xref: /freebsd/usr.sbin/bhyve/pci_ahci.c (revision 0572ccaa4543b0abef8ef81e384c1d04de9f3da1)
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  * ATA commands
99  */
100 #define	ATA_SF_ENAB_SATA_SF		0x10
101 #define		ATA_SATA_SF_AN		0x05
102 #define	ATA_SF_DIS_SATA_SF		0x90
103 
104 /*
105  * Debug printf
106  */
107 #ifdef AHCI_DEBUG
108 static FILE *dbg;
109 #define DPRINTF(format, arg...)	do{fprintf(dbg, format, ##arg);fflush(dbg);}while(0)
110 #else
111 #define DPRINTF(format, arg...)
112 #endif
113 #define WPRINTF(format, arg...) printf(format, ##arg)
114 
115 struct ahci_ioreq {
116 	struct blockif_req io_req;
117 	struct ahci_port *io_pr;
118 	STAILQ_ENTRY(ahci_ioreq) io_list;
119 	uint8_t *cfis;
120 	uint32_t len;
121 	uint32_t done;
122 	int slot;
123 	int prdtl;
124 };
125 
126 struct ahci_port {
127 	struct blockif_ctxt *bctx;
128 	struct pci_ahci_softc *pr_sc;
129 	uint8_t *cmd_lst;
130 	uint8_t *rfis;
131 	int atapi;
132 	int reset;
133 	int mult_sectors;
134 	uint8_t xfermode;
135 	uint8_t sense_key;
136 	uint8_t asc;
137 	uint32_t pending;
138 
139 	uint32_t clb;
140 	uint32_t clbu;
141 	uint32_t fb;
142 	uint32_t fbu;
143 	uint32_t is;
144 	uint32_t ie;
145 	uint32_t cmd;
146 	uint32_t unused0;
147 	uint32_t tfd;
148 	uint32_t sig;
149 	uint32_t ssts;
150 	uint32_t sctl;
151 	uint32_t serr;
152 	uint32_t sact;
153 	uint32_t ci;
154 	uint32_t sntf;
155 	uint32_t fbs;
156 
157 	/*
158 	 * i/o request info
159 	 */
160 	struct ahci_ioreq *ioreq;
161 	int ioqsz;
162 	STAILQ_HEAD(ahci_fhead, ahci_ioreq) iofhd;
163 };
164 
165 struct ahci_cmd_hdr {
166 	uint16_t flags;
167 	uint16_t prdtl;
168 	uint32_t prdbc;
169 	uint64_t ctba;
170 	uint32_t reserved[4];
171 };
172 
173 struct ahci_prdt_entry {
174 	uint64_t dba;
175 	uint32_t reserved;
176 #define	DBCMASK		0x3fffff
177 	uint32_t dbc;
178 };
179 
180 struct pci_ahci_softc {
181 	struct pci_devinst *asc_pi;
182 	pthread_mutex_t	mtx;
183 	int ports;
184 	uint32_t cap;
185 	uint32_t ghc;
186 	uint32_t is;
187 	uint32_t pi;
188 	uint32_t vs;
189 	uint32_t ccc_ctl;
190 	uint32_t ccc_pts;
191 	uint32_t em_loc;
192 	uint32_t em_ctl;
193 	uint32_t cap2;
194 	uint32_t bohc;
195 	uint32_t lintr;
196 	struct ahci_port port[MAX_PORTS];
197 };
198 #define	ahci_ctx(sc)	((sc)->asc_pi->pi_vmctx)
199 
200 static inline void lba_to_msf(uint8_t *buf, int lba)
201 {
202 	lba += 150;
203 	buf[0] = (lba / 75) / 60;
204 	buf[1] = (lba / 75) % 60;
205 	buf[2] = lba % 75;
206 }
207 
208 /*
209  * generate HBA intr depending on whether or not ports within
210  * the controller have an interrupt pending.
211  */
212 static void
213 ahci_generate_intr(struct pci_ahci_softc *sc)
214 {
215 	struct pci_devinst *pi;
216 	int i;
217 
218 	pi = sc->asc_pi;
219 
220 	for (i = 0; i < sc->ports; i++) {
221 		struct ahci_port *pr;
222 		pr = &sc->port[i];
223 		if (pr->is & pr->ie)
224 			sc->is |= (1 << i);
225 	}
226 
227 	DPRINTF("%s %x\n", __func__, sc->is);
228 
229 	if (sc->is && (sc->ghc & AHCI_GHC_IE)) {
230 		if (pci_msi_enabled(pi)) {
231 			/*
232 			 * Generate an MSI interrupt on every edge
233 			 */
234 			pci_generate_msi(pi, 0);
235 		} else if (!sc->lintr) {
236 			/*
237 			 * Only generate a pin-based interrupt if one wasn't
238 			 * in progress
239 			 */
240 			sc->lintr = 1;
241 			pci_lintr_assert(pi);
242 		}
243 	} else if (sc->lintr) {
244 		/*
245 		 * No interrupts: deassert pin-based signal if it had
246 		 * been asserted
247 		 */
248 		pci_lintr_deassert(pi);
249 		sc->lintr = 0;
250 	}
251 }
252 
253 static void
254 ahci_write_fis(struct ahci_port *p, enum sata_fis_type ft, uint8_t *fis)
255 {
256 	int offset, len, irq;
257 
258 	if (p->rfis == NULL || !(p->cmd & AHCI_P_CMD_FRE))
259 		return;
260 
261 	switch (ft) {
262 	case FIS_TYPE_REGD2H:
263 		offset = 0x40;
264 		len = 20;
265 		irq = AHCI_P_IX_DHR;
266 		break;
267 	case FIS_TYPE_SETDEVBITS:
268 		offset = 0x58;
269 		len = 8;
270 		irq = AHCI_P_IX_SDB;
271 		break;
272 	case FIS_TYPE_PIOSETUP:
273 		offset = 0x20;
274 		len = 20;
275 		irq = 0;
276 		break;
277 	default:
278 		WPRINTF("unsupported fis type %d\n", ft);
279 		return;
280 	}
281 	memcpy(p->rfis + offset, fis, len);
282 	if (irq) {
283 		p->is |= irq;
284 		ahci_generate_intr(p->pr_sc);
285 	}
286 }
287 
288 static void
289 ahci_write_fis_piosetup(struct ahci_port *p)
290 {
291 	uint8_t fis[20];
292 
293 	memset(fis, 0, sizeof(fis));
294 	fis[0] = FIS_TYPE_PIOSETUP;
295 	ahci_write_fis(p, FIS_TYPE_PIOSETUP, fis);
296 }
297 
298 static void
299 ahci_write_fis_sdb(struct ahci_port *p, int slot, uint32_t tfd)
300 {
301 	uint8_t fis[8];
302 	uint8_t error;
303 
304 	error = (tfd >> 8) & 0xff;
305 	memset(fis, 0, sizeof(fis));
306 	fis[0] = error;
307 	fis[2] = tfd & 0x77;
308 	*(uint32_t *)(fis + 4) = (1 << slot);
309 	if (fis[2] & ATA_S_ERROR)
310 		p->is |= AHCI_P_IX_TFE;
311 	p->tfd = tfd;
312 	ahci_write_fis(p, FIS_TYPE_SETDEVBITS, fis);
313 }
314 
315 static void
316 ahci_write_fis_d2h(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t tfd)
317 {
318 	uint8_t fis[20];
319 	uint8_t error;
320 
321 	error = (tfd >> 8) & 0xff;
322 	memset(fis, 0, sizeof(fis));
323 	fis[0] = FIS_TYPE_REGD2H;
324 	fis[1] = (1 << 6);
325 	fis[2] = tfd & 0xff;
326 	fis[3] = error;
327 	fis[4] = cfis[4];
328 	fis[5] = cfis[5];
329 	fis[6] = cfis[6];
330 	fis[7] = cfis[7];
331 	fis[8] = cfis[8];
332 	fis[9] = cfis[9];
333 	fis[10] = cfis[10];
334 	fis[11] = cfis[11];
335 	fis[12] = cfis[12];
336 	fis[13] = cfis[13];
337 	if (fis[2] & ATA_S_ERROR)
338 		p->is |= AHCI_P_IX_TFE;
339 	p->tfd = tfd;
340 	p->ci &= ~(1 << slot);
341 	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
342 }
343 
344 static void
345 ahci_write_reset_fis_d2h(struct ahci_port *p)
346 {
347 	uint8_t fis[20];
348 
349 	memset(fis, 0, sizeof(fis));
350 	fis[0] = FIS_TYPE_REGD2H;
351 	fis[3] = 1;
352 	fis[4] = 1;
353 	if (p->atapi) {
354 		fis[5] = 0x14;
355 		fis[6] = 0xeb;
356 	}
357 	fis[12] = 1;
358 	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
359 }
360 
361 static void
362 ahci_port_reset(struct ahci_port *pr)
363 {
364 	pr->sctl = 0;
365 	pr->serr = 0;
366 	pr->sact = 0;
367 	pr->xfermode = ATA_UDMA6;
368 	pr->mult_sectors = 128;
369 
370 	if (!pr->bctx) {
371 		pr->ssts = ATA_SS_DET_NO_DEVICE;
372 		pr->sig = 0xFFFFFFFF;
373 		pr->tfd = 0x7F;
374 		return;
375 	}
376 	pr->ssts = ATA_SS_DET_PHY_ONLINE | ATA_SS_SPD_GEN2 |
377 		ATA_SS_IPM_ACTIVE;
378 	pr->tfd = (1 << 8) | ATA_S_DSC | ATA_S_DMA;
379 	if (!pr->atapi) {
380 		pr->sig = PxSIG_ATA;
381 		pr->tfd |= ATA_S_READY;
382 	} else
383 		pr->sig = PxSIG_ATAPI;
384 	ahci_write_reset_fis_d2h(pr);
385 }
386 
387 static void
388 ahci_reset(struct pci_ahci_softc *sc)
389 {
390 	int i;
391 
392 	sc->ghc = AHCI_GHC_AE;
393 	sc->is = 0;
394 
395 	if (sc->lintr) {
396 		pci_lintr_deassert(sc->asc_pi);
397 		sc->lintr = 0;
398 	}
399 
400 	for (i = 0; i < sc->ports; i++) {
401 		sc->port[i].ie = 0;
402 		sc->port[i].is = 0;
403 		ahci_port_reset(&sc->port[i]);
404 	}
405 }
406 
407 static void
408 ata_string(uint8_t *dest, const char *src, int len)
409 {
410 	int i;
411 
412 	for (i = 0; i < len; i++) {
413 		if (*src)
414 			dest[i ^ 1] = *src++;
415 		else
416 			dest[i ^ 1] = ' ';
417 	}
418 }
419 
420 static void
421 atapi_string(uint8_t *dest, const char *src, int len)
422 {
423 	int i;
424 
425 	for (i = 0; i < len; i++) {
426 		if (*src)
427 			dest[i] = *src++;
428 		else
429 			dest[i] = ' ';
430 	}
431 }
432 
433 static void
434 ahci_handle_dma(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t done,
435     int seek)
436 {
437 	struct ahci_ioreq *aior;
438 	struct blockif_req *breq;
439 	struct pci_ahci_softc *sc;
440 	struct ahci_prdt_entry *prdt;
441 	struct ahci_cmd_hdr *hdr;
442 	uint64_t lba;
443 	uint32_t len;
444 	int i, err, iovcnt, ncq, readop;
445 
446 	sc = p->pr_sc;
447 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
448 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
449 	ncq = 0;
450 	readop = 1;
451 
452 	prdt += seek;
453 	if (cfis[2] == ATA_WRITE_DMA || cfis[2] == ATA_WRITE_DMA48 ||
454 			cfis[2] == ATA_WRITE_FPDMA_QUEUED)
455 		readop = 0;
456 
457 	if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
458 			cfis[2] == ATA_READ_FPDMA_QUEUED) {
459 		lba = ((uint64_t)cfis[10] << 40) |
460 			((uint64_t)cfis[9] << 32) |
461 			((uint64_t)cfis[8] << 24) |
462 			((uint64_t)cfis[6] << 16) |
463 			((uint64_t)cfis[5] << 8) |
464 			cfis[4];
465 		len = cfis[11] << 8 | cfis[3];
466 		if (!len)
467 			len = 65536;
468 		ncq = 1;
469 	} else if (cfis[2] == ATA_READ_DMA48 || cfis[2] == ATA_WRITE_DMA48) {
470 		lba = ((uint64_t)cfis[10] << 40) |
471 			((uint64_t)cfis[9] << 32) |
472 			((uint64_t)cfis[8] << 24) |
473 			((uint64_t)cfis[6] << 16) |
474 			((uint64_t)cfis[5] << 8) |
475 			cfis[4];
476 		len = cfis[13] << 8 | cfis[12];
477 		if (!len)
478 			len = 65536;
479 	} else {
480 		lba = ((cfis[7] & 0xf) << 24) | (cfis[6] << 16) |
481 			(cfis[5] << 8) | cfis[4];
482 		len = cfis[12];
483 		if (!len)
484 			len = 256;
485 	}
486 	lba *= blockif_sectsz(p->bctx);
487 	len *= blockif_sectsz(p->bctx);
488 
489 	/*
490 	 * Pull request off free list
491 	 */
492 	aior = STAILQ_FIRST(&p->iofhd);
493 	assert(aior != NULL);
494 	STAILQ_REMOVE_HEAD(&p->iofhd, io_list);
495 	aior->cfis = cfis;
496 	aior->slot = slot;
497 	aior->len = len;
498 	aior->done = done;
499 	breq = &aior->io_req;
500 	breq->br_offset = lba + done;
501 	iovcnt = hdr->prdtl - seek;
502 	if (iovcnt > BLOCKIF_IOV_MAX) {
503 		aior->prdtl = iovcnt - BLOCKIF_IOV_MAX;
504 		iovcnt = BLOCKIF_IOV_MAX;
505 		/*
506 		 * Mark this command in-flight.
507 		 */
508 		p->pending |= 1 << slot;
509 	} else
510 		aior->prdtl = 0;
511 	breq->br_iovcnt = iovcnt;
512 
513 	/*
514 	 * Build up the iovec based on the prdt
515 	 */
516 	for (i = 0; i < iovcnt; i++) {
517 		uint32_t dbcsz;
518 
519 		dbcsz = (prdt->dbc & DBCMASK) + 1;
520 		breq->br_iov[i].iov_base = paddr_guest2host(ahci_ctx(sc),
521 		    prdt->dba, dbcsz);
522 		breq->br_iov[i].iov_len = dbcsz;
523 		aior->done += dbcsz;
524 		prdt++;
525 	}
526 	if (readop)
527 		err = blockif_read(p->bctx, breq);
528 	else
529 		err = blockif_write(p->bctx, breq);
530 	assert(err == 0);
531 
532 	if (ncq)
533 		p->ci &= ~(1 << slot);
534 }
535 
536 static void
537 ahci_handle_flush(struct ahci_port *p, int slot, uint8_t *cfis)
538 {
539 	struct ahci_ioreq *aior;
540 	struct blockif_req *breq;
541 	int err;
542 
543 	/*
544 	 * Pull request off free list
545 	 */
546 	aior = STAILQ_FIRST(&p->iofhd);
547 	assert(aior != NULL);
548 	STAILQ_REMOVE_HEAD(&p->iofhd, io_list);
549 	aior->cfis = cfis;
550 	aior->slot = slot;
551 	aior->len = 0;
552 	aior->done = 0;
553 	aior->prdtl = 0;
554 	breq = &aior->io_req;
555 
556 	err = blockif_flush(p->bctx, breq);
557 	assert(err == 0);
558 }
559 
560 static inline void
561 write_prdt(struct ahci_port *p, int slot, uint8_t *cfis,
562 		void *buf, int size)
563 {
564 	struct ahci_cmd_hdr *hdr;
565 	struct ahci_prdt_entry *prdt;
566 	void *from;
567 	int i, len;
568 
569 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
570 	len = size;
571 	from = buf;
572 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
573 	for (i = 0; i < hdr->prdtl && len; i++) {
574 		uint8_t *ptr;
575 		uint32_t dbcsz;
576 		int sublen;
577 
578 		dbcsz = (prdt->dbc & DBCMASK) + 1;
579 		ptr = paddr_guest2host(ahci_ctx(p->pr_sc), prdt->dba, dbcsz);
580 		sublen = len < dbcsz ? len : dbcsz;
581 		memcpy(ptr, from, sublen);
582 		len -= sublen;
583 		from += sublen;
584 		prdt++;
585 	}
586 	hdr->prdbc = size - len;
587 }
588 
589 static void
590 handle_identify(struct ahci_port *p, int slot, uint8_t *cfis)
591 {
592 	struct ahci_cmd_hdr *hdr;
593 
594 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
595 	if (p->atapi || hdr->prdtl == 0) {
596 		p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
597 		p->is |= AHCI_P_IX_TFE;
598 	} else {
599 		uint16_t buf[256];
600 		uint64_t sectors;
601 
602 		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
603 		memset(buf, 0, sizeof(buf));
604 		buf[0] = 0x0040;
605 		/* TODO emulate different serial? */
606 		ata_string((uint8_t *)(buf+10), "123456", 20);
607 		ata_string((uint8_t *)(buf+23), "001", 8);
608 		ata_string((uint8_t *)(buf+27), "BHYVE SATA DISK", 40);
609 		buf[47] = (0x8000 | 128);
610 		buf[48] = 0x1;
611 		buf[49] = (1 << 8 | 1 << 9 | 1 << 11);
612 		buf[50] = (1 << 14);
613 		buf[53] = (1 << 1 | 1 << 2);
614 		if (p->mult_sectors)
615 			buf[59] = (0x100 | p->mult_sectors);
616 		buf[60] = sectors;
617 		buf[61] = (sectors >> 16);
618 		buf[63] = 0x7;
619 		if (p->xfermode & ATA_WDMA0)
620 			buf[63] |= (1 << ((p->xfermode & 7) + 8));
621 		buf[64] = 0x3;
622 		buf[65] = 100;
623 		buf[66] = 100;
624 		buf[67] = 100;
625 		buf[68] = 100;
626 		buf[75] = 31;
627 		buf[76] = (1 << 8 | 1 << 2);
628 		buf[80] = 0x1f0;
629 		buf[81] = 0x28;
630 		buf[82] = (1 << 5 | 1 << 14);
631 		buf[83] = (1 << 10 | 1 << 12 | 1 << 13 | 1 << 14);
632 		buf[84] = (1 << 14);
633 		buf[85] = (1 << 5 | 1 << 14);
634 		buf[86] = (1 << 10 | 1 << 12 | 1 << 13);
635 		buf[87] = (1 << 14);
636 		buf[88] = 0x7f;
637 		if (p->xfermode & ATA_UDMA0)
638 			buf[88] |= (1 << ((p->xfermode & 7) + 8));
639 		buf[93] = (1 | 1 <<14);
640 		buf[100] = sectors;
641 		buf[101] = (sectors >> 16);
642 		buf[102] = (sectors >> 32);
643 		buf[103] = (sectors >> 48);
644 		ahci_write_fis_piosetup(p);
645 		write_prdt(p, slot, cfis, (void *)buf, sizeof(buf));
646 		p->tfd = ATA_S_DSC | ATA_S_READY;
647 		p->is |= AHCI_P_IX_DP;
648 	}
649 	p->ci &= ~(1 << slot);
650 	ahci_generate_intr(p->pr_sc);
651 }
652 
653 static void
654 handle_atapi_identify(struct ahci_port *p, int slot, uint8_t *cfis)
655 {
656 	if (!p->atapi) {
657 		p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
658 		p->is |= AHCI_P_IX_TFE;
659 	} else {
660 		uint16_t buf[256];
661 
662 		memset(buf, 0, sizeof(buf));
663 		buf[0] = (2 << 14 | 5 << 8 | 1 << 7 | 2 << 5);
664 		/* TODO emulate different serial? */
665 		ata_string((uint8_t *)(buf+10), "123456", 20);
666 		ata_string((uint8_t *)(buf+23), "001", 8);
667 		ata_string((uint8_t *)(buf+27), "BHYVE SATA DVD ROM", 40);
668 		buf[49] = (1 << 9 | 1 << 8);
669 		buf[50] = (1 << 14 | 1);
670 		buf[53] = (1 << 2 | 1 << 1);
671 		buf[62] = 0x3f;
672 		buf[63] = 7;
673 		buf[64] = 3;
674 		buf[65] = 100;
675 		buf[66] = 100;
676 		buf[67] = 100;
677 		buf[68] = 100;
678 		buf[76] = (1 << 2 | 1 << 1);
679 		buf[78] = (1 << 5);
680 		buf[80] = (0x1f << 4);
681 		buf[82] = (1 << 4);
682 		buf[83] = (1 << 14);
683 		buf[84] = (1 << 14);
684 		buf[85] = (1 << 4);
685 		buf[87] = (1 << 14);
686 		buf[88] = (1 << 14 | 0x7f);
687 		ahci_write_fis_piosetup(p);
688 		write_prdt(p, slot, cfis, (void *)buf, sizeof(buf));
689 		p->tfd = ATA_S_DSC | ATA_S_READY;
690 		p->is |= AHCI_P_IX_DHR;
691 	}
692 	p->ci &= ~(1 << slot);
693 	ahci_generate_intr(p->pr_sc);
694 }
695 
696 static void
697 atapi_inquiry(struct ahci_port *p, int slot, uint8_t *cfis)
698 {
699 	uint8_t buf[36];
700 	uint8_t *acmd;
701 	int len;
702 
703 	acmd = cfis + 0x40;
704 
705 	buf[0] = 0x05;
706 	buf[1] = 0x80;
707 	buf[2] = 0x00;
708 	buf[3] = 0x21;
709 	buf[4] = 31;
710 	buf[5] = 0;
711 	buf[6] = 0;
712 	buf[7] = 0;
713 	atapi_string(buf + 8, "BHYVE", 8);
714 	atapi_string(buf + 16, "BHYVE DVD-ROM", 16);
715 	atapi_string(buf + 32, "001", 4);
716 
717 	len = sizeof(buf);
718 	if (len > acmd[4])
719 		len = acmd[4];
720 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
721 	write_prdt(p, slot, cfis, buf, len);
722 	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
723 }
724 
725 static void
726 atapi_read_capacity(struct ahci_port *p, int slot, uint8_t *cfis)
727 {
728 	uint8_t buf[8];
729 	uint64_t sectors;
730 
731 	sectors = blockif_size(p->bctx) / 2048;
732 	be32enc(buf, sectors - 1);
733 	be32enc(buf + 4, 2048);
734 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
735 	write_prdt(p, slot, cfis, buf, sizeof(buf));
736 	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
737 }
738 
739 static void
740 atapi_read_toc(struct ahci_port *p, int slot, uint8_t *cfis)
741 {
742 	uint8_t *acmd;
743 	uint8_t format;
744 	int len;
745 
746 	acmd = cfis + 0x40;
747 
748 	len = be16dec(acmd + 7);
749 	format = acmd[9] >> 6;
750 	switch (format) {
751 	case 0:
752 	{
753 		int msf, size;
754 		uint64_t sectors;
755 		uint8_t start_track, buf[20], *bp;
756 
757 		msf = (acmd[1] >> 1) & 1;
758 		start_track = acmd[6];
759 		if (start_track > 1 && start_track != 0xaa) {
760 			uint32_t tfd;
761 			p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
762 			p->asc = 0x24;
763 			tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
764 			cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
765 			ahci_write_fis_d2h(p, slot, cfis, tfd);
766 			return;
767 		}
768 		bp = buf + 2;
769 		*bp++ = 1;
770 		*bp++ = 1;
771 		if (start_track <= 1) {
772 			*bp++ = 0;
773 			*bp++ = 0x14;
774 			*bp++ = 1;
775 			*bp++ = 0;
776 			if (msf) {
777 				*bp++ = 0;
778 				lba_to_msf(bp, 0);
779 				bp += 3;
780 			} else {
781 				*bp++ = 0;
782 				*bp++ = 0;
783 				*bp++ = 0;
784 				*bp++ = 0;
785 			}
786 		}
787 		*bp++ = 0;
788 		*bp++ = 0x14;
789 		*bp++ = 0xaa;
790 		*bp++ = 0;
791 		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
792 		sectors >>= 2;
793 		if (msf) {
794 			*bp++ = 0;
795 			lba_to_msf(bp, sectors);
796 			bp += 3;
797 		} else {
798 			be32enc(bp, sectors);
799 			bp += 4;
800 		}
801 		size = bp - buf;
802 		be16enc(buf, size - 2);
803 		if (len > size)
804 			len = size;
805 		write_prdt(p, slot, cfis, buf, len);
806 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
807 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
808 		break;
809 	}
810 	case 1:
811 	{
812 		uint8_t buf[12];
813 
814 		memset(buf, 0, sizeof(buf));
815 		buf[1] = 0xa;
816 		buf[2] = 0x1;
817 		buf[3] = 0x1;
818 		if (len > sizeof(buf))
819 			len = sizeof(buf);
820 		write_prdt(p, slot, cfis, buf, len);
821 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
822 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
823 		break;
824 	}
825 	case 2:
826 	{
827 		int msf, size;
828 		uint64_t sectors;
829 		uint8_t start_track, *bp, buf[50];
830 
831 		msf = (acmd[1] >> 1) & 1;
832 		start_track = acmd[6];
833 		bp = buf + 2;
834 		*bp++ = 1;
835 		*bp++ = 1;
836 
837 		*bp++ = 1;
838 		*bp++ = 0x14;
839 		*bp++ = 0;
840 		*bp++ = 0xa0;
841 		*bp++ = 0;
842 		*bp++ = 0;
843 		*bp++ = 0;
844 		*bp++ = 0;
845 		*bp++ = 1;
846 		*bp++ = 0;
847 		*bp++ = 0;
848 
849 		*bp++ = 1;
850 		*bp++ = 0x14;
851 		*bp++ = 0;
852 		*bp++ = 0xa1;
853 		*bp++ = 0;
854 		*bp++ = 0;
855 		*bp++ = 0;
856 		*bp++ = 0;
857 		*bp++ = 1;
858 		*bp++ = 0;
859 		*bp++ = 0;
860 
861 		*bp++ = 1;
862 		*bp++ = 0x14;
863 		*bp++ = 0;
864 		*bp++ = 0xa2;
865 		*bp++ = 0;
866 		*bp++ = 0;
867 		*bp++ = 0;
868 		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
869 		sectors >>= 2;
870 		if (msf) {
871 			*bp++ = 0;
872 			lba_to_msf(bp, sectors);
873 			bp += 3;
874 		} else {
875 			be32enc(bp, sectors);
876 			bp += 4;
877 		}
878 
879 		*bp++ = 1;
880 		*bp++ = 0x14;
881 		*bp++ = 0;
882 		*bp++ = 1;
883 		*bp++ = 0;
884 		*bp++ = 0;
885 		*bp++ = 0;
886 		if (msf) {
887 			*bp++ = 0;
888 			lba_to_msf(bp, 0);
889 			bp += 3;
890 		} else {
891 			*bp++ = 0;
892 			*bp++ = 0;
893 			*bp++ = 0;
894 			*bp++ = 0;
895 		}
896 
897 		size = bp - buf;
898 		be16enc(buf, size - 2);
899 		if (len > size)
900 			len = size;
901 		write_prdt(p, slot, cfis, buf, len);
902 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
903 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
904 		break;
905 	}
906 	default:
907 	{
908 		uint32_t tfd;
909 
910 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
911 		p->asc = 0x24;
912 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
913 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
914 		ahci_write_fis_d2h(p, slot, cfis, tfd);
915 		break;
916 	}
917 	}
918 }
919 
920 static void
921 atapi_read(struct ahci_port *p, int slot, uint8_t *cfis,
922 		uint32_t done, int seek)
923 {
924 	struct ahci_ioreq *aior;
925 	struct ahci_cmd_hdr *hdr;
926 	struct ahci_prdt_entry *prdt;
927 	struct blockif_req *breq;
928 	struct pci_ahci_softc *sc;
929 	uint8_t *acmd;
930 	uint64_t lba;
931 	uint32_t len;
932 	int i, err, iovcnt;
933 
934 	sc = p->pr_sc;
935 	acmd = cfis + 0x40;
936 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
937 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
938 
939 	prdt += seek;
940 	lba = be32dec(acmd + 2);
941 	if (acmd[0] == READ_10)
942 		len = be16dec(acmd + 7);
943 	else
944 		len = be32dec(acmd + 6);
945 	if (len == 0) {
946 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
947 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
948 	}
949 	lba *= 2048;
950 	len *= 2048;
951 
952 	/*
953 	 * Pull request off free list
954 	 */
955 	aior = STAILQ_FIRST(&p->iofhd);
956 	assert(aior != NULL);
957 	STAILQ_REMOVE_HEAD(&p->iofhd, io_list);
958 	aior->cfis = cfis;
959 	aior->slot = slot;
960 	aior->len = len;
961 	aior->done = done;
962 	breq = &aior->io_req;
963 	breq->br_offset = lba + done;
964 	iovcnt = hdr->prdtl - seek;
965 	if (iovcnt > BLOCKIF_IOV_MAX) {
966 		aior->prdtl = iovcnt - BLOCKIF_IOV_MAX;
967 		iovcnt = BLOCKIF_IOV_MAX;
968 	} else
969 		aior->prdtl = 0;
970 	breq->br_iovcnt = iovcnt;
971 
972 	/*
973 	 * Build up the iovec based on the prdt
974 	 */
975 	for (i = 0; i < iovcnt; i++) {
976 		uint32_t dbcsz;
977 
978 		dbcsz = (prdt->dbc & DBCMASK) + 1;
979 		breq->br_iov[i].iov_base = paddr_guest2host(ahci_ctx(sc),
980 		    prdt->dba, dbcsz);
981 		breq->br_iov[i].iov_len = dbcsz;
982 		aior->done += dbcsz;
983 		prdt++;
984 	}
985 	err = blockif_read(p->bctx, breq);
986 	assert(err == 0);
987 }
988 
989 static void
990 atapi_request_sense(struct ahci_port *p, int slot, uint8_t *cfis)
991 {
992 	uint8_t buf[64];
993 	uint8_t *acmd;
994 	int len;
995 
996 	acmd = cfis + 0x40;
997 	len = acmd[4];
998 	if (len > sizeof(buf))
999 		len = sizeof(buf);
1000 	memset(buf, 0, len);
1001 	buf[0] = 0x70 | (1 << 7);
1002 	buf[2] = p->sense_key;
1003 	buf[7] = 10;
1004 	buf[12] = p->asc;
1005 	write_prdt(p, slot, cfis, buf, len);
1006 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1007 	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1008 }
1009 
1010 static void
1011 atapi_start_stop_unit(struct ahci_port *p, int slot, uint8_t *cfis)
1012 {
1013 	uint8_t *acmd = cfis + 0x40;
1014 	uint32_t tfd;
1015 
1016 	switch (acmd[4] & 3) {
1017 	case 0:
1018 	case 1:
1019 	case 3:
1020 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1021 		tfd = ATA_S_READY | ATA_S_DSC;
1022 		break;
1023 	case 2:
1024 		/* TODO eject media */
1025 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1026 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1027 		p->asc = 0x53;
1028 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1029 		break;
1030 	}
1031 	ahci_write_fis_d2h(p, slot, cfis, tfd);
1032 }
1033 
1034 static void
1035 atapi_mode_sense(struct ahci_port *p, int slot, uint8_t *cfis)
1036 {
1037 	uint8_t *acmd;
1038 	uint32_t tfd;
1039 	uint8_t pc, code;
1040 	int len;
1041 
1042 	acmd = cfis + 0x40;
1043 	len = be16dec(acmd + 7);
1044 	pc = acmd[2] >> 6;
1045 	code = acmd[2] & 0x3f;
1046 
1047 	switch (pc) {
1048 	case 0:
1049 		switch (code) {
1050 		case MODEPAGE_RW_ERROR_RECOVERY:
1051 		{
1052 			uint8_t buf[16];
1053 
1054 			if (len > sizeof(buf))
1055 				len = sizeof(buf);
1056 
1057 			memset(buf, 0, sizeof(buf));
1058 			be16enc(buf, 16 - 2);
1059 			buf[2] = 0x70;
1060 			buf[8] = 0x01;
1061 			buf[9] = 16 - 10;
1062 			buf[11] = 0x05;
1063 			write_prdt(p, slot, cfis, buf, len);
1064 			tfd = ATA_S_READY | ATA_S_DSC;
1065 			break;
1066 		}
1067 		case MODEPAGE_CD_CAPABILITIES:
1068 		{
1069 			uint8_t buf[30];
1070 
1071 			if (len > sizeof(buf))
1072 				len = sizeof(buf);
1073 
1074 			memset(buf, 0, sizeof(buf));
1075 			be16enc(buf, 30 - 2);
1076 			buf[2] = 0x70;
1077 			buf[8] = 0x2A;
1078 			buf[9] = 30 - 10;
1079 			buf[10] = 0x08;
1080 			buf[12] = 0x71;
1081 			be16enc(&buf[18], 2);
1082 			be16enc(&buf[20], 512);
1083 			write_prdt(p, slot, cfis, buf, len);
1084 			tfd = ATA_S_READY | ATA_S_DSC;
1085 			break;
1086 		}
1087 		default:
1088 			goto error;
1089 			break;
1090 		}
1091 		break;
1092 	case 3:
1093 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1094 		p->asc = 0x39;
1095 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1096 		break;
1097 error:
1098 	case 1:
1099 	case 2:
1100 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1101 		p->asc = 0x24;
1102 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1103 		break;
1104 	}
1105 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1106 	ahci_write_fis_d2h(p, slot, cfis, tfd);
1107 }
1108 
1109 static void
1110 atapi_get_event_status_notification(struct ahci_port *p, int slot,
1111     uint8_t *cfis)
1112 {
1113 	uint8_t *acmd;
1114 	uint32_t tfd;
1115 
1116 	acmd = cfis + 0x40;
1117 
1118 	/* we don't support asynchronous operation */
1119 	if (!(acmd[1] & 1)) {
1120 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1121 		p->asc = 0x24;
1122 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1123 	} else {
1124 		uint8_t buf[8];
1125 		int len;
1126 
1127 		len = be16dec(acmd + 7);
1128 		if (len > sizeof(buf))
1129 			len = sizeof(buf);
1130 
1131 		memset(buf, 0, sizeof(buf));
1132 		be16enc(buf, 8 - 2);
1133 		buf[2] = 0x04;
1134 		buf[3] = 0x10;
1135 		buf[5] = 0x02;
1136 		write_prdt(p, slot, cfis, buf, len);
1137 		tfd = ATA_S_READY | ATA_S_DSC;
1138 	}
1139 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1140 	ahci_write_fis_d2h(p, slot, cfis, tfd);
1141 }
1142 
1143 static void
1144 handle_packet_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1145 {
1146 	uint8_t *acmd;
1147 
1148 	acmd = cfis + 0x40;
1149 
1150 #ifdef AHCI_DEBUG
1151 	{
1152 		int i;
1153 		DPRINTF("ACMD:");
1154 		for (i = 0; i < 16; i++)
1155 			DPRINTF("%02x ", acmd[i]);
1156 		DPRINTF("\n");
1157 	}
1158 #endif
1159 
1160 	switch (acmd[0]) {
1161 	case TEST_UNIT_READY:
1162 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1163 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1164 		break;
1165 	case INQUIRY:
1166 		atapi_inquiry(p, slot, cfis);
1167 		break;
1168 	case READ_CAPACITY:
1169 		atapi_read_capacity(p, slot, cfis);
1170 		break;
1171 	case PREVENT_ALLOW:
1172 		/* TODO */
1173 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1174 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1175 		break;
1176 	case READ_TOC:
1177 		atapi_read_toc(p, slot, cfis);
1178 		break;
1179 	case READ_10:
1180 	case READ_12:
1181 		atapi_read(p, slot, cfis, 0, 0);
1182 		break;
1183 	case REQUEST_SENSE:
1184 		atapi_request_sense(p, slot, cfis);
1185 		break;
1186 	case START_STOP_UNIT:
1187 		atapi_start_stop_unit(p, slot, cfis);
1188 		break;
1189 	case MODE_SENSE_10:
1190 		atapi_mode_sense(p, slot, cfis);
1191 		break;
1192 	case GET_EVENT_STATUS_NOTIFICATION:
1193 		atapi_get_event_status_notification(p, slot, cfis);
1194 		break;
1195 	default:
1196 		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1197 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1198 		p->asc = 0x20;
1199 		ahci_write_fis_d2h(p, slot, cfis, (p->sense_key << 12) |
1200 				ATA_S_READY | ATA_S_ERROR);
1201 		break;
1202 	}
1203 }
1204 
1205 static void
1206 ahci_handle_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1207 {
1208 
1209 	switch (cfis[2]) {
1210 	case ATA_ATA_IDENTIFY:
1211 		handle_identify(p, slot, cfis);
1212 		break;
1213 	case ATA_SETFEATURES:
1214 	{
1215 		switch (cfis[3]) {
1216 		case ATA_SF_ENAB_SATA_SF:
1217 			switch (cfis[12]) {
1218 			case ATA_SATA_SF_AN:
1219 				p->tfd = ATA_S_DSC | ATA_S_READY;
1220 				break;
1221 			default:
1222 				p->tfd = ATA_S_ERROR | ATA_S_READY;
1223 				p->tfd |= (ATA_ERROR_ABORT << 8);
1224 				break;
1225 			}
1226 			break;
1227 		case ATA_SF_ENAB_WCACHE:
1228 		case ATA_SF_DIS_WCACHE:
1229 		case ATA_SF_ENAB_RCACHE:
1230 		case ATA_SF_DIS_RCACHE:
1231 			p->tfd = ATA_S_DSC | ATA_S_READY;
1232 			break;
1233 		case ATA_SF_SETXFER:
1234 		{
1235 			switch (cfis[12] & 0xf8) {
1236 			case ATA_PIO:
1237 			case ATA_PIO0:
1238 				break;
1239 			case ATA_WDMA0:
1240 			case ATA_UDMA0:
1241 				p->xfermode = (cfis[12] & 0x7);
1242 				break;
1243 			}
1244 			p->tfd = ATA_S_DSC | ATA_S_READY;
1245 			break;
1246 		}
1247 		default:
1248 			p->tfd = ATA_S_ERROR | ATA_S_READY;
1249 			p->tfd |= (ATA_ERROR_ABORT << 8);
1250 			break;
1251 		}
1252 		ahci_write_fis_d2h(p, slot, cfis, p->tfd);
1253 		break;
1254 	}
1255 	case ATA_SET_MULTI:
1256 		if (cfis[12] != 0 &&
1257 			(cfis[12] > 128 || (cfis[12] & (cfis[12] - 1)))) {
1258 			p->tfd = ATA_S_ERROR | ATA_S_READY;
1259 			p->tfd |= (ATA_ERROR_ABORT << 8);
1260 		} else {
1261 			p->mult_sectors = cfis[12];
1262 			p->tfd = ATA_S_DSC | ATA_S_READY;
1263 		}
1264 		p->is |= AHCI_P_IX_DP;
1265 		p->ci &= ~(1 << slot);
1266 		ahci_generate_intr(p->pr_sc);
1267 		break;
1268 	case ATA_READ_DMA:
1269 	case ATA_WRITE_DMA:
1270 	case ATA_READ_DMA48:
1271 	case ATA_WRITE_DMA48:
1272 	case ATA_READ_FPDMA_QUEUED:
1273 	case ATA_WRITE_FPDMA_QUEUED:
1274 		ahci_handle_dma(p, slot, cfis, 0, 0);
1275 		break;
1276 	case ATA_FLUSHCACHE:
1277 	case ATA_FLUSHCACHE48:
1278 		ahci_handle_flush(p, slot, cfis);
1279 		break;
1280 	case ATA_STANDBY_CMD:
1281 		break;
1282 	case ATA_NOP:
1283 	case ATA_STANDBY_IMMEDIATE:
1284 	case ATA_IDLE_IMMEDIATE:
1285 	case ATA_SLEEP:
1286 		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1287 		break;
1288 	case ATA_ATAPI_IDENTIFY:
1289 		handle_atapi_identify(p, slot, cfis);
1290 		break;
1291 	case ATA_PACKET_CMD:
1292 		if (!p->atapi) {
1293 			p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1294 			p->is |= AHCI_P_IX_TFE;
1295 			p->ci &= ~(1 << slot);
1296 			ahci_generate_intr(p->pr_sc);
1297 		} else
1298 			handle_packet_cmd(p, slot, cfis);
1299 		break;
1300 	default:
1301 		WPRINTF("Unsupported cmd:%02x\n", cfis[2]);
1302 		p->tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1303 		p->is |= AHCI_P_IX_TFE;
1304 		p->ci &= ~(1 << slot);
1305 		ahci_generate_intr(p->pr_sc);
1306 		break;
1307 	}
1308 }
1309 
1310 static void
1311 ahci_handle_slot(struct ahci_port *p, int slot)
1312 {
1313 	struct ahci_cmd_hdr *hdr;
1314 	struct ahci_prdt_entry *prdt;
1315 	struct pci_ahci_softc *sc;
1316 	uint8_t *cfis;
1317 	int cfl;
1318 
1319 	sc = p->pr_sc;
1320 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1321 	cfl = (hdr->flags & 0x1f) * 4;
1322 	cfis = paddr_guest2host(ahci_ctx(sc), hdr->ctba,
1323 			0x80 + hdr->prdtl * sizeof(struct ahci_prdt_entry));
1324 	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
1325 
1326 #ifdef AHCI_DEBUG
1327 	DPRINTF("\ncfis:");
1328 	for (i = 0; i < cfl; i++) {
1329 		if (i % 10 == 0)
1330 			DPRINTF("\n");
1331 		DPRINTF("%02x ", cfis[i]);
1332 	}
1333 	DPRINTF("\n");
1334 
1335 	for (i = 0; i < hdr->prdtl; i++) {
1336 		DPRINTF("%d@%08"PRIx64"\n", prdt->dbc & 0x3fffff, prdt->dba);
1337 		prdt++;
1338 	}
1339 #endif
1340 
1341 	if (cfis[0] != FIS_TYPE_REGH2D) {
1342 		WPRINTF("Not a H2D FIS:%02x\n", cfis[0]);
1343 		return;
1344 	}
1345 
1346 	if (cfis[1] & 0x80) {
1347 		ahci_handle_cmd(p, slot, cfis);
1348 	} else {
1349 		if (cfis[15] & (1 << 2))
1350 			p->reset = 1;
1351 		else if (p->reset) {
1352 			p->reset = 0;
1353 			ahci_port_reset(p);
1354 		}
1355 		p->ci &= ~(1 << slot);
1356 	}
1357 }
1358 
1359 static void
1360 ahci_handle_port(struct ahci_port *p)
1361 {
1362 	int i;
1363 
1364 	if (!(p->cmd & AHCI_P_CMD_ST))
1365 		return;
1366 
1367 	/*
1368 	 * Search for any new commands to issue ignoring those that
1369 	 * are already in-flight.
1370 	 */
1371 	for (i = 0; (i < 32) && p->ci; i++) {
1372 		if ((p->ci & (1 << i)) && !(p->pending & (1 << i)))
1373 			ahci_handle_slot(p, i);
1374 	}
1375 }
1376 
1377 /*
1378  * blockif callback routine - this runs in the context of the blockif
1379  * i/o thread, so the mutex needs to be acquired.
1380  */
1381 static void
1382 ata_ioreq_cb(struct blockif_req *br, int err)
1383 {
1384 	struct ahci_cmd_hdr *hdr;
1385 	struct ahci_ioreq *aior;
1386 	struct ahci_port *p;
1387 	struct pci_ahci_softc *sc;
1388 	uint32_t tfd;
1389 	uint8_t *cfis;
1390 	int pending, slot, ncq;
1391 
1392 	DPRINTF("%s %d\n", __func__, err);
1393 
1394 	ncq = 0;
1395 	aior = br->br_param;
1396 	p = aior->io_pr;
1397 	cfis = aior->cfis;
1398 	slot = aior->slot;
1399 	pending = aior->prdtl;
1400 	sc = p->pr_sc;
1401 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1402 
1403 	if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
1404 			cfis[2] == ATA_READ_FPDMA_QUEUED)
1405 		ncq = 1;
1406 
1407 	pthread_mutex_lock(&sc->mtx);
1408 
1409 	/*
1410 	 * Move the blockif request back to the free list
1411 	 */
1412 	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_list);
1413 
1414 	if (pending && !err) {
1415 		ahci_handle_dma(p, slot, cfis, aior->done,
1416 		    hdr->prdtl - pending);
1417 		goto out;
1418 	}
1419 
1420 	if (!err && aior->done == aior->len) {
1421 		tfd = ATA_S_READY | ATA_S_DSC;
1422 		if (ncq)
1423 			hdr->prdbc = 0;
1424 		else
1425 			hdr->prdbc = aior->len;
1426 	} else {
1427 		tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1428 		hdr->prdbc = 0;
1429 		if (ncq)
1430 			p->serr |= (1 << slot);
1431 	}
1432 
1433 	/*
1434 	 * This command is now complete.
1435 	 */
1436 	p->pending &= ~(1 << slot);
1437 
1438 	if (ncq) {
1439 		p->sact &= ~(1 << slot);
1440 		ahci_write_fis_sdb(p, slot, tfd);
1441 	} else
1442 		ahci_write_fis_d2h(p, slot, cfis, tfd);
1443 
1444 out:
1445 	pthread_mutex_unlock(&sc->mtx);
1446 	DPRINTF("%s exit\n", __func__);
1447 }
1448 
1449 static void
1450 atapi_ioreq_cb(struct blockif_req *br, int err)
1451 {
1452 	struct ahci_cmd_hdr *hdr;
1453 	struct ahci_ioreq *aior;
1454 	struct ahci_port *p;
1455 	struct pci_ahci_softc *sc;
1456 	uint8_t *cfis;
1457 	uint32_t tfd;
1458 	int pending, slot;
1459 
1460 	DPRINTF("%s %d\n", __func__, err);
1461 
1462 	aior = br->br_param;
1463 	p = aior->io_pr;
1464 	cfis = aior->cfis;
1465 	slot = aior->slot;
1466 	pending = aior->prdtl;
1467 	sc = p->pr_sc;
1468 	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + aior->slot * AHCI_CL_SIZE);
1469 
1470 	pthread_mutex_lock(&sc->mtx);
1471 
1472 	/*
1473 	 * Move the blockif request back to the free list
1474 	 */
1475 	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_list);
1476 
1477 	if (pending && !err) {
1478 		atapi_read(p, slot, cfis, aior->done, hdr->prdtl - pending);
1479 		goto out;
1480 	}
1481 
1482 	if (!err && aior->done == aior->len) {
1483 		tfd = ATA_S_READY | ATA_S_DSC;
1484 		hdr->prdbc = aior->len;
1485 	} else {
1486 		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1487 		p->asc = 0x21;
1488 		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1489 		hdr->prdbc = 0;
1490 	}
1491 
1492 	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1493 	ahci_write_fis_d2h(p, slot, cfis, tfd);
1494 
1495 out:
1496 	pthread_mutex_unlock(&sc->mtx);
1497 	DPRINTF("%s exit\n", __func__);
1498 }
1499 
1500 static void
1501 pci_ahci_ioreq_init(struct ahci_port *pr)
1502 {
1503 	struct ahci_ioreq *vr;
1504 	int i;
1505 
1506 	pr->ioqsz = blockif_queuesz(pr->bctx);
1507 	pr->ioreq = calloc(pr->ioqsz, sizeof(struct ahci_ioreq));
1508 	STAILQ_INIT(&pr->iofhd);
1509 
1510 	/*
1511 	 * Add all i/o request entries to the free queue
1512 	 */
1513 	for (i = 0; i < pr->ioqsz; i++) {
1514 		vr = &pr->ioreq[i];
1515 		vr->io_pr = pr;
1516 		if (!pr->atapi)
1517 			vr->io_req.br_callback = ata_ioreq_cb;
1518 		else
1519 			vr->io_req.br_callback = atapi_ioreq_cb;
1520 		vr->io_req.br_param = vr;
1521 		STAILQ_INSERT_TAIL(&pr->iofhd, vr, io_list);
1522 	}
1523 }
1524 
1525 static void
1526 pci_ahci_port_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
1527 {
1528 	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
1529 	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
1530 	struct ahci_port *p = &sc->port[port];
1531 
1532 	DPRINTF("pci_ahci_port %d: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
1533 		port, offset, value);
1534 
1535 	switch (offset) {
1536 	case AHCI_P_CLB:
1537 		p->clb = value;
1538 		break;
1539 	case AHCI_P_CLBU:
1540 		p->clbu = value;
1541 		break;
1542 	case AHCI_P_FB:
1543 		p->fb = value;
1544 		break;
1545 	case AHCI_P_FBU:
1546 		p->fbu = value;
1547 		break;
1548 	case AHCI_P_IS:
1549 		p->is &= ~value;
1550 		break;
1551 	case AHCI_P_IE:
1552 		p->ie = value & 0xFDC000FF;
1553 		ahci_generate_intr(sc);
1554 		break;
1555 	case AHCI_P_CMD:
1556 	{
1557 		p->cmd = value;
1558 
1559 		if (!(value & AHCI_P_CMD_ST)) {
1560 			p->cmd &= ~(AHCI_P_CMD_CR | AHCI_P_CMD_CCS_MASK);
1561 			p->ci = 0;
1562 			p->sact = 0;
1563 		} else {
1564 			uint64_t clb;
1565 
1566 			p->cmd |= AHCI_P_CMD_CR;
1567 			clb = (uint64_t)p->clbu << 32 | p->clb;
1568 			p->cmd_lst = paddr_guest2host(ahci_ctx(sc), clb,
1569 					AHCI_CL_SIZE * AHCI_MAX_SLOTS);
1570 		}
1571 
1572 		if (value & AHCI_P_CMD_FRE) {
1573 			uint64_t fb;
1574 
1575 			p->cmd |= AHCI_P_CMD_FR;
1576 			fb = (uint64_t)p->fbu << 32 | p->fb;
1577 			/* we don't support FBSCP, so rfis size is 256Bytes */
1578 			p->rfis = paddr_guest2host(ahci_ctx(sc), fb, 256);
1579 		} else {
1580 			p->cmd &= ~AHCI_P_CMD_FR;
1581 		}
1582 
1583 		if (value & AHCI_P_CMD_CLO) {
1584 			p->tfd = 0;
1585 			p->cmd &= ~AHCI_P_CMD_CLO;
1586 		}
1587 
1588 		ahci_handle_port(p);
1589 		break;
1590 	}
1591 	case AHCI_P_TFD:
1592 	case AHCI_P_SIG:
1593 	case AHCI_P_SSTS:
1594 		WPRINTF("pci_ahci_port: read only registers 0x%"PRIx64"\n", offset);
1595 		break;
1596 	case AHCI_P_SCTL:
1597 		if (!(p->cmd & AHCI_P_CMD_ST)) {
1598 			if (value & ATA_SC_DET_RESET)
1599 				ahci_port_reset(p);
1600 			p->sctl = value;
1601 		}
1602 		break;
1603 	case AHCI_P_SERR:
1604 		p->serr &= ~value;
1605 		break;
1606 	case AHCI_P_SACT:
1607 		p->sact |= value;
1608 		break;
1609 	case AHCI_P_CI:
1610 		p->ci |= value;
1611 		ahci_handle_port(p);
1612 		break;
1613 	case AHCI_P_SNTF:
1614 	case AHCI_P_FBS:
1615 	default:
1616 		break;
1617 	}
1618 }
1619 
1620 static void
1621 pci_ahci_host_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
1622 {
1623 	DPRINTF("pci_ahci_host: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
1624 		offset, value);
1625 
1626 	switch (offset) {
1627 	case AHCI_CAP:
1628 	case AHCI_PI:
1629 	case AHCI_VS:
1630 	case AHCI_CAP2:
1631 		DPRINTF("pci_ahci_host: read only registers 0x%"PRIx64"\n", offset);
1632 		break;
1633 	case AHCI_GHC:
1634 		if (value & AHCI_GHC_HR)
1635 			ahci_reset(sc);
1636 		else if (value & AHCI_GHC_IE) {
1637 			sc->ghc |= AHCI_GHC_IE;
1638 			ahci_generate_intr(sc);
1639 		}
1640 		break;
1641 	case AHCI_IS:
1642 		sc->is &= ~value;
1643 		ahci_generate_intr(sc);
1644 		break;
1645 	default:
1646 		break;
1647 	}
1648 }
1649 
1650 static void
1651 pci_ahci_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
1652 		int baridx, uint64_t offset, int size, uint64_t value)
1653 {
1654 	struct pci_ahci_softc *sc = pi->pi_arg;
1655 
1656 	assert(baridx == 5);
1657 	assert(size == 4);
1658 
1659 	pthread_mutex_lock(&sc->mtx);
1660 
1661 	if (offset < AHCI_OFFSET)
1662 		pci_ahci_host_write(sc, offset, value);
1663 	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
1664 		pci_ahci_port_write(sc, offset, value);
1665 	else
1666 		WPRINTF("pci_ahci: unknown i/o write offset 0x%"PRIx64"\n", offset);
1667 
1668 	pthread_mutex_unlock(&sc->mtx);
1669 }
1670 
1671 static uint64_t
1672 pci_ahci_host_read(struct pci_ahci_softc *sc, uint64_t offset)
1673 {
1674 	uint32_t value;
1675 
1676 	switch (offset) {
1677 	case AHCI_CAP:
1678 	case AHCI_GHC:
1679 	case AHCI_IS:
1680 	case AHCI_PI:
1681 	case AHCI_VS:
1682 	case AHCI_CCCC:
1683 	case AHCI_CCCP:
1684 	case AHCI_EM_LOC:
1685 	case AHCI_EM_CTL:
1686 	case AHCI_CAP2:
1687 	{
1688 		uint32_t *p = &sc->cap;
1689 		p += (offset - AHCI_CAP) / sizeof(uint32_t);
1690 		value = *p;
1691 		break;
1692 	}
1693 	default:
1694 		value = 0;
1695 		break;
1696 	}
1697 	DPRINTF("pci_ahci_host: read offset 0x%"PRIx64" value 0x%x\n",
1698 		offset, value);
1699 
1700 	return (value);
1701 }
1702 
1703 static uint64_t
1704 pci_ahci_port_read(struct pci_ahci_softc *sc, uint64_t offset)
1705 {
1706 	uint32_t value;
1707 	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
1708 	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
1709 
1710 	switch (offset) {
1711 	case AHCI_P_CLB:
1712 	case AHCI_P_CLBU:
1713 	case AHCI_P_FB:
1714 	case AHCI_P_FBU:
1715 	case AHCI_P_IS:
1716 	case AHCI_P_IE:
1717 	case AHCI_P_CMD:
1718 	case AHCI_P_TFD:
1719 	case AHCI_P_SIG:
1720 	case AHCI_P_SSTS:
1721 	case AHCI_P_SCTL:
1722 	case AHCI_P_SERR:
1723 	case AHCI_P_SACT:
1724 	case AHCI_P_CI:
1725 	case AHCI_P_SNTF:
1726 	case AHCI_P_FBS:
1727 	{
1728 		uint32_t *p= &sc->port[port].clb;
1729 		p += (offset - AHCI_P_CLB) / sizeof(uint32_t);
1730 		value = *p;
1731 		break;
1732 	}
1733 	default:
1734 		value = 0;
1735 		break;
1736 	}
1737 
1738 	DPRINTF("pci_ahci_port %d: read offset 0x%"PRIx64" value 0x%x\n",
1739 		port, offset, value);
1740 
1741 	return value;
1742 }
1743 
1744 static uint64_t
1745 pci_ahci_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
1746     uint64_t offset, int size)
1747 {
1748 	struct pci_ahci_softc *sc = pi->pi_arg;
1749 	uint32_t value;
1750 
1751 	assert(baridx == 5);
1752 	assert(size == 4);
1753 
1754 	pthread_mutex_lock(&sc->mtx);
1755 
1756 	if (offset < AHCI_OFFSET)
1757 		value = pci_ahci_host_read(sc, offset);
1758 	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
1759 		value = pci_ahci_port_read(sc, offset);
1760 	else {
1761 		value = 0;
1762 		WPRINTF("pci_ahci: unknown i/o read offset 0x%"PRIx64"\n", offset);
1763 	}
1764 
1765 	pthread_mutex_unlock(&sc->mtx);
1766 
1767 	return (value);
1768 }
1769 
1770 static int
1771 pci_ahci_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts, int atapi)
1772 {
1773 	char bident[sizeof("XX:X:X")];
1774 	struct blockif_ctxt *bctxt;
1775 	struct pci_ahci_softc *sc;
1776 	int ret, slots;
1777 
1778 	ret = 0;
1779 
1780 	if (opts == NULL) {
1781 		fprintf(stderr, "pci_ahci: backing device required\n");
1782 		return (1);
1783 	}
1784 
1785 #ifdef AHCI_DEBUG
1786 	dbg = fopen("/tmp/log", "w+");
1787 #endif
1788 
1789 	sc = calloc(1, sizeof(struct pci_ahci_softc));
1790 	pi->pi_arg = sc;
1791 	sc->asc_pi = pi;
1792 	sc->ports = MAX_PORTS;
1793 
1794 	/*
1795 	 * Only use port 0 for a backing device. All other ports will be
1796 	 * marked as unused
1797 	 */
1798 	sc->port[0].atapi = atapi;
1799 
1800 	/*
1801 	 * Attempt to open the backing image. Use the PCI
1802 	 * slot/func for the identifier string.
1803 	 */
1804 	snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func);
1805 	bctxt = blockif_open(opts, bident);
1806 	if (bctxt == NULL) {
1807 		ret = 1;
1808 		goto open_fail;
1809 	}
1810 	sc->port[0].bctx = bctxt;
1811 	sc->port[0].pr_sc = sc;
1812 
1813 	/*
1814 	 * Allocate blockif request structures and add them
1815 	 * to the free list
1816 	 */
1817 	pci_ahci_ioreq_init(&sc->port[0]);
1818 
1819 	pthread_mutex_init(&sc->mtx, NULL);
1820 
1821 	/* Intel ICH8 AHCI */
1822 	slots = sc->port[0].ioqsz;
1823 	if (slots > 32)
1824 		slots = 32;
1825 	--slots;
1826 	sc->cap = AHCI_CAP_64BIT | AHCI_CAP_SNCQ | AHCI_CAP_SSNTF |
1827 	    AHCI_CAP_SMPS | AHCI_CAP_SSS | AHCI_CAP_SALP |
1828 	    AHCI_CAP_SAL | AHCI_CAP_SCLO | (0x3 << AHCI_CAP_ISS_SHIFT)|
1829 	    AHCI_CAP_PMD | AHCI_CAP_SSC | AHCI_CAP_PSC |
1830 	    (slots << AHCI_CAP_NCS_SHIFT) | AHCI_CAP_SXS | (sc->ports - 1);
1831 
1832 	/* Only port 0 implemented */
1833 	sc->pi = 1;
1834 	sc->vs = 0x10300;
1835 	sc->cap2 = AHCI_CAP2_APST;
1836 	ahci_reset(sc);
1837 
1838 	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x2821);
1839 	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x8086);
1840 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
1841 	pci_set_cfgdata8(pi, PCIR_SUBCLASS, PCIS_STORAGE_SATA);
1842 	pci_set_cfgdata8(pi, PCIR_PROGIF, PCIP_STORAGE_SATA_AHCI_1_0);
1843 	pci_emul_add_msicap(pi, 1);
1844 	pci_emul_alloc_bar(pi, 5, PCIBAR_MEM32,
1845 	    AHCI_OFFSET + sc->ports * AHCI_STEP);
1846 
1847 	pci_lintr_request(pi);
1848 
1849 open_fail:
1850 	if (ret) {
1851 		blockif_close(sc->port[0].bctx);
1852 		free(sc);
1853 	}
1854 
1855 	return (ret);
1856 }
1857 
1858 static int
1859 pci_ahci_hd_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1860 {
1861 
1862 	return (pci_ahci_init(ctx, pi, opts, 0));
1863 }
1864 
1865 static int
1866 pci_ahci_atapi_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
1867 {
1868 
1869 	return (pci_ahci_init(ctx, pi, opts, 1));
1870 }
1871 
1872 /*
1873  * Use separate emulation names to distinguish drive and atapi devices
1874  */
1875 struct pci_devemu pci_de_ahci_hd = {
1876 	.pe_emu =	"ahci-hd",
1877 	.pe_init =	pci_ahci_hd_init,
1878 	.pe_barwrite =	pci_ahci_write,
1879 	.pe_barread =	pci_ahci_read
1880 };
1881 PCI_EMUL_SET(pci_de_ahci_hd);
1882 
1883 struct pci_devemu pci_de_ahci_cd = {
1884 	.pe_emu =	"ahci-cd",
1885 	.pe_init =	pci_ahci_atapi_init,
1886 	.pe_barwrite =	pci_ahci_write,
1887 	.pe_barread =	pci_ahci_read
1888 };
1889 PCI_EMUL_SET(pci_de_ahci_cd);
1890