xref: /openbmc/qemu/hw/scsi/vmw_pvscsi.c (revision 3a2e982d)
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 "qemu/main-loop.h"
31 #include "qemu/module.h"
32 #include "hw/scsi/scsi.h"
33 #include "migration/vmstate.h"
34 #include "scsi/constants.h"
35 #include "hw/pci/msi.h"
36 #include "hw/qdev-properties.h"
37 #include "vmw_pvscsi.h"
38 #include "trace.h"
39 #include "qom/object.h"
40 
41 
42 #define PVSCSI_USE_64BIT         (true)
43 #define PVSCSI_PER_VECTOR_MASK   (false)
44 
45 #define PVSCSI_MAX_DEVS                   (64)
46 #define PVSCSI_MSIX_NUM_VECTORS           (1)
47 
48 #define PVSCSI_MAX_SG_ELEM                2048
49 
50 #define PVSCSI_MAX_CMD_DATA_WORDS \
51     (sizeof(PVSCSICmdDescSetupRings)/sizeof(uint32_t))
52 
53 #define RS_GET_FIELD(m, field) \
54     (ldl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
55                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field)))
56 #define RS_SET_FIELD(m, field, val) \
57     (stl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
58                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field), val))
59 
60 struct PVSCSIClass {
61     PCIDeviceClass parent_class;
62     DeviceRealize parent_dc_realize;
63 };
64 
65 #define TYPE_PVSCSI "pvscsi"
66 OBJECT_DECLARE_TYPE(PVSCSIState, PVSCSIClass, PVSCSI)
67 
68 
69 /* Compatibility flags for migration */
70 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT 0
71 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION \
72     (1 << PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT)
73 #define PVSCSI_COMPAT_DISABLE_PCIE_BIT 1
74 #define PVSCSI_COMPAT_DISABLE_PCIE \
75     (1 << PVSCSI_COMPAT_DISABLE_PCIE_BIT)
76 
77 #define PVSCSI_USE_OLD_PCI_CONFIGURATION(s) \
78     ((s)->compat_flags & PVSCSI_COMPAT_OLD_PCI_CONFIGURATION)
79 #define PVSCSI_MSI_OFFSET(s) \
80     (PVSCSI_USE_OLD_PCI_CONFIGURATION(s) ? 0x50 : 0x7c)
81 #define PVSCSI_EXP_EP_OFFSET (0x40)
82 
83 typedef struct PVSCSIRingInfo {
84     uint64_t            rs_pa;
85     uint32_t            txr_len_mask;
86     uint32_t            rxr_len_mask;
87     uint32_t            msg_len_mask;
88     uint64_t            req_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
89     uint64_t            cmp_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
90     uint64_t            msg_ring_pages_pa[PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES];
91     uint64_t            consumed_ptr;
92     uint64_t            filled_cmp_ptr;
93     uint64_t            filled_msg_ptr;
94 } PVSCSIRingInfo;
95 
96 typedef struct PVSCSISGState {
97     hwaddr elemAddr;
98     hwaddr dataAddr;
99     uint32_t resid;
100 } PVSCSISGState;
101 
102 typedef QTAILQ_HEAD(, PVSCSIRequest) PVSCSIRequestList;
103 
104 struct PVSCSIState {
105     PCIDevice parent_obj;
106     MemoryRegion io_space;
107     SCSIBus bus;
108     QEMUBH *completion_worker;
109     PVSCSIRequestList pending_queue;
110     PVSCSIRequestList completion_queue;
111 
112     uint64_t reg_interrupt_status;        /* Interrupt status register value */
113     uint64_t reg_interrupt_enabled;       /* Interrupt mask register value   */
114     uint64_t reg_command_status;          /* Command status register value   */
115 
116     /* Command data adoption mechanism */
117     uint64_t curr_cmd;                   /* Last command arrived             */
118     uint32_t curr_cmd_data_cntr;         /* Amount of data for last command  */
119 
120     /* Collector for current command data */
121     uint32_t curr_cmd_data[PVSCSI_MAX_CMD_DATA_WORDS];
122 
123     uint8_t rings_info_valid;            /* Whether data rings initialized   */
124     uint8_t msg_ring_info_valid;         /* Whether message ring initialized */
125     uint8_t use_msg;                     /* Whether to use message ring      */
126 
127     uint8_t msi_used;                    /* For migration compatibility      */
128     PVSCSIRingInfo rings;                /* Data transfer rings manager      */
129     uint32_t resetting;                  /* Reset in progress                */
130 
131     uint32_t compat_flags;
132 };
133 
134 typedef struct PVSCSIRequest {
135     SCSIRequest *sreq;
136     PVSCSIState *dev;
137     uint8_t sense_key;
138     uint8_t completed;
139     int lun;
140     QEMUSGList sgl;
141     PVSCSISGState sg;
142     struct PVSCSIRingReqDesc req;
143     struct PVSCSIRingCmpDesc cmp;
144     QTAILQ_ENTRY(PVSCSIRequest) next;
145 } PVSCSIRequest;
146 
147 /* Integer binary logarithm */
148 static int
149 pvscsi_log2(uint32_t input)
150 {
151     int log = 0;
152     assert(input > 0);
153     while (input >> ++log) {
154     }
155     return log;
156 }
157 
158 static void
159 pvscsi_ring_init_data(PVSCSIRingInfo *m, PVSCSICmdDescSetupRings *ri)
160 {
161     int i;
162     uint32_t txr_len_log2, rxr_len_log2;
163     uint32_t req_ring_size, cmp_ring_size;
164     m->rs_pa = ri->ringsStatePPN << VMW_PAGE_SHIFT;
165 
166     req_ring_size = ri->reqRingNumPages * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
167     cmp_ring_size = ri->cmpRingNumPages * PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
168     txr_len_log2 = pvscsi_log2(req_ring_size - 1);
169     rxr_len_log2 = pvscsi_log2(cmp_ring_size - 1);
170 
171     m->txr_len_mask = MASK(txr_len_log2);
172     m->rxr_len_mask = MASK(rxr_len_log2);
173 
174     m->consumed_ptr = 0;
175     m->filled_cmp_ptr = 0;
176 
177     for (i = 0; i < ri->reqRingNumPages; i++) {
178         m->req_ring_pages_pa[i] = ri->reqRingPPNs[i] << VMW_PAGE_SHIFT;
179     }
180 
181     for (i = 0; i < ri->cmpRingNumPages; i++) {
182         m->cmp_ring_pages_pa[i] = ri->cmpRingPPNs[i] << VMW_PAGE_SHIFT;
183     }
184 
185     RS_SET_FIELD(m, reqProdIdx, 0);
186     RS_SET_FIELD(m, reqConsIdx, 0);
187     RS_SET_FIELD(m, reqNumEntriesLog2, txr_len_log2);
188 
189     RS_SET_FIELD(m, cmpProdIdx, 0);
190     RS_SET_FIELD(m, cmpConsIdx, 0);
191     RS_SET_FIELD(m, cmpNumEntriesLog2, rxr_len_log2);
192 
193     trace_pvscsi_ring_init_data(txr_len_log2, rxr_len_log2);
194 
195     /* Flush ring state page changes */
196     smp_wmb();
197 }
198 
199 static int
200 pvscsi_ring_init_msg(PVSCSIRingInfo *m, PVSCSICmdDescSetupMsgRing *ri)
201 {
202     int i;
203     uint32_t len_log2;
204     uint32_t ring_size;
205 
206     if (!ri->numPages || ri->numPages > PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES) {
207         return -1;
208     }
209     ring_size = ri->numPages * PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
210     len_log2 = pvscsi_log2(ring_size - 1);
211 
212     m->msg_len_mask = MASK(len_log2);
213 
214     m->filled_msg_ptr = 0;
215 
216     for (i = 0; i < ri->numPages; i++) {
217         m->msg_ring_pages_pa[i] = ri->ringPPNs[i] << VMW_PAGE_SHIFT;
218     }
219 
220     RS_SET_FIELD(m, msgProdIdx, 0);
221     RS_SET_FIELD(m, msgConsIdx, 0);
222     RS_SET_FIELD(m, msgNumEntriesLog2, len_log2);
223 
224     trace_pvscsi_ring_init_msg(len_log2);
225 
226     /* Flush ring state page changes */
227     smp_wmb();
228 
229     return 0;
230 }
231 
232 static void
233 pvscsi_ring_cleanup(PVSCSIRingInfo *mgr)
234 {
235     mgr->rs_pa = 0;
236     mgr->txr_len_mask = 0;
237     mgr->rxr_len_mask = 0;
238     mgr->msg_len_mask = 0;
239     mgr->consumed_ptr = 0;
240     mgr->filled_cmp_ptr = 0;
241     mgr->filled_msg_ptr = 0;
242     memset(mgr->req_ring_pages_pa, 0, sizeof(mgr->req_ring_pages_pa));
243     memset(mgr->cmp_ring_pages_pa, 0, sizeof(mgr->cmp_ring_pages_pa));
244     memset(mgr->msg_ring_pages_pa, 0, sizeof(mgr->msg_ring_pages_pa));
245 }
246 
247 static hwaddr
248 pvscsi_ring_pop_req_descr(PVSCSIRingInfo *mgr)
249 {
250     uint32_t ready_ptr = RS_GET_FIELD(mgr, reqProdIdx);
251     uint32_t ring_size = PVSCSI_MAX_NUM_PAGES_REQ_RING
252                             * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
253 
254     if (ready_ptr != mgr->consumed_ptr
255         && ready_ptr - mgr->consumed_ptr < ring_size) {
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, cmp_desc, sizeof(*cmp_desc));
405 }
406 
407 static void
408 pvscsi_msg_ring_put(PVSCSIState *s, struct PVSCSIRingMsgDesc *msg_desc)
409 {
410     hwaddr msg_descr_pa;
411 
412     msg_descr_pa = pvscsi_ring_pop_msg_descr(&s->rings);
413     trace_pvscsi_msg_ring_put(msg_descr_pa);
414     cpu_physical_memory_write(msg_descr_pa, msg_desc, sizeof(*msg_desc));
415 }
416 
417 static void
418 pvscsi_process_completion_queue(void *opaque)
419 {
420     PVSCSIState *s = opaque;
421     PVSCSIRequest *pvscsi_req;
422     bool has_completed = false;
423 
424     while (!QTAILQ_EMPTY(&s->completion_queue)) {
425         pvscsi_req = QTAILQ_FIRST(&s->completion_queue);
426         QTAILQ_REMOVE(&s->completion_queue, pvscsi_req, next);
427         pvscsi_cmp_ring_put(s, &pvscsi_req->cmp);
428         g_free(pvscsi_req);
429         has_completed = true;
430     }
431 
432     if (has_completed) {
433         pvscsi_ring_flush_cmp(&s->rings);
434         pvscsi_raise_completion_interrupt(s);
435     }
436 }
437 
438 static void
439 pvscsi_reset_adapter(PVSCSIState *s)
440 {
441     s->resetting++;
442     qbus_reset_all(BUS(&s->bus));
443     s->resetting--;
444     pvscsi_process_completion_queue(s);
445     assert(QTAILQ_EMPTY(&s->pending_queue));
446     pvscsi_reset_state(s);
447 }
448 
449 static void
450 pvscsi_schedule_completion_processing(PVSCSIState *s)
451 {
452     /* Try putting more complete requests on the ring. */
453     if (!QTAILQ_EMPTY(&s->completion_queue)) {
454         qemu_bh_schedule(s->completion_worker);
455     }
456 }
457 
458 static void
459 pvscsi_complete_request(PVSCSIState *s, PVSCSIRequest *r)
460 {
461     assert(!r->completed);
462 
463     trace_pvscsi_complete_request(r->cmp.context, r->cmp.dataLen,
464                                   r->sense_key);
465     if (r->sreq != NULL) {
466         scsi_req_unref(r->sreq);
467         r->sreq = NULL;
468     }
469     r->completed = 1;
470     QTAILQ_REMOVE(&s->pending_queue, r, next);
471     QTAILQ_INSERT_TAIL(&s->completion_queue, r, next);
472     pvscsi_schedule_completion_processing(s);
473 }
474 
475 static QEMUSGList *pvscsi_get_sg_list(SCSIRequest *r)
476 {
477     PVSCSIRequest *req = r->hba_private;
478 
479     trace_pvscsi_get_sg_list(req->sgl.nsg, req->sgl.size);
480 
481     return &req->sgl;
482 }
483 
484 static void
485 pvscsi_get_next_sg_elem(PVSCSISGState *sg)
486 {
487     struct PVSCSISGElement elem;
488 
489     cpu_physical_memory_read(sg->elemAddr, &elem, sizeof(elem));
490     if ((elem.flags & ~PVSCSI_KNOWN_FLAGS) != 0) {
491         /*
492             * There is PVSCSI_SGE_FLAG_CHAIN_ELEMENT flag described in
493             * header file but its value is unknown. This flag requires
494             * additional processing, so we put warning here to catch it
495             * some day and make proper implementation
496             */
497         trace_pvscsi_get_next_sg_elem(elem.flags);
498     }
499 
500     sg->elemAddr += sizeof(elem);
501     sg->dataAddr = elem.addr;
502     sg->resid = elem.length;
503 }
504 
505 static void
506 pvscsi_write_sense(PVSCSIRequest *r, uint8_t *sense, int len)
507 {
508     r->cmp.senseLen = MIN(r->req.senseLen, len);
509     r->sense_key = sense[(sense[0] & 2) ? 1 : 2];
510     cpu_physical_memory_write(r->req.senseAddr, sense, r->cmp.senseLen);
511 }
512 
513 static void
514 pvscsi_command_complete(SCSIRequest *req, size_t resid)
515 {
516     PVSCSIRequest *pvscsi_req = req->hba_private;
517     PVSCSIState *s;
518 
519     if (!pvscsi_req) {
520         trace_pvscsi_command_complete_not_found(req->tag);
521         return;
522     }
523     s = pvscsi_req->dev;
524 
525     if (resid) {
526         /* Short transfer.  */
527         trace_pvscsi_command_complete_data_run();
528         pvscsi_req->cmp.hostStatus = BTSTAT_DATARUN;
529     }
530 
531     pvscsi_req->cmp.scsiStatus = req->status;
532     if (pvscsi_req->cmp.scsiStatus == CHECK_CONDITION) {
533         uint8_t sense[SCSI_SENSE_BUF_SIZE];
534         int sense_len =
535             scsi_req_get_sense(pvscsi_req->sreq, sense, sizeof(sense));
536 
537         trace_pvscsi_command_complete_sense_len(sense_len);
538         pvscsi_write_sense(pvscsi_req, sense, sense_len);
539     }
540     qemu_sglist_destroy(&pvscsi_req->sgl);
541     pvscsi_complete_request(s, pvscsi_req);
542 }
543 
544 static void
545 pvscsi_send_msg(PVSCSIState *s, SCSIDevice *dev, uint32_t msg_type)
546 {
547     if (s->msg_ring_info_valid && pvscsi_ring_msg_has_room(&s->rings)) {
548         PVSCSIMsgDescDevStatusChanged msg = {0};
549 
550         msg.type = msg_type;
551         msg.bus = dev->channel;
552         msg.target = dev->id;
553         msg.lun[1] = dev->lun;
554 
555         pvscsi_msg_ring_put(s, (PVSCSIRingMsgDesc *)&msg);
556         pvscsi_ring_flush_msg(&s->rings);
557         pvscsi_raise_message_interrupt(s);
558     }
559 }
560 
561 static void
562 pvscsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
563 {
564     PVSCSIState *s = PVSCSI(hotplug_dev);
565 
566     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_ADDED);
567 }
568 
569 static void
570 pvscsi_hot_unplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
571 {
572     PVSCSIState *s = PVSCSI(hotplug_dev);
573 
574     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_REMOVED);
575     qdev_simple_device_unplug_cb(hotplug_dev, dev, errp);
576 }
577 
578 static void
579 pvscsi_request_cancelled(SCSIRequest *req)
580 {
581     PVSCSIRequest *pvscsi_req = req->hba_private;
582     PVSCSIState *s = pvscsi_req->dev;
583 
584     if (pvscsi_req->completed) {
585         return;
586     }
587 
588    if (pvscsi_req->dev->resetting) {
589        pvscsi_req->cmp.hostStatus = BTSTAT_BUSRESET;
590     } else {
591        pvscsi_req->cmp.hostStatus = BTSTAT_ABORTQUEUE;
592     }
593 
594     pvscsi_complete_request(s, pvscsi_req);
595 }
596 
597 static SCSIDevice*
598 pvscsi_device_find(PVSCSIState *s, int channel, int target,
599                    uint8_t *requested_lun, uint8_t *target_lun)
600 {
601     if (requested_lun[0] || requested_lun[2] || requested_lun[3] ||
602         requested_lun[4] || requested_lun[5] || requested_lun[6] ||
603         requested_lun[7] || (target > PVSCSI_MAX_DEVS)) {
604         return NULL;
605     } else {
606         *target_lun = requested_lun[1];
607         return scsi_device_find(&s->bus, channel, target, *target_lun);
608     }
609 }
610 
611 static PVSCSIRequest *
612 pvscsi_queue_pending_descriptor(PVSCSIState *s, SCSIDevice **d,
613                                 struct PVSCSIRingReqDesc *descr)
614 {
615     PVSCSIRequest *pvscsi_req;
616     uint8_t lun;
617 
618     pvscsi_req = g_malloc0(sizeof(*pvscsi_req));
619     pvscsi_req->dev = s;
620     pvscsi_req->req = *descr;
621     pvscsi_req->cmp.context = pvscsi_req->req.context;
622     QTAILQ_INSERT_TAIL(&s->pending_queue, pvscsi_req, next);
623 
624     *d = pvscsi_device_find(s, descr->bus, descr->target, descr->lun, &lun);
625     if (*d) {
626         pvscsi_req->lun = lun;
627     }
628 
629     return pvscsi_req;
630 }
631 
632 static void
633 pvscsi_convert_sglist(PVSCSIRequest *r)
634 {
635     uint32_t chunk_size, elmcnt = 0;
636     uint64_t data_length = r->req.dataLen;
637     PVSCSISGState sg = r->sg;
638     while (data_length && elmcnt < PVSCSI_MAX_SG_ELEM) {
639         while (!sg.resid && elmcnt++ < PVSCSI_MAX_SG_ELEM) {
640             pvscsi_get_next_sg_elem(&sg);
641             trace_pvscsi_convert_sglist(r->req.context, r->sg.dataAddr,
642                                         r->sg.resid);
643         }
644         chunk_size = MIN(data_length, sg.resid);
645         if (chunk_size) {
646             qemu_sglist_add(&r->sgl, sg.dataAddr, chunk_size);
647         }
648 
649         sg.dataAddr += chunk_size;
650         data_length -= chunk_size;
651         sg.resid -= chunk_size;
652     }
653 }
654 
655 static void
656 pvscsi_build_sglist(PVSCSIState *s, PVSCSIRequest *r)
657 {
658     PCIDevice *d = PCI_DEVICE(s);
659 
660     pci_dma_sglist_init(&r->sgl, d, 1);
661     if (r->req.flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
662         pvscsi_convert_sglist(r);
663     } else {
664         qemu_sglist_add(&r->sgl, r->req.dataAddr, r->req.dataLen);
665     }
666 }
667 
668 static void
669 pvscsi_process_request_descriptor(PVSCSIState *s,
670                                   struct PVSCSIRingReqDesc *descr)
671 {
672     SCSIDevice *d;
673     PVSCSIRequest *r = pvscsi_queue_pending_descriptor(s, &d, descr);
674     int64_t n;
675 
676     trace_pvscsi_process_req_descr(descr->cdb[0], descr->context);
677 
678     if (!d) {
679         r->cmp.hostStatus = BTSTAT_SELTIMEO;
680         trace_pvscsi_process_req_descr_unknown_device();
681         pvscsi_complete_request(s, r);
682         return;
683     }
684 
685     if (descr->flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
686         r->sg.elemAddr = descr->dataAddr;
687     }
688 
689     r->sreq = scsi_req_new(d, descr->context, r->lun, descr->cdb, r);
690     if (r->sreq->cmd.mode == SCSI_XFER_FROM_DEV &&
691         (descr->flags & PVSCSI_FLAG_CMD_DIR_TODEVICE)) {
692         r->cmp.hostStatus = BTSTAT_BADMSG;
693         trace_pvscsi_process_req_descr_invalid_dir();
694         scsi_req_cancel(r->sreq);
695         return;
696     }
697     if (r->sreq->cmd.mode == SCSI_XFER_TO_DEV &&
698         (descr->flags & PVSCSI_FLAG_CMD_DIR_TOHOST)) {
699         r->cmp.hostStatus = BTSTAT_BADMSG;
700         trace_pvscsi_process_req_descr_invalid_dir();
701         scsi_req_cancel(r->sreq);
702         return;
703     }
704 
705     pvscsi_build_sglist(s, r);
706     n = scsi_req_enqueue(r->sreq);
707 
708     if (n) {
709         scsi_req_continue(r->sreq);
710     }
711 }
712 
713 static void
714 pvscsi_process_io(PVSCSIState *s)
715 {
716     PVSCSIRingReqDesc descr;
717     hwaddr next_descr_pa;
718 
719     if (!s->rings_info_valid) {
720         return;
721     }
722 
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->cmpRingPPNs[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     if (!rc->reqRingNumPages
783         || rc->reqRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES
784         || !rc->cmpRingNumPages
785         || rc->cmpRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES) {
786         return PVSCSI_COMMAND_PROCESSING_FAILED;
787     }
788 
789     pvscsi_dbg_dump_tx_rings_config(rc);
790     pvscsi_ring_init_data(&s->rings, rc);
791 
792     s->rings_info_valid = TRUE;
793     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
794 }
795 
796 static uint64_t
797 pvscsi_on_cmd_abort(PVSCSIState *s)
798 {
799     PVSCSICmdDescAbortCmd *cmd = (PVSCSICmdDescAbortCmd *) s->curr_cmd_data;
800     PVSCSIRequest *r, *next;
801 
802     trace_pvscsi_on_cmd_abort(cmd->context, cmd->target);
803 
804     QTAILQ_FOREACH_SAFE(r, &s->pending_queue, next, next) {
805         if (r->req.context == cmd->context) {
806             break;
807         }
808     }
809     if (r) {
810         assert(!r->completed);
811         r->cmp.hostStatus = BTSTAT_ABORTQUEUE;
812         scsi_req_cancel(r->sreq);
813     }
814 
815     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
816 }
817 
818 static uint64_t
819 pvscsi_on_cmd_unknown(PVSCSIState *s)
820 {
821     trace_pvscsi_on_cmd_unknown_data(s->curr_cmd_data[0]);
822     return PVSCSI_COMMAND_PROCESSING_FAILED;
823 }
824 
825 static uint64_t
826 pvscsi_on_cmd_reset_device(PVSCSIState *s)
827 {
828     uint8_t target_lun = 0;
829     struct PVSCSICmdDescResetDevice *cmd =
830         (struct PVSCSICmdDescResetDevice *) s->curr_cmd_data;
831     SCSIDevice *sdev;
832 
833     sdev = pvscsi_device_find(s, 0, cmd->target, cmd->lun, &target_lun);
834 
835     trace_pvscsi_on_cmd_reset_dev(cmd->target, (int) target_lun, sdev);
836 
837     if (sdev != NULL) {
838         s->resetting++;
839         device_legacy_reset(&sdev->qdev);
840         s->resetting--;
841         return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
842     }
843 
844     return PVSCSI_COMMAND_PROCESSING_FAILED;
845 }
846 
847 static uint64_t
848 pvscsi_on_cmd_reset_bus(PVSCSIState *s)
849 {
850     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_RESET_BUS");
851 
852     s->resetting++;
853     qbus_reset_all(BUS(&s->bus));
854     s->resetting--;
855     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
856 }
857 
858 static uint64_t
859 pvscsi_on_cmd_setup_msg_ring(PVSCSIState *s)
860 {
861     PVSCSICmdDescSetupMsgRing *rc =
862         (PVSCSICmdDescSetupMsgRing *) s->curr_cmd_data;
863 
864     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_MSG_RING");
865 
866     if (!s->use_msg) {
867         return PVSCSI_COMMAND_PROCESSING_FAILED;
868     }
869 
870     if (s->rings_info_valid) {
871         if (pvscsi_ring_init_msg(&s->rings, rc) < 0) {
872             return PVSCSI_COMMAND_PROCESSING_FAILED;
873         }
874         s->msg_ring_info_valid = TRUE;
875     }
876     return sizeof(PVSCSICmdDescSetupMsgRing) / sizeof(uint32_t);
877 }
878 
879 static uint64_t
880 pvscsi_on_cmd_adapter_reset(PVSCSIState *s)
881 {
882     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_ADAPTER_RESET");
883 
884     pvscsi_reset_adapter(s);
885     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
886 }
887 
888 static const struct {
889     int       data_size;
890     uint64_t  (*handler_fn)(PVSCSIState *s);
891 } pvscsi_commands[] = {
892     [PVSCSI_CMD_FIRST] = {
893         .data_size = 0,
894         .handler_fn = pvscsi_on_cmd_unknown,
895     },
896 
897     /* Not implemented, data size defined based on what arrives on windows */
898     [PVSCSI_CMD_CONFIG] = {
899         .data_size = 6 * sizeof(uint32_t),
900         .handler_fn = pvscsi_on_cmd_config,
901     },
902 
903     /* Command not implemented, data size is unknown */
904     [PVSCSI_CMD_ISSUE_SCSI] = {
905         .data_size = 0,
906         .handler_fn = pvscsi_on_issue_scsi,
907     },
908 
909     /* Command not implemented, data size is unknown */
910     [PVSCSI_CMD_DEVICE_UNPLUG] = {
911         .data_size = 0,
912         .handler_fn = pvscsi_on_cmd_unplug,
913     },
914 
915     [PVSCSI_CMD_SETUP_RINGS] = {
916         .data_size = sizeof(PVSCSICmdDescSetupRings),
917         .handler_fn = pvscsi_on_cmd_setup_rings,
918     },
919 
920     [PVSCSI_CMD_RESET_DEVICE] = {
921         .data_size = sizeof(struct PVSCSICmdDescResetDevice),
922         .handler_fn = pvscsi_on_cmd_reset_device,
923     },
924 
925     [PVSCSI_CMD_RESET_BUS] = {
926         .data_size = 0,
927         .handler_fn = pvscsi_on_cmd_reset_bus,
928     },
929 
930     [PVSCSI_CMD_SETUP_MSG_RING] = {
931         .data_size = sizeof(PVSCSICmdDescSetupMsgRing),
932         .handler_fn = pvscsi_on_cmd_setup_msg_ring,
933     },
934 
935     [PVSCSI_CMD_ADAPTER_RESET] = {
936         .data_size = 0,
937         .handler_fn = pvscsi_on_cmd_adapter_reset,
938     },
939 
940     [PVSCSI_CMD_ABORT_CMD] = {
941         .data_size = sizeof(struct PVSCSICmdDescAbortCmd),
942         .handler_fn = pvscsi_on_cmd_abort,
943     },
944 };
945 
946 static void
947 pvscsi_do_command_processing(PVSCSIState *s)
948 {
949     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
950 
951     assert(s->curr_cmd < PVSCSI_CMD_LAST);
952     if (bytes_arrived >= pvscsi_commands[s->curr_cmd].data_size) {
953         s->reg_command_status = pvscsi_commands[s->curr_cmd].handler_fn(s);
954         s->curr_cmd = PVSCSI_CMD_FIRST;
955         s->curr_cmd_data_cntr   = 0;
956     }
957 }
958 
959 static void
960 pvscsi_on_command_data(PVSCSIState *s, uint32_t value)
961 {
962     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
963 
964     assert(bytes_arrived < sizeof(s->curr_cmd_data));
965     s->curr_cmd_data[s->curr_cmd_data_cntr++] = value;
966 
967     pvscsi_do_command_processing(s);
968 }
969 
970 static void
971 pvscsi_on_command(PVSCSIState *s, uint64_t cmd_id)
972 {
973     if ((cmd_id > PVSCSI_CMD_FIRST) && (cmd_id < PVSCSI_CMD_LAST)) {
974         s->curr_cmd = cmd_id;
975     } else {
976         s->curr_cmd = PVSCSI_CMD_FIRST;
977         trace_pvscsi_on_cmd_unknown(cmd_id);
978     }
979 
980     s->curr_cmd_data_cntr = 0;
981     s->reg_command_status = PVSCSI_COMMAND_NOT_ENOUGH_DATA;
982 
983     pvscsi_do_command_processing(s);
984 }
985 
986 static void
987 pvscsi_io_write(void *opaque, hwaddr addr,
988                 uint64_t val, unsigned size)
989 {
990     PVSCSIState *s = opaque;
991 
992     switch (addr) {
993     case PVSCSI_REG_OFFSET_COMMAND:
994         pvscsi_on_command(s, val);
995         break;
996 
997     case PVSCSI_REG_OFFSET_COMMAND_DATA:
998         pvscsi_on_command_data(s, (uint32_t) val);
999         break;
1000 
1001     case PVSCSI_REG_OFFSET_INTR_STATUS:
1002         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_STATUS", val);
1003         s->reg_interrupt_status &= ~val;
1004         pvscsi_update_irq_status(s);
1005         pvscsi_schedule_completion_processing(s);
1006         break;
1007 
1008     case PVSCSI_REG_OFFSET_INTR_MASK:
1009         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_MASK", val);
1010         s->reg_interrupt_enabled = val;
1011         pvscsi_update_irq_status(s);
1012         break;
1013 
1014     case PVSCSI_REG_OFFSET_KICK_NON_RW_IO:
1015         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_NON_RW_IO", val);
1016         pvscsi_process_io(s);
1017         break;
1018 
1019     case PVSCSI_REG_OFFSET_KICK_RW_IO:
1020         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_RW_IO", val);
1021         pvscsi_process_io(s);
1022         break;
1023 
1024     case PVSCSI_REG_OFFSET_DEBUG:
1025         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_DEBUG", val);
1026         break;
1027 
1028     default:
1029         trace_pvscsi_io_write_unknown(addr, size, val);
1030         break;
1031     }
1032 
1033 }
1034 
1035 static uint64_t
1036 pvscsi_io_read(void *opaque, hwaddr addr, unsigned size)
1037 {
1038     PVSCSIState *s = opaque;
1039 
1040     switch (addr) {
1041     case PVSCSI_REG_OFFSET_INTR_STATUS:
1042         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_STATUS",
1043                              s->reg_interrupt_status);
1044         return s->reg_interrupt_status;
1045 
1046     case PVSCSI_REG_OFFSET_INTR_MASK:
1047         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_MASK",
1048                              s->reg_interrupt_status);
1049         return s->reg_interrupt_enabled;
1050 
1051     case PVSCSI_REG_OFFSET_COMMAND_STATUS:
1052         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_COMMAND_STATUS",
1053                              s->reg_interrupt_status);
1054         return s->reg_command_status;
1055 
1056     default:
1057         trace_pvscsi_io_read_unknown(addr, size);
1058         return 0;
1059     }
1060 }
1061 
1062 
1063 static void
1064 pvscsi_init_msi(PVSCSIState *s)
1065 {
1066     int res;
1067     PCIDevice *d = PCI_DEVICE(s);
1068 
1069     res = msi_init(d, PVSCSI_MSI_OFFSET(s), PVSCSI_MSIX_NUM_VECTORS,
1070                    PVSCSI_USE_64BIT, PVSCSI_PER_VECTOR_MASK, NULL);
1071     if (res < 0) {
1072         trace_pvscsi_init_msi_fail(res);
1073         s->msi_used = false;
1074     } else {
1075         s->msi_used = true;
1076     }
1077 }
1078 
1079 static void
1080 pvscsi_cleanup_msi(PVSCSIState *s)
1081 {
1082     PCIDevice *d = PCI_DEVICE(s);
1083 
1084     msi_uninit(d);
1085 }
1086 
1087 static const MemoryRegionOps pvscsi_ops = {
1088         .read = pvscsi_io_read,
1089         .write = pvscsi_io_write,
1090         .endianness = DEVICE_LITTLE_ENDIAN,
1091         .impl = {
1092                 .min_access_size = 4,
1093                 .max_access_size = 4,
1094         },
1095 };
1096 
1097 static const struct SCSIBusInfo pvscsi_scsi_info = {
1098         .tcq = true,
1099         .max_target = PVSCSI_MAX_DEVS,
1100         .max_channel = 0,
1101         .max_lun = 0,
1102 
1103         .get_sg_list = pvscsi_get_sg_list,
1104         .complete = pvscsi_command_complete,
1105         .cancel = pvscsi_request_cancelled,
1106 };
1107 
1108 static void
1109 pvscsi_realizefn(PCIDevice *pci_dev, Error **errp)
1110 {
1111     PVSCSIState *s = PVSCSI(pci_dev);
1112 
1113     trace_pvscsi_state("init");
1114 
1115     /* PCI subsystem ID, subsystem vendor ID, revision */
1116     if (PVSCSI_USE_OLD_PCI_CONFIGURATION(s)) {
1117         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID, 0x1000);
1118     } else {
1119         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1120                      PCI_VENDOR_ID_VMWARE);
1121         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1122                      PCI_DEVICE_ID_VMWARE_PVSCSI);
1123         pci_config_set_revision(pci_dev->config, 0x2);
1124     }
1125 
1126     /* PCI latency timer = 255 */
1127     pci_dev->config[PCI_LATENCY_TIMER] = 0xff;
1128 
1129     /* Interrupt pin A */
1130     pci_config_set_interrupt_pin(pci_dev->config, 1);
1131 
1132     memory_region_init_io(&s->io_space, OBJECT(s), &pvscsi_ops, s,
1133                           "pvscsi-io", PVSCSI_MEM_SPACE_SIZE);
1134     pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->io_space);
1135 
1136     pvscsi_init_msi(s);
1137 
1138     if (pci_is_express(pci_dev) && pci_bus_is_express(pci_get_bus(pci_dev))) {
1139         pcie_endpoint_cap_init(pci_dev, PVSCSI_EXP_EP_OFFSET);
1140     }
1141 
1142     s->completion_worker = qemu_bh_new(pvscsi_process_completion_queue, s);
1143 
1144     scsi_bus_new(&s->bus, sizeof(s->bus), DEVICE(pci_dev),
1145                  &pvscsi_scsi_info, NULL);
1146     /* override default SCSI bus hotplug-handler, with pvscsi's one */
1147     qbus_set_hotplug_handler(BUS(&s->bus), OBJECT(s));
1148     pvscsi_reset_state(s);
1149 }
1150 
1151 static void
1152 pvscsi_uninit(PCIDevice *pci_dev)
1153 {
1154     PVSCSIState *s = PVSCSI(pci_dev);
1155 
1156     trace_pvscsi_state("uninit");
1157     qemu_bh_delete(s->completion_worker);
1158 
1159     pvscsi_cleanup_msi(s);
1160 }
1161 
1162 static void
1163 pvscsi_reset(DeviceState *dev)
1164 {
1165     PCIDevice *d = PCI_DEVICE(dev);
1166     PVSCSIState *s = PVSCSI(d);
1167 
1168     trace_pvscsi_state("reset");
1169     pvscsi_reset_adapter(s);
1170 }
1171 
1172 static int
1173 pvscsi_pre_save(void *opaque)
1174 {
1175     PVSCSIState *s = (PVSCSIState *) opaque;
1176 
1177     trace_pvscsi_state("presave");
1178 
1179     assert(QTAILQ_EMPTY(&s->pending_queue));
1180     assert(QTAILQ_EMPTY(&s->completion_queue));
1181 
1182     return 0;
1183 }
1184 
1185 static int
1186 pvscsi_post_load(void *opaque, int version_id)
1187 {
1188     trace_pvscsi_state("postload");
1189     return 0;
1190 }
1191 
1192 static bool pvscsi_vmstate_need_pcie_device(void *opaque)
1193 {
1194     PVSCSIState *s = PVSCSI(opaque);
1195 
1196     return !(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE);
1197 }
1198 
1199 static bool pvscsi_vmstate_test_pci_device(void *opaque, int version_id)
1200 {
1201     return !pvscsi_vmstate_need_pcie_device(opaque);
1202 }
1203 
1204 static const VMStateDescription vmstate_pvscsi_pcie_device = {
1205     .name = "pvscsi/pcie",
1206     .needed = pvscsi_vmstate_need_pcie_device,
1207     .fields = (VMStateField[]) {
1208         VMSTATE_PCI_DEVICE(parent_obj, PVSCSIState),
1209         VMSTATE_END_OF_LIST()
1210     }
1211 };
1212 
1213 static const VMStateDescription vmstate_pvscsi = {
1214     .name = "pvscsi",
1215     .version_id = 0,
1216     .minimum_version_id = 0,
1217     .pre_save = pvscsi_pre_save,
1218     .post_load = pvscsi_post_load,
1219     .fields = (VMStateField[]) {
1220         VMSTATE_STRUCT_TEST(parent_obj, PVSCSIState,
1221                             pvscsi_vmstate_test_pci_device, 0,
1222                             vmstate_pci_device, PCIDevice),
1223         VMSTATE_UINT8(msi_used, PVSCSIState),
1224         VMSTATE_UINT32(resetting, PVSCSIState),
1225         VMSTATE_UINT64(reg_interrupt_status, PVSCSIState),
1226         VMSTATE_UINT64(reg_interrupt_enabled, PVSCSIState),
1227         VMSTATE_UINT64(reg_command_status, PVSCSIState),
1228         VMSTATE_UINT64(curr_cmd, PVSCSIState),
1229         VMSTATE_UINT32(curr_cmd_data_cntr, PVSCSIState),
1230         VMSTATE_UINT32_ARRAY(curr_cmd_data, PVSCSIState,
1231                              ARRAY_SIZE(((PVSCSIState *)NULL)->curr_cmd_data)),
1232         VMSTATE_UINT8(rings_info_valid, PVSCSIState),
1233         VMSTATE_UINT8(msg_ring_info_valid, PVSCSIState),
1234         VMSTATE_UINT8(use_msg, PVSCSIState),
1235 
1236         VMSTATE_UINT64(rings.rs_pa, PVSCSIState),
1237         VMSTATE_UINT32(rings.txr_len_mask, PVSCSIState),
1238         VMSTATE_UINT32(rings.rxr_len_mask, PVSCSIState),
1239         VMSTATE_UINT64_ARRAY(rings.req_ring_pages_pa, PVSCSIState,
1240                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1241         VMSTATE_UINT64_ARRAY(rings.cmp_ring_pages_pa, PVSCSIState,
1242                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1243         VMSTATE_UINT64(rings.consumed_ptr, PVSCSIState),
1244         VMSTATE_UINT64(rings.filled_cmp_ptr, PVSCSIState),
1245 
1246         VMSTATE_END_OF_LIST()
1247     },
1248     .subsections = (const VMStateDescription*[]) {
1249         &vmstate_pvscsi_pcie_device,
1250         NULL
1251     }
1252 };
1253 
1254 static Property pvscsi_properties[] = {
1255     DEFINE_PROP_UINT8("use_msg", PVSCSIState, use_msg, 1),
1256     DEFINE_PROP_BIT("x-old-pci-configuration", PVSCSIState, compat_flags,
1257                     PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT, false),
1258     DEFINE_PROP_BIT("x-disable-pcie", PVSCSIState, compat_flags,
1259                     PVSCSI_COMPAT_DISABLE_PCIE_BIT, false),
1260     DEFINE_PROP_END_OF_LIST(),
1261 };
1262 
1263 static void pvscsi_realize(DeviceState *qdev, Error **errp)
1264 {
1265     PVSCSIClass *pvs_c = PVSCSI_GET_CLASS(qdev);
1266     PCIDevice *pci_dev = PCI_DEVICE(qdev);
1267     PVSCSIState *s = PVSCSI(qdev);
1268 
1269     if (!(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE)) {
1270         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1271     }
1272 
1273     pvs_c->parent_dc_realize(qdev, errp);
1274 }
1275 
1276 static void pvscsi_class_init(ObjectClass *klass, void *data)
1277 {
1278     DeviceClass *dc = DEVICE_CLASS(klass);
1279     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1280     PVSCSIClass *pvs_k = PVSCSI_CLASS(klass);
1281     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
1282 
1283     k->realize = pvscsi_realizefn;
1284     k->exit = pvscsi_uninit;
1285     k->vendor_id = PCI_VENDOR_ID_VMWARE;
1286     k->device_id = PCI_DEVICE_ID_VMWARE_PVSCSI;
1287     k->class_id = PCI_CLASS_STORAGE_SCSI;
1288     k->subsystem_id = 0x1000;
1289     device_class_set_parent_realize(dc, pvscsi_realize,
1290                                     &pvs_k->parent_dc_realize);
1291     dc->reset = pvscsi_reset;
1292     dc->vmsd = &vmstate_pvscsi;
1293     device_class_set_props(dc, pvscsi_properties);
1294     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1295     hc->unplug = pvscsi_hot_unplug;
1296     hc->plug = pvscsi_hotplug;
1297 }
1298 
1299 static const TypeInfo pvscsi_info = {
1300     .name          = TYPE_PVSCSI,
1301     .parent        = TYPE_PCI_DEVICE,
1302     .class_size    = sizeof(PVSCSIClass),
1303     .instance_size = sizeof(PVSCSIState),
1304     .class_init    = pvscsi_class_init,
1305     .interfaces = (InterfaceInfo[]) {
1306         { TYPE_HOTPLUG_HANDLER },
1307         { INTERFACE_PCIE_DEVICE },
1308         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
1309         { }
1310     }
1311 };
1312 
1313 static void
1314 pvscsi_register_types(void)
1315 {
1316     type_register_static(&pvscsi_info);
1317 }
1318 
1319 type_init(pvscsi_register_types);
1320