2 * Copyright (c) 2006 Yahoo!, Inc.
4 * Written by: John Baldwin <jhb@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
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.
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
42 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mutex.h>
47 #include <sys/systm.h>
48 #include <machine/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>
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
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. */
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.
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)
92 static MALLOC_DEFINE(M_MSI, "msi", "PCI MSI");
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.
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
105 * For MSI-X, each message is isolated.
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) */
120 static void msi_create_source(void);
121 static void msi_enable_source(struct intsrc *isrc);
122 static void msi_disable_source(struct intsrc *isrc, int eoi);
123 static void msi_eoi_source(struct intsrc *isrc);
124 static void msi_enable_intr(struct intsrc *isrc);
125 static void msi_disable_intr(struct intsrc *isrc);
126 static int msi_vector(struct intsrc *isrc);
127 static int msi_source_pending(struct intsrc *isrc);
128 static int msi_config_intr(struct intsrc *isrc, enum intr_trigger trig,
129 enum intr_polarity pol);
130 static int msi_assign_cpu(struct intsrc *isrc, u_int apic_id);
132 struct 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,
137 static int msi_enabled;
138 static int msi_last_irq;
139 static struct mtx msi_lock;
142 msi_enable_source(struct intsrc *isrc)
147 msi_disable_source(struct intsrc *isrc, int eoi)
155 msi_eoi_source(struct intsrc *isrc)
162 msi_enable_intr(struct intsrc *isrc)
164 struct msi_intsrc *msi = (struct msi_intsrc *)isrc;
166 apic_enable_vector(msi->msi_cpu, msi->msi_vector);
170 msi_disable_intr(struct intsrc *isrc)
172 struct msi_intsrc *msi = (struct msi_intsrc *)isrc;
174 apic_disable_vector(msi->msi_cpu, msi->msi_vector);
178 msi_vector(struct intsrc *isrc)
180 struct msi_intsrc *msi = (struct msi_intsrc *)isrc;
182 return (msi->msi_irq);
186 msi_source_pending(struct intsrc *isrc)
193 msi_config_intr(struct intsrc *isrc, enum intr_trigger trig,
194 enum intr_polarity pol)
201 msi_assign_cpu(struct intsrc *isrc, u_int apic_id)
203 struct msi_intsrc *sib, *msi = (struct msi_intsrc *)isrc;
209 * Only allow CPUs to be assigned to the first message for an
212 if (msi->msi_first != msi)
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)
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);
227 vector = apic_alloc_vector(apic_id, msi->msi_irq);
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);
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);
247 "msi: Assigning MSI IRQ %d to local APIC %u vector %u\n",
248 sib->msi_irq, sib->msi_cpu, sib->msi_vector);
250 pci_remap_msi_irq(msi->msi_dev, msi->msi_irq);
253 * Free the old vector after the new one is established. This is done
254 * to prevent races where we could miss an interrupt.
256 if (msi->msi_intsrc.is_handlers > 0)
257 apic_disable_vector(old_id, old_vector);
258 apic_free_vector(old_id, old_vector, msi->msi_irq);
259 for (i = 1; i < msi->msi_count; i++) {
260 sib = (struct msi_intsrc *)intr_lookup_source(msi->msi_irqs[i]);
261 if (sib->msi_intsrc.is_handlers > 0)
262 apic_disable_vector(old_id, old_vector + i);
263 apic_free_vector(old_id, old_vector + i, msi->msi_irqs[i]);
272 /* Check if we have a supported CPU. */
273 switch (cpu_vendor_id) {
274 case CPU_VENDOR_INTEL:
277 case CPU_VENDOR_CENTAUR:
278 if (I386_CPU_FAMILY(cpu_id) == 0x6 &&
279 I386_CPU_MODEL(cpu_id) >= 0xf)
287 intr_register_pic(&msi_pic);
288 mtx_init(&msi_lock, "msi", NULL, MTX_DEF);
292 msi_create_source(void)
294 struct msi_intsrc *msi;
298 if (msi_last_irq >= NUM_MSI_INTS) {
299 mtx_unlock(&msi_lock);
302 irq = msi_last_irq + FIRST_MSI_INT;
304 mtx_unlock(&msi_lock);
306 msi = malloc(sizeof(struct msi_intsrc), M_MSI, M_WAITOK | M_ZERO);
307 msi->msi_intsrc.is_pic = &msi_pic;
309 intr_register_source(&msi->msi_intsrc);
314 * Try to allocate 'count' interrupt sources with contiguous IDT values.
317 msi_alloc(device_t dev, int count, int maxcount, int *irqs)
319 struct msi_intsrc *msi, *fsrc;
321 int cnt, i, *mirqs, vector;
327 mirqs = malloc(count * sizeof(*mirqs), M_MSI, M_WAITOK);
333 /* Try to find 'count' free IRQs. */
335 for (i = FIRST_MSI_INT; i < FIRST_MSI_INT + NUM_MSI_INTS; i++) {
336 msi = (struct msi_intsrc *)intr_lookup_source(i);
338 /* End of allocated sources, so break. */
342 /* If this is a free one, save its IRQ in the array. */
343 if (msi->msi_dev == NULL) {
351 /* Do we need to create some new sources? */
353 /* If we would exceed the max, give up. */
354 if (i + (count - cnt) > FIRST_MSI_INT + NUM_MSI_INTS) {
355 mtx_unlock(&msi_lock);
359 mtx_unlock(&msi_lock);
361 /* We need count - cnt more sources. */
362 while (cnt < count) {
369 /* Ok, we now have the IRQs allocated. */
370 KASSERT(cnt == count, ("count mismatch"));
372 /* Allocate 'count' IDT vectors. */
373 cpu = intr_next_cpu();
374 vector = apic_alloc_vectors(cpu, irqs, count, maxcount);
376 mtx_unlock(&msi_lock);
381 /* Assign IDT vectors and make these messages owned by 'dev'. */
382 fsrc = (struct msi_intsrc *)intr_lookup_source(irqs[0]);
383 for (i = 0; i < count; i++) {
384 msi = (struct msi_intsrc *)intr_lookup_source(irqs[i]);
387 msi->msi_vector = vector + i;
390 "msi: routing MSI IRQ %d to local APIC %u vector %u\n",
391 msi->msi_irq, msi->msi_cpu, msi->msi_vector);
392 msi->msi_first = fsrc;
393 KASSERT(msi->msi_intsrc.is_handlers == 0,
394 ("dead MSI has handlers"));
396 fsrc->msi_count = count;
397 fsrc->msi_maxcount = maxcount;
399 bcopy(irqs, mirqs, count * sizeof(*mirqs));
400 fsrc->msi_irqs = mirqs;
401 mtx_unlock(&msi_lock);
407 msi_release(int *irqs, int count)
409 struct msi_intsrc *msi, *first;
413 first = (struct msi_intsrc *)intr_lookup_source(irqs[0]);
415 mtx_unlock(&msi_lock);
419 /* Make sure this isn't an MSI-X message. */
420 if (first->msi_msix) {
421 mtx_unlock(&msi_lock);
425 /* Make sure this message is allocated to a group. */
426 if (first->msi_first == NULL) {
427 mtx_unlock(&msi_lock);
432 * Make sure this is the start of a group and that we are releasing
435 if (first->msi_first != first || first->msi_count != count) {
436 mtx_unlock(&msi_lock);
439 KASSERT(first->msi_dev != NULL, ("unowned group"));
441 /* Clear all the extra messages in the group. */
442 for (i = 1; i < count; i++) {
443 msi = (struct msi_intsrc *)intr_lookup_source(irqs[i]);
444 KASSERT(msi->msi_first == first, ("message not in group"));
445 KASSERT(msi->msi_dev == first->msi_dev, ("owner mismatch"));
446 msi->msi_first = NULL;
448 apic_free_vector(msi->msi_cpu, msi->msi_vector, msi->msi_irq);
452 /* Clear out the first message. */
453 first->msi_first = NULL;
454 first->msi_dev = NULL;
455 apic_free_vector(first->msi_cpu, first->msi_vector, first->msi_irq);
456 first->msi_vector = 0;
457 first->msi_count = 0;
458 first->msi_maxcount = 0;
459 free(first->msi_irqs, M_MSI);
460 first->msi_irqs = NULL;
462 mtx_unlock(&msi_lock);
467 msi_map(int irq, uint64_t *addr, uint32_t *data)
469 struct msi_intsrc *msi;
472 msi = (struct msi_intsrc *)intr_lookup_source(irq);
474 mtx_unlock(&msi_lock);
478 /* Make sure this message is allocated to a device. */
479 if (msi->msi_dev == NULL) {
480 mtx_unlock(&msi_lock);
485 * If this message isn't an MSI-X message, make sure it's part
486 * of a group, and switch to the first message in the
489 if (!msi->msi_msix) {
490 if (msi->msi_first == NULL) {
491 mtx_unlock(&msi_lock);
494 msi = msi->msi_first;
497 *addr = INTEL_ADDR(msi);
498 *data = INTEL_DATA(msi);
499 mtx_unlock(&msi_lock);
504 msix_alloc(device_t dev, int *irq)
506 struct msi_intsrc *msi;
516 /* Find a free IRQ. */
517 for (i = FIRST_MSI_INT; i < FIRST_MSI_INT + NUM_MSI_INTS; i++) {
518 msi = (struct msi_intsrc *)intr_lookup_source(i);
520 /* End of allocated sources, so break. */
524 /* Stop at the first free source. */
525 if (msi->msi_dev == NULL)
529 /* Do we need to create a new source? */
531 /* If we would exceed the max, give up. */
532 if (i + 1 > FIRST_MSI_INT + NUM_MSI_INTS) {
533 mtx_unlock(&msi_lock);
536 mtx_unlock(&msi_lock);
538 /* Create a new source. */
543 /* Allocate an IDT vector. */
544 cpu = intr_next_cpu();
545 vector = apic_alloc_vector(cpu, i);
547 mtx_unlock(&msi_lock);
551 printf("msi: routing MSI-X IRQ %d to local APIC %u vector %u\n",
552 msi->msi_irq, cpu, vector);
557 msi->msi_first = msi;
558 msi->msi_vector = vector;
561 msi->msi_maxcount = 1;
562 msi->msi_irqs = NULL;
564 KASSERT(msi->msi_intsrc.is_handlers == 0, ("dead MSI-X has handlers"));
565 mtx_unlock(&msi_lock);
572 msix_release(int irq)
574 struct msi_intsrc *msi;
577 msi = (struct msi_intsrc *)intr_lookup_source(irq);
579 mtx_unlock(&msi_lock);
583 /* Make sure this is an MSI-X message. */
584 if (!msi->msi_msix) {
585 mtx_unlock(&msi_lock);
589 KASSERT(msi->msi_dev != NULL, ("unowned message"));
591 /* Clear out the message. */
592 msi->msi_first = NULL;
594 apic_free_vector(msi->msi_cpu, msi->msi_vector, msi->msi_irq);
598 msi->msi_maxcount = 0;
600 mtx_unlock(&msi_lock);