Home | History | Annotate | Line # | Download | only in pci
ld_virtio.c revision 1.32
      1 /*	$NetBSD: ld_virtio.c,v 1.32 2023/03/23 03:55:11 yamaguchi 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: ld_virtio.c,v 1.32 2023/03/23 03:55:11 yamaguchi Exp $");
     30 
     31 #include <sys/param.h>
     32 #include <sys/systm.h>
     33 #include <sys/kernel.h>
     34 #include <sys/buf.h>
     35 #include <sys/bufq.h>
     36 #include <sys/bus.h>
     37 #include <sys/device.h>
     38 #include <sys/disk.h>
     39 #include <sys/mutex.h>
     40 #include <sys/module.h>
     41 
     42 #include <dev/ldvar.h>
     43 #include <dev/pci/virtioreg.h>
     44 #include <dev/pci/virtiovar.h>
     45 
     46 #include "ioconf.h"
     47 
     48 /*
     49  * ld_virtioreg:
     50  */
     51 /* Configuration registers */
     52 #define VIRTIO_BLK_CONFIG_CAPACITY	0 /* 64bit */
     53 #define VIRTIO_BLK_CONFIG_SIZE_MAX	8 /* 32bit */
     54 #define VIRTIO_BLK_CONFIG_SEG_MAX	12 /* 32bit */
     55 #define VIRTIO_BLK_CONFIG_GEOMETRY_C	16 /* 16bit */
     56 #define VIRTIO_BLK_CONFIG_GEOMETRY_H	18 /* 8bit */
     57 #define VIRTIO_BLK_CONFIG_GEOMETRY_S	19 /* 8bit */
     58 #define VIRTIO_BLK_CONFIG_BLK_SIZE	20 /* 32bit */
     59 #define VIRTIO_BLK_CONFIG_WRITEBACK	32 /* 8bit */
     60 
     61 /* Feature bits */
     62 #define VIRTIO_BLK_F_BARRIER	(1<<0)
     63 #define VIRTIO_BLK_F_SIZE_MAX	(1<<1)
     64 #define VIRTIO_BLK_F_SEG_MAX	(1<<2)
     65 #define VIRTIO_BLK_F_GEOMETRY	(1<<4)
     66 #define VIRTIO_BLK_F_RO		(1<<5)
     67 #define VIRTIO_BLK_F_BLK_SIZE	(1<<6)
     68 #define VIRTIO_BLK_F_SCSI	(1<<7)
     69 #define VIRTIO_BLK_F_FLUSH	(1<<9)
     70 #define VIRTIO_BLK_F_TOPOLOGY	(1<<10)
     71 #define VIRTIO_BLK_F_CONFIG_WCE	(1<<11)
     72 
     73 /*
     74  * Each block request uses at least two segments - one for the header
     75  * and one for the status.
     76 */
     77 #define	VIRTIO_BLK_MIN_SEGMENTS	2
     78 
     79 #define VIRTIO_BLK_FLAG_BITS			\
     80 	VIRTIO_COMMON_FLAG_BITS			\
     81 	"b\x0b" "CONFIG_WCE\0"			\
     82 	"b\x0a" "TOPOLOGY\0"			\
     83 	"b\x09" "FLUSH\0"			\
     84 	"b\x07" "SCSI\0"			\
     85 	"b\x06" "BLK_SIZE\0"			\
     86 	"b\x05" "RO\0"				\
     87 	"b\x04" "GEOMETRY\0"			\
     88 	"b\x02" "SEG_MAX\0"			\
     89 	"b\x01" "SIZE_MAX\0"			\
     90 	"b\x00" "BARRIER\0"
     91 
     92 /* Command */
     93 #define VIRTIO_BLK_T_IN		0
     94 #define VIRTIO_BLK_T_OUT	1
     95 #define VIRTIO_BLK_T_FLUSH	4
     96 #define VIRTIO_BLK_T_BARRIER	0x80000000
     97 
     98 /* Sector */
     99 #define VIRTIO_BLK_BSIZE	512
    100 
    101 /* Status */
    102 #define VIRTIO_BLK_S_OK		0
    103 #define VIRTIO_BLK_S_IOERR	1
    104 #define VIRTIO_BLK_S_UNSUPP	2
    105 
    106 /* Request header structure */
    107 struct virtio_blk_req_hdr {
    108 	uint32_t	type;	/* VIRTIO_BLK_T_* */
    109 	uint32_t	ioprio;
    110 	uint64_t	sector;
    111 } __packed;
    112 /* payload and 1 byte status follows */
    113 
    114 
    115 /*
    116  * ld_virtiovar:
    117  */
    118 struct virtio_blk_req {
    119 	struct virtio_blk_req_hdr	vr_hdr;
    120 	uint8_t				vr_status;
    121 	struct buf			*vr_bp;
    122 #define DUMMY_VR_BP				((void *)1)
    123 	bus_dmamap_t			vr_cmdsts;
    124 	bus_dmamap_t			vr_payload;
    125 };
    126 
    127 struct ld_virtio_softc {
    128 	struct ld_softc		sc_ld;
    129 	device_t		sc_dev;
    130 
    131 	struct virtio_softc	*sc_virtio;
    132 	struct virtqueue	sc_vq;
    133 
    134 	struct virtio_blk_req	*sc_reqs;
    135 	bus_dma_segment_t	sc_reqs_seg;
    136 
    137 	int			sc_readonly;
    138 
    139 	enum {
    140 		SYNC_FREE, SYNC_BUSY, SYNC_DONE
    141 	}			sc_sync_use;
    142 	kcondvar_t		sc_sync_wait;
    143 	kmutex_t		sc_sync_wait_lock;
    144 	uint8_t			sc_sync_status;
    145 };
    146 
    147 static int	ld_virtio_match(device_t, cfdata_t, void *);
    148 static void	ld_virtio_attach(device_t, device_t, void *);
    149 static int	ld_virtio_detach(device_t, int);
    150 
    151 CFATTACH_DECL_NEW(ld_virtio, sizeof(struct ld_virtio_softc),
    152     ld_virtio_match, ld_virtio_attach, ld_virtio_detach, NULL);
    153 
    154 static int
    155 ld_virtio_match(device_t parent, cfdata_t match, void *aux)
    156 {
    157 	struct virtio_attach_args *va = aux;
    158 
    159 	if (va->sc_childdevid == VIRTIO_DEVICE_ID_BLOCK)
    160 		return 1;
    161 
    162 	return 0;
    163 }
    164 
    165 static int ld_virtio_vq_done(struct virtqueue *);
    166 static int ld_virtio_dump(struct ld_softc *, void *, int, int);
    167 static int ld_virtio_start(struct ld_softc *, struct buf *);
    168 static int ld_virtio_ioctl(struct ld_softc *, u_long, void *, int32_t, bool);
    169 
    170 static int
    171 ld_virtio_alloc_reqs(struct ld_virtio_softc *sc, int qsize)
    172 {
    173 	int allocsize, r, rsegs, i;
    174 	struct ld_softc *ld = &sc->sc_ld;
    175 	void *vaddr;
    176 
    177 	allocsize = sizeof(struct virtio_blk_req) * qsize;
    178 	r = bus_dmamem_alloc(virtio_dmat(sc->sc_virtio), allocsize, 0, 0,
    179 			     &sc->sc_reqs_seg, 1, &rsegs, BUS_DMA_WAITOK);
    180 	if (r != 0) {
    181 		aprint_error_dev(sc->sc_dev,
    182 				 "DMA memory allocation failed, size %d, "
    183 				 "error code %d\n", allocsize, r);
    184 		goto err_none;
    185 	}
    186 	r = bus_dmamem_map(virtio_dmat(sc->sc_virtio),
    187 			   &sc->sc_reqs_seg, 1, allocsize,
    188 			   &vaddr, BUS_DMA_WAITOK);
    189 	if (r != 0) {
    190 		aprint_error_dev(sc->sc_dev,
    191 				 "DMA memory map failed, "
    192 				 "error code %d\n", r);
    193 		goto err_dmamem_alloc;
    194 	}
    195 	sc->sc_reqs = vaddr;
    196 	memset(vaddr, 0, allocsize);
    197 	for (i = 0; i < qsize; i++) {
    198 		struct virtio_blk_req *vr = &sc->sc_reqs[i];
    199 		r = bus_dmamap_create(virtio_dmat(sc->sc_virtio),
    200 				      offsetof(struct virtio_blk_req, vr_bp),
    201 				      1,
    202 				      offsetof(struct virtio_blk_req, vr_bp),
    203 				      0,
    204 				      BUS_DMA_WAITOK|BUS_DMA_ALLOCNOW,
    205 				      &vr->vr_cmdsts);
    206 		if (r != 0) {
    207 			aprint_error_dev(sc->sc_dev,
    208 					 "command dmamap creation failed, "
    209 					 "error code %d\n", r);
    210 			goto err_reqs;
    211 		}
    212 		r = bus_dmamap_load(virtio_dmat(sc->sc_virtio), vr->vr_cmdsts,
    213 				    &vr->vr_hdr,
    214 				    offsetof(struct virtio_blk_req, vr_bp),
    215 				    NULL, BUS_DMA_WAITOK);
    216 		if (r != 0) {
    217 			aprint_error_dev(sc->sc_dev,
    218 					 "command dmamap load failed, "
    219 					 "error code %d\n", r);
    220 			goto err_reqs;
    221 		}
    222 		r = bus_dmamap_create(virtio_dmat(sc->sc_virtio),
    223 				      ld->sc_maxxfer,
    224 				      (ld->sc_maxxfer / NBPG) +
    225 				      VIRTIO_BLK_MIN_SEGMENTS,
    226 				      ld->sc_maxxfer,
    227 				      0,
    228 				      BUS_DMA_WAITOK|BUS_DMA_ALLOCNOW,
    229 				      &vr->vr_payload);
    230 		if (r != 0) {
    231 			aprint_error_dev(sc->sc_dev,
    232 					 "payload dmamap creation failed, "
    233 					 "error code %d\n", r);
    234 			goto err_reqs;
    235 		}
    236 	}
    237 	return 0;
    238 
    239 err_reqs:
    240 	for (i = 0; i < qsize; i++) {
    241 		struct virtio_blk_req *vr = &sc->sc_reqs[i];
    242 		if (vr->vr_cmdsts) {
    243 			bus_dmamap_destroy(virtio_dmat(sc->sc_virtio),
    244 					   vr->vr_cmdsts);
    245 			vr->vr_cmdsts = 0;
    246 		}
    247 		if (vr->vr_payload) {
    248 			bus_dmamap_destroy(virtio_dmat(sc->sc_virtio),
    249 					   vr->vr_payload);
    250 			vr->vr_payload = 0;
    251 		}
    252 	}
    253 	bus_dmamem_unmap(virtio_dmat(sc->sc_virtio), sc->sc_reqs, allocsize);
    254 err_dmamem_alloc:
    255 	bus_dmamem_free(virtio_dmat(sc->sc_virtio), &sc->sc_reqs_seg, 1);
    256 err_none:
    257 	return -1;
    258 }
    259 
    260 static void
    261 ld_virtio_attach(device_t parent, device_t self, void *aux)
    262 {
    263 	struct ld_virtio_softc *sc = device_private(self);
    264 	struct ld_softc *ld = &sc->sc_ld;
    265 	struct virtio_softc *vsc = device_private(parent);
    266 	uint64_t features;
    267 	int qsize, maxxfersize, maxnsegs;
    268 
    269 	if (virtio_child(vsc) != NULL) {
    270 		aprint_normal(": child already attached for %s; "
    271 			      "something wrong...\n", device_xname(parent));
    272 		return;
    273 	}
    274 
    275 	sc->sc_dev = self;
    276 	sc->sc_virtio = vsc;
    277 
    278 	virtio_child_attach_start(vsc, self, IPL_BIO,
    279 	    (VIRTIO_BLK_F_SIZE_MAX | VIRTIO_BLK_F_SEG_MAX |
    280 	     VIRTIO_BLK_F_GEOMETRY | VIRTIO_BLK_F_RO | VIRTIO_BLK_F_BLK_SIZE |
    281 	     VIRTIO_BLK_F_FLUSH | VIRTIO_BLK_F_CONFIG_WCE),
    282 	    VIRTIO_BLK_FLAG_BITS);
    283 
    284 	features = virtio_features(vsc);
    285 	if (features == 0)
    286 		goto err;
    287 
    288 	if (features & VIRTIO_BLK_F_RO)
    289 		sc->sc_readonly = 1;
    290 	else
    291 		sc->sc_readonly = 0;
    292 
    293 	if (features & VIRTIO_BLK_F_BLK_SIZE) {
    294 		ld->sc_secsize = virtio_read_device_config_4(vsc,
    295 					VIRTIO_BLK_CONFIG_BLK_SIZE);
    296 	} else
    297 		ld->sc_secsize = VIRTIO_BLK_BSIZE;
    298 
    299 	/* At least genfs_io assumes maxxfer == MAXPHYS. */
    300 	if (features & VIRTIO_BLK_F_SIZE_MAX) {
    301 		maxxfersize = virtio_read_device_config_4(vsc,
    302 		    VIRTIO_BLK_CONFIG_SIZE_MAX);
    303 		if (maxxfersize < MAXPHYS) {
    304 			aprint_error_dev(sc->sc_dev,
    305 			    "Too small SIZE_MAX %dK minimum is %dK\n",
    306 			    maxxfersize / 1024, MAXPHYS / 1024);
    307 			// goto err;
    308 			maxxfersize = MAXPHYS;
    309 		} else if (maxxfersize > MAXPHYS) {
    310 			aprint_normal_dev(sc->sc_dev,
    311 			    "Clip SEG_MAX from %dK to %dK\n",
    312 			    maxxfersize / 1024,
    313 			    MAXPHYS / 1024);
    314 			maxxfersize = MAXPHYS;
    315 		}
    316 	} else
    317 		maxxfersize = MAXPHYS;
    318 
    319 	if (features & VIRTIO_BLK_F_SEG_MAX) {
    320 		maxnsegs = virtio_read_device_config_4(vsc,
    321 		    VIRTIO_BLK_CONFIG_SEG_MAX);
    322 		if (maxnsegs < VIRTIO_BLK_MIN_SEGMENTS) {
    323 			aprint_error_dev(sc->sc_dev,
    324 			    "Too small SEG_MAX %d minimum is %d\n",
    325 			    maxnsegs, VIRTIO_BLK_MIN_SEGMENTS);
    326 			maxnsegs = maxxfersize / NBPG;
    327 			// goto err;
    328 		}
    329 	} else
    330 		maxnsegs = maxxfersize / NBPG;
    331 
    332 	/* 2 for the minimum size */
    333 	maxnsegs += VIRTIO_BLK_MIN_SEGMENTS;
    334 
    335 	virtio_init_vq_vqdone(vsc, &sc->sc_vq, 0,
    336 	    ld_virtio_vq_done);
    337 
    338 	if (virtio_alloc_vq(vsc, &sc->sc_vq, maxxfersize, maxnsegs,
    339 	    "I/O request") != 0) {
    340 		goto err;
    341 	}
    342 	qsize = sc->sc_vq.vq_num;
    343 
    344 	if (virtio_child_attach_finish(vsc, &sc->sc_vq, 1,
    345 	    NULL, VIRTIO_F_INTR_MSIX) != 0)
    346 		goto err;
    347 
    348 	ld->sc_dv = self;
    349 	ld->sc_secperunit = virtio_read_device_config_8(vsc,
    350 	    VIRTIO_BLK_CONFIG_CAPACITY) / (ld->sc_secsize / VIRTIO_BLK_BSIZE);
    351 	ld->sc_maxxfer = maxxfersize;
    352 	if (features & VIRTIO_BLK_F_GEOMETRY) {
    353 		ld->sc_ncylinders = virtio_read_device_config_2(vsc,
    354 					VIRTIO_BLK_CONFIG_GEOMETRY_C);
    355 		ld->sc_nheads     = virtio_read_device_config_1(vsc,
    356 					VIRTIO_BLK_CONFIG_GEOMETRY_H);
    357 		ld->sc_nsectors   = virtio_read_device_config_1(vsc,
    358 					VIRTIO_BLK_CONFIG_GEOMETRY_S);
    359 	}
    360 	ld->sc_maxqueuecnt = qsize - 1; /* reserve slot for dumps, flushes */
    361 
    362 	if (ld_virtio_alloc_reqs(sc, qsize) < 0)
    363 		goto err;
    364 
    365 	cv_init(&sc->sc_sync_wait, "vblksync");
    366 	mutex_init(&sc->sc_sync_wait_lock, MUTEX_DEFAULT, IPL_BIO);
    367 	sc->sc_sync_use = SYNC_FREE;
    368 
    369 	ld->sc_dump = ld_virtio_dump;
    370 	ld->sc_start = ld_virtio_start;
    371 	ld->sc_ioctl = ld_virtio_ioctl;
    372 
    373 	ld->sc_flags = LDF_ENABLED | LDF_MPSAFE;
    374 	ldattach(ld, BUFQ_DISK_DEFAULT_STRAT);
    375 
    376 	return;
    377 
    378 err:
    379 	virtio_child_attach_failed(vsc);
    380 	return;
    381 }
    382 
    383 static int
    384 ld_virtio_start(struct ld_softc *ld, struct buf *bp)
    385 {
    386 	/* splbio */
    387 	struct ld_virtio_softc *sc = device_private(ld->sc_dv);
    388 	struct virtio_softc *vsc = sc->sc_virtio;
    389 	struct virtqueue *vq = &sc->sc_vq;
    390 	struct virtio_blk_req *vr;
    391 	int r;
    392 	int isread = (bp->b_flags & B_READ);
    393 	int slot;
    394 
    395 	if (sc->sc_readonly && !isread)
    396 		return EIO;
    397 
    398 	r = virtio_enqueue_prep(vsc, vq, &slot);
    399 	if (r != 0)
    400 		return r;
    401 
    402 	vr = &sc->sc_reqs[slot];
    403 	KASSERT(vr->vr_bp == NULL);
    404 
    405 	r = bus_dmamap_load(virtio_dmat(vsc), vr->vr_payload,
    406 			    bp->b_data, bp->b_bcount, NULL,
    407 			    ((isread?BUS_DMA_READ:BUS_DMA_WRITE)
    408 			     |BUS_DMA_NOWAIT));
    409 	if (r != 0) {
    410 		aprint_error_dev(sc->sc_dev,
    411 		    "payload dmamap failed, error code %d\n", r);
    412 		virtio_enqueue_abort(vsc, vq, slot);
    413 		return r;
    414 	}
    415 
    416 	r = virtio_enqueue_reserve(vsc, vq, slot, vr->vr_payload->dm_nsegs +
    417 	    VIRTIO_BLK_MIN_SEGMENTS);
    418 	if (r != 0) {
    419 		bus_dmamap_unload(virtio_dmat(vsc), vr->vr_payload);
    420 		return r;
    421 	}
    422 
    423 	vr->vr_bp = bp;
    424 	vr->vr_hdr.type   = virtio_rw32(vsc,
    425 			isread ? VIRTIO_BLK_T_IN : VIRTIO_BLK_T_OUT);
    426 	vr->vr_hdr.ioprio = virtio_rw32(vsc, 0);
    427 	vr->vr_hdr.sector = virtio_rw64(vsc,
    428 			bp->b_rawblkno * sc->sc_ld.sc_secsize /
    429 			VIRTIO_BLK_BSIZE);
    430 
    431 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    432 			0, sizeof(struct virtio_blk_req_hdr),
    433 			BUS_DMASYNC_PREWRITE);
    434 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_payload,
    435 			0, bp->b_bcount,
    436 			isread?BUS_DMASYNC_PREREAD:BUS_DMASYNC_PREWRITE);
    437 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    438 			offsetof(struct virtio_blk_req, vr_status),
    439 			sizeof(uint8_t),
    440 			BUS_DMASYNC_PREREAD);
    441 
    442 	virtio_enqueue_p(vsc, vq, slot, vr->vr_cmdsts,
    443 			 0, sizeof(struct virtio_blk_req_hdr),
    444 			 true);
    445 	virtio_enqueue(vsc, vq, slot, vr->vr_payload, !isread);
    446 	virtio_enqueue_p(vsc, vq, slot, vr->vr_cmdsts,
    447 			 offsetof(struct virtio_blk_req, vr_status),
    448 			 sizeof(uint8_t),
    449 			 false);
    450 	virtio_enqueue_commit(vsc, vq, slot, true);
    451 
    452 	return 0;
    453 }
    454 
    455 static void
    456 ld_virtio_vq_done1(struct ld_virtio_softc *sc, struct virtio_softc *vsc,
    457 		   struct virtqueue *vq, int slot)
    458 {
    459 	struct virtio_blk_req *vr = &sc->sc_reqs[slot];
    460 	struct buf *bp = vr->vr_bp;
    461 
    462 	vr->vr_bp = NULL;
    463 
    464 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    465 			0, sizeof(struct virtio_blk_req_hdr),
    466 			BUS_DMASYNC_POSTWRITE);
    467 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    468 			sizeof(struct virtio_blk_req_hdr), sizeof(uint8_t),
    469 			BUS_DMASYNC_POSTREAD);
    470 	if (bp == DUMMY_VR_BP) {
    471 		mutex_enter(&sc->sc_sync_wait_lock);
    472 		sc->sc_sync_status = vr->vr_status;
    473 		sc->sc_sync_use = SYNC_DONE;
    474 		cv_broadcast(&sc->sc_sync_wait);
    475 		mutex_exit(&sc->sc_sync_wait_lock);
    476 		virtio_dequeue_commit(vsc, vq, slot);
    477 		return;
    478 	}
    479 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_payload,
    480 			0, bp->b_bcount,
    481 			(bp->b_flags & B_READ)?BUS_DMASYNC_POSTREAD
    482 					      :BUS_DMASYNC_POSTWRITE);
    483 	bus_dmamap_unload(virtio_dmat(vsc), vr->vr_payload);
    484 
    485 	if (vr->vr_status != VIRTIO_BLK_S_OK) {
    486 		bp->b_error = EIO;
    487 		bp->b_resid = bp->b_bcount;
    488 	} else {
    489 		bp->b_error = 0;
    490 		bp->b_resid = 0;
    491 	}
    492 
    493 	virtio_dequeue_commit(vsc, vq, slot);
    494 
    495 	lddone(&sc->sc_ld, bp);
    496 }
    497 
    498 static int
    499 ld_virtio_vq_done(struct virtqueue *vq)
    500 {
    501 	struct virtio_softc *vsc = vq->vq_owner;
    502 	struct ld_virtio_softc *sc = device_private(virtio_child(vsc));
    503 	int r = 0;
    504 	int slot;
    505 
    506 again:
    507 	if (virtio_dequeue(vsc, vq, &slot, NULL))
    508 		return r;
    509 	r = 1;
    510 
    511 	ld_virtio_vq_done1(sc, vsc, vq, slot);
    512 	goto again;
    513 }
    514 
    515 static int
    516 ld_virtio_dump(struct ld_softc *ld, void *data, int blkno, int blkcnt)
    517 {
    518 	struct ld_virtio_softc *sc = device_private(ld->sc_dv);
    519 	struct virtio_softc *vsc = sc->sc_virtio;
    520 	struct virtqueue *vq = &sc->sc_vq;
    521 	struct virtio_blk_req *vr;
    522 	int slot, r;
    523 
    524 	if (sc->sc_readonly)
    525 		return EIO;
    526 
    527 	r = virtio_enqueue_prep(vsc, vq, &slot);
    528 	if (r != 0) {
    529 		if (r == EAGAIN) { /* no free slot; dequeue first */
    530 			delay(100);
    531 			ld_virtio_vq_done(vq);
    532 			r = virtio_enqueue_prep(vsc, vq, &slot);
    533 			if (r != 0)
    534 				return r;
    535 		}
    536 		return r;
    537 	}
    538 	vr = &sc->sc_reqs[slot];
    539 	r = bus_dmamap_load(virtio_dmat(vsc), vr->vr_payload,
    540 			    data, blkcnt*ld->sc_secsize, NULL,
    541 			    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
    542 	if (r != 0)
    543 		return r;
    544 
    545 	r = virtio_enqueue_reserve(vsc, vq, slot, vr->vr_payload->dm_nsegs +
    546 	    VIRTIO_BLK_MIN_SEGMENTS);
    547 	if (r != 0) {
    548 		bus_dmamap_unload(virtio_dmat(vsc), vr->vr_payload);
    549 		return r;
    550 	}
    551 
    552 	vr->vr_bp = (void*)0xdeadbeef;
    553 	vr->vr_hdr.type   = virtio_rw32(vsc, VIRTIO_BLK_T_OUT);
    554 	vr->vr_hdr.ioprio = virtio_rw32(vsc, 0);
    555 	vr->vr_hdr.sector = virtio_rw64(vsc,
    556 			(daddr_t) blkno * ld->sc_secsize /
    557 			VIRTIO_BLK_BSIZE);
    558 
    559 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    560 			0, sizeof(struct virtio_blk_req_hdr),
    561 			BUS_DMASYNC_PREWRITE);
    562 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_payload,
    563 			0, blkcnt*ld->sc_secsize,
    564 			BUS_DMASYNC_PREWRITE);
    565 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    566 			offsetof(struct virtio_blk_req, vr_status),
    567 			sizeof(uint8_t),
    568 			BUS_DMASYNC_PREREAD);
    569 
    570 	virtio_enqueue_p(vsc, vq, slot, vr->vr_cmdsts,
    571 			 0, sizeof(struct virtio_blk_req_hdr),
    572 			 true);
    573 	virtio_enqueue(vsc, vq, slot, vr->vr_payload, true);
    574 	virtio_enqueue_p(vsc, vq, slot, vr->vr_cmdsts,
    575 			 offsetof(struct virtio_blk_req, vr_status),
    576 			 sizeof(uint8_t),
    577 			 false);
    578 	virtio_enqueue_commit(vsc, vq, slot, true);
    579 
    580 	for ( ; ; ) {
    581 		int dslot;
    582 
    583 		r = virtio_dequeue(vsc, vq, &dslot, NULL);
    584 		if (r != 0)
    585 			continue;
    586 		if (dslot != slot) {
    587 			ld_virtio_vq_done1(sc, vsc, vq, dslot);
    588 			continue;
    589 		} else
    590 			break;
    591 	}
    592 
    593 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    594 			0, sizeof(struct virtio_blk_req_hdr),
    595 			BUS_DMASYNC_POSTWRITE);
    596 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_payload,
    597 			0, blkcnt*ld->sc_secsize,
    598 			BUS_DMASYNC_POSTWRITE);
    599 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    600 			offsetof(struct virtio_blk_req, vr_status),
    601 			sizeof(uint8_t),
    602 			BUS_DMASYNC_POSTREAD);
    603 	if (vr->vr_status == VIRTIO_BLK_S_OK)
    604 		r = 0;
    605 	else
    606 		r = EIO;
    607 	virtio_dequeue_commit(vsc, vq, slot);
    608 
    609 	return r;
    610 }
    611 
    612 static int
    613 ld_virtio_detach(device_t self, int flags)
    614 {
    615 	struct ld_virtio_softc *sc = device_private(self);
    616 	struct ld_softc *ld = &sc->sc_ld;
    617 	bus_dma_tag_t dmat = virtio_dmat(sc->sc_virtio);
    618 	int r, i, qsize;
    619 
    620 	qsize = sc->sc_vq.vq_num;
    621 	r = ldbegindetach(ld, flags);
    622 	if (r != 0)
    623 		return r;
    624 	virtio_reset(sc->sc_virtio);
    625 	virtio_free_vq(sc->sc_virtio, &sc->sc_vq);
    626 
    627 	for (i = 0; i < qsize; i++) {
    628 		bus_dmamap_destroy(dmat,
    629 				   sc->sc_reqs[i].vr_cmdsts);
    630 		bus_dmamap_destroy(dmat,
    631 				   sc->sc_reqs[i].vr_payload);
    632 	}
    633 	bus_dmamem_unmap(dmat, sc->sc_reqs,
    634 			 sizeof(struct virtio_blk_req) * qsize);
    635 	bus_dmamem_free(dmat, &sc->sc_reqs_seg, 1);
    636 
    637 	ldenddetach(ld);
    638 
    639 	cv_destroy(&sc->sc_sync_wait);
    640 	mutex_destroy(&sc->sc_sync_wait_lock);
    641 
    642 	virtio_child_detach(sc->sc_virtio);
    643 
    644 	return 0;
    645 }
    646 
    647 static int
    648 ld_virtio_flush(struct ld_softc *ld, bool poll)
    649 {
    650 	struct ld_virtio_softc * const sc = device_private(ld->sc_dv);
    651 	struct virtio_softc * const vsc = sc->sc_virtio;
    652 	const uint64_t features = virtio_features(vsc);
    653 	struct virtqueue *vq = &sc->sc_vq;
    654 	struct virtio_blk_req *vr;
    655 	int slot;
    656 	int r;
    657 
    658 	if ((features & VIRTIO_BLK_F_FLUSH) == 0)
    659 		return 0;
    660 
    661 	mutex_enter(&sc->sc_sync_wait_lock);
    662 	while (sc->sc_sync_use != SYNC_FREE) {
    663 		if (poll) {
    664 			mutex_exit(&sc->sc_sync_wait_lock);
    665 			ld_virtio_vq_done(vq);
    666 			mutex_enter(&sc->sc_sync_wait_lock);
    667 			continue;
    668 		}
    669 		cv_wait(&sc->sc_sync_wait, &sc->sc_sync_wait_lock);
    670 	}
    671 	sc->sc_sync_use = SYNC_BUSY;
    672 	mutex_exit(&sc->sc_sync_wait_lock);
    673 
    674 	r = virtio_enqueue_prep(vsc, vq, &slot);
    675 	if (r != 0) {
    676 		return r;
    677 	}
    678 
    679 	vr = &sc->sc_reqs[slot];
    680 	KASSERT(vr->vr_bp == NULL);
    681 
    682 	r = virtio_enqueue_reserve(vsc, vq, slot, VIRTIO_BLK_MIN_SEGMENTS);
    683 	if (r != 0) {
    684 		return r;
    685 	}
    686 
    687 	vr->vr_bp = DUMMY_VR_BP;
    688 	vr->vr_hdr.type   = virtio_rw32(vsc, VIRTIO_BLK_T_FLUSH);
    689 	vr->vr_hdr.ioprio = virtio_rw32(vsc, 0);
    690 	vr->vr_hdr.sector = virtio_rw64(vsc, 0);
    691 
    692 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    693 			0, sizeof(struct virtio_blk_req_hdr),
    694 			BUS_DMASYNC_PREWRITE);
    695 	bus_dmamap_sync(virtio_dmat(vsc), vr->vr_cmdsts,
    696 			offsetof(struct virtio_blk_req, vr_status),
    697 			sizeof(uint8_t),
    698 			BUS_DMASYNC_PREREAD);
    699 
    700 	virtio_enqueue_p(vsc, vq, slot, vr->vr_cmdsts,
    701 			 0, sizeof(struct virtio_blk_req_hdr),
    702 			 true);
    703 	virtio_enqueue_p(vsc, vq, slot, vr->vr_cmdsts,
    704 			 offsetof(struct virtio_blk_req, vr_status),
    705 			 sizeof(uint8_t),
    706 			 false);
    707 	virtio_enqueue_commit(vsc, vq, slot, true);
    708 
    709 	mutex_enter(&sc->sc_sync_wait_lock);
    710 	while (sc->sc_sync_use != SYNC_DONE) {
    711 		if (poll) {
    712 			mutex_exit(&sc->sc_sync_wait_lock);
    713 			ld_virtio_vq_done(vq);
    714 			mutex_enter(&sc->sc_sync_wait_lock);
    715 			continue;
    716 		}
    717 		cv_wait(&sc->sc_sync_wait, &sc->sc_sync_wait_lock);
    718 	}
    719 
    720 	if (sc->sc_sync_status == VIRTIO_BLK_S_OK)
    721 		r = 0;
    722 	else
    723 		r = EIO;
    724 
    725 	sc->sc_sync_use = SYNC_FREE;
    726 	cv_broadcast(&sc->sc_sync_wait);
    727 	mutex_exit(&sc->sc_sync_wait_lock);
    728 
    729 	return r;
    730 }
    731 
    732 static int
    733 ld_virtio_getcache(struct ld_softc *ld, int *bitsp)
    734 {
    735 	struct ld_virtio_softc * const sc = device_private(ld->sc_dv);
    736 	struct virtio_softc * const vsc = sc->sc_virtio;
    737 	const uint64_t features = virtio_features(vsc);
    738 
    739 	*bitsp = DKCACHE_READ;
    740 	if ((features & VIRTIO_BLK_F_CONFIG_WCE) != 0)
    741 		*bitsp |= DKCACHE_WCHANGE;
    742 	if (virtio_read_device_config_1(vsc,
    743 	    VIRTIO_BLK_CONFIG_WRITEBACK) != 0x00)
    744 		*bitsp |= DKCACHE_WRITE;
    745 
    746 	return 0;
    747 }
    748 
    749 static int
    750 ld_virtio_setcache(struct ld_softc *ld, int bits)
    751 {
    752 	struct ld_virtio_softc * const sc = device_private(ld->sc_dv);
    753 	struct virtio_softc * const vsc = sc->sc_virtio;
    754 	const uint8_t wce = (bits & DKCACHE_WRITE) ? 0x01 : 0x00;
    755 
    756 	virtio_write_device_config_1(vsc,
    757 	    VIRTIO_BLK_CONFIG_WRITEBACK, wce);
    758 	if (virtio_read_device_config_1(vsc,
    759 	    VIRTIO_BLK_CONFIG_WRITEBACK) != wce)
    760 		return EIO;
    761 
    762 	return 0;
    763 }
    764 
    765 static int
    766 ld_virtio_ioctl(struct ld_softc *ld, u_long cmd, void *addr, int32_t flag, bool poll)
    767 {
    768 	int error;
    769 
    770 	switch (cmd) {
    771 	case DIOCCACHESYNC:
    772 		error = ld_virtio_flush(ld, poll);
    773 		break;
    774 
    775 	case DIOCGCACHE:
    776 		error = ld_virtio_getcache(ld, (int *)addr);
    777 		break;
    778 
    779 	case DIOCSCACHE:
    780 		error = ld_virtio_setcache(ld, *(int *)addr);
    781 		break;
    782 
    783 	default:
    784 		error = EPASSTHROUGH;
    785 		break;
    786 	}
    787 
    788 	return error;
    789 }
    790 
    791 MODULE(MODULE_CLASS_DRIVER, ld_virtio, "ld,virtio");
    792 
    793 #ifdef _MODULE
    794 /*
    795  * XXX Don't allow ioconf.c to redefine the "struct cfdriver ld_cd"
    796  * XXX it will be defined in the common-code module
    797  */
    798 #undef  CFDRIVER_DECL
    799 #define CFDRIVER_DECL(name, class, attr)
    800 #include "ioconf.c"
    801 #endif
    802 
    803 static int
    804 ld_virtio_modcmd(modcmd_t cmd, void *opaque)
    805 {
    806 #ifdef _MODULE
    807 	/*
    808 	 * We ignore the cfdriver_vec[] that ioconf provides, since
    809 	 * the cfdrivers are attached already.
    810 	 */
    811 	static struct cfdriver * const no_cfdriver_vec[] = { NULL };
    812 #endif
    813 	int error = 0;
    814 
    815 #ifdef _MODULE
    816 	switch (cmd) {
    817 	case MODULE_CMD_INIT:
    818 		error = config_init_component(no_cfdriver_vec,
    819 		    cfattach_ioconf_ld_virtio, cfdata_ioconf_ld_virtio);
    820 		break;
    821 	case MODULE_CMD_FINI:
    822 		error = config_fini_component(no_cfdriver_vec,
    823 		    cfattach_ioconf_ld_virtio, cfdata_ioconf_ld_virtio);
    824 		break;
    825 	default:
    826 		error = ENOTTY;
    827 		break;
    828 	}
    829 #endif
    830 
    831 	return error;
    832 }
    833