xref: /openbmc/qemu/hw/ide/ahci.c (revision a78b1299)
1 /*
2  * QEMU AHCI Emulation
3  *
4  * Copyright (c) 2010 qiaochong@loongson.cn
5  * Copyright (c) 2010 Roland Elek <elek.roland@gmail.com>
6  * Copyright (c) 2010 Sebastian Herbszt <herbszt@gmx.de>
7  * Copyright (c) 2010 Alexander Graf <agraf@suse.de>
8  *
9  * This library is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2 of the License, or (at your option) any later version.
13  *
14  * This library is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
21  *
22  */
23 
24 #include "qemu/osdep.h"
25 #include "hw/hw.h"
26 #include "hw/pci/msi.h"
27 #include "hw/pci/pci.h"
28 
29 #include "qemu/error-report.h"
30 #include "sysemu/block-backend.h"
31 #include "sysemu/dma.h"
32 #include "hw/ide/internal.h"
33 #include "hw/ide/pci.h"
34 #include "hw/ide/ahci_internal.h"
35 
36 #include "trace.h"
37 
38 static void check_cmd(AHCIState *s, int port);
39 static int handle_cmd(AHCIState *s, int port, uint8_t slot);
40 static void ahci_reset_port(AHCIState *s, int port);
41 static bool ahci_write_fis_d2h(AHCIDevice *ad);
42 static void ahci_init_d2h(AHCIDevice *ad);
43 static int ahci_dma_prepare_buf(IDEDMA *dma, int32_t limit);
44 static bool ahci_map_clb_address(AHCIDevice *ad);
45 static bool ahci_map_fis_address(AHCIDevice *ad);
46 static void ahci_unmap_clb_address(AHCIDevice *ad);
47 static void ahci_unmap_fis_address(AHCIDevice *ad);
48 
49 static const char *AHCIPortIRQ_lookup[AHCI_PORT_IRQ__COUNT] = {
50     [AHCI_PORT_IRQ_BIT_DHRS] = "DHRS",
51     [AHCI_PORT_IRQ_BIT_PSS]  = "PSS",
52     [AHCI_PORT_IRQ_BIT_DSS]  = "DSS",
53     [AHCI_PORT_IRQ_BIT_SDBS] = "SDBS",
54     [AHCI_PORT_IRQ_BIT_UFS]  = "UFS",
55     [AHCI_PORT_IRQ_BIT_DPS]  = "DPS",
56     [AHCI_PORT_IRQ_BIT_PCS]  = "PCS",
57     [AHCI_PORT_IRQ_BIT_DMPS] = "DMPS",
58     [8 ... 21]               = "RESERVED",
59     [AHCI_PORT_IRQ_BIT_PRCS] = "PRCS",
60     [AHCI_PORT_IRQ_BIT_IPMS] = "IPMS",
61     [AHCI_PORT_IRQ_BIT_OFS]  = "OFS",
62     [25]                     = "RESERVED",
63     [AHCI_PORT_IRQ_BIT_INFS] = "INFS",
64     [AHCI_PORT_IRQ_BIT_IFS]  = "IFS",
65     [AHCI_PORT_IRQ_BIT_HBDS] = "HBDS",
66     [AHCI_PORT_IRQ_BIT_HBFS] = "HBFS",
67     [AHCI_PORT_IRQ_BIT_TFES] = "TFES",
68     [AHCI_PORT_IRQ_BIT_CPDS] = "CPDS"
69 };
70 
71 static uint32_t  ahci_port_read(AHCIState *s, int port, int offset)
72 {
73     uint32_t val;
74     AHCIPortRegs *pr;
75     pr = &s->dev[port].port_regs;
76 
77     switch (offset) {
78     case PORT_LST_ADDR:
79         val = pr->lst_addr;
80         break;
81     case PORT_LST_ADDR_HI:
82         val = pr->lst_addr_hi;
83         break;
84     case PORT_FIS_ADDR:
85         val = pr->fis_addr;
86         break;
87     case PORT_FIS_ADDR_HI:
88         val = pr->fis_addr_hi;
89         break;
90     case PORT_IRQ_STAT:
91         val = pr->irq_stat;
92         break;
93     case PORT_IRQ_MASK:
94         val = pr->irq_mask;
95         break;
96     case PORT_CMD:
97         val = pr->cmd;
98         break;
99     case PORT_TFDATA:
100         val = pr->tfdata;
101         break;
102     case PORT_SIG:
103         val = pr->sig;
104         break;
105     case PORT_SCR_STAT:
106         if (s->dev[port].port.ifs[0].blk) {
107             val = SATA_SCR_SSTATUS_DET_DEV_PRESENT_PHY_UP |
108                   SATA_SCR_SSTATUS_SPD_GEN1 | SATA_SCR_SSTATUS_IPM_ACTIVE;
109         } else {
110             val = SATA_SCR_SSTATUS_DET_NODEV;
111         }
112         break;
113     case PORT_SCR_CTL:
114         val = pr->scr_ctl;
115         break;
116     case PORT_SCR_ERR:
117         val = pr->scr_err;
118         break;
119     case PORT_SCR_ACT:
120         val = pr->scr_act;
121         break;
122     case PORT_CMD_ISSUE:
123         val = pr->cmd_issue;
124         break;
125     case PORT_RESERVED:
126     default:
127         val = 0;
128     }
129 
130     trace_ahci_port_read(s, port, offset, val);
131     return val;
132 }
133 
134 static void ahci_irq_raise(AHCIState *s, AHCIDevice *dev)
135 {
136     DeviceState *dev_state = s->container;
137     PCIDevice *pci_dev = (PCIDevice *) object_dynamic_cast(OBJECT(dev_state),
138                                                            TYPE_PCI_DEVICE);
139 
140     trace_ahci_irq_raise(s);
141 
142     if (pci_dev && msi_enabled(pci_dev)) {
143         msi_notify(pci_dev, 0);
144     } else {
145         qemu_irq_raise(s->irq);
146     }
147 }
148 
149 static void ahci_irq_lower(AHCIState *s, AHCIDevice *dev)
150 {
151     DeviceState *dev_state = s->container;
152     PCIDevice *pci_dev = (PCIDevice *) object_dynamic_cast(OBJECT(dev_state),
153                                                            TYPE_PCI_DEVICE);
154 
155     trace_ahci_irq_lower(s);
156 
157     if (!pci_dev || !msi_enabled(pci_dev)) {
158         qemu_irq_lower(s->irq);
159     }
160 }
161 
162 static void ahci_check_irq(AHCIState *s)
163 {
164     int i;
165     uint32_t old_irq = s->control_regs.irqstatus;
166 
167     s->control_regs.irqstatus = 0;
168     for (i = 0; i < s->ports; i++) {
169         AHCIPortRegs *pr = &s->dev[i].port_regs;
170         if (pr->irq_stat & pr->irq_mask) {
171             s->control_regs.irqstatus |= (1 << i);
172         }
173     }
174     trace_ahci_check_irq(s, old_irq, s->control_regs.irqstatus);
175     if (s->control_regs.irqstatus &&
176         (s->control_regs.ghc & HOST_CTL_IRQ_EN)) {
177             ahci_irq_raise(s, NULL);
178     } else {
179         ahci_irq_lower(s, NULL);
180     }
181 }
182 
183 static void ahci_trigger_irq(AHCIState *s, AHCIDevice *d,
184                              enum AHCIPortIRQ irqbit)
185 {
186     g_assert((unsigned)irqbit < 32);
187     uint32_t irq = 1U << irqbit;
188     uint32_t irqstat = d->port_regs.irq_stat | irq;
189 
190     trace_ahci_trigger_irq(s, d->port_no,
191                            AHCIPortIRQ_lookup[irqbit], irq,
192                            d->port_regs.irq_stat, irqstat,
193                            irqstat & d->port_regs.irq_mask);
194 
195     d->port_regs.irq_stat = irqstat;
196     ahci_check_irq(s);
197 }
198 
199 static void map_page(AddressSpace *as, uint8_t **ptr, uint64_t addr,
200                      uint32_t wanted)
201 {
202     hwaddr len = wanted;
203 
204     if (*ptr) {
205         dma_memory_unmap(as, *ptr, len, DMA_DIRECTION_FROM_DEVICE, len);
206     }
207 
208     *ptr = dma_memory_map(as, addr, &len, DMA_DIRECTION_FROM_DEVICE);
209     if (len < wanted) {
210         dma_memory_unmap(as, *ptr, len, DMA_DIRECTION_FROM_DEVICE, len);
211         *ptr = NULL;
212     }
213 }
214 
215 /**
216  * Check the cmd register to see if we should start or stop
217  * the DMA or FIS RX engines.
218  *
219  * @ad: Device to dis/engage.
220  *
221  * @return 0 on success, -1 on error.
222  */
223 static int ahci_cond_start_engines(AHCIDevice *ad)
224 {
225     AHCIPortRegs *pr = &ad->port_regs;
226     bool cmd_start = pr->cmd & PORT_CMD_START;
227     bool cmd_on    = pr->cmd & PORT_CMD_LIST_ON;
228     bool fis_start = pr->cmd & PORT_CMD_FIS_RX;
229     bool fis_on    = pr->cmd & PORT_CMD_FIS_ON;
230 
231     if (cmd_start && !cmd_on) {
232         if (!ahci_map_clb_address(ad)) {
233             pr->cmd &= ~PORT_CMD_START;
234             error_report("AHCI: Failed to start DMA engine: "
235                          "bad command list buffer address");
236             return -1;
237         }
238     } else if (!cmd_start && cmd_on) {
239         ahci_unmap_clb_address(ad);
240     }
241 
242     if (fis_start && !fis_on) {
243         if (!ahci_map_fis_address(ad)) {
244             pr->cmd &= ~PORT_CMD_FIS_RX;
245             error_report("AHCI: Failed to start FIS receive engine: "
246                          "bad FIS receive buffer address");
247             return -1;
248         }
249     } else if (!fis_start && fis_on) {
250         ahci_unmap_fis_address(ad);
251     }
252 
253     return 0;
254 }
255 
256 static void  ahci_port_write(AHCIState *s, int port, int offset, uint32_t val)
257 {
258     AHCIPortRegs *pr = &s->dev[port].port_regs;
259 
260     trace_ahci_port_write(s, port, offset, val);
261     switch (offset) {
262         case PORT_LST_ADDR:
263             pr->lst_addr = val;
264             break;
265         case PORT_LST_ADDR_HI:
266             pr->lst_addr_hi = val;
267             break;
268         case PORT_FIS_ADDR:
269             pr->fis_addr = val;
270             break;
271         case PORT_FIS_ADDR_HI:
272             pr->fis_addr_hi = val;
273             break;
274         case PORT_IRQ_STAT:
275             pr->irq_stat &= ~val;
276             ahci_check_irq(s);
277             break;
278         case PORT_IRQ_MASK:
279             pr->irq_mask = val & 0xfdc000ff;
280             ahci_check_irq(s);
281             break;
282         case PORT_CMD:
283             /* Block any Read-only fields from being set;
284              * including LIST_ON and FIS_ON.
285              * The spec requires to set ICC bits to zero after the ICC change
286              * is done. We don't support ICC state changes, therefore always
287              * force the ICC bits to zero.
288              */
289             pr->cmd = (pr->cmd & PORT_CMD_RO_MASK) |
290                       (val & ~(PORT_CMD_RO_MASK|PORT_CMD_ICC_MASK));
291 
292             /* Check FIS RX and CLB engines */
293             ahci_cond_start_engines(&s->dev[port]);
294 
295             /* XXX usually the FIS would be pending on the bus here and
296                    issuing deferred until the OS enables FIS receival.
297                    Instead, we only submit it once - which works in most
298                    cases, but is a hack. */
299             if ((pr->cmd & PORT_CMD_FIS_ON) &&
300                 !s->dev[port].init_d2h_sent) {
301                 ahci_init_d2h(&s->dev[port]);
302             }
303 
304             check_cmd(s, port);
305             break;
306         case PORT_TFDATA:
307             /* Read Only. */
308             break;
309         case PORT_SIG:
310             /* Read Only */
311             break;
312         case PORT_SCR_STAT:
313             /* Read Only */
314             break;
315         case PORT_SCR_CTL:
316             if (((pr->scr_ctl & AHCI_SCR_SCTL_DET) == 1) &&
317                 ((val & AHCI_SCR_SCTL_DET) == 0)) {
318                 ahci_reset_port(s, port);
319             }
320             pr->scr_ctl = val;
321             break;
322         case PORT_SCR_ERR:
323             pr->scr_err &= ~val;
324             break;
325         case PORT_SCR_ACT:
326             /* RW1 */
327             pr->scr_act |= val;
328             break;
329         case PORT_CMD_ISSUE:
330             pr->cmd_issue |= val;
331             check_cmd(s, port);
332             break;
333         default:
334             break;
335     }
336 }
337 
338 static uint64_t ahci_mem_read_32(void *opaque, hwaddr addr)
339 {
340     AHCIState *s = opaque;
341     uint32_t val = 0;
342 
343     if (addr < AHCI_GENERIC_HOST_CONTROL_REGS_MAX_ADDR) {
344         switch (addr) {
345         case HOST_CAP:
346             val = s->control_regs.cap;
347             break;
348         case HOST_CTL:
349             val = s->control_regs.ghc;
350             break;
351         case HOST_IRQ_STAT:
352             val = s->control_regs.irqstatus;
353             break;
354         case HOST_PORTS_IMPL:
355             val = s->control_regs.impl;
356             break;
357         case HOST_VERSION:
358             val = s->control_regs.version;
359             break;
360         }
361     } else if ((addr >= AHCI_PORT_REGS_START_ADDR) &&
362                (addr < (AHCI_PORT_REGS_START_ADDR +
363                 (s->ports * AHCI_PORT_ADDR_OFFSET_LEN)))) {
364         val = ahci_port_read(s, (addr - AHCI_PORT_REGS_START_ADDR) >> 7,
365                              addr & AHCI_PORT_ADDR_OFFSET_MASK);
366     }
367 
368     trace_ahci_mem_read_32(s, addr, val);
369     return val;
370 }
371 
372 
373 /**
374  * AHCI 1.3 section 3 ("HBA Memory Registers")
375  * Support unaligned 8/16/32 bit reads, and 64 bit aligned reads.
376  * Caller is responsible for masking unwanted higher order bytes.
377  */
378 static uint64_t ahci_mem_read(void *opaque, hwaddr addr, unsigned size)
379 {
380     hwaddr aligned = addr & ~0x3;
381     int ofst = addr - aligned;
382     uint64_t lo = ahci_mem_read_32(opaque, aligned);
383     uint64_t hi;
384     uint64_t val;
385 
386     /* if < 8 byte read does not cross 4 byte boundary */
387     if (ofst + size <= 4) {
388         val = lo >> (ofst * 8);
389     } else {
390         g_assert_cmpint(size, >, 1);
391 
392         /* If the 64bit read is unaligned, we will produce undefined
393          * results. AHCI does not support unaligned 64bit reads. */
394         hi = ahci_mem_read_32(opaque, aligned + 4);
395         val = (hi << 32 | lo) >> (ofst * 8);
396     }
397 
398     trace_ahci_mem_read(opaque, size, addr, val);
399     return val;
400 }
401 
402 
403 static void ahci_mem_write(void *opaque, hwaddr addr,
404                            uint64_t val, unsigned size)
405 {
406     AHCIState *s = opaque;
407 
408     trace_ahci_mem_write(s, size, addr, val);
409 
410     /* Only aligned reads are allowed on AHCI */
411     if (addr & 3) {
412         fprintf(stderr, "ahci: Mis-aligned write to addr 0x"
413                 TARGET_FMT_plx "\n", addr);
414         return;
415     }
416 
417     if (addr < AHCI_GENERIC_HOST_CONTROL_REGS_MAX_ADDR) {
418         switch (addr) {
419             case HOST_CAP: /* R/WO, RO */
420                 /* FIXME handle R/WO */
421                 break;
422             case HOST_CTL: /* R/W */
423                 if (val & HOST_CTL_RESET) {
424                     ahci_reset(s);
425                 } else {
426                     s->control_regs.ghc = (val & 0x3) | HOST_CTL_AHCI_EN;
427                     ahci_check_irq(s);
428                 }
429                 break;
430             case HOST_IRQ_STAT: /* R/WC, RO */
431                 s->control_regs.irqstatus &= ~val;
432                 ahci_check_irq(s);
433                 break;
434             case HOST_PORTS_IMPL: /* R/WO, RO */
435                 /* FIXME handle R/WO */
436                 break;
437             case HOST_VERSION: /* RO */
438                 /* FIXME report write? */
439                 break;
440             default:
441                 trace_ahci_mem_write_unknown(s, size, addr, val);
442         }
443     } else if ((addr >= AHCI_PORT_REGS_START_ADDR) &&
444                (addr < (AHCI_PORT_REGS_START_ADDR +
445                 (s->ports * AHCI_PORT_ADDR_OFFSET_LEN)))) {
446         ahci_port_write(s, (addr - AHCI_PORT_REGS_START_ADDR) >> 7,
447                         addr & AHCI_PORT_ADDR_OFFSET_MASK, val);
448     }
449 
450 }
451 
452 static const MemoryRegionOps ahci_mem_ops = {
453     .read = ahci_mem_read,
454     .write = ahci_mem_write,
455     .endianness = DEVICE_LITTLE_ENDIAN,
456 };
457 
458 static uint64_t ahci_idp_read(void *opaque, hwaddr addr,
459                               unsigned size)
460 {
461     AHCIState *s = opaque;
462 
463     if (addr == s->idp_offset) {
464         /* index register */
465         return s->idp_index;
466     } else if (addr == s->idp_offset + 4) {
467         /* data register - do memory read at location selected by index */
468         return ahci_mem_read(opaque, s->idp_index, size);
469     } else {
470         return 0;
471     }
472 }
473 
474 static void ahci_idp_write(void *opaque, hwaddr addr,
475                            uint64_t val, unsigned size)
476 {
477     AHCIState *s = opaque;
478 
479     if (addr == s->idp_offset) {
480         /* index register - mask off reserved bits */
481         s->idp_index = (uint32_t)val & ((AHCI_MEM_BAR_SIZE - 1) & ~3);
482     } else if (addr == s->idp_offset + 4) {
483         /* data register - do memory write at location selected by index */
484         ahci_mem_write(opaque, s->idp_index, val, size);
485     }
486 }
487 
488 static const MemoryRegionOps ahci_idp_ops = {
489     .read = ahci_idp_read,
490     .write = ahci_idp_write,
491     .endianness = DEVICE_LITTLE_ENDIAN,
492 };
493 
494 
495 static void ahci_reg_init(AHCIState *s)
496 {
497     int i;
498 
499     s->control_regs.cap = (s->ports - 1) |
500                           (AHCI_NUM_COMMAND_SLOTS << 8) |
501                           (AHCI_SUPPORTED_SPEED_GEN1 << AHCI_SUPPORTED_SPEED) |
502                           HOST_CAP_NCQ | HOST_CAP_AHCI | HOST_CAP_64;
503 
504     s->control_regs.impl = (1 << s->ports) - 1;
505 
506     s->control_regs.version = AHCI_VERSION_1_0;
507 
508     for (i = 0; i < s->ports; i++) {
509         s->dev[i].port_state = STATE_RUN;
510     }
511 }
512 
513 static void check_cmd(AHCIState *s, int port)
514 {
515     AHCIPortRegs *pr = &s->dev[port].port_regs;
516     uint8_t slot;
517 
518     if ((pr->cmd & PORT_CMD_START) && pr->cmd_issue) {
519         for (slot = 0; (slot < 32) && pr->cmd_issue; slot++) {
520             if ((pr->cmd_issue & (1U << slot)) &&
521                 !handle_cmd(s, port, slot)) {
522                 pr->cmd_issue &= ~(1U << slot);
523             }
524         }
525     }
526 }
527 
528 static void ahci_check_cmd_bh(void *opaque)
529 {
530     AHCIDevice *ad = opaque;
531 
532     qemu_bh_delete(ad->check_bh);
533     ad->check_bh = NULL;
534 
535     if ((ad->busy_slot != -1) &&
536         !(ad->port.ifs[0].status & (BUSY_STAT|DRQ_STAT))) {
537         /* no longer busy */
538         ad->port_regs.cmd_issue &= ~(1 << ad->busy_slot);
539         ad->busy_slot = -1;
540     }
541 
542     check_cmd(ad->hba, ad->port_no);
543 }
544 
545 static void ahci_init_d2h(AHCIDevice *ad)
546 {
547     IDEState *ide_state = &ad->port.ifs[0];
548     AHCIPortRegs *pr = &ad->port_regs;
549 
550     if (ad->init_d2h_sent) {
551         return;
552     }
553 
554     if (ahci_write_fis_d2h(ad)) {
555         ad->init_d2h_sent = true;
556         /* We're emulating receiving the first Reg H2D Fis from the device;
557          * Update the SIG register, but otherwise proceed as normal. */
558         pr->sig = ((uint32_t)ide_state->hcyl << 24) |
559             (ide_state->lcyl << 16) |
560             (ide_state->sector << 8) |
561             (ide_state->nsector & 0xFF);
562     }
563 }
564 
565 static void ahci_set_signature(AHCIDevice *ad, uint32_t sig)
566 {
567     IDEState *s = &ad->port.ifs[0];
568     s->hcyl = sig >> 24 & 0xFF;
569     s->lcyl = sig >> 16 & 0xFF;
570     s->sector = sig >> 8 & 0xFF;
571     s->nsector = sig & 0xFF;
572 
573     trace_ahci_set_signature(ad->hba, ad->port_no, s->nsector, s->sector,
574                              s->lcyl, s->hcyl, sig);
575 }
576 
577 static void ahci_reset_port(AHCIState *s, int port)
578 {
579     AHCIDevice *d = &s->dev[port];
580     AHCIPortRegs *pr = &d->port_regs;
581     IDEState *ide_state = &d->port.ifs[0];
582     int i;
583 
584     trace_ahci_reset_port(s, port);
585 
586     ide_bus_reset(&d->port);
587     ide_state->ncq_queues = AHCI_MAX_CMDS;
588 
589     pr->scr_stat = 0;
590     pr->scr_err = 0;
591     pr->scr_act = 0;
592     pr->tfdata = 0x7F;
593     pr->sig = 0xFFFFFFFF;
594     d->busy_slot = -1;
595     d->init_d2h_sent = false;
596 
597     ide_state = &s->dev[port].port.ifs[0];
598     if (!ide_state->blk) {
599         return;
600     }
601 
602     /* reset ncq queue */
603     for (i = 0; i < AHCI_MAX_CMDS; i++) {
604         NCQTransferState *ncq_tfs = &s->dev[port].ncq_tfs[i];
605         ncq_tfs->halt = false;
606         if (!ncq_tfs->used) {
607             continue;
608         }
609 
610         if (ncq_tfs->aiocb) {
611             blk_aio_cancel(ncq_tfs->aiocb);
612             ncq_tfs->aiocb = NULL;
613         }
614 
615         /* Maybe we just finished the request thanks to blk_aio_cancel() */
616         if (!ncq_tfs->used) {
617             continue;
618         }
619 
620         qemu_sglist_destroy(&ncq_tfs->sglist);
621         ncq_tfs->used = 0;
622     }
623 
624     s->dev[port].port_state = STATE_RUN;
625     if (ide_state->drive_kind == IDE_CD) {
626         ahci_set_signature(d, SATA_SIGNATURE_CDROM);\
627         ide_state->status = SEEK_STAT | WRERR_STAT | READY_STAT;
628     } else {
629         ahci_set_signature(d, SATA_SIGNATURE_DISK);
630         ide_state->status = SEEK_STAT | WRERR_STAT;
631     }
632 
633     ide_state->error = 1;
634     ahci_init_d2h(d);
635 }
636 
637 /* Buffer pretty output based on a raw FIS structure. */
638 static char *ahci_pretty_buffer_fis(uint8_t *fis, int cmd_len)
639 {
640     int i;
641     GString *s = g_string_new("FIS:");
642 
643     for (i = 0; i < cmd_len; i++) {
644         if ((i & 0xf) == 0) {
645             g_string_append_printf(s, "\n0x%02x: ", i);
646         }
647         g_string_append_printf(s, "%02x ", fis[i]);
648     }
649     g_string_append_c(s, '\n');
650 
651     return g_string_free(s, FALSE);
652 }
653 
654 static bool ahci_map_fis_address(AHCIDevice *ad)
655 {
656     AHCIPortRegs *pr = &ad->port_regs;
657     map_page(ad->hba->as, &ad->res_fis,
658              ((uint64_t)pr->fis_addr_hi << 32) | pr->fis_addr, 256);
659     if (ad->res_fis != NULL) {
660         pr->cmd |= PORT_CMD_FIS_ON;
661         return true;
662     }
663 
664     pr->cmd &= ~PORT_CMD_FIS_ON;
665     return false;
666 }
667 
668 static void ahci_unmap_fis_address(AHCIDevice *ad)
669 {
670     if (ad->res_fis == NULL) {
671         trace_ahci_unmap_fis_address_null(ad->hba, ad->port_no);
672         return;
673     }
674     ad->port_regs.cmd &= ~PORT_CMD_FIS_ON;
675     dma_memory_unmap(ad->hba->as, ad->res_fis, 256,
676                      DMA_DIRECTION_FROM_DEVICE, 256);
677     ad->res_fis = NULL;
678 }
679 
680 static bool ahci_map_clb_address(AHCIDevice *ad)
681 {
682     AHCIPortRegs *pr = &ad->port_regs;
683     ad->cur_cmd = NULL;
684     map_page(ad->hba->as, &ad->lst,
685              ((uint64_t)pr->lst_addr_hi << 32) | pr->lst_addr, 1024);
686     if (ad->lst != NULL) {
687         pr->cmd |= PORT_CMD_LIST_ON;
688         return true;
689     }
690 
691     pr->cmd &= ~PORT_CMD_LIST_ON;
692     return false;
693 }
694 
695 static void ahci_unmap_clb_address(AHCIDevice *ad)
696 {
697     if (ad->lst == NULL) {
698         trace_ahci_unmap_clb_address_null(ad->hba, ad->port_no);
699         return;
700     }
701     ad->port_regs.cmd &= ~PORT_CMD_LIST_ON;
702     dma_memory_unmap(ad->hba->as, ad->lst, 1024,
703                      DMA_DIRECTION_FROM_DEVICE, 1024);
704     ad->lst = NULL;
705 }
706 
707 static void ahci_write_fis_sdb(AHCIState *s, NCQTransferState *ncq_tfs)
708 {
709     AHCIDevice *ad = ncq_tfs->drive;
710     AHCIPortRegs *pr = &ad->port_regs;
711     IDEState *ide_state;
712     SDBFIS *sdb_fis;
713 
714     if (!ad->res_fis ||
715         !(pr->cmd & PORT_CMD_FIS_RX)) {
716         return;
717     }
718 
719     sdb_fis = (SDBFIS *)&ad->res_fis[RES_FIS_SDBFIS];
720     ide_state = &ad->port.ifs[0];
721 
722     sdb_fis->type = SATA_FIS_TYPE_SDB;
723     /* Interrupt pending & Notification bit */
724     sdb_fis->flags = 0x40; /* Interrupt bit, always 1 for NCQ */
725     sdb_fis->status = ide_state->status & 0x77;
726     sdb_fis->error = ide_state->error;
727     /* update SAct field in SDB_FIS */
728     sdb_fis->payload = cpu_to_le32(ad->finished);
729 
730     /* Update shadow registers (except BSY 0x80 and DRQ 0x08) */
731     pr->tfdata = (ad->port.ifs[0].error << 8) |
732         (ad->port.ifs[0].status & 0x77) |
733         (pr->tfdata & 0x88);
734     pr->scr_act &= ~ad->finished;
735     ad->finished = 0;
736 
737     /* Trigger IRQ if interrupt bit is set (which currently, it always is) */
738     if (sdb_fis->flags & 0x40) {
739         ahci_trigger_irq(s, ad, AHCI_PORT_IRQ_BIT_SDBS);
740     }
741 }
742 
743 static void ahci_write_fis_pio(AHCIDevice *ad, uint16_t len)
744 {
745     AHCIPortRegs *pr = &ad->port_regs;
746     uint8_t *pio_fis;
747     IDEState *s = &ad->port.ifs[0];
748 
749     if (!ad->res_fis || !(pr->cmd & PORT_CMD_FIS_RX)) {
750         return;
751     }
752 
753     pio_fis = &ad->res_fis[RES_FIS_PSFIS];
754 
755     pio_fis[0] = SATA_FIS_TYPE_PIO_SETUP;
756     pio_fis[1] = (ad->hba->control_regs.irqstatus ? (1 << 6) : 0);
757     pio_fis[2] = s->status;
758     pio_fis[3] = s->error;
759 
760     pio_fis[4] = s->sector;
761     pio_fis[5] = s->lcyl;
762     pio_fis[6] = s->hcyl;
763     pio_fis[7] = s->select;
764     pio_fis[8] = s->hob_sector;
765     pio_fis[9] = s->hob_lcyl;
766     pio_fis[10] = s->hob_hcyl;
767     pio_fis[11] = 0;
768     pio_fis[12] = s->nsector & 0xFF;
769     pio_fis[13] = (s->nsector >> 8) & 0xFF;
770     pio_fis[14] = 0;
771     pio_fis[15] = s->status;
772     pio_fis[16] = len & 255;
773     pio_fis[17] = len >> 8;
774     pio_fis[18] = 0;
775     pio_fis[19] = 0;
776 
777     /* Update shadow registers: */
778     pr->tfdata = (ad->port.ifs[0].error << 8) |
779         ad->port.ifs[0].status;
780 
781     if (pio_fis[2] & ERR_STAT) {
782         ahci_trigger_irq(ad->hba, ad, AHCI_PORT_IRQ_BIT_TFES);
783     }
784 
785     ahci_trigger_irq(ad->hba, ad, AHCI_PORT_IRQ_BIT_PSS);
786 }
787 
788 static bool ahci_write_fis_d2h(AHCIDevice *ad)
789 {
790     AHCIPortRegs *pr = &ad->port_regs;
791     uint8_t *d2h_fis;
792     int i;
793     IDEState *s = &ad->port.ifs[0];
794 
795     if (!ad->res_fis || !(pr->cmd & PORT_CMD_FIS_RX)) {
796         return false;
797     }
798 
799     d2h_fis = &ad->res_fis[RES_FIS_RFIS];
800 
801     d2h_fis[0] = SATA_FIS_TYPE_REGISTER_D2H;
802     d2h_fis[1] = (ad->hba->control_regs.irqstatus ? (1 << 6) : 0);
803     d2h_fis[2] = s->status;
804     d2h_fis[3] = s->error;
805 
806     d2h_fis[4] = s->sector;
807     d2h_fis[5] = s->lcyl;
808     d2h_fis[6] = s->hcyl;
809     d2h_fis[7] = s->select;
810     d2h_fis[8] = s->hob_sector;
811     d2h_fis[9] = s->hob_lcyl;
812     d2h_fis[10] = s->hob_hcyl;
813     d2h_fis[11] = 0;
814     d2h_fis[12] = s->nsector & 0xFF;
815     d2h_fis[13] = (s->nsector >> 8) & 0xFF;
816     for (i = 14; i < 20; i++) {
817         d2h_fis[i] = 0;
818     }
819 
820     /* Update shadow registers: */
821     pr->tfdata = (ad->port.ifs[0].error << 8) |
822         ad->port.ifs[0].status;
823 
824     if (d2h_fis[2] & ERR_STAT) {
825         ahci_trigger_irq(ad->hba, ad, AHCI_PORT_IRQ_BIT_TFES);
826     }
827 
828     ahci_trigger_irq(ad->hba, ad, AHCI_PORT_IRQ_BIT_DHRS);
829     return true;
830 }
831 
832 static int prdt_tbl_entry_size(const AHCI_SG *tbl)
833 {
834     /* flags_size is zero-based */
835     return (le32_to_cpu(tbl->flags_size) & AHCI_PRDT_SIZE_MASK) + 1;
836 }
837 
838 /**
839  * Fetch entries in a guest-provided PRDT and convert it into a QEMU SGlist.
840  * @ad: The AHCIDevice for whom we are building the SGList.
841  * @sglist: The SGList target to add PRD entries to.
842  * @cmd: The AHCI Command Header that describes where the PRDT is.
843  * @limit: The remaining size of the S/ATA transaction, in bytes.
844  * @offset: The number of bytes already transferred, in bytes.
845  *
846  * The AHCI PRDT can describe up to 256GiB. S/ATA only support transactions of
847  * up to 32MiB as of ATA8-ACS3 rev 1b, assuming a 512 byte sector size. We stop
848  * building the sglist from the PRDT as soon as we hit @limit bytes,
849  * which is <= INT32_MAX/2GiB.
850  */
851 static int ahci_populate_sglist(AHCIDevice *ad, QEMUSGList *sglist,
852                                 AHCICmdHdr *cmd, int64_t limit, uint64_t offset)
853 {
854     uint16_t opts = le16_to_cpu(cmd->opts);
855     uint16_t prdtl = le16_to_cpu(cmd->prdtl);
856     uint64_t cfis_addr = le64_to_cpu(cmd->tbl_addr);
857     uint64_t prdt_addr = cfis_addr + 0x80;
858     dma_addr_t prdt_len = (prdtl * sizeof(AHCI_SG));
859     dma_addr_t real_prdt_len = prdt_len;
860     uint8_t *prdt;
861     int i;
862     int r = 0;
863     uint64_t sum = 0;
864     int off_idx = -1;
865     int64_t off_pos = -1;
866     int tbl_entry_size;
867     IDEBus *bus = &ad->port;
868     BusState *qbus = BUS(bus);
869 
870     trace_ahci_populate_sglist(ad->hba, ad->port_no);
871 
872     if (!prdtl) {
873         trace_ahci_populate_sglist_no_prdtl(ad->hba, ad->port_no, opts);
874         return -1;
875     }
876 
877     /* map PRDT */
878     if (!(prdt = dma_memory_map(ad->hba->as, prdt_addr, &prdt_len,
879                                 DMA_DIRECTION_TO_DEVICE))){
880         trace_ahci_populate_sglist_no_map(ad->hba, ad->port_no);
881         return -1;
882     }
883 
884     if (prdt_len < real_prdt_len) {
885         trace_ahci_populate_sglist_short_map(ad->hba, ad->port_no);
886         r = -1;
887         goto out;
888     }
889 
890     /* Get entries in the PRDT, init a qemu sglist accordingly */
891     if (prdtl > 0) {
892         AHCI_SG *tbl = (AHCI_SG *)prdt;
893         sum = 0;
894         for (i = 0; i < prdtl; i++) {
895             tbl_entry_size = prdt_tbl_entry_size(&tbl[i]);
896             if (offset < (sum + tbl_entry_size)) {
897                 off_idx = i;
898                 off_pos = offset - sum;
899                 break;
900             }
901             sum += tbl_entry_size;
902         }
903         if ((off_idx == -1) || (off_pos < 0) || (off_pos > tbl_entry_size)) {
904             trace_ahci_populate_sglist_bad_offset(ad->hba, ad->port_no,
905                                                   off_idx, off_pos);
906             r = -1;
907             goto out;
908         }
909 
910         qemu_sglist_init(sglist, qbus->parent, (prdtl - off_idx),
911                          ad->hba->as);
912         qemu_sglist_add(sglist, le64_to_cpu(tbl[off_idx].addr) + off_pos,
913                         MIN(prdt_tbl_entry_size(&tbl[off_idx]) - off_pos,
914                             limit));
915 
916         for (i = off_idx + 1; i < prdtl && sglist->size < limit; i++) {
917             qemu_sglist_add(sglist, le64_to_cpu(tbl[i].addr),
918                             MIN(prdt_tbl_entry_size(&tbl[i]),
919                                 limit - sglist->size));
920         }
921     }
922 
923 out:
924     dma_memory_unmap(ad->hba->as, prdt, prdt_len,
925                      DMA_DIRECTION_TO_DEVICE, prdt_len);
926     return r;
927 }
928 
929 static void ncq_err(NCQTransferState *ncq_tfs)
930 {
931     IDEState *ide_state = &ncq_tfs->drive->port.ifs[0];
932 
933     ide_state->error = ABRT_ERR;
934     ide_state->status = READY_STAT | ERR_STAT;
935     ncq_tfs->drive->port_regs.scr_err |= (1 << ncq_tfs->tag);
936     qemu_sglist_destroy(&ncq_tfs->sglist);
937     ncq_tfs->used = 0;
938 }
939 
940 static void ncq_finish(NCQTransferState *ncq_tfs)
941 {
942     /* If we didn't error out, set our finished bit. Errored commands
943      * do not get a bit set for the SDB FIS ACT register, nor do they
944      * clear the outstanding bit in scr_act (PxSACT). */
945     if (!(ncq_tfs->drive->port_regs.scr_err & (1 << ncq_tfs->tag))) {
946         ncq_tfs->drive->finished |= (1 << ncq_tfs->tag);
947     }
948 
949     ahci_write_fis_sdb(ncq_tfs->drive->hba, ncq_tfs);
950 
951     trace_ncq_finish(ncq_tfs->drive->hba, ncq_tfs->drive->port_no,
952                      ncq_tfs->tag);
953 
954     block_acct_done(blk_get_stats(ncq_tfs->drive->port.ifs[0].blk),
955                     &ncq_tfs->acct);
956     qemu_sglist_destroy(&ncq_tfs->sglist);
957     ncq_tfs->used = 0;
958 }
959 
960 static void ncq_cb(void *opaque, int ret)
961 {
962     NCQTransferState *ncq_tfs = (NCQTransferState *)opaque;
963     IDEState *ide_state = &ncq_tfs->drive->port.ifs[0];
964 
965     ncq_tfs->aiocb = NULL;
966     if (ret == -ECANCELED) {
967         return;
968     }
969 
970     if (ret < 0) {
971         bool is_read = ncq_tfs->cmd == READ_FPDMA_QUEUED;
972         BlockErrorAction action = blk_get_error_action(ide_state->blk,
973                                                        is_read, -ret);
974         if (action == BLOCK_ERROR_ACTION_STOP) {
975             ncq_tfs->halt = true;
976             ide_state->bus->error_status = IDE_RETRY_HBA;
977         } else if (action == BLOCK_ERROR_ACTION_REPORT) {
978             ncq_err(ncq_tfs);
979         }
980         blk_error_action(ide_state->blk, action, is_read, -ret);
981     } else {
982         ide_state->status = READY_STAT | SEEK_STAT;
983     }
984 
985     if (!ncq_tfs->halt) {
986         ncq_finish(ncq_tfs);
987     }
988 }
989 
990 static int is_ncq(uint8_t ata_cmd)
991 {
992     /* Based on SATA 3.2 section 13.6.3.2 */
993     switch (ata_cmd) {
994     case READ_FPDMA_QUEUED:
995     case WRITE_FPDMA_QUEUED:
996     case NCQ_NON_DATA:
997     case RECEIVE_FPDMA_QUEUED:
998     case SEND_FPDMA_QUEUED:
999         return 1;
1000     default:
1001         return 0;
1002     }
1003 }
1004 
1005 static void execute_ncq_command(NCQTransferState *ncq_tfs)
1006 {
1007     AHCIDevice *ad = ncq_tfs->drive;
1008     IDEState *ide_state = &ad->port.ifs[0];
1009     int port = ad->port_no;
1010 
1011     g_assert(is_ncq(ncq_tfs->cmd));
1012     ncq_tfs->halt = false;
1013 
1014     switch (ncq_tfs->cmd) {
1015     case READ_FPDMA_QUEUED:
1016         trace_execute_ncq_command_read(ad->hba, port, ncq_tfs->tag,
1017                                        ncq_tfs->sector_count, ncq_tfs->lba);
1018         dma_acct_start(ide_state->blk, &ncq_tfs->acct,
1019                        &ncq_tfs->sglist, BLOCK_ACCT_READ);
1020         ncq_tfs->aiocb = dma_blk_read(ide_state->blk, &ncq_tfs->sglist,
1021                                       ncq_tfs->lba << BDRV_SECTOR_BITS,
1022                                       BDRV_SECTOR_SIZE,
1023                                       ncq_cb, ncq_tfs);
1024         break;
1025     case WRITE_FPDMA_QUEUED:
1026         trace_execute_ncq_command_read(ad->hba, port, ncq_tfs->tag,
1027                                        ncq_tfs->sector_count, ncq_tfs->lba);
1028         dma_acct_start(ide_state->blk, &ncq_tfs->acct,
1029                        &ncq_tfs->sglist, BLOCK_ACCT_WRITE);
1030         ncq_tfs->aiocb = dma_blk_write(ide_state->blk, &ncq_tfs->sglist,
1031                                        ncq_tfs->lba << BDRV_SECTOR_BITS,
1032                                        BDRV_SECTOR_SIZE,
1033                                        ncq_cb, ncq_tfs);
1034         break;
1035     default:
1036         trace_execute_ncq_command_unsup(ad->hba, port,
1037                                         ncq_tfs->tag, ncq_tfs->cmd);
1038         ncq_err(ncq_tfs);
1039     }
1040 }
1041 
1042 
1043 static void process_ncq_command(AHCIState *s, int port, uint8_t *cmd_fis,
1044                                 uint8_t slot)
1045 {
1046     AHCIDevice *ad = &s->dev[port];
1047     NCQFrame *ncq_fis = (NCQFrame*)cmd_fis;
1048     uint8_t tag = ncq_fis->tag >> 3;
1049     NCQTransferState *ncq_tfs = &ad->ncq_tfs[tag];
1050     size_t size;
1051 
1052     g_assert(is_ncq(ncq_fis->command));
1053     if (ncq_tfs->used) {
1054         /* error - already in use */
1055         fprintf(stderr, "%s: tag %d already used\n", __func__, tag);
1056         return;
1057     }
1058 
1059     ncq_tfs->used = 1;
1060     ncq_tfs->drive = ad;
1061     ncq_tfs->slot = slot;
1062     ncq_tfs->cmdh = &((AHCICmdHdr *)ad->lst)[slot];
1063     ncq_tfs->cmd = ncq_fis->command;
1064     ncq_tfs->lba = ((uint64_t)ncq_fis->lba5 << 40) |
1065                    ((uint64_t)ncq_fis->lba4 << 32) |
1066                    ((uint64_t)ncq_fis->lba3 << 24) |
1067                    ((uint64_t)ncq_fis->lba2 << 16) |
1068                    ((uint64_t)ncq_fis->lba1 << 8) |
1069                    (uint64_t)ncq_fis->lba0;
1070     ncq_tfs->tag = tag;
1071 
1072     /* Sanity-check the NCQ packet */
1073     if (tag != slot) {
1074         trace_process_ncq_command_mismatch(s, port, tag, slot);
1075     }
1076 
1077     if (ncq_fis->aux0 || ncq_fis->aux1 || ncq_fis->aux2 || ncq_fis->aux3) {
1078         trace_process_ncq_command_aux(s, port, tag);
1079     }
1080     if (ncq_fis->prio || ncq_fis->icc) {
1081         trace_process_ncq_command_prioicc(s, port, tag);
1082     }
1083     if (ncq_fis->fua & NCQ_FIS_FUA_MASK) {
1084         trace_process_ncq_command_fua(s, port, tag);
1085     }
1086     if (ncq_fis->tag & NCQ_FIS_RARC_MASK) {
1087         trace_process_ncq_command_rarc(s, port, tag);
1088     }
1089 
1090     ncq_tfs->sector_count = ((ncq_fis->sector_count_high << 8) |
1091                              ncq_fis->sector_count_low);
1092     if (!ncq_tfs->sector_count) {
1093         ncq_tfs->sector_count = 0x10000;
1094     }
1095     size = ncq_tfs->sector_count * 512;
1096     ahci_populate_sglist(ad, &ncq_tfs->sglist, ncq_tfs->cmdh, size, 0);
1097 
1098     if (ncq_tfs->sglist.size < size) {
1099         error_report("ahci: PRDT length for NCQ command (0x%zx) "
1100                      "is smaller than the requested size (0x%zx)",
1101                      ncq_tfs->sglist.size, size);
1102         ncq_err(ncq_tfs);
1103         ahci_trigger_irq(ad->hba, ad, AHCI_PORT_IRQ_BIT_OFS);
1104         return;
1105     } else if (ncq_tfs->sglist.size != size) {
1106         trace_process_ncq_command_large(s, port, tag,
1107                                         ncq_tfs->sglist.size, size);
1108     }
1109 
1110     trace_process_ncq_command(s, port, tag,
1111                               ncq_fis->command,
1112                               ncq_tfs->lba,
1113                               ncq_tfs->lba + ncq_tfs->sector_count - 1);
1114     execute_ncq_command(ncq_tfs);
1115 }
1116 
1117 static AHCICmdHdr *get_cmd_header(AHCIState *s, uint8_t port, uint8_t slot)
1118 {
1119     if (port >= s->ports || slot >= AHCI_MAX_CMDS) {
1120         return NULL;
1121     }
1122 
1123     return s->dev[port].lst ? &((AHCICmdHdr *)s->dev[port].lst)[slot] : NULL;
1124 }
1125 
1126 static void handle_reg_h2d_fis(AHCIState *s, int port,
1127                                uint8_t slot, uint8_t *cmd_fis)
1128 {
1129     IDEState *ide_state = &s->dev[port].port.ifs[0];
1130     AHCICmdHdr *cmd = get_cmd_header(s, port, slot);
1131     uint16_t opts = le16_to_cpu(cmd->opts);
1132 
1133     if (cmd_fis[1] & 0x0F) {
1134         trace_handle_reg_h2d_fis_pmp(s, port, cmd_fis[1],
1135                                      cmd_fis[2], cmd_fis[3]);
1136         return;
1137     }
1138 
1139     if (cmd_fis[1] & 0x70) {
1140         trace_handle_reg_h2d_fis_res(s, port, cmd_fis[1],
1141                                      cmd_fis[2], cmd_fis[3]);
1142         return;
1143     }
1144 
1145     if (!(cmd_fis[1] & SATA_FIS_REG_H2D_UPDATE_COMMAND_REGISTER)) {
1146         switch (s->dev[port].port_state) {
1147         case STATE_RUN:
1148             if (cmd_fis[15] & ATA_SRST) {
1149                 s->dev[port].port_state = STATE_RESET;
1150             }
1151             break;
1152         case STATE_RESET:
1153             if (!(cmd_fis[15] & ATA_SRST)) {
1154                 ahci_reset_port(s, port);
1155             }
1156             break;
1157         }
1158         return;
1159     }
1160 
1161     /* Check for NCQ command */
1162     if (is_ncq(cmd_fis[2])) {
1163         process_ncq_command(s, port, cmd_fis, slot);
1164         return;
1165     }
1166 
1167     /* Decompose the FIS:
1168      * AHCI does not interpret FIS packets, it only forwards them.
1169      * SATA 1.0 describes how to decode LBA28 and CHS FIS packets.
1170      * Later specifications, e.g, SATA 3.2, describe LBA48 FIS packets.
1171      *
1172      * ATA4 describes sector number for LBA28/CHS commands.
1173      * ATA6 describes sector number for LBA48 commands.
1174      * ATA8 deprecates CHS fully, describing only LBA28/48.
1175      *
1176      * We dutifully convert the FIS into IDE registers, and allow the
1177      * core layer to interpret them as needed. */
1178     ide_state->feature = cmd_fis[3];
1179     ide_state->sector = cmd_fis[4];      /* LBA 7:0 */
1180     ide_state->lcyl = cmd_fis[5];        /* LBA 15:8  */
1181     ide_state->hcyl = cmd_fis[6];        /* LBA 23:16 */
1182     ide_state->select = cmd_fis[7];      /* LBA 27:24 (LBA28) */
1183     ide_state->hob_sector = cmd_fis[8];  /* LBA 31:24 */
1184     ide_state->hob_lcyl = cmd_fis[9];    /* LBA 39:32 */
1185     ide_state->hob_hcyl = cmd_fis[10];   /* LBA 47:40 */
1186     ide_state->hob_feature = cmd_fis[11];
1187     ide_state->nsector = (int64_t)((cmd_fis[13] << 8) | cmd_fis[12]);
1188     /* 14, 16, 17, 18, 19: Reserved (SATA 1.0) */
1189     /* 15: Only valid when UPDATE_COMMAND not set. */
1190 
1191     /* Copy the ACMD field (ATAPI packet, if any) from the AHCI command
1192      * table to ide_state->io_buffer */
1193     if (opts & AHCI_CMD_ATAPI) {
1194         memcpy(ide_state->io_buffer, &cmd_fis[AHCI_COMMAND_TABLE_ACMD], 0x10);
1195         if (trace_event_get_state_backends(TRACE_HANDLE_REG_H2D_FIS_DUMP)) {
1196             char *pretty_fis = ahci_pretty_buffer_fis(ide_state->io_buffer, 0x10);
1197             trace_handle_reg_h2d_fis_dump(s, port, pretty_fis);
1198             g_free(pretty_fis);
1199         }
1200         s->dev[port].done_atapi_packet = false;
1201         /* XXX send PIO setup FIS */
1202     }
1203 
1204     ide_state->error = 0;
1205 
1206     /* Reset transferred byte counter */
1207     cmd->status = 0;
1208 
1209     /* We're ready to process the command in FIS byte 2. */
1210     ide_exec_cmd(&s->dev[port].port, cmd_fis[2]);
1211 }
1212 
1213 static int handle_cmd(AHCIState *s, int port, uint8_t slot)
1214 {
1215     IDEState *ide_state;
1216     uint64_t tbl_addr;
1217     AHCICmdHdr *cmd;
1218     uint8_t *cmd_fis;
1219     dma_addr_t cmd_len;
1220 
1221     if (s->dev[port].port.ifs[0].status & (BUSY_STAT|DRQ_STAT)) {
1222         /* Engine currently busy, try again later */
1223         trace_handle_cmd_busy(s, port);
1224         return -1;
1225     }
1226 
1227     if (!s->dev[port].lst) {
1228         trace_handle_cmd_nolist(s, port);
1229         return -1;
1230     }
1231     cmd = get_cmd_header(s, port, slot);
1232     /* remember current slot handle for later */
1233     s->dev[port].cur_cmd = cmd;
1234 
1235     /* The device we are working for */
1236     ide_state = &s->dev[port].port.ifs[0];
1237     if (!ide_state->blk) {
1238         trace_handle_cmd_badport(s, port);
1239         return -1;
1240     }
1241 
1242     tbl_addr = le64_to_cpu(cmd->tbl_addr);
1243     cmd_len = 0x80;
1244     cmd_fis = dma_memory_map(s->as, tbl_addr, &cmd_len,
1245                              DMA_DIRECTION_FROM_DEVICE);
1246     if (!cmd_fis) {
1247         trace_handle_cmd_badfis(s, port);
1248         return -1;
1249     } else if (cmd_len != 0x80) {
1250         ahci_trigger_irq(s, &s->dev[port], AHCI_PORT_IRQ_BIT_HBFS);
1251         trace_handle_cmd_badmap(s, port, cmd_len);
1252         goto out;
1253     }
1254     if (trace_event_get_state_backends(TRACE_HANDLE_CMD_FIS_DUMP)) {
1255         char *pretty_fis = ahci_pretty_buffer_fis(cmd_fis, 0x80);
1256         trace_handle_cmd_fis_dump(s, port, pretty_fis);
1257         g_free(pretty_fis);
1258     }
1259     switch (cmd_fis[0]) {
1260         case SATA_FIS_TYPE_REGISTER_H2D:
1261             handle_reg_h2d_fis(s, port, slot, cmd_fis);
1262             break;
1263         default:
1264             trace_handle_cmd_unhandled_fis(s, port,
1265                                            cmd_fis[0], cmd_fis[1], cmd_fis[2]);
1266             break;
1267     }
1268 
1269 out:
1270     dma_memory_unmap(s->as, cmd_fis, cmd_len, DMA_DIRECTION_FROM_DEVICE,
1271                      cmd_len);
1272 
1273     if (s->dev[port].port.ifs[0].status & (BUSY_STAT|DRQ_STAT)) {
1274         /* async command, complete later */
1275         s->dev[port].busy_slot = slot;
1276         return -1;
1277     }
1278 
1279     /* done handling the command */
1280     return 0;
1281 }
1282 
1283 /* DMA dev <-> ram */
1284 static void ahci_start_transfer(IDEDMA *dma)
1285 {
1286     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1287     IDEState *s = &ad->port.ifs[0];
1288     uint32_t size = (uint32_t)(s->data_end - s->data_ptr);
1289     /* write == ram -> device */
1290     uint16_t opts = le16_to_cpu(ad->cur_cmd->opts);
1291     int is_write = opts & AHCI_CMD_WRITE;
1292     int is_atapi = opts & AHCI_CMD_ATAPI;
1293     int has_sglist = 0;
1294 
1295     if (is_atapi && !ad->done_atapi_packet) {
1296         /* already prepopulated iobuffer */
1297         ad->done_atapi_packet = true;
1298         size = 0;
1299         goto out;
1300     }
1301 
1302     if (ahci_dma_prepare_buf(dma, size)) {
1303         has_sglist = 1;
1304     }
1305 
1306     trace_ahci_start_transfer(ad->hba, ad->port_no, is_write ? "writ" : "read",
1307                               size, is_atapi ? "atapi" : "ata",
1308                               has_sglist ? "" : "o");
1309 
1310     if (has_sglist && size) {
1311         if (is_write) {
1312             dma_buf_write(s->data_ptr, size, &s->sg);
1313         } else {
1314             dma_buf_read(s->data_ptr, size, &s->sg);
1315         }
1316     }
1317 
1318 out:
1319     /* declare that we processed everything */
1320     s->data_ptr = s->data_end;
1321 
1322     /* Update number of transferred bytes, destroy sglist */
1323     dma_buf_commit(s, size);
1324 
1325     s->end_transfer_func(s);
1326 
1327     if (!(s->status & DRQ_STAT)) {
1328         /* done with PIO send/receive */
1329         ahci_write_fis_pio(ad, le32_to_cpu(ad->cur_cmd->status));
1330     }
1331 }
1332 
1333 static void ahci_start_dma(IDEDMA *dma, IDEState *s,
1334                            BlockCompletionFunc *dma_cb)
1335 {
1336     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1337     trace_ahci_start_dma(ad->hba, ad->port_no);
1338     s->io_buffer_offset = 0;
1339     dma_cb(s, 0);
1340 }
1341 
1342 static void ahci_restart_dma(IDEDMA *dma)
1343 {
1344     /* Nothing to do, ahci_start_dma already resets s->io_buffer_offset.  */
1345 }
1346 
1347 /**
1348  * IDE/PIO restarts are handled by the core layer, but NCQ commands
1349  * need an extra kick from the AHCI HBA.
1350  */
1351 static void ahci_restart(IDEDMA *dma)
1352 {
1353     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1354     int i;
1355 
1356     for (i = 0; i < AHCI_MAX_CMDS; i++) {
1357         NCQTransferState *ncq_tfs = &ad->ncq_tfs[i];
1358         if (ncq_tfs->halt) {
1359             execute_ncq_command(ncq_tfs);
1360         }
1361     }
1362 }
1363 
1364 /**
1365  * Called in DMA and PIO R/W chains to read the PRDT.
1366  * Not shared with NCQ pathways.
1367  */
1368 static int32_t ahci_dma_prepare_buf(IDEDMA *dma, int32_t limit)
1369 {
1370     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1371     IDEState *s = &ad->port.ifs[0];
1372 
1373     if (ahci_populate_sglist(ad, &s->sg, ad->cur_cmd,
1374                              limit, s->io_buffer_offset) == -1) {
1375         trace_ahci_dma_prepare_buf_fail(ad->hba, ad->port_no);
1376         return -1;
1377     }
1378     s->io_buffer_size = s->sg.size;
1379 
1380     trace_ahci_dma_prepare_buf(ad->hba, ad->port_no, limit, s->io_buffer_size);
1381     return s->io_buffer_size;
1382 }
1383 
1384 /**
1385  * Updates the command header with a bytes-read value.
1386  * Called via dma_buf_commit, for both DMA and PIO paths.
1387  * sglist destruction is handled within dma_buf_commit.
1388  */
1389 static void ahci_commit_buf(IDEDMA *dma, uint32_t tx_bytes)
1390 {
1391     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1392 
1393     tx_bytes += le32_to_cpu(ad->cur_cmd->status);
1394     ad->cur_cmd->status = cpu_to_le32(tx_bytes);
1395 }
1396 
1397 static int ahci_dma_rw_buf(IDEDMA *dma, int is_write)
1398 {
1399     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1400     IDEState *s = &ad->port.ifs[0];
1401     uint8_t *p = s->io_buffer + s->io_buffer_index;
1402     int l = s->io_buffer_size - s->io_buffer_index;
1403 
1404     if (ahci_populate_sglist(ad, &s->sg, ad->cur_cmd, l, s->io_buffer_offset)) {
1405         return 0;
1406     }
1407 
1408     if (is_write) {
1409         dma_buf_read(p, l, &s->sg);
1410     } else {
1411         dma_buf_write(p, l, &s->sg);
1412     }
1413 
1414     /* free sglist, update byte count */
1415     dma_buf_commit(s, l);
1416     s->io_buffer_index += l;
1417 
1418     trace_ahci_dma_rw_buf(ad->hba, ad->port_no, l);
1419     return 1;
1420 }
1421 
1422 static void ahci_cmd_done(IDEDMA *dma)
1423 {
1424     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1425 
1426     trace_ahci_cmd_done(ad->hba, ad->port_no);
1427 
1428     /* update d2h status */
1429     ahci_write_fis_d2h(ad);
1430 
1431     if (!ad->check_bh) {
1432         /* maybe we still have something to process, check later */
1433         ad->check_bh = qemu_bh_new(ahci_check_cmd_bh, ad);
1434         qemu_bh_schedule(ad->check_bh);
1435     }
1436 }
1437 
1438 static void ahci_irq_set(void *opaque, int n, int level)
1439 {
1440 }
1441 
1442 static const IDEDMAOps ahci_dma_ops = {
1443     .start_dma = ahci_start_dma,
1444     .restart = ahci_restart,
1445     .restart_dma = ahci_restart_dma,
1446     .start_transfer = ahci_start_transfer,
1447     .prepare_buf = ahci_dma_prepare_buf,
1448     .commit_buf = ahci_commit_buf,
1449     .rw_buf = ahci_dma_rw_buf,
1450     .cmd_done = ahci_cmd_done,
1451 };
1452 
1453 void ahci_init(AHCIState *s, DeviceState *qdev)
1454 {
1455     s->container = qdev;
1456     /* XXX BAR size should be 1k, but that breaks, so bump it to 4k for now */
1457     memory_region_init_io(&s->mem, OBJECT(qdev), &ahci_mem_ops, s,
1458                           "ahci", AHCI_MEM_BAR_SIZE);
1459     memory_region_init_io(&s->idp, OBJECT(qdev), &ahci_idp_ops, s,
1460                           "ahci-idp", 32);
1461 }
1462 
1463 void ahci_realize(AHCIState *s, DeviceState *qdev, AddressSpace *as, int ports)
1464 {
1465     qemu_irq *irqs;
1466     int i;
1467 
1468     s->as = as;
1469     s->ports = ports;
1470     s->dev = g_new0(AHCIDevice, ports);
1471     ahci_reg_init(s);
1472     irqs = qemu_allocate_irqs(ahci_irq_set, s, s->ports);
1473     for (i = 0; i < s->ports; i++) {
1474         AHCIDevice *ad = &s->dev[i];
1475 
1476         ide_bus_new(&ad->port, sizeof(ad->port), qdev, i, 1);
1477         ide_init2(&ad->port, irqs[i]);
1478 
1479         ad->hba = s;
1480         ad->port_no = i;
1481         ad->port.dma = &ad->dma;
1482         ad->port.dma->ops = &ahci_dma_ops;
1483         ide_register_restart_cb(&ad->port);
1484     }
1485     g_free(irqs);
1486 }
1487 
1488 void ahci_uninit(AHCIState *s)
1489 {
1490     int i, j;
1491 
1492     for (i = 0; i < s->ports; i++) {
1493         AHCIDevice *ad = &s->dev[i];
1494 
1495         for (j = 0; j < 2; j++) {
1496             IDEState *s = &ad->port.ifs[j];
1497 
1498             ide_exit(s);
1499         }
1500         object_unparent(OBJECT(&ad->port));
1501     }
1502 
1503     g_free(s->dev);
1504 }
1505 
1506 void ahci_reset(AHCIState *s)
1507 {
1508     AHCIPortRegs *pr;
1509     int i;
1510 
1511     trace_ahci_reset(s);
1512 
1513     s->control_regs.irqstatus = 0;
1514     /* AHCI Enable (AE)
1515      * The implementation of this bit is dependent upon the value of the
1516      * CAP.SAM bit. If CAP.SAM is '0', then GHC.AE shall be read-write and
1517      * shall have a reset value of '0'. If CAP.SAM is '1', then AE shall be
1518      * read-only and shall have a reset value of '1'.
1519      *
1520      * We set HOST_CAP_AHCI so we must enable AHCI at reset.
1521      */
1522     s->control_regs.ghc = HOST_CTL_AHCI_EN;
1523 
1524     for (i = 0; i < s->ports; i++) {
1525         pr = &s->dev[i].port_regs;
1526         pr->irq_stat = 0;
1527         pr->irq_mask = 0;
1528         pr->scr_ctl = 0;
1529         pr->cmd = PORT_CMD_SPIN_UP | PORT_CMD_POWER_ON;
1530         ahci_reset_port(s, i);
1531     }
1532 }
1533 
1534 static const VMStateDescription vmstate_ncq_tfs = {
1535     .name = "ncq state",
1536     .version_id = 1,
1537     .fields = (VMStateField[]) {
1538         VMSTATE_UINT32(sector_count, NCQTransferState),
1539         VMSTATE_UINT64(lba, NCQTransferState),
1540         VMSTATE_UINT8(tag, NCQTransferState),
1541         VMSTATE_UINT8(cmd, NCQTransferState),
1542         VMSTATE_UINT8(slot, NCQTransferState),
1543         VMSTATE_BOOL(used, NCQTransferState),
1544         VMSTATE_BOOL(halt, NCQTransferState),
1545         VMSTATE_END_OF_LIST()
1546     },
1547 };
1548 
1549 static const VMStateDescription vmstate_ahci_device = {
1550     .name = "ahci port",
1551     .version_id = 1,
1552     .fields = (VMStateField[]) {
1553         VMSTATE_IDE_BUS(port, AHCIDevice),
1554         VMSTATE_IDE_DRIVE(port.ifs[0], AHCIDevice),
1555         VMSTATE_UINT32(port_state, AHCIDevice),
1556         VMSTATE_UINT32(finished, AHCIDevice),
1557         VMSTATE_UINT32(port_regs.lst_addr, AHCIDevice),
1558         VMSTATE_UINT32(port_regs.lst_addr_hi, AHCIDevice),
1559         VMSTATE_UINT32(port_regs.fis_addr, AHCIDevice),
1560         VMSTATE_UINT32(port_regs.fis_addr_hi, AHCIDevice),
1561         VMSTATE_UINT32(port_regs.irq_stat, AHCIDevice),
1562         VMSTATE_UINT32(port_regs.irq_mask, AHCIDevice),
1563         VMSTATE_UINT32(port_regs.cmd, AHCIDevice),
1564         VMSTATE_UINT32(port_regs.tfdata, AHCIDevice),
1565         VMSTATE_UINT32(port_regs.sig, AHCIDevice),
1566         VMSTATE_UINT32(port_regs.scr_stat, AHCIDevice),
1567         VMSTATE_UINT32(port_regs.scr_ctl, AHCIDevice),
1568         VMSTATE_UINT32(port_regs.scr_err, AHCIDevice),
1569         VMSTATE_UINT32(port_regs.scr_act, AHCIDevice),
1570         VMSTATE_UINT32(port_regs.cmd_issue, AHCIDevice),
1571         VMSTATE_BOOL(done_atapi_packet, AHCIDevice),
1572         VMSTATE_INT32(busy_slot, AHCIDevice),
1573         VMSTATE_BOOL(init_d2h_sent, AHCIDevice),
1574         VMSTATE_STRUCT_ARRAY(ncq_tfs, AHCIDevice, AHCI_MAX_CMDS,
1575                              1, vmstate_ncq_tfs, NCQTransferState),
1576         VMSTATE_END_OF_LIST()
1577     },
1578 };
1579 
1580 static int ahci_state_post_load(void *opaque, int version_id)
1581 {
1582     int i, j;
1583     struct AHCIDevice *ad;
1584     NCQTransferState *ncq_tfs;
1585     AHCIPortRegs *pr;
1586     AHCIState *s = opaque;
1587 
1588     for (i = 0; i < s->ports; i++) {
1589         ad = &s->dev[i];
1590         pr = &ad->port_regs;
1591 
1592         if (!(pr->cmd & PORT_CMD_START) && (pr->cmd & PORT_CMD_LIST_ON)) {
1593             error_report("AHCI: DMA engine should be off, but status bit "
1594                          "indicates it is still running.");
1595             return -1;
1596         }
1597         if (!(pr->cmd & PORT_CMD_FIS_RX) && (pr->cmd & PORT_CMD_FIS_ON)) {
1598             error_report("AHCI: FIS RX engine should be off, but status bit "
1599                          "indicates it is still running.");
1600             return -1;
1601         }
1602 
1603         /* After a migrate, the DMA/FIS engines are "off" and
1604          * need to be conditionally restarted */
1605         pr->cmd &= ~(PORT_CMD_LIST_ON | PORT_CMD_FIS_ON);
1606         if (ahci_cond_start_engines(ad) != 0) {
1607             return -1;
1608         }
1609 
1610         for (j = 0; j < AHCI_MAX_CMDS; j++) {
1611             ncq_tfs = &ad->ncq_tfs[j];
1612             ncq_tfs->drive = ad;
1613 
1614             if (ncq_tfs->used != ncq_tfs->halt) {
1615                 return -1;
1616             }
1617             if (!ncq_tfs->halt) {
1618                 continue;
1619             }
1620             if (!is_ncq(ncq_tfs->cmd)) {
1621                 return -1;
1622             }
1623             if (ncq_tfs->slot != ncq_tfs->tag) {
1624                 return -1;
1625             }
1626             /* If ncq_tfs->halt is justly set, the engine should be engaged,
1627              * and the command list buffer should be mapped. */
1628             ncq_tfs->cmdh = get_cmd_header(s, i, ncq_tfs->slot);
1629             if (!ncq_tfs->cmdh) {
1630                 return -1;
1631             }
1632             ahci_populate_sglist(ncq_tfs->drive, &ncq_tfs->sglist,
1633                                  ncq_tfs->cmdh, ncq_tfs->sector_count * 512,
1634                                  0);
1635             if (ncq_tfs->sector_count != ncq_tfs->sglist.size >> 9) {
1636                 return -1;
1637             }
1638         }
1639 
1640 
1641         /*
1642          * If an error is present, ad->busy_slot will be valid and not -1.
1643          * In this case, an operation is waiting to resume and will re-check
1644          * for additional AHCI commands to execute upon completion.
1645          *
1646          * In the case where no error was present, busy_slot will be -1,
1647          * and we should check to see if there are additional commands waiting.
1648          */
1649         if (ad->busy_slot == -1) {
1650             check_cmd(s, i);
1651         } else {
1652             /* We are in the middle of a command, and may need to access
1653              * the command header in guest memory again. */
1654             if (ad->busy_slot < 0 || ad->busy_slot >= AHCI_MAX_CMDS) {
1655                 return -1;
1656             }
1657             ad->cur_cmd = get_cmd_header(s, i, ad->busy_slot);
1658         }
1659     }
1660 
1661     return 0;
1662 }
1663 
1664 const VMStateDescription vmstate_ahci = {
1665     .name = "ahci",
1666     .version_id = 1,
1667     .post_load = ahci_state_post_load,
1668     .fields = (VMStateField[]) {
1669         VMSTATE_STRUCT_VARRAY_POINTER_INT32(dev, AHCIState, ports,
1670                                      vmstate_ahci_device, AHCIDevice),
1671         VMSTATE_UINT32(control_regs.cap, AHCIState),
1672         VMSTATE_UINT32(control_regs.ghc, AHCIState),
1673         VMSTATE_UINT32(control_regs.irqstatus, AHCIState),
1674         VMSTATE_UINT32(control_regs.impl, AHCIState),
1675         VMSTATE_UINT32(control_regs.version, AHCIState),
1676         VMSTATE_UINT32(idp_index, AHCIState),
1677         VMSTATE_INT32_EQUAL(ports, AHCIState, NULL),
1678         VMSTATE_END_OF_LIST()
1679     },
1680 };
1681 
1682 static const VMStateDescription vmstate_sysbus_ahci = {
1683     .name = "sysbus-ahci",
1684     .fields = (VMStateField[]) {
1685         VMSTATE_AHCI(ahci, SysbusAHCIState),
1686         VMSTATE_END_OF_LIST()
1687     },
1688 };
1689 
1690 static void sysbus_ahci_reset(DeviceState *dev)
1691 {
1692     SysbusAHCIState *s = SYSBUS_AHCI(dev);
1693 
1694     ahci_reset(&s->ahci);
1695 }
1696 
1697 static void sysbus_ahci_init(Object *obj)
1698 {
1699     SysbusAHCIState *s = SYSBUS_AHCI(obj);
1700     SysBusDevice *sbd = SYS_BUS_DEVICE(obj);
1701 
1702     ahci_init(&s->ahci, DEVICE(obj));
1703 
1704     sysbus_init_mmio(sbd, &s->ahci.mem);
1705     sysbus_init_irq(sbd, &s->ahci.irq);
1706 }
1707 
1708 static void sysbus_ahci_realize(DeviceState *dev, Error **errp)
1709 {
1710     SysbusAHCIState *s = SYSBUS_AHCI(dev);
1711 
1712     ahci_realize(&s->ahci, dev, &address_space_memory, s->num_ports);
1713 }
1714 
1715 static Property sysbus_ahci_properties[] = {
1716     DEFINE_PROP_UINT32("num-ports", SysbusAHCIState, num_ports, 1),
1717     DEFINE_PROP_END_OF_LIST(),
1718 };
1719 
1720 static void sysbus_ahci_class_init(ObjectClass *klass, void *data)
1721 {
1722     DeviceClass *dc = DEVICE_CLASS(klass);
1723 
1724     dc->realize = sysbus_ahci_realize;
1725     dc->vmsd = &vmstate_sysbus_ahci;
1726     dc->props = sysbus_ahci_properties;
1727     dc->reset = sysbus_ahci_reset;
1728     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1729 }
1730 
1731 static const TypeInfo sysbus_ahci_info = {
1732     .name          = TYPE_SYSBUS_AHCI,
1733     .parent        = TYPE_SYS_BUS_DEVICE,
1734     .instance_size = sizeof(SysbusAHCIState),
1735     .instance_init = sysbus_ahci_init,
1736     .class_init    = sysbus_ahci_class_init,
1737 };
1738 
1739 static void sysbus_ahci_register_types(void)
1740 {
1741     type_register_static(&sysbus_ahci_info);
1742 }
1743 
1744 type_init(sysbus_ahci_register_types)
1745 
1746 int32_t ahci_get_num_ports(PCIDevice *dev)
1747 {
1748     AHCIPCIState *d = ICH_AHCI(dev);
1749     AHCIState *ahci = &d->ahci;
1750 
1751     return ahci->ports;
1752 }
1753 
1754 void ahci_ide_create_devs(PCIDevice *dev, DriveInfo **hd)
1755 {
1756     AHCIPCIState *d = ICH_AHCI(dev);
1757     AHCIState *ahci = &d->ahci;
1758     int i;
1759 
1760     for (i = 0; i < ahci->ports; i++) {
1761         if (hd[i] == NULL) {
1762             continue;
1763         }
1764         ide_create_drive(&ahci->dev[i].port, 0, hd[i]);
1765     }
1766 
1767 }
1768