1 /* 2 * Copyright (c) 2007, Intel Corporation. 3 * 4 * This work is licensed under the terms of the GNU GPL, version 2. See 5 * the COPYING file in the top-level directory. 6 * 7 * Jiang Yunhong <yunhong.jiang@intel.com> 8 * 9 * This file implements direct PCI assignment to a HVM guest 10 */ 11 12 #include <sys/mman.h> 13 14 #include "hw/xen/xen_backend.h" 15 #include "xen_pt.h" 16 #include "hw/i386/apic-msidef.h" 17 18 19 #define XEN_PT_AUTO_ASSIGN -1 20 21 /* shift count for gflags */ 22 #define XEN_PT_GFLAGS_SHIFT_DEST_ID 0 23 #define XEN_PT_GFLAGS_SHIFT_RH 8 24 #define XEN_PT_GFLAGS_SHIFT_DM 9 25 #define XEN_PT_GFLAGSSHIFT_DELIV_MODE 12 26 #define XEN_PT_GFLAGSSHIFT_TRG_MODE 15 27 28 29 /* 30 * Helpers 31 */ 32 33 static inline uint8_t msi_vector(uint32_t data) 34 { 35 return (data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT; 36 } 37 38 static inline uint8_t msi_dest_id(uint32_t addr) 39 { 40 return (addr & MSI_ADDR_DEST_ID_MASK) >> MSI_ADDR_DEST_ID_SHIFT; 41 } 42 43 static inline uint32_t msi_ext_dest_id(uint32_t addr_hi) 44 { 45 return addr_hi & 0xffffff00; 46 } 47 48 static uint32_t msi_gflags(uint32_t data, uint64_t addr) 49 { 50 uint32_t result = 0; 51 int rh, dm, dest_id, deliv_mode, trig_mode; 52 53 rh = (addr >> MSI_ADDR_REDIRECTION_SHIFT) & 0x1; 54 dm = (addr >> MSI_ADDR_DEST_MODE_SHIFT) & 0x1; 55 dest_id = msi_dest_id(addr); 56 deliv_mode = (data >> MSI_DATA_DELIVERY_MODE_SHIFT) & 0x7; 57 trig_mode = (data >> MSI_DATA_TRIGGER_SHIFT) & 0x1; 58 59 result = dest_id | (rh << XEN_PT_GFLAGS_SHIFT_RH) 60 | (dm << XEN_PT_GFLAGS_SHIFT_DM) 61 | (deliv_mode << XEN_PT_GFLAGSSHIFT_DELIV_MODE) 62 | (trig_mode << XEN_PT_GFLAGSSHIFT_TRG_MODE); 63 64 return result; 65 } 66 67 static inline uint64_t msi_addr64(XenPTMSI *msi) 68 { 69 return (uint64_t)msi->addr_hi << 32 | msi->addr_lo; 70 } 71 72 static int msi_msix_enable(XenPCIPassthroughState *s, 73 uint32_t address, 74 uint16_t flag, 75 bool enable) 76 { 77 uint16_t val = 0; 78 79 if (!address) { 80 return -1; 81 } 82 83 xen_host_pci_get_word(&s->real_device, address, &val); 84 if (enable) { 85 val |= flag; 86 } else { 87 val &= ~flag; 88 } 89 xen_host_pci_set_word(&s->real_device, address, val); 90 return 0; 91 } 92 93 static int msi_msix_setup(XenPCIPassthroughState *s, 94 uint64_t addr, 95 uint32_t data, 96 int *ppirq, 97 bool is_msix, 98 int msix_entry, 99 bool is_not_mapped) 100 { 101 uint8_t gvec = msi_vector(data); 102 int rc = 0; 103 104 assert((!is_msix && msix_entry == 0) || is_msix); 105 106 if (gvec == 0) { 107 /* if gvec is 0, the guest is asking for a particular pirq that 108 * is passed as dest_id */ 109 *ppirq = msi_ext_dest_id(addr >> 32) | msi_dest_id(addr); 110 if (!*ppirq) { 111 /* this probably identifies an misconfiguration of the guest, 112 * try the emulated path */ 113 *ppirq = XEN_PT_UNASSIGNED_PIRQ; 114 } else { 115 XEN_PT_LOG(&s->dev, "requested pirq %d for MSI%s" 116 " (vec: %#x, entry: %#x)\n", 117 *ppirq, is_msix ? "-X" : "", gvec, msix_entry); 118 } 119 } 120 121 if (is_not_mapped) { 122 uint64_t table_base = 0; 123 124 if (is_msix) { 125 table_base = s->msix->table_base; 126 } 127 128 rc = xc_physdev_map_pirq_msi(xen_xc, xen_domid, XEN_PT_AUTO_ASSIGN, 129 ppirq, PCI_DEVFN(s->real_device.dev, 130 s->real_device.func), 131 s->real_device.bus, 132 msix_entry, table_base); 133 if (rc) { 134 XEN_PT_ERR(&s->dev, 135 "Mapping of MSI%s (rc: %i, vec: %#x, entry %#x)\n", 136 is_msix ? "-X" : "", rc, gvec, msix_entry); 137 return rc; 138 } 139 } 140 141 return 0; 142 } 143 static int msi_msix_update(XenPCIPassthroughState *s, 144 uint64_t addr, 145 uint32_t data, 146 int pirq, 147 bool is_msix, 148 int msix_entry, 149 int *old_pirq) 150 { 151 PCIDevice *d = &s->dev; 152 uint8_t gvec = msi_vector(data); 153 uint32_t gflags = msi_gflags(data, addr); 154 int rc = 0; 155 uint64_t table_addr = 0; 156 157 XEN_PT_LOG(d, "Updating MSI%s with pirq %d gvec %#x gflags %#x" 158 " (entry: %#x)\n", 159 is_msix ? "-X" : "", pirq, gvec, gflags, msix_entry); 160 161 if (is_msix) { 162 table_addr = s->msix->mmio_base_addr; 163 } 164 165 rc = xc_domain_update_msi_irq(xen_xc, xen_domid, gvec, 166 pirq, gflags, table_addr); 167 168 if (rc) { 169 XEN_PT_ERR(d, "Updating of MSI%s failed. (rc: %d)\n", 170 is_msix ? "-X" : "", rc); 171 172 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, *old_pirq)) { 173 XEN_PT_ERR(d, "Unmapping of MSI%s pirq %d failed.\n", 174 is_msix ? "-X" : "", *old_pirq); 175 } 176 *old_pirq = XEN_PT_UNASSIGNED_PIRQ; 177 } 178 return rc; 179 } 180 181 static int msi_msix_disable(XenPCIPassthroughState *s, 182 uint64_t addr, 183 uint32_t data, 184 int pirq, 185 bool is_msix, 186 bool is_binded) 187 { 188 PCIDevice *d = &s->dev; 189 uint8_t gvec = msi_vector(data); 190 uint32_t gflags = msi_gflags(data, addr); 191 int rc = 0; 192 193 if (pirq == XEN_PT_UNASSIGNED_PIRQ) { 194 return 0; 195 } 196 197 if (is_binded) { 198 XEN_PT_LOG(d, "Unbind MSI%s with pirq %d, gvec %#x\n", 199 is_msix ? "-X" : "", pirq, gvec); 200 rc = xc_domain_unbind_msi_irq(xen_xc, xen_domid, gvec, pirq, gflags); 201 if (rc) { 202 XEN_PT_ERR(d, "Unbinding of MSI%s failed. (pirq: %d, gvec: %#x)\n", 203 is_msix ? "-X" : "", pirq, gvec); 204 return rc; 205 } 206 } 207 208 XEN_PT_LOG(d, "Unmap MSI%s pirq %d\n", is_msix ? "-X" : "", pirq); 209 rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, pirq); 210 if (rc) { 211 XEN_PT_ERR(d, "Unmapping of MSI%s pirq %d failed. (rc: %i)\n", 212 is_msix ? "-X" : "", pirq, rc); 213 return rc; 214 } 215 216 return 0; 217 } 218 219 /* 220 * MSI virtualization functions 221 */ 222 223 int xen_pt_msi_set_enable(XenPCIPassthroughState *s, bool enable) 224 { 225 XEN_PT_LOG(&s->dev, "%s MSI.\n", enable ? "enabling" : "disabling"); 226 227 if (!s->msi) { 228 return -1; 229 } 230 231 return msi_msix_enable(s, s->msi->ctrl_offset, PCI_MSI_FLAGS_ENABLE, 232 enable); 233 } 234 235 /* setup physical msi, but don't enable it */ 236 int xen_pt_msi_setup(XenPCIPassthroughState *s) 237 { 238 int pirq = XEN_PT_UNASSIGNED_PIRQ; 239 int rc = 0; 240 XenPTMSI *msi = s->msi; 241 242 if (msi->initialized) { 243 XEN_PT_ERR(&s->dev, 244 "Setup physical MSI when it has been properly initialized.\n"); 245 return -1; 246 } 247 248 rc = msi_msix_setup(s, msi_addr64(msi), msi->data, &pirq, false, 0, true); 249 if (rc) { 250 return rc; 251 } 252 253 if (pirq < 0) { 254 XEN_PT_ERR(&s->dev, "Invalid pirq number: %d.\n", pirq); 255 return -1; 256 } 257 258 msi->pirq = pirq; 259 XEN_PT_LOG(&s->dev, "MSI mapped with pirq %d.\n", pirq); 260 261 return 0; 262 } 263 264 int xen_pt_msi_update(XenPCIPassthroughState *s) 265 { 266 XenPTMSI *msi = s->msi; 267 return msi_msix_update(s, msi_addr64(msi), msi->data, msi->pirq, 268 false, 0, &msi->pirq); 269 } 270 271 void xen_pt_msi_disable(XenPCIPassthroughState *s) 272 { 273 XenPTMSI *msi = s->msi; 274 275 if (!msi) { 276 return; 277 } 278 279 xen_pt_msi_set_enable(s, false); 280 281 msi_msix_disable(s, msi_addr64(msi), msi->data, msi->pirq, false, 282 msi->initialized); 283 284 /* clear msi info */ 285 msi->flags &= ~PCI_MSI_FLAGS_ENABLE; 286 msi->initialized = false; 287 msi->mapped = false; 288 msi->pirq = XEN_PT_UNASSIGNED_PIRQ; 289 } 290 291 /* 292 * MSI-X virtualization functions 293 */ 294 295 static int msix_set_enable(XenPCIPassthroughState *s, bool enabled) 296 { 297 XEN_PT_LOG(&s->dev, "%s MSI-X.\n", enabled ? "enabling" : "disabling"); 298 299 if (!s->msix) { 300 return -1; 301 } 302 303 return msi_msix_enable(s, s->msix->ctrl_offset, PCI_MSIX_FLAGS_ENABLE, 304 enabled); 305 } 306 307 static int xen_pt_msix_update_one(XenPCIPassthroughState *s, int entry_nr) 308 { 309 XenPTMSIXEntry *entry = NULL; 310 int pirq; 311 int rc; 312 313 if (entry_nr < 0 || entry_nr >= s->msix->total_entries) { 314 return -EINVAL; 315 } 316 317 entry = &s->msix->msix_entry[entry_nr]; 318 319 if (!entry->updated) { 320 return 0; 321 } 322 323 pirq = entry->pirq; 324 325 rc = msi_msix_setup(s, entry->addr, entry->data, &pirq, true, entry_nr, 326 entry->pirq == XEN_PT_UNASSIGNED_PIRQ); 327 if (rc) { 328 return rc; 329 } 330 if (entry->pirq == XEN_PT_UNASSIGNED_PIRQ) { 331 entry->pirq = pirq; 332 } 333 334 rc = msi_msix_update(s, entry->addr, entry->data, pirq, true, 335 entry_nr, &entry->pirq); 336 337 if (!rc) { 338 entry->updated = false; 339 } 340 341 return rc; 342 } 343 344 int xen_pt_msix_update(XenPCIPassthroughState *s) 345 { 346 XenPTMSIX *msix = s->msix; 347 int i; 348 349 for (i = 0; i < msix->total_entries; i++) { 350 xen_pt_msix_update_one(s, i); 351 } 352 353 return 0; 354 } 355 356 void xen_pt_msix_disable(XenPCIPassthroughState *s) 357 { 358 int i = 0; 359 360 msix_set_enable(s, false); 361 362 for (i = 0; i < s->msix->total_entries; i++) { 363 XenPTMSIXEntry *entry = &s->msix->msix_entry[i]; 364 365 msi_msix_disable(s, entry->addr, entry->data, entry->pirq, true, true); 366 367 /* clear MSI-X info */ 368 entry->pirq = XEN_PT_UNASSIGNED_PIRQ; 369 entry->updated = false; 370 } 371 } 372 373 int xen_pt_msix_update_remap(XenPCIPassthroughState *s, int bar_index) 374 { 375 XenPTMSIXEntry *entry; 376 int i, ret; 377 378 if (!(s->msix && s->msix->bar_index == bar_index)) { 379 return 0; 380 } 381 382 for (i = 0; i < s->msix->total_entries; i++) { 383 entry = &s->msix->msix_entry[i]; 384 if (entry->pirq != XEN_PT_UNASSIGNED_PIRQ) { 385 ret = xc_domain_unbind_pt_irq(xen_xc, xen_domid, entry->pirq, 386 PT_IRQ_TYPE_MSI, 0, 0, 0, 0); 387 if (ret) { 388 XEN_PT_ERR(&s->dev, "unbind MSI-X entry %d failed\n", 389 entry->pirq); 390 } 391 entry->updated = true; 392 } 393 } 394 return xen_pt_msix_update(s); 395 } 396 397 static uint32_t get_entry_value(XenPTMSIXEntry *e, int offset) 398 { 399 switch (offset) { 400 case PCI_MSIX_ENTRY_LOWER_ADDR: 401 return e->addr & UINT32_MAX; 402 case PCI_MSIX_ENTRY_UPPER_ADDR: 403 return e->addr >> 32; 404 case PCI_MSIX_ENTRY_DATA: 405 return e->data; 406 case PCI_MSIX_ENTRY_VECTOR_CTRL: 407 return e->vector_ctrl; 408 default: 409 return 0; 410 } 411 } 412 413 static void set_entry_value(XenPTMSIXEntry *e, int offset, uint32_t val) 414 { 415 switch (offset) { 416 case PCI_MSIX_ENTRY_LOWER_ADDR: 417 e->addr = (e->addr & ((uint64_t)UINT32_MAX << 32)) | val; 418 break; 419 case PCI_MSIX_ENTRY_UPPER_ADDR: 420 e->addr = (uint64_t)val << 32 | (e->addr & UINT32_MAX); 421 break; 422 case PCI_MSIX_ENTRY_DATA: 423 e->data = val; 424 break; 425 case PCI_MSIX_ENTRY_VECTOR_CTRL: 426 e->vector_ctrl = val; 427 break; 428 } 429 } 430 431 static void pci_msix_write(void *opaque, hwaddr addr, 432 uint64_t val, unsigned size) 433 { 434 XenPCIPassthroughState *s = opaque; 435 XenPTMSIX *msix = s->msix; 436 XenPTMSIXEntry *entry; 437 int entry_nr, offset; 438 439 entry_nr = addr / PCI_MSIX_ENTRY_SIZE; 440 if (entry_nr < 0 || entry_nr >= msix->total_entries) { 441 XEN_PT_ERR(&s->dev, "asked MSI-X entry '%i' invalid!\n", entry_nr); 442 return; 443 } 444 entry = &msix->msix_entry[entry_nr]; 445 offset = addr % PCI_MSIX_ENTRY_SIZE; 446 447 if (offset != PCI_MSIX_ENTRY_VECTOR_CTRL) { 448 const volatile uint32_t *vec_ctrl; 449 450 if (get_entry_value(entry, offset) == val 451 && entry->pirq != XEN_PT_UNASSIGNED_PIRQ) { 452 return; 453 } 454 455 /* 456 * If Xen intercepts the mask bit access, entry->vec_ctrl may not be 457 * up-to-date. Read from hardware directly. 458 */ 459 vec_ctrl = s->msix->phys_iomem_base + entry_nr * PCI_MSIX_ENTRY_SIZE 460 + PCI_MSIX_ENTRY_VECTOR_CTRL; 461 462 if (msix->enabled && !(*vec_ctrl & PCI_MSIX_ENTRY_CTRL_MASKBIT)) { 463 XEN_PT_ERR(&s->dev, "Can't update msix entry %d since MSI-X is" 464 " already enabled.\n", entry_nr); 465 return; 466 } 467 468 entry->updated = true; 469 } 470 471 set_entry_value(entry, offset, val); 472 473 if (offset == PCI_MSIX_ENTRY_VECTOR_CTRL) { 474 if (msix->enabled && !(val & PCI_MSIX_ENTRY_CTRL_MASKBIT)) { 475 xen_pt_msix_update_one(s, entry_nr); 476 } 477 } 478 } 479 480 static uint64_t pci_msix_read(void *opaque, hwaddr addr, 481 unsigned size) 482 { 483 XenPCIPassthroughState *s = opaque; 484 XenPTMSIX *msix = s->msix; 485 int entry_nr, offset; 486 487 entry_nr = addr / PCI_MSIX_ENTRY_SIZE; 488 if (entry_nr < 0) { 489 XEN_PT_ERR(&s->dev, "asked MSI-X entry '%i' invalid!\n", entry_nr); 490 return 0; 491 } 492 493 offset = addr % PCI_MSIX_ENTRY_SIZE; 494 495 if (addr < msix->total_entries * PCI_MSIX_ENTRY_SIZE) { 496 return get_entry_value(&msix->msix_entry[entry_nr], offset); 497 } else { 498 /* Pending Bit Array (PBA) */ 499 return *(uint32_t *)(msix->phys_iomem_base + addr); 500 } 501 } 502 503 static const MemoryRegionOps pci_msix_ops = { 504 .read = pci_msix_read, 505 .write = pci_msix_write, 506 .endianness = DEVICE_NATIVE_ENDIAN, 507 .valid = { 508 .min_access_size = 4, 509 .max_access_size = 4, 510 .unaligned = false, 511 }, 512 }; 513 514 int xen_pt_msix_init(XenPCIPassthroughState *s, uint32_t base) 515 { 516 uint8_t id = 0; 517 uint16_t control = 0; 518 uint32_t table_off = 0; 519 int i, total_entries, bar_index; 520 XenHostPCIDevice *hd = &s->real_device; 521 PCIDevice *d = &s->dev; 522 int fd = -1; 523 XenPTMSIX *msix = NULL; 524 int rc = 0; 525 526 rc = xen_host_pci_get_byte(hd, base + PCI_CAP_LIST_ID, &id); 527 if (rc) { 528 return rc; 529 } 530 531 if (id != PCI_CAP_ID_MSIX) { 532 XEN_PT_ERR(d, "Invalid id %#x base %#x\n", id, base); 533 return -1; 534 } 535 536 xen_host_pci_get_word(hd, base + PCI_MSIX_FLAGS, &control); 537 total_entries = control & PCI_MSIX_FLAGS_QSIZE; 538 total_entries += 1; 539 540 s->msix = g_malloc0(sizeof (XenPTMSIX) 541 + total_entries * sizeof (XenPTMSIXEntry)); 542 msix = s->msix; 543 544 msix->total_entries = total_entries; 545 for (i = 0; i < total_entries; i++) { 546 msix->msix_entry[i].pirq = XEN_PT_UNASSIGNED_PIRQ; 547 } 548 549 memory_region_init_io(&msix->mmio, OBJECT(s), &pci_msix_ops, 550 s, "xen-pci-pt-msix", 551 (total_entries * PCI_MSIX_ENTRY_SIZE 552 + XC_PAGE_SIZE - 1) 553 & XC_PAGE_MASK); 554 555 xen_host_pci_get_long(hd, base + PCI_MSIX_TABLE, &table_off); 556 bar_index = msix->bar_index = table_off & PCI_MSIX_FLAGS_BIRMASK; 557 table_off = table_off & ~PCI_MSIX_FLAGS_BIRMASK; 558 msix->table_base = s->real_device.io_regions[bar_index].base_addr; 559 XEN_PT_LOG(d, "get MSI-X table BAR base 0x%"PRIx64"\n", msix->table_base); 560 561 fd = open("/dev/mem", O_RDWR); 562 if (fd == -1) { 563 rc = -errno; 564 XEN_PT_ERR(d, "Can't open /dev/mem: %s\n", strerror(errno)); 565 goto error_out; 566 } 567 XEN_PT_LOG(d, "table_off = %#x, total_entries = %d\n", 568 table_off, total_entries); 569 msix->table_offset_adjust = table_off & 0x0fff; 570 msix->phys_iomem_base = 571 mmap(NULL, 572 total_entries * PCI_MSIX_ENTRY_SIZE + msix->table_offset_adjust, 573 PROT_READ, 574 MAP_SHARED | MAP_LOCKED, 575 fd, 576 msix->table_base + table_off - msix->table_offset_adjust); 577 close(fd); 578 if (msix->phys_iomem_base == MAP_FAILED) { 579 rc = -errno; 580 XEN_PT_ERR(d, "Can't map physical MSI-X table: %s\n", strerror(errno)); 581 goto error_out; 582 } 583 msix->phys_iomem_base = (char *)msix->phys_iomem_base 584 + msix->table_offset_adjust; 585 586 XEN_PT_LOG(d, "mapping physical MSI-X table to %p\n", 587 msix->phys_iomem_base); 588 589 memory_region_add_subregion_overlap(&s->bar[bar_index], table_off, 590 &msix->mmio, 591 2); /* Priority: pci default + 1 */ 592 593 return 0; 594 595 error_out: 596 memory_region_destroy(&msix->mmio); 597 g_free(s->msix); 598 s->msix = NULL; 599 return rc; 600 } 601 602 void xen_pt_msix_delete(XenPCIPassthroughState *s) 603 { 604 XenPTMSIX *msix = s->msix; 605 606 if (!msix) { 607 return; 608 } 609 610 /* unmap the MSI-X memory mapped register area */ 611 if (msix->phys_iomem_base) { 612 XEN_PT_LOG(&s->dev, "unmapping physical MSI-X table from %p\n", 613 msix->phys_iomem_base); 614 munmap(msix->phys_iomem_base, msix->total_entries * PCI_MSIX_ENTRY_SIZE 615 + msix->table_offset_adjust); 616 } 617 618 memory_region_del_subregion(&s->bar[msix->bar_index], &msix->mmio); 619 memory_region_destroy(&msix->mmio); 620 621 g_free(s->msix); 622 s->msix = NULL; 623 } 624