1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * PCI MSI/MSI-X — Exported APIs for device drivers
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  * Copyright (C) 2022 Linutronix GmbH
9  */
10 
11 #include <linux/export.h>
12 #include <linux/irq.h>
13 
14 #include "msi.h"
15 
16 /**
17  * pci_enable_msi() - Enable MSI interrupt mode on device
18  * @dev: the PCI device to operate on
19  *
20  * Legacy device driver API to enable MSI interrupts mode on device and
21  * allocate a single interrupt vector. On success, the allocated vector
22  * Linux IRQ will be saved at @dev->irq. The driver must invoke
23  * pci_disable_msi() on cleanup.
24  *
25  * NOTE: The newer pci_alloc_irq_vectors() / pci_free_irq_vectors() API
26  * pair should, in general, be used instead.
27  *
28  * Return: 0 on success, errno otherwise
29  */
30 int pci_enable_msi(struct pci_dev *dev)
31 {
32 	int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
33 	if (rc < 0)
34 		return rc;
35 	return 0;
36 }
37 EXPORT_SYMBOL(pci_enable_msi);
38 
39 /**
40  * pci_disable_msi() - Disable MSI interrupt mode on device
41  * @dev: the PCI device to operate on
42  *
43  * Legacy device driver API to disable MSI interrupt mode on device,
44  * free earlier allocated interrupt vectors, and restore INTx emulation.
45  * The PCI device Linux IRQ (@dev->irq) is restored to its default
46  * pin-assertion IRQ. This is the cleanup pair of pci_enable_msi().
47  *
48  * NOTE: The newer pci_alloc_irq_vectors() / pci_free_irq_vectors() API
49  * pair should, in general, be used instead.
50  */
51 void pci_disable_msi(struct pci_dev *dev)
52 {
53 	if (!pci_msi_enabled() || !dev || !dev->msi_enabled)
54 		return;
55 
56 	guard(msi_descs_lock)(&dev->dev);
57 	pci_msi_shutdown(dev);
58 	pci_free_msi_irqs(dev);
59 }
60 EXPORT_SYMBOL(pci_disable_msi);
61 
62 /**
63  * pci_msix_vec_count() - Get number of MSI-X interrupt vectors on device
64  * @dev: the PCI device to operate on
65  *
66  * Return: number of MSI-X interrupt vectors available on this device
67  * (i.e., the device's MSI-X capability structure "table size"), -EINVAL
68  * if the device is not MSI-X capable, other errnos otherwise.
69  */
70 int pci_msix_vec_count(struct pci_dev *dev)
71 {
72 	u16 control;
73 
74 	if (!dev->msix_cap)
75 		return -EINVAL;
76 
77 	pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
78 	return msix_table_size(control);
79 }
80 EXPORT_SYMBOL(pci_msix_vec_count);
81 
82 /**
83  * pci_enable_msix_range() - Enable MSI-X interrupt mode on device
84  * @dev:     the PCI device to operate on
85  * @entries: input/output parameter, array of MSI-X configuration entries
86  * @minvec:  minimum required number of MSI-X vectors
87  * @maxvec:  maximum desired number of MSI-X vectors
88  *
89  * Legacy device driver API to enable MSI-X interrupt mode on device and
90  * configure its MSI-X capability structure as appropriate.  The passed
91  * @entries array must have each of its members "entry" field set to a
92  * desired (valid) MSI-X vector number, where the range of valid MSI-X
93  * vector numbers can be queried through pci_msix_vec_count().  If
94  * successful, the driver must invoke pci_disable_msix() on cleanup.
95  *
96  * NOTE: The newer pci_alloc_irq_vectors() / pci_free_irq_vectors() API
97  * pair should, in general, be used instead.
98  *
99  * Return: number of MSI-X vectors allocated (which might be smaller
100  * than @maxvecs), where Linux IRQ numbers for such allocated vectors
101  * are saved back in the @entries array elements' "vector" field. Return
102  * -ENOSPC if less than @minvecs interrupt vectors are available.
103  * Return -EINVAL if one of the passed @entries members "entry" field
104  * was invalid or a duplicate, or if plain MSI interrupts mode was
105  * earlier enabled on device. Return other errnos otherwise.
106  */
107 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
108 			  int minvec, int maxvec)
109 {
110 	return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
111 }
112 EXPORT_SYMBOL(pci_enable_msix_range);
113 
114 /**
115  * pci_msix_can_alloc_dyn - Query whether dynamic allocation after enabling
116  *			    MSI-X is supported
117  *
118  * @dev:	PCI device to operate on
119  *
120  * Return: True if supported, false otherwise
121  */
122 bool pci_msix_can_alloc_dyn(struct pci_dev *dev)
123 {
124 	if (!dev->msix_cap)
125 		return false;
126 
127 	return pci_msi_domain_supports(dev, MSI_FLAG_PCI_MSIX_ALLOC_DYN, DENY_LEGACY);
128 }
129 EXPORT_SYMBOL_GPL(pci_msix_can_alloc_dyn);
130 
131 /**
132  * pci_msix_alloc_irq_at - Allocate an MSI-X interrupt after enabling MSI-X
133  *			   at a given MSI-X vector index or any free vector index
134  *
135  * @dev:	PCI device to operate on
136  * @index:	Index to allocate. If @index == MSI_ANY_INDEX this allocates
137  *		the next free index in the MSI-X table
138  * @affdesc:	Optional pointer to an affinity descriptor structure. NULL otherwise
139  *
140  * Return: A struct msi_map
141  *
142  *	On success msi_map::index contains the allocated index (>= 0) and
143  *	msi_map::virq contains the allocated Linux interrupt number (> 0).
144  *
145  *	On fail msi_map::index contains the error code and msi_map::virq
146  *	is set to 0.
147  */
148 struct msi_map pci_msix_alloc_irq_at(struct pci_dev *dev, unsigned int index,
149 				     const struct irq_affinity_desc *affdesc)
150 {
151 	struct msi_map map = { .index = -ENOTSUPP };
152 
153 	if (!dev->msix_enabled)
154 		return map;
155 
156 	if (!pci_msix_can_alloc_dyn(dev))
157 		return map;
158 
159 	return msi_domain_alloc_irq_at(&dev->dev, MSI_DEFAULT_DOMAIN, index, affdesc, NULL);
160 }
161 EXPORT_SYMBOL_GPL(pci_msix_alloc_irq_at);
162 
163 /**
164  * pci_msix_free_irq - Free an interrupt on a PCI/MSI-X interrupt domain
165  *
166  * @dev:	The PCI device to operate on
167  * @map:	A struct msi_map describing the interrupt to free
168  *
169  * Undo an interrupt vector allocation. Does not disable MSI-X.
170  */
171 void pci_msix_free_irq(struct pci_dev *dev, struct msi_map map)
172 {
173 	if (WARN_ON_ONCE(map.index < 0 || map.virq <= 0))
174 		return;
175 	if (WARN_ON_ONCE(!pci_msix_can_alloc_dyn(dev)))
176 		return;
177 	msi_domain_free_irqs_range(&dev->dev, MSI_DEFAULT_DOMAIN, map.index, map.index);
178 }
179 EXPORT_SYMBOL_GPL(pci_msix_free_irq);
180 
181 /**
182  * pci_disable_msix() - Disable MSI-X interrupt mode on device
183  * @dev: the PCI device to operate on
184  *
185  * Legacy device driver API to disable MSI-X interrupt mode on device,
186  * free earlier-allocated interrupt vectors, and restore INTx.
187  * The PCI device Linux IRQ (@dev->irq) is restored to its default pin
188  * assertion IRQ. This is the cleanup pair of pci_enable_msix_range().
189  *
190  * NOTE: The newer pci_alloc_irq_vectors() / pci_free_irq_vectors() API
191  * pair should, in general, be used instead.
192  */
193 void pci_disable_msix(struct pci_dev *dev)
194 {
195 	if (!pci_msi_enabled() || !dev || !dev->msix_enabled)
196 		return;
197 
198 	guard(msi_descs_lock)(&dev->dev);
199 	pci_msix_shutdown(dev);
200 	pci_free_msi_irqs(dev);
201 }
202 EXPORT_SYMBOL(pci_disable_msix);
203 
204 /**
205  * pci_alloc_irq_vectors() - Allocate multiple device interrupt vectors
206  * @dev:      the PCI device to operate on
207  * @min_vecs: minimum required number of vectors (must be >= 1)
208  * @max_vecs: maximum desired number of vectors
209  * @flags:    One or more of:
210  *
211  *            * %PCI_IRQ_MSIX      Allow trying MSI-X vector allocations
212  *            * %PCI_IRQ_MSI       Allow trying MSI vector allocations
213  *
214  *            * %PCI_IRQ_INTX      Allow trying INTx interrupts, if and
215  *              only if @min_vecs == 1
216  *
217  *            * %PCI_IRQ_AFFINITY  Auto-manage IRQs affinity by spreading
218  *              the vectors around available CPUs
219  *
220  * Allocate up to @max_vecs interrupt vectors on device. MSI-X irq
221  * vector allocation has a higher precedence over plain MSI, which has a
222  * higher precedence over legacy INTx emulation.
223  *
224  * Upon a successful allocation, the caller should use pci_irq_vector()
225  * to get the Linux IRQ number to be passed to request_threaded_irq().
226  * The driver must call pci_free_irq_vectors() on cleanup.
227  *
228  * Return: number of allocated vectors (which might be smaller than
229  * @max_vecs), -ENOSPC if less than @min_vecs interrupt vectors are
230  * available, other errnos otherwise.
231  */
232 int pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
233 			  unsigned int max_vecs, unsigned int flags)
234 {
235 	return pci_alloc_irq_vectors_affinity(dev, min_vecs, max_vecs,
236 					      flags, NULL);
237 }
238 EXPORT_SYMBOL(pci_alloc_irq_vectors);
239 
240 /**
241  * pci_alloc_irq_vectors_affinity() - Allocate multiple device interrupt
242  *                                    vectors with affinity requirements
243  * @dev:      the PCI device to operate on
244  * @min_vecs: minimum required number of vectors (must be >= 1)
245  * @max_vecs: maximum desired number of vectors
246  * @flags:    allocation flags, as in pci_alloc_irq_vectors()
247  * @affd:     affinity requirements (can be %NULL).
248  *
249  * Same as pci_alloc_irq_vectors(), but with the extra @affd parameter.
250  * Check that function docs, and &struct irq_affinity, for more details.
251  */
252 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
253 				   unsigned int max_vecs, unsigned int flags,
254 				   struct irq_affinity *affd)
255 {
256 	struct irq_affinity msi_default_affd = {0};
257 	int nvecs = -ENOSPC;
258 
259 	if (flags & PCI_IRQ_AFFINITY) {
260 		if (!affd)
261 			affd = &msi_default_affd;
262 	} else {
263 		if (WARN_ON(affd))
264 			affd = NULL;
265 	}
266 
267 	if (flags & PCI_IRQ_MSIX) {
268 		nvecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
269 						affd, flags);
270 		if (nvecs > 0)
271 			return nvecs;
272 	}
273 
274 	if (flags & PCI_IRQ_MSI) {
275 		nvecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
276 		if (nvecs > 0)
277 			return nvecs;
278 	}
279 
280 	/* use INTx IRQ if allowed */
281 	if (flags & PCI_IRQ_INTX) {
282 		if (min_vecs == 1 && dev->irq) {
283 			/*
284 			 * Invoke the affinity spreading logic to ensure that
285 			 * the device driver can adjust queue configuration
286 			 * for the single interrupt case.
287 			 */
288 			if (affd)
289 				irq_create_affinity_masks(1, affd);
290 			pci_intx(dev, 1);
291 			return 1;
292 		}
293 	}
294 
295 	return nvecs;
296 }
297 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
298 
299 /**
300  * pci_irq_vector() - Get Linux IRQ number of a device interrupt vector
301  * @dev: the PCI device to operate on
302  * @nr:  device-relative interrupt vector index (0-based); has different
303  *       meanings, depending on interrupt mode:
304  *
305  *         * MSI-X     the index in the MSI-X vector table
306  *         * MSI       the index of the enabled MSI vectors
307  *         * INTx      must be 0
308  *
309  * Return: the Linux IRQ number, or -EINVAL if @nr is out of range
310  */
311 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
312 {
313 	unsigned int irq;
314 
315 	if (!dev->msi_enabled && !dev->msix_enabled)
316 		return !nr ? dev->irq : -EINVAL;
317 
318 	irq = msi_get_virq(&dev->dev, nr);
319 	return irq ? irq : -EINVAL;
320 }
321 EXPORT_SYMBOL(pci_irq_vector);
322 
323 /**
324  * pci_irq_get_affinity() - Get a device interrupt vector affinity
325  * @dev: the PCI device to operate on
326  * @nr:  device-relative interrupt vector index (0-based); has different
327  *       meanings, depending on interrupt mode:
328  *
329  *         * MSI-X     the index in the MSI-X vector table
330  *         * MSI       the index of the enabled MSI vectors
331  *         * INTx      must be 0
332  *
333  * Return: MSI/MSI-X vector affinity, NULL if @nr is out of range or if
334  * the MSI(-X) vector was allocated without explicit affinity
335  * requirements (e.g., by pci_enable_msi(), pci_enable_msix_range(), or
336  * pci_alloc_irq_vectors() without the %PCI_IRQ_AFFINITY flag). Return a
337  * generic set of CPU IDs representing all possible CPUs available
338  * during system boot if the device is in legacy INTx mode.
339  */
340 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
341 {
342 	int idx, irq = pci_irq_vector(dev, nr);
343 	struct msi_desc *desc;
344 
345 	if (WARN_ON_ONCE(irq <= 0))
346 		return NULL;
347 
348 	desc = irq_get_msi_desc(irq);
349 	/* Non-MSI does not have the information handy */
350 	if (!desc)
351 		return cpu_possible_mask;
352 
353 	/* MSI[X] interrupts can be allocated without affinity descriptor */
354 	if (!desc->affinity)
355 		return NULL;
356 
357 	/*
358 	 * MSI has a mask array in the descriptor.
359 	 * MSI-X has a single mask.
360 	 */
361 	idx = dev->msi_enabled ? nr : 0;
362 	return &desc->affinity[idx].mask;
363 }
364 EXPORT_SYMBOL(pci_irq_get_affinity);
365 
366 /**
367  * pci_free_irq_vectors() - Free previously allocated IRQs for a device
368  * @dev: the PCI device to operate on
369  *
370  * Undo the interrupt vector allocations and possible device MSI/MSI-X
371  * enablement earlier done through pci_alloc_irq_vectors_affinity() or
372  * pci_alloc_irq_vectors().
373  */
374 void pci_free_irq_vectors(struct pci_dev *dev)
375 {
376 	pci_disable_msix(dev);
377 	pci_disable_msi(dev);
378 }
379 EXPORT_SYMBOL(pci_free_irq_vectors);
380 
381 /**
382  * pci_restore_msi_state() - Restore cached MSI(-X) state on device
383  * @dev: the PCI device to operate on
384  *
385  * Write the Linux-cached MSI(-X) state back on device. This is
386  * typically useful upon system resume, or after an error-recovery PCI
387  * adapter reset.
388  */
389 void pci_restore_msi_state(struct pci_dev *dev)
390 {
391 	__pci_restore_msi_state(dev);
392 	__pci_restore_msix_state(dev);
393 }
394 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
395 
396 /**
397  * pci_msi_enabled() - Are MSI(-X) interrupts enabled system-wide?
398  *
399  * Return: true if MSI has not been globally disabled through ACPI FADT,
400  * PCI bridge quirks, or the "pci=nomsi" kernel command-line option.
401  */
402 bool pci_msi_enabled(void)
403 {
404 	return pci_msi_enable;
405 }
406 EXPORT_SYMBOL(pci_msi_enabled);
407