xref: /src/sys/amd64/vmm/io/ppt.c (revision 36b855f1892575cbfe1cd5455b989bfc8ae07502)
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/bus.h>
32 #include <sys/kernel.h>
33 #include <sys/lock.h>
34 #include <sys/malloc.h>
35 #include <sys/module.h>
36 #include <sys/pciio.h>
37 #include <sys/rman.h>
38 #include <sys/smp.h>
39 #include <sys/sx.h>
40 #include <sys/sysctl.h>
41 
42 #include <dev/pci/pcivar.h>
43 #include <dev/pci/pcireg.h>
44 
45 #include <machine/resource.h>
46 #include <machine/vmm.h>
47 #include <machine/vmm_dev.h>
48 
49 #include <dev/vmm/vmm_ktr.h>
50 
51 #include "vmm_lapic.h"
52 
53 #include "iommu.h"
54 #include "ppt.h"
55 
56 /* XXX locking */
57 
58 #define	MAX_MSIMSGS	32
59 
60 /*
61  * If the MSI-X table is located in the middle of a BAR then that MMIO
62  * region gets split into two segments - one segment above the MSI-X table
63  * and the other segment below the MSI-X table - with a hole in place of
64  * the MSI-X table so accesses to it can be trapped and emulated.
65  *
66  * So, allocate a MMIO segment for each BAR register + 1 additional segment.
67  */
68 #define	MAX_MMIOSEGS	((PCIR_MAX_BAR_0 + 1) + 1)
69 
70 MALLOC_DEFINE(M_PPTMSIX, "pptmsix", "Passthru MSI-X resources");
71 
72 static struct sx ppt_mtx;
73 SX_SYSINIT(ppt_mtx, &ppt_mtx, "ppt_mtx");
74 #define	PPT_LOCK()		sx_xlock(&ppt_mtx)
75 #define	PPT_UNLOCK()		sx_xunlock(&ppt_mtx)
76 #define	PPT_ASSERT_LOCKED()	sx_assert(&ppt_mtx, SA_XLOCKED)
77 
78 struct pptintr_arg {				/* pptintr(pptintr_arg) */
79 	struct pptdev	*pptdev;
80 	uint64_t	addr;
81 	uint64_t	msg_data;
82 };
83 
84 struct pptseg {
85 	vm_paddr_t	gpa;
86 	size_t		len;
87 	int		wired;
88 };
89 
90 struct pptdev {
91 	device_t	dev;
92 	struct vm	*vm;			/* owner of this device */
93 	TAILQ_ENTRY(pptdev)	next;
94 	struct pptseg mmio[MAX_MMIOSEGS];
95 	struct {
96 		int	num_msgs;		/* guest state */
97 
98 		int	startrid;		/* host state */
99 		struct resource *res[MAX_MSIMSGS];
100 		void	*cookie[MAX_MSIMSGS];
101 		struct pptintr_arg arg[MAX_MSIMSGS];
102 	} msi;
103 
104 	struct {
105 		int num_msgs;
106 		int startrid;
107 		int msix_table_rid;
108 		int msix_pba_rid;
109 		struct resource *msix_table_res;
110 		struct resource *msix_pba_res;
111 		struct resource **res;
112 		void **cookie;
113 		struct pptintr_arg *arg;
114 	} msix;
115 };
116 
117 SYSCTL_DECL(_hw_vmm);
118 SYSCTL_NODE(_hw_vmm, OID_AUTO, ppt, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
119     "bhyve passthru devices");
120 
121 static int num_pptdevs;
122 SYSCTL_INT(_hw_vmm_ppt, OID_AUTO, devices, CTLFLAG_RD, &num_pptdevs, 0,
123     "number of pci passthru devices");
124 
125 static TAILQ_HEAD(, pptdev) pptdev_list = TAILQ_HEAD_INITIALIZER(pptdev_list);
126 
127 static int
ppt_probe(device_t dev)128 ppt_probe(device_t dev)
129 {
130 	int bus, slot, func;
131 	struct pci_devinfo *dinfo;
132 
133 	dinfo = (struct pci_devinfo *)device_get_ivars(dev);
134 
135 	bus = pci_get_bus(dev);
136 	slot = pci_get_slot(dev);
137 	func = pci_get_function(dev);
138 
139 	/*
140 	 * To qualify as a pci passthrough device a device must:
141 	 * - be allowed by administrator to be used in this role
142 	 * - be an endpoint device
143 	 */
144 	if ((dinfo->cfg.hdrtype & PCIM_HDRTYPE) != PCIM_HDRTYPE_NORMAL)
145 		return (ENXIO);
146 	else if (vmm_is_pptdev(bus, slot, func))
147 		return (0);
148 	else
149 		/*
150 		 * Returning BUS_PROBE_NOWILDCARD here matches devices that the
151 		 * SR-IOV infrastructure specified as "ppt" passthrough devices.
152 		 * All normal devices that did not have "ppt" specified as their
153 		 * driver will not be matched by this.
154 		 */
155 		return (BUS_PROBE_NOWILDCARD);
156 }
157 
158 static int
ppt_attach(device_t dev)159 ppt_attach(device_t dev)
160 {
161 	struct pptdev *ppt;
162 	uint16_t cmd, cmd1;
163 	int error;
164 
165 	ppt = device_get_softc(dev);
166 
167 	PPT_LOCK();
168 	cmd1 = cmd = pci_read_config(dev, PCIR_COMMAND, 2);
169 	cmd &= ~(PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
170 	pci_write_config(dev, PCIR_COMMAND, cmd, 2);
171 	error = iommu_remove_device(iommu_host_domain(), dev, pci_get_rid(dev));
172 	if (error != 0) {
173 		pci_write_config(dev, PCIR_COMMAND, cmd1, 2);
174 		PPT_UNLOCK();
175 		return (error);
176 	}
177 	num_pptdevs++;
178 	TAILQ_INSERT_TAIL(&pptdev_list, ppt, next);
179 	ppt->dev = dev;
180 	PPT_UNLOCK();
181 
182 	if (bootverbose)
183 		device_printf(dev, "attached\n");
184 
185 	return (0);
186 }
187 
188 static int
ppt_detach(device_t dev)189 ppt_detach(device_t dev)
190 {
191 	struct pptdev *ppt;
192 	int error;
193 
194 	error = 0;
195 	ppt = device_get_softc(dev);
196 
197 	PPT_LOCK();
198 	if (ppt->vm != NULL) {
199 		error = EBUSY;
200 		goto out;
201 	}
202 	if (iommu_host_domain() != NULL) {
203 		error = iommu_add_device(iommu_host_domain(), dev,
204 		    pci_get_rid(dev));
205 		if (error != 0)
206 			goto out;
207 	}
208 	num_pptdevs--;
209 	TAILQ_REMOVE(&pptdev_list, ppt, next);
210 out:
211 	PPT_UNLOCK();
212 
213 	return (error);
214 }
215 
216 static device_method_t ppt_methods[] = {
217 	/* Device interface */
218 	DEVMETHOD(device_probe,		ppt_probe),
219 	DEVMETHOD(device_attach,	ppt_attach),
220 	DEVMETHOD(device_detach,	ppt_detach),
221 	{0, 0}
222 };
223 
224 DEFINE_CLASS_0(ppt, ppt_driver, ppt_methods, sizeof(struct pptdev));
225 DRIVER_MODULE(ppt, pci, ppt_driver, NULL, NULL);
226 
227 static int
ppt_find(struct vm * vm,int bus,int slot,int func,struct pptdev ** pptp)228 ppt_find(struct vm *vm, int bus, int slot, int func, struct pptdev **pptp)
229 {
230 	device_t dev;
231 	struct pptdev *ppt;
232 	int b, s, f;
233 
234 	PPT_ASSERT_LOCKED();
235 
236 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
237 		dev = ppt->dev;
238 		b = pci_get_bus(dev);
239 		s = pci_get_slot(dev);
240 		f = pci_get_function(dev);
241 		if (bus == b && slot == s && func == f)
242 			break;
243 	}
244 
245 	if (ppt == NULL)
246 		return (ENOENT);
247 	if (ppt->vm != vm)		/* Make sure we own this device */
248 		return (EBUSY);
249 	*pptp = ppt;
250 	return (0);
251 }
252 
253 static void
ppt_unmap_all_mmio(struct vm * vm,struct pptdev * ppt)254 ppt_unmap_all_mmio(struct vm *vm, struct pptdev *ppt)
255 {
256 	int i;
257 	struct pptseg *seg;
258 
259 	for (i = 0; i < MAX_MMIOSEGS; i++) {
260 		seg = &ppt->mmio[i];
261 		if (seg->len == 0)
262 			continue;
263 		(void)vm_unmap_mmio(vm, seg->gpa, seg->len);
264 		bzero(seg, sizeof(struct pptseg));
265 	}
266 }
267 
268 static void
ppt_teardown_msi(struct pptdev * ppt)269 ppt_teardown_msi(struct pptdev *ppt)
270 {
271 	int i, rid;
272 	void *cookie;
273 	struct resource *res;
274 
275 	if (ppt->msi.num_msgs == 0)
276 		return;
277 
278 	for (i = 0; i < ppt->msi.num_msgs; i++) {
279 		rid = ppt->msi.startrid + i;
280 		res = ppt->msi.res[i];
281 		cookie = ppt->msi.cookie[i];
282 
283 		if (cookie != NULL)
284 			bus_teardown_intr(ppt->dev, res, cookie);
285 
286 		if (res != NULL)
287 			bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res);
288 
289 		ppt->msi.res[i] = NULL;
290 		ppt->msi.cookie[i] = NULL;
291 	}
292 
293 	if (ppt->msi.startrid == 1)
294 		pci_release_msi(ppt->dev);
295 
296 	ppt->msi.num_msgs = 0;
297 }
298 
299 static void
ppt_teardown_msix_intr(struct pptdev * ppt,int idx)300 ppt_teardown_msix_intr(struct pptdev *ppt, int idx)
301 {
302 	int rid;
303 	struct resource *res;
304 	void *cookie;
305 
306 	rid = ppt->msix.startrid + idx;
307 	res = ppt->msix.res[idx];
308 	cookie = ppt->msix.cookie[idx];
309 
310 	if (cookie != NULL)
311 		bus_teardown_intr(ppt->dev, res, cookie);
312 
313 	if (res != NULL)
314 		bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, res);
315 
316 	ppt->msix.res[idx] = NULL;
317 	ppt->msix.cookie[idx] = NULL;
318 }
319 
320 static void
ppt_teardown_msix(struct pptdev * ppt)321 ppt_teardown_msix(struct pptdev *ppt)
322 {
323 	int i;
324 
325 	if (ppt->msix.num_msgs == 0)
326 		return;
327 
328 	for (i = 0; i < ppt->msix.num_msgs; i++)
329 		ppt_teardown_msix_intr(ppt, i);
330 
331 	free(ppt->msix.res, M_PPTMSIX);
332 	free(ppt->msix.cookie, M_PPTMSIX);
333 	free(ppt->msix.arg, M_PPTMSIX);
334 
335 	pci_release_msi(ppt->dev);
336 
337 	if (ppt->msix.msix_table_res) {
338 		bus_release_resource(ppt->dev, SYS_RES_MEMORY,
339 				     ppt->msix.msix_table_rid,
340 				     ppt->msix.msix_table_res);
341 		ppt->msix.msix_table_res = NULL;
342 		ppt->msix.msix_table_rid = 0;
343 	}
344 	if (ppt->msix.msix_pba_res) {
345 		bus_release_resource(ppt->dev, SYS_RES_MEMORY,
346 				     ppt->msix.msix_pba_rid,
347 				     ppt->msix.msix_pba_res);
348 		ppt->msix.msix_pba_res = NULL;
349 		ppt->msix.msix_pba_rid = 0;
350 	}
351 
352 	ppt->msix.num_msgs = 0;
353 }
354 
355 int
ppt_assigned_devices(struct vm * vm)356 ppt_assigned_devices(struct vm *vm)
357 {
358 	struct pptdev *ppt;
359 	int num;
360 
361 	num = 0;
362 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
363 		if (ppt->vm == vm)
364 			num++;
365 	}
366 	return (num);
367 }
368 
369 bool
ppt_is_mmio(struct vm * vm,vm_paddr_t gpa)370 ppt_is_mmio(struct vm *vm, vm_paddr_t gpa)
371 {
372 	int i;
373 	struct pptdev *ppt;
374 	struct pptseg *seg;
375 
376 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
377 		if (ppt->vm != vm)
378 			continue;
379 
380 		for (i = 0; i < MAX_MMIOSEGS; i++) {
381 			seg = &ppt->mmio[i];
382 			if (seg->len == 0)
383 				continue;
384 			if (gpa >= seg->gpa && gpa < seg->gpa + seg->len)
385 				return (true);
386 		}
387 	}
388 
389 	return (false);
390 }
391 
392 static void
ppt_pci_reset(device_t dev)393 ppt_pci_reset(device_t dev)
394 {
395 
396 	if (pcie_flr(dev,
397 	     max(pcie_get_max_completion_timeout(dev) / 1000, 10), true))
398 		return;
399 
400 	pci_power_reset(dev);
401 }
402 
403 static uint16_t
ppt_bar_enables(struct pptdev * ppt)404 ppt_bar_enables(struct pptdev *ppt)
405 {
406 	struct pci_map *pm;
407 	uint16_t cmd;
408 
409 	cmd = 0;
410 	for (pm = pci_first_bar(ppt->dev); pm != NULL; pm = pci_next_bar(pm)) {
411 		if (PCI_BAR_IO(pm->pm_value))
412 			cmd |= PCIM_CMD_PORTEN;
413 		if (PCI_BAR_MEM(pm->pm_value))
414 			cmd |= PCIM_CMD_MEMEN;
415 	}
416 	return (cmd);
417 }
418 
419 int
ppt_assign_device(struct vm * vm,int bus,int slot,int func)420 ppt_assign_device(struct vm *vm, int bus, int slot, int func)
421 {
422 	struct pptdev *ppt;
423 	int error;
424 	uint16_t cmd;
425 
426 	PPT_LOCK();
427 	/* Passing NULL requires the device to be unowned. */
428 	error = ppt_find(NULL, bus, slot, func, &ppt);
429 	if (error != 0)
430 		goto out;
431 
432 	pci_save_state(ppt->dev);
433 	ppt_pci_reset(ppt->dev);
434 	pci_restore_state(ppt->dev);
435 	error = iommu_add_device(vm_iommu_domain(vm), ppt->dev,
436 	    pci_get_rid(ppt->dev));
437 	if (error != 0)
438 		goto out;
439 	ppt->vm = vm;
440 	cmd = pci_read_config(ppt->dev, PCIR_COMMAND, 2);
441 	cmd |= PCIM_CMD_BUSMASTEREN | ppt_bar_enables(ppt);
442 	pci_write_config(ppt->dev, PCIR_COMMAND, cmd, 2);
443 out:
444 	PPT_UNLOCK();
445 	return (error);
446 }
447 
448 int
ppt_unassign_device(struct vm * vm,int bus,int slot,int func)449 ppt_unassign_device(struct vm *vm, int bus, int slot, int func)
450 {
451 	struct pptdev *ppt;
452 	int error;
453 	uint16_t cmd;
454 
455 	PPT_LOCK();
456 	error = ppt_find(vm, bus, slot, func, &ppt);
457 	if (error != 0)
458 		goto out;
459 
460 	cmd = pci_read_config(ppt->dev, PCIR_COMMAND, 2);
461 	cmd &= ~(PCIM_CMD_PORTEN | PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
462 	pci_write_config(ppt->dev, PCIR_COMMAND, cmd, 2);
463 	pci_save_state(ppt->dev);
464 	ppt_pci_reset(ppt->dev);
465 	pci_restore_state(ppt->dev);
466 	ppt_unmap_all_mmio(vm, ppt);
467 	ppt_teardown_msi(ppt);
468 	ppt_teardown_msix(ppt);
469 	error = iommu_remove_device(vm_iommu_domain(vm), ppt->dev,
470 	    pci_get_rid(ppt->dev));
471 	ppt->vm = NULL;
472 out:
473 	PPT_UNLOCK();
474 	return (error);
475 }
476 
477 int
ppt_unassign_all(struct vm * vm)478 ppt_unassign_all(struct vm *vm)
479 {
480 	struct pptdev *ppt;
481 	int bus, slot, func;
482 	device_t dev;
483 
484 	TAILQ_FOREACH(ppt, &pptdev_list, next) {
485 		if (ppt->vm == vm) {
486 			dev = ppt->dev;
487 			bus = pci_get_bus(dev);
488 			slot = pci_get_slot(dev);
489 			func = pci_get_function(dev);
490 			vm_unassign_pptdev(vm, bus, slot, func);
491 		}
492 	}
493 
494 	return (0);
495 }
496 
497 static bool
ppt_valid_bar_mapping(struct pptdev * ppt,vm_paddr_t hpa,size_t len)498 ppt_valid_bar_mapping(struct pptdev *ppt, vm_paddr_t hpa, size_t len)
499 {
500 	struct pci_map *pm;
501 	pci_addr_t base, size;
502 
503 	for (pm = pci_first_bar(ppt->dev); pm != NULL; pm = pci_next_bar(pm)) {
504 		if (!PCI_BAR_MEM(pm->pm_value))
505 			continue;
506 		base = pm->pm_value & PCIM_BAR_MEM_BASE;
507 		size = (pci_addr_t)1 << pm->pm_size;
508 		if (hpa >= base && hpa + len <= base + size)
509 			return (true);
510 	}
511 	return (false);
512 }
513 
514 int
ppt_map_mmio(struct vm * vm,int bus,int slot,int func,vm_paddr_t gpa,size_t len,vm_paddr_t hpa)515 ppt_map_mmio(struct vm *vm, int bus, int slot, int func,
516 	     vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
517 {
518 	int i, error;
519 	struct pptseg *seg;
520 	struct pptdev *ppt;
521 
522 	if (len % PAGE_SIZE != 0 || len == 0 || gpa % PAGE_SIZE != 0 ||
523 	    hpa % PAGE_SIZE != 0 || gpa + len < gpa || hpa + len < hpa)
524 		return (EINVAL);
525 
526 	PPT_LOCK();
527 	error = ppt_find(vm, bus, slot, func, &ppt);
528 	if (error)
529 		goto out;
530 
531 	if (!ppt_valid_bar_mapping(ppt, hpa, len)) {
532 		error = EINVAL;
533 		goto out;
534 	}
535 
536 	error = ENOSPC;
537 	for (i = 0; i < MAX_MMIOSEGS; i++) {
538 		seg = &ppt->mmio[i];
539 		if (seg->len == 0) {
540 			error = vm_map_mmio(vm, gpa, len, hpa);
541 			if (error == 0) {
542 				seg->gpa = gpa;
543 				seg->len = len;
544 			}
545 			break;
546 		}
547 	}
548 out:
549 	PPT_UNLOCK();
550 	return (error);
551 }
552 
553 int
ppt_unmap_mmio(struct vm * vm,int bus,int slot,int func,vm_paddr_t gpa,size_t len)554 ppt_unmap_mmio(struct vm *vm, int bus, int slot, int func,
555 	       vm_paddr_t gpa, size_t len)
556 {
557 	int i, error;
558 	struct pptseg *seg;
559 	struct pptdev *ppt;
560 
561 	PPT_LOCK();
562 	error = ppt_find(vm, bus, slot, func, &ppt);
563 	if (error)
564 		goto out;
565 
566 	error = ENOENT;
567 	for (i = 0; i < MAX_MMIOSEGS; i++) {
568 		seg = &ppt->mmio[i];
569 		if (seg->gpa == gpa && seg->len == len) {
570 			error = vm_unmap_mmio(vm, seg->gpa, seg->len);
571 			if (error == 0) {
572 				seg->gpa = 0;
573 				seg->len = 0;
574 			}
575 			break;
576 		}
577 	}
578 out:
579 	PPT_UNLOCK();
580 	return (ENOENT);
581 }
582 
583 static int
pptintr(void * arg)584 pptintr(void *arg)
585 {
586 	struct pptdev *ppt;
587 	struct pptintr_arg *pptarg;
588 
589 	pptarg = arg;
590 	ppt = pptarg->pptdev;
591 
592 	if (ppt->vm != NULL)
593 		lapic_intr_msi(ppt->vm, pptarg->addr, pptarg->msg_data);
594 	else {
595 		/*
596 		 * XXX
597 		 * This is not expected to happen - panic?
598 		 */
599 	}
600 
601 	/*
602 	 * For legacy interrupts give other filters a chance in case
603 	 * the interrupt was not generated by the passthrough device.
604 	 */
605 	if (ppt->msi.startrid == 0)
606 		return (FILTER_STRAY);
607 	else
608 		return (FILTER_HANDLED);
609 }
610 
611 int
ppt_setup_msi(struct vm * vm,int bus,int slot,int func,uint64_t addr,uint64_t msg,int numvec)612 ppt_setup_msi(struct vm *vm, int bus, int slot, int func,
613 	      uint64_t addr, uint64_t msg, int numvec)
614 {
615 	int i, rid, flags;
616 	int msi_count, startrid, error, tmp;
617 	struct pptdev *ppt;
618 
619 	if (numvec < 0 || numvec > MAX_MSIMSGS)
620 		return (EINVAL);
621 
622 	PPT_LOCK();
623 	error = ppt_find(vm, bus, slot, func, &ppt);
624 	if (error)
625 		goto out;
626 
627 	/* Reject attempts to enable MSI while MSI-X is active. */
628 	if (ppt->msix.num_msgs != 0 && numvec != 0) {
629 		error = EBUSY;
630 		goto out;
631 	}
632 
633 	/* Free any allocated resources */
634 	ppt_teardown_msi(ppt);
635 
636 	if (numvec == 0)		/* nothing more to do */
637 		goto out;
638 
639 	flags = RF_ACTIVE;
640 	msi_count = pci_msi_count(ppt->dev);
641 	if (msi_count == 0) {
642 		startrid = 0;		/* legacy interrupt */
643 		msi_count = 1;
644 		flags |= RF_SHAREABLE;
645 	} else
646 		startrid = 1;		/* MSI */
647 
648 	/*
649 	 * The device must be capable of supporting the number of vectors
650 	 * the guest wants to allocate.
651 	 */
652 	if (numvec > msi_count) {
653 		error = EINVAL;
654 		goto out;
655 	}
656 
657 	/*
658 	 * Make sure that we can allocate all the MSI vectors that are needed
659 	 * by the guest.
660 	 */
661 	if (startrid == 1) {
662 		tmp = numvec;
663 		error = pci_alloc_msi(ppt->dev, &tmp);
664 		if (error)
665 			goto out;
666 		else if (tmp != numvec) {
667 			pci_release_msi(ppt->dev);
668 			error = ENOSPC;
669 			goto out;
670 		} else {
671 			/* success */
672 		}
673 	}
674 
675 	ppt->msi.startrid = startrid;
676 
677 	/*
678 	 * Allocate the irq resource and attach it to the interrupt handler.
679 	 */
680 	for (i = 0; i < numvec; i++) {
681 		ppt->msi.num_msgs = i + 1;
682 		ppt->msi.cookie[i] = NULL;
683 
684 		rid = startrid + i;
685 		ppt->msi.res[i] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ,
686 							 &rid, flags);
687 		if (ppt->msi.res[i] == NULL)
688 			break;
689 
690 		ppt->msi.arg[i].pptdev = ppt;
691 		ppt->msi.arg[i].addr = addr;
692 		ppt->msi.arg[i].msg_data = msg + i;
693 
694 		error = bus_setup_intr(ppt->dev, ppt->msi.res[i],
695 				       INTR_TYPE_NET | INTR_MPSAFE,
696 				       pptintr, NULL, &ppt->msi.arg[i],
697 				       &ppt->msi.cookie[i]);
698 		if (error != 0)
699 			break;
700 	}
701 
702 	if (i < numvec) {
703 		ppt_teardown_msi(ppt);
704 		error = ENXIO;
705 	}
706 
707 out:
708 	PPT_UNLOCK();
709 	return (error);
710 }
711 
712 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)713 ppt_setup_msix(struct vm *vm, int bus, int slot, int func,
714 	       int idx, uint64_t addr, uint64_t msg, uint32_t vector_control)
715 {
716 	struct pptdev *ppt;
717 	struct pci_devinfo *dinfo;
718 	int numvec, alloced, rid, error;
719 	size_t res_size, cookie_size, arg_size;
720 
721 	PPT_LOCK();
722 	error = ppt_find(vm, bus, slot, func, &ppt);
723 	if (error)
724 		goto out;
725 
726 	/* Reject attempts to enable MSI-X while MSI is active. */
727 	if (ppt->msi.num_msgs != 0) {
728 		error = EBUSY;
729 		goto out;
730 	}
731 
732 	dinfo = device_get_ivars(ppt->dev);
733 	if (dinfo == NULL) {
734 		error = ENXIO;
735 		goto out;
736 	}
737 
738 	/*
739 	 * First-time configuration:
740 	 * 	Allocate the MSI-X table
741 	 *	Allocate the IRQ resources
742 	 *	Set up some variables in ppt->msix
743 	 */
744 	if (ppt->msix.num_msgs == 0) {
745 		numvec = pci_msix_count(ppt->dev);
746 		if (numvec <= 0) {
747 			error = EINVAL;
748 			goto out;
749 		}
750 
751 		ppt->msix.startrid = 1;
752 		ppt->msix.num_msgs = numvec;
753 
754 		res_size = numvec * sizeof(ppt->msix.res[0]);
755 		cookie_size = numvec * sizeof(ppt->msix.cookie[0]);
756 		arg_size = numvec * sizeof(ppt->msix.arg[0]);
757 
758 		ppt->msix.res = malloc(res_size, M_PPTMSIX, M_WAITOK | M_ZERO);
759 		ppt->msix.cookie = malloc(cookie_size, M_PPTMSIX,
760 					  M_WAITOK | M_ZERO);
761 		ppt->msix.arg = malloc(arg_size, M_PPTMSIX, M_WAITOK | M_ZERO);
762 
763 		rid = dinfo->cfg.msix.msix_table_bar;
764 		ppt->msix.msix_table_res = bus_alloc_resource_any(ppt->dev,
765 					       SYS_RES_MEMORY, &rid, RF_ACTIVE);
766 
767 		if (ppt->msix.msix_table_res == NULL) {
768 			ppt_teardown_msix(ppt);
769 			error = ENOSPC;
770 			goto out;
771 		}
772 		ppt->msix.msix_table_rid = rid;
773 
774 		if (dinfo->cfg.msix.msix_table_bar !=
775 		    dinfo->cfg.msix.msix_pba_bar) {
776 			rid = dinfo->cfg.msix.msix_pba_bar;
777 			ppt->msix.msix_pba_res = bus_alloc_resource_any(
778 			    ppt->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
779 
780 			if (ppt->msix.msix_pba_res == NULL) {
781 				ppt_teardown_msix(ppt);
782 				error = ENOSPC;
783 				goto out;
784 			}
785 			ppt->msix.msix_pba_rid = rid;
786 		}
787 
788 		alloced = numvec;
789 		error = pci_alloc_msix(ppt->dev, &alloced);
790 		if (error || alloced != numvec) {
791 			ppt_teardown_msix(ppt);
792 			if (error == 0)
793 				error = ENOSPC;
794 			goto out;
795 		}
796 	}
797 
798 	if (idx >= ppt->msix.num_msgs) {
799 		error = EINVAL;
800 		goto out;
801 	}
802 
803 	if ((vector_control & PCIM_MSIX_VCTRL_MASK) == 0) {
804 		/* Tear down the IRQ if it's already set up */
805 		ppt_teardown_msix_intr(ppt, idx);
806 
807 		/* Allocate the IRQ resource */
808 		ppt->msix.cookie[idx] = NULL;
809 		rid = ppt->msix.startrid + idx;
810 		ppt->msix.res[idx] = bus_alloc_resource_any(ppt->dev, SYS_RES_IRQ,
811 							    &rid, RF_ACTIVE);
812 		if (ppt->msix.res[idx] == NULL) {
813 			error = ENXIO;
814 			goto out;
815 		}
816 
817 		ppt->msix.arg[idx].pptdev = ppt;
818 		ppt->msix.arg[idx].addr = addr;
819 		ppt->msix.arg[idx].msg_data = msg;
820 
821 		/* Setup the MSI-X interrupt */
822 		error = bus_setup_intr(ppt->dev, ppt->msix.res[idx],
823 				       INTR_TYPE_NET | INTR_MPSAFE,
824 				       pptintr, NULL, &ppt->msix.arg[idx],
825 				       &ppt->msix.cookie[idx]);
826 		if (error != 0) {
827 			bus_release_resource(ppt->dev, SYS_RES_IRQ, rid, ppt->msix.res[idx]);
828 			ppt->msix.cookie[idx] = NULL;
829 			ppt->msix.res[idx] = NULL;
830 			error = ENXIO;
831 			goto out;
832 		}
833 	} else {
834 		/* Masked, tear it down if it's already been set up */
835 		ppt_teardown_msix_intr(ppt, idx);
836 	}
837 out:
838 	PPT_UNLOCK();
839 	return (error);
840 }
841 
842 int
ppt_disable_msix(struct vm * vm,int bus,int slot,int func)843 ppt_disable_msix(struct vm *vm, int bus, int slot, int func)
844 {
845 	struct pptdev *ppt;
846 	int error;
847 
848 	PPT_LOCK();
849 	error = ppt_find(vm, bus, slot, func, &ppt);
850 	if (error != 0) {
851 		PPT_UNLOCK();
852 		return (error);
853 	}
854 	ppt_teardown_msix(ppt);
855 	PPT_UNLOCK();
856 	return (0);
857 }
858