xref: /freebsd/sys/amd64/vmm/io/ppt.c (revision 7214e0469234d90edd6deda4b181cd4862a9ace0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/kernel.h>
32 #include <sys/malloc.h>
33 #include <sys/module.h>
34 #include <sys/bus.h>
35 #include <sys/pciio.h>
36 #include <sys/rman.h>
37 #include <sys/smp.h>
38 #include <sys/sysctl.h>
39 
40 #include <dev/pci/pcivar.h>
41 #include <dev/pci/pcireg.h>
42 
43 #include <machine/resource.h>
44 #include <machine/vmm.h>
45 #include <machine/vmm_dev.h>
46 
47 #include <dev/vmm/vmm_ktr.h>
48 
49 #include "vmm_lapic.h"
50 
51 #include "iommu.h"
52 #include "ppt.h"
53 
54 /* XXX locking */
55 
56 #define	MAX_MSIMSGS	32
57 
58 /*
59  * If the MSI-X table is located in the middle of a BAR then that MMIO
60  * region gets split into two segments - one segment above the MSI-X table
61  * and the other segment below the MSI-X table - with a hole in place of
62  * the MSI-X table so accesses to it can be trapped and emulated.
63  *
64  * So, allocate a MMIO segment for each BAR register + 1 additional segment.
65  */
66 #define	MAX_MMIOSEGS	((PCIR_MAX_BAR_0 + 1) + 1)
67 
68 MALLOC_DEFINE(M_PPTMSIX, "pptmsix", "Passthru MSI-X resources");
69 
70 struct pptintr_arg {				/* pptintr(pptintr_arg) */
71 	struct pptdev	*pptdev;
72 	uint64_t	addr;
73 	uint64_t	msg_data;
74 };
75 
76 struct pptseg {
77 	vm_paddr_t	gpa;
78 	size_t		len;
79 	int		wired;
80 };
81 
82 struct pptdev {
83 	device_t	dev;
84 	struct vm	*vm;			/* owner of this device */
85 	TAILQ_ENTRY(pptdev)	next;
86 	struct pptseg mmio[MAX_MMIOSEGS];
87 	struct {
88 		int	num_msgs;		/* guest state */
89 
90 		int	startrid;		/* host state */
91 		struct resource *res[MAX_MSIMSGS];
92 		void	*cookie[MAX_MSIMSGS];
93 		struct pptintr_arg arg[MAX_MSIMSGS];
94 	} msi;
95 
96 	struct {
97 		int num_msgs;
98 		int startrid;
99 		int msix_table_rid;
100 		int msix_pba_rid;
101 		struct resource *msix_table_res;
102 		struct resource *msix_pba_res;
103 		struct resource **res;
104 		void **cookie;
105 		struct pptintr_arg *arg;
106 	} msix;
107 };
108 
109 SYSCTL_DECL(_hw_vmm);
110 SYSCTL_NODE(_hw_vmm, OID_AUTO, ppt, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
111     "bhyve passthru devices");
112 
113 static int num_pptdevs;
114 SYSCTL_INT(_hw_vmm_ppt, OID_AUTO, devices, CTLFLAG_RD, &num_pptdevs, 0,
115     "number of pci passthru devices");
116 
117 static TAILQ_HEAD(, pptdev) pptdev_list = TAILQ_HEAD_INITIALIZER(pptdev_list);
118 
119 static int
ppt_probe(device_t dev)120 ppt_probe(device_t dev)
121 {
122 	int bus, slot, func;
123 	struct pci_devinfo *dinfo;
124 
125 	dinfo = (struct pci_devinfo *)device_get_ivars(dev);
126 
127 	bus = pci_get_bus(dev);
128 	slot = pci_get_slot(dev);
129 	func = pci_get_function(dev);
130 
131 	/*
132 	 * To qualify as a pci passthrough device a device must:
133 	 * - be allowed by administrator to be used in this role
134 	 * - be an endpoint device
135 	 */
136 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
137 		return (ENXIO);
138 	else if (vmm_is_pptdev(bus, slot, func))
139 		return (0);
140 	else
141 		/*
142 		 * Returning BUS_PROBE_NOWILDCARD here matches devices that the
143 		 * SR-IOV infrastructure specified as "ppt" passthrough devices.
144 		 * All normal devices that did not have "ppt" specified as their
145 		 * driver will not be matched by this.
146 		 */
147 		return (BUS_PROBE_NOWILDCARD);
148 }
149 
150 static int
ppt_attach(device_t dev)151 ppt_attach(device_t dev)
152 {
153 	struct pptdev *ppt;
154 	uint16_t cmd, cmd1;
155 	int error;
156 
157 	ppt = device_get_softc(dev);
158 
159 	cmd1 = cmd = pci_read_config(dev, PCIR_COMMAND, 2);
160 	cmd &= ~(PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
161 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
162 	error = iommu_remove_device(iommu_host_domain(), dev, pci_get_rid(dev));
163 	if (error != 0) {
164 		pci_write_config(dev, PCIR_COMMAND, cmd1, 2);
165 		return (error);
166 	}
167 	num_pptdevs++;
168 	TAILQ_INSERT_TAIL(&pptdev_list, ppt, next);
169 	ppt->dev = dev;
170 
171 	if (bootverbose)
172 		device_printf(dev, "attached\n");
173 
174 	return (0);
175 }
176 
177 static int
ppt_detach(device_t dev)178 ppt_detach(device_t dev)
179 {
180 	struct pptdev *ppt;
181 	int error;
182 
183 	ppt = device_get_softc(dev);
184 
185 	if (ppt->vm != NULL)
186 		return (EBUSY);
187 	if (iommu_host_domain() != NULL) {
188 		error = iommu_add_device(iommu_host_domain(), dev,
189 		    pci_get_rid(dev));
190 	} else {
191 		error = 0;
192 	}
193 	if (error != 0)
194 		return (error);
195 	num_pptdevs--;
196 	TAILQ_REMOVE(&pptdev_list, ppt, next);
197 
198 	return (0);
199 }
200 
201 static device_method_t ppt_methods[] = {
202 	/* Device interface */
203 	DEVMETHOD(device_probe,		ppt_probe),
204 	DEVMETHOD(device_attach,	ppt_attach),
205 	DEVMETHOD(device_detach,	ppt_detach),
206 	{0, 0}
207 };
208 
209 DEFINE_CLASS_0(ppt, ppt_driver, ppt_methods, sizeof(struct pptdev));
210 DRIVER_MODULE(ppt, pci, ppt_driver, NULL, NULL);
211 
212 static int
ppt_find(struct vm * vm,int bus,int slot,int func,struct pptdev ** pptp)213 ppt_find(struct vm *vm, int bus, int slot, int func, struct pptdev **pptp)
214 {
215 	device_t dev;
216 	struct pptdev *ppt;
217 	int b, s, f;
218 
219 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
220 		dev = ppt->dev;
221 		b = pci_get_bus(dev);
222 		s = pci_get_slot(dev);
223 		f = pci_get_function(dev);
224 		if (bus == b && slot == s && func == f)
225 			break;
226 	}
227 
228 	if (ppt == NULL)
229 		return (ENOENT);
230 	if (ppt->vm != vm)		/* Make sure we own this device */
231 		return (EBUSY);
232 	*pptp = ppt;
233 	return (0);
234 }
235 
236 static void
ppt_unmap_all_mmio(struct vm * vm,struct pptdev * ppt)237 ppt_unmap_all_mmio(struct vm *vm, struct pptdev *ppt)
238 {
239 	int i;
240 	struct pptseg *seg;
241 
242 	for (i = 0; i < MAX_MMIOSEGS; i++) {
243 		seg = &ppt->mmio[i];
244 		if (seg->len == 0)
245 			continue;
246 		(void)vm_unmap_mmio(vm, seg->gpa, seg->len);
247 		bzero(seg, sizeof(struct pptseg));
248 	}
249 }
250 
251 static void
ppt_teardown_msi(struct pptdev * ppt)252 ppt_teardown_msi(struct pptdev *ppt)
253 {
254 	int i, rid;
255 	void *cookie;
256 	struct resource *res;
257 
258 	if (ppt->msi.num_msgs == 0)
259 		return;
260 
261 	for (i = 0; i < ppt->msi.num_msgs; i++) {
262 		rid = ppt->msi.startrid + i;
263 		res = ppt->msi.res[i];
264 		cookie = ppt->msi.cookie[i];
265 
266 		if (cookie != NULL)
267 			bus_teardown_intr(ppt->dev, res, cookie);
268 
269 		if (res != NULL)
270 			bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res);
271 
272 		ppt->msi.res[i] = NULL;
273 		ppt->msi.cookie[i] = NULL;
274 	}
275 
276 	if (ppt->msi.startrid == 1)
277 		pci_release_msi(ppt->dev);
278 
279 	ppt->msi.num_msgs = 0;
280 }
281 
282 static void
ppt_teardown_msix_intr(struct pptdev * ppt,int idx)283 ppt_teardown_msix_intr(struct pptdev *ppt, int idx)
284 {
285 	int rid;
286 	struct resource *res;
287 	void *cookie;
288 
289 	rid = ppt->msix.startrid + idx;
290 	res = ppt->msix.res[idx];
291 	cookie = ppt->msix.cookie[idx];
292 
293 	if (cookie != NULL)
294 		bus_teardown_intr(ppt->dev, res, cookie);
295 
296 	if (res != NULL)
297 		bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res);
298 
299 	ppt->msix.res[idx] = NULL;
300 	ppt->msix.cookie[idx] = NULL;
301 }
302 
303 static void
ppt_teardown_msix(struct pptdev * ppt)304 ppt_teardown_msix(struct pptdev *ppt)
305 {
306 	int i;
307 
308 	if (ppt->msix.num_msgs == 0)
309 		return;
310 
311 	for (i = 0; i < ppt->msix.num_msgs; i++)
312 		ppt_teardown_msix_intr(ppt, i);
313 
314 	free(ppt->msix.res, M_PPTMSIX);
315 	free(ppt->msix.cookie, M_PPTMSIX);
316 	free(ppt->msix.arg, M_PPTMSIX);
317 
318 	pci_release_msi(ppt->dev);
319 
320 	if (ppt->msix.msix_table_res) {
321 		bus_release_resource(ppt->dev, SYS_RES_MEMORY,
322 				     ppt->msix.msix_table_rid,
323 				     ppt->msix.msix_table_res);
324 		ppt->msix.msix_table_res = NULL;
325 		ppt->msix.msix_table_rid = 0;
326 	}
327 	if (ppt->msix.msix_pba_res) {
328 		bus_release_resource(ppt->dev, SYS_RES_MEMORY,
329 				     ppt->msix.msix_pba_rid,
330 				     ppt->msix.msix_pba_res);
331 		ppt->msix.msix_pba_res = NULL;
332 		ppt->msix.msix_pba_rid = 0;
333 	}
334 
335 	ppt->msix.num_msgs = 0;
336 }
337 
338 int
ppt_assigned_devices(struct vm * vm)339 ppt_assigned_devices(struct vm *vm)
340 {
341 	struct pptdev *ppt;
342 	int num;
343 
344 	num = 0;
345 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
346 		if (ppt->vm == vm)
347 			num++;
348 	}
349 	return (num);
350 }
351 
352 bool
ppt_is_mmio(struct vm * vm,vm_paddr_t gpa)353 ppt_is_mmio(struct vm *vm, vm_paddr_t gpa)
354 {
355 	int i;
356 	struct pptdev *ppt;
357 	struct pptseg *seg;
358 
359 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
360 		if (ppt->vm != vm)
361 			continue;
362 
363 		for (i = 0; i < MAX_MMIOSEGS; i++) {
364 			seg = &ppt->mmio[i];
365 			if (seg->len == 0)
366 				continue;
367 			if (gpa >= seg->gpa && gpa < seg->gpa + seg->len)
368 				return (true);
369 		}
370 	}
371 
372 	return (false);
373 }
374 
375 static void
ppt_pci_reset(device_t dev)376 ppt_pci_reset(device_t dev)
377 {
378 
379 	if (pcie_flr(dev,
380 	     max(pcie_get_max_completion_timeout(dev) / 1000, 10), true))
381 		return;
382 
383 	pci_power_reset(dev);
384 }
385 
386 static uint16_t
ppt_bar_enables(struct pptdev * ppt)387 ppt_bar_enables(struct pptdev *ppt)
388 {
389 	struct pci_map *pm;
390 	uint16_t cmd;
391 
392 	cmd = 0;
393 	for (pm = pci_first_bar(ppt->dev); pm != NULL; pm = pci_next_bar(pm)) {
394 		if (PCI_BAR_IO(pm->pm_value))
395 			cmd |= PCIM_CMD_PORTEN;
396 		if (PCI_BAR_MEM(pm->pm_value))
397 			cmd |= PCIM_CMD_MEMEN;
398 	}
399 	return (cmd);
400 }
401 
402 int
ppt_assign_device(struct vm * vm,int bus,int slot,int func)403 ppt_assign_device(struct vm *vm, int bus, int slot, int func)
404 {
405 	struct pptdev *ppt;
406 	int error;
407 	uint16_t cmd;
408 
409 	/* Passing NULL requires the device to be unowned. */
410 	error = ppt_find(NULL, bus, slot, func, &ppt);
411 	if (error)
412 		return (error);
413 
414 	pci_save_state(ppt->dev);
415 	ppt_pci_reset(ppt->dev);
416 	pci_restore_state(ppt->dev);
417 	error = iommu_add_device(vm_iommu_domain(vm), ppt->dev,
418 	    pci_get_rid(ppt->dev));
419 	if (error != 0)
420 		return (error);
421 	ppt->vm = vm;
422 	cmd = pci_read_config(ppt->dev, PCIR_COMMAND, 2);
423 	cmd |= PCIM_CMD_BUSMASTEREN | ppt_bar_enables(ppt);
424 	pci_write_config(ppt->dev, PCIR_COMMAND, cmd, 2);
425 	return (0);
426 }
427 
428 int
ppt_unassign_device(struct vm * vm,int bus,int slot,int func)429 ppt_unassign_device(struct vm *vm, int bus, int slot, int func)
430 {
431 	struct pptdev *ppt;
432 	int error;
433 	uint16_t cmd;
434 
435 	error = ppt_find(vm, bus, slot, func, &ppt);
436 	if (error)
437 		return (error);
438 
439 	cmd = pci_read_config(ppt->dev, PCIR_COMMAND, 2);
440 	cmd &= ~(PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
441 	pci_write_config(ppt->dev, PCIR_COMMAND, cmd, 2);
442 	pci_save_state(ppt->dev);
443 	ppt_pci_reset(ppt->dev);
444 	pci_restore_state(ppt->dev);
445 	ppt_unmap_all_mmio(vm, ppt);
446 	ppt_teardown_msi(ppt);
447 	ppt_teardown_msix(ppt);
448 	error = iommu_remove_device(vm_iommu_domain(vm), ppt->dev,
449 	    pci_get_rid(ppt->dev));
450 	ppt->vm = NULL;
451 	return (error);
452 }
453 
454 int
ppt_unassign_all(struct vm * vm)455 ppt_unassign_all(struct vm *vm)
456 {
457 	struct pptdev *ppt;
458 	int bus, slot, func;
459 	device_t dev;
460 
461 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
462 		if (ppt->vm == vm) {
463 			dev = ppt->dev;
464 			bus = pci_get_bus(dev);
465 			slot = pci_get_slot(dev);
466 			func = pci_get_function(dev);
467 			vm_unassign_pptdev(vm, bus, slot, func);
468 		}
469 	}
470 
471 	return (0);
472 }
473 
474 static bool
ppt_valid_bar_mapping(struct pptdev * ppt,vm_paddr_t hpa,size_t len)475 ppt_valid_bar_mapping(struct pptdev *ppt, vm_paddr_t hpa, size_t len)
476 {
477 	struct pci_map *pm;
478 	pci_addr_t base, size;
479 
480 	for (pm = pci_first_bar(ppt->dev); pm != NULL; pm = pci_next_bar(pm)) {
481 		if (!PCI_BAR_MEM(pm->pm_value))
482 			continue;
483 		base = pm->pm_value & PCIM_BAR_MEM_BASE;
484 		size = (pci_addr_t)1 << pm->pm_size;
485 		if (hpa >= base && hpa + len <= base + size)
486 			return (true);
487 	}
488 	return (false);
489 }
490 
491 int
ppt_map_mmio(struct vm * vm,int bus,int slot,int func,vm_paddr_t gpa,size_t len,vm_paddr_t hpa)492 ppt_map_mmio(struct vm *vm, int bus, int slot, int func,
493 	     vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
494 {
495 	int i, error;
496 	struct pptseg *seg;
497 	struct pptdev *ppt;
498 
499 	if (len % PAGE_SIZE != 0 || len == 0 || gpa % PAGE_SIZE != 0 ||
500 	    hpa % PAGE_SIZE != 0 || gpa + len < gpa || hpa + len < hpa)
501 		return (EINVAL);
502 
503 	error = ppt_find(vm, bus, slot, func, &ppt);
504 	if (error)
505 		return (error);
506 
507 	if (!ppt_valid_bar_mapping(ppt, hpa, len))
508 		return (EINVAL);
509 
510 	for (i = 0; i < MAX_MMIOSEGS; i++) {
511 		seg = &ppt->mmio[i];
512 		if (seg->len == 0) {
513 			error = vm_map_mmio(vm, gpa, len, hpa);
514 			if (error == 0) {
515 				seg->gpa = gpa;
516 				seg->len = len;
517 			}
518 			return (error);
519 		}
520 	}
521 	return (ENOSPC);
522 }
523 
524 int
ppt_unmap_mmio(struct vm * vm,int bus,int slot,int func,vm_paddr_t gpa,size_t len)525 ppt_unmap_mmio(struct vm *vm, int bus, int slot, int func,
526 	       vm_paddr_t gpa, size_t len)
527 {
528 	int i, error;
529 	struct pptseg *seg;
530 	struct pptdev *ppt;
531 
532 	error = ppt_find(vm, bus, slot, func, &ppt);
533 	if (error)
534 		return (error);
535 
536 	for (i = 0; i < MAX_MMIOSEGS; i++) {
537 		seg = &ppt->mmio[i];
538 		if (seg->gpa == gpa && seg->len == len) {
539 			error = vm_unmap_mmio(vm, seg->gpa, seg->len);
540 			if (error == 0) {
541 				seg->gpa = 0;
542 				seg->len = 0;
543 			}
544 			return (error);
545 		}
546 	}
547 	return (ENOENT);
548 }
549 
550 static int
pptintr(void * arg)551 pptintr(void *arg)
552 {
553 	struct pptdev *ppt;
554 	struct pptintr_arg *pptarg;
555 
556 	pptarg = arg;
557 	ppt = pptarg->pptdev;
558 
559 	if (ppt->vm != NULL)
560 		lapic_intr_msi(ppt->vm, pptarg->addr, pptarg->msg_data);
561 	else {
562 		/*
563 		 * XXX
564 		 * This is not expected to happen - panic?
565 		 */
566 	}
567 
568 	/*
569 	 * For legacy interrupts give other filters a chance in case
570 	 * the interrupt was not generated by the passthrough device.
571 	 */
572 	if (ppt->msi.startrid == 0)
573 		return (FILTER_STRAY);
574 	else
575 		return (FILTER_HANDLED);
576 }
577 
578 int
ppt_setup_msi(struct vm * vm,int bus,int slot,int func,uint64_t addr,uint64_t msg,int numvec)579 ppt_setup_msi(struct vm *vm, int bus, int slot, int func,
580 	      uint64_t addr, uint64_t msg, int numvec)
581 {
582 	int i, rid, flags;
583 	int msi_count, startrid, error, tmp;
584 	struct pptdev *ppt;
585 
586 	if (numvec < 0 || numvec > MAX_MSIMSGS)
587 		return (EINVAL);
588 
589 	error = ppt_find(vm, bus, slot, func, &ppt);
590 	if (error)
591 		return (error);
592 
593 	/* Reject attempts to enable MSI while MSI-X is active. */
594 	if (ppt->msix.num_msgs != 0 && numvec != 0)
595 		return (EBUSY);
596 
597 	/* Free any allocated resources */
598 	ppt_teardown_msi(ppt);
599 
600 	if (numvec == 0)		/* nothing more to do */
601 		return (0);
602 
603 	flags = RF_ACTIVE;
604 	msi_count = pci_msi_count(ppt->dev);
605 	if (msi_count == 0) {
606 		startrid = 0;		/* legacy interrupt */
607 		msi_count = 1;
608 		flags |= RF_SHAREABLE;
609 	} else
610 		startrid = 1;		/* MSI */
611 
612 	/*
613 	 * The device must be capable of supporting the number of vectors
614 	 * the guest wants to allocate.
615 	 */
616 	if (numvec > msi_count)
617 		return (EINVAL);
618 
619 	/*
620 	 * Make sure that we can allocate all the MSI vectors that are needed
621 	 * by the guest.
622 	 */
623 	if (startrid == 1) {
624 		tmp = numvec;
625 		error = pci_alloc_msi(ppt->dev, &tmp);
626 		if (error)
627 			return (error);
628 		else if (tmp != numvec) {
629 			pci_release_msi(ppt->dev);
630 			return (ENOSPC);
631 		} else {
632 			/* success */
633 		}
634 	}
635 
636 	ppt->msi.startrid = startrid;
637 
638 	/*
639 	 * Allocate the irq resource and attach it to the interrupt handler.
640 	 */
641 	for (i = 0; i < numvec; i++) {
642 		ppt->msi.num_msgs = i + 1;
643 		ppt->msi.cookie[i] = NULL;
644 
645 		rid = startrid + i;
646 		ppt->msi.res[i] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ,
647 							 &rid, flags);
648 		if (ppt->msi.res[i] == NULL)
649 			break;
650 
651 		ppt->msi.arg[i].pptdev = ppt;
652 		ppt->msi.arg[i].addr = addr;
653 		ppt->msi.arg[i].msg_data = msg + i;
654 
655 		error = bus_setup_intr(ppt->dev, ppt->msi.res[i],
656 				       INTR_TYPE_NET | INTR_MPSAFE,
657 				       pptintr, NULL, &ppt->msi.arg[i],
658 				       &ppt->msi.cookie[i]);
659 		if (error != 0)
660 			break;
661 	}
662 
663 	if (i < numvec) {
664 		ppt_teardown_msi(ppt);
665 		return (ENXIO);
666 	}
667 
668 	return (0);
669 }
670 
671 int
ppt_setup_msix(struct vm * vm,int bus,int slot,int func,int idx,uint64_t addr,uint64_t msg,uint32_t vector_control)672 ppt_setup_msix(struct vm *vm, int bus, int slot, int func,
673 	       int idx, uint64_t addr, uint64_t msg, uint32_t vector_control)
674 {
675 	struct pptdev *ppt;
676 	struct pci_devinfo *dinfo;
677 	int numvec, alloced, rid, error;
678 	size_t res_size, cookie_size, arg_size;
679 
680 	error = ppt_find(vm, bus, slot, func, &ppt);
681 	if (error)
682 		return (error);
683 
684 	/* Reject attempts to enable MSI-X while MSI is active. */
685 	if (ppt->msi.num_msgs != 0)
686 		return (EBUSY);
687 
688 	dinfo = device_get_ivars(ppt->dev);
689 	if (!dinfo)
690 		return (ENXIO);
691 
692 	/*
693 	 * First-time configuration:
694 	 * 	Allocate the MSI-X table
695 	 *	Allocate the IRQ resources
696 	 *	Set up some variables in ppt->msix
697 	 */
698 	if (ppt->msix.num_msgs == 0) {
699 		numvec = pci_msix_count(ppt->dev);
700 		if (numvec <= 0)
701 			return (EINVAL);
702 
703 		ppt->msix.startrid = 1;
704 		ppt->msix.num_msgs = numvec;
705 
706 		res_size = numvec * sizeof(ppt->msix.res[0]);
707 		cookie_size = numvec * sizeof(ppt->msix.cookie[0]);
708 		arg_size = numvec * sizeof(ppt->msix.arg[0]);
709 
710 		ppt->msix.res = malloc(res_size, M_PPTMSIX, M_WAITOK | M_ZERO);
711 		ppt->msix.cookie = malloc(cookie_size, M_PPTMSIX,
712 					  M_WAITOK | M_ZERO);
713 		ppt->msix.arg = malloc(arg_size, M_PPTMSIX, M_WAITOK | M_ZERO);
714 
715 		rid = dinfo->cfg.msix.msix_table_bar;
716 		ppt->msix.msix_table_res = bus_alloc_resource_any(ppt->dev,
717 					       SYS_RES_MEMORY, &rid, RF_ACTIVE);
718 
719 		if (ppt->msix.msix_table_res == NULL) {
720 			ppt_teardown_msix(ppt);
721 			return (ENOSPC);
722 		}
723 		ppt->msix.msix_table_rid = rid;
724 
725 		if (dinfo->cfg.msix.msix_table_bar !=
726 		    dinfo->cfg.msix.msix_pba_bar) {
727 			rid = dinfo->cfg.msix.msix_pba_bar;
728 			ppt->msix.msix_pba_res = bus_alloc_resource_any(
729 			    ppt->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
730 
731 			if (ppt->msix.msix_pba_res == NULL) {
732 				ppt_teardown_msix(ppt);
733 				return (ENOSPC);
734 			}
735 			ppt->msix.msix_pba_rid = rid;
736 		}
737 
738 		alloced = numvec;
739 		error = pci_alloc_msix(ppt->dev, &alloced);
740 		if (error || alloced != numvec) {
741 			ppt_teardown_msix(ppt);
742 			return (error == 0 ? ENOSPC: error);
743 		}
744 	}
745 
746 	if (idx >= ppt->msix.num_msgs)
747 		return (EINVAL);
748 
749 	if ((vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
750 		/* Tear down the IRQ if it's already set up */
751 		ppt_teardown_msix_intr(ppt, idx);
752 
753 		/* Allocate the IRQ resource */
754 		ppt->msix.cookie[idx] = NULL;
755 		rid = ppt->msix.startrid + idx;
756 		ppt->msix.res[idx] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ,
757 							    &rid, RF_ACTIVE);
758 		if (ppt->msix.res[idx] == NULL)
759 			return (ENXIO);
760 
761 		ppt->msix.arg[idx].pptdev = ppt;
762 		ppt->msix.arg[idx].addr = addr;
763 		ppt->msix.arg[idx].msg_data = msg;
764 
765 		/* Setup the MSI-X interrupt */
766 		error = bus_setup_intr(ppt->dev, ppt->msix.res[idx],
767 				       INTR_TYPE_NET | INTR_MPSAFE,
768 				       pptintr, NULL, &ppt->msix.arg[idx],
769 				       &ppt->msix.cookie[idx]);
770 
771 		if (error != 0) {
772 			bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, ppt->msix.res[idx]);
773 			ppt->msix.cookie[idx] = NULL;
774 			ppt->msix.res[idx] = NULL;
775 			return (ENXIO);
776 		}
777 	} else {
778 		/* Masked, tear it down if it's already been set up */
779 		ppt_teardown_msix_intr(ppt, idx);
780 	}
781 
782 	return (0);
783 }
784 
785 int
ppt_disable_msix(struct vm * vm,int bus,int slot,int func)786 ppt_disable_msix(struct vm *vm, int bus, int slot, int func)
787 {
788 	struct pptdev *ppt;
789 	int error;
790 
791 	error = ppt_find(vm, bus, slot, func, &ppt);
792 	if (error)
793 		return (error);
794 
795 	ppt_teardown_msix(ppt);
796 	return (0);
797 }
798