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