xref: /kernel/linux/linux-5.10/arch/x86/kernel/apic/msi.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Support of MSI, HPET and DMAR interrupts.
4 *
5 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
6 *	Moved from arch/x86/kernel/apic/io_apic.c.
7 * Jiang Liu <jiang.liu@linux.intel.com>
8 *	Convert to hierarchical irqdomain
9 */
10#include <linux/mm.h>
11#include <linux/interrupt.h>
12#include <linux/irq.h>
13#include <linux/pci.h>
14#include <linux/dmar.h>
15#include <linux/hpet.h>
16#include <linux/msi.h>
17#include <asm/irqdomain.h>
18#include <asm/msidef.h>
19#include <asm/hpet.h>
20#include <asm/hw_irq.h>
21#include <asm/apic.h>
22#include <asm/irq_remapping.h>
23
24struct irq_domain *x86_pci_msi_default_domain __ro_after_init;
25
26static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg)
27{
28	msg->address_hi = MSI_ADDR_BASE_HI;
29
30	if (x2apic_enabled())
31		msg->address_hi |= MSI_ADDR_EXT_DEST_ID(cfg->dest_apicid);
32
33	msg->address_lo =
34		MSI_ADDR_BASE_LO |
35		((apic->irq_dest_mode == 0) ?
36			MSI_ADDR_DEST_MODE_PHYSICAL :
37			MSI_ADDR_DEST_MODE_LOGICAL) |
38		MSI_ADDR_REDIRECTION_CPU |
39		MSI_ADDR_DEST_ID(cfg->dest_apicid);
40
41	msg->data =
42		MSI_DATA_TRIGGER_EDGE |
43		MSI_DATA_LEVEL_ASSERT |
44		MSI_DATA_DELIVERY_FIXED |
45		MSI_DATA_VECTOR(cfg->vector);
46}
47
48void x86_vector_msi_compose_msg(struct irq_data *data, struct msi_msg *msg)
49{
50	__irq_msi_compose_msg(irqd_cfg(data), msg);
51}
52
53static void irq_msi_update_msg(struct irq_data *irqd, struct irq_cfg *cfg)
54{
55	struct msi_msg msg[2] = { [1] = { }, };
56
57	__irq_msi_compose_msg(cfg, msg);
58	irq_data_get_irq_chip(irqd)->irq_write_msi_msg(irqd, msg);
59}
60
61static int
62msi_set_affinity(struct irq_data *irqd, const struct cpumask *mask, bool force)
63{
64	struct irq_cfg old_cfg, *cfg = irqd_cfg(irqd);
65	struct irq_data *parent = irqd->parent_data;
66	unsigned int cpu;
67	int ret;
68
69	/* Save the current configuration */
70	cpu = cpumask_first(irq_data_get_effective_affinity_mask(irqd));
71	old_cfg = *cfg;
72
73	/* Allocate a new target vector */
74	ret = parent->chip->irq_set_affinity(parent, mask, force);
75	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
76		return ret;
77
78	/*
79	 * For non-maskable and non-remapped MSI interrupts the migration
80	 * to a different destination CPU and a different vector has to be
81	 * done careful to handle the possible stray interrupt which can be
82	 * caused by the non-atomic update of the address/data pair.
83	 *
84	 * Direct update is possible when:
85	 * - The MSI is maskable (remapped MSI does not use this code path)).
86	 *   The quirk bit is not set in this case.
87	 * - The new vector is the same as the old vector
88	 * - The old vector is MANAGED_IRQ_SHUTDOWN_VECTOR (interrupt starts up)
89	 * - The interrupt is not yet started up
90	 * - The new destination CPU is the same as the old destination CPU
91	 */
92	if (!irqd_msi_nomask_quirk(irqd) ||
93	    cfg->vector == old_cfg.vector ||
94	    old_cfg.vector == MANAGED_IRQ_SHUTDOWN_VECTOR ||
95	    !irqd_is_started(irqd) ||
96	    cfg->dest_apicid == old_cfg.dest_apicid) {
97		irq_msi_update_msg(irqd, cfg);
98		return ret;
99	}
100
101	/*
102	 * Paranoia: Validate that the interrupt target is the local
103	 * CPU.
104	 */
105	if (WARN_ON_ONCE(cpu != smp_processor_id())) {
106		irq_msi_update_msg(irqd, cfg);
107		return ret;
108	}
109
110	/*
111	 * Redirect the interrupt to the new vector on the current CPU
112	 * first. This might cause a spurious interrupt on this vector if
113	 * the device raises an interrupt right between this update and the
114	 * update to the final destination CPU.
115	 *
116	 * If the vector is in use then the installed device handler will
117	 * denote it as spurious which is no harm as this is a rare event
118	 * and interrupt handlers have to cope with spurious interrupts
119	 * anyway. If the vector is unused, then it is marked so it won't
120	 * trigger the 'No irq handler for vector' warning in
121	 * common_interrupt().
122	 *
123	 * This requires to hold vector lock to prevent concurrent updates to
124	 * the affected vector.
125	 */
126	lock_vector_lock();
127
128	/*
129	 * Mark the new target vector on the local CPU if it is currently
130	 * unused. Reuse the VECTOR_RETRIGGERED state which is also used in
131	 * the CPU hotplug path for a similar purpose. This cannot be
132	 * undone here as the current CPU has interrupts disabled and
133	 * cannot handle the interrupt before the whole set_affinity()
134	 * section is done. In the CPU unplug case, the current CPU is
135	 * about to vanish and will not handle any interrupts anymore. The
136	 * vector is cleaned up when the CPU comes online again.
137	 */
138	if (IS_ERR_OR_NULL(this_cpu_read(vector_irq[cfg->vector])))
139		this_cpu_write(vector_irq[cfg->vector], VECTOR_RETRIGGERED);
140
141	/* Redirect it to the new vector on the local CPU temporarily */
142	old_cfg.vector = cfg->vector;
143	irq_msi_update_msg(irqd, &old_cfg);
144
145	/* Now transition it to the target CPU */
146	irq_msi_update_msg(irqd, cfg);
147
148	/*
149	 * All interrupts after this point are now targeted at the new
150	 * vector/CPU.
151	 *
152	 * Drop vector lock before testing whether the temporary assignment
153	 * to the local CPU was hit by an interrupt raised in the device,
154	 * because the retrigger function acquires vector lock again.
155	 */
156	unlock_vector_lock();
157
158	/*
159	 * Check whether the transition raced with a device interrupt and
160	 * is pending in the local APICs IRR. It is safe to do this outside
161	 * of vector lock as the irq_desc::lock of this interrupt is still
162	 * held and interrupts are disabled: The check is not accessing the
163	 * underlying vector store. It's just checking the local APIC's
164	 * IRR.
165	 */
166	if (lapic_vector_set_in_irr(cfg->vector))
167		irq_data_get_irq_chip(irqd)->irq_retrigger(irqd);
168
169	return ret;
170}
171
172/*
173 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
174 * which implement the MSI or MSI-X Capability Structure.
175 */
176static struct irq_chip pci_msi_controller = {
177	.name			= "PCI-MSI",
178	.irq_unmask		= pci_msi_unmask_irq,
179	.irq_mask		= pci_msi_mask_irq,
180	.irq_ack		= irq_chip_ack_parent,
181	.irq_retrigger		= irq_chip_retrigger_hierarchy,
182	.irq_set_affinity	= msi_set_affinity,
183	.flags			= IRQCHIP_SKIP_SET_WAKE |
184				  IRQCHIP_AFFINITY_PRE_STARTUP,
185};
186
187int pci_msi_prepare(struct irq_domain *domain, struct device *dev, int nvec,
188		    msi_alloc_info_t *arg)
189{
190	struct pci_dev *pdev = to_pci_dev(dev);
191	struct msi_desc *desc = first_pci_msi_entry(pdev);
192
193	init_irq_alloc_info(arg, NULL);
194	if (desc->msi_attrib.is_msix) {
195		arg->type = X86_IRQ_ALLOC_TYPE_PCI_MSIX;
196	} else {
197		arg->type = X86_IRQ_ALLOC_TYPE_PCI_MSI;
198		arg->flags |= X86_IRQ_ALLOC_CONTIGUOUS_VECTORS;
199	}
200
201	return 0;
202}
203EXPORT_SYMBOL_GPL(pci_msi_prepare);
204
205static struct msi_domain_ops pci_msi_domain_ops = {
206	.msi_prepare	= pci_msi_prepare,
207};
208
209static struct msi_domain_info pci_msi_domain_info = {
210	.flags		= MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
211			  MSI_FLAG_PCI_MSIX,
212	.ops		= &pci_msi_domain_ops,
213	.chip		= &pci_msi_controller,
214	.handler	= handle_edge_irq,
215	.handler_name	= "edge",
216};
217
218struct irq_domain * __init native_create_pci_msi_domain(void)
219{
220	struct fwnode_handle *fn;
221	struct irq_domain *d;
222
223	if (disable_apic)
224		return NULL;
225
226	fn = irq_domain_alloc_named_fwnode("PCI-MSI");
227	if (!fn)
228		return NULL;
229
230	d = pci_msi_create_irq_domain(fn, &pci_msi_domain_info,
231				      x86_vector_domain);
232	if (!d) {
233		irq_domain_free_fwnode(fn);
234		pr_warn("Failed to initialize PCI-MSI irqdomain.\n");
235	} else {
236		d->flags |= IRQ_DOMAIN_MSI_NOMASK_QUIRK;
237	}
238	return d;
239}
240
241void __init x86_create_pci_msi_domain(void)
242{
243	x86_pci_msi_default_domain = x86_init.irqs.create_pci_msi_domain();
244}
245
246#ifdef CONFIG_IRQ_REMAP
247static struct irq_chip pci_msi_ir_controller = {
248	.name			= "IR-PCI-MSI",
249	.irq_unmask		= pci_msi_unmask_irq,
250	.irq_mask		= pci_msi_mask_irq,
251	.irq_ack		= irq_chip_ack_parent,
252	.irq_retrigger		= irq_chip_retrigger_hierarchy,
253	.flags			= IRQCHIP_SKIP_SET_WAKE |
254				  IRQCHIP_AFFINITY_PRE_STARTUP,
255};
256
257static struct msi_domain_info pci_msi_ir_domain_info = {
258	.flags		= MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
259			  MSI_FLAG_MULTI_PCI_MSI | MSI_FLAG_PCI_MSIX,
260	.ops		= &pci_msi_domain_ops,
261	.chip		= &pci_msi_ir_controller,
262	.handler	= handle_edge_irq,
263	.handler_name	= "edge",
264};
265
266struct irq_domain *arch_create_remap_msi_irq_domain(struct irq_domain *parent,
267						    const char *name, int id)
268{
269	struct fwnode_handle *fn;
270	struct irq_domain *d;
271
272	fn = irq_domain_alloc_named_id_fwnode(name, id);
273	if (!fn)
274		return NULL;
275	d = pci_msi_create_irq_domain(fn, &pci_msi_ir_domain_info, parent);
276	if (!d)
277		irq_domain_free_fwnode(fn);
278	return d;
279}
280#endif
281
282#ifdef CONFIG_DMAR_TABLE
283static void dmar_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
284{
285	dmar_msi_write(data->irq, msg);
286}
287
288static struct irq_chip dmar_msi_controller = {
289	.name			= "DMAR-MSI",
290	.irq_unmask		= dmar_msi_unmask,
291	.irq_mask		= dmar_msi_mask,
292	.irq_ack		= irq_chip_ack_parent,
293	.irq_set_affinity	= msi_domain_set_affinity,
294	.irq_retrigger		= irq_chip_retrigger_hierarchy,
295	.irq_write_msi_msg	= dmar_msi_write_msg,
296	.flags			= IRQCHIP_SKIP_SET_WAKE |
297				  IRQCHIP_AFFINITY_PRE_STARTUP,
298};
299
300static int dmar_msi_init(struct irq_domain *domain,
301			 struct msi_domain_info *info, unsigned int virq,
302			 irq_hw_number_t hwirq, msi_alloc_info_t *arg)
303{
304	irq_domain_set_info(domain, virq, arg->devid, info->chip, NULL,
305			    handle_edge_irq, arg->data, "edge");
306
307	return 0;
308}
309
310static struct msi_domain_ops dmar_msi_domain_ops = {
311	.msi_init	= dmar_msi_init,
312};
313
314static struct msi_domain_info dmar_msi_domain_info = {
315	.ops		= &dmar_msi_domain_ops,
316	.chip		= &dmar_msi_controller,
317	.flags		= MSI_FLAG_USE_DEF_DOM_OPS,
318};
319
320static struct irq_domain *dmar_get_irq_domain(void)
321{
322	static struct irq_domain *dmar_domain;
323	static DEFINE_MUTEX(dmar_lock);
324	struct fwnode_handle *fn;
325
326	mutex_lock(&dmar_lock);
327	if (dmar_domain)
328		goto out;
329
330	fn = irq_domain_alloc_named_fwnode("DMAR-MSI");
331	if (fn) {
332		dmar_domain = msi_create_irq_domain(fn, &dmar_msi_domain_info,
333						    x86_vector_domain);
334		if (!dmar_domain)
335			irq_domain_free_fwnode(fn);
336	}
337out:
338	mutex_unlock(&dmar_lock);
339	return dmar_domain;
340}
341
342int dmar_alloc_hwirq(int id, int node, void *arg)
343{
344	struct irq_domain *domain = dmar_get_irq_domain();
345	struct irq_alloc_info info;
346
347	if (!domain)
348		return -1;
349
350	init_irq_alloc_info(&info, NULL);
351	info.type = X86_IRQ_ALLOC_TYPE_DMAR;
352	info.devid = id;
353	info.hwirq = id;
354	info.data = arg;
355
356	return irq_domain_alloc_irqs(domain, 1, node, &info);
357}
358
359void dmar_free_hwirq(int irq)
360{
361	irq_domain_free_irqs(irq, 1);
362}
363#endif
364
365/*
366 * MSI message composition
367 */
368#ifdef CONFIG_HPET_TIMER
369static inline int hpet_dev_id(struct irq_domain *domain)
370{
371	struct msi_domain_info *info = msi_get_domain_info(domain);
372
373	return (int)(long)info->data;
374}
375
376static void hpet_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
377{
378	hpet_msi_write(irq_data_get_irq_handler_data(data), msg);
379}
380
381static struct irq_chip hpet_msi_controller __ro_after_init = {
382	.name = "HPET-MSI",
383	.irq_unmask = hpet_msi_unmask,
384	.irq_mask = hpet_msi_mask,
385	.irq_ack = irq_chip_ack_parent,
386	.irq_set_affinity = msi_domain_set_affinity,
387	.irq_retrigger = irq_chip_retrigger_hierarchy,
388	.irq_write_msi_msg = hpet_msi_write_msg,
389	.flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_AFFINITY_PRE_STARTUP,
390};
391
392static int hpet_msi_init(struct irq_domain *domain,
393			 struct msi_domain_info *info, unsigned int virq,
394			 irq_hw_number_t hwirq, msi_alloc_info_t *arg)
395{
396	irq_set_status_flags(virq, IRQ_MOVE_PCNTXT);
397	irq_domain_set_info(domain, virq, arg->hwirq, info->chip, NULL,
398			    handle_edge_irq, arg->data, "edge");
399
400	return 0;
401}
402
403static void hpet_msi_free(struct irq_domain *domain,
404			  struct msi_domain_info *info, unsigned int virq)
405{
406	irq_clear_status_flags(virq, IRQ_MOVE_PCNTXT);
407}
408
409static struct msi_domain_ops hpet_msi_domain_ops = {
410	.msi_init	= hpet_msi_init,
411	.msi_free	= hpet_msi_free,
412};
413
414static struct msi_domain_info hpet_msi_domain_info = {
415	.ops		= &hpet_msi_domain_ops,
416	.chip		= &hpet_msi_controller,
417	.flags		= MSI_FLAG_USE_DEF_DOM_OPS,
418};
419
420struct irq_domain *hpet_create_irq_domain(int hpet_id)
421{
422	struct msi_domain_info *domain_info;
423	struct irq_domain *parent, *d;
424	struct irq_alloc_info info;
425	struct fwnode_handle *fn;
426
427	if (x86_vector_domain == NULL)
428		return NULL;
429
430	domain_info = kzalloc(sizeof(*domain_info), GFP_KERNEL);
431	if (!domain_info)
432		return NULL;
433
434	*domain_info = hpet_msi_domain_info;
435	domain_info->data = (void *)(long)hpet_id;
436
437	init_irq_alloc_info(&info, NULL);
438	info.type = X86_IRQ_ALLOC_TYPE_HPET_GET_PARENT;
439	info.devid = hpet_id;
440	parent = irq_remapping_get_irq_domain(&info);
441	if (parent == NULL)
442		parent = x86_vector_domain;
443	else
444		hpet_msi_controller.name = "IR-HPET-MSI";
445
446	fn = irq_domain_alloc_named_id_fwnode(hpet_msi_controller.name,
447					      hpet_id);
448	if (!fn) {
449		kfree(domain_info);
450		return NULL;
451	}
452
453	d = msi_create_irq_domain(fn, domain_info, parent);
454	if (!d) {
455		irq_domain_free_fwnode(fn);
456		kfree(domain_info);
457	}
458	return d;
459}
460
461int hpet_assign_irq(struct irq_domain *domain, struct hpet_channel *hc,
462		    int dev_num)
463{
464	struct irq_alloc_info info;
465
466	init_irq_alloc_info(&info, NULL);
467	info.type = X86_IRQ_ALLOC_TYPE_HPET;
468	info.data = hc;
469	info.devid = hpet_dev_id(domain);
470	info.hwirq = dev_num;
471
472	return irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
473}
474#endif
475