1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * PCI Message Signaled Interrupt (MSI)
4  *
5  * Copyright (C) 2003-2004 Intel
6  * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
7  * Copyright (C) 2016 Christoph Hellwig.
8  */
9 
10 #include <linux/err.h>
11 #include <linux/mm.h>
12 #include <linux/irq.h>
13 #include <linux/interrupt.h>
14 #include <linux/export.h>
15 #include <linux/ioport.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
18 #include <linux/msi.h>
19 #include <linux/smp.h>
20 #include <linux/errno.h>
21 #include <linux/io.h>
22 #include <linux/acpi_iort.h>
23 #include <linux/slab.h>
24 #include <linux/irqdomain.h>
25 #include <linux/of_irq.h>
26 
27 #include "pci.h"
28 
29 #ifdef CONFIG_PCI_MSI
30 
31 static int pci_msi_enable = 1;
32 int pci_msi_ignore_mask;
33 
34 #define msix_table_size(flags)	((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
35 
36 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
pci_msi_setup_msi_irqs(struct pci_dev * dev,int nvec,int type)37 static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
38 {
39 	struct irq_domain *domain;
40 
41 	domain = dev_get_msi_domain(&dev->dev);
42 	if (domain && irq_domain_is_hierarchy(domain))
43 		return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
44 
45 	return arch_setup_msi_irqs(dev, nvec, type);
46 }
47 
pci_msi_teardown_msi_irqs(struct pci_dev * dev)48 static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
49 {
50 	struct irq_domain *domain;
51 
52 	domain = dev_get_msi_domain(&dev->dev);
53 	if (domain && irq_domain_is_hierarchy(domain))
54 		msi_domain_free_irqs(domain, &dev->dev);
55 	else
56 		arch_teardown_msi_irqs(dev);
57 }
58 #else
59 #define pci_msi_setup_msi_irqs		arch_setup_msi_irqs
60 #define pci_msi_teardown_msi_irqs	arch_teardown_msi_irqs
61 #endif
62 
63 #ifdef CONFIG_PCI_MSI_ARCH_FALLBACKS
64 /* Arch hooks */
arch_setup_msi_irq(struct pci_dev * dev,struct msi_desc * desc)65 int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
66 {
67 	return -EINVAL;
68 }
69 
arch_teardown_msi_irq(unsigned int irq)70 void __weak arch_teardown_msi_irq(unsigned int irq)
71 {
72 }
73 
arch_setup_msi_irqs(struct pci_dev * dev,int nvec,int type)74 int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
75 {
76 	struct msi_desc *entry;
77 	int ret;
78 
79 	/*
80 	 * If an architecture wants to support multiple MSI, it needs to
81 	 * override arch_setup_msi_irqs()
82 	 */
83 	if (type == PCI_CAP_ID_MSI && nvec > 1)
84 		return 1;
85 
86 	for_each_pci_msi_entry(entry, dev) {
87 		ret = arch_setup_msi_irq(dev, entry);
88 		if (ret < 0)
89 			return ret;
90 		if (ret > 0)
91 			return -ENOSPC;
92 	}
93 
94 	return 0;
95 }
96 
arch_teardown_msi_irqs(struct pci_dev * dev)97 void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
98 {
99 	int i;
100 	struct msi_desc *entry;
101 
102 	for_each_pci_msi_entry(entry, dev)
103 		if (entry->irq)
104 			for (i = 0; i < entry->nvec_used; i++)
105 				arch_teardown_msi_irq(entry->irq + i);
106 }
107 #endif /* CONFIG_PCI_MSI_ARCH_FALLBACKS */
108 
default_restore_msi_irq(struct pci_dev * dev,int irq)109 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
110 {
111 	struct msi_desc *entry;
112 
113 	entry = NULL;
114 	if (dev->msix_enabled) {
115 		for_each_pci_msi_entry(entry, dev) {
116 			if (irq == entry->irq)
117 				break;
118 		}
119 	} else if (dev->msi_enabled)  {
120 		entry = irq_get_msi_desc(irq);
121 	}
122 
123 	if (entry)
124 		__pci_write_msi_msg(entry, &entry->msg);
125 }
126 
arch_restore_msi_irqs(struct pci_dev * dev)127 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
128 {
129 	return default_restore_msi_irqs(dev);
130 }
131 
132 /*
133  * PCI 2.3 does not specify mask bits for each MSI interrupt.  Attempting to
134  * mask all MSI interrupts by clearing the MSI enable bit does not work
135  * reliably as devices without an INTx disable bit will then generate a
136  * level IRQ which will never be cleared.
137  */
msi_multi_mask(struct msi_desc * desc)138 static inline __attribute_const__ u32 msi_multi_mask(struct msi_desc *desc)
139 {
140 	/* Don't shift by >= width of type */
141 	if (desc->msi_attrib.multi_cap >= 5)
142 		return 0xffffffff;
143 	return (1 << (1 << desc->msi_attrib.multi_cap)) - 1;
144 }
145 
pci_msi_update_mask(struct msi_desc * desc,u32 clear,u32 set)146 static noinline void pci_msi_update_mask(struct msi_desc *desc, u32 clear, u32 set)
147 {
148 	raw_spinlock_t *lock = &desc->dev->msi_lock;
149 	unsigned long flags;
150 
151 	raw_spin_lock_irqsave(lock, flags);
152 	desc->msi_mask &= ~clear;
153 	desc->msi_mask |= set;
154 	pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
155 			       desc->msi_mask);
156 	raw_spin_unlock_irqrestore(lock, flags);
157 }
158 
pci_msi_mask(struct msi_desc * desc,u32 mask)159 static inline void pci_msi_mask(struct msi_desc *desc, u32 mask)
160 {
161 	pci_msi_update_mask(desc, 0, mask);
162 }
163 
pci_msi_unmask(struct msi_desc * desc,u32 mask)164 static inline void pci_msi_unmask(struct msi_desc *desc, u32 mask)
165 {
166 	pci_msi_update_mask(desc, mask, 0);
167 }
168 
pci_msix_desc_addr(struct msi_desc * desc)169 static inline void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
170 {
171 	return desc->mask_base + desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
172 }
173 
174 /*
175  * This internal function does not flush PCI writes to the device.  All
176  * users must ensure that they read from the device before either assuming
177  * that the device state is up to date, or returning out of this file.
178  * It does not affect the msi_desc::msix_ctrl cache either. Use with care!
179  */
pci_msix_write_vector_ctrl(struct msi_desc * desc,u32 ctrl)180 static void pci_msix_write_vector_ctrl(struct msi_desc *desc, u32 ctrl)
181 {
182 	void __iomem *desc_addr = pci_msix_desc_addr(desc);
183 
184 	writel(ctrl, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
185 }
186 
pci_msix_mask(struct msi_desc * desc)187 static inline void pci_msix_mask(struct msi_desc *desc)
188 {
189 	desc->msix_ctrl |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
190 	pci_msix_write_vector_ctrl(desc, desc->msix_ctrl);
191 	/* Flush write to device */
192 	readl(desc->mask_base);
193 }
194 
pci_msix_unmask(struct msi_desc * desc)195 static inline void pci_msix_unmask(struct msi_desc *desc)
196 {
197 	desc->msix_ctrl &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
198 	pci_msix_write_vector_ctrl(desc, desc->msix_ctrl);
199 }
200 
__pci_msi_mask_desc(struct msi_desc * desc,u32 mask)201 static void __pci_msi_mask_desc(struct msi_desc *desc, u32 mask)
202 {
203 	if (pci_msi_ignore_mask || desc->msi_attrib.is_virtual)
204 		return;
205 
206 	if (desc->msi_attrib.is_msix)
207 		pci_msix_mask(desc);
208 	else if (desc->msi_attrib.maskbit)
209 		pci_msi_mask(desc, mask);
210 }
211 
__pci_msi_unmask_desc(struct msi_desc * desc,u32 mask)212 static void __pci_msi_unmask_desc(struct msi_desc *desc, u32 mask)
213 {
214 	if (pci_msi_ignore_mask || desc->msi_attrib.is_virtual)
215 		return;
216 
217 	if (desc->msi_attrib.is_msix)
218 		pci_msix_unmask(desc);
219 	else if (desc->msi_attrib.maskbit)
220 		pci_msi_unmask(desc, mask);
221 }
222 
223 /**
224  * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
225  * @data:	pointer to irqdata associated to that interrupt
226  */
pci_msi_mask_irq(struct irq_data * data)227 void pci_msi_mask_irq(struct irq_data *data)
228 {
229 	struct msi_desc *desc = irq_data_get_msi_desc(data);
230 
231 	__pci_msi_mask_desc(desc, BIT(data->irq - desc->irq));
232 }
233 EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
234 
235 /**
236  * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
237  * @data:	pointer to irqdata associated to that interrupt
238  */
pci_msi_unmask_irq(struct irq_data * data)239 void pci_msi_unmask_irq(struct irq_data *data)
240 {
241 	struct msi_desc *desc = irq_data_get_msi_desc(data);
242 
243 	__pci_msi_unmask_desc(desc, BIT(data->irq - desc->irq));
244 }
245 EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
246 
default_restore_msi_irqs(struct pci_dev * dev)247 void default_restore_msi_irqs(struct pci_dev *dev)
248 {
249 	struct msi_desc *entry;
250 
251 	for_each_pci_msi_entry(entry, dev)
252 		default_restore_msi_irq(dev, entry->irq);
253 }
254 
__pci_read_msi_msg(struct msi_desc * entry,struct msi_msg * msg)255 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
256 {
257 	struct pci_dev *dev = msi_desc_to_pci_dev(entry);
258 
259 	BUG_ON(dev->current_state != PCI_D0);
260 
261 	if (entry->msi_attrib.is_msix) {
262 		void __iomem *base = pci_msix_desc_addr(entry);
263 
264 		if (WARN_ON_ONCE(entry->msi_attrib.is_virtual))
265 			return;
266 
267 		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
268 		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
269 		msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
270 	} else {
271 		int pos = dev->msi_cap;
272 		u16 data;
273 
274 		pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
275 				      &msg->address_lo);
276 		if (entry->msi_attrib.is_64) {
277 			pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
278 					      &msg->address_hi);
279 			pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
280 		} else {
281 			msg->address_hi = 0;
282 			pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
283 		}
284 		msg->data = data;
285 	}
286 }
287 
__pci_write_msi_msg(struct msi_desc * entry,struct msi_msg * msg)288 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
289 {
290 	struct pci_dev *dev = msi_desc_to_pci_dev(entry);
291 
292 	if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
293 		/* Don't touch the hardware now */
294 	} else if (entry->msi_attrib.is_msix) {
295 		void __iomem *base = pci_msix_desc_addr(entry);
296 		u32 ctrl = entry->msix_ctrl;
297 		bool unmasked = !(ctrl & PCI_MSIX_ENTRY_CTRL_MASKBIT);
298 
299 		if (entry->msi_attrib.is_virtual)
300 			goto skip;
301 
302 		/*
303 		 * The specification mandates that the entry is masked
304 		 * when the message is modified:
305 		 *
306 		 * "If software changes the Address or Data value of an
307 		 * entry while the entry is unmasked, the result is
308 		 * undefined."
309 		 */
310 		if (unmasked)
311 			pci_msix_write_vector_ctrl(entry, ctrl | PCI_MSIX_ENTRY_CTRL_MASKBIT);
312 
313 		writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
314 		writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
315 		writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
316 
317 		if (unmasked)
318 			pci_msix_write_vector_ctrl(entry, ctrl);
319 
320 		/* Ensure that the writes are visible in the device */
321 		readl(base + PCI_MSIX_ENTRY_DATA);
322 	} else {
323 		int pos = dev->msi_cap;
324 		u16 msgctl;
325 
326 		pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
327 		msgctl &= ~PCI_MSI_FLAGS_QSIZE;
328 		msgctl |= entry->msi_attrib.multiple << 4;
329 		pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
330 
331 		pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
332 				       msg->address_lo);
333 		if (entry->msi_attrib.is_64) {
334 			pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
335 					       msg->address_hi);
336 			pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
337 					      msg->data);
338 		} else {
339 			pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
340 					      msg->data);
341 		}
342 		/* Ensure that the writes are visible in the device */
343 		pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
344 	}
345 
346 skip:
347 	entry->msg = *msg;
348 
349 	if (entry->write_msi_msg)
350 		entry->write_msi_msg(entry, entry->write_msi_msg_data);
351 
352 }
353 
pci_write_msi_msg(unsigned int irq,struct msi_msg * msg)354 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
355 {
356 	struct msi_desc *entry = irq_get_msi_desc(irq);
357 
358 	__pci_write_msi_msg(entry, msg);
359 }
360 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
361 
free_msi_irqs(struct pci_dev * dev)362 static void free_msi_irqs(struct pci_dev *dev)
363 {
364 	struct list_head *msi_list = dev_to_msi_list(&dev->dev);
365 	struct msi_desc *entry, *tmp;
366 	int i;
367 
368 	for_each_pci_msi_entry(entry, dev)
369 		if (entry->irq)
370 			for (i = 0; i < entry->nvec_used; i++)
371 				BUG_ON(irq_has_action(entry->irq + i));
372 
373 	pci_msi_teardown_msi_irqs(dev);
374 
375 	list_for_each_entry_safe(entry, tmp, msi_list, list) {
376 		if (entry->msi_attrib.is_msix) {
377 			if (list_is_last(&entry->list, msi_list))
378 				iounmap(entry->mask_base);
379 		}
380 
381 		list_del(&entry->list);
382 		free_msi_entry(entry);
383 	}
384 
385 	if (dev->msi_irq_groups) {
386 		msi_destroy_sysfs(&dev->dev, dev->msi_irq_groups);
387 		dev->msi_irq_groups = NULL;
388 	}
389 }
390 
pci_intx_for_msi(struct pci_dev * dev,int enable)391 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
392 {
393 	if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
394 		pci_intx(dev, enable);
395 }
396 
pci_msi_set_enable(struct pci_dev * dev,int enable)397 static void pci_msi_set_enable(struct pci_dev *dev, int enable)
398 {
399 	u16 control;
400 
401 	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
402 	control &= ~PCI_MSI_FLAGS_ENABLE;
403 	if (enable)
404 		control |= PCI_MSI_FLAGS_ENABLE;
405 	pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
406 }
407 
__pci_restore_msi_state(struct pci_dev * dev)408 static void __pci_restore_msi_state(struct pci_dev *dev)
409 {
410 	u16 control;
411 	struct msi_desc *entry;
412 
413 	if (!dev->msi_enabled)
414 		return;
415 
416 	entry = irq_get_msi_desc(dev->irq);
417 
418 	pci_intx_for_msi(dev, 0);
419 	pci_msi_set_enable(dev, 0);
420 	arch_restore_msi_irqs(dev);
421 
422 	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
423 	pci_msi_update_mask(entry, 0, 0);
424 	control &= ~PCI_MSI_FLAGS_QSIZE;
425 	control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
426 	pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
427 }
428 
pci_msix_clear_and_set_ctrl(struct pci_dev * dev,u16 clear,u16 set)429 static void pci_msix_clear_and_set_ctrl(struct pci_dev *dev, u16 clear, u16 set)
430 {
431 	u16 ctrl;
432 
433 	pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl);
434 	ctrl &= ~clear;
435 	ctrl |= set;
436 	pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, ctrl);
437 }
438 
__pci_restore_msix_state(struct pci_dev * dev)439 static void __pci_restore_msix_state(struct pci_dev *dev)
440 {
441 	struct msi_desc *entry;
442 
443 	if (!dev->msix_enabled)
444 		return;
445 	BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
446 
447 	/* route the table */
448 	pci_intx_for_msi(dev, 0);
449 	pci_msix_clear_and_set_ctrl(dev, 0,
450 				PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
451 
452 	arch_restore_msi_irqs(dev);
453 	for_each_pci_msi_entry(entry, dev)
454 		pci_msix_write_vector_ctrl(entry, entry->msix_ctrl);
455 
456 	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
457 }
458 
pci_restore_msi_state(struct pci_dev * dev)459 void pci_restore_msi_state(struct pci_dev *dev)
460 {
461 	__pci_restore_msi_state(dev);
462 	__pci_restore_msix_state(dev);
463 }
464 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
465 
466 static struct msi_desc *
msi_setup_entry(struct pci_dev * dev,int nvec,struct irq_affinity * affd)467 msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
468 {
469 	struct irq_affinity_desc *masks = NULL;
470 	struct msi_desc *entry;
471 	u16 control;
472 
473 	if (affd)
474 		masks = irq_create_affinity_masks(nvec, affd);
475 
476 	/* MSI Entry Initialization */
477 	entry = alloc_msi_entry(&dev->dev, nvec, masks);
478 	if (!entry)
479 		goto out;
480 
481 	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
482 
483 	entry->msi_attrib.is_msix	= 0;
484 	entry->msi_attrib.is_64		= !!(control & PCI_MSI_FLAGS_64BIT);
485 	entry->msi_attrib.is_virtual    = 0;
486 	entry->msi_attrib.entry_nr	= 0;
487 	entry->msi_attrib.maskbit	= !!(control & PCI_MSI_FLAGS_MASKBIT);
488 	entry->msi_attrib.default_irq	= dev->irq;	/* Save IOAPIC IRQ */
489 	entry->msi_attrib.multi_cap	= (control & PCI_MSI_FLAGS_QMASK) >> 1;
490 	entry->msi_attrib.multiple	= ilog2(__roundup_pow_of_two(nvec));
491 
492 	if (control & PCI_MSI_FLAGS_64BIT)
493 		entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
494 	else
495 		entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
496 
497 	/* Save the initial mask status */
498 	if (entry->msi_attrib.maskbit)
499 		pci_read_config_dword(dev, entry->mask_pos, &entry->msi_mask);
500 
501 out:
502 	kfree(masks);
503 	return entry;
504 }
505 
msi_verify_entries(struct pci_dev * dev)506 static int msi_verify_entries(struct pci_dev *dev)
507 {
508 	struct msi_desc *entry;
509 
510 	if (!dev->no_64bit_msi)
511 		return 0;
512 
513 	for_each_pci_msi_entry(entry, dev) {
514 		if (entry->msg.address_hi) {
515 			pci_err(dev, "arch assigned 64-bit MSI address %#x%08x but device only supports 32 bits\n",
516 				entry->msg.address_hi, entry->msg.address_lo);
517 			return -EIO;
518 		}
519 	}
520 	return 0;
521 }
522 
523 /**
524  * msi_capability_init - configure device's MSI capability structure
525  * @dev: pointer to the pci_dev data structure of MSI device function
526  * @nvec: number of interrupts to allocate
527  * @affd: description of automatic IRQ affinity assignments (may be %NULL)
528  *
529  * Setup the MSI capability structure of the device with the requested
530  * number of interrupts.  A return value of zero indicates the successful
531  * setup of an entry with the new MSI IRQ.  A negative return value indicates
532  * an error, and a positive return value indicates the number of interrupts
533  * which could have been allocated.
534  */
msi_capability_init(struct pci_dev * dev,int nvec,struct irq_affinity * affd)535 static int msi_capability_init(struct pci_dev *dev, int nvec,
536 			       struct irq_affinity *affd)
537 {
538 	const struct attribute_group **groups;
539 	struct msi_desc *entry;
540 	int ret;
541 
542 	pci_msi_set_enable(dev, 0);	/* Disable MSI during set up */
543 
544 	entry = msi_setup_entry(dev, nvec, affd);
545 	if (!entry)
546 		return -ENOMEM;
547 
548 	/* All MSIs are unmasked by default; mask them all */
549 	pci_msi_mask(entry, msi_multi_mask(entry));
550 
551 	list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
552 
553 	/* Configure MSI capability structure */
554 	ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
555 	if (ret)
556 		goto err;
557 
558 	ret = msi_verify_entries(dev);
559 	if (ret)
560 		goto err;
561 
562 	groups = msi_populate_sysfs(&dev->dev);
563 	if (IS_ERR(groups)) {
564 		ret = PTR_ERR(groups);
565 		goto err;
566 	}
567 
568 	dev->msi_irq_groups = groups;
569 
570 	/* Set MSI enabled bits	*/
571 	pci_intx_for_msi(dev, 0);
572 	pci_msi_set_enable(dev, 1);
573 	dev->msi_enabled = 1;
574 
575 	pcibios_free_irq(dev);
576 	dev->irq = entry->irq;
577 	return 0;
578 
579 err:
580 	pci_msi_unmask(entry, msi_multi_mask(entry));
581 	free_msi_irqs(dev);
582 	return ret;
583 }
584 
msix_map_region(struct pci_dev * dev,unsigned nr_entries)585 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
586 {
587 	resource_size_t phys_addr;
588 	u32 table_offset;
589 	unsigned long flags;
590 	u8 bir;
591 
592 	pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
593 			      &table_offset);
594 	bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
595 	flags = pci_resource_flags(dev, bir);
596 	if (!flags || (flags & IORESOURCE_UNSET))
597 		return NULL;
598 
599 	table_offset &= PCI_MSIX_TABLE_OFFSET;
600 	phys_addr = pci_resource_start(dev, bir) + table_offset;
601 
602 	return ioremap(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
603 }
604 
msix_setup_entries(struct pci_dev * dev,void __iomem * base,struct msix_entry * entries,int nvec,struct irq_affinity * affd)605 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
606 			      struct msix_entry *entries, int nvec,
607 			      struct irq_affinity *affd)
608 {
609 	struct irq_affinity_desc *curmsk, *masks = NULL;
610 	struct msi_desc *entry;
611 	void __iomem *addr;
612 	int ret, i;
613 	int vec_count = pci_msix_vec_count(dev);
614 
615 	if (affd)
616 		masks = irq_create_affinity_masks(nvec, affd);
617 
618 	for (i = 0, curmsk = masks; i < nvec; i++) {
619 		entry = alloc_msi_entry(&dev->dev, 1, curmsk);
620 		if (!entry) {
621 			if (!i)
622 				iounmap(base);
623 			else
624 				free_msi_irqs(dev);
625 			/* No enough memory. Don't try again */
626 			ret = -ENOMEM;
627 			goto out;
628 		}
629 
630 		entry->msi_attrib.is_msix	= 1;
631 		entry->msi_attrib.is_64		= 1;
632 
633 		if (entries)
634 			entry->msi_attrib.entry_nr = entries[i].entry;
635 		else
636 			entry->msi_attrib.entry_nr = i;
637 
638 		entry->msi_attrib.is_virtual =
639 			entry->msi_attrib.entry_nr >= vec_count;
640 
641 		entry->msi_attrib.default_irq	= dev->irq;
642 		entry->mask_base		= base;
643 
644 		if (!entry->msi_attrib.is_virtual) {
645 			addr = pci_msix_desc_addr(entry);
646 			entry->msix_ctrl = readl(addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
647 		}
648 
649 		list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
650 		if (masks)
651 			curmsk++;
652 	}
653 	ret = 0;
654 out:
655 	kfree(masks);
656 	return ret;
657 }
658 
msix_update_entries(struct pci_dev * dev,struct msix_entry * entries)659 static void msix_update_entries(struct pci_dev *dev, struct msix_entry *entries)
660 {
661 	struct msi_desc *entry;
662 
663 	for_each_pci_msi_entry(entry, dev) {
664 		if (entries) {
665 			entries->vector = entry->irq;
666 			entries++;
667 		}
668 	}
669 }
670 
msix_mask_all(void __iomem * base,int tsize)671 static void msix_mask_all(void __iomem *base, int tsize)
672 {
673 	u32 ctrl = PCI_MSIX_ENTRY_CTRL_MASKBIT;
674 	int i;
675 
676 	if (pci_msi_ignore_mask)
677 		return;
678 
679 	for (i = 0; i < tsize; i++, base += PCI_MSIX_ENTRY_SIZE)
680 		writel(ctrl, base + PCI_MSIX_ENTRY_VECTOR_CTRL);
681 }
682 
683 /**
684  * msix_capability_init - configure device's MSI-X capability
685  * @dev: pointer to the pci_dev data structure of MSI-X device function
686  * @entries: pointer to an array of struct msix_entry entries
687  * @nvec: number of @entries
688  * @affd: Optional pointer to enable automatic affinity assignment
689  *
690  * Setup the MSI-X capability structure of device function with a
691  * single MSI-X IRQ. A return of zero indicates the successful setup of
692  * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
693  **/
msix_capability_init(struct pci_dev * dev,struct msix_entry * entries,int nvec,struct irq_affinity * affd)694 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
695 				int nvec, struct irq_affinity *affd)
696 {
697 	const struct attribute_group **groups;
698 	void __iomem *base;
699 	int ret, tsize;
700 	u16 control;
701 
702 	/*
703 	 * Some devices require MSI-X to be enabled before the MSI-X
704 	 * registers can be accessed.  Mask all the vectors to prevent
705 	 * interrupts coming in before they're fully set up.
706 	 */
707 	pci_msix_clear_and_set_ctrl(dev, 0, PCI_MSIX_FLAGS_MASKALL |
708 				    PCI_MSIX_FLAGS_ENABLE);
709 
710 	pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
711 	/* Request & Map MSI-X table region */
712 	tsize = msix_table_size(control);
713 	base = msix_map_region(dev, tsize);
714 	if (!base) {
715 		ret = -ENOMEM;
716 		goto out_disable;
717 	}
718 
719 	/* Ensure that all table entries are masked. */
720 	msix_mask_all(base, tsize);
721 
722 	ret = msix_setup_entries(dev, base, entries, nvec, affd);
723 	if (ret)
724 		goto out_disable;
725 
726 	ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
727 	if (ret)
728 		goto out_avail;
729 
730 	/* Check if all MSI entries honor device restrictions */
731 	ret = msi_verify_entries(dev);
732 	if (ret)
733 		goto out_free;
734 
735 	msix_update_entries(dev, entries);
736 
737 	groups = msi_populate_sysfs(&dev->dev);
738 	if (IS_ERR(groups)) {
739 		ret = PTR_ERR(groups);
740 		goto out_free;
741 	}
742 
743 	dev->msi_irq_groups = groups;
744 
745 	/* Set MSI-X enabled bits and unmask the function */
746 	pci_intx_for_msi(dev, 0);
747 	dev->msix_enabled = 1;
748 	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
749 
750 	pcibios_free_irq(dev);
751 	return 0;
752 
753 out_avail:
754 	if (ret < 0) {
755 		/*
756 		 * If we had some success, report the number of IRQs
757 		 * we succeeded in setting up.
758 		 */
759 		struct msi_desc *entry;
760 		int avail = 0;
761 
762 		for_each_pci_msi_entry(entry, dev) {
763 			if (entry->irq != 0)
764 				avail++;
765 		}
766 		if (avail != 0)
767 			ret = avail;
768 	}
769 
770 out_free:
771 	free_msi_irqs(dev);
772 
773 out_disable:
774 	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
775 
776 	return ret;
777 }
778 
779 /**
780  * pci_msi_supported - check whether MSI may be enabled on a device
781  * @dev: pointer to the pci_dev data structure of MSI device function
782  * @nvec: how many MSIs have been requested?
783  *
784  * Look at global flags, the device itself, and its parent buses
785  * to determine if MSI/-X are supported for the device. If MSI/-X is
786  * supported return 1, else return 0.
787  **/
pci_msi_supported(struct pci_dev * dev,int nvec)788 static int pci_msi_supported(struct pci_dev *dev, int nvec)
789 {
790 	struct pci_bus *bus;
791 
792 	/* MSI must be globally enabled and supported by the device */
793 	if (!pci_msi_enable)
794 		return 0;
795 
796 	if (!dev || dev->no_msi)
797 		return 0;
798 
799 	/*
800 	 * You can't ask to have 0 or less MSIs configured.
801 	 *  a) it's stupid ..
802 	 *  b) the list manipulation code assumes nvec >= 1.
803 	 */
804 	if (nvec < 1)
805 		return 0;
806 
807 	/*
808 	 * Any bridge which does NOT route MSI transactions from its
809 	 * secondary bus to its primary bus must set NO_MSI flag on
810 	 * the secondary pci_bus.
811 	 *
812 	 * The NO_MSI flag can either be set directly by:
813 	 * - arch-specific PCI host bus controller drivers (deprecated)
814 	 * - quirks for specific PCI bridges
815 	 *
816 	 * or indirectly by platform-specific PCI host bridge drivers by
817 	 * advertising the 'msi_domain' property, which results in
818 	 * the NO_MSI flag when no MSI domain is found for this bridge
819 	 * at probe time.
820 	 */
821 	for (bus = dev->bus; bus; bus = bus->parent)
822 		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
823 			return 0;
824 
825 	return 1;
826 }
827 
828 /**
829  * pci_msi_vec_count - Return the number of MSI vectors a device can send
830  * @dev: device to report about
831  *
832  * This function returns the number of MSI vectors a device requested via
833  * Multiple Message Capable register. It returns a negative errno if the
834  * device is not capable sending MSI interrupts. Otherwise, the call succeeds
835  * and returns a power of two, up to a maximum of 2^5 (32), according to the
836  * MSI specification.
837  **/
pci_msi_vec_count(struct pci_dev * dev)838 int pci_msi_vec_count(struct pci_dev *dev)
839 {
840 	int ret;
841 	u16 msgctl;
842 
843 	if (!dev->msi_cap)
844 		return -EINVAL;
845 
846 	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
847 	ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
848 
849 	return ret;
850 }
851 EXPORT_SYMBOL(pci_msi_vec_count);
852 
pci_msi_shutdown(struct pci_dev * dev)853 static void pci_msi_shutdown(struct pci_dev *dev)
854 {
855 	struct msi_desc *desc;
856 
857 	if (!pci_msi_enable || !dev || !dev->msi_enabled)
858 		return;
859 
860 	BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
861 	desc = first_pci_msi_entry(dev);
862 
863 	pci_msi_set_enable(dev, 0);
864 	pci_intx_for_msi(dev, 1);
865 	dev->msi_enabled = 0;
866 
867 	/* Return the device with MSI unmasked as initial states */
868 	pci_msi_unmask(desc, msi_multi_mask(desc));
869 
870 	/* Restore dev->irq to its default pin-assertion IRQ */
871 	dev->irq = desc->msi_attrib.default_irq;
872 	pcibios_alloc_irq(dev);
873 }
874 
pci_disable_msi(struct pci_dev * dev)875 void pci_disable_msi(struct pci_dev *dev)
876 {
877 	if (!pci_msi_enable || !dev || !dev->msi_enabled)
878 		return;
879 
880 	pci_msi_shutdown(dev);
881 	free_msi_irqs(dev);
882 }
883 EXPORT_SYMBOL(pci_disable_msi);
884 
885 /**
886  * pci_msix_vec_count - return the number of device's MSI-X table entries
887  * @dev: pointer to the pci_dev data structure of MSI-X device function
888  * This function returns the number of device's MSI-X table entries and
889  * therefore the number of MSI-X vectors device is capable of sending.
890  * It returns a negative errno if the device is not capable of sending MSI-X
891  * interrupts.
892  **/
pci_msix_vec_count(struct pci_dev * dev)893 int pci_msix_vec_count(struct pci_dev *dev)
894 {
895 	u16 control;
896 
897 	if (!dev->msix_cap)
898 		return -EINVAL;
899 
900 	pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
901 	return msix_table_size(control);
902 }
903 EXPORT_SYMBOL(pci_msix_vec_count);
904 
__pci_enable_msix(struct pci_dev * dev,struct msix_entry * entries,int nvec,struct irq_affinity * affd,int flags)905 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
906 			     int nvec, struct irq_affinity *affd, int flags)
907 {
908 	int nr_entries;
909 	int i, j;
910 
911 	if (!pci_msi_supported(dev, nvec) || dev->current_state != PCI_D0)
912 		return -EINVAL;
913 
914 	nr_entries = pci_msix_vec_count(dev);
915 	if (nr_entries < 0)
916 		return nr_entries;
917 	if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
918 		return nr_entries;
919 
920 	if (entries) {
921 		/* Check for any invalid entries */
922 		for (i = 0; i < nvec; i++) {
923 			if (entries[i].entry >= nr_entries)
924 				return -EINVAL;		/* invalid entry */
925 			for (j = i + 1; j < nvec; j++) {
926 				if (entries[i].entry == entries[j].entry)
927 					return -EINVAL;	/* duplicate entry */
928 			}
929 		}
930 	}
931 
932 	/* Check whether driver already requested for MSI IRQ */
933 	if (dev->msi_enabled) {
934 		pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
935 		return -EINVAL;
936 	}
937 	return msix_capability_init(dev, entries, nvec, affd);
938 }
939 
pci_msix_shutdown(struct pci_dev * dev)940 static void pci_msix_shutdown(struct pci_dev *dev)
941 {
942 	struct msi_desc *entry;
943 
944 	if (!pci_msi_enable || !dev || !dev->msix_enabled)
945 		return;
946 
947 	if (pci_dev_is_disconnected(dev)) {
948 		dev->msix_enabled = 0;
949 		return;
950 	}
951 
952 	/* Return the device with MSI-X masked as initial states */
953 	for_each_pci_msi_entry(entry, dev)
954 		pci_msix_mask(entry);
955 
956 	pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
957 	pci_intx_for_msi(dev, 1);
958 	dev->msix_enabled = 0;
959 	pcibios_alloc_irq(dev);
960 }
961 
pci_disable_msix(struct pci_dev * dev)962 void pci_disable_msix(struct pci_dev *dev)
963 {
964 	if (!pci_msi_enable || !dev || !dev->msix_enabled)
965 		return;
966 
967 	pci_msix_shutdown(dev);
968 	free_msi_irqs(dev);
969 }
970 EXPORT_SYMBOL(pci_disable_msix);
971 
pci_no_msi(void)972 void pci_no_msi(void)
973 {
974 	pci_msi_enable = 0;
975 }
976 
977 /**
978  * pci_msi_enabled - is MSI enabled?
979  *
980  * Returns true if MSI has not been disabled by the command-line option
981  * pci=nomsi.
982  **/
pci_msi_enabled(void)983 int pci_msi_enabled(void)
984 {
985 	return pci_msi_enable;
986 }
987 EXPORT_SYMBOL(pci_msi_enabled);
988 
__pci_enable_msi_range(struct pci_dev * dev,int minvec,int maxvec,struct irq_affinity * affd)989 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
990 				  struct irq_affinity *affd)
991 {
992 	int nvec;
993 	int rc;
994 
995 	if (!pci_msi_supported(dev, minvec) || dev->current_state != PCI_D0)
996 		return -EINVAL;
997 
998 	/* Check whether driver already requested MSI-X IRQs */
999 	if (dev->msix_enabled) {
1000 		pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1001 		return -EINVAL;
1002 	}
1003 
1004 	if (maxvec < minvec)
1005 		return -ERANGE;
1006 
1007 	if (WARN_ON_ONCE(dev->msi_enabled))
1008 		return -EINVAL;
1009 
1010 	nvec = pci_msi_vec_count(dev);
1011 	if (nvec < 0)
1012 		return nvec;
1013 	if (nvec < minvec)
1014 		return -ENOSPC;
1015 
1016 	if (nvec > maxvec)
1017 		nvec = maxvec;
1018 
1019 	for (;;) {
1020 		if (affd) {
1021 			nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1022 			if (nvec < minvec)
1023 				return -ENOSPC;
1024 		}
1025 
1026 		rc = msi_capability_init(dev, nvec, affd);
1027 		if (rc == 0)
1028 			return nvec;
1029 
1030 		if (rc < 0)
1031 			return rc;
1032 		if (rc < minvec)
1033 			return -ENOSPC;
1034 
1035 		nvec = rc;
1036 	}
1037 }
1038 
1039 /* deprecated, don't use */
pci_enable_msi(struct pci_dev * dev)1040 int pci_enable_msi(struct pci_dev *dev)
1041 {
1042 	int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1043 	if (rc < 0)
1044 		return rc;
1045 	return 0;
1046 }
1047 EXPORT_SYMBOL(pci_enable_msi);
1048 
__pci_enable_msix_range(struct pci_dev * dev,struct msix_entry * entries,int minvec,int maxvec,struct irq_affinity * affd,int flags)1049 static int __pci_enable_msix_range(struct pci_dev *dev,
1050 				   struct msix_entry *entries, int minvec,
1051 				   int maxvec, struct irq_affinity *affd,
1052 				   int flags)
1053 {
1054 	int rc, nvec = maxvec;
1055 
1056 	if (maxvec < minvec)
1057 		return -ERANGE;
1058 
1059 	if (WARN_ON_ONCE(dev->msix_enabled))
1060 		return -EINVAL;
1061 
1062 	for (;;) {
1063 		if (affd) {
1064 			nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1065 			if (nvec < minvec)
1066 				return -ENOSPC;
1067 		}
1068 
1069 		rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1070 		if (rc == 0)
1071 			return nvec;
1072 
1073 		if (rc < 0)
1074 			return rc;
1075 		if (rc < minvec)
1076 			return -ENOSPC;
1077 
1078 		nvec = rc;
1079 	}
1080 }
1081 
1082 /**
1083  * pci_enable_msix_range - configure device's MSI-X capability structure
1084  * @dev: pointer to the pci_dev data structure of MSI-X device function
1085  * @entries: pointer to an array of MSI-X entries
1086  * @minvec: minimum number of MSI-X IRQs requested
1087  * @maxvec: maximum number of MSI-X IRQs requested
1088  *
1089  * Setup the MSI-X capability structure of device function with a maximum
1090  * possible number of interrupts in the range between @minvec and @maxvec
1091  * upon its software driver call to request for MSI-X mode enabled on its
1092  * hardware device function. It returns a negative errno if an error occurs.
1093  * If it succeeds, it returns the actual number of interrupts allocated and
1094  * indicates the successful configuration of MSI-X capability structure
1095  * with new allocated MSI-X interrupts.
1096  **/
pci_enable_msix_range(struct pci_dev * dev,struct msix_entry * entries,int minvec,int maxvec)1097 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1098 		int minvec, int maxvec)
1099 {
1100 	return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1101 }
1102 EXPORT_SYMBOL(pci_enable_msix_range);
1103 
1104 /**
1105  * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1106  * @dev:		PCI device to operate on
1107  * @min_vecs:		minimum number of vectors required (must be >= 1)
1108  * @max_vecs:		maximum (desired) number of vectors
1109  * @flags:		flags or quirks for the allocation
1110  * @affd:		optional description of the affinity requirements
1111  *
1112  * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1113  * vectors if available, and fall back to a single legacy vector
1114  * if neither is available.  Return the number of vectors allocated,
1115  * (which might be smaller than @max_vecs) if successful, or a negative
1116  * error code on error. If less than @min_vecs interrupt vectors are
1117  * available for @dev the function will fail with -ENOSPC.
1118  *
1119  * To get the Linux IRQ number used for a vector that can be passed to
1120  * request_irq() use the pci_irq_vector() helper.
1121  */
pci_alloc_irq_vectors_affinity(struct pci_dev * dev,unsigned int min_vecs,unsigned int max_vecs,unsigned int flags,struct irq_affinity * affd)1122 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1123 				   unsigned int max_vecs, unsigned int flags,
1124 				   struct irq_affinity *affd)
1125 {
1126 	struct irq_affinity msi_default_affd = {0};
1127 	int nvecs = -ENOSPC;
1128 
1129 	if (flags & PCI_IRQ_AFFINITY) {
1130 		if (!affd)
1131 			affd = &msi_default_affd;
1132 	} else {
1133 		if (WARN_ON(affd))
1134 			affd = NULL;
1135 	}
1136 
1137 	if (flags & PCI_IRQ_MSIX) {
1138 		nvecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1139 						affd, flags);
1140 		if (nvecs > 0)
1141 			return nvecs;
1142 	}
1143 
1144 	if (flags & PCI_IRQ_MSI) {
1145 		nvecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1146 		if (nvecs > 0)
1147 			return nvecs;
1148 	}
1149 
1150 	/* use legacy IRQ if allowed */
1151 	if (flags & PCI_IRQ_LEGACY) {
1152 		if (min_vecs == 1 && dev->irq) {
1153 			/*
1154 			 * Invoke the affinity spreading logic to ensure that
1155 			 * the device driver can adjust queue configuration
1156 			 * for the single interrupt case.
1157 			 */
1158 			if (affd)
1159 				irq_create_affinity_masks(1, affd);
1160 			pci_intx(dev, 1);
1161 			return 1;
1162 		}
1163 	}
1164 
1165 	return nvecs;
1166 }
1167 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1168 
1169 /**
1170  * pci_free_irq_vectors - free previously allocated IRQs for a device
1171  * @dev:		PCI device to operate on
1172  *
1173  * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1174  */
pci_free_irq_vectors(struct pci_dev * dev)1175 void pci_free_irq_vectors(struct pci_dev *dev)
1176 {
1177 	pci_disable_msix(dev);
1178 	pci_disable_msi(dev);
1179 }
1180 EXPORT_SYMBOL(pci_free_irq_vectors);
1181 
1182 /**
1183  * pci_irq_vector - return Linux IRQ number of a device vector
1184  * @dev: PCI device to operate on
1185  * @nr: device-relative interrupt vector index (0-based).
1186  */
pci_irq_vector(struct pci_dev * dev,unsigned int nr)1187 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1188 {
1189 	if (dev->msix_enabled) {
1190 		struct msi_desc *entry;
1191 		int i = 0;
1192 
1193 		for_each_pci_msi_entry(entry, dev) {
1194 			if (i == nr)
1195 				return entry->irq;
1196 			i++;
1197 		}
1198 		WARN_ON_ONCE(1);
1199 		return -EINVAL;
1200 	}
1201 
1202 	if (dev->msi_enabled) {
1203 		struct msi_desc *entry = first_pci_msi_entry(dev);
1204 
1205 		if (WARN_ON_ONCE(nr >= entry->nvec_used))
1206 			return -EINVAL;
1207 	} else {
1208 		if (WARN_ON_ONCE(nr > 0))
1209 			return -EINVAL;
1210 	}
1211 
1212 	return dev->irq + nr;
1213 }
1214 EXPORT_SYMBOL(pci_irq_vector);
1215 
1216 /**
1217  * pci_irq_get_affinity - return the affinity of a particular MSI vector
1218  * @dev:	PCI device to operate on
1219  * @nr:		device-relative interrupt vector index (0-based).
1220  */
pci_irq_get_affinity(struct pci_dev * dev,int nr)1221 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1222 {
1223 	if (dev->msix_enabled) {
1224 		struct msi_desc *entry;
1225 		int i = 0;
1226 
1227 		for_each_pci_msi_entry(entry, dev) {
1228 			if (i == nr)
1229 				return &entry->affinity->mask;
1230 			i++;
1231 		}
1232 		WARN_ON_ONCE(1);
1233 		return NULL;
1234 	} else if (dev->msi_enabled) {
1235 		struct msi_desc *entry = first_pci_msi_entry(dev);
1236 
1237 		if (WARN_ON_ONCE(!entry || !entry->affinity ||
1238 				 nr >= entry->nvec_used))
1239 			return NULL;
1240 
1241 		return &entry->affinity[nr].mask;
1242 	} else {
1243 		return cpu_possible_mask;
1244 	}
1245 }
1246 EXPORT_SYMBOL(pci_irq_get_affinity);
1247 
msi_desc_to_pci_dev(struct msi_desc * desc)1248 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1249 {
1250 	return to_pci_dev(desc->dev);
1251 }
1252 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1253 
msi_desc_to_pci_sysdata(struct msi_desc * desc)1254 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1255 {
1256 	struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1257 
1258 	return dev->bus->sysdata;
1259 }
1260 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1261 
1262 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1263 /**
1264  * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1265  * @irq_data:	Pointer to interrupt data of the MSI interrupt
1266  * @msg:	Pointer to the message
1267  */
pci_msi_domain_write_msg(struct irq_data * irq_data,struct msi_msg * msg)1268 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1269 {
1270 	struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1271 
1272 	/*
1273 	 * For MSI-X desc->irq is always equal to irq_data->irq. For
1274 	 * MSI only the first interrupt of MULTI MSI passes the test.
1275 	 */
1276 	if (desc->irq == irq_data->irq)
1277 		__pci_write_msi_msg(desc, msg);
1278 }
1279 
1280 /**
1281  * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1282  * @desc:	Pointer to the MSI descriptor
1283  *
1284  * The ID number is only used within the irqdomain.
1285  */
pci_msi_domain_calc_hwirq(struct msi_desc * desc)1286 static irq_hw_number_t pci_msi_domain_calc_hwirq(struct msi_desc *desc)
1287 {
1288 	struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1289 
1290 	return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1291 		pci_dev_id(dev) << 11 |
1292 		(pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1293 }
1294 
pci_msi_desc_is_multi_msi(struct msi_desc * desc)1295 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1296 {
1297 	return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1298 }
1299 
1300 /**
1301  * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1302  * 			      for @dev
1303  * @domain:	The interrupt domain to check
1304  * @info:	The domain info for verification
1305  * @dev:	The device to check
1306  *
1307  * Returns:
1308  *  0 if the functionality is supported
1309  *  1 if Multi MSI is requested, but the domain does not support it
1310  *  -ENOTSUPP otherwise
1311  */
pci_msi_domain_check_cap(struct irq_domain * domain,struct msi_domain_info * info,struct device * dev)1312 int pci_msi_domain_check_cap(struct irq_domain *domain,
1313 			     struct msi_domain_info *info, struct device *dev)
1314 {
1315 	struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1316 
1317 	/* Special handling to support __pci_enable_msi_range() */
1318 	if (pci_msi_desc_is_multi_msi(desc) &&
1319 	    !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1320 		return 1;
1321 	else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1322 		return -ENOTSUPP;
1323 
1324 	return 0;
1325 }
1326 
pci_msi_domain_handle_error(struct irq_domain * domain,struct msi_desc * desc,int error)1327 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1328 				       struct msi_desc *desc, int error)
1329 {
1330 	/* Special handling to support __pci_enable_msi_range() */
1331 	if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1332 		return 1;
1333 
1334 	return error;
1335 }
1336 
pci_msi_domain_set_desc(msi_alloc_info_t * arg,struct msi_desc * desc)1337 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1338 				    struct msi_desc *desc)
1339 {
1340 	arg->desc = desc;
1341 	arg->hwirq = pci_msi_domain_calc_hwirq(desc);
1342 }
1343 
1344 static struct msi_domain_ops pci_msi_domain_ops_default = {
1345 	.set_desc	= pci_msi_domain_set_desc,
1346 	.msi_check	= pci_msi_domain_check_cap,
1347 	.handle_error	= pci_msi_domain_handle_error,
1348 };
1349 
pci_msi_domain_update_dom_ops(struct msi_domain_info * info)1350 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1351 {
1352 	struct msi_domain_ops *ops = info->ops;
1353 
1354 	if (ops == NULL) {
1355 		info->ops = &pci_msi_domain_ops_default;
1356 	} else {
1357 		if (ops->set_desc == NULL)
1358 			ops->set_desc = pci_msi_domain_set_desc;
1359 		if (ops->msi_check == NULL)
1360 			ops->msi_check = pci_msi_domain_check_cap;
1361 		if (ops->handle_error == NULL)
1362 			ops->handle_error = pci_msi_domain_handle_error;
1363 	}
1364 }
1365 
pci_msi_domain_update_chip_ops(struct msi_domain_info * info)1366 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1367 {
1368 	struct irq_chip *chip = info->chip;
1369 
1370 	BUG_ON(!chip);
1371 	if (!chip->irq_write_msi_msg)
1372 		chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1373 	if (!chip->irq_mask)
1374 		chip->irq_mask = pci_msi_mask_irq;
1375 	if (!chip->irq_unmask)
1376 		chip->irq_unmask = pci_msi_unmask_irq;
1377 }
1378 
1379 /**
1380  * pci_msi_create_irq_domain - Create a MSI interrupt domain
1381  * @fwnode:	Optional fwnode of the interrupt controller
1382  * @info:	MSI domain info
1383  * @parent:	Parent irq domain
1384  *
1385  * Updates the domain and chip ops and creates a MSI interrupt domain.
1386  *
1387  * Returns:
1388  * A domain pointer or NULL in case of failure.
1389  */
pci_msi_create_irq_domain(struct fwnode_handle * fwnode,struct msi_domain_info * info,struct irq_domain * parent)1390 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1391 					     struct msi_domain_info *info,
1392 					     struct irq_domain *parent)
1393 {
1394 	struct irq_domain *domain;
1395 
1396 	if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1397 		info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1398 
1399 	if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1400 		pci_msi_domain_update_dom_ops(info);
1401 	if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1402 		pci_msi_domain_update_chip_ops(info);
1403 
1404 	info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1405 	if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1406 		info->flags |= MSI_FLAG_MUST_REACTIVATE;
1407 
1408 	/* PCI-MSI is oneshot-safe */
1409 	info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1410 
1411 	domain = msi_create_irq_domain(fwnode, info, parent);
1412 	if (!domain)
1413 		return NULL;
1414 
1415 	irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1416 	return domain;
1417 }
1418 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1419 
1420 /*
1421  * Users of the generic MSI infrastructure expect a device to have a single ID,
1422  * so with DMA aliases we have to pick the least-worst compromise. Devices with
1423  * DMA phantom functions tend to still emit MSIs from the real function number,
1424  * so we ignore those and only consider topological aliases where either the
1425  * alias device or RID appears on a different bus number. We also make the
1426  * reasonable assumption that bridges are walked in an upstream direction (so
1427  * the last one seen wins), and the much braver assumption that the most likely
1428  * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1429  * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1430  * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1431  * for taking ownership all we can really do is close our eyes and hope...
1432  */
get_msi_id_cb(struct pci_dev * pdev,u16 alias,void * data)1433 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1434 {
1435 	u32 *pa = data;
1436 	u8 bus = PCI_BUS_NUM(*pa);
1437 
1438 	if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1439 		*pa = alias;
1440 
1441 	return 0;
1442 }
1443 
1444 /**
1445  * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1446  * @domain:	The interrupt domain
1447  * @pdev:	The PCI device.
1448  *
1449  * The RID for a device is formed from the alias, with a firmware
1450  * supplied mapping applied
1451  *
1452  * Returns: The RID.
1453  */
pci_msi_domain_get_msi_rid(struct irq_domain * domain,struct pci_dev * pdev)1454 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1455 {
1456 	struct device_node *of_node;
1457 	u32 rid = pci_dev_id(pdev);
1458 
1459 	pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1460 
1461 	of_node = irq_domain_get_of_node(domain);
1462 	rid = of_node ? of_msi_map_id(&pdev->dev, of_node, rid) :
1463 			iort_msi_map_id(&pdev->dev, rid);
1464 
1465 	return rid;
1466 }
1467 
1468 /**
1469  * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1470  * @pdev:	The PCI device
1471  *
1472  * Use the firmware data to find a device-specific MSI domain
1473  * (i.e. not one that is set as a default).
1474  *
1475  * Returns: The corresponding MSI domain or NULL if none has been found.
1476  */
pci_msi_get_device_domain(struct pci_dev * pdev)1477 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1478 {
1479 	struct irq_domain *dom;
1480 	u32 rid = pci_dev_id(pdev);
1481 
1482 	pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1483 	dom = of_msi_map_get_device_domain(&pdev->dev, rid, DOMAIN_BUS_PCI_MSI);
1484 	if (!dom)
1485 		dom = iort_get_device_domain(&pdev->dev, rid,
1486 					     DOMAIN_BUS_PCI_MSI);
1487 	return dom;
1488 }
1489 
1490 /**
1491  * pci_dev_has_special_msi_domain - Check whether the device is handled by
1492  *				    a non-standard PCI-MSI domain
1493  * @pdev:	The PCI device to check.
1494  *
1495  * Returns: True if the device irqdomain or the bus irqdomain is
1496  * non-standard PCI/MSI.
1497  */
pci_dev_has_special_msi_domain(struct pci_dev * pdev)1498 bool pci_dev_has_special_msi_domain(struct pci_dev *pdev)
1499 {
1500 	struct irq_domain *dom = dev_get_msi_domain(&pdev->dev);
1501 
1502 	if (!dom)
1503 		dom = dev_get_msi_domain(&pdev->bus->dev);
1504 
1505 	if (!dom)
1506 		return true;
1507 
1508 	return dom->bus_token != DOMAIN_BUS_PCI_MSI;
1509 }
1510 
1511 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */
1512 #endif /* CONFIG_PCI_MSI */
1513 
pci_msi_init(struct pci_dev * dev)1514 void pci_msi_init(struct pci_dev *dev)
1515 {
1516 	u16 ctrl;
1517 
1518 	/*
1519 	 * Disable the MSI hardware to avoid screaming interrupts
1520 	 * during boot.  This is the power on reset default so
1521 	 * usually this should be a noop.
1522 	 */
1523 	dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1524 	if (!dev->msi_cap)
1525 		return;
1526 
1527 	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &ctrl);
1528 	if (ctrl & PCI_MSI_FLAGS_ENABLE)
1529 		pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS,
1530 				      ctrl & ~PCI_MSI_FLAGS_ENABLE);
1531 
1532 	if (!(ctrl & PCI_MSI_FLAGS_64BIT))
1533 		dev->no_64bit_msi = 1;
1534 }
1535 
pci_msix_init(struct pci_dev * dev)1536 void pci_msix_init(struct pci_dev *dev)
1537 {
1538 	u16 ctrl;
1539 
1540 	dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1541 	if (!dev->msix_cap)
1542 		return;
1543 
1544 	pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl);
1545 	if (ctrl & PCI_MSIX_FLAGS_ENABLE)
1546 		pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS,
1547 				      ctrl & ~PCI_MSIX_FLAGS_ENABLE);
1548 }
1549