1/*-
2 * Copyright (c) 2013  Zhixiang Yu <zcore@freebsd.org>
3 * Copyright (c) 2015-2016 Alexander Motin <mav@FreeBSD.org>
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 *
27 * $FreeBSD: stable/10/usr.sbin/bhyve/pci_ahci.c 341606 2018-12-05 21:49:39Z emaste $
28 */
29
30#include <sys/cdefs.h>
31__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_ahci.c 341606 2018-12-05 21:49:39Z emaste $");
32
33#include <sys/param.h>
34#include <sys/linker_set.h>
35#include <sys/stat.h>
36#include <sys/uio.h>
37#include <sys/ioctl.h>
38#include <sys/disk.h>
39#include <sys/ata.h>
40#include <sys/endian.h>
41
42#include <errno.h>
43#include <fcntl.h>
44#include <stdio.h>
45#include <stdlib.h>
46#include <stdint.h>
47#include <string.h>
48#include <strings.h>
49#include <unistd.h>
50#include <assert.h>
51#include <pthread.h>
52#include <pthread_np.h>
53#include <inttypes.h>
54#include <md5.h>
55
56#include "bhyverun.h"
57#include "pci_emul.h"
58#include "ahci.h"
59#include "block_if.h"
60
61#define	DEF_PORTS	6	/* Intel ICH8 AHCI supports 6 ports */
62#define	MAX_PORTS	32	/* AHCI supports 32 ports */
63
64#define	PxSIG_ATA	0x00000101 /* ATA drive */
65#define	PxSIG_ATAPI	0xeb140101 /* ATAPI drive */
66
67enum sata_fis_type {
68	FIS_TYPE_REGH2D		= 0x27,	/* Register FIS - host to device */
69	FIS_TYPE_REGD2H		= 0x34,	/* Register FIS - device to host */
70	FIS_TYPE_DMAACT		= 0x39,	/* DMA activate FIS - device to host */
71	FIS_TYPE_DMASETUP	= 0x41,	/* DMA setup FIS - bidirectional */
72	FIS_TYPE_DATA		= 0x46,	/* Data FIS - bidirectional */
73	FIS_TYPE_BIST		= 0x58,	/* BIST activate FIS - bidirectional */
74	FIS_TYPE_PIOSETUP	= 0x5F,	/* PIO setup FIS - device to host */
75	FIS_TYPE_SETDEVBITS	= 0xA1,	/* Set dev bits FIS - device to host */
76};
77
78/*
79 * SCSI opcodes
80 */
81#define	TEST_UNIT_READY		0x00
82#define	REQUEST_SENSE		0x03
83#define	INQUIRY			0x12
84#define	START_STOP_UNIT		0x1B
85#define	PREVENT_ALLOW		0x1E
86#define	READ_CAPACITY		0x25
87#define	READ_10			0x28
88#define	POSITION_TO_ELEMENT	0x2B
89#define	READ_TOC		0x43
90#define	GET_EVENT_STATUS_NOTIFICATION 0x4A
91#define	MODE_SENSE_10		0x5A
92#define	REPORT_LUNS		0xA0
93#define	READ_12			0xA8
94#define	READ_CD			0xBE
95
96/*
97 * SCSI mode page codes
98 */
99#define	MODEPAGE_RW_ERROR_RECOVERY	0x01
100#define	MODEPAGE_CD_CAPABILITIES	0x2A
101
102/*
103 * ATA commands
104 */
105#define	ATA_SF_ENAB_SATA_SF		0x10
106#define		ATA_SATA_SF_AN		0x05
107#define	ATA_SF_DIS_SATA_SF		0x90
108
109/*
110 * Debug printf
111 */
112#ifdef AHCI_DEBUG
113static FILE *dbg;
114#define DPRINTF(format, arg...)	do{fprintf(dbg, format, ##arg);fflush(dbg);}while(0)
115#else
116#define DPRINTF(format, arg...)
117#endif
118#define WPRINTF(format, arg...) printf(format, ##arg)
119
120struct ahci_ioreq {
121	struct blockif_req io_req;
122	struct ahci_port *io_pr;
123	STAILQ_ENTRY(ahci_ioreq) io_flist;
124	TAILQ_ENTRY(ahci_ioreq) io_blist;
125	uint8_t *cfis;
126	uint32_t len;
127	uint32_t done;
128	int slot;
129	int more;
130};
131
132struct ahci_port {
133	struct blockif_ctxt *bctx;
134	struct pci_ahci_softc *pr_sc;
135	uint8_t *cmd_lst;
136	uint8_t *rfis;
137	char ident[20 + 1];
138	int port;
139	int atapi;
140	int reset;
141	int waitforclear;
142	int mult_sectors;
143	uint8_t xfermode;
144	uint8_t err_cfis[20];
145	uint8_t sense_key;
146	uint8_t asc;
147	u_int ccs;
148	uint32_t pending;
149
150	uint32_t clb;
151	uint32_t clbu;
152	uint32_t fb;
153	uint32_t fbu;
154	uint32_t is;
155	uint32_t ie;
156	uint32_t cmd;
157	uint32_t unused0;
158	uint32_t tfd;
159	uint32_t sig;
160	uint32_t ssts;
161	uint32_t sctl;
162	uint32_t serr;
163	uint32_t sact;
164	uint32_t ci;
165	uint32_t sntf;
166	uint32_t fbs;
167
168	/*
169	 * i/o request info
170	 */
171	struct ahci_ioreq *ioreq;
172	int ioqsz;
173	STAILQ_HEAD(ahci_fhead, ahci_ioreq) iofhd;
174	TAILQ_HEAD(ahci_bhead, ahci_ioreq) iobhd;
175};
176
177struct ahci_cmd_hdr {
178	uint16_t flags;
179	uint16_t prdtl;
180	uint32_t prdbc;
181	uint64_t ctba;
182	uint32_t reserved[4];
183};
184
185struct ahci_prdt_entry {
186	uint64_t dba;
187	uint32_t reserved;
188#define	DBCMASK		0x3fffff
189	uint32_t dbc;
190};
191
192struct pci_ahci_softc {
193	struct pci_devinst *asc_pi;
194	pthread_mutex_t	mtx;
195	int ports;
196	uint32_t cap;
197	uint32_t ghc;
198	uint32_t is;
199	uint32_t pi;
200	uint32_t vs;
201	uint32_t ccc_ctl;
202	uint32_t ccc_pts;
203	uint32_t em_loc;
204	uint32_t em_ctl;
205	uint32_t cap2;
206	uint32_t bohc;
207	uint32_t lintr;
208	struct ahci_port port[MAX_PORTS];
209};
210#define	ahci_ctx(sc)	((sc)->asc_pi->pi_vmctx)
211
212static void ahci_handle_port(struct ahci_port *p);
213
214static inline void lba_to_msf(uint8_t *buf, int lba)
215{
216	lba += 150;
217	buf[0] = (lba / 75) / 60;
218	buf[1] = (lba / 75) % 60;
219	buf[2] = lba % 75;
220}
221
222/*
223 * Generate HBA interrupts on global IS register write.
224 */
225static void
226ahci_generate_intr(struct pci_ahci_softc *sc, uint32_t mask)
227{
228	struct pci_devinst *pi = sc->asc_pi;
229	struct ahci_port *p;
230	int i, nmsg;
231	uint32_t mmask;
232
233	/* Update global IS from PxIS/PxIE. */
234	for (i = 0; i < sc->ports; i++) {
235		p = &sc->port[i];
236		if (p->is & p->ie)
237			sc->is |= (1 << i);
238	}
239	DPRINTF("%s(%08x) %08x\n", __func__, mask, sc->is);
240
241	/* If there is nothing enabled -- clear legacy interrupt and exit. */
242	if (sc->is == 0 || (sc->ghc & AHCI_GHC_IE) == 0) {
243		if (sc->lintr) {
244			pci_lintr_deassert(pi);
245			sc->lintr = 0;
246		}
247		return;
248	}
249
250	/* If there is anything and no MSI -- assert legacy interrupt. */
251	nmsg = pci_msi_maxmsgnum(pi);
252	if (nmsg == 0) {
253		if (!sc->lintr) {
254			sc->lintr = 1;
255			pci_lintr_assert(pi);
256		}
257		return;
258	}
259
260	/* Assert respective MSIs for ports that were touched. */
261	for (i = 0; i < nmsg; i++) {
262		if (sc->ports <= nmsg || i < nmsg - 1)
263			mmask = 1 << i;
264		else
265			mmask = 0xffffffff << i;
266		if (sc->is & mask && mmask & mask)
267			pci_generate_msi(pi, i);
268	}
269}
270
271/*
272 * Generate HBA interrupt on specific port event.
273 */
274static void
275ahci_port_intr(struct ahci_port *p)
276{
277	struct pci_ahci_softc *sc = p->pr_sc;
278	struct pci_devinst *pi = sc->asc_pi;
279	int nmsg;
280
281	DPRINTF("%s(%d) %08x/%08x %08x\n", __func__,
282	    p->port, p->is, p->ie, sc->is);
283
284	/* If there is nothing enabled -- we are done. */
285	if ((p->is & p->ie) == 0)
286		return;
287
288	/* In case of non-shared MSI always generate interrupt. */
289	nmsg = pci_msi_maxmsgnum(pi);
290	if (sc->ports <= nmsg || p->port < nmsg - 1) {
291		sc->is |= (1 << p->port);
292		if ((sc->ghc & AHCI_GHC_IE) == 0)
293			return;
294		pci_generate_msi(pi, p->port);
295		return;
296	}
297
298	/* If IS for this port is already set -- do nothing. */
299	if (sc->is & (1 << p->port))
300		return;
301
302	sc->is |= (1 << p->port);
303
304	/* If interrupts are enabled -- generate one. */
305	if ((sc->ghc & AHCI_GHC_IE) == 0)
306		return;
307	if (nmsg > 0) {
308		pci_generate_msi(pi, nmsg - 1);
309	} else if (!sc->lintr) {
310		sc->lintr = 1;
311		pci_lintr_assert(pi);
312	}
313}
314
315static void
316ahci_write_fis(struct ahci_port *p, enum sata_fis_type ft, uint8_t *fis)
317{
318	int offset, len, irq;
319
320	if (p->rfis == NULL || !(p->cmd & AHCI_P_CMD_FRE))
321		return;
322
323	switch (ft) {
324	case FIS_TYPE_REGD2H:
325		offset = 0x40;
326		len = 20;
327		irq = (fis[1] & (1 << 6)) ? AHCI_P_IX_DHR : 0;
328		break;
329	case FIS_TYPE_SETDEVBITS:
330		offset = 0x58;
331		len = 8;
332		irq = (fis[1] & (1 << 6)) ? AHCI_P_IX_SDB : 0;
333		break;
334	case FIS_TYPE_PIOSETUP:
335		offset = 0x20;
336		len = 20;
337		irq = (fis[1] & (1 << 6)) ? AHCI_P_IX_PS : 0;
338		break;
339	default:
340		WPRINTF("unsupported fis type %d\n", ft);
341		return;
342	}
343	if (fis[2] & ATA_S_ERROR) {
344		p->waitforclear = 1;
345		irq |= AHCI_P_IX_TFE;
346	}
347	memcpy(p->rfis + offset, fis, len);
348	if (irq) {
349		if (~p->is & irq) {
350			p->is |= irq;
351			ahci_port_intr(p);
352		}
353	}
354}
355
356static void
357ahci_write_fis_piosetup(struct ahci_port *p)
358{
359	uint8_t fis[20];
360
361	memset(fis, 0, sizeof(fis));
362	fis[0] = FIS_TYPE_PIOSETUP;
363	ahci_write_fis(p, FIS_TYPE_PIOSETUP, fis);
364}
365
366static void
367ahci_write_fis_sdb(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t tfd)
368{
369	uint8_t fis[8];
370	uint8_t error;
371
372	error = (tfd >> 8) & 0xff;
373	tfd &= 0x77;
374	memset(fis, 0, sizeof(fis));
375	fis[0] = FIS_TYPE_SETDEVBITS;
376	fis[1] = (1 << 6);
377	fis[2] = tfd;
378	fis[3] = error;
379	if (fis[2] & ATA_S_ERROR) {
380		p->err_cfis[0] = slot;
381		p->err_cfis[2] = tfd;
382		p->err_cfis[3] = error;
383		memcpy(&p->err_cfis[4], cfis + 4, 16);
384	} else {
385		*(uint32_t *)(fis + 4) = (1 << slot);
386		p->sact &= ~(1 << slot);
387	}
388	p->tfd &= ~0x77;
389	p->tfd |= tfd;
390	ahci_write_fis(p, FIS_TYPE_SETDEVBITS, fis);
391}
392
393static void
394ahci_write_fis_d2h(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t tfd)
395{
396	uint8_t fis[20];
397	uint8_t error;
398
399	error = (tfd >> 8) & 0xff;
400	memset(fis, 0, sizeof(fis));
401	fis[0] = FIS_TYPE_REGD2H;
402	fis[1] = (1 << 6);
403	fis[2] = tfd & 0xff;
404	fis[3] = error;
405	fis[4] = cfis[4];
406	fis[5] = cfis[5];
407	fis[6] = cfis[6];
408	fis[7] = cfis[7];
409	fis[8] = cfis[8];
410	fis[9] = cfis[9];
411	fis[10] = cfis[10];
412	fis[11] = cfis[11];
413	fis[12] = cfis[12];
414	fis[13] = cfis[13];
415	if (fis[2] & ATA_S_ERROR) {
416		p->err_cfis[0] = 0x80;
417		p->err_cfis[2] = tfd & 0xff;
418		p->err_cfis[3] = error;
419		memcpy(&p->err_cfis[4], cfis + 4, 16);
420	} else
421		p->ci &= ~(1 << slot);
422	p->tfd = tfd;
423	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
424}
425
426static void
427ahci_write_fis_d2h_ncq(struct ahci_port *p, int slot)
428{
429	uint8_t fis[20];
430
431	p->tfd = ATA_S_READY | ATA_S_DSC;
432	memset(fis, 0, sizeof(fis));
433	fis[0] = FIS_TYPE_REGD2H;
434	fis[1] = 0;			/* No interrupt */
435	fis[2] = p->tfd;		/* Status */
436	fis[3] = 0;			/* No error */
437	p->ci &= ~(1 << slot);
438	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
439}
440
441static void
442ahci_write_reset_fis_d2h(struct ahci_port *p)
443{
444	uint8_t fis[20];
445
446	memset(fis, 0, sizeof(fis));
447	fis[0] = FIS_TYPE_REGD2H;
448	fis[3] = 1;
449	fis[4] = 1;
450	if (p->atapi) {
451		fis[5] = 0x14;
452		fis[6] = 0xeb;
453	}
454	fis[12] = 1;
455	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
456}
457
458static void
459ahci_check_stopped(struct ahci_port *p)
460{
461	/*
462	 * If we are no longer processing the command list and nothing
463	 * is in-flight, clear the running bit, the current command
464	 * slot, the command issue and active bits.
465	 */
466	if (!(p->cmd & AHCI_P_CMD_ST)) {
467		if (p->pending == 0) {
468			p->ccs = 0;
469			p->cmd &= ~(AHCI_P_CMD_CR | AHCI_P_CMD_CCS_MASK);
470			p->ci = 0;
471			p->sact = 0;
472			p->waitforclear = 0;
473		}
474	}
475}
476
477static void
478ahci_port_stop(struct ahci_port *p)
479{
480	struct ahci_ioreq *aior;
481	uint8_t *cfis;
482	int slot;
483	int ncq;
484	int error;
485
486	assert(pthread_mutex_isowned_np(&p->pr_sc->mtx));
487
488	TAILQ_FOREACH(aior, &p->iobhd, io_blist) {
489		/*
490		 * Try to cancel the outstanding blockif request.
491		 */
492		error = blockif_cancel(p->bctx, &aior->io_req);
493		if (error != 0)
494			continue;
495
496		slot = aior->slot;
497		cfis = aior->cfis;
498		if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
499		    cfis[2] == ATA_READ_FPDMA_QUEUED ||
500		    cfis[2] == ATA_SEND_FPDMA_QUEUED)
501			ncq = 1;
502
503		if (ncq)
504			p->sact &= ~(1 << slot);
505		else
506			p->ci &= ~(1 << slot);
507
508		/*
509		 * This command is now done.
510		 */
511		p->pending &= ~(1 << slot);
512
513		/*
514		 * Delete the blockif request from the busy list
515		 */
516		TAILQ_REMOVE(&p->iobhd, aior, io_blist);
517
518		/*
519		 * Move the blockif request back to the free list
520		 */
521		STAILQ_INSERT_TAIL(&p->iofhd, aior, io_flist);
522	}
523
524	ahci_check_stopped(p);
525}
526
527static void
528ahci_port_reset(struct ahci_port *pr)
529{
530	pr->serr = 0;
531	pr->sact = 0;
532	pr->xfermode = ATA_UDMA6;
533	pr->mult_sectors = 128;
534
535	if (!pr->bctx) {
536		pr->ssts = ATA_SS_DET_NO_DEVICE;
537		pr->sig = 0xFFFFFFFF;
538		pr->tfd = 0x7F;
539		return;
540	}
541	pr->ssts = ATA_SS_DET_PHY_ONLINE | ATA_SS_IPM_ACTIVE;
542	if (pr->sctl & ATA_SC_SPD_MASK)
543		pr->ssts |= (pr->sctl & ATA_SC_SPD_MASK);
544	else
545		pr->ssts |= ATA_SS_SPD_GEN3;
546	pr->tfd = (1 << 8) | ATA_S_DSC | ATA_S_DMA;
547	if (!pr->atapi) {
548		pr->sig = PxSIG_ATA;
549		pr->tfd |= ATA_S_READY;
550	} else
551		pr->sig = PxSIG_ATAPI;
552	ahci_write_reset_fis_d2h(pr);
553}
554
555static void
556ahci_reset(struct pci_ahci_softc *sc)
557{
558	int i;
559
560	sc->ghc = AHCI_GHC_AE;
561	sc->is = 0;
562
563	if (sc->lintr) {
564		pci_lintr_deassert(sc->asc_pi);
565		sc->lintr = 0;
566	}
567
568	for (i = 0; i < sc->ports; i++) {
569		sc->port[i].ie = 0;
570		sc->port[i].is = 0;
571		sc->port[i].cmd = (AHCI_P_CMD_SUD | AHCI_P_CMD_POD);
572		if (sc->port[i].bctx)
573			sc->port[i].cmd |= AHCI_P_CMD_CPS;
574		sc->port[i].sctl = 0;
575		ahci_port_reset(&sc->port[i]);
576	}
577}
578
579static void
580ata_string(uint8_t *dest, const char *src, int len)
581{
582	int i;
583
584	for (i = 0; i < len; i++) {
585		if (*src)
586			dest[i ^ 1] = *src++;
587		else
588			dest[i ^ 1] = ' ';
589	}
590}
591
592static void
593atapi_string(uint8_t *dest, const char *src, int len)
594{
595	int i;
596
597	for (i = 0; i < len; i++) {
598		if (*src)
599			dest[i] = *src++;
600		else
601			dest[i] = ' ';
602	}
603}
604
605/*
606 * Build up the iovec based on the PRDT, 'done' and 'len'.
607 */
608static void
609ahci_build_iov(struct ahci_port *p, struct ahci_ioreq *aior,
610    struct ahci_prdt_entry *prdt, uint16_t prdtl)
611{
612	struct blockif_req *breq = &aior->io_req;
613	int i, j, skip, todo, left, extra;
614	uint32_t dbcsz;
615
616	/* Copy part of PRDT between 'done' and 'len' bytes into the iov. */
617	skip = aior->done;
618	left = aior->len - aior->done;
619	todo = 0;
620	for (i = 0, j = 0; i < prdtl && j < BLOCKIF_IOV_MAX && left > 0;
621	    i++, prdt++) {
622		dbcsz = (prdt->dbc & DBCMASK) + 1;
623		/* Skip already done part of the PRDT */
624		if (dbcsz <= skip) {
625			skip -= dbcsz;
626			continue;
627		}
628		dbcsz -= skip;
629		if (dbcsz > left)
630			dbcsz = left;
631		breq->br_iov[j].iov_base = paddr_guest2host(ahci_ctx(p->pr_sc),
632		    prdt->dba + skip, dbcsz);
633		breq->br_iov[j].iov_len = dbcsz;
634		todo += dbcsz;
635		left -= dbcsz;
636		skip = 0;
637		j++;
638	}
639
640	/* If we got limited by IOV length, round I/O down to sector size. */
641	if (j == BLOCKIF_IOV_MAX) {
642		extra = todo % blockif_sectsz(p->bctx);
643		todo -= extra;
644		assert(todo > 0);
645		while (extra > 0) {
646			if (breq->br_iov[j - 1].iov_len > extra) {
647				breq->br_iov[j - 1].iov_len -= extra;
648				break;
649			}
650			extra -= breq->br_iov[j - 1].iov_len;
651			j--;
652		}
653	}
654
655	breq->br_iovcnt = j;
656	breq->br_resid = todo;
657	aior->done += todo;
658	aior->more = (aior->done < aior->len && i < prdtl);
659}
660
661static void
662ahci_handle_rw(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t done)
663{
664	struct ahci_ioreq *aior;
665	struct blockif_req *breq;
666	struct ahci_prdt_entry *prdt;
667	struct ahci_cmd_hdr *hdr;
668	uint64_t lba;
669	uint32_t len;
670	int err, first, ncq, readop;
671
672	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
673	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
674	ncq = 0;
675	readop = 1;
676	first = (done == 0);
677
678	if (cfis[2] == ATA_WRITE || cfis[2] == ATA_WRITE48 ||
679	    cfis[2] == ATA_WRITE_MUL || cfis[2] == ATA_WRITE_MUL48 ||
680	    cfis[2] == ATA_WRITE_DMA || cfis[2] == ATA_WRITE_DMA48 ||
681	    cfis[2] == ATA_WRITE_FPDMA_QUEUED)
682		readop = 0;
683
684	if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
685	    cfis[2] == ATA_READ_FPDMA_QUEUED) {
686		lba = ((uint64_t)cfis[10] << 40) |
687			((uint64_t)cfis[9] << 32) |
688			((uint64_t)cfis[8] << 24) |
689			((uint64_t)cfis[6] << 16) |
690			((uint64_t)cfis[5] << 8) |
691			cfis[4];
692		len = cfis[11] << 8 | cfis[3];
693		if (!len)
694			len = 65536;
695		ncq = 1;
696	} else if (cfis[2] == ATA_READ48 || cfis[2] == ATA_WRITE48 ||
697	    cfis[2] == ATA_READ_MUL48 || cfis[2] == ATA_WRITE_MUL48 ||
698	    cfis[2] == ATA_READ_DMA48 || cfis[2] == ATA_WRITE_DMA48) {
699		lba = ((uint64_t)cfis[10] << 40) |
700			((uint64_t)cfis[9] << 32) |
701			((uint64_t)cfis[8] << 24) |
702			((uint64_t)cfis[6] << 16) |
703			((uint64_t)cfis[5] << 8) |
704			cfis[4];
705		len = cfis[13] << 8 | cfis[12];
706		if (!len)
707			len = 65536;
708	} else {
709		lba = ((cfis[7] & 0xf) << 24) | (cfis[6] << 16) |
710			(cfis[5] << 8) | cfis[4];
711		len = cfis[12];
712		if (!len)
713			len = 256;
714	}
715	lba *= blockif_sectsz(p->bctx);
716	len *= blockif_sectsz(p->bctx);
717
718	/* Pull request off free list */
719	aior = STAILQ_FIRST(&p->iofhd);
720	assert(aior != NULL);
721	STAILQ_REMOVE_HEAD(&p->iofhd, io_flist);
722
723	aior->cfis = cfis;
724	aior->slot = slot;
725	aior->len = len;
726	aior->done = done;
727	breq = &aior->io_req;
728	breq->br_offset = lba + done;
729	ahci_build_iov(p, aior, prdt, hdr->prdtl);
730
731	/* Mark this command in-flight. */
732	p->pending |= 1 << slot;
733
734	/* Stuff request onto busy list. */
735	TAILQ_INSERT_HEAD(&p->iobhd, aior, io_blist);
736
737	if (ncq && first)
738		ahci_write_fis_d2h_ncq(p, slot);
739
740	if (readop)
741		err = blockif_read(p->bctx, breq);
742	else
743		err = blockif_write(p->bctx, breq);
744	assert(err == 0);
745}
746
747static void
748ahci_handle_flush(struct ahci_port *p, int slot, uint8_t *cfis)
749{
750	struct ahci_ioreq *aior;
751	struct blockif_req *breq;
752	int err;
753
754	/*
755	 * Pull request off free list
756	 */
757	aior = STAILQ_FIRST(&p->iofhd);
758	assert(aior != NULL);
759	STAILQ_REMOVE_HEAD(&p->iofhd, io_flist);
760	aior->cfis = cfis;
761	aior->slot = slot;
762	aior->len = 0;
763	aior->done = 0;
764	aior->more = 0;
765	breq = &aior->io_req;
766
767	/*
768	 * Mark this command in-flight.
769	 */
770	p->pending |= 1 << slot;
771
772	/*
773	 * Stuff request onto busy list
774	 */
775	TAILQ_INSERT_HEAD(&p->iobhd, aior, io_blist);
776
777	err = blockif_flush(p->bctx, breq);
778	assert(err == 0);
779}
780
781static inline void
782read_prdt(struct ahci_port *p, int slot, uint8_t *cfis,
783		void *buf, int size)
784{
785	struct ahci_cmd_hdr *hdr;
786	struct ahci_prdt_entry *prdt;
787	void *to;
788	int i, len;
789
790	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
791	len = size;
792	to = buf;
793	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
794	for (i = 0; i < hdr->prdtl && len; i++) {
795		uint8_t *ptr;
796		uint32_t dbcsz;
797		int sublen;
798
799		dbcsz = (prdt->dbc & DBCMASK) + 1;
800		ptr = paddr_guest2host(ahci_ctx(p->pr_sc), prdt->dba, dbcsz);
801		sublen = MIN(len, dbcsz);
802		memcpy(to, ptr, sublen);
803		len -= sublen;
804		to += sublen;
805		prdt++;
806	}
807}
808
809static void
810ahci_handle_dsm_trim(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t done)
811{
812	struct ahci_ioreq *aior;
813	struct blockif_req *breq;
814	uint8_t *entry;
815	uint64_t elba;
816	uint32_t len, elen;
817	int err, first, ncq;
818	uint8_t buf[512];
819
820	first = (done == 0);
821	if (cfis[2] == ATA_DATA_SET_MANAGEMENT) {
822		len = (uint16_t)cfis[13] << 8 | cfis[12];
823		len *= 512;
824		ncq = 0;
825	} else { /* ATA_SEND_FPDMA_QUEUED */
826		len = (uint16_t)cfis[11] << 8 | cfis[3];
827		len *= 512;
828		ncq = 1;
829	}
830	read_prdt(p, slot, cfis, buf, sizeof(buf));
831
832next:
833	entry = &buf[done];
834	elba = ((uint64_t)entry[5] << 40) |
835		((uint64_t)entry[4] << 32) |
836		((uint64_t)entry[3] << 24) |
837		((uint64_t)entry[2] << 16) |
838		((uint64_t)entry[1] << 8) |
839		entry[0];
840	elen = (uint16_t)entry[7] << 8 | entry[6];
841	done += 8;
842	if (elen == 0) {
843		if (done >= len) {
844			if (ncq) {
845				if (first)
846					ahci_write_fis_d2h_ncq(p, slot);
847				ahci_write_fis_sdb(p, slot, cfis,
848				    ATA_S_READY | ATA_S_DSC);
849			} else {
850				ahci_write_fis_d2h(p, slot, cfis,
851				    ATA_S_READY | ATA_S_DSC);
852			}
853			p->pending &= ~(1 << slot);
854			ahci_check_stopped(p);
855			if (!first)
856				ahci_handle_port(p);
857			return;
858		}
859		goto next;
860	}
861
862	/*
863	 * Pull request off free list
864	 */
865	aior = STAILQ_FIRST(&p->iofhd);
866	assert(aior != NULL);
867	STAILQ_REMOVE_HEAD(&p->iofhd, io_flist);
868	aior->cfis = cfis;
869	aior->slot = slot;
870	aior->len = len;
871	aior->done = done;
872	aior->more = (len != done);
873
874	breq = &aior->io_req;
875	breq->br_offset = elba * blockif_sectsz(p->bctx);
876	breq->br_resid = elen * blockif_sectsz(p->bctx);
877
878	/*
879	 * Mark this command in-flight.
880	 */
881	p->pending |= 1 << slot;
882
883	/*
884	 * Stuff request onto busy list
885	 */
886	TAILQ_INSERT_HEAD(&p->iobhd, aior, io_blist);
887
888	if (ncq && first)
889		ahci_write_fis_d2h_ncq(p, slot);
890
891	err = blockif_delete(p->bctx, breq);
892	assert(err == 0);
893}
894
895static inline void
896write_prdt(struct ahci_port *p, int slot, uint8_t *cfis,
897		void *buf, int size)
898{
899	struct ahci_cmd_hdr *hdr;
900	struct ahci_prdt_entry *prdt;
901	void *from;
902	int i, len;
903
904	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
905	len = size;
906	from = buf;
907	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
908	for (i = 0; i < hdr->prdtl && len; i++) {
909		uint8_t *ptr;
910		uint32_t dbcsz;
911		int sublen;
912
913		dbcsz = (prdt->dbc & DBCMASK) + 1;
914		ptr = paddr_guest2host(ahci_ctx(p->pr_sc), prdt->dba, dbcsz);
915		sublen = MIN(len, dbcsz);
916		memcpy(ptr, from, sublen);
917		len -= sublen;
918		from += sublen;
919		prdt++;
920	}
921	hdr->prdbc = size - len;
922}
923
924static void
925ahci_checksum(uint8_t *buf, int size)
926{
927	int i;
928	uint8_t sum = 0;
929
930	for (i = 0; i < size - 1; i++)
931		sum += buf[i];
932	buf[size - 1] = 0x100 - sum;
933}
934
935static void
936ahci_handle_read_log(struct ahci_port *p, int slot, uint8_t *cfis)
937{
938	struct ahci_cmd_hdr *hdr;
939	uint32_t buf[128];
940	uint8_t *buf8 = (uint8_t *)buf;
941	uint16_t *buf16 = (uint16_t *)buf;
942
943	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
944	if (p->atapi || hdr->prdtl == 0 || cfis[5] != 0 ||
945	    cfis[9] != 0 || cfis[12] != 1 || cfis[13] != 0) {
946		ahci_write_fis_d2h(p, slot, cfis,
947		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
948		return;
949	}
950
951	memset(buf, 0, sizeof(buf));
952	if (cfis[4] == 0x00) {	/* Log directory */
953		buf16[0x00] = 1; /* Version -- 1 */
954		buf16[0x10] = 1; /* NCQ Command Error Log -- 1 page */
955		buf16[0x13] = 1; /* SATA NCQ Send and Receive Log -- 1 page */
956	} else if (cfis[4] == 0x10) {	/* NCQ Command Error Log */
957		memcpy(buf8, p->err_cfis, sizeof(p->err_cfis));
958		ahci_checksum(buf8, sizeof(buf));
959	} else if (cfis[4] == 0x13) {	/* SATA NCQ Send and Receive Log */
960		if (blockif_candelete(p->bctx) && !blockif_is_ro(p->bctx)) {
961			buf[0x00] = 1;	/* SFQ DSM supported */
962			buf[0x01] = 1;	/* SFQ DSM TRIM supported */
963		}
964	} else {
965		ahci_write_fis_d2h(p, slot, cfis,
966		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
967		return;
968	}
969
970	if (cfis[2] == ATA_READ_LOG_EXT)
971		ahci_write_fis_piosetup(p);
972	write_prdt(p, slot, cfis, (void *)buf, sizeof(buf));
973	ahci_write_fis_d2h(p, slot, cfis, ATA_S_DSC | ATA_S_READY);
974}
975
976static void
977handle_identify(struct ahci_port *p, int slot, uint8_t *cfis)
978{
979	struct ahci_cmd_hdr *hdr;
980
981	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
982	if (p->atapi || hdr->prdtl == 0) {
983		ahci_write_fis_d2h(p, slot, cfis,
984		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
985	} else {
986		uint16_t buf[256];
987		uint64_t sectors;
988		int sectsz, psectsz, psectoff, candelete, ro;
989		uint16_t cyl;
990		uint8_t sech, heads;
991
992		ro = blockif_is_ro(p->bctx);
993		candelete = blockif_candelete(p->bctx);
994		sectsz = blockif_sectsz(p->bctx);
995		sectors = blockif_size(p->bctx) / sectsz;
996		blockif_chs(p->bctx, &cyl, &heads, &sech);
997		blockif_psectsz(p->bctx, &psectsz, &psectoff);
998		memset(buf, 0, sizeof(buf));
999		buf[0] = 0x0040;
1000		buf[1] = cyl;
1001		buf[3] = heads;
1002		buf[6] = sech;
1003		ata_string((uint8_t *)(buf+10), p->ident, 20);
1004		ata_string((uint8_t *)(buf+23), "001", 8);
1005		ata_string((uint8_t *)(buf+27), "BHYVE SATA DISK", 40);
1006		buf[47] = (0x8000 | 128);
1007		buf[48] = 0;
1008		buf[49] = (1 << 8 | 1 << 9 | 1 << 11);
1009		buf[50] = (1 << 14);
1010		buf[53] = (1 << 1 | 1 << 2);
1011		if (p->mult_sectors)
1012			buf[59] = (0x100 | p->mult_sectors);
1013		if (sectors <= 0x0fffffff) {
1014			buf[60] = sectors;
1015			buf[61] = (sectors >> 16);
1016		} else {
1017			buf[60] = 0xffff;
1018			buf[61] = 0x0fff;
1019		}
1020		buf[63] = 0x7;
1021		if (p->xfermode & ATA_WDMA0)
1022			buf[63] |= (1 << ((p->xfermode & 7) + 8));
1023		buf[64] = 0x3;
1024		buf[65] = 120;
1025		buf[66] = 120;
1026		buf[67] = 120;
1027		buf[68] = 120;
1028		buf[69] = 0;
1029		buf[75] = 31;
1030		buf[76] = (ATA_SATA_GEN1 | ATA_SATA_GEN2 | ATA_SATA_GEN3 |
1031			   ATA_SUPPORT_NCQ);
1032		buf[77] = (ATA_SUPPORT_RCVSND_FPDMA_QUEUED |
1033			   (p->ssts & ATA_SS_SPD_MASK) >> 3);
1034		buf[80] = 0x3f0;
1035		buf[81] = 0x28;
1036		buf[82] = (ATA_SUPPORT_POWERMGT | ATA_SUPPORT_WRITECACHE|
1037			   ATA_SUPPORT_LOOKAHEAD | ATA_SUPPORT_NOP);
1038		buf[83] = (ATA_SUPPORT_ADDRESS48 | ATA_SUPPORT_FLUSHCACHE |
1039			   ATA_SUPPORT_FLUSHCACHE48 | 1 << 14);
1040		buf[84] = (1 << 14);
1041		buf[85] = (ATA_SUPPORT_POWERMGT | ATA_SUPPORT_WRITECACHE|
1042			   ATA_SUPPORT_LOOKAHEAD | ATA_SUPPORT_NOP);
1043		buf[86] = (ATA_SUPPORT_ADDRESS48 | ATA_SUPPORT_FLUSHCACHE |
1044			   ATA_SUPPORT_FLUSHCACHE48 | 1 << 15);
1045		buf[87] = (1 << 14);
1046		buf[88] = 0x7f;
1047		if (p->xfermode & ATA_UDMA0)
1048			buf[88] |= (1 << ((p->xfermode & 7) + 8));
1049		buf[100] = sectors;
1050		buf[101] = (sectors >> 16);
1051		buf[102] = (sectors >> 32);
1052		buf[103] = (sectors >> 48);
1053		if (candelete && !ro) {
1054			buf[69] |= ATA_SUPPORT_RZAT | ATA_SUPPORT_DRAT;
1055			buf[105] = 1;
1056			buf[169] = ATA_SUPPORT_DSM_TRIM;
1057		}
1058		buf[106] = 0x4000;
1059		buf[209] = 0x4000;
1060		if (psectsz > sectsz) {
1061			buf[106] |= 0x2000;
1062			buf[106] |= ffsl(psectsz / sectsz) - 1;
1063			buf[209] |= (psectoff / sectsz);
1064		}
1065		if (sectsz > 512) {
1066			buf[106] |= 0x1000;
1067			buf[117] = sectsz / 2;
1068			buf[118] = ((sectsz / 2) >> 16);
1069		}
1070		buf[119] = (ATA_SUPPORT_RWLOGDMAEXT | 1 << 14);
1071		buf[120] = (ATA_SUPPORT_RWLOGDMAEXT | 1 << 14);
1072		buf[222] = 0x1020;
1073		buf[255] = 0x00a5;
1074		ahci_checksum((uint8_t *)buf, sizeof(buf));
1075		ahci_write_fis_piosetup(p);
1076		write_prdt(p, slot, cfis, (void *)buf, sizeof(buf));
1077		ahci_write_fis_d2h(p, slot, cfis, ATA_S_DSC | ATA_S_READY);
1078	}
1079}
1080
1081static void
1082handle_atapi_identify(struct ahci_port *p, int slot, uint8_t *cfis)
1083{
1084	if (!p->atapi) {
1085		ahci_write_fis_d2h(p, slot, cfis,
1086		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1087	} else {
1088		uint16_t buf[256];
1089
1090		memset(buf, 0, sizeof(buf));
1091		buf[0] = (2 << 14 | 5 << 8 | 1 << 7 | 2 << 5);
1092		ata_string((uint8_t *)(buf+10), p->ident, 20);
1093		ata_string((uint8_t *)(buf+23), "001", 8);
1094		ata_string((uint8_t *)(buf+27), "BHYVE SATA DVD ROM", 40);
1095		buf[49] = (1 << 9 | 1 << 8);
1096		buf[50] = (1 << 14 | 1);
1097		buf[53] = (1 << 2 | 1 << 1);
1098		buf[62] = 0x3f;
1099		buf[63] = 7;
1100		if (p->xfermode & ATA_WDMA0)
1101			buf[63] |= (1 << ((p->xfermode & 7) + 8));
1102		buf[64] = 3;
1103		buf[65] = 120;
1104		buf[66] = 120;
1105		buf[67] = 120;
1106		buf[68] = 120;
1107		buf[76] = (ATA_SATA_GEN1 | ATA_SATA_GEN2 | ATA_SATA_GEN3);
1108		buf[77] = ((p->ssts & ATA_SS_SPD_MASK) >> 3);
1109		buf[78] = (1 << 5);
1110		buf[80] = 0x3f0;
1111		buf[82] = (ATA_SUPPORT_POWERMGT | ATA_SUPPORT_PACKET |
1112			   ATA_SUPPORT_RESET | ATA_SUPPORT_NOP);
1113		buf[83] = (1 << 14);
1114		buf[84] = (1 << 14);
1115		buf[85] = (ATA_SUPPORT_POWERMGT | ATA_SUPPORT_PACKET |
1116			   ATA_SUPPORT_RESET | ATA_SUPPORT_NOP);
1117		buf[87] = (1 << 14);
1118		buf[88] = 0x7f;
1119		if (p->xfermode & ATA_UDMA0)
1120			buf[88] |= (1 << ((p->xfermode & 7) + 8));
1121		buf[222] = 0x1020;
1122		buf[255] = 0x00a5;
1123		ahci_checksum((uint8_t *)buf, sizeof(buf));
1124		ahci_write_fis_piosetup(p);
1125		write_prdt(p, slot, cfis, (void *)buf, sizeof(buf));
1126		ahci_write_fis_d2h(p, slot, cfis, ATA_S_DSC | ATA_S_READY);
1127	}
1128}
1129
1130static void
1131atapi_inquiry(struct ahci_port *p, int slot, uint8_t *cfis)
1132{
1133	uint8_t buf[36];
1134	uint8_t *acmd;
1135	int len;
1136	uint32_t tfd;
1137
1138	acmd = cfis + 0x40;
1139
1140	if (acmd[1] & 1) {		/* VPD */
1141		if (acmd[2] == 0) {	/* Supported VPD pages */
1142			buf[0] = 0x05;
1143			buf[1] = 0;
1144			buf[2] = 0;
1145			buf[3] = 1;
1146			buf[4] = 0;
1147			len = 4 + buf[3];
1148		} else {
1149			p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1150			p->asc = 0x24;
1151			tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1152			cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1153			ahci_write_fis_d2h(p, slot, cfis, tfd);
1154			return;
1155		}
1156	} else {
1157		buf[0] = 0x05;
1158		buf[1] = 0x80;
1159		buf[2] = 0x00;
1160		buf[3] = 0x21;
1161		buf[4] = 31;
1162		buf[5] = 0;
1163		buf[6] = 0;
1164		buf[7] = 0;
1165		atapi_string(buf + 8, "BHYVE", 8);
1166		atapi_string(buf + 16, "BHYVE DVD-ROM", 16);
1167		atapi_string(buf + 32, "001", 4);
1168		len = sizeof(buf);
1169	}
1170
1171	if (len > acmd[4])
1172		len = acmd[4];
1173	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1174	write_prdt(p, slot, cfis, buf, len);
1175	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1176}
1177
1178static void
1179atapi_read_capacity(struct ahci_port *p, int slot, uint8_t *cfis)
1180{
1181	uint8_t buf[8];
1182	uint64_t sectors;
1183
1184	sectors = blockif_size(p->bctx) / 2048;
1185	be32enc(buf, sectors - 1);
1186	be32enc(buf + 4, 2048);
1187	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1188	write_prdt(p, slot, cfis, buf, sizeof(buf));
1189	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1190}
1191
1192static void
1193atapi_read_toc(struct ahci_port *p, int slot, uint8_t *cfis)
1194{
1195	uint8_t *acmd;
1196	uint8_t format;
1197	int len;
1198
1199	acmd = cfis + 0x40;
1200
1201	len = be16dec(acmd + 7);
1202	format = acmd[9] >> 6;
1203	switch (format) {
1204	case 0:
1205	{
1206		int msf, size;
1207		uint64_t sectors;
1208		uint8_t start_track, buf[20], *bp;
1209
1210		msf = (acmd[1] >> 1) & 1;
1211		start_track = acmd[6];
1212		if (start_track > 1 && start_track != 0xaa) {
1213			uint32_t tfd;
1214			p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1215			p->asc = 0x24;
1216			tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1217			cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1218			ahci_write_fis_d2h(p, slot, cfis, tfd);
1219			return;
1220		}
1221		bp = buf + 2;
1222		*bp++ = 1;
1223		*bp++ = 1;
1224		if (start_track <= 1) {
1225			*bp++ = 0;
1226			*bp++ = 0x14;
1227			*bp++ = 1;
1228			*bp++ = 0;
1229			if (msf) {
1230				*bp++ = 0;
1231				lba_to_msf(bp, 0);
1232				bp += 3;
1233			} else {
1234				*bp++ = 0;
1235				*bp++ = 0;
1236				*bp++ = 0;
1237				*bp++ = 0;
1238			}
1239		}
1240		*bp++ = 0;
1241		*bp++ = 0x14;
1242		*bp++ = 0xaa;
1243		*bp++ = 0;
1244		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
1245		sectors >>= 2;
1246		if (msf) {
1247			*bp++ = 0;
1248			lba_to_msf(bp, sectors);
1249			bp += 3;
1250		} else {
1251			be32enc(bp, sectors);
1252			bp += 4;
1253		}
1254		size = bp - buf;
1255		be16enc(buf, size - 2);
1256		if (len > size)
1257			len = size;
1258		write_prdt(p, slot, cfis, buf, len);
1259		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1260		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1261		break;
1262	}
1263	case 1:
1264	{
1265		uint8_t buf[12];
1266
1267		memset(buf, 0, sizeof(buf));
1268		buf[1] = 0xa;
1269		buf[2] = 0x1;
1270		buf[3] = 0x1;
1271		if (len > sizeof(buf))
1272			len = sizeof(buf);
1273		write_prdt(p, slot, cfis, buf, len);
1274		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1275		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1276		break;
1277	}
1278	case 2:
1279	{
1280		int msf, size;
1281		uint64_t sectors;
1282		uint8_t start_track, *bp, buf[50];
1283
1284		msf = (acmd[1] >> 1) & 1;
1285		start_track = acmd[6];
1286		bp = buf + 2;
1287		*bp++ = 1;
1288		*bp++ = 1;
1289
1290		*bp++ = 1;
1291		*bp++ = 0x14;
1292		*bp++ = 0;
1293		*bp++ = 0xa0;
1294		*bp++ = 0;
1295		*bp++ = 0;
1296		*bp++ = 0;
1297		*bp++ = 0;
1298		*bp++ = 1;
1299		*bp++ = 0;
1300		*bp++ = 0;
1301
1302		*bp++ = 1;
1303		*bp++ = 0x14;
1304		*bp++ = 0;
1305		*bp++ = 0xa1;
1306		*bp++ = 0;
1307		*bp++ = 0;
1308		*bp++ = 0;
1309		*bp++ = 0;
1310		*bp++ = 1;
1311		*bp++ = 0;
1312		*bp++ = 0;
1313
1314		*bp++ = 1;
1315		*bp++ = 0x14;
1316		*bp++ = 0;
1317		*bp++ = 0xa2;
1318		*bp++ = 0;
1319		*bp++ = 0;
1320		*bp++ = 0;
1321		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
1322		sectors >>= 2;
1323		if (msf) {
1324			*bp++ = 0;
1325			lba_to_msf(bp, sectors);
1326			bp += 3;
1327		} else {
1328			be32enc(bp, sectors);
1329			bp += 4;
1330		}
1331
1332		*bp++ = 1;
1333		*bp++ = 0x14;
1334		*bp++ = 0;
1335		*bp++ = 1;
1336		*bp++ = 0;
1337		*bp++ = 0;
1338		*bp++ = 0;
1339		if (msf) {
1340			*bp++ = 0;
1341			lba_to_msf(bp, 0);
1342			bp += 3;
1343		} else {
1344			*bp++ = 0;
1345			*bp++ = 0;
1346			*bp++ = 0;
1347			*bp++ = 0;
1348		}
1349
1350		size = bp - buf;
1351		be16enc(buf, size - 2);
1352		if (len > size)
1353			len = size;
1354		write_prdt(p, slot, cfis, buf, len);
1355		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1356		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1357		break;
1358	}
1359	default:
1360	{
1361		uint32_t tfd;
1362
1363		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1364		p->asc = 0x24;
1365		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1366		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1367		ahci_write_fis_d2h(p, slot, cfis, tfd);
1368		break;
1369	}
1370	}
1371}
1372
1373static void
1374atapi_report_luns(struct ahci_port *p, int slot, uint8_t *cfis)
1375{
1376	uint8_t buf[16];
1377
1378	memset(buf, 0, sizeof(buf));
1379	buf[3] = 8;
1380
1381	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1382	write_prdt(p, slot, cfis, buf, sizeof(buf));
1383	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1384}
1385
1386static void
1387atapi_read(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t done)
1388{
1389	struct ahci_ioreq *aior;
1390	struct ahci_cmd_hdr *hdr;
1391	struct ahci_prdt_entry *prdt;
1392	struct blockif_req *breq;
1393	struct pci_ahci_softc *sc;
1394	uint8_t *acmd;
1395	uint64_t lba;
1396	uint32_t len;
1397	int err;
1398
1399	sc = p->pr_sc;
1400	acmd = cfis + 0x40;
1401	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1402	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
1403
1404	lba = be32dec(acmd + 2);
1405	if (acmd[0] == READ_10)
1406		len = be16dec(acmd + 7);
1407	else
1408		len = be32dec(acmd + 6);
1409	if (len == 0) {
1410		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1411		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1412	}
1413	lba *= 2048;
1414	len *= 2048;
1415
1416	/*
1417	 * Pull request off free list
1418	 */
1419	aior = STAILQ_FIRST(&p->iofhd);
1420	assert(aior != NULL);
1421	STAILQ_REMOVE_HEAD(&p->iofhd, io_flist);
1422	aior->cfis = cfis;
1423	aior->slot = slot;
1424	aior->len = len;
1425	aior->done = done;
1426	breq = &aior->io_req;
1427	breq->br_offset = lba + done;
1428	ahci_build_iov(p, aior, prdt, hdr->prdtl);
1429
1430	/* Mark this command in-flight. */
1431	p->pending |= 1 << slot;
1432
1433	/* Stuff request onto busy list. */
1434	TAILQ_INSERT_HEAD(&p->iobhd, aior, io_blist);
1435
1436	err = blockif_read(p->bctx, breq);
1437	assert(err == 0);
1438}
1439
1440static void
1441atapi_request_sense(struct ahci_port *p, int slot, uint8_t *cfis)
1442{
1443	uint8_t buf[64];
1444	uint8_t *acmd;
1445	int len;
1446
1447	acmd = cfis + 0x40;
1448	len = acmd[4];
1449	if (len > sizeof(buf))
1450		len = sizeof(buf);
1451	memset(buf, 0, len);
1452	buf[0] = 0x70 | (1 << 7);
1453	buf[2] = p->sense_key;
1454	buf[7] = 10;
1455	buf[12] = p->asc;
1456	write_prdt(p, slot, cfis, buf, len);
1457	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1458	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1459}
1460
1461static void
1462atapi_start_stop_unit(struct ahci_port *p, int slot, uint8_t *cfis)
1463{
1464	uint8_t *acmd = cfis + 0x40;
1465	uint32_t tfd;
1466
1467	switch (acmd[4] & 3) {
1468	case 0:
1469	case 1:
1470	case 3:
1471		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1472		tfd = ATA_S_READY | ATA_S_DSC;
1473		break;
1474	case 2:
1475		/* TODO eject media */
1476		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1477		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1478		p->asc = 0x53;
1479		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1480		break;
1481	}
1482	ahci_write_fis_d2h(p, slot, cfis, tfd);
1483}
1484
1485static void
1486atapi_mode_sense(struct ahci_port *p, int slot, uint8_t *cfis)
1487{
1488	uint8_t *acmd;
1489	uint32_t tfd;
1490	uint8_t pc, code;
1491	int len;
1492
1493	acmd = cfis + 0x40;
1494	len = be16dec(acmd + 7);
1495	pc = acmd[2] >> 6;
1496	code = acmd[2] & 0x3f;
1497
1498	switch (pc) {
1499	case 0:
1500		switch (code) {
1501		case MODEPAGE_RW_ERROR_RECOVERY:
1502		{
1503			uint8_t buf[16];
1504
1505			if (len > sizeof(buf))
1506				len = sizeof(buf);
1507
1508			memset(buf, 0, sizeof(buf));
1509			be16enc(buf, 16 - 2);
1510			buf[2] = 0x70;
1511			buf[8] = 0x01;
1512			buf[9] = 16 - 10;
1513			buf[11] = 0x05;
1514			write_prdt(p, slot, cfis, buf, len);
1515			tfd = ATA_S_READY | ATA_S_DSC;
1516			break;
1517		}
1518		case MODEPAGE_CD_CAPABILITIES:
1519		{
1520			uint8_t buf[30];
1521
1522			if (len > sizeof(buf))
1523				len = sizeof(buf);
1524
1525			memset(buf, 0, sizeof(buf));
1526			be16enc(buf, 30 - 2);
1527			buf[2] = 0x70;
1528			buf[8] = 0x2A;
1529			buf[9] = 30 - 10;
1530			buf[10] = 0x08;
1531			buf[12] = 0x71;
1532			be16enc(&buf[18], 2);
1533			be16enc(&buf[20], 512);
1534			write_prdt(p, slot, cfis, buf, len);
1535			tfd = ATA_S_READY | ATA_S_DSC;
1536			break;
1537		}
1538		default:
1539			goto error;
1540			break;
1541		}
1542		break;
1543	case 3:
1544		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1545		p->asc = 0x39;
1546		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1547		break;
1548error:
1549	case 1:
1550	case 2:
1551		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1552		p->asc = 0x24;
1553		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1554		break;
1555	}
1556	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1557	ahci_write_fis_d2h(p, slot, cfis, tfd);
1558}
1559
1560static void
1561atapi_get_event_status_notification(struct ahci_port *p, int slot,
1562    uint8_t *cfis)
1563{
1564	uint8_t *acmd;
1565	uint32_t tfd;
1566
1567	acmd = cfis + 0x40;
1568
1569	/* we don't support asynchronous operation */
1570	if (!(acmd[1] & 1)) {
1571		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1572		p->asc = 0x24;
1573		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1574	} else {
1575		uint8_t buf[8];
1576		int len;
1577
1578		len = be16dec(acmd + 7);
1579		if (len > sizeof(buf))
1580			len = sizeof(buf);
1581
1582		memset(buf, 0, sizeof(buf));
1583		be16enc(buf, 8 - 2);
1584		buf[2] = 0x04;
1585		buf[3] = 0x10;
1586		buf[5] = 0x02;
1587		write_prdt(p, slot, cfis, buf, len);
1588		tfd = ATA_S_READY | ATA_S_DSC;
1589	}
1590	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1591	ahci_write_fis_d2h(p, slot, cfis, tfd);
1592}
1593
1594static void
1595handle_packet_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1596{
1597	uint8_t *acmd;
1598
1599	acmd = cfis + 0x40;
1600
1601#ifdef AHCI_DEBUG
1602	{
1603		int i;
1604		DPRINTF("ACMD:");
1605		for (i = 0; i < 16; i++)
1606			DPRINTF("%02x ", acmd[i]);
1607		DPRINTF("\n");
1608	}
1609#endif
1610
1611	switch (acmd[0]) {
1612	case TEST_UNIT_READY:
1613		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1614		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1615		break;
1616	case INQUIRY:
1617		atapi_inquiry(p, slot, cfis);
1618		break;
1619	case READ_CAPACITY:
1620		atapi_read_capacity(p, slot, cfis);
1621		break;
1622	case PREVENT_ALLOW:
1623		/* TODO */
1624		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1625		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1626		break;
1627	case READ_TOC:
1628		atapi_read_toc(p, slot, cfis);
1629		break;
1630	case REPORT_LUNS:
1631		atapi_report_luns(p, slot, cfis);
1632		break;
1633	case READ_10:
1634	case READ_12:
1635		atapi_read(p, slot, cfis, 0);
1636		break;
1637	case REQUEST_SENSE:
1638		atapi_request_sense(p, slot, cfis);
1639		break;
1640	case START_STOP_UNIT:
1641		atapi_start_stop_unit(p, slot, cfis);
1642		break;
1643	case MODE_SENSE_10:
1644		atapi_mode_sense(p, slot, cfis);
1645		break;
1646	case GET_EVENT_STATUS_NOTIFICATION:
1647		atapi_get_event_status_notification(p, slot, cfis);
1648		break;
1649	default:
1650		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1651		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1652		p->asc = 0x20;
1653		ahci_write_fis_d2h(p, slot, cfis, (p->sense_key << 12) |
1654				ATA_S_READY | ATA_S_ERROR);
1655		break;
1656	}
1657}
1658
1659static void
1660ahci_handle_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1661{
1662
1663	p->tfd |= ATA_S_BUSY;
1664	switch (cfis[2]) {
1665	case ATA_ATA_IDENTIFY:
1666		handle_identify(p, slot, cfis);
1667		break;
1668	case ATA_SETFEATURES:
1669	{
1670		switch (cfis[3]) {
1671		case ATA_SF_ENAB_SATA_SF:
1672			switch (cfis[12]) {
1673			case ATA_SATA_SF_AN:
1674				p->tfd = ATA_S_DSC | ATA_S_READY;
1675				break;
1676			default:
1677				p->tfd = ATA_S_ERROR | ATA_S_READY;
1678				p->tfd |= (ATA_ERROR_ABORT << 8);
1679				break;
1680			}
1681			break;
1682		case ATA_SF_ENAB_WCACHE:
1683		case ATA_SF_DIS_WCACHE:
1684		case ATA_SF_ENAB_RCACHE:
1685		case ATA_SF_DIS_RCACHE:
1686			p->tfd = ATA_S_DSC | ATA_S_READY;
1687			break;
1688		case ATA_SF_SETXFER:
1689		{
1690			switch (cfis[12] & 0xf8) {
1691			case ATA_PIO:
1692			case ATA_PIO0:
1693				break;
1694			case ATA_WDMA0:
1695			case ATA_UDMA0:
1696				p->xfermode = (cfis[12] & 0x7);
1697				break;
1698			}
1699			p->tfd = ATA_S_DSC | ATA_S_READY;
1700			break;
1701		}
1702		default:
1703			p->tfd = ATA_S_ERROR | ATA_S_READY;
1704			p->tfd |= (ATA_ERROR_ABORT << 8);
1705			break;
1706		}
1707		ahci_write_fis_d2h(p, slot, cfis, p->tfd);
1708		break;
1709	}
1710	case ATA_SET_MULTI:
1711		if (cfis[12] != 0 &&
1712			(cfis[12] > 128 || (cfis[12] & (cfis[12] - 1)))) {
1713			p->tfd = ATA_S_ERROR | ATA_S_READY;
1714			p->tfd |= (ATA_ERROR_ABORT << 8);
1715		} else {
1716			p->mult_sectors = cfis[12];
1717			p->tfd = ATA_S_DSC | ATA_S_READY;
1718		}
1719		ahci_write_fis_d2h(p, slot, cfis, p->tfd);
1720		break;
1721	case ATA_READ:
1722	case ATA_WRITE:
1723	case ATA_READ48:
1724	case ATA_WRITE48:
1725	case ATA_READ_MUL:
1726	case ATA_WRITE_MUL:
1727	case ATA_READ_MUL48:
1728	case ATA_WRITE_MUL48:
1729	case ATA_READ_DMA:
1730	case ATA_WRITE_DMA:
1731	case ATA_READ_DMA48:
1732	case ATA_WRITE_DMA48:
1733	case ATA_READ_FPDMA_QUEUED:
1734	case ATA_WRITE_FPDMA_QUEUED:
1735		ahci_handle_rw(p, slot, cfis, 0);
1736		break;
1737	case ATA_FLUSHCACHE:
1738	case ATA_FLUSHCACHE48:
1739		ahci_handle_flush(p, slot, cfis);
1740		break;
1741	case ATA_DATA_SET_MANAGEMENT:
1742		if (cfis[11] == 0 && cfis[3] == ATA_DSM_TRIM &&
1743		    cfis[13] == 0 && cfis[12] == 1) {
1744			ahci_handle_dsm_trim(p, slot, cfis, 0);
1745			break;
1746		}
1747		ahci_write_fis_d2h(p, slot, cfis,
1748		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1749		break;
1750	case ATA_SEND_FPDMA_QUEUED:
1751		if ((cfis[13] & 0x1f) == ATA_SFPDMA_DSM &&
1752		    cfis[17] == 0 && cfis[16] == ATA_DSM_TRIM &&
1753		    cfis[11] == 0 && cfis[3] == 1) {
1754			ahci_handle_dsm_trim(p, slot, cfis, 0);
1755			break;
1756		}
1757		ahci_write_fis_d2h(p, slot, cfis,
1758		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1759		break;
1760	case ATA_READ_LOG_EXT:
1761	case ATA_READ_LOG_DMA_EXT:
1762		ahci_handle_read_log(p, slot, cfis);
1763		break;
1764	case ATA_SECURITY_FREEZE_LOCK:
1765	case ATA_SMART_CMD:
1766	case ATA_NOP:
1767		ahci_write_fis_d2h(p, slot, cfis,
1768		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1769		break;
1770	case ATA_CHECK_POWER_MODE:
1771		cfis[12] = 0xff;	/* always on */
1772		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1773		break;
1774	case ATA_STANDBY_CMD:
1775	case ATA_STANDBY_IMMEDIATE:
1776	case ATA_IDLE_CMD:
1777	case ATA_IDLE_IMMEDIATE:
1778	case ATA_SLEEP:
1779	case ATA_READ_VERIFY:
1780	case ATA_READ_VERIFY48:
1781		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1782		break;
1783	case ATA_ATAPI_IDENTIFY:
1784		handle_atapi_identify(p, slot, cfis);
1785		break;
1786	case ATA_PACKET_CMD:
1787		if (!p->atapi) {
1788			ahci_write_fis_d2h(p, slot, cfis,
1789			    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1790		} else
1791			handle_packet_cmd(p, slot, cfis);
1792		break;
1793	default:
1794		WPRINTF("Unsupported cmd:%02x\n", cfis[2]);
1795		ahci_write_fis_d2h(p, slot, cfis,
1796		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1797		break;
1798	}
1799}
1800
1801static void
1802ahci_handle_slot(struct ahci_port *p, int slot)
1803{
1804	struct ahci_cmd_hdr *hdr;
1805#ifdef AHCI_DEBUG
1806	struct ahci_prdt_entry *prdt;
1807#endif
1808	struct pci_ahci_softc *sc;
1809	uint8_t *cfis;
1810#ifdef AHCI_DEBUG
1811	int cfl, i;
1812#endif
1813
1814	sc = p->pr_sc;
1815	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1816#ifdef AHCI_DEBUG
1817	cfl = (hdr->flags & 0x1f) * 4;
1818#endif
1819	cfis = paddr_guest2host(ahci_ctx(sc), hdr->ctba,
1820			0x80 + hdr->prdtl * sizeof(struct ahci_prdt_entry));
1821#ifdef AHCI_DEBUG
1822	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
1823
1824	DPRINTF("\ncfis:");
1825	for (i = 0; i < cfl; i++) {
1826		if (i % 10 == 0)
1827			DPRINTF("\n");
1828		DPRINTF("%02x ", cfis[i]);
1829	}
1830	DPRINTF("\n");
1831
1832	for (i = 0; i < hdr->prdtl; i++) {
1833		DPRINTF("%d@%08"PRIx64"\n", prdt->dbc & 0x3fffff, prdt->dba);
1834		prdt++;
1835	}
1836#endif
1837
1838	if (cfis[0] != FIS_TYPE_REGH2D) {
1839		WPRINTF("Not a H2D FIS:%02x\n", cfis[0]);
1840		return;
1841	}
1842
1843	if (cfis[1] & 0x80) {
1844		ahci_handle_cmd(p, slot, cfis);
1845	} else {
1846		if (cfis[15] & (1 << 2))
1847			p->reset = 1;
1848		else if (p->reset) {
1849			p->reset = 0;
1850			ahci_port_reset(p);
1851		}
1852		p->ci &= ~(1 << slot);
1853	}
1854}
1855
1856static void
1857ahci_handle_port(struct ahci_port *p)
1858{
1859
1860	if (!(p->cmd & AHCI_P_CMD_ST))
1861		return;
1862
1863	/*
1864	 * Search for any new commands to issue ignoring those that
1865	 * are already in-flight.  Stop if device is busy or in error.
1866	 */
1867	for (; (p->ci & ~p->pending) != 0; p->ccs = ((p->ccs + 1) & 31)) {
1868		if ((p->tfd & (ATA_S_BUSY | ATA_S_DRQ)) != 0)
1869			break;
1870		if (p->waitforclear)
1871			break;
1872		if ((p->ci & ~p->pending & (1 << p->ccs)) != 0) {
1873			p->cmd &= ~AHCI_P_CMD_CCS_MASK;
1874			p->cmd |= p->ccs << AHCI_P_CMD_CCS_SHIFT;
1875			ahci_handle_slot(p, p->ccs);
1876		}
1877	}
1878}
1879
1880/*
1881 * blockif callback routine - this runs in the context of the blockif
1882 * i/o thread, so the mutex needs to be acquired.
1883 */
1884static void
1885ata_ioreq_cb(struct blockif_req *br, int err)
1886{
1887	struct ahci_cmd_hdr *hdr;
1888	struct ahci_ioreq *aior;
1889	struct ahci_port *p;
1890	struct pci_ahci_softc *sc;
1891	uint32_t tfd;
1892	uint8_t *cfis;
1893	int slot, ncq, dsm;
1894
1895	DPRINTF("%s %d\n", __func__, err);
1896
1897	ncq = dsm = 0;
1898	aior = br->br_param;
1899	p = aior->io_pr;
1900	cfis = aior->cfis;
1901	slot = aior->slot;
1902	sc = p->pr_sc;
1903	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1904
1905	if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
1906	    cfis[2] == ATA_READ_FPDMA_QUEUED ||
1907	    cfis[2] == ATA_SEND_FPDMA_QUEUED)
1908		ncq = 1;
1909	if (cfis[2] == ATA_DATA_SET_MANAGEMENT ||
1910	    (cfis[2] == ATA_SEND_FPDMA_QUEUED &&
1911	     (cfis[13] & 0x1f) == ATA_SFPDMA_DSM))
1912		dsm = 1;
1913
1914	pthread_mutex_lock(&sc->mtx);
1915
1916	/*
1917	 * Delete the blockif request from the busy list
1918	 */
1919	TAILQ_REMOVE(&p->iobhd, aior, io_blist);
1920
1921	/*
1922	 * Move the blockif request back to the free list
1923	 */
1924	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_flist);
1925
1926	if (!err)
1927		hdr->prdbc = aior->done;
1928
1929	if (!err && aior->more) {
1930		if (dsm)
1931			ahci_handle_dsm_trim(p, slot, cfis, aior->done);
1932		else
1933			ahci_handle_rw(p, slot, cfis, aior->done);
1934		goto out;
1935	}
1936
1937	if (!err)
1938		tfd = ATA_S_READY | ATA_S_DSC;
1939	else
1940		tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1941	if (ncq)
1942		ahci_write_fis_sdb(p, slot, cfis, tfd);
1943	else
1944		ahci_write_fis_d2h(p, slot, cfis, tfd);
1945
1946	/*
1947	 * This command is now complete.
1948	 */
1949	p->pending &= ~(1 << slot);
1950
1951	ahci_check_stopped(p);
1952	ahci_handle_port(p);
1953out:
1954	pthread_mutex_unlock(&sc->mtx);
1955	DPRINTF("%s exit\n", __func__);
1956}
1957
1958static void
1959atapi_ioreq_cb(struct blockif_req *br, int err)
1960{
1961	struct ahci_cmd_hdr *hdr;
1962	struct ahci_ioreq *aior;
1963	struct ahci_port *p;
1964	struct pci_ahci_softc *sc;
1965	uint8_t *cfis;
1966	uint32_t tfd;
1967	int slot;
1968
1969	DPRINTF("%s %d\n", __func__, err);
1970
1971	aior = br->br_param;
1972	p = aior->io_pr;
1973	cfis = aior->cfis;
1974	slot = aior->slot;
1975	sc = p->pr_sc;
1976	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + aior->slot * AHCI_CL_SIZE);
1977
1978	pthread_mutex_lock(&sc->mtx);
1979
1980	/*
1981	 * Delete the blockif request from the busy list
1982	 */
1983	TAILQ_REMOVE(&p->iobhd, aior, io_blist);
1984
1985	/*
1986	 * Move the blockif request back to the free list
1987	 */
1988	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_flist);
1989
1990	if (!err)
1991		hdr->prdbc = aior->done;
1992
1993	if (!err && aior->more) {
1994		atapi_read(p, slot, cfis, aior->done);
1995		goto out;
1996	}
1997
1998	if (!err) {
1999		tfd = ATA_S_READY | ATA_S_DSC;
2000	} else {
2001		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
2002		p->asc = 0x21;
2003		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
2004	}
2005	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
2006	ahci_write_fis_d2h(p, slot, cfis, tfd);
2007
2008	/*
2009	 * This command is now complete.
2010	 */
2011	p->pending &= ~(1 << slot);
2012
2013	ahci_check_stopped(p);
2014	ahci_handle_port(p);
2015out:
2016	pthread_mutex_unlock(&sc->mtx);
2017	DPRINTF("%s exit\n", __func__);
2018}
2019
2020static void
2021pci_ahci_ioreq_init(struct ahci_port *pr)
2022{
2023	struct ahci_ioreq *vr;
2024	int i;
2025
2026	pr->ioqsz = blockif_queuesz(pr->bctx);
2027	pr->ioreq = calloc(pr->ioqsz, sizeof(struct ahci_ioreq));
2028	STAILQ_INIT(&pr->iofhd);
2029
2030	/*
2031	 * Add all i/o request entries to the free queue
2032	 */
2033	for (i = 0; i < pr->ioqsz; i++) {
2034		vr = &pr->ioreq[i];
2035		vr->io_pr = pr;
2036		if (!pr->atapi)
2037			vr->io_req.br_callback = ata_ioreq_cb;
2038		else
2039			vr->io_req.br_callback = atapi_ioreq_cb;
2040		vr->io_req.br_param = vr;
2041		STAILQ_INSERT_TAIL(&pr->iofhd, vr, io_flist);
2042	}
2043
2044	TAILQ_INIT(&pr->iobhd);
2045}
2046
2047static void
2048pci_ahci_port_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
2049{
2050	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
2051	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
2052	struct ahci_port *p = &sc->port[port];
2053
2054	DPRINTF("pci_ahci_port %d: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
2055		port, offset, value);
2056
2057	switch (offset) {
2058	case AHCI_P_CLB:
2059		p->clb = value;
2060		break;
2061	case AHCI_P_CLBU:
2062		p->clbu = value;
2063		break;
2064	case AHCI_P_FB:
2065		p->fb = value;
2066		break;
2067	case AHCI_P_FBU:
2068		p->fbu = value;
2069		break;
2070	case AHCI_P_IS:
2071		p->is &= ~value;
2072		ahci_port_intr(p);
2073		break;
2074	case AHCI_P_IE:
2075		p->ie = value & 0xFDC000FF;
2076		ahci_port_intr(p);
2077		break;
2078	case AHCI_P_CMD:
2079	{
2080		p->cmd &= ~(AHCI_P_CMD_ST | AHCI_P_CMD_SUD | AHCI_P_CMD_POD |
2081		    AHCI_P_CMD_CLO | AHCI_P_CMD_FRE | AHCI_P_CMD_APSTE |
2082		    AHCI_P_CMD_ATAPI | AHCI_P_CMD_DLAE | AHCI_P_CMD_ALPE |
2083		    AHCI_P_CMD_ASP | AHCI_P_CMD_ICC_MASK);
2084		p->cmd |= (AHCI_P_CMD_ST | AHCI_P_CMD_SUD | AHCI_P_CMD_POD |
2085		    AHCI_P_CMD_CLO | AHCI_P_CMD_FRE | AHCI_P_CMD_APSTE |
2086		    AHCI_P_CMD_ATAPI | AHCI_P_CMD_DLAE | AHCI_P_CMD_ALPE |
2087		    AHCI_P_CMD_ASP | AHCI_P_CMD_ICC_MASK) & value;
2088
2089		if (!(value & AHCI_P_CMD_ST)) {
2090			ahci_port_stop(p);
2091		} else {
2092			uint64_t clb;
2093
2094			p->cmd |= AHCI_P_CMD_CR;
2095			clb = (uint64_t)p->clbu << 32 | p->clb;
2096			p->cmd_lst = paddr_guest2host(ahci_ctx(sc), clb,
2097					AHCI_CL_SIZE * AHCI_MAX_SLOTS);
2098		}
2099
2100		if (value & AHCI_P_CMD_FRE) {
2101			uint64_t fb;
2102
2103			p->cmd |= AHCI_P_CMD_FR;
2104			fb = (uint64_t)p->fbu << 32 | p->fb;
2105			/* we don't support FBSCP, so rfis size is 256Bytes */
2106			p->rfis = paddr_guest2host(ahci_ctx(sc), fb, 256);
2107		} else {
2108			p->cmd &= ~AHCI_P_CMD_FR;
2109		}
2110
2111		if (value & AHCI_P_CMD_CLO) {
2112			p->tfd &= ~(ATA_S_BUSY | ATA_S_DRQ);
2113			p->cmd &= ~AHCI_P_CMD_CLO;
2114		}
2115
2116		if (value & AHCI_P_CMD_ICC_MASK) {
2117			p->cmd &= ~AHCI_P_CMD_ICC_MASK;
2118		}
2119
2120		ahci_handle_port(p);
2121		break;
2122	}
2123	case AHCI_P_TFD:
2124	case AHCI_P_SIG:
2125	case AHCI_P_SSTS:
2126		WPRINTF("pci_ahci_port: read only registers 0x%"PRIx64"\n", offset);
2127		break;
2128	case AHCI_P_SCTL:
2129		p->sctl = value;
2130		if (!(p->cmd & AHCI_P_CMD_ST)) {
2131			if (value & ATA_SC_DET_RESET)
2132				ahci_port_reset(p);
2133		}
2134		break;
2135	case AHCI_P_SERR:
2136		p->serr &= ~value;
2137		break;
2138	case AHCI_P_SACT:
2139		p->sact |= value;
2140		break;
2141	case AHCI_P_CI:
2142		p->ci |= value;
2143		ahci_handle_port(p);
2144		break;
2145	case AHCI_P_SNTF:
2146	case AHCI_P_FBS:
2147	default:
2148		break;
2149	}
2150}
2151
2152static void
2153pci_ahci_host_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
2154{
2155	DPRINTF("pci_ahci_host: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
2156		offset, value);
2157
2158	switch (offset) {
2159	case AHCI_CAP:
2160	case AHCI_PI:
2161	case AHCI_VS:
2162	case AHCI_CAP2:
2163		DPRINTF("pci_ahci_host: read only registers 0x%"PRIx64"\n", offset);
2164		break;
2165	case AHCI_GHC:
2166		if (value & AHCI_GHC_HR) {
2167			ahci_reset(sc);
2168			break;
2169		}
2170		if (value & AHCI_GHC_IE)
2171			sc->ghc |= AHCI_GHC_IE;
2172		else
2173			sc->ghc &= ~AHCI_GHC_IE;
2174		ahci_generate_intr(sc, 0xffffffff);
2175		break;
2176	case AHCI_IS:
2177		sc->is &= ~value;
2178		ahci_generate_intr(sc, value);
2179		break;
2180	default:
2181		break;
2182	}
2183}
2184
2185static void
2186pci_ahci_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
2187		int baridx, uint64_t offset, int size, uint64_t value)
2188{
2189	struct pci_ahci_softc *sc = pi->pi_arg;
2190
2191	assert(baridx == 5);
2192	assert((offset % 4) == 0 && size == 4);
2193
2194	pthread_mutex_lock(&sc->mtx);
2195
2196	if (offset < AHCI_OFFSET)
2197		pci_ahci_host_write(sc, offset, value);
2198	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
2199		pci_ahci_port_write(sc, offset, value);
2200	else
2201		WPRINTF("pci_ahci: unknown i/o write offset 0x%"PRIx64"\n", offset);
2202
2203	pthread_mutex_unlock(&sc->mtx);
2204}
2205
2206static uint64_t
2207pci_ahci_host_read(struct pci_ahci_softc *sc, uint64_t offset)
2208{
2209	uint32_t value;
2210
2211	switch (offset) {
2212	case AHCI_CAP:
2213	case AHCI_GHC:
2214	case AHCI_IS:
2215	case AHCI_PI:
2216	case AHCI_VS:
2217	case AHCI_CCCC:
2218	case AHCI_CCCP:
2219	case AHCI_EM_LOC:
2220	case AHCI_EM_CTL:
2221	case AHCI_CAP2:
2222	{
2223		uint32_t *p = &sc->cap;
2224		p += (offset - AHCI_CAP) / sizeof(uint32_t);
2225		value = *p;
2226		break;
2227	}
2228	default:
2229		value = 0;
2230		break;
2231	}
2232	DPRINTF("pci_ahci_host: read offset 0x%"PRIx64" value 0x%x\n",
2233		offset, value);
2234
2235	return (value);
2236}
2237
2238static uint64_t
2239pci_ahci_port_read(struct pci_ahci_softc *sc, uint64_t offset)
2240{
2241	uint32_t value;
2242	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
2243	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
2244
2245	switch (offset) {
2246	case AHCI_P_CLB:
2247	case AHCI_P_CLBU:
2248	case AHCI_P_FB:
2249	case AHCI_P_FBU:
2250	case AHCI_P_IS:
2251	case AHCI_P_IE:
2252	case AHCI_P_CMD:
2253	case AHCI_P_TFD:
2254	case AHCI_P_SIG:
2255	case AHCI_P_SSTS:
2256	case AHCI_P_SCTL:
2257	case AHCI_P_SERR:
2258	case AHCI_P_SACT:
2259	case AHCI_P_CI:
2260	case AHCI_P_SNTF:
2261	case AHCI_P_FBS:
2262	{
2263		uint32_t *p= &sc->port[port].clb;
2264		p += (offset - AHCI_P_CLB) / sizeof(uint32_t);
2265		value = *p;
2266		break;
2267	}
2268	default:
2269		value = 0;
2270		break;
2271	}
2272
2273	DPRINTF("pci_ahci_port %d: read offset 0x%"PRIx64" value 0x%x\n",
2274		port, offset, value);
2275
2276	return value;
2277}
2278
2279static uint64_t
2280pci_ahci_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2281    uint64_t regoff, int size)
2282{
2283	struct pci_ahci_softc *sc = pi->pi_arg;
2284	uint64_t offset;
2285	uint32_t value;
2286
2287	assert(baridx == 5);
2288	assert(size == 1 || size == 2 || size == 4);
2289	assert((regoff & (size - 1)) == 0);
2290
2291	pthread_mutex_lock(&sc->mtx);
2292
2293	offset = regoff & ~0x3;	    /* round down to a multiple of 4 bytes */
2294	if (offset < AHCI_OFFSET)
2295		value = pci_ahci_host_read(sc, offset);
2296	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
2297		value = pci_ahci_port_read(sc, offset);
2298	else {
2299		value = 0;
2300		WPRINTF("pci_ahci: unknown i/o read offset 0x%"PRIx64"\n",
2301		    regoff);
2302	}
2303	value >>= 8 * (regoff & 0x3);
2304
2305	pthread_mutex_unlock(&sc->mtx);
2306
2307	return (value);
2308}
2309
2310static int
2311pci_ahci_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts, int atapi)
2312{
2313	char bident[sizeof("XX:XX:XX")];
2314	struct blockif_ctxt *bctxt;
2315	struct pci_ahci_softc *sc;
2316	int ret, slots, p;
2317	MD5_CTX mdctx;
2318	u_char digest[16];
2319	char *next, *next2;
2320
2321	ret = 0;
2322
2323#ifdef AHCI_DEBUG
2324	dbg = fopen("/tmp/log", "w+");
2325#endif
2326
2327	sc = calloc(1, sizeof(struct pci_ahci_softc));
2328	pi->pi_arg = sc;
2329	sc->asc_pi = pi;
2330	pthread_mutex_init(&sc->mtx, NULL);
2331	sc->ports = 0;
2332	sc->pi = 0;
2333	slots = 32;
2334
2335	for (p = 0; p < MAX_PORTS && opts != NULL; p++, opts = next) {
2336		/* Identify and cut off type of present port. */
2337		if (strncmp(opts, "hd:", 3) == 0) {
2338			atapi = 0;
2339			opts += 3;
2340		} else if (strncmp(opts, "cd:", 3) == 0) {
2341			atapi = 1;
2342			opts += 3;
2343		}
2344
2345		/* Find and cut off the next port options. */
2346		next = strstr(opts, ",hd:");
2347		next2 = strstr(opts, ",cd:");
2348		if (next == NULL || (next2 != NULL && next2 < next))
2349			next = next2;
2350		if (next != NULL) {
2351			next[0] = 0;
2352			next++;
2353		}
2354
2355		if (opts[0] == 0)
2356			continue;
2357
2358		/*
2359		 * Attempt to open the backing image. Use the PCI slot/func
2360		 * and the port number for the identifier string.
2361		 */
2362		snprintf(bident, sizeof(bident), "%d:%d:%d", pi->pi_slot,
2363		    pi->pi_func, p);
2364		bctxt = blockif_open(opts, bident);
2365		if (bctxt == NULL) {
2366			sc->ports = p;
2367			ret = 1;
2368			goto open_fail;
2369		}
2370		sc->port[p].bctx = bctxt;
2371		sc->port[p].pr_sc = sc;
2372		sc->port[p].port = p;
2373		sc->port[p].atapi = atapi;
2374
2375		/*
2376		 * Create an identifier for the backing file.
2377		 * Use parts of the md5 sum of the filename
2378		 */
2379		MD5Init(&mdctx);
2380		MD5Update(&mdctx, opts, strlen(opts));
2381		MD5Final(digest, &mdctx);
2382		sprintf(sc->port[p].ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
2383		    digest[0], digest[1], digest[2], digest[3], digest[4],
2384		    digest[5]);
2385
2386		/*
2387		 * Allocate blockif request structures and add them
2388		 * to the free list
2389		 */
2390		pci_ahci_ioreq_init(&sc->port[p]);
2391
2392		sc->pi |= (1 << p);
2393		if (sc->port[p].ioqsz < slots)
2394			slots = sc->port[p].ioqsz;
2395	}
2396	sc->ports = p;
2397
2398	/* Intel ICH8 AHCI */
2399	--slots;
2400	if (sc->ports < DEF_PORTS)
2401		sc->ports = DEF_PORTS;
2402	sc->cap = AHCI_CAP_64BIT | AHCI_CAP_SNCQ | AHCI_CAP_SSNTF |
2403	    AHCI_CAP_SMPS | AHCI_CAP_SSS | AHCI_CAP_SALP |
2404	    AHCI_CAP_SAL | AHCI_CAP_SCLO | (0x3 << AHCI_CAP_ISS_SHIFT)|
2405	    AHCI_CAP_PMD | AHCI_CAP_SSC | AHCI_CAP_PSC |
2406	    (slots << AHCI_CAP_NCS_SHIFT) | AHCI_CAP_SXS | (sc->ports - 1);
2407
2408	sc->vs = 0x10300;
2409	sc->cap2 = AHCI_CAP2_APST;
2410	ahci_reset(sc);
2411
2412	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x2821);
2413	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x8086);
2414	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
2415	pci_set_cfgdata8(pi, PCIR_SUBCLASS, PCIS_STORAGE_SATA);
2416	pci_set_cfgdata8(pi, PCIR_PROGIF, PCIP_STORAGE_SATA_AHCI_1_0);
2417	p = MIN(sc->ports, 16);
2418	p = flsl(p) - ((p & (p - 1)) ? 0 : 1);
2419	pci_emul_add_msicap(pi, 1 << p);
2420	pci_emul_alloc_bar(pi, 5, PCIBAR_MEM32,
2421	    AHCI_OFFSET + sc->ports * AHCI_STEP);
2422
2423	pci_lintr_request(pi);
2424
2425open_fail:
2426	if (ret) {
2427		for (p = 0; p < sc->ports; p++) {
2428			if (sc->port[p].bctx != NULL)
2429				blockif_close(sc->port[p].bctx);
2430		}
2431		free(sc);
2432	}
2433
2434	return (ret);
2435}
2436
2437static int
2438pci_ahci_hd_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
2439{
2440
2441	return (pci_ahci_init(ctx, pi, opts, 0));
2442}
2443
2444static int
2445pci_ahci_atapi_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
2446{
2447
2448	return (pci_ahci_init(ctx, pi, opts, 1));
2449}
2450
2451/*
2452 * Use separate emulation names to distinguish drive and atapi devices
2453 */
2454struct pci_devemu pci_de_ahci = {
2455	.pe_emu =	"ahci",
2456	.pe_init =	pci_ahci_hd_init,
2457	.pe_barwrite =	pci_ahci_write,
2458	.pe_barread =	pci_ahci_read
2459};
2460PCI_EMUL_SET(pci_de_ahci);
2461
2462struct pci_devemu pci_de_ahci_hd = {
2463	.pe_emu =	"ahci-hd",
2464	.pe_init =	pci_ahci_hd_init,
2465	.pe_barwrite =	pci_ahci_write,
2466	.pe_barread =	pci_ahci_read
2467};
2468PCI_EMUL_SET(pci_de_ahci_hd);
2469
2470struct pci_devemu pci_de_ahci_cd = {
2471	.pe_emu =	"ahci-cd",
2472	.pe_init =	pci_ahci_atapi_init,
2473	.pe_barwrite =	pci_ahci_write,
2474	.pe_barread =	pci_ahci_read
2475};
2476PCI_EMUL_SET(pci_de_ahci_cd);
2477