1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2013  Zhixiang Yu <zcore@freebsd.org>
5 * Copyright (c) 2015-2016 Alexander Motin <mav@FreeBSD.org>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * $FreeBSD: stable/11/usr.sbin/bhyve/pci_ahci.c 347436 2019-05-10 16:36:38Z jhb $
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: stable/11/usr.sbin/bhyve/pci_ahci.c 347436 2019-05-10 16:36:38Z jhb $");
34
35#include <sys/param.h>
36#include <sys/linker_set.h>
37#include <sys/stat.h>
38#include <sys/uio.h>
39#include <sys/ioctl.h>
40#include <sys/disk.h>
41#include <sys/ata.h>
42#include <sys/endian.h>
43
44#include <errno.h>
45#include <fcntl.h>
46#include <stdio.h>
47#include <stdlib.h>
48#include <stdint.h>
49#include <string.h>
50#include <strings.h>
51#include <unistd.h>
52#include <assert.h>
53#include <pthread.h>
54#include <pthread_np.h>
55#include <inttypes.h>
56#include <md5.h>
57
58#include "bhyverun.h"
59#include "pci_emul.h"
60#include "ahci.h"
61#include "block_if.h"
62
63#define	DEF_PORTS	6	/* Intel ICH8 AHCI supports 6 ports */
64#define	MAX_PORTS	32	/* AHCI supports 32 ports */
65
66#define	PxSIG_ATA	0x00000101 /* ATA drive */
67#define	PxSIG_ATAPI	0xeb140101 /* ATAPI drive */
68
69enum sata_fis_type {
70	FIS_TYPE_REGH2D		= 0x27,	/* Register FIS - host to device */
71	FIS_TYPE_REGD2H		= 0x34,	/* Register FIS - device to host */
72	FIS_TYPE_DMAACT		= 0x39,	/* DMA activate FIS - device to host */
73	FIS_TYPE_DMASETUP	= 0x41,	/* DMA setup FIS - bidirectional */
74	FIS_TYPE_DATA		= 0x46,	/* Data FIS - bidirectional */
75	FIS_TYPE_BIST		= 0x58,	/* BIST activate FIS - bidirectional */
76	FIS_TYPE_PIOSETUP	= 0x5F,	/* PIO setup FIS - device to host */
77	FIS_TYPE_SETDEVBITS	= 0xA1,	/* Set dev bits FIS - device to host */
78};
79
80/*
81 * SCSI opcodes
82 */
83#define	TEST_UNIT_READY		0x00
84#define	REQUEST_SENSE		0x03
85#define	INQUIRY			0x12
86#define	START_STOP_UNIT		0x1B
87#define	PREVENT_ALLOW		0x1E
88#define	READ_CAPACITY		0x25
89#define	READ_10			0x28
90#define	POSITION_TO_ELEMENT	0x2B
91#define	READ_TOC		0x43
92#define	GET_EVENT_STATUS_NOTIFICATION 0x4A
93#define	MODE_SENSE_10		0x5A
94#define	REPORT_LUNS		0xA0
95#define	READ_12			0xA8
96#define	READ_CD			0xBE
97
98/*
99 * SCSI mode page codes
100 */
101#define	MODEPAGE_RW_ERROR_RECOVERY	0x01
102#define	MODEPAGE_CD_CAPABILITIES	0x2A
103
104/*
105 * ATA commands
106 */
107#define	ATA_SF_ENAB_SATA_SF		0x10
108#define	ATA_SATA_SF_AN			0x05
109#define	ATA_SF_DIS_SATA_SF		0x90
110
111/*
112 * Debug printf
113 */
114#ifdef AHCI_DEBUG
115static FILE *dbg;
116#define DPRINTF(format, arg...)	do{fprintf(dbg, format, ##arg);fflush(dbg);}while(0)
117#else
118#define DPRINTF(format, arg...)
119#endif
120#define WPRINTF(format, arg...) printf(format, ##arg)
121
122#define AHCI_PORT_IDENT 20 + 1
123
124struct ahci_ioreq {
125	struct blockif_req io_req;
126	struct ahci_port *io_pr;
127	STAILQ_ENTRY(ahci_ioreq) io_flist;
128	TAILQ_ENTRY(ahci_ioreq) io_blist;
129	uint8_t *cfis;
130	uint32_t len;
131	uint32_t done;
132	int slot;
133	int more;
134};
135
136struct ahci_port {
137	struct blockif_ctxt *bctx;
138	struct pci_ahci_softc *pr_sc;
139	uint8_t *cmd_lst;
140	uint8_t *rfis;
141	char ident[AHCI_PORT_IDENT];
142	int port;
143	int atapi;
144	int reset;
145	int waitforclear;
146	int mult_sectors;
147	uint8_t xfermode;
148	uint8_t err_cfis[20];
149	uint8_t sense_key;
150	uint8_t asc;
151	u_int ccs;
152	uint32_t pending;
153
154	uint32_t clb;
155	uint32_t clbu;
156	uint32_t fb;
157	uint32_t fbu;
158	uint32_t is;
159	uint32_t ie;
160	uint32_t cmd;
161	uint32_t unused0;
162	uint32_t tfd;
163	uint32_t sig;
164	uint32_t ssts;
165	uint32_t sctl;
166	uint32_t serr;
167	uint32_t sact;
168	uint32_t ci;
169	uint32_t sntf;
170	uint32_t fbs;
171
172	/*
173	 * i/o request info
174	 */
175	struct ahci_ioreq *ioreq;
176	int ioqsz;
177	STAILQ_HEAD(ahci_fhead, ahci_ioreq) iofhd;
178	TAILQ_HEAD(ahci_bhead, ahci_ioreq) iobhd;
179};
180
181struct ahci_cmd_hdr {
182	uint16_t flags;
183	uint16_t prdtl;
184	uint32_t prdbc;
185	uint64_t ctba;
186	uint32_t reserved[4];
187};
188
189struct ahci_prdt_entry {
190	uint64_t dba;
191	uint32_t reserved;
192#define	DBCMASK		0x3fffff
193	uint32_t dbc;
194};
195
196struct pci_ahci_softc {
197	struct pci_devinst *asc_pi;
198	pthread_mutex_t	mtx;
199	int ports;
200	uint32_t cap;
201	uint32_t ghc;
202	uint32_t is;
203	uint32_t pi;
204	uint32_t vs;
205	uint32_t ccc_ctl;
206	uint32_t ccc_pts;
207	uint32_t em_loc;
208	uint32_t em_ctl;
209	uint32_t cap2;
210	uint32_t bohc;
211	uint32_t lintr;
212	struct ahci_port port[MAX_PORTS];
213};
214#define	ahci_ctx(sc)	((sc)->asc_pi->pi_vmctx)
215
216static void ahci_handle_port(struct ahci_port *p);
217
218static inline void lba_to_msf(uint8_t *buf, int lba)
219{
220	lba += 150;
221	buf[0] = (lba / 75) / 60;
222	buf[1] = (lba / 75) % 60;
223	buf[2] = lba % 75;
224}
225
226/*
227 * Generate HBA interrupts on global IS register write.
228 */
229static void
230ahci_generate_intr(struct pci_ahci_softc *sc, uint32_t mask)
231{
232	struct pci_devinst *pi = sc->asc_pi;
233	struct ahci_port *p;
234	int i, nmsg;
235	uint32_t mmask;
236
237	/* Update global IS from PxIS/PxIE. */
238	for (i = 0; i < sc->ports; i++) {
239		p = &sc->port[i];
240		if (p->is & p->ie)
241			sc->is |= (1 << i);
242	}
243	DPRINTF("%s(%08x) %08x\n", __func__, mask, sc->is);
244
245	/* If there is nothing enabled -- clear legacy interrupt and exit. */
246	if (sc->is == 0 || (sc->ghc & AHCI_GHC_IE) == 0) {
247		if (sc->lintr) {
248			pci_lintr_deassert(pi);
249			sc->lintr = 0;
250		}
251		return;
252	}
253
254	/* If there is anything and no MSI -- assert legacy interrupt. */
255	nmsg = pci_msi_maxmsgnum(pi);
256	if (nmsg == 0) {
257		if (!sc->lintr) {
258			sc->lintr = 1;
259			pci_lintr_assert(pi);
260		}
261		return;
262	}
263
264	/* Assert respective MSIs for ports that were touched. */
265	for (i = 0; i < nmsg; i++) {
266		if (sc->ports <= nmsg || i < nmsg - 1)
267			mmask = 1 << i;
268		else
269			mmask = 0xffffffff << i;
270		if (sc->is & mask && mmask & mask)
271			pci_generate_msi(pi, i);
272	}
273}
274
275/*
276 * Generate HBA interrupt on specific port event.
277 */
278static void
279ahci_port_intr(struct ahci_port *p)
280{
281	struct pci_ahci_softc *sc = p->pr_sc;
282	struct pci_devinst *pi = sc->asc_pi;
283	int nmsg;
284
285	DPRINTF("%s(%d) %08x/%08x %08x\n", __func__,
286	    p->port, p->is, p->ie, sc->is);
287
288	/* If there is nothing enabled -- we are done. */
289	if ((p->is & p->ie) == 0)
290		return;
291
292	/* In case of non-shared MSI always generate interrupt. */
293	nmsg = pci_msi_maxmsgnum(pi);
294	if (sc->ports <= nmsg || p->port < nmsg - 1) {
295		sc->is |= (1 << p->port);
296		if ((sc->ghc & AHCI_GHC_IE) == 0)
297			return;
298		pci_generate_msi(pi, p->port);
299		return;
300	}
301
302	/* If IS for this port is already set -- do nothing. */
303	if (sc->is & (1 << p->port))
304		return;
305
306	sc->is |= (1 << p->port);
307
308	/* If interrupts are enabled -- generate one. */
309	if ((sc->ghc & AHCI_GHC_IE) == 0)
310		return;
311	if (nmsg > 0) {
312		pci_generate_msi(pi, nmsg - 1);
313	} else if (!sc->lintr) {
314		sc->lintr = 1;
315		pci_lintr_assert(pi);
316	}
317}
318
319static void
320ahci_write_fis(struct ahci_port *p, enum sata_fis_type ft, uint8_t *fis)
321{
322	int offset, len, irq;
323
324	if (p->rfis == NULL || !(p->cmd & AHCI_P_CMD_FRE))
325		return;
326
327	switch (ft) {
328	case FIS_TYPE_REGD2H:
329		offset = 0x40;
330		len = 20;
331		irq = (fis[1] & (1 << 6)) ? AHCI_P_IX_DHR : 0;
332		break;
333	case FIS_TYPE_SETDEVBITS:
334		offset = 0x58;
335		len = 8;
336		irq = (fis[1] & (1 << 6)) ? AHCI_P_IX_SDB : 0;
337		break;
338	case FIS_TYPE_PIOSETUP:
339		offset = 0x20;
340		len = 20;
341		irq = (fis[1] & (1 << 6)) ? AHCI_P_IX_PS : 0;
342		break;
343	default:
344		WPRINTF("unsupported fis type %d\n", ft);
345		return;
346	}
347	if (fis[2] & ATA_S_ERROR) {
348		p->waitforclear = 1;
349		irq |= AHCI_P_IX_TFE;
350	}
351	memcpy(p->rfis + offset, fis, len);
352	if (irq) {
353		if (~p->is & irq) {
354			p->is |= irq;
355			ahci_port_intr(p);
356		}
357	}
358}
359
360static void
361ahci_write_fis_piosetup(struct ahci_port *p)
362{
363	uint8_t fis[20];
364
365	memset(fis, 0, sizeof(fis));
366	fis[0] = FIS_TYPE_PIOSETUP;
367	ahci_write_fis(p, FIS_TYPE_PIOSETUP, fis);
368}
369
370static void
371ahci_write_fis_sdb(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t tfd)
372{
373	uint8_t fis[8];
374	uint8_t error;
375
376	error = (tfd >> 8) & 0xff;
377	tfd &= 0x77;
378	memset(fis, 0, sizeof(fis));
379	fis[0] = FIS_TYPE_SETDEVBITS;
380	fis[1] = (1 << 6);
381	fis[2] = tfd;
382	fis[3] = error;
383	if (fis[2] & ATA_S_ERROR) {
384		p->err_cfis[0] = slot;
385		p->err_cfis[2] = tfd;
386		p->err_cfis[3] = error;
387		memcpy(&p->err_cfis[4], cfis + 4, 16);
388	} else {
389		*(uint32_t *)(fis + 4) = (1 << slot);
390		p->sact &= ~(1 << slot);
391	}
392	p->tfd &= ~0x77;
393	p->tfd |= tfd;
394	ahci_write_fis(p, FIS_TYPE_SETDEVBITS, fis);
395}
396
397static void
398ahci_write_fis_d2h(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t tfd)
399{
400	uint8_t fis[20];
401	uint8_t error;
402
403	error = (tfd >> 8) & 0xff;
404	memset(fis, 0, sizeof(fis));
405	fis[0] = FIS_TYPE_REGD2H;
406	fis[1] = (1 << 6);
407	fis[2] = tfd & 0xff;
408	fis[3] = error;
409	fis[4] = cfis[4];
410	fis[5] = cfis[5];
411	fis[6] = cfis[6];
412	fis[7] = cfis[7];
413	fis[8] = cfis[8];
414	fis[9] = cfis[9];
415	fis[10] = cfis[10];
416	fis[11] = cfis[11];
417	fis[12] = cfis[12];
418	fis[13] = cfis[13];
419	if (fis[2] & ATA_S_ERROR) {
420		p->err_cfis[0] = 0x80;
421		p->err_cfis[2] = tfd & 0xff;
422		p->err_cfis[3] = error;
423		memcpy(&p->err_cfis[4], cfis + 4, 16);
424	} else
425		p->ci &= ~(1 << slot);
426	p->tfd = tfd;
427	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
428}
429
430static void
431ahci_write_fis_d2h_ncq(struct ahci_port *p, int slot)
432{
433	uint8_t fis[20];
434
435	p->tfd = ATA_S_READY | ATA_S_DSC;
436	memset(fis, 0, sizeof(fis));
437	fis[0] = FIS_TYPE_REGD2H;
438	fis[1] = 0;			/* No interrupt */
439	fis[2] = p->tfd;		/* Status */
440	fis[3] = 0;			/* No error */
441	p->ci &= ~(1 << slot);
442	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
443}
444
445static void
446ahci_write_reset_fis_d2h(struct ahci_port *p)
447{
448	uint8_t fis[20];
449
450	memset(fis, 0, sizeof(fis));
451	fis[0] = FIS_TYPE_REGD2H;
452	fis[3] = 1;
453	fis[4] = 1;
454	if (p->atapi) {
455		fis[5] = 0x14;
456		fis[6] = 0xeb;
457	}
458	fis[12] = 1;
459	ahci_write_fis(p, FIS_TYPE_REGD2H, fis);
460}
461
462static void
463ahci_check_stopped(struct ahci_port *p)
464{
465	/*
466	 * If we are no longer processing the command list and nothing
467	 * is in-flight, clear the running bit, the current command
468	 * slot, the command issue and active bits.
469	 */
470	if (!(p->cmd & AHCI_P_CMD_ST)) {
471		if (p->pending == 0) {
472			p->ccs = 0;
473			p->cmd &= ~(AHCI_P_CMD_CR | AHCI_P_CMD_CCS_MASK);
474			p->ci = 0;
475			p->sact = 0;
476			p->waitforclear = 0;
477		}
478	}
479}
480
481static void
482ahci_port_stop(struct ahci_port *p)
483{
484	struct ahci_ioreq *aior;
485	uint8_t *cfis;
486	int slot;
487	int error;
488
489	assert(pthread_mutex_isowned_np(&p->pr_sc->mtx));
490
491	TAILQ_FOREACH(aior, &p->iobhd, io_blist) {
492		/*
493		 * Try to cancel the outstanding blockif request.
494		 */
495		error = blockif_cancel(p->bctx, &aior->io_req);
496		if (error != 0)
497			continue;
498
499		slot = aior->slot;
500		cfis = aior->cfis;
501		if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
502		    cfis[2] == ATA_READ_FPDMA_QUEUED ||
503		    cfis[2] == ATA_SEND_FPDMA_QUEUED)
504			p->sact &= ~(1 << slot);	/* NCQ */
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 *bp, buf[50];
1283
1284		msf = (acmd[1] >> 1) & 1;
1285		bp = buf + 2;
1286		*bp++ = 1;
1287		*bp++ = 1;
1288
1289		*bp++ = 1;
1290		*bp++ = 0x14;
1291		*bp++ = 0;
1292		*bp++ = 0xa0;
1293		*bp++ = 0;
1294		*bp++ = 0;
1295		*bp++ = 0;
1296		*bp++ = 0;
1297		*bp++ = 1;
1298		*bp++ = 0;
1299		*bp++ = 0;
1300
1301		*bp++ = 1;
1302		*bp++ = 0x14;
1303		*bp++ = 0;
1304		*bp++ = 0xa1;
1305		*bp++ = 0;
1306		*bp++ = 0;
1307		*bp++ = 0;
1308		*bp++ = 0;
1309		*bp++ = 1;
1310		*bp++ = 0;
1311		*bp++ = 0;
1312
1313		*bp++ = 1;
1314		*bp++ = 0x14;
1315		*bp++ = 0;
1316		*bp++ = 0xa2;
1317		*bp++ = 0;
1318		*bp++ = 0;
1319		*bp++ = 0;
1320		sectors = blockif_size(p->bctx) / blockif_sectsz(p->bctx);
1321		sectors >>= 2;
1322		if (msf) {
1323			*bp++ = 0;
1324			lba_to_msf(bp, sectors);
1325			bp += 3;
1326		} else {
1327			be32enc(bp, sectors);
1328			bp += 4;
1329		}
1330
1331		*bp++ = 1;
1332		*bp++ = 0x14;
1333		*bp++ = 0;
1334		*bp++ = 1;
1335		*bp++ = 0;
1336		*bp++ = 0;
1337		*bp++ = 0;
1338		if (msf) {
1339			*bp++ = 0;
1340			lba_to_msf(bp, 0);
1341			bp += 3;
1342		} else {
1343			*bp++ = 0;
1344			*bp++ = 0;
1345			*bp++ = 0;
1346			*bp++ = 0;
1347		}
1348
1349		size = bp - buf;
1350		be16enc(buf, size - 2);
1351		if (len > size)
1352			len = size;
1353		write_prdt(p, slot, cfis, buf, len);
1354		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1355		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1356		break;
1357	}
1358	default:
1359	{
1360		uint32_t tfd;
1361
1362		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1363		p->asc = 0x24;
1364		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1365		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1366		ahci_write_fis_d2h(p, slot, cfis, tfd);
1367		break;
1368	}
1369	}
1370}
1371
1372static void
1373atapi_report_luns(struct ahci_port *p, int slot, uint8_t *cfis)
1374{
1375	uint8_t buf[16];
1376
1377	memset(buf, 0, sizeof(buf));
1378	buf[3] = 8;
1379
1380	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1381	write_prdt(p, slot, cfis, buf, sizeof(buf));
1382	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1383}
1384
1385static void
1386atapi_read(struct ahci_port *p, int slot, uint8_t *cfis, uint32_t done)
1387{
1388	struct ahci_ioreq *aior;
1389	struct ahci_cmd_hdr *hdr;
1390	struct ahci_prdt_entry *prdt;
1391	struct blockif_req *breq;
1392	uint8_t *acmd;
1393	uint64_t lba;
1394	uint32_t len;
1395	int err;
1396
1397	acmd = cfis + 0x40;
1398	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1399	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
1400
1401	lba = be32dec(acmd + 2);
1402	if (acmd[0] == READ_10)
1403		len = be16dec(acmd + 7);
1404	else
1405		len = be32dec(acmd + 6);
1406	if (len == 0) {
1407		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1408		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1409	}
1410	lba *= 2048;
1411	len *= 2048;
1412
1413	/*
1414	 * Pull request off free list
1415	 */
1416	aior = STAILQ_FIRST(&p->iofhd);
1417	assert(aior != NULL);
1418	STAILQ_REMOVE_HEAD(&p->iofhd, io_flist);
1419	aior->cfis = cfis;
1420	aior->slot = slot;
1421	aior->len = len;
1422	aior->done = done;
1423	breq = &aior->io_req;
1424	breq->br_offset = lba + done;
1425	ahci_build_iov(p, aior, prdt, hdr->prdtl);
1426
1427	/* Mark this command in-flight. */
1428	p->pending |= 1 << slot;
1429
1430	/* Stuff request onto busy list. */
1431	TAILQ_INSERT_HEAD(&p->iobhd, aior, io_blist);
1432
1433	err = blockif_read(p->bctx, breq);
1434	assert(err == 0);
1435}
1436
1437static void
1438atapi_request_sense(struct ahci_port *p, int slot, uint8_t *cfis)
1439{
1440	uint8_t buf[64];
1441	uint8_t *acmd;
1442	int len;
1443
1444	acmd = cfis + 0x40;
1445	len = acmd[4];
1446	if (len > sizeof(buf))
1447		len = sizeof(buf);
1448	memset(buf, 0, len);
1449	buf[0] = 0x70 | (1 << 7);
1450	buf[2] = p->sense_key;
1451	buf[7] = 10;
1452	buf[12] = p->asc;
1453	write_prdt(p, slot, cfis, buf, len);
1454	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1455	ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1456}
1457
1458static void
1459atapi_start_stop_unit(struct ahci_port *p, int slot, uint8_t *cfis)
1460{
1461	uint8_t *acmd = cfis + 0x40;
1462	uint32_t tfd;
1463
1464	switch (acmd[4] & 3) {
1465	case 0:
1466	case 1:
1467	case 3:
1468		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1469		tfd = ATA_S_READY | ATA_S_DSC;
1470		break;
1471	case 2:
1472		/* TODO eject media */
1473		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1474		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1475		p->asc = 0x53;
1476		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1477		break;
1478	}
1479	ahci_write_fis_d2h(p, slot, cfis, tfd);
1480}
1481
1482static void
1483atapi_mode_sense(struct ahci_port *p, int slot, uint8_t *cfis)
1484{
1485	uint8_t *acmd;
1486	uint32_t tfd;
1487	uint8_t pc, code;
1488	int len;
1489
1490	acmd = cfis + 0x40;
1491	len = be16dec(acmd + 7);
1492	pc = acmd[2] >> 6;
1493	code = acmd[2] & 0x3f;
1494
1495	switch (pc) {
1496	case 0:
1497		switch (code) {
1498		case MODEPAGE_RW_ERROR_RECOVERY:
1499		{
1500			uint8_t buf[16];
1501
1502			if (len > sizeof(buf))
1503				len = sizeof(buf);
1504
1505			memset(buf, 0, sizeof(buf));
1506			be16enc(buf, 16 - 2);
1507			buf[2] = 0x70;
1508			buf[8] = 0x01;
1509			buf[9] = 16 - 10;
1510			buf[11] = 0x05;
1511			write_prdt(p, slot, cfis, buf, len);
1512			tfd = ATA_S_READY | ATA_S_DSC;
1513			break;
1514		}
1515		case MODEPAGE_CD_CAPABILITIES:
1516		{
1517			uint8_t buf[30];
1518
1519			if (len > sizeof(buf))
1520				len = sizeof(buf);
1521
1522			memset(buf, 0, sizeof(buf));
1523			be16enc(buf, 30 - 2);
1524			buf[2] = 0x70;
1525			buf[8] = 0x2A;
1526			buf[9] = 30 - 10;
1527			buf[10] = 0x08;
1528			buf[12] = 0x71;
1529			be16enc(&buf[18], 2);
1530			be16enc(&buf[20], 512);
1531			write_prdt(p, slot, cfis, buf, len);
1532			tfd = ATA_S_READY | ATA_S_DSC;
1533			break;
1534		}
1535		default:
1536			goto error;
1537			break;
1538		}
1539		break;
1540	case 3:
1541		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1542		p->asc = 0x39;
1543		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1544		break;
1545error:
1546	case 1:
1547	case 2:
1548		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1549		p->asc = 0x24;
1550		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1551		break;
1552	}
1553	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1554	ahci_write_fis_d2h(p, slot, cfis, tfd);
1555}
1556
1557static void
1558atapi_get_event_status_notification(struct ahci_port *p, int slot,
1559    uint8_t *cfis)
1560{
1561	uint8_t *acmd;
1562	uint32_t tfd;
1563
1564	acmd = cfis + 0x40;
1565
1566	/* we don't support asynchronous operation */
1567	if (!(acmd[1] & 1)) {
1568		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1569		p->asc = 0x24;
1570		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
1571	} else {
1572		uint8_t buf[8];
1573		int len;
1574
1575		len = be16dec(acmd + 7);
1576		if (len > sizeof(buf))
1577			len = sizeof(buf);
1578
1579		memset(buf, 0, sizeof(buf));
1580		be16enc(buf, 8 - 2);
1581		buf[2] = 0x04;
1582		buf[3] = 0x10;
1583		buf[5] = 0x02;
1584		write_prdt(p, slot, cfis, buf, len);
1585		tfd = ATA_S_READY | ATA_S_DSC;
1586	}
1587	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1588	ahci_write_fis_d2h(p, slot, cfis, tfd);
1589}
1590
1591static void
1592handle_packet_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1593{
1594	uint8_t *acmd;
1595
1596	acmd = cfis + 0x40;
1597
1598#ifdef AHCI_DEBUG
1599	{
1600		int i;
1601		DPRINTF("ACMD:");
1602		for (i = 0; i < 16; i++)
1603			DPRINTF("%02x ", acmd[i]);
1604		DPRINTF("\n");
1605	}
1606#endif
1607
1608	switch (acmd[0]) {
1609	case TEST_UNIT_READY:
1610		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1611		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1612		break;
1613	case INQUIRY:
1614		atapi_inquiry(p, slot, cfis);
1615		break;
1616	case READ_CAPACITY:
1617		atapi_read_capacity(p, slot, cfis);
1618		break;
1619	case PREVENT_ALLOW:
1620		/* TODO */
1621		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1622		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1623		break;
1624	case READ_TOC:
1625		atapi_read_toc(p, slot, cfis);
1626		break;
1627	case REPORT_LUNS:
1628		atapi_report_luns(p, slot, cfis);
1629		break;
1630	case READ_10:
1631	case READ_12:
1632		atapi_read(p, slot, cfis, 0);
1633		break;
1634	case REQUEST_SENSE:
1635		atapi_request_sense(p, slot, cfis);
1636		break;
1637	case START_STOP_UNIT:
1638		atapi_start_stop_unit(p, slot, cfis);
1639		break;
1640	case MODE_SENSE_10:
1641		atapi_mode_sense(p, slot, cfis);
1642		break;
1643	case GET_EVENT_STATUS_NOTIFICATION:
1644		atapi_get_event_status_notification(p, slot, cfis);
1645		break;
1646	default:
1647		cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
1648		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1649		p->asc = 0x20;
1650		ahci_write_fis_d2h(p, slot, cfis, (p->sense_key << 12) |
1651				ATA_S_READY | ATA_S_ERROR);
1652		break;
1653	}
1654}
1655
1656static void
1657ahci_handle_cmd(struct ahci_port *p, int slot, uint8_t *cfis)
1658{
1659
1660	p->tfd |= ATA_S_BUSY;
1661	switch (cfis[2]) {
1662	case ATA_ATA_IDENTIFY:
1663		handle_identify(p, slot, cfis);
1664		break;
1665	case ATA_SETFEATURES:
1666	{
1667		switch (cfis[3]) {
1668		case ATA_SF_ENAB_SATA_SF:
1669			switch (cfis[12]) {
1670			case ATA_SATA_SF_AN:
1671				p->tfd = ATA_S_DSC | ATA_S_READY;
1672				break;
1673			default:
1674				p->tfd = ATA_S_ERROR | ATA_S_READY;
1675				p->tfd |= (ATA_ERROR_ABORT << 8);
1676				break;
1677			}
1678			break;
1679		case ATA_SF_ENAB_WCACHE:
1680		case ATA_SF_DIS_WCACHE:
1681		case ATA_SF_ENAB_RCACHE:
1682		case ATA_SF_DIS_RCACHE:
1683			p->tfd = ATA_S_DSC | ATA_S_READY;
1684			break;
1685		case ATA_SF_SETXFER:
1686		{
1687			switch (cfis[12] & 0xf8) {
1688			case ATA_PIO:
1689			case ATA_PIO0:
1690				break;
1691			case ATA_WDMA0:
1692			case ATA_UDMA0:
1693				p->xfermode = (cfis[12] & 0x7);
1694				break;
1695			}
1696			p->tfd = ATA_S_DSC | ATA_S_READY;
1697			break;
1698		}
1699		default:
1700			p->tfd = ATA_S_ERROR | ATA_S_READY;
1701			p->tfd |= (ATA_ERROR_ABORT << 8);
1702			break;
1703		}
1704		ahci_write_fis_d2h(p, slot, cfis, p->tfd);
1705		break;
1706	}
1707	case ATA_SET_MULTI:
1708		if (cfis[12] != 0 &&
1709			(cfis[12] > 128 || (cfis[12] & (cfis[12] - 1)))) {
1710			p->tfd = ATA_S_ERROR | ATA_S_READY;
1711			p->tfd |= (ATA_ERROR_ABORT << 8);
1712		} else {
1713			p->mult_sectors = cfis[12];
1714			p->tfd = ATA_S_DSC | ATA_S_READY;
1715		}
1716		ahci_write_fis_d2h(p, slot, cfis, p->tfd);
1717		break;
1718	case ATA_READ:
1719	case ATA_WRITE:
1720	case ATA_READ48:
1721	case ATA_WRITE48:
1722	case ATA_READ_MUL:
1723	case ATA_WRITE_MUL:
1724	case ATA_READ_MUL48:
1725	case ATA_WRITE_MUL48:
1726	case ATA_READ_DMA:
1727	case ATA_WRITE_DMA:
1728	case ATA_READ_DMA48:
1729	case ATA_WRITE_DMA48:
1730	case ATA_READ_FPDMA_QUEUED:
1731	case ATA_WRITE_FPDMA_QUEUED:
1732		ahci_handle_rw(p, slot, cfis, 0);
1733		break;
1734	case ATA_FLUSHCACHE:
1735	case ATA_FLUSHCACHE48:
1736		ahci_handle_flush(p, slot, cfis);
1737		break;
1738	case ATA_DATA_SET_MANAGEMENT:
1739		if (cfis[11] == 0 && cfis[3] == ATA_DSM_TRIM &&
1740		    cfis[13] == 0 && cfis[12] == 1) {
1741			ahci_handle_dsm_trim(p, slot, cfis, 0);
1742			break;
1743		}
1744		ahci_write_fis_d2h(p, slot, cfis,
1745		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1746		break;
1747	case ATA_SEND_FPDMA_QUEUED:
1748		if ((cfis[13] & 0x1f) == ATA_SFPDMA_DSM &&
1749		    cfis[17] == 0 && cfis[16] == ATA_DSM_TRIM &&
1750		    cfis[11] == 0 && cfis[3] == 1) {
1751			ahci_handle_dsm_trim(p, slot, cfis, 0);
1752			break;
1753		}
1754		ahci_write_fis_d2h(p, slot, cfis,
1755		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1756		break;
1757	case ATA_READ_LOG_EXT:
1758	case ATA_READ_LOG_DMA_EXT:
1759		ahci_handle_read_log(p, slot, cfis);
1760		break;
1761	case ATA_SECURITY_FREEZE_LOCK:
1762	case ATA_SMART_CMD:
1763	case ATA_NOP:
1764		ahci_write_fis_d2h(p, slot, cfis,
1765		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1766		break;
1767	case ATA_CHECK_POWER_MODE:
1768		cfis[12] = 0xff;	/* always on */
1769		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1770		break;
1771	case ATA_STANDBY_CMD:
1772	case ATA_STANDBY_IMMEDIATE:
1773	case ATA_IDLE_CMD:
1774	case ATA_IDLE_IMMEDIATE:
1775	case ATA_SLEEP:
1776	case ATA_READ_VERIFY:
1777	case ATA_READ_VERIFY48:
1778		ahci_write_fis_d2h(p, slot, cfis, ATA_S_READY | ATA_S_DSC);
1779		break;
1780	case ATA_ATAPI_IDENTIFY:
1781		handle_atapi_identify(p, slot, cfis);
1782		break;
1783	case ATA_PACKET_CMD:
1784		if (!p->atapi) {
1785			ahci_write_fis_d2h(p, slot, cfis,
1786			    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1787		} else
1788			handle_packet_cmd(p, slot, cfis);
1789		break;
1790	default:
1791		WPRINTF("Unsupported cmd:%02x\n", cfis[2]);
1792		ahci_write_fis_d2h(p, slot, cfis,
1793		    (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR);
1794		break;
1795	}
1796}
1797
1798static void
1799ahci_handle_slot(struct ahci_port *p, int slot)
1800{
1801	struct ahci_cmd_hdr *hdr;
1802#ifdef AHCI_DEBUG
1803	struct ahci_prdt_entry *prdt;
1804#endif
1805	struct pci_ahci_softc *sc;
1806	uint8_t *cfis;
1807#ifdef AHCI_DEBUG
1808	int cfl, i;
1809#endif
1810
1811	sc = p->pr_sc;
1812	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1813#ifdef AHCI_DEBUG
1814	cfl = (hdr->flags & 0x1f) * 4;
1815#endif
1816	cfis = paddr_guest2host(ahci_ctx(sc), hdr->ctba,
1817			0x80 + hdr->prdtl * sizeof(struct ahci_prdt_entry));
1818#ifdef AHCI_DEBUG
1819	prdt = (struct ahci_prdt_entry *)(cfis + 0x80);
1820
1821	DPRINTF("\ncfis:");
1822	for (i = 0; i < cfl; i++) {
1823		if (i % 10 == 0)
1824			DPRINTF("\n");
1825		DPRINTF("%02x ", cfis[i]);
1826	}
1827	DPRINTF("\n");
1828
1829	for (i = 0; i < hdr->prdtl; i++) {
1830		DPRINTF("%d@%08"PRIx64"\n", prdt->dbc & 0x3fffff, prdt->dba);
1831		prdt++;
1832	}
1833#endif
1834
1835	if (cfis[0] != FIS_TYPE_REGH2D) {
1836		WPRINTF("Not a H2D FIS:%02x\n", cfis[0]);
1837		return;
1838	}
1839
1840	if (cfis[1] & 0x80) {
1841		ahci_handle_cmd(p, slot, cfis);
1842	} else {
1843		if (cfis[15] & (1 << 2))
1844			p->reset = 1;
1845		else if (p->reset) {
1846			p->reset = 0;
1847			ahci_port_reset(p);
1848		}
1849		p->ci &= ~(1 << slot);
1850	}
1851}
1852
1853static void
1854ahci_handle_port(struct ahci_port *p)
1855{
1856
1857	if (!(p->cmd & AHCI_P_CMD_ST))
1858		return;
1859
1860	/*
1861	 * Search for any new commands to issue ignoring those that
1862	 * are already in-flight.  Stop if device is busy or in error.
1863	 */
1864	for (; (p->ci & ~p->pending) != 0; p->ccs = ((p->ccs + 1) & 31)) {
1865		if ((p->tfd & (ATA_S_BUSY | ATA_S_DRQ)) != 0)
1866			break;
1867		if (p->waitforclear)
1868			break;
1869		if ((p->ci & ~p->pending & (1 << p->ccs)) != 0) {
1870			p->cmd &= ~AHCI_P_CMD_CCS_MASK;
1871			p->cmd |= p->ccs << AHCI_P_CMD_CCS_SHIFT;
1872			ahci_handle_slot(p, p->ccs);
1873		}
1874	}
1875}
1876
1877/*
1878 * blockif callback routine - this runs in the context of the blockif
1879 * i/o thread, so the mutex needs to be acquired.
1880 */
1881static void
1882ata_ioreq_cb(struct blockif_req *br, int err)
1883{
1884	struct ahci_cmd_hdr *hdr;
1885	struct ahci_ioreq *aior;
1886	struct ahci_port *p;
1887	struct pci_ahci_softc *sc;
1888	uint32_t tfd;
1889	uint8_t *cfis;
1890	int slot, ncq, dsm;
1891
1892	DPRINTF("%s %d\n", __func__, err);
1893
1894	ncq = dsm = 0;
1895	aior = br->br_param;
1896	p = aior->io_pr;
1897	cfis = aior->cfis;
1898	slot = aior->slot;
1899	sc = p->pr_sc;
1900	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + slot * AHCI_CL_SIZE);
1901
1902	if (cfis[2] == ATA_WRITE_FPDMA_QUEUED ||
1903	    cfis[2] == ATA_READ_FPDMA_QUEUED ||
1904	    cfis[2] == ATA_SEND_FPDMA_QUEUED)
1905		ncq = 1;
1906	if (cfis[2] == ATA_DATA_SET_MANAGEMENT ||
1907	    (cfis[2] == ATA_SEND_FPDMA_QUEUED &&
1908	     (cfis[13] & 0x1f) == ATA_SFPDMA_DSM))
1909		dsm = 1;
1910
1911	pthread_mutex_lock(&sc->mtx);
1912
1913	/*
1914	 * Delete the blockif request from the busy list
1915	 */
1916	TAILQ_REMOVE(&p->iobhd, aior, io_blist);
1917
1918	/*
1919	 * Move the blockif request back to the free list
1920	 */
1921	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_flist);
1922
1923	if (!err)
1924		hdr->prdbc = aior->done;
1925
1926	if (!err && aior->more) {
1927		if (dsm)
1928			ahci_handle_dsm_trim(p, slot, cfis, aior->done);
1929		else
1930			ahci_handle_rw(p, slot, cfis, aior->done);
1931		goto out;
1932	}
1933
1934	if (!err)
1935		tfd = ATA_S_READY | ATA_S_DSC;
1936	else
1937		tfd = (ATA_E_ABORT << 8) | ATA_S_READY | ATA_S_ERROR;
1938	if (ncq)
1939		ahci_write_fis_sdb(p, slot, cfis, tfd);
1940	else
1941		ahci_write_fis_d2h(p, slot, cfis, tfd);
1942
1943	/*
1944	 * This command is now complete.
1945	 */
1946	p->pending &= ~(1 << slot);
1947
1948	ahci_check_stopped(p);
1949	ahci_handle_port(p);
1950out:
1951	pthread_mutex_unlock(&sc->mtx);
1952	DPRINTF("%s exit\n", __func__);
1953}
1954
1955static void
1956atapi_ioreq_cb(struct blockif_req *br, int err)
1957{
1958	struct ahci_cmd_hdr *hdr;
1959	struct ahci_ioreq *aior;
1960	struct ahci_port *p;
1961	struct pci_ahci_softc *sc;
1962	uint8_t *cfis;
1963	uint32_t tfd;
1964	int slot;
1965
1966	DPRINTF("%s %d\n", __func__, err);
1967
1968	aior = br->br_param;
1969	p = aior->io_pr;
1970	cfis = aior->cfis;
1971	slot = aior->slot;
1972	sc = p->pr_sc;
1973	hdr = (struct ahci_cmd_hdr *)(p->cmd_lst + aior->slot * AHCI_CL_SIZE);
1974
1975	pthread_mutex_lock(&sc->mtx);
1976
1977	/*
1978	 * Delete the blockif request from the busy list
1979	 */
1980	TAILQ_REMOVE(&p->iobhd, aior, io_blist);
1981
1982	/*
1983	 * Move the blockif request back to the free list
1984	 */
1985	STAILQ_INSERT_TAIL(&p->iofhd, aior, io_flist);
1986
1987	if (!err)
1988		hdr->prdbc = aior->done;
1989
1990	if (!err && aior->more) {
1991		atapi_read(p, slot, cfis, aior->done);
1992		goto out;
1993	}
1994
1995	if (!err) {
1996		tfd = ATA_S_READY | ATA_S_DSC;
1997	} else {
1998		p->sense_key = ATA_SENSE_ILLEGAL_REQUEST;
1999		p->asc = 0x21;
2000		tfd = (p->sense_key << 12) | ATA_S_READY | ATA_S_ERROR;
2001	}
2002	cfis[4] = (cfis[4] & ~7) | ATA_I_CMD | ATA_I_IN;
2003	ahci_write_fis_d2h(p, slot, cfis, tfd);
2004
2005	/*
2006	 * This command is now complete.
2007	 */
2008	p->pending &= ~(1 << slot);
2009
2010	ahci_check_stopped(p);
2011	ahci_handle_port(p);
2012out:
2013	pthread_mutex_unlock(&sc->mtx);
2014	DPRINTF("%s exit\n", __func__);
2015}
2016
2017static void
2018pci_ahci_ioreq_init(struct ahci_port *pr)
2019{
2020	struct ahci_ioreq *vr;
2021	int i;
2022
2023	pr->ioqsz = blockif_queuesz(pr->bctx);
2024	pr->ioreq = calloc(pr->ioqsz, sizeof(struct ahci_ioreq));
2025	STAILQ_INIT(&pr->iofhd);
2026
2027	/*
2028	 * Add all i/o request entries to the free queue
2029	 */
2030	for (i = 0; i < pr->ioqsz; i++) {
2031		vr = &pr->ioreq[i];
2032		vr->io_pr = pr;
2033		if (!pr->atapi)
2034			vr->io_req.br_callback = ata_ioreq_cb;
2035		else
2036			vr->io_req.br_callback = atapi_ioreq_cb;
2037		vr->io_req.br_param = vr;
2038		STAILQ_INSERT_TAIL(&pr->iofhd, vr, io_flist);
2039	}
2040
2041	TAILQ_INIT(&pr->iobhd);
2042}
2043
2044static void
2045pci_ahci_port_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
2046{
2047	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
2048	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
2049	struct ahci_port *p = &sc->port[port];
2050
2051	DPRINTF("pci_ahci_port %d: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
2052		port, offset, value);
2053
2054	switch (offset) {
2055	case AHCI_P_CLB:
2056		p->clb = value;
2057		break;
2058	case AHCI_P_CLBU:
2059		p->clbu = value;
2060		break;
2061	case AHCI_P_FB:
2062		p->fb = value;
2063		break;
2064	case AHCI_P_FBU:
2065		p->fbu = value;
2066		break;
2067	case AHCI_P_IS:
2068		p->is &= ~value;
2069		ahci_port_intr(p);
2070		break;
2071	case AHCI_P_IE:
2072		p->ie = value & 0xFDC000FF;
2073		ahci_port_intr(p);
2074		break;
2075	case AHCI_P_CMD:
2076	{
2077		p->cmd &= ~(AHCI_P_CMD_ST | AHCI_P_CMD_SUD | AHCI_P_CMD_POD |
2078		    AHCI_P_CMD_CLO | AHCI_P_CMD_FRE | AHCI_P_CMD_APSTE |
2079		    AHCI_P_CMD_ATAPI | AHCI_P_CMD_DLAE | AHCI_P_CMD_ALPE |
2080		    AHCI_P_CMD_ASP | AHCI_P_CMD_ICC_MASK);
2081		p->cmd |= (AHCI_P_CMD_ST | AHCI_P_CMD_SUD | AHCI_P_CMD_POD |
2082		    AHCI_P_CMD_CLO | AHCI_P_CMD_FRE | AHCI_P_CMD_APSTE |
2083		    AHCI_P_CMD_ATAPI | AHCI_P_CMD_DLAE | AHCI_P_CMD_ALPE |
2084		    AHCI_P_CMD_ASP | AHCI_P_CMD_ICC_MASK) & value;
2085
2086		if (!(value & AHCI_P_CMD_ST)) {
2087			ahci_port_stop(p);
2088		} else {
2089			uint64_t clb;
2090
2091			p->cmd |= AHCI_P_CMD_CR;
2092			clb = (uint64_t)p->clbu << 32 | p->clb;
2093			p->cmd_lst = paddr_guest2host(ahci_ctx(sc), clb,
2094					AHCI_CL_SIZE * AHCI_MAX_SLOTS);
2095		}
2096
2097		if (value & AHCI_P_CMD_FRE) {
2098			uint64_t fb;
2099
2100			p->cmd |= AHCI_P_CMD_FR;
2101			fb = (uint64_t)p->fbu << 32 | p->fb;
2102			/* we don't support FBSCP, so rfis size is 256Bytes */
2103			p->rfis = paddr_guest2host(ahci_ctx(sc), fb, 256);
2104		} else {
2105			p->cmd &= ~AHCI_P_CMD_FR;
2106		}
2107
2108		if (value & AHCI_P_CMD_CLO) {
2109			p->tfd &= ~(ATA_S_BUSY | ATA_S_DRQ);
2110			p->cmd &= ~AHCI_P_CMD_CLO;
2111		}
2112
2113		if (value & AHCI_P_CMD_ICC_MASK) {
2114			p->cmd &= ~AHCI_P_CMD_ICC_MASK;
2115		}
2116
2117		ahci_handle_port(p);
2118		break;
2119	}
2120	case AHCI_P_TFD:
2121	case AHCI_P_SIG:
2122	case AHCI_P_SSTS:
2123		WPRINTF("pci_ahci_port: read only registers 0x%"PRIx64"\n", offset);
2124		break;
2125	case AHCI_P_SCTL:
2126		p->sctl = value;
2127		if (!(p->cmd & AHCI_P_CMD_ST)) {
2128			if (value & ATA_SC_DET_RESET)
2129				ahci_port_reset(p);
2130		}
2131		break;
2132	case AHCI_P_SERR:
2133		p->serr &= ~value;
2134		break;
2135	case AHCI_P_SACT:
2136		p->sact |= value;
2137		break;
2138	case AHCI_P_CI:
2139		p->ci |= value;
2140		ahci_handle_port(p);
2141		break;
2142	case AHCI_P_SNTF:
2143	case AHCI_P_FBS:
2144	default:
2145		break;
2146	}
2147}
2148
2149static void
2150pci_ahci_host_write(struct pci_ahci_softc *sc, uint64_t offset, uint64_t value)
2151{
2152	DPRINTF("pci_ahci_host: write offset 0x%"PRIx64" value 0x%"PRIx64"\n",
2153		offset, value);
2154
2155	switch (offset) {
2156	case AHCI_CAP:
2157	case AHCI_PI:
2158	case AHCI_VS:
2159	case AHCI_CAP2:
2160		DPRINTF("pci_ahci_host: read only registers 0x%"PRIx64"\n", offset);
2161		break;
2162	case AHCI_GHC:
2163		if (value & AHCI_GHC_HR) {
2164			ahci_reset(sc);
2165			break;
2166		}
2167		if (value & AHCI_GHC_IE)
2168			sc->ghc |= AHCI_GHC_IE;
2169		else
2170			sc->ghc &= ~AHCI_GHC_IE;
2171		ahci_generate_intr(sc, 0xffffffff);
2172		break;
2173	case AHCI_IS:
2174		sc->is &= ~value;
2175		ahci_generate_intr(sc, value);
2176		break;
2177	default:
2178		break;
2179	}
2180}
2181
2182static void
2183pci_ahci_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
2184		int baridx, uint64_t offset, int size, uint64_t value)
2185{
2186	struct pci_ahci_softc *sc = pi->pi_arg;
2187
2188	assert(baridx == 5);
2189	assert((offset % 4) == 0 && size == 4);
2190
2191	pthread_mutex_lock(&sc->mtx);
2192
2193	if (offset < AHCI_OFFSET)
2194		pci_ahci_host_write(sc, offset, value);
2195	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
2196		pci_ahci_port_write(sc, offset, value);
2197	else
2198		WPRINTF("pci_ahci: unknown i/o write offset 0x%"PRIx64"\n", offset);
2199
2200	pthread_mutex_unlock(&sc->mtx);
2201}
2202
2203static uint64_t
2204pci_ahci_host_read(struct pci_ahci_softc *sc, uint64_t offset)
2205{
2206	uint32_t value;
2207
2208	switch (offset) {
2209	case AHCI_CAP:
2210	case AHCI_GHC:
2211	case AHCI_IS:
2212	case AHCI_PI:
2213	case AHCI_VS:
2214	case AHCI_CCCC:
2215	case AHCI_CCCP:
2216	case AHCI_EM_LOC:
2217	case AHCI_EM_CTL:
2218	case AHCI_CAP2:
2219	{
2220		uint32_t *p = &sc->cap;
2221		p += (offset - AHCI_CAP) / sizeof(uint32_t);
2222		value = *p;
2223		break;
2224	}
2225	default:
2226		value = 0;
2227		break;
2228	}
2229	DPRINTF("pci_ahci_host: read offset 0x%"PRIx64" value 0x%x\n",
2230		offset, value);
2231
2232	return (value);
2233}
2234
2235static uint64_t
2236pci_ahci_port_read(struct pci_ahci_softc *sc, uint64_t offset)
2237{
2238	uint32_t value;
2239	int port = (offset - AHCI_OFFSET) / AHCI_STEP;
2240	offset = (offset - AHCI_OFFSET) % AHCI_STEP;
2241
2242	switch (offset) {
2243	case AHCI_P_CLB:
2244	case AHCI_P_CLBU:
2245	case AHCI_P_FB:
2246	case AHCI_P_FBU:
2247	case AHCI_P_IS:
2248	case AHCI_P_IE:
2249	case AHCI_P_CMD:
2250	case AHCI_P_TFD:
2251	case AHCI_P_SIG:
2252	case AHCI_P_SSTS:
2253	case AHCI_P_SCTL:
2254	case AHCI_P_SERR:
2255	case AHCI_P_SACT:
2256	case AHCI_P_CI:
2257	case AHCI_P_SNTF:
2258	case AHCI_P_FBS:
2259	{
2260		uint32_t *p= &sc->port[port].clb;
2261		p += (offset - AHCI_P_CLB) / sizeof(uint32_t);
2262		value = *p;
2263		break;
2264	}
2265	default:
2266		value = 0;
2267		break;
2268	}
2269
2270	DPRINTF("pci_ahci_port %d: read offset 0x%"PRIx64" value 0x%x\n",
2271		port, offset, value);
2272
2273	return value;
2274}
2275
2276static uint64_t
2277pci_ahci_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi, int baridx,
2278    uint64_t regoff, int size)
2279{
2280	struct pci_ahci_softc *sc = pi->pi_arg;
2281	uint64_t offset;
2282	uint32_t value;
2283
2284	assert(baridx == 5);
2285	assert(size == 1 || size == 2 || size == 4);
2286	assert((regoff & (size - 1)) == 0);
2287
2288	pthread_mutex_lock(&sc->mtx);
2289
2290	offset = regoff & ~0x3;	    /* round down to a multiple of 4 bytes */
2291	if (offset < AHCI_OFFSET)
2292		value = pci_ahci_host_read(sc, offset);
2293	else if (offset < AHCI_OFFSET + sc->ports * AHCI_STEP)
2294		value = pci_ahci_port_read(sc, offset);
2295	else {
2296		value = 0;
2297		WPRINTF("pci_ahci: unknown i/o read offset 0x%"PRIx64"\n",
2298		    regoff);
2299	}
2300	value >>= 8 * (regoff & 0x3);
2301
2302	pthread_mutex_unlock(&sc->mtx);
2303
2304	return (value);
2305}
2306
2307static int
2308pci_ahci_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts, int atapi)
2309{
2310	char bident[sizeof("XX:XX:XX")];
2311	struct blockif_ctxt *bctxt;
2312	struct pci_ahci_softc *sc;
2313	int ret, slots, p;
2314	MD5_CTX mdctx;
2315	u_char digest[16];
2316	char *next, *next2;
2317
2318	ret = 0;
2319
2320#ifdef AHCI_DEBUG
2321	dbg = fopen("/tmp/log", "w+");
2322#endif
2323
2324	sc = calloc(1, sizeof(struct pci_ahci_softc));
2325	pi->pi_arg = sc;
2326	sc->asc_pi = pi;
2327	pthread_mutex_init(&sc->mtx, NULL);
2328	sc->ports = 0;
2329	sc->pi = 0;
2330	slots = 32;
2331
2332	for (p = 0; p < MAX_PORTS && opts != NULL; p++, opts = next) {
2333		/* Identify and cut off type of present port. */
2334		if (strncmp(opts, "hd:", 3) == 0) {
2335			atapi = 0;
2336			opts += 3;
2337		} else if (strncmp(opts, "cd:", 3) == 0) {
2338			atapi = 1;
2339			opts += 3;
2340		}
2341
2342		/* Find and cut off the next port options. */
2343		next = strstr(opts, ",hd:");
2344		next2 = strstr(opts, ",cd:");
2345		if (next == NULL || (next2 != NULL && next2 < next))
2346			next = next2;
2347		if (next != NULL) {
2348			next[0] = 0;
2349			next++;
2350		}
2351
2352		if (opts[0] == 0)
2353			continue;
2354
2355		/*
2356		 * Attempt to open the backing image. Use the PCI slot/func
2357		 * and the port number for the identifier string.
2358		 */
2359		snprintf(bident, sizeof(bident), "%d:%d:%d", pi->pi_slot,
2360		    pi->pi_func, p);
2361		bctxt = blockif_open(opts, bident);
2362		if (bctxt == NULL) {
2363			sc->ports = p;
2364			ret = 1;
2365			goto open_fail;
2366		}
2367		sc->port[p].bctx = bctxt;
2368		sc->port[p].pr_sc = sc;
2369		sc->port[p].port = p;
2370		sc->port[p].atapi = atapi;
2371
2372		/*
2373		 * Create an identifier for the backing file.
2374		 * Use parts of the md5 sum of the filename
2375		 */
2376		MD5Init(&mdctx);
2377		MD5Update(&mdctx, opts, strlen(opts));
2378		MD5Final(digest, &mdctx);
2379		snprintf(sc->port[p].ident, AHCI_PORT_IDENT,
2380		    "BHYVE-%02X%02X-%02X%02X-%02X%02X",
2381		    digest[0], digest[1], digest[2], digest[3], digest[4],
2382		    digest[5]);
2383
2384		/*
2385		 * Allocate blockif request structures and add them
2386		 * to the free list
2387		 */
2388		pci_ahci_ioreq_init(&sc->port[p]);
2389
2390		sc->pi |= (1 << p);
2391		if (sc->port[p].ioqsz < slots)
2392			slots = sc->port[p].ioqsz;
2393	}
2394	sc->ports = p;
2395
2396	/* Intel ICH8 AHCI */
2397	--slots;
2398	if (sc->ports < DEF_PORTS)
2399		sc->ports = DEF_PORTS;
2400	sc->cap = AHCI_CAP_64BIT | AHCI_CAP_SNCQ | AHCI_CAP_SSNTF |
2401	    AHCI_CAP_SMPS | AHCI_CAP_SSS | AHCI_CAP_SALP |
2402	    AHCI_CAP_SAL | AHCI_CAP_SCLO | (0x3 << AHCI_CAP_ISS_SHIFT)|
2403	    AHCI_CAP_PMD | AHCI_CAP_SSC | AHCI_CAP_PSC |
2404	    (slots << AHCI_CAP_NCS_SHIFT) | AHCI_CAP_SXS | (sc->ports - 1);
2405
2406	sc->vs = 0x10300;
2407	sc->cap2 = AHCI_CAP2_APST;
2408	ahci_reset(sc);
2409
2410	pci_set_cfgdata16(pi, PCIR_DEVICE, 0x2821);
2411	pci_set_cfgdata16(pi, PCIR_VENDOR, 0x8086);
2412	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
2413	pci_set_cfgdata8(pi, PCIR_SUBCLASS, PCIS_STORAGE_SATA);
2414	pci_set_cfgdata8(pi, PCIR_PROGIF, PCIP_STORAGE_SATA_AHCI_1_0);
2415	p = MIN(sc->ports, 16);
2416	p = flsl(p) - ((p & (p - 1)) ? 0 : 1);
2417	pci_emul_add_msicap(pi, 1 << p);
2418	pci_emul_alloc_bar(pi, 5, PCIBAR_MEM32,
2419	    AHCI_OFFSET + sc->ports * AHCI_STEP);
2420
2421	pci_lintr_request(pi);
2422
2423open_fail:
2424	if (ret) {
2425		for (p = 0; p < sc->ports; p++) {
2426			if (sc->port[p].bctx != NULL)
2427				blockif_close(sc->port[p].bctx);
2428		}
2429		free(sc);
2430	}
2431
2432	return (ret);
2433}
2434
2435static int
2436pci_ahci_hd_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
2437{
2438
2439	return (pci_ahci_init(ctx, pi, opts, 0));
2440}
2441
2442static int
2443pci_ahci_atapi_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
2444{
2445
2446	return (pci_ahci_init(ctx, pi, opts, 1));
2447}
2448
2449/*
2450 * Use separate emulation names to distinguish drive and atapi devices
2451 */
2452struct pci_devemu pci_de_ahci = {
2453	.pe_emu =	"ahci",
2454	.pe_init =	pci_ahci_hd_init,
2455	.pe_barwrite =	pci_ahci_write,
2456	.pe_barread =	pci_ahci_read
2457};
2458PCI_EMUL_SET(pci_de_ahci);
2459
2460struct pci_devemu pci_de_ahci_hd = {
2461	.pe_emu =	"ahci-hd",
2462	.pe_init =	pci_ahci_hd_init,
2463	.pe_barwrite =	pci_ahci_write,
2464	.pe_barread =	pci_ahci_read
2465};
2466PCI_EMUL_SET(pci_de_ahci_hd);
2467
2468struct pci_devemu pci_de_ahci_cd = {
2469	.pe_emu =	"ahci-cd",
2470	.pe_init =	pci_ahci_atapi_init,
2471	.pe_barwrite =	pci_ahci_write,
2472	.pe_barread =	pci_ahci_read
2473};
2474PCI_EMUL_SET(pci_de_ahci_cd);
2475