xref: /openbmc/qemu/hw/scsi/vmw_pvscsi.c (revision b70ce101)
1 /*
2  * QEMU VMWARE PVSCSI paravirtual SCSI bus
3  *
4  * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
5  *
6  * Developed by Daynix Computing LTD (http://www.daynix.com)
7  *
8  * Based on implementation by Paolo Bonzini
9  * http://lists.gnu.org/archive/html/qemu-devel/2011-08/msg00729.html
10  *
11  * Authors:
12  * Paolo Bonzini <pbonzini@redhat.com>
13  * Dmitry Fleytman <dmitry@daynix.com>
14  * Yan Vugenfirer <yan@daynix.com>
15  *
16  * This work is licensed under the terms of the GNU GPL, version 2.
17  * See the COPYING file in the top-level directory.
18  *
19  * NOTE about MSI-X:
20  * MSI-X support has been removed for the moment because it leads Windows OS
21  * to crash on startup. The crash happens because Windows driver requires
22  * MSI-X shared memory to be part of the same BAR used for rings state
23  * registers, etc. This is not supported by QEMU infrastructure so separate
24  * BAR created from MSI-X purposes. Windows driver fails to deal with 2 BARs.
25  *
26  */
27 
28 #include "qemu/osdep.h"
29 #include "qapi/error.h"
30 #include "hw/scsi/scsi.h"
31 #include <block/scsi.h>
32 #include "hw/pci/msi.h"
33 #include "vmw_pvscsi.h"
34 #include "trace.h"
35 
36 
37 #define PVSCSI_USE_64BIT         (true)
38 #define PVSCSI_PER_VECTOR_MASK   (false)
39 
40 #define PVSCSI_MAX_DEVS                   (64)
41 #define PVSCSI_MSIX_NUM_VECTORS           (1)
42 
43 #define PVSCSI_MAX_CMD_DATA_WORDS \
44     (sizeof(PVSCSICmdDescSetupRings)/sizeof(uint32_t))
45 
46 #define RS_GET_FIELD(m, field) \
47     (ldl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
48                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field)))
49 #define RS_SET_FIELD(m, field, val) \
50     (stl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
51                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field), val))
52 
53 typedef struct PVSCSIClass {
54     PCIDeviceClass parent_class;
55     DeviceRealize parent_dc_realize;
56 } PVSCSIClass;
57 
58 #define TYPE_PVSCSI "pvscsi"
59 #define PVSCSI(obj) OBJECT_CHECK(PVSCSIState, (obj), TYPE_PVSCSI)
60 
61 #define PVSCSI_DEVICE_CLASS(klass) \
62     OBJECT_CLASS_CHECK(PVSCSIClass, (klass), TYPE_PVSCSI)
63 #define PVSCSI_DEVICE_GET_CLASS(obj) \
64     OBJECT_GET_CLASS(PVSCSIClass, (obj), TYPE_PVSCSI)
65 
66 /* Compatibility flags for migration */
67 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT 0
68 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION \
69     (1 << PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT)
70 #define PVSCSI_COMPAT_DISABLE_PCIE_BIT 1
71 #define PVSCSI_COMPAT_DISABLE_PCIE \
72     (1 << PVSCSI_COMPAT_DISABLE_PCIE_BIT)
73 
74 #define PVSCSI_USE_OLD_PCI_CONFIGURATION(s) \
75     ((s)->compat_flags & PVSCSI_COMPAT_OLD_PCI_CONFIGURATION)
76 #define PVSCSI_MSI_OFFSET(s) \
77     (PVSCSI_USE_OLD_PCI_CONFIGURATION(s) ? 0x50 : 0x7c)
78 #define PVSCSI_EXP_EP_OFFSET (0x40)
79 
80 typedef struct PVSCSIRingInfo {
81     uint64_t            rs_pa;
82     uint32_t            txr_len_mask;
83     uint32_t            rxr_len_mask;
84     uint32_t            msg_len_mask;
85     uint64_t            req_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
86     uint64_t            cmp_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
87     uint64_t            msg_ring_pages_pa[PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES];
88     uint64_t            consumed_ptr;
89     uint64_t            filled_cmp_ptr;
90     uint64_t            filled_msg_ptr;
91 } PVSCSIRingInfo;
92 
93 typedef struct PVSCSISGState {
94     hwaddr elemAddr;
95     hwaddr dataAddr;
96     uint32_t resid;
97 } PVSCSISGState;
98 
99 typedef QTAILQ_HEAD(, PVSCSIRequest) PVSCSIRequestList;
100 
101 typedef struct {
102     PCIDevice parent_obj;
103     MemoryRegion io_space;
104     SCSIBus bus;
105     QEMUBH *completion_worker;
106     PVSCSIRequestList pending_queue;
107     PVSCSIRequestList completion_queue;
108 
109     uint64_t reg_interrupt_status;        /* Interrupt status register value */
110     uint64_t reg_interrupt_enabled;       /* Interrupt mask register value   */
111     uint64_t reg_command_status;          /* Command status register value   */
112 
113     /* Command data adoption mechanism */
114     uint64_t curr_cmd;                   /* Last command arrived             */
115     uint32_t curr_cmd_data_cntr;         /* Amount of data for last command  */
116 
117     /* Collector for current command data */
118     uint32_t curr_cmd_data[PVSCSI_MAX_CMD_DATA_WORDS];
119 
120     uint8_t rings_info_valid;            /* Whether data rings initialized   */
121     uint8_t msg_ring_info_valid;         /* Whether message ring initialized */
122     uint8_t use_msg;                     /* Whether to use message ring      */
123 
124     uint8_t msi_used;                    /* For migration compatibility      */
125     PVSCSIRingInfo rings;                /* Data transfer rings manager      */
126     uint32_t resetting;                  /* Reset in progress                */
127 
128     uint32_t compat_flags;
129 } PVSCSIState;
130 
131 typedef struct PVSCSIRequest {
132     SCSIRequest *sreq;
133     PVSCSIState *dev;
134     uint8_t sense_key;
135     uint8_t completed;
136     int lun;
137     QEMUSGList sgl;
138     PVSCSISGState sg;
139     struct PVSCSIRingReqDesc req;
140     struct PVSCSIRingCmpDesc cmp;
141     QTAILQ_ENTRY(PVSCSIRequest) next;
142 } PVSCSIRequest;
143 
144 /* Integer binary logarithm */
145 static int
146 pvscsi_log2(uint32_t input)
147 {
148     int log = 0;
149     assert(input > 0);
150     while (input >> ++log) {
151     }
152     return log;
153 }
154 
155 static int
156 pvscsi_ring_init_data(PVSCSIRingInfo *m, PVSCSICmdDescSetupRings *ri)
157 {
158     int i;
159     uint32_t txr_len_log2, rxr_len_log2;
160     uint32_t req_ring_size, cmp_ring_size;
161     m->rs_pa = ri->ringsStatePPN << VMW_PAGE_SHIFT;
162 
163     if ((ri->reqRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES)
164         || (ri->cmpRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES)) {
165         return -1;
166     }
167     req_ring_size = ri->reqRingNumPages * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
168     cmp_ring_size = ri->cmpRingNumPages * PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
169     txr_len_log2 = pvscsi_log2(req_ring_size - 1);
170     rxr_len_log2 = pvscsi_log2(cmp_ring_size - 1);
171 
172     m->txr_len_mask = MASK(txr_len_log2);
173     m->rxr_len_mask = MASK(rxr_len_log2);
174 
175     m->consumed_ptr = 0;
176     m->filled_cmp_ptr = 0;
177 
178     for (i = 0; i < ri->reqRingNumPages; i++) {
179         m->req_ring_pages_pa[i] = ri->reqRingPPNs[i] << VMW_PAGE_SHIFT;
180     }
181 
182     for (i = 0; i < ri->cmpRingNumPages; i++) {
183         m->cmp_ring_pages_pa[i] = ri->cmpRingPPNs[i] << VMW_PAGE_SHIFT;
184     }
185 
186     RS_SET_FIELD(m, reqProdIdx, 0);
187     RS_SET_FIELD(m, reqConsIdx, 0);
188     RS_SET_FIELD(m, reqNumEntriesLog2, txr_len_log2);
189 
190     RS_SET_FIELD(m, cmpProdIdx, 0);
191     RS_SET_FIELD(m, cmpConsIdx, 0);
192     RS_SET_FIELD(m, cmpNumEntriesLog2, rxr_len_log2);
193 
194     trace_pvscsi_ring_init_data(txr_len_log2, rxr_len_log2);
195 
196     /* Flush ring state page changes */
197     smp_wmb();
198 
199     return 0;
200 }
201 
202 static int
203 pvscsi_ring_init_msg(PVSCSIRingInfo *m, PVSCSICmdDescSetupMsgRing *ri)
204 {
205     int i;
206     uint32_t len_log2;
207     uint32_t ring_size;
208 
209     if (ri->numPages > PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES) {
210         return -1;
211     }
212     ring_size = ri->numPages * PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
213     len_log2 = pvscsi_log2(ring_size - 1);
214 
215     m->msg_len_mask = MASK(len_log2);
216 
217     m->filled_msg_ptr = 0;
218 
219     for (i = 0; i < ri->numPages; i++) {
220         m->msg_ring_pages_pa[i] = ri->ringPPNs[i] << VMW_PAGE_SHIFT;
221     }
222 
223     RS_SET_FIELD(m, msgProdIdx, 0);
224     RS_SET_FIELD(m, msgConsIdx, 0);
225     RS_SET_FIELD(m, msgNumEntriesLog2, len_log2);
226 
227     trace_pvscsi_ring_init_msg(len_log2);
228 
229     /* Flush ring state page changes */
230     smp_wmb();
231 
232     return 0;
233 }
234 
235 static void
236 pvscsi_ring_cleanup(PVSCSIRingInfo *mgr)
237 {
238     mgr->rs_pa = 0;
239     mgr->txr_len_mask = 0;
240     mgr->rxr_len_mask = 0;
241     mgr->msg_len_mask = 0;
242     mgr->consumed_ptr = 0;
243     mgr->filled_cmp_ptr = 0;
244     mgr->filled_msg_ptr = 0;
245     memset(mgr->req_ring_pages_pa, 0, sizeof(mgr->req_ring_pages_pa));
246     memset(mgr->cmp_ring_pages_pa, 0, sizeof(mgr->cmp_ring_pages_pa));
247     memset(mgr->msg_ring_pages_pa, 0, sizeof(mgr->msg_ring_pages_pa));
248 }
249 
250 static hwaddr
251 pvscsi_ring_pop_req_descr(PVSCSIRingInfo *mgr)
252 {
253     uint32_t ready_ptr = RS_GET_FIELD(mgr, reqProdIdx);
254 
255     if (ready_ptr != mgr->consumed_ptr) {
256         uint32_t next_ready_ptr =
257             mgr->consumed_ptr++ & mgr->txr_len_mask;
258         uint32_t next_ready_page =
259             next_ready_ptr / PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
260         uint32_t inpage_idx =
261             next_ready_ptr % PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
262 
263         return mgr->req_ring_pages_pa[next_ready_page] +
264                inpage_idx * sizeof(PVSCSIRingReqDesc);
265     } else {
266         return 0;
267     }
268 }
269 
270 static void
271 pvscsi_ring_flush_req(PVSCSIRingInfo *mgr)
272 {
273     RS_SET_FIELD(mgr, reqConsIdx, mgr->consumed_ptr);
274 }
275 
276 static hwaddr
277 pvscsi_ring_pop_cmp_descr(PVSCSIRingInfo *mgr)
278 {
279     /*
280      * According to Linux driver code it explicitly verifies that number
281      * of requests being processed by device is less then the size of
282      * completion queue, so device may omit completion queue overflow
283      * conditions check. We assume that this is true for other (Windows)
284      * drivers as well.
285      */
286 
287     uint32_t free_cmp_ptr =
288         mgr->filled_cmp_ptr++ & mgr->rxr_len_mask;
289     uint32_t free_cmp_page =
290         free_cmp_ptr / PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
291     uint32_t inpage_idx =
292         free_cmp_ptr % PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
293     return mgr->cmp_ring_pages_pa[free_cmp_page] +
294            inpage_idx * sizeof(PVSCSIRingCmpDesc);
295 }
296 
297 static hwaddr
298 pvscsi_ring_pop_msg_descr(PVSCSIRingInfo *mgr)
299 {
300     uint32_t free_msg_ptr =
301         mgr->filled_msg_ptr++ & mgr->msg_len_mask;
302     uint32_t free_msg_page =
303         free_msg_ptr / PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
304     uint32_t inpage_idx =
305         free_msg_ptr % PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
306     return mgr->msg_ring_pages_pa[free_msg_page] +
307            inpage_idx * sizeof(PVSCSIRingMsgDesc);
308 }
309 
310 static void
311 pvscsi_ring_flush_cmp(PVSCSIRingInfo *mgr)
312 {
313     /* Flush descriptor changes */
314     smp_wmb();
315 
316     trace_pvscsi_ring_flush_cmp(mgr->filled_cmp_ptr);
317 
318     RS_SET_FIELD(mgr, cmpProdIdx, mgr->filled_cmp_ptr);
319 }
320 
321 static bool
322 pvscsi_ring_msg_has_room(PVSCSIRingInfo *mgr)
323 {
324     uint32_t prodIdx = RS_GET_FIELD(mgr, msgProdIdx);
325     uint32_t consIdx = RS_GET_FIELD(mgr, msgConsIdx);
326 
327     return (prodIdx - consIdx) < (mgr->msg_len_mask + 1);
328 }
329 
330 static void
331 pvscsi_ring_flush_msg(PVSCSIRingInfo *mgr)
332 {
333     /* Flush descriptor changes */
334     smp_wmb();
335 
336     trace_pvscsi_ring_flush_msg(mgr->filled_msg_ptr);
337 
338     RS_SET_FIELD(mgr, msgProdIdx, mgr->filled_msg_ptr);
339 }
340 
341 static void
342 pvscsi_reset_state(PVSCSIState *s)
343 {
344     s->curr_cmd = PVSCSI_CMD_FIRST;
345     s->curr_cmd_data_cntr = 0;
346     s->reg_command_status = PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
347     s->reg_interrupt_status = 0;
348     pvscsi_ring_cleanup(&s->rings);
349     s->rings_info_valid = FALSE;
350     s->msg_ring_info_valid = FALSE;
351     QTAILQ_INIT(&s->pending_queue);
352     QTAILQ_INIT(&s->completion_queue);
353 }
354 
355 static void
356 pvscsi_update_irq_status(PVSCSIState *s)
357 {
358     PCIDevice *d = PCI_DEVICE(s);
359     bool should_raise = s->reg_interrupt_enabled & s->reg_interrupt_status;
360 
361     trace_pvscsi_update_irq_level(should_raise, s->reg_interrupt_enabled,
362                                   s->reg_interrupt_status);
363 
364     if (msi_enabled(d)) {
365         if (should_raise) {
366             trace_pvscsi_update_irq_msi();
367             msi_notify(d, PVSCSI_VECTOR_COMPLETION);
368         }
369         return;
370     }
371 
372     pci_set_irq(d, !!should_raise);
373 }
374 
375 static void
376 pvscsi_raise_completion_interrupt(PVSCSIState *s)
377 {
378     s->reg_interrupt_status |= PVSCSI_INTR_CMPL_0;
379 
380     /* Memory barrier to flush interrupt status register changes*/
381     smp_wmb();
382 
383     pvscsi_update_irq_status(s);
384 }
385 
386 static void
387 pvscsi_raise_message_interrupt(PVSCSIState *s)
388 {
389     s->reg_interrupt_status |= PVSCSI_INTR_MSG_0;
390 
391     /* Memory barrier to flush interrupt status register changes*/
392     smp_wmb();
393 
394     pvscsi_update_irq_status(s);
395 }
396 
397 static void
398 pvscsi_cmp_ring_put(PVSCSIState *s, struct PVSCSIRingCmpDesc *cmp_desc)
399 {
400     hwaddr cmp_descr_pa;
401 
402     cmp_descr_pa = pvscsi_ring_pop_cmp_descr(&s->rings);
403     trace_pvscsi_cmp_ring_put(cmp_descr_pa);
404     cpu_physical_memory_write(cmp_descr_pa, (void *)cmp_desc,
405                               sizeof(*cmp_desc));
406 }
407 
408 static void
409 pvscsi_msg_ring_put(PVSCSIState *s, struct PVSCSIRingMsgDesc *msg_desc)
410 {
411     hwaddr msg_descr_pa;
412 
413     msg_descr_pa = pvscsi_ring_pop_msg_descr(&s->rings);
414     trace_pvscsi_msg_ring_put(msg_descr_pa);
415     cpu_physical_memory_write(msg_descr_pa, (void *)msg_desc,
416                               sizeof(*msg_desc));
417 }
418 
419 static void
420 pvscsi_process_completion_queue(void *opaque)
421 {
422     PVSCSIState *s = opaque;
423     PVSCSIRequest *pvscsi_req;
424     bool has_completed = false;
425 
426     while (!QTAILQ_EMPTY(&s->completion_queue)) {
427         pvscsi_req = QTAILQ_FIRST(&s->completion_queue);
428         QTAILQ_REMOVE(&s->completion_queue, pvscsi_req, next);
429         pvscsi_cmp_ring_put(s, &pvscsi_req->cmp);
430         g_free(pvscsi_req);
431         has_completed = true;
432     }
433 
434     if (has_completed) {
435         pvscsi_ring_flush_cmp(&s->rings);
436         pvscsi_raise_completion_interrupt(s);
437     }
438 }
439 
440 static void
441 pvscsi_reset_adapter(PVSCSIState *s)
442 {
443     s->resetting++;
444     qbus_reset_all_fn(&s->bus);
445     s->resetting--;
446     pvscsi_process_completion_queue(s);
447     assert(QTAILQ_EMPTY(&s->pending_queue));
448     pvscsi_reset_state(s);
449 }
450 
451 static void
452 pvscsi_schedule_completion_processing(PVSCSIState *s)
453 {
454     /* Try putting more complete requests on the ring. */
455     if (!QTAILQ_EMPTY(&s->completion_queue)) {
456         qemu_bh_schedule(s->completion_worker);
457     }
458 }
459 
460 static void
461 pvscsi_complete_request(PVSCSIState *s, PVSCSIRequest *r)
462 {
463     assert(!r->completed);
464 
465     trace_pvscsi_complete_request(r->cmp.context, r->cmp.dataLen,
466                                   r->sense_key);
467     if (r->sreq != NULL) {
468         scsi_req_unref(r->sreq);
469         r->sreq = NULL;
470     }
471     r->completed = 1;
472     QTAILQ_REMOVE(&s->pending_queue, r, next);
473     QTAILQ_INSERT_TAIL(&s->completion_queue, r, next);
474     pvscsi_schedule_completion_processing(s);
475 }
476 
477 static QEMUSGList *pvscsi_get_sg_list(SCSIRequest *r)
478 {
479     PVSCSIRequest *req = r->hba_private;
480 
481     trace_pvscsi_get_sg_list(req->sgl.nsg, req->sgl.size);
482 
483     return &req->sgl;
484 }
485 
486 static void
487 pvscsi_get_next_sg_elem(PVSCSISGState *sg)
488 {
489     struct PVSCSISGElement elem;
490 
491     cpu_physical_memory_read(sg->elemAddr, (void *)&elem, sizeof(elem));
492     if ((elem.flags & ~PVSCSI_KNOWN_FLAGS) != 0) {
493         /*
494             * There is PVSCSI_SGE_FLAG_CHAIN_ELEMENT flag described in
495             * header file but its value is unknown. This flag requires
496             * additional processing, so we put warning here to catch it
497             * some day and make proper implementation
498             */
499         trace_pvscsi_get_next_sg_elem(elem.flags);
500     }
501 
502     sg->elemAddr += sizeof(elem);
503     sg->dataAddr = elem.addr;
504     sg->resid = elem.length;
505 }
506 
507 static void
508 pvscsi_write_sense(PVSCSIRequest *r, uint8_t *sense, int len)
509 {
510     r->cmp.senseLen = MIN(r->req.senseLen, len);
511     r->sense_key = sense[(sense[0] & 2) ? 1 : 2];
512     cpu_physical_memory_write(r->req.senseAddr, sense, r->cmp.senseLen);
513 }
514 
515 static void
516 pvscsi_command_complete(SCSIRequest *req, uint32_t status, size_t resid)
517 {
518     PVSCSIRequest *pvscsi_req = req->hba_private;
519     PVSCSIState *s;
520 
521     if (!pvscsi_req) {
522         trace_pvscsi_command_complete_not_found(req->tag);
523         return;
524     }
525     s = pvscsi_req->dev;
526 
527     if (resid) {
528         /* Short transfer.  */
529         trace_pvscsi_command_complete_data_run();
530         pvscsi_req->cmp.hostStatus = BTSTAT_DATARUN;
531     }
532 
533     pvscsi_req->cmp.scsiStatus = status;
534     if (pvscsi_req->cmp.scsiStatus == CHECK_CONDITION) {
535         uint8_t sense[SCSI_SENSE_BUF_SIZE];
536         int sense_len =
537             scsi_req_get_sense(pvscsi_req->sreq, sense, sizeof(sense));
538 
539         trace_pvscsi_command_complete_sense_len(sense_len);
540         pvscsi_write_sense(pvscsi_req, sense, sense_len);
541     }
542     qemu_sglist_destroy(&pvscsi_req->sgl);
543     pvscsi_complete_request(s, pvscsi_req);
544 }
545 
546 static void
547 pvscsi_send_msg(PVSCSIState *s, SCSIDevice *dev, uint32_t msg_type)
548 {
549     if (s->msg_ring_info_valid && pvscsi_ring_msg_has_room(&s->rings)) {
550         PVSCSIMsgDescDevStatusChanged msg = {0};
551 
552         msg.type = msg_type;
553         msg.bus = dev->channel;
554         msg.target = dev->id;
555         msg.lun[1] = dev->lun;
556 
557         pvscsi_msg_ring_put(s, (PVSCSIRingMsgDesc *)&msg);
558         pvscsi_ring_flush_msg(&s->rings);
559         pvscsi_raise_message_interrupt(s);
560     }
561 }
562 
563 static void
564 pvscsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
565 {
566     PVSCSIState *s = PVSCSI(hotplug_dev);
567 
568     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_ADDED);
569 }
570 
571 static void
572 pvscsi_hot_unplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
573 {
574     PVSCSIState *s = PVSCSI(hotplug_dev);
575 
576     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_REMOVED);
577     qdev_simple_device_unplug_cb(hotplug_dev, dev, errp);
578 }
579 
580 static void
581 pvscsi_request_cancelled(SCSIRequest *req)
582 {
583     PVSCSIRequest *pvscsi_req = req->hba_private;
584     PVSCSIState *s = pvscsi_req->dev;
585 
586     if (pvscsi_req->completed) {
587         return;
588     }
589 
590    if (pvscsi_req->dev->resetting) {
591        pvscsi_req->cmp.hostStatus = BTSTAT_BUSRESET;
592     } else {
593        pvscsi_req->cmp.hostStatus = BTSTAT_ABORTQUEUE;
594     }
595 
596     pvscsi_complete_request(s, pvscsi_req);
597 }
598 
599 static SCSIDevice*
600 pvscsi_device_find(PVSCSIState *s, int channel, int target,
601                    uint8_t *requested_lun, uint8_t *target_lun)
602 {
603     if (requested_lun[0] || requested_lun[2] || requested_lun[3] ||
604         requested_lun[4] || requested_lun[5] || requested_lun[6] ||
605         requested_lun[7] || (target > PVSCSI_MAX_DEVS)) {
606         return NULL;
607     } else {
608         *target_lun = requested_lun[1];
609         return scsi_device_find(&s->bus, channel, target, *target_lun);
610     }
611 }
612 
613 static PVSCSIRequest *
614 pvscsi_queue_pending_descriptor(PVSCSIState *s, SCSIDevice **d,
615                                 struct PVSCSIRingReqDesc *descr)
616 {
617     PVSCSIRequest *pvscsi_req;
618     uint8_t lun;
619 
620     pvscsi_req = g_malloc0(sizeof(*pvscsi_req));
621     pvscsi_req->dev = s;
622     pvscsi_req->req = *descr;
623     pvscsi_req->cmp.context = pvscsi_req->req.context;
624     QTAILQ_INSERT_TAIL(&s->pending_queue, pvscsi_req, next);
625 
626     *d = pvscsi_device_find(s, descr->bus, descr->target, descr->lun, &lun);
627     if (*d) {
628         pvscsi_req->lun = lun;
629     }
630 
631     return pvscsi_req;
632 }
633 
634 static void
635 pvscsi_convert_sglist(PVSCSIRequest *r)
636 {
637     int chunk_size;
638     uint64_t data_length = r->req.dataLen;
639     PVSCSISGState sg = r->sg;
640     while (data_length) {
641         while (!sg.resid) {
642             pvscsi_get_next_sg_elem(&sg);
643             trace_pvscsi_convert_sglist(r->req.context, r->sg.dataAddr,
644                                         r->sg.resid);
645         }
646         assert(data_length > 0);
647         chunk_size = MIN((unsigned) data_length, sg.resid);
648         if (chunk_size) {
649             qemu_sglist_add(&r->sgl, sg.dataAddr, chunk_size);
650         }
651 
652         sg.dataAddr += chunk_size;
653         data_length -= chunk_size;
654         sg.resid -= chunk_size;
655     }
656 }
657 
658 static void
659 pvscsi_build_sglist(PVSCSIState *s, PVSCSIRequest *r)
660 {
661     PCIDevice *d = PCI_DEVICE(s);
662 
663     pci_dma_sglist_init(&r->sgl, d, 1);
664     if (r->req.flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
665         pvscsi_convert_sglist(r);
666     } else {
667         qemu_sglist_add(&r->sgl, r->req.dataAddr, r->req.dataLen);
668     }
669 }
670 
671 static void
672 pvscsi_process_request_descriptor(PVSCSIState *s,
673                                   struct PVSCSIRingReqDesc *descr)
674 {
675     SCSIDevice *d;
676     PVSCSIRequest *r = pvscsi_queue_pending_descriptor(s, &d, descr);
677     int64_t n;
678 
679     trace_pvscsi_process_req_descr(descr->cdb[0], descr->context);
680 
681     if (!d) {
682         r->cmp.hostStatus = BTSTAT_SELTIMEO;
683         trace_pvscsi_process_req_descr_unknown_device();
684         pvscsi_complete_request(s, r);
685         return;
686     }
687 
688     if (descr->flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
689         r->sg.elemAddr = descr->dataAddr;
690     }
691 
692     r->sreq = scsi_req_new(d, descr->context, r->lun, descr->cdb, r);
693     if (r->sreq->cmd.mode == SCSI_XFER_FROM_DEV &&
694         (descr->flags & PVSCSI_FLAG_CMD_DIR_TODEVICE)) {
695         r->cmp.hostStatus = BTSTAT_BADMSG;
696         trace_pvscsi_process_req_descr_invalid_dir();
697         scsi_req_cancel(r->sreq);
698         return;
699     }
700     if (r->sreq->cmd.mode == SCSI_XFER_TO_DEV &&
701         (descr->flags & PVSCSI_FLAG_CMD_DIR_TOHOST)) {
702         r->cmp.hostStatus = BTSTAT_BADMSG;
703         trace_pvscsi_process_req_descr_invalid_dir();
704         scsi_req_cancel(r->sreq);
705         return;
706     }
707 
708     pvscsi_build_sglist(s, r);
709     n = scsi_req_enqueue(r->sreq);
710 
711     if (n) {
712         scsi_req_continue(r->sreq);
713     }
714 }
715 
716 static void
717 pvscsi_process_io(PVSCSIState *s)
718 {
719     PVSCSIRingReqDesc descr;
720     hwaddr next_descr_pa;
721 
722     assert(s->rings_info_valid);
723     while ((next_descr_pa = pvscsi_ring_pop_req_descr(&s->rings)) != 0) {
724 
725         /* Only read after production index verification */
726         smp_rmb();
727 
728         trace_pvscsi_process_io(next_descr_pa);
729         cpu_physical_memory_read(next_descr_pa, &descr, sizeof(descr));
730         pvscsi_process_request_descriptor(s, &descr);
731     }
732 
733     pvscsi_ring_flush_req(&s->rings);
734 }
735 
736 static void
737 pvscsi_dbg_dump_tx_rings_config(PVSCSICmdDescSetupRings *rc)
738 {
739     int i;
740     trace_pvscsi_tx_rings_ppn("Rings State", rc->ringsStatePPN);
741 
742     trace_pvscsi_tx_rings_num_pages("Request Ring", rc->reqRingNumPages);
743     for (i = 0; i < rc->reqRingNumPages; i++) {
744         trace_pvscsi_tx_rings_ppn("Request Ring", rc->reqRingPPNs[i]);
745     }
746 
747     trace_pvscsi_tx_rings_num_pages("Confirm Ring", rc->cmpRingNumPages);
748     for (i = 0; i < rc->cmpRingNumPages; i++) {
749         trace_pvscsi_tx_rings_ppn("Confirm Ring", rc->reqRingPPNs[i]);
750     }
751 }
752 
753 static uint64_t
754 pvscsi_on_cmd_config(PVSCSIState *s)
755 {
756     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_CONFIG");
757     return PVSCSI_COMMAND_PROCESSING_FAILED;
758 }
759 
760 static uint64_t
761 pvscsi_on_cmd_unplug(PVSCSIState *s)
762 {
763     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_DEVICE_UNPLUG");
764     return PVSCSI_COMMAND_PROCESSING_FAILED;
765 }
766 
767 static uint64_t
768 pvscsi_on_issue_scsi(PVSCSIState *s)
769 {
770     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_ISSUE_SCSI");
771     return PVSCSI_COMMAND_PROCESSING_FAILED;
772 }
773 
774 static uint64_t
775 pvscsi_on_cmd_setup_rings(PVSCSIState *s)
776 {
777     PVSCSICmdDescSetupRings *rc =
778         (PVSCSICmdDescSetupRings *) s->curr_cmd_data;
779 
780     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_RINGS");
781 
782     pvscsi_dbg_dump_tx_rings_config(rc);
783     if (pvscsi_ring_init_data(&s->rings, rc) < 0) {
784         return PVSCSI_COMMAND_PROCESSING_FAILED;
785     }
786 
787     s->rings_info_valid = TRUE;
788     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
789 }
790 
791 static uint64_t
792 pvscsi_on_cmd_abort(PVSCSIState *s)
793 {
794     PVSCSICmdDescAbortCmd *cmd = (PVSCSICmdDescAbortCmd *) s->curr_cmd_data;
795     PVSCSIRequest *r, *next;
796 
797     trace_pvscsi_on_cmd_abort(cmd->context, cmd->target);
798 
799     QTAILQ_FOREACH_SAFE(r, &s->pending_queue, next, next) {
800         if (r->req.context == cmd->context) {
801             break;
802         }
803     }
804     if (r) {
805         assert(!r->completed);
806         r->cmp.hostStatus = BTSTAT_ABORTQUEUE;
807         scsi_req_cancel(r->sreq);
808     }
809 
810     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
811 }
812 
813 static uint64_t
814 pvscsi_on_cmd_unknown(PVSCSIState *s)
815 {
816     trace_pvscsi_on_cmd_unknown_data(s->curr_cmd_data[0]);
817     return PVSCSI_COMMAND_PROCESSING_FAILED;
818 }
819 
820 static uint64_t
821 pvscsi_on_cmd_reset_device(PVSCSIState *s)
822 {
823     uint8_t target_lun = 0;
824     struct PVSCSICmdDescResetDevice *cmd =
825         (struct PVSCSICmdDescResetDevice *) s->curr_cmd_data;
826     SCSIDevice *sdev;
827 
828     sdev = pvscsi_device_find(s, 0, cmd->target, cmd->lun, &target_lun);
829 
830     trace_pvscsi_on_cmd_reset_dev(cmd->target, (int) target_lun, sdev);
831 
832     if (sdev != NULL) {
833         s->resetting++;
834         device_reset(&sdev->qdev);
835         s->resetting--;
836         return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
837     }
838 
839     return PVSCSI_COMMAND_PROCESSING_FAILED;
840 }
841 
842 static uint64_t
843 pvscsi_on_cmd_reset_bus(PVSCSIState *s)
844 {
845     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_RESET_BUS");
846 
847     s->resetting++;
848     qbus_reset_all_fn(&s->bus);
849     s->resetting--;
850     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
851 }
852 
853 static uint64_t
854 pvscsi_on_cmd_setup_msg_ring(PVSCSIState *s)
855 {
856     PVSCSICmdDescSetupMsgRing *rc =
857         (PVSCSICmdDescSetupMsgRing *) s->curr_cmd_data;
858 
859     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_MSG_RING");
860 
861     if (!s->use_msg) {
862         return PVSCSI_COMMAND_PROCESSING_FAILED;
863     }
864 
865     if (s->rings_info_valid) {
866         if (pvscsi_ring_init_msg(&s->rings, rc) < 0) {
867             return PVSCSI_COMMAND_PROCESSING_FAILED;
868         }
869         s->msg_ring_info_valid = TRUE;
870     }
871     return sizeof(PVSCSICmdDescSetupMsgRing) / sizeof(uint32_t);
872 }
873 
874 static uint64_t
875 pvscsi_on_cmd_adapter_reset(PVSCSIState *s)
876 {
877     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_ADAPTER_RESET");
878 
879     pvscsi_reset_adapter(s);
880     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
881 }
882 
883 static const struct {
884     int       data_size;
885     uint64_t  (*handler_fn)(PVSCSIState *s);
886 } pvscsi_commands[] = {
887     [PVSCSI_CMD_FIRST] = {
888         .data_size = 0,
889         .handler_fn = pvscsi_on_cmd_unknown,
890     },
891 
892     /* Not implemented, data size defined based on what arrives on windows */
893     [PVSCSI_CMD_CONFIG] = {
894         .data_size = 6 * sizeof(uint32_t),
895         .handler_fn = pvscsi_on_cmd_config,
896     },
897 
898     /* Command not implemented, data size is unknown */
899     [PVSCSI_CMD_ISSUE_SCSI] = {
900         .data_size = 0,
901         .handler_fn = pvscsi_on_issue_scsi,
902     },
903 
904     /* Command not implemented, data size is unknown */
905     [PVSCSI_CMD_DEVICE_UNPLUG] = {
906         .data_size = 0,
907         .handler_fn = pvscsi_on_cmd_unplug,
908     },
909 
910     [PVSCSI_CMD_SETUP_RINGS] = {
911         .data_size = sizeof(PVSCSICmdDescSetupRings),
912         .handler_fn = pvscsi_on_cmd_setup_rings,
913     },
914 
915     [PVSCSI_CMD_RESET_DEVICE] = {
916         .data_size = sizeof(struct PVSCSICmdDescResetDevice),
917         .handler_fn = pvscsi_on_cmd_reset_device,
918     },
919 
920     [PVSCSI_CMD_RESET_BUS] = {
921         .data_size = 0,
922         .handler_fn = pvscsi_on_cmd_reset_bus,
923     },
924 
925     [PVSCSI_CMD_SETUP_MSG_RING] = {
926         .data_size = sizeof(PVSCSICmdDescSetupMsgRing),
927         .handler_fn = pvscsi_on_cmd_setup_msg_ring,
928     },
929 
930     [PVSCSI_CMD_ADAPTER_RESET] = {
931         .data_size = 0,
932         .handler_fn = pvscsi_on_cmd_adapter_reset,
933     },
934 
935     [PVSCSI_CMD_ABORT_CMD] = {
936         .data_size = sizeof(struct PVSCSICmdDescAbortCmd),
937         .handler_fn = pvscsi_on_cmd_abort,
938     },
939 };
940 
941 static void
942 pvscsi_do_command_processing(PVSCSIState *s)
943 {
944     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
945 
946     assert(s->curr_cmd < PVSCSI_CMD_LAST);
947     if (bytes_arrived >= pvscsi_commands[s->curr_cmd].data_size) {
948         s->reg_command_status = pvscsi_commands[s->curr_cmd].handler_fn(s);
949         s->curr_cmd = PVSCSI_CMD_FIRST;
950         s->curr_cmd_data_cntr   = 0;
951     }
952 }
953 
954 static void
955 pvscsi_on_command_data(PVSCSIState *s, uint32_t value)
956 {
957     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
958 
959     assert(bytes_arrived < sizeof(s->curr_cmd_data));
960     s->curr_cmd_data[s->curr_cmd_data_cntr++] = value;
961 
962     pvscsi_do_command_processing(s);
963 }
964 
965 static void
966 pvscsi_on_command(PVSCSIState *s, uint64_t cmd_id)
967 {
968     if ((cmd_id > PVSCSI_CMD_FIRST) && (cmd_id < PVSCSI_CMD_LAST)) {
969         s->curr_cmd = cmd_id;
970     } else {
971         s->curr_cmd = PVSCSI_CMD_FIRST;
972         trace_pvscsi_on_cmd_unknown(cmd_id);
973     }
974 
975     s->curr_cmd_data_cntr = 0;
976     s->reg_command_status = PVSCSI_COMMAND_NOT_ENOUGH_DATA;
977 
978     pvscsi_do_command_processing(s);
979 }
980 
981 static void
982 pvscsi_io_write(void *opaque, hwaddr addr,
983                 uint64_t val, unsigned size)
984 {
985     PVSCSIState *s = opaque;
986 
987     switch (addr) {
988     case PVSCSI_REG_OFFSET_COMMAND:
989         pvscsi_on_command(s, val);
990         break;
991 
992     case PVSCSI_REG_OFFSET_COMMAND_DATA:
993         pvscsi_on_command_data(s, (uint32_t) val);
994         break;
995 
996     case PVSCSI_REG_OFFSET_INTR_STATUS:
997         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_STATUS", val);
998         s->reg_interrupt_status &= ~val;
999         pvscsi_update_irq_status(s);
1000         pvscsi_schedule_completion_processing(s);
1001         break;
1002 
1003     case PVSCSI_REG_OFFSET_INTR_MASK:
1004         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_MASK", val);
1005         s->reg_interrupt_enabled = val;
1006         pvscsi_update_irq_status(s);
1007         break;
1008 
1009     case PVSCSI_REG_OFFSET_KICK_NON_RW_IO:
1010         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_NON_RW_IO", val);
1011         pvscsi_process_io(s);
1012         break;
1013 
1014     case PVSCSI_REG_OFFSET_KICK_RW_IO:
1015         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_RW_IO", val);
1016         pvscsi_process_io(s);
1017         break;
1018 
1019     case PVSCSI_REG_OFFSET_DEBUG:
1020         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_DEBUG", val);
1021         break;
1022 
1023     default:
1024         trace_pvscsi_io_write_unknown(addr, size, val);
1025         break;
1026     }
1027 
1028 }
1029 
1030 static uint64_t
1031 pvscsi_io_read(void *opaque, hwaddr addr, unsigned size)
1032 {
1033     PVSCSIState *s = opaque;
1034 
1035     switch (addr) {
1036     case PVSCSI_REG_OFFSET_INTR_STATUS:
1037         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_STATUS",
1038                              s->reg_interrupt_status);
1039         return s->reg_interrupt_status;
1040 
1041     case PVSCSI_REG_OFFSET_INTR_MASK:
1042         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_MASK",
1043                              s->reg_interrupt_status);
1044         return s->reg_interrupt_enabled;
1045 
1046     case PVSCSI_REG_OFFSET_COMMAND_STATUS:
1047         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_COMMAND_STATUS",
1048                              s->reg_interrupt_status);
1049         return s->reg_command_status;
1050 
1051     default:
1052         trace_pvscsi_io_read_unknown(addr, size);
1053         return 0;
1054     }
1055 }
1056 
1057 
1058 static void
1059 pvscsi_init_msi(PVSCSIState *s)
1060 {
1061     int res;
1062     PCIDevice *d = PCI_DEVICE(s);
1063 
1064     res = msi_init(d, PVSCSI_MSI_OFFSET(s), PVSCSI_MSIX_NUM_VECTORS,
1065                    PVSCSI_USE_64BIT, PVSCSI_PER_VECTOR_MASK, NULL);
1066     if (res < 0) {
1067         trace_pvscsi_init_msi_fail(res);
1068         s->msi_used = false;
1069     } else {
1070         s->msi_used = true;
1071     }
1072 }
1073 
1074 static void
1075 pvscsi_cleanup_msi(PVSCSIState *s)
1076 {
1077     PCIDevice *d = PCI_DEVICE(s);
1078 
1079     msi_uninit(d);
1080 }
1081 
1082 static const MemoryRegionOps pvscsi_ops = {
1083         .read = pvscsi_io_read,
1084         .write = pvscsi_io_write,
1085         .endianness = DEVICE_LITTLE_ENDIAN,
1086         .impl = {
1087                 .min_access_size = 4,
1088                 .max_access_size = 4,
1089         },
1090 };
1091 
1092 static const struct SCSIBusInfo pvscsi_scsi_info = {
1093         .tcq = true,
1094         .max_target = PVSCSI_MAX_DEVS,
1095         .max_channel = 0,
1096         .max_lun = 0,
1097 
1098         .get_sg_list = pvscsi_get_sg_list,
1099         .complete = pvscsi_command_complete,
1100         .cancel = pvscsi_request_cancelled,
1101 };
1102 
1103 static int
1104 pvscsi_init(PCIDevice *pci_dev)
1105 {
1106     PVSCSIState *s = PVSCSI(pci_dev);
1107 
1108     trace_pvscsi_state("init");
1109 
1110     /* PCI subsystem ID, subsystem vendor ID, revision */
1111     if (PVSCSI_USE_OLD_PCI_CONFIGURATION(s)) {
1112         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID, 0x1000);
1113     } else {
1114         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1115                      PCI_VENDOR_ID_VMWARE);
1116         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1117                      PCI_DEVICE_ID_VMWARE_PVSCSI);
1118         pci_config_set_revision(pci_dev->config, 0x2);
1119     }
1120 
1121     /* PCI latency timer = 255 */
1122     pci_dev->config[PCI_LATENCY_TIMER] = 0xff;
1123 
1124     /* Interrupt pin A */
1125     pci_config_set_interrupt_pin(pci_dev->config, 1);
1126 
1127     memory_region_init_io(&s->io_space, OBJECT(s), &pvscsi_ops, s,
1128                           "pvscsi-io", PVSCSI_MEM_SPACE_SIZE);
1129     pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->io_space);
1130 
1131     pvscsi_init_msi(s);
1132 
1133     if (pci_is_express(pci_dev) && pci_bus_is_express(pci_dev->bus)) {
1134         pcie_endpoint_cap_init(pci_dev, PVSCSI_EXP_EP_OFFSET);
1135     }
1136 
1137     s->completion_worker = qemu_bh_new(pvscsi_process_completion_queue, s);
1138     if (!s->completion_worker) {
1139         pvscsi_cleanup_msi(s);
1140         return -ENOMEM;
1141     }
1142 
1143     scsi_bus_new(&s->bus, sizeof(s->bus), DEVICE(pci_dev),
1144                  &pvscsi_scsi_info, NULL);
1145     /* override default SCSI bus hotplug-handler, with pvscsi's one */
1146     qbus_set_hotplug_handler(BUS(&s->bus), DEVICE(s), &error_abort);
1147     pvscsi_reset_state(s);
1148 
1149     return 0;
1150 }
1151 
1152 static void
1153 pvscsi_uninit(PCIDevice *pci_dev)
1154 {
1155     PVSCSIState *s = PVSCSI(pci_dev);
1156 
1157     trace_pvscsi_state("uninit");
1158     qemu_bh_delete(s->completion_worker);
1159 
1160     pvscsi_cleanup_msi(s);
1161 }
1162 
1163 static void
1164 pvscsi_reset(DeviceState *dev)
1165 {
1166     PCIDevice *d = PCI_DEVICE(dev);
1167     PVSCSIState *s = PVSCSI(d);
1168 
1169     trace_pvscsi_state("reset");
1170     pvscsi_reset_adapter(s);
1171 }
1172 
1173 static void
1174 pvscsi_pre_save(void *opaque)
1175 {
1176     PVSCSIState *s = (PVSCSIState *) opaque;
1177 
1178     trace_pvscsi_state("presave");
1179 
1180     assert(QTAILQ_EMPTY(&s->pending_queue));
1181     assert(QTAILQ_EMPTY(&s->completion_queue));
1182 }
1183 
1184 static int
1185 pvscsi_post_load(void *opaque, int version_id)
1186 {
1187     trace_pvscsi_state("postload");
1188     return 0;
1189 }
1190 
1191 static bool pvscsi_vmstate_need_pcie_device(void *opaque)
1192 {
1193     PVSCSIState *s = PVSCSI(opaque);
1194 
1195     return !(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE);
1196 }
1197 
1198 static bool pvscsi_vmstate_test_pci_device(void *opaque, int version_id)
1199 {
1200     return !pvscsi_vmstate_need_pcie_device(opaque);
1201 }
1202 
1203 static const VMStateDescription vmstate_pvscsi_pcie_device = {
1204     .name = "pvscsi/pcie",
1205     .needed = pvscsi_vmstate_need_pcie_device,
1206     .fields = (VMStateField[]) {
1207         VMSTATE_PCIE_DEVICE(parent_obj, PVSCSIState),
1208         VMSTATE_END_OF_LIST()
1209     }
1210 };
1211 
1212 static const VMStateDescription vmstate_pvscsi = {
1213     .name = "pvscsi",
1214     .version_id = 0,
1215     .minimum_version_id = 0,
1216     .pre_save = pvscsi_pre_save,
1217     .post_load = pvscsi_post_load,
1218     .fields = (VMStateField[]) {
1219         VMSTATE_STRUCT_TEST(parent_obj, PVSCSIState,
1220                             pvscsi_vmstate_test_pci_device, 0,
1221                             vmstate_pci_device, PCIDevice),
1222         VMSTATE_UINT8(msi_used, PVSCSIState),
1223         VMSTATE_UINT32(resetting, PVSCSIState),
1224         VMSTATE_UINT64(reg_interrupt_status, PVSCSIState),
1225         VMSTATE_UINT64(reg_interrupt_enabled, PVSCSIState),
1226         VMSTATE_UINT64(reg_command_status, PVSCSIState),
1227         VMSTATE_UINT64(curr_cmd, PVSCSIState),
1228         VMSTATE_UINT32(curr_cmd_data_cntr, PVSCSIState),
1229         VMSTATE_UINT32_ARRAY(curr_cmd_data, PVSCSIState,
1230                              ARRAY_SIZE(((PVSCSIState *)NULL)->curr_cmd_data)),
1231         VMSTATE_UINT8(rings_info_valid, PVSCSIState),
1232         VMSTATE_UINT8(msg_ring_info_valid, PVSCSIState),
1233         VMSTATE_UINT8(use_msg, PVSCSIState),
1234 
1235         VMSTATE_UINT64(rings.rs_pa, PVSCSIState),
1236         VMSTATE_UINT32(rings.txr_len_mask, PVSCSIState),
1237         VMSTATE_UINT32(rings.rxr_len_mask, PVSCSIState),
1238         VMSTATE_UINT64_ARRAY(rings.req_ring_pages_pa, PVSCSIState,
1239                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1240         VMSTATE_UINT64_ARRAY(rings.cmp_ring_pages_pa, PVSCSIState,
1241                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1242         VMSTATE_UINT64(rings.consumed_ptr, PVSCSIState),
1243         VMSTATE_UINT64(rings.filled_cmp_ptr, PVSCSIState),
1244 
1245         VMSTATE_END_OF_LIST()
1246     },
1247     .subsections = (const VMStateDescription*[]) {
1248         &vmstate_pvscsi_pcie_device,
1249         NULL
1250     }
1251 };
1252 
1253 static Property pvscsi_properties[] = {
1254     DEFINE_PROP_UINT8("use_msg", PVSCSIState, use_msg, 1),
1255     DEFINE_PROP_BIT("x-old-pci-configuration", PVSCSIState, compat_flags,
1256                     PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT, false),
1257     DEFINE_PROP_BIT("x-disable-pcie", PVSCSIState, compat_flags,
1258                     PVSCSI_COMPAT_DISABLE_PCIE_BIT, false),
1259     DEFINE_PROP_END_OF_LIST(),
1260 };
1261 
1262 static void pvscsi_realize(DeviceState *qdev, Error **errp)
1263 {
1264     PVSCSIClass *pvs_c = PVSCSI_DEVICE_GET_CLASS(qdev);
1265     PCIDevice *pci_dev = PCI_DEVICE(qdev);
1266     PVSCSIState *s = PVSCSI(qdev);
1267 
1268     if (!(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE)) {
1269         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1270     }
1271 
1272     pvs_c->parent_dc_realize(qdev, errp);
1273 }
1274 
1275 static void pvscsi_class_init(ObjectClass *klass, void *data)
1276 {
1277     DeviceClass *dc = DEVICE_CLASS(klass);
1278     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1279     PVSCSIClass *pvs_k = PVSCSI_DEVICE_CLASS(klass);
1280     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
1281 
1282     k->init = pvscsi_init;
1283     k->exit = pvscsi_uninit;
1284     k->vendor_id = PCI_VENDOR_ID_VMWARE;
1285     k->device_id = PCI_DEVICE_ID_VMWARE_PVSCSI;
1286     k->class_id = PCI_CLASS_STORAGE_SCSI;
1287     k->subsystem_id = 0x1000;
1288     pvs_k->parent_dc_realize = dc->realize;
1289     dc->realize = pvscsi_realize;
1290     dc->reset = pvscsi_reset;
1291     dc->vmsd = &vmstate_pvscsi;
1292     dc->props = pvscsi_properties;
1293     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1294     hc->unplug = pvscsi_hot_unplug;
1295     hc->plug = pvscsi_hotplug;
1296 }
1297 
1298 static const TypeInfo pvscsi_info = {
1299     .name          = TYPE_PVSCSI,
1300     .parent        = TYPE_PCI_DEVICE,
1301     .class_size    = sizeof(PVSCSIClass),
1302     .instance_size = sizeof(PVSCSIState),
1303     .class_init    = pvscsi_class_init,
1304     .interfaces = (InterfaceInfo[]) {
1305         { TYPE_HOTPLUG_HANDLER },
1306         { }
1307     }
1308 };
1309 
1310 static void
1311 pvscsi_register_types(void)
1312 {
1313     type_register_static(&pvscsi_info);
1314 }
1315 
1316 type_init(pvscsi_register_types);
1317