Home | History | Annotate | Line # | Download | only in pci
virtio_pci.c revision 1.40
      1 /* $NetBSD: virtio_pci.c,v 1.40 2023/03/31 07:34:26 yamaguchi Exp $ */
      2 
      3 /*
      4  * Copyright (c) 2020 The NetBSD Foundation, Inc.
      5  * Copyright (c) 2012 Stefan Fritsch.
      6  * Copyright (c) 2010 Minoura Makoto.
      7  * All rights reserved.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  *
     18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     28  */
     29 
     30 #include <sys/cdefs.h>
     31 __KERNEL_RCSID(0, "$NetBSD: virtio_pci.c,v 1.40 2023/03/31 07:34:26 yamaguchi Exp $");
     32 
     33 #include <sys/param.h>
     34 #include <sys/systm.h>
     35 #include <sys/kmem.h>
     36 #include <sys/module.h>
     37 #include <sys/endian.h>
     38 #include <sys/interrupt.h>
     39 #include <sys/syslog.h>
     40 
     41 #include <sys/device.h>
     42 
     43 #include <dev/pci/pcidevs.h>
     44 #include <dev/pci/pcireg.h>
     45 #include <dev/pci/pcivar.h>
     46 
     47 #include <dev/pci/virtioreg.h> /* XXX: move to non-pci */
     48 #include <dev/pci/virtio_pcireg.h>
     49 
     50 #define VIRTIO_PRIVATE
     51 #include <dev/pci/virtiovar.h> /* XXX: move to non-pci */
     52 
     53 
     54 #define VIRTIO_PCI_LOG(_sc, _use_log, _fmt, _args...)	\
     55 do {							\
     56 	if ((_use_log)) {				\
     57 		log(LOG_DEBUG, "%s: " _fmt,		\
     58 		    device_xname((_sc)->sc_dev),	\
     59 		    ##_args);				\
     60 	} else {					\
     61 		aprint_error_dev((_sc)->sc_dev,		\
     62 		    _fmt, ##_args);			\
     63 	}						\
     64 } while(0)
     65 
     66 static int	virtio_pci_match(device_t, cfdata_t, void *);
     67 static void	virtio_pci_attach(device_t, device_t, void *);
     68 static int	virtio_pci_rescan(device_t, const char *, const int *);
     69 static int	virtio_pci_detach(device_t, int);
     70 
     71 
     72 #define NMAPREG		((PCI_MAPREG_END - PCI_MAPREG_START) / \
     73 				sizeof(pcireg_t))
     74 struct virtio_pci_softc {
     75 	struct virtio_softc	sc_sc;
     76 	bool			sc_intr_pervq;
     77 
     78 	/* IO space */
     79 	bus_space_tag_t		sc_iot;
     80 	bus_space_handle_t	sc_ioh;
     81 	bus_size_t		sc_iosize;
     82 	bus_size_t		sc_mapped_iosize;
     83 
     84 	/* BARs */
     85 	bus_space_tag_t		sc_bars_iot[NMAPREG];
     86 	bus_space_handle_t	sc_bars_ioh[NMAPREG];
     87 	bus_size_t		sc_bars_iosize[NMAPREG];
     88 
     89 	/* notify space */
     90 	bus_space_tag_t		sc_notify_iot;
     91 	bus_space_handle_t	sc_notify_ioh;
     92 	bus_size_t		sc_notify_iosize;
     93 	uint32_t		sc_notify_off_multiplier;
     94 
     95 	/* isr space */
     96 	bus_space_tag_t		sc_isr_iot;
     97 	bus_space_handle_t	sc_isr_ioh;
     98 	bus_size_t		sc_isr_iosize;
     99 
    100 	/* generic */
    101 	struct pci_attach_args	sc_pa;
    102 	pci_intr_handle_t	*sc_ihp;
    103 	void			**sc_ihs;
    104 	int			sc_ihs_num;
    105 	int			sc_devcfg_offset;	/* for 0.9 */
    106 };
    107 
    108 static int	virtio_pci_attach_09(device_t, void *);
    109 static void	virtio_pci_kick_09(struct virtio_softc *, uint16_t);
    110 static uint16_t	virtio_pci_read_queue_size_09(struct virtio_softc *, uint16_t);
    111 static void	virtio_pci_setup_queue_09(struct virtio_softc *, uint16_t, uint64_t);
    112 static void	virtio_pci_set_status_09(struct virtio_softc *, int);
    113 static void	virtio_pci_negotiate_features_09(struct virtio_softc *, uint64_t);
    114 
    115 static int	virtio_pci_attach_10(device_t, void *);
    116 static void	virtio_pci_kick_10(struct virtio_softc *, uint16_t);
    117 static uint16_t	virtio_pci_read_queue_size_10(struct virtio_softc *, uint16_t);
    118 static void	virtio_pci_setup_queue_10(struct virtio_softc *, uint16_t, uint64_t);
    119 static void	virtio_pci_set_status_10(struct virtio_softc *, int);
    120 static void	virtio_pci_negotiate_features_10(struct virtio_softc *, uint64_t);
    121 static int	virtio_pci_find_cap(struct virtio_pci_softc *psc, int cfg_type, void *buf, int buflen);
    122 
    123 static int	virtio_pci_alloc_interrupts(struct virtio_softc *);
    124 static void	virtio_pci_free_interrupts(struct virtio_softc *);
    125 static int	virtio_pci_adjust_config_region(struct virtio_pci_softc *psc);
    126 static int	virtio_pci_intr(void *arg);
    127 static int	virtio_pci_msix_queue_intr(void *);
    128 static int	virtio_pci_msix_config_intr(void *);
    129 static int	virtio_pci_setup_interrupts_09(struct virtio_softc *, int);
    130 static int	virtio_pci_setup_interrupts_10(struct virtio_softc *, int);
    131 static int	virtio_pci_establish_msix_interrupts(struct virtio_softc *,
    132 		    struct pci_attach_args *);
    133 static int	virtio_pci_establish_intx_interrupt(struct virtio_softc *,
    134 		    struct pci_attach_args *);
    135 static bool	virtio_pci_msix_enabled(struct virtio_pci_softc *);
    136 
    137 #define VIRTIO_MSIX_CONFIG_VECTOR_INDEX	0
    138 #define VIRTIO_MSIX_QUEUE_VECTOR_INDEX	1
    139 
    140 /*
    141  * When using PCI attached virtio on aarch64-eb under Qemu, the IO space
    142  * suddenly read BIG_ENDIAN where it should stay LITTLE_ENDIAN. The data read
    143  * 1 byte at a time seem OK but reading bigger lengths result in swapped
    144  * endian. This is most notable on reading 8 byters since we can't use
    145  * bus_space_{read,write}_8().
    146  */
    147 
    148 #if defined(__aarch64__) && BYTE_ORDER == BIG_ENDIAN
    149 #	define READ_ENDIAN_09	BIG_ENDIAN	/* should be LITTLE_ENDIAN */
    150 #	define READ_ENDIAN_10	BIG_ENDIAN
    151 #	define STRUCT_ENDIAN_09	BIG_ENDIAN
    152 #	define STRUCT_ENDIAN_10	LITTLE_ENDIAN
    153 #elif BYTE_ORDER == BIG_ENDIAN
    154 #	define READ_ENDIAN_09	LITTLE_ENDIAN
    155 #	define READ_ENDIAN_10	BIG_ENDIAN
    156 #	define STRUCT_ENDIAN_09	BIG_ENDIAN
    157 #	define STRUCT_ENDIAN_10	LITTLE_ENDIAN
    158 #else /* little endian */
    159 #	define READ_ENDIAN_09	LITTLE_ENDIAN
    160 #	define READ_ENDIAN_10	LITTLE_ENDIAN
    161 #	define STRUCT_ENDIAN_09	LITTLE_ENDIAN
    162 #	define STRUCT_ENDIAN_10	LITTLE_ENDIAN
    163 #endif
    164 
    165 
    166 CFATTACH_DECL3_NEW(virtio_pci, sizeof(struct virtio_pci_softc),
    167     virtio_pci_match, virtio_pci_attach, virtio_pci_detach, NULL,
    168     virtio_pci_rescan, NULL, DVF_DETACH_SHUTDOWN);
    169 
    170 static const struct virtio_ops virtio_pci_ops_09 = {
    171 	.kick = virtio_pci_kick_09,
    172 	.read_queue_size = virtio_pci_read_queue_size_09,
    173 	.setup_queue = virtio_pci_setup_queue_09,
    174 	.set_status = virtio_pci_set_status_09,
    175 	.neg_features = virtio_pci_negotiate_features_09,
    176 	.alloc_interrupts = virtio_pci_alloc_interrupts,
    177 	.free_interrupts = virtio_pci_free_interrupts,
    178 	.setup_interrupts = virtio_pci_setup_interrupts_09,
    179 };
    180 
    181 static const struct virtio_ops virtio_pci_ops_10 = {
    182 	.kick = virtio_pci_kick_10,
    183 	.read_queue_size = virtio_pci_read_queue_size_10,
    184 	.setup_queue = virtio_pci_setup_queue_10,
    185 	.set_status = virtio_pci_set_status_10,
    186 	.neg_features = virtio_pci_negotiate_features_10,
    187 	.alloc_interrupts = virtio_pci_alloc_interrupts,
    188 	.free_interrupts = virtio_pci_free_interrupts,
    189 	.setup_interrupts = virtio_pci_setup_interrupts_10,
    190 };
    191 
    192 static int
    193 virtio_pci_match(device_t parent, cfdata_t match, void *aux)
    194 {
    195 	struct pci_attach_args *pa;
    196 
    197 	pa = (struct pci_attach_args *)aux;
    198 	switch (PCI_VENDOR(pa->pa_id)) {
    199 	case PCI_VENDOR_QUMRANET:
    200 		/* Transitional devices MUST have a PCI Revision ID of 0. */
    201 		if (((PCI_PRODUCT_QUMRANET_VIRTIO_1000 <=
    202 		      PCI_PRODUCT(pa->pa_id)) &&
    203 		     (PCI_PRODUCT(pa->pa_id) <=
    204 		      PCI_PRODUCT_QUMRANET_VIRTIO_103F)) &&
    205 	              PCI_REVISION(pa->pa_class) == 0)
    206 			return 1;
    207 		/*
    208 		 * Non-transitional devices SHOULD have a PCI Revision
    209 		 * ID of 1 or higher.  Drivers MUST match any PCI
    210 		 * Revision ID value.
    211 		 */
    212 		if (((PCI_PRODUCT_QUMRANET_VIRTIO_1040 <=
    213 		      PCI_PRODUCT(pa->pa_id)) &&
    214 		     (PCI_PRODUCT(pa->pa_id) <=
    215 		      PCI_PRODUCT_QUMRANET_VIRTIO_107F)) &&
    216 		      /* XXX: TODO */
    217 		      PCI_REVISION(pa->pa_class) == 1)
    218 			return 1;
    219 		break;
    220 	}
    221 
    222 	return 0;
    223 }
    224 
    225 static void
    226 virtio_pci_attach(device_t parent, device_t self, void *aux)
    227 {
    228 	struct virtio_pci_softc * const psc = device_private(self);
    229 	struct virtio_softc * const sc = &psc->sc_sc;
    230 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
    231 	pci_chipset_tag_t pc = pa->pa_pc;
    232 	pcitag_t tag = pa->pa_tag;
    233 	int revision;
    234 	int ret;
    235 	pcireg_t id;
    236 	pcireg_t csr;
    237 
    238 	revision = PCI_REVISION(pa->pa_class);
    239 	switch (revision) {
    240 	case 0:
    241 		/* subsystem ID shows what I am */
    242 		id = PCI_SUBSYS_ID(pci_conf_read(pc, tag, PCI_SUBSYS_ID_REG));
    243 		break;
    244 	case 1:
    245 		/* pci product number shows what I am */
    246 		id = PCI_PRODUCT(pa->pa_id) - PCI_PRODUCT_QUMRANET_VIRTIO_1040;
    247 		break;
    248 	default:
    249 		aprint_normal(": unknown revision 0x%02x; giving up\n",
    250 			      revision);
    251 		return;
    252 	}
    253 
    254 	aprint_normal("\n");
    255 	aprint_naive("\n");
    256 	virtio_print_device_type(self, id, revision);
    257 
    258 	csr = pci_conf_read(pc, tag, PCI_COMMAND_STATUS_REG);
    259 	csr |= PCI_COMMAND_MASTER_ENABLE | PCI_COMMAND_IO_ENABLE;
    260 	pci_conf_write(pc, tag, PCI_COMMAND_STATUS_REG, csr);
    261 
    262 	sc->sc_dev = self;
    263 	psc->sc_pa = *pa;
    264 	psc->sc_iot = pa->pa_iot;
    265 
    266 	sc->sc_dmat = pa->pa_dmat;
    267 	if (pci_dma64_available(pa))
    268 		sc->sc_dmat = pa->pa_dmat64;
    269 
    270 	/* attach is dependent on revision */
    271 	ret = 0;
    272 	if (revision == 1) {
    273 		/* try to attach 1.0 */
    274 		ret = virtio_pci_attach_10(self, aux);
    275 	}
    276 	if (ret == 0 && revision == 0) {
    277 		/* revision 0 means 0.9 only or both 0.9 and 1.0 */
    278 		ret = virtio_pci_attach_09(self, aux);
    279 	}
    280 	if (ret) {
    281 		aprint_error_dev(self, "cannot attach (%d)\n", ret);
    282 		return;
    283 	}
    284 	KASSERT(sc->sc_ops);
    285 
    286 	/* preset config region */
    287 	psc->sc_devcfg_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
    288 	if (virtio_pci_adjust_config_region(psc))
    289 		return;
    290 
    291 	/* generic */
    292 	virtio_device_reset(sc);
    293 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
    294 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
    295 
    296 	sc->sc_childdevid = id;
    297 	sc->sc_child = NULL;
    298 	virtio_pci_rescan(self, NULL, NULL);
    299 	return;
    300 }
    301 
    302 /* ARGSUSED */
    303 static int
    304 virtio_pci_rescan(device_t self, const char *ifattr, const int *locs)
    305 {
    306 	struct virtio_pci_softc * const psc = device_private(self);
    307 	struct virtio_softc * const sc = &psc->sc_sc;
    308 	struct virtio_attach_args va;
    309 
    310 	if (sc->sc_child)	/* Child already attached? */
    311 		return 0;
    312 
    313 	memset(&va, 0, sizeof(va));
    314 	va.sc_childdevid = sc->sc_childdevid;
    315 
    316 	config_found(self, &va, NULL, CFARGS_NONE);
    317 
    318 	if (virtio_attach_failed(sc))
    319 		return 0;
    320 
    321 	return 0;
    322 }
    323 
    324 
    325 static int
    326 virtio_pci_detach(device_t self, int flags)
    327 {
    328 	struct virtio_pci_softc * const psc = device_private(self);
    329 	struct virtio_softc * const sc = &psc->sc_sc;
    330 	int r;
    331 
    332 	r = config_detach_children(self, flags);
    333 	if (r != 0)
    334 		return r;
    335 
    336 	/* Check that child detached properly */
    337 	KASSERT(ISSET(sc->sc_child_flags, VIRTIO_CHILD_DETACHED));
    338 	KASSERT(sc->sc_vqs == NULL);
    339 	KASSERT(psc->sc_ihs_num == 0);
    340 
    341 	if (psc->sc_iosize)
    342 		bus_space_unmap(psc->sc_iot, psc->sc_ioh,
    343 			psc->sc_mapped_iosize);
    344 	psc->sc_iosize = 0;
    345 
    346 	return 0;
    347 }
    348 
    349 
    350 static int
    351 virtio_pci_attach_09(device_t self, void *aux)
    352 	//struct virtio_pci_softc *psc, struct pci_attach_args *pa)
    353 {
    354 	struct virtio_pci_softc * const psc = device_private(self);
    355 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
    356 	struct virtio_softc * const sc = &psc->sc_sc;
    357 //	pci_chipset_tag_t pc = pa->pa_pc;
    358 //	pcitag_t tag = pa->pa_tag;
    359 
    360 	/* complete IO region */
    361 	if (pci_mapreg_map(pa, PCI_MAPREG_START, PCI_MAPREG_TYPE_IO, 0,
    362 			   &psc->sc_iot, &psc->sc_ioh, NULL, &psc->sc_iosize)) {
    363 		aprint_error_dev(self, "can't map i/o space\n");
    364 		return EIO;
    365 	}
    366 	psc->sc_mapped_iosize = psc->sc_iosize;
    367 
    368 	/* queue space */
    369 	if (bus_space_subregion(psc->sc_iot, psc->sc_ioh,
    370 			VIRTIO_CONFIG_QUEUE_NOTIFY, 2, &psc->sc_notify_ioh)) {
    371 		aprint_error_dev(self, "can't map notify i/o space\n");
    372 		return EIO;
    373 	}
    374 	psc->sc_notify_iosize = 2;
    375 	psc->sc_notify_iot = psc->sc_iot;
    376 
    377 	/* ISR space */
    378 	if (bus_space_subregion(psc->sc_iot, psc->sc_ioh,
    379 			VIRTIO_CONFIG_ISR_STATUS, 1, &psc->sc_isr_ioh)) {
    380 		aprint_error_dev(self, "can't map isr i/o space\n");
    381 		return EIO;
    382 	}
    383 	psc->sc_isr_iosize = 1;
    384 	psc->sc_isr_iot = psc->sc_iot;
    385 
    386 	/* set our version 0.9 ops */
    387 	sc->sc_ops = &virtio_pci_ops_09;
    388 	sc->sc_bus_endian    = READ_ENDIAN_09;
    389 	sc->sc_struct_endian = STRUCT_ENDIAN_09;
    390 	return 0;
    391 }
    392 
    393 
    394 static int
    395 virtio_pci_attach_10(device_t self, void *aux)
    396 {
    397 	struct virtio_pci_softc * const psc = device_private(self);
    398 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
    399 	struct virtio_softc * const sc = &psc->sc_sc;
    400 	pci_chipset_tag_t pc = pa->pa_pc;
    401 	pcitag_t tag = pa->pa_tag;
    402 
    403 	struct virtio_pci_cap common, isr, device;
    404 	struct virtio_pci_notify_cap notify;
    405 	int have_device_cfg = 0;
    406 	bus_size_t bars[NMAPREG] = { 0 };
    407 	int bars_idx[NMAPREG] = { 0 };
    408 	struct virtio_pci_cap *caps[] = { &common, &isr, &device, &notify.cap };
    409 	int i, j, ret = 0;
    410 
    411 	if (virtio_pci_find_cap(psc, VIRTIO_PCI_CAP_COMMON_CFG,
    412 			&common, sizeof(common)))
    413 		return ENODEV;
    414 	if (virtio_pci_find_cap(psc, VIRTIO_PCI_CAP_NOTIFY_CFG,
    415 			&notify, sizeof(notify)))
    416 		return ENODEV;
    417 	if (virtio_pci_find_cap(psc, VIRTIO_PCI_CAP_ISR_CFG,
    418 			&isr, sizeof(isr)))
    419 		return ENODEV;
    420 	if (virtio_pci_find_cap(psc, VIRTIO_PCI_CAP_DEVICE_CFG,
    421 			&device, sizeof(device)))
    422 		memset(&device, 0, sizeof(device));
    423 	else
    424 		have_device_cfg = 1;
    425 
    426 	/* Figure out which bars we need to map */
    427 	for (i = 0; i < __arraycount(caps); i++) {
    428 		int bar = caps[i]->bar;
    429 		bus_size_t len = caps[i]->offset + caps[i]->length;
    430 		if (caps[i]->length == 0)
    431 			continue;
    432 		if (bars[bar] < len)
    433 			bars[bar] = len;
    434 	}
    435 
    436 	for (i = j = 0; i < __arraycount(bars); i++) {
    437 		int reg;
    438 		pcireg_t type;
    439 		if (bars[i] == 0)
    440 			continue;
    441 		reg = PCI_BAR(i);
    442 		type = pci_mapreg_type(pc, tag, reg);
    443 		if (pci_mapreg_map(pa, reg, type, 0,
    444 				&psc->sc_bars_iot[j], &psc->sc_bars_ioh[j],
    445 				NULL, &psc->sc_bars_iosize[j])) {
    446 			aprint_error_dev(self, "can't map bar %u \n", i);
    447 			ret = EIO;
    448 			goto err;
    449 		}
    450 		aprint_debug_dev(self,
    451 		    "bar[%d]: iot %p, size 0x%" PRIxBUSSIZE "\n",
    452 		    j, psc->sc_bars_iot[j], psc->sc_bars_iosize[j]);
    453 		bars_idx[i] = j;
    454 		j++;
    455 	}
    456 
    457 	i = bars_idx[notify.cap.bar];
    458 	if (bus_space_subregion(psc->sc_bars_iot[i], psc->sc_bars_ioh[i],
    459 			notify.cap.offset, notify.cap.length,
    460 			&psc->sc_notify_ioh)) {
    461 		aprint_error_dev(self, "can't map notify i/o space\n");
    462 		ret = EIO;
    463 		goto err;
    464 	}
    465 	psc->sc_notify_iosize = notify.cap.length;
    466 	psc->sc_notify_iot = psc->sc_bars_iot[i];
    467 	psc->sc_notify_off_multiplier = le32toh(notify.notify_off_multiplier);
    468 
    469 	if (have_device_cfg) {
    470 		i = bars_idx[device.bar];
    471 		if (bus_space_subregion(psc->sc_bars_iot[i], psc->sc_bars_ioh[i],
    472 				device.offset, device.length,
    473 				&sc->sc_devcfg_ioh)) {
    474 			aprint_error_dev(self, "can't map devcfg i/o space\n");
    475 			ret = EIO;
    476 			goto err;
    477 		}
    478 		aprint_debug_dev(self,
    479 			"device.offset = 0x%x, device.length = 0x%x\n",
    480 			device.offset, device.length);
    481 		sc->sc_devcfg_iosize = device.length;
    482 		sc->sc_devcfg_iot = psc->sc_bars_iot[i];
    483 	}
    484 
    485 	i = bars_idx[isr.bar];
    486 	if (bus_space_subregion(psc->sc_bars_iot[i], psc->sc_bars_ioh[i],
    487 			isr.offset, isr.length, &psc->sc_isr_ioh)) {
    488 		aprint_error_dev(self, "can't map isr i/o space\n");
    489 		ret = EIO;
    490 		goto err;
    491 	}
    492 	psc->sc_isr_iosize = isr.length;
    493 	psc->sc_isr_iot = psc->sc_bars_iot[i];
    494 
    495 	i = bars_idx[common.bar];
    496 	if (bus_space_subregion(psc->sc_bars_iot[i], psc->sc_bars_ioh[i],
    497 			common.offset, common.length, &psc->sc_ioh)) {
    498 		aprint_error_dev(self, "can't map common i/o space\n");
    499 		ret = EIO;
    500 		goto err;
    501 	}
    502 	psc->sc_iosize = common.length;
    503 	psc->sc_iot = psc->sc_bars_iot[i];
    504 	psc->sc_mapped_iosize = psc->sc_bars_iosize[i];
    505 
    506 	psc->sc_sc.sc_version_1 = 1;
    507 
    508 	/* set our version 1.0 ops */
    509 	sc->sc_ops = &virtio_pci_ops_10;
    510 	sc->sc_bus_endian    = READ_ENDIAN_10;
    511 	sc->sc_struct_endian = STRUCT_ENDIAN_10;
    512 	return 0;
    513 
    514 err:
    515 	/* undo our pci_mapreg_map()s */
    516 	for (i = 0; i < __arraycount(bars); i++) {
    517 		if (psc->sc_bars_iosize[i] == 0)
    518 			continue;
    519 		bus_space_unmap(psc->sc_bars_iot[i], psc->sc_bars_ioh[i],
    520 				psc->sc_bars_iosize[i]);
    521 	}
    522 	return ret;
    523 }
    524 
    525 /* v1.0 attach helper */
    526 static int
    527 virtio_pci_find_cap(struct virtio_pci_softc *psc, int cfg_type, void *buf, int buflen)
    528 {
    529 	device_t self = psc->sc_sc.sc_dev;
    530 	pci_chipset_tag_t pc = psc->sc_pa.pa_pc;
    531 	pcitag_t tag = psc->sc_pa.pa_tag;
    532 	unsigned int offset, i, len;
    533 	union {
    534 		pcireg_t reg[8];
    535 		struct virtio_pci_cap vcap;
    536 	} *v = buf;
    537 
    538 	if (buflen < sizeof(struct virtio_pci_cap))
    539 		return ERANGE;
    540 
    541 	if (!pci_get_capability(pc, tag, PCI_CAP_VENDSPEC, &offset, &v->reg[0]))
    542 		return ENOENT;
    543 
    544 	do {
    545 		for (i = 0; i < 4; i++)
    546 			v->reg[i] =
    547 				le32toh(pci_conf_read(pc, tag, offset + i * 4));
    548 		if (v->vcap.cfg_type == cfg_type)
    549 			break;
    550 		offset = v->vcap.cap_next;
    551 	} while (offset != 0);
    552 
    553 	if (offset == 0)
    554 		return ENOENT;
    555 
    556 	if (v->vcap.cap_len > sizeof(struct virtio_pci_cap)) {
    557 		len = roundup(v->vcap.cap_len, sizeof(pcireg_t));
    558 		if (len > buflen) {
    559 			aprint_error_dev(self, "%s cap too large\n", __func__);
    560 			return ERANGE;
    561 		}
    562 		for (i = 4; i < len / sizeof(pcireg_t);  i++)
    563 			v->reg[i] =
    564 				le32toh(pci_conf_read(pc, tag, offset + i * 4));
    565 	}
    566 
    567 	/* endian fixup */
    568 	v->vcap.offset = le32toh(v->vcap.offset);
    569 	v->vcap.length = le32toh(v->vcap.length);
    570 	return 0;
    571 }
    572 
    573 
    574 /* -------------------------------------
    575  * Version 0.9 support
    576  * -------------------------------------*/
    577 
    578 static void
    579 virtio_pci_kick_09(struct virtio_softc *sc, uint16_t idx)
    580 {
    581 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    582 
    583 	bus_space_write_2(psc->sc_notify_iot, psc->sc_notify_ioh, 0, idx);
    584 }
    585 
    586 /* only applicable for v 0.9 but also called for 1.0 */
    587 static int
    588 virtio_pci_adjust_config_region(struct virtio_pci_softc *psc)
    589 {
    590 	struct virtio_softc * const sc = &psc->sc_sc;
    591 	device_t self = sc->sc_dev;
    592 
    593 	if (psc->sc_sc.sc_version_1)
    594 		return 0;
    595 
    596 	sc->sc_devcfg_iosize = psc->sc_iosize - psc->sc_devcfg_offset;
    597 	sc->sc_devcfg_iot = psc->sc_iot;
    598 	if (bus_space_subregion(psc->sc_iot, psc->sc_ioh,
    599 			psc->sc_devcfg_offset, sc->sc_devcfg_iosize,
    600 			&sc->sc_devcfg_ioh)) {
    601 		aprint_error_dev(self, "can't map config i/o space\n");
    602 		return EIO;
    603 	}
    604 
    605 	return 0;
    606 }
    607 
    608 static uint16_t
    609 virtio_pci_read_queue_size_09(struct virtio_softc *sc, uint16_t idx)
    610 {
    611 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    612 
    613 	bus_space_write_2(psc->sc_iot, psc->sc_ioh,
    614 	    VIRTIO_CONFIG_QUEUE_SELECT, idx);
    615 	return bus_space_read_2(psc->sc_iot, psc->sc_ioh,
    616 	    VIRTIO_CONFIG_QUEUE_SIZE);
    617 }
    618 
    619 static void
    620 virtio_pci_setup_queue_09(struct virtio_softc *sc, uint16_t idx, uint64_t addr)
    621 {
    622 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    623 
    624 	bus_space_write_2(psc->sc_iot, psc->sc_ioh,
    625 	    VIRTIO_CONFIG_QUEUE_SELECT, idx);
    626 	bus_space_write_4(psc->sc_iot, psc->sc_ioh,
    627 	    VIRTIO_CONFIG_QUEUE_ADDRESS, addr / VIRTIO_PAGE_SIZE);
    628 
    629 	if (psc->sc_ihs_num > 1) {
    630 		int vec = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    631 		if (psc->sc_intr_pervq)
    632 			vec += idx;
    633 		bus_space_write_2(psc->sc_iot, psc->sc_ioh,
    634 		    VIRTIO_CONFIG_MSI_QUEUE_VECTOR, vec);
    635 	}
    636 }
    637 
    638 static void
    639 virtio_pci_set_status_09(struct virtio_softc *sc, int status)
    640 {
    641 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    642 	int old = 0;
    643 
    644 	if (status != 0) {
    645 	    old = bus_space_read_1(psc->sc_iot, psc->sc_ioh,
    646 		VIRTIO_CONFIG_DEVICE_STATUS);
    647 	}
    648 	bus_space_write_1(psc->sc_iot, psc->sc_ioh,
    649 	    VIRTIO_CONFIG_DEVICE_STATUS, status|old);
    650 }
    651 
    652 static void
    653 virtio_pci_negotiate_features_09(struct virtio_softc *sc, uint64_t guest_features)
    654 {
    655 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    656 	uint32_t r;
    657 
    658 	r = bus_space_read_4(psc->sc_iot, psc->sc_ioh,
    659 	    VIRTIO_CONFIG_DEVICE_FEATURES);
    660 
    661 	r &= guest_features;
    662 
    663 	bus_space_write_4(psc->sc_iot, psc->sc_ioh,
    664 	    VIRTIO_CONFIG_GUEST_FEATURES, r);
    665 
    666 	sc->sc_active_features = r;
    667 }
    668 
    669 /* -------------------------------------
    670  * Version 1.0 support
    671  * -------------------------------------*/
    672 
    673 static void
    674 virtio_pci_kick_10(struct virtio_softc *sc, uint16_t idx)
    675 {
    676 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    677 	unsigned offset = sc->sc_vqs[idx].vq_notify_off *
    678 		psc->sc_notify_off_multiplier;
    679 
    680 	bus_space_write_2(psc->sc_notify_iot, psc->sc_notify_ioh, offset, idx);
    681 }
    682 
    683 
    684 static uint16_t
    685 virtio_pci_read_queue_size_10(struct virtio_softc *sc, uint16_t idx)
    686 {
    687 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    688 	bus_space_tag_t	   iot = psc->sc_iot;
    689 	bus_space_handle_t ioh = psc->sc_ioh;
    690 
    691 	bus_space_write_2(iot, ioh, VIRTIO_CONFIG1_QUEUE_SELECT, idx);
    692 	return bus_space_read_2(iot, ioh, VIRTIO_CONFIG1_QUEUE_SIZE);
    693 }
    694 
    695 /*
    696  * By definition little endian only in v1.0.  NB: "MAY" in the text
    697  * below refers to "independently" (i.e. the order of accesses) not
    698  * "32-bit" (which is restricted by the earlier "MUST").
    699  *
    700  * 4.1.3.1 Driver Requirements: PCI Device Layout
    701  *
    702  * For device configuration access, the driver MUST use ... 32-bit
    703  * wide and aligned accesses for ... 64-bit wide fields.  For 64-bit
    704  * fields, the driver MAY access each of the high and low 32-bit parts
    705  * of the field independently.
    706  */
    707 static __inline void
    708 virtio_pci_bus_space_write_8(bus_space_tag_t iot, bus_space_handle_t ioh,
    709      bus_size_t offset, uint64_t value)
    710 {
    711 #if _QUAD_HIGHWORD
    712 	bus_space_write_4(iot, ioh, offset, BUS_ADDR_LO32(value));
    713 	bus_space_write_4(iot, ioh, offset + 4, BUS_ADDR_HI32(value));
    714 #else
    715 	bus_space_write_4(iot, ioh, offset, BUS_ADDR_HI32(value));
    716 	bus_space_write_4(iot, ioh, offset + 4, BUS_ADDR_LO32(value));
    717 #endif
    718 }
    719 
    720 static void
    721 virtio_pci_setup_queue_10(struct virtio_softc *sc, uint16_t idx, uint64_t addr)
    722 {
    723 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    724 	struct virtqueue *vq = &sc->sc_vqs[idx];
    725 	bus_space_tag_t	   iot = psc->sc_iot;
    726 	bus_space_handle_t ioh = psc->sc_ioh;
    727 	KASSERT(vq->vq_index == idx);
    728 
    729 	bus_space_write_2(iot, ioh, VIRTIO_CONFIG1_QUEUE_SELECT, vq->vq_index);
    730 	if (addr == 0) {
    731 		bus_space_write_2(iot, ioh, VIRTIO_CONFIG1_QUEUE_ENABLE, 0);
    732 		virtio_pci_bus_space_write_8(iot, ioh,
    733 		    VIRTIO_CONFIG1_QUEUE_DESC,   0);
    734 		virtio_pci_bus_space_write_8(iot, ioh,
    735 		    VIRTIO_CONFIG1_QUEUE_AVAIL,  0);
    736 		virtio_pci_bus_space_write_8(iot, ioh,
    737 		    VIRTIO_CONFIG1_QUEUE_USED,   0);
    738 	} else {
    739 		virtio_pci_bus_space_write_8(iot, ioh,
    740 			VIRTIO_CONFIG1_QUEUE_DESC, addr);
    741 		virtio_pci_bus_space_write_8(iot, ioh,
    742 			VIRTIO_CONFIG1_QUEUE_AVAIL, addr + vq->vq_availoffset);
    743 		virtio_pci_bus_space_write_8(iot, ioh,
    744 			VIRTIO_CONFIG1_QUEUE_USED, addr + vq->vq_usedoffset);
    745 		bus_space_write_2(iot, ioh,
    746 			VIRTIO_CONFIG1_QUEUE_ENABLE, 1);
    747 		vq->vq_notify_off = bus_space_read_2(iot, ioh,
    748 			VIRTIO_CONFIG1_QUEUE_NOTIFY_OFF);
    749 	}
    750 
    751 	if (psc->sc_ihs_num > 1) {
    752 		int vec = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    753 		if (psc->sc_intr_pervq)
    754 			vec += idx;
    755 		bus_space_write_2(iot, ioh,
    756 			VIRTIO_CONFIG1_QUEUE_MSIX_VECTOR, vec);
    757 	}
    758 }
    759 
    760 static void
    761 virtio_pci_set_status_10(struct virtio_softc *sc, int status)
    762 {
    763 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    764 	bus_space_tag_t	   iot = psc->sc_iot;
    765 	bus_space_handle_t ioh = psc->sc_ioh;
    766 	int old = 0;
    767 
    768 	if (status)
    769 		old = bus_space_read_1(iot, ioh, VIRTIO_CONFIG1_DEVICE_STATUS);
    770 	bus_space_write_1(iot, ioh, VIRTIO_CONFIG1_DEVICE_STATUS, status | old);
    771 }
    772 
    773 void
    774 virtio_pci_negotiate_features_10(struct virtio_softc *sc, uint64_t guest_features)
    775 {
    776 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    777 	device_t self          =  sc->sc_dev;
    778 	bus_space_tag_t	   iot = psc->sc_iot;
    779 	bus_space_handle_t ioh = psc->sc_ioh;
    780 	uint64_t host, negotiated, device_status;
    781 
    782 	guest_features |= VIRTIO_F_VERSION_1;
    783 	/* notify on empty is 0.9 only */
    784 	guest_features &= ~VIRTIO_F_NOTIFY_ON_EMPTY;
    785 	sc->sc_active_features = 0;
    786 
    787 	bus_space_write_4(iot, ioh, VIRTIO_CONFIG1_DEVICE_FEATURE_SELECT, 0);
    788 	host = bus_space_read_4(iot, ioh, VIRTIO_CONFIG1_DEVICE_FEATURE);
    789 	bus_space_write_4(iot, ioh, VIRTIO_CONFIG1_DEVICE_FEATURE_SELECT, 1);
    790 	host |= (uint64_t)
    791 		bus_space_read_4(iot, ioh, VIRTIO_CONFIG1_DEVICE_FEATURE) << 32;
    792 
    793 	negotiated = host & guest_features;
    794 
    795 	bus_space_write_4(iot, ioh, VIRTIO_CONFIG1_DRIVER_FEATURE_SELECT, 0);
    796 	bus_space_write_4(iot, ioh, VIRTIO_CONFIG1_DRIVER_FEATURE,
    797 			negotiated & 0xffffffff);
    798 	bus_space_write_4(iot, ioh, VIRTIO_CONFIG1_DRIVER_FEATURE_SELECT, 1);
    799 	bus_space_write_4(iot, ioh, VIRTIO_CONFIG1_DRIVER_FEATURE,
    800 			negotiated >> 32);
    801 	virtio_pci_set_status_10(sc, VIRTIO_CONFIG_DEVICE_STATUS_FEATURES_OK);
    802 
    803 	device_status = bus_space_read_1(iot, ioh, VIRTIO_CONFIG1_DEVICE_STATUS);
    804 	if ((device_status & VIRTIO_CONFIG_DEVICE_STATUS_FEATURES_OK) == 0) {
    805 		aprint_error_dev(self, "feature negotiation failed\n");
    806 		bus_space_write_1(iot, ioh, VIRTIO_CONFIG1_DEVICE_STATUS,
    807 				VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
    808 		return;
    809 	}
    810 
    811 	if ((negotiated & VIRTIO_F_VERSION_1) == 0) {
    812 		aprint_error_dev(self, "host rejected version 1\n");
    813 		bus_space_write_1(iot, ioh, VIRTIO_CONFIG1_DEVICE_STATUS,
    814 				VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
    815 		return;
    816 	}
    817 
    818 	sc->sc_active_features = negotiated;
    819 	return;
    820 }
    821 
    822 
    823 /* -------------------------------------
    824  * Generic PCI interrupt code
    825  * -------------------------------------*/
    826 
    827 static int
    828 virtio_pci_setup_interrupts_10(struct virtio_softc *sc, int reinit)
    829 {
    830 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    831 	bus_space_tag_t	   iot = psc->sc_iot;
    832 	bus_space_handle_t ioh = psc->sc_ioh;
    833 	int vector, ret, qid;
    834 
    835 	if (!virtio_pci_msix_enabled(psc))
    836 		return 0;
    837 
    838 	vector = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    839 	bus_space_write_2(iot, ioh,
    840 		VIRTIO_CONFIG1_CONFIG_MSIX_VECTOR, vector);
    841 	ret = bus_space_read_2(iot, ioh, VIRTIO_CONFIG1_CONFIG_MSIX_VECTOR);
    842 	if (ret != vector) {
    843 		VIRTIO_PCI_LOG(sc, reinit,
    844 		    "can't set config msix vector\n");
    845 		return -1;
    846 	}
    847 
    848 	for (qid = 0; qid < sc->sc_nvqs; qid++) {
    849 		vector = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    850 
    851 		if (psc->sc_intr_pervq)
    852 			vector += qid;
    853 		bus_space_write_2(iot, ioh, VIRTIO_CONFIG1_QUEUE_SELECT, qid);
    854 		bus_space_write_2(iot, ioh, VIRTIO_CONFIG1_QUEUE_MSIX_VECTOR,
    855 			vector);
    856 		ret = bus_space_read_2(iot, ioh,
    857 			VIRTIO_CONFIG1_QUEUE_MSIX_VECTOR);
    858 		if (ret != vector) {
    859 			VIRTIO_PCI_LOG(sc, reinit, "can't set queue %d "
    860 			    "msix vector\n", qid);
    861 			return -1;
    862 		}
    863 	}
    864 
    865 	return 0;
    866 }
    867 
    868 static int
    869 virtio_pci_setup_interrupts_09(struct virtio_softc *sc, int reinit)
    870 {
    871 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    872 	int offset, vector, ret, qid;
    873 
    874 	if (!virtio_pci_msix_enabled(psc))
    875 		return 0;
    876 
    877 	offset = VIRTIO_CONFIG_MSI_CONFIG_VECTOR;
    878 	vector = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    879 
    880 	bus_space_write_2(psc->sc_iot, psc->sc_ioh, offset, vector);
    881 	ret = bus_space_read_2(psc->sc_iot, psc->sc_ioh, offset);
    882 	if (ret != vector) {
    883 		aprint_debug_dev(sc->sc_dev, "%s: expected=%d, actual=%d\n",
    884 		    __func__, vector, ret);
    885 		VIRTIO_PCI_LOG(sc, reinit,
    886 		    "can't set config msix vector\n");
    887 		return -1;
    888 	}
    889 
    890 	for (qid = 0; qid < sc->sc_nvqs; qid++) {
    891 		offset = VIRTIO_CONFIG_QUEUE_SELECT;
    892 		bus_space_write_2(psc->sc_iot, psc->sc_ioh, offset, qid);
    893 
    894 		offset = VIRTIO_CONFIG_MSI_QUEUE_VECTOR;
    895 		vector = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    896 
    897 		if (psc->sc_intr_pervq)
    898 			vector += qid;
    899 
    900 		bus_space_write_2(psc->sc_iot, psc->sc_ioh, offset, vector);
    901 		ret = bus_space_read_2(psc->sc_iot, psc->sc_ioh, offset);
    902 		if (ret != vector) {
    903 			aprint_debug_dev(sc->sc_dev, "%s[qid=%d]:"
    904 			    " expected=%d, actual=%d\n",
    905 			    __func__, qid, vector, ret);
    906 			VIRTIO_PCI_LOG(sc, reinit, "can't set queue %d "
    907 			    "msix vector\n", qid);
    908 			return -1;
    909 		}
    910 	}
    911 
    912 	return 0;
    913 }
    914 
    915 static int
    916 virtio_pci_establish_msix_interrupts(struct virtio_softc *sc,
    917     struct pci_attach_args *pa)
    918 {
    919 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
    920 	device_t self = sc->sc_dev;
    921 	pci_chipset_tag_t pc = pa->pa_pc;
    922 	struct virtqueue *vq;
    923 	char intrbuf[PCI_INTRSTR_LEN];
    924 	char intr_xname[INTRDEVNAMEBUF];
    925 	char const *intrstr;
    926 	int idx, qid, n;
    927 
    928 	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    929 	if (sc->sc_flags & VIRTIO_F_INTR_MPSAFE)
    930 		pci_intr_setattr(pc, &psc->sc_ihp[idx], PCI_INTR_MPSAFE, true);
    931 
    932 	snprintf(intr_xname, sizeof(intr_xname), "%s config",
    933 	    device_xname(sc->sc_dev));
    934 
    935 	psc->sc_ihs[idx] = pci_intr_establish_xname(pc, psc->sc_ihp[idx],
    936 	    sc->sc_ipl, virtio_pci_msix_config_intr, sc, intr_xname);
    937 	if (psc->sc_ihs[idx] == NULL) {
    938 		aprint_error_dev(self, "couldn't establish MSI-X for config\n");
    939 		goto error;
    940 	}
    941 
    942 	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    943 	if (psc->sc_intr_pervq) {
    944 		for (qid = 0; qid < sc->sc_nvqs; qid++) {
    945 			n = idx + qid;
    946 			vq = &sc->sc_vqs[qid];
    947 
    948 			snprintf(intr_xname, sizeof(intr_xname), "%s vq#%d",
    949 			    device_xname(sc->sc_dev), qid);
    950 
    951 			if (sc->sc_flags & VIRTIO_F_INTR_MPSAFE) {
    952 				pci_intr_setattr(pc, &psc->sc_ihp[n],
    953 				    PCI_INTR_MPSAFE, true);
    954 			}
    955 
    956 			psc->sc_ihs[n] = pci_intr_establish_xname(pc, psc->sc_ihp[n],
    957 			    sc->sc_ipl, vq->vq_intrhand, vq->vq_intrhand_arg, intr_xname);
    958 			if (psc->sc_ihs[n] == NULL) {
    959 				aprint_error_dev(self, "couldn't establish MSI-X for a vq\n");
    960 				goto error;
    961 			}
    962 		}
    963 	} else {
    964 		if (sc->sc_flags & VIRTIO_F_INTR_MPSAFE)
    965 			pci_intr_setattr(pc, &psc->sc_ihp[idx], PCI_INTR_MPSAFE, true);
    966 
    967 		snprintf(intr_xname, sizeof(intr_xname), "%s queues",
    968 		    device_xname(sc->sc_dev));
    969 		psc->sc_ihs[idx] = pci_intr_establish_xname(pc, psc->sc_ihp[idx],
    970 		    sc->sc_ipl, virtio_pci_msix_queue_intr, sc, intr_xname);
    971 		if (psc->sc_ihs[idx] == NULL) {
    972 			aprint_error_dev(self, "couldn't establish MSI-X for queues\n");
    973 			goto error;
    974 		}
    975 	}
    976 
    977 	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    978 	intrstr = pci_intr_string(pc, psc->sc_ihp[idx], intrbuf, sizeof(intrbuf));
    979 	aprint_normal_dev(self, "config interrupting at %s\n", intrstr);
    980 	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    981 	if (psc->sc_intr_pervq) {
    982 		kcpuset_t *affinity;
    983 		int affinity_to, r;
    984 
    985 		kcpuset_create(&affinity, false);
    986 
    987 		for (qid = 0; qid < sc->sc_nvqs; qid++) {
    988 			n = idx + qid;
    989 			affinity_to = (qid / 2) % ncpu;
    990 
    991 			intrstr = pci_intr_string(pc, psc->sc_ihp[n],
    992 			    intrbuf, sizeof(intrbuf));
    993 
    994 			kcpuset_zero(affinity);
    995 			kcpuset_set(affinity, affinity_to);
    996 			r = interrupt_distribute(psc->sc_ihs[n], affinity, NULL);
    997 			if (r == 0) {
    998 				aprint_normal_dev(self,
    999 				    "for vq #%d interrupting at %s affinity to %u\n",
   1000 				    qid, intrstr, affinity_to);
   1001 			} else {
   1002 				aprint_normal_dev(self,
   1003 				    "for vq #%d interrupting at %s\n",
   1004 				    qid, intrstr);
   1005 			}
   1006 		}
   1007 
   1008 		kcpuset_destroy(affinity);
   1009 	} else {
   1010 		intrstr = pci_intr_string(pc, psc->sc_ihp[idx], intrbuf, sizeof(intrbuf));
   1011 		aprint_normal_dev(self, "queues interrupting at %s\n", intrstr);
   1012 	}
   1013 
   1014 	return 0;
   1015 
   1016 error:
   1017 	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
   1018 	if (psc->sc_ihs[idx] != NULL)
   1019 		pci_intr_disestablish(psc->sc_pa.pa_pc, psc->sc_ihs[idx]);
   1020 	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
   1021 	if (psc->sc_intr_pervq) {
   1022 		for (qid = 0; qid < sc->sc_nvqs; qid++) {
   1023 			n = idx + qid;
   1024 			if (psc->sc_ihs[n] == NULL)
   1025 				continue;
   1026 			pci_intr_disestablish(psc->sc_pa.pa_pc, psc->sc_ihs[n]);
   1027 		}
   1028 
   1029 	} else {
   1030 		if (psc->sc_ihs[idx] != NULL)
   1031 			pci_intr_disestablish(psc->sc_pa.pa_pc, psc->sc_ihs[idx]);
   1032 	}
   1033 
   1034 	return -1;
   1035 }
   1036 
   1037 static int
   1038 virtio_pci_establish_intx_interrupt(struct virtio_softc *sc,
   1039     struct pci_attach_args *pa)
   1040 {
   1041 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
   1042 	device_t self = sc->sc_dev;
   1043 	pci_chipset_tag_t pc = pa->pa_pc;
   1044 	char intrbuf[PCI_INTRSTR_LEN];
   1045 	char const *intrstr;
   1046 
   1047 	if (sc->sc_flags & VIRTIO_F_INTR_MPSAFE)
   1048 		pci_intr_setattr(pc, &psc->sc_ihp[0], PCI_INTR_MPSAFE, true);
   1049 
   1050 	psc->sc_ihs[0] = pci_intr_establish_xname(pc, psc->sc_ihp[0],
   1051 	    sc->sc_ipl, virtio_pci_intr, sc, device_xname(sc->sc_dev));
   1052 	if (psc->sc_ihs[0] == NULL) {
   1053 		aprint_error_dev(self, "couldn't establish INTx\n");
   1054 		return -1;
   1055 	}
   1056 
   1057 	intrstr = pci_intr_string(pc, psc->sc_ihp[0], intrbuf, sizeof(intrbuf));
   1058 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
   1059 
   1060 	return 0;
   1061 }
   1062 
   1063 static int
   1064 virtio_pci_alloc_interrupts(struct virtio_softc *sc)
   1065 {
   1066 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
   1067 	device_t self = sc->sc_dev;
   1068 	pci_chipset_tag_t pc = psc->sc_pa.pa_pc;
   1069 	pcitag_t tag = psc->sc_pa.pa_tag;
   1070 	int error;
   1071 	int nmsix;
   1072 	int off;
   1073 	int counts[PCI_INTR_TYPE_SIZE];
   1074 	pci_intr_type_t max_type;
   1075 	pcireg_t ctl;
   1076 
   1077 	nmsix = pci_msix_count(psc->sc_pa.pa_pc, psc->sc_pa.pa_tag);
   1078 	aprint_debug_dev(self, "pci_msix_count=%d\n", nmsix);
   1079 
   1080 	/* We need at least two: one for config and the other for queues */
   1081 	if ((sc->sc_flags & VIRTIO_F_INTR_MSIX) == 0 || nmsix < 2) {
   1082 		/* Try INTx only */
   1083 		max_type = PCI_INTR_TYPE_INTX;
   1084 		counts[PCI_INTR_TYPE_INTX] = 1;
   1085 	} else {
   1086 		/* Try MSI-X first and INTx second */
   1087 		if (ISSET(sc->sc_flags, VIRTIO_F_INTR_PERVQ) &&
   1088 		    sc->sc_nvqs + VIRTIO_MSIX_QUEUE_VECTOR_INDEX <= nmsix) {
   1089 			nmsix = sc->sc_nvqs + VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
   1090 		} else {
   1091 			nmsix = 2;
   1092 		}
   1093 
   1094 		max_type = PCI_INTR_TYPE_MSIX;
   1095 		counts[PCI_INTR_TYPE_MSIX] = nmsix;
   1096 		counts[PCI_INTR_TYPE_MSI] = 0;
   1097 		counts[PCI_INTR_TYPE_INTX] = 1;
   1098 	}
   1099 
   1100 retry:
   1101 	error = pci_intr_alloc(&psc->sc_pa, &psc->sc_ihp, counts, max_type);
   1102 	if (error != 0) {
   1103 		aprint_error_dev(self, "couldn't map interrupt\n");
   1104 		return -1;
   1105 	}
   1106 
   1107 	if (pci_intr_type(pc, psc->sc_ihp[0]) == PCI_INTR_TYPE_MSIX) {
   1108 		psc->sc_intr_pervq = nmsix > 2 ? true : false;
   1109 		psc->sc_ihs = kmem_zalloc(sizeof(*psc->sc_ihs) * nmsix,
   1110 		    KM_SLEEP);
   1111 
   1112 		error = virtio_pci_establish_msix_interrupts(sc, &psc->sc_pa);
   1113 		if (error != 0) {
   1114 			kmem_free(psc->sc_ihs, sizeof(*psc->sc_ihs) * nmsix);
   1115 			pci_intr_release(pc, psc->sc_ihp, nmsix);
   1116 
   1117 			/* Retry INTx */
   1118 			max_type = PCI_INTR_TYPE_INTX;
   1119 			counts[PCI_INTR_TYPE_INTX] = 1;
   1120 			goto retry;
   1121 		}
   1122 
   1123 		psc->sc_ihs_num = nmsix;
   1124 		psc->sc_devcfg_offset = VIRTIO_CONFIG_DEVICE_CONFIG_MSI;
   1125 		virtio_pci_adjust_config_region(psc);
   1126 	} else if (pci_intr_type(pc, psc->sc_ihp[0]) == PCI_INTR_TYPE_INTX) {
   1127 		psc->sc_intr_pervq = false;
   1128 		psc->sc_ihs = kmem_zalloc(sizeof(*psc->sc_ihs) * 1,
   1129 		    KM_SLEEP);
   1130 
   1131 		error = virtio_pci_establish_intx_interrupt(sc, &psc->sc_pa);
   1132 		if (error != 0) {
   1133 			kmem_free(psc->sc_ihs, sizeof(*psc->sc_ihs) * 1);
   1134 			pci_intr_release(pc, psc->sc_ihp, 1);
   1135 			return -1;
   1136 		}
   1137 
   1138 		psc->sc_ihs_num = 1;
   1139 		psc->sc_devcfg_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
   1140 		virtio_pci_adjust_config_region(psc);
   1141 
   1142 		error = pci_get_capability(pc, tag, PCI_CAP_MSIX, &off, NULL);
   1143 		if (error != 0) {
   1144 			ctl = pci_conf_read(pc, tag, off + PCI_MSIX_CTL);
   1145 			ctl &= ~PCI_MSIX_CTL_ENABLE;
   1146 			pci_conf_write(pc, tag, off + PCI_MSIX_CTL, ctl);
   1147 		}
   1148 	}
   1149 
   1150 	if (!psc->sc_intr_pervq)
   1151 		CLR(sc->sc_flags, VIRTIO_F_INTR_PERVQ);
   1152 	return 0;
   1153 }
   1154 
   1155 static void
   1156 virtio_pci_free_interrupts(struct virtio_softc *sc)
   1157 {
   1158 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
   1159 
   1160 	for (int i = 0; i < psc->sc_ihs_num; i++) {
   1161 		if (psc->sc_ihs[i] == NULL)
   1162 			continue;
   1163 		pci_intr_disestablish(psc->sc_pa.pa_pc, psc->sc_ihs[i]);
   1164 		psc->sc_ihs[i] = NULL;
   1165 	}
   1166 
   1167 	if (psc->sc_ihs_num > 0)
   1168 		pci_intr_release(psc->sc_pa.pa_pc, psc->sc_ihp, psc->sc_ihs_num);
   1169 
   1170 	if (psc->sc_ihs != NULL) {
   1171 		kmem_free(psc->sc_ihs, sizeof(*psc->sc_ihs) * psc->sc_ihs_num);
   1172 		psc->sc_ihs = NULL;
   1173 	}
   1174 	psc->sc_ihs_num = 0;
   1175 }
   1176 
   1177 static bool
   1178 virtio_pci_msix_enabled(struct virtio_pci_softc *psc)
   1179 {
   1180 	pci_chipset_tag_t pc = psc->sc_pa.pa_pc;
   1181 
   1182 	if (pci_intr_type(pc, psc->sc_ihp[0]) == PCI_INTR_TYPE_MSIX)
   1183 		return true;
   1184 
   1185 	return false;
   1186 }
   1187 
   1188 /*
   1189  * Interrupt handler.
   1190  */
   1191 static int
   1192 virtio_pci_intr(void *arg)
   1193 {
   1194 	struct virtio_softc *sc = arg;
   1195 	struct virtio_pci_softc * const psc = (struct virtio_pci_softc *)sc;
   1196 	int isr, r = 0;
   1197 
   1198 	/* check and ack the interrupt */
   1199 	isr = bus_space_read_1(psc->sc_isr_iot, psc->sc_isr_ioh, 0);
   1200 	if (isr == 0)
   1201 		return 0;
   1202 	if ((isr & VIRTIO_CONFIG_ISR_CONFIG_CHANGE) &&
   1203 	    (sc->sc_config_change != NULL))
   1204 		r = (sc->sc_config_change)(sc);
   1205 	if (sc->sc_intrhand != NULL) {
   1206 		if (sc->sc_soft_ih != NULL)
   1207 			softint_schedule(sc->sc_soft_ih);
   1208 		else
   1209 			r |= (sc->sc_intrhand)(sc);
   1210 	}
   1211 
   1212 	return r;
   1213 }
   1214 
   1215 static int
   1216 virtio_pci_msix_queue_intr(void *arg)
   1217 {
   1218 	struct virtio_softc *sc = arg;
   1219 	int r = 0;
   1220 
   1221 	if (sc->sc_intrhand != NULL) {
   1222 		if (sc->sc_soft_ih != NULL)
   1223 			softint_schedule(sc->sc_soft_ih);
   1224 		else
   1225 			r |= (sc->sc_intrhand)(sc);
   1226 	}
   1227 
   1228 	return r;
   1229 }
   1230 
   1231 static int
   1232 virtio_pci_msix_config_intr(void *arg)
   1233 {
   1234 	struct virtio_softc *sc = arg;
   1235 	int r = 0;
   1236 
   1237 	if (sc->sc_config_change != NULL)
   1238 		r = (sc->sc_config_change)(sc);
   1239 	return r;
   1240 }
   1241 
   1242 MODULE(MODULE_CLASS_DRIVER, virtio_pci, "pci,virtio");
   1243 
   1244 #ifdef _MODULE
   1245 #include "ioconf.c"
   1246 #endif
   1247 
   1248 static int
   1249 virtio_pci_modcmd(modcmd_t cmd, void *opaque)
   1250 {
   1251 	int error = 0;
   1252 
   1253 #ifdef _MODULE
   1254 	switch (cmd) {
   1255 	case MODULE_CMD_INIT:
   1256 		error = config_init_component(cfdriver_ioconf_virtio_pci,
   1257 		    cfattach_ioconf_virtio_pci, cfdata_ioconf_virtio_pci);
   1258 		break;
   1259 	case MODULE_CMD_FINI:
   1260 		error = config_fini_component(cfdriver_ioconf_virtio_pci,
   1261 		    cfattach_ioconf_virtio_pci, cfdata_ioconf_virtio_pci);
   1262 		break;
   1263 	default:
   1264 		error = ENOTTY;
   1265 		break;
   1266 	}
   1267 #endif
   1268 
   1269 	return error;
   1270 }
   1271