xref: /freebsd/usr.sbin/bhyve/pci_ahci.c (revision 4abd7edcbde21ba7a089c7d1a0bba8f87ebece06)
1 /*-
2  * Copyright (c) 2013  Zhixiang Yu <zcore@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/stat.h>
35 #include <sys/uio.h>
36 #include <sys/ioctl.h>
37 #include <sys/disk.h>
38 #include <sys/ata.h>
39 #include <sys/endian.h>
40 
41 #include <errno.h>
42 #include <fcntl.h>
43 #include <stdio.h>
44 #include <stdlib.h>
45 #include <stdint.h>
46 #include <string.h>
47 #include <strings.h>
48 #include <unistd.h>
49 #include <assert.h>
50 #include <pthread.h>
51 #include <inttypes.h>
52 
53 #include "bhyverun.h"
54 #include "pci_emul.h"
55 #include "ahci.h"
56 #include "block_if.h"
57 
58 #define	MAX_PORTS	6	/* Intel ICH8 AHCI supports 6 ports */
59 
60 #define	PxSIG_ATA	0x00000101 /* ATA drive */
61 #define	PxSIG_ATAPI	0xeb140101 /* ATAPI drive */
62 
63 enum sata_fis_type {
64 	FIS_TYPE_REGH2D		= 0x27,	/* Register FIS - host to device */
65 	FIS_TYPE_REGD2H		= 0x34,	/* Register FIS - device to host */
66 	FIS_TYPE_DMAACT		= 0x39,	/* DMA activate FIS - device to host */
67 	FIS_TYPE_DMASETUP	= 0x41,	/* DMA setup FIS - bidirectional */
68 	FIS_TYPE_DATA		= 0x46,	/* Data FIS - bidirectional */
69 	FIS_TYPE_BIST		= 0x58,	/* BIST activate FIS - bidirectional */
70 	FIS_TYPE_PIOSETUP	= 0x5F,	/* PIO setup FIS - device to host */
71 	FIS_TYPE_SETDEVBITS	= 0xA1,	/* Set dev bits FIS - device to host */
72 };
73 
74 /*
75  * SCSI opcodes
76  */
77 #define	TEST_UNIT_READY		0x00
78 #define	REQUEST_SENSE		0x03
79 #define	INQUIRY			0x12
80 #define	START_STOP_UNIT		0x1B
81 #define	PREVENT_ALLOW		0x1E
82 #define	READ_CAPACITY		0x25
83 #define	READ_10			0x28
84 #define	POSITION_TO_ELEMENT	0x2B
85 #define	READ_TOC		0x43
86 #define	GET_EVENT_STATUS_NOTIFICATION 0x4A
87 #define	MODE_SENSE_10		0x5A
88 #define	READ_12			0xA8
89 #define	READ_CD			0xBE
90 
91 /*
92  * SCSI mode page codes
93  */
94 #define	MODEPAGE_RW_ERROR_RECOVERY	0x01
95 #define	MODEPAGE_CD_CAPABILITIES	0x2A
96 
97 /*
98  * Debug printf
99  */
100 #ifdef AHCI_DEBUG
101 static FILE *dbg;
102 #define DPRINTF(format, arg...)	do{fprintf(dbg, format, ##arg);fflush(dbg);}while(0)
103 #else
104 #define DPRINTF(format, arg...)
105 #endif
106 #define WPRINTF(format, arg...) printf(format, ##arg)
107 
108 struct ahci_ioreq {
109 	struct blockif_req io_req;
110 	struct ahci_port *io_pr;
111 	STAILQ_ENTRY(ahci_ioreq) io_list;
112 	uint8_t *cfis;
113 	uint32_t len;
114 	uint32_t done;
115 	int slot;
116 	int prdtl;
117 };
118 
119 struct ahci_port {
120 	struct blockif_ctxt *bctx;
121 	struct pci_ahci_softc *pr_sc;
122 	uint8_t *cmd_lst;
123 	uint8_t *rfis;
124 	int atapi;
125 	int reset;
126 	int mult_sectors;
127 	uint8_t xfermode;
128 	uint8_t sense_key;
129 	uint8_t asc;
130 
131 	uint32_t clb;
132 	uint32_t clbu;
133 	uint32_t fb;
134 	uint32_t fbu;
135 	uint32_t is;
136 	uint32_t ie;
137 	uint32_t cmd;
138 	uint32_t unused0;
139 	uint32_t tfd;
140 	uint32_t sig;
141 	uint32_t ssts;
142 	uint32_t sctl;
143 	uint32_t serr;
144 	uint32_t sact;
145 	uint32_t ci;
146 	uint32_t sntf;
147 	uint32_t fbs;
148 
149 	/*
150 	 * i/o request info
151 	 */
152 	struct ahci_ioreq *ioreq;
153 	int ioqsz;
154 	STAILQ_HEAD(ahci_fhead, ahci_ioreq) iofhd;
155 };
156 
157 struct ahci_cmd_hdr {
158 	uint16_t flags;
159 	uint16_t prdtl;
160 	uint32_t prdbc;
161 	uint64_t ctba;
162 	uint32_t reserved[4];
163 };
164 
165 struct ahci_prdt_entry {
166 	uint64_t dba;
167 	uint32_t reserved;
168 #define	DBCMASK		0x3fffff
169 	uint32_t dbc;
170 };
171 
172 struct pci_ahci_softc {
173 	struct pci_devinst *asc_pi;
174 	pthread_mutex_t	mtx;
175 	int ports;
176 	uint32_t cap;
177 	uint32_t ghc;
178 	uint32_t is;
179 	uint32_t pi;
180 	uint32_t vs;
181 	uint32_t ccc_ctl;
182 	uint32_t ccc_pts;
183 	uint32_t em_loc;
184 	uint32_t em_ctl;
185 	uint32_t cap2;
186 	uint32_t bohc;
187 	struct ahci_port port[MAX_PORTS];
188 };
189 #define	ahci_ctx(sc)	((sc)->asc_pi->pi_vmctx)
190 
191 static inline void lba_to_msf(uint8_t *buf, int lba)
192 {
193 	lba += 150;
194 	buf[0] = (lba / 75) / 60;
195 	buf[1] = (lba / 75) % 60;
196 	buf[2] = lba % 75;
197 }
198 
199 /*
200  * generate HBA intr depending on whether or not ports within
201  * the controller have an interrupt pending.
202  */
203 static void
204 ahci_generate_intr(struct pci_ahci_softc *sc)
205 {
206 	int i;
207 
208 	for (i = 0; i < sc->ports; i++) {
209 		struct ahci_port *pr;
210 		pr = &sc->port[i];
211 		if (pr->is & pr->ie)
212 			sc->is |= (1 << i);
213 	}
214 
215 	DPRINTF("%s %x\n", __func__, sc->is);
216 
217 	if (sc->is && (sc->ghc & AHCI_GHC_IE))
218 		pci_generate_msi(sc->asc_pi, 0);
219 }
220 
221 static void
222 ahci_write_fis(struct ahci_port *p, enum sata_fis_type ft, uint8_t *fis)
223 {
224 	int offset, len, irq;
225 
226 	if (p->rfis == NULL || !(p->cmd & AHCI_P_CMD_FRE))
227 		return;
228 
229 	switch (ft) {
230 	case FIS_TYPE_REGD2H:
231 		offset = 0x40;
232 		len = 20;
233 		irq = AHCI_P_IX_DHR;
234 		break;
235 	case FIS_TYPE_SETDEVBITS:
236 		offset = 0x58;
237 		len = 8;
238 		irq = AHCI_P_IX_SDB;
239 		break;
240 	case FIS_TYPE_PIOSETUP:
241 		offset = 0x20;
242 		len = 20;
243 		irq = 0;
244 		break;
245 	default:
246 		WPRINTF("unsupported fis type %d\n", ft);
247 		return;
248 	}
249 	memcpy(p->rfis + offset, fis, len);
250 	if (irq) {
251 		p->is |= irq;
252 		ahci_generate_intr(p->pr_sc);
253 	}
254 }
255 
256 static void
257 ahci_write_fis_sdb(struct ahci_port *p, int slot, uint32_t tfd)
258 {
259 	uint8_t fis[8];
260 	uint8_t error;
261 
262 	error = (tfd >> 8) & 0xff;
263 	memset(fis, 0, sizeof(fis));
264 	fis[0] = error;
265 	fis[2] = tfd & 0x77;
266 	*(uint32_t *)(fis + 4) = (1 << slot);
267 	if (fis[2] & ATA_S_ERROR)
268 		p->is |= AHCI_P_IX_TFE;
269 	p->tfd = tfd;
270 	ahci_write_fis(p, FIS_TYPE_SETDEVBITS, fis);
271 }
272 
273 static void
274 ahci_write_fis_d2h(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t tfd)
275 {
276 	uint8_t fis[20];
277 	uint8_t error;
278 
279 	error = (tfd >> 8) & 0xff;
280 	memset(fis, 0, sizeof(fis));
281 	fis[0] = FIS_TYPE_REGD2H;
282 	fis[1] = (1 << 6);
283 	fis[2] = tfd & 0xff;
284 	fis[3] = error;
285 	fis[4] = cfis[4];
286 	fis[5] = cfis[5];
287 	fis[6] = cfis[6];
288 	fis[7] = cfis[7];
289 	fis[8] = cfis[8];
290 	fis[9] = cfis[9];
291 	fis[10] = cfis[10];
292 	fis[11] = cfis[11];
293 	fis[12] = cfis[12];
294 	fis[13] = cfis[13];
295 	if (fis[2] & ATA_S_ERROR)
296 		p->is |= AHCI_P_IX_TFE;
297 	p->tfd = tfd;
298 	p->ci &= ~(1 << slot);
299 	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
300 }
301 
302 static void
303 ahci_write_reset_fis_d2h(struct ahci_port *p)
304 {
305 	uint8_t fis[20];
306 
307 	memset(fis, 0, sizeof(fis));
308 	fis[0] = FIS_TYPE_REGD2H;
309 	fis[3] = 1;
310 	fis[4] = 1;
311 	if (p->atapi) {
312 		fis[5] = 0x14;
313 		fis[6] = 0xeb;
314 	}
315 	fis[12] = 1;
316 	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
317 }
318 
319 static void
320 ahci_port_reset(struct ahci_port *pr)
321 {
322 	pr->sctl = 0;
323 	pr->serr = 0;
324 	pr->sact = 0;
325 	pr->xfermode = ATA_UDMA6;
326 	pr->mult_sectors = 128;
327 
328 	if (!pr->bctx) {
329 		pr->ssts = ATA_SS_DET_NO_DEVICE;
330 		pr->sig = 0xFFFFFFFF;
331 		pr->tfd = 0x7F;
332 		return;
333 	}
334 	pr->ssts = ATA_SS_DET_PHY_ONLINE | ATA_SS_SPD_GEN2 |
335 		ATA_SS_IPM_ACTIVE;
336 	pr->tfd = (1 << 8) | ATA_S_DSC | ATA_S_DMA;
337 	if (!pr->atapi) {
338 		pr->sig = PxSIG_ATA;
339 		pr->tfd |= ATA_S_READY;
340 	} else
341 		pr->sig = PxSIG_ATAPI;
342 	ahci_write_reset_fis_d2h(pr);
343 }
344 
345 static void
346 ahci_reset(struct pci_ahci_softc *sc)
347 {
348 	int i;
349 
350 	sc->ghc = AHCI_GHC_AE;
351 	sc->is = 0;
352 	for (i = 0; i < sc->ports; i++) {
353 		sc->port[i].ie = 0;
354 		sc->port[i].is = 0;
355 		ahci_port_reset(&sc->port[i]);
356 	}
357 }
358 
359 static void
360 ata_string(uint8_t *dest, const char *src, int len)
361 {
362 	int i;
363 
364 	for (i = 0; i < len; i++) {
365 		if (*src)
366 			dest[i ^ 1] = *src++;
367 		else
368 			dest[i ^ 1] = ' ';
369 	}
370 }
371 
372 static void
373 atapi_string(uint8_t *dest, const char *src, int len)
374 {
375 	int i;
376 
377 	for (i = 0; i < len; i++) {
378 		if (*src)
379 			dest[i] = *src++;
380 		else
381 			dest[i] = ' ';
382 	}
383 }
384 
385 static void
386 ahci_handle_dma(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t done,
387     int seek)
388 {
389 	struct ahci_ioreq *aior;
390 	struct blockif_req *breq;
391 	struct pci_ahci_softc *sc;
392 	struct ahci_prdt_entry *prdt;
393 	struct ahci_cmd_hdr *hdr;
394 	uint64_t lba;
395 	uint32_t len;
396 	int i, err, iovcnt, ncq, readop;
397 
398 	sc = p->pr_sc;
399 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
400 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
401 	ncq = 0;
402 	readop = 1;
403 
404 	prdt += seek;
405 	if (cfis[2] == ATA_WRITE_DMA || cfis[2] == ATA_WRITE_DMA48 ||
406 			cfis[2] == ATA_WRITE_FPDMA_QUEUED)
407 		readop = 0;
408 
409 	if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
410 			cfis[2] == ATA_READ_FPDMA_QUEUED) {
411 		lba = ((uint64_t)cfis[10] << 40) |
412 			((uint64_t)cfis[9] << 32) |
413 			((uint64_t)cfis[8] << 24) |
414 			((uint64_t)cfis[6] << 16) |
415 			((uint64_t)cfis[5] << 8) |
416 			cfis[4];
417 		len = cfis[11] << 8 | cfis[3];
418 		if (!len)
419 			len = 65536;
420 		ncq = 1;
421 	} else if (cfis[2] == ATA_READ_DMA48 || cfis[2] == ATA_WRITE_DMA48) {
422 		lba = ((uint64_t)cfis[10] << 40) |
423 			((uint64_t)cfis[9] << 32) |
424 			((uint64_t)cfis[8] << 24) |
425 			((uint64_t)cfis[6] << 16) |
426 			((uint64_t)cfis[5] << 8) |
427 			cfis[4];
428 		len = cfis[13] << 8 | cfis[12];
429 		if (!len)
430 			len = 65536;
431 	} else {
432 		lba = ((cfis[7] & 0xf) << 24) | (cfis[6] << 16) |
433 			(cfis[5] << 8) | cfis[4];
434 		len = cfis[12];
435 		if (!len)
436 			len = 256;
437 	}
438 	lba *= blockif_sectsz(p->bctx);
439 	len *= blockif_sectsz(p->bctx);
440 
441 	/*
442 	 * Pull request off free list
443 	 */
444 	aior = STAILQ_FIRST(&p->iofhd);
445 	assert(aior != NULL);
446 	STAILQ_REMOVE_HEAD(&p->iofhd, io_list);
447 	aior->cfis = cfis;
448 	aior->slot = slot;
449 	aior->len = len;
450 	aior->done = done;
451 	breq = &aior->io_req;
452 	breq->br_offset = lba + done;
453 	iovcnt = hdr->prdtl - seek;
454 	if (iovcnt > BLOCKIF_IOV_MAX) {
455 		aior->prdtl = iovcnt - BLOCKIF_IOV_MAX;
456 		iovcnt = BLOCKIF_IOV_MAX;
457 	} else
458 		aior->prdtl = 0;
459 	breq->br_iovcnt = iovcnt;
460 
461 	/*
462 	 * Build up the iovec based on the prdt
463 	 */
464 	for (i = 0; i < iovcnt; i++) {
465 		uint32_t dbcsz;
466 
467 		dbcsz = (prdt->dbc & DBCMASK) + 1;
468 		breq->br_iov[i].iov_base = paddr_guest2host(ahci_ctx(sc),
469 		    prdt->dba, dbcsz);
470 		breq->br_iov[i].iov_len = dbcsz;
471 		aior->done += dbcsz;
472 		prdt++;
473 	}
474 	if (readop)
475 		err = blockif_read(p->bctx, breq);
476 	else
477 		err = blockif_write(p->bctx, breq);
478 	assert(err == 0);
479 
480 	if (!aior->prdtl && ncq)
481 		p->ci &= ~(1 << slot);
482 }
483 
484 static void
485 ahci_handle_flush(struct ahci_port *p, int slot, uint8_t *cfis)
486 {
487 	struct ahci_ioreq *aior;
488 	struct blockif_req *breq;
489 	int err;
490 
491 	/*
492 	 * Pull request off free list
493 	 */
494 	aior = STAILQ_FIRST(&p->iofhd);
495 	assert(aior != NULL);
496 	STAILQ_REMOVE_HEAD(&p->iofhd, io_list);
497 	aior->cfis = cfis;
498 	aior->slot = slot;
499 	aior->len = 0;
500 	breq = &aior->io_req;
501 
502 	err = blockif_flush(p->bctx, breq);
503 	assert(err == 0);
504 }
505 
506 static inline void
507 write_prdt(struct ahci_port *p, int slot, uint8_t *cfis,
508 		void *buf, int size)
509 {
510 	struct ahci_cmd_hdr *hdr;
511 	struct ahci_prdt_entry *prdt;
512 	void *from;
513 	int i, len;
514 
515 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
516 	len = size;
517 	from = buf;
518 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
519 	for (i = 0; i < hdr->prdtl && len; i++) {
520 		uint8_t *ptr;
521 		uint32_t dbcsz;
522 
523 		dbcsz = (prdt->dbc & DBCMASK) + 1;
524 		ptr = paddr_guest2host(ahci_ctx(p->pr_sc), prdt->dba, dbcsz);
525 		memcpy(ptr, from, dbcsz);
526 		len -= dbcsz;
527 		from += dbcsz;
528 		prdt++;
529 	}
530 	hdr->prdbc = size - len;
531 }
532 
533 static void
534 handle_identify(struct ahci_port *p, int slot, uint8_t *cfis)
535 {
536 	struct ahci_cmd_hdr *hdr;
537 
538 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
539 	if (p->atapi || hdr->prdtl == 0) {
540 		p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
541 		p->is |= AHCI_P_IX_TFE;
542 	} else {
543 		uint16_t buf[256];
544 		uint64_t sectors;
545 
546 		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
547 		memset(buf, 0, sizeof(buf));
548 		buf[0] = 0x0040;
549 		/* TODO emulate different serial? */
550 		ata_string((uint8_t *)(buf+10), "123456", 20);
551 		ata_string((uint8_t *)(buf+23), "001", 8);
552 		ata_string((uint8_t *)(buf+27), "BHYVE SATA DISK", 40);
553 		buf[47] = (0x8000 | 128);
554 		buf[48] = 0x1;
555 		buf[49] = (1 << 8 | 1 << 9 | 1 << 11);
556 		buf[50] = (1 << 14);
557 		buf[53] = (1 << 1 | 1 << 2);
558 		if (p->mult_sectors)
559 			buf[59] = (0x100 | p->mult_sectors);
560 		buf[60] = sectors;
561 		buf[61] = (sectors >> 16);
562 		buf[63] = 0x7;
563 		if (p->xfermode & ATA_WDMA0)
564 			buf[63] |= (1 << ((p->xfermode & 7) + 8));
565 		buf[64] = 0x3;
566 		buf[65] = 100;
567 		buf[66] = 100;
568 		buf[67] = 100;
569 		buf[68] = 100;
570 		buf[75] = 31;
571 		buf[76] = (1 << 8 | 1 << 2);
572 		buf[80] = 0x1f0;
573 		buf[81] = 0x28;
574 		buf[82] = (1 << 5 | 1 << 14);
575 		buf[83] = (1 << 10 | 1 << 12 | 1 << 13 | 1 << 14);
576 		buf[84] = (1 << 14);
577 		buf[85] = (1 << 5 | 1 << 14);
578 		buf[86] = (1 << 10 | 1 << 12 | 1 << 13);
579 		buf[87] = (1 << 14);
580 		buf[88] = 0x7f;
581 		if (p->xfermode & ATA_UDMA0)
582 			buf[88] |= (1 << ((p->xfermode & 7) + 8));
583 		buf[93] = (1 | 1 <<14);
584 		buf[100] = sectors;
585 		buf[101] = (sectors >> 16);
586 		buf[102] = (sectors >> 32);
587 		buf[103] = (sectors >> 48);
588 		write_prdt(p, slot, cfis, (void *)buf, sizeof(buf));
589 		p->tfd = ATA_S_DSC | ATA_S_READY;
590 		p->is |= AHCI_P_IX_DP;
591 	}
592 	p->ci &= ~(1 << slot);
593 	ahci_generate_intr(p->pr_sc);
594 }
595 
596 static void
597 handle_atapi_identify(struct ahci_port *p, int slot, uint8_t *cfis)
598 {
599 	if (!p->atapi) {
600 		p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
601 		p->is |= AHCI_P_IX_TFE;
602 	} else {
603 		uint16_t buf[256];
604 
605 		memset(buf, 0, sizeof(buf));
606 		buf[0] = (2 << 14 | 5 << 8 | 1 << 7 | 2 << 5);
607 		/* TODO emulate different serial? */
608 		ata_string((uint8_t *)(buf+10), "123456", 20);
609 		ata_string((uint8_t *)(buf+23), "001", 8);
610 		ata_string((uint8_t *)(buf+27), "BHYVE SATA DVD ROM", 40);
611 		buf[49] = (1 << 9 | 1 << 8);
612 		buf[50] = (1 << 14 | 1);
613 		buf[53] = (1 << 2 | 1 << 1);
614 		buf[62] = 0x3f;
615 		buf[63] = 7;
616 		buf[64] = 3;
617 		buf[65] = 100;
618 		buf[66] = 100;
619 		buf[67] = 100;
620 		buf[68] = 100;
621 		buf[76] = (1 << 2 | 1 << 1);
622 		buf[78] = (1 << 5);
623 		buf[80] = (0x1f << 4);
624 		buf[82] = (1 << 4);
625 		buf[83] = (1 << 14);
626 		buf[84] = (1 << 14);
627 		buf[85] = (1 << 4);
628 		buf[87] = (1 << 14);
629 		buf[88] = (1 << 14 | 0x7f);
630 		write_prdt(p, slot, cfis, (void *)buf, sizeof(buf));
631 		p->tfd = ATA_S_DSC | ATA_S_READY;
632 		p->is |= AHCI_P_IX_DHR;
633 	}
634 	p->ci &= ~(1 << slot);
635 	ahci_generate_intr(p->pr_sc);
636 }
637 
638 static void
639 atapi_inquiry(struct ahci_port *p, int slot, uint8_t *cfis)
640 {
641 	uint8_t buf[36];
642 	uint8_t *acmd;
643 	int len;
644 
645 	acmd = cfis + 0x40;
646 
647 	buf[0] = 0x05;
648 	buf[1] = 0x80;
649 	buf[2] = 0x00;
650 	buf[3] = 0x21;
651 	buf[4] = 31;
652 	buf[5] = 0;
653 	buf[6] = 0;
654 	buf[7] = 0;
655 	atapi_string(buf + 8, "BHYVE", 8);
656 	atapi_string(buf + 16, "BHYVE DVD-ROM", 16);
657 	atapi_string(buf + 32, "001", 4);
658 
659 	len = sizeof(buf);
660 	if (len > acmd[4])
661 		len = acmd[4];
662 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
663 	write_prdt(p, slot, cfis, buf, len);
664 	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
665 }
666 
667 static void
668 atapi_read_capacity(struct ahci_port *p, int slot, uint8_t *cfis)
669 {
670 	uint8_t buf[8];
671 	uint64_t sectors;
672 
673 	sectors = blockif_size(p->bctx) / 2048;
674 	be32enc(buf, sectors - 1);
675 	be32enc(buf + 4, 2048);
676 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
677 	write_prdt(p, slot, cfis, buf, sizeof(buf));
678 	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
679 }
680 
681 static void
682 atapi_read_toc(struct ahci_port *p, int slot, uint8_t *cfis)
683 {
684 	uint8_t *acmd;
685 	uint8_t format;
686 	int len;
687 
688 	acmd = cfis + 0x40;
689 
690 	len = be16dec(acmd + 7);
691 	format = acmd[9] >> 6;
692 	switch (format) {
693 	case 0:
694 	{
695 		int msf, size;
696 		uint64_t sectors;
697 		uint8_t start_track, buf[20], *bp;
698 
699 		msf = (acmd[1] >> 1) & 1;
700 		start_track = acmd[6];
701 		if (start_track > 1 && start_track != 0xaa) {
702 			uint32_t tfd;
703 			p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
704 			p->asc = 0x24;
705 			tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
706 			cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
707 			ahci_write_fis_d2h(p, slot, cfis, tfd);
708 			return;
709 		}
710 		bp = buf + 2;
711 		*bp++ = 1;
712 		*bp++ = 1;
713 		if (start_track <= 1) {
714 			*bp++ = 0;
715 			*bp++ = 0x14;
716 			*bp++ = 1;
717 			*bp++ = 0;
718 			if (msf) {
719 				*bp++ = 0;
720 				lba_to_msf(bp, 0);
721 				bp += 3;
722 			} else {
723 				*bp++ = 0;
724 				*bp++ = 0;
725 				*bp++ = 0;
726 				*bp++ = 0;
727 			}
728 		}
729 		*bp++ = 0;
730 		*bp++ = 0x14;
731 		*bp++ = 0xaa;
732 		*bp++ = 0;
733 		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
734 		sectors >>= 2;
735 		if (msf) {
736 			*bp++ = 0;
737 			lba_to_msf(bp, sectors);
738 			bp += 3;
739 		} else {
740 			be32enc(bp, sectors);
741 			bp += 4;
742 		}
743 		size = bp - buf;
744 		be16enc(buf, size - 2);
745 		if (len > size)
746 			len = size;
747 		write_prdt(p, slot, cfis, buf, len);
748 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
749 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
750 		break;
751 	}
752 	case 1:
753 	{
754 		uint8_t buf[12];
755 
756 		memset(buf, 0, sizeof(buf));
757 		buf[1] = 0xa;
758 		buf[2] = 0x1;
759 		buf[3] = 0x1;
760 		if (len > sizeof(buf))
761 			len = sizeof(buf);
762 		write_prdt(p, slot, cfis, buf, len);
763 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
764 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
765 		break;
766 	}
767 	case 2:
768 	{
769 		int msf, size;
770 		uint64_t sectors;
771 		uint8_t start_track, *bp, buf[50];
772 
773 		msf = (acmd[1] >> 1) & 1;
774 		start_track = acmd[6];
775 		bp = buf + 2;
776 		*bp++ = 1;
777 		*bp++ = 1;
778 
779 		*bp++ = 1;
780 		*bp++ = 0x14;
781 		*bp++ = 0;
782 		*bp++ = 0xa0;
783 		*bp++ = 0;
784 		*bp++ = 0;
785 		*bp++ = 0;
786 		*bp++ = 0;
787 		*bp++ = 1;
788 		*bp++ = 0;
789 		*bp++ = 0;
790 
791 		*bp++ = 1;
792 		*bp++ = 0x14;
793 		*bp++ = 0;
794 		*bp++ = 0xa1;
795 		*bp++ = 0;
796 		*bp++ = 0;
797 		*bp++ = 0;
798 		*bp++ = 0;
799 		*bp++ = 1;
800 		*bp++ = 0;
801 		*bp++ = 0;
802 
803 		*bp++ = 1;
804 		*bp++ = 0x14;
805 		*bp++ = 0;
806 		*bp++ = 0xa2;
807 		*bp++ = 0;
808 		*bp++ = 0;
809 		*bp++ = 0;
810 		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
811 		sectors >>= 2;
812 		if (msf) {
813 			*bp++ = 0;
814 			lba_to_msf(bp, sectors);
815 			bp += 3;
816 		} else {
817 			be32enc(bp, sectors);
818 			bp += 4;
819 		}
820 
821 		*bp++ = 1;
822 		*bp++ = 0x14;
823 		*bp++ = 0;
824 		*bp++ = 1;
825 		*bp++ = 0;
826 		*bp++ = 0;
827 		*bp++ = 0;
828 		if (msf) {
829 			*bp++ = 0;
830 			lba_to_msf(bp, 0);
831 			bp += 3;
832 		} else {
833 			*bp++ = 0;
834 			*bp++ = 0;
835 			*bp++ = 0;
836 			*bp++ = 0;
837 		}
838 
839 		size = bp - buf;
840 		be16enc(buf, size - 2);
841 		if (len > size)
842 			len = size;
843 		write_prdt(p, slot, cfis, buf, len);
844 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
845 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
846 		break;
847 	}
848 	default:
849 	{
850 		uint32_t tfd;
851 
852 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
853 		p->asc = 0x24;
854 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
855 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
856 		ahci_write_fis_d2h(p, slot, cfis, tfd);
857 		break;
858 	}
859 	}
860 }
861 
862 static void
863 atapi_read(struct ahci_port *p, int slot, uint8_t *cfis,
864 		uint32_t done, int seek)
865 {
866 	struct ahci_ioreq *aior;
867 	struct ahci_cmd_hdr *hdr;
868 	struct ahci_prdt_entry *prdt;
869 	struct blockif_req *breq;
870 	struct pci_ahci_softc *sc;
871 	uint8_t *acmd;
872 	uint64_t lba;
873 	uint32_t len;
874 	int i, err, iovcnt;
875 
876 	sc = p->pr_sc;
877 	acmd = cfis + 0x40;
878 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
879 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
880 
881 	prdt += seek;
882 	lba = be32dec(acmd + 2);
883 	if (acmd[0] == READ_10)
884 		len = be16dec(acmd + 7);
885 	else
886 		len = be32dec(acmd + 6);
887 	if (len == 0) {
888 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
889 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
890 	}
891 	lba *= 2048;
892 	len *= 2048;
893 
894 	/*
895 	 * Pull request off free list
896 	 */
897 	aior = STAILQ_FIRST(&p->iofhd);
898 	assert(aior != NULL);
899 	STAILQ_REMOVE_HEAD(&p->iofhd, io_list);
900 	aior->cfis = cfis;
901 	aior->slot = slot;
902 	aior->len = len;
903 	aior->done = done;
904 	breq = &aior->io_req;
905 	breq->br_offset = lba + done;
906 	iovcnt = hdr->prdtl - seek;
907 	if (iovcnt > BLOCKIF_IOV_MAX) {
908 		aior->prdtl = iovcnt - BLOCKIF_IOV_MAX;
909 		iovcnt = BLOCKIF_IOV_MAX;
910 	} else
911 		aior->prdtl = 0;
912 	breq->br_iovcnt = iovcnt;
913 
914 	/*
915 	 * Build up the iovec based on the prdt
916 	 */
917 	for (i = 0; i < iovcnt; i++) {
918 		uint32_t dbcsz;
919 
920 		dbcsz = (prdt->dbc & DBCMASK) + 1;
921 		breq->br_iov[i].iov_base = paddr_guest2host(ahci_ctx(sc),
922 		    prdt->dba, dbcsz);
923 		breq->br_iov[i].iov_len = dbcsz;
924 		aior->done += dbcsz;
925 		prdt++;
926 	}
927 	err = blockif_read(p->bctx, breq);
928 	assert(err == 0);
929 }
930 
931 static void
932 atapi_request_sense(struct ahci_port *p, int slot, uint8_t *cfis)
933 {
934 	uint8_t buf[64];
935 	uint8_t *acmd;
936 	int len;
937 
938 	acmd = cfis + 0x40;
939 	len = acmd[4];
940 	if (len > sizeof(buf))
941 		len = sizeof(buf);
942 	memset(buf, 0, len);
943 	buf[0] = 0x70 | (1 << 7);
944 	buf[2] = p->sense_key;
945 	buf[7] = 10;
946 	buf[12] = p->asc;
947 	write_prdt(p, slot, cfis, buf, len);
948 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
949 	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
950 }
951 
952 static void
953 atapi_start_stop_unit(struct ahci_port *p, int slot, uint8_t *cfis)
954 {
955 	uint8_t *acmd = cfis + 0x40;
956 	uint32_t tfd;
957 
958 	switch (acmd[4] & 3) {
959 	case 0:
960 	case 1:
961 	case 3:
962 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
963 		tfd = ATA_S_READY | ATA_S_DSC;
964 		break;
965 	case 2:
966 		/* TODO eject media */
967 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
968 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
969 		p->asc = 0x53;
970 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
971 		break;
972 	}
973 	ahci_write_fis_d2h(p, slot, cfis, tfd);
974 }
975 
976 static void
977 atapi_mode_sense(struct ahci_port *p, int slot, uint8_t *cfis)
978 {
979 	uint8_t *acmd;
980 	uint32_t tfd;
981 	uint8_t pc, code;
982 	int len;
983 
984 	acmd = cfis + 0x40;
985 	len = be16dec(acmd + 7);
986 	pc = acmd[2] >> 6;
987 	code = acmd[2] & 0x3f;
988 
989 	switch (pc) {
990 	case 0:
991 		switch (code) {
992 		case MODEPAGE_RW_ERROR_RECOVERY:
993 		{
994 			uint8_t buf[16];
995 
996 			if (len > sizeof(buf))
997 				len = sizeof(buf);
998 
999 			memset(buf, 0, sizeof(buf));
1000 			be16enc(buf, 16 - 2);
1001 			buf[2] = 0x70;
1002 			buf[8] = 0x01;
1003 			buf[9] = 16 - 10;
1004 			buf[11] = 0x05;
1005 			write_prdt(p, slot, cfis, buf, len);
1006 			tfd = ATA_S_READY | ATA_S_DSC;
1007 			break;
1008 		}
1009 		case MODEPAGE_CD_CAPABILITIES:
1010 		{
1011 			uint8_t buf[30];
1012 
1013 			if (len > sizeof(buf))
1014 				len = sizeof(buf);
1015 
1016 			memset(buf, 0, sizeof(buf));
1017 			be16enc(buf, 30 - 2);
1018 			buf[2] = 0x70;
1019 			buf[8] = 0x2A;
1020 			buf[9] = 30 - 10;
1021 			buf[10] = 0x08;
1022 			buf[12] = 0x71;
1023 			be16enc(&buf[18], 2);
1024 			be16enc(&buf[20], 512);
1025 			write_prdt(p, slot, cfis, buf, len);
1026 			tfd = ATA_S_READY | ATA_S_DSC;
1027 			break;
1028 		}
1029 		default:
1030 			goto error;
1031 			break;
1032 		}
1033 		break;
1034 	case 3:
1035 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1036 		p->asc = 0x39;
1037 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1038 		break;
1039 error:
1040 	case 1:
1041 	case 2:
1042 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1043 		p->asc = 0x24;
1044 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1045 		break;
1046 	}
1047 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1048 	ahci_write_fis_d2h(p, slot, cfis, tfd);
1049 }
1050 
1051 static void
1052 atapi_get_event_status_notification(struct ahci_port *p, int slot,
1053     uint8_t *cfis)
1054 {
1055 	uint8_t *acmd;
1056 	uint32_t tfd;
1057 
1058 	acmd = cfis + 0x40;
1059 
1060 	/* we don't support asynchronous operation */
1061 	if (!(acmd[1] & 1)) {
1062 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1063 		p->asc = 0x24;
1064 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1065 	} else {
1066 		uint8_t buf[8];
1067 		int len;
1068 
1069 		len = be16dec(acmd + 7);
1070 		if (len > sizeof(buf))
1071 			len = sizeof(buf);
1072 
1073 		memset(buf, 0, sizeof(buf));
1074 		be16enc(buf, 8 - 2);
1075 		buf[2] = 0x04;
1076 		buf[3] = 0x10;
1077 		buf[5] = 0x02;
1078 		write_prdt(p, slot, cfis, buf, len);
1079 		tfd = ATA_S_READY | ATA_S_DSC;
1080 	}
1081 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1082 	ahci_write_fis_d2h(p, slot, cfis, tfd);
1083 }
1084 
1085 static void
1086 handle_packet_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1087 {
1088 	uint8_t *acmd;
1089 
1090 	acmd = cfis + 0x40;
1091 
1092 #ifdef AHCI_DEBUG
1093 	{
1094 		int i;
1095 		DPRINTF("ACMD:");
1096 		for (i = 0; i < 16; i++)
1097 			DPRINTF("%02x ", acmd[i]);
1098 		DPRINTF("\n");
1099 	}
1100 #endif
1101 
1102 	switch (acmd[0]) {
1103 	case TEST_UNIT_READY:
1104 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1105 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1106 		break;
1107 	case INQUIRY:
1108 		atapi_inquiry(p, slot, cfis);
1109 		break;
1110 	case READ_CAPACITY:
1111 		atapi_read_capacity(p, slot, cfis);
1112 		break;
1113 	case PREVENT_ALLOW:
1114 		/* TODO */
1115 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1116 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1117 		break;
1118 	case READ_TOC:
1119 		atapi_read_toc(p, slot, cfis);
1120 		break;
1121 	case READ_10:
1122 	case READ_12:
1123 		atapi_read(p, slot, cfis, 0, 0);
1124 		break;
1125 	case REQUEST_SENSE:
1126 		atapi_request_sense(p, slot, cfis);
1127 		break;
1128 	case START_STOP_UNIT:
1129 		atapi_start_stop_unit(p, slot, cfis);
1130 		break;
1131 	case MODE_SENSE_10:
1132 		atapi_mode_sense(p, slot, cfis);
1133 		break;
1134 	case GET_EVENT_STATUS_NOTIFICATION:
1135 		atapi_get_event_status_notification(p, slot, cfis);
1136 		break;
1137 	default:
1138 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1139 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1140 		p->asc = 0x20;
1141 		ahci_write_fis_d2h(p, slot, cfis, (p->sense_key << 12) |
1142 				ATA_S_READY | ATA_S_ERROR);
1143 		break;
1144 	}
1145 }
1146 
1147 static void
1148 ahci_handle_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1149 {
1150 
1151 	switch (cfis[2]) {
1152 	case ATA_ATA_IDENTIFY:
1153 		handle_identify(p, slot, cfis);
1154 		break;
1155 	case ATA_SETFEATURES:
1156 	{
1157 		switch (cfis[3]) {
1158 		case ATA_SF_ENAB_WCACHE:
1159 		case ATA_SF_DIS_WCACHE:
1160 		case ATA_SF_ENAB_RCACHE:
1161 		case ATA_SF_DIS_RCACHE:
1162 			p->tfd = ATA_S_DSC | ATA_S_READY;
1163 			break;
1164 		case ATA_SF_SETXFER:
1165 		{
1166 			switch (cfis[12] & 0xf8) {
1167 			case ATA_PIO:
1168 			case ATA_PIO0:
1169 				break;
1170 			case ATA_WDMA0:
1171 			case ATA_UDMA0:
1172 				p->xfermode = (cfis[12] & 0x7);
1173 				break;
1174 			}
1175 			p->tfd = ATA_S_DSC | ATA_S_READY;
1176 			break;
1177 		}
1178 		default:
1179 			p->tfd = ATA_S_ERROR | ATA_S_READY;
1180 			p->tfd |= (ATA_ERROR_ABORT << 8);
1181 			break;
1182 		}
1183 		p->is |= AHCI_P_IX_DP;
1184 		p->ci &= ~(1 << slot);
1185 		ahci_generate_intr(p->pr_sc);
1186 		break;
1187 	}
1188 	case ATA_SET_MULTI:
1189 		if (cfis[12] != 0 &&
1190 			(cfis[12] > 128 || (cfis[12] & (cfis[12] - 1)))) {
1191 			p->tfd = ATA_S_ERROR | ATA_S_READY;
1192 			p->tfd |= (ATA_ERROR_ABORT << 8);
1193 		} else {
1194 			p->mult_sectors = cfis[12];
1195 			p->tfd = ATA_S_DSC | ATA_S_READY;
1196 		}
1197 		p->is |= AHCI_P_IX_DP;
1198 		p->ci &= ~(1 << slot);
1199 		ahci_generate_intr(p->pr_sc);
1200 		break;
1201 	case ATA_READ_DMA:
1202 	case ATA_WRITE_DMA:
1203 	case ATA_READ_DMA48:
1204 	case ATA_WRITE_DMA48:
1205 	case ATA_READ_FPDMA_QUEUED:
1206 	case ATA_WRITE_FPDMA_QUEUED:
1207 		ahci_handle_dma(p, slot, cfis, 0, 0);
1208 		break;
1209 	case ATA_FLUSHCACHE:
1210 	case ATA_FLUSHCACHE48:
1211 		ahci_handle_flush(p, slot, cfis);
1212 		break;
1213 	case ATA_STANDBY_CMD:
1214 		break;
1215 	case ATA_NOP:
1216 	case ATA_STANDBY_IMMEDIATE:
1217 	case ATA_IDLE_IMMEDIATE:
1218 	case ATA_SLEEP:
1219 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1220 		break;
1221 	case ATA_ATAPI_IDENTIFY:
1222 		handle_atapi_identify(p, slot, cfis);
1223 		break;
1224 	case ATA_PACKET_CMD:
1225 		if (!p->atapi) {
1226 			p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1227 			p->is |= AHCI_P_IX_TFE;
1228 			p->ci &= ~(1 << slot);
1229 			ahci_generate_intr(p->pr_sc);
1230 		} else
1231 			handle_packet_cmd(p, slot, cfis);
1232 		break;
1233 	default:
1234 		WPRINTF("Unsupported cmd:%02x\n", cfis[2]);
1235 		p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1236 		p->is |= AHCI_P_IX_TFE;
1237 		p->ci &= ~(1 << slot);
1238 		ahci_generate_intr(p->pr_sc);
1239 		break;
1240 	}
1241 }
1242 
1243 static void
1244 ahci_handle_slot(struct ahci_port *p, int slot)
1245 {
1246 	struct ahci_cmd_hdr *hdr;
1247 	struct ahci_prdt_entry *prdt;
1248 	struct pci_ahci_softc *sc;
1249 	uint8_t *cfis;
1250 	int cfl;
1251 
1252 	sc = p->pr_sc;
1253 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1254 	cfl = (hdr->flags & 0x1f) * 4;
1255 	cfis = paddr_guest2host(ahci_ctx(sc), hdr->ctba,
1256 			0x80 + hdr->prdtl * sizeof(struct ahci_prdt_entry));
1257 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
1258 
1259 #ifdef AHCI_DEBUG
1260 	DPRINTF("\ncfis:");
1261 	for (i = 0; i < cfl; i++) {
1262 		if (i % 10 == 0)
1263 			DPRINTF("\n");
1264 		DPRINTF("%02x ", cfis[i]);
1265 	}
1266 	DPRINTF("\n");
1267 
1268 	for (i = 0; i < hdr->prdtl; i++) {
1269 		DPRINTF("%d@%08"PRIx64"\n", prdt->dbc & 0x3fffff, prdt->dba);
1270 		prdt++;
1271 	}
1272 #endif
1273 
1274 	if (cfis[0] != FIS_TYPE_REGH2D) {
1275 		WPRINTF("Not a H2D FIS:%02x\n", cfis[0]);
1276 		return;
1277 	}
1278 
1279 	if (cfis[1] & 0x80) {
1280 		ahci_handle_cmd(p, slot, cfis);
1281 	} else {
1282 		if (cfis[15] & (1 << 2))
1283 			p->reset = 1;
1284 		else if (p->reset) {
1285 			p->reset = 0;
1286 			ahci_port_reset(p);
1287 		}
1288 		p->ci &= ~(1 << slot);
1289 	}
1290 }
1291 
1292 static void
1293 ahci_handle_port(struct ahci_port *p)
1294 {
1295 	int i;
1296 
1297 	if (!(p->cmd & AHCI_P_CMD_ST))
1298 		return;
1299 
1300 	for (i = 0; (i < 32) && p->ci; i++) {
1301 		if (p->ci & (1 << i))
1302 			ahci_handle_slot(p, i);
1303 	}
1304 }
1305 
1306 /*
1307  * blockif callback routine - this runs in the context of the blockif
1308  * i/o thread, so the mutex needs to be acquired.
1309  */
1310 static void
1311 ata_ioreq_cb(struct blockif_req *br, int err)
1312 {
1313 	struct ahci_cmd_hdr *hdr;
1314 	struct ahci_ioreq *aior;
1315 	struct ahci_port *p;
1316 	struct pci_ahci_softc *sc;
1317 	uint32_t tfd;
1318 	uint8_t *cfis;
1319 	int pending, slot, ncq;
1320 
1321 	DPRINTF("%s %d\n", __func__, err);
1322 
1323 	ncq = 0;
1324 	aior = br->br_param;
1325 	p = aior->io_pr;
1326 	cfis = aior->cfis;
1327 	slot = aior->slot;
1328 	pending = aior->prdtl;
1329 	sc = p->pr_sc;
1330 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1331 
1332 	if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
1333 			cfis[2] == ATA_READ_FPDMA_QUEUED)
1334 		ncq = 1;
1335 
1336 	pthread_mutex_lock(&sc->mtx);
1337 
1338 	/*
1339 	 * Move the blockif request back to the free list
1340 	 */
1341 	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_list);
1342 
1343 	if (pending && !err) {
1344 		ahci_handle_dma(p, slot, cfis, aior->done,
1345 		    hdr->prdtl - pending);
1346 		goto out;
1347 	}
1348 
1349 	if (!err && aior->done == aior->len) {
1350 		tfd = ATA_S_READY | ATA_S_DSC;
1351 		if (ncq)
1352 			hdr->prdbc = 0;
1353 		else
1354 			hdr->prdbc = aior->len;
1355 	} else {
1356 		tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1357 		hdr->prdbc = 0;
1358 		if (ncq)
1359 			p->serr |= (1 << slot);
1360 	}
1361 
1362 	if (ncq) {
1363 		p->sact &= ~(1 << slot);
1364 		ahci_write_fis_sdb(p, slot, tfd);
1365 	} else
1366 		ahci_write_fis_d2h(p, slot, cfis, tfd);
1367 
1368 out:
1369 	pthread_mutex_unlock(&sc->mtx);
1370 	DPRINTF("%s exit\n", __func__);
1371 }
1372 
1373 static void
1374 atapi_ioreq_cb(struct blockif_req *br, int err)
1375 {
1376 	struct ahci_cmd_hdr *hdr;
1377 	struct ahci_ioreq *aior;
1378 	struct ahci_port *p;
1379 	struct pci_ahci_softc *sc;
1380 	uint8_t *cfis;
1381 	uint32_t tfd;
1382 	int pending, slot;
1383 
1384 	DPRINTF("%s %d\n", __func__, err);
1385 
1386 	aior = br->br_param;
1387 	p = aior->io_pr;
1388 	cfis = aior->cfis;
1389 	slot = aior->slot;
1390 	pending = aior->prdtl;
1391 	sc = p->pr_sc;
1392 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + aior->slot * AHCI_CL_SIZE);
1393 
1394 	pthread_mutex_lock(&sc->mtx);
1395 
1396 	/*
1397 	 * Move the blockif request back to the free list
1398 	 */
1399 	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_list);
1400 
1401 	if (pending && !err) {
1402 		atapi_read(p, slot, cfis, aior->done, hdr->prdtl - pending);
1403 		goto out;
1404 	}
1405 
1406 	if (!err && aior->done == aior->len) {
1407 		tfd = ATA_S_READY | ATA_S_DSC;
1408 		hdr->prdbc = aior->len;
1409 	} else {
1410 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1411 		p->asc = 0x21;
1412 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1413 		hdr->prdbc = 0;
1414 	}
1415 
1416 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1417 	ahci_write_fis_d2h(p, slot, cfis, tfd);
1418 
1419 out:
1420 	pthread_mutex_unlock(&sc->mtx);
1421 	DPRINTF("%s exit\n", __func__);
1422 }
1423 
1424 static void
1425 pci_ahci_ioreq_init(struct ahci_port *pr)
1426 {
1427 	struct ahci_ioreq *vr;
1428 	int i;
1429 
1430 	pr->ioqsz = blockif_queuesz(pr->bctx);
1431 	pr->ioreq = calloc(pr->ioqsz, sizeof(struct ahci_ioreq));
1432 	STAILQ_INIT(&pr->iofhd);
1433 
1434 	/*
1435 	 * Add all i/o request entries to the free queue
1436 	 */
1437 	for (i = 0; i < pr->ioqsz; i++) {
1438 		vr = &pr->ioreq[i];
1439 		vr->io_pr = pr;
1440 		if (!pr->atapi)
1441 			vr->io_req.br_callback = ata_ioreq_cb;
1442 		else
1443 			vr->io_req.br_callback = atapi_ioreq_cb;
1444 		vr->io_req.br_param = vr;
1445 		STAILQ_INSERT_TAIL(&pr->iofhd, vr, io_list);
1446 	}
1447 }
1448 
1449 static void
1450 pci_ahci_port_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
1451 {
1452 	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
1453 	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
1454 	struct ahci_port *p = &sc->port[port];
1455 
1456 	DPRINTF("pci_ahci_port %d: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
1457 		port, offset, value);
1458 
1459 	switch (offset) {
1460 	case AHCI_P_CLB:
1461 		p->clb = value;
1462 		break;
1463 	case AHCI_P_CLBU:
1464 		p->clbu = value;
1465 		break;
1466 	case AHCI_P_FB:
1467 		p->fb = value;
1468 		break;
1469 	case AHCI_P_FBU:
1470 		p->fbu = value;
1471 		break;
1472 	case AHCI_P_IS:
1473 		p->is &= ~value;
1474 		break;
1475 	case AHCI_P_IE:
1476 		p->ie = value & 0xFDC000FF;
1477 		ahci_generate_intr(sc);
1478 		break;
1479 	case AHCI_P_CMD:
1480 	{
1481 		p->cmd = value;
1482 
1483 		if (!(value & AHCI_P_CMD_ST)) {
1484 			p->cmd &= ~(AHCI_P_CMD_CR | AHCI_P_CMD_CCS_MASK);
1485 			p->ci = 0;
1486 			p->sact = 0;
1487 		} else {
1488 			uint64_t clb;
1489 
1490 			p->cmd |= AHCI_P_CMD_CR;
1491 			clb = (uint64_t)p->clbu << 32 | p->clb;
1492 			p->cmd_lst = paddr_guest2host(ahci_ctx(sc), clb,
1493 					AHCI_CL_SIZE * AHCI_MAX_SLOTS);
1494 		}
1495 
1496 		if (value & AHCI_P_CMD_FRE) {
1497 			uint64_t fb;
1498 
1499 			p->cmd |= AHCI_P_CMD_FR;
1500 			fb = (uint64_t)p->fbu << 32 | p->fb;
1501 			/* we don't support FBSCP, so rfis size is 256Bytes */
1502 			p->rfis = paddr_guest2host(ahci_ctx(sc), fb, 256);
1503 		} else {
1504 			p->cmd &= ~AHCI_P_CMD_FR;
1505 		}
1506 
1507 		if (value & AHCI_P_CMD_CLO) {
1508 			p->tfd = 0;
1509 			p->cmd &= ~AHCI_P_CMD_CLO;
1510 		}
1511 
1512 		ahci_handle_port(p);
1513 		break;
1514 	}
1515 	case AHCI_P_TFD:
1516 	case AHCI_P_SIG:
1517 	case AHCI_P_SSTS:
1518 		WPRINTF("pci_ahci_port: read only registers 0x%"PRIx64"\n", offset);
1519 		break;
1520 	case AHCI_P_SCTL:
1521 		if (!(p->cmd & AHCI_P_CMD_ST)) {
1522 			if (value & ATA_SC_DET_RESET)
1523 				ahci_port_reset(p);
1524 			p->sctl = value;
1525 		}
1526 		break;
1527 	case AHCI_P_SERR:
1528 		p->serr &= ~value;
1529 		break;
1530 	case AHCI_P_SACT:
1531 		p->sact |= value;
1532 		break;
1533 	case AHCI_P_CI:
1534 		p->ci |= value;
1535 		ahci_handle_port(p);
1536 		break;
1537 	case AHCI_P_SNTF:
1538 	case AHCI_P_FBS:
1539 	default:
1540 		break;
1541 	}
1542 }
1543 
1544 static void
1545 pci_ahci_host_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
1546 {
1547 	DPRINTF("pci_ahci_host: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
1548 		offset, value);
1549 
1550 	switch (offset) {
1551 	case AHCI_CAP:
1552 	case AHCI_PI:
1553 	case AHCI_VS:
1554 	case AHCI_CAP2:
1555 		DPRINTF("pci_ahci_host: read only registers 0x%"PRIx64"\n", offset);
1556 		break;
1557 	case AHCI_GHC:
1558 		if (value & AHCI_GHC_HR)
1559 			ahci_reset(sc);
1560 		else if (value & AHCI_GHC_IE) {
1561 			sc->ghc |= AHCI_GHC_IE;
1562 			ahci_generate_intr(sc);
1563 		}
1564 		break;
1565 	case AHCI_IS:
1566 		sc->is &= ~value;
1567 		ahci_generate_intr(sc);
1568 		break;
1569 	default:
1570 		break;
1571 	}
1572 }
1573 
1574 static void
1575 pci_ahci_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
1576 		int baridx, uint64_t offset, int size, uint64_t value)
1577 {
1578 	struct pci_ahci_softc *sc = pi->pi_arg;
1579 
1580 	assert(baridx == 5);
1581 	assert(size == 4);
1582 
1583 	pthread_mutex_lock(&sc->mtx);
1584 
1585 	if (offset < AHCI_OFFSET)
1586 		pci_ahci_host_write(sc, offset, value);
1587 	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
1588 		pci_ahci_port_write(sc, offset, value);
1589 	else
1590 		WPRINTF("pci_ahci: unknown i/o write offset 0x%"PRIx64"\n", offset);
1591 
1592 	pthread_mutex_unlock(&sc->mtx);
1593 }
1594 
1595 static uint64_t
1596 pci_ahci_host_read(struct pci_ahci_softc *sc, uint64_t offset)
1597 {
1598 	uint32_t value;
1599 
1600 	switch (offset) {
1601 	case AHCI_CAP:
1602 	case AHCI_GHC:
1603 	case AHCI_IS:
1604 	case AHCI_PI:
1605 	case AHCI_VS:
1606 	case AHCI_CCCC:
1607 	case AHCI_CCCP:
1608 	case AHCI_EM_LOC:
1609 	case AHCI_EM_CTL:
1610 	case AHCI_CAP2:
1611 	{
1612 		uint32_t *p = &sc->cap;
1613 		p += (offset - AHCI_CAP) / sizeof(uint32_t);
1614 		value = *p;
1615 		break;
1616 	}
1617 	default:
1618 		value = 0;
1619 		break;
1620 	}
1621 	DPRINTF("pci_ahci_host: read offset 0x%"PRIx64" value 0x%x\n",
1622 		offset, value);
1623 
1624 	return (value);
1625 }
1626 
1627 static uint64_t
1628 pci_ahci_port_read(struct pci_ahci_softc *sc, uint64_t offset)
1629 {
1630 	uint32_t value;
1631 	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
1632 	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
1633 
1634 	switch (offset) {
1635 	case AHCI_P_CLB:
1636 	case AHCI_P_CLBU:
1637 	case AHCI_P_FB:
1638 	case AHCI_P_FBU:
1639 	case AHCI_P_IS:
1640 	case AHCI_P_IE:
1641 	case AHCI_P_CMD:
1642 	case AHCI_P_TFD:
1643 	case AHCI_P_SIG:
1644 	case AHCI_P_SSTS:
1645 	case AHCI_P_SCTL:
1646 	case AHCI_P_SERR:
1647 	case AHCI_P_SACT:
1648 	case AHCI_P_CI:
1649 	case AHCI_P_SNTF:
1650 	case AHCI_P_FBS:
1651 	{
1652 		uint32_t *p= &sc->port[port].clb;
1653 		p += (offset - AHCI_P_CLB) / sizeof(uint32_t);
1654 		value = *p;
1655 		break;
1656 	}
1657 	default:
1658 		value = 0;
1659 		break;
1660 	}
1661 
1662 	DPRINTF("pci_ahci_port %d: read offset 0x%"PRIx64" value 0x%x\n",
1663 		port, offset, value);
1664 
1665 	return value;
1666 }
1667 
1668 static uint64_t
1669 pci_ahci_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1670     uint64_t offset, int size)
1671 {
1672 	struct pci_ahci_softc *sc = pi->pi_arg;
1673 	uint32_t value;
1674 
1675 	assert(baridx == 5);
1676 	assert(size == 4);
1677 
1678 	pthread_mutex_lock(&sc->mtx);
1679 
1680 	if (offset < AHCI_OFFSET)
1681 		value = pci_ahci_host_read(sc, offset);
1682 	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
1683 		value = pci_ahci_port_read(sc, offset);
1684 	else {
1685 		value = 0;
1686 		WPRINTF("pci_ahci: unknown i/o read offset 0x%"PRIx64"\n", offset);
1687 	}
1688 
1689 	pthread_mutex_unlock(&sc->mtx);
1690 
1691 	return (value);
1692 }
1693 
1694 static int
1695 pci_ahci_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts, int atapi)
1696 {
1697 	char bident[sizeof("XX:X:X")];
1698 	struct blockif_ctxt *bctxt;
1699 	struct pci_ahci_softc *sc;
1700 	int ret, slots;
1701 
1702 	ret = 0;
1703 
1704 	if (opts == NULL) {
1705 		fprintf(stderr, "pci_ahci: backing device required\n");
1706 		return (1);
1707 	}
1708 
1709 #ifdef AHCI_DEBUG
1710 	dbg = fopen("/tmp/log", "w+");
1711 #endif
1712 
1713        	sc = malloc(sizeof(struct pci_ahci_softc));
1714 	memset(sc, 0, sizeof(struct pci_ahci_softc));
1715 	pi->pi_arg = sc;
1716 	sc->asc_pi = pi;
1717 	sc->ports = MAX_PORTS;
1718 
1719 	/*
1720 	 * Only use port 0 for a backing device. All other ports will be
1721 	 * marked as unused
1722 	 */
1723 	sc->port[0].atapi = atapi;
1724 
1725 	/*
1726 	 * Attempt to open the backing image. Use the PCI
1727 	 * slot/func for the identifier string.
1728 	 */
1729 	snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func);
1730 	bctxt = blockif_open(opts, bident);
1731 	if (bctxt == NULL) {
1732 		ret = 1;
1733 		goto open_fail;
1734 	}
1735 	sc->port[0].bctx = bctxt;
1736 	sc->port[0].pr_sc = sc;
1737 
1738 	/*
1739 	 * Allocate blockif request structures and add them
1740 	 * to the free list
1741 	 */
1742 	pci_ahci_ioreq_init(&sc->port[0]);
1743 
1744 	pthread_mutex_init(&sc->mtx, NULL);
1745 
1746 	/* Intel ICH8 AHCI */
1747 	slots = sc->port[0].ioqsz;
1748 	if (slots > 32)
1749 		slots = 32;
1750 	--slots;
1751 	sc->cap = AHCI_CAP_64BIT | AHCI_CAP_SNCQ | AHCI_CAP_SSNTF |
1752 	    AHCI_CAP_SMPS | AHCI_CAP_SSS | AHCI_CAP_SALP |
1753 	    AHCI_CAP_SAL | AHCI_CAP_SCLO | (0x3 << AHCI_CAP_ISS_SHIFT)|
1754 	    AHCI_CAP_PMD | AHCI_CAP_SSC | AHCI_CAP_PSC |
1755 	    (slots << AHCI_CAP_NCS_SHIFT) | AHCI_CAP_SXS | (sc->ports - 1);
1756 
1757 	/* Only port 0 implemented */
1758 	sc->pi = 1;
1759 	sc->vs = 0x10300;
1760 	sc->cap2 = AHCI_CAP2_APST;
1761 	ahci_reset(sc);
1762 
1763 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x2821);
1764 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x8086);
1765 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
1766 	pci_set_cfgdata8(pi, PCIR_SUBCLASS, PCIS_STORAGE_SATA);
1767 	pci_set_cfgdata8(pi, PCIR_PROGIF, PCIP_STORAGE_SATA_AHCI_1_0);
1768 	pci_emul_add_msicap(pi, 1);
1769 	pci_emul_alloc_bar(pi, 5, PCIBAR_MEM32,
1770 	    AHCI_OFFSET + sc->ports * AHCI_STEP);
1771 
1772 open_fail:
1773 	if (ret) {
1774 		blockif_close(sc->port[0].bctx);
1775 		free(sc);
1776 	}
1777 
1778 	return (ret);
1779 }
1780 
1781 static int
1782 pci_ahci_hd_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1783 {
1784 
1785 	return (pci_ahci_init(ctx, pi, opts, 0));
1786 }
1787 
1788 static int
1789 pci_ahci_atapi_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1790 {
1791 
1792 	return (pci_ahci_init(ctx, pi, opts, 1));
1793 }
1794 
1795 /*
1796  * Use separate emulation names to distinguish drive and atapi devices
1797  */
1798 struct pci_devemu pci_de_ahci_hd = {
1799 	.pe_emu =	"ahci-hd",
1800 	.pe_init =	pci_ahci_hd_init,
1801 	.pe_barwrite =	pci_ahci_write,
1802 	.pe_barread =	pci_ahci_read
1803 };
1804 PCI_EMUL_SET(pci_de_ahci_hd);
1805 
1806 struct pci_devemu pci_de_ahci_cd = {
1807 	.pe_emu =	"ahci-cd",
1808 	.pe_init =	pci_ahci_atapi_init,
1809 	.pe_barwrite =	pci_ahci_write,
1810 	.pe_barread =	pci_ahci_read
1811 };
1812 PCI_EMUL_SET(pci_de_ahci_cd);
1813