xref: /linux/drivers/pci/controller/vmd.c (revision 19d0070a2792181f79df01277fe00b83b9f7eda7)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Volume Management Device driver
4  * Copyright (c) 2015, Intel Corporation.
5  */
6 
7 #include <linux/device.h>
8 #include <linux/interrupt.h>
9 #include <linux/irq.h>
10 #include <linux/kernel.h>
11 #include <linux/module.h>
12 #include <linux/msi.h>
13 #include <linux/pci.h>
14 #include <linux/srcu.h>
15 #include <linux/rculist.h>
16 #include <linux/rcupdate.h>
17 
18 #include <asm/irqdomain.h>
19 #include <asm/device.h>
20 #include <asm/msi.h>
21 #include <asm/msidef.h>
22 
23 #define VMD_CFGBAR	0
24 #define VMD_MEMBAR1	2
25 #define VMD_MEMBAR2	4
26 
27 #define PCI_REG_VMCAP		0x40
28 #define BUS_RESTRICT_CAP(vmcap)	(vmcap & 0x1)
29 #define PCI_REG_VMCONFIG	0x44
30 #define BUS_RESTRICT_CFG(vmcfg)	((vmcfg >> 8) & 0x3)
31 #define PCI_REG_VMLOCK		0x70
32 #define MB2_SHADOW_EN(vmlock)	(vmlock & 0x2)
33 
34 #define MB2_SHADOW_OFFSET	0x2000
35 #define MB2_SHADOW_SIZE		16
36 
37 enum vmd_features {
38 	/*
39 	 * Device may contain registers which hint the physical location of the
40 	 * membars, in order to allow proper address translation during
41 	 * resource assignment to enable guest virtualization
42 	 */
43 	VMD_FEAT_HAS_MEMBAR_SHADOW	= (1 << 0),
44 
45 	/*
46 	 * Device may provide root port configuration information which limits
47 	 * bus numbering
48 	 */
49 	VMD_FEAT_HAS_BUS_RESTRICTIONS	= (1 << 1),
50 };
51 
52 /*
53  * Lock for manipulating VMD IRQ lists.
54  */
55 static DEFINE_RAW_SPINLOCK(list_lock);
56 
57 /**
58  * struct vmd_irq - private data to map driver IRQ to the VMD shared vector
59  * @node:	list item for parent traversal.
60  * @irq:	back pointer to parent.
61  * @enabled:	true if driver enabled IRQ
62  * @virq:	the virtual IRQ value provided to the requesting driver.
63  *
64  * Every MSI/MSI-X IRQ requested for a device in a VMD domain will be mapped to
65  * a VMD IRQ using this structure.
66  */
67 struct vmd_irq {
68 	struct list_head	node;
69 	struct vmd_irq_list	*irq;
70 	bool			enabled;
71 	unsigned int		virq;
72 };
73 
74 /**
75  * struct vmd_irq_list - list of driver requested IRQs mapping to a VMD vector
76  * @irq_list:	the list of irq's the VMD one demuxes to.
77  * @srcu:	SRCU struct for local synchronization.
78  * @count:	number of child IRQs assigned to this vector; used to track
79  *		sharing.
80  */
81 struct vmd_irq_list {
82 	struct list_head	irq_list;
83 	struct srcu_struct	srcu;
84 	unsigned int		count;
85 };
86 
87 struct vmd_dev {
88 	struct pci_dev		*dev;
89 
90 	spinlock_t		cfg_lock;
91 	char __iomem		*cfgbar;
92 
93 	int msix_count;
94 	struct vmd_irq_list	*irqs;
95 
96 	struct pci_sysdata	sysdata;
97 	struct resource		resources[3];
98 	struct irq_domain	*irq_domain;
99 	struct pci_bus		*bus;
100 	u8			busn_start;
101 };
102 
103 static inline struct vmd_dev *vmd_from_bus(struct pci_bus *bus)
104 {
105 	return container_of(bus->sysdata, struct vmd_dev, sysdata);
106 }
107 
108 static inline unsigned int index_from_irqs(struct vmd_dev *vmd,
109 					   struct vmd_irq_list *irqs)
110 {
111 	return irqs - vmd->irqs;
112 }
113 
114 /*
115  * Drivers managing a device in a VMD domain allocate their own IRQs as before,
116  * but the MSI entry for the hardware it's driving will be programmed with a
117  * destination ID for the VMD MSI-X table.  The VMD muxes interrupts in its
118  * domain into one of its own, and the VMD driver de-muxes these for the
119  * handlers sharing that VMD IRQ.  The vmd irq_domain provides the operations
120  * and irq_chip to set this up.
121  */
122 static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
123 {
124 	struct vmd_irq *vmdirq = data->chip_data;
125 	struct vmd_irq_list *irq = vmdirq->irq;
126 	struct vmd_dev *vmd = irq_data_get_irq_handler_data(data);
127 
128 	msg->address_hi = MSI_ADDR_BASE_HI;
129 	msg->address_lo = MSI_ADDR_BASE_LO |
130 			  MSI_ADDR_DEST_ID(index_from_irqs(vmd, irq));
131 	msg->data = 0;
132 }
133 
134 /*
135  * We rely on MSI_FLAG_USE_DEF_CHIP_OPS to set the IRQ mask/unmask ops.
136  */
137 static void vmd_irq_enable(struct irq_data *data)
138 {
139 	struct vmd_irq *vmdirq = data->chip_data;
140 	unsigned long flags;
141 
142 	raw_spin_lock_irqsave(&list_lock, flags);
143 	WARN_ON(vmdirq->enabled);
144 	list_add_tail_rcu(&vmdirq->node, &vmdirq->irq->irq_list);
145 	vmdirq->enabled = true;
146 	raw_spin_unlock_irqrestore(&list_lock, flags);
147 
148 	data->chip->irq_unmask(data);
149 }
150 
151 static void vmd_irq_disable(struct irq_data *data)
152 {
153 	struct vmd_irq *vmdirq = data->chip_data;
154 	unsigned long flags;
155 
156 	data->chip->irq_mask(data);
157 
158 	raw_spin_lock_irqsave(&list_lock, flags);
159 	if (vmdirq->enabled) {
160 		list_del_rcu(&vmdirq->node);
161 		vmdirq->enabled = false;
162 	}
163 	raw_spin_unlock_irqrestore(&list_lock, flags);
164 }
165 
166 /*
167  * XXX: Stubbed until we develop acceptable way to not create conflicts with
168  * other devices sharing the same vector.
169  */
170 static int vmd_irq_set_affinity(struct irq_data *data,
171 				const struct cpumask *dest, bool force)
172 {
173 	return -EINVAL;
174 }
175 
176 static struct irq_chip vmd_msi_controller = {
177 	.name			= "VMD-MSI",
178 	.irq_enable		= vmd_irq_enable,
179 	.irq_disable		= vmd_irq_disable,
180 	.irq_compose_msi_msg	= vmd_compose_msi_msg,
181 	.irq_set_affinity	= vmd_irq_set_affinity,
182 };
183 
184 static irq_hw_number_t vmd_get_hwirq(struct msi_domain_info *info,
185 				     msi_alloc_info_t *arg)
186 {
187 	return 0;
188 }
189 
190 /*
191  * XXX: We can be even smarter selecting the best IRQ once we solve the
192  * affinity problem.
193  */
194 static struct vmd_irq_list *vmd_next_irq(struct vmd_dev *vmd, struct msi_desc *desc)
195 {
196 	int i, best = 1;
197 	unsigned long flags;
198 
199 	if (vmd->msix_count == 1)
200 		return &vmd->irqs[0];
201 
202 	/*
203 	 * White list for fast-interrupt handlers. All others will share the
204 	 * "slow" interrupt vector.
205 	 */
206 	switch (msi_desc_to_pci_dev(desc)->class) {
207 	case PCI_CLASS_STORAGE_EXPRESS:
208 		break;
209 	default:
210 		return &vmd->irqs[0];
211 	}
212 
213 	raw_spin_lock_irqsave(&list_lock, flags);
214 	for (i = 1; i < vmd->msix_count; i++)
215 		if (vmd->irqs[i].count < vmd->irqs[best].count)
216 			best = i;
217 	vmd->irqs[best].count++;
218 	raw_spin_unlock_irqrestore(&list_lock, flags);
219 
220 	return &vmd->irqs[best];
221 }
222 
223 static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
224 			unsigned int virq, irq_hw_number_t hwirq,
225 			msi_alloc_info_t *arg)
226 {
227 	struct msi_desc *desc = arg->desc;
228 	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
229 	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
230 	unsigned int index, vector;
231 
232 	if (!vmdirq)
233 		return -ENOMEM;
234 
235 	INIT_LIST_HEAD(&vmdirq->node);
236 	vmdirq->irq = vmd_next_irq(vmd, desc);
237 	vmdirq->virq = virq;
238 	index = index_from_irqs(vmd, vmdirq->irq);
239 	vector = pci_irq_vector(vmd->dev, index);
240 
241 	irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
242 			    handle_untracked_irq, vmd, NULL);
243 	return 0;
244 }
245 
246 static void vmd_msi_free(struct irq_domain *domain,
247 			struct msi_domain_info *info, unsigned int virq)
248 {
249 	struct vmd_irq *vmdirq = irq_get_chip_data(virq);
250 	unsigned long flags;
251 
252 	synchronize_srcu(&vmdirq->irq->srcu);
253 
254 	/* XXX: Potential optimization to rebalance */
255 	raw_spin_lock_irqsave(&list_lock, flags);
256 	vmdirq->irq->count--;
257 	raw_spin_unlock_irqrestore(&list_lock, flags);
258 
259 	kfree(vmdirq);
260 }
261 
262 static int vmd_msi_prepare(struct irq_domain *domain, struct device *dev,
263 			   int nvec, msi_alloc_info_t *arg)
264 {
265 	struct pci_dev *pdev = to_pci_dev(dev);
266 	struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
267 
268 	if (nvec > vmd->msix_count)
269 		return vmd->msix_count;
270 
271 	memset(arg, 0, sizeof(*arg));
272 	return 0;
273 }
274 
275 static void vmd_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
276 {
277 	arg->desc = desc;
278 }
279 
280 static struct msi_domain_ops vmd_msi_domain_ops = {
281 	.get_hwirq	= vmd_get_hwirq,
282 	.msi_init	= vmd_msi_init,
283 	.msi_free	= vmd_msi_free,
284 	.msi_prepare	= vmd_msi_prepare,
285 	.set_desc	= vmd_set_desc,
286 };
287 
288 static struct msi_domain_info vmd_msi_domain_info = {
289 	.flags		= MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
290 			  MSI_FLAG_PCI_MSIX,
291 	.ops		= &vmd_msi_domain_ops,
292 	.chip		= &vmd_msi_controller,
293 };
294 
295 static char __iomem *vmd_cfg_addr(struct vmd_dev *vmd, struct pci_bus *bus,
296 				  unsigned int devfn, int reg, int len)
297 {
298 	char __iomem *addr = vmd->cfgbar +
299 			     ((bus->number - vmd->busn_start) << 20) +
300 			     (devfn << 12) + reg;
301 
302 	if ((addr - vmd->cfgbar) + len >=
303 	    resource_size(&vmd->dev->resource[VMD_CFGBAR]))
304 		return NULL;
305 
306 	return addr;
307 }
308 
309 /*
310  * CPU may deadlock if config space is not serialized on some versions of this
311  * hardware, so all config space access is done under a spinlock.
312  */
313 static int vmd_pci_read(struct pci_bus *bus, unsigned int devfn, int reg,
314 			int len, u32 *value)
315 {
316 	struct vmd_dev *vmd = vmd_from_bus(bus);
317 	char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
318 	unsigned long flags;
319 	int ret = 0;
320 
321 	if (!addr)
322 		return -EFAULT;
323 
324 	spin_lock_irqsave(&vmd->cfg_lock, flags);
325 	switch (len) {
326 	case 1:
327 		*value = readb(addr);
328 		break;
329 	case 2:
330 		*value = readw(addr);
331 		break;
332 	case 4:
333 		*value = readl(addr);
334 		break;
335 	default:
336 		ret = -EINVAL;
337 		break;
338 	}
339 	spin_unlock_irqrestore(&vmd->cfg_lock, flags);
340 	return ret;
341 }
342 
343 /*
344  * VMD h/w converts non-posted config writes to posted memory writes. The
345  * read-back in this function forces the completion so it returns only after
346  * the config space was written, as expected.
347  */
348 static int vmd_pci_write(struct pci_bus *bus, unsigned int devfn, int reg,
349 			 int len, u32 value)
350 {
351 	struct vmd_dev *vmd = vmd_from_bus(bus);
352 	char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
353 	unsigned long flags;
354 	int ret = 0;
355 
356 	if (!addr)
357 		return -EFAULT;
358 
359 	spin_lock_irqsave(&vmd->cfg_lock, flags);
360 	switch (len) {
361 	case 1:
362 		writeb(value, addr);
363 		readb(addr);
364 		break;
365 	case 2:
366 		writew(value, addr);
367 		readw(addr);
368 		break;
369 	case 4:
370 		writel(value, addr);
371 		readl(addr);
372 		break;
373 	default:
374 		ret = -EINVAL;
375 		break;
376 	}
377 	spin_unlock_irqrestore(&vmd->cfg_lock, flags);
378 	return ret;
379 }
380 
381 static struct pci_ops vmd_ops = {
382 	.read		= vmd_pci_read,
383 	.write		= vmd_pci_write,
384 };
385 
386 static void vmd_attach_resources(struct vmd_dev *vmd)
387 {
388 	vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
389 	vmd->dev->resource[VMD_MEMBAR2].child = &vmd->resources[2];
390 }
391 
392 static void vmd_detach_resources(struct vmd_dev *vmd)
393 {
394 	vmd->dev->resource[VMD_MEMBAR1].child = NULL;
395 	vmd->dev->resource[VMD_MEMBAR2].child = NULL;
396 }
397 
398 /*
399  * VMD domains start at 0x10000 to not clash with ACPI _SEG domains.
400  * Per ACPI r6.0, sec 6.5.6,  _SEG returns an integer, of which the lower
401  * 16 bits are the PCI Segment Group (domain) number.  Other bits are
402  * currently reserved.
403  */
404 static int vmd_find_free_domain(void)
405 {
406 	int domain = 0xffff;
407 	struct pci_bus *bus = NULL;
408 
409 	while ((bus = pci_find_next_bus(bus)) != NULL)
410 		domain = max_t(int, domain, pci_domain_nr(bus));
411 	return domain + 1;
412 }
413 
414 static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
415 {
416 	struct pci_sysdata *sd = &vmd->sysdata;
417 	struct fwnode_handle *fn;
418 	struct resource *res;
419 	u32 upper_bits;
420 	unsigned long flags;
421 	LIST_HEAD(resources);
422 	resource_size_t offset[2] = {0};
423 	resource_size_t membar2_offset = 0x2000;
424 	struct pci_bus *child;
425 
426 	/*
427 	 * Shadow registers may exist in certain VMD device ids which allow
428 	 * guests to correctly assign host physical addresses to the root ports
429 	 * and child devices. These registers will either return the host value
430 	 * or 0, depending on an enable bit in the VMD device.
431 	 */
432 	if (features & VMD_FEAT_HAS_MEMBAR_SHADOW) {
433 		u32 vmlock;
434 		int ret;
435 
436 		membar2_offset = MB2_SHADOW_OFFSET + MB2_SHADOW_SIZE;
437 		ret = pci_read_config_dword(vmd->dev, PCI_REG_VMLOCK, &vmlock);
438 		if (ret || vmlock == ~0)
439 			return -ENODEV;
440 
441 		if (MB2_SHADOW_EN(vmlock)) {
442 			void __iomem *membar2;
443 
444 			membar2 = pci_iomap(vmd->dev, VMD_MEMBAR2, 0);
445 			if (!membar2)
446 				return -ENOMEM;
447 			offset[0] = vmd->dev->resource[VMD_MEMBAR1].start -
448 					(readq(membar2 + MB2_SHADOW_OFFSET) &
449 					 PCI_BASE_ADDRESS_MEM_MASK);
450 			offset[1] = vmd->dev->resource[VMD_MEMBAR2].start -
451 					(readq(membar2 + MB2_SHADOW_OFFSET + 8) &
452 					 PCI_BASE_ADDRESS_MEM_MASK);
453 			pci_iounmap(vmd->dev, membar2);
454 		}
455 	}
456 
457 	/*
458 	 * Certain VMD devices may have a root port configuration option which
459 	 * limits the bus range to between 0-127, 128-255, or 224-255
460 	 */
461 	if (features & VMD_FEAT_HAS_BUS_RESTRICTIONS) {
462 		u16 reg16;
463 
464 		pci_read_config_word(vmd->dev, PCI_REG_VMCAP, &reg16);
465 		if (BUS_RESTRICT_CAP(reg16)) {
466 			pci_read_config_word(vmd->dev, PCI_REG_VMCONFIG,
467 					     &reg16);
468 
469 			switch (BUS_RESTRICT_CFG(reg16)) {
470 			case 1:
471 				vmd->busn_start = 128;
472 				break;
473 			case 2:
474 				vmd->busn_start = 224;
475 				break;
476 			case 3:
477 				pci_err(vmd->dev, "Unknown Bus Offset Setting\n");
478 				return -ENODEV;
479 			default:
480 				break;
481 			}
482 		}
483 	}
484 
485 	res = &vmd->dev->resource[VMD_CFGBAR];
486 	vmd->resources[0] = (struct resource) {
487 		.name  = "VMD CFGBAR",
488 		.start = vmd->busn_start,
489 		.end   = vmd->busn_start + (resource_size(res) >> 20) - 1,
490 		.flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED,
491 	};
492 
493 	/*
494 	 * If the window is below 4GB, clear IORESOURCE_MEM_64 so we can
495 	 * put 32-bit resources in the window.
496 	 *
497 	 * There's no hardware reason why a 64-bit window *couldn't*
498 	 * contain a 32-bit resource, but pbus_size_mem() computes the
499 	 * bridge window size assuming a 64-bit window will contain no
500 	 * 32-bit resources.  __pci_assign_resource() enforces that
501 	 * artificial restriction to make sure everything will fit.
502 	 *
503 	 * The only way we could use a 64-bit non-prefetchable MEMBAR is
504 	 * if its address is <4GB so that we can convert it to a 32-bit
505 	 * resource.  To be visible to the host OS, all VMD endpoints must
506 	 * be initially configured by platform BIOS, which includes setting
507 	 * up these resources.  We can assume the device is configured
508 	 * according to the platform needs.
509 	 */
510 	res = &vmd->dev->resource[VMD_MEMBAR1];
511 	upper_bits = upper_32_bits(res->end);
512 	flags = res->flags & ~IORESOURCE_SIZEALIGN;
513 	if (!upper_bits)
514 		flags &= ~IORESOURCE_MEM_64;
515 	vmd->resources[1] = (struct resource) {
516 		.name  = "VMD MEMBAR1",
517 		.start = res->start,
518 		.end   = res->end,
519 		.flags = flags,
520 		.parent = res,
521 	};
522 
523 	res = &vmd->dev->resource[VMD_MEMBAR2];
524 	upper_bits = upper_32_bits(res->end);
525 	flags = res->flags & ~IORESOURCE_SIZEALIGN;
526 	if (!upper_bits)
527 		flags &= ~IORESOURCE_MEM_64;
528 	vmd->resources[2] = (struct resource) {
529 		.name  = "VMD MEMBAR2",
530 		.start = res->start + membar2_offset,
531 		.end   = res->end,
532 		.flags = flags,
533 		.parent = res,
534 	};
535 
536 	sd->vmd_dev = vmd->dev;
537 	sd->domain = vmd_find_free_domain();
538 	if (sd->domain < 0)
539 		return sd->domain;
540 
541 	sd->node = pcibus_to_node(vmd->dev->bus);
542 
543 	fn = irq_domain_alloc_named_id_fwnode("VMD-MSI", vmd->sysdata.domain);
544 	if (!fn)
545 		return -ENODEV;
546 
547 	vmd->irq_domain = pci_msi_create_irq_domain(fn, &vmd_msi_domain_info,
548 						    x86_vector_domain);
549 	if (!vmd->irq_domain) {
550 		irq_domain_free_fwnode(fn);
551 		return -ENODEV;
552 	}
553 
554 	pci_add_resource(&resources, &vmd->resources[0]);
555 	pci_add_resource_offset(&resources, &vmd->resources[1], offset[0]);
556 	pci_add_resource_offset(&resources, &vmd->resources[2], offset[1]);
557 
558 	vmd->bus = pci_create_root_bus(&vmd->dev->dev, vmd->busn_start,
559 				       &vmd_ops, sd, &resources);
560 	if (!vmd->bus) {
561 		pci_free_resource_list(&resources);
562 		irq_domain_remove(vmd->irq_domain);
563 		irq_domain_free_fwnode(fn);
564 		return -ENODEV;
565 	}
566 
567 	vmd_attach_resources(vmd);
568 	dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
569 
570 	pci_scan_child_bus(vmd->bus);
571 	pci_assign_unassigned_bus_resources(vmd->bus);
572 
573 	/*
574 	 * VMD root buses are virtual and don't return true on pci_is_pcie()
575 	 * and will fail pcie_bus_configure_settings() early. It can instead be
576 	 * run on each of the real root ports.
577 	 */
578 	list_for_each_entry(child, &vmd->bus->children, node)
579 		pcie_bus_configure_settings(child);
580 
581 	pci_bus_add_devices(vmd->bus);
582 
583 	WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
584 			       "domain"), "Can't create symlink to domain\n");
585 	return 0;
586 }
587 
588 static irqreturn_t vmd_irq(int irq, void *data)
589 {
590 	struct vmd_irq_list *irqs = data;
591 	struct vmd_irq *vmdirq;
592 	int idx;
593 
594 	idx = srcu_read_lock(&irqs->srcu);
595 	list_for_each_entry_rcu(vmdirq, &irqs->irq_list, node)
596 		generic_handle_irq(vmdirq->virq);
597 	srcu_read_unlock(&irqs->srcu, idx);
598 
599 	return IRQ_HANDLED;
600 }
601 
602 static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
603 {
604 	struct vmd_dev *vmd;
605 	int i, err;
606 
607 	if (resource_size(&dev->resource[VMD_CFGBAR]) < (1 << 20))
608 		return -ENOMEM;
609 
610 	vmd = devm_kzalloc(&dev->dev, sizeof(*vmd), GFP_KERNEL);
611 	if (!vmd)
612 		return -ENOMEM;
613 
614 	vmd->dev = dev;
615 	err = pcim_enable_device(dev);
616 	if (err < 0)
617 		return err;
618 
619 	vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
620 	if (!vmd->cfgbar)
621 		return -ENOMEM;
622 
623 	pci_set_master(dev);
624 	if (dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(64)) &&
625 	    dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(32)))
626 		return -ENODEV;
627 
628 	vmd->msix_count = pci_msix_vec_count(dev);
629 	if (vmd->msix_count < 0)
630 		return -ENODEV;
631 
632 	vmd->msix_count = pci_alloc_irq_vectors(dev, 1, vmd->msix_count,
633 					PCI_IRQ_MSIX);
634 	if (vmd->msix_count < 0)
635 		return vmd->msix_count;
636 
637 	vmd->irqs = devm_kcalloc(&dev->dev, vmd->msix_count, sizeof(*vmd->irqs),
638 				 GFP_KERNEL);
639 	if (!vmd->irqs)
640 		return -ENOMEM;
641 
642 	for (i = 0; i < vmd->msix_count; i++) {
643 		err = init_srcu_struct(&vmd->irqs[i].srcu);
644 		if (err)
645 			return err;
646 
647 		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
648 		err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
649 				       vmd_irq, IRQF_NO_THREAD,
650 				       "vmd", &vmd->irqs[i]);
651 		if (err)
652 			return err;
653 	}
654 
655 	spin_lock_init(&vmd->cfg_lock);
656 	pci_set_drvdata(dev, vmd);
657 	err = vmd_enable_domain(vmd, (unsigned long) id->driver_data);
658 	if (err)
659 		return err;
660 
661 	dev_info(&vmd->dev->dev, "Bound to PCI domain %04x\n",
662 		 vmd->sysdata.domain);
663 	return 0;
664 }
665 
666 static void vmd_cleanup_srcu(struct vmd_dev *vmd)
667 {
668 	int i;
669 
670 	for (i = 0; i < vmd->msix_count; i++)
671 		cleanup_srcu_struct(&vmd->irqs[i].srcu);
672 }
673 
674 static void vmd_remove(struct pci_dev *dev)
675 {
676 	struct vmd_dev *vmd = pci_get_drvdata(dev);
677 	struct fwnode_handle *fn = vmd->irq_domain->fwnode;
678 
679 	sysfs_remove_link(&vmd->dev->dev.kobj, "domain");
680 	pci_stop_root_bus(vmd->bus);
681 	pci_remove_root_bus(vmd->bus);
682 	vmd_cleanup_srcu(vmd);
683 	vmd_detach_resources(vmd);
684 	irq_domain_remove(vmd->irq_domain);
685 	irq_domain_free_fwnode(fn);
686 }
687 
688 #ifdef CONFIG_PM_SLEEP
689 static int vmd_suspend(struct device *dev)
690 {
691 	struct pci_dev *pdev = to_pci_dev(dev);
692 	struct vmd_dev *vmd = pci_get_drvdata(pdev);
693 	int i;
694 
695 	for (i = 0; i < vmd->msix_count; i++)
696 		devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
697 
698 	pci_save_state(pdev);
699 	return 0;
700 }
701 
702 static int vmd_resume(struct device *dev)
703 {
704 	struct pci_dev *pdev = to_pci_dev(dev);
705 	struct vmd_dev *vmd = pci_get_drvdata(pdev);
706 	int err, i;
707 
708 	for (i = 0; i < vmd->msix_count; i++) {
709 		err = devm_request_irq(dev, pci_irq_vector(pdev, i),
710 				       vmd_irq, IRQF_NO_THREAD,
711 				       "vmd", &vmd->irqs[i]);
712 		if (err)
713 			return err;
714 	}
715 
716 	pci_restore_state(pdev);
717 	return 0;
718 }
719 #endif
720 static SIMPLE_DEV_PM_OPS(vmd_dev_pm_ops, vmd_suspend, vmd_resume);
721 
722 static const struct pci_device_id vmd_ids[] = {
723 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_VMD_201D),},
724 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_VMD_28C0),
725 		.driver_data = VMD_FEAT_HAS_MEMBAR_SHADOW |
726 				VMD_FEAT_HAS_BUS_RESTRICTIONS,},
727 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, 0x467f),
728 		.driver_data = VMD_FEAT_HAS_BUS_RESTRICTIONS,},
729 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, 0x4c3d),
730 		.driver_data = VMD_FEAT_HAS_BUS_RESTRICTIONS,},
731 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_VMD_9A0B),
732 		.driver_data = VMD_FEAT_HAS_BUS_RESTRICTIONS,},
733 	{0,}
734 };
735 MODULE_DEVICE_TABLE(pci, vmd_ids);
736 
737 static struct pci_driver vmd_drv = {
738 	.name		= "vmd",
739 	.id_table	= vmd_ids,
740 	.probe		= vmd_probe,
741 	.remove		= vmd_remove,
742 	.driver		= {
743 		.pm	= &vmd_dev_pm_ops,
744 	},
745 };
746 module_pci_driver(vmd_drv);
747 
748 MODULE_AUTHOR("Intel Corporation");
749 MODULE_LICENSE("GPL v2");
750 MODULE_VERSION("0.6");
751