1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2020 Intel Corporation 4 * Author: Johannes Berg <johannes@sipsolutions.net> 5 */ 6 #include <linux/module.h> 7 #include <linux/pci.h> 8 #include <linux/virtio.h> 9 #include <linux/virtio_config.h> 10 #include <linux/logic_iomem.h> 11 #include <linux/of_platform.h> 12 #include <linux/irqdomain.h> 13 #include <linux/virtio_pcidev.h> 14 #include <linux/virtio-uml.h> 15 #include <linux/delay.h> 16 #include <linux/msi.h> 17 #include <asm/unaligned.h> 18 #include <irq_kern.h> 19 20 #define MAX_DEVICES 8 21 #define MAX_MSI_VECTORS 32 22 #define CFG_SPACE_SIZE 4096 23 24 /* for MSI-X we have a 32-bit payload */ 25 #define MAX_IRQ_MSG_SIZE (sizeof(struct virtio_pcidev_msg) + sizeof(u32)) 26 #define NUM_IRQ_MSGS 10 27 28 #define HANDLE_NO_FREE(ptr) ((void *)((unsigned long)(ptr) | 1)) 29 #define HANDLE_IS_NO_FREE(ptr) ((unsigned long)(ptr) & 1) 30 31 struct um_pci_device { 32 struct virtio_device *vdev; 33 34 /* for now just standard BARs */ 35 u8 resptr[PCI_STD_NUM_BARS]; 36 37 struct virtqueue *cmd_vq, *irq_vq; 38 39 #define UM_PCI_STAT_WAITING 0 40 unsigned long status; 41 42 int irq; 43 44 bool platform; 45 }; 46 47 struct um_pci_device_reg { 48 struct um_pci_device *dev; 49 void __iomem *iomem; 50 }; 51 52 static struct pci_host_bridge *bridge; 53 static DEFINE_MUTEX(um_pci_mtx); 54 static struct um_pci_device *um_pci_platform_device; 55 static struct um_pci_device_reg um_pci_devices[MAX_DEVICES]; 56 static struct fwnode_handle *um_pci_fwnode; 57 static struct irq_domain *um_pci_inner_domain; 58 static struct irq_domain *um_pci_msi_domain; 59 static unsigned long um_pci_msi_used[BITS_TO_LONGS(MAX_MSI_VECTORS)]; 60 61 static unsigned int um_pci_max_delay_us = 40000; 62 module_param_named(max_delay_us, um_pci_max_delay_us, uint, 0644); 63 64 struct um_pci_message_buffer { 65 struct virtio_pcidev_msg hdr; 66 u8 data[8]; 67 }; 68 69 static struct um_pci_message_buffer __percpu *um_pci_msg_bufs; 70 71 static int um_pci_send_cmd(struct um_pci_device *dev, 72 struct virtio_pcidev_msg *cmd, 73 unsigned int cmd_size, 74 const void *extra, unsigned int extra_size, 75 void *out, unsigned int out_size) 76 { 77 struct scatterlist out_sg, extra_sg, in_sg; 78 struct scatterlist *sgs_list[] = { 79 [0] = &out_sg, 80 [1] = extra ? &extra_sg : &in_sg, 81 [2] = extra ? &in_sg : NULL, 82 }; 83 struct um_pci_message_buffer *buf; 84 int delay_count = 0; 85 int ret, len; 86 bool posted; 87 88 if (WARN_ON(cmd_size < sizeof(*cmd) || cmd_size > sizeof(*buf))) 89 return -EINVAL; 90 91 switch (cmd->op) { 92 case VIRTIO_PCIDEV_OP_CFG_WRITE: 93 case VIRTIO_PCIDEV_OP_MMIO_WRITE: 94 case VIRTIO_PCIDEV_OP_MMIO_MEMSET: 95 /* in PCI, writes are posted, so don't wait */ 96 posted = !out; 97 WARN_ON(!posted); 98 break; 99 default: 100 posted = false; 101 break; 102 } 103 104 buf = get_cpu_var(um_pci_msg_bufs); 105 if (buf) 106 memcpy(buf, cmd, cmd_size); 107 108 if (posted) { 109 u8 *ncmd = kmalloc(cmd_size + extra_size, GFP_ATOMIC); 110 111 if (ncmd) { 112 memcpy(ncmd, cmd, cmd_size); 113 if (extra) 114 memcpy(ncmd + cmd_size, extra, extra_size); 115 cmd = (void *)ncmd; 116 cmd_size += extra_size; 117 extra = NULL; 118 extra_size = 0; 119 } else { 120 /* try without allocating memory */ 121 posted = false; 122 cmd = (void *)buf; 123 } 124 } else { 125 cmd = (void *)buf; 126 } 127 128 sg_init_one(&out_sg, cmd, cmd_size); 129 if (extra) 130 sg_init_one(&extra_sg, extra, extra_size); 131 if (out) 132 sg_init_one(&in_sg, out, out_size); 133 134 /* add to internal virtio queue */ 135 ret = virtqueue_add_sgs(dev->cmd_vq, sgs_list, 136 extra ? 2 : 1, 137 out ? 1 : 0, 138 posted ? cmd : HANDLE_NO_FREE(cmd), 139 GFP_ATOMIC); 140 if (ret) { 141 if (posted) 142 kfree(cmd); 143 goto out; 144 } 145 146 if (posted) { 147 virtqueue_kick(dev->cmd_vq); 148 ret = 0; 149 goto out; 150 } 151 152 /* kick and poll for getting a response on the queue */ 153 set_bit(UM_PCI_STAT_WAITING, &dev->status); 154 virtqueue_kick(dev->cmd_vq); 155 156 while (1) { 157 void *completed = virtqueue_get_buf(dev->cmd_vq, &len); 158 159 if (completed == HANDLE_NO_FREE(cmd)) 160 break; 161 162 if (completed && !HANDLE_IS_NO_FREE(completed)) 163 kfree(completed); 164 165 if (WARN_ONCE(virtqueue_is_broken(dev->cmd_vq) || 166 ++delay_count > um_pci_max_delay_us, 167 "um virt-pci delay: %d", delay_count)) { 168 ret = -EIO; 169 break; 170 } 171 udelay(1); 172 } 173 clear_bit(UM_PCI_STAT_WAITING, &dev->status); 174 175 out: 176 put_cpu_var(um_pci_msg_bufs); 177 return ret; 178 } 179 180 static unsigned long um_pci_cfgspace_read(void *priv, unsigned int offset, 181 int size) 182 { 183 struct um_pci_device_reg *reg = priv; 184 struct um_pci_device *dev = reg->dev; 185 struct virtio_pcidev_msg hdr = { 186 .op = VIRTIO_PCIDEV_OP_CFG_READ, 187 .size = size, 188 .addr = offset, 189 }; 190 /* buf->data is maximum size - we may only use parts of it */ 191 struct um_pci_message_buffer *buf; 192 u8 *data; 193 unsigned long ret = ULONG_MAX; 194 size_t bytes = sizeof(buf->data); 195 196 if (!dev) 197 return ULONG_MAX; 198 199 buf = get_cpu_var(um_pci_msg_bufs); 200 data = buf->data; 201 202 if (buf) 203 memset(data, 0xff, bytes); 204 205 switch (size) { 206 case 1: 207 case 2: 208 case 4: 209 #ifdef CONFIG_64BIT 210 case 8: 211 #endif 212 break; 213 default: 214 WARN(1, "invalid config space read size %d\n", size); 215 goto out; 216 } 217 218 if (um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, data, bytes)) 219 goto out; 220 221 switch (size) { 222 case 1: 223 ret = data[0]; 224 break; 225 case 2: 226 ret = le16_to_cpup((void *)data); 227 break; 228 case 4: 229 ret = le32_to_cpup((void *)data); 230 break; 231 #ifdef CONFIG_64BIT 232 case 8: 233 ret = le64_to_cpup((void *)data); 234 break; 235 #endif 236 default: 237 break; 238 } 239 240 out: 241 put_cpu_var(um_pci_msg_bufs); 242 return ret; 243 } 244 245 static void um_pci_cfgspace_write(void *priv, unsigned int offset, int size, 246 unsigned long val) 247 { 248 struct um_pci_device_reg *reg = priv; 249 struct um_pci_device *dev = reg->dev; 250 struct { 251 struct virtio_pcidev_msg hdr; 252 /* maximum size - we may only use parts of it */ 253 u8 data[8]; 254 } msg = { 255 .hdr = { 256 .op = VIRTIO_PCIDEV_OP_CFG_WRITE, 257 .size = size, 258 .addr = offset, 259 }, 260 }; 261 262 if (!dev) 263 return; 264 265 switch (size) { 266 case 1: 267 msg.data[0] = (u8)val; 268 break; 269 case 2: 270 put_unaligned_le16(val, (void *)msg.data); 271 break; 272 case 4: 273 put_unaligned_le32(val, (void *)msg.data); 274 break; 275 #ifdef CONFIG_64BIT 276 case 8: 277 put_unaligned_le64(val, (void *)msg.data); 278 break; 279 #endif 280 default: 281 WARN(1, "invalid config space write size %d\n", size); 282 return; 283 } 284 285 WARN_ON(um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0)); 286 } 287 288 static const struct logic_iomem_ops um_pci_device_cfgspace_ops = { 289 .read = um_pci_cfgspace_read, 290 .write = um_pci_cfgspace_write, 291 }; 292 293 static void um_pci_bar_copy_from(void *priv, void *buffer, 294 unsigned int offset, int size) 295 { 296 u8 *resptr = priv; 297 struct um_pci_device *dev = container_of(resptr - *resptr, 298 struct um_pci_device, 299 resptr[0]); 300 struct virtio_pcidev_msg hdr = { 301 .op = VIRTIO_PCIDEV_OP_MMIO_READ, 302 .bar = *resptr, 303 .size = size, 304 .addr = offset, 305 }; 306 307 memset(buffer, 0xff, size); 308 309 um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, buffer, size); 310 } 311 312 static unsigned long um_pci_bar_read(void *priv, unsigned int offset, 313 int size) 314 { 315 /* buf->data is maximum size - we may only use parts of it */ 316 struct um_pci_message_buffer *buf; 317 u8 *data; 318 unsigned long ret = ULONG_MAX; 319 320 buf = get_cpu_var(um_pci_msg_bufs); 321 data = buf->data; 322 323 switch (size) { 324 case 1: 325 case 2: 326 case 4: 327 #ifdef CONFIG_64BIT 328 case 8: 329 #endif 330 break; 331 default: 332 WARN(1, "invalid config space read size %d\n", size); 333 goto out; 334 } 335 336 um_pci_bar_copy_from(priv, data, offset, size); 337 338 switch (size) { 339 case 1: 340 ret = data[0]; 341 break; 342 case 2: 343 ret = le16_to_cpup((void *)data); 344 break; 345 case 4: 346 ret = le32_to_cpup((void *)data); 347 break; 348 #ifdef CONFIG_64BIT 349 case 8: 350 ret = le64_to_cpup((void *)data); 351 break; 352 #endif 353 default: 354 break; 355 } 356 357 out: 358 put_cpu_var(um_pci_msg_bufs); 359 return ret; 360 } 361 362 static void um_pci_bar_copy_to(void *priv, unsigned int offset, 363 const void *buffer, int size) 364 { 365 u8 *resptr = priv; 366 struct um_pci_device *dev = container_of(resptr - *resptr, 367 struct um_pci_device, 368 resptr[0]); 369 struct virtio_pcidev_msg hdr = { 370 .op = VIRTIO_PCIDEV_OP_MMIO_WRITE, 371 .bar = *resptr, 372 .size = size, 373 .addr = offset, 374 }; 375 376 um_pci_send_cmd(dev, &hdr, sizeof(hdr), buffer, size, NULL, 0); 377 } 378 379 static void um_pci_bar_write(void *priv, unsigned int offset, int size, 380 unsigned long val) 381 { 382 /* maximum size - we may only use parts of it */ 383 u8 data[8]; 384 385 switch (size) { 386 case 1: 387 data[0] = (u8)val; 388 break; 389 case 2: 390 put_unaligned_le16(val, (void *)data); 391 break; 392 case 4: 393 put_unaligned_le32(val, (void *)data); 394 break; 395 #ifdef CONFIG_64BIT 396 case 8: 397 put_unaligned_le64(val, (void *)data); 398 break; 399 #endif 400 default: 401 WARN(1, "invalid config space write size %d\n", size); 402 return; 403 } 404 405 um_pci_bar_copy_to(priv, offset, data, size); 406 } 407 408 static void um_pci_bar_set(void *priv, unsigned int offset, u8 value, int size) 409 { 410 u8 *resptr = priv; 411 struct um_pci_device *dev = container_of(resptr - *resptr, 412 struct um_pci_device, 413 resptr[0]); 414 struct { 415 struct virtio_pcidev_msg hdr; 416 u8 data; 417 } msg = { 418 .hdr = { 419 .op = VIRTIO_PCIDEV_OP_CFG_WRITE, 420 .bar = *resptr, 421 .size = size, 422 .addr = offset, 423 }, 424 .data = value, 425 }; 426 427 um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0); 428 } 429 430 static const struct logic_iomem_ops um_pci_device_bar_ops = { 431 .read = um_pci_bar_read, 432 .write = um_pci_bar_write, 433 .set = um_pci_bar_set, 434 .copy_from = um_pci_bar_copy_from, 435 .copy_to = um_pci_bar_copy_to, 436 }; 437 438 static void __iomem *um_pci_map_bus(struct pci_bus *bus, unsigned int devfn, 439 int where) 440 { 441 struct um_pci_device_reg *dev; 442 unsigned int busn = bus->number; 443 444 if (busn > 0) 445 return NULL; 446 447 /* not allowing functions for now ... */ 448 if (devfn % 8) 449 return NULL; 450 451 if (devfn / 8 >= ARRAY_SIZE(um_pci_devices)) 452 return NULL; 453 454 dev = &um_pci_devices[devfn / 8]; 455 if (!dev) 456 return NULL; 457 458 return (void __iomem *)((unsigned long)dev->iomem + where); 459 } 460 461 static struct pci_ops um_pci_ops = { 462 .map_bus = um_pci_map_bus, 463 .read = pci_generic_config_read, 464 .write = pci_generic_config_write, 465 }; 466 467 static void um_pci_rescan(void) 468 { 469 pci_lock_rescan_remove(); 470 pci_rescan_bus(bridge->bus); 471 pci_unlock_rescan_remove(); 472 } 473 474 static void um_pci_irq_vq_addbuf(struct virtqueue *vq, void *buf, bool kick) 475 { 476 struct scatterlist sg[1]; 477 478 sg_init_one(sg, buf, MAX_IRQ_MSG_SIZE); 479 if (virtqueue_add_inbuf(vq, sg, 1, buf, GFP_ATOMIC)) 480 kfree(buf); 481 else if (kick) 482 virtqueue_kick(vq); 483 } 484 485 static void um_pci_handle_irq_message(struct virtqueue *vq, 486 struct virtio_pcidev_msg *msg) 487 { 488 struct virtio_device *vdev = vq->vdev; 489 struct um_pci_device *dev = vdev->priv; 490 491 if (!dev->irq) 492 return; 493 494 /* we should properly chain interrupts, but on ARCH=um we don't care */ 495 496 switch (msg->op) { 497 case VIRTIO_PCIDEV_OP_INT: 498 generic_handle_irq(dev->irq); 499 break; 500 case VIRTIO_PCIDEV_OP_MSI: 501 /* our MSI message is just the interrupt number */ 502 if (msg->size == sizeof(u32)) 503 generic_handle_irq(le32_to_cpup((void *)msg->data)); 504 else 505 generic_handle_irq(le16_to_cpup((void *)msg->data)); 506 break; 507 case VIRTIO_PCIDEV_OP_PME: 508 /* nothing to do - we already woke up due to the message */ 509 break; 510 default: 511 dev_err(&vdev->dev, "unexpected virt-pci message %d\n", msg->op); 512 break; 513 } 514 } 515 516 static void um_pci_cmd_vq_cb(struct virtqueue *vq) 517 { 518 struct virtio_device *vdev = vq->vdev; 519 struct um_pci_device *dev = vdev->priv; 520 void *cmd; 521 int len; 522 523 if (test_bit(UM_PCI_STAT_WAITING, &dev->status)) 524 return; 525 526 while ((cmd = virtqueue_get_buf(vq, &len))) { 527 if (WARN_ON(HANDLE_IS_NO_FREE(cmd))) 528 continue; 529 kfree(cmd); 530 } 531 } 532 533 static void um_pci_irq_vq_cb(struct virtqueue *vq) 534 { 535 struct virtio_pcidev_msg *msg; 536 int len; 537 538 while ((msg = virtqueue_get_buf(vq, &len))) { 539 if (len >= sizeof(*msg)) 540 um_pci_handle_irq_message(vq, msg); 541 542 /* recycle the message buffer */ 543 um_pci_irq_vq_addbuf(vq, msg, true); 544 } 545 } 546 547 /* Copied from arch/x86/kernel/devicetree.c */ 548 struct device_node *pcibios_get_phb_of_node(struct pci_bus *bus) 549 { 550 struct device_node *np; 551 552 for_each_node_by_type(np, "pci") { 553 const void *prop; 554 unsigned int bus_min; 555 556 prop = of_get_property(np, "bus-range", NULL); 557 if (!prop) 558 continue; 559 bus_min = be32_to_cpup(prop); 560 if (bus->number == bus_min) 561 return np; 562 } 563 return NULL; 564 } 565 566 static int um_pci_init_vqs(struct um_pci_device *dev) 567 { 568 struct virtqueue *vqs[2]; 569 static const char *const names[2] = { "cmd", "irq" }; 570 vq_callback_t *cbs[2] = { um_pci_cmd_vq_cb, um_pci_irq_vq_cb }; 571 int err, i; 572 573 err = virtio_find_vqs(dev->vdev, 2, vqs, cbs, names, NULL); 574 if (err) 575 return err; 576 577 dev->cmd_vq = vqs[0]; 578 dev->irq_vq = vqs[1]; 579 580 virtio_device_ready(dev->vdev); 581 582 for (i = 0; i < NUM_IRQ_MSGS; i++) { 583 void *msg = kzalloc(MAX_IRQ_MSG_SIZE, GFP_KERNEL); 584 585 if (msg) 586 um_pci_irq_vq_addbuf(dev->irq_vq, msg, false); 587 } 588 589 virtqueue_kick(dev->irq_vq); 590 591 return 0; 592 } 593 594 static void __um_pci_virtio_platform_remove(struct virtio_device *vdev, 595 struct um_pci_device *dev) 596 { 597 virtio_reset_device(vdev); 598 vdev->config->del_vqs(vdev); 599 600 mutex_lock(&um_pci_mtx); 601 um_pci_platform_device = NULL; 602 mutex_unlock(&um_pci_mtx); 603 604 kfree(dev); 605 } 606 607 static int um_pci_virtio_platform_probe(struct virtio_device *vdev, 608 struct um_pci_device *dev) 609 { 610 int ret; 611 612 dev->platform = true; 613 614 mutex_lock(&um_pci_mtx); 615 616 if (um_pci_platform_device) { 617 mutex_unlock(&um_pci_mtx); 618 ret = -EBUSY; 619 goto out_free; 620 } 621 622 ret = um_pci_init_vqs(dev); 623 if (ret) { 624 mutex_unlock(&um_pci_mtx); 625 goto out_free; 626 } 627 628 um_pci_platform_device = dev; 629 630 mutex_unlock(&um_pci_mtx); 631 632 ret = of_platform_default_populate(vdev->dev.of_node, NULL, &vdev->dev); 633 if (ret) 634 __um_pci_virtio_platform_remove(vdev, dev); 635 636 return ret; 637 638 out_free: 639 kfree(dev); 640 return ret; 641 } 642 643 static int um_pci_virtio_probe(struct virtio_device *vdev) 644 { 645 struct um_pci_device *dev; 646 int i, free = -1; 647 int err = -ENOSPC; 648 649 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 650 if (!dev) 651 return -ENOMEM; 652 653 dev->vdev = vdev; 654 vdev->priv = dev; 655 656 if (of_device_is_compatible(vdev->dev.of_node, "simple-bus")) 657 return um_pci_virtio_platform_probe(vdev, dev); 658 659 mutex_lock(&um_pci_mtx); 660 for (i = 0; i < MAX_DEVICES; i++) { 661 if (um_pci_devices[i].dev) 662 continue; 663 free = i; 664 break; 665 } 666 667 if (free < 0) 668 goto error; 669 670 err = um_pci_init_vqs(dev); 671 if (err) 672 goto error; 673 674 dev->irq = irq_alloc_desc(numa_node_id()); 675 if (dev->irq < 0) { 676 err = dev->irq; 677 goto err_reset; 678 } 679 um_pci_devices[free].dev = dev; 680 vdev->priv = dev; 681 682 mutex_unlock(&um_pci_mtx); 683 684 device_set_wakeup_enable(&vdev->dev, true); 685 686 /* 687 * In order to do suspend-resume properly, don't allow VQs 688 * to be suspended. 689 */ 690 virtio_uml_set_no_vq_suspend(vdev, true); 691 692 um_pci_rescan(); 693 return 0; 694 err_reset: 695 virtio_reset_device(vdev); 696 vdev->config->del_vqs(vdev); 697 error: 698 mutex_unlock(&um_pci_mtx); 699 kfree(dev); 700 return err; 701 } 702 703 static void um_pci_virtio_remove(struct virtio_device *vdev) 704 { 705 struct um_pci_device *dev = vdev->priv; 706 int i; 707 708 if (dev->platform) { 709 of_platform_depopulate(&vdev->dev); 710 __um_pci_virtio_platform_remove(vdev, dev); 711 return; 712 } 713 714 device_set_wakeup_enable(&vdev->dev, false); 715 716 mutex_lock(&um_pci_mtx); 717 for (i = 0; i < MAX_DEVICES; i++) { 718 if (um_pci_devices[i].dev != dev) 719 continue; 720 721 um_pci_devices[i].dev = NULL; 722 irq_free_desc(dev->irq); 723 724 break; 725 } 726 mutex_unlock(&um_pci_mtx); 727 728 if (i < MAX_DEVICES) { 729 struct pci_dev *pci_dev; 730 731 pci_dev = pci_get_slot(bridge->bus, i); 732 if (pci_dev) 733 pci_stop_and_remove_bus_device_locked(pci_dev); 734 } 735 736 /* Stop all virtqueues */ 737 virtio_reset_device(vdev); 738 dev->cmd_vq = NULL; 739 dev->irq_vq = NULL; 740 vdev->config->del_vqs(vdev); 741 742 kfree(dev); 743 } 744 745 static struct virtio_device_id id_table[] = { 746 { CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID, VIRTIO_DEV_ANY_ID }, 747 { 0 }, 748 }; 749 MODULE_DEVICE_TABLE(virtio, id_table); 750 751 static struct virtio_driver um_pci_virtio_driver = { 752 .driver.name = "virtio-pci", 753 .driver.owner = THIS_MODULE, 754 .id_table = id_table, 755 .probe = um_pci_virtio_probe, 756 .remove = um_pci_virtio_remove, 757 }; 758 759 static struct resource virt_cfgspace_resource = { 760 .name = "PCI config space", 761 .start = 0xf0000000 - MAX_DEVICES * CFG_SPACE_SIZE, 762 .end = 0xf0000000 - 1, 763 .flags = IORESOURCE_MEM, 764 }; 765 766 static long um_pci_map_cfgspace(unsigned long offset, size_t size, 767 const struct logic_iomem_ops **ops, 768 void **priv) 769 { 770 if (WARN_ON(size > CFG_SPACE_SIZE || offset % CFG_SPACE_SIZE)) 771 return -EINVAL; 772 773 if (offset / CFG_SPACE_SIZE < MAX_DEVICES) { 774 *ops = &um_pci_device_cfgspace_ops; 775 *priv = &um_pci_devices[offset / CFG_SPACE_SIZE]; 776 return 0; 777 } 778 779 WARN(1, "cannot map offset 0x%lx/0x%zx\n", offset, size); 780 return -ENOENT; 781 } 782 783 static const struct logic_iomem_region_ops um_pci_cfgspace_ops = { 784 .map = um_pci_map_cfgspace, 785 }; 786 787 static struct resource virt_iomem_resource = { 788 .name = "PCI iomem", 789 .start = 0xf0000000, 790 .end = 0xffffffff, 791 .flags = IORESOURCE_MEM, 792 }; 793 794 struct um_pci_map_iomem_data { 795 unsigned long offset; 796 size_t size; 797 const struct logic_iomem_ops **ops; 798 void **priv; 799 long ret; 800 }; 801 802 static int um_pci_map_iomem_walk(struct pci_dev *pdev, void *_data) 803 { 804 struct um_pci_map_iomem_data *data = _data; 805 struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8]; 806 struct um_pci_device *dev; 807 int i; 808 809 if (!reg->dev) 810 return 0; 811 812 for (i = 0; i < ARRAY_SIZE(dev->resptr); i++) { 813 struct resource *r = &pdev->resource[i]; 814 815 if ((r->flags & IORESOURCE_TYPE_BITS) != IORESOURCE_MEM) 816 continue; 817 818 /* 819 * must be the whole or part of the resource, 820 * not allowed to only overlap 821 */ 822 if (data->offset < r->start || data->offset > r->end) 823 continue; 824 if (data->offset + data->size - 1 > r->end) 825 continue; 826 827 dev = reg->dev; 828 *data->ops = &um_pci_device_bar_ops; 829 dev->resptr[i] = i; 830 *data->priv = &dev->resptr[i]; 831 data->ret = data->offset - r->start; 832 833 /* no need to continue */ 834 return 1; 835 } 836 837 return 0; 838 } 839 840 static long um_pci_map_iomem(unsigned long offset, size_t size, 841 const struct logic_iomem_ops **ops, 842 void **priv) 843 { 844 struct um_pci_map_iomem_data data = { 845 /* we want the full address here */ 846 .offset = offset + virt_iomem_resource.start, 847 .size = size, 848 .ops = ops, 849 .priv = priv, 850 .ret = -ENOENT, 851 }; 852 853 pci_walk_bus(bridge->bus, um_pci_map_iomem_walk, &data); 854 return data.ret; 855 } 856 857 static const struct logic_iomem_region_ops um_pci_iomem_ops = { 858 .map = um_pci_map_iomem, 859 }; 860 861 static void um_pci_compose_msi_msg(struct irq_data *data, struct msi_msg *msg) 862 { 863 /* 864 * This is a very low address and not actually valid 'physical' memory 865 * in UML, so we can simply map MSI(-X) vectors to there, it cannot be 866 * legitimately written to by the device in any other way. 867 * We use the (virtual) IRQ number here as the message to simplify the 868 * code that receives the message, where for now we simply trust the 869 * device to send the correct message. 870 */ 871 msg->address_hi = 0; 872 msg->address_lo = 0xa0000; 873 msg->data = data->irq; 874 } 875 876 static struct irq_chip um_pci_msi_bottom_irq_chip = { 877 .name = "UM virtio MSI", 878 .irq_compose_msi_msg = um_pci_compose_msi_msg, 879 }; 880 881 static int um_pci_inner_domain_alloc(struct irq_domain *domain, 882 unsigned int virq, unsigned int nr_irqs, 883 void *args) 884 { 885 unsigned long bit; 886 887 WARN_ON(nr_irqs != 1); 888 889 mutex_lock(&um_pci_mtx); 890 bit = find_first_zero_bit(um_pci_msi_used, MAX_MSI_VECTORS); 891 if (bit >= MAX_MSI_VECTORS) { 892 mutex_unlock(&um_pci_mtx); 893 return -ENOSPC; 894 } 895 896 set_bit(bit, um_pci_msi_used); 897 mutex_unlock(&um_pci_mtx); 898 899 irq_domain_set_info(domain, virq, bit, &um_pci_msi_bottom_irq_chip, 900 domain->host_data, handle_simple_irq, 901 NULL, NULL); 902 903 return 0; 904 } 905 906 static void um_pci_inner_domain_free(struct irq_domain *domain, 907 unsigned int virq, unsigned int nr_irqs) 908 { 909 struct irq_data *d = irq_domain_get_irq_data(domain, virq); 910 911 mutex_lock(&um_pci_mtx); 912 913 if (!test_bit(d->hwirq, um_pci_msi_used)) 914 pr_err("trying to free unused MSI#%lu\n", d->hwirq); 915 else 916 __clear_bit(d->hwirq, um_pci_msi_used); 917 918 mutex_unlock(&um_pci_mtx); 919 } 920 921 static const struct irq_domain_ops um_pci_inner_domain_ops = { 922 .alloc = um_pci_inner_domain_alloc, 923 .free = um_pci_inner_domain_free, 924 }; 925 926 static struct irq_chip um_pci_msi_irq_chip = { 927 .name = "UM virtio PCIe MSI", 928 .irq_mask = pci_msi_mask_irq, 929 .irq_unmask = pci_msi_unmask_irq, 930 }; 931 932 static struct msi_domain_info um_pci_msi_domain_info = { 933 .flags = MSI_FLAG_USE_DEF_DOM_OPS | 934 MSI_FLAG_USE_DEF_CHIP_OPS | 935 MSI_FLAG_PCI_MSIX, 936 .chip = &um_pci_msi_irq_chip, 937 }; 938 939 static struct resource busn_resource = { 940 .name = "PCI busn", 941 .start = 0, 942 .end = 0, 943 .flags = IORESOURCE_BUS, 944 }; 945 946 static int um_pci_map_irq(const struct pci_dev *pdev, u8 slot, u8 pin) 947 { 948 struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8]; 949 950 if (WARN_ON(!reg->dev)) 951 return -EINVAL; 952 953 /* Yes, we map all pins to the same IRQ ... doesn't matter for now. */ 954 return reg->dev->irq; 955 } 956 957 void *pci_root_bus_fwnode(struct pci_bus *bus) 958 { 959 return um_pci_fwnode; 960 } 961 962 static long um_pci_map_platform(unsigned long offset, size_t size, 963 const struct logic_iomem_ops **ops, 964 void **priv) 965 { 966 if (!um_pci_platform_device) 967 return -ENOENT; 968 969 *ops = &um_pci_device_bar_ops; 970 *priv = &um_pci_platform_device->resptr[0]; 971 972 return 0; 973 } 974 975 static const struct logic_iomem_region_ops um_pci_platform_ops = { 976 .map = um_pci_map_platform, 977 }; 978 979 static struct resource virt_platform_resource = { 980 .name = "platform", 981 .start = 0x10000000, 982 .end = 0x1fffffff, 983 .flags = IORESOURCE_MEM, 984 }; 985 986 static int __init um_pci_init(void) 987 { 988 int err, i; 989 990 WARN_ON(logic_iomem_add_region(&virt_cfgspace_resource, 991 &um_pci_cfgspace_ops)); 992 WARN_ON(logic_iomem_add_region(&virt_iomem_resource, 993 &um_pci_iomem_ops)); 994 WARN_ON(logic_iomem_add_region(&virt_platform_resource, 995 &um_pci_platform_ops)); 996 997 if (WARN(CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID < 0, 998 "No virtio device ID configured for PCI - no PCI support\n")) 999 return 0; 1000 1001 um_pci_msg_bufs = alloc_percpu(struct um_pci_message_buffer); 1002 if (!um_pci_msg_bufs) 1003 return -ENOMEM; 1004 1005 bridge = pci_alloc_host_bridge(0); 1006 if (!bridge) { 1007 err = -ENOMEM; 1008 goto free; 1009 } 1010 1011 um_pci_fwnode = irq_domain_alloc_named_fwnode("um-pci"); 1012 if (!um_pci_fwnode) { 1013 err = -ENOMEM; 1014 goto free; 1015 } 1016 1017 um_pci_inner_domain = __irq_domain_add(um_pci_fwnode, MAX_MSI_VECTORS, 1018 MAX_MSI_VECTORS, 0, 1019 &um_pci_inner_domain_ops, NULL); 1020 if (!um_pci_inner_domain) { 1021 err = -ENOMEM; 1022 goto free; 1023 } 1024 1025 um_pci_msi_domain = pci_msi_create_irq_domain(um_pci_fwnode, 1026 &um_pci_msi_domain_info, 1027 um_pci_inner_domain); 1028 if (!um_pci_msi_domain) { 1029 err = -ENOMEM; 1030 goto free; 1031 } 1032 1033 pci_add_resource(&bridge->windows, &virt_iomem_resource); 1034 pci_add_resource(&bridge->windows, &busn_resource); 1035 bridge->ops = &um_pci_ops; 1036 bridge->map_irq = um_pci_map_irq; 1037 1038 for (i = 0; i < MAX_DEVICES; i++) { 1039 resource_size_t start; 1040 1041 start = virt_cfgspace_resource.start + i * CFG_SPACE_SIZE; 1042 um_pci_devices[i].iomem = ioremap(start, CFG_SPACE_SIZE); 1043 if (WARN(!um_pci_devices[i].iomem, "failed to map %d\n", i)) { 1044 err = -ENOMEM; 1045 goto free; 1046 } 1047 } 1048 1049 err = pci_host_probe(bridge); 1050 if (err) 1051 goto free; 1052 1053 err = register_virtio_driver(&um_pci_virtio_driver); 1054 if (err) 1055 goto free; 1056 return 0; 1057 free: 1058 if (um_pci_inner_domain) 1059 irq_domain_remove(um_pci_inner_domain); 1060 if (um_pci_fwnode) 1061 irq_domain_free_fwnode(um_pci_fwnode); 1062 if (bridge) { 1063 pci_free_resource_list(&bridge->windows); 1064 pci_free_host_bridge(bridge); 1065 } 1066 free_percpu(um_pci_msg_bufs); 1067 return err; 1068 } 1069 module_init(um_pci_init); 1070 1071 static void __exit um_pci_exit(void) 1072 { 1073 unregister_virtio_driver(&um_pci_virtio_driver); 1074 irq_domain_remove(um_pci_msi_domain); 1075 irq_domain_remove(um_pci_inner_domain); 1076 pci_free_resource_list(&bridge->windows); 1077 pci_free_host_bridge(bridge); 1078 free_percpu(um_pci_msg_bufs); 1079 } 1080 module_exit(um_pci_exit); 1081