xref: /openbmc/qemu/hw/ide/ahci.c (revision 52f91c37)
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 <hw/hw.h>
25 #include <hw/pci/msi.h>
26 #include <hw/i386/pc.h>
27 #include <hw/pci/pci.h>
28 #include <hw/sysbus.h>
29 
30 #include "monitor/monitor.h"
31 #include "sysemu/dma.h"
32 #include "internal.h"
33 #include <hw/ide/pci.h>
34 #include <hw/ide/ahci.h>
35 
36 /* #define DEBUG_AHCI */
37 
38 #ifdef DEBUG_AHCI
39 #define DPRINTF(port, fmt, ...) \
40 do { fprintf(stderr, "ahci: %s: [%d] ", __FUNCTION__, port); \
41      fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
42 #else
43 #define DPRINTF(port, fmt, ...) do {} while(0)
44 #endif
45 
46 static void check_cmd(AHCIState *s, int port);
47 static int handle_cmd(AHCIState *s,int port,int slot);
48 static void ahci_reset_port(AHCIState *s, int port);
49 static void ahci_write_fis_d2h(AHCIDevice *ad, uint8_t *cmd_fis);
50 static void ahci_init_d2h(AHCIDevice *ad);
51 
52 static uint32_t  ahci_port_read(AHCIState *s, int port, int offset)
53 {
54     uint32_t val;
55     AHCIPortRegs *pr;
56     pr = &s->dev[port].port_regs;
57 
58     switch (offset) {
59     case PORT_LST_ADDR:
60         val = pr->lst_addr;
61         break;
62     case PORT_LST_ADDR_HI:
63         val = pr->lst_addr_hi;
64         break;
65     case PORT_FIS_ADDR:
66         val = pr->fis_addr;
67         break;
68     case PORT_FIS_ADDR_HI:
69         val = pr->fis_addr_hi;
70         break;
71     case PORT_IRQ_STAT:
72         val = pr->irq_stat;
73         break;
74     case PORT_IRQ_MASK:
75         val = pr->irq_mask;
76         break;
77     case PORT_CMD:
78         val = pr->cmd;
79         break;
80     case PORT_TFDATA:
81         val = ((uint16_t)s->dev[port].port.ifs[0].error << 8) |
82               s->dev[port].port.ifs[0].status;
83         break;
84     case PORT_SIG:
85         val = pr->sig;
86         break;
87     case PORT_SCR_STAT:
88         if (s->dev[port].port.ifs[0].bs) {
89             val = SATA_SCR_SSTATUS_DET_DEV_PRESENT_PHY_UP |
90                   SATA_SCR_SSTATUS_SPD_GEN1 | SATA_SCR_SSTATUS_IPM_ACTIVE;
91         } else {
92             val = SATA_SCR_SSTATUS_DET_NODEV;
93         }
94         break;
95     case PORT_SCR_CTL:
96         val = pr->scr_ctl;
97         break;
98     case PORT_SCR_ERR:
99         val = pr->scr_err;
100         break;
101     case PORT_SCR_ACT:
102         pr->scr_act &= ~s->dev[port].finished;
103         s->dev[port].finished = 0;
104         val = pr->scr_act;
105         break;
106     case PORT_CMD_ISSUE:
107         val = pr->cmd_issue;
108         break;
109     case PORT_RESERVED:
110     default:
111         val = 0;
112     }
113     DPRINTF(port, "offset: 0x%x val: 0x%x\n", offset, val);
114     return val;
115 
116 }
117 
118 static void ahci_irq_raise(AHCIState *s, AHCIDevice *dev)
119 {
120     AHCIPCIState *d = container_of(s, AHCIPCIState, ahci);
121     PCIDevice *pci_dev =
122         (PCIDevice *)object_dynamic_cast(OBJECT(d), TYPE_PCI_DEVICE);
123 
124     DPRINTF(0, "raise irq\n");
125 
126     if (pci_dev && msi_enabled(pci_dev)) {
127         msi_notify(pci_dev, 0);
128     } else {
129         qemu_irq_raise(s->irq);
130     }
131 }
132 
133 static void ahci_irq_lower(AHCIState *s, AHCIDevice *dev)
134 {
135     AHCIPCIState *d = container_of(s, AHCIPCIState, ahci);
136     PCIDevice *pci_dev =
137         (PCIDevice *)object_dynamic_cast(OBJECT(d), TYPE_PCI_DEVICE);
138 
139     DPRINTF(0, "lower irq\n");
140 
141     if (!pci_dev || !msi_enabled(pci_dev)) {
142         qemu_irq_lower(s->irq);
143     }
144 }
145 
146 static void ahci_check_irq(AHCIState *s)
147 {
148     int i;
149 
150     DPRINTF(-1, "check irq %#x\n", s->control_regs.irqstatus);
151 
152     s->control_regs.irqstatus = 0;
153     for (i = 0; i < s->ports; i++) {
154         AHCIPortRegs *pr = &s->dev[i].port_regs;
155         if (pr->irq_stat & pr->irq_mask) {
156             s->control_regs.irqstatus |= (1 << i);
157         }
158     }
159 
160     if (s->control_regs.irqstatus &&
161         (s->control_regs.ghc & HOST_CTL_IRQ_EN)) {
162             ahci_irq_raise(s, NULL);
163     } else {
164         ahci_irq_lower(s, NULL);
165     }
166 }
167 
168 static void ahci_trigger_irq(AHCIState *s, AHCIDevice *d,
169                              int irq_type)
170 {
171     DPRINTF(d->port_no, "trigger irq %#x -> %x\n",
172             irq_type, d->port_regs.irq_mask & irq_type);
173 
174     d->port_regs.irq_stat |= irq_type;
175     ahci_check_irq(s);
176 }
177 
178 static void map_page(uint8_t **ptr, uint64_t addr, uint32_t wanted)
179 {
180     hwaddr len = wanted;
181 
182     if (*ptr) {
183         cpu_physical_memory_unmap(*ptr, len, 1, len);
184     }
185 
186     *ptr = cpu_physical_memory_map(addr, &len, 1);
187     if (len < wanted) {
188         cpu_physical_memory_unmap(*ptr, len, 1, len);
189         *ptr = NULL;
190     }
191 }
192 
193 static void  ahci_port_write(AHCIState *s, int port, int offset, uint32_t val)
194 {
195     AHCIPortRegs *pr = &s->dev[port].port_regs;
196 
197     DPRINTF(port, "offset: 0x%x val: 0x%x\n", offset, val);
198     switch (offset) {
199         case PORT_LST_ADDR:
200             pr->lst_addr = val;
201             map_page(&s->dev[port].lst,
202                      ((uint64_t)pr->lst_addr_hi << 32) | pr->lst_addr, 1024);
203             s->dev[port].cur_cmd = NULL;
204             break;
205         case PORT_LST_ADDR_HI:
206             pr->lst_addr_hi = val;
207             map_page(&s->dev[port].lst,
208                      ((uint64_t)pr->lst_addr_hi << 32) | pr->lst_addr, 1024);
209             s->dev[port].cur_cmd = NULL;
210             break;
211         case PORT_FIS_ADDR:
212             pr->fis_addr = val;
213             map_page(&s->dev[port].res_fis,
214                      ((uint64_t)pr->fis_addr_hi << 32) | pr->fis_addr, 256);
215             break;
216         case PORT_FIS_ADDR_HI:
217             pr->fis_addr_hi = val;
218             map_page(&s->dev[port].res_fis,
219                      ((uint64_t)pr->fis_addr_hi << 32) | pr->fis_addr, 256);
220             break;
221         case PORT_IRQ_STAT:
222             pr->irq_stat &= ~val;
223             ahci_check_irq(s);
224             break;
225         case PORT_IRQ_MASK:
226             pr->irq_mask = val & 0xfdc000ff;
227             ahci_check_irq(s);
228             break;
229         case PORT_CMD:
230             pr->cmd = val & ~(PORT_CMD_LIST_ON | PORT_CMD_FIS_ON);
231 
232             if (pr->cmd & PORT_CMD_START) {
233                 pr->cmd |= PORT_CMD_LIST_ON;
234             }
235 
236             if (pr->cmd & PORT_CMD_FIS_RX) {
237                 pr->cmd |= PORT_CMD_FIS_ON;
238             }
239 
240             /* XXX usually the FIS would be pending on the bus here and
241                    issuing deferred until the OS enables FIS receival.
242                    Instead, we only submit it once - which works in most
243                    cases, but is a hack. */
244             if ((pr->cmd & PORT_CMD_FIS_ON) &&
245                 !s->dev[port].init_d2h_sent) {
246                 ahci_init_d2h(&s->dev[port]);
247                 s->dev[port].init_d2h_sent = true;
248             }
249 
250             check_cmd(s, port);
251             break;
252         case PORT_TFDATA:
253             s->dev[port].port.ifs[0].error = (val >> 8) & 0xff;
254             s->dev[port].port.ifs[0].status = val & 0xff;
255             break;
256         case PORT_SIG:
257             pr->sig = val;
258             break;
259         case PORT_SCR_STAT:
260             pr->scr_stat = val;
261             break;
262         case PORT_SCR_CTL:
263             if (((pr->scr_ctl & AHCI_SCR_SCTL_DET) == 1) &&
264                 ((val & AHCI_SCR_SCTL_DET) == 0)) {
265                 ahci_reset_port(s, port);
266             }
267             pr->scr_ctl = val;
268             break;
269         case PORT_SCR_ERR:
270             pr->scr_err &= ~val;
271             break;
272         case PORT_SCR_ACT:
273             /* RW1 */
274             pr->scr_act |= val;
275             break;
276         case PORT_CMD_ISSUE:
277             pr->cmd_issue |= val;
278             check_cmd(s, port);
279             break;
280         default:
281             break;
282     }
283 }
284 
285 static uint64_t ahci_mem_read(void *opaque, hwaddr addr,
286                               unsigned size)
287 {
288     AHCIState *s = opaque;
289     uint32_t val = 0;
290 
291     if (addr < AHCI_GENERIC_HOST_CONTROL_REGS_MAX_ADDR) {
292         switch (addr) {
293         case HOST_CAP:
294             val = s->control_regs.cap;
295             break;
296         case HOST_CTL:
297             val = s->control_regs.ghc;
298             break;
299         case HOST_IRQ_STAT:
300             val = s->control_regs.irqstatus;
301             break;
302         case HOST_PORTS_IMPL:
303             val = s->control_regs.impl;
304             break;
305         case HOST_VERSION:
306             val = s->control_regs.version;
307             break;
308         }
309 
310         DPRINTF(-1, "(addr 0x%08X), val 0x%08X\n", (unsigned) addr, val);
311     } else if ((addr >= AHCI_PORT_REGS_START_ADDR) &&
312                (addr < (AHCI_PORT_REGS_START_ADDR +
313                 (s->ports * AHCI_PORT_ADDR_OFFSET_LEN)))) {
314         val = ahci_port_read(s, (addr - AHCI_PORT_REGS_START_ADDR) >> 7,
315                              addr & AHCI_PORT_ADDR_OFFSET_MASK);
316     }
317 
318     return val;
319 }
320 
321 
322 
323 static void ahci_mem_write(void *opaque, hwaddr addr,
324                            uint64_t val, unsigned size)
325 {
326     AHCIState *s = opaque;
327 
328     /* Only aligned reads are allowed on AHCI */
329     if (addr & 3) {
330         fprintf(stderr, "ahci: Mis-aligned write to addr 0x"
331                 TARGET_FMT_plx "\n", addr);
332         return;
333     }
334 
335     if (addr < AHCI_GENERIC_HOST_CONTROL_REGS_MAX_ADDR) {
336         DPRINTF(-1, "(addr 0x%08X), val 0x%08"PRIX64"\n", (unsigned) addr, val);
337 
338         switch (addr) {
339             case HOST_CAP: /* R/WO, RO */
340                 /* FIXME handle R/WO */
341                 break;
342             case HOST_CTL: /* R/W */
343                 if (val & HOST_CTL_RESET) {
344                     DPRINTF(-1, "HBA Reset\n");
345                     ahci_reset(s);
346                 } else {
347                     s->control_regs.ghc = (val & 0x3) | HOST_CTL_AHCI_EN;
348                     ahci_check_irq(s);
349                 }
350                 break;
351             case HOST_IRQ_STAT: /* R/WC, RO */
352                 s->control_regs.irqstatus &= ~val;
353                 ahci_check_irq(s);
354                 break;
355             case HOST_PORTS_IMPL: /* R/WO, RO */
356                 /* FIXME handle R/WO */
357                 break;
358             case HOST_VERSION: /* RO */
359                 /* FIXME report write? */
360                 break;
361             default:
362                 DPRINTF(-1, "write to unknown register 0x%x\n", (unsigned)addr);
363         }
364     } else if ((addr >= AHCI_PORT_REGS_START_ADDR) &&
365                (addr < (AHCI_PORT_REGS_START_ADDR +
366                 (s->ports * AHCI_PORT_ADDR_OFFSET_LEN)))) {
367         ahci_port_write(s, (addr - AHCI_PORT_REGS_START_ADDR) >> 7,
368                         addr & AHCI_PORT_ADDR_OFFSET_MASK, val);
369     }
370 
371 }
372 
373 static const MemoryRegionOps ahci_mem_ops = {
374     .read = ahci_mem_read,
375     .write = ahci_mem_write,
376     .endianness = DEVICE_LITTLE_ENDIAN,
377 };
378 
379 static uint64_t ahci_idp_read(void *opaque, hwaddr addr,
380                               unsigned size)
381 {
382     AHCIState *s = opaque;
383 
384     if (addr == s->idp_offset) {
385         /* index register */
386         return s->idp_index;
387     } else if (addr == s->idp_offset + 4) {
388         /* data register - do memory read at location selected by index */
389         return ahci_mem_read(opaque, s->idp_index, size);
390     } else {
391         return 0;
392     }
393 }
394 
395 static void ahci_idp_write(void *opaque, hwaddr addr,
396                            uint64_t val, unsigned size)
397 {
398     AHCIState *s = opaque;
399 
400     if (addr == s->idp_offset) {
401         /* index register - mask off reserved bits */
402         s->idp_index = (uint32_t)val & ((AHCI_MEM_BAR_SIZE - 1) & ~3);
403     } else if (addr == s->idp_offset + 4) {
404         /* data register - do memory write at location selected by index */
405         ahci_mem_write(opaque, s->idp_index, val, size);
406     }
407 }
408 
409 static const MemoryRegionOps ahci_idp_ops = {
410     .read = ahci_idp_read,
411     .write = ahci_idp_write,
412     .endianness = DEVICE_LITTLE_ENDIAN,
413 };
414 
415 
416 static void ahci_reg_init(AHCIState *s)
417 {
418     int i;
419 
420     s->control_regs.cap = (s->ports - 1) |
421                           (AHCI_NUM_COMMAND_SLOTS << 8) |
422                           (AHCI_SUPPORTED_SPEED_GEN1 << AHCI_SUPPORTED_SPEED) |
423                           HOST_CAP_NCQ | HOST_CAP_AHCI;
424 
425     s->control_regs.impl = (1 << s->ports) - 1;
426 
427     s->control_regs.version = AHCI_VERSION_1_0;
428 
429     for (i = 0; i < s->ports; i++) {
430         s->dev[i].port_state = STATE_RUN;
431     }
432 }
433 
434 static void check_cmd(AHCIState *s, int port)
435 {
436     AHCIPortRegs *pr = &s->dev[port].port_regs;
437     int slot;
438 
439     if ((pr->cmd & PORT_CMD_START) && pr->cmd_issue) {
440         for (slot = 0; (slot < 32) && pr->cmd_issue; slot++) {
441             if ((pr->cmd_issue & (1U << slot)) &&
442                 !handle_cmd(s, port, slot)) {
443                 pr->cmd_issue &= ~(1U << slot);
444             }
445         }
446     }
447 }
448 
449 static void ahci_check_cmd_bh(void *opaque)
450 {
451     AHCIDevice *ad = opaque;
452 
453     qemu_bh_delete(ad->check_bh);
454     ad->check_bh = NULL;
455 
456     if ((ad->busy_slot != -1) &&
457         !(ad->port.ifs[0].status & (BUSY_STAT|DRQ_STAT))) {
458         /* no longer busy */
459         ad->port_regs.cmd_issue &= ~(1 << ad->busy_slot);
460         ad->busy_slot = -1;
461     }
462 
463     check_cmd(ad->hba, ad->port_no);
464 }
465 
466 static void ahci_init_d2h(AHCIDevice *ad)
467 {
468     uint8_t init_fis[20];
469     IDEState *ide_state = &ad->port.ifs[0];
470 
471     memset(init_fis, 0, sizeof(init_fis));
472 
473     init_fis[4] = 1;
474     init_fis[12] = 1;
475 
476     if (ide_state->drive_kind == IDE_CD) {
477         init_fis[5] = ide_state->lcyl;
478         init_fis[6] = ide_state->hcyl;
479     }
480 
481     ahci_write_fis_d2h(ad, init_fis);
482 }
483 
484 static void ahci_reset_port(AHCIState *s, int port)
485 {
486     AHCIDevice *d = &s->dev[port];
487     AHCIPortRegs *pr = &d->port_regs;
488     IDEState *ide_state = &d->port.ifs[0];
489     int i;
490 
491     DPRINTF(port, "reset port\n");
492 
493     ide_bus_reset(&d->port);
494     ide_state->ncq_queues = AHCI_MAX_CMDS;
495 
496     pr->scr_stat = 0;
497     pr->scr_err = 0;
498     pr->scr_act = 0;
499     d->busy_slot = -1;
500     d->init_d2h_sent = false;
501 
502     ide_state = &s->dev[port].port.ifs[0];
503     if (!ide_state->bs) {
504         return;
505     }
506 
507     /* reset ncq queue */
508     for (i = 0; i < AHCI_MAX_CMDS; i++) {
509         NCQTransferState *ncq_tfs = &s->dev[port].ncq_tfs[i];
510         if (!ncq_tfs->used) {
511             continue;
512         }
513 
514         if (ncq_tfs->aiocb) {
515             bdrv_aio_cancel(ncq_tfs->aiocb);
516             ncq_tfs->aiocb = NULL;
517         }
518 
519         /* Maybe we just finished the request thanks to bdrv_aio_cancel() */
520         if (!ncq_tfs->used) {
521             continue;
522         }
523 
524         qemu_sglist_destroy(&ncq_tfs->sglist);
525         ncq_tfs->used = 0;
526     }
527 
528     s->dev[port].port_state = STATE_RUN;
529     if (!ide_state->bs) {
530         s->dev[port].port_regs.sig = 0;
531         ide_state->status = SEEK_STAT | WRERR_STAT;
532     } else if (ide_state->drive_kind == IDE_CD) {
533         s->dev[port].port_regs.sig = SATA_SIGNATURE_CDROM;
534         ide_state->lcyl = 0x14;
535         ide_state->hcyl = 0xeb;
536         DPRINTF(port, "set lcyl = %d\n", ide_state->lcyl);
537         ide_state->status = SEEK_STAT | WRERR_STAT | READY_STAT;
538     } else {
539         s->dev[port].port_regs.sig = SATA_SIGNATURE_DISK;
540         ide_state->status = SEEK_STAT | WRERR_STAT;
541     }
542 
543     ide_state->error = 1;
544     ahci_init_d2h(d);
545 }
546 
547 static void debug_print_fis(uint8_t *fis, int cmd_len)
548 {
549 #ifdef DEBUG_AHCI
550     int i;
551 
552     fprintf(stderr, "fis:");
553     for (i = 0; i < cmd_len; i++) {
554         if ((i & 0xf) == 0) {
555             fprintf(stderr, "\n%02x:",i);
556         }
557         fprintf(stderr, "%02x ",fis[i]);
558     }
559     fprintf(stderr, "\n");
560 #endif
561 }
562 
563 static void ahci_write_fis_sdb(AHCIState *s, int port, uint32_t finished)
564 {
565     AHCIPortRegs *pr = &s->dev[port].port_regs;
566     IDEState *ide_state;
567     uint8_t *sdb_fis;
568 
569     if (!s->dev[port].res_fis ||
570         !(pr->cmd & PORT_CMD_FIS_RX)) {
571         return;
572     }
573 
574     sdb_fis = &s->dev[port].res_fis[RES_FIS_SDBFIS];
575     ide_state = &s->dev[port].port.ifs[0];
576 
577     /* clear memory */
578     *(uint32_t*)sdb_fis = 0;
579 
580     /* write values */
581     sdb_fis[0] = ide_state->error;
582     sdb_fis[2] = ide_state->status & 0x77;
583     s->dev[port].finished |= finished;
584     *(uint32_t*)(sdb_fis + 4) = cpu_to_le32(s->dev[port].finished);
585 
586     ahci_trigger_irq(s, &s->dev[port], PORT_IRQ_STAT_SDBS);
587 }
588 
589 static void ahci_write_fis_d2h(AHCIDevice *ad, uint8_t *cmd_fis)
590 {
591     AHCIPortRegs *pr = &ad->port_regs;
592     uint8_t *d2h_fis;
593     int i;
594     dma_addr_t cmd_len = 0x80;
595     int cmd_mapped = 0;
596 
597     if (!ad->res_fis || !(pr->cmd & PORT_CMD_FIS_RX)) {
598         return;
599     }
600 
601     if (!cmd_fis) {
602         /* map cmd_fis */
603         uint64_t tbl_addr = le64_to_cpu(ad->cur_cmd->tbl_addr);
604         cmd_fis = dma_memory_map(ad->hba->as, tbl_addr, &cmd_len,
605                                  DMA_DIRECTION_TO_DEVICE);
606         cmd_mapped = 1;
607     }
608 
609     d2h_fis = &ad->res_fis[RES_FIS_RFIS];
610 
611     d2h_fis[0] = 0x34;
612     d2h_fis[1] = (ad->hba->control_regs.irqstatus ? (1 << 6) : 0);
613     d2h_fis[2] = ad->port.ifs[0].status;
614     d2h_fis[3] = ad->port.ifs[0].error;
615 
616     d2h_fis[4] = cmd_fis[4];
617     d2h_fis[5] = cmd_fis[5];
618     d2h_fis[6] = cmd_fis[6];
619     d2h_fis[7] = cmd_fis[7];
620     d2h_fis[8] = cmd_fis[8];
621     d2h_fis[9] = cmd_fis[9];
622     d2h_fis[10] = cmd_fis[10];
623     d2h_fis[11] = cmd_fis[11];
624     d2h_fis[12] = cmd_fis[12];
625     d2h_fis[13] = cmd_fis[13];
626     for (i = 14; i < 20; i++) {
627         d2h_fis[i] = 0;
628     }
629 
630     if (d2h_fis[2] & ERR_STAT) {
631         ahci_trigger_irq(ad->hba, ad, PORT_IRQ_STAT_TFES);
632     }
633 
634     ahci_trigger_irq(ad->hba, ad, PORT_IRQ_D2H_REG_FIS);
635 
636     if (cmd_mapped) {
637         dma_memory_unmap(ad->hba->as, cmd_fis, cmd_len,
638                          DMA_DIRECTION_TO_DEVICE, cmd_len);
639     }
640 }
641 
642 static int ahci_populate_sglist(AHCIDevice *ad, QEMUSGList *sglist, int offset)
643 {
644     AHCICmdHdr *cmd = ad->cur_cmd;
645     uint32_t opts = le32_to_cpu(cmd->opts);
646     uint64_t prdt_addr = le64_to_cpu(cmd->tbl_addr) + 0x80;
647     int sglist_alloc_hint = opts >> AHCI_CMD_HDR_PRDT_LEN;
648     dma_addr_t prdt_len = (sglist_alloc_hint * sizeof(AHCI_SG));
649     dma_addr_t real_prdt_len = prdt_len;
650     uint8_t *prdt;
651     int i;
652     int r = 0;
653     int sum = 0;
654     int off_idx = -1;
655     int off_pos = -1;
656     int tbl_entry_size;
657     IDEBus *bus = &ad->port;
658     BusState *qbus = BUS(bus);
659 
660     if (!sglist_alloc_hint) {
661         DPRINTF(ad->port_no, "no sg list given by guest: 0x%08x\n", opts);
662         return -1;
663     }
664 
665     /* map PRDT */
666     if (!(prdt = dma_memory_map(ad->hba->as, prdt_addr, &prdt_len,
667                                 DMA_DIRECTION_TO_DEVICE))){
668         DPRINTF(ad->port_no, "map failed\n");
669         return -1;
670     }
671 
672     if (prdt_len < real_prdt_len) {
673         DPRINTF(ad->port_no, "mapped less than expected\n");
674         r = -1;
675         goto out;
676     }
677 
678     /* Get entries in the PRDT, init a qemu sglist accordingly */
679     if (sglist_alloc_hint > 0) {
680         AHCI_SG *tbl = (AHCI_SG *)prdt;
681         sum = 0;
682         for (i = 0; i < sglist_alloc_hint; i++) {
683             /* flags_size is zero-based */
684             tbl_entry_size = (le32_to_cpu(tbl[i].flags_size) + 1);
685             if (offset <= (sum + tbl_entry_size)) {
686                 off_idx = i;
687                 off_pos = offset - sum;
688                 break;
689             }
690             sum += tbl_entry_size;
691         }
692         if ((off_idx == -1) || (off_pos < 0) || (off_pos > tbl_entry_size)) {
693             DPRINTF(ad->port_no, "%s: Incorrect offset! "
694                             "off_idx: %d, off_pos: %d\n",
695                             __func__, off_idx, off_pos);
696             r = -1;
697             goto out;
698         }
699 
700         qemu_sglist_init(sglist, qbus->parent, (sglist_alloc_hint - off_idx),
701                          ad->hba->as);
702         qemu_sglist_add(sglist, le64_to_cpu(tbl[off_idx].addr + off_pos),
703                         le32_to_cpu(tbl[off_idx].flags_size) + 1 - off_pos);
704 
705         for (i = off_idx + 1; i < sglist_alloc_hint; i++) {
706             /* flags_size is zero-based */
707             qemu_sglist_add(sglist, le64_to_cpu(tbl[i].addr),
708                             le32_to_cpu(tbl[i].flags_size) + 1);
709         }
710     }
711 
712 out:
713     dma_memory_unmap(ad->hba->as, prdt, prdt_len,
714                      DMA_DIRECTION_TO_DEVICE, prdt_len);
715     return r;
716 }
717 
718 static void ncq_cb(void *opaque, int ret)
719 {
720     NCQTransferState *ncq_tfs = (NCQTransferState *)opaque;
721     IDEState *ide_state = &ncq_tfs->drive->port.ifs[0];
722 
723     /* Clear bit for this tag in SActive */
724     ncq_tfs->drive->port_regs.scr_act &= ~(1 << ncq_tfs->tag);
725 
726     if (ret < 0) {
727         /* error */
728         ide_state->error = ABRT_ERR;
729         ide_state->status = READY_STAT | ERR_STAT;
730         ncq_tfs->drive->port_regs.scr_err |= (1 << ncq_tfs->tag);
731     } else {
732         ide_state->status = READY_STAT | SEEK_STAT;
733     }
734 
735     ahci_write_fis_sdb(ncq_tfs->drive->hba, ncq_tfs->drive->port_no,
736                        (1 << ncq_tfs->tag));
737 
738     DPRINTF(ncq_tfs->drive->port_no, "NCQ transfer tag %d finished\n",
739             ncq_tfs->tag);
740 
741     bdrv_acct_done(ncq_tfs->drive->port.ifs[0].bs, &ncq_tfs->acct);
742     qemu_sglist_destroy(&ncq_tfs->sglist);
743     ncq_tfs->used = 0;
744 }
745 
746 static void process_ncq_command(AHCIState *s, int port, uint8_t *cmd_fis,
747                                 int slot)
748 {
749     NCQFrame *ncq_fis = (NCQFrame*)cmd_fis;
750     uint8_t tag = ncq_fis->tag >> 3;
751     NCQTransferState *ncq_tfs = &s->dev[port].ncq_tfs[tag];
752 
753     if (ncq_tfs->used) {
754         /* error - already in use */
755         fprintf(stderr, "%s: tag %d already used\n", __FUNCTION__, tag);
756         return;
757     }
758 
759     ncq_tfs->used = 1;
760     ncq_tfs->drive = &s->dev[port];
761     ncq_tfs->slot = slot;
762     ncq_tfs->lba = ((uint64_t)ncq_fis->lba5 << 40) |
763                    ((uint64_t)ncq_fis->lba4 << 32) |
764                    ((uint64_t)ncq_fis->lba3 << 24) |
765                    ((uint64_t)ncq_fis->lba2 << 16) |
766                    ((uint64_t)ncq_fis->lba1 << 8) |
767                    (uint64_t)ncq_fis->lba0;
768 
769     /* Note: We calculate the sector count, but don't currently rely on it.
770      * The total size of the DMA buffer tells us the transfer size instead. */
771     ncq_tfs->sector_count = ((uint16_t)ncq_fis->sector_count_high << 8) |
772                                 ncq_fis->sector_count_low;
773 
774     DPRINTF(port, "NCQ transfer LBA from %"PRId64" to %"PRId64", "
775             "drive max %"PRId64"\n",
776             ncq_tfs->lba, ncq_tfs->lba + ncq_tfs->sector_count - 2,
777             s->dev[port].port.ifs[0].nb_sectors - 1);
778 
779     ahci_populate_sglist(&s->dev[port], &ncq_tfs->sglist, 0);
780     ncq_tfs->tag = tag;
781 
782     switch(ncq_fis->command) {
783         case READ_FPDMA_QUEUED:
784             DPRINTF(port, "NCQ reading %d sectors from LBA %"PRId64", "
785                     "tag %d\n",
786                     ncq_tfs->sector_count-1, ncq_tfs->lba, ncq_tfs->tag);
787 
788             DPRINTF(port, "tag %d aio read %"PRId64"\n",
789                     ncq_tfs->tag, ncq_tfs->lba);
790 
791             dma_acct_start(ncq_tfs->drive->port.ifs[0].bs, &ncq_tfs->acct,
792                            &ncq_tfs->sglist, BDRV_ACCT_READ);
793             ncq_tfs->aiocb = dma_bdrv_read(ncq_tfs->drive->port.ifs[0].bs,
794                                            &ncq_tfs->sglist, ncq_tfs->lba,
795                                            ncq_cb, ncq_tfs);
796             break;
797         case WRITE_FPDMA_QUEUED:
798             DPRINTF(port, "NCQ writing %d sectors to LBA %"PRId64", tag %d\n",
799                     ncq_tfs->sector_count-1, ncq_tfs->lba, ncq_tfs->tag);
800 
801             DPRINTF(port, "tag %d aio write %"PRId64"\n",
802                     ncq_tfs->tag, ncq_tfs->lba);
803 
804             dma_acct_start(ncq_tfs->drive->port.ifs[0].bs, &ncq_tfs->acct,
805                            &ncq_tfs->sglist, BDRV_ACCT_WRITE);
806             ncq_tfs->aiocb = dma_bdrv_write(ncq_tfs->drive->port.ifs[0].bs,
807                                             &ncq_tfs->sglist, ncq_tfs->lba,
808                                             ncq_cb, ncq_tfs);
809             break;
810         default:
811             DPRINTF(port, "error: tried to process non-NCQ command as NCQ\n");
812             qemu_sglist_destroy(&ncq_tfs->sglist);
813             break;
814     }
815 }
816 
817 static int handle_cmd(AHCIState *s, int port, int slot)
818 {
819     IDEState *ide_state;
820     uint32_t opts;
821     uint64_t tbl_addr;
822     AHCICmdHdr *cmd;
823     uint8_t *cmd_fis;
824     dma_addr_t cmd_len;
825 
826     if (s->dev[port].port.ifs[0].status & (BUSY_STAT|DRQ_STAT)) {
827         /* Engine currently busy, try again later */
828         DPRINTF(port, "engine busy\n");
829         return -1;
830     }
831 
832     cmd = &((AHCICmdHdr *)s->dev[port].lst)[slot];
833 
834     if (!s->dev[port].lst) {
835         DPRINTF(port, "error: lst not given but cmd handled");
836         return -1;
837     }
838 
839     /* remember current slot handle for later */
840     s->dev[port].cur_cmd = cmd;
841 
842     opts = le32_to_cpu(cmd->opts);
843     tbl_addr = le64_to_cpu(cmd->tbl_addr);
844 
845     cmd_len = 0x80;
846     cmd_fis = dma_memory_map(s->as, tbl_addr, &cmd_len,
847                              DMA_DIRECTION_FROM_DEVICE);
848 
849     if (!cmd_fis) {
850         DPRINTF(port, "error: guest passed us an invalid cmd fis\n");
851         return -1;
852     }
853 
854     /* The device we are working for */
855     ide_state = &s->dev[port].port.ifs[0];
856 
857     if (!ide_state->bs) {
858         DPRINTF(port, "error: guest accessed unused port");
859         goto out;
860     }
861 
862     debug_print_fis(cmd_fis, 0x90);
863     //debug_print_fis(cmd_fis, (opts & AHCI_CMD_HDR_CMD_FIS_LEN) * 4);
864 
865     switch (cmd_fis[0]) {
866         case SATA_FIS_TYPE_REGISTER_H2D:
867             break;
868         default:
869             DPRINTF(port, "unknown command cmd_fis[0]=%02x cmd_fis[1]=%02x "
870                           "cmd_fis[2]=%02x\n", cmd_fis[0], cmd_fis[1],
871                           cmd_fis[2]);
872             goto out;
873             break;
874     }
875 
876     switch (cmd_fis[1]) {
877         case SATA_FIS_REG_H2D_UPDATE_COMMAND_REGISTER:
878             break;
879         case 0:
880             break;
881         default:
882             DPRINTF(port, "unknown command cmd_fis[0]=%02x cmd_fis[1]=%02x "
883                           "cmd_fis[2]=%02x\n", cmd_fis[0], cmd_fis[1],
884                           cmd_fis[2]);
885             goto out;
886             break;
887     }
888 
889     switch (s->dev[port].port_state) {
890         case STATE_RUN:
891             if (cmd_fis[15] & ATA_SRST) {
892                 s->dev[port].port_state = STATE_RESET;
893             }
894             break;
895         case STATE_RESET:
896             if (!(cmd_fis[15] & ATA_SRST)) {
897                 ahci_reset_port(s, port);
898             }
899             break;
900     }
901 
902     if (cmd_fis[1] == SATA_FIS_REG_H2D_UPDATE_COMMAND_REGISTER) {
903 
904         /* Check for NCQ command */
905         if ((cmd_fis[2] == READ_FPDMA_QUEUED) ||
906             (cmd_fis[2] == WRITE_FPDMA_QUEUED)) {
907             process_ncq_command(s, port, cmd_fis, slot);
908             goto out;
909         }
910 
911         /* Decompose the FIS  */
912         ide_state->nsector = (int64_t)((cmd_fis[13] << 8) | cmd_fis[12]);
913         ide_state->feature = cmd_fis[3];
914         if (!ide_state->nsector) {
915             ide_state->nsector = 256;
916         }
917 
918         if (ide_state->drive_kind != IDE_CD) {
919             /*
920              * We set the sector depending on the sector defined in the FIS.
921              * Unfortunately, the spec isn't exactly obvious on this one.
922              *
923              * Apparently LBA48 commands set fis bytes 10,9,8,6,5,4 to the
924              * 48 bit sector number. ATA_CMD_READ_DMA_EXT is an example for
925              * such a command.
926              *
927              * Non-LBA48 commands however use 7[lower 4 bits],6,5,4 to define a
928              * 28-bit sector number. ATA_CMD_READ_DMA is an example for such
929              * a command.
930              *
931              * Since the spec doesn't explicitly state what each field should
932              * do, I simply assume non-used fields as reserved and OR everything
933              * together, independent of the command.
934              */
935             ide_set_sector(ide_state, ((uint64_t)cmd_fis[10] << 40)
936                                     | ((uint64_t)cmd_fis[9] << 32)
937                                     /* This is used for LBA48 commands */
938                                     | ((uint64_t)cmd_fis[8] << 24)
939                                     /* This is used for non-LBA48 commands */
940                                     | ((uint64_t)(cmd_fis[7] & 0xf) << 24)
941                                     | ((uint64_t)cmd_fis[6] << 16)
942                                     | ((uint64_t)cmd_fis[5] << 8)
943                                     | cmd_fis[4]);
944         }
945 
946         /* Copy the ACMD field (ATAPI packet, if any) from the AHCI command
947          * table to ide_state->io_buffer
948          */
949         if (opts & AHCI_CMD_ATAPI) {
950             memcpy(ide_state->io_buffer, &cmd_fis[AHCI_COMMAND_TABLE_ACMD], 0x10);
951             ide_state->lcyl = 0x14;
952             ide_state->hcyl = 0xeb;
953             debug_print_fis(ide_state->io_buffer, 0x10);
954             ide_state->feature = IDE_FEATURE_DMA;
955             s->dev[port].done_atapi_packet = false;
956             /* XXX send PIO setup FIS */
957         }
958 
959         ide_state->error = 0;
960 
961         /* Reset transferred byte counter */
962         cmd->status = 0;
963 
964         /* We're ready to process the command in FIS byte 2. */
965         ide_exec_cmd(&s->dev[port].port, cmd_fis[2]);
966 
967         if ((s->dev[port].port.ifs[0].status & (READY_STAT|DRQ_STAT|BUSY_STAT)) ==
968             READY_STAT) {
969             ahci_write_fis_d2h(&s->dev[port], cmd_fis);
970         }
971     }
972 
973 out:
974     dma_memory_unmap(s->as, cmd_fis, cmd_len, DMA_DIRECTION_FROM_DEVICE,
975                      cmd_len);
976 
977     if (s->dev[port].port.ifs[0].status & (BUSY_STAT|DRQ_STAT)) {
978         /* async command, complete later */
979         s->dev[port].busy_slot = slot;
980         return -1;
981     }
982 
983     /* done handling the command */
984     return 0;
985 }
986 
987 /* DMA dev <-> ram */
988 static int ahci_start_transfer(IDEDMA *dma)
989 {
990     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
991     IDEState *s = &ad->port.ifs[0];
992     uint32_t size = (uint32_t)(s->data_end - s->data_ptr);
993     /* write == ram -> device */
994     uint32_t opts = le32_to_cpu(ad->cur_cmd->opts);
995     int is_write = opts & AHCI_CMD_WRITE;
996     int is_atapi = opts & AHCI_CMD_ATAPI;
997     int has_sglist = 0;
998 
999     if (is_atapi && !ad->done_atapi_packet) {
1000         /* already prepopulated iobuffer */
1001         ad->done_atapi_packet = true;
1002         goto out;
1003     }
1004 
1005     if (!ahci_populate_sglist(ad, &s->sg, 0)) {
1006         has_sglist = 1;
1007     }
1008 
1009     DPRINTF(ad->port_no, "%sing %d bytes on %s w/%s sglist\n",
1010             is_write ? "writ" : "read", size, is_atapi ? "atapi" : "ata",
1011             has_sglist ? "" : "o");
1012 
1013     if (has_sglist && size) {
1014         if (is_write) {
1015             dma_buf_write(s->data_ptr, size, &s->sg);
1016         } else {
1017             dma_buf_read(s->data_ptr, size, &s->sg);
1018         }
1019     }
1020 
1021     /* update number of transferred bytes */
1022     ad->cur_cmd->status = cpu_to_le32(le32_to_cpu(ad->cur_cmd->status) + size);
1023 
1024 out:
1025     /* declare that we processed everything */
1026     s->data_ptr = s->data_end;
1027 
1028     if (has_sglist) {
1029         qemu_sglist_destroy(&s->sg);
1030     }
1031 
1032     s->end_transfer_func(s);
1033 
1034     if (!(s->status & DRQ_STAT)) {
1035         /* done with DMA */
1036         ahci_trigger_irq(ad->hba, ad, PORT_IRQ_STAT_DSS);
1037     }
1038 
1039     return 0;
1040 }
1041 
1042 static void ahci_start_dma(IDEDMA *dma, IDEState *s,
1043                            BlockDriverCompletionFunc *dma_cb)
1044 {
1045 #ifdef DEBUG_AHCI
1046     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1047 #endif
1048     DPRINTF(ad->port_no, "\n");
1049     s->io_buffer_offset = 0;
1050     dma_cb(s, 0);
1051 }
1052 
1053 static int ahci_dma_prepare_buf(IDEDMA *dma, int is_write)
1054 {
1055     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1056     IDEState *s = &ad->port.ifs[0];
1057 
1058     ahci_populate_sglist(ad, &s->sg, 0);
1059     s->io_buffer_size = s->sg.size;
1060 
1061     DPRINTF(ad->port_no, "len=%#x\n", s->io_buffer_size);
1062     return s->io_buffer_size != 0;
1063 }
1064 
1065 static int ahci_dma_rw_buf(IDEDMA *dma, int is_write)
1066 {
1067     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1068     IDEState *s = &ad->port.ifs[0];
1069     uint8_t *p = s->io_buffer + s->io_buffer_index;
1070     int l = s->io_buffer_size - s->io_buffer_index;
1071 
1072     if (ahci_populate_sglist(ad, &s->sg, s->io_buffer_offset)) {
1073         return 0;
1074     }
1075 
1076     if (is_write) {
1077         dma_buf_read(p, l, &s->sg);
1078     } else {
1079         dma_buf_write(p, l, &s->sg);
1080     }
1081 
1082     /* free sglist that was created in ahci_populate_sglist() */
1083     qemu_sglist_destroy(&s->sg);
1084 
1085     /* update number of transferred bytes */
1086     ad->cur_cmd->status = cpu_to_le32(le32_to_cpu(ad->cur_cmd->status) + l);
1087     s->io_buffer_index += l;
1088     s->io_buffer_offset += l;
1089 
1090     DPRINTF(ad->port_no, "len=%#x\n", l);
1091 
1092     return 1;
1093 }
1094 
1095 static int ahci_dma_set_unit(IDEDMA *dma, int unit)
1096 {
1097     /* only a single unit per link */
1098     return 0;
1099 }
1100 
1101 static int ahci_dma_add_status(IDEDMA *dma, int status)
1102 {
1103     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1104     DPRINTF(ad->port_no, "set status: %x\n", status);
1105 
1106     if (status & BM_STATUS_INT) {
1107         ahci_trigger_irq(ad->hba, ad, PORT_IRQ_STAT_DSS);
1108     }
1109 
1110     return 0;
1111 }
1112 
1113 static int ahci_dma_set_inactive(IDEDMA *dma)
1114 {
1115     return 0;
1116 }
1117 
1118 static int ahci_async_cmd_done(IDEDMA *dma)
1119 {
1120     AHCIDevice *ad = DO_UPCAST(AHCIDevice, dma, dma);
1121 
1122     DPRINTF(ad->port_no, "async cmd done\n");
1123 
1124     /* update d2h status */
1125     ahci_write_fis_d2h(ad, NULL);
1126 
1127     if (!ad->check_bh) {
1128         /* maybe we still have something to process, check later */
1129         ad->check_bh = qemu_bh_new(ahci_check_cmd_bh, ad);
1130         qemu_bh_schedule(ad->check_bh);
1131     }
1132 
1133     return 0;
1134 }
1135 
1136 static void ahci_irq_set(void *opaque, int n, int level)
1137 {
1138 }
1139 
1140 static void ahci_dma_restart_cb(void *opaque, int running, RunState state)
1141 {
1142 }
1143 
1144 static int ahci_dma_reset(IDEDMA *dma)
1145 {
1146     return 0;
1147 }
1148 
1149 static const IDEDMAOps ahci_dma_ops = {
1150     .start_dma = ahci_start_dma,
1151     .start_transfer = ahci_start_transfer,
1152     .prepare_buf = ahci_dma_prepare_buf,
1153     .rw_buf = ahci_dma_rw_buf,
1154     .set_unit = ahci_dma_set_unit,
1155     .add_status = ahci_dma_add_status,
1156     .set_inactive = ahci_dma_set_inactive,
1157     .async_cmd_done = ahci_async_cmd_done,
1158     .restart_cb = ahci_dma_restart_cb,
1159     .reset = ahci_dma_reset,
1160 };
1161 
1162 void ahci_init(AHCIState *s, DeviceState *qdev, AddressSpace *as, int ports)
1163 {
1164     qemu_irq *irqs;
1165     int i;
1166 
1167     s->as = as;
1168     s->ports = ports;
1169     s->dev = g_malloc0(sizeof(AHCIDevice) * ports);
1170     ahci_reg_init(s);
1171     /* XXX BAR size should be 1k, but that breaks, so bump it to 4k for now */
1172     memory_region_init_io(&s->mem, OBJECT(qdev), &ahci_mem_ops, s,
1173                           "ahci", AHCI_MEM_BAR_SIZE);
1174     memory_region_init_io(&s->idp, OBJECT(qdev), &ahci_idp_ops, s,
1175                           "ahci-idp", 32);
1176 
1177     irqs = qemu_allocate_irqs(ahci_irq_set, s, s->ports);
1178 
1179     for (i = 0; i < s->ports; i++) {
1180         AHCIDevice *ad = &s->dev[i];
1181 
1182         ide_bus_new(&ad->port, sizeof(ad->port), qdev, i, 1);
1183         ide_init2(&ad->port, irqs[i]);
1184 
1185         ad->hba = s;
1186         ad->port_no = i;
1187         ad->port.dma = &ad->dma;
1188         ad->port.dma->ops = &ahci_dma_ops;
1189     }
1190 }
1191 
1192 void ahci_uninit(AHCIState *s)
1193 {
1194     memory_region_destroy(&s->mem);
1195     memory_region_destroy(&s->idp);
1196     g_free(s->dev);
1197 }
1198 
1199 void ahci_reset(AHCIState *s)
1200 {
1201     AHCIPortRegs *pr;
1202     int i;
1203 
1204     s->control_regs.irqstatus = 0;
1205     /* AHCI Enable (AE)
1206      * The implementation of this bit is dependent upon the value of the
1207      * CAP.SAM bit. If CAP.SAM is '0', then GHC.AE shall be read-write and
1208      * shall have a reset value of '0'. If CAP.SAM is '1', then AE shall be
1209      * read-only and shall have a reset value of '1'.
1210      *
1211      * We set HOST_CAP_AHCI so we must enable AHCI at reset.
1212      */
1213     s->control_regs.ghc = HOST_CTL_AHCI_EN;
1214 
1215     for (i = 0; i < s->ports; i++) {
1216         pr = &s->dev[i].port_regs;
1217         pr->irq_stat = 0;
1218         pr->irq_mask = 0;
1219         pr->scr_ctl = 0;
1220         pr->cmd = PORT_CMD_SPIN_UP | PORT_CMD_POWER_ON;
1221         ahci_reset_port(s, i);
1222     }
1223 }
1224 
1225 static const VMStateDescription vmstate_ahci_device = {
1226     .name = "ahci port",
1227     .version_id = 1,
1228     .fields = (VMStateField []) {
1229         VMSTATE_IDE_BUS(port, AHCIDevice),
1230         VMSTATE_UINT32(port_state, AHCIDevice),
1231         VMSTATE_UINT32(finished, AHCIDevice),
1232         VMSTATE_UINT32(port_regs.lst_addr, AHCIDevice),
1233         VMSTATE_UINT32(port_regs.lst_addr_hi, AHCIDevice),
1234         VMSTATE_UINT32(port_regs.fis_addr, AHCIDevice),
1235         VMSTATE_UINT32(port_regs.fis_addr_hi, AHCIDevice),
1236         VMSTATE_UINT32(port_regs.irq_stat, AHCIDevice),
1237         VMSTATE_UINT32(port_regs.irq_mask, AHCIDevice),
1238         VMSTATE_UINT32(port_regs.cmd, AHCIDevice),
1239         VMSTATE_UINT32(port_regs.tfdata, AHCIDevice),
1240         VMSTATE_UINT32(port_regs.sig, AHCIDevice),
1241         VMSTATE_UINT32(port_regs.scr_stat, AHCIDevice),
1242         VMSTATE_UINT32(port_regs.scr_ctl, AHCIDevice),
1243         VMSTATE_UINT32(port_regs.scr_err, AHCIDevice),
1244         VMSTATE_UINT32(port_regs.scr_act, AHCIDevice),
1245         VMSTATE_UINT32(port_regs.cmd_issue, AHCIDevice),
1246         VMSTATE_BOOL(done_atapi_packet, AHCIDevice),
1247         VMSTATE_INT32(busy_slot, AHCIDevice),
1248         VMSTATE_BOOL(init_d2h_sent, AHCIDevice),
1249         VMSTATE_END_OF_LIST()
1250     },
1251 };
1252 
1253 static int ahci_state_post_load(void *opaque, int version_id)
1254 {
1255     int i;
1256     struct AHCIDevice *ad;
1257     AHCIState *s = opaque;
1258 
1259     for (i = 0; i < s->ports; i++) {
1260         ad = &s->dev[i];
1261         AHCIPortRegs *pr = &ad->port_regs;
1262 
1263         map_page(&ad->lst,
1264                  ((uint64_t)pr->lst_addr_hi << 32) | pr->lst_addr, 1024);
1265         map_page(&ad->res_fis,
1266                  ((uint64_t)pr->fis_addr_hi << 32) | pr->fis_addr, 256);
1267         /*
1268          * All pending i/o should be flushed out on a migrate. However,
1269          * we might not have cleared the busy_slot since this is done
1270          * in a bh. Also, issue i/o against any slots that are pending.
1271          */
1272         if ((ad->busy_slot != -1) &&
1273             !(ad->port.ifs[0].status & (BUSY_STAT|DRQ_STAT))) {
1274             pr->cmd_issue &= ~(1 << ad->busy_slot);
1275             ad->busy_slot = -1;
1276         }
1277         check_cmd(s, i);
1278     }
1279 
1280     return 0;
1281 }
1282 
1283 const VMStateDescription vmstate_ahci = {
1284     .name = "ahci",
1285     .version_id = 1,
1286     .post_load = ahci_state_post_load,
1287     .fields = (VMStateField []) {
1288         VMSTATE_STRUCT_VARRAY_POINTER_INT32(dev, AHCIState, ports,
1289                                      vmstate_ahci_device, AHCIDevice),
1290         VMSTATE_UINT32(control_regs.cap, AHCIState),
1291         VMSTATE_UINT32(control_regs.ghc, AHCIState),
1292         VMSTATE_UINT32(control_regs.irqstatus, AHCIState),
1293         VMSTATE_UINT32(control_regs.impl, AHCIState),
1294         VMSTATE_UINT32(control_regs.version, AHCIState),
1295         VMSTATE_UINT32(idp_index, AHCIState),
1296         VMSTATE_INT32_EQUAL(ports, AHCIState),
1297         VMSTATE_END_OF_LIST()
1298     },
1299 };
1300 
1301 #define TYPE_SYSBUS_AHCI "sysbus-ahci"
1302 #define SYSBUS_AHCI(obj) OBJECT_CHECK(SysbusAHCIState, (obj), TYPE_SYSBUS_AHCI)
1303 
1304 typedef struct SysbusAHCIState {
1305     /*< private >*/
1306     SysBusDevice parent_obj;
1307     /*< public >*/
1308 
1309     AHCIState ahci;
1310     uint32_t num_ports;
1311 } SysbusAHCIState;
1312 
1313 static const VMStateDescription vmstate_sysbus_ahci = {
1314     .name = "sysbus-ahci",
1315     .unmigratable = 1, /* Still buggy under I/O load */
1316     .fields = (VMStateField []) {
1317         VMSTATE_AHCI(ahci, SysbusAHCIState),
1318         VMSTATE_END_OF_LIST()
1319     },
1320 };
1321 
1322 static void sysbus_ahci_reset(DeviceState *dev)
1323 {
1324     SysbusAHCIState *s = SYSBUS_AHCI(dev);
1325 
1326     ahci_reset(&s->ahci);
1327 }
1328 
1329 static void sysbus_ahci_realize(DeviceState *dev, Error **errp)
1330 {
1331     SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
1332     SysbusAHCIState *s = SYSBUS_AHCI(dev);
1333 
1334     ahci_init(&s->ahci, dev, &address_space_memory, s->num_ports);
1335 
1336     sysbus_init_mmio(sbd, &s->ahci.mem);
1337     sysbus_init_irq(sbd, &s->ahci.irq);
1338 }
1339 
1340 static Property sysbus_ahci_properties[] = {
1341     DEFINE_PROP_UINT32("num-ports", SysbusAHCIState, num_ports, 1),
1342     DEFINE_PROP_END_OF_LIST(),
1343 };
1344 
1345 static void sysbus_ahci_class_init(ObjectClass *klass, void *data)
1346 {
1347     DeviceClass *dc = DEVICE_CLASS(klass);
1348 
1349     dc->realize = sysbus_ahci_realize;
1350     dc->vmsd = &vmstate_sysbus_ahci;
1351     dc->props = sysbus_ahci_properties;
1352     dc->reset = sysbus_ahci_reset;
1353     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1354 }
1355 
1356 static const TypeInfo sysbus_ahci_info = {
1357     .name          = TYPE_SYSBUS_AHCI,
1358     .parent        = TYPE_SYS_BUS_DEVICE,
1359     .instance_size = sizeof(SysbusAHCIState),
1360     .class_init    = sysbus_ahci_class_init,
1361 };
1362 
1363 static void sysbus_ahci_register_types(void)
1364 {
1365     type_register_static(&sysbus_ahci_info);
1366 }
1367 
1368 type_init(sysbus_ahci_register_types)
1369