1/*-
2 * Copyright (c) 2006 Yahoo!, Inc.
3 * All rights reserved.
4 * Written by: John Baldwin <jhb@FreeBSD.org>
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 3. Neither the name of the author nor the names of any co-contributors
15 *    may be used to endorse or promote products derived from this software
16 *    without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31/*
32 * Support for PCI Message Signalled Interrupts (MSI).  MSI interrupts on
33 * x86 are basically APIC messages that the northbridge delivers directly
34 * to the local APICs as if they had come from an I/O APIC.
35 */
36
37#include <sys/cdefs.h>
38__FBSDID("$FreeBSD$");
39
40#include <sys/param.h>
41#include <sys/bus.h>
42#include <sys/kernel.h>
43#include <sys/lock.h>
44#include <sys/malloc.h>
45#include <sys/mutex.h>
46#include <sys/sx.h>
47#include <sys/systm.h>
48#include <x86/apicreg.h>
49#include <machine/cputypes.h>
50#include <machine/md_var.h>
51#include <machine/frame.h>
52#include <machine/intr_machdep.h>
53#include <machine/apicvar.h>
54#include <machine/specialreg.h>
55#include <dev/pci/pcivar.h>
56
57/* Fields in address for Intel MSI messages. */
58#define	MSI_INTEL_ADDR_DEST		0x000ff000
59#define	MSI_INTEL_ADDR_RH		0x00000008
60# define MSI_INTEL_ADDR_RH_ON		0x00000008
61# define MSI_INTEL_ADDR_RH_OFF		0x00000000
62#define	MSI_INTEL_ADDR_DM		0x00000004
63# define MSI_INTEL_ADDR_DM_PHYSICAL	0x00000000
64# define MSI_INTEL_ADDR_DM_LOGICAL	0x00000004
65
66/* Fields in data for Intel MSI messages. */
67#define	MSI_INTEL_DATA_TRGRMOD		IOART_TRGRMOD	/* Trigger mode. */
68# define MSI_INTEL_DATA_TRGREDG		IOART_TRGREDG
69# define MSI_INTEL_DATA_TRGRLVL		IOART_TRGRLVL
70#define	MSI_INTEL_DATA_LEVEL		0x00004000	/* Polarity. */
71# define MSI_INTEL_DATA_DEASSERT	0x00000000
72# define MSI_INTEL_DATA_ASSERT		0x00004000
73#define	MSI_INTEL_DATA_DELMOD		IOART_DELMOD	/* Delivery mode. */
74# define MSI_INTEL_DATA_DELFIXED	IOART_DELFIXED
75# define MSI_INTEL_DATA_DELLOPRI	IOART_DELLOPRI
76# define MSI_INTEL_DATA_DELSMI		IOART_DELSMI
77# define MSI_INTEL_DATA_DELNMI		IOART_DELNMI
78# define MSI_INTEL_DATA_DELINIT		IOART_DELINIT
79# define MSI_INTEL_DATA_DELEXINT	IOART_DELEXINT
80#define	MSI_INTEL_DATA_INTVEC		IOART_INTVEC	/* Interrupt vector. */
81
82/*
83 * Build Intel MSI message and data values from a source.  AMD64 systems
84 * seem to be compatible, so we use the same function for both.
85 */
86#define	INTEL_ADDR(msi)							\
87	(MSI_INTEL_ADDR_BASE | (msi)->msi_cpu << 12 |			\
88	    MSI_INTEL_ADDR_RH_OFF | MSI_INTEL_ADDR_DM_PHYSICAL)
89#define	INTEL_DATA(msi)							\
90	(MSI_INTEL_DATA_TRGREDG | MSI_INTEL_DATA_DELFIXED | (msi)->msi_vector)
91
92static MALLOC_DEFINE(M_MSI, "msi", "PCI MSI");
93
94/*
95 * MSI sources are bunched into groups.  This is because MSI forces
96 * all of the messages to share the address and data registers and
97 * thus certain properties (such as the local APIC ID target on x86).
98 * Each group has a 'first' source that contains information global to
99 * the group.  These fields are marked with (g) below.
100 *
101 * Note that local APIC ID is kind of special.  Each message will be
102 * assigned an ID by the system; however, a group will use the ID from
103 * the first message.
104 *
105 * For MSI-X, each message is isolated.
106 */
107struct msi_intsrc {
108	struct intsrc msi_intsrc;
109	device_t msi_dev;		/* Owning device. (g) */
110	struct msi_intsrc *msi_first;	/* First source in group. */
111	u_int msi_irq;			/* IRQ cookie. */
112	u_int msi_msix;			/* MSI-X message. */
113	u_int msi_vector:8;		/* IDT vector. */
114	u_int msi_cpu:8;		/* Local APIC ID. (g) */
115	u_int msi_count:8;		/* Messages in this group. (g) */
116	u_int msi_maxcount:8;		/* Alignment for this group. (g) */
117	int *msi_irqs;			/* Group's IRQ list. (g) */
118};
119
120static void	msi_create_source(void);
121static void	msi_enable_source(struct intsrc *isrc);
122static void	msi_disable_source(struct intsrc *isrc, int eoi);
123static void	msi_eoi_source(struct intsrc *isrc);
124static void	msi_enable_intr(struct intsrc *isrc);
125static void	msi_disable_intr(struct intsrc *isrc);
126static int	msi_vector(struct intsrc *isrc);
127static int	msi_source_pending(struct intsrc *isrc);
128static int	msi_config_intr(struct intsrc *isrc, enum intr_trigger trig,
129		    enum intr_polarity pol);
130static int	msi_assign_cpu(struct intsrc *isrc, u_int apic_id);
131
132struct pic msi_pic = { msi_enable_source, msi_disable_source, msi_eoi_source,
133		       msi_enable_intr, msi_disable_intr, msi_vector,
134		       msi_source_pending, NULL, NULL, msi_config_intr,
135		       msi_assign_cpu };
136
137static int msi_enabled;
138static int msi_last_irq;
139static struct mtx msi_lock;
140
141static void
142msi_enable_source(struct intsrc *isrc)
143{
144}
145
146static void
147msi_disable_source(struct intsrc *isrc, int eoi)
148{
149
150	if (eoi == PIC_EOI)
151		lapic_eoi();
152}
153
154static void
155msi_eoi_source(struct intsrc *isrc)
156{
157
158	lapic_eoi();
159}
160
161static void
162msi_enable_intr(struct intsrc *isrc)
163{
164	struct msi_intsrc *msi = (struct msi_intsrc *)isrc;
165
166	apic_enable_vector(msi->msi_cpu, msi->msi_vector);
167}
168
169static void
170msi_disable_intr(struct intsrc *isrc)
171{
172	struct msi_intsrc *msi = (struct msi_intsrc *)isrc;
173
174	apic_disable_vector(msi->msi_cpu, msi->msi_vector);
175}
176
177static int
178msi_vector(struct intsrc *isrc)
179{
180	struct msi_intsrc *msi = (struct msi_intsrc *)isrc;
181
182	return (msi->msi_irq);
183}
184
185static int
186msi_source_pending(struct intsrc *isrc)
187{
188
189	return (0);
190}
191
192static int
193msi_config_intr(struct intsrc *isrc, enum intr_trigger trig,
194    enum intr_polarity pol)
195{
196
197	return (ENODEV);
198}
199
200static int
201msi_assign_cpu(struct intsrc *isrc, u_int apic_id)
202{
203	struct msi_intsrc *sib, *msi = (struct msi_intsrc *)isrc;
204	int old_vector;
205	u_int old_id;
206	int i, vector;
207
208	/*
209	 * Only allow CPUs to be assigned to the first message for an
210	 * MSI group.
211	 */
212	if (msi->msi_first != msi)
213		return (EINVAL);
214
215	/* Store information to free existing irq. */
216	old_vector = msi->msi_vector;
217	old_id = msi->msi_cpu;
218	if (old_id == apic_id)
219		return (0);
220
221	/* Allocate IDT vectors on this cpu. */
222	if (msi->msi_count > 1) {
223		KASSERT(msi->msi_msix == 0, ("MSI-X message group"));
224		vector = apic_alloc_vectors(apic_id, msi->msi_irqs,
225		    msi->msi_count, msi->msi_maxcount);
226	} else
227		vector = apic_alloc_vector(apic_id, msi->msi_irq);
228	if (vector == 0)
229		return (ENOSPC);
230
231	msi->msi_cpu = apic_id;
232	msi->msi_vector = vector;
233	if (msi->msi_intsrc.is_handlers > 0)
234		apic_enable_vector(msi->msi_cpu, msi->msi_vector);
235	if (bootverbose)
236		printf("msi: Assigning %s IRQ %d to local APIC %u vector %u\n",
237		    msi->msi_msix ? "MSI-X" : "MSI", msi->msi_irq,
238		    msi->msi_cpu, msi->msi_vector);
239	for (i = 1; i < msi->msi_count; i++) {
240		sib = (struct msi_intsrc *)intr_lookup_source(msi->msi_irqs[i]);
241		sib->msi_cpu = apic_id;
242		sib->msi_vector = vector + i;
243		if (sib->msi_intsrc.is_handlers > 0)
244			apic_enable_vector(sib->msi_cpu, sib->msi_vector);
245		if (bootverbose)
246			printf(
247		    "msi: Assigning MSI IRQ %d to local APIC %u vector %u\n",
248			    sib->msi_irq, sib->msi_cpu, sib->msi_vector);
249	}
250	BUS_REMAP_INTR(device_get_parent(msi->msi_dev), msi->msi_dev,
251	    msi->msi_irq);
252
253	/*
254	 * Free the old vector after the new one is established.  This is done
255	 * to prevent races where we could miss an interrupt.
256	 */
257	if (msi->msi_intsrc.is_handlers > 0)
258		apic_disable_vector(old_id, old_vector);
259	apic_free_vector(old_id, old_vector, msi->msi_irq);
260	for (i = 1; i < msi->msi_count; i++) {
261		sib = (struct msi_intsrc *)intr_lookup_source(msi->msi_irqs[i]);
262		if (sib->msi_intsrc.is_handlers > 0)
263			apic_disable_vector(old_id, old_vector + i);
264		apic_free_vector(old_id, old_vector + i, msi->msi_irqs[i]);
265	}
266	return (0);
267}
268
269void
270msi_init(void)
271{
272
273	/* Check if we have a supported CPU. */
274	switch (cpu_vendor_id) {
275	case CPU_VENDOR_INTEL:
276	case CPU_VENDOR_AMD:
277		break;
278	case CPU_VENDOR_CENTAUR:
279		if (CPUID_TO_FAMILY(cpu_id) == 0x6 &&
280		    CPUID_TO_MODEL(cpu_id) >= 0xf)
281			break;
282		/* FALLTHROUGH */
283	default:
284		return;
285	}
286
287	msi_enabled = 1;
288	intr_register_pic(&msi_pic);
289	mtx_init(&msi_lock, "msi", NULL, MTX_DEF);
290}
291
292static void
293msi_create_source(void)
294{
295	struct msi_intsrc *msi;
296	u_int irq;
297
298	mtx_lock(&msi_lock);
299	if (msi_last_irq >= NUM_MSI_INTS) {
300		mtx_unlock(&msi_lock);
301		return;
302	}
303	irq = msi_last_irq + FIRST_MSI_INT;
304	msi_last_irq++;
305	mtx_unlock(&msi_lock);
306
307	msi = malloc(sizeof(struct msi_intsrc), M_MSI, M_WAITOK | M_ZERO);
308	msi->msi_intsrc.is_pic = &msi_pic;
309	msi->msi_irq = irq;
310	intr_register_source(&msi->msi_intsrc);
311	nexus_add_irq(irq);
312}
313
314/*
315 * Try to allocate 'count' interrupt sources with contiguous IDT values.
316 */
317int
318msi_alloc(device_t dev, int count, int maxcount, int *irqs)
319{
320	struct msi_intsrc *msi, *fsrc;
321	u_int cpu;
322	int cnt, i, *mirqs, vector;
323
324	if (!msi_enabled)
325		return (ENXIO);
326
327	if (count > 1)
328		mirqs = malloc(count * sizeof(*mirqs), M_MSI, M_WAITOK);
329	else
330		mirqs = NULL;
331again:
332	mtx_lock(&msi_lock);
333
334	/* Try to find 'count' free IRQs. */
335	cnt = 0;
336	for (i = FIRST_MSI_INT; i < FIRST_MSI_INT + NUM_MSI_INTS; i++) {
337		msi = (struct msi_intsrc *)intr_lookup_source(i);
338
339		/* End of allocated sources, so break. */
340		if (msi == NULL)
341			break;
342
343		/* If this is a free one, save its IRQ in the array. */
344		if (msi->msi_dev == NULL) {
345			irqs[cnt] = i;
346			cnt++;
347			if (cnt == count)
348				break;
349		}
350	}
351
352	/* Do we need to create some new sources? */
353	if (cnt < count) {
354		/* If we would exceed the max, give up. */
355		if (i + (count - cnt) > FIRST_MSI_INT + NUM_MSI_INTS) {
356			mtx_unlock(&msi_lock);
357			free(mirqs, M_MSI);
358			return (ENXIO);
359		}
360		mtx_unlock(&msi_lock);
361
362		/* We need count - cnt more sources. */
363		while (cnt < count) {
364			msi_create_source();
365			cnt++;
366		}
367		goto again;
368	}
369
370	/* Ok, we now have the IRQs allocated. */
371	KASSERT(cnt == count, ("count mismatch"));
372
373	/* Allocate 'count' IDT vectors. */
374	cpu = intr_next_cpu();
375	vector = apic_alloc_vectors(cpu, irqs, count, maxcount);
376	if (vector == 0) {
377		mtx_unlock(&msi_lock);
378		free(mirqs, M_MSI);
379		return (ENOSPC);
380	}
381
382	/* Assign IDT vectors and make these messages owned by 'dev'. */
383	fsrc = (struct msi_intsrc *)intr_lookup_source(irqs[0]);
384	for (i = 0; i < count; i++) {
385		msi = (struct msi_intsrc *)intr_lookup_source(irqs[i]);
386		msi->msi_cpu = cpu;
387		msi->msi_dev = dev;
388		msi->msi_vector = vector + i;
389		if (bootverbose)
390			printf(
391		    "msi: routing MSI IRQ %d to local APIC %u vector %u\n",
392			    msi->msi_irq, msi->msi_cpu, msi->msi_vector);
393		msi->msi_first = fsrc;
394		KASSERT(msi->msi_intsrc.is_handlers == 0,
395		    ("dead MSI has handlers"));
396	}
397	fsrc->msi_count = count;
398	fsrc->msi_maxcount = maxcount;
399	if (count > 1)
400		bcopy(irqs, mirqs, count * sizeof(*mirqs));
401	fsrc->msi_irqs = mirqs;
402	mtx_unlock(&msi_lock);
403
404	return (0);
405}
406
407int
408msi_release(int *irqs, int count)
409{
410	struct msi_intsrc *msi, *first;
411	int i;
412
413	mtx_lock(&msi_lock);
414	first = (struct msi_intsrc *)intr_lookup_source(irqs[0]);
415	if (first == NULL) {
416		mtx_unlock(&msi_lock);
417		return (ENOENT);
418	}
419
420	/* Make sure this isn't an MSI-X message. */
421	if (first->msi_msix) {
422		mtx_unlock(&msi_lock);
423		return (EINVAL);
424	}
425
426	/* Make sure this message is allocated to a group. */
427	if (first->msi_first == NULL) {
428		mtx_unlock(&msi_lock);
429		return (ENXIO);
430	}
431
432	/*
433	 * Make sure this is the start of a group and that we are releasing
434	 * the entire group.
435	 */
436	if (first->msi_first != first || first->msi_count != count) {
437		mtx_unlock(&msi_lock);
438		return (EINVAL);
439	}
440	KASSERT(first->msi_dev != NULL, ("unowned group"));
441
442	/* Clear all the extra messages in the group. */
443	for (i = 1; i < count; i++) {
444		msi = (struct msi_intsrc *)intr_lookup_source(irqs[i]);
445		KASSERT(msi->msi_first == first, ("message not in group"));
446		KASSERT(msi->msi_dev == first->msi_dev, ("owner mismatch"));
447		msi->msi_first = NULL;
448		msi->msi_dev = NULL;
449		apic_free_vector(msi->msi_cpu, msi->msi_vector, msi->msi_irq);
450		msi->msi_vector = 0;
451	}
452
453	/* Clear out the first message. */
454	first->msi_first = NULL;
455	first->msi_dev = NULL;
456	apic_free_vector(first->msi_cpu, first->msi_vector, first->msi_irq);
457	first->msi_vector = 0;
458	first->msi_count = 0;
459	first->msi_maxcount = 0;
460	free(first->msi_irqs, M_MSI);
461	first->msi_irqs = NULL;
462
463	mtx_unlock(&msi_lock);
464	return (0);
465}
466
467int
468msi_map(int irq, uint64_t *addr, uint32_t *data)
469{
470	struct msi_intsrc *msi;
471
472	mtx_lock(&msi_lock);
473	msi = (struct msi_intsrc *)intr_lookup_source(irq);
474	if (msi == NULL) {
475		mtx_unlock(&msi_lock);
476		return (ENOENT);
477	}
478
479	/* Make sure this message is allocated to a device. */
480	if (msi->msi_dev == NULL) {
481		mtx_unlock(&msi_lock);
482		return (ENXIO);
483	}
484
485	/*
486	 * If this message isn't an MSI-X message, make sure it's part
487	 * of a group, and switch to the first message in the
488	 * group.
489	 */
490	if (!msi->msi_msix) {
491		if (msi->msi_first == NULL) {
492			mtx_unlock(&msi_lock);
493			return (ENXIO);
494		}
495		msi = msi->msi_first;
496	}
497
498	*addr = INTEL_ADDR(msi);
499	*data = INTEL_DATA(msi);
500	mtx_unlock(&msi_lock);
501	return (0);
502}
503
504int
505msix_alloc(device_t dev, int *irq)
506{
507	struct msi_intsrc *msi;
508	u_int cpu;
509	int i, vector;
510
511	if (!msi_enabled)
512		return (ENXIO);
513
514again:
515	mtx_lock(&msi_lock);
516
517	/* Find a free IRQ. */
518	for (i = FIRST_MSI_INT; i < FIRST_MSI_INT + NUM_MSI_INTS; i++) {
519		msi = (struct msi_intsrc *)intr_lookup_source(i);
520
521		/* End of allocated sources, so break. */
522		if (msi == NULL)
523			break;
524
525		/* Stop at the first free source. */
526		if (msi->msi_dev == NULL)
527			break;
528	}
529
530	/* Do we need to create a new source? */
531	if (msi == NULL) {
532		/* If we would exceed the max, give up. */
533		if (i + 1 > FIRST_MSI_INT + NUM_MSI_INTS) {
534			mtx_unlock(&msi_lock);
535			return (ENXIO);
536		}
537		mtx_unlock(&msi_lock);
538
539		/* Create a new source. */
540		msi_create_source();
541		goto again;
542	}
543
544	/* Allocate an IDT vector. */
545	cpu = intr_next_cpu();
546	vector = apic_alloc_vector(cpu, i);
547	if (vector == 0) {
548		mtx_unlock(&msi_lock);
549		return (ENOSPC);
550	}
551	if (bootverbose)
552		printf("msi: routing MSI-X IRQ %d to local APIC %u vector %u\n",
553		    msi->msi_irq, cpu, vector);
554
555	/* Setup source. */
556	msi->msi_cpu = cpu;
557	msi->msi_dev = dev;
558	msi->msi_first = msi;
559	msi->msi_vector = vector;
560	msi->msi_msix = 1;
561	msi->msi_count = 1;
562	msi->msi_maxcount = 1;
563	msi->msi_irqs = NULL;
564
565	KASSERT(msi->msi_intsrc.is_handlers == 0, ("dead MSI-X has handlers"));
566	mtx_unlock(&msi_lock);
567
568	*irq = i;
569	return (0);
570}
571
572int
573msix_release(int irq)
574{
575	struct msi_intsrc *msi;
576
577	mtx_lock(&msi_lock);
578	msi = (struct msi_intsrc *)intr_lookup_source(irq);
579	if (msi == NULL) {
580		mtx_unlock(&msi_lock);
581		return (ENOENT);
582	}
583
584	/* Make sure this is an MSI-X message. */
585	if (!msi->msi_msix) {
586		mtx_unlock(&msi_lock);
587		return (EINVAL);
588	}
589
590	KASSERT(msi->msi_dev != NULL, ("unowned message"));
591
592	/* Clear out the message. */
593	msi->msi_first = NULL;
594	msi->msi_dev = NULL;
595	apic_free_vector(msi->msi_cpu, msi->msi_vector, msi->msi_irq);
596	msi->msi_vector = 0;
597	msi->msi_msix = 0;
598	msi->msi_count = 0;
599	msi->msi_maxcount = 0;
600
601	mtx_unlock(&msi_lock);
602	return (0);
603}
604