xref: /linux/drivers/pci/controller/vmd.c (revision 561add0da6d3d07c9bccb0832fb6ed5619167d26)
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 /*
208  * XXX: Stubbed until we develop acceptable way to not create conflicts with
209  * other devices sharing the same vector.
210  */
211 static int vmd_irq_set_affinity(struct irq_data *data,
212 				const struct cpumask *dest, bool force)
213 {
214 	return -EINVAL;
215 }
216 
217 static struct irq_chip vmd_msi_controller = {
218 	.name			= "VMD-MSI",
219 	.irq_enable		= vmd_irq_enable,
220 	.irq_disable		= vmd_irq_disable,
221 	.irq_compose_msi_msg	= vmd_compose_msi_msg,
222 	.irq_set_affinity	= vmd_irq_set_affinity,
223 };
224 
225 static irq_hw_number_t vmd_get_hwirq(struct msi_domain_info *info,
226 				     msi_alloc_info_t *arg)
227 {
228 	return 0;
229 }
230 
231 /*
232  * XXX: We can be even smarter selecting the best IRQ once we solve the
233  * affinity problem.
234  */
235 static struct vmd_irq_list *vmd_next_irq(struct vmd_dev *vmd, struct msi_desc *desc)
236 {
237 	unsigned long flags;
238 	int i, best;
239 
240 	if (vmd->msix_count == 1 + vmd->first_vec)
241 		return &vmd->irqs[vmd->first_vec];
242 
243 	/*
244 	 * White list for fast-interrupt handlers. All others will share the
245 	 * "slow" interrupt vector.
246 	 */
247 	switch (msi_desc_to_pci_dev(desc)->class) {
248 	case PCI_CLASS_STORAGE_EXPRESS:
249 		break;
250 	default:
251 		return &vmd->irqs[vmd->first_vec];
252 	}
253 
254 	raw_spin_lock_irqsave(&list_lock, flags);
255 	best = vmd->first_vec + 1;
256 	for (i = best; i < vmd->msix_count; i++)
257 		if (vmd->irqs[i].count < vmd->irqs[best].count)
258 			best = i;
259 	vmd->irqs[best].count++;
260 	raw_spin_unlock_irqrestore(&list_lock, flags);
261 
262 	return &vmd->irqs[best];
263 }
264 
265 static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
266 			unsigned int virq, irq_hw_number_t hwirq,
267 			msi_alloc_info_t *arg)
268 {
269 	struct msi_desc *desc = arg->desc;
270 	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
271 	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
272 
273 	if (!vmdirq)
274 		return -ENOMEM;
275 
276 	INIT_LIST_HEAD(&vmdirq->node);
277 	vmdirq->irq = vmd_next_irq(vmd, desc);
278 	vmdirq->virq = virq;
279 
280 	irq_domain_set_info(domain, virq, vmdirq->irq->virq, info->chip, vmdirq,
281 			    handle_untracked_irq, vmd, NULL);
282 	return 0;
283 }
284 
285 static void vmd_msi_free(struct irq_domain *domain,
286 			struct msi_domain_info *info, unsigned int virq)
287 {
288 	struct vmd_irq *vmdirq = irq_get_chip_data(virq);
289 	unsigned long flags;
290 
291 	synchronize_srcu(&vmdirq->irq->srcu);
292 
293 	/* XXX: Potential optimization to rebalance */
294 	raw_spin_lock_irqsave(&list_lock, flags);
295 	vmdirq->irq->count--;
296 	raw_spin_unlock_irqrestore(&list_lock, flags);
297 
298 	kfree(vmdirq);
299 }
300 
301 static int vmd_msi_prepare(struct irq_domain *domain, struct device *dev,
302 			   int nvec, msi_alloc_info_t *arg)
303 {
304 	struct pci_dev *pdev = to_pci_dev(dev);
305 	struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
306 
307 	if (nvec > vmd->msix_count)
308 		return vmd->msix_count;
309 
310 	memset(arg, 0, sizeof(*arg));
311 	return 0;
312 }
313 
314 static void vmd_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
315 {
316 	arg->desc = desc;
317 }
318 
319 static struct msi_domain_ops vmd_msi_domain_ops = {
320 	.get_hwirq	= vmd_get_hwirq,
321 	.msi_init	= vmd_msi_init,
322 	.msi_free	= vmd_msi_free,
323 	.msi_prepare	= vmd_msi_prepare,
324 	.set_desc	= vmd_set_desc,
325 };
326 
327 static struct msi_domain_info vmd_msi_domain_info = {
328 	.flags		= MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
329 			  MSI_FLAG_PCI_MSIX,
330 	.ops		= &vmd_msi_domain_ops,
331 	.chip		= &vmd_msi_controller,
332 };
333 
334 static void vmd_set_msi_remapping(struct vmd_dev *vmd, bool enable)
335 {
336 	u16 reg;
337 
338 	pci_read_config_word(vmd->dev, PCI_REG_VMCONFIG, &reg);
339 	reg = enable ? (reg & ~VMCONFIG_MSI_REMAP) :
340 		       (reg | VMCONFIG_MSI_REMAP);
341 	pci_write_config_word(vmd->dev, PCI_REG_VMCONFIG, reg);
342 }
343 
344 static int vmd_create_irq_domain(struct vmd_dev *vmd)
345 {
346 	struct fwnode_handle *fn;
347 
348 	fn = irq_domain_alloc_named_id_fwnode("VMD-MSI", vmd->sysdata.domain);
349 	if (!fn)
350 		return -ENODEV;
351 
352 	vmd->irq_domain = pci_msi_create_irq_domain(fn, &vmd_msi_domain_info, NULL);
353 	if (!vmd->irq_domain) {
354 		irq_domain_free_fwnode(fn);
355 		return -ENODEV;
356 	}
357 
358 	return 0;
359 }
360 
361 static void vmd_remove_irq_domain(struct vmd_dev *vmd)
362 {
363 	/*
364 	 * Some production BIOS won't enable remapping between soft reboots.
365 	 * Ensure remapping is restored before unloading the driver.
366 	 */
367 	if (!vmd->msix_count)
368 		vmd_set_msi_remapping(vmd, true);
369 
370 	if (vmd->irq_domain) {
371 		struct fwnode_handle *fn = vmd->irq_domain->fwnode;
372 
373 		irq_domain_remove(vmd->irq_domain);
374 		irq_domain_free_fwnode(fn);
375 	}
376 }
377 
378 static void __iomem *vmd_cfg_addr(struct vmd_dev *vmd, struct pci_bus *bus,
379 				  unsigned int devfn, int reg, int len)
380 {
381 	unsigned int busnr_ecam = bus->number - vmd->busn_start;
382 	u32 offset = PCIE_ECAM_OFFSET(busnr_ecam, devfn, reg);
383 
384 	if (offset + len >= resource_size(&vmd->dev->resource[VMD_CFGBAR]))
385 		return NULL;
386 
387 	return vmd->cfgbar + offset;
388 }
389 
390 /*
391  * CPU may deadlock if config space is not serialized on some versions of this
392  * hardware, so all config space access is done under a spinlock.
393  */
394 static int vmd_pci_read(struct pci_bus *bus, unsigned int devfn, int reg,
395 			int len, u32 *value)
396 {
397 	struct vmd_dev *vmd = vmd_from_bus(bus);
398 	void __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
399 	unsigned long flags;
400 	int ret = 0;
401 
402 	if (!addr)
403 		return -EFAULT;
404 
405 	spin_lock_irqsave(&vmd->cfg_lock, flags);
406 	switch (len) {
407 	case 1:
408 		*value = readb(addr);
409 		break;
410 	case 2:
411 		*value = readw(addr);
412 		break;
413 	case 4:
414 		*value = readl(addr);
415 		break;
416 	default:
417 		ret = -EINVAL;
418 		break;
419 	}
420 	spin_unlock_irqrestore(&vmd->cfg_lock, flags);
421 	return ret;
422 }
423 
424 /*
425  * VMD h/w converts non-posted config writes to posted memory writes. The
426  * read-back in this function forces the completion so it returns only after
427  * the config space was written, as expected.
428  */
429 static int vmd_pci_write(struct pci_bus *bus, unsigned int devfn, int reg,
430 			 int len, u32 value)
431 {
432 	struct vmd_dev *vmd = vmd_from_bus(bus);
433 	void __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
434 	unsigned long flags;
435 	int ret = 0;
436 
437 	if (!addr)
438 		return -EFAULT;
439 
440 	spin_lock_irqsave(&vmd->cfg_lock, flags);
441 	switch (len) {
442 	case 1:
443 		writeb(value, addr);
444 		readb(addr);
445 		break;
446 	case 2:
447 		writew(value, addr);
448 		readw(addr);
449 		break;
450 	case 4:
451 		writel(value, addr);
452 		readl(addr);
453 		break;
454 	default:
455 		ret = -EINVAL;
456 		break;
457 	}
458 	spin_unlock_irqrestore(&vmd->cfg_lock, flags);
459 	return ret;
460 }
461 
462 static struct pci_ops vmd_ops = {
463 	.read		= vmd_pci_read,
464 	.write		= vmd_pci_write,
465 };
466 
467 #ifdef CONFIG_ACPI
468 static struct acpi_device *vmd_acpi_find_companion(struct pci_dev *pci_dev)
469 {
470 	struct pci_host_bridge *bridge;
471 	u32 busnr, addr;
472 
473 	if (pci_dev->bus->ops != &vmd_ops)
474 		return NULL;
475 
476 	bridge = pci_find_host_bridge(pci_dev->bus);
477 	busnr = pci_dev->bus->number - bridge->bus->number;
478 	/*
479 	 * The address computation below is only applicable to relative bus
480 	 * numbers below 32.
481 	 */
482 	if (busnr > 31)
483 		return NULL;
484 
485 	addr = (busnr << 24) | ((u32)pci_dev->devfn << 16) | 0x8000FFFFU;
486 
487 	dev_dbg(&pci_dev->dev, "Looking for ACPI companion (address 0x%x)\n",
488 		addr);
489 
490 	return acpi_find_child_device(ACPI_COMPANION(bridge->dev.parent), addr,
491 				      false);
492 }
493 
494 static bool hook_installed;
495 
496 static void vmd_acpi_begin(void)
497 {
498 	if (pci_acpi_set_companion_lookup_hook(vmd_acpi_find_companion))
499 		return;
500 
501 	hook_installed = true;
502 }
503 
504 static void vmd_acpi_end(void)
505 {
506 	if (!hook_installed)
507 		return;
508 
509 	pci_acpi_clear_companion_lookup_hook();
510 	hook_installed = false;
511 }
512 #else
513 static inline void vmd_acpi_begin(void) { }
514 static inline void vmd_acpi_end(void) { }
515 #endif /* CONFIG_ACPI */
516 
517 static void vmd_domain_reset(struct vmd_dev *vmd)
518 {
519 	u16 bus, max_buses = resource_size(&vmd->resources[0]);
520 	u8 dev, functions, fn, hdr_type;
521 	char __iomem *base;
522 
523 	for (bus = 0; bus < max_buses; bus++) {
524 		for (dev = 0; dev < 32; dev++) {
525 			base = vmd->cfgbar + PCIE_ECAM_OFFSET(bus,
526 						PCI_DEVFN(dev, 0), 0);
527 
528 			hdr_type = readb(base + PCI_HEADER_TYPE) &
529 					 PCI_HEADER_TYPE_MASK;
530 
531 			functions = (hdr_type & 0x80) ? 8 : 1;
532 			for (fn = 0; fn < functions; fn++) {
533 				base = vmd->cfgbar + PCIE_ECAM_OFFSET(bus,
534 						PCI_DEVFN(dev, fn), 0);
535 
536 				hdr_type = readb(base + PCI_HEADER_TYPE) &
537 						PCI_HEADER_TYPE_MASK;
538 
539 				if (hdr_type != PCI_HEADER_TYPE_BRIDGE ||
540 				    (readw(base + PCI_CLASS_DEVICE) !=
541 				     PCI_CLASS_BRIDGE_PCI))
542 					continue;
543 
544 				/*
545 				 * Temporarily disable the I/O range before updating
546 				 * PCI_IO_BASE.
547 				 */
548 				writel(0x0000ffff, base + PCI_IO_BASE_UPPER16);
549 				/* Update lower 16 bits of I/O base/limit */
550 				writew(0x00f0, base + PCI_IO_BASE);
551 				/* Update upper 16 bits of I/O base/limit */
552 				writel(0, base + PCI_IO_BASE_UPPER16);
553 
554 				/* MMIO Base/Limit */
555 				writel(0x0000fff0, base + PCI_MEMORY_BASE);
556 
557 				/* Prefetchable MMIO Base/Limit */
558 				writel(0, base + PCI_PREF_LIMIT_UPPER32);
559 				writel(0x0000fff0, base + PCI_PREF_MEMORY_BASE);
560 				writel(0xffffffff, base + PCI_PREF_BASE_UPPER32);
561 			}
562 		}
563 	}
564 }
565 
566 static void vmd_attach_resources(struct vmd_dev *vmd)
567 {
568 	vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
569 	vmd->dev->resource[VMD_MEMBAR2].child = &vmd->resources[2];
570 }
571 
572 static void vmd_detach_resources(struct vmd_dev *vmd)
573 {
574 	vmd->dev->resource[VMD_MEMBAR1].child = NULL;
575 	vmd->dev->resource[VMD_MEMBAR2].child = NULL;
576 }
577 
578 /*
579  * VMD domains start at 0x10000 to not clash with ACPI _SEG domains.
580  * Per ACPI r6.0, sec 6.5.6,  _SEG returns an integer, of which the lower
581  * 16 bits are the PCI Segment Group (domain) number.  Other bits are
582  * currently reserved.
583  */
584 static int vmd_find_free_domain(void)
585 {
586 	int domain = 0xffff;
587 	struct pci_bus *bus = NULL;
588 
589 	while ((bus = pci_find_next_bus(bus)) != NULL)
590 		domain = max_t(int, domain, pci_domain_nr(bus));
591 	return domain + 1;
592 }
593 
594 static int vmd_get_phys_offsets(struct vmd_dev *vmd, bool native_hint,
595 				resource_size_t *offset1,
596 				resource_size_t *offset2)
597 {
598 	struct pci_dev *dev = vmd->dev;
599 	u64 phys1, phys2;
600 
601 	if (native_hint) {
602 		u32 vmlock;
603 		int ret;
604 
605 		ret = pci_read_config_dword(dev, PCI_REG_VMLOCK, &vmlock);
606 		if (ret || PCI_POSSIBLE_ERROR(vmlock))
607 			return -ENODEV;
608 
609 		if (MB2_SHADOW_EN(vmlock)) {
610 			void __iomem *membar2;
611 
612 			membar2 = pci_iomap(dev, VMD_MEMBAR2, 0);
613 			if (!membar2)
614 				return -ENOMEM;
615 			phys1 = readq(membar2 + MB2_SHADOW_OFFSET);
616 			phys2 = readq(membar2 + MB2_SHADOW_OFFSET + 8);
617 			pci_iounmap(dev, membar2);
618 		} else
619 			return 0;
620 	} else {
621 		/* Hypervisor-Emulated Vendor-Specific Capability */
622 		int pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
623 		u32 reg, regu;
624 
625 		pci_read_config_dword(dev, pos + 4, &reg);
626 
627 		/* "SHDW" */
628 		if (pos && reg == 0x53484457) {
629 			pci_read_config_dword(dev, pos + 8, &reg);
630 			pci_read_config_dword(dev, pos + 12, &regu);
631 			phys1 = (u64) regu << 32 | reg;
632 
633 			pci_read_config_dword(dev, pos + 16, &reg);
634 			pci_read_config_dword(dev, pos + 20, &regu);
635 			phys2 = (u64) regu << 32 | reg;
636 		} else
637 			return 0;
638 	}
639 
640 	*offset1 = dev->resource[VMD_MEMBAR1].start -
641 			(phys1 & PCI_BASE_ADDRESS_MEM_MASK);
642 	*offset2 = dev->resource[VMD_MEMBAR2].start -
643 			(phys2 & PCI_BASE_ADDRESS_MEM_MASK);
644 
645 	return 0;
646 }
647 
648 static int vmd_get_bus_number_start(struct vmd_dev *vmd)
649 {
650 	struct pci_dev *dev = vmd->dev;
651 	u16 reg;
652 
653 	pci_read_config_word(dev, PCI_REG_VMCAP, &reg);
654 	if (BUS_RESTRICT_CAP(reg)) {
655 		pci_read_config_word(dev, PCI_REG_VMCONFIG, &reg);
656 
657 		switch (BUS_RESTRICT_CFG(reg)) {
658 		case 0:
659 			vmd->busn_start = 0;
660 			break;
661 		case 1:
662 			vmd->busn_start = 128;
663 			break;
664 		case 2:
665 			vmd->busn_start = 224;
666 			break;
667 		default:
668 			pci_err(dev, "Unknown Bus Offset Setting (%d)\n",
669 				BUS_RESTRICT_CFG(reg));
670 			return -ENODEV;
671 		}
672 	}
673 
674 	return 0;
675 }
676 
677 static irqreturn_t vmd_irq(int irq, void *data)
678 {
679 	struct vmd_irq_list *irqs = data;
680 	struct vmd_irq *vmdirq;
681 	int idx;
682 
683 	idx = srcu_read_lock(&irqs->srcu);
684 	list_for_each_entry_rcu(vmdirq, &irqs->irq_list, node)
685 		generic_handle_irq(vmdirq->virq);
686 	srcu_read_unlock(&irqs->srcu, idx);
687 
688 	return IRQ_HANDLED;
689 }
690 
691 static int vmd_alloc_irqs(struct vmd_dev *vmd)
692 {
693 	struct pci_dev *dev = vmd->dev;
694 	int i, err;
695 
696 	vmd->msix_count = pci_msix_vec_count(dev);
697 	if (vmd->msix_count < 0)
698 		return -ENODEV;
699 
700 	vmd->msix_count = pci_alloc_irq_vectors(dev, vmd->first_vec + 1,
701 						vmd->msix_count, PCI_IRQ_MSIX);
702 	if (vmd->msix_count < 0)
703 		return vmd->msix_count;
704 
705 	vmd->irqs = devm_kcalloc(&dev->dev, vmd->msix_count, sizeof(*vmd->irqs),
706 				 GFP_KERNEL);
707 	if (!vmd->irqs)
708 		return -ENOMEM;
709 
710 	for (i = 0; i < vmd->msix_count; i++) {
711 		err = init_srcu_struct(&vmd->irqs[i].srcu);
712 		if (err)
713 			return err;
714 
715 		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
716 		vmd->irqs[i].virq = pci_irq_vector(dev, i);
717 		err = devm_request_irq(&dev->dev, vmd->irqs[i].virq,
718 				       vmd_irq, IRQF_NO_THREAD,
719 				       vmd->name, &vmd->irqs[i]);
720 		if (err)
721 			return err;
722 	}
723 
724 	return 0;
725 }
726 
727 /*
728  * Since VMD is an aperture to regular PCIe root ports, only allow it to
729  * control features that the OS is allowed to control on the physical PCI bus.
730  */
731 static void vmd_copy_host_bridge_flags(struct pci_host_bridge *root_bridge,
732 				       struct pci_host_bridge *vmd_bridge)
733 {
734 	vmd_bridge->native_pcie_hotplug = root_bridge->native_pcie_hotplug;
735 	vmd_bridge->native_shpc_hotplug = root_bridge->native_shpc_hotplug;
736 	vmd_bridge->native_aer = root_bridge->native_aer;
737 	vmd_bridge->native_pme = root_bridge->native_pme;
738 	vmd_bridge->native_ltr = root_bridge->native_ltr;
739 	vmd_bridge->native_dpc = root_bridge->native_dpc;
740 }
741 
742 /*
743  * Enable ASPM and LTR settings on devices that aren't configured by BIOS.
744  */
745 static int vmd_pm_enable_quirk(struct pci_dev *pdev, void *userdata)
746 {
747 	unsigned long features = *(unsigned long *)userdata;
748 	u16 ltr = VMD_BIOS_PM_QUIRK_LTR;
749 	u32 ltr_reg;
750 	int pos;
751 
752 	if (!(features & VMD_FEAT_BIOS_PM_QUIRK))
753 		return 0;
754 
755 	pci_enable_link_state(pdev, PCIE_LINK_STATE_ALL);
756 
757 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_LTR);
758 	if (!pos)
759 		return 0;
760 
761 	/*
762 	 * Skip if the max snoop LTR is non-zero, indicating BIOS has set it
763 	 * so the LTR quirk is not needed.
764 	 */
765 	pci_read_config_dword(pdev, pos + PCI_LTR_MAX_SNOOP_LAT, &ltr_reg);
766 	if (!!(ltr_reg & (PCI_LTR_VALUE_MASK | PCI_LTR_SCALE_MASK)))
767 		return 0;
768 
769 	/*
770 	 * Set the default values to the maximum required by the platform to
771 	 * allow the deepest power management savings. Write as a DWORD where
772 	 * the lower word is the max snoop latency and the upper word is the
773 	 * max non-snoop latency.
774 	 */
775 	ltr_reg = (ltr << 16) | ltr;
776 	pci_write_config_dword(pdev, pos + PCI_LTR_MAX_SNOOP_LAT, ltr_reg);
777 	pci_info(pdev, "VMD: Default LTR value set by driver\n");
778 
779 	return 0;
780 }
781 
782 static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
783 {
784 	struct pci_sysdata *sd = &vmd->sysdata;
785 	struct resource *res;
786 	u32 upper_bits;
787 	unsigned long flags;
788 	LIST_HEAD(resources);
789 	resource_size_t offset[2] = {0};
790 	resource_size_t membar2_offset = 0x2000;
791 	struct pci_bus *child;
792 	struct pci_dev *dev;
793 	int ret;
794 
795 	/*
796 	 * Shadow registers may exist in certain VMD device ids which allow
797 	 * guests to correctly assign host physical addresses to the root ports
798 	 * and child devices. These registers will either return the host value
799 	 * or 0, depending on an enable bit in the VMD device.
800 	 */
801 	if (features & VMD_FEAT_HAS_MEMBAR_SHADOW) {
802 		membar2_offset = MB2_SHADOW_OFFSET + MB2_SHADOW_SIZE;
803 		ret = vmd_get_phys_offsets(vmd, true, &offset[0], &offset[1]);
804 		if (ret)
805 			return ret;
806 	} else if (features & VMD_FEAT_HAS_MEMBAR_SHADOW_VSCAP) {
807 		ret = vmd_get_phys_offsets(vmd, false, &offset[0], &offset[1]);
808 		if (ret)
809 			return ret;
810 	}
811 
812 	/*
813 	 * Certain VMD devices may have a root port configuration option which
814 	 * limits the bus range to between 0-127, 128-255, or 224-255
815 	 */
816 	if (features & VMD_FEAT_HAS_BUS_RESTRICTIONS) {
817 		ret = vmd_get_bus_number_start(vmd);
818 		if (ret)
819 			return ret;
820 	}
821 
822 	res = &vmd->dev->resource[VMD_CFGBAR];
823 	vmd->resources[0] = (struct resource) {
824 		.name  = "VMD CFGBAR",
825 		.start = vmd->busn_start,
826 		.end   = vmd->busn_start + (resource_size(res) >> 20) - 1,
827 		.flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED,
828 	};
829 
830 	/*
831 	 * If the window is below 4GB, clear IORESOURCE_MEM_64 so we can
832 	 * put 32-bit resources in the window.
833 	 *
834 	 * There's no hardware reason why a 64-bit window *couldn't*
835 	 * contain a 32-bit resource, but pbus_size_mem() computes the
836 	 * bridge window size assuming a 64-bit window will contain no
837 	 * 32-bit resources.  __pci_assign_resource() enforces that
838 	 * artificial restriction to make sure everything will fit.
839 	 *
840 	 * The only way we could use a 64-bit non-prefetchable MEMBAR is
841 	 * if its address is <4GB so that we can convert it to a 32-bit
842 	 * resource.  To be visible to the host OS, all VMD endpoints must
843 	 * be initially configured by platform BIOS, which includes setting
844 	 * up these resources.  We can assume the device is configured
845 	 * according to the platform needs.
846 	 */
847 	res = &vmd->dev->resource[VMD_MEMBAR1];
848 	upper_bits = upper_32_bits(res->end);
849 	flags = res->flags & ~IORESOURCE_SIZEALIGN;
850 	if (!upper_bits)
851 		flags &= ~IORESOURCE_MEM_64;
852 	vmd->resources[1] = (struct resource) {
853 		.name  = "VMD MEMBAR1",
854 		.start = res->start,
855 		.end   = res->end,
856 		.flags = flags,
857 		.parent = res,
858 	};
859 
860 	res = &vmd->dev->resource[VMD_MEMBAR2];
861 	upper_bits = upper_32_bits(res->end);
862 	flags = res->flags & ~IORESOURCE_SIZEALIGN;
863 	if (!upper_bits)
864 		flags &= ~IORESOURCE_MEM_64;
865 	vmd->resources[2] = (struct resource) {
866 		.name  = "VMD MEMBAR2",
867 		.start = res->start + membar2_offset,
868 		.end   = res->end,
869 		.flags = flags,
870 		.parent = res,
871 	};
872 
873 	sd->vmd_dev = vmd->dev;
874 	sd->domain = vmd_find_free_domain();
875 	if (sd->domain < 0)
876 		return sd->domain;
877 
878 	sd->node = pcibus_to_node(vmd->dev->bus);
879 
880 	/*
881 	 * Currently MSI remapping must be enabled in guest passthrough mode
882 	 * due to some missing interrupt remapping plumbing. This is probably
883 	 * acceptable because the guest is usually CPU-limited and MSI
884 	 * remapping doesn't become a performance bottleneck.
885 	 */
886 	if (!(features & VMD_FEAT_CAN_BYPASS_MSI_REMAP) ||
887 	    offset[0] || offset[1]) {
888 		ret = vmd_alloc_irqs(vmd);
889 		if (ret)
890 			return ret;
891 
892 		vmd_set_msi_remapping(vmd, true);
893 
894 		ret = vmd_create_irq_domain(vmd);
895 		if (ret)
896 			return ret;
897 
898 		/*
899 		 * Override the IRQ domain bus token so the domain can be
900 		 * distinguished from a regular PCI/MSI domain.
901 		 */
902 		irq_domain_update_bus_token(vmd->irq_domain, DOMAIN_BUS_VMD_MSI);
903 	} else {
904 		vmd_set_msi_remapping(vmd, false);
905 	}
906 
907 	pci_add_resource(&resources, &vmd->resources[0]);
908 	pci_add_resource_offset(&resources, &vmd->resources[1], offset[0]);
909 	pci_add_resource_offset(&resources, &vmd->resources[2], offset[1]);
910 
911 	vmd->bus = pci_create_root_bus(&vmd->dev->dev, vmd->busn_start,
912 				       &vmd_ops, sd, &resources);
913 	if (!vmd->bus) {
914 		pci_free_resource_list(&resources);
915 		vmd_remove_irq_domain(vmd);
916 		return -ENODEV;
917 	}
918 
919 	vmd_copy_host_bridge_flags(pci_find_host_bridge(vmd->dev->bus),
920 				   to_pci_host_bridge(vmd->bus->bridge));
921 
922 	vmd_attach_resources(vmd);
923 	if (vmd->irq_domain)
924 		dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
925 	else
926 		dev_set_msi_domain(&vmd->bus->dev,
927 				   dev_get_msi_domain(&vmd->dev->dev));
928 
929 	vmd_acpi_begin();
930 
931 	pci_scan_child_bus(vmd->bus);
932 	vmd_domain_reset(vmd);
933 
934 	/* When Intel VMD is enabled, the OS does not discover the Root Ports
935 	 * owned by Intel VMD within the MMCFG space. pci_reset_bus() applies
936 	 * a reset to the parent of the PCI device supplied as argument. This
937 	 * is why we pass a child device, so the reset can be triggered at
938 	 * the Intel bridge level and propagated to all the children in the
939 	 * hierarchy.
940 	 */
941 	list_for_each_entry(child, &vmd->bus->children, node) {
942 		if (!list_empty(&child->devices)) {
943 			dev = list_first_entry(&child->devices,
944 					       struct pci_dev, bus_list);
945 			ret = pci_reset_bus(dev);
946 			if (ret)
947 				pci_warn(dev, "can't reset device: %d\n", ret);
948 
949 			break;
950 		}
951 	}
952 
953 	pci_assign_unassigned_bus_resources(vmd->bus);
954 
955 	pci_walk_bus(vmd->bus, vmd_pm_enable_quirk, &features);
956 
957 	/*
958 	 * VMD root buses are virtual and don't return true on pci_is_pcie()
959 	 * and will fail pcie_bus_configure_settings() early. It can instead be
960 	 * run on each of the real root ports.
961 	 */
962 	list_for_each_entry(child, &vmd->bus->children, node)
963 		pcie_bus_configure_settings(child);
964 
965 	pci_bus_add_devices(vmd->bus);
966 
967 	vmd_acpi_end();
968 
969 	WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
970 			       "domain"), "Can't create symlink to domain\n");
971 	return 0;
972 }
973 
974 static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
975 {
976 	unsigned long features = (unsigned long) id->driver_data;
977 	struct vmd_dev *vmd;
978 	int err;
979 
980 	if (resource_size(&dev->resource[VMD_CFGBAR]) < (1 << 20))
981 		return -ENOMEM;
982 
983 	vmd = devm_kzalloc(&dev->dev, sizeof(*vmd), GFP_KERNEL);
984 	if (!vmd)
985 		return -ENOMEM;
986 
987 	vmd->dev = dev;
988 	vmd->instance = ida_simple_get(&vmd_instance_ida, 0, 0, GFP_KERNEL);
989 	if (vmd->instance < 0)
990 		return vmd->instance;
991 
992 	vmd->name = devm_kasprintf(&dev->dev, GFP_KERNEL, "vmd%d",
993 				   vmd->instance);
994 	if (!vmd->name) {
995 		err = -ENOMEM;
996 		goto out_release_instance;
997 	}
998 
999 	err = pcim_enable_device(dev);
1000 	if (err < 0)
1001 		goto out_release_instance;
1002 
1003 	vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
1004 	if (!vmd->cfgbar) {
1005 		err = -ENOMEM;
1006 		goto out_release_instance;
1007 	}
1008 
1009 	pci_set_master(dev);
1010 	if (dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(64)) &&
1011 	    dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(32))) {
1012 		err = -ENODEV;
1013 		goto out_release_instance;
1014 	}
1015 
1016 	if (features & VMD_FEAT_OFFSET_FIRST_VECTOR)
1017 		vmd->first_vec = 1;
1018 
1019 	spin_lock_init(&vmd->cfg_lock);
1020 	pci_set_drvdata(dev, vmd);
1021 	err = vmd_enable_domain(vmd, features);
1022 	if (err)
1023 		goto out_release_instance;
1024 
1025 	dev_info(&vmd->dev->dev, "Bound to PCI domain %04x\n",
1026 		 vmd->sysdata.domain);
1027 	return 0;
1028 
1029  out_release_instance:
1030 	ida_simple_remove(&vmd_instance_ida, vmd->instance);
1031 	return err;
1032 }
1033 
1034 static void vmd_cleanup_srcu(struct vmd_dev *vmd)
1035 {
1036 	int i;
1037 
1038 	for (i = 0; i < vmd->msix_count; i++)
1039 		cleanup_srcu_struct(&vmd->irqs[i].srcu);
1040 }
1041 
1042 static void vmd_remove(struct pci_dev *dev)
1043 {
1044 	struct vmd_dev *vmd = pci_get_drvdata(dev);
1045 
1046 	sysfs_remove_link(&vmd->dev->dev.kobj, "domain");
1047 	pci_stop_root_bus(vmd->bus);
1048 	pci_remove_root_bus(vmd->bus);
1049 	vmd_cleanup_srcu(vmd);
1050 	vmd_detach_resources(vmd);
1051 	vmd_remove_irq_domain(vmd);
1052 	ida_simple_remove(&vmd_instance_ida, vmd->instance);
1053 }
1054 
1055 static void vmd_shutdown(struct pci_dev *dev)
1056 {
1057         struct vmd_dev *vmd = pci_get_drvdata(dev);
1058 
1059         vmd_remove_irq_domain(vmd);
1060 }
1061 
1062 #ifdef CONFIG_PM_SLEEP
1063 static int vmd_suspend(struct device *dev)
1064 {
1065 	struct pci_dev *pdev = to_pci_dev(dev);
1066 	struct vmd_dev *vmd = pci_get_drvdata(pdev);
1067 	int i;
1068 
1069 	for (i = 0; i < vmd->msix_count; i++)
1070 		devm_free_irq(dev, vmd->irqs[i].virq, &vmd->irqs[i]);
1071 
1072 	return 0;
1073 }
1074 
1075 static int vmd_resume(struct device *dev)
1076 {
1077 	struct pci_dev *pdev = to_pci_dev(dev);
1078 	struct vmd_dev *vmd = pci_get_drvdata(pdev);
1079 	int err, i;
1080 
1081        if (vmd->irq_domain)
1082                vmd_set_msi_remapping(vmd, true);
1083        else
1084                vmd_set_msi_remapping(vmd, false);
1085 
1086 	for (i = 0; i < vmd->msix_count; i++) {
1087 		err = devm_request_irq(dev, vmd->irqs[i].virq,
1088 				       vmd_irq, IRQF_NO_THREAD,
1089 				       vmd->name, &vmd->irqs[i]);
1090 		if (err)
1091 			return err;
1092 	}
1093 
1094 	return 0;
1095 }
1096 #endif
1097 static SIMPLE_DEV_PM_OPS(vmd_dev_pm_ops, vmd_suspend, vmd_resume);
1098 
1099 static const struct pci_device_id vmd_ids[] = {
1100 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_VMD_201D),
1101 		.driver_data = VMD_FEAT_HAS_MEMBAR_SHADOW_VSCAP,},
1102 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_VMD_28C0),
1103 		.driver_data = VMD_FEAT_HAS_MEMBAR_SHADOW |
1104 				VMD_FEAT_HAS_BUS_RESTRICTIONS |
1105 				VMD_FEAT_CAN_BYPASS_MSI_REMAP,},
1106 	{PCI_VDEVICE(INTEL, 0x467f),
1107 		.driver_data = VMD_FEATS_CLIENT,},
1108 	{PCI_VDEVICE(INTEL, 0x4c3d),
1109 		.driver_data = VMD_FEATS_CLIENT,},
1110 	{PCI_VDEVICE(INTEL, 0xa77f),
1111 		.driver_data = VMD_FEATS_CLIENT,},
1112 	{PCI_VDEVICE(INTEL, 0x7d0b),
1113 		.driver_data = VMD_FEATS_CLIENT,},
1114 	{PCI_VDEVICE(INTEL, 0xad0b),
1115 		.driver_data = VMD_FEATS_CLIENT,},
1116 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_VMD_9A0B),
1117 		.driver_data = VMD_FEATS_CLIENT,},
1118 	{0,}
1119 };
1120 MODULE_DEVICE_TABLE(pci, vmd_ids);
1121 
1122 static struct pci_driver vmd_drv = {
1123 	.name		= "vmd",
1124 	.id_table	= vmd_ids,
1125 	.probe		= vmd_probe,
1126 	.remove		= vmd_remove,
1127 	.shutdown	= vmd_shutdown,
1128 	.driver		= {
1129 		.pm	= &vmd_dev_pm_ops,
1130 	},
1131 };
1132 module_pci_driver(vmd_drv);
1133 
1134 MODULE_AUTHOR("Intel Corporation");
1135 MODULE_LICENSE("GPL v2");
1136 MODULE_VERSION("0.6");
1137