1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. 24 */ 25 26/* 27 * Copyright (c) 2009, Intel Corporation. 28 * All rights reserved. 29 */ 30 31 32#include <sys/apic.h> 33#include <vm/hat_i86.h> 34#include <sys/sysmacros.h> 35#include <sys/smp_impldefs.h> 36#include <sys/immu.h> 37 38 39typedef struct intrmap_private { 40 immu_t *ir_immu; 41 immu_inv_wait_t ir_inv_wait; 42 uint16_t ir_idx; 43 uint32_t ir_sid_svt_sq; 44} intrmap_private_t; 45 46#define INTRMAP_PRIVATE(intrmap) ((intrmap_private_t *)intrmap) 47 48/* interrupt remapping table entry */ 49typedef struct intrmap_rte { 50 uint64_t lo; 51 uint64_t hi; 52} intrmap_rte_t; 53 54#define IRTE_HIGH(sid_svt_sq) (sid_svt_sq) 55#define IRTE_LOW(dst, vector, dlm, tm, rh, dm, fpd, p) \ 56 (((uint64_t)(dst) << 32) | \ 57 ((uint64_t)(vector) << 16) | \ 58 ((uint64_t)(dlm) << 5) | \ 59 ((uint64_t)(tm) << 4) | \ 60 ((uint64_t)(rh) << 3) | \ 61 ((uint64_t)(dm) << 2) | \ 62 ((uint64_t)(fpd) << 1) | \ 63 (p)) 64 65typedef enum { 66 SVT_NO_VERIFY = 0, /* no verification */ 67 SVT_ALL_VERIFY, /* using sid and sq to verify */ 68 SVT_BUS_VERIFY, /* verify #startbus and #endbus */ 69 SVT_RSVD 70} intrmap_svt_t; 71 72typedef enum { 73 SQ_VERIFY_ALL = 0, /* verify all 16 bits */ 74 SQ_VERIFY_IGR_1, /* ignore bit 3 */ 75 SQ_VERIFY_IGR_2, /* ignore bit 2-3 */ 76 SQ_VERIFY_IGR_3 /* ignore bit 1-3 */ 77} intrmap_sq_t; 78 79/* 80 * S field of the Interrupt Remapping Table Address Register 81 * the size of the interrupt remapping table is 1 << (immu_intrmap_irta_s + 1) 82 */ 83static uint_t intrmap_irta_s = INTRMAP_MAX_IRTA_SIZE; 84 85/* 86 * If true, arrange to suppress broadcast EOI by setting edge-triggered mode 87 * even for level-triggered interrupts in the interrupt-remapping engine. 88 * If false, broadcast EOI can still be suppressed if the CPU supports the 89 * APIC_SVR_SUPPRESS_BROADCAST_EOI bit. In both cases, the IOAPIC is still 90 * programmed with the correct trigger mode, and pcplusmp must send an EOI 91 * to the IOAPIC by writing to the IOAPIC's EOI register to make up for the 92 * missing broadcast EOI. 93 */ 94static int intrmap_suppress_brdcst_eoi = 0; 95 96/* 97 * whether verify the source id of interrupt request 98 */ 99static int intrmap_enable_sid_verify = 0; 100 101/* fault types for DVMA remapping */ 102static char *immu_dvma_faults[] = { 103 "Reserved", 104 "The present field in root-entry is Clear", 105 "The present field in context-entry is Clear", 106 "Hardware detected invalid programming of a context-entry", 107 "The DMA request attempted to access an address beyond max support", 108 "The Write field in a page-table entry is Clear when DMA write", 109 "The Read field in a page-table entry is Clear when DMA read", 110 "Access the next level page table resulted in error", 111 "Access the root-entry table resulted in error", 112 "Access the context-entry table resulted in error", 113 "Reserved field not initialized to zero in a present root-entry", 114 "Reserved field not initialized to zero in a present context-entry", 115 "Reserved field not initialized to zero in a present page-table entry", 116 "DMA blocked due to the Translation Type field in context-entry", 117 "Incorrect fault event reason number", 118}; 119#define DVMA_MAX_FAULTS (sizeof (immu_dvma_faults)/(sizeof (char *))) - 1 120 121/* fault types for interrupt remapping */ 122static char *immu_intrmap_faults[] = { 123 "reserved field set in IRTE", 124 "interrupt_index exceed the intr-remap table size", 125 "present field in IRTE is clear", 126 "hardware access intr-remap table address resulted in error", 127 "reserved field set in IRTE, include various conditional", 128 "hardware blocked an interrupt request in Compatibility format", 129 "remappable interrupt request blocked due to verification failure" 130}; 131#define INTRMAP_MAX_FAULTS \ 132 (sizeof (immu_intrmap_faults) / (sizeof (char *))) - 1 133 134/* Function prototypes */ 135static int immu_intrmap_init(int apic_mode); 136static void immu_intrmap_switchon(int suppress_brdcst_eoi); 137static void immu_intrmap_alloc(void **intrmap_private_tbl, dev_info_t *dip, 138 uint16_t type, int count, uchar_t ioapic_index); 139static void immu_intrmap_map(void *intrmap_private, void *intrmap_data, 140 uint16_t type, int count); 141static void immu_intrmap_free(void **intrmap_privatep); 142static void immu_intrmap_rdt(void *intrmap_private, ioapic_rdt_t *irdt); 143static void immu_intrmap_msi(void *intrmap_private, msi_regs_t *mregs); 144 145static struct apic_intrmap_ops intrmap_ops = { 146 immu_intrmap_init, 147 immu_intrmap_switchon, 148 immu_intrmap_alloc, 149 immu_intrmap_map, 150 immu_intrmap_free, 151 immu_intrmap_rdt, 152 immu_intrmap_msi, 153}; 154 155/* apic mode, APIC/X2APIC */ 156static int intrmap_apic_mode = LOCAL_APIC; 157 158 159/* 160 * helper functions 161 */ 162static uint_t 163bitset_find_free(bitset_t *b, uint_t post) 164{ 165 uint_t i; 166 uint_t cap = bitset_capacity(b); 167 168 if (post == cap) 169 post = 0; 170 171 ASSERT(post < cap); 172 173 for (i = post; i < cap; i++) { 174 if (!bitset_in_set(b, i)) 175 return (i); 176 } 177 178 for (i = 0; i < post; i++) { 179 if (!bitset_in_set(b, i)) 180 return (i); 181 } 182 183 return (INTRMAP_IDX_FULL); /* no free index */ 184} 185 186/* 187 * helper function to find 'count' contigous free 188 * interrupt remapping table entries 189 */ 190static uint_t 191bitset_find_multi_free(bitset_t *b, uint_t post, uint_t count) 192{ 193 uint_t i, j; 194 uint_t cap = bitset_capacity(b); 195 196 if (post == INTRMAP_IDX_FULL) { 197 return (INTRMAP_IDX_FULL); 198 } 199 200 if (count > cap) 201 return (INTRMAP_IDX_FULL); 202 203 ASSERT(post < cap); 204 205 for (i = post; (i + count) <= cap; i++) { 206 for (j = 0; j < count; j++) { 207 if (bitset_in_set(b, (i + j))) { 208 i = i + j; 209 break; 210 } 211 if (j == count - 1) 212 return (i); 213 } 214 } 215 216 for (i = 0; (i < post) && ((i + count) <= cap); i++) { 217 for (j = 0; j < count; j++) { 218 if (bitset_in_set(b, (i + j))) { 219 i = i + j; 220 break; 221 } 222 if (j == count - 1) 223 return (i); 224 } 225 } 226 227 return (INTRMAP_IDX_FULL); /* no free index */ 228} 229 230/* alloc one interrupt remapping table entry */ 231static int 232alloc_tbl_entry(intrmap_t *intrmap) 233{ 234 uint32_t idx; 235 236 for (;;) { 237 mutex_enter(&intrmap->intrmap_lock); 238 idx = intrmap->intrmap_free; 239 if (idx != INTRMAP_IDX_FULL) { 240 bitset_add(&intrmap->intrmap_map, idx); 241 intrmap->intrmap_free = 242 bitset_find_free(&intrmap->intrmap_map, idx + 1); 243 mutex_exit(&intrmap->intrmap_lock); 244 break; 245 } 246 247 /* no free intr entry, use compatible format intr */ 248 mutex_exit(&intrmap->intrmap_lock); 249 250 if (intrmap_apic_mode != LOCAL_X2APIC) { 251 break; 252 } 253 254 /* 255 * x2apic mode not allowed compatible 256 * interrupt 257 */ 258 delay(IMMU_ALLOC_RESOURCE_DELAY); 259 } 260 261 return (idx); 262} 263 264/* alloc 'cnt' contigous interrupt remapping table entries */ 265static int 266alloc_tbl_multi_entries(intrmap_t *intrmap, uint_t cnt) 267{ 268 uint_t idx, pos, i; 269 270 for (; ; ) { 271 mutex_enter(&intrmap->intrmap_lock); 272 pos = intrmap->intrmap_free; 273 idx = bitset_find_multi_free(&intrmap->intrmap_map, pos, cnt); 274 275 if (idx != INTRMAP_IDX_FULL) { 276 if (idx <= pos && pos < (idx + cnt)) { 277 intrmap->intrmap_free = bitset_find_free( 278 &intrmap->intrmap_map, idx + cnt); 279 } 280 for (i = 0; i < cnt; i++) { 281 bitset_add(&intrmap->intrmap_map, idx + i); 282 } 283 mutex_exit(&intrmap->intrmap_lock); 284 break; 285 } 286 287 mutex_exit(&intrmap->intrmap_lock); 288 289 if (intrmap_apic_mode != LOCAL_X2APIC) { 290 break; 291 } 292 293 /* x2apic mode not allowed comapitible interrupt */ 294 delay(IMMU_ALLOC_RESOURCE_DELAY); 295 } 296 297 return (idx); 298} 299 300/* init interrupt remapping table */ 301static int 302init_unit(immu_t *immu) 303{ 304 intrmap_t *intrmap; 305 size_t size; 306 307 ddi_dma_attr_t intrmap_dma_attr = { 308 DMA_ATTR_V0, 309 0U, 310 0xffffffffffffffffULL, 311 0xffffffffU, 312 MMU_PAGESIZE, /* page aligned */ 313 0x1, 314 0x1, 315 0xffffffffU, 316 0xffffffffffffffffULL, 317 1, 318 4, 319 0 320 }; 321 322 ddi_device_acc_attr_t intrmap_acc_attr = { 323 DDI_DEVICE_ATTR_V0, 324 DDI_NEVERSWAP_ACC, 325 DDI_STRICTORDER_ACC 326 }; 327 328 /* 329 * Using interrupt remapping implies using the queue 330 * invalidation interface. According to Intel, 331 * hardware that supports interrupt remapping should 332 * also support QI. 333 */ 334 ASSERT(IMMU_ECAP_GET_QI(immu->immu_regs_excap)); 335 336 if (intrmap_apic_mode == LOCAL_X2APIC) { 337 if (!IMMU_ECAP_GET_EIM(immu->immu_regs_excap)) { 338 return (DDI_FAILURE); 339 } 340 } 341 342 if (intrmap_irta_s > INTRMAP_MAX_IRTA_SIZE) { 343 intrmap_irta_s = INTRMAP_MAX_IRTA_SIZE; 344 } 345 346 intrmap = kmem_zalloc(sizeof (intrmap_t), KM_SLEEP); 347 348 if (ddi_dma_alloc_handle(immu->immu_dip, 349 &intrmap_dma_attr, 350 DDI_DMA_SLEEP, 351 NULL, 352 &(intrmap->intrmap_dma_hdl)) != DDI_SUCCESS) { 353 kmem_free(intrmap, sizeof (intrmap_t)); 354 return (DDI_FAILURE); 355 } 356 357 intrmap->intrmap_size = 1 << (intrmap_irta_s + 1); 358 size = intrmap->intrmap_size * INTRMAP_RTE_SIZE; 359 if (ddi_dma_mem_alloc(intrmap->intrmap_dma_hdl, 360 size, 361 &intrmap_acc_attr, 362 DDI_DMA_CONSISTENT | IOMEM_DATA_UNCACHED, 363 DDI_DMA_SLEEP, 364 NULL, 365 &(intrmap->intrmap_vaddr), 366 &size, 367 &(intrmap->intrmap_acc_hdl)) != DDI_SUCCESS) { 368 ddi_dma_free_handle(&(intrmap->intrmap_dma_hdl)); 369 kmem_free(intrmap, sizeof (intrmap_t)); 370 return (DDI_FAILURE); 371 } 372 373 ASSERT(!((uintptr_t)intrmap->intrmap_vaddr & MMU_PAGEOFFSET)); 374 bzero(intrmap->intrmap_vaddr, size); 375 intrmap->intrmap_paddr = pfn_to_pa( 376 hat_getpfnum(kas.a_hat, intrmap->intrmap_vaddr)); 377 378 mutex_init(&(intrmap->intrmap_lock), NULL, MUTEX_DRIVER, NULL); 379 bitset_init(&intrmap->intrmap_map); 380 bitset_resize(&intrmap->intrmap_map, intrmap->intrmap_size); 381 intrmap->intrmap_free = 0; 382 383 immu->immu_intrmap = intrmap; 384 385 return (DDI_SUCCESS); 386} 387 388static immu_t * 389get_immu(dev_info_t *dip, uint16_t type, uchar_t ioapic_index) 390{ 391 immu_t *immu = NULL; 392 393 if (!DDI_INTR_IS_MSI_OR_MSIX(type)) { 394 immu = immu_dmar_ioapic_immu(ioapic_index); 395 } else { 396 if (dip != NULL) 397 immu = immu_dmar_get_immu(dip); 398 } 399 400 return (immu); 401} 402 403static int 404get_top_pcibridge(dev_info_t *dip, void *arg) 405{ 406 dev_info_t **topdipp = arg; 407 immu_devi_t *immu_devi; 408 409 mutex_enter(&(DEVI(dip)->devi_lock)); 410 immu_devi = DEVI(dip)->devi_iommu; 411 mutex_exit(&(DEVI(dip)->devi_lock)); 412 413 if (immu_devi == NULL || immu_devi->imd_pcib_type == IMMU_PCIB_BAD || 414 immu_devi->imd_pcib_type == IMMU_PCIB_ENDPOINT) { 415 return (DDI_WALK_CONTINUE); 416 } 417 418 *topdipp = dip; 419 420 return (DDI_WALK_CONTINUE); 421} 422 423static dev_info_t * 424intrmap_top_pcibridge(dev_info_t *rdip) 425{ 426 dev_info_t *top_pcibridge = NULL; 427 428 if (immu_walk_ancestor(rdip, NULL, get_top_pcibridge, 429 &top_pcibridge, NULL, 0) != DDI_SUCCESS) { 430 return (NULL); 431 } 432 433 return (top_pcibridge); 434} 435 436/* function to get interrupt request source id */ 437static uint32_t 438get_sid(dev_info_t *dip, uint16_t type, uchar_t ioapic_index) 439{ 440 dev_info_t *pdip; 441 immu_devi_t *immu_devi; 442 uint16_t sid; 443 uchar_t svt, sq; 444 445 if (!intrmap_enable_sid_verify) { 446 return (0); 447 } 448 449 if (!DDI_INTR_IS_MSI_OR_MSIX(type)) { 450 /* for interrupt through I/O APIC */ 451 sid = immu_dmar_ioapic_sid(ioapic_index); 452 svt = SVT_ALL_VERIFY; 453 sq = SQ_VERIFY_ALL; 454 } else { 455 /* MSI/MSI-X interrupt */ 456 ASSERT(dip); 457 pdip = intrmap_top_pcibridge(dip); 458 ASSERT(pdip); 459 immu_devi = DEVI(pdip)->devi_iommu; 460 ASSERT(immu_devi); 461 if (immu_devi->imd_pcib_type == IMMU_PCIB_PCIE_PCI) { 462 /* device behind pcie to pci bridge */ 463 sid = (immu_devi->imd_bus << 8) | immu_devi->imd_sec; 464 svt = SVT_BUS_VERIFY; 465 sq = SQ_VERIFY_ALL; 466 } else { 467 /* pcie device or device behind pci to pci bridge */ 468 sid = (immu_devi->imd_bus << 8) | 469 immu_devi->imd_devfunc; 470 svt = SVT_ALL_VERIFY; 471 sq = SQ_VERIFY_ALL; 472 } 473 } 474 475 return (sid | (svt << 18) | (sq << 16)); 476} 477 478static void 479intrmap_enable(immu_t *immu) 480{ 481 intrmap_t *intrmap; 482 uint64_t irta_reg; 483 484 intrmap = immu->immu_intrmap; 485 486 irta_reg = intrmap->intrmap_paddr | intrmap_irta_s; 487 if (intrmap_apic_mode == LOCAL_X2APIC) { 488 irta_reg |= (0x1 << 11); 489 } 490 491 immu_regs_intrmap_enable(immu, irta_reg); 492} 493 494/* ####################################################################### */ 495 496/* 497 * immu_intr_handler() 498 * the fault event handler for a single immu unit 499 */ 500int 501immu_intr_handler(immu_t *immu) 502{ 503 uint32_t status; 504 int index, fault_reg_offset; 505 int max_fault_index; 506 boolean_t found_fault; 507 dev_info_t *idip; 508 509 mutex_enter(&(immu->immu_intr_lock)); 510 mutex_enter(&(immu->immu_regs_lock)); 511 512 /* read the fault status */ 513 status = immu_regs_get32(immu, IMMU_REG_FAULT_STS); 514 515 idip = immu->immu_dip; 516 ASSERT(idip); 517 518 /* check if we have a pending fault for this immu unit */ 519 if ((status & IMMU_FAULT_STS_PPF) == 0) { 520 mutex_exit(&(immu->immu_regs_lock)); 521 mutex_exit(&(immu->immu_intr_lock)); 522 return (DDI_INTR_UNCLAIMED); 523 } 524 525 /* 526 * handle all primary pending faults 527 */ 528 index = IMMU_FAULT_GET_INDEX(status); 529 max_fault_index = IMMU_CAP_GET_NFR(immu->immu_regs_cap) - 1; 530 fault_reg_offset = IMMU_CAP_GET_FRO(immu->immu_regs_cap); 531 532 found_fault = B_FALSE; 533 _NOTE(CONSTCOND) 534 while (1) { 535 uint64_t val; 536 uint8_t fault_reason; 537 uint8_t fault_type; 538 uint16_t sid; 539 uint64_t pg_addr; 540 uint64_t idx; 541 542 /* read the higher 64bits */ 543 val = immu_regs_get64(immu, fault_reg_offset + index * 16 + 8); 544 545 /* check if this fault register has pending fault */ 546 if (!IMMU_FRR_GET_F(val)) { 547 break; 548 } 549 550 found_fault = B_TRUE; 551 552 /* get the fault reason, fault type and sid */ 553 fault_reason = IMMU_FRR_GET_FR(val); 554 fault_type = IMMU_FRR_GET_FT(val); 555 sid = IMMU_FRR_GET_SID(val); 556 557 /* read the first 64bits */ 558 val = immu_regs_get64(immu, fault_reg_offset + index * 16); 559 pg_addr = val & IMMU_PAGEMASK; 560 idx = val >> 48; 561 562 /* clear the fault */ 563 immu_regs_put32(immu, fault_reg_offset + index * 16 + 12, 564 (((uint32_t)1) << 31)); 565 566 /* report the fault info */ 567 if (fault_reason < 0x20) { 568 /* immu-remapping fault */ 569 ddi_err(DER_WARN, idip, 570 "generated a fault event when translating DMA %s\n" 571 "\t on address 0x%" PRIx64 " for PCI(%d, %d, %d), " 572 "the reason is:\n\t %s", 573 fault_type ? "read" : "write", pg_addr, 574 (sid >> 8) & 0xff, (sid >> 3) & 0x1f, sid & 0x7, 575 immu_dvma_faults[MIN(fault_reason, 576 DVMA_MAX_FAULTS)]); 577 immu_print_fault_info(sid, pg_addr); 578 } else if (fault_reason < 0x27) { 579 /* intr-remapping fault */ 580 ddi_err(DER_WARN, idip, 581 "generated a fault event when translating " 582 "interrupt request\n" 583 "\t on index 0x%" PRIx64 " for PCI(%d, %d, %d), " 584 "the reason is:\n\t %s", 585 idx, 586 (sid >> 8) & 0xff, (sid >> 3) & 0x1f, sid & 0x7, 587 immu_intrmap_faults[MIN((fault_reason - 0x20), 588 INTRMAP_MAX_FAULTS)]); 589 } else { 590 ddi_err(DER_WARN, idip, "Unknown fault reason: 0x%x", 591 fault_reason); 592 } 593 594 index++; 595 if (index > max_fault_index) 596 index = 0; 597 } 598 599 /* Clear the fault */ 600 if (!found_fault) { 601 ddi_err(DER_MODE, idip, 602 "Fault register set but no fault present"); 603 } 604 immu_regs_put32(immu, IMMU_REG_FAULT_STS, 1); 605 mutex_exit(&(immu->immu_regs_lock)); 606 mutex_exit(&(immu->immu_intr_lock)); 607 return (DDI_INTR_CLAIMED); 608} 609/* ######################################################################### */ 610 611/* 612 * Interrupt remap entry points 613 */ 614 615/* initialize interrupt remapping */ 616static int 617immu_intrmap_init(int apic_mode) 618{ 619 immu_t *immu; 620 int error = DDI_FAILURE; 621 622 if (immu_intrmap_enable == B_FALSE) { 623 return (DDI_SUCCESS); 624 } 625 626 intrmap_apic_mode = apic_mode; 627 628 immu = list_head(&immu_list); 629 for (; immu; immu = list_next(&immu_list, immu)) { 630 if ((immu->immu_intrmap_running == B_TRUE) && 631 IMMU_ECAP_GET_IR(immu->immu_regs_excap)) { 632 if (init_unit(immu) == DDI_SUCCESS) { 633 error = DDI_SUCCESS; 634 } 635 } 636 } 637 638 /* 639 * if all IOMMU units disable intr remapping, 640 * return FAILURE 641 */ 642 return (error); 643} 644 645 646 647/* enable interrupt remapping */ 648static void 649immu_intrmap_switchon(int suppress_brdcst_eoi) 650{ 651 immu_t *immu; 652 653 654 intrmap_suppress_brdcst_eoi = suppress_brdcst_eoi; 655 656 immu = list_head(&immu_list); 657 for (; immu; immu = list_next(&immu_list, immu)) { 658 if (immu->immu_intrmap_setup == B_TRUE) { 659 intrmap_enable(immu); 660 } 661 } 662} 663 664/* alloc remapping entry for the interrupt */ 665static void 666immu_intrmap_alloc(void **intrmap_private_tbl, dev_info_t *dip, 667 uint16_t type, int count, uchar_t ioapic_index) 668{ 669 immu_t *immu; 670 intrmap_t *intrmap; 671 immu_inv_wait_t *iwp; 672 uint32_t idx, i; 673 uint32_t sid_svt_sq; 674 intrmap_private_t *intrmap_private; 675 676 if (intrmap_private_tbl[0] == INTRMAP_DISABLE || 677 intrmap_private_tbl[0] != NULL) { 678 return; 679 } 680 681 intrmap_private_tbl[0] = 682 kmem_zalloc(sizeof (intrmap_private_t), KM_SLEEP); 683 intrmap_private = INTRMAP_PRIVATE(intrmap_private_tbl[0]); 684 685 immu = get_immu(dip, type, ioapic_index); 686 if ((immu != NULL) && (immu->immu_intrmap_running == B_TRUE)) { 687 intrmap_private->ir_immu = immu; 688 } else { 689 goto intrmap_disable; 690 } 691 692 intrmap = immu->immu_intrmap; 693 694 if (count == 1) { 695 idx = alloc_tbl_entry(intrmap); 696 } else { 697 idx = alloc_tbl_multi_entries(intrmap, count); 698 } 699 700 if (idx == INTRMAP_IDX_FULL) { 701 goto intrmap_disable; 702 } 703 704 intrmap_private->ir_idx = idx; 705 706 sid_svt_sq = intrmap_private->ir_sid_svt_sq = 707 get_sid(dip, type, ioapic_index); 708 iwp = &intrmap_private->ir_inv_wait; 709 immu_init_inv_wait(iwp, "intrmaplocal", B_TRUE); 710 711 if (count == 1) { 712 if (IMMU_CAP_GET_CM(immu->immu_regs_cap)) { 713 immu_qinv_intr_one_cache(immu, idx, iwp); 714 } else { 715 immu_regs_wbf_flush(immu); 716 } 717 return; 718 } 719 720 for (i = 1; i < count; i++) { 721 intrmap_private_tbl[i] = 722 kmem_zalloc(sizeof (intrmap_private_t), KM_SLEEP); 723 724 INTRMAP_PRIVATE(intrmap_private_tbl[i])->ir_immu = immu; 725 INTRMAP_PRIVATE(intrmap_private_tbl[i])->ir_sid_svt_sq = 726 sid_svt_sq; 727 INTRMAP_PRIVATE(intrmap_private_tbl[i])->ir_idx = idx + i; 728 } 729 730 if (IMMU_CAP_GET_CM(immu->immu_regs_cap)) { 731 immu_qinv_intr_caches(immu, idx, count, iwp); 732 } else { 733 immu_regs_wbf_flush(immu); 734 } 735 736 return; 737 738intrmap_disable: 739 kmem_free(intrmap_private_tbl[0], sizeof (intrmap_private_t)); 740 intrmap_private_tbl[0] = INTRMAP_DISABLE; 741} 742 743 744/* remapping the interrupt */ 745static void 746immu_intrmap_map(void *intrmap_private, void *intrmap_data, uint16_t type, 747 int count) 748{ 749 immu_t *immu; 750 immu_inv_wait_t *iwp; 751 intrmap_t *intrmap; 752 ioapic_rdt_t *irdt = (ioapic_rdt_t *)intrmap_data; 753 msi_regs_t *mregs = (msi_regs_t *)intrmap_data; 754 intrmap_rte_t irte; 755 uint_t idx, i; 756 uint32_t dst, sid_svt_sq; 757 uchar_t vector, dlm, tm, rh, dm; 758 759 if (intrmap_private == INTRMAP_DISABLE) 760 return; 761 762 idx = INTRMAP_PRIVATE(intrmap_private)->ir_idx; 763 immu = INTRMAP_PRIVATE(intrmap_private)->ir_immu; 764 iwp = &INTRMAP_PRIVATE(intrmap_private)->ir_inv_wait; 765 intrmap = immu->immu_intrmap; 766 sid_svt_sq = INTRMAP_PRIVATE(intrmap_private)->ir_sid_svt_sq; 767 768 if (!DDI_INTR_IS_MSI_OR_MSIX(type)) { 769 dm = RDT_DM(irdt->ir_lo); 770 rh = 0; 771 tm = RDT_TM(irdt->ir_lo); 772 dlm = RDT_DLM(irdt->ir_lo); 773 dst = irdt->ir_hi; 774 775 /* 776 * Mark the IRTE's TM as Edge to suppress broadcast EOI. 777 */ 778 if (intrmap_suppress_brdcst_eoi) { 779 tm = TRIGGER_MODE_EDGE; 780 } 781 782 vector = RDT_VECTOR(irdt->ir_lo); 783 } else { 784 dm = MSI_ADDR_DM_PHYSICAL; 785 rh = MSI_ADDR_RH_FIXED; 786 tm = TRIGGER_MODE_EDGE; 787 dlm = 0; 788 dst = mregs->mr_addr; 789 790 vector = mregs->mr_data & 0xff; 791 } 792 793 if (intrmap_apic_mode == LOCAL_APIC) 794 dst = (dst & 0xFF) << 8; 795 796 if (count == 1) { 797 irte.lo = IRTE_LOW(dst, vector, dlm, tm, rh, dm, 0, 1); 798 irte.hi = IRTE_HIGH(sid_svt_sq); 799 800 /* set interrupt remapping table entry */ 801 bcopy(&irte, intrmap->intrmap_vaddr + 802 idx * INTRMAP_RTE_SIZE, 803 INTRMAP_RTE_SIZE); 804 805 immu_qinv_intr_one_cache(immu, idx, iwp); 806 807 } else { 808 for (i = 0; i < count; i++) { 809 irte.lo = IRTE_LOW(dst, vector, dlm, tm, rh, dm, 0, 1); 810 irte.hi = IRTE_HIGH(sid_svt_sq); 811 812 /* set interrupt remapping table entry */ 813 bcopy(&irte, intrmap->intrmap_vaddr + 814 idx * INTRMAP_RTE_SIZE, 815 INTRMAP_RTE_SIZE); 816 vector++; 817 idx++; 818 } 819 820 immu_qinv_intr_caches(immu, idx, count, iwp); 821 } 822} 823 824/* free the remapping entry */ 825static void 826immu_intrmap_free(void **intrmap_privatep) 827{ 828 immu_t *immu; 829 immu_inv_wait_t *iwp; 830 intrmap_t *intrmap; 831 uint32_t idx; 832 833 if (*intrmap_privatep == INTRMAP_DISABLE || *intrmap_privatep == NULL) { 834 *intrmap_privatep = NULL; 835 return; 836 } 837 838 immu = INTRMAP_PRIVATE(*intrmap_privatep)->ir_immu; 839 iwp = &INTRMAP_PRIVATE(*intrmap_privatep)->ir_inv_wait; 840 intrmap = immu->immu_intrmap; 841 idx = INTRMAP_PRIVATE(*intrmap_privatep)->ir_idx; 842 843 bzero(intrmap->intrmap_vaddr + idx * INTRMAP_RTE_SIZE, 844 INTRMAP_RTE_SIZE); 845 846 immu_qinv_intr_one_cache(immu, idx, iwp); 847 848 mutex_enter(&intrmap->intrmap_lock); 849 bitset_del(&intrmap->intrmap_map, idx); 850 if (intrmap->intrmap_free == INTRMAP_IDX_FULL) { 851 intrmap->intrmap_free = idx; 852 } 853 mutex_exit(&intrmap->intrmap_lock); 854 855 kmem_free(*intrmap_privatep, sizeof (intrmap_private_t)); 856 *intrmap_privatep = NULL; 857} 858 859/* record the ioapic rdt entry */ 860static void 861immu_intrmap_rdt(void *intrmap_private, ioapic_rdt_t *irdt) 862{ 863 uint32_t rdt_entry, tm, pol, idx, vector; 864 865 rdt_entry = irdt->ir_lo; 866 867 if (intrmap_private != INTRMAP_DISABLE && intrmap_private != NULL) { 868 idx = INTRMAP_PRIVATE(intrmap_private)->ir_idx; 869 tm = RDT_TM(rdt_entry); 870 pol = RDT_POL(rdt_entry); 871 vector = RDT_VECTOR(rdt_entry); 872 irdt->ir_lo = (tm << INTRMAP_IOAPIC_TM_SHIFT) | 873 (pol << INTRMAP_IOAPIC_POL_SHIFT) | 874 ((idx >> 15) << INTRMAP_IOAPIC_IDX15_SHIFT) | 875 vector; 876 irdt->ir_hi = (idx << INTRMAP_IOAPIC_IDX_SHIFT) | 877 (1 << INTRMAP_IOAPIC_FORMAT_SHIFT); 878 } else { 879 irdt->ir_hi <<= APIC_ID_BIT_OFFSET; 880 } 881} 882 883/* record the msi interrupt structure */ 884/*ARGSUSED*/ 885static void 886immu_intrmap_msi(void *intrmap_private, msi_regs_t *mregs) 887{ 888 uint_t idx; 889 890 if (intrmap_private != INTRMAP_DISABLE && intrmap_private != NULL) { 891 idx = INTRMAP_PRIVATE(intrmap_private)->ir_idx; 892 893 mregs->mr_data = 0; 894 mregs->mr_addr = MSI_ADDR_HDR | 895 ((idx & 0x7fff) << INTRMAP_MSI_IDX_SHIFT) | 896 (1 << INTRMAP_MSI_FORMAT_SHIFT) | 897 (1 << INTRMAP_MSI_SHV_SHIFT) | 898 ((idx >> 15) << INTRMAP_MSI_IDX15_SHIFT); 899 } else { 900 mregs->mr_addr = MSI_ADDR_HDR | 901 (MSI_ADDR_RH_FIXED << MSI_ADDR_RH_SHIFT) | 902 (MSI_ADDR_DM_PHYSICAL << MSI_ADDR_DM_SHIFT) | 903 (mregs->mr_addr << MSI_ADDR_DEST_SHIFT); 904 mregs->mr_data = (MSI_DATA_TM_EDGE << MSI_DATA_TM_SHIFT) | 905 mregs->mr_data; 906 } 907} 908 909/* ######################################################################### */ 910/* 911 * Functions exported by immu_intr.c 912 */ 913void 914immu_intrmap_setup(list_t *listp) 915{ 916 immu_t *immu; 917 918 /* 919 * Check if ACPI DMAR tables say that 920 * interrupt remapping is supported 921 */ 922 if (immu_dmar_intrmap_supported() == B_FALSE) { 923 return; 924 } 925 926 /* 927 * Check if interrupt remapping is disabled. 928 */ 929 if (immu_intrmap_enable == B_FALSE) { 930 return; 931 } 932 933 psm_vt_ops = &intrmap_ops; 934 935 immu = list_head(listp); 936 for (; immu; immu = list_next(listp, immu)) { 937 mutex_init(&(immu->immu_intrmap_lock), NULL, 938 MUTEX_DEFAULT, NULL); 939 mutex_enter(&(immu->immu_intrmap_lock)); 940 immu_init_inv_wait(&immu->immu_intrmap_inv_wait, 941 "intrmapglobal", B_TRUE); 942 immu->immu_intrmap_setup = B_TRUE; 943 mutex_exit(&(immu->immu_intrmap_lock)); 944 } 945} 946 947void 948immu_intrmap_startup(immu_t *immu) 949{ 950 /* do nothing */ 951 mutex_enter(&(immu->immu_intrmap_lock)); 952 if (immu->immu_intrmap_setup == B_TRUE) { 953 immu->immu_intrmap_running = B_TRUE; 954 } 955 mutex_exit(&(immu->immu_intrmap_lock)); 956} 957 958/* 959 * Register a Intel IOMMU unit (i.e. DMAR unit's) 960 * interrupt handler 961 */ 962void 963immu_intr_register(immu_t *immu) 964{ 965 int irq, vect; 966 char intr_handler_name[IMMU_MAXNAMELEN]; 967 uint32_t msi_data; 968 uint32_t uaddr; 969 uint32_t msi_addr; 970 uint32_t localapic_id = 0; 971 972 if (psm_get_localapicid) 973 localapic_id = psm_get_localapicid(0); 974 975 msi_addr = (MSI_ADDR_HDR | 976 ((localapic_id & 0xFF) << MSI_ADDR_DEST_SHIFT) | 977 (MSI_ADDR_RH_FIXED << MSI_ADDR_RH_SHIFT) | 978 (MSI_ADDR_DM_PHYSICAL << MSI_ADDR_DM_SHIFT)); 979 980 if (intrmap_apic_mode == LOCAL_X2APIC) { 981 uaddr = localapic_id & 0xFFFFFF00; 982 } else { 983 uaddr = 0; 984 } 985 986 /* Dont need to hold immu_intr_lock since we are in boot */ 987 irq = vect = psm_get_ipivect(IMMU_INTR_IPL, -1); 988 if (psm_xlate_vector_by_irq != NULL) 989 vect = psm_xlate_vector_by_irq(irq); 990 991 msi_data = ((MSI_DATA_DELIVERY_FIXED << 992 MSI_DATA_DELIVERY_SHIFT) | vect); 993 994 (void) snprintf(intr_handler_name, sizeof (intr_handler_name), 995 "%s-intr-handler", immu->immu_name); 996 997 (void) add_avintr((void *)NULL, IMMU_INTR_IPL, 998 (avfunc)(immu_intr_handler), intr_handler_name, irq, 999 (caddr_t)immu, NULL, NULL, NULL); 1000 1001 immu_regs_intr_enable(immu, msi_addr, msi_data, uaddr); 1002 1003 (void) immu_intr_handler(immu); 1004} 1005