Home | History | Annotate | Line # | Download | only in pci
virtio.c revision 1.7.2.8
      1 /*	$NetBSD: virtio.c,v 1.7.2.8 2017/08/28 17:52:25 skrll Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2010 Minoura Makoto.
      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 THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     26  */
     27 
     28 #include <sys/cdefs.h>
     29 __KERNEL_RCSID(0, "$NetBSD: virtio.c,v 1.7.2.8 2017/08/28 17:52:25 skrll Exp $");
     30 
     31 #include <sys/param.h>
     32 #include <sys/systm.h>
     33 #include <sys/kernel.h>
     34 #include <sys/atomic.h>
     35 #include <sys/bus.h>
     36 #include <sys/device.h>
     37 #include <sys/kmem.h>
     38 #include <sys/module.h>
     39 
     40 #include <dev/pci/pcidevs.h>
     41 #include <dev/pci/pcireg.h>
     42 #include <dev/pci/pcivar.h>
     43 
     44 #define VIRTIO_PRIVATE
     45 
     46 #include <dev/pci/virtioreg.h> /* XXX: move to non-pci */
     47 #include <dev/pci/virtiovar.h> /* XXX: move to non-pci */
     48 
     49 #define MINSEG_INDIRECT		2 /* use indirect if nsegs >= this value */
     50 
     51 static int	virtio_intr(void *arg);
     52 static int	virtio_msix_queue_intr(void *);
     53 static int	virtio_msix_config_intr(void *);
     54 static int	virtio_setup_msix_vectors(struct virtio_softc *);
     55 static int	virtio_setup_msix_interrupts(struct virtio_softc *,
     56 		    struct pci_attach_args *);
     57 static int	virtio_setup_intx_interrupt(struct virtio_softc *,
     58 		    struct pci_attach_args *);
     59 static int	virtio_setup_interrupts(struct virtio_softc *);
     60 static void	virtio_free_interrupts(struct virtio_softc *);
     61 static void	virtio_soft_intr(void *arg);
     62 static void	virtio_init_vq(struct virtio_softc *,
     63 		    struct virtqueue *, const bool);
     64 
     65 
     66 /* we use the legacy virtio spec, so the pci registers are host native
     67  * byte order, not pci (i.e. LE) byte order */
     68 static inline uint16_t
     69 nbo_bus_space_read_2(bus_space_tag_t space, bus_space_handle_t handle,
     70          bus_size_t offset)
     71 {
     72 	return le16toh(bus_space_read_2(space, handle, offset));
     73 }
     74 
     75 static inline uint32_t
     76 nbo_bus_space_read_4(bus_space_tag_t space, bus_space_handle_t handle,
     77 	bus_size_t offset)
     78 {
     79 	return le32toh(bus_space_read_4(space, handle, offset));
     80 }
     81 
     82 static void
     83 nbo_bus_space_write_2(bus_space_tag_t space, bus_space_handle_t handle,
     84 	bus_size_t offset, uint16_t value)
     85 {
     86 	bus_space_write_2(space, handle, offset, htole16(value));
     87 }
     88 
     89 static void
     90 nbo_bus_space_write_4(bus_space_tag_t space, bus_space_handle_t handle,
     91 	bus_size_t offset, uint32_t value)
     92 {
     93 	bus_space_write_4(space, handle, offset, htole32(value));
     94 }
     95 
     96 /* some functions access registers at 4 byte offset for little/high halves */
     97 #if BYTE_ORDER == BIG_ENDIAN
     98 #define REG_HI_OFF	0
     99 #define	REG_LO_OFF	4
    100 #else
    101 #define REG_HI_OFF	4
    102 #define	REG_LO_OFF	0
    103 #endif
    104 
    105 void
    106 virtio_set_status(struct virtio_softc *sc, int status)
    107 {
    108 	int old = 0;
    109 
    110 	if (status != 0)
    111 		old = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
    112 				       VIRTIO_CONFIG_DEVICE_STATUS);
    113 	bus_space_write_1(sc->sc_iot, sc->sc_ioh, VIRTIO_CONFIG_DEVICE_STATUS,
    114 			  status|old);
    115 }
    116 
    117 #define VIRTIO_MSIX_CONFIG_VECTOR_INDEX	0
    118 #define VIRTIO_MSIX_QUEUE_VECTOR_INDEX	1
    119 
    120 static int
    121 virtio_setup_msix_vectors(struct virtio_softc *sc)
    122 {
    123 	int offset, vector, ret, qid;
    124 
    125 	offset = VIRTIO_CONFIG_MSI_CONFIG_VECTOR;
    126 	vector = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    127 
    128 	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh, offset, vector);
    129 	ret = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh, offset);
    130 	aprint_debug_dev(sc->sc_dev, "expected=%d, actual=%d\n",
    131 	    vector, ret);
    132 	if (ret != vector)
    133 		return -1;
    134 
    135 	for (qid = 0; qid < sc->sc_nvqs; qid++) {
    136 		offset = VIRTIO_CONFIG_QUEUE_SELECT;
    137 		nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh, offset, qid);
    138 
    139 		offset = VIRTIO_CONFIG_MSI_QUEUE_VECTOR;
    140 		vector = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    141 
    142 		nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh, offset, vector);
    143 		ret = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh, offset);
    144 		aprint_debug_dev(sc->sc_dev, "expected=%d, actual=%d\n",
    145 		    vector, ret);
    146 		if (ret != vector)
    147 			return -1;
    148 	}
    149 
    150 	return 0;
    151 }
    152 
    153 static int
    154 virtio_setup_msix_interrupts(struct virtio_softc *sc,
    155     struct pci_attach_args *pa)
    156 {
    157 	device_t self = sc->sc_dev;
    158 	pci_chipset_tag_t pc = pa->pa_pc;
    159 	char intrbuf[PCI_INTRSTR_LEN];
    160 	char const *intrstr;
    161 	int idx;
    162 
    163 	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    164 	if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
    165 		pci_intr_setattr(pc, &sc->sc_ihp[idx], PCI_INTR_MPSAFE, true);
    166 
    167 	sc->sc_ihs[idx] = pci_intr_establish_xname(pc, sc->sc_ihp[idx],
    168 	    sc->sc_ipl, virtio_msix_config_intr, sc, device_xname(sc->sc_dev));
    169 	if (sc->sc_ihs[idx] == NULL) {
    170 		aprint_error_dev(self, "couldn't establish MSI-X for config\n");
    171 		goto error;
    172 	}
    173 
    174 	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    175 	if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
    176 		pci_intr_setattr(pc, &sc->sc_ihp[idx], PCI_INTR_MPSAFE, true);
    177 
    178 	sc->sc_ihs[idx] = pci_intr_establish_xname(pc, sc->sc_ihp[idx],
    179 	    sc->sc_ipl, virtio_msix_queue_intr, sc, device_xname(sc->sc_dev));
    180 	if (sc->sc_ihs[idx] == NULL) {
    181 		aprint_error_dev(self, "couldn't establish MSI-X for queues\n");
    182 		goto error;
    183 	}
    184 
    185 	if (virtio_setup_msix_vectors(sc) != 0) {
    186 		aprint_error_dev(self, "couldn't setup MSI-X vectors\n");
    187 		goto error;
    188 	}
    189 
    190 	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    191 	intrstr = pci_intr_string(pc, sc->sc_ihp[idx], intrbuf, sizeof(intrbuf));
    192 	aprint_normal_dev(self, "config interrupting at %s\n", intrstr);
    193 	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    194 	intrstr = pci_intr_string(pc, sc->sc_ihp[idx], intrbuf, sizeof(intrbuf));
    195 	aprint_normal_dev(self, "queues interrupting at %s\n", intrstr);
    196 
    197 	return 0;
    198 
    199 error:
    200 	idx = VIRTIO_MSIX_CONFIG_VECTOR_INDEX;
    201 	if (sc->sc_ihs[idx] != NULL)
    202 		pci_intr_disestablish(sc->sc_pc, sc->sc_ihs[idx]);
    203 	idx = VIRTIO_MSIX_QUEUE_VECTOR_INDEX;
    204 	if (sc->sc_ihs[idx] != NULL)
    205 		pci_intr_disestablish(sc->sc_pc, sc->sc_ihs[idx]);
    206 
    207 	return -1;
    208 }
    209 
    210 static int
    211 virtio_setup_intx_interrupt(struct virtio_softc *sc,
    212     struct pci_attach_args *pa)
    213 {
    214 	device_t self = sc->sc_dev;
    215 	pci_chipset_tag_t pc = pa->pa_pc;
    216 	char intrbuf[PCI_INTRSTR_LEN];
    217 	char const *intrstr;
    218 
    219 	if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
    220 		pci_intr_setattr(pc, &sc->sc_ihp[0], PCI_INTR_MPSAFE, true);
    221 
    222 	sc->sc_ihs[0] = pci_intr_establish_xname(pc, sc->sc_ihp[0],
    223 	    sc->sc_ipl, virtio_intr, sc, device_xname(sc->sc_dev));
    224 	if (sc->sc_ihs[0] == NULL) {
    225 		aprint_error_dev(self, "couldn't establish INTx\n");
    226 		return -1;
    227 	}
    228 
    229 	intrstr = pci_intr_string(pc, sc->sc_ihp[0], intrbuf, sizeof(intrbuf));
    230 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
    231 
    232 	return 0;
    233 }
    234 
    235 static int
    236 virtio_setup_interrupts(struct virtio_softc *sc)
    237 {
    238 	device_t self = sc->sc_dev;
    239 	pci_chipset_tag_t pc = sc->sc_pa.pa_pc;
    240 	int error;
    241 	int nmsix;
    242 	int counts[PCI_INTR_TYPE_SIZE];
    243 	pci_intr_type_t max_type;
    244 
    245 	nmsix = pci_msix_count(sc->sc_pa.pa_pc, sc->sc_pa.pa_tag);
    246 	aprint_debug_dev(self, "pci_msix_count=%d\n", nmsix);
    247 
    248 	/* We need at least two: one for config and the other for queues */
    249 	if ((sc->sc_flags & VIRTIO_F_PCI_INTR_MSIX) == 0 || nmsix < 2) {
    250 		/* Try INTx only */
    251 		max_type = PCI_INTR_TYPE_INTX;
    252 		counts[PCI_INTR_TYPE_INTX] = 1;
    253 	} else {
    254 		/* Try MSI-X first and INTx second */
    255 		max_type = PCI_INTR_TYPE_MSIX;
    256 		counts[PCI_INTR_TYPE_MSIX] = 2;
    257 		counts[PCI_INTR_TYPE_MSI] = 0;
    258 		counts[PCI_INTR_TYPE_INTX] = 1;
    259 	}
    260 
    261  retry:
    262 	error = pci_intr_alloc(&sc->sc_pa, &sc->sc_ihp, counts, max_type);
    263 	if (error != 0) {
    264 		aprint_error_dev(self, "couldn't map interrupt\n");
    265 		return -1;
    266 	}
    267 
    268 	if (pci_intr_type(pc, sc->sc_ihp[0]) == PCI_INTR_TYPE_MSIX) {
    269 		sc->sc_ihs = kmem_alloc(sizeof(*sc->sc_ihs) * 2,
    270 		    KM_SLEEP);
    271 
    272 		error = virtio_setup_msix_interrupts(sc, &sc->sc_pa);
    273 		if (error != 0) {
    274 			kmem_free(sc->sc_ihs, sizeof(*sc->sc_ihs) * 2);
    275 			pci_intr_release(pc, sc->sc_ihp, 2);
    276 
    277 			/* Retry INTx */
    278 			max_type = PCI_INTR_TYPE_INTX;
    279 			counts[PCI_INTR_TYPE_INTX] = 1;
    280 			goto retry;
    281 		}
    282 
    283 		sc->sc_ihs_num = 2;
    284 		sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_MSI;
    285 	} else if (pci_intr_type(pc, sc->sc_ihp[0]) == PCI_INTR_TYPE_INTX) {
    286 		sc->sc_ihs = kmem_alloc(sizeof(*sc->sc_ihs) * 1,
    287 		    KM_SLEEP);
    288 
    289 		error = virtio_setup_intx_interrupt(sc, &sc->sc_pa);
    290 		if (error != 0) {
    291 			kmem_free(sc->sc_ihs, sizeof(*sc->sc_ihs) * 1);
    292 			pci_intr_release(pc, sc->sc_ihp, 1);
    293 			return -1;
    294 		}
    295 
    296 		sc->sc_ihs_num = 1;
    297 		sc->sc_config_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
    298 	}
    299 
    300 	KASSERT(sc->sc_soft_ih == NULL);
    301 	if (sc->sc_flags & VIRTIO_F_PCI_INTR_SOFTINT) {
    302 		u_int flags = SOFTINT_NET;
    303 		if (sc->sc_flags & VIRTIO_F_PCI_INTR_MPSAFE)
    304 			flags |= SOFTINT_MPSAFE;
    305 
    306 		sc->sc_soft_ih = softint_establish(flags, virtio_soft_intr, sc);
    307 		if (sc->sc_soft_ih == NULL) {
    308 			virtio_free_interrupts(sc);
    309 			aprint_error_dev(sc->sc_dev,
    310 			    "failed to establish soft interrupt\n");
    311 			return -1;
    312 		}
    313 	}
    314 
    315 	return 0;
    316 }
    317 
    318 static void
    319 virtio_free_interrupts(struct virtio_softc *sc)
    320 {
    321 	for (int i = 0; i < sc->sc_ihs_num; i++) {
    322 		if (sc->sc_ihs[i] == NULL)
    323 			continue;
    324 		pci_intr_disestablish(sc->sc_pc, sc->sc_ihs[i]);
    325 		sc->sc_ihs[i] = NULL;
    326 	}
    327 
    328 	if (sc->sc_ihs_num > 0)
    329 		pci_intr_release(sc->sc_pc, sc->sc_ihp, sc->sc_ihs_num);
    330 
    331 	if (sc->sc_soft_ih) {
    332 		softint_disestablish(sc->sc_soft_ih);
    333 		sc->sc_soft_ih = NULL;
    334 	}
    335 
    336 	if (sc->sc_ihs != NULL) {
    337 		kmem_free(sc->sc_ihs, sizeof(*sc->sc_ihs) * sc->sc_ihs_num);
    338 		sc->sc_ihs = NULL;
    339 	}
    340 	sc->sc_ihs_num = 0;
    341 }
    342 
    343 
    344 
    345 /*
    346  * Reset the device.
    347  */
    348 /*
    349  * To reset the device to a known state, do following:
    350  *	virtio_reset(sc);	     // this will stop the device activity
    351  *	<dequeue finished requests>; // virtio_dequeue() still can be called
    352  *	<revoke pending requests in the vqs if any>;
    353  *	virtio_reinit_begin(sc);     // dequeue prohibitted
    354  *	newfeatures = virtio_negotiate_features(sc, requestedfeatures);
    355  *	<some other initialization>;
    356  *	virtio_reinit_end(sc);	     // device activated; enqueue allowed
    357  * Once attached, feature negotiation can only be allowed after virtio_reset.
    358  */
    359 void
    360 virtio_reset(struct virtio_softc *sc)
    361 {
    362 	virtio_device_reset(sc);
    363 }
    364 
    365 void
    366 virtio_reinit_start(struct virtio_softc *sc)
    367 {
    368 	int i;
    369 
    370 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
    371 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER);
    372 	for (i = 0; i < sc->sc_nvqs; i++) {
    373 		int n;
    374 		struct virtqueue *vq = &sc->sc_vqs[i];
    375 		nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
    376 				  VIRTIO_CONFIG_QUEUE_SELECT,
    377 				  vq->vq_index);
    378 		n = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh,
    379 				     VIRTIO_CONFIG_QUEUE_SIZE);
    380 		if (n == 0)	/* vq disappeared */
    381 			continue;
    382 		if (n != vq->vq_num) {
    383 			panic("%s: virtqueue size changed, vq index %d\n",
    384 			      device_xname(sc->sc_dev),
    385 			      vq->vq_index);
    386 		}
    387 		virtio_init_vq(sc, vq, true);
    388 		nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    389 				  VIRTIO_CONFIG_QUEUE_ADDRESS,
    390 				  (vq->vq_dmamap->dm_segs[0].ds_addr
    391 				   / VIRTIO_PAGE_SIZE));
    392 	}
    393 
    394 	/* MSI-X should have more than one handles where INTx has just one */
    395 	if (sc->sc_ihs_num > 1) {
    396 		if (virtio_setup_msix_vectors(sc) != 0) {
    397 			aprint_error_dev(sc->sc_dev,
    398 			    "couldn't setup MSI-X vectors\n");
    399 			return;
    400 		}
    401 	}
    402 }
    403 
    404 void
    405 virtio_reinit_end(struct virtio_softc *sc)
    406 {
    407 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
    408 }
    409 
    410 /*
    411  * Feature negotiation.
    412  */
    413 uint32_t
    414 virtio_negotiate_features(struct virtio_softc *sc, uint32_t guest_features)
    415 {
    416 	uint32_t r;
    417 
    418 	if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) &&
    419 	    !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */
    420 		guest_features |= VIRTIO_F_RING_INDIRECT_DESC;
    421 	r = nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
    422 			     VIRTIO_CONFIG_DEVICE_FEATURES);
    423 	r &= guest_features;
    424 	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    425 			  VIRTIO_CONFIG_GUEST_FEATURES, r);
    426 	sc->sc_features = r;
    427 	if (r & VIRTIO_F_RING_INDIRECT_DESC)
    428 		sc->sc_indirect = true;
    429 	else
    430 		sc->sc_indirect = false;
    431 
    432 	return r;
    433 }
    434 
    435 /*
    436  * Device configuration registers.
    437  */
    438 uint8_t
    439 virtio_read_device_config_1(struct virtio_softc *sc, int index)
    440 {
    441 	return bus_space_read_1(sc->sc_iot, sc->sc_ioh,
    442 				sc->sc_config_offset + index);
    443 }
    444 
    445 uint16_t
    446 virtio_read_device_config_2(struct virtio_softc *sc, int index)
    447 {
    448 	return nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh,
    449 				sc->sc_config_offset + index);
    450 }
    451 
    452 uint32_t
    453 virtio_read_device_config_4(struct virtio_softc *sc, int index)
    454 {
    455 	return nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
    456 				sc->sc_config_offset + index);
    457 }
    458 
    459 uint64_t
    460 virtio_read_device_config_8(struct virtio_softc *sc, int index)
    461 {
    462 	uint64_t r;
    463 
    464 	r = nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
    465 			     sc->sc_config_offset + index + REG_HI_OFF);
    466 	r <<= 32;
    467 	r |= nbo_bus_space_read_4(sc->sc_iot, sc->sc_ioh,
    468 			      sc->sc_config_offset + index + REG_LO_OFF);
    469 
    470 	return r;
    471 }
    472 
    473 void
    474 virtio_write_device_config_1(struct virtio_softc *sc,
    475 			     int index, uint8_t value)
    476 {
    477 	bus_space_write_1(sc->sc_iot, sc->sc_ioh,
    478 			  sc->sc_config_offset + index, value);
    479 }
    480 
    481 void
    482 virtio_write_device_config_2(struct virtio_softc *sc,
    483 			     int index, uint16_t value)
    484 {
    485 	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
    486 			  sc->sc_config_offset + index, value);
    487 }
    488 
    489 void
    490 virtio_write_device_config_4(struct virtio_softc *sc,
    491 			     int index, uint32_t value)
    492 {
    493 	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    494 			  sc->sc_config_offset + index, value);
    495 }
    496 
    497 void
    498 virtio_write_device_config_8(struct virtio_softc *sc,
    499 			     int index, uint64_t value)
    500 {
    501 	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    502 			  sc->sc_config_offset + index + REG_LO_OFF,
    503 			  value & 0xffffffff);
    504 	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    505 			  sc->sc_config_offset + index + REG_HI_OFF,
    506 			  value >> 32);
    507 }
    508 
    509 /*
    510  * Interrupt handler.
    511  */
    512 static int
    513 virtio_intr(void *arg)
    514 {
    515 	struct virtio_softc *sc = arg;
    516 	int isr, r = 0;
    517 
    518 	/* check and ack the interrupt */
    519 	isr = bus_space_read_1(sc->sc_iot, sc->sc_ioh,
    520 			       VIRTIO_CONFIG_ISR_STATUS);
    521 	if (isr == 0)
    522 		return 0;
    523 	if ((isr & VIRTIO_CONFIG_ISR_CONFIG_CHANGE) &&
    524 	    (sc->sc_config_change != NULL))
    525 		r = (sc->sc_config_change)(sc);
    526 	if (sc->sc_intrhand != NULL) {
    527 		if (sc->sc_soft_ih != NULL)
    528 			softint_schedule(sc->sc_soft_ih);
    529 		else
    530 			r |= (sc->sc_intrhand)(sc);
    531 	}
    532 
    533 	return r;
    534 }
    535 
    536 static int
    537 virtio_msix_queue_intr(void *arg)
    538 {
    539 	struct virtio_softc *sc = arg;
    540 	int r = 0;
    541 
    542 	if (sc->sc_intrhand != NULL) {
    543 		if (sc->sc_soft_ih != NULL)
    544 			softint_schedule(sc->sc_soft_ih);
    545 		else
    546 			r |= (sc->sc_intrhand)(sc);
    547 	}
    548 
    549 	return r;
    550 }
    551 
    552 static int
    553 virtio_msix_config_intr(void *arg)
    554 {
    555 	struct virtio_softc *sc = arg;
    556 	int r = 0;
    557 
    558 	if (sc->sc_config_change != NULL)
    559 		r = (sc->sc_config_change)(sc);
    560 	return r;
    561 }
    562 
    563 static void
    564 virtio_soft_intr(void *arg)
    565 {
    566 	struct virtio_softc *sc = arg;
    567 
    568 	KASSERT(sc->sc_intrhand != NULL);
    569 
    570 	(sc->sc_intrhand)(sc);
    571 }
    572 
    573 /*
    574  * dmamap sync operations for a virtqueue.
    575  */
    576 static inline void
    577 vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops)
    578 {
    579 	/* availoffset == sizeof(vring_desc)*vq_num */
    580 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset,
    581 			ops);
    582 }
    583 
    584 static inline void
    585 vq_sync_aring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
    586 {
    587 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
    588 			vq->vq_availoffset,
    589 			offsetof(struct vring_avail, ring)
    590 			 + vq->vq_num * sizeof(uint16_t),
    591 			ops);
    592 }
    593 
    594 static inline void
    595 vq_sync_uring(struct virtio_softc *sc, struct virtqueue *vq, int ops)
    596 {
    597 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
    598 			vq->vq_usedoffset,
    599 			offsetof(struct vring_used, ring)
    600 			 + vq->vq_num * sizeof(struct vring_used_elem),
    601 			ops);
    602 }
    603 
    604 static inline void
    605 vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot,
    606 		     int ops)
    607 {
    608 	int offset = vq->vq_indirectoffset
    609 		      + sizeof(struct vring_desc) * vq->vq_maxnsegs * slot;
    610 
    611 	bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap,
    612 			offset, sizeof(struct vring_desc) * vq->vq_maxnsegs,
    613 			ops);
    614 }
    615 
    616 /*
    617  * Can be used as sc_intrhand.
    618  */
    619 /*
    620  * Scan vq, bus_dmamap_sync for the vqs (not for the payload),
    621  * and calls (*vq_done)() if some entries are consumed.
    622  */
    623 int
    624 virtio_vq_intr(struct virtio_softc *sc)
    625 {
    626 	struct virtqueue *vq;
    627 	int i, r = 0;
    628 
    629 	for (i = 0; i < sc->sc_nvqs; i++) {
    630 		vq = &sc->sc_vqs[i];
    631 		if (vq->vq_queued) {
    632 			vq->vq_queued = 0;
    633 			vq_sync_aring(sc, vq, BUS_DMASYNC_POSTWRITE);
    634 		}
    635 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
    636 		membar_consumer();
    637 		if (vq->vq_used_idx != vq->vq_used->idx) {
    638 			if (vq->vq_done)
    639 				r |= (vq->vq_done)(vq);
    640 		}
    641 	}
    642 
    643 	return r;
    644 }
    645 
    646 /*
    647  * Start/stop vq interrupt.  No guarantee.
    648  */
    649 void
    650 virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
    651 {
    652 	vq->vq_avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
    653 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
    654 	vq->vq_queued++;
    655 }
    656 
    657 void
    658 virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq)
    659 {
    660 	vq->vq_avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
    661 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
    662 	vq->vq_queued++;
    663 }
    664 
    665 /*
    666  * Initialize vq structure.
    667  */
    668 static void
    669 virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq,
    670     const bool reinit)
    671 {
    672 	int i, j;
    673 	int vq_size = vq->vq_num;
    674 
    675 	memset(vq->vq_vaddr, 0, vq->vq_bytesize);
    676 
    677 	/* build the indirect descriptor chain */
    678 	if (vq->vq_indirect != NULL) {
    679 		struct vring_desc *vd;
    680 
    681 		for (i = 0; i < vq_size; i++) {
    682 			vd = vq->vq_indirect;
    683 			vd += vq->vq_maxnsegs * i;
    684 			for (j = 0; j < vq->vq_maxnsegs-1; j++) {
    685 				vd[j].next = j + 1;
    686 			}
    687 		}
    688 	}
    689 
    690 	/* free slot management */
    691 	SIMPLEQ_INIT(&vq->vq_freelist);
    692 	for (i = 0; i < vq_size; i++) {
    693 		SIMPLEQ_INSERT_TAIL(&vq->vq_freelist,
    694 				    &vq->vq_entries[i], qe_list);
    695 		vq->vq_entries[i].qe_index = i;
    696 	}
    697 	if (!reinit)
    698 		mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl);
    699 
    700 	/* enqueue/dequeue status */
    701 	vq->vq_avail_idx = 0;
    702 	vq->vq_used_idx = 0;
    703 	vq->vq_queued = 0;
    704 	if (!reinit) {
    705 		mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl);
    706 		mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl);
    707 	}
    708 	vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
    709 	vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
    710 	vq->vq_queued++;
    711 }
    712 
    713 /*
    714  * Allocate/free a vq.
    715  */
    716 int
    717 virtio_alloc_vq(struct virtio_softc *sc, struct virtqueue *vq, int index,
    718     int maxsegsize, int maxnsegs, const char *name)
    719 {
    720 	int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0;
    721 	int rsegs, r;
    722 #define VIRTQUEUE_ALIGN(n)	(((n)+(VIRTIO_PAGE_SIZE-1))&	\
    723 				 ~(VIRTIO_PAGE_SIZE-1))
    724 
    725 	/* Make sure callers allocate vqs in order */
    726 	KASSERT(sc->sc_nvqs == index);
    727 
    728 	memset(vq, 0, sizeof(*vq));
    729 
    730 	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
    731 			  VIRTIO_CONFIG_QUEUE_SELECT, index);
    732 	vq_size = nbo_bus_space_read_2(sc->sc_iot, sc->sc_ioh,
    733 				   VIRTIO_CONFIG_QUEUE_SIZE);
    734 	if (vq_size == 0) {
    735 		aprint_error_dev(sc->sc_dev,
    736 				 "virtqueue not exist, index %d for %s\n",
    737 				 index, name);
    738 		goto err;
    739 	}
    740 	/* allocsize1: descriptor table + avail ring + pad */
    741 	allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc)*vq_size
    742 				     + sizeof(uint16_t)*(2+vq_size));
    743 	/* allocsize2: used ring + pad */
    744 	allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t)*2
    745 				     + sizeof(struct vring_used_elem)*vq_size);
    746 	/* allocsize3: indirect table */
    747 	if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT)
    748 		allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size;
    749 	else
    750 		allocsize3 = 0;
    751 	allocsize = allocsize1 + allocsize2 + allocsize3;
    752 
    753 	/* alloc and map the memory */
    754 	r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0,
    755 			     &vq->vq_segs[0], 1, &rsegs, BUS_DMA_NOWAIT);
    756 	if (r != 0) {
    757 		aprint_error_dev(sc->sc_dev,
    758 				 "virtqueue %d for %s allocation failed, "
    759 				 "error code %d\n", index, name, r);
    760 		goto err;
    761 	}
    762 	r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], 1, allocsize,
    763 			   &vq->vq_vaddr, BUS_DMA_NOWAIT);
    764 	if (r != 0) {
    765 		aprint_error_dev(sc->sc_dev,
    766 				 "virtqueue %d for %s map failed, "
    767 				 "error code %d\n", index, name, r);
    768 		goto err;
    769 	}
    770 	r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0,
    771 			      BUS_DMA_NOWAIT, &vq->vq_dmamap);
    772 	if (r != 0) {
    773 		aprint_error_dev(sc->sc_dev,
    774 				 "virtqueue %d for %s dmamap creation failed, "
    775 				 "error code %d\n", index, name, r);
    776 		goto err;
    777 	}
    778 	r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap,
    779 			    vq->vq_vaddr, allocsize, NULL, BUS_DMA_NOWAIT);
    780 	if (r != 0) {
    781 		aprint_error_dev(sc->sc_dev,
    782 				 "virtqueue %d for %s dmamap load failed, "
    783 				 "error code %d\n", index, name, r);
    784 		goto err;
    785 	}
    786 
    787 	/* set the vq address */
    788 	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    789 			  VIRTIO_CONFIG_QUEUE_ADDRESS,
    790 			  (vq->vq_dmamap->dm_segs[0].ds_addr
    791 			   / VIRTIO_PAGE_SIZE));
    792 
    793 	/* remember addresses and offsets for later use */
    794 	vq->vq_owner = sc;
    795 	vq->vq_num = vq_size;
    796 	vq->vq_index = index;
    797 	vq->vq_desc = vq->vq_vaddr;
    798 	vq->vq_availoffset = sizeof(struct vring_desc)*vq_size;
    799 	vq->vq_avail = (void*)(((char*)vq->vq_desc) + vq->vq_availoffset);
    800 	vq->vq_usedoffset = allocsize1;
    801 	vq->vq_used = (void*)(((char*)vq->vq_desc) + vq->vq_usedoffset);
    802 	if (allocsize3 > 0) {
    803 		vq->vq_indirectoffset = allocsize1 + allocsize2;
    804 		vq->vq_indirect = (void*)(((char*)vq->vq_desc)
    805 					  + vq->vq_indirectoffset);
    806 	}
    807 	vq->vq_bytesize = allocsize;
    808 	vq->vq_maxsegsize = maxsegsize;
    809 	vq->vq_maxnsegs = maxnsegs;
    810 
    811 	/* free slot management */
    812 	vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry)*vq_size,
    813 				     KM_NOSLEEP);
    814 	if (vq->vq_entries == NULL) {
    815 		r = ENOMEM;
    816 		goto err;
    817 	}
    818 
    819 	virtio_init_vq(sc, vq, false);
    820 
    821 	aprint_verbose_dev(sc->sc_dev,
    822 			   "allocated %u byte for virtqueue %d for %s, "
    823 			   "size %d\n", allocsize, index, name, vq_size);
    824 	if (allocsize3 > 0)
    825 		aprint_verbose_dev(sc->sc_dev,
    826 				   "using %d byte (%d entries) "
    827 				   "indirect descriptors\n",
    828 				   allocsize3, maxnsegs * vq_size);
    829 
    830 	sc->sc_nvqs++;
    831 
    832 	return 0;
    833 
    834 err:
    835 	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    836 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
    837 	if (vq->vq_dmamap)
    838 		bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
    839 	if (vq->vq_vaddr)
    840 		bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize);
    841 	if (vq->vq_segs[0].ds_addr)
    842 		bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
    843 	memset(vq, 0, sizeof(*vq));
    844 
    845 	return -1;
    846 }
    847 
    848 int
    849 virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq)
    850 {
    851 	struct vq_entry *qe;
    852 	int i = 0;
    853 
    854 	/* device must be already deactivated */
    855 	/* confirm the vq is empty */
    856 	SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) {
    857 		i++;
    858 	}
    859 	if (i != vq->vq_num) {
    860 		printf("%s: freeing non-empty vq, index %d\n",
    861 		       device_xname(sc->sc_dev), vq->vq_index);
    862 		return EBUSY;
    863 	}
    864 
    865 	/* tell device that there's no virtqueue any longer */
    866 	nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
    867 			  VIRTIO_CONFIG_QUEUE_SELECT, vq->vq_index);
    868 	nbo_bus_space_write_4(sc->sc_iot, sc->sc_ioh,
    869 			  VIRTIO_CONFIG_QUEUE_ADDRESS, 0);
    870 
    871 	kmem_free(vq->vq_entries, sizeof(*vq->vq_entries) * vq->vq_num);
    872 	bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap);
    873 	bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap);
    874 	bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize);
    875 	bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1);
    876 	mutex_destroy(&vq->vq_freelist_lock);
    877 	mutex_destroy(&vq->vq_uring_lock);
    878 	mutex_destroy(&vq->vq_aring_lock);
    879 	memset(vq, 0, sizeof(*vq));
    880 
    881 	sc->sc_nvqs--;
    882 
    883 	return 0;
    884 }
    885 
    886 /*
    887  * Free descriptor management.
    888  */
    889 static struct vq_entry *
    890 vq_alloc_entry(struct virtqueue *vq)
    891 {
    892 	struct vq_entry *qe;
    893 
    894 	mutex_enter(&vq->vq_freelist_lock);
    895 	if (SIMPLEQ_EMPTY(&vq->vq_freelist)) {
    896 		mutex_exit(&vq->vq_freelist_lock);
    897 		return NULL;
    898 	}
    899 	qe = SIMPLEQ_FIRST(&vq->vq_freelist);
    900 	SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list);
    901 	mutex_exit(&vq->vq_freelist_lock);
    902 
    903 	return qe;
    904 }
    905 
    906 static void
    907 vq_free_entry(struct virtqueue *vq, struct vq_entry *qe)
    908 {
    909 	mutex_enter(&vq->vq_freelist_lock);
    910 	SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list);
    911 	mutex_exit(&vq->vq_freelist_lock);
    912 
    913 	return;
    914 }
    915 
    916 /*
    917  * Enqueue several dmamaps as a single request.
    918  */
    919 /*
    920  * Typical usage:
    921  *  <queue size> number of followings are stored in arrays
    922  *  - command blocks (in dmamem) should be pre-allocated and mapped
    923  *  - dmamaps for command blocks should be pre-allocated and loaded
    924  *  - dmamaps for payload should be pre-allocated
    925  *      r = virtio_enqueue_prep(sc, vq, &slot);		// allocate a slot
    926  *	if (r)		// currently 0 or EAGAIN
    927  *	  return r;
    928  *	r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..);
    929  *	if (r) {
    930  *	  virtio_enqueue_abort(sc, vq, slot);
    931  *	  return r;
    932  *	}
    933  *	r = virtio_enqueue_reserve(sc, vq, slot,
    934  *				   dmamap_payload[slot]->dm_nsegs+1);
    935  *							// ^ +1 for command
    936  *	if (r) {	// currently 0 or EAGAIN
    937  *	  bus_dmamap_unload(dmat, dmamap_payload[slot]);
    938  *	  return r;					// do not call abort()
    939  *	}
    940  *	<setup and prepare commands>
    941  *	bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE);
    942  *	bus_dmamap_sync(dmat, dmamap_payload[slot],...);
    943  *	virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false);
    944  *	virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite);
    945  *	virtio_enqueue_commit(sc, vq, slot, true);
    946  */
    947 
    948 /*
    949  * enqueue_prep: allocate a slot number
    950  */
    951 int
    952 virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp)
    953 {
    954 	struct vq_entry *qe1;
    955 
    956 	KASSERT(slotp != NULL);
    957 
    958 	qe1 = vq_alloc_entry(vq);
    959 	if (qe1 == NULL)
    960 		return EAGAIN;
    961 	/* next slot is not allocated yet */
    962 	qe1->qe_next = -1;
    963 	*slotp = qe1->qe_index;
    964 
    965 	return 0;
    966 }
    967 
    968 /*
    969  * enqueue_reserve: allocate remaining slots and build the descriptor chain.
    970  */
    971 int
    972 virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq,
    973 		       int slot, int nsegs)
    974 {
    975 	int indirect;
    976 	struct vq_entry *qe1 = &vq->vq_entries[slot];
    977 
    978 	KASSERT(qe1->qe_next == -1);
    979 	KASSERT(1 <= nsegs && nsegs <= vq->vq_num);
    980 
    981 	if ((vq->vq_indirect != NULL) &&
    982 	    (nsegs >= MINSEG_INDIRECT) &&
    983 	    (nsegs <= vq->vq_maxnsegs))
    984 		indirect = 1;
    985 	else
    986 		indirect = 0;
    987 	qe1->qe_indirect = indirect;
    988 
    989 	if (indirect) {
    990 		struct vring_desc *vd;
    991 		int i;
    992 
    993 		vd = &vq->vq_desc[qe1->qe_index];
    994 		vd->addr = vq->vq_dmamap->dm_segs[0].ds_addr
    995 			+ vq->vq_indirectoffset;
    996 		vd->addr += sizeof(struct vring_desc)
    997 			* vq->vq_maxnsegs * qe1->qe_index;
    998 		vd->len = sizeof(struct vring_desc) * nsegs;
    999 		vd->flags = VRING_DESC_F_INDIRECT;
   1000 
   1001 		vd = vq->vq_indirect;
   1002 		vd += vq->vq_maxnsegs * qe1->qe_index;
   1003 		qe1->qe_desc_base = vd;
   1004 
   1005 		for (i = 0; i < nsegs-1; i++) {
   1006 			vd[i].flags = VRING_DESC_F_NEXT;
   1007 		}
   1008 		vd[i].flags = 0;
   1009 		qe1->qe_next = 0;
   1010 
   1011 		return 0;
   1012 	} else {
   1013 		struct vring_desc *vd;
   1014 		struct vq_entry *qe;
   1015 		int i, s;
   1016 
   1017 		vd = &vq->vq_desc[0];
   1018 		qe1->qe_desc_base = vd;
   1019 		qe1->qe_next = qe1->qe_index;
   1020 		s = slot;
   1021 		for (i = 0; i < nsegs - 1; i++) {
   1022 			qe = vq_alloc_entry(vq);
   1023 			if (qe == NULL) {
   1024 				vd[s].flags = 0;
   1025 				virtio_enqueue_abort(sc, vq, slot);
   1026 				return EAGAIN;
   1027 			}
   1028 			vd[s].flags = VRING_DESC_F_NEXT;
   1029 			vd[s].next = qe->qe_index;
   1030 			s = qe->qe_index;
   1031 		}
   1032 		vd[s].flags = 0;
   1033 
   1034 		return 0;
   1035 	}
   1036 }
   1037 
   1038 /*
   1039  * enqueue: enqueue a single dmamap.
   1040  */
   1041 int
   1042 virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot,
   1043 	       bus_dmamap_t dmamap, bool write)
   1044 {
   1045 	struct vq_entry *qe1 = &vq->vq_entries[slot];
   1046 	struct vring_desc *vd = qe1->qe_desc_base;
   1047 	int i;
   1048 	int s = qe1->qe_next;
   1049 
   1050 	KASSERT(s >= 0);
   1051 	KASSERT(dmamap->dm_nsegs > 0);
   1052 
   1053 	for (i = 0; i < dmamap->dm_nsegs; i++) {
   1054 		vd[s].addr = dmamap->dm_segs[i].ds_addr;
   1055 		vd[s].len = dmamap->dm_segs[i].ds_len;
   1056 		if (!write)
   1057 			vd[s].flags |= VRING_DESC_F_WRITE;
   1058 		s = vd[s].next;
   1059 	}
   1060 	qe1->qe_next = s;
   1061 
   1062 	return 0;
   1063 }
   1064 
   1065 int
   1066 virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot,
   1067 		 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len,
   1068 		 bool write)
   1069 {
   1070 	struct vq_entry *qe1 = &vq->vq_entries[slot];
   1071 	struct vring_desc *vd = qe1->qe_desc_base;
   1072 	int s = qe1->qe_next;
   1073 
   1074 	KASSERT(s >= 0);
   1075 	KASSERT(dmamap->dm_nsegs == 1); /* XXX */
   1076 	KASSERT((dmamap->dm_segs[0].ds_len > start) &&
   1077 		(dmamap->dm_segs[0].ds_len >= start + len));
   1078 
   1079 	vd[s].addr = dmamap->dm_segs[0].ds_addr + start;
   1080 	vd[s].len = len;
   1081 	if (!write)
   1082 		vd[s].flags |= VRING_DESC_F_WRITE;
   1083 	qe1->qe_next = vd[s].next;
   1084 
   1085 	return 0;
   1086 }
   1087 
   1088 /*
   1089  * enqueue_commit: add it to the aring.
   1090  */
   1091 int
   1092 virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot,
   1093 		      bool notifynow)
   1094 {
   1095 	struct vq_entry *qe1;
   1096 
   1097 	if (slot < 0) {
   1098 		mutex_enter(&vq->vq_aring_lock);
   1099 		goto notify;
   1100 	}
   1101 	vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE);
   1102 	qe1 = &vq->vq_entries[slot];
   1103 	if (qe1->qe_indirect)
   1104 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE);
   1105 	mutex_enter(&vq->vq_aring_lock);
   1106 	vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = slot;
   1107 
   1108 notify:
   1109 	if (notifynow) {
   1110 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
   1111 		vq_sync_uring(sc, vq, BUS_DMASYNC_PREREAD);
   1112 		membar_producer();
   1113 		vq->vq_avail->idx = vq->vq_avail_idx;
   1114 		vq_sync_aring(sc, vq, BUS_DMASYNC_PREWRITE);
   1115 		membar_producer();
   1116 		vq->vq_queued++;
   1117 		vq_sync_uring(sc, vq, BUS_DMASYNC_POSTREAD);
   1118 		membar_consumer();
   1119 		if (!(vq->vq_used->flags & VRING_USED_F_NO_NOTIFY))
   1120 			nbo_bus_space_write_2(sc->sc_iot, sc->sc_ioh,
   1121 					  VIRTIO_CONFIG_QUEUE_NOTIFY,
   1122 					  vq->vq_index);
   1123 	}
   1124 	mutex_exit(&vq->vq_aring_lock);
   1125 
   1126 	return 0;
   1127 }
   1128 
   1129 /*
   1130  * enqueue_abort: rollback.
   1131  */
   1132 int
   1133 virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot)
   1134 {
   1135 	struct vq_entry *qe = &vq->vq_entries[slot];
   1136 	struct vring_desc *vd;
   1137 	int s;
   1138 
   1139 	if (qe->qe_next < 0) {
   1140 		vq_free_entry(vq, qe);
   1141 		return 0;
   1142 	}
   1143 
   1144 	s = slot;
   1145 	vd = &vq->vq_desc[0];
   1146 	while (vd[s].flags & VRING_DESC_F_NEXT) {
   1147 		s = vd[s].next;
   1148 		vq_free_entry(vq, qe);
   1149 		qe = &vq->vq_entries[s];
   1150 	}
   1151 	vq_free_entry(vq, qe);
   1152 	return 0;
   1153 }
   1154 
   1155 /*
   1156  * Dequeue a request.
   1157  */
   1158 /*
   1159  * dequeue: dequeue a request from uring; dmamap_sync for uring is
   1160  *	    already done in the interrupt handler.
   1161  */
   1162 int
   1163 virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq,
   1164 	       int *slotp, int *lenp)
   1165 {
   1166 	uint16_t slot, usedidx;
   1167 	struct vq_entry *qe;
   1168 
   1169 	if (vq->vq_used_idx == vq->vq_used->idx)
   1170 		return ENOENT;
   1171 	mutex_enter(&vq->vq_uring_lock);
   1172 	usedidx = vq->vq_used_idx++;
   1173 	mutex_exit(&vq->vq_uring_lock);
   1174 	usedidx %= vq->vq_num;
   1175 	slot = vq->vq_used->ring[usedidx].id;
   1176 	qe = &vq->vq_entries[slot];
   1177 
   1178 	if (qe->qe_indirect)
   1179 		vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE);
   1180 
   1181 	if (slotp)
   1182 		*slotp = slot;
   1183 	if (lenp)
   1184 		*lenp = vq->vq_used->ring[usedidx].len;
   1185 
   1186 	return 0;
   1187 }
   1188 
   1189 /*
   1190  * dequeue_commit: complete dequeue; the slot is recycled for future use.
   1191  *                 if you forget to call this the slot will be leaked.
   1192  */
   1193 int
   1194 virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot)
   1195 {
   1196 	struct vq_entry *qe = &vq->vq_entries[slot];
   1197 	struct vring_desc *vd = &vq->vq_desc[0];
   1198 	int s = slot;
   1199 
   1200 	while (vd[s].flags & VRING_DESC_F_NEXT) {
   1201 		s = vd[s].next;
   1202 		vq_free_entry(vq, qe);
   1203 		qe = &vq->vq_entries[s];
   1204 	}
   1205 	vq_free_entry(vq, qe);
   1206 
   1207 	return 0;
   1208 }
   1209 
   1210 /*
   1211  * Attach a child, fill all the members.
   1212  */
   1213 void
   1214 virtio_child_attach_start(struct virtio_softc *sc, device_t child, int ipl,
   1215 		    struct virtqueue *vqs,
   1216 		    virtio_callback config_change,
   1217 		    virtio_callback intr_hand,
   1218 		    int req_flags, int req_features, const char *feat_bits)
   1219 {
   1220 	char buf[256];
   1221 	int features;
   1222 
   1223 	sc->sc_child = child;
   1224 	sc->sc_ipl = ipl;
   1225 	sc->sc_vqs = vqs;
   1226 	sc->sc_config_change = config_change;
   1227 	sc->sc_intrhand = intr_hand;
   1228 	sc->sc_flags = req_flags;
   1229 
   1230 	features = virtio_negotiate_features(sc, req_features);
   1231 	snprintb(buf, sizeof(buf), feat_bits, features);
   1232 	aprint_normal(": Features: %s\n", buf);
   1233 	aprint_naive("\n");
   1234 }
   1235 
   1236 int
   1237 virtio_child_attach_finish(struct virtio_softc *sc)
   1238 {
   1239 	int r;
   1240 
   1241 	r = virtio_setup_interrupts(sc);
   1242 	if (r != 0) {
   1243 		aprint_error_dev(sc->sc_dev, "failed to setup interrupts\n");
   1244 		virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
   1245 		return 1;
   1246 	}
   1247 
   1248 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK);
   1249 
   1250 	return 0;
   1251 }
   1252 
   1253 void
   1254 virtio_child_detach(struct virtio_softc *sc)
   1255 {
   1256 	sc->sc_child = NULL;
   1257 	sc->sc_vqs = NULL;
   1258 
   1259 	virtio_device_reset(sc);
   1260 
   1261 	virtio_free_interrupts(sc);
   1262 }
   1263 
   1264 void
   1265 virtio_child_attach_failed(struct virtio_softc *sc)
   1266 {
   1267 	virtio_child_detach(sc);
   1268 
   1269 	virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED);
   1270 
   1271 	sc->sc_child = VIRTIO_CHILD_FAILED;
   1272 }
   1273 
   1274 bus_dma_tag_t
   1275 virtio_dmat(struct virtio_softc *sc)
   1276 {
   1277 	return sc->sc_dmat;
   1278 }
   1279 
   1280 device_t
   1281 virtio_child(struct virtio_softc *sc)
   1282 {
   1283 	return sc->sc_child;
   1284 }
   1285 
   1286 int
   1287 virtio_intrhand(struct virtio_softc *sc)
   1288 {
   1289 	return (sc->sc_intrhand)(sc);
   1290 }
   1291 
   1292 uint32_t
   1293 virtio_features(struct virtio_softc *sc)
   1294 {
   1295 	return sc->sc_features;
   1296 }
   1297 
   1298 MODULE(MODULE_CLASS_DRIVER, virtio, "pci");
   1299 
   1300 #ifdef _MODULE
   1301 #include "ioconf.c"
   1302 #endif
   1303 
   1304 static int
   1305 virtio_modcmd(modcmd_t cmd, void *opaque)
   1306 {
   1307 	int error = 0;
   1308 
   1309 #ifdef _MODULE
   1310 	switch (cmd) {
   1311 	case MODULE_CMD_INIT:
   1312 		error = config_init_component(cfdriver_ioconf_virtio,
   1313 		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
   1314 		break;
   1315 	case MODULE_CMD_FINI:
   1316 		error = config_fini_component(cfdriver_ioconf_virtio,
   1317 		    cfattach_ioconf_virtio, cfdata_ioconf_virtio);
   1318 		break;
   1319 	default:
   1320 		error = ENOTTY;
   1321 		break;
   1322 	}
   1323 #endif
   1324 
   1325 	return error;
   1326 }
   1327