xref: /linux/drivers/pci/controller/vmd.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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/pci-acpi.h>
15 #include <linux/pci-ecam.h>
16 #include <linux/srcu.h>
17 #include <linux/rculist.h>
18 #include <linux/rcupdate.h>
19 
20 #include <asm/irqdomain.h>
21 
22 #define VMD_CFGBAR	0
23 #define VMD_MEMBAR1	2
24 #define VMD_MEMBAR2	4
25 
26 #define PCI_REG_VMCAP		0x40
27 #define BUS_RESTRICT_CAP(vmcap)	(vmcap & 0x1)
28 #define PCI_REG_VMCONFIG	0x44
29 #define BUS_RESTRICT_CFG(vmcfg)	((vmcfg >> 8) & 0x3)
30 #define VMCONFIG_MSI_REMAP	0x2
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 	 * Device contains physical location shadow registers in
53 	 * vendor-specific capability space
54 	 */
55 	VMD_FEAT_HAS_MEMBAR_SHADOW_VSCAP	= (1 << 2),
56 
57 	/*
58 	 * Device may use MSI-X vector 0 for software triggering and will not
59 	 * be used for MSI remapping
60 	 */
61 	VMD_FEAT_OFFSET_FIRST_VECTOR		= (1 << 3),
62 
63 	/*
64 	 * Device can bypass remapping MSI-X transactions into its MSI-X table,
65 	 * avoiding the requirement of a VMD MSI domain for child device
66 	 * interrupt handling.
67 	 */
68 	VMD_FEAT_CAN_BYPASS_MSI_REMAP		= (1 << 4),
69 
70 	/*
71 	 * Enable ASPM on the PCIE root ports and set the default LTR of the
72 	 * storage devices on platforms where these values are not configured by
73 	 * BIOS. This is needed for laptops, which require these settings for
74 	 * proper power management of the SoC.
75 	 */
76 	VMD_FEAT_BIOS_PM_QUIRK		= (1 << 5),
77 };
78 
79 #define VMD_BIOS_PM_QUIRK_LTR	0x1003	/* 3145728 ns */
80 
81 #define VMD_FEATS_CLIENT	(VMD_FEAT_HAS_MEMBAR_SHADOW_VSCAP |	\
82 				 VMD_FEAT_HAS_BUS_RESTRICTIONS |	\
83 				 VMD_FEAT_OFFSET_FIRST_VECTOR |		\
84 				 VMD_FEAT_BIOS_PM_QUIRK)
85 
86 static DEFINE_IDA(vmd_instance_ida);
87 
88 /*
89  * Lock for manipulating VMD IRQ lists.
90  */
91 static DEFINE_RAW_SPINLOCK(list_lock);
92 
93 /**
94  * struct vmd_irq - private data to map driver IRQ to the VMD shared vector
95  * @node:	list item for parent traversal.
96  * @irq:	back pointer to parent.
97  * @enabled:	true if driver enabled IRQ
98  * @virq:	the virtual IRQ value provided to the requesting driver.
99  *
100  * Every MSI/MSI-X IRQ requested for a device in a VMD domain will be mapped to
101  * a VMD IRQ using this structure.
102  */
103 struct vmd_irq {
104 	struct list_head	node;
105 	struct vmd_irq_list	*irq;
106 	bool			enabled;
107 	unsigned int		virq;
108 };
109 
110 /**
111  * struct vmd_irq_list - list of driver requested IRQs mapping to a VMD vector
112  * @irq_list:	the list of irq's the VMD one demuxes to.
113  * @srcu:	SRCU struct for local synchronization.
114  * @count:	number of child IRQs assigned to this vector; used to track
115  *		sharing.
116  * @virq:	The underlying VMD Linux interrupt number
117  */
118 struct vmd_irq_list {
119 	struct list_head	irq_list;
120 	struct srcu_struct	srcu;
121 	unsigned int		count;
122 	unsigned int		virq;
123 };
124 
125 struct vmd_dev {
126 	struct pci_dev		*dev;
127 
128 	spinlock_t		cfg_lock;
129 	void __iomem		*cfgbar;
130 
131 	int msix_count;
132 	struct vmd_irq_list	*irqs;
133 
134 	struct pci_sysdata	sysdata;
135 	struct resource		resources[3];
136 	struct irq_domain	*irq_domain;
137 	struct pci_bus		*bus;
138 	u8			busn_start;
139 	u8			first_vec;
140 	char			*name;
141 	int			instance;
142 };
143 
144 static inline struct vmd_dev *vmd_from_bus(struct pci_bus *bus)
145 {
146 	return container_of(bus->sysdata, struct vmd_dev, sysdata);
147 }
148 
149 static inline unsigned int index_from_irqs(struct vmd_dev *vmd,
150 					   struct vmd_irq_list *irqs)
151 {
152 	return irqs - vmd->irqs;
153 }
154 
155 /*
156  * Drivers managing a device in a VMD domain allocate their own IRQs as before,
157  * but the MSI entry for the hardware it's driving will be programmed with a
158  * destination ID for the VMD MSI-X table.  The VMD muxes interrupts in its
159  * domain into one of its own, and the VMD driver de-muxes these for the
160  * handlers sharing that VMD IRQ.  The vmd irq_domain provides the operations
161  * and irq_chip to set this up.
162  */
163 static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
164 {
165 	struct vmd_irq *vmdirq = data->chip_data;
166 	struct vmd_irq_list *irq = vmdirq->irq;
167 	struct vmd_dev *vmd = irq_data_get_irq_handler_data(data);
168 
169 	memset(msg, 0, sizeof(*msg));
170 	msg->address_hi = X86_MSI_BASE_ADDRESS_HIGH;
171 	msg->arch_addr_lo.base_address = X86_MSI_BASE_ADDRESS_LOW;
172 	msg->arch_addr_lo.destid_0_7 = index_from_irqs(vmd, irq);
173 }
174 
175 /*
176  * We rely on MSI_FLAG_USE_DEF_CHIP_OPS to set the IRQ mask/unmask ops.
177  */
178 static void vmd_irq_enable(struct irq_data *data)
179 {
180 	struct vmd_irq *vmdirq = data->chip_data;
181 	unsigned long flags;
182 
183 	raw_spin_lock_irqsave(&list_lock, flags);
184 	WARN_ON(vmdirq->enabled);
185 	list_add_tail_rcu(&vmdirq->node, &vmdirq->irq->irq_list);
186 	vmdirq->enabled = true;
187 	raw_spin_unlock_irqrestore(&list_lock, flags);
188 
189 	data->chip->irq_unmask(data);
190 }
191 
192 static void vmd_irq_disable(struct irq_data *data)
193 {
194 	struct vmd_irq *vmdirq = data->chip_data;
195 	unsigned long flags;
196 
197 	data->chip->irq_mask(data);
198 
199 	raw_spin_lock_irqsave(&list_lock, flags);
200 	if (vmdirq->enabled) {
201 		list_del_rcu(&vmdirq->node);
202 		vmdirq->enabled = false;
203 	}
204 	raw_spin_unlock_irqrestore(&list_lock, flags);
205 }
206 
207 static struct irq_chip vmd_msi_controller = {
208 	.name			= "VMD-MSI",
209 	.irq_enable		= vmd_irq_enable,
210 	.irq_disable		= vmd_irq_disable,
211 	.irq_compose_msi_msg	= vmd_compose_msi_msg,
212 };
213 
214 static irq_hw_number_t vmd_get_hwirq(struct msi_domain_info *info,
215 				     msi_alloc_info_t *arg)
216 {
217 	return 0;
218 }
219 
220 /*
221  * XXX: We can be even smarter selecting the best IRQ once we solve the
222  * affinity problem.
223  */
224 static struct vmd_irq_list *vmd_next_irq(struct vmd_dev *vmd, struct msi_desc *desc)
225 {
226 	unsigned long flags;
227 	int i, best;
228 
229 	if (vmd->msix_count == 1 + vmd->first_vec)
230 		return &vmd->irqs[vmd->first_vec];
231 
232 	/*
233 	 * White list for fast-interrupt handlers. All others will share the
234 	 * "slow" interrupt vector.
235 	 */
236 	switch (msi_desc_to_pci_dev(desc)->class) {
237 	case PCI_CLASS_STORAGE_EXPRESS:
238 		break;
239 	default:
240 		return &vmd->irqs[vmd->first_vec];
241 	}
242 
243 	raw_spin_lock_irqsave(&list_lock, flags);
244 	best = vmd->first_vec + 1;
245 	for (i = best; i < vmd->msix_count; i++)
246 		if (vmd->irqs[i].count < vmd->irqs[best].count)
247 			best = i;
248 	vmd->irqs[best].count++;
249 	raw_spin_unlock_irqrestore(&list_lock, flags);
250 
251 	return &vmd->irqs[best];
252 }
253 
254 static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
255 			unsigned int virq, irq_hw_number_t hwirq,
256 			msi_alloc_info_t *arg)
257 {
258 	struct msi_desc *desc = arg->desc;
259 	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
260 	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
261 
262 	if (!vmdirq)
263 		return -ENOMEM;
264 
265 	INIT_LIST_HEAD(&vmdirq->node);
266 	vmdirq->irq = vmd_next_irq(vmd, desc);
267 	vmdirq->virq = virq;
268 
269 	irq_domain_set_info(domain, virq, vmdirq->irq->virq, info->chip, vmdirq,
270 			    handle_untracked_irq, vmd, NULL);
271 	return 0;
272 }
273 
274 static void vmd_msi_free(struct irq_domain *domain,
275 			struct msi_domain_info *info, unsigned int virq)
276 {
277 	struct vmd_irq *vmdirq = irq_get_chip_data(virq);
278 	unsigned long flags;
279 
280 	synchronize_srcu(&vmdirq->irq->srcu);
281 
282 	/* XXX: Potential optimization to rebalance */
283 	raw_spin_lock_irqsave(&list_lock, flags);
284 	vmdirq->irq->count--;
285 	raw_spin_unlock_irqrestore(&list_lock, flags);
286 
287 	kfree(vmdirq);
288 }
289 
290 static int vmd_msi_prepare(struct irq_domain *domain, struct device *dev,
291 			   int nvec, msi_alloc_info_t *arg)
292 {
293 	struct pci_dev *pdev = to_pci_dev(dev);
294 	struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
295 
296 	if (nvec > vmd->msix_count)
297 		return vmd->msix_count;
298 
299 	memset(arg, 0, sizeof(*arg));
300 	return 0;
301 }
302 
303 static void vmd_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
304 {
305 	arg->desc = desc;
306 }
307 
308 static struct msi_domain_ops vmd_msi_domain_ops = {
309 	.get_hwirq	= vmd_get_hwirq,
310 	.msi_init	= vmd_msi_init,
311 	.msi_free	= vmd_msi_free,
312 	.msi_prepare	= vmd_msi_prepare,
313 	.set_desc	= vmd_set_desc,
314 };
315 
316 static struct msi_domain_info vmd_msi_domain_info = {
317 	.flags		= MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
318 			  MSI_FLAG_NO_AFFINITY | MSI_FLAG_PCI_MSIX,
319 	.ops		= &vmd_msi_domain_ops,
320 	.chip		= &vmd_msi_controller,
321 };
322 
323 static void vmd_set_msi_remapping(struct vmd_dev *vmd, bool enable)
324 {
325 	u16 reg;
326 
327 	pci_read_config_word(vmd->dev, PCI_REG_VMCONFIG, &reg);
328 	reg = enable ? (reg & ~VMCONFIG_MSI_REMAP) :
329 		       (reg | VMCONFIG_MSI_REMAP);
330 	pci_write_config_word(vmd->dev, PCI_REG_VMCONFIG, reg);
331 }
332 
333 static int vmd_create_irq_domain(struct vmd_dev *vmd)
334 {
335 	struct fwnode_handle *fn;
336 
337 	fn = irq_domain_alloc_named_id_fwnode("VMD-MSI", vmd->sysdata.domain);
338 	if (!fn)
339 		return -ENODEV;
340 
341 	vmd->irq_domain = pci_msi_create_irq_domain(fn, &vmd_msi_domain_info, NULL);
342 	if (!vmd->irq_domain) {
343 		irq_domain_free_fwnode(fn);
344 		return -ENODEV;
345 	}
346 
347 	return 0;
348 }
349 
350 static void vmd_remove_irq_domain(struct vmd_dev *vmd)
351 {
352 	/*
353 	 * Some production BIOS won't enable remapping between soft reboots.
354 	 * Ensure remapping is restored before unloading the driver.
355 	 */
356 	if (!vmd->msix_count)
357 		vmd_set_msi_remapping(vmd, true);
358 
359 	if (vmd->irq_domain) {
360 		struct fwnode_handle *fn = vmd->irq_domain->fwnode;
361 
362 		irq_domain_remove(vmd->irq_domain);
363 		irq_domain_free_fwnode(fn);
364 	}
365 }
366 
367 static void __iomem *vmd_cfg_addr(struct vmd_dev *vmd, struct pci_bus *bus,
368 				  unsigned int devfn, int reg, int len)
369 {
370 	unsigned int busnr_ecam = bus->number - vmd->busn_start;
371 	u32 offset = PCIE_ECAM_OFFSET(busnr_ecam, devfn, reg);
372 
373 	if (offset + len >= resource_size(&vmd->dev->resource[VMD_CFGBAR]))
374 		return NULL;
375 
376 	return vmd->cfgbar + offset;
377 }
378 
379 /*
380  * CPU may deadlock if config space is not serialized on some versions of this
381  * hardware, so all config space access is done under a spinlock.
382  */
383 static int vmd_pci_read(struct pci_bus *bus, unsigned int devfn, int reg,
384 			int len, u32 *value)
385 {
386 	struct vmd_dev *vmd = vmd_from_bus(bus);
387 	void __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
388 	unsigned long flags;
389 	int ret = 0;
390 
391 	if (!addr)
392 		return -EFAULT;
393 
394 	spin_lock_irqsave(&vmd->cfg_lock, flags);
395 	switch (len) {
396 	case 1:
397 		*value = readb(addr);
398 		break;
399 	case 2:
400 		*value = readw(addr);
401 		break;
402 	case 4:
403 		*value = readl(addr);
404 		break;
405 	default:
406 		ret = -EINVAL;
407 		break;
408 	}
409 	spin_unlock_irqrestore(&vmd->cfg_lock, flags);
410 	return ret;
411 }
412 
413 /*
414  * VMD h/w converts non-posted config writes to posted memory writes. The
415  * read-back in this function forces the completion so it returns only after
416  * the config space was written, as expected.
417  */
418 static int vmd_pci_write(struct pci_bus *bus, unsigned int devfn, int reg,
419 			 int len, u32 value)
420 {
421 	struct vmd_dev *vmd = vmd_from_bus(bus);
422 	void __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
423 	unsigned long flags;
424 	int ret = 0;
425 
426 	if (!addr)
427 		return -EFAULT;
428 
429 	spin_lock_irqsave(&vmd->cfg_lock, flags);
430 	switch (len) {
431 	case 1:
432 		writeb(value, addr);
433 		readb(addr);
434 		break;
435 	case 2:
436 		writew(value, addr);
437 		readw(addr);
438 		break;
439 	case 4:
440 		writel(value, addr);
441 		readl(addr);
442 		break;
443 	default:
444 		ret = -EINVAL;
445 		break;
446 	}
447 	spin_unlock_irqrestore(&vmd->cfg_lock, flags);
448 	return ret;
449 }
450 
451 static struct pci_ops vmd_ops = {
452 	.read		= vmd_pci_read,
453 	.write		= vmd_pci_write,
454 };
455 
456 #ifdef CONFIG_ACPI
457 static struct acpi_device *vmd_acpi_find_companion(struct pci_dev *pci_dev)
458 {
459 	struct pci_host_bridge *bridge;
460 	u32 busnr, addr;
461 
462 	if (pci_dev->bus->ops != &vmd_ops)
463 		return NULL;
464 
465 	bridge = pci_find_host_bridge(pci_dev->bus);
466 	busnr = pci_dev->bus->number - bridge->bus->number;
467 	/*
468 	 * The address computation below is only applicable to relative bus
469 	 * numbers below 32.
470 	 */
471 	if (busnr > 31)
472 		return NULL;
473 
474 	addr = (busnr << 24) | ((u32)pci_dev->devfn << 16) | 0x8000FFFFU;
475 
476 	dev_dbg(&pci_dev->dev, "Looking for ACPI companion (address 0x%x)\n",
477 		addr);
478 
479 	return acpi_find_child_device(ACPI_COMPANION(bridge->dev.parent), addr,
480 				      false);
481 }
482 
483 static bool hook_installed;
484 
485 static void vmd_acpi_begin(void)
486 {
487 	if (pci_acpi_set_companion_lookup_hook(vmd_acpi_find_companion))
488 		return;
489 
490 	hook_installed = true;
491 }
492 
493 static void vmd_acpi_end(void)
494 {
495 	if (!hook_installed)
496 		return;
497 
498 	pci_acpi_clear_companion_lookup_hook();
499 	hook_installed = false;
500 }
501 #else
502 static inline void vmd_acpi_begin(void) { }
503 static inline void vmd_acpi_end(void) { }
504 #endif /* CONFIG_ACPI */
505 
506 static void vmd_domain_reset(struct vmd_dev *vmd)
507 {
508 	u16 bus, max_buses = resource_size(&vmd->resources[0]);
509 	u8 dev, functions, fn, hdr_type;
510 	char __iomem *base;
511 
512 	for (bus = 0; bus < max_buses; bus++) {
513 		for (dev = 0; dev < 32; dev++) {
514 			base = vmd->cfgbar + PCIE_ECAM_OFFSET(bus,
515 						PCI_DEVFN(dev, 0), 0);
516 
517 			hdr_type = readb(base + PCI_HEADER_TYPE);
518 
519 			functions = (hdr_type & PCI_HEADER_TYPE_MFD) ? 8 : 1;
520 			for (fn = 0; fn < functions; fn++) {
521 				base = vmd->cfgbar + PCIE_ECAM_OFFSET(bus,
522 						PCI_DEVFN(dev, fn), 0);
523 
524 				hdr_type = readb(base + PCI_HEADER_TYPE) &
525 						PCI_HEADER_TYPE_MASK;
526 
527 				if (hdr_type != PCI_HEADER_TYPE_BRIDGE ||
528 				    (readw(base + PCI_CLASS_DEVICE) !=
529 				     PCI_CLASS_BRIDGE_PCI))
530 					continue;
531 
532 				/*
533 				 * Temporarily disable the I/O range before updating
534 				 * PCI_IO_BASE.
535 				 */
536 				writel(0x0000ffff, base + PCI_IO_BASE_UPPER16);
537 				/* Update lower 16 bits of I/O base/limit */
538 				writew(0x00f0, base + PCI_IO_BASE);
539 				/* Update upper 16 bits of I/O base/limit */
540 				writel(0, base + PCI_IO_BASE_UPPER16);
541 
542 				/* MMIO Base/Limit */
543 				writel(0x0000fff0, base + PCI_MEMORY_BASE);
544 
545 				/* Prefetchable MMIO Base/Limit */
546 				writel(0, base + PCI_PREF_LIMIT_UPPER32);
547 				writel(0x0000fff0, base + PCI_PREF_MEMORY_BASE);
548 				writel(0xffffffff, base + PCI_PREF_BASE_UPPER32);
549 			}
550 		}
551 	}
552 }
553 
554 static void vmd_attach_resources(struct vmd_dev *vmd)
555 {
556 	vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
557 	vmd->dev->resource[VMD_MEMBAR2].child = &vmd->resources[2];
558 }
559 
560 static void vmd_detach_resources(struct vmd_dev *vmd)
561 {
562 	vmd->dev->resource[VMD_MEMBAR1].child = NULL;
563 	vmd->dev->resource[VMD_MEMBAR2].child = NULL;
564 }
565 
566 /*
567  * VMD domains start at 0x10000 to not clash with ACPI _SEG domains.
568  * Per ACPI r6.0, sec 6.5.6,  _SEG returns an integer, of which the lower
569  * 16 bits are the PCI Segment Group (domain) number.  Other bits are
570  * currently reserved.
571  */
572 static int vmd_find_free_domain(void)
573 {
574 	int domain = 0xffff;
575 	struct pci_bus *bus = NULL;
576 
577 	while ((bus = pci_find_next_bus(bus)) != NULL)
578 		domain = max_t(int, domain, pci_domain_nr(bus));
579 	return domain + 1;
580 }
581 
582 static int vmd_get_phys_offsets(struct vmd_dev *vmd, bool native_hint,
583 				resource_size_t *offset1,
584 				resource_size_t *offset2)
585 {
586 	struct pci_dev *dev = vmd->dev;
587 	u64 phys1, phys2;
588 
589 	if (native_hint) {
590 		u32 vmlock;
591 		int ret;
592 
593 		ret = pci_read_config_dword(dev, PCI_REG_VMLOCK, &vmlock);
594 		if (ret || PCI_POSSIBLE_ERROR(vmlock))
595 			return -ENODEV;
596 
597 		if (MB2_SHADOW_EN(vmlock)) {
598 			void __iomem *membar2;
599 
600 			membar2 = pci_iomap(dev, VMD_MEMBAR2, 0);
601 			if (!membar2)
602 				return -ENOMEM;
603 			phys1 = readq(membar2 + MB2_SHADOW_OFFSET);
604 			phys2 = readq(membar2 + MB2_SHADOW_OFFSET + 8);
605 			pci_iounmap(dev, membar2);
606 		} else
607 			return 0;
608 	} else {
609 		/* Hypervisor-Emulated Vendor-Specific Capability */
610 		int pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
611 		u32 reg, regu;
612 
613 		pci_read_config_dword(dev, pos + 4, &reg);
614 
615 		/* "SHDW" */
616 		if (pos && reg == 0x53484457) {
617 			pci_read_config_dword(dev, pos + 8, &reg);
618 			pci_read_config_dword(dev, pos + 12, &regu);
619 			phys1 = (u64) regu << 32 | reg;
620 
621 			pci_read_config_dword(dev, pos + 16, &reg);
622 			pci_read_config_dword(dev, pos + 20, &regu);
623 			phys2 = (u64) regu << 32 | reg;
624 		} else
625 			return 0;
626 	}
627 
628 	*offset1 = dev->resource[VMD_MEMBAR1].start -
629 			(phys1 & PCI_BASE_ADDRESS_MEM_MASK);
630 	*offset2 = dev->resource[VMD_MEMBAR2].start -
631 			(phys2 & PCI_BASE_ADDRESS_MEM_MASK);
632 
633 	return 0;
634 }
635 
636 static int vmd_get_bus_number_start(struct vmd_dev *vmd)
637 {
638 	struct pci_dev *dev = vmd->dev;
639 	u16 reg;
640 
641 	pci_read_config_word(dev, PCI_REG_VMCAP, &reg);
642 	if (BUS_RESTRICT_CAP(reg)) {
643 		pci_read_config_word(dev, PCI_REG_VMCONFIG, &reg);
644 
645 		switch (BUS_RESTRICT_CFG(reg)) {
646 		case 0:
647 			vmd->busn_start = 0;
648 			break;
649 		case 1:
650 			vmd->busn_start = 128;
651 			break;
652 		case 2:
653 			vmd->busn_start = 224;
654 			break;
655 		default:
656 			pci_err(dev, "Unknown Bus Offset Setting (%d)\n",
657 				BUS_RESTRICT_CFG(reg));
658 			return -ENODEV;
659 		}
660 	}
661 
662 	return 0;
663 }
664 
665 static irqreturn_t vmd_irq(int irq, void *data)
666 {
667 	struct vmd_irq_list *irqs = data;
668 	struct vmd_irq *vmdirq;
669 	int idx;
670 
671 	idx = srcu_read_lock(&irqs->srcu);
672 	list_for_each_entry_rcu(vmdirq, &irqs->irq_list, node)
673 		generic_handle_irq(vmdirq->virq);
674 	srcu_read_unlock(&irqs->srcu, idx);
675 
676 	return IRQ_HANDLED;
677 }
678 
679 static int vmd_alloc_irqs(struct vmd_dev *vmd)
680 {
681 	struct pci_dev *dev = vmd->dev;
682 	int i, err;
683 
684 	vmd->msix_count = pci_msix_vec_count(dev);
685 	if (vmd->msix_count < 0)
686 		return -ENODEV;
687 
688 	vmd->msix_count = pci_alloc_irq_vectors(dev, vmd->first_vec + 1,
689 						vmd->msix_count, PCI_IRQ_MSIX);
690 	if (vmd->msix_count < 0)
691 		return vmd->msix_count;
692 
693 	vmd->irqs = devm_kcalloc(&dev->dev, vmd->msix_count, sizeof(*vmd->irqs),
694 				 GFP_KERNEL);
695 	if (!vmd->irqs)
696 		return -ENOMEM;
697 
698 	for (i = 0; i < vmd->msix_count; i++) {
699 		err = init_srcu_struct(&vmd->irqs[i].srcu);
700 		if (err)
701 			return err;
702 
703 		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
704 		vmd->irqs[i].virq = pci_irq_vector(dev, i);
705 		err = devm_request_irq(&dev->dev, vmd->irqs[i].virq,
706 				       vmd_irq, IRQF_NO_THREAD,
707 				       vmd->name, &vmd->irqs[i]);
708 		if (err)
709 			return err;
710 	}
711 
712 	return 0;
713 }
714 
715 /*
716  * Since VMD is an aperture to regular PCIe root ports, only allow it to
717  * control features that the OS is allowed to control on the physical PCI bus.
718  */
719 static void vmd_copy_host_bridge_flags(struct pci_host_bridge *root_bridge,
720 				       struct pci_host_bridge *vmd_bridge)
721 {
722 	vmd_bridge->native_pcie_hotplug = root_bridge->native_pcie_hotplug;
723 	vmd_bridge->native_shpc_hotplug = root_bridge->native_shpc_hotplug;
724 	vmd_bridge->native_aer = root_bridge->native_aer;
725 	vmd_bridge->native_pme = root_bridge->native_pme;
726 	vmd_bridge->native_ltr = root_bridge->native_ltr;
727 	vmd_bridge->native_dpc = root_bridge->native_dpc;
728 }
729 
730 /*
731  * Enable ASPM and LTR settings on devices that aren't configured by BIOS.
732  */
733 static int vmd_pm_enable_quirk(struct pci_dev *pdev, void *userdata)
734 {
735 	unsigned long features = *(unsigned long *)userdata;
736 	u16 ltr = VMD_BIOS_PM_QUIRK_LTR;
737 	u32 ltr_reg;
738 	int pos;
739 
740 	if (!(features & VMD_FEAT_BIOS_PM_QUIRK))
741 		return 0;
742 
743 	pci_enable_link_state_locked(pdev, PCIE_LINK_STATE_ALL);
744 
745 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_LTR);
746 	if (!pos)
747 		return 0;
748 
749 	/*
750 	 * Skip if the max snoop LTR is non-zero, indicating BIOS has set it
751 	 * so the LTR quirk is not needed.
752 	 */
753 	pci_read_config_dword(pdev, pos + PCI_LTR_MAX_SNOOP_LAT, &ltr_reg);
754 	if (!!(ltr_reg & (PCI_LTR_VALUE_MASK | PCI_LTR_SCALE_MASK)))
755 		return 0;
756 
757 	/*
758 	 * Set the default values to the maximum required by the platform to
759 	 * allow the deepest power management savings. Write as a DWORD where
760 	 * the lower word is the max snoop latency and the upper word is the
761 	 * max non-snoop latency.
762 	 */
763 	ltr_reg = (ltr << 16) | ltr;
764 	pci_write_config_dword(pdev, pos + PCI_LTR_MAX_SNOOP_LAT, ltr_reg);
765 	pci_info(pdev, "VMD: Default LTR value set by driver\n");
766 
767 	return 0;
768 }
769 
770 static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
771 {
772 	struct pci_sysdata *sd = &vmd->sysdata;
773 	struct resource *res;
774 	u32 upper_bits;
775 	unsigned long flags;
776 	LIST_HEAD(resources);
777 	resource_size_t offset[2] = {0};
778 	resource_size_t membar2_offset = 0x2000;
779 	struct pci_bus *child;
780 	struct pci_dev *dev;
781 	int ret;
782 
783 	/*
784 	 * Shadow registers may exist in certain VMD device ids which allow
785 	 * guests to correctly assign host physical addresses to the root ports
786 	 * and child devices. These registers will either return the host value
787 	 * or 0, depending on an enable bit in the VMD device.
788 	 */
789 	if (features & VMD_FEAT_HAS_MEMBAR_SHADOW) {
790 		membar2_offset = MB2_SHADOW_OFFSET + MB2_SHADOW_SIZE;
791 		ret = vmd_get_phys_offsets(vmd, true, &offset[0], &offset[1]);
792 		if (ret)
793 			return ret;
794 	} else if (features & VMD_FEAT_HAS_MEMBAR_SHADOW_VSCAP) {
795 		ret = vmd_get_phys_offsets(vmd, false, &offset[0], &offset[1]);
796 		if (ret)
797 			return ret;
798 	}
799 
800 	/*
801 	 * Certain VMD devices may have a root port configuration option which
802 	 * limits the bus range to between 0-127, 128-255, or 224-255
803 	 */
804 	if (features & VMD_FEAT_HAS_BUS_RESTRICTIONS) {
805 		ret = vmd_get_bus_number_start(vmd);
806 		if (ret)
807 			return ret;
808 	}
809 
810 	res = &vmd->dev->resource[VMD_CFGBAR];
811 	vmd->resources[0] = (struct resource) {
812 		.name  = "VMD CFGBAR",
813 		.start = vmd->busn_start,
814 		.end   = vmd->busn_start + (resource_size(res) >> 20) - 1,
815 		.flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED,
816 	};
817 
818 	/*
819 	 * If the window is below 4GB, clear IORESOURCE_MEM_64 so we can
820 	 * put 32-bit resources in the window.
821 	 *
822 	 * There's no hardware reason why a 64-bit window *couldn't*
823 	 * contain a 32-bit resource, but pbus_size_mem() computes the
824 	 * bridge window size assuming a 64-bit window will contain no
825 	 * 32-bit resources.  __pci_assign_resource() enforces that
826 	 * artificial restriction to make sure everything will fit.
827 	 *
828 	 * The only way we could use a 64-bit non-prefetchable MEMBAR is
829 	 * if its address is <4GB so that we can convert it to a 32-bit
830 	 * resource.  To be visible to the host OS, all VMD endpoints must
831 	 * be initially configured by platform BIOS, which includes setting
832 	 * up these resources.  We can assume the device is configured
833 	 * according to the platform needs.
834 	 */
835 	res = &vmd->dev->resource[VMD_MEMBAR1];
836 	upper_bits = upper_32_bits(res->end);
837 	flags = res->flags & ~IORESOURCE_SIZEALIGN;
838 	if (!upper_bits)
839 		flags &= ~IORESOURCE_MEM_64;
840 	vmd->resources[1] = (struct resource) {
841 		.name  = "VMD MEMBAR1",
842 		.start = res->start,
843 		.end   = res->end,
844 		.flags = flags,
845 		.parent = res,
846 	};
847 
848 	res = &vmd->dev->resource[VMD_MEMBAR2];
849 	upper_bits = upper_32_bits(res->end);
850 	flags = res->flags & ~IORESOURCE_SIZEALIGN;
851 	if (!upper_bits)
852 		flags &= ~IORESOURCE_MEM_64;
853 	vmd->resources[2] = (struct resource) {
854 		.name  = "VMD MEMBAR2",
855 		.start = res->start + membar2_offset,
856 		.end   = res->end,
857 		.flags = flags,
858 		.parent = res,
859 	};
860 
861 	sd->vmd_dev = vmd->dev;
862 	sd->domain = vmd_find_free_domain();
863 	if (sd->domain < 0)
864 		return sd->domain;
865 
866 	sd->node = pcibus_to_node(vmd->dev->bus);
867 
868 	/*
869 	 * Currently MSI remapping must be enabled in guest passthrough mode
870 	 * due to some missing interrupt remapping plumbing. This is probably
871 	 * acceptable because the guest is usually CPU-limited and MSI
872 	 * remapping doesn't become a performance bottleneck.
873 	 */
874 	if (!(features & VMD_FEAT_CAN_BYPASS_MSI_REMAP) ||
875 	    offset[0] || offset[1]) {
876 		ret = vmd_alloc_irqs(vmd);
877 		if (ret)
878 			return ret;
879 
880 		vmd_set_msi_remapping(vmd, true);
881 
882 		ret = vmd_create_irq_domain(vmd);
883 		if (ret)
884 			return ret;
885 
886 		/*
887 		 * Override the IRQ domain bus token so the domain can be
888 		 * distinguished from a regular PCI/MSI domain.
889 		 */
890 		irq_domain_update_bus_token(vmd->irq_domain, DOMAIN_BUS_VMD_MSI);
891 	} else {
892 		vmd_set_msi_remapping(vmd, false);
893 	}
894 
895 	pci_add_resource(&resources, &vmd->resources[0]);
896 	pci_add_resource_offset(&resources, &vmd->resources[1], offset[0]);
897 	pci_add_resource_offset(&resources, &vmd->resources[2], offset[1]);
898 
899 	vmd->bus = pci_create_root_bus(&vmd->dev->dev, vmd->busn_start,
900 				       &vmd_ops, sd, &resources);
901 	if (!vmd->bus) {
902 		pci_free_resource_list(&resources);
903 		vmd_remove_irq_domain(vmd);
904 		return -ENODEV;
905 	}
906 
907 	vmd_copy_host_bridge_flags(pci_find_host_bridge(vmd->dev->bus),
908 				   to_pci_host_bridge(vmd->bus->bridge));
909 
910 	vmd_attach_resources(vmd);
911 	if (vmd->irq_domain)
912 		dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
913 	else
914 		dev_set_msi_domain(&vmd->bus->dev,
915 				   dev_get_msi_domain(&vmd->dev->dev));
916 
917 	WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
918 			       "domain"), "Can't create symlink to domain\n");
919 
920 	vmd_acpi_begin();
921 
922 	pci_scan_child_bus(vmd->bus);
923 	vmd_domain_reset(vmd);
924 
925 	/* When Intel VMD is enabled, the OS does not discover the Root Ports
926 	 * owned by Intel VMD within the MMCFG space. pci_reset_bus() applies
927 	 * a reset to the parent of the PCI device supplied as argument. This
928 	 * is why we pass a child device, so the reset can be triggered at
929 	 * the Intel bridge level and propagated to all the children in the
930 	 * hierarchy.
931 	 */
932 	list_for_each_entry(child, &vmd->bus->children, node) {
933 		if (!list_empty(&child->devices)) {
934 			dev = list_first_entry(&child->devices,
935 					       struct pci_dev, bus_list);
936 			ret = pci_reset_bus(dev);
937 			if (ret)
938 				pci_warn(dev, "can't reset device: %d\n", ret);
939 
940 			break;
941 		}
942 	}
943 
944 	pci_assign_unassigned_bus_resources(vmd->bus);
945 
946 	pci_walk_bus(vmd->bus, vmd_pm_enable_quirk, &features);
947 
948 	/*
949 	 * VMD root buses are virtual and don't return true on pci_is_pcie()
950 	 * and will fail pcie_bus_configure_settings() early. It can instead be
951 	 * run on each of the real root ports.
952 	 */
953 	list_for_each_entry(child, &vmd->bus->children, node)
954 		pcie_bus_configure_settings(child);
955 
956 	pci_bus_add_devices(vmd->bus);
957 
958 	vmd_acpi_end();
959 	return 0;
960 }
961 
962 static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
963 {
964 	unsigned long features = (unsigned long) id->driver_data;
965 	struct vmd_dev *vmd;
966 	int err;
967 
968 	if (resource_size(&dev->resource[VMD_CFGBAR]) < (1 << 20))
969 		return -ENOMEM;
970 
971 	vmd = devm_kzalloc(&dev->dev, sizeof(*vmd), GFP_KERNEL);
972 	if (!vmd)
973 		return -ENOMEM;
974 
975 	vmd->dev = dev;
976 	vmd->instance = ida_alloc(&vmd_instance_ida, GFP_KERNEL);
977 	if (vmd->instance < 0)
978 		return vmd->instance;
979 
980 	vmd->name = devm_kasprintf(&dev->dev, GFP_KERNEL, "vmd%d",
981 				   vmd->instance);
982 	if (!vmd->name) {
983 		err = -ENOMEM;
984 		goto out_release_instance;
985 	}
986 
987 	err = pcim_enable_device(dev);
988 	if (err < 0)
989 		goto out_release_instance;
990 
991 	vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
992 	if (!vmd->cfgbar) {
993 		err = -ENOMEM;
994 		goto out_release_instance;
995 	}
996 
997 	pci_set_master(dev);
998 	if (dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(64)) &&
999 	    dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(32))) {
1000 		err = -ENODEV;
1001 		goto out_release_instance;
1002 	}
1003 
1004 	if (features & VMD_FEAT_OFFSET_FIRST_VECTOR)
1005 		vmd->first_vec = 1;
1006 
1007 	spin_lock_init(&vmd->cfg_lock);
1008 	pci_set_drvdata(dev, vmd);
1009 	err = vmd_enable_domain(vmd, features);
1010 	if (err)
1011 		goto out_release_instance;
1012 
1013 	dev_info(&vmd->dev->dev, "Bound to PCI domain %04x\n",
1014 		 vmd->sysdata.domain);
1015 	return 0;
1016 
1017  out_release_instance:
1018 	ida_free(&vmd_instance_ida, vmd->instance);
1019 	return err;
1020 }
1021 
1022 static void vmd_cleanup_srcu(struct vmd_dev *vmd)
1023 {
1024 	int i;
1025 
1026 	for (i = 0; i < vmd->msix_count; i++)
1027 		cleanup_srcu_struct(&vmd->irqs[i].srcu);
1028 }
1029 
1030 static void vmd_remove(struct pci_dev *dev)
1031 {
1032 	struct vmd_dev *vmd = pci_get_drvdata(dev);
1033 
1034 	pci_stop_root_bus(vmd->bus);
1035 	sysfs_remove_link(&vmd->dev->dev.kobj, "domain");
1036 	pci_remove_root_bus(vmd->bus);
1037 	vmd_cleanup_srcu(vmd);
1038 	vmd_detach_resources(vmd);
1039 	vmd_remove_irq_domain(vmd);
1040 	ida_free(&vmd_instance_ida, vmd->instance);
1041 }
1042 
1043 static void vmd_shutdown(struct pci_dev *dev)
1044 {
1045 	struct vmd_dev *vmd = pci_get_drvdata(dev);
1046 
1047 	vmd_remove_irq_domain(vmd);
1048 }
1049 
1050 #ifdef CONFIG_PM_SLEEP
1051 static int vmd_suspend(struct device *dev)
1052 {
1053 	struct pci_dev *pdev = to_pci_dev(dev);
1054 	struct vmd_dev *vmd = pci_get_drvdata(pdev);
1055 	int i;
1056 
1057 	for (i = 0; i < vmd->msix_count; i++)
1058 		devm_free_irq(dev, vmd->irqs[i].virq, &vmd->irqs[i]);
1059 
1060 	return 0;
1061 }
1062 
1063 static int vmd_resume(struct device *dev)
1064 {
1065 	struct pci_dev *pdev = to_pci_dev(dev);
1066 	struct vmd_dev *vmd = pci_get_drvdata(pdev);
1067 	int err, i;
1068 
1069 	vmd_set_msi_remapping(vmd, !!vmd->irq_domain);
1070 
1071 	for (i = 0; i < vmd->msix_count; i++) {
1072 		err = devm_request_irq(dev, vmd->irqs[i].virq,
1073 				       vmd_irq, IRQF_NO_THREAD,
1074 				       vmd->name, &vmd->irqs[i]);
1075 		if (err)
1076 			return err;
1077 	}
1078 
1079 	return 0;
1080 }
1081 #endif
1082 static SIMPLE_DEV_PM_OPS(vmd_dev_pm_ops, vmd_suspend, vmd_resume);
1083 
1084 static const struct pci_device_id vmd_ids[] = {
1085 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_VMD_201D),
1086 		.driver_data = VMD_FEAT_HAS_MEMBAR_SHADOW_VSCAP,},
1087 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_VMD_28C0),
1088 		.driver_data = VMD_FEAT_HAS_MEMBAR_SHADOW |
1089 				VMD_FEAT_HAS_BUS_RESTRICTIONS |
1090 				VMD_FEAT_CAN_BYPASS_MSI_REMAP,},
1091 	{PCI_VDEVICE(INTEL, 0x467f),
1092 		.driver_data = VMD_FEATS_CLIENT,},
1093 	{PCI_VDEVICE(INTEL, 0x4c3d),
1094 		.driver_data = VMD_FEATS_CLIENT,},
1095 	{PCI_VDEVICE(INTEL, 0xa77f),
1096 		.driver_data = VMD_FEATS_CLIENT,},
1097 	{PCI_VDEVICE(INTEL, 0x7d0b),
1098 		.driver_data = VMD_FEATS_CLIENT,},
1099 	{PCI_VDEVICE(INTEL, 0xad0b),
1100 		.driver_data = VMD_FEATS_CLIENT,},
1101 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_VMD_9A0B),
1102 		.driver_data = VMD_FEATS_CLIENT,},
1103 	{0,}
1104 };
1105 MODULE_DEVICE_TABLE(pci, vmd_ids);
1106 
1107 static struct pci_driver vmd_drv = {
1108 	.name		= "vmd",
1109 	.id_table	= vmd_ids,
1110 	.probe		= vmd_probe,
1111 	.remove		= vmd_remove,
1112 	.shutdown	= vmd_shutdown,
1113 	.driver		= {
1114 		.pm	= &vmd_dev_pm_ops,
1115 	},
1116 };
1117 module_pci_driver(vmd_drv);
1118 
1119 MODULE_AUTHOR("Intel Corporation");
1120 MODULE_DESCRIPTION("Volume Management Device driver");
1121 MODULE_LICENSE("GPL v2");
1122 MODULE_VERSION("0.6");
1123