xref: /openbmc/qemu/hw/scsi/vmw_pvscsi.c (revision f0984d40)
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(pval, m, field) \
54     ldl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
55                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field), \
56                  pval, MEMTXATTRS_UNSPECIFIED)
57 #define RS_SET_FIELD(m, field, val) \
58     (stl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
59                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field), val, \
60                  MEMTXATTRS_UNSPECIFIED))
61 
62 struct PVSCSIClass {
63     PCIDeviceClass parent_class;
64     DeviceRealize parent_dc_realize;
65 };
66 
67 #define TYPE_PVSCSI "pvscsi"
68 OBJECT_DECLARE_TYPE(PVSCSIState, PVSCSIClass, PVSCSI)
69 
70 
71 /* Compatibility flags for migration */
72 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT 0
73 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION \
74     (1 << PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT)
75 #define PVSCSI_COMPAT_DISABLE_PCIE_BIT 1
76 #define PVSCSI_COMPAT_DISABLE_PCIE \
77     (1 << PVSCSI_COMPAT_DISABLE_PCIE_BIT)
78 
79 #define PVSCSI_USE_OLD_PCI_CONFIGURATION(s) \
80     ((s)->compat_flags & PVSCSI_COMPAT_OLD_PCI_CONFIGURATION)
81 #define PVSCSI_MSI_OFFSET(s) \
82     (PVSCSI_USE_OLD_PCI_CONFIGURATION(s) ? 0x50 : 0x7c)
83 #define PVSCSI_EXP_EP_OFFSET (0x40)
84 
85 typedef struct PVSCSIRingInfo {
86     uint64_t            rs_pa;
87     uint32_t            txr_len_mask;
88     uint32_t            rxr_len_mask;
89     uint32_t            msg_len_mask;
90     uint64_t            req_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
91     uint64_t            cmp_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
92     uint64_t            msg_ring_pages_pa[PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES];
93     uint64_t            consumed_ptr;
94     uint64_t            filled_cmp_ptr;
95     uint64_t            filled_msg_ptr;
96 } PVSCSIRingInfo;
97 
98 typedef struct PVSCSISGState {
99     hwaddr elemAddr;
100     hwaddr dataAddr;
101     uint32_t resid;
102 } PVSCSISGState;
103 
104 typedef QTAILQ_HEAD(, PVSCSIRequest) PVSCSIRequestList;
105 
106 struct PVSCSIState {
107     PCIDevice parent_obj;
108     MemoryRegion io_space;
109     SCSIBus bus;
110     QEMUBH *completion_worker;
111     PVSCSIRequestList pending_queue;
112     PVSCSIRequestList completion_queue;
113 
114     uint64_t reg_interrupt_status;        /* Interrupt status register value */
115     uint64_t reg_interrupt_enabled;       /* Interrupt mask register value   */
116     uint64_t reg_command_status;          /* Command status register value   */
117 
118     /* Command data adoption mechanism */
119     uint64_t curr_cmd;                   /* Last command arrived             */
120     uint32_t curr_cmd_data_cntr;         /* Amount of data for last command  */
121 
122     /* Collector for current command data */
123     uint32_t curr_cmd_data[PVSCSI_MAX_CMD_DATA_WORDS];
124 
125     uint8_t rings_info_valid;            /* Whether data rings initialized   */
126     uint8_t msg_ring_info_valid;         /* Whether message ring initialized */
127     uint8_t use_msg;                     /* Whether to use message ring      */
128 
129     uint8_t msi_used;                    /* For migration compatibility      */
130     PVSCSIRingInfo rings;                /* Data transfer rings manager      */
131     uint32_t resetting;                  /* Reset in progress                */
132 
133     uint32_t compat_flags;
134 };
135 
136 typedef struct PVSCSIRequest {
137     SCSIRequest *sreq;
138     PVSCSIState *dev;
139     uint8_t sense_key;
140     uint8_t completed;
141     int lun;
142     QEMUSGList sgl;
143     PVSCSISGState sg;
144     struct PVSCSIRingReqDesc req;
145     struct PVSCSIRingCmpDesc cmp;
146     QTAILQ_ENTRY(PVSCSIRequest) next;
147 } PVSCSIRequest;
148 
149 /* Integer binary logarithm */
150 static int
151 pvscsi_log2(uint32_t input)
152 {
153     int log = 0;
154     assert(input > 0);
155     while (input >> ++log) {
156     }
157     return log;
158 }
159 
160 static void
161 pvscsi_ring_init_data(PVSCSIRingInfo *m, PVSCSICmdDescSetupRings *ri)
162 {
163     int i;
164     uint32_t txr_len_log2, rxr_len_log2;
165     uint32_t req_ring_size, cmp_ring_size;
166     m->rs_pa = ri->ringsStatePPN << VMW_PAGE_SHIFT;
167 
168     req_ring_size = ri->reqRingNumPages * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
169     cmp_ring_size = ri->cmpRingNumPages * PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
170     txr_len_log2 = pvscsi_log2(req_ring_size - 1);
171     rxr_len_log2 = pvscsi_log2(cmp_ring_size - 1);
172 
173     m->txr_len_mask = MASK(txr_len_log2);
174     m->rxr_len_mask = MASK(rxr_len_log2);
175 
176     m->consumed_ptr = 0;
177     m->filled_cmp_ptr = 0;
178 
179     for (i = 0; i < ri->reqRingNumPages; i++) {
180         m->req_ring_pages_pa[i] = ri->reqRingPPNs[i] << VMW_PAGE_SHIFT;
181     }
182 
183     for (i = 0; i < ri->cmpRingNumPages; i++) {
184         m->cmp_ring_pages_pa[i] = ri->cmpRingPPNs[i] << VMW_PAGE_SHIFT;
185     }
186 
187     RS_SET_FIELD(m, reqProdIdx, 0);
188     RS_SET_FIELD(m, reqConsIdx, 0);
189     RS_SET_FIELD(m, reqNumEntriesLog2, txr_len_log2);
190 
191     RS_SET_FIELD(m, cmpProdIdx, 0);
192     RS_SET_FIELD(m, cmpConsIdx, 0);
193     RS_SET_FIELD(m, cmpNumEntriesLog2, rxr_len_log2);
194 
195     trace_pvscsi_ring_init_data(txr_len_log2, rxr_len_log2);
196 
197     /* Flush ring state page changes */
198     smp_wmb();
199 }
200 
201 static int
202 pvscsi_ring_init_msg(PVSCSIRingInfo *m, PVSCSICmdDescSetupMsgRing *ri)
203 {
204     int i;
205     uint32_t len_log2;
206     uint32_t ring_size;
207 
208     if (!ri->numPages || ri->numPages > PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES) {
209         return -1;
210     }
211     ring_size = ri->numPages * PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
212     len_log2 = pvscsi_log2(ring_size - 1);
213 
214     m->msg_len_mask = MASK(len_log2);
215 
216     m->filled_msg_ptr = 0;
217 
218     for (i = 0; i < ri->numPages; i++) {
219         m->msg_ring_pages_pa[i] = ri->ringPPNs[i] << VMW_PAGE_SHIFT;
220     }
221 
222     RS_SET_FIELD(m, msgProdIdx, 0);
223     RS_SET_FIELD(m, msgConsIdx, 0);
224     RS_SET_FIELD(m, msgNumEntriesLog2, len_log2);
225 
226     trace_pvscsi_ring_init_msg(len_log2);
227 
228     /* Flush ring state page changes */
229     smp_wmb();
230 
231     return 0;
232 }
233 
234 static void
235 pvscsi_ring_cleanup(PVSCSIRingInfo *mgr)
236 {
237     mgr->rs_pa = 0;
238     mgr->txr_len_mask = 0;
239     mgr->rxr_len_mask = 0;
240     mgr->msg_len_mask = 0;
241     mgr->consumed_ptr = 0;
242     mgr->filled_cmp_ptr = 0;
243     mgr->filled_msg_ptr = 0;
244     memset(mgr->req_ring_pages_pa, 0, sizeof(mgr->req_ring_pages_pa));
245     memset(mgr->cmp_ring_pages_pa, 0, sizeof(mgr->cmp_ring_pages_pa));
246     memset(mgr->msg_ring_pages_pa, 0, sizeof(mgr->msg_ring_pages_pa));
247 }
248 
249 static hwaddr
250 pvscsi_ring_pop_req_descr(PVSCSIRingInfo *mgr)
251 {
252     uint32_t ready_ptr;
253     uint32_t ring_size = PVSCSI_MAX_NUM_PAGES_REQ_RING
254                             * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
255 
256     RS_GET_FIELD(&ready_ptr, mgr, reqProdIdx);
257     if (ready_ptr != mgr->consumed_ptr
258         && ready_ptr - mgr->consumed_ptr < ring_size) {
259         uint32_t next_ready_ptr =
260             mgr->consumed_ptr++ & mgr->txr_len_mask;
261         uint32_t next_ready_page =
262             next_ready_ptr / PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
263         uint32_t inpage_idx =
264             next_ready_ptr % PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
265 
266         return mgr->req_ring_pages_pa[next_ready_page] +
267                inpage_idx * sizeof(PVSCSIRingReqDesc);
268     } else {
269         return 0;
270     }
271 }
272 
273 static void
274 pvscsi_ring_flush_req(PVSCSIRingInfo *mgr)
275 {
276     RS_SET_FIELD(mgr, reqConsIdx, mgr->consumed_ptr);
277 }
278 
279 static hwaddr
280 pvscsi_ring_pop_cmp_descr(PVSCSIRingInfo *mgr)
281 {
282     /*
283      * According to Linux driver code it explicitly verifies that number
284      * of requests being processed by device is less then the size of
285      * completion queue, so device may omit completion queue overflow
286      * conditions check. We assume that this is true for other (Windows)
287      * drivers as well.
288      */
289 
290     uint32_t free_cmp_ptr =
291         mgr->filled_cmp_ptr++ & mgr->rxr_len_mask;
292     uint32_t free_cmp_page =
293         free_cmp_ptr / PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
294     uint32_t inpage_idx =
295         free_cmp_ptr % PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
296     return mgr->cmp_ring_pages_pa[free_cmp_page] +
297            inpage_idx * sizeof(PVSCSIRingCmpDesc);
298 }
299 
300 static hwaddr
301 pvscsi_ring_pop_msg_descr(PVSCSIRingInfo *mgr)
302 {
303     uint32_t free_msg_ptr =
304         mgr->filled_msg_ptr++ & mgr->msg_len_mask;
305     uint32_t free_msg_page =
306         free_msg_ptr / PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
307     uint32_t inpage_idx =
308         free_msg_ptr % PVSCSI_MAX_NUM_MSG_ENTRIES_PER_PAGE;
309     return mgr->msg_ring_pages_pa[free_msg_page] +
310            inpage_idx * sizeof(PVSCSIRingMsgDesc);
311 }
312 
313 static void
314 pvscsi_ring_flush_cmp(PVSCSIRingInfo *mgr)
315 {
316     /* Flush descriptor changes */
317     smp_wmb();
318 
319     trace_pvscsi_ring_flush_cmp(mgr->filled_cmp_ptr);
320 
321     RS_SET_FIELD(mgr, cmpProdIdx, mgr->filled_cmp_ptr);
322 }
323 
324 static bool
325 pvscsi_ring_msg_has_room(PVSCSIRingInfo *mgr)
326 {
327     uint32_t prodIdx;
328     uint32_t consIdx;
329 
330     RS_GET_FIELD(&prodIdx, mgr, msgProdIdx);
331     RS_GET_FIELD(&consIdx, mgr, msgConsIdx);
332 
333     return (prodIdx - consIdx) < (mgr->msg_len_mask + 1);
334 }
335 
336 static void
337 pvscsi_ring_flush_msg(PVSCSIRingInfo *mgr)
338 {
339     /* Flush descriptor changes */
340     smp_wmb();
341 
342     trace_pvscsi_ring_flush_msg(mgr->filled_msg_ptr);
343 
344     RS_SET_FIELD(mgr, msgProdIdx, mgr->filled_msg_ptr);
345 }
346 
347 static void
348 pvscsi_reset_state(PVSCSIState *s)
349 {
350     s->curr_cmd = PVSCSI_CMD_FIRST;
351     s->curr_cmd_data_cntr = 0;
352     s->reg_command_status = PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
353     s->reg_interrupt_status = 0;
354     pvscsi_ring_cleanup(&s->rings);
355     s->rings_info_valid = FALSE;
356     s->msg_ring_info_valid = FALSE;
357     QTAILQ_INIT(&s->pending_queue);
358     QTAILQ_INIT(&s->completion_queue);
359 }
360 
361 static void
362 pvscsi_update_irq_status(PVSCSIState *s)
363 {
364     PCIDevice *d = PCI_DEVICE(s);
365     bool should_raise = s->reg_interrupt_enabled & s->reg_interrupt_status;
366 
367     trace_pvscsi_update_irq_level(should_raise, s->reg_interrupt_enabled,
368                                   s->reg_interrupt_status);
369 
370     if (msi_enabled(d)) {
371         if (should_raise) {
372             trace_pvscsi_update_irq_msi();
373             msi_notify(d, PVSCSI_VECTOR_COMPLETION);
374         }
375         return;
376     }
377 
378     pci_set_irq(d, !!should_raise);
379 }
380 
381 static void
382 pvscsi_raise_completion_interrupt(PVSCSIState *s)
383 {
384     s->reg_interrupt_status |= PVSCSI_INTR_CMPL_0;
385 
386     /* Memory barrier to flush interrupt status register changes*/
387     smp_wmb();
388 
389     pvscsi_update_irq_status(s);
390 }
391 
392 static void
393 pvscsi_raise_message_interrupt(PVSCSIState *s)
394 {
395     s->reg_interrupt_status |= PVSCSI_INTR_MSG_0;
396 
397     /* Memory barrier to flush interrupt status register changes*/
398     smp_wmb();
399 
400     pvscsi_update_irq_status(s);
401 }
402 
403 static void
404 pvscsi_cmp_ring_put(PVSCSIState *s, struct PVSCSIRingCmpDesc *cmp_desc)
405 {
406     hwaddr cmp_descr_pa;
407 
408     cmp_descr_pa = pvscsi_ring_pop_cmp_descr(&s->rings);
409     trace_pvscsi_cmp_ring_put(cmp_descr_pa);
410     cpu_physical_memory_write(cmp_descr_pa, cmp_desc, sizeof(*cmp_desc));
411 }
412 
413 static void
414 pvscsi_msg_ring_put(PVSCSIState *s, struct PVSCSIRingMsgDesc *msg_desc)
415 {
416     hwaddr msg_descr_pa;
417 
418     msg_descr_pa = pvscsi_ring_pop_msg_descr(&s->rings);
419     trace_pvscsi_msg_ring_put(msg_descr_pa);
420     cpu_physical_memory_write(msg_descr_pa, msg_desc, sizeof(*msg_desc));
421 }
422 
423 static void
424 pvscsi_process_completion_queue(void *opaque)
425 {
426     PVSCSIState *s = opaque;
427     PVSCSIRequest *pvscsi_req;
428     bool has_completed = false;
429 
430     while (!QTAILQ_EMPTY(&s->completion_queue)) {
431         pvscsi_req = QTAILQ_FIRST(&s->completion_queue);
432         QTAILQ_REMOVE(&s->completion_queue, pvscsi_req, next);
433         pvscsi_cmp_ring_put(s, &pvscsi_req->cmp);
434         g_free(pvscsi_req);
435         has_completed = true;
436     }
437 
438     if (has_completed) {
439         pvscsi_ring_flush_cmp(&s->rings);
440         pvscsi_raise_completion_interrupt(s);
441     }
442 }
443 
444 static void
445 pvscsi_reset_adapter(PVSCSIState *s)
446 {
447     s->resetting++;
448     bus_cold_reset(BUS(&s->bus));
449     s->resetting--;
450     pvscsi_process_completion_queue(s);
451     assert(QTAILQ_EMPTY(&s->pending_queue));
452     pvscsi_reset_state(s);
453 }
454 
455 static void
456 pvscsi_schedule_completion_processing(PVSCSIState *s)
457 {
458     /* Try putting more complete requests on the ring. */
459     if (!QTAILQ_EMPTY(&s->completion_queue)) {
460         qemu_bh_schedule(s->completion_worker);
461     }
462 }
463 
464 static void
465 pvscsi_complete_request(PVSCSIState *s, PVSCSIRequest *r)
466 {
467     assert(!r->completed);
468 
469     trace_pvscsi_complete_request(r->cmp.context, r->cmp.dataLen,
470                                   r->sense_key);
471     if (r->sreq != NULL) {
472         scsi_req_unref(r->sreq);
473         r->sreq = NULL;
474     }
475     r->completed = 1;
476     QTAILQ_REMOVE(&s->pending_queue, r, next);
477     QTAILQ_INSERT_TAIL(&s->completion_queue, r, next);
478     pvscsi_schedule_completion_processing(s);
479 }
480 
481 static QEMUSGList *pvscsi_get_sg_list(SCSIRequest *r)
482 {
483     PVSCSIRequest *req = r->hba_private;
484 
485     trace_pvscsi_get_sg_list(req->sgl.nsg, req->sgl.size);
486 
487     return &req->sgl;
488 }
489 
490 static void
491 pvscsi_get_next_sg_elem(PVSCSISGState *sg)
492 {
493     struct PVSCSISGElement elem;
494 
495     cpu_physical_memory_read(sg->elemAddr, &elem, sizeof(elem));
496     if ((elem.flags & ~PVSCSI_KNOWN_FLAGS) != 0) {
497         /*
498             * There is PVSCSI_SGE_FLAG_CHAIN_ELEMENT flag described in
499             * header file but its value is unknown. This flag requires
500             * additional processing, so we put warning here to catch it
501             * some day and make proper implementation
502             */
503         trace_pvscsi_get_next_sg_elem(elem.flags);
504     }
505 
506     sg->elemAddr += sizeof(elem);
507     sg->dataAddr = elem.addr;
508     sg->resid = elem.length;
509 }
510 
511 static void
512 pvscsi_write_sense(PVSCSIRequest *r, uint8_t *sense, int len)
513 {
514     r->cmp.senseLen = MIN(r->req.senseLen, len);
515     r->sense_key = sense[(sense[0] & 2) ? 1 : 2];
516     cpu_physical_memory_write(r->req.senseAddr, sense, r->cmp.senseLen);
517 }
518 
519 static void
520 pvscsi_command_failed(SCSIRequest *req)
521 {
522     PVSCSIRequest *pvscsi_req = req->hba_private;
523     PVSCSIState *s;
524 
525     if (!pvscsi_req) {
526         trace_pvscsi_command_complete_not_found(req->tag);
527         return;
528     }
529     s = pvscsi_req->dev;
530 
531     switch (req->host_status) {
532     case SCSI_HOST_NO_LUN:
533         pvscsi_req->cmp.hostStatus = BTSTAT_LUNMISMATCH;
534         break;
535     case SCSI_HOST_BUSY:
536         pvscsi_req->cmp.hostStatus = BTSTAT_ABORTQUEUE;
537         break;
538     case SCSI_HOST_TIME_OUT:
539     case SCSI_HOST_ABORTED:
540         pvscsi_req->cmp.hostStatus = BTSTAT_SENTRST;
541         break;
542     case SCSI_HOST_BAD_RESPONSE:
543         pvscsi_req->cmp.hostStatus = BTSTAT_SELTIMEO;
544         break;
545     case SCSI_HOST_RESET:
546         pvscsi_req->cmp.hostStatus = BTSTAT_BUSRESET;
547         break;
548     default:
549         pvscsi_req->cmp.hostStatus = BTSTAT_HASOFTWARE;
550         break;
551     }
552     pvscsi_req->cmp.scsiStatus = GOOD;
553     qemu_sglist_destroy(&pvscsi_req->sgl);
554     pvscsi_complete_request(s, pvscsi_req);
555 }
556 
557 static void
558 pvscsi_command_complete(SCSIRequest *req, size_t resid)
559 {
560     PVSCSIRequest *pvscsi_req = req->hba_private;
561     PVSCSIState *s;
562 
563     if (!pvscsi_req) {
564         trace_pvscsi_command_complete_not_found(req->tag);
565         return;
566     }
567     s = pvscsi_req->dev;
568 
569     if (resid) {
570         /* Short transfer.  */
571         trace_pvscsi_command_complete_data_run();
572         pvscsi_req->cmp.hostStatus = BTSTAT_DATARUN;
573     }
574 
575     pvscsi_req->cmp.scsiStatus = req->status;
576     if (pvscsi_req->cmp.scsiStatus == CHECK_CONDITION) {
577         uint8_t sense[SCSI_SENSE_BUF_SIZE];
578         int sense_len =
579             scsi_req_get_sense(pvscsi_req->sreq, sense, sizeof(sense));
580 
581         trace_pvscsi_command_complete_sense_len(sense_len);
582         pvscsi_write_sense(pvscsi_req, sense, sense_len);
583     }
584     qemu_sglist_destroy(&pvscsi_req->sgl);
585     pvscsi_complete_request(s, pvscsi_req);
586 }
587 
588 static void
589 pvscsi_send_msg(PVSCSIState *s, SCSIDevice *dev, uint32_t msg_type)
590 {
591     if (s->msg_ring_info_valid && pvscsi_ring_msg_has_room(&s->rings)) {
592         PVSCSIMsgDescDevStatusChanged msg = {0};
593 
594         msg.type = msg_type;
595         msg.bus = dev->channel;
596         msg.target = dev->id;
597         msg.lun[1] = dev->lun;
598 
599         pvscsi_msg_ring_put(s, (PVSCSIRingMsgDesc *)&msg);
600         pvscsi_ring_flush_msg(&s->rings);
601         pvscsi_raise_message_interrupt(s);
602     }
603 }
604 
605 static void
606 pvscsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
607 {
608     PVSCSIState *s = PVSCSI(hotplug_dev);
609 
610     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_ADDED);
611 }
612 
613 static void
614 pvscsi_hot_unplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
615 {
616     PVSCSIState *s = PVSCSI(hotplug_dev);
617 
618     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_REMOVED);
619     qdev_simple_device_unplug_cb(hotplug_dev, dev, errp);
620 }
621 
622 static void
623 pvscsi_request_cancelled(SCSIRequest *req)
624 {
625     PVSCSIRequest *pvscsi_req = req->hba_private;
626     PVSCSIState *s = pvscsi_req->dev;
627 
628     if (pvscsi_req->completed) {
629         return;
630     }
631 
632    if (pvscsi_req->dev->resetting) {
633        pvscsi_req->cmp.hostStatus = BTSTAT_BUSRESET;
634     } else {
635        pvscsi_req->cmp.hostStatus = BTSTAT_ABORTQUEUE;
636     }
637 
638     pvscsi_complete_request(s, pvscsi_req);
639 }
640 
641 static SCSIDevice*
642 pvscsi_device_find(PVSCSIState *s, int channel, int target,
643                    uint8_t *requested_lun, uint8_t *target_lun)
644 {
645     if (requested_lun[0] || requested_lun[2] || requested_lun[3] ||
646         requested_lun[4] || requested_lun[5] || requested_lun[6] ||
647         requested_lun[7] || (target > PVSCSI_MAX_DEVS)) {
648         return NULL;
649     } else {
650         *target_lun = requested_lun[1];
651         return scsi_device_find(&s->bus, channel, target, *target_lun);
652     }
653 }
654 
655 static PVSCSIRequest *
656 pvscsi_queue_pending_descriptor(PVSCSIState *s, SCSIDevice **d,
657                                 struct PVSCSIRingReqDesc *descr)
658 {
659     PVSCSIRequest *pvscsi_req;
660     uint8_t lun;
661 
662     pvscsi_req = g_malloc0(sizeof(*pvscsi_req));
663     pvscsi_req->dev = s;
664     pvscsi_req->req = *descr;
665     pvscsi_req->cmp.context = pvscsi_req->req.context;
666     QTAILQ_INSERT_TAIL(&s->pending_queue, pvscsi_req, next);
667 
668     *d = pvscsi_device_find(s, descr->bus, descr->target, descr->lun, &lun);
669     if (*d) {
670         pvscsi_req->lun = lun;
671     }
672 
673     return pvscsi_req;
674 }
675 
676 static void
677 pvscsi_convert_sglist(PVSCSIRequest *r)
678 {
679     uint32_t chunk_size, elmcnt = 0;
680     uint64_t data_length = r->req.dataLen;
681     PVSCSISGState sg = r->sg;
682     while (data_length && elmcnt < PVSCSI_MAX_SG_ELEM) {
683         while (!sg.resid && elmcnt++ < PVSCSI_MAX_SG_ELEM) {
684             pvscsi_get_next_sg_elem(&sg);
685             trace_pvscsi_convert_sglist(r->req.context, r->sg.dataAddr,
686                                         r->sg.resid);
687         }
688         chunk_size = MIN(data_length, sg.resid);
689         if (chunk_size) {
690             qemu_sglist_add(&r->sgl, sg.dataAddr, chunk_size);
691         }
692 
693         sg.dataAddr += chunk_size;
694         data_length -= chunk_size;
695         sg.resid -= chunk_size;
696     }
697 }
698 
699 static void
700 pvscsi_build_sglist(PVSCSIState *s, PVSCSIRequest *r)
701 {
702     PCIDevice *d = PCI_DEVICE(s);
703 
704     pci_dma_sglist_init(&r->sgl, d, 1);
705     if (r->req.flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
706         pvscsi_convert_sglist(r);
707     } else {
708         qemu_sglist_add(&r->sgl, r->req.dataAddr, r->req.dataLen);
709     }
710 }
711 
712 static void
713 pvscsi_process_request_descriptor(PVSCSIState *s,
714                                   struct PVSCSIRingReqDesc *descr)
715 {
716     SCSIDevice *d;
717     PVSCSIRequest *r = pvscsi_queue_pending_descriptor(s, &d, descr);
718     int64_t n;
719 
720     trace_pvscsi_process_req_descr(descr->cdb[0], descr->context);
721 
722     if (!d) {
723         r->cmp.hostStatus = BTSTAT_SELTIMEO;
724         trace_pvscsi_process_req_descr_unknown_device();
725         pvscsi_complete_request(s, r);
726         return;
727     }
728 
729     if (descr->flags & PVSCSI_FLAG_CMD_WITH_SG_LIST) {
730         r->sg.elemAddr = descr->dataAddr;
731     }
732 
733     r->sreq = scsi_req_new(d, descr->context, r->lun, descr->cdb, descr->cdbLen, r);
734     if (r->sreq->cmd.mode == SCSI_XFER_FROM_DEV &&
735         (descr->flags & PVSCSI_FLAG_CMD_DIR_TODEVICE)) {
736         r->cmp.hostStatus = BTSTAT_BADMSG;
737         trace_pvscsi_process_req_descr_invalid_dir();
738         scsi_req_cancel(r->sreq);
739         return;
740     }
741     if (r->sreq->cmd.mode == SCSI_XFER_TO_DEV &&
742         (descr->flags & PVSCSI_FLAG_CMD_DIR_TOHOST)) {
743         r->cmp.hostStatus = BTSTAT_BADMSG;
744         trace_pvscsi_process_req_descr_invalid_dir();
745         scsi_req_cancel(r->sreq);
746         return;
747     }
748 
749     pvscsi_build_sglist(s, r);
750     n = scsi_req_enqueue(r->sreq);
751 
752     if (n) {
753         scsi_req_continue(r->sreq);
754     }
755 }
756 
757 static void
758 pvscsi_process_io(PVSCSIState *s)
759 {
760     PVSCSIRingReqDesc descr;
761     hwaddr next_descr_pa;
762 
763     if (!s->rings_info_valid) {
764         return;
765     }
766 
767     while ((next_descr_pa = pvscsi_ring_pop_req_descr(&s->rings)) != 0) {
768 
769         /* Only read after production index verification */
770         smp_rmb();
771 
772         trace_pvscsi_process_io(next_descr_pa);
773         cpu_physical_memory_read(next_descr_pa, &descr, sizeof(descr));
774         pvscsi_process_request_descriptor(s, &descr);
775     }
776 
777     pvscsi_ring_flush_req(&s->rings);
778 }
779 
780 static void
781 pvscsi_dbg_dump_tx_rings_config(PVSCSICmdDescSetupRings *rc)
782 {
783     int i;
784     trace_pvscsi_tx_rings_ppn("Rings State", rc->ringsStatePPN);
785 
786     trace_pvscsi_tx_rings_num_pages("Request Ring", rc->reqRingNumPages);
787     for (i = 0; i < rc->reqRingNumPages; i++) {
788         trace_pvscsi_tx_rings_ppn("Request Ring", rc->reqRingPPNs[i]);
789     }
790 
791     trace_pvscsi_tx_rings_num_pages("Confirm Ring", rc->cmpRingNumPages);
792     for (i = 0; i < rc->cmpRingNumPages; i++) {
793         trace_pvscsi_tx_rings_ppn("Confirm Ring", rc->cmpRingPPNs[i]);
794     }
795 }
796 
797 static uint64_t
798 pvscsi_on_cmd_config(PVSCSIState *s)
799 {
800     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_CONFIG");
801     return PVSCSI_COMMAND_PROCESSING_FAILED;
802 }
803 
804 static uint64_t
805 pvscsi_on_cmd_unplug(PVSCSIState *s)
806 {
807     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_DEVICE_UNPLUG");
808     return PVSCSI_COMMAND_PROCESSING_FAILED;
809 }
810 
811 static uint64_t
812 pvscsi_on_issue_scsi(PVSCSIState *s)
813 {
814     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_ISSUE_SCSI");
815     return PVSCSI_COMMAND_PROCESSING_FAILED;
816 }
817 
818 static uint64_t
819 pvscsi_on_cmd_setup_rings(PVSCSIState *s)
820 {
821     PVSCSICmdDescSetupRings *rc =
822         (PVSCSICmdDescSetupRings *) s->curr_cmd_data;
823 
824     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_RINGS");
825 
826     if (!rc->reqRingNumPages
827         || rc->reqRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES
828         || !rc->cmpRingNumPages
829         || rc->cmpRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES) {
830         return PVSCSI_COMMAND_PROCESSING_FAILED;
831     }
832 
833     pvscsi_dbg_dump_tx_rings_config(rc);
834     pvscsi_ring_init_data(&s->rings, rc);
835 
836     s->rings_info_valid = TRUE;
837     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
838 }
839 
840 static uint64_t
841 pvscsi_on_cmd_abort(PVSCSIState *s)
842 {
843     PVSCSICmdDescAbortCmd *cmd = (PVSCSICmdDescAbortCmd *) s->curr_cmd_data;
844     PVSCSIRequest *r, *next;
845 
846     trace_pvscsi_on_cmd_abort(cmd->context, cmd->target);
847 
848     QTAILQ_FOREACH_SAFE(r, &s->pending_queue, next, next) {
849         if (r->req.context == cmd->context) {
850             break;
851         }
852     }
853     if (r) {
854         assert(!r->completed);
855         r->cmp.hostStatus = BTSTAT_ABORTQUEUE;
856         scsi_req_cancel(r->sreq);
857     }
858 
859     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
860 }
861 
862 static uint64_t
863 pvscsi_on_cmd_unknown(PVSCSIState *s)
864 {
865     trace_pvscsi_on_cmd_unknown_data(s->curr_cmd_data[0]);
866     return PVSCSI_COMMAND_PROCESSING_FAILED;
867 }
868 
869 static uint64_t
870 pvscsi_on_cmd_reset_device(PVSCSIState *s)
871 {
872     uint8_t target_lun = 0;
873     struct PVSCSICmdDescResetDevice *cmd =
874         (struct PVSCSICmdDescResetDevice *) s->curr_cmd_data;
875     SCSIDevice *sdev;
876 
877     sdev = pvscsi_device_find(s, 0, cmd->target, cmd->lun, &target_lun);
878 
879     trace_pvscsi_on_cmd_reset_dev(cmd->target, (int) target_lun, sdev);
880 
881     if (sdev != NULL) {
882         s->resetting++;
883         device_cold_reset(&sdev->qdev);
884         s->resetting--;
885         return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
886     }
887 
888     return PVSCSI_COMMAND_PROCESSING_FAILED;
889 }
890 
891 static uint64_t
892 pvscsi_on_cmd_reset_bus(PVSCSIState *s)
893 {
894     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_RESET_BUS");
895 
896     s->resetting++;
897     bus_cold_reset(BUS(&s->bus));
898     s->resetting--;
899     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
900 }
901 
902 static uint64_t
903 pvscsi_on_cmd_setup_msg_ring(PVSCSIState *s)
904 {
905     PVSCSICmdDescSetupMsgRing *rc =
906         (PVSCSICmdDescSetupMsgRing *) s->curr_cmd_data;
907 
908     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_MSG_RING");
909 
910     if (!s->use_msg) {
911         return PVSCSI_COMMAND_PROCESSING_FAILED;
912     }
913 
914     if (s->rings_info_valid) {
915         if (pvscsi_ring_init_msg(&s->rings, rc) < 0) {
916             return PVSCSI_COMMAND_PROCESSING_FAILED;
917         }
918         s->msg_ring_info_valid = TRUE;
919     }
920     return sizeof(PVSCSICmdDescSetupMsgRing) / sizeof(uint32_t);
921 }
922 
923 static uint64_t
924 pvscsi_on_cmd_adapter_reset(PVSCSIState *s)
925 {
926     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_ADAPTER_RESET");
927 
928     pvscsi_reset_adapter(s);
929     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
930 }
931 
932 static const struct {
933     int       data_size;
934     uint64_t  (*handler_fn)(PVSCSIState *s);
935 } pvscsi_commands[] = {
936     [PVSCSI_CMD_FIRST] = {
937         .data_size = 0,
938         .handler_fn = pvscsi_on_cmd_unknown,
939     },
940 
941     /* Not implemented, data size defined based on what arrives on windows */
942     [PVSCSI_CMD_CONFIG] = {
943         .data_size = 6 * sizeof(uint32_t),
944         .handler_fn = pvscsi_on_cmd_config,
945     },
946 
947     /* Command not implemented, data size is unknown */
948     [PVSCSI_CMD_ISSUE_SCSI] = {
949         .data_size = 0,
950         .handler_fn = pvscsi_on_issue_scsi,
951     },
952 
953     /* Command not implemented, data size is unknown */
954     [PVSCSI_CMD_DEVICE_UNPLUG] = {
955         .data_size = 0,
956         .handler_fn = pvscsi_on_cmd_unplug,
957     },
958 
959     [PVSCSI_CMD_SETUP_RINGS] = {
960         .data_size = sizeof(PVSCSICmdDescSetupRings),
961         .handler_fn = pvscsi_on_cmd_setup_rings,
962     },
963 
964     [PVSCSI_CMD_RESET_DEVICE] = {
965         .data_size = sizeof(struct PVSCSICmdDescResetDevice),
966         .handler_fn = pvscsi_on_cmd_reset_device,
967     },
968 
969     [PVSCSI_CMD_RESET_BUS] = {
970         .data_size = 0,
971         .handler_fn = pvscsi_on_cmd_reset_bus,
972     },
973 
974     [PVSCSI_CMD_SETUP_MSG_RING] = {
975         .data_size = sizeof(PVSCSICmdDescSetupMsgRing),
976         .handler_fn = pvscsi_on_cmd_setup_msg_ring,
977     },
978 
979     [PVSCSI_CMD_ADAPTER_RESET] = {
980         .data_size = 0,
981         .handler_fn = pvscsi_on_cmd_adapter_reset,
982     },
983 
984     [PVSCSI_CMD_ABORT_CMD] = {
985         .data_size = sizeof(struct PVSCSICmdDescAbortCmd),
986         .handler_fn = pvscsi_on_cmd_abort,
987     },
988 };
989 
990 static void
991 pvscsi_do_command_processing(PVSCSIState *s)
992 {
993     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
994 
995     assert(s->curr_cmd < PVSCSI_CMD_LAST);
996     if (bytes_arrived >= pvscsi_commands[s->curr_cmd].data_size) {
997         s->reg_command_status = pvscsi_commands[s->curr_cmd].handler_fn(s);
998         s->curr_cmd = PVSCSI_CMD_FIRST;
999         s->curr_cmd_data_cntr   = 0;
1000     }
1001 }
1002 
1003 static void
1004 pvscsi_on_command_data(PVSCSIState *s, uint32_t value)
1005 {
1006     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
1007 
1008     assert(bytes_arrived < sizeof(s->curr_cmd_data));
1009     s->curr_cmd_data[s->curr_cmd_data_cntr++] = value;
1010 
1011     pvscsi_do_command_processing(s);
1012 }
1013 
1014 static void
1015 pvscsi_on_command(PVSCSIState *s, uint64_t cmd_id)
1016 {
1017     if ((cmd_id > PVSCSI_CMD_FIRST) && (cmd_id < PVSCSI_CMD_LAST)) {
1018         s->curr_cmd = cmd_id;
1019     } else {
1020         s->curr_cmd = PVSCSI_CMD_FIRST;
1021         trace_pvscsi_on_cmd_unknown(cmd_id);
1022     }
1023 
1024     s->curr_cmd_data_cntr = 0;
1025     s->reg_command_status = PVSCSI_COMMAND_NOT_ENOUGH_DATA;
1026 
1027     pvscsi_do_command_processing(s);
1028 }
1029 
1030 static void
1031 pvscsi_io_write(void *opaque, hwaddr addr,
1032                 uint64_t val, unsigned size)
1033 {
1034     PVSCSIState *s = opaque;
1035 
1036     switch (addr) {
1037     case PVSCSI_REG_OFFSET_COMMAND:
1038         pvscsi_on_command(s, val);
1039         break;
1040 
1041     case PVSCSI_REG_OFFSET_COMMAND_DATA:
1042         pvscsi_on_command_data(s, (uint32_t) val);
1043         break;
1044 
1045     case PVSCSI_REG_OFFSET_INTR_STATUS:
1046         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_STATUS", val);
1047         s->reg_interrupt_status &= ~val;
1048         pvscsi_update_irq_status(s);
1049         pvscsi_schedule_completion_processing(s);
1050         break;
1051 
1052     case PVSCSI_REG_OFFSET_INTR_MASK:
1053         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_MASK", val);
1054         s->reg_interrupt_enabled = val;
1055         pvscsi_update_irq_status(s);
1056         break;
1057 
1058     case PVSCSI_REG_OFFSET_KICK_NON_RW_IO:
1059         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_NON_RW_IO", val);
1060         pvscsi_process_io(s);
1061         break;
1062 
1063     case PVSCSI_REG_OFFSET_KICK_RW_IO:
1064         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_RW_IO", val);
1065         pvscsi_process_io(s);
1066         break;
1067 
1068     case PVSCSI_REG_OFFSET_DEBUG:
1069         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_DEBUG", val);
1070         break;
1071 
1072     default:
1073         trace_pvscsi_io_write_unknown(addr, size, val);
1074         break;
1075     }
1076 
1077 }
1078 
1079 static uint64_t
1080 pvscsi_io_read(void *opaque, hwaddr addr, unsigned size)
1081 {
1082     PVSCSIState *s = opaque;
1083 
1084     switch (addr) {
1085     case PVSCSI_REG_OFFSET_INTR_STATUS:
1086         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_STATUS",
1087                              s->reg_interrupt_status);
1088         return s->reg_interrupt_status;
1089 
1090     case PVSCSI_REG_OFFSET_INTR_MASK:
1091         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_MASK",
1092                              s->reg_interrupt_status);
1093         return s->reg_interrupt_enabled;
1094 
1095     case PVSCSI_REG_OFFSET_COMMAND_STATUS:
1096         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_COMMAND_STATUS",
1097                              s->reg_interrupt_status);
1098         return s->reg_command_status;
1099 
1100     default:
1101         trace_pvscsi_io_read_unknown(addr, size);
1102         return 0;
1103     }
1104 }
1105 
1106 
1107 static void
1108 pvscsi_init_msi(PVSCSIState *s)
1109 {
1110     int res;
1111     PCIDevice *d = PCI_DEVICE(s);
1112 
1113     res = msi_init(d, PVSCSI_MSI_OFFSET(s), PVSCSI_MSIX_NUM_VECTORS,
1114                    PVSCSI_USE_64BIT, PVSCSI_PER_VECTOR_MASK, NULL);
1115     if (res < 0) {
1116         trace_pvscsi_init_msi_fail(res);
1117         s->msi_used = false;
1118     } else {
1119         s->msi_used = true;
1120     }
1121 }
1122 
1123 static void
1124 pvscsi_cleanup_msi(PVSCSIState *s)
1125 {
1126     PCIDevice *d = PCI_DEVICE(s);
1127 
1128     msi_uninit(d);
1129 }
1130 
1131 static const MemoryRegionOps pvscsi_ops = {
1132         .read = pvscsi_io_read,
1133         .write = pvscsi_io_write,
1134         .endianness = DEVICE_LITTLE_ENDIAN,
1135         .impl = {
1136                 .min_access_size = 4,
1137                 .max_access_size = 4,
1138         },
1139 };
1140 
1141 static const struct SCSIBusInfo pvscsi_scsi_info = {
1142         .tcq = true,
1143         .max_target = PVSCSI_MAX_DEVS,
1144         .max_channel = 0,
1145         .max_lun = 0,
1146 
1147         .get_sg_list = pvscsi_get_sg_list,
1148         .complete = pvscsi_command_complete,
1149         .cancel = pvscsi_request_cancelled,
1150         .fail = pvscsi_command_failed,
1151 };
1152 
1153 static void
1154 pvscsi_realizefn(PCIDevice *pci_dev, Error **errp)
1155 {
1156     PVSCSIState *s = PVSCSI(pci_dev);
1157 
1158     trace_pvscsi_state("init");
1159 
1160     /* PCI subsystem ID, subsystem vendor ID, revision */
1161     if (PVSCSI_USE_OLD_PCI_CONFIGURATION(s)) {
1162         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID, 0x1000);
1163     } else {
1164         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1165                      PCI_VENDOR_ID_VMWARE);
1166         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1167                      PCI_DEVICE_ID_VMWARE_PVSCSI);
1168         pci_config_set_revision(pci_dev->config, 0x2);
1169     }
1170 
1171     /* PCI latency timer = 255 */
1172     pci_dev->config[PCI_LATENCY_TIMER] = 0xff;
1173 
1174     /* Interrupt pin A */
1175     pci_config_set_interrupt_pin(pci_dev->config, 1);
1176 
1177     memory_region_init_io(&s->io_space, OBJECT(s), &pvscsi_ops, s,
1178                           "pvscsi-io", PVSCSI_MEM_SPACE_SIZE);
1179     pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->io_space);
1180 
1181     pvscsi_init_msi(s);
1182 
1183     if (pci_is_express(pci_dev) && pci_bus_is_express(pci_get_bus(pci_dev))) {
1184         pcie_endpoint_cap_init(pci_dev, PVSCSI_EXP_EP_OFFSET);
1185     }
1186 
1187     s->completion_worker = qemu_bh_new(pvscsi_process_completion_queue, s);
1188 
1189     scsi_bus_init(&s->bus, sizeof(s->bus), DEVICE(pci_dev), &pvscsi_scsi_info);
1190     /* override default SCSI bus hotplug-handler, with pvscsi's one */
1191     qbus_set_hotplug_handler(BUS(&s->bus), OBJECT(s));
1192     pvscsi_reset_state(s);
1193 }
1194 
1195 static void
1196 pvscsi_uninit(PCIDevice *pci_dev)
1197 {
1198     PVSCSIState *s = PVSCSI(pci_dev);
1199 
1200     trace_pvscsi_state("uninit");
1201     qemu_bh_delete(s->completion_worker);
1202 
1203     pvscsi_cleanup_msi(s);
1204 }
1205 
1206 static void
1207 pvscsi_reset(DeviceState *dev)
1208 {
1209     PCIDevice *d = PCI_DEVICE(dev);
1210     PVSCSIState *s = PVSCSI(d);
1211 
1212     trace_pvscsi_state("reset");
1213     pvscsi_reset_adapter(s);
1214 }
1215 
1216 static int
1217 pvscsi_pre_save(void *opaque)
1218 {
1219     PVSCSIState *s = (PVSCSIState *) opaque;
1220 
1221     trace_pvscsi_state("presave");
1222 
1223     assert(QTAILQ_EMPTY(&s->pending_queue));
1224     assert(QTAILQ_EMPTY(&s->completion_queue));
1225 
1226     return 0;
1227 }
1228 
1229 static int
1230 pvscsi_post_load(void *opaque, int version_id)
1231 {
1232     trace_pvscsi_state("postload");
1233     return 0;
1234 }
1235 
1236 static bool pvscsi_vmstate_need_pcie_device(void *opaque)
1237 {
1238     PVSCSIState *s = PVSCSI(opaque);
1239 
1240     return !(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE);
1241 }
1242 
1243 static bool pvscsi_vmstate_test_pci_device(void *opaque, int version_id)
1244 {
1245     return !pvscsi_vmstate_need_pcie_device(opaque);
1246 }
1247 
1248 static const VMStateDescription vmstate_pvscsi_pcie_device = {
1249     .name = "pvscsi/pcie",
1250     .needed = pvscsi_vmstate_need_pcie_device,
1251     .fields = (VMStateField[]) {
1252         VMSTATE_PCI_DEVICE(parent_obj, PVSCSIState),
1253         VMSTATE_END_OF_LIST()
1254     }
1255 };
1256 
1257 static const VMStateDescription vmstate_pvscsi = {
1258     .name = "pvscsi",
1259     .version_id = 0,
1260     .minimum_version_id = 0,
1261     .pre_save = pvscsi_pre_save,
1262     .post_load = pvscsi_post_load,
1263     .fields = (VMStateField[]) {
1264         VMSTATE_STRUCT_TEST(parent_obj, PVSCSIState,
1265                             pvscsi_vmstate_test_pci_device, 0,
1266                             vmstate_pci_device, PCIDevice),
1267         VMSTATE_UINT8(msi_used, PVSCSIState),
1268         VMSTATE_UINT32(resetting, PVSCSIState),
1269         VMSTATE_UINT64(reg_interrupt_status, PVSCSIState),
1270         VMSTATE_UINT64(reg_interrupt_enabled, PVSCSIState),
1271         VMSTATE_UINT64(reg_command_status, PVSCSIState),
1272         VMSTATE_UINT64(curr_cmd, PVSCSIState),
1273         VMSTATE_UINT32(curr_cmd_data_cntr, PVSCSIState),
1274         VMSTATE_UINT32_ARRAY(curr_cmd_data, PVSCSIState,
1275                              ARRAY_SIZE(((PVSCSIState *)NULL)->curr_cmd_data)),
1276         VMSTATE_UINT8(rings_info_valid, PVSCSIState),
1277         VMSTATE_UINT8(msg_ring_info_valid, PVSCSIState),
1278         VMSTATE_UINT8(use_msg, PVSCSIState),
1279 
1280         VMSTATE_UINT64(rings.rs_pa, PVSCSIState),
1281         VMSTATE_UINT32(rings.txr_len_mask, PVSCSIState),
1282         VMSTATE_UINT32(rings.rxr_len_mask, PVSCSIState),
1283         VMSTATE_UINT64_ARRAY(rings.req_ring_pages_pa, PVSCSIState,
1284                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1285         VMSTATE_UINT64_ARRAY(rings.cmp_ring_pages_pa, PVSCSIState,
1286                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1287         VMSTATE_UINT64(rings.consumed_ptr, PVSCSIState),
1288         VMSTATE_UINT64(rings.filled_cmp_ptr, PVSCSIState),
1289 
1290         VMSTATE_END_OF_LIST()
1291     },
1292     .subsections = (const VMStateDescription*[]) {
1293         &vmstate_pvscsi_pcie_device,
1294         NULL
1295     }
1296 };
1297 
1298 static Property pvscsi_properties[] = {
1299     DEFINE_PROP_UINT8("use_msg", PVSCSIState, use_msg, 1),
1300     DEFINE_PROP_BIT("x-old-pci-configuration", PVSCSIState, compat_flags,
1301                     PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT, false),
1302     DEFINE_PROP_BIT("x-disable-pcie", PVSCSIState, compat_flags,
1303                     PVSCSI_COMPAT_DISABLE_PCIE_BIT, false),
1304     DEFINE_PROP_END_OF_LIST(),
1305 };
1306 
1307 static void pvscsi_realize(DeviceState *qdev, Error **errp)
1308 {
1309     PVSCSIClass *pvs_c = PVSCSI_GET_CLASS(qdev);
1310     PCIDevice *pci_dev = PCI_DEVICE(qdev);
1311     PVSCSIState *s = PVSCSI(qdev);
1312 
1313     if (!(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE)) {
1314         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1315     }
1316 
1317     pvs_c->parent_dc_realize(qdev, errp);
1318 }
1319 
1320 static void pvscsi_class_init(ObjectClass *klass, void *data)
1321 {
1322     DeviceClass *dc = DEVICE_CLASS(klass);
1323     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1324     PVSCSIClass *pvs_k = PVSCSI_CLASS(klass);
1325     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
1326 
1327     k->realize = pvscsi_realizefn;
1328     k->exit = pvscsi_uninit;
1329     k->vendor_id = PCI_VENDOR_ID_VMWARE;
1330     k->device_id = PCI_DEVICE_ID_VMWARE_PVSCSI;
1331     k->class_id = PCI_CLASS_STORAGE_SCSI;
1332     k->subsystem_id = 0x1000;
1333     device_class_set_parent_realize(dc, pvscsi_realize,
1334                                     &pvs_k->parent_dc_realize);
1335     dc->reset = pvscsi_reset;
1336     dc->vmsd = &vmstate_pvscsi;
1337     device_class_set_props(dc, pvscsi_properties);
1338     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1339     hc->unplug = pvscsi_hot_unplug;
1340     hc->plug = pvscsi_hotplug;
1341 }
1342 
1343 static const TypeInfo pvscsi_info = {
1344     .name          = TYPE_PVSCSI,
1345     .parent        = TYPE_PCI_DEVICE,
1346     .class_size    = sizeof(PVSCSIClass),
1347     .instance_size = sizeof(PVSCSIState),
1348     .class_init    = pvscsi_class_init,
1349     .interfaces = (InterfaceInfo[]) {
1350         { TYPE_HOTPLUG_HANDLER },
1351         { INTERFACE_PCIE_DEVICE },
1352         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
1353         { }
1354     }
1355 };
1356 
1357 static void
1358 pvscsi_register_types(void)
1359 {
1360     type_register_static(&pvscsi_info);
1361 }
1362 
1363 type_init(pvscsi_register_types);
1364