Linux Audio

Check our new training course

Loading...
v3.1
 
   1/*
   2 * Xen PCI Frontend.
   3 *
   4 *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
   5 */
   6#include <linux/module.h>
   7#include <linux/init.h>
   8#include <linux/mm.h>
   9#include <xen/xenbus.h>
  10#include <xen/events.h>
  11#include <xen/grant_table.h>
  12#include <xen/page.h>
  13#include <linux/spinlock.h>
  14#include <linux/pci.h>
  15#include <linux/msi.h>
  16#include <xen/interface/io/pciif.h>
  17#include <asm/xen/pci.h>
  18#include <linux/interrupt.h>
  19#include <linux/atomic.h>
  20#include <linux/workqueue.h>
  21#include <linux/bitops.h>
  22#include <linux/time.h>
 
 
 
 
 
  23
  24#define INVALID_GRANT_REF (0)
  25#define INVALID_EVTCHN    (-1)
  26
  27struct pci_bus_entry {
  28	struct list_head list;
  29	struct pci_bus *bus;
  30};
  31
  32#define _PDEVB_op_active		(0)
  33#define PDEVB_op_active			(1 << (_PDEVB_op_active))
  34
  35struct pcifront_device {
  36	struct xenbus_device *xdev;
  37	struct list_head root_buses;
  38
  39	int evtchn;
  40	int gnt_ref;
  41
  42	int irq;
  43
  44	/* Lock this when doing any operations in sh_info */
  45	spinlock_t sh_info_lock;
  46	struct xen_pci_sharedinfo *sh_info;
  47	struct work_struct op_work;
  48	unsigned long flags;
  49
  50};
  51
  52struct pcifront_sd {
  53	int domain;
  54	struct pcifront_device *pdev;
  55};
  56
  57static inline struct pcifront_device *
  58pcifront_get_pdev(struct pcifront_sd *sd)
  59{
  60	return sd->pdev;
  61}
  62
  63static inline void pcifront_init_sd(struct pcifront_sd *sd,
  64				    unsigned int domain, unsigned int bus,
  65				    struct pcifront_device *pdev)
  66{
  67	sd->domain = domain;
 
 
  68	sd->pdev = pdev;
  69}
  70
  71static DEFINE_SPINLOCK(pcifront_dev_lock);
  72static struct pcifront_device *pcifront_dev;
  73
  74static int verbose_request;
  75module_param(verbose_request, int, 0644);
  76
  77static int errno_to_pcibios_err(int errno)
  78{
  79	switch (errno) {
  80	case XEN_PCI_ERR_success:
  81		return PCIBIOS_SUCCESSFUL;
  82
  83	case XEN_PCI_ERR_dev_not_found:
  84		return PCIBIOS_DEVICE_NOT_FOUND;
  85
  86	case XEN_PCI_ERR_invalid_offset:
  87	case XEN_PCI_ERR_op_failed:
  88		return PCIBIOS_BAD_REGISTER_NUMBER;
  89
  90	case XEN_PCI_ERR_not_implemented:
  91		return PCIBIOS_FUNC_NOT_SUPPORTED;
  92
  93	case XEN_PCI_ERR_access_denied:
  94		return PCIBIOS_SET_FAILED;
  95	}
  96	return errno;
  97}
  98
  99static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
 100{
 101	if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
 102		&& !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
 103		dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
 104		schedule_work(&pdev->op_work);
 105	}
 106}
 107
 108static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
 109{
 110	int err = 0;
 111	struct xen_pci_op *active_op = &pdev->sh_info->op;
 112	unsigned long irq_flags;
 113	evtchn_port_t port = pdev->evtchn;
 114	unsigned irq = pdev->irq;
 115	s64 ns, ns_timeout;
 116	struct timeval tv;
 117
 118	spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
 119
 120	memcpy(active_op, op, sizeof(struct xen_pci_op));
 121
 122	/* Go */
 123	wmb();
 124	set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
 125	notify_remote_via_evtchn(port);
 126
 127	/*
 128	 * We set a poll timeout of 3 seconds but give up on return after
 129	 * 2 seconds. It is better to time out too late rather than too early
 130	 * (in the latter case we end up continually re-executing poll() with a
 131	 * timeout in the past). 1s difference gives plenty of slack for error.
 132	 */
 133	do_gettimeofday(&tv);
 134	ns_timeout = timeval_to_ns(&tv) + 2 * (s64)NSEC_PER_SEC;
 135
 136	xen_clear_irq_pending(irq);
 137
 138	while (test_bit(_XEN_PCIF_active,
 139			(unsigned long *)&pdev->sh_info->flags)) {
 140		xen_poll_irq_timeout(irq, jiffies + 3*HZ);
 141		xen_clear_irq_pending(irq);
 142		do_gettimeofday(&tv);
 143		ns = timeval_to_ns(&tv);
 144		if (ns > ns_timeout) {
 145			dev_err(&pdev->xdev->dev,
 146				"pciback not responding!!!\n");
 147			clear_bit(_XEN_PCIF_active,
 148				  (unsigned long *)&pdev->sh_info->flags);
 149			err = XEN_PCI_ERR_dev_not_found;
 150			goto out;
 151		}
 152	}
 153
 154	/*
 155	* We might lose backend service request since we
 156	* reuse same evtchn with pci_conf backend response. So re-schedule
 157	* aer pcifront service.
 158	*/
 159	if (test_bit(_XEN_PCIB_active,
 160			(unsigned long *)&pdev->sh_info->flags)) {
 161		dev_err(&pdev->xdev->dev,
 162			"schedule aer pcifront service\n");
 163		schedule_pcifront_aer_op(pdev);
 164	}
 165
 166	memcpy(op, active_op, sizeof(struct xen_pci_op));
 167
 168	err = op->err;
 169out:
 170	spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
 171	return err;
 172}
 173
 174/* Access to this function is spinlocked in drivers/pci/access.c */
 175static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
 176			     int where, int size, u32 *val)
 177{
 178	int err = 0;
 179	struct xen_pci_op op = {
 180		.cmd    = XEN_PCI_OP_conf_read,
 181		.domain = pci_domain_nr(bus),
 182		.bus    = bus->number,
 183		.devfn  = devfn,
 184		.offset = where,
 185		.size   = size,
 186	};
 187	struct pcifront_sd *sd = bus->sysdata;
 188	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 189
 190	if (verbose_request)
 191		dev_info(&pdev->xdev->dev,
 192			 "read dev=%04x:%02x:%02x.%01x - offset %x size %d\n",
 193			 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
 194			 PCI_FUNC(devfn), where, size);
 195
 196	err = do_pci_op(pdev, &op);
 197
 198	if (likely(!err)) {
 199		if (verbose_request)
 200			dev_info(&pdev->xdev->dev, "read got back value %x\n",
 201				 op.value);
 202
 203		*val = op.value;
 204	} else if (err == -ENODEV) {
 205		/* No device here, pretend that it just returned 0 */
 206		err = 0;
 207		*val = 0;
 208	}
 209
 210	return errno_to_pcibios_err(err);
 211}
 212
 213/* Access to this function is spinlocked in drivers/pci/access.c */
 214static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
 215			      int where, int size, u32 val)
 216{
 217	struct xen_pci_op op = {
 218		.cmd    = XEN_PCI_OP_conf_write,
 219		.domain = pci_domain_nr(bus),
 220		.bus    = bus->number,
 221		.devfn  = devfn,
 222		.offset = where,
 223		.size   = size,
 224		.value  = val,
 225	};
 226	struct pcifront_sd *sd = bus->sysdata;
 227	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 228
 229	if (verbose_request)
 230		dev_info(&pdev->xdev->dev,
 231			 "write dev=%04x:%02x:%02x.%01x - "
 232			 "offset %x size %d val %x\n",
 233			 pci_domain_nr(bus), bus->number,
 234			 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
 235
 236	return errno_to_pcibios_err(do_pci_op(pdev, &op));
 237}
 238
 239struct pci_ops pcifront_bus_ops = {
 240	.read = pcifront_bus_read,
 241	.write = pcifront_bus_write,
 242};
 243
 244#ifdef CONFIG_PCI_MSI
 245static int pci_frontend_enable_msix(struct pci_dev *dev,
 246				    int vector[], int nvec)
 247{
 248	int err;
 249	int i;
 250	struct xen_pci_op op = {
 251		.cmd    = XEN_PCI_OP_enable_msix,
 252		.domain = pci_domain_nr(dev->bus),
 253		.bus = dev->bus->number,
 254		.devfn = dev->devfn,
 255		.value = nvec,
 256	};
 257	struct pcifront_sd *sd = dev->bus->sysdata;
 258	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 259	struct msi_desc *entry;
 260
 261	if (nvec > SH_INFO_MAX_VEC) {
 262		dev_err(&dev->dev, "too much vector for pci frontend: %x."
 263				   " Increase SH_INFO_MAX_VEC.\n", nvec);
 264		return -EINVAL;
 265	}
 266
 267	i = 0;
 268	list_for_each_entry(entry, &dev->msi_list, list) {
 269		op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
 270		/* Vector is useless at this point. */
 271		op.msix_entries[i].vector = -1;
 272		i++;
 273	}
 274
 275	err = do_pci_op(pdev, &op);
 276
 277	if (likely(!err)) {
 278		if (likely(!op.value)) {
 279			/* we get the result */
 280			for (i = 0; i < nvec; i++) {
 281				if (op.msix_entries[i].vector <= 0) {
 282					dev_warn(&dev->dev, "MSI-X entry %d is invalid: %d!\n",
 283						i, op.msix_entries[i].vector);
 284					err = -EINVAL;
 285					vector[i] = -1;
 286					continue;
 287				}
 288				vector[i] = op.msix_entries[i].vector;
 289			}
 290		} else {
 291			printk(KERN_DEBUG "enable msix get value %x\n",
 292				op.value);
 293		}
 294	} else {
 295		dev_err(&dev->dev, "enable msix get err %x\n", err);
 296	}
 297	return err;
 298}
 299
 300static void pci_frontend_disable_msix(struct pci_dev *dev)
 301{
 302	int err;
 303	struct xen_pci_op op = {
 304		.cmd    = XEN_PCI_OP_disable_msix,
 305		.domain = pci_domain_nr(dev->bus),
 306		.bus = dev->bus->number,
 307		.devfn = dev->devfn,
 308	};
 309	struct pcifront_sd *sd = dev->bus->sysdata;
 310	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 311
 312	err = do_pci_op(pdev, &op);
 313
 314	/* What should do for error ? */
 315	if (err)
 316		dev_err(&dev->dev, "pci_disable_msix get err %x\n", err);
 317}
 318
 319static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
 320{
 321	int err;
 322	struct xen_pci_op op = {
 323		.cmd    = XEN_PCI_OP_enable_msi,
 324		.domain = pci_domain_nr(dev->bus),
 325		.bus = dev->bus->number,
 326		.devfn = dev->devfn,
 327	};
 328	struct pcifront_sd *sd = dev->bus->sysdata;
 329	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 330
 331	err = do_pci_op(pdev, &op);
 332	if (likely(!err)) {
 333		vector[0] = op.value;
 334		if (op.value <= 0) {
 335			dev_warn(&dev->dev, "MSI entry is invalid: %d!\n",
 336				op.value);
 337			err = -EINVAL;
 338			vector[0] = -1;
 339		}
 340	} else {
 341		dev_err(&dev->dev, "pci frontend enable msi failed for dev "
 342				    "%x:%x\n", op.bus, op.devfn);
 343		err = -EINVAL;
 344	}
 345	return err;
 346}
 347
 348static void pci_frontend_disable_msi(struct pci_dev *dev)
 349{
 350	int err;
 351	struct xen_pci_op op = {
 352		.cmd    = XEN_PCI_OP_disable_msi,
 353		.domain = pci_domain_nr(dev->bus),
 354		.bus = dev->bus->number,
 355		.devfn = dev->devfn,
 356	};
 357	struct pcifront_sd *sd = dev->bus->sysdata;
 358	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 359
 360	err = do_pci_op(pdev, &op);
 361	if (err == XEN_PCI_ERR_dev_not_found) {
 362		/* XXX No response from backend, what shall we do? */
 363		printk(KERN_DEBUG "get no response from backend for disable MSI\n");
 364		return;
 365	}
 366	if (err)
 367		/* how can pciback notify us fail? */
 368		printk(KERN_DEBUG "get fake response frombackend\n");
 369}
 370
 371static struct xen_pci_frontend_ops pci_frontend_ops = {
 372	.enable_msi = pci_frontend_enable_msi,
 373	.disable_msi = pci_frontend_disable_msi,
 374	.enable_msix = pci_frontend_enable_msix,
 375	.disable_msix = pci_frontend_disable_msix,
 376};
 377
 378static void pci_frontend_registrar(int enable)
 379{
 380	if (enable)
 381		xen_pci_frontend = &pci_frontend_ops;
 382	else
 383		xen_pci_frontend = NULL;
 384};
 385#else
 386static inline void pci_frontend_registrar(int enable) { };
 387#endif /* CONFIG_PCI_MSI */
 388
 389/* Claim resources for the PCI frontend as-is, backend won't allow changes */
 390static int pcifront_claim_resource(struct pci_dev *dev, void *data)
 391{
 392	struct pcifront_device *pdev = data;
 393	int i;
 394	struct resource *r;
 395
 396	for (i = 0; i < PCI_NUM_RESOURCES; i++) {
 397		r = &dev->resource[i];
 398
 399		if (!r->parent && r->start && r->flags) {
 400			dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
 401				pci_name(dev), i);
 402			if (pci_claim_resource(dev, i)) {
 403				dev_err(&pdev->xdev->dev, "Could not claim "
 404					"resource %s/%d! Device offline. Try "
 405					"giving less than 4GB to domain.\n",
 406					pci_name(dev), i);
 407			}
 408		}
 409	}
 410
 411	return 0;
 412}
 413
 414static int __devinit pcifront_scan_bus(struct pcifront_device *pdev,
 415				unsigned int domain, unsigned int bus,
 416				struct pci_bus *b)
 417{
 418	struct pci_dev *d;
 419	unsigned int devfn;
 420
 421	/* Scan the bus for functions and add.
 
 422	 * We omit handling of PCI bridge attachment because pciback prevents
 423	 * bridges from being exported.
 424	 */
 425	for (devfn = 0; devfn < 0x100; devfn++) {
 426		d = pci_get_slot(b, devfn);
 427		if (d) {
 428			/* Device is already known. */
 429			pci_dev_put(d);
 430			continue;
 431		}
 432
 433		d = pci_scan_single_device(b, devfn);
 434		if (d)
 435			dev_info(&pdev->xdev->dev, "New device on "
 436				 "%04x:%02x:%02x.%02x found.\n", domain, bus,
 437				 PCI_SLOT(devfn), PCI_FUNC(devfn));
 438	}
 439
 440	return 0;
 441}
 442
 443static int __devinit pcifront_scan_root(struct pcifront_device *pdev,
 444				 unsigned int domain, unsigned int bus)
 445{
 446	struct pci_bus *b;
 
 447	struct pcifront_sd *sd = NULL;
 448	struct pci_bus_entry *bus_entry = NULL;
 449	int err = 0;
 
 
 
 
 
 450
 451#ifndef CONFIG_PCI_DOMAINS
 452	if (domain != 0) {
 453		dev_err(&pdev->xdev->dev,
 454			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
 455		dev_err(&pdev->xdev->dev,
 456			"Please compile with CONFIG_PCI_DOMAINS\n");
 457		err = -EINVAL;
 458		goto err_out;
 459	}
 460#endif
 461
 462	dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
 463		 domain, bus);
 464
 465	bus_entry = kmalloc(sizeof(*bus_entry), GFP_KERNEL);
 466	sd = kmalloc(sizeof(*sd), GFP_KERNEL);
 467	if (!bus_entry || !sd) {
 468		err = -ENOMEM;
 469		goto err_out;
 470	}
 
 
 
 471	pcifront_init_sd(sd, domain, bus, pdev);
 472
 473	b = pci_scan_bus_parented(&pdev->xdev->dev, bus,
 474				  &pcifront_bus_ops, sd);
 
 
 475	if (!b) {
 476		dev_err(&pdev->xdev->dev,
 477			"Error creating PCI Frontend Bus!\n");
 478		err = -ENOMEM;
 
 
 479		goto err_out;
 480	}
 481
 482	bus_entry->bus = b;
 483
 484	list_add(&bus_entry->list, &pdev->root_buses);
 485
 486	/* pci_scan_bus_parented skips devices which do not have a have
 487	* devfn==0. The pcifront_scan_bus enumerates all devfn. */
 
 
 488	err = pcifront_scan_bus(pdev, domain, bus, b);
 489
 490	/* Claim resources before going "live" with our devices */
 491	pci_walk_bus(b, pcifront_claim_resource, pdev);
 492
 493	/* Create SysFS and notify udev of the devices. Aka: "going live" */
 494	pci_bus_add_devices(b);
 495
 
 496	return err;
 497
 498err_out:
 499	kfree(bus_entry);
 500	kfree(sd);
 501
 502	return err;
 503}
 504
 505static int __devinit pcifront_rescan_root(struct pcifront_device *pdev,
 506				   unsigned int domain, unsigned int bus)
 507{
 508	int err;
 509	struct pci_bus *b;
 510
 511#ifndef CONFIG_PCI_DOMAINS
 512	if (domain != 0) {
 513		dev_err(&pdev->xdev->dev,
 514			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
 515		dev_err(&pdev->xdev->dev,
 516			"Please compile with CONFIG_PCI_DOMAINS\n");
 517		return -EINVAL;
 518	}
 519#endif
 520
 521	dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
 522		 domain, bus);
 523
 524	b = pci_find_bus(domain, bus);
 525	if (!b)
 526		/* If the bus is unknown, create it. */
 527		return pcifront_scan_root(pdev, domain, bus);
 528
 
 
 
 529	err = pcifront_scan_bus(pdev, domain, bus, b);
 530
 531	/* Claim resources before going "live" with our devices */
 532	pci_walk_bus(b, pcifront_claim_resource, pdev);
 533
 534	/* Create SysFS and notify udev of the devices. Aka: "going live" */
 535	pci_bus_add_devices(b);
 536
 537	return err;
 538}
 539
 540static void free_root_bus_devs(struct pci_bus *bus)
 541{
 542	struct pci_dev *dev;
 543
 544	while (!list_empty(&bus->devices)) {
 545		dev = container_of(bus->devices.next, struct pci_dev,
 546				   bus_list);
 547		dev_dbg(&dev->dev, "removing device\n");
 548		pci_remove_bus_device(dev);
 549	}
 550}
 551
 552static void pcifront_free_roots(struct pcifront_device *pdev)
 553{
 554	struct pci_bus_entry *bus_entry, *t;
 555
 556	dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
 557
 
 558	list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
 559		list_del(&bus_entry->list);
 560
 561		free_root_bus_devs(bus_entry->bus);
 562
 563		kfree(bus_entry->bus->sysdata);
 564
 565		device_unregister(bus_entry->bus->bridge);
 566		pci_remove_bus(bus_entry->bus);
 567
 568		kfree(bus_entry);
 569	}
 
 570}
 571
 572static pci_ers_result_t pcifront_common_process(int cmd,
 573						struct pcifront_device *pdev,
 574						pci_channel_state_t state)
 575{
 576	pci_ers_result_t result;
 577	struct pci_driver *pdrv;
 578	int bus = pdev->sh_info->aer_op.bus;
 579	int devfn = pdev->sh_info->aer_op.devfn;
 
 580	struct pci_dev *pcidev;
 581	int flag = 0;
 582
 583	dev_dbg(&pdev->xdev->dev,
 584		"pcifront AER process: cmd %x (bus:%x, devfn%x)",
 585		cmd, bus, devfn);
 586	result = PCI_ERS_RESULT_NONE;
 587
 588	pcidev = pci_get_bus_and_slot(bus, devfn);
 589	if (!pcidev || !pcidev->driver) {
 590		dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
 591		if (pcidev)
 592			pci_dev_put(pcidev);
 593		return result;
 594	}
 595	pdrv = pcidev->driver;
 596
 597	if (get_driver(&pdrv->driver)) {
 598		if (pdrv->err_handler && pdrv->err_handler->error_detected) {
 599			dev_dbg(&pcidev->dev,
 600				"trying to call AER service\n");
 601			if (pcidev) {
 602				flag = 1;
 603				switch (cmd) {
 604				case XEN_PCI_OP_aer_detected:
 605					result = pdrv->err_handler->
 606						 error_detected(pcidev, state);
 607					break;
 608				case XEN_PCI_OP_aer_mmio:
 609					result = pdrv->err_handler->
 610						 mmio_enabled(pcidev);
 611					break;
 612				case XEN_PCI_OP_aer_slotreset:
 613					result = pdrv->err_handler->
 614						 slot_reset(pcidev);
 615					break;
 616				case XEN_PCI_OP_aer_resume:
 617					pdrv->err_handler->resume(pcidev);
 618					break;
 619				default:
 620					dev_err(&pdev->xdev->dev,
 621						"bad request in aer recovery "
 622						"operation!\n");
 623
 624				}
 625			}
 
 
 
 
 
 
 
 
 
 
 
 
 
 626		}
 627		put_driver(&pdrv->driver);
 628	}
 629	if (!flag)
 630		result = PCI_ERS_RESULT_NONE;
 631
 632	return result;
 633}
 634
 635
 636static void pcifront_do_aer(struct work_struct *data)
 637{
 638	struct pcifront_device *pdev =
 639		container_of(data, struct pcifront_device, op_work);
 640	int cmd = pdev->sh_info->aer_op.cmd;
 641	pci_channel_state_t state =
 642		(pci_channel_state_t)pdev->sh_info->aer_op.err;
 643
 644	/*If a pci_conf op is in progress,
 645		we have to wait until it is done before service aer op*/
 
 
 646	dev_dbg(&pdev->xdev->dev,
 647		"pcifront service aer bus %x devfn %x\n",
 648		pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
 649
 650	pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
 651
 652	/* Post the operation to the guest. */
 653	wmb();
 654	clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
 655	notify_remote_via_evtchn(pdev->evtchn);
 656
 657	/*in case of we lost an aer request in four lines time_window*/
 658	smp_mb__before_clear_bit();
 659	clear_bit(_PDEVB_op_active, &pdev->flags);
 660	smp_mb__after_clear_bit();
 661
 662	schedule_pcifront_aer_op(pdev);
 663
 664}
 665
 666static irqreturn_t pcifront_handler_aer(int irq, void *dev)
 667{
 668	struct pcifront_device *pdev = dev;
 
 669	schedule_pcifront_aer_op(pdev);
 670	return IRQ_HANDLED;
 671}
 672static int pcifront_connect(struct pcifront_device *pdev)
 673{
 674	int err = 0;
 675
 676	spin_lock(&pcifront_dev_lock);
 677
 678	if (!pcifront_dev) {
 679		dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
 680		pcifront_dev = pdev;
 681	} else {
 682		dev_err(&pdev->xdev->dev, "PCI frontend already installed!\n");
 683		err = -EEXIST;
 684	}
 685
 686	spin_unlock(&pcifront_dev_lock);
 687
 
 
 
 
 
 688	return err;
 689}
 690
 691static void pcifront_disconnect(struct pcifront_device *pdev)
 692{
 693	spin_lock(&pcifront_dev_lock);
 694
 695	if (pdev == pcifront_dev) {
 696		dev_info(&pdev->xdev->dev,
 697			 "Disconnecting PCI Frontend Buses\n");
 698		pcifront_dev = NULL;
 699	}
 700
 701	spin_unlock(&pcifront_dev_lock);
 702}
 703static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
 704{
 705	struct pcifront_device *pdev;
 706
 707	pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
 708	if (pdev == NULL)
 709		goto out;
 710
 711	pdev->sh_info =
 712	    (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
 713	if (pdev->sh_info == NULL) {
 714		kfree(pdev);
 715		pdev = NULL;
 716		goto out;
 717	}
 718	pdev->sh_info->flags = 0;
 719
 720	/*Flag for registering PV AER handler*/
 721	set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
 722
 723	dev_set_drvdata(&xdev->dev, pdev);
 724	pdev->xdev = xdev;
 725
 726	INIT_LIST_HEAD(&pdev->root_buses);
 727
 728	spin_lock_init(&pdev->sh_info_lock);
 729
 730	pdev->evtchn = INVALID_EVTCHN;
 731	pdev->gnt_ref = INVALID_GRANT_REF;
 732	pdev->irq = -1;
 733
 734	INIT_WORK(&pdev->op_work, pcifront_do_aer);
 735
 736	dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
 737		pdev, pdev->sh_info);
 738out:
 739	return pdev;
 740}
 741
 742static void free_pdev(struct pcifront_device *pdev)
 743{
 744	dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
 745
 746	pcifront_free_roots(pdev);
 747
 748	cancel_work_sync(&pdev->op_work);
 749
 750	if (pdev->irq >= 0)
 751		unbind_from_irqhandler(pdev->irq, pdev);
 752
 753	if (pdev->evtchn != INVALID_EVTCHN)
 754		xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
 755
 756	if (pdev->gnt_ref != INVALID_GRANT_REF)
 757		gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
 758					  (unsigned long)pdev->sh_info);
 759	else
 760		free_page((unsigned long)pdev->sh_info);
 761
 762	dev_set_drvdata(&pdev->xdev->dev, NULL);
 763
 764	kfree(pdev);
 765}
 766
 767static int pcifront_publish_info(struct pcifront_device *pdev)
 768{
 769	int err = 0;
 770	struct xenbus_transaction trans;
 771
 772	err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info));
 773	if (err < 0)
 774		goto out;
 775
 776	pdev->gnt_ref = err;
 777
 778	err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
 779	if (err)
 780		goto out;
 781
 782	err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
 783		0, "pcifront", pdev);
 784
 785	if (err < 0)
 786		return err;
 787
 788	pdev->irq = err;
 789
 790do_publish:
 791	err = xenbus_transaction_start(&trans);
 792	if (err) {
 793		xenbus_dev_fatal(pdev->xdev, err,
 794				 "Error writing configuration for backend "
 795				 "(start transaction)");
 796		goto out;
 797	}
 798
 799	err = xenbus_printf(trans, pdev->xdev->nodename,
 800			    "pci-op-ref", "%u", pdev->gnt_ref);
 801	if (!err)
 802		err = xenbus_printf(trans, pdev->xdev->nodename,
 803				    "event-channel", "%u", pdev->evtchn);
 804	if (!err)
 805		err = xenbus_printf(trans, pdev->xdev->nodename,
 806				    "magic", XEN_PCI_MAGIC);
 807
 808	if (err) {
 809		xenbus_transaction_end(trans, 1);
 810		xenbus_dev_fatal(pdev->xdev, err,
 811				 "Error writing configuration for backend");
 812		goto out;
 813	} else {
 814		err = xenbus_transaction_end(trans, 0);
 815		if (err == -EAGAIN)
 816			goto do_publish;
 817		else if (err) {
 818			xenbus_dev_fatal(pdev->xdev, err,
 819					 "Error completing transaction "
 820					 "for backend");
 821			goto out;
 822		}
 823	}
 824
 825	xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
 826
 827	dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
 828
 829out:
 830	return err;
 831}
 832
 833static int __devinit pcifront_try_connect(struct pcifront_device *pdev)
 834{
 835	int err = -EFAULT;
 836	int i, num_roots, len;
 837	char str[64];
 838	unsigned int domain, bus;
 839
 840
 841	/* Only connect once */
 842	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
 843	    XenbusStateInitialised)
 844		goto out;
 845
 846	err = pcifront_connect(pdev);
 847	if (err) {
 848		xenbus_dev_fatal(pdev->xdev, err,
 849				 "Error connecting PCI Frontend");
 850		goto out;
 851	}
 852
 853	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
 854			   "root_num", "%d", &num_roots);
 855	if (err == -ENOENT) {
 856		xenbus_dev_error(pdev->xdev, err,
 857				 "No PCI Roots found, trying 0000:00");
 858		err = pcifront_scan_root(pdev, 0, 0);
 
 
 
 
 
 859		num_roots = 0;
 860	} else if (err != 1) {
 861		if (err == 0)
 862			err = -EINVAL;
 863		xenbus_dev_fatal(pdev->xdev, err,
 864				 "Error reading number of PCI roots");
 865		goto out;
 866	}
 867
 868	for (i = 0; i < num_roots; i++) {
 869		len = snprintf(str, sizeof(str), "root-%d", i);
 870		if (unlikely(len >= (sizeof(str) - 1))) {
 871			err = -ENOMEM;
 872			goto out;
 873		}
 874
 875		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
 876				   "%x:%x", &domain, &bus);
 877		if (err != 2) {
 878			if (err >= 0)
 879				err = -EINVAL;
 880			xenbus_dev_fatal(pdev->xdev, err,
 881					 "Error reading PCI root %d", i);
 882			goto out;
 883		}
 884
 885		err = pcifront_scan_root(pdev, domain, bus);
 886		if (err) {
 887			xenbus_dev_fatal(pdev->xdev, err,
 888					 "Error scanning PCI root %04x:%02x",
 889					 domain, bus);
 890			goto out;
 891		}
 892	}
 893
 894	err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
 
 895
 896out:
 897	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 898}
 899
 900static int pcifront_try_disconnect(struct pcifront_device *pdev)
 901{
 902	int err = 0;
 903	enum xenbus_state prev_state;
 904
 905
 906	prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
 907
 908	if (prev_state >= XenbusStateClosing)
 909		goto out;
 910
 911	if (prev_state == XenbusStateConnected) {
 912		pcifront_free_roots(pdev);
 913		pcifront_disconnect(pdev);
 914	}
 915
 916	err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
 917
 918out:
 919
 920	return err;
 921}
 922
 923static int __devinit pcifront_attach_devices(struct pcifront_device *pdev)
 924{
 925	int err = -EFAULT;
 926	int i, num_roots, len;
 927	unsigned int domain, bus;
 928	char str[64];
 929
 930	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
 931	    XenbusStateReconfiguring)
 932		goto out;
 933
 934	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
 935			   "root_num", "%d", &num_roots);
 936	if (err == -ENOENT) {
 937		xenbus_dev_error(pdev->xdev, err,
 938				 "No PCI Roots found, trying 0000:00");
 939		err = pcifront_rescan_root(pdev, 0, 0);
 940		num_roots = 0;
 941	} else if (err != 1) {
 942		if (err == 0)
 943			err = -EINVAL;
 944		xenbus_dev_fatal(pdev->xdev, err,
 945				 "Error reading number of PCI roots");
 946		goto out;
 947	}
 948
 949	for (i = 0; i < num_roots; i++) {
 950		len = snprintf(str, sizeof(str), "root-%d", i);
 951		if (unlikely(len >= (sizeof(str) - 1))) {
 952			err = -ENOMEM;
 953			goto out;
 954		}
 955
 956		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
 957				   "%x:%x", &domain, &bus);
 958		if (err != 2) {
 959			if (err >= 0)
 960				err = -EINVAL;
 961			xenbus_dev_fatal(pdev->xdev, err,
 962					 "Error reading PCI root %d", i);
 963			goto out;
 964		}
 965
 966		err = pcifront_rescan_root(pdev, domain, bus);
 967		if (err) {
 968			xenbus_dev_fatal(pdev->xdev, err,
 969					 "Error scanning PCI root %04x:%02x",
 970					 domain, bus);
 971			goto out;
 972		}
 973	}
 974
 975	xenbus_switch_state(pdev->xdev, XenbusStateConnected);
 976
 977out:
 978	return err;
 979}
 980
 981static int pcifront_detach_devices(struct pcifront_device *pdev)
 982{
 983	int err = 0;
 984	int i, num_devs;
 
 985	unsigned int domain, bus, slot, func;
 986	struct pci_bus *pci_bus;
 987	struct pci_dev *pci_dev;
 988	char str[64];
 989
 990	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
 991	    XenbusStateConnected)
 
 
 
 
 
 
 
 
 
 
 
 992		goto out;
 
 993
 994	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
 995			   &num_devs);
 996	if (err != 1) {
 997		if (err >= 0)
 998			err = -EINVAL;
 999		xenbus_dev_fatal(pdev->xdev, err,
1000				 "Error reading number of PCI devices");
1001		goto out;
1002	}
1003
1004	/* Find devices being detached and remove them. */
1005	for (i = 0; i < num_devs; i++) {
1006		int l, state;
 
1007		l = snprintf(str, sizeof(str), "state-%d", i);
1008		if (unlikely(l >= (sizeof(str) - 1))) {
1009			err = -ENOMEM;
1010			goto out;
1011		}
1012		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d",
1013				   &state);
1014		if (err != 1)
1015			state = XenbusStateUnknown;
1016
1017		if (state != XenbusStateClosing)
1018			continue;
1019
1020		/* Remove device. */
1021		l = snprintf(str, sizeof(str), "vdev-%d", i);
1022		if (unlikely(l >= (sizeof(str) - 1))) {
1023			err = -ENOMEM;
1024			goto out;
1025		}
1026		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1027				   "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1028		if (err != 4) {
1029			if (err >= 0)
1030				err = -EINVAL;
1031			xenbus_dev_fatal(pdev->xdev, err,
1032					 "Error reading PCI device %d", i);
1033			goto out;
1034		}
1035
1036		pci_bus = pci_find_bus(domain, bus);
1037		if (!pci_bus) {
1038			dev_dbg(&pdev->xdev->dev, "Cannot get bus %04x:%02x\n",
1039				domain, bus);
1040			continue;
1041		}
1042		pci_dev = pci_get_slot(pci_bus, PCI_DEVFN(slot, func));
1043		if (!pci_dev) {
1044			dev_dbg(&pdev->xdev->dev,
1045				"Cannot get PCI device %04x:%02x:%02x.%02x\n",
1046				domain, bus, slot, func);
1047			continue;
1048		}
1049		pci_remove_bus_device(pci_dev);
 
1050		pci_dev_put(pci_dev);
 
1051
1052		dev_dbg(&pdev->xdev->dev,
1053			"PCI device %04x:%02x:%02x.%02x removed.\n",
1054			domain, bus, slot, func);
1055	}
1056
 
1057	err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1058
1059out:
1060	return err;
1061}
1062
1063static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev,
1064						  enum xenbus_state be_state)
1065{
1066	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1067
1068	switch (be_state) {
1069	case XenbusStateUnknown:
1070	case XenbusStateInitialising:
1071	case XenbusStateInitWait:
1072	case XenbusStateInitialised:
1073	case XenbusStateClosed:
1074		break;
1075
1076	case XenbusStateConnected:
1077		pcifront_try_connect(pdev);
1078		break;
1079
 
 
 
 
1080	case XenbusStateClosing:
1081		dev_warn(&xdev->dev, "backend going away!\n");
1082		pcifront_try_disconnect(pdev);
1083		break;
1084
1085	case XenbusStateReconfiguring:
1086		pcifront_detach_devices(pdev);
1087		break;
1088
1089	case XenbusStateReconfigured:
1090		pcifront_attach_devices(pdev);
1091		break;
1092	}
1093}
1094
1095static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1096				 const struct xenbus_device_id *id)
1097{
1098	int err = 0;
1099	struct pcifront_device *pdev = alloc_pdev(xdev);
1100
1101	if (pdev == NULL) {
1102		err = -ENOMEM;
1103		xenbus_dev_fatal(xdev, err,
1104				 "Error allocating pcifront_device struct");
1105		goto out;
1106	}
1107
1108	err = pcifront_publish_info(pdev);
1109	if (err)
1110		free_pdev(pdev);
1111
1112out:
1113	return err;
1114}
1115
1116static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1117{
1118	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
 
1119	if (pdev)
1120		free_pdev(pdev);
1121
1122	return 0;
1123}
1124
1125static const struct xenbus_device_id xenpci_ids[] = {
1126	{"pci"},
1127	{""},
1128};
1129
1130static struct xenbus_driver xenbus_pcifront_driver = {
1131	.name			= "pcifront",
1132	.owner			= THIS_MODULE,
1133	.ids			= xenpci_ids,
1134	.probe			= pcifront_xenbus_probe,
1135	.remove			= pcifront_xenbus_remove,
1136	.otherend_changed	= pcifront_backend_changed,
1137};
1138
1139static int __init pcifront_init(void)
1140{
1141	if (!xen_pv_domain() || xen_initial_domain())
1142		return -ENODEV;
1143
 
 
 
1144	pci_frontend_registrar(1 /* enable */);
1145
1146	return xenbus_register_frontend(&xenbus_pcifront_driver);
1147}
1148
1149static void __exit pcifront_cleanup(void)
1150{
1151	xenbus_unregister_driver(&xenbus_pcifront_driver);
1152	pci_frontend_registrar(0 /* disable */);
1153}
1154module_init(pcifront_init);
1155module_exit(pcifront_cleanup);
1156
1157MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1158MODULE_LICENSE("GPL");
1159MODULE_ALIAS("xen:pci");
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Xen PCI Frontend
   4 *
   5 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
   6 */
   7#include <linux/module.h>
   8#include <linux/init.h>
   9#include <linux/mm.h>
  10#include <xen/xenbus.h>
  11#include <xen/events.h>
  12#include <xen/grant_table.h>
  13#include <xen/page.h>
  14#include <linux/spinlock.h>
  15#include <linux/pci.h>
  16#include <linux/msi.h>
  17#include <xen/interface/io/pciif.h>
  18#include <asm/xen/pci.h>
  19#include <linux/interrupt.h>
  20#include <linux/atomic.h>
  21#include <linux/workqueue.h>
  22#include <linux/bitops.h>
  23#include <linux/time.h>
  24#include <linux/ktime.h>
  25#include <linux/swiotlb.h>
  26#include <xen/platform_pci.h>
  27
  28#include <asm/xen/swiotlb-xen.h>
  29
 
  30#define INVALID_EVTCHN    (-1)
  31
  32struct pci_bus_entry {
  33	struct list_head list;
  34	struct pci_bus *bus;
  35};
  36
  37#define _PDEVB_op_active		(0)
  38#define PDEVB_op_active			(1 << (_PDEVB_op_active))
  39
  40struct pcifront_device {
  41	struct xenbus_device *xdev;
  42	struct list_head root_buses;
  43
  44	int evtchn;
  45	grant_ref_t gnt_ref;
  46
  47	int irq;
  48
  49	/* Lock this when doing any operations in sh_info */
  50	spinlock_t sh_info_lock;
  51	struct xen_pci_sharedinfo *sh_info;
  52	struct work_struct op_work;
  53	unsigned long flags;
  54
  55};
  56
  57struct pcifront_sd {
  58	struct pci_sysdata sd;
  59	struct pcifront_device *pdev;
  60};
  61
  62static inline struct pcifront_device *
  63pcifront_get_pdev(struct pcifront_sd *sd)
  64{
  65	return sd->pdev;
  66}
  67
  68static inline void pcifront_init_sd(struct pcifront_sd *sd,
  69				    unsigned int domain, unsigned int bus,
  70				    struct pcifront_device *pdev)
  71{
  72	/* Because we do not expose that information via XenBus. */
  73	sd->sd.node = first_online_node;
  74	sd->sd.domain = domain;
  75	sd->pdev = pdev;
  76}
  77
  78static DEFINE_SPINLOCK(pcifront_dev_lock);
  79static struct pcifront_device *pcifront_dev;
  80
 
 
 
  81static int errno_to_pcibios_err(int errno)
  82{
  83	switch (errno) {
  84	case XEN_PCI_ERR_success:
  85		return PCIBIOS_SUCCESSFUL;
  86
  87	case XEN_PCI_ERR_dev_not_found:
  88		return PCIBIOS_DEVICE_NOT_FOUND;
  89
  90	case XEN_PCI_ERR_invalid_offset:
  91	case XEN_PCI_ERR_op_failed:
  92		return PCIBIOS_BAD_REGISTER_NUMBER;
  93
  94	case XEN_PCI_ERR_not_implemented:
  95		return PCIBIOS_FUNC_NOT_SUPPORTED;
  96
  97	case XEN_PCI_ERR_access_denied:
  98		return PCIBIOS_SET_FAILED;
  99	}
 100	return errno;
 101}
 102
 103static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
 104{
 105	if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
 106		&& !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
 107		dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
 108		schedule_work(&pdev->op_work);
 109	}
 110}
 111
 112static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
 113{
 114	int err = 0;
 115	struct xen_pci_op *active_op = &pdev->sh_info->op;
 116	unsigned long irq_flags;
 117	evtchn_port_t port = pdev->evtchn;
 118	unsigned int irq = pdev->irq;
 119	s64 ns, ns_timeout;
 
 120
 121	spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
 122
 123	memcpy(active_op, op, sizeof(struct xen_pci_op));
 124
 125	/* Go */
 126	wmb();
 127	set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
 128	notify_remote_via_evtchn(port);
 129
 130	/*
 131	 * We set a poll timeout of 3 seconds but give up on return after
 132	 * 2 seconds. It is better to time out too late rather than too early
 133	 * (in the latter case we end up continually re-executing poll() with a
 134	 * timeout in the past). 1s difference gives plenty of slack for error.
 135	 */
 136	ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC;
 
 137
 138	xen_clear_irq_pending(irq);
 139
 140	while (test_bit(_XEN_PCIF_active,
 141			(unsigned long *)&pdev->sh_info->flags)) {
 142		xen_poll_irq_timeout(irq, jiffies + 3*HZ);
 143		xen_clear_irq_pending(irq);
 144		ns = ktime_get_ns();
 
 145		if (ns > ns_timeout) {
 146			dev_err(&pdev->xdev->dev,
 147				"pciback not responding!!!\n");
 148			clear_bit(_XEN_PCIF_active,
 149				  (unsigned long *)&pdev->sh_info->flags);
 150			err = XEN_PCI_ERR_dev_not_found;
 151			goto out;
 152		}
 153	}
 154
 155	/*
 156	 * We might lose backend service request since we
 157	 * reuse same evtchn with pci_conf backend response. So re-schedule
 158	 * aer pcifront service.
 159	 */
 160	if (test_bit(_XEN_PCIB_active,
 161			(unsigned long *)&pdev->sh_info->flags)) {
 162		dev_err(&pdev->xdev->dev,
 163			"schedule aer pcifront service\n");
 164		schedule_pcifront_aer_op(pdev);
 165	}
 166
 167	memcpy(op, active_op, sizeof(struct xen_pci_op));
 168
 169	err = op->err;
 170out:
 171	spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
 172	return err;
 173}
 174
 175/* Access to this function is spinlocked in drivers/pci/access.c */
 176static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
 177			     int where, int size, u32 *val)
 178{
 179	int err = 0;
 180	struct xen_pci_op op = {
 181		.cmd    = XEN_PCI_OP_conf_read,
 182		.domain = pci_domain_nr(bus),
 183		.bus    = bus->number,
 184		.devfn  = devfn,
 185		.offset = where,
 186		.size   = size,
 187	};
 188	struct pcifront_sd *sd = bus->sysdata;
 189	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 190
 191	dev_dbg(&pdev->xdev->dev,
 192		"read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
 193		pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
 194		PCI_FUNC(devfn), where, size);
 
 195
 196	err = do_pci_op(pdev, &op);
 197
 198	if (likely(!err)) {
 199		dev_dbg(&pdev->xdev->dev, "read got back value %x\n",
 200			op.value);
 
 201
 202		*val = op.value;
 203	} else if (err == -ENODEV) {
 204		/* No device here, pretend that it just returned 0 */
 205		err = 0;
 206		*val = 0;
 207	}
 208
 209	return errno_to_pcibios_err(err);
 210}
 211
 212/* Access to this function is spinlocked in drivers/pci/access.c */
 213static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
 214			      int where, int size, u32 val)
 215{
 216	struct xen_pci_op op = {
 217		.cmd    = XEN_PCI_OP_conf_write,
 218		.domain = pci_domain_nr(bus),
 219		.bus    = bus->number,
 220		.devfn  = devfn,
 221		.offset = where,
 222		.size   = size,
 223		.value  = val,
 224	};
 225	struct pcifront_sd *sd = bus->sysdata;
 226	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 227
 228	dev_dbg(&pdev->xdev->dev,
 229		"write dev=%04x:%02x:%02x.%d - offset %x size %d val %x\n",
 230		pci_domain_nr(bus), bus->number,
 231		PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
 
 
 232
 233	return errno_to_pcibios_err(do_pci_op(pdev, &op));
 234}
 235
 236static struct pci_ops pcifront_bus_ops = {
 237	.read = pcifront_bus_read,
 238	.write = pcifront_bus_write,
 239};
 240
 241#ifdef CONFIG_PCI_MSI
 242static int pci_frontend_enable_msix(struct pci_dev *dev,
 243				    int vector[], int nvec)
 244{
 245	int err;
 246	int i;
 247	struct xen_pci_op op = {
 248		.cmd    = XEN_PCI_OP_enable_msix,
 249		.domain = pci_domain_nr(dev->bus),
 250		.bus = dev->bus->number,
 251		.devfn = dev->devfn,
 252		.value = nvec,
 253	};
 254	struct pcifront_sd *sd = dev->bus->sysdata;
 255	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 256	struct msi_desc *entry;
 257
 258	if (nvec > SH_INFO_MAX_VEC) {
 259		pci_err(dev, "too many vectors (0x%x) for PCI frontend:"
 260				   " Increase SH_INFO_MAX_VEC\n", nvec);
 261		return -EINVAL;
 262	}
 263
 264	i = 0;
 265	msi_for_each_desc(entry, &dev->dev, MSI_DESC_NOTASSOCIATED) {
 266		op.msix_entries[i].entry = entry->msi_index;
 267		/* Vector is useless at this point. */
 268		op.msix_entries[i].vector = -1;
 269		i++;
 270	}
 271
 272	err = do_pci_op(pdev, &op);
 273
 274	if (likely(!err)) {
 275		if (likely(!op.value)) {
 276			/* we get the result */
 277			for (i = 0; i < nvec; i++) {
 278				if (op.msix_entries[i].vector <= 0) {
 279					pci_warn(dev, "MSI-X entry %d is invalid: %d!\n",
 280						i, op.msix_entries[i].vector);
 281					err = -EINVAL;
 282					vector[i] = -1;
 283					continue;
 284				}
 285				vector[i] = op.msix_entries[i].vector;
 286			}
 287		} else {
 288			pr_info("enable msix get value %x\n", op.value);
 289			err = op.value;
 290		}
 291	} else {
 292		pci_err(dev, "enable msix get err %x\n", err);
 293	}
 294	return err;
 295}
 296
 297static void pci_frontend_disable_msix(struct pci_dev *dev)
 298{
 299	int err;
 300	struct xen_pci_op op = {
 301		.cmd    = XEN_PCI_OP_disable_msix,
 302		.domain = pci_domain_nr(dev->bus),
 303		.bus = dev->bus->number,
 304		.devfn = dev->devfn,
 305	};
 306	struct pcifront_sd *sd = dev->bus->sysdata;
 307	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 308
 309	err = do_pci_op(pdev, &op);
 310
 311	/* What should do for error ? */
 312	if (err)
 313		pci_err(dev, "pci_disable_msix get err %x\n", err);
 314}
 315
 316static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
 317{
 318	int err;
 319	struct xen_pci_op op = {
 320		.cmd    = XEN_PCI_OP_enable_msi,
 321		.domain = pci_domain_nr(dev->bus),
 322		.bus = dev->bus->number,
 323		.devfn = dev->devfn,
 324	};
 325	struct pcifront_sd *sd = dev->bus->sysdata;
 326	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 327
 328	err = do_pci_op(pdev, &op);
 329	if (likely(!err)) {
 330		vector[0] = op.value;
 331		if (op.value <= 0) {
 332			pci_warn(dev, "MSI entry is invalid: %d!\n",
 333				op.value);
 334			err = -EINVAL;
 335			vector[0] = -1;
 336		}
 337	} else {
 338		pci_err(dev, "pci frontend enable msi failed for dev "
 339				    "%x:%x\n", op.bus, op.devfn);
 340		err = -EINVAL;
 341	}
 342	return err;
 343}
 344
 345static void pci_frontend_disable_msi(struct pci_dev *dev)
 346{
 347	int err;
 348	struct xen_pci_op op = {
 349		.cmd    = XEN_PCI_OP_disable_msi,
 350		.domain = pci_domain_nr(dev->bus),
 351		.bus = dev->bus->number,
 352		.devfn = dev->devfn,
 353	};
 354	struct pcifront_sd *sd = dev->bus->sysdata;
 355	struct pcifront_device *pdev = pcifront_get_pdev(sd);
 356
 357	err = do_pci_op(pdev, &op);
 358	if (err == XEN_PCI_ERR_dev_not_found) {
 359		/* XXX No response from backend, what shall we do? */
 360		pr_info("get no response from backend for disable MSI\n");
 361		return;
 362	}
 363	if (err)
 364		/* how can pciback notify us fail? */
 365		pr_info("get fake response from backend\n");
 366}
 367
 368static struct xen_pci_frontend_ops pci_frontend_ops = {
 369	.enable_msi = pci_frontend_enable_msi,
 370	.disable_msi = pci_frontend_disable_msi,
 371	.enable_msix = pci_frontend_enable_msix,
 372	.disable_msix = pci_frontend_disable_msix,
 373};
 374
 375static void pci_frontend_registrar(int enable)
 376{
 377	if (enable)
 378		xen_pci_frontend = &pci_frontend_ops;
 379	else
 380		xen_pci_frontend = NULL;
 381};
 382#else
 383static inline void pci_frontend_registrar(int enable) { };
 384#endif /* CONFIG_PCI_MSI */
 385
 386/* Claim resources for the PCI frontend as-is, backend won't allow changes */
 387static int pcifront_claim_resource(struct pci_dev *dev, void *data)
 388{
 389	struct pcifront_device *pdev = data;
 390	int i;
 391	struct resource *r;
 392
 393	for (i = 0; i < PCI_NUM_RESOURCES; i++) {
 394		r = &dev->resource[i];
 395
 396		if (!r->parent && r->start && r->flags) {
 397			dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
 398				pci_name(dev), i);
 399			if (pci_claim_resource(dev, i)) {
 400				dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! "
 401					"Device offline. Try using e820_host=1 in the guest config.\n",
 
 402					pci_name(dev), i);
 403			}
 404		}
 405	}
 406
 407	return 0;
 408}
 409
 410static int pcifront_scan_bus(struct pcifront_device *pdev,
 411				unsigned int domain, unsigned int bus,
 412				struct pci_bus *b)
 413{
 414	struct pci_dev *d;
 415	unsigned int devfn;
 416
 417	/*
 418	 * Scan the bus for functions and add.
 419	 * We omit handling of PCI bridge attachment because pciback prevents
 420	 * bridges from being exported.
 421	 */
 422	for (devfn = 0; devfn < 0x100; devfn++) {
 423		d = pci_get_slot(b, devfn);
 424		if (d) {
 425			/* Device is already known. */
 426			pci_dev_put(d);
 427			continue;
 428		}
 429
 430		d = pci_scan_single_device(b, devfn);
 431		if (d)
 432			dev_info(&pdev->xdev->dev, "New device on "
 433				 "%04x:%02x:%02x.%d found.\n", domain, bus,
 434				 PCI_SLOT(devfn), PCI_FUNC(devfn));
 435	}
 436
 437	return 0;
 438}
 439
 440static int pcifront_scan_root(struct pcifront_device *pdev,
 441				 unsigned int domain, unsigned int bus)
 442{
 443	struct pci_bus *b;
 444	LIST_HEAD(resources);
 445	struct pcifront_sd *sd = NULL;
 446	struct pci_bus_entry *bus_entry = NULL;
 447	int err = 0;
 448	static struct resource busn_res = {
 449		.start = 0,
 450		.end = 255,
 451		.flags = IORESOURCE_BUS,
 452	};
 453
 454#ifndef CONFIG_PCI_DOMAINS
 455	if (domain != 0) {
 456		dev_err(&pdev->xdev->dev,
 457			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
 458		dev_err(&pdev->xdev->dev,
 459			"Please compile with CONFIG_PCI_DOMAINS\n");
 460		err = -EINVAL;
 461		goto err_out;
 462	}
 463#endif
 464
 465	dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
 466		 domain, bus);
 467
 468	bus_entry = kzalloc(sizeof(*bus_entry), GFP_KERNEL);
 469	sd = kzalloc(sizeof(*sd), GFP_KERNEL);
 470	if (!bus_entry || !sd) {
 471		err = -ENOMEM;
 472		goto err_out;
 473	}
 474	pci_add_resource(&resources, &ioport_resource);
 475	pci_add_resource(&resources, &iomem_resource);
 476	pci_add_resource(&resources, &busn_res);
 477	pcifront_init_sd(sd, domain, bus, pdev);
 478
 479	pci_lock_rescan_remove();
 480
 481	b = pci_scan_root_bus(&pdev->xdev->dev, bus,
 482				  &pcifront_bus_ops, sd, &resources);
 483	if (!b) {
 484		dev_err(&pdev->xdev->dev,
 485			"Error creating PCI Frontend Bus!\n");
 486		err = -ENOMEM;
 487		pci_unlock_rescan_remove();
 488		pci_free_resource_list(&resources);
 489		goto err_out;
 490	}
 491
 492	bus_entry->bus = b;
 493
 494	list_add(&bus_entry->list, &pdev->root_buses);
 495
 496	/*
 497	 * pci_scan_root_bus skips devices which do not have a
 498	 * devfn==0. The pcifront_scan_bus enumerates all devfn.
 499	 */
 500	err = pcifront_scan_bus(pdev, domain, bus, b);
 501
 502	/* Claim resources before going "live" with our devices */
 503	pci_walk_bus(b, pcifront_claim_resource, pdev);
 504
 505	/* Create SysFS and notify udev of the devices. Aka: "going live" */
 506	pci_bus_add_devices(b);
 507
 508	pci_unlock_rescan_remove();
 509	return err;
 510
 511err_out:
 512	kfree(bus_entry);
 513	kfree(sd);
 514
 515	return err;
 516}
 517
 518static int pcifront_rescan_root(struct pcifront_device *pdev,
 519				   unsigned int domain, unsigned int bus)
 520{
 521	int err;
 522	struct pci_bus *b;
 523
 
 
 
 
 
 
 
 
 
 
 
 
 
 524	b = pci_find_bus(domain, bus);
 525	if (!b)
 526		/* If the bus is unknown, create it. */
 527		return pcifront_scan_root(pdev, domain, bus);
 528
 529	dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
 530		 domain, bus);
 531
 532	err = pcifront_scan_bus(pdev, domain, bus, b);
 533
 534	/* Claim resources before going "live" with our devices */
 535	pci_walk_bus(b, pcifront_claim_resource, pdev);
 536
 537	/* Create SysFS and notify udev of the devices. Aka: "going live" */
 538	pci_bus_add_devices(b);
 539
 540	return err;
 541}
 542
 543static void free_root_bus_devs(struct pci_bus *bus)
 544{
 545	struct pci_dev *dev;
 546
 547	while (!list_empty(&bus->devices)) {
 548		dev = container_of(bus->devices.next, struct pci_dev,
 549				   bus_list);
 550		pci_dbg(dev, "removing device\n");
 551		pci_stop_and_remove_bus_device(dev);
 552	}
 553}
 554
 555static void pcifront_free_roots(struct pcifront_device *pdev)
 556{
 557	struct pci_bus_entry *bus_entry, *t;
 558
 559	dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
 560
 561	pci_lock_rescan_remove();
 562	list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
 563		list_del(&bus_entry->list);
 564
 565		free_root_bus_devs(bus_entry->bus);
 566
 567		kfree(bus_entry->bus->sysdata);
 568
 569		device_unregister(bus_entry->bus->bridge);
 570		pci_remove_bus(bus_entry->bus);
 571
 572		kfree(bus_entry);
 573	}
 574	pci_unlock_rescan_remove();
 575}
 576
 577static pci_ers_result_t pcifront_common_process(int cmd,
 578						struct pcifront_device *pdev,
 579						pci_channel_state_t state)
 580{
 
 581	struct pci_driver *pdrv;
 582	int bus = pdev->sh_info->aer_op.bus;
 583	int devfn = pdev->sh_info->aer_op.devfn;
 584	int domain = pdev->sh_info->aer_op.domain;
 585	struct pci_dev *pcidev;
 
 586
 587	dev_dbg(&pdev->xdev->dev,
 588		"pcifront AER process: cmd %x (bus:%x, devfn%x)",
 589		cmd, bus, devfn);
 
 590
 591	pcidev = pci_get_domain_bus_and_slot(domain, bus, devfn);
 592	if (!pcidev || !pcidev->dev.driver) {
 593		dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
 594		pci_dev_put(pcidev);
 595		return PCI_ERS_RESULT_NONE;
 596	}
 597	pdrv = to_pci_driver(pcidev->dev.driver);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 598
 599	if (pdrv->err_handler && pdrv->err_handler->error_detected) {
 600		pci_dbg(pcidev, "trying to call AER service\n");
 601		switch (cmd) {
 602		case XEN_PCI_OP_aer_detected:
 603			return pdrv->err_handler->error_detected(pcidev, state);
 604		case XEN_PCI_OP_aer_mmio:
 605			return pdrv->err_handler->mmio_enabled(pcidev);
 606		case XEN_PCI_OP_aer_slotreset:
 607			return pdrv->err_handler->slot_reset(pcidev);
 608		case XEN_PCI_OP_aer_resume:
 609			pdrv->err_handler->resume(pcidev);
 610			return PCI_ERS_RESULT_NONE;
 611		default:
 612			dev_err(&pdev->xdev->dev,
 613				"bad request in aer recovery operation!\n");
 614		}
 
 615	}
 
 
 616
 617	return PCI_ERS_RESULT_NONE;
 618}
 619
 620
 621static void pcifront_do_aer(struct work_struct *data)
 622{
 623	struct pcifront_device *pdev =
 624		container_of(data, struct pcifront_device, op_work);
 625	int cmd = pdev->sh_info->aer_op.cmd;
 626	pci_channel_state_t state =
 627		(pci_channel_state_t)pdev->sh_info->aer_op.err;
 628
 629	/*
 630	 * If a pci_conf op is in progress, we have to wait until it is done
 631	 * before service aer op
 632	 */
 633	dev_dbg(&pdev->xdev->dev,
 634		"pcifront service aer bus %x devfn %x\n",
 635		pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
 636
 637	pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
 638
 639	/* Post the operation to the guest. */
 640	wmb();
 641	clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
 642	notify_remote_via_evtchn(pdev->evtchn);
 643
 644	/*in case of we lost an aer request in four lines time_window*/
 645	smp_mb__before_atomic();
 646	clear_bit(_PDEVB_op_active, &pdev->flags);
 647	smp_mb__after_atomic();
 648
 649	schedule_pcifront_aer_op(pdev);
 650
 651}
 652
 653static irqreturn_t pcifront_handler_aer(int irq, void *dev)
 654{
 655	struct pcifront_device *pdev = dev;
 656
 657	schedule_pcifront_aer_op(pdev);
 658	return IRQ_HANDLED;
 659}
 660static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
 661{
 662	int err = 0;
 663
 664	spin_lock(&pcifront_dev_lock);
 665
 666	if (!pcifront_dev) {
 667		dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
 668		pcifront_dev = pdev;
 669	} else
 
 670		err = -EEXIST;
 
 671
 672	spin_unlock(&pcifront_dev_lock);
 673
 674	if (!err && !is_swiotlb_active(&pdev->xdev->dev)) {
 675		err = pci_xen_swiotlb_init_late();
 676		if (err)
 677			dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
 678	}
 679	return err;
 680}
 681
 682static void pcifront_disconnect(struct pcifront_device *pdev)
 683{
 684	spin_lock(&pcifront_dev_lock);
 685
 686	if (pdev == pcifront_dev) {
 687		dev_info(&pdev->xdev->dev,
 688			 "Disconnecting PCI Frontend Buses\n");
 689		pcifront_dev = NULL;
 690	}
 691
 692	spin_unlock(&pcifront_dev_lock);
 693}
 694static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
 695{
 696	struct pcifront_device *pdev;
 697
 698	pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
 699	if (pdev == NULL)
 700		goto out;
 701
 702	if (xenbus_setup_ring(xdev, GFP_KERNEL, (void **)&pdev->sh_info, 1,
 703			      &pdev->gnt_ref)) {
 
 704		kfree(pdev);
 705		pdev = NULL;
 706		goto out;
 707	}
 708	pdev->sh_info->flags = 0;
 709
 710	/*Flag for registering PV AER handler*/
 711	set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
 712
 713	dev_set_drvdata(&xdev->dev, pdev);
 714	pdev->xdev = xdev;
 715
 716	INIT_LIST_HEAD(&pdev->root_buses);
 717
 718	spin_lock_init(&pdev->sh_info_lock);
 719
 720	pdev->evtchn = INVALID_EVTCHN;
 
 721	pdev->irq = -1;
 722
 723	INIT_WORK(&pdev->op_work, pcifront_do_aer);
 724
 725	dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
 726		pdev, pdev->sh_info);
 727out:
 728	return pdev;
 729}
 730
 731static void free_pdev(struct pcifront_device *pdev)
 732{
 733	dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
 734
 735	pcifront_free_roots(pdev);
 736
 737	cancel_work_sync(&pdev->op_work);
 738
 739	if (pdev->irq >= 0)
 740		unbind_from_irqhandler(pdev->irq, pdev);
 741
 742	if (pdev->evtchn != INVALID_EVTCHN)
 743		xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
 744
 745	xenbus_teardown_ring((void **)&pdev->sh_info, 1, &pdev->gnt_ref);
 
 
 
 
 746
 747	dev_set_drvdata(&pdev->xdev->dev, NULL);
 748
 749	kfree(pdev);
 750}
 751
 752static int pcifront_publish_info(struct pcifront_device *pdev)
 753{
 754	int err = 0;
 755	struct xenbus_transaction trans;
 756
 
 
 
 
 
 
 757	err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
 758	if (err)
 759		goto out;
 760
 761	err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
 762		0, "pcifront", pdev);
 763
 764	if (err < 0)
 765		return err;
 766
 767	pdev->irq = err;
 768
 769do_publish:
 770	err = xenbus_transaction_start(&trans);
 771	if (err) {
 772		xenbus_dev_fatal(pdev->xdev, err,
 773				 "Error writing configuration for backend "
 774				 "(start transaction)");
 775		goto out;
 776	}
 777
 778	err = xenbus_printf(trans, pdev->xdev->nodename,
 779			    "pci-op-ref", "%u", pdev->gnt_ref);
 780	if (!err)
 781		err = xenbus_printf(trans, pdev->xdev->nodename,
 782				    "event-channel", "%u", pdev->evtchn);
 783	if (!err)
 784		err = xenbus_printf(trans, pdev->xdev->nodename,
 785				    "magic", XEN_PCI_MAGIC);
 786
 787	if (err) {
 788		xenbus_transaction_end(trans, 1);
 789		xenbus_dev_fatal(pdev->xdev, err,
 790				 "Error writing configuration for backend");
 791		goto out;
 792	} else {
 793		err = xenbus_transaction_end(trans, 0);
 794		if (err == -EAGAIN)
 795			goto do_publish;
 796		else if (err) {
 797			xenbus_dev_fatal(pdev->xdev, err,
 798					 "Error completing transaction "
 799					 "for backend");
 800			goto out;
 801		}
 802	}
 803
 804	xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
 805
 806	dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
 807
 808out:
 809	return err;
 810}
 811
 812static void pcifront_connect(struct pcifront_device *pdev)
 813{
 814	int err;
 815	int i, num_roots, len;
 816	char str[64];
 817	unsigned int domain, bus;
 818
 
 
 
 
 
 
 
 
 
 
 
 
 
 819	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
 820			   "root_num", "%d", &num_roots);
 821	if (err == -ENOENT) {
 822		xenbus_dev_error(pdev->xdev, err,
 823				 "No PCI Roots found, trying 0000:00");
 824		err = pcifront_rescan_root(pdev, 0, 0);
 825		if (err) {
 826			xenbus_dev_fatal(pdev->xdev, err,
 827					 "Error scanning PCI root 0000:00");
 828			return;
 829		}
 830		num_roots = 0;
 831	} else if (err != 1) {
 832		xenbus_dev_fatal(pdev->xdev, err >= 0 ? -EINVAL : err,
 
 
 833				 "Error reading number of PCI roots");
 834		return;
 835	}
 836
 837	for (i = 0; i < num_roots; i++) {
 838		len = snprintf(str, sizeof(str), "root-%d", i);
 839		if (unlikely(len >= (sizeof(str) - 1)))
 840			return;
 
 
 841
 842		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
 843				   "%x:%x", &domain, &bus);
 844		if (err != 2) {
 845			xenbus_dev_fatal(pdev->xdev, err >= 0 ? -EINVAL : err,
 
 
 846					 "Error reading PCI root %d", i);
 847			return;
 848		}
 849
 850		err = pcifront_rescan_root(pdev, domain, bus);
 851		if (err) {
 852			xenbus_dev_fatal(pdev->xdev, err,
 853					 "Error scanning PCI root %04x:%02x",
 854					 domain, bus);
 855			return;
 856		}
 857	}
 858
 859	xenbus_switch_state(pdev->xdev, XenbusStateConnected);
 860}
 861
 862static void pcifront_try_connect(struct pcifront_device *pdev)
 863{
 864	int err;
 865
 866	/* Only connect once */
 867	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
 868	    XenbusStateInitialised)
 869		return;
 870
 871	err = pcifront_connect_and_init_dma(pdev);
 872	if (err && err != -EEXIST) {
 873		xenbus_dev_fatal(pdev->xdev, err,
 874				 "Error setting up PCI Frontend");
 875		return;
 876	}
 877
 878	pcifront_connect(pdev);
 879}
 880
 881static int pcifront_try_disconnect(struct pcifront_device *pdev)
 882{
 883	int err = 0;
 884	enum xenbus_state prev_state;
 885
 886
 887	prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
 888
 889	if (prev_state >= XenbusStateClosing)
 890		goto out;
 891
 892	if (prev_state == XenbusStateConnected) {
 893		pcifront_free_roots(pdev);
 894		pcifront_disconnect(pdev);
 895	}
 896
 897	err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
 898
 899out:
 900
 901	return err;
 902}
 903
 904static void pcifront_attach_devices(struct pcifront_device *pdev)
 905{
 906	if (xenbus_read_driver_state(pdev->xdev->nodename) ==
 
 
 
 
 
 907	    XenbusStateReconfiguring)
 908		pcifront_connect(pdev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 909}
 910
 911static int pcifront_detach_devices(struct pcifront_device *pdev)
 912{
 913	int err = 0;
 914	int i, num_devs;
 915	enum xenbus_state state;
 916	unsigned int domain, bus, slot, func;
 
 917	struct pci_dev *pci_dev;
 918	char str[64];
 919
 920	state = xenbus_read_driver_state(pdev->xdev->nodename);
 921	if (state == XenbusStateInitialised) {
 922		dev_dbg(&pdev->xdev->dev, "Handle skipped connect.\n");
 923		/* We missed Connected and need to initialize. */
 924		err = pcifront_connect_and_init_dma(pdev);
 925		if (err && err != -EEXIST) {
 926			xenbus_dev_fatal(pdev->xdev, err,
 927					 "Error setting up PCI Frontend");
 928			goto out;
 929		}
 930
 931		goto out_switch_state;
 932	} else if (state != XenbusStateConnected) {
 933		goto out;
 934	}
 935
 936	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
 937			   &num_devs);
 938	if (err != 1) {
 939		if (err >= 0)
 940			err = -EINVAL;
 941		xenbus_dev_fatal(pdev->xdev, err,
 942				 "Error reading number of PCI devices");
 943		goto out;
 944	}
 945
 946	/* Find devices being detached and remove them. */
 947	for (i = 0; i < num_devs; i++) {
 948		int l, state;
 949
 950		l = snprintf(str, sizeof(str), "state-%d", i);
 951		if (unlikely(l >= (sizeof(str) - 1))) {
 952			err = -ENOMEM;
 953			goto out;
 954		}
 955		state = xenbus_read_unsigned(pdev->xdev->otherend, str,
 956					     XenbusStateUnknown);
 
 
 957
 958		if (state != XenbusStateClosing)
 959			continue;
 960
 961		/* Remove device. */
 962		l = snprintf(str, sizeof(str), "vdev-%d", i);
 963		if (unlikely(l >= (sizeof(str) - 1))) {
 964			err = -ENOMEM;
 965			goto out;
 966		}
 967		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
 968				   "%x:%x:%x.%x", &domain, &bus, &slot, &func);
 969		if (err != 4) {
 970			if (err >= 0)
 971				err = -EINVAL;
 972			xenbus_dev_fatal(pdev->xdev, err,
 973					 "Error reading PCI device %d", i);
 974			goto out;
 975		}
 976
 977		pci_dev = pci_get_domain_bus_and_slot(domain, bus,
 978				PCI_DEVFN(slot, func));
 
 
 
 
 
 979		if (!pci_dev) {
 980			dev_dbg(&pdev->xdev->dev,
 981				"Cannot get PCI device %04x:%02x:%02x.%d\n",
 982				domain, bus, slot, func);
 983			continue;
 984		}
 985		pci_lock_rescan_remove();
 986		pci_stop_and_remove_bus_device(pci_dev);
 987		pci_dev_put(pci_dev);
 988		pci_unlock_rescan_remove();
 989
 990		dev_dbg(&pdev->xdev->dev,
 991			"PCI device %04x:%02x:%02x.%d removed.\n",
 992			domain, bus, slot, func);
 993	}
 994
 995 out_switch_state:
 996	err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
 997
 998out:
 999	return err;
1000}
1001
1002static void pcifront_backend_changed(struct xenbus_device *xdev,
1003						  enum xenbus_state be_state)
1004{
1005	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1006
1007	switch (be_state) {
1008	case XenbusStateUnknown:
1009	case XenbusStateInitialising:
1010	case XenbusStateInitWait:
1011	case XenbusStateInitialised:
 
1012		break;
1013
1014	case XenbusStateConnected:
1015		pcifront_try_connect(pdev);
1016		break;
1017
1018	case XenbusStateClosed:
1019		if (xdev->state == XenbusStateClosed)
1020			break;
1021		fallthrough;	/* Missed the backend's CLOSING state */
1022	case XenbusStateClosing:
1023		dev_warn(&xdev->dev, "backend going away!\n");
1024		pcifront_try_disconnect(pdev);
1025		break;
1026
1027	case XenbusStateReconfiguring:
1028		pcifront_detach_devices(pdev);
1029		break;
1030
1031	case XenbusStateReconfigured:
1032		pcifront_attach_devices(pdev);
1033		break;
1034	}
1035}
1036
1037static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1038				 const struct xenbus_device_id *id)
1039{
1040	int err = 0;
1041	struct pcifront_device *pdev = alloc_pdev(xdev);
1042
1043	if (pdev == NULL) {
1044		err = -ENOMEM;
1045		xenbus_dev_fatal(xdev, err,
1046				 "Error allocating pcifront_device struct");
1047		goto out;
1048	}
1049
1050	err = pcifront_publish_info(pdev);
1051	if (err)
1052		free_pdev(pdev);
1053
1054out:
1055	return err;
1056}
1057
1058static void pcifront_xenbus_remove(struct xenbus_device *xdev)
1059{
1060	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1061
1062	if (pdev)
1063		free_pdev(pdev);
 
 
1064}
1065
1066static const struct xenbus_device_id xenpci_ids[] = {
1067	{"pci"},
1068	{""},
1069};
1070
1071static struct xenbus_driver xenpci_driver = {
1072	.name			= "pcifront",
 
1073	.ids			= xenpci_ids,
1074	.probe			= pcifront_xenbus_probe,
1075	.remove			= pcifront_xenbus_remove,
1076	.otherend_changed	= pcifront_backend_changed,
1077};
1078
1079static int __init pcifront_init(void)
1080{
1081	if (!xen_pv_domain() || xen_initial_domain())
1082		return -ENODEV;
1083
1084	if (!xen_has_pv_devices())
1085		return -ENODEV;
1086
1087	pci_frontend_registrar(1 /* enable */);
1088
1089	return xenbus_register_frontend(&xenpci_driver);
1090}
1091
1092static void __exit pcifront_cleanup(void)
1093{
1094	xenbus_unregister_driver(&xenpci_driver);
1095	pci_frontend_registrar(0 /* disable */);
1096}
1097module_init(pcifront_init);
1098module_exit(pcifront_cleanup);
1099
1100MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1101MODULE_LICENSE("GPL");
1102MODULE_ALIAS("xen:pci");