 volatile uint32_t ntb_ctl; 260 volatile uint32_t lnk_sta; 261 262 uint64_t db_valid_mask; 263 uint64_t db_link_mask; 264 uint64_t db_mask; 265 uint64_t fake_db; /* NTB_SB01BASE_LOCKUP*/ 266 uint64_t force_db; /* NTB_SB01BASE_LOCKUP*/ 267 268 int last_ts; /* ticks @ last irq */ 269 270 const struct ntb_reg *reg; 271 const struct ntb_alt_reg *self_reg; 272 const struct ntb_alt_reg *peer_reg; 273 const struct ntb_xlat_reg *xlat_reg; 274 }; 275 276 #ifdef __i386__ 277 static __inline uint64_t 278 bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle, 279 bus_size_t offset) 280 { 281 282 return (bus_space_read_4(tag, handle, offset) | 283 ((uint64_t)bus_space_read_4(tag, handle, offset + 4)) << 32); 284 } 285 286 static __inline void 287 bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t handle, 288 bus_size_t offset, uint64_t val) 289 { 290 291 bus_space_write_4(tag, handle, offset, val); 292 bus_space_write_4(tag, handle, offset + 4, val >> 32); 293 } 294 #endif 295 296 #define intel_ntb_bar_read(SIZE, bar, offset) \ 297 bus_space_read_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \ 298 ntb->bar_info[(bar)].pci_bus_handle, (offset)) 299 #define intel_ntb_bar_write(SIZE, bar, offset, val) \ 300 bus_space_write_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \ 301 ntb->bar_info[(bar)].pci_bus_handle, (offset), (val)) 302 #define intel_ntb_reg_read(SIZE, offset) \ 303 intel_ntb_bar_read(SIZE, NTB_CONFIG_BAR, offset) 304 #define intel_ntb_reg_write(SIZE, offset, val) \ 305 intel_ntb_bar_write(SIZE, NTB_CONFIG_BAR, offset, val) 306 #define intel_ntb_mw_read(SIZE, offset) \ 307 intel_ntb_bar_read(SIZE, intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx), \ 308 offset) 309 #define intel_ntb_mw_write(SIZE, offset, val) \ 310 intel_ntb_bar_write(SIZE, intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx), \ 311 offset, val) 312 313 static int intel_ntb_probe(device_t device); 314 static int intel_ntb_attach(device_t device); 315 static int intel_ntb_detach(device_t device); 316 static uint64_t intel_ntb_db_valid_mask(device_t dev); 317 static void intel_ntb_spad_clear(device_t dev); 318 static uint64_t intel_ntb_db_vector_mask(device_t dev, uint32_t vector); 319 static bool intel_ntb_link_is_up(device_t dev, enum ntb_speed *speed, 320 enum ntb_width *width); 321 static int intel_ntb_link_enable(device_t dev, enum ntb_speed speed, 322 enum ntb_width width); 323 static int intel_ntb_link_disable(device_t dev); 324 static int intel_ntb_spad_read(device_t dev, unsigned int idx, uint32_t *val); 325 static int intel_ntb_peer_spad_write(device_t dev, unsigned int idx, uint32_t val); 326 327 static unsigned intel_ntb_user_mw_to_idx(struct ntb_softc *, unsigned uidx); 328 static inline enum ntb_bar intel_ntb_mw_to_bar(struct ntb_softc *, unsigned mw); 329 static inline bool bar_is_64bit(struct ntb_softc *, enum ntb_bar); 330 static inline void bar_get_xlat_params(struct ntb_softc *, enum ntb_bar, 331 uint32_t *base, uint32_t *xlat, uint32_t *lmt); 332 static int intel_ntb_map_pci_bars(struct ntb_softc *ntb); 333 static int intel_ntb_mw_set_wc_internal(struct ntb_softc *, unsigned idx, 334 vm_memattr_t); 335 static void print_map_success(struct ntb_softc *, struct ntb_pci_bar_info *, 336 const char *); 337 static int map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar); 338 static int map_memory_window_bar(struct ntb_softc *ntb, 339 struct ntb_pci_bar_info *bar); 340 static void intel_ntb_unmap_pci_bar(struct ntb_softc *ntb); 341 static int intel_ntb_remap_msix(device_t, uint32_t desired, uint32_t avail); 342 static int intel_ntb_init_isr(struct ntb_softc *ntb); 343 static int intel_ntb_xeon_gen3_init_isr(struct ntb_softc *ntb); 344 static int intel_ntb_setup_legacy_interrupt(struct ntb_softc *ntb); 345 static int intel_ntb_setup_msix(struct ntb_softc *ntb, uint32_t num_vectors); 346 static void intel_ntb_teardown_interrupts(struct ntb_softc *ntb); 347 static inline uint64_t intel_ntb_vec_mask(struct ntb_softc *, uint64_t db_vector); 348 static void intel_ntb_interrupt(struct ntb_softc *, uint32_t vec); 349 static void ndev_vec_isr(void *arg); 350 static void ndev_irq_isr(void *arg); 351 static inline uint64_t db_ioread(struct ntb_softc *, uint64_t regoff); 352 static inline void db_iowrite(struct ntb_softc *, uint64_t regoff, uint64_t); 353 static inline void db_iowrite_raw(struct ntb_softc *, uint64_t regoff, uint64_t); 354 static int intel_ntb_create_msix_vec(struct ntb_softc *ntb, uint32_t num_vectors); 355 static void intel_ntb_free_msix_vec(struct ntb_softc *ntb); 356 static void intel_ntb_get_msix_info(struct ntb_softc *ntb); 357 static void intel_ntb_exchange_msix(void *); 358 static struct ntb_hw_info *intel_ntb_get_device_info(uint32_t device_id); 359 static void intel_ntb_detect_max_mw(struct ntb_softc *ntb); 360 static int intel_ntb_detect_xeon(struct ntb_softc *ntb); 361 static int intel_ntb_detect_xeon_gen3(struct ntb_softc *ntb); 362 static int intel_ntb_detect_atom(struct ntb_softc *ntb); 363 static int intel_ntb_xeon_init_dev(struct ntb_softc *ntb); 364 static int intel_ntb_xeon_gen3_init_dev(struct ntb_softc *ntb); 365 static int intel_ntb_atom_init_dev(struct ntb_softc *ntb); 366 static void intel_ntb_teardown_xeon(struct ntb_softc *ntb); 367 static void configure_atom_secondary_side_bars(struct ntb_softc *ntb); 368 static void xeon_reset_sbar_size(struct ntb_softc *, enum ntb_bar idx, 369 enum ntb_bar regbar); 370 static void xeon_set_sbar_base_and_limit(struct ntb_softc *, 371 uint64_t base_addr, enum ntb_bar idx, enum ntb_bar regbar); 372 static void xeon_set_pbar_xlat(struct ntb_softc *, uint64_t base_addr, 373 enum ntb_bar idx); 374 static int xeon_setup_b2b_mw(struct ntb_softc *, 375 const struct ntb_b2b_addr *addr, const struct ntb_b2b_addr *peer_addr); 376 static int xeon_gen3_setup_b2b_mw(struct ntb_softc *); 377 static int intel_ntb_mw_set_trans(device_t dev, unsigned idx, bus_addr_t addr, 378 size_t size); 379 static inline bool link_is_up(struct ntb_softc *ntb); 380 static inline bool _xeon_link_is_up(struct ntb_softc *ntb); 381 static inline bool atom_link_is_err(struct ntb_softc *ntb); 382 static inline enum ntb_speed intel_ntb_link_sta_speed(struct ntb_softc *); 383 static inline enum ntb_width intel_ntb_link_sta_width(struct ntb_softc *); 384 static void atom_link_hb(void *arg); 385 static void recover_atom_link(void *arg); 386 static bool intel_ntb_poll_link(struct ntb_softc *ntb); 387 static void save_bar_parameters(struct ntb_pci_bar_info *bar); 388 static void intel_ntb_sysctl_init(struct ntb_softc *); 389 static int sysctl_handle_features(SYSCTL_HANDLER_ARGS); 390 static int sysctl_handle_link_admin(SYSCTL_HANDLER_ARGS); 391 static int sysctl_handle_link_status_human(SYSCTL_HANDLER_ARGS); 392 static int sysctl_handle_link_status(SYSCTL_HANDLER_ARGS); 393 static int sysctl_handle_register(SYSCTL_HANDLER_ARGS); 394 395 static unsigned g_ntb_hw_debug_level; 396 SYSCTL_UINT(_hw_ntb, OID_AUTO, debug_level, CTLFLAG_RWTUN, 397 &g_ntb_hw_debug_level, 0, "ntb_hw log level -- higher is more verbose"); 398 #define intel_ntb_printf(lvl, ...) do { \ 399 if ((lvl) <= g_ntb_hw_debug_level) { \ 400 device_printf(ntb->device, __VA_ARGS__); \ 401 } \ 402 } while (0) 403 404 #define _NTB_PAT_UC 0 405 #define _NTB_PAT_WC 1 406 #define _NTB_PAT_WT 4 407 #define _NTB_PAT_WP 5 408 #define _NTB_PAT_WB 6 409 #define _NTB_PAT_UCM 7 410 static unsigned g_ntb_mw_pat = _NTB_PAT_UC; 411 SYSCTL_UINT(_hw_ntb, OID_AUTO, default_mw_pat, CTLFLAG_RDTUN, 412 &g_ntb_mw_pat, 0, "Configure the default memory window cache flags (PAT): " 413 "UC: " __XSTRING(_NTB_PAT_UC) ", " 414 "WC: " __XSTRING(_NTB_PAT_WC) ", " 415 "WT: " __XSTRING(_NTB_PAT_WT) ", " 416 "WP: " __XSTRING(_NTB_PAT_WP) ", " 417 "WB: " __XSTRING(_NTB_PAT_WB) ", " 418 "UC-: " __XSTRING(_NTB_PAT_UCM)); 419 420 static inline vm_memattr_t 421 intel_ntb_pat_flags(void) 422 { 423 424 switch (g_ntb_mw_pat) { 425 case _NTB_PAT_WC: 426 return (VM_MEMATTR_WRITE_COMBINING); 427 case _NTB_PAT_WT: 428 return (VM_MEMATTR_WRITE_THROUGH); 429 case _NTB_PAT_WP: 430 return (VM_MEMATTR_WRITE_PROTECTED); 431 case _NTB_PAT_WB: 432 return (VM_MEMATTR_WRITE_BACK); 433 case _NTB_PAT_UCM: 434 return (VM_MEMATTR_WEAK_UNCACHEABLE); 435 case _NTB_PAT_UC: 436 /* FALLTHROUGH */ 437 default: 438 return (VM_MEMATTR_UNCACHEABLE); 439 } 440 } 441 442 /* 443 * Well, this obviously doesn't belong here, but it doesn't seem to exist 444 * anywhere better yet. 445 */ 446 static inline const char * 447 intel_ntb_vm_memattr_to_str(vm_memattr_t pat) 448 { 449 450 switch (pat) { 451 case VM_MEMATTR_WRITE_COMBINING: 452 return ("WRITE_COMBINING"); 453 case VM_MEMATTR_WRITE_THROUGH: 454 return ("WRITE_THROUGH"); 455 case VM_MEMATTR_WRITE_PROTECTED: 456 return ("WRITE_PROTECTED"); 457 case VM_MEMATTR_WRITE_BACK: 458 return ("WRITE_BACK"); 459 case VM_MEMATTR_WEAK_UNCACHEABLE: 460 return ("UNCACHED"); 461 case VM_MEMATTR_UNCACHEABLE: 462 return ("UNCACHEABLE"); 463 default: 464 return ("UNKNOWN"); 465 } 466 } 467 468 static int g_ntb_msix_idx = 1; 469 SYSCTL_INT(_hw_ntb, OID_AUTO, msix_mw_idx, CTLFLAG_RDTUN, &g_ntb_msix_idx, 470 0, "Use this memory window to access the peer MSIX message complex on " 471 "certain Xeon-based NTB systems, as a workaround for a hardware errata. " 472 "Like b2b_mw_idx, negative values index from the last available memory " 473 "window. (Applies on Xeon platforms with SB01BASE_LOCKUP errata.)"); 474 475 static int g_ntb_mw_idx = -1; 476 SYSCTL_INT(_hw_ntb, OID_AUTO, b2b_mw_idx, CTLFLAG_RDTUN, &g_ntb_mw_idx, 477 0, "Use this memory window to access the peer NTB registers. A " 478 "non-negative value starts from the first MW index; a negative value " 479 "starts from the last MW index. The default is -1, i.e., the last " 480 "available memory window. Both sides of the NTB MUST set the same " 481 "value here! (Applies on Xeon platforms with SDOORBELL_LOCKUP errata.)"); 482 483 /* Hardware owns the low 16 bits of features. */ 484 #define NTB_BAR_SIZE_4K (1 << 0) 485 #define NTB_SDOORBELL_LOCKUP (1 << 1) 486 #define NTB_SB01BASE_LOCKUP (1 << 2) 487 #define NTB_B2BDOORBELL_BIT14 (1 << 3) 488 /* Software/configuration owns the top 16 bits. */ 489 #define NTB_SPLIT_BAR (1ull << 16) 490 #define NTB_ONE_MSIX (1ull << 17) 491 492 #define NTB_FEATURES_STR \ 493 "\20\21SPLIT_BAR4\04B2B_DOORBELL_BIT14\03SB01BASE_LOCKUP" \ 494 "\02SDOORBELL_LOCKUP\01BAR_SIZE_4K" 495 496 static struct ntb_hw_info pci_ids[] = { 497 /* XXX: PS/SS IDs left out until they are supported. */ 498 { 0x0C4E8086, "BWD Atom Processor S1200 Non-Transparent Bridge B2B", 499 NTB_ATOM, 0 }, 500 501 { 0x37258086, "JSF Xeon C35xx/C55xx Non-Transparent Bridge B2B", 502 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 }, 503 { 0x3C0D8086, "SNB Xeon E5/Core i7 Non-Transparent Bridge B2B", 504 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 }, 505 { 0x0E0D8086, "IVT Xeon E5 V2 Non-Transparent Bridge B2B", 506 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 | 507 NTB_SB01BASE_LOCKUP | NTB_BAR_SIZE_4K }, 508 { 0x2F0D8086, "HSX Xeon E5 V3 Non-Transparent Bridge B2B", 509 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 | 510 NTB_SB01BASE_LOCKUP }, 511 { 0x6F0D8086, "BDX Xeon E5 V4 Non-Transparent Bridge B2B", 512 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 | 513 NTB_SB01BASE_LOCKUP }, 514 515 { 0x201C8086, "SKL Xeon E5 V5 Non-Transparent Bridge B2B", 516 NTB_XEON_GEN3, 0 }, 517 }; 518 519 static const struct ntb_reg atom_reg = { 520 .ntb_ctl = ATOM_NTBCNTL_OFFSET, 521 .lnk_sta = ATOM_LINK_STATUS_OFFSET, 522 .db_size = sizeof(uint64_t), 523 .mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2 }, 524 }; 525 526 static const struct ntb_alt_reg atom_pri_reg = { 527 .db_bell = ATOM_PDOORBELL_OFFSET, 528 .db_mask = ATOM_PDBMSK_OFFSET, 529 .spad = ATOM_SPAD_OFFSET, 530 }; 531 532 static const struct ntb_alt_reg atom_b2b_reg = { 533 .db_bell = ATOM_B2B_DOORBELL_OFFSET, 534 .spad = ATOM_B2B_SPAD_OFFSET, 535 }; 536 537 static const struct ntb_xlat_reg atom_sec_xlat = { 538 #if 0 539 /* "FIXME" says the Linux driver. */ 540 .bar0_base = ATOM_SBAR0BASE_OFFSET, 541 .bar2_base = ATOM_SBAR2BASE_OFFSET, 542 .bar4_base = ATOM_SBAR4BASE_OFFSET, 543 544 .bar2_limit = ATOM_SBAR2LMT_OFFSET, 545 .bar4_limit = ATOM_SBAR4LMT_OFFSET, 546 #endif 547 548 .bar2_xlat = ATOM_SBAR2XLAT_OFFSET, 549 .bar4_xlat = ATOM_SBAR4XLAT_OFFSET, 550 }; 551 552 static const struct ntb_reg xeon_reg = { 553 .ntb_ctl = XEON_NTBCNTL_OFFSET, 554 .lnk_sta = XEON_LINK_STATUS_OFFSET, 555 .db_size = sizeof(uint16_t), 556 .mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2, NTB_B2B_BAR_3 }, 557 }; 558 559 static const struct ntb_alt_reg xeon_pri_reg = { 560 .db_bell = XEON_PDOORBELL_OFFSET, 561 .db_mask = XEON_PDBMSK_OFFSET, 562 .spad = XEON_SPAD_OFFSET, 563 }; 564 565 static const struct ntb_alt_reg xeon_b2b_reg = { 566 .db_bell = XEON_B2B_DOORBELL_OFFSET, 567 .spad = XEON_B2B_SPAD_OFFSET, 568 }; 569 570 static const struct ntb_xlat_reg xeon_sec_xlat = { 571 .bar0_base = XEON_SBAR0BASE_OFFSET, 572 .bar2_base = XEON_SBAR2BASE_OFFSET, 573 .bar4_base = XEON_SBAR4BASE_OFFSET, 574 .bar5_base = XEON_SBAR5BASE_OFFSET, 575 576 .bar2_limit = XEON_SBAR2LMT_OFFSET, 577 .bar4_limit = XEON_SBAR4LMT_OFFSET, 578 .bar5_limit = XEON_SBAR5LMT_OFFSET, 579 580 .bar2_xlat = XEON_SBAR2XLAT_OFFSET, 581 .bar4_xlat = XEON_SBAR4XLAT_OFFSET, 582 .bar5_xlat = XEON_SBAR5XLAT_OFFSET, 583 }; 584 585 static struct ntb_b2b_addr xeon_b2b_usd_addr = { 586 .bar0_addr = XEON_B2B_BAR0_ADDR, 587 .bar2_addr64 = XEON_B2B_BAR2_ADDR64, 588 .bar4_addr64 = XEON_B2B_BAR4_ADDR64, 589 .bar4_addr32 = XEON_B2B_BAR4_ADDR32, 590 .bar5_addr32 = XEON_B2B_BAR5_ADDR32, 591 }; 592 593 static struct ntb_b2b_addr xeon_b2b_dsd_addr = { 594 .bar0_addr = XEON_B2B_BAR0_ADDR, 595 .bar2_addr64 = XEON_B2B_BAR2_ADDR64, 596 .bar4_addr64 = XEON_B2B_BAR4_ADDR64, 597 .bar4_addr32 = XEON_B2B_BAR4_ADDR32, 598 .bar5_addr32 = XEON_B2B_BAR5_ADDR32, 599 }; 600 601 static const struct ntb_reg xeon_gen3_reg = { 602 .ntb_ctl = XEON_GEN3_REG_IMNTB_CTRL, 603 .lnk_sta = XEON_GEN3_INT_LNK_STS_OFFSET, 604 .db_size = sizeof(uint32_t), 605 .mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2 }, 606 }; 607 608 static const struct ntb_alt_reg xeon_gen3_pri_reg = { 609 .db_bell = XEON_GEN3_REG_EMDOORBELL, 610 .db_mask = XEON_GEN3_REG_IMINT_DISABLE, 611 .spad = XEON_GEN3_REG_IMSPAD, 612 }; 613 614 static const struct ntb_alt_reg xeon_gen3_b2b_reg = { 615 .db_bell = XEON_GEN3_REG_IMDOORBELL, 616 .db_mask = XEON_GEN3_REG_EMINT_DISABLE, 617 .spad = XEON_GEN3_REG_IMB2B_SSPAD, 618 }; 619 620 static const struct ntb_xlat_reg xeon_gen3_sec_xlat = { 621 .bar0_base = XEON_GEN3_EXT_REG_BAR0BASE, 622 .bar2_base = XEON_GEN3_EXT_REG_BAR1BASE, 623 .bar4_base = XEON_GEN3_EXT_REG_BAR2BASE, 624 625 .bar2_limit = XEON_GEN3_REG_IMBAR1XLIMIT, 626 .bar4_limit = XEON_GEN3_REG_IMBAR2XLIMIT, 627 628 .bar2_xlat = XEON_GEN3_REG_IMBAR1XBASE, 629 .bar4_xlat = XEON_GEN3_REG_IMBAR2XBASE, 630 }; 631 632 SYSCTL_NODE(_hw_ntb, OID_AUTO, xeon_b2b, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 633 "B2B MW segment overrides -- MUST be the same on both sides"); 634 635 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar2_addr64, CTLFLAG_RDTUN, 636 &xeon_b2b_usd_addr.bar2_addr64, 0, "If using B2B topology on Xeon " 637 "hardware, use this 64-bit address on the bus between the NTB devices for " 638 "the window at BAR2, on the upstream side of the link. MUST be the same " 639 "address on both sides."); 640 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar4_addr64, CTLFLAG_RDTUN, 641 &xeon_b2b_usd_addr.bar4_addr64, 0, "See usd_bar2_addr64, but BAR4."); 642 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar4_addr32, CTLFLAG_RDTUN, 643 &xeon_b2b_usd_addr.bar4_addr32, 0, "See usd_bar2_addr64, but BAR4 " 644 "(split-BAR mode)."); 645 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar5_addr32, CTLFLAG_RDTUN, 646 &xeon_b2b_usd_addr.bar5_addr32, 0, "See usd_bar2_addr64, but BAR5 " 647 "(split-BAR mode)."); 648 649 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar2_addr64, CTLFLAG_RDTUN, 650 &xeon_b2b_dsd_addr.bar2_addr64, 0, "If using B2B topology on Xeon " 651 "hardware, use this 64-bit address on the bus between the NTB devices for " 652 "the window at BAR2, on the downstream side of the link. MUST be the same" 653 " address on both sides."); 654 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar4_addr64, CTLFLAG_RDTUN, 655 &xeon_b2b_dsd_addr.bar4_addr64, 0, "See dsd_bar2_addr64, but BAR4."); 656 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar4_addr32, CTLFLAG_RDTUN, 657 &xeon_b2b_dsd_addr.bar4_addr32, 0, "See dsd_bar2_addr64, but BAR4 " 658 "(split-BAR mode)."); 659 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar5_addr32, CTLFLAG_RDTUN, 660 &xeon_b2b_dsd_addr.bar5_addr32, 0, "See dsd_bar2_addr64, but BAR5 " 661 "(split-BAR mode)."); 662 663 /* 664 * OS <-> Driver interface structures 665 */ 666 MALLOC_DEFINE(M_NTB, "ntb_hw", "ntb_hw driver memory allocations"); 667 668 /* 669 * OS <-> Driver linkage functions 670 */ 671 static int 672 intel_ntb_probe(device_t device) 673 { 674 struct ntb_hw_info *p; 675 676 p = intel_ntb_get_device_info(pci_get_devid(device)); 677 if (p == NULL) 678 return (ENXIO); 679 680 device_set_desc(device, p->desc); 681 return (0); 682 } 683 684 static int 685 intel_ntb_attach(device_t device) 686 { 687 struct ntb_softc *ntb; 688 struct ntb_hw_info *p; 689 int error; 690 691 ntb = device_get_softc(device); 692 p = intel_ntb_get_device_info(pci_get_devid(device)); 693 694 ntb->device = device; 695 ntb->type = p->type; 696 ntb->features = p->features; 697 ntb->b2b_mw_idx = B2B_MW_DISABLED; 698 ntb->msix_mw_idx = B2B_MW_DISABLED; 699 700 /* Heartbeat timer for NTB_ATOM since there is no link interrupt */ 701 callout_init(&ntb->heartbeat_timer, 1); 702 callout_init(&ntb->lr_timer, 1); 703 callout_init(&ntb->peer_msix_work, 1); 704 mtx_init(&ntb->db_mask_lock, "ntb hw bits", NULL, MTX_SPIN); 705 706 if (ntb->type == NTB_ATOM) 707 error = intel_ntb_detect_atom(ntb); 708 else if (ntb->type == NTB_XEON_GEN3) 709 error = intel_ntb_detect_xeon_gen3(ntb); 710 else 711 error = intel_ntb_detect_xeon(ntb); 712 if (error != 0) 713 goto out; 714 715 intel_ntb_detect_max_mw(ntb); 716 717 pci_enable_busmaster(ntb->device); 718 719 error = intel_ntb_map_pci_bars(ntb); 720 if (error != 0) 721 goto out; 722 if (ntb->type == NTB_ATOM) 723 error = intel_ntb_atom_init_dev(ntb); 724 else if (ntb->type == NTB_XEON_GEN3) 725 error = intel_ntb_xeon_gen3_init_dev(ntb); 726 else 727 error = intel_ntb_xeon_init_dev(ntb); 728 if (error != 0) 729 goto out; 730 731 intel_ntb_spad_clear(device); 732 733 intel_ntb_poll_link(ntb); 734 735 intel_ntb_sysctl_init(ntb); 736 737 /* Attach children to this controller */ 738 error = ntb_register_device(device); 739 740 out: 741 if (error != 0) 742 intel_ntb_detach(device); 743 return (error); 744 } 745 746 static int 747 intel_ntb_detach(device_t device) 748 { 749 struct ntb_softc *ntb; 750 751 ntb = device_get_softc(device); 752 753 /* Detach & delete all children */ 754 ntb_unregister_device(device); 755 756 if (ntb->self_reg != NULL) { 757 DB_MASK_LOCK(ntb); 758 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_valid_mask); 759 DB_MASK_UNLOCK(ntb); 760 } 761 callout_drain(&ntb->heartbeat_timer); 762 callout_drain(&ntb->lr_timer); 763 callout_drain(&ntb->peer_msix_work); 764 pci_disable_busmaster(ntb->device); 765 if (ntb->type == NTB_XEON_GEN1) 766 intel_ntb_teardown_xeon(ntb); 767 intel_ntb_teardown_interrupts(ntb); 768 769 mtx_destroy(&ntb->db_mask_lock); 770 771 intel_ntb_unmap_pci_bar(ntb); 772 773 return (0); 774 } 775 776 /* 777 * Driver internal routines 778 */ 779 static inline enum ntb_bar 780 intel_ntb_mw_to_bar(struct ntb_softc *ntb, unsigned mw) 781 { 782 783 KASSERT(mw < ntb->mw_count, 784 ("%s: mw:%u > count:%u", __func__, mw, (unsigned)ntb->mw_count)); 785 KASSERT(ntb->reg->mw_bar[mw] != 0, ("invalid mw")); 786 787 return (ntb->reg->mw_bar[mw]); 788 } 789 790 static inline bool 791 bar_is_64bit(struct ntb_softc *ntb, enum ntb_bar bar) 792 { 793 /* XXX This assertion could be stronger. */ 794 KASSERT(bar < NTB_MAX_BARS, ("bogus bar")); 795 return (bar < NTB_B2B_BAR_2 || !HAS_FEATURE(ntb, NTB_SPLIT_BAR)); 796 } 797 798 static inline void 799 bar_get_xlat_params(struct ntb_softc *ntb, enum ntb_bar bar, uint32_t *base, 800 uint32_t *xlat, uint32_t *lmt) 801 { 802 uint32_t basev, lmtv, xlatv; 803 804 switch (bar) { 805 case NTB_B2B_BAR_1: 806 basev = ntb->xlat_reg->bar2_base; 807 lmtv = ntb->xlat_reg->bar2_limit; 808 xlatv = ntb->xlat_reg->bar2_xlat; 809 break; 810 case NTB_B2B_BAR_2: 811 basev = ntb->xlat_reg->bar4_base; 812 lmtv = ntb->xlat_reg->bar4_limit; 813 xlatv = ntb->xlat_reg->bar4_xlat; 814 break; 815 case NTB_B2B_BAR_3: 816 basev = ntb->xlat_reg->bar5_base; 817 lmtv = ntb->xlat_reg->bar5_limit; 818 xlatv = ntb->xlat_reg->bar5_xlat; 819 break; 820 default: 821 KASSERT(bar >= NTB_B2B_BAR_1 && bar < NTB_MAX_BARS, 822 ("bad bar")); 823 basev = lmtv = xlatv = 0; 824 break; 825 } 826 827 if (base != NULL) 828 *base = basev; 829 if (xlat != NULL) 830 *xlat = xlatv; 831 if (lmt != NULL) 832 *lmt = lmtv; 833 } 834 835 static int 836 intel_ntb_map_pci_bars(struct ntb_softc *ntb) 837 { 838 struct ntb_pci_bar_info *bar; 839 int rc; 840 841 bar = &ntb->bar_info[NTB_CONFIG_BAR]; 842 bar->pci_resource_id = PCIR_BAR(0); 843 rc = map_mmr_bar(ntb, bar); 844 if (rc != 0) 845 goto out; 846 847 /* 848 * At least on Xeon v4 NTB device leaks to host some remote side 849 * BAR0 writes supposed to update scratchpad registers. I am not 850 * sure why it happens, but it may be related to the fact that 851 * on a link side BAR0 is 32KB, while on a host side it is 64KB. 852 * Without this hack DMAR blocks those accesses as not allowed. 853 */ 854 if (bus_dma_tag_create(bus_get_dma_tag(ntb->device), 1, 0, 855 BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, 856 bar->size, 1, bar->size, 0, NULL, NULL, &ntb->bar0_dma_tag)) { 857 device_printf(ntb->device, "Unable to create BAR0 tag\n"); 858 return (ENOMEM); 859 } 860 if (bus_dmamap_create(ntb->bar0_dma_tag, 0, &ntb->bar0_dma_map)) { 861 device_printf(ntb->device, "Unable to create BAR0 map\n"); 862 return (ENOMEM); 863 } 864 if (bus_dma_iommu_load_ident(ntb->bar0_dma_tag, ntb->bar0_dma_map, 865 bar->pbase, bar->size, 0)) { 866 device_printf(ntb->device, "Unable to load BAR0 map\n"); 867 return (ENOMEM); 868 } 869 870 bar = &ntb->bar_info[NTB_B2B_BAR_1]; 871 bar->pci_resource_id = PCIR_BAR(2); 872 rc = map_memory_window_bar(ntb, bar); 873 if (rc != 0) 874 goto out; 875 if (ntb->type == NTB_XEON_GEN3) { 876 bar->psz_off = XEON_GEN3_INT_REG_IMBAR1SZ; 877 bar->ssz_off = XEON_GEN3_INT_REG_EMBAR1SZ; 878 bar->pbarxlat_off = XEON_GEN3_REG_EMBAR1XBASE; 879 } else { 880 bar->psz_off = XEON_PBAR23SZ_OFFSET; 881 bar->ssz_off = XEON_SBAR23SZ_OFFSET; 882 bar->pbarxlat_off = XEON_PBAR2XLAT_OFFSET; 883 } 884 885 bar = &ntb->bar_info[NTB_B2B_BAR_2]; 886 bar->pci_resource_id = PCIR_BAR(4); 887 rc = map_memory_window_bar(ntb, bar); 888 if (rc != 0) 889 goto out; 890 if (ntb->type == NTB_XEON_GEN3) { 891 bar->psz_off = XEON_GEN3_INT_REG_IMBAR2SZ; 892 bar->ssz_off = XEON_GEN3_INT_REG_EMBAR2SZ; 893 bar->pbarxlat_off = XEON_GEN3_REG_EMBAR2XBASE; 894 } else { 895 bar->psz_off = XEON_PBAR4SZ_OFFSET; 896 bar->ssz_off = XEON_SBAR4SZ_OFFSET; 897 bar->pbarxlat_off = XEON_PBAR4XLAT_OFFSET; 898 } 899 900 if (!HAS_FEATURE(ntb, NTB_SPLIT_BAR)) 901 goto out; 902 903 if (ntb->type == NTB_XEON_GEN3) { 904 device_printf(ntb->device, "no split bar support\n"); 905 return (ENXIO); 906 } 907 908 bar = &ntb->bar_info[NTB_B2B_BAR_3]; 909 bar->pci_resource_id = PCIR_BAR(5); 910 rc = map_memory_window_bar(ntb, bar); 911 bar->psz_off = XEON_PBAR5SZ_OFFSET; 912 bar->ssz_off = XEON_SBAR5SZ_OFFSET; 913 bar->pbarxlat_off = XEON_PBAR5XLAT_OFFSET; 914 915 out: 916 if (rc != 0) 917 device_printf(ntb->device, 918 "unable to allocate pci resource\n"); 919 return (rc); 920 } 921 922 static void 923 print_map_success(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar, 924 const char *kind) 925 { 926 927 device_printf(ntb->device, 928 "Mapped BAR%d v:[%p-%p] p:[%p-%p] (0x%jx bytes) (%s)\n", 929 PCI_RID2BAR(bar->pci_resource_id), bar->vbase, 930 (char *)bar->vbase + bar->size - 1, 931 (void *)bar->pbase, (void *)(bar->pbase + bar->size - 1), 932 (uintmax_t)bar->size, kind); 933 } 934 935 static int 936 map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar) 937 { 938 939 bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY, 940 &bar->pci_resource_id, RF_ACTIVE); 941 if (bar->pci_resource == NULL) 942 return (ENXIO); 943 944 save_bar_parameters(bar); 945 bar->map_mode = VM_MEMATTR_UNCACHEABLE; 946 print_map_success(ntb, bar, "mmr"); 947 return (0); 948 } 949 950 static int 951 map_memory_window_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar) 952 { 953 int rc; 954 vm_memattr_t mapmode; 955 uint8_t bar_size_bits = 0; 956 957 bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY, 958 &bar->pci_resource_id, RF_ACTIVE); 959 960 if (bar->pci_resource == NULL) 961 return (ENXIO); 962 963 save_bar_parameters(bar); 964 /* 965 * Ivytown NTB BAR sizes are misreported by the hardware due to a 966 * hardware issue. To work around this, query the size it should be 967 * configured to by the device and modify the resource to correspond to 968 * this new size. The BIOS on systems with this problem is required to 969 * provide enough address space to allow the driver to make this change 970 * safely. 971 * 972 * Ideally I could have just specified the size when I allocated the 973 * resource like: 974 * bus_alloc_resource(ntb->device, 975 * SYS_RES_MEMORY, &bar->pci_resource_id, 0ul, ~0ul, 976 * 1ul << bar_size_bits, RF_ACTIVE); 977 * but the PCI driver does not honor the size in this call, so we have 978 * to modify it after the fact. 979 */ 980 if (HAS_FEATURE(ntb, NTB_BAR_SIZE_4K)) { 981 if (bar->pci_resource_id == PCIR_BAR(2)) 982 bar_size_bits = pci_read_config(ntb->device, 983 XEON_PBAR23SZ_OFFSET, 1); 984 else 985 bar_size_bits = pci_read_config(ntb->device, 986 XEON_PBAR45SZ_OFFSET, 1); 987 988 rc = bus_adjust_resource(ntb->device, SYS_RES_MEMORY, 989 bar->pci_resource, bar->pbase, 990 bar->pbase + (1ul << bar_size_bits) - 1); 991 if (rc != 0) { 992 device_printf(ntb->device, 993 "unable to resize bar\n"); 994 return (rc); 995 } 996 997 save_bar_parameters(bar); 998 } 999 1000 bar->map_mode = VM_MEMATTR_UNCACHEABLE; 1001 print_map_success(ntb, bar, "mw"); 1002 1003 /* 1004 * Optionally, mark MW BARs as anything other than UC to improve 1005 * performance. 1006 */ 1007 mapmode = intel_ntb_pat_flags(); 1008 if (mapmode == bar->map_mode) 1009 return (0); 1010 1011 rc = pmap_change_attr((vm_offset_t)bar->vbase, bar->size, mapmode); 1012 if (rc == 0) { 1013 bar->map_mode = mapmode; 1014 device_printf(ntb->device, 1015 "Marked BAR%d v:[%p-%p] p:[%p-%p] as " 1016 "%s.\n", 1017 PCI_RID2BAR(bar->pci_resource_id), bar->vbase, 1018 (char *)bar->vbase + bar->size - 1, 1019 (void *)bar->pbase, (void *)(bar->pbase + bar->size - 1), 1020 intel_ntb_vm_memattr_to_str(mapmode)); 1021 } else 1022 device_printf(ntb->device, 1023 "Unable to mark BAR%d v:[%p-%p] p:[%p-%p] as " 1024 "%s: %d\n", 1025 PCI_RID2BAR(bar->pci_resource_id), bar->vbase, 1026 (char *)bar->vbase + bar->size - 1, 1027 (void *)bar->pbase, (void *)(bar->pbase + bar->size - 1), 1028 intel_ntb_vm_memattr_to_str(mapmode), rc); 1029 /* Proceed anyway */ 1030 return (0); 1031 } 1032 1033 static void 1034 intel_ntb_unmap_pci_bar(struct ntb_softc *ntb) 1035 { 1036 struct ntb_pci_bar_info *bar; 1037 int i; 1038 1039 if (ntb->bar0_dma_map != NULL) { 1040 bus_dmamap_unload(ntb->bar0_dma_tag, ntb->bar0_dma_map); 1041 bus_dmamap_destroy(ntb->bar0_dma_tag, ntb->bar0_dma_map); 1042 } 1043 if (ntb->bar0_dma_tag != NULL) 1044 bus_dma_tag_destroy(ntb->bar0_dma_tag); 1045 for (i = 0; i < NTB_MAX_BARS; i++) { 1046 bar = &ntb->bar_info[i]; 1047 if (bar->pci_resource != NULL) 1048 bus_release_resource(ntb->device, SYS_RES_MEMORY, 1049 bar->pci_resource_id, bar->pci_resource); 1050 } 1051 } 1052 1053 static int 1054 intel_ntb_setup_msix(struct ntb_softc *ntb, uint32_t num_vectors) 1055 { 1056 uint32_t i; 1057 int rc; 1058 1059 for (i = 0; i < num_vectors; i++) { 1060 ntb->int_info[i].rid = i + 1; 1061 ntb->int_info[i].res = bus_alloc_resource_any(ntb->device, 1062 SYS_RES_IRQ, &ntb->int_info[i].rid, RF_ACTIVE); 1063 if (ntb->int_info[i].res == NULL) { 1064 device_printf(ntb->device, 1065 "bus_alloc_resource failed\n"); 1066 return (ENOMEM); 1067 } 1068 ntb->int_info[i].tag = NULL; 1069 ntb->allocated_interrupts++; 1070 rc = bus_setup_intr(ntb->device, ntb->int_info[i].res, 1071 INTR_MPSAFE | INTR_TYPE_MISC, NULL, ndev_vec_isr, 1072 &ntb->msix_vec[i], &ntb->int_info[i].tag); 1073 if (rc != 0) { 1074 device_printf(ntb->device, "bus_setup_intr failed\n"); 1075 return (ENXIO); 1076 } 1077 } 1078 return (0); 1079 } 1080 1081 /* 1082 * The Linux NTB driver drops from MSI-X to legacy INTx if a unique vector 1083 * cannot be allocated for each MSI-X message. JHB seems to think remapping 1084 * should be okay. This tunable should enable us to test that hypothesis 1085 * when someone gets their hands on some Xeon hardware. 1086 */ 1087 static int ntb_force_remap_mode; 1088 SYSCTL_INT(_hw_ntb, OID_AUTO, force_remap_mode, CTLFLAG_RDTUN, 1089 &ntb_force_remap_mode, 0, "If enabled, force MSI-X messages to be remapped" 1090 " to a smaller number of ithreads, even if the desired number are " 1091 "available"); 1092 1093 /* 1094 * In case it is NOT ok, give consumers an abort button. 1095 */ 1096 static int ntb_prefer_intx; 1097 SYSCTL_INT(_hw_ntb, OID_AUTO, prefer_intx_to_remap, CTLFLAG_RDTUN, 1098 &ntb_prefer_intx, 0, "If enabled, prefer to use legacy INTx mode rather " 1099 "than remapping MSI-X messages over available slots (match Linux driver " 1100 "behavior)"); 1101 1102 /* 1103 * Remap the desired number of MSI-X messages to available ithreads in a simple 1104 * round-robin fashion. 1105 */ 1106 static int 1107 intel_ntb_remap_msix(device_t dev, uint32_t desired, uint32_t avail) 1108 { 1109 u_int *vectors; 1110 uint32_t i; 1111 int rc; 1112 1113 if (ntb_prefer_intx != 0) 1114 return (ENXIO); 1115 1116 vectors = malloc(desired * sizeof(*vectors), M_NTB, M_ZERO | M_WAITOK); 1117 1118 for (i = 0; i < desired; i++) 1119 vectors[i] = (i % avail) + 1; 1120 1121 rc = pci_remap_msix(dev, desired, vectors); 1122 free(vectors, M_NTB); 1123 return (rc); 1124 } 1125 1126 static int 1127 intel_ntb_xeon_gen3_init_isr(struct ntb_softc *ntb) 1128 { 1129 uint64_t i, reg; 1130 uint32_t desired_vectors, num_vectors; 1131 int rc; 1132 1133 ntb->allocated_interrupts = 0; 1134 ntb->last_ts = ticks; 1135 1136 /* Mask all the interrupts, including hardware interrupt */ 1137 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_DISABLE, ~0ULL); 1138 1139 /* Clear Interrupt Status */ 1140 reg = intel_ntb_reg_read(8, XEON_GEN3_REG_IMINT_STATUS); 1141 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_STATUS, reg); 1142 1143 num_vectors = desired_vectors = MIN(pci_msix_count(ntb->device), 1144 XEON_GEN3_DB_MSIX_VECTOR_COUNT); 1145 1146 rc = pci_alloc_msix(ntb->device, &num_vectors); 1147 if (rc != 0) { 1148 device_printf(ntb->device, 1149 "Interrupt allocation failed %d\n", rc); 1150 return (rc); 1151 } 1152 if (desired_vectors != num_vectors) { 1153 device_printf(ntb->device, "Couldn't get %d vectors\n", 1154 XEON_GEN3_DB_MSIX_VECTOR_COUNT); 1155 return (ENXIO); 1156 } 1157 /* 32 db + 1 hardware */ 1158 if (num_vectors == XEON_GEN3_DB_MSIX_VECTOR_COUNT) { 1159 /* Program INTVECXX source register */ 1160 for (i = 0; i < XEON_GEN3_DB_MSIX_VECTOR_COUNT; i++) { 1161 /* interrupt source i for vector i */ 1162 intel_ntb_reg_write(1, XEON_GEN3_REG_IMINTVEC00 + i, i); 1163 if (i == (XEON_GEN3_DB_MSIX_VECTOR_COUNT - 1)) { 1164 intel_ntb_reg_write(1, 1165 XEON_GEN3_REG_IMINTVEC00 + i, 1166 XEON_GEN3_LINK_VECTOR_INDEX); 1167 } 1168 } 1169 1170 intel_ntb_create_msix_vec(ntb, num_vectors); 1171 rc = intel_ntb_setup_msix(ntb, num_vectors); 1172 1173 /* enable all interrupts */ 1174 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_DISABLE, 0ULL); 1175 } else { 1176 device_printf(ntb->device, "need to remap interrupts, giving up.\n"); 1177 return (ENXIO); 1178 } 1179 1180 return (0); 1181 } 1182 1183 static int 1184 intel_ntb_init_isr(struct ntb_softc *ntb) 1185 { 1186 uint32_t desired_vectors, num_vectors; 1187 int rc; 1188 1189 ntb->allocated_interrupts = 0; 1190 ntb->last_ts = ticks; 1191 1192 /* 1193 * Mask all doorbell interrupts. (Except link events!) 1194 */ 1195 DB_MASK_LOCK(ntb); 1196 ntb->db_mask = ntb->db_valid_mask; 1197 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask); 1198 DB_MASK_UNLOCK(ntb); 1199 1200 num_vectors = desired_vectors = MIN(pci_msix_count(ntb->device), 1201 ntb->db_count); 1202 if (desired_vectors >= 1) { 1203 rc = pci_alloc_msix(ntb->device, &num_vectors); 1204 1205 if (ntb_force_remap_mode != 0 && rc == 0 && 1206 num_vectors == desired_vectors) 1207 num_vectors--; 1208 1209 if (rc == 0 && num_vectors < desired_vectors) { 1210 rc = intel_ntb_remap_msix(ntb->device, desired_vectors, 1211 num_vectors); 1212 if (rc == 0) 1213 num_vectors = desired_vectors; 1214 else 1215 pci_release_msi(ntb->device); 1216 } 1217 if (rc != 0) 1218 num_vectors = 1; 1219 } else 1220 num_vectors = 1; 1221 1222 if (ntb->type == NTB_XEON_GEN1 && num_vectors < ntb->db_vec_count) { 1223 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 1224 device_printf(ntb->device, 1225 "Errata workaround does not support MSI or INTX\n"); 1226 return (EINVAL); 1227 } 1228 1229 ntb->db_vec_count = 1; 1230 ntb->db_vec_shift = XEON_DB_TOTAL_SHIFT; 1231 rc = intel_ntb_setup_legacy_interrupt(ntb); 1232 } else { 1233 if (num_vectors - 1 != XEON_NONLINK_DB_MSIX_BITS && 1234 HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 1235 device_printf(ntb->device, 1236 "Errata workaround expects %d doorbell bits\n", 1237 XEON_NONLINK_DB_MSIX_BITS); 1238 return (EINVAL); 1239 } 1240 1241 intel_ntb_create_msix_vec(ntb, num_vectors); 1242 rc = intel_ntb_setup_msix(ntb, num_vectors); 1243 } 1244 if (rc != 0) { 1245 device_printf(ntb->device, 1246 "Error allocating interrupts: %d\n", rc); 1247 intel_ntb_free_msix_vec(ntb); 1248 } 1249 1250 return (rc); 1251 } 1252 1253 static int 1254 intel_ntb_setup_legacy_interrupt(struct ntb_softc *ntb) 1255 { 1256 int rc; 1257 1258 ntb->int_info[0].rid = 0; 1259 ntb->int_info[0].res = bus_alloc_resource_any(ntb->device, SYS_RES_IRQ, 1260 &ntb->int_info[0].rid, RF_SHAREABLE|RF_ACTIVE); 1261 if (ntb->int_info[0].res == NULL) { 1262 device_printf(ntb->device, "bus_alloc_resource failed\n"); 1263 return (ENOMEM); 1264 } 1265 1266 ntb->int_info[0].tag = NULL; 1267 ntb->allocated_interrupts = 1; 1268 1269 rc = bus_setup_intr(ntb->device, ntb->int_info[0].res, 1270 INTR_MPSAFE | INTR_TYPE_MISC, NULL, ndev_irq_isr, 1271 ntb, &ntb->int_info[0].tag); 1272 if (rc != 0) { 1273 device_printf(ntb->device, "bus_setup_intr failed\n"); 1274 return (ENXIO); 1275 } 1276 1277 return (0); 1278 } 1279 1280 static void 1281 intel_ntb_teardown_interrupts(struct ntb_softc *ntb) 1282 { 1283 struct ntb_int_info *current_int; 1284 int i; 1285 1286 for (i = 0; i < ntb->allocated_interrupts; i++) { 1287 current_int = &ntb->int_info[i]; 1288 if (current_int->tag != NULL) 1289 bus_teardown_intr(ntb->device, current_int->res, 1290 current_int->tag); 1291 1292 if (current_int->res != NULL) 1293 bus_release_resource(ntb->device, SYS_RES_IRQ, 1294 rman_get_rid(current_int->res), current_int->res); 1295 } 1296 1297 intel_ntb_free_msix_vec(ntb); 1298 pci_release_msi(ntb->device); 1299 } 1300 1301 static inline uint64_t 1302 db_ioread(struct ntb_softc *ntb, uint64_t regoff) 1303 { 1304 1305 switch (ntb->type) { 1306 case NTB_ATOM: 1307 case NTB_XEON_GEN3: 1308 return (intel_ntb_reg_read(8, regoff)); 1309 case NTB_XEON_GEN1: 1310 return (intel_ntb_reg_read(2, regoff)); 1311 } 1312 __assert_unreachable(); 1313 } 1314 1315 static inline void 1316 db_iowrite(struct ntb_softc *ntb, uint64_t regoff, uint64_t val) 1317 { 1318 1319 KASSERT((val & ~ntb->db_valid_mask) == 0, 1320 ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__, 1321 (uintmax_t)(val & ~ntb->db_valid_mask), 1322 (uintmax_t)ntb->db_valid_mask)); 1323 1324 if (regoff == ntb->self_reg->db_mask) 1325 DB_MASK_ASSERT(ntb, MA_OWNED); 1326 db_iowrite_raw(ntb, regoff, val); 1327 } 1328 1329 static inline void 1330 db_iowrite_raw(struct ntb_softc *ntb, uint64_t regoff, uint64_t val) 1331 { 1332 1333 switch (ntb->type) { 1334 case NTB_ATOM: 1335 case NTB_XEON_GEN3: 1336 intel_ntb_reg_write(8, regoff, val); 1337 break; 1338 case NTB_XEON_GEN1: 1339 intel_ntb_reg_write(2, regoff, (uint16_t)val); 1340 break; 1341 } 1342 } 1343 1344 static void 1345 intel_ntb_db_set_mask(device_t dev, uint64_t bits) 1346 { 1347 struct ntb_softc *ntb = device_get_softc(dev); 1348 1349 DB_MASK_LOCK(ntb); 1350 ntb->db_mask |= bits; 1351 if (!HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) 1352 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask); 1353 DB_MASK_UNLOCK(ntb); 1354 } 1355 1356 static void 1357 intel_ntb_db_clear_mask(device_t dev, uint64_t bits) 1358 { 1359 struct ntb_softc *ntb = device_get_softc(dev); 1360 uint64_t ibits; 1361 int i; 1362 1363 KASSERT((bits & ~ntb->db_valid_mask) == 0, 1364 ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__, 1365 (uintmax_t)(bits & ~ntb->db_valid_mask), 1366 (uintmax_t)ntb->db_valid_mask)); 1367 1368 DB_MASK_LOCK(ntb); 1369 ibits = ntb->fake_db & ntb->db_mask & bits; 1370 ntb->db_mask &= ~bits; 1371 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 1372 /* Simulate fake interrupts if unmasked DB bits are set. */ 1373 ntb->force_db |= ibits; 1374 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) { 1375 if ((ibits & intel_ntb_db_vector_mask(dev, i)) != 0) 1376 swi_sched(ntb->int_info[i].tag, 0); 1377 } 1378 } else { 1379 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask); 1380 } 1381 DB_MASK_UNLOCK(ntb); 1382 } 1383 1384 static uint64_t 1385 intel_ntb_db_read(device_t dev) 1386 { 1387 struct ntb_softc *ntb = device_get_softc(dev); 1388 1389 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) 1390 return (ntb->fake_db); 1391 if (ntb->type == NTB_XEON_GEN3) 1392 return (intel_ntb_reg_read(8, XEON_GEN3_REG_IMINT_STATUS)); 1393 else 1394 return (db_ioread(ntb, ntb->self_reg->db_bell)); 1395 } 1396 1397 static void 1398 intel_ntb_db_clear(device_t dev, uint64_t bits) 1399 { 1400 struct ntb_softc *ntb = device_get_softc(dev); 1401 1402 KASSERT((bits & ~ntb->db_valid_mask) == 0, 1403 ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__, 1404 (uintmax_t)(bits & ~ntb->db_valid_mask), 1405 (uintmax_t)ntb->db_valid_mask)); 1406 1407 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 1408 DB_MASK_LOCK(ntb); 1409 ntb->fake_db &= ~bits; 1410 DB_MASK_UNLOCK(ntb); 1411 return; 1412 } 1413 1414 if (ntb->type == NTB_XEON_GEN3) 1415 intel_ntb_reg_write(4, XEON_GEN3_REG_IMINT_STATUS, 1416 (uint32_t)bits); 1417 else 1418 db_iowrite(ntb, ntb->self_reg->db_bell, bits); 1419 } 1420 1421 static inline uint64_t 1422 intel_ntb_vec_mask(struct ntb_softc *ntb, uint64_t db_vector) 1423 { 1424 uint64_t shift, mask; 1425 1426 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 1427 /* 1428 * Remap vectors in custom way to make at least first 1429 * three doorbells to not generate stray events. 1430 * This breaks Linux compatibility (if one existed) 1431 * when more then one DB is used (not by if_ntb). 1432 */ 1433 if (db_vector < XEON_NONLINK_DB_MSIX_BITS - 1) 1434 return (1 << db_vector); 1435 if (db_vector == XEON_NONLINK_DB_MSIX_BITS - 1) 1436 return (0x7ffc); 1437 } 1438 1439 shift = ntb->db_vec_shift; 1440 mask = (1ull << shift) - 1; 1441 return (mask << (shift * db_vector)); 1442 } 1443 1444 static void 1445 intel_ntb_interrupt(struct ntb_softc *ntb, uint32_t vec) 1446 { 1447 uint64_t vec_mask; 1448 1449 ntb->last_ts = ticks; 1450 vec_mask = intel_ntb_vec_mask(ntb, vec); 1451 1452 if (ntb->type == NTB_XEON_GEN3 && vec == XEON_GEN3_LINK_VECTOR_INDEX) 1453 vec_mask |= ntb->db_link_mask; 1454 if ((vec_mask & ntb->db_link_mask) != 0) { 1455 if (intel_ntb_poll_link(ntb)) 1456 ntb_link_event(ntb->device); 1457 if (ntb->type == NTB_XEON_GEN3) 1458 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_STATUS, 1459 intel_ntb_reg_read(8, XEON_GEN3_REG_IMINT_STATUS)); 1460 } 1461 1462 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP) && 1463 (vec_mask & ntb->db_link_mask) == 0) { 1464 DB_MASK_LOCK(ntb); 1465 1466 /* 1467 * Do not report same DB events again if not cleared yet, 1468 * unless the mask was just cleared for them and this 1469 * interrupt handler call can be the consequence of it. 1470 */ 1471 vec_mask &= ~ntb->fake_db | ntb->force_db; 1472 ntb->force_db &= ~vec_mask; 1473 1474 /* Update our internal doorbell register. */ 1475 ntb->fake_db |= vec_mask; 1476 1477 /* Do not report masked DB events. */ 1478 vec_mask &= ~ntb->db_mask; 1479 1480 DB_MASK_UNLOCK(ntb); 1481 } 1482 1483 if ((vec_mask & ntb->db_valid_mask) != 0) 1484 ntb_db_event(ntb->device, vec); 1485 } 1486 1487 static void 1488 ndev_vec_isr(void *arg) 1489 { 1490 struct ntb_vec *nvec = arg; 1491 1492 intel_ntb_interrupt(nvec->ntb, nvec->num); 1493 } 1494 1495 static void 1496 ndev_irq_isr(void *arg) 1497 { 1498 /* If we couldn't set up MSI-X, we only have the one vector. */ 1499 intel_ntb_interrupt(arg, 0); 1500 } 1501 1502 static int 1503 intel_ntb_create_msix_vec(struct ntb_softc *ntb, uint32_t num_vectors) 1504 { 1505 uint32_t i; 1506 1507 ntb->msix_vec = malloc(num_vectors * sizeof(*ntb->msix_vec), M_NTB, 1508 M_ZERO | M_WAITOK); 1509 for (i = 0; i < num_vectors; i++) { 1510 ntb->msix_vec[i].num = i; 1511 ntb->msix_vec[i].ntb = ntb; 1512 } 1513 1514 return (0); 1515 } 1516 1517 static void 1518 intel_ntb_free_msix_vec(struct ntb_softc *ntb) 1519 { 1520 1521 if (ntb->msix_vec == NULL) 1522 return; 1523 1524 free(ntb->msix_vec, M_NTB); 1525 ntb->msix_vec = NULL; 1526 } 1527 1528 static void 1529 intel_ntb_get_msix_info(struct ntb_softc *ntb) 1530 { 1531 struct pci_devinfo *dinfo; 1532 struct pcicfg_msix *msix; 1533 uint32_t laddr, data, i, offset; 1534 1535 dinfo = device_get_ivars(ntb->device); 1536 msix = &dinfo->cfg.msix; 1537 1538 CTASSERT(XEON_NONLINK_DB_MSIX_BITS == nitems(ntb->msix_data)); 1539 1540 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) { 1541 offset = msix->msix_table_offset + i * PCI_MSIX_ENTRY_SIZE; 1542 1543 laddr = bus_read_4(msix->msix_table_res, offset + 1544 PCI_MSIX_ENTRY_LOWER_ADDR); 1545 intel_ntb_printf(2, "local MSIX addr(%u): 0x%x\n", i, laddr); 1546 1547 KASSERT((laddr & MSI_INTEL_ADDR_BASE) == MSI_INTEL_ADDR_BASE, 1548 ("local MSIX addr 0x%x not in MSI base 0x%x", laddr, 1549 MSI_INTEL_ADDR_BASE)); 1550 ntb->msix_data[i].nmd_ofs = laddr; 1551 1552 data = bus_read_4(msix->msix_table_res, offset + 1553 PCI_MSIX_ENTRY_DATA); 1554 intel_ntb_printf(2, "local MSIX data(%u): 0x%x\n", i, data); 1555 1556 ntb->msix_data[i].nmd_data = data; 1557 } 1558 } 1559 1560 static struct ntb_hw_info * 1561 intel_ntb_get_device_info(uint32_t device_id) 1562 { 1563 struct ntb_hw_info *ep; 1564 1565 for (ep = pci_ids; ep < &pci_ids[nitems(pci_ids)]; ep++) { 1566 if (ep->device_id == device_id) 1567 return (ep); 1568 } 1569 return (NULL); 1570 } 1571 1572 static void 1573 intel_ntb_teardown_xeon(struct ntb_softc *ntb) 1574 { 1575 1576 if (ntb->reg != NULL) 1577 intel_ntb_link_disable(ntb->device); 1578 } 1579 1580 static void 1581 intel_ntb_detect_max_mw(struct ntb_softc *ntb) 1582 { 1583 1584 switch (ntb->type) { 1585 case NTB_ATOM: 1586 ntb->mw_count = ATOM_MW_COUNT; 1587 break; 1588 case NTB_XEON_GEN1: 1589 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) 1590 ntb->mw_count = XEON_HSX_SPLIT_MW_COUNT; 1591 else 1592 ntb->mw_count = XEON_SNB_MW_COUNT; 1593 break; 1594 case NTB_XEON_GEN3: 1595 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) 1596 ntb->mw_count = XEON_GEN3_SPLIT_MW_COUNT; 1597 else 1598 ntb->mw_count = XEON_GEN3_MW_COUNT; 1599 break; 1600 } 1601 } 1602 1603 static int 1604 intel_ntb_detect_xeon(struct ntb_softc *ntb) 1605 { 1606 uint8_t ppd, conn_type; 1607 1608 ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 1); 1609 ntb->ppd = ppd; 1610 1611 if ((ppd & XEON_PPD_DEV_TYPE) != 0) 1612 ntb->dev_type = NTB_DEV_DSD; 1613 else 1614 ntb->dev_type = NTB_DEV_USD; 1615 1616 if ((ppd & XEON_PPD_SPLIT_BAR) != 0) 1617 ntb->features |= NTB_SPLIT_BAR; 1618 1619 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP) && 1620 !HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 1621 device_printf(ntb->device, 1622 "Can not apply SB01BASE_LOCKUP workaround " 1623 "with split BARs disabled!\n"); 1624 device_printf(ntb->device, 1625 "Expect system hangs under heavy NTB traffic!\n"); 1626 ntb->features &= ~NTB_SB01BASE_LOCKUP; 1627 } 1628 1629 /* 1630 * SDOORBELL errata workaround gets in the way of SB01BASE_LOCKUP 1631 * errata workaround; only do one at a time. 1632 */ 1633 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) 1634 ntb->features &= ~NTB_SDOORBELL_LOCKUP; 1635 1636 conn_type = ppd & XEON_PPD_CONN_TYPE; 1637 switch (conn_type) { 1638 case NTB_CONN_B2B: 1639 ntb->conn_type = conn_type; 1640 break; 1641 case NTB_CONN_RP: 1642 case NTB_CONN_TRANSPARENT: 1643 default: 1644 device_printf(ntb->device, "Unsupported connection type: %u\n", 1645 (unsigned)conn_type); 1646 return (ENXIO); 1647 } 1648 return (0); 1649 } 1650 1651 static int 1652 intel_ntb_detect_atom(struct ntb_softc *ntb) 1653 { 1654 uint32_t ppd, conn_type; 1655 1656 ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 4); 1657 ntb->ppd = ppd; 1658 1659 if ((ppd & ATOM_PPD_DEV_TYPE) != 0) 1660 ntb->dev_type = NTB_DEV_DSD; 1661 else 1662 ntb->dev_type = NTB_DEV_USD; 1663 1664 conn_type = (ppd & ATOM_PPD_CONN_TYPE) >> 8; 1665 switch (conn_type) { 1666 case NTB_CONN_B2B: 1667 ntb->conn_type = conn_type; 1668 break; 1669 default: 1670 device_printf(ntb->device, "Unsupported NTB configuration\n"); 1671 return (ENXIO); 1672 } 1673 return (0); 1674 } 1675 1676 static int 1677 intel_ntb_detect_xeon_gen3(struct ntb_softc *ntb) 1678 { 1679 uint8_t ppd, conn_type; 1680 1681 ppd = pci_read_config(ntb->device, XEON_GEN3_INT_REG_PPD, 1); 1682 ntb->ppd = ppd; 1683 1684 /* check port definition */ 1685 conn_type = XEON_GEN3_REG_PPD_PORT_DEF_F(ppd); 1686 switch (conn_type) { 1687 case NTB_CONN_B2B: 1688 ntb->conn_type = conn_type; 1689 break; 1690 default: 1691 device_printf(ntb->device, "Unsupported connection type: %u\n", 1692 conn_type); 1693 return (ENXIO); 1694 } 1695 1696 /* check cross link configuration status */ 1697 if (XEON_GEN3_REG_PPD_CONF_STS_F(ppd)) { 1698 /* NTB Port is configured as DSD/USP */ 1699 ntb->dev_type = NTB_DEV_DSD; 1700 } else { 1701 /* NTB Port is configured as USD/DSP */ 1702 ntb->dev_type = NTB_DEV_USD; 1703 } 1704 1705 if (XEON_GEN3_REG_PPD_ONE_MSIX_F(ppd)) { 1706 /* 1707 * This bit when set, causes only a single MSI-X message to be 1708 * generated if MSI-X is enabled. 1709 */ 1710 ntb->features |= NTB_ONE_MSIX; 1711 } 1712 1713 if (XEON_GEN3_REG_PPD_BAR45_SPL_F(ppd)) { 1714 /* BARs 4 and 5 are presented as two 32b non-prefetchable BARs */ 1715 ntb->features |= NTB_SPLIT_BAR; 1716 } 1717 1718 device_printf(ntb->device, "conn type 0x%02x, dev type 0x%02x," 1719 "features 0x%02x\n", ntb->conn_type, ntb->dev_type, ntb->features); 1720 1721 return (0); 1722 } 1723 1724 static int 1725 intel_ntb_xeon_init_dev(struct ntb_softc *ntb) 1726 { 1727 int rc; 1728 1729 ntb->spad_count = XEON_SPAD_COUNT; 1730 ntb->db_count = XEON_DB_COUNT; 1731 ntb->db_link_mask = XEON_DB_LINK_BIT; 1732 ntb->db_vec_count = XEON_DB_MSIX_VECTOR_COUNT; 1733 ntb->db_vec_shift = XEON_DB_MSIX_VECTOR_SHIFT; 1734 1735 if (ntb->conn_type != NTB_CONN_B2B) { 1736 device_printf(ntb->device, "Connection type %d not supported\n", 1737 ntb->conn_type); 1738 return (ENXIO); 1739 } 1740 1741 ntb->reg = &xeon_reg; 1742 ntb->self_reg = &xeon_pri_reg; 1743 ntb->peer_reg = &xeon_b2b_reg; 1744 ntb->xlat_reg = &xeon_sec_xlat; 1745 1746 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 1747 ntb->force_db = ntb->fake_db = 0; 1748 ntb->msix_mw_idx = (ntb->mw_count + g_ntb_msix_idx) % 1749 ntb->mw_count; 1750 intel_ntb_printf(2, "Setting up MSIX mw idx %d means %u\n", 1751 g_ntb_msix_idx, ntb->msix_mw_idx); 1752 rc = intel_ntb_mw_set_wc_internal(ntb, ntb->msix_mw_idx, 1753 VM_MEMATTR_UNCACHEABLE); 1754 KASSERT(rc == 0, ("shouldn't fail")); 1755 } else if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) { 1756 /* 1757 * There is a Xeon hardware errata related to writes to SDOORBELL or 1758 * B2BDOORBELL in conjunction with inbound access to NTB MMIO space, 1759 * which may hang the system. To workaround this, use a memory 1760 * window to access the interrupt and scratch pad registers on the 1761 * remote system. 1762 */ 1763 ntb->b2b_mw_idx = (ntb->mw_count + g_ntb_mw_idx) % 1764 ntb->mw_count; 1765 intel_ntb_printf(2, "Setting up b2b mw idx %d means %u\n", 1766 g_ntb_mw_idx, ntb->b2b_mw_idx); 1767 rc = intel_ntb_mw_set_wc_internal(ntb, ntb->b2b_mw_idx, 1768 VM_MEMATTR_UNCACHEABLE); 1769 KASSERT(rc == 0, ("shouldn't fail")); 1770 } else if (HAS_FEATURE(ntb, NTB_B2BDOORBELL_BIT14)) 1771 /* 1772 * HW Errata on bit 14 of b2bdoorbell register. Writes will not be 1773 * mirrored to the remote system. Shrink the number of bits by one, 1774 * since bit 14 is the last bit. 1775 * 1776 * On REGS_THRU_MW errata mode, we don't use the b2bdoorbell register 1777 * anyway. Nor for non-B2B connection types. 1778 */ 1779 ntb->db_count = XEON_DB_COUNT - 1; 1780 1781 ntb->db_valid_mask = (1ull << ntb->db_count) - 1; 1782 1783 if (ntb->dev_type == NTB_DEV_USD) 1784 rc = xeon_setup_b2b_mw(ntb, &xeon_b2b_dsd_addr, 1785 &xeon_b2b_usd_addr); 1786 else 1787 rc = xeon_setup_b2b_mw(ntb, &xeon_b2b_usd_addr, 1788 &xeon_b2b_dsd_addr); 1789 if (rc != 0) 1790 return (rc); 1791 1792 /* Enable Bus Master and Memory Space on the secondary side */ 1793 intel_ntb_reg_write(2, XEON_SPCICMD_OFFSET, 1794 PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN); 1795 1796 /* 1797 * Mask all doorbell interrupts. 1798 */ 1799 DB_MASK_LOCK(ntb); 1800 ntb->db_mask = ntb->db_valid_mask; 1801 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask); 1802 DB_MASK_UNLOCK(ntb); 1803 1804 rc = intel_ntb_init_isr(ntb); 1805 return (rc); 1806 } 1807 1808 static int 1809 intel_ntb_xeon_gen3_init_dev(struct ntb_softc *ntb) 1810 { 1811 int rc; 1812 1813 ntb->spad_count = XEON_GEN3_SPAD_COUNT; 1814 ntb->db_count = XEON_GEN3_DB_COUNT; 1815 ntb->db_link_mask = XEON_GEN3_DB_LINK_BIT; 1816 ntb->db_vec_count = XEON_GEN3_DB_MSIX_VECTOR_COUNT; 1817 ntb->db_vec_shift = XEON_GEN3_DB_MSIX_VECTOR_SHIFT; 1818 1819 if (ntb->conn_type != NTB_CONN_B2B) { 1820 device_printf(ntb->device, "Connection type %d not supported\n", 1821 ntb->conn_type); 1822 return (ENXIO); 1823 } 1824 1825 ntb->reg = &xeon_gen3_reg; 1826 ntb->self_reg = &xeon_gen3_pri_reg; 1827 ntb->peer_reg = &xeon_gen3_b2b_reg; 1828 ntb->xlat_reg = &xeon_gen3_sec_xlat; 1829 1830 ntb->db_valid_mask = (1ULL << ntb->db_count) - 1; 1831 1832 xeon_gen3_setup_b2b_mw(ntb); 1833 1834 /* Enable Bus Master and Memory Space on the External Side */ 1835 intel_ntb_reg_write(2, XEON_GEN3_EXT_REG_PCI_CMD, 1836 PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN); 1837 1838 /* Setup Interrupt */ 1839 rc = intel_ntb_xeon_gen3_init_isr(ntb); 1840 1841 return (rc); 1842 } 1843 1844 static int 1845 intel_ntb_atom_init_dev(struct ntb_softc *ntb) 1846 { 1847 int error; 1848 1849 KASSERT(ntb->conn_type == NTB_CONN_B2B, 1850 ("Unsupported NTB configuration (%d)\n", ntb->conn_type)); 1851 1852 ntb->spad_count = ATOM_SPAD_COUNT; 1853 ntb->db_count = ATOM_DB_COUNT; 1854 ntb->db_vec_count = ATOM_DB_MSIX_VECTOR_COUNT; 1855 ntb->db_vec_shift = ATOM_DB_MSIX_VECTOR_SHIFT; 1856 ntb->db_valid_mask = (1ull << ntb->db_count) - 1; 1857 1858 ntb->reg = &atom_reg; 1859 ntb->self_reg = &atom_pri_reg; 1860 ntb->peer_reg = &atom_b2b_reg; 1861 ntb->xlat_reg = &atom_sec_xlat; 1862 1863 /* 1864 * FIXME - MSI-X bug on early Atom HW, remove once internal issue is 1865 * resolved. Mask transaction layer internal parity errors. 1866 */ 1867 pci_write_config(ntb->device, 0xFC, 0x4, 4); 1868 1869 configure_atom_secondary_side_bars(ntb); 1870 1871 /* Enable Bus Master and Memory Space on the secondary side */ 1872 intel_ntb_reg_write(2, ATOM_SPCICMD_OFFSET, 1873 PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN); 1874 1875 error = intel_ntb_init_isr(ntb); 1876 if (error != 0) 1877 return (error); 1878 1879 /* Initiate PCI-E link training */ 1880 intel_ntb_link_enable(ntb->device, NTB_SPEED_AUTO, NTB_WIDTH_AUTO); 1881 1882 callout_reset(&ntb->heartbeat_timer, 0, atom_link_hb, ntb); 1883 1884 return (0); 1885 } 1886 1887 /* XXX: Linux driver doesn't seem to do any of this for Atom. */ 1888 static void 1889 configure_atom_secondary_side_bars(struct ntb_softc *ntb) 1890 { 1891 1892 if (ntb->dev_type == NTB_DEV_USD) { 1893 intel_ntb_reg_write(8, ATOM_PBAR2XLAT_OFFSET, 1894 XEON_B2B_BAR2_ADDR64); 1895 intel_ntb_reg_write(8, ATOM_PBAR4XLAT_OFFSET, 1896 XEON_B2B_BAR4_ADDR64); 1897 intel_ntb_reg_write(8, ATOM_MBAR23_OFFSET, XEON_B2B_BAR2_ADDR64); 1898 intel_ntb_reg_write(8, ATOM_MBAR45_OFFSET, XEON_B2B_BAR4_ADDR64); 1899 } else { 1900 intel_ntb_reg_write(8, ATOM_PBAR2XLAT_OFFSET, 1901 XEON_B2B_BAR2_ADDR64); 1902 intel_ntb_reg_write(8, ATOM_PBAR4XLAT_OFFSET, 1903 XEON_B2B_BAR4_ADDR64); 1904 intel_ntb_reg_write(8, ATOM_MBAR23_OFFSET, XEON_B2B_BAR2_ADDR64); 1905 intel_ntb_reg_write(8, ATOM_MBAR45_OFFSET, XEON_B2B_BAR4_ADDR64); 1906 } 1907 } 1908 1909 /* 1910 * When working around Xeon SDOORBELL errata by remapping remote registers in a 1911 * MW, limit the B2B MW to half a MW. By sharing a MW, half the shared MW 1912 * remains for use by a higher layer. 1913 * 1914 * Will only be used if working around SDOORBELL errata and the BIOS-configured 1915 * MW size is sufficiently large. 1916 */ 1917 static unsigned int ntb_b2b_mw_share; 1918 SYSCTL_UINT(_hw_ntb, OID_AUTO, b2b_mw_share, CTLFLAG_RDTUN, &ntb_b2b_mw_share, 1919 0, "If enabled (non-zero), prefer to share half of the B2B peer register " 1920 "MW with higher level consumers. Both sides of the NTB MUST set the same " 1921 "value here."); 1922 1923 static void 1924 xeon_reset_sbar_size(struct ntb_softc *ntb, enum ntb_bar idx, 1925 enum ntb_bar regbar) 1926 { 1927 struct ntb_pci_bar_info *bar; 1928 uint8_t bar_sz; 1929 1930 if (!HAS_FEATURE(ntb, NTB_SPLIT_BAR) && idx >= NTB_B2B_BAR_3) 1931 return; 1932 1933 bar = &ntb->bar_info[idx]; 1934 bar_sz = pci_read_config(ntb->device, bar->psz_off, 1); 1935 if (idx == regbar) { 1936 if (ntb->b2b_off != 0) 1937 bar_sz--; 1938 else 1939 bar_sz = 0; 1940 } 1941 pci_write_config(ntb->device, bar->ssz_off, bar_sz, 1); 1942 bar_sz = pci_read_config(ntb->device, bar->ssz_off, 1); 1943 (void)bar_sz; 1944 } 1945 1946 static void 1947 xeon_set_sbar_base_and_limit(struct ntb_softc *ntb, uint64_t bar_addr, 1948 enum ntb_bar idx, enum ntb_bar regbar) 1949 { 1950 uint64_t reg_val; 1951 uint32_t base_reg, lmt_reg; 1952 1953 bar_get_xlat_params(ntb, idx, &base_reg, NULL, &lmt_reg); 1954 if (idx == regbar) { 1955 if (ntb->b2b_off) 1956 bar_addr += ntb->b2b_off; 1957 else 1958 bar_addr = 0; 1959 } 1960 1961 if (!bar_is_64bit(ntb, idx)) { 1962 intel_ntb_reg_write(4, base_reg, bar_addr); 1963 reg_val = intel_ntb_reg_read(4, base_reg); 1964 (void)reg_val; 1965 1966 intel_ntb_reg_write(4, lmt_reg, bar_addr); 1967 reg_val = intel_ntb_reg_read(4, lmt_reg); 1968 (void)reg_val; 1969 } else { 1970 intel_ntb_reg_write(8, base_reg, bar_addr); 1971 reg_val = intel_ntb_reg_read(8, base_reg); 1972 (void)reg_val; 1973 1974 intel_ntb_reg_write(8, lmt_reg, bar_addr); 1975 reg_val = intel_ntb_reg_read(8, lmt_reg); 1976 (void)reg_val; 1977 } 1978 } 1979 1980 static void 1981 xeon_set_pbar_xlat(struct ntb_softc *ntb, uint64_t base_addr, enum ntb_bar idx) 1982 { 1983 struct ntb_pci_bar_info *bar; 1984 1985 bar = &ntb->bar_info[idx]; 1986 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR) && idx >= NTB_B2B_BAR_2) { 1987 intel_ntb_reg_write(4, bar->pbarxlat_off, base_addr); 1988 base_addr = intel_ntb_reg_read(4, bar->pbarxlat_off); 1989 } else { 1990 intel_ntb_reg_write(8, bar->pbarxlat_off, base_addr); 1991 base_addr = intel_ntb_reg_read(8, bar->pbarxlat_off); 1992 } 1993 (void)base_addr; 1994 } 1995 1996 static int 1997 xeon_setup_b2b_mw(struct ntb_softc *ntb, const struct ntb_b2b_addr *addr, 1998 const struct ntb_b2b_addr *peer_addr) 1999 { 2000 struct ntb_pci_bar_info *b2b_bar; 2001 vm_size_t bar_size; 2002 uint64_t bar_addr; 2003 enum ntb_bar b2b_bar_num, i; 2004 2005 if (ntb->b2b_mw_idx == B2B_MW_DISABLED) { 2006 b2b_bar = NULL; 2007 b2b_bar_num = NTB_CONFIG_BAR; 2008 ntb->b2b_off = 0; 2009 } else { 2010 b2b_bar_num = intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx); 2011 KASSERT(b2b_bar_num > 0 && b2b_bar_num < NTB_MAX_BARS, 2012 ("invalid b2b mw bar")); 2013 2014 b2b_bar = &ntb->bar_info[b2b_bar_num]; 2015 bar_size = b2b_bar->size; 2016 2017 if (ntb_b2b_mw_share != 0 && 2018 (bar_size >> 1) >= XEON_B2B_MIN_SIZE) 2019 ntb->b2b_off = bar_size >> 1; 2020 else if (bar_size >= XEON_B2B_MIN_SIZE) { 2021 ntb->b2b_off = 0; 2022 } else { 2023 device_printf(ntb->device, 2024 "B2B bar size is too small!\n"); 2025 return (EIO); 2026 } 2027 } 2028 2029 /* 2030 * Reset the secondary bar sizes to match the primary bar sizes. 2031 * (Except, disable or halve the size of the B2B secondary bar.) 2032 */ 2033 for (i = NTB_B2B_BAR_1; i < NTB_MAX_BARS; i++) 2034 xeon_reset_sbar_size(ntb, i, b2b_bar_num); 2035 2036 bar_addr = 0; 2037 if (b2b_bar_num == NTB_CONFIG_BAR) 2038 bar_addr = addr->bar0_addr; 2039 else if (b2b_bar_num == NTB_B2B_BAR_1) 2040 bar_addr = addr->bar2_addr64; 2041 else if (b2b_bar_num == NTB_B2B_BAR_2 && !HAS_FEATURE(ntb, NTB_SPLIT_BAR)) 2042 bar_addr = addr->bar4_addr64; 2043 else if (b2b_bar_num == NTB_B2B_BAR_2) 2044 bar_addr = addr->bar4_addr32; 2045 else if (b2b_bar_num == NTB_B2B_BAR_3) 2046 bar_addr = addr->bar5_addr32; 2047 else 2048 KASSERT(false, ("invalid bar")); 2049 2050 intel_ntb_reg_write(8, XEON_SBAR0BASE_OFFSET, bar_addr); 2051 2052 /* 2053 * Other SBARs are normally hit by the PBAR xlat, except for the b2b 2054 * register BAR. The B2B BAR is either disabled above or configured 2055 * half-size. It starts at PBAR xlat + offset. 2056 * 2057 * Also set up incoming BAR limits == base (zero length window). 2058 */ 2059 xeon_set_sbar_base_and_limit(ntb, addr->bar2_addr64, NTB_B2B_BAR_1, 2060 b2b_bar_num); 2061 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 2062 xeon_set_sbar_base_and_limit(ntb, addr->bar4_addr32, 2063 NTB_B2B_BAR_2, b2b_bar_num); 2064 xeon_set_sbar_base_and_limit(ntb, addr->bar5_addr32, 2065 NTB_B2B_BAR_3, b2b_bar_num); 2066 } else 2067 xeon_set_sbar_base_and_limit(ntb, addr->bar4_addr64, 2068 NTB_B2B_BAR_2, b2b_bar_num); 2069 2070 /* Zero incoming translation addrs */ 2071 intel_ntb_reg_write(8, XEON_SBAR2XLAT_OFFSET, 0); 2072 intel_ntb_reg_write(8, XEON_SBAR4XLAT_OFFSET, 0); 2073 2074 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 2075 uint32_t xlat_reg, lmt_reg; 2076 enum ntb_bar bar_num; 2077 2078 /* 2079 * We point the chosen MSIX MW BAR xlat to remote LAPIC for 2080 * workaround 2081 */ 2082 bar_num = intel_ntb_mw_to_bar(ntb, ntb->msix_mw_idx); 2083 bar_get_xlat_params(ntb, bar_num, NULL, &xlat_reg, &lmt_reg); 2084 if (bar_is_64bit(ntb, bar_num)) { 2085 intel_ntb_reg_write(8, xlat_reg, MSI_INTEL_ADDR_BASE); 2086 ntb->msix_xlat = intel_ntb_reg_read(8, xlat_reg); 2087 intel_ntb_reg_write(8, lmt_reg, 0); 2088 } else { 2089 intel_ntb_reg_write(4, xlat_reg, MSI_INTEL_ADDR_BASE); 2090 ntb->msix_xlat = intel_ntb_reg_read(4, xlat_reg); 2091 intel_ntb_reg_write(4, lmt_reg, 0); 2092 } 2093 2094 ntb->peer_lapic_bar = &ntb->bar_info[bar_num]; 2095 } 2096 (void)intel_ntb_reg_read(8, XEON_SBAR2XLAT_OFFSET); 2097 (void)intel_ntb_reg_read(8, XEON_SBAR4XLAT_OFFSET); 2098 2099 /* Zero outgoing translation limits (whole bar size windows) */ 2100 intel_ntb_reg_write(8, XEON_PBAR2LMT_OFFSET, 0); 2101 intel_ntb_reg_write(8, XEON_PBAR4LMT_OFFSET, 0); 2102 2103 /* Set outgoing translation offsets */ 2104 xeon_set_pbar_xlat(ntb, peer_addr->bar2_addr64, NTB_B2B_BAR_1); 2105 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 2106 xeon_set_pbar_xlat(ntb, peer_addr->bar4_addr32, NTB_B2B_BAR_2); 2107 xeon_set_pbar_xlat(ntb, peer_addr->bar5_addr32, NTB_B2B_BAR_3); 2108 } else 2109 xeon_set_pbar_xlat(ntb, peer_addr->bar4_addr64, NTB_B2B_BAR_2); 2110 2111 /* Set the translation offset for B2B registers */ 2112 bar_addr = 0; 2113 if (b2b_bar_num == NTB_CONFIG_BAR) 2114 bar_addr = peer_addr->bar0_addr; 2115 else if (b2b_bar_num == NTB_B2B_BAR_1) 2116 bar_addr = peer_addr->bar2_addr64; 2117 else if (b2b_bar_num == NTB_B2B_BAR_2 && !HAS_FEATURE(ntb, NTB_SPLIT_BAR)) 2118 bar_addr = peer_addr->bar4_addr64; 2119 else if (b2b_bar_num == NTB_B2B_BAR_2) 2120 bar_addr = peer_addr->bar4_addr32; 2121 else if (b2b_bar_num == NTB_B2B_BAR_3) 2122 bar_addr = peer_addr->bar5_addr32; 2123 else 2124 KASSERT(false, ("invalid bar")); 2125 2126 /* 2127 * B2B_XLAT_OFFSET is a 64-bit register but can only be written 32 bits 2128 * at a time. 2129 */ 2130 intel_ntb_reg_write(4, XEON_B2B_XLAT_OFFSETL, bar_addr & 0xffffffff); 2131 intel_ntb_reg_write(4, XEON_B2B_XLAT_OFFSETU, bar_addr >> 32); 2132 return (0); 2133 } 2134 2135 static int 2136 xeon_gen3_setup_b2b_mw(struct ntb_softc *ntb) 2137 { 2138 uint64_t reg; 2139 uint32_t embarsz, imbarsz; 2140 2141 /* IMBAR1SZ should be equal to EMBAR1SZ */ 2142 embarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_EMBAR1SZ, 1); 2143 imbarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_IMBAR1SZ, 1); 2144 if (embarsz != imbarsz) { 2145 device_printf(ntb->device, 2146 "IMBAR1SZ (%u) should be equal to EMBAR1SZ (%u)\n", 2147 imbarsz, embarsz); 2148 return (EIO); 2149 } 2150 2151 /* IMBAR2SZ should be equal to EMBAR2SZ */ 2152 embarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_EMBAR2SZ, 1); 2153 imbarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_IMBAR2SZ, 1); 2154 if (embarsz != imbarsz) { 2155 device_printf(ntb->device, 2156 "IMBAR2SZ (%u) should be equal to EMBAR2SZ (%u)\n", 2157 imbarsz, embarsz); 2158 return (EIO); 2159 } 2160 2161 /* Client will provide the incoming IMBAR1/2XBASE, zero it for now */ 2162 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR1XBASE, 0); 2163 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR2XBASE, 0); 2164 2165 /* 2166 * If the value in EMBAR1LIMIT is set equal to the value in EMBAR1, 2167 * the memory window for EMBAR1 is disabled. 2168 * Note: It is needed to avoid malacious access. 2169 */ 2170 reg = pci_read_config(ntb->device, XEON_GEN3_EXT_REG_BAR1BASE, 8); 2171 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR1XLIMIT, reg); 2172 2173 reg = pci_read_config(ntb->device, XEON_GEN3_EXT_REG_BAR2BASE, 8); 2174 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR2XLIMIT, reg); 2175 2176 return (0); 2177 } 2178 2179 static inline bool 2180 _xeon_link_is_up(struct ntb_softc *ntb) 2181 { 2182 2183 if (ntb->conn_type == NTB_CONN_TRANSPARENT) 2184 return (true); 2185 return ((ntb->lnk_sta & NTB_LINK_STATUS_ACTIVE) != 0); 2186 } 2187 2188 static inline bool 2189 link_is_up(struct ntb_softc *ntb) 2190 { 2191 2192 if (ntb->type == NTB_XEON_GEN1 || ntb->type == NTB_XEON_GEN3) 2193 return (_xeon_link_is_up(ntb) && (ntb->peer_msix_good || 2194 !HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP))); 2195 2196 KASSERT(ntb->type == NTB_ATOM, ("ntb type")); 2197 return ((ntb->ntb_ctl & ATOM_CNTL_LINK_DOWN) == 0); 2198 } 2199 2200 static inline bool 2201 atom_link_is_err(struct ntb_softc *ntb) 2202 { 2203 uint32_t status; 2204 2205 KASSERT(ntb->type == NTB_ATOM, ("ntb type")); 2206 2207 status = intel_ntb_reg_read(4, ATOM_LTSSMSTATEJMP_OFFSET); 2208 if ((status & ATOM_LTSSMSTATEJMP_FORCEDETECT) != 0) 2209 return (true); 2210 2211 status = intel_ntb_reg_read(4, ATOM_IBSTERRRCRVSTS0_OFFSET); 2212 return ((status & ATOM_IBIST_ERR_OFLOW) != 0); 2213 } 2214 2215 /* Atom does not have link status interrupt, poll on that platform */ 2216 static void 2217 atom_link_hb(void *arg) 2218 { 2219 struct ntb_softc *ntb = arg; 2220 sbintime_t timo, poll_ts; 2221 2222 timo = NTB_HB_TIMEOUT * hz; 2223 poll_ts = ntb->last_ts + timo; 2224 2225 /* 2226 * Delay polling the link status if an interrupt was received, unless 2227 * the cached link status says the link is down. 2228 */ 2229 if ((sbintime_t)ticks - poll_ts < 0 && link_is_up(ntb)) { 2230 timo = poll_ts - ticks; 2231 goto out; 2232 } 2233 2234 if (intel_ntb_poll_link(ntb)) 2235 ntb_link_event(ntb->device); 2236 2237 if (!link_is_up(ntb) && atom_link_is_err(ntb)) { 2238 /* Link is down with error, proceed with recovery */ 2239 callout_reset(&ntb->lr_timer, 0, recover_atom_link, ntb); 2240 return; 2241 } 2242 2243 out: 2244 callout_reset(&ntb->heartbeat_timer, timo, atom_link_hb, ntb); 2245 } 2246 2247 static void 2248 atom_perform_link_restart(struct ntb_softc *ntb) 2249 { 2250 uint32_t status; 2251 2252 /* Driver resets the NTB ModPhy lanes - magic! */ 2253 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG6, 0xe0); 2254 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG4, 0x40); 2255 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG4, 0x60); 2256 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG6, 0x60); 2257 2258 /* Driver waits 100ms to allow the NTB ModPhy to settle */ 2259 pause("ModPhy", hz / 10); 2260 2261 /* Clear AER Errors, write to clear */ 2262 status = intel_ntb_reg_read(4, ATOM_ERRCORSTS_OFFSET); 2263 status &= PCIM_AER_COR_REPLAY_ROLLOVER; 2264 intel_ntb_reg_write(4, ATOM_ERRCORSTS_OFFSET, status); 2265 2266 /* Clear unexpected electrical idle event in LTSSM, write to clear */ 2267 status = intel_ntb_reg_read(4, ATOM_LTSSMERRSTS0_OFFSET); 2268 status |= ATOM_LTSSMERRSTS0_UNEXPECTEDEI; 2269 intel_ntb_reg_write(4, ATOM_LTSSMERRSTS0_OFFSET, status); 2270 2271 /* Clear DeSkew Buffer error, write to clear */ 2272 status = intel_ntb_reg_read(4, ATOM_DESKEWSTS_OFFSET); 2273 status |= ATOM_DESKEWSTS_DBERR; 2274 intel_ntb_reg_write(4, ATOM_DESKEWSTS_OFFSET, status); 2275 2276 status = intel_ntb_reg_read(4, ATOM_IBSTERRRCRVSTS0_OFFSET); 2277 status &= ATOM_IBIST_ERR_OFLOW; 2278 intel_ntb_reg_write(4, ATOM_IBSTERRRCRVSTS0_OFFSET, status); 2279 2280 /* Releases the NTB state machine to allow the link to retrain */ 2281 status = intel_ntb_reg_read(4, ATOM_LTSSMSTATEJMP_OFFSET); 2282 status &= ~ATOM_LTSSMSTATEJMP_FORCEDETECT; 2283 intel_ntb_reg_write(4, ATOM_LTSSMSTATEJMP_OFFSET, status); 2284 } 2285 2286 static int 2287 intel_ntb_port_number(device_t dev) 2288 { 2289 struct ntb_softc *ntb = device_get_softc(dev); 2290 2291 return (ntb->dev_type == NTB_DEV_USD ? 0 : 1); 2292 } 2293 2294 static int 2295 intel_ntb_peer_port_count(device_t dev) 2296 { 2297 2298 return (1); 2299 } 2300 2301 static int 2302 intel_ntb_peer_port_number(device_t dev, int pidx) 2303 { 2304 struct ntb_softc *ntb = device_get_softc(dev); 2305 2306 if (pidx != 0) 2307 return (-EINVAL); 2308 2309 return (ntb->dev_type == NTB_DEV_USD ? 1 : 0); 2310 } 2311 2312 static int 2313 intel_ntb_peer_port_idx(device_t dev, int port) 2314 { 2315 int peer_port; 2316 2317 peer_port = intel_ntb_peer_port_number(dev, 0); 2318 if (peer_port == -EINVAL || port != peer_port) 2319 return (-EINVAL); 2320 2321 return (0); 2322 } 2323 2324 static int 2325 intel_ntb_link_enable(device_t dev, enum ntb_speed speed __unused, 2326 enum ntb_width width __unused) 2327 { 2328 struct ntb_softc *ntb = device_get_softc(dev); 2329 uint32_t cntl; 2330 2331 intel_ntb_printf(2, "%s\n", __func__); 2332 2333 if (ntb->type == NTB_ATOM) { 2334 pci_write_config(ntb->device, NTB_PPD_OFFSET, 2335 ntb->ppd | ATOM_PPD_INIT_LINK, 4); 2336 return (0); 2337 } 2338 2339 if (ntb->conn_type == NTB_CONN_TRANSPARENT) { 2340 ntb_link_event(dev); 2341 return (0); 2342 } 2343 2344 cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl); 2345 cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK); 2346 cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP; 2347 cntl |= NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP; 2348 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) 2349 cntl |= NTB_CNTL_P2S_BAR5_SNOOP | NTB_CNTL_S2P_BAR5_SNOOP; 2350 intel_ntb_reg_write(4, ntb->reg->ntb_ctl, cntl); 2351 return (0); 2352 } 2353 2354 static int 2355 intel_ntb_link_disable(device_t dev) 2356 { 2357 struct ntb_softc *ntb = device_get_softc(dev); 2358 uint32_t cntl; 2359 2360 intel_ntb_printf(2, "%s\n", __func__); 2361 2362 if (ntb->conn_type == NTB_CONN_TRANSPARENT) { 2363 ntb_link_event(dev); 2364 return (0); 2365 } 2366 2367 cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl); 2368 cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP); 2369 cntl &= ~(NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP); 2370 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) 2371 cntl &= ~(NTB_CNTL_P2S_BAR5_SNOOP | NTB_CNTL_S2P_BAR5_SNOOP); 2372 cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK; 2373 intel_ntb_reg_write(4, ntb->reg->ntb_ctl, cntl); 2374 return (0); 2375 } 2376 2377 static bool 2378 intel_ntb_link_enabled(device_t dev) 2379 { 2380 struct ntb_softc *ntb = device_get_softc(dev); 2381 uint32_t cntl; 2382 2383 if (ntb->type == NTB_ATOM) { 2384 cntl = pci_read_config(ntb->device, NTB_PPD_OFFSET, 4); 2385 return ((cntl & ATOM_PPD_INIT_LINK) != 0); 2386 } 2387 2388 if (ntb->conn_type == NTB_CONN_TRANSPARENT) 2389 return (true); 2390 2391 cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl); 2392 return ((cntl & NTB_CNTL_LINK_DISABLE) == 0); 2393 } 2394 2395 static void 2396 recover_atom_link(void *arg) 2397 { 2398 struct ntb_softc *ntb = arg; 2399 unsigned speed, width, oldspeed, oldwidth; 2400 uint32_t status32; 2401 2402 atom_perform_link_restart(ntb); 2403 2404 /* 2405 * There is a potential race between the 2 NTB devices recovering at 2406 * the same time. If the times are the same, the link will not recover 2407 * and the driver will be stuck in this loop forever. Add a random 2408 * interval to the recovery time to prevent this race. 2409 */ 2410 status32 = arc4random() % ATOM_LINK_RECOVERY_TIME; 2411 pause("Link", (ATOM_LINK_RECOVERY_TIME + status32) * hz / 1000); 2412 2413 if (atom_link_is_err(ntb)) 2414 goto retry; 2415 2416 status32 = intel_ntb_reg_read(4, ntb->reg->ntb_ctl); 2417 if ((status32 & ATOM_CNTL_LINK_DOWN) != 0) 2418 goto out; 2419 2420 status32 = intel_ntb_reg_read(4, ntb->reg->lnk_sta); 2421 width = NTB_LNK_STA_WIDTH(status32); 2422 speed = status32 & NTB_LINK_SPEED_MASK; 2423 2424 oldwidth = NTB_LNK_STA_WIDTH(ntb->lnk_sta); 2425 oldspeed = ntb->lnk_sta & NTB_LINK_SPEED_MASK; 2426 if (oldwidth != width || oldspeed != speed) 2427 goto retry; 2428 2429 out: 2430 callout_reset(&ntb->heartbeat_timer, NTB_HB_TIMEOUT * hz, atom_link_hb, 2431 ntb); 2432 return; 2433 2434 retry: 2435 callout_reset(&ntb->lr_timer, NTB_HB_TIMEOUT * hz, recover_atom_link, 2436 ntb); 2437 } 2438 2439 /* 2440 * Polls the HW link status register(s); returns true if something has changed. 2441 */ 2442 static bool 2443 intel_ntb_poll_link(struct ntb_softc *ntb) 2444 { 2445 uint32_t ntb_cntl; 2446 uint16_t reg_val; 2447 2448 if (ntb->type == NTB_ATOM) { 2449 ntb_cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl); 2450 if (ntb_cntl == ntb->ntb_ctl) 2451 return (false); 2452 2453 ntb->ntb_ctl = ntb_cntl; 2454 ntb->lnk_sta = intel_ntb_reg_read(4, ntb->reg->lnk_sta); 2455 } else { 2456 if (ntb->type == NTB_XEON_GEN1) 2457 db_iowrite_raw(ntb, ntb->self_reg->db_bell, 2458 ntb->db_link_mask); 2459 2460 reg_val = pci_read_config(ntb->device, ntb->reg->lnk_sta, 2); 2461 if (reg_val == ntb->lnk_sta) 2462 return (false); 2463 2464 ntb->lnk_sta = reg_val; 2465 2466 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 2467 if (_xeon_link_is_up(ntb)) { 2468 if (!ntb->peer_msix_good) { 2469 callout_reset(&ntb->peer_msix_work, 0, 2470 intel_ntb_exchange_msix, ntb); 2471 return (false); 2472 } 2473 } else { 2474 ntb->peer_msix_good = false; 2475 ntb->peer_msix_done = false; 2476 } 2477 } 2478 } 2479 return (true); 2480 } 2481 2482 static inline enum ntb_speed 2483 intel_ntb_link_sta_speed(struct ntb_softc *ntb) 2484 { 2485 2486 if (!link_is_up(ntb)) 2487 return (NTB_SPEED_NONE); 2488 return (ntb->lnk_sta & NTB_LINK_SPEED_MASK); 2489 } 2490 2491 static inline enum ntb_width 2492 intel_ntb_link_sta_width(struct ntb_softc *ntb) 2493 { 2494 2495 if (!link_is_up(ntb)) 2496 return (NTB_WIDTH_NONE); 2497 return (NTB_LNK_STA_WIDTH(ntb->lnk_sta)); 2498 } 2499 2500 SYSCTL_NODE(_hw_ntb, OID_AUTO, debug_info, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 2501 "Driver state, statistics, and HW registers"); 2502 2503 #define NTB_REGSZ_MASK (3ul << 30) 2504 #define NTB_REG_64 (1ul << 30) 2505 #define NTB_REG_32 (2ul << 30) 2506 #define NTB_REG_16 (3ul << 30) 2507 #define NTB_REG_8 (0ul << 30) 2508 2509 #define NTB_DB_READ (1ul << 29) 2510 #define NTB_PCI_REG (1ul << 28) 2511 #define NTB_REGFLAGS_MASK (NTB_REGSZ_MASK | NTB_DB_READ | NTB_PCI_REG) 2512 2513 static void 2514 intel_ntb_sysctl_init(struct ntb_softc *ntb) 2515 { 2516 struct sysctl_oid_list *globals, *tree_par, *regpar, *statpar, *errpar; 2517 struct sysctl_ctx_list *ctx; 2518 struct sysctl_oid *tree, *tmptree; 2519 2520 ctx = device_get_sysctl_ctx(ntb->device); 2521 globals = SYSCTL_CHILDREN(device_get_sysctl_tree(ntb->device)); 2522 2523 SYSCTL_ADD_PROC(ctx, globals, OID_AUTO, "link_status", 2524 CTLFLAG_RD | CTLTYPE_STRING | CTLFLAG_MPSAFE, ntb, 0, 2525 sysctl_handle_link_status_human, "A", 2526 "Link status (human readable)"); 2527 SYSCTL_ADD_PROC(ctx, globals, OID_AUTO, "active", 2528 CTLFLAG_RD | CTLTYPE_UINT | CTLFLAG_MPSAFE, ntb, 0, 2529 sysctl_handle_link_status, "IU", 2530 "Link status (1=active, 0=inactive)"); 2531 SYSCTL_ADD_PROC(ctx, globals, OID_AUTO, "admin_up", 2532 CTLFLAG_RW | CTLTYPE_UINT | CTLFLAG_MPSAFE, ntb, 0, 2533 sysctl_handle_link_admin, "IU", 2534 "Set/get interface status (1=UP, 0=DOWN)"); 2535 2536 tree = SYSCTL_ADD_NODE(ctx, globals, OID_AUTO, "debug_info", 2537 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 2538 "Driver state, statistics, and HW registers"); 2539 tree_par = SYSCTL_CHILDREN(tree); 2540 2541 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "conn_type", CTLFLAG_RD, 2542 &ntb->conn_type, 0, "0 - Transparent; 1 - B2B; 2 - Root Port"); 2543 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "dev_type", CTLFLAG_RD, 2544 &ntb->dev_type, 0, "0 - USD; 1 - DSD"); 2545 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "ppd", CTLFLAG_RD, 2546 &ntb->ppd, 0, "Raw PPD register (cached)"); 2547 2548 if (ntb->b2b_mw_idx != B2B_MW_DISABLED) { 2549 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "b2b_idx", CTLFLAG_RD, 2550 &ntb->b2b_mw_idx, 0, 2551 "Index of the MW used for B2B remote register access"); 2552 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "b2b_off", 2553 CTLFLAG_RD, &ntb->b2b_off, 2554 "If non-zero, offset of B2B register region in shared MW"); 2555 } 2556 2557 SYSCTL_ADD_PROC(ctx, tree_par, OID_AUTO, "features", 2558 CTLFLAG_RD | CTLTYPE_STRING | CTLFLAG_MPSAFE, ntb, 0, 2559 sysctl_handle_features, "A", "Features/errata of this NTB device"); 2560 2561 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "ntb_ctl", CTLFLAG_RD, 2562 __DEVOLATILE(uint32_t *, &ntb->ntb_ctl), 0, 2563 "NTB CTL register (cached)"); 2564 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "lnk_sta", CTLFLAG_RD, 2565 __DEVOLATILE(uint32_t *, &ntb->lnk_sta), 0, 2566 "LNK STA register (cached)"); 2567 2568 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "mw_count", CTLFLAG_RD, 2569 &ntb->mw_count, 0, "MW count"); 2570 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "spad_count", CTLFLAG_RD, 2571 &ntb->spad_count, 0, "Scratchpad count"); 2572 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_count", CTLFLAG_RD, 2573 &ntb->db_count, 0, "Doorbell count"); 2574 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_vec_count", CTLFLAG_RD, 2575 &ntb->db_vec_count, 0, "Doorbell vector count"); 2576 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_vec_shift", CTLFLAG_RD, 2577 &ntb->db_vec_shift, 0, "Doorbell vector shift"); 2578 2579 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_valid_mask", CTLFLAG_RD, 2580 &ntb->db_valid_mask, "Doorbell valid mask"); 2581 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_link_mask", CTLFLAG_RD, 2582 &ntb->db_link_mask, "Doorbell link mask"); 2583 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_mask", CTLFLAG_RD, 2584 &ntb->db_mask, "Doorbell mask (cached)"); 2585 2586 tmptree = SYSCTL_ADD_NODE(ctx, tree_par, OID_AUTO, "registers", 2587 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 2588 "Raw HW registers (big-endian)"); 2589 regpar = SYSCTL_CHILDREN(tmptree); 2590 2591 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "ntbcntl", 2592 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2593 NTB_REG_32 | ntb->reg->ntb_ctl, sysctl_handle_register, "IU", 2594 "NTB Control register"); 2595 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnkcap", 2596 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2597 NTB_REG_32 | 0x19c, sysctl_handle_register, "IU", 2598 "NTB Link Capabilities"); 2599 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnkcon", 2600 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2601 NTB_REG_32 | 0x1a0, sysctl_handle_register, "IU", 2602 "NTB Link Control register"); 2603 2604 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "db_mask", 2605 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2606 NTB_REG_64 | NTB_DB_READ | ntb->self_reg->db_mask, 2607 sysctl_handle_register, "QU", "Doorbell mask register"); 2608 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "db_bell", 2609 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2610 NTB_REG_64 | NTB_DB_READ | ntb->self_reg->db_bell, 2611 sysctl_handle_register, "QU", "Doorbell register"); 2612 2613 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat23", 2614 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2615 NTB_REG_64 | ntb->xlat_reg->bar2_xlat, 2616 sysctl_handle_register, "QU", "Incoming XLAT23 register"); 2617 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 2618 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat4", 2619 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2620 NTB_REG_32 | ntb->xlat_reg->bar4_xlat, 2621 sysctl_handle_register, "IU", "Incoming XLAT4 register"); 2622 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat5", 2623 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2624 NTB_REG_32 | ntb->xlat_reg->bar5_xlat, 2625 sysctl_handle_register, "IU", "Incoming XLAT5 register"); 2626 } else { 2627 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat45", 2628 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2629 NTB_REG_64 | ntb->xlat_reg->bar4_xlat, 2630 sysctl_handle_register, "QU", "Incoming XLAT45 register"); 2631 } 2632 2633 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt23", 2634 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2635 NTB_REG_64 | ntb->xlat_reg->bar2_limit, 2636 sysctl_handle_register, "QU", "Incoming LMT23 register"); 2637 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 2638 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt4", 2639 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2640 NTB_REG_32 | ntb->xlat_reg->bar4_limit, 2641 sysctl_handle_register, "IU", "Incoming LMT4 register"); 2642 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt5", 2643 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2644 NTB_REG_32 | ntb->xlat_reg->bar5_limit, 2645 sysctl_handle_register, "IU", "Incoming LMT5 register"); 2646 } else { 2647 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt45", 2648 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2649 NTB_REG_64 | ntb->xlat_reg->bar4_limit, 2650 sysctl_handle_register, "QU", "Incoming LMT45 register"); 2651 } 2652 2653 if (ntb->type == NTB_ATOM) 2654 return; 2655 2656 tmptree = SYSCTL_ADD_NODE(ctx, regpar, OID_AUTO, "xeon_stats", 2657 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Xeon HW statistics"); 2658 statpar = SYSCTL_CHILDREN(tmptree); 2659 SYSCTL_ADD_PROC(ctx, statpar, OID_AUTO, "upstream_mem_miss", 2660 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2661 NTB_REG_16 | XEON_USMEMMISS_OFFSET, 2662 sysctl_handle_register, "SU", "Upstream Memory Miss"); 2663 2664 tmptree = SYSCTL_ADD_NODE(ctx, regpar, OID_AUTO, "xeon_hw_err", 2665 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Xeon HW errors"); 2666 errpar = SYSCTL_CHILDREN(tmptree); 2667 2668 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "ppd", 2669 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2670 NTB_REG_8 | NTB_PCI_REG | NTB_PPD_OFFSET, 2671 sysctl_handle_register, "CU", "PPD"); 2672 2673 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar23_sz", 2674 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2675 NTB_REG_8 | NTB_PCI_REG | XEON_PBAR23SZ_OFFSET, 2676 sysctl_handle_register, "CU", "PBAR23 SZ (log2)"); 2677 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar4_sz", 2678 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2679 NTB_REG_8 | NTB_PCI_REG | XEON_PBAR4SZ_OFFSET, 2680 sysctl_handle_register, "CU", "PBAR4 SZ (log2)"); 2681 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar5_sz", 2682 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2683 NTB_REG_8 | NTB_PCI_REG | XEON_PBAR5SZ_OFFSET, 2684 sysctl_handle_register, "CU", "PBAR5 SZ (log2)"); 2685 2686 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar23_sz", 2687 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2688 NTB_REG_8 | NTB_PCI_REG | XEON_SBAR23SZ_OFFSET, 2689 sysctl_handle_register, "CU", "SBAR23 SZ (log2)"); 2690 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar4_sz", 2691 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2692 NTB_REG_8 | NTB_PCI_REG | XEON_SBAR4SZ_OFFSET, 2693 sysctl_handle_register, "CU", "SBAR4 SZ (log2)"); 2694 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar5_sz", 2695 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2696 NTB_REG_8 | NTB_PCI_REG | XEON_SBAR5SZ_OFFSET, 2697 sysctl_handle_register, "CU", "SBAR5 SZ (log2)"); 2698 2699 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "devsts", 2700 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2701 NTB_REG_16 | NTB_PCI_REG | XEON_DEVSTS_OFFSET, 2702 sysctl_handle_register, "SU", "DEVSTS"); 2703 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnksts", 2704 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2705 NTB_REG_16 | NTB_PCI_REG | XEON_LINK_STATUS_OFFSET, 2706 sysctl_handle_register, "SU", "LNKSTS"); 2707 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "slnksts", 2708 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2709 NTB_REG_16 | NTB_PCI_REG | XEON_SLINK_STATUS_OFFSET, 2710 sysctl_handle_register, "SU", "SLNKSTS"); 2711 2712 SYSCTL_ADD_PROC(ctx, errpar, OID_AUTO, "uncerrsts", 2713 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2714 NTB_REG_32 | NTB_PCI_REG | XEON_UNCERRSTS_OFFSET, 2715 sysctl_handle_register, "IU", "UNCERRSTS"); 2716 SYSCTL_ADD_PROC(ctx, errpar, OID_AUTO, "corerrsts", 2717 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2718 NTB_REG_32 | NTB_PCI_REG | XEON_CORERRSTS_OFFSET, 2719 sysctl_handle_register, "IU", "CORERRSTS"); 2720 2721 if (ntb->conn_type != NTB_CONN_B2B) 2722 return; 2723 2724 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat01l", 2725 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2726 NTB_REG_32 | XEON_B2B_XLAT_OFFSETL, 2727 sysctl_handle_register, "IU", "Outgoing XLAT0L register"); 2728 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat01u", 2729 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2730 NTB_REG_32 | XEON_B2B_XLAT_OFFSETU, 2731 sysctl_handle_register, "IU", "Outgoing XLAT0U register"); 2732 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat23", 2733 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2734 NTB_REG_64 | ntb->bar_info[NTB_B2B_BAR_1].pbarxlat_off, 2735 sysctl_handle_register, "QU", "Outgoing XLAT23 register"); 2736 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 2737 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat4", 2738 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2739 NTB_REG_32 | ntb->bar_info[NTB_B2B_BAR_2].pbarxlat_off, 2740 sysctl_handle_register, "IU", "Outgoing XLAT4 register"); 2741 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat5", 2742 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2743 NTB_REG_32 | ntb->bar_info[NTB_B2B_BAR_3].pbarxlat_off, 2744 sysctl_handle_register, "IU", "Outgoing XLAT5 register"); 2745 } else { 2746 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat45", 2747 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2748 NTB_REG_64 | ntb->bar_info[NTB_B2B_BAR_2].pbarxlat_off, 2749 sysctl_handle_register, "QU", "Outgoing XLAT45 register"); 2750 } 2751 2752 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt23", 2753 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2754 NTB_REG_64 | XEON_PBAR2LMT_OFFSET, 2755 sysctl_handle_register, "QU", "Outgoing LMT23 register"); 2756 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 2757 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt4", 2758 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2759 NTB_REG_32 | XEON_PBAR4LMT_OFFSET, 2760 sysctl_handle_register, "IU", "Outgoing LMT4 register"); 2761 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt5", 2762 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2763 NTB_REG_32 | XEON_PBAR5LMT_OFFSET, 2764 sysctl_handle_register, "IU", "Outgoing LMT5 register"); 2765 } else { 2766 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt45", 2767 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2768 NTB_REG_64 | XEON_PBAR4LMT_OFFSET, 2769 sysctl_handle_register, "QU", "Outgoing LMT45 register"); 2770 } 2771 2772 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar01_base", 2773 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2774 NTB_REG_64 | ntb->xlat_reg->bar0_base, 2775 sysctl_handle_register, "QU", "Secondary BAR01 base register"); 2776 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar23_base", 2777 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2778 NTB_REG_64 | ntb->xlat_reg->bar2_base, 2779 sysctl_handle_register, "QU", "Secondary BAR23 base register"); 2780 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) { 2781 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar4_base", 2782 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2783 NTB_REG_32 | ntb->xlat_reg->bar4_base, 2784 sysctl_handle_register, "IU", 2785 "Secondary BAR4 base register"); 2786 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar5_base", 2787 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2788 NTB_REG_32 | ntb->xlat_reg->bar5_base, 2789 sysctl_handle_register, "IU", 2790 "Secondary BAR5 base register"); 2791 } else { 2792 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar45_base", 2793 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb, 2794 NTB_REG_64 | ntb->xlat_reg->bar4_base, 2795 sysctl_handle_register, "QU", 2796 "Secondary BAR45 base register"); 2797 } 2798 } 2799 2800 static int 2801 sysctl_handle_features(SYSCTL_HANDLER_ARGS) 2802 { 2803 struct ntb_softc *ntb = arg1; 2804 struct sbuf sb; 2805 int error; 2806 2807 sbuf_new_for_sysctl(&sb, NULL, 256, req); 2808 2809 sbuf_printf(&sb, "%b", ntb->features, NTB_FEATURES_STR); 2810 error = sbuf_finish(&sb); 2811 sbuf_delete(&sb); 2812 2813 if (error || !req->newptr) 2814 return (error); 2815 return (EINVAL); 2816 } 2817 2818 static int 2819 sysctl_handle_link_admin(SYSCTL_HANDLER_ARGS) 2820 { 2821 struct ntb_softc *ntb = arg1; 2822 unsigned old, new; 2823 int error; 2824 2825 old = intel_ntb_link_enabled(ntb->device); 2826 2827 error = SYSCTL_OUT(req, &old, sizeof(old)); 2828 if (error != 0 || req->newptr == NULL) 2829 return (error); 2830 2831 error = SYSCTL_IN(req, &new, sizeof(new)); 2832 if (error != 0) 2833 return (error); 2834 2835 intel_ntb_printf(0, "Admin set interface state to '%sabled'\n", 2836 (new != 0)? "en" : "dis"); 2837 2838 if (new != 0) 2839 error = intel_ntb_link_enable(ntb->device, NTB_SPEED_AUTO, NTB_WIDTH_AUTO); 2840 else 2841 error = intel_ntb_link_disable(ntb->device); 2842 return (error); 2843 } 2844 2845 static int 2846 sysctl_handle_link_status_human(SYSCTL_HANDLER_ARGS) 2847 { 2848 struct ntb_softc *ntb = arg1; 2849 struct sbuf sb; 2850 enum ntb_speed speed; 2851 enum ntb_width width; 2852 int error; 2853 2854 sbuf_new_for_sysctl(&sb, NULL, 32, req); 2855 2856 if (intel_ntb_link_is_up(ntb->device, &speed, &width)) 2857 sbuf_printf(&sb, "up / PCIe Gen %u / Width x%u", 2858 (unsigned)speed, (unsigned)width); 2859 else 2860 sbuf_printf(&sb, "down"); 2861 2862 error = sbuf_finish(&sb); 2863 sbuf_delete(&sb); 2864 2865 if (error || !req->newptr) 2866 return (error); 2867 return (EINVAL); 2868 } 2869 2870 static int 2871 sysctl_handle_link_status(SYSCTL_HANDLER_ARGS) 2872 { 2873 struct ntb_softc *ntb = arg1; 2874 unsigned res; 2875 int error; 2876 2877 res = intel_ntb_link_is_up(ntb->device, NULL, NULL); 2878 2879 error = SYSCTL_OUT(req, &res, sizeof(res)); 2880 if (error || !req->newptr) 2881 return (error); 2882 return (EINVAL); 2883 } 2884 2885 static int 2886 sysctl_handle_register(SYSCTL_HANDLER_ARGS) 2887 { 2888 struct ntb_softc *ntb; 2889 const void *outp; 2890 uintptr_t sz; 2891 uint64_t umv; 2892 char be[sizeof(umv)]; 2893 size_t outsz; 2894 uint32_t reg; 2895 bool db, pci; 2896 int error; 2897 2898 ntb = arg1; 2899 reg = arg2 & ~NTB_REGFLAGS_MASK; 2900 sz = arg2 & NTB_REGSZ_MASK; 2901 db = (arg2 & NTB_DB_READ) != 0; 2902 pci = (arg2 & NTB_PCI_REG) != 0; 2903 2904 KASSERT(!(db && pci), ("bogus")); 2905 2906 if (db) { 2907 KASSERT(sz == NTB_REG_64, ("bogus")); 2908 umv = db_ioread(ntb, reg); 2909 outsz = sizeof(uint64_t); 2910 } else { 2911 switch (sz) { 2912 case NTB_REG_64: 2913 if (pci) 2914 umv = pci_read_config(ntb->device, reg, 8); 2915 else 2916 umv = intel_ntb_reg_read(8, reg); 2917 outsz = sizeof(uint64_t); 2918 break; 2919 case NTB_REG_32: 2920 if (pci) 2921 umv = pci_read_config(ntb->device, reg, 4); 2922 else 2923 umv = intel_ntb_reg_read(4, reg); 2924 outsz = sizeof(uint32_t); 2925 break; 2926 case NTB_REG_16: 2927 if (pci) 2928 umv = pci_read_config(ntb->device, reg, 2); 2929 else 2930 umv = intel_ntb_reg_read(2, reg); 2931 outsz = sizeof(uint16_t); 2932 break; 2933 case NTB_REG_8: 2934 if (pci) 2935 umv = pci_read_config(ntb->device, reg, 1); 2936 else 2937 umv = intel_ntb_reg_read(1, reg); 2938 outsz = sizeof(uint8_t); 2939 break; 2940 default: 2941 panic("bogus"); 2942 break; 2943 } 2944 } 2945 2946 /* Encode bigendian so that sysctl -x is legible. */ 2947 be64enc(be, umv); 2948 outp = ((char *)be) + sizeof(umv) - outsz; 2949 2950 error = SYSCTL_OUT(req, outp, outsz); 2951 if (error || !req->newptr) 2952 return (error); 2953 return (EINVAL); 2954 } 2955 2956 static unsigned 2957 intel_ntb_user_mw_to_idx(struct ntb_softc *ntb, unsigned uidx) 2958 { 2959 2960 if ((ntb->b2b_mw_idx != B2B_MW_DISABLED && ntb->b2b_off == 0 && 2961 uidx >= ntb->b2b_mw_idx) || 2962 (ntb->msix_mw_idx != B2B_MW_DISABLED && uidx >= ntb->msix_mw_idx)) 2963 uidx++; 2964 if ((ntb->b2b_mw_idx != B2B_MW_DISABLED && ntb->b2b_off == 0 && 2965 uidx >= ntb->b2b_mw_idx) && 2966 (ntb->msix_mw_idx != B2B_MW_DISABLED && uidx >= ntb->msix_mw_idx)) 2967 uidx++; 2968 return (uidx); 2969 } 2970 2971 #ifndef EARLY_AP_STARTUP 2972 static int msix_ready; 2973 2974 static void 2975 intel_ntb_msix_ready(void *arg __unused) 2976 { 2977 2978 msix_ready = 1; 2979 } 2980 SYSINIT(intel_ntb_msix_ready, SI_SUB_SMP, SI_ORDER_ANY, 2981 intel_ntb_msix_ready, NULL); 2982 #endif 2983 2984 static void 2985 intel_ntb_exchange_msix(void *ctx) 2986 { 2987 struct ntb_softc *ntb; 2988 uint32_t val; 2989 unsigned i; 2990 2991 ntb = ctx; 2992 2993 if (ntb->peer_msix_good) 2994 goto msix_good; 2995 if (ntb->peer_msix_done) 2996 goto msix_done; 2997 2998 #ifndef EARLY_AP_STARTUP 2999 /* Block MSIX negotiation until SMP started and IRQ reshuffled. */ 3000 if (!msix_ready) 3001 goto reschedule; 3002 #endif 3003 3004 intel_ntb_get_msix_info(ntb); 3005 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) { 3006 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_DATA0 + i, 3007 ntb->msix_data[i].nmd_data); 3008 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_OFS0 + i, 3009 ntb->msix_data[i].nmd_ofs - ntb->msix_xlat); 3010 } 3011 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_GUARD, NTB_MSIX_VER_GUARD); 3012 3013 intel_ntb_spad_read(ntb->device, NTB_MSIX_GUARD, &val); 3014 if (val != NTB_MSIX_VER_GUARD) 3015 goto reschedule; 3016 3017 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) { 3018 intel_ntb_spad_read(ntb->device, NTB_MSIX_DATA0 + i, &val); 3019 intel_ntb_printf(2, "remote MSIX data(%u): 0x%x\n", i, val); 3020 ntb->peer_msix_data[i].nmd_data = val; 3021 intel_ntb_spad_read(ntb->device, NTB_MSIX_OFS0 + i, &val); 3022 intel_ntb_printf(2, "remote MSIX addr(%u): 0x%x\n", i, val); 3023 ntb->peer_msix_data[i].nmd_ofs = val; 3024 } 3025 3026 ntb->peer_msix_done = true; 3027 3028 msix_done: 3029 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_DONE, NTB_MSIX_RECEIVED); 3030 intel_ntb_spad_read(ntb->device, NTB_MSIX_DONE, &val); 3031 if (val != NTB_MSIX_RECEIVED) 3032 goto reschedule; 3033 3034 intel_ntb_spad_clear(ntb->device); 3035 ntb->peer_msix_good = true; 3036 /* Give peer time to see our NTB_MSIX_RECEIVED. */ 3037 goto reschedule; 3038 3039 msix_good: 3040 intel_ntb_poll_link(ntb); 3041 ntb_link_event(ntb->device); 3042 return; 3043 3044 reschedule: 3045 ntb->lnk_sta = pci_read_config(ntb->device, ntb->reg->lnk_sta, 2); 3046 if (_xeon_link_is_up(ntb)) { 3047 callout_reset(&ntb->peer_msix_work, 3048 hz * (ntb->peer_msix_good ? 2 : 1) / 10, 3049 intel_ntb_exchange_msix, ntb); 3050 } else 3051 intel_ntb_spad_clear(ntb->device); 3052 } 3053 3054 /* 3055 * Public API to the rest of the OS 3056 */ 3057 3058 static uint8_t 3059 intel_ntb_spad_count(device_t dev) 3060 { 3061 struct ntb_softc *ntb = device_get_softc(dev); 3062 3063 return (ntb->spad_count); 3064 } 3065 3066 static uint8_t 3067 intel_ntb_mw_count(device_t dev) 3068 { 3069 struct ntb_softc *ntb = device_get_softc(dev); 3070 uint8_t res; 3071 3072 res = ntb->mw_count; 3073 if (ntb->b2b_mw_idx != B2B_MW_DISABLED && ntb->b2b_off == 0) 3074 res--; 3075 if (ntb->msix_mw_idx != B2B_MW_DISABLED) 3076 res--; 3077 return (res); 3078 } 3079 3080 static int 3081 intel_ntb_spad_write(device_t dev, unsigned int idx, uint32_t val) 3082 { 3083 struct ntb_softc *ntb = device_get_softc(dev); 3084 3085 if (idx >= ntb->spad_count) 3086 return (EINVAL); 3087 3088 intel_ntb_reg_write(4, ntb->self_reg->spad + idx * 4, val); 3089 3090 return (0); 3091 } 3092 3093 /* 3094 * Zeros the local scratchpad. 3095 */ 3096 static void 3097 intel_ntb_spad_clear(device_t dev) 3098 { 3099 struct ntb_softc *ntb = device_get_softc(dev); 3100 unsigned i; 3101 3102 for (i = 0; i < ntb->spad_count; i++) 3103 intel_ntb_spad_write(dev, i, 0); 3104 } 3105 3106 static int 3107 intel_ntb_spad_read(device_t dev, unsigned int idx, uint32_t *val) 3108 { 3109 struct ntb_softc *ntb = device_get_softc(dev); 3110 3111 if (idx >= ntb->spad_count) 3112 return (EINVAL); 3113 3114 *val = intel_ntb_reg_read(4, ntb->self_reg->spad + idx * 4); 3115 3116 return (0); 3117 } 3118 3119 static int 3120 intel_ntb_peer_spad_write(device_t dev, unsigned int idx, uint32_t val) 3121 { 3122 struct ntb_softc *ntb = device_get_softc(dev); 3123 3124 if (idx >= ntb->spad_count) 3125 return (EINVAL); 3126 3127 if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) 3128 intel_ntb_mw_write(4, XEON_SPAD_OFFSET + idx * 4, val); 3129 else 3130 intel_ntb_reg_write(4, ntb->peer_reg->spad + idx * 4, val); 3131 3132 return (0); 3133 } 3134 3135 static int 3136 intel_ntb_peer_spad_read(device_t dev, unsigned int idx, uint32_t *val) 3137 { 3138 struct ntb_softc *ntb = device_get_softc(dev); 3139 3140 if (idx >= ntb->spad_count) 3141 return (EINVAL); 3142 3143 if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) 3144 *val = intel_ntb_mw_read(4, XEON_SPAD_OFFSET + idx * 4); 3145 else 3146 *val = intel_ntb_reg_read(4, ntb->peer_reg->spad + idx * 4); 3147 3148 return (0); 3149 } 3150 3151 static int 3152 intel_ntb_mw_get_range(device_t dev, unsigned mw_idx, vm_paddr_t *base, 3153 caddr_t *vbase, size_t *size, size_t *align, size_t *align_size, 3154 bus_addr_t *plimit) 3155 { 3156 struct ntb_softc *ntb = device_get_softc(dev); 3157 struct ntb_pci_bar_info *bar; 3158 bus_addr_t limit; 3159 size_t bar_b2b_off; 3160 enum ntb_bar bar_num; 3161 3162 if (mw_idx >= intel_ntb_mw_count(dev)) 3163 return (EINVAL); 3164 mw_idx = intel_ntb_user_mw_to_idx(ntb, mw_idx); 3165 3166 bar_num = intel_ntb_mw_to_bar(ntb, mw_idx); 3167 bar = &ntb->bar_info[bar_num]; 3168 bar_b2b_off = 0; 3169 if (mw_idx == ntb->b2b_mw_idx) { 3170 KASSERT(ntb->b2b_off != 0, 3171 ("user shouldn't get non-shared b2b mw")); 3172 bar_b2b_off = ntb->b2b_off; 3173 } 3174 3175 if (bar_is_64bit(ntb, bar_num)) 3176 limit = BUS_SPACE_MAXADDR; 3177 else 3178 limit = BUS_SPACE_MAXADDR_32BIT; 3179 3180 if (base != NULL) 3181 *base = bar->pbase + bar_b2b_off; 3182 if (vbase != NULL) 3183 *vbase = bar->vbase + bar_b2b_off; 3184 if (size != NULL) 3185 *size = bar->size - bar_b2b_off; 3186 if (align != NULL) 3187 *align = bar->size; 3188 if (align_size != NULL) 3189 *align_size = 1; 3190 if (plimit != NULL) 3191 *plimit = limit; 3192 return (0); 3193 } 3194 3195 static int 3196 intel_ntb_mw_set_trans(device_t dev, unsigned idx, bus_addr_t addr, size_t size) 3197 { 3198 struct ntb_softc *ntb = device_get_softc(dev); 3199 struct ntb_pci_bar_info *bar; 3200 uint64_t base, limit, reg_val; 3201 size_t bar_size, mw_size; 3202 uint32_t base_reg, xlat_reg, limit_reg; 3203 enum ntb_bar bar_num; 3204 3205 if (idx >= intel_ntb_mw_count(dev)) 3206 return (EINVAL); 3207 idx = intel_ntb_user_mw_to_idx(ntb, idx); 3208 3209 bar_num = intel_ntb_mw_to_bar(ntb, idx); 3210 bar = &ntb->bar_info[bar_num]; 3211 3212 bar_size = bar->size; 3213 if (idx == ntb->b2b_mw_idx) 3214 mw_size = bar_size - ntb->b2b_off; 3215 else 3216 mw_size = bar_size; 3217 3218 /* Hardware requires that addr is aligned to bar size */ 3219 if ((addr & (bar_size - 1)) != 0) 3220 return (EINVAL); 3221 3222 if (size > mw_size) 3223 return (EINVAL); 3224 3225 bar_get_xlat_params(ntb, bar_num, &base_reg, &xlat_reg, &limit_reg); 3226 3227 limit = 0; 3228 if (bar_is_64bit(ntb, bar_num)) { 3229 base = intel_ntb_reg_read(8, base_reg) & BAR_HIGH_MASK; 3230 3231 if (limit_reg != 0 && size != mw_size) 3232 limit = base + size; 3233 else 3234 limit = base + mw_size; 3235 3236 /* Set and verify translation address */ 3237 intel_ntb_reg_write(8, xlat_reg, addr); 3238 reg_val = intel_ntb_reg_read(8, xlat_reg) & BAR_HIGH_MASK; 3239 if (reg_val != addr) { 3240 intel_ntb_reg_write(8, xlat_reg, 0); 3241 return (EIO); 3242 } 3243 3244 /* Set and verify the limit */ 3245 intel_ntb_reg_write(8, limit_reg, limit); 3246 reg_val = intel_ntb_reg_read(8, limit_reg) & BAR_HIGH_MASK; 3247 if (reg_val != limit) { 3248 intel_ntb_reg_write(8, limit_reg, base); 3249 intel_ntb_reg_write(8, xlat_reg, 0); 3250 return (EIO); 3251 } 3252 3253 if (ntb->type == NTB_XEON_GEN3) { 3254 limit = base + size; 3255 3256 /* set EMBAR1/2XLIMIT */ 3257 if (!idx) 3258 intel_ntb_reg_write(8, 3259 XEON_GEN3_REG_EMBAR1XLIMIT, limit); 3260 else 3261 intel_ntb_reg_write(8, 3262 XEON_GEN3_REG_EMBAR2XLIMIT, limit); 3263 } 3264 } else { 3265 /* Configure 32-bit (split) BAR MW */ 3266 if (ntb->type == NTB_XEON_GEN3) 3267 return (EIO); 3268 3269 if ((addr & UINT32_MAX) != addr) 3270 return (ERANGE); 3271 if (((addr + size) & UINT32_MAX) != (addr + size)) 3272 return (ERANGE); 3273 3274 base = intel_ntb_reg_read(4, base_reg) & BAR_HIGH_MASK; 3275 3276 if (limit_reg != 0 && size != mw_size) 3277 limit = base + size; 3278 3279 /* Set and verify translation address */ 3280 intel_ntb_reg_write(4, xlat_reg, addr); 3281 reg_val = intel_ntb_reg_read(4, xlat_reg) & BAR_HIGH_MASK; 3282 if (reg_val != addr) { 3283 intel_ntb_reg_write(4, xlat_reg, 0); 3284 return (EIO); 3285 } 3286 3287 /* Set and verify the limit */ 3288 intel_ntb_reg_write(4, limit_reg, limit); 3289 reg_val = intel_ntb_reg_read(4, limit_reg) & BAR_HIGH_MASK; 3290 if (reg_val != limit) { 3291 intel_ntb_reg_write(4, limit_reg, base); 3292 intel_ntb_reg_write(4, xlat_reg, 0); 3293 return (EIO); 3294 } 3295 } 3296 return (0); 3297 } 3298 3299 static int 3300 intel_ntb_mw_clear_trans(device_t dev, unsigned mw_idx) 3301 { 3302 3303 return (intel_ntb_mw_set_trans(dev, mw_idx, 0, 0)); 3304 } 3305 3306 static int 3307 intel_ntb_mw_get_wc(device_t dev, unsigned idx, vm_memattr_t *mode) 3308 { 3309 struct ntb_softc *ntb = device_get_softc(dev); 3310 struct ntb_pci_bar_info *bar; 3311 3312 if (idx >= intel_ntb_mw_count(dev)) 3313 return (EINVAL); 3314 idx = intel_ntb_user_mw_to_idx(ntb, idx); 3315 3316 bar = &ntb->bar_info[intel_ntb_mw_to_bar(ntb, idx)]; 3317 *mode = bar->map_mode; 3318 return (0); 3319 } 3320 3321 static int 3322 intel_ntb_mw_set_wc(device_t dev, unsigned idx, vm_memattr_t mode) 3323 { 3324 struct ntb_softc *ntb = device_get_softc(dev); 3325 3326 if (idx >= intel_ntb_mw_count(dev)) 3327 return (EINVAL); 3328 3329 idx = intel_ntb_user_mw_to_idx(ntb, idx); 3330 return (intel_ntb_mw_set_wc_internal(ntb, idx, mode)); 3331 } 3332 3333 static int 3334 intel_ntb_mw_set_wc_internal(struct ntb_softc *ntb, unsigned idx, vm_memattr_t mode) 3335 { 3336 struct ntb_pci_bar_info *bar; 3337 int rc; 3338 3339 bar = &ntb->bar_info[intel_ntb_mw_to_bar(ntb, idx)]; 3340 if (bar->map_mode == mode) 3341 return (0); 3342 3343 rc = pmap_change_attr((vm_offset_t)bar->vbase, bar->size, mode); 3344 if (rc == 0) 3345 bar->map_mode = mode; 3346 3347 return (rc); 3348 } 3349 3350 static void 3351 intel_ntb_peer_db_set(device_t dev, uint64_t bits) 3352 { 3353 struct ntb_softc *ntb = device_get_softc(dev); 3354 uint64_t db; 3355 3356 if ((bits & ~ntb->db_valid_mask) != 0) { 3357 device_printf(ntb->device, "Invalid doorbell bits %#jx\n", 3358 (uintmax_t)bits); 3359 return; 3360 } 3361 3362 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) { 3363 struct ntb_pci_bar_info *lapic; 3364 unsigned i; 3365 3366 lapic = ntb->peer_lapic_bar; 3367 3368 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) { 3369 if ((bits & intel_ntb_db_vector_mask(dev, i)) != 0) 3370 bus_space_write_4(lapic->pci_bus_tag, 3371 lapic->pci_bus_handle, 3372 ntb->peer_msix_data[i].nmd_ofs, 3373 ntb->peer_msix_data[i].nmd_data); 3374 } 3375 return; 3376 } 3377 3378 if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) { 3379 intel_ntb_mw_write(2, XEON_PDOORBELL_OFFSET, bits); 3380 return; 3381 } 3382 3383 if (ntb->type == NTB_XEON_GEN3) { 3384 while (bits != 0) { 3385 db = ffsll(bits); 3386 3387 intel_ntb_reg_write(1, 3388 ntb->peer_reg->db_bell + (db - 1) * 4, 0x1); 3389 3390 bits = bits & (bits - 1); 3391 } 3392 } else { 3393 db_iowrite(ntb, ntb->peer_reg->db_bell, bits); 3394 } 3395 } 3396 3397 static int 3398 intel_ntb_peer_db_addr(device_t dev, bus_addr_t *db_addr, vm_size_t *db_size) 3399 { 3400 struct ntb_softc *ntb = device_get_softc(dev); 3401 struct ntb_pci_bar_info *bar; 3402 uint64_t regoff; 3403 3404 KASSERT((db_addr != NULL && db_size != NULL), ("must be non-NULL")); 3405 3406 if (!HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) { 3407 bar = &ntb->bar_info[NTB_CONFIG_BAR]; 3408 regoff = ntb->peer_reg->db_bell; 3409 } else { 3410 KASSERT(ntb->b2b_mw_idx != B2B_MW_DISABLED, 3411 ("invalid b2b idx")); 3412 3413 bar = &ntb->bar_info[intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx)]; 3414 regoff = XEON_PDOORBELL_OFFSET; 3415 } 3416 KASSERT(bar->pci_bus_tag != X86_BUS_SPACE_IO, ("uh oh")); 3417 3418 /* HACK: Specific to current x86 bus implementation. */ 3419 *db_addr = ((uint64_t)bar->pci_bus_handle + regoff); 3420 *db_size = ntb->reg->db_size; 3421 return (0); 3422 } 3423 3424 static uint64_t 3425 intel_ntb_db_valid_mask(device_t dev) 3426 { 3427 struct ntb_softc *ntb = device_get_softc(dev); 3428 3429 return (ntb->db_valid_mask); 3430 } 3431 3432 static int 3433 intel_ntb_db_vector_count(device_t dev) 3434 { 3435 struct ntb_softc *ntb = device_get_softc(dev); 3436 3437 return (ntb->db_vec_count); 3438 } 3439 3440 static uint64_t 3441 intel_ntb_db_vector_mask(device_t dev, uint32_t vector) 3442 { 3443 struct ntb_softc *ntb = device_get_softc(dev); 3444 3445 if (vector > ntb->db_vec_count) 3446 return (0); 3447 return (ntb->db_valid_mask & intel_ntb_vec_mask(ntb, vector)); 3448 } 3449 3450 static bool 3451 intel_ntb_link_is_up(device_t dev, enum ntb_speed *speed, enum ntb_width *width) 3452 { 3453 struct ntb_softc *ntb = device_get_softc(dev); 3454 3455 if (speed != NULL) 3456 *speed = intel_ntb_link_sta_speed(ntb); 3457 if (width != NULL) 3458 *width = intel_ntb_link_sta_width(ntb); 3459 return (link_is_up(ntb)); 3460 } 3461 3462 static void 3463 save_bar_parameters(struct ntb_pci_bar_info *bar) 3464 { 3465 3466 bar->pci_bus_tag = rman_get_bustag(bar->pci_resource); 3467 bar->pci_bus_handle = rman_get_bushandle(bar->pci_resource); 3468 bar->pbase = rman_get_start(bar->pci_resource); 3469 bar->size = rman_get_size(bar->pci_resource); 3470 bar->vbase = rman_get_virtual(bar->pci_resource); 3471 } 3472 3473 static device_method_t ntb_intel_methods[] = { 3474 /* Device interface */ 3475 DEVMETHOD(device_probe, intel_ntb_probe), 3476 DEVMETHOD(device_attach, intel_ntb_attach), 3477 DEVMETHOD(device_detach, intel_ntb_detach), 3478 /* Bus interface */ 3479 DEVMETHOD(bus_child_location, ntb_child_location), 3480 DEVMETHOD(bus_print_child, ntb_print_child), 3481 DEVMETHOD(bus_get_dma_tag, ntb_get_dma_tag), 3482 /* NTB interface */ 3483 DEVMETHOD(ntb_port_number, intel_ntb_port_number), 3484 DEVMETHOD(ntb_peer_port_count, intel_ntb_peer_port_count), 3485 DEVMETHOD(ntb_peer_port_number, intel_ntb_peer_port_number), 3486 DEVMETHOD(ntb_peer_port_idx, intel_ntb_peer_port_idx), 3487 DEVMETHOD(ntb_link_is_up, intel_ntb_link_is_up), 3488 DEVMETHOD(ntb_link_enable, intel_ntb_link_enable), 3489 DEVMETHOD(ntb_link_disable, intel_ntb_link_disable), 3490 DEVMETHOD(ntb_link_enabled, intel_ntb_link_enabled), 3491 DEVMETHOD(ntb_mw_count, intel_ntb_mw_count), 3492 DEVMETHOD(ntb_mw_get_range, intel_ntb_mw_get_range), 3493 DEVMETHOD(ntb_mw_set_trans, intel_ntb_mw_set_trans), 3494 DEVMETHOD(ntb_mw_clear_trans, intel_ntb_mw_clear_trans), 3495 DEVMETHOD(ntb_mw_get_wc, intel_ntb_mw_get_wc), 3496 DEVMETHOD(ntb_mw_set_wc, intel_ntb_mw_set_wc), 3497 DEVMETHOD(ntb_spad_count, intel_ntb_spad_count), 3498 DEVMETHOD(ntb_spad_clear, intel_ntb_spad_clear), 3499 DEVMETHOD(ntb_spad_write, intel_ntb_spad_write), 3500 DEVMETHOD(ntb_spad_read, intel_ntb_spad_read), 3501 DEVMETHOD(ntb_peer_spad_write, intel_ntb_peer_spad_write), 3502 DEVMETHOD(ntb_peer_spad_read, intel_ntb_peer_spad_read), 3503 DEVMETHOD(ntb_db_valid_mask, intel_ntb_db_valid_mask), 3504 DEVMETHOD(ntb_db_vector_count, intel_ntb_db_vector_count), 3505 DEVMETHOD(ntb_db_vector_mask, intel_ntb_db_vector_mask), 3506 DEVMETHOD(ntb_db_clear, intel_ntb_db_clear), 3507 DEVMETHOD(ntb_db_clear_mask, intel_ntb_db_clear_mask), 3508 DEVMETHOD(ntb_db_read, intel_ntb_db_read), 3509 DEVMETHOD(ntb_db_set_mask, intel_ntb_db_set_mask), 3510 DEVMETHOD(ntb_peer_db_addr, intel_ntb_peer_db_addr), 3511 DEVMETHOD(ntb_peer_db_set, intel_ntb_peer_db_set), 3512 DEVMETHOD_END 3513 }; 3514 3515 static DEFINE_CLASS_0(ntb_hw, ntb_intel_driver, ntb_intel_methods, 3516 sizeof(struct ntb_softc)); 3517 DRIVER_MODULE(ntb_hw_intel, pci, ntb_intel_driver, ntb_hw_devclass, NULL, NULL); 3518 MODULE_DEPEND(ntb_hw_intel, ntb, 1, 1, 1); 3519 MODULE_VERSION(ntb_hw_intel, 1); 3520 MODULE_PNP_INFO("W32:vendor/device;D:#", pci, ntb_hw_intel, pci_ids, 3521 nitems(pci_ids)); 3522