xref: /openbmc/qemu/hw/scsi/vmw_pvscsi.c (revision f7160f32)
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 
40 
41 #define PVSCSI_USE_64BIT         (true)
42 #define PVSCSI_PER_VECTOR_MASK   (false)
43 
44 #define PVSCSI_MAX_DEVS                   (64)
45 #define PVSCSI_MSIX_NUM_VECTORS           (1)
46 
47 #define PVSCSI_MAX_SG_ELEM                2048
48 
49 #define PVSCSI_MAX_CMD_DATA_WORDS \
50     (sizeof(PVSCSICmdDescSetupRings)/sizeof(uint32_t))
51 
52 #define RS_GET_FIELD(m, field) \
53     (ldl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
54                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field)))
55 #define RS_SET_FIELD(m, field, val) \
56     (stl_le_pci_dma(&container_of(m, PVSCSIState, rings)->parent_obj, \
57                  (m)->rs_pa + offsetof(struct PVSCSIRingsState, field), val))
58 
59 typedef struct PVSCSIClass {
60     PCIDeviceClass parent_class;
61     DeviceRealize parent_dc_realize;
62 } PVSCSIClass;
63 
64 #define TYPE_PVSCSI "pvscsi"
65 #define PVSCSI(obj) OBJECT_CHECK(PVSCSIState, (obj), TYPE_PVSCSI)
66 
67 #define PVSCSI_DEVICE_CLASS(klass) \
68     OBJECT_CLASS_CHECK(PVSCSIClass, (klass), TYPE_PVSCSI)
69 #define PVSCSI_DEVICE_GET_CLASS(obj) \
70     OBJECT_GET_CLASS(PVSCSIClass, (obj), TYPE_PVSCSI)
71 
72 /* Compatibility flags for migration */
73 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT 0
74 #define PVSCSI_COMPAT_OLD_PCI_CONFIGURATION \
75     (1 << PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT)
76 #define PVSCSI_COMPAT_DISABLE_PCIE_BIT 1
77 #define PVSCSI_COMPAT_DISABLE_PCIE \
78     (1 << PVSCSI_COMPAT_DISABLE_PCIE_BIT)
79 
80 #define PVSCSI_USE_OLD_PCI_CONFIGURATION(s) \
81     ((s)->compat_flags & PVSCSI_COMPAT_OLD_PCI_CONFIGURATION)
82 #define PVSCSI_MSI_OFFSET(s) \
83     (PVSCSI_USE_OLD_PCI_CONFIGURATION(s) ? 0x50 : 0x7c)
84 #define PVSCSI_EXP_EP_OFFSET (0x40)
85 
86 typedef struct PVSCSIRingInfo {
87     uint64_t            rs_pa;
88     uint32_t            txr_len_mask;
89     uint32_t            rxr_len_mask;
90     uint32_t            msg_len_mask;
91     uint64_t            req_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
92     uint64_t            cmp_ring_pages_pa[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
93     uint64_t            msg_ring_pages_pa[PVSCSI_SETUP_MSG_RING_MAX_NUM_PAGES];
94     uint64_t            consumed_ptr;
95     uint64_t            filled_cmp_ptr;
96     uint64_t            filled_msg_ptr;
97 } PVSCSIRingInfo;
98 
99 typedef struct PVSCSISGState {
100     hwaddr elemAddr;
101     hwaddr dataAddr;
102     uint32_t resid;
103 } PVSCSISGState;
104 
105 typedef QTAILQ_HEAD(, PVSCSIRequest) PVSCSIRequestList;
106 
107 typedef struct {
108     PCIDevice parent_obj;
109     MemoryRegion io_space;
110     SCSIBus bus;
111     QEMUBH *completion_worker;
112     PVSCSIRequestList pending_queue;
113     PVSCSIRequestList completion_queue;
114 
115     uint64_t reg_interrupt_status;        /* Interrupt status register value */
116     uint64_t reg_interrupt_enabled;       /* Interrupt mask register value   */
117     uint64_t reg_command_status;          /* Command status register value   */
118 
119     /* Command data adoption mechanism */
120     uint64_t curr_cmd;                   /* Last command arrived             */
121     uint32_t curr_cmd_data_cntr;         /* Amount of data for last command  */
122 
123     /* Collector for current command data */
124     uint32_t curr_cmd_data[PVSCSI_MAX_CMD_DATA_WORDS];
125 
126     uint8_t rings_info_valid;            /* Whether data rings initialized   */
127     uint8_t msg_ring_info_valid;         /* Whether message ring initialized */
128     uint8_t use_msg;                     /* Whether to use message ring      */
129 
130     uint8_t msi_used;                    /* For migration compatibility      */
131     PVSCSIRingInfo rings;                /* Data transfer rings manager      */
132     uint32_t resetting;                  /* Reset in progress                */
133 
134     uint32_t compat_flags;
135 } PVSCSIState;
136 
137 typedef struct PVSCSIRequest {
138     SCSIRequest *sreq;
139     PVSCSIState *dev;
140     uint8_t sense_key;
141     uint8_t completed;
142     int lun;
143     QEMUSGList sgl;
144     PVSCSISGState sg;
145     struct PVSCSIRingReqDesc req;
146     struct PVSCSIRingCmpDesc cmp;
147     QTAILQ_ENTRY(PVSCSIRequest) next;
148 } PVSCSIRequest;
149 
150 /* Integer binary logarithm */
151 static int
152 pvscsi_log2(uint32_t input)
153 {
154     int log = 0;
155     assert(input > 0);
156     while (input >> ++log) {
157     }
158     return log;
159 }
160 
161 static void
162 pvscsi_ring_init_data(PVSCSIRingInfo *m, PVSCSICmdDescSetupRings *ri)
163 {
164     int i;
165     uint32_t txr_len_log2, rxr_len_log2;
166     uint32_t req_ring_size, cmp_ring_size;
167     m->rs_pa = ri->ringsStatePPN << VMW_PAGE_SHIFT;
168 
169     req_ring_size = ri->reqRingNumPages * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
170     cmp_ring_size = ri->cmpRingNumPages * PVSCSI_MAX_NUM_CMP_ENTRIES_PER_PAGE;
171     txr_len_log2 = pvscsi_log2(req_ring_size - 1);
172     rxr_len_log2 = pvscsi_log2(cmp_ring_size - 1);
173 
174     m->txr_len_mask = MASK(txr_len_log2);
175     m->rxr_len_mask = MASK(rxr_len_log2);
176 
177     m->consumed_ptr = 0;
178     m->filled_cmp_ptr = 0;
179 
180     for (i = 0; i < ri->reqRingNumPages; i++) {
181         m->req_ring_pages_pa[i] = ri->reqRingPPNs[i] << VMW_PAGE_SHIFT;
182     }
183 
184     for (i = 0; i < ri->cmpRingNumPages; i++) {
185         m->cmp_ring_pages_pa[i] = ri->cmpRingPPNs[i] << VMW_PAGE_SHIFT;
186     }
187 
188     RS_SET_FIELD(m, reqProdIdx, 0);
189     RS_SET_FIELD(m, reqConsIdx, 0);
190     RS_SET_FIELD(m, reqNumEntriesLog2, txr_len_log2);
191 
192     RS_SET_FIELD(m, cmpProdIdx, 0);
193     RS_SET_FIELD(m, cmpConsIdx, 0);
194     RS_SET_FIELD(m, cmpNumEntriesLog2, rxr_len_log2);
195 
196     trace_pvscsi_ring_init_data(txr_len_log2, rxr_len_log2);
197 
198     /* Flush ring state page changes */
199     smp_wmb();
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 || 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     uint32_t ring_size = PVSCSI_MAX_NUM_PAGES_REQ_RING
255                             * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
256 
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 = RS_GET_FIELD(mgr, msgProdIdx);
328     uint32_t consIdx = RS_GET_FIELD(mgr, msgConsIdx);
329 
330     return (prodIdx - consIdx) < (mgr->msg_len_mask + 1);
331 }
332 
333 static void
334 pvscsi_ring_flush_msg(PVSCSIRingInfo *mgr)
335 {
336     /* Flush descriptor changes */
337     smp_wmb();
338 
339     trace_pvscsi_ring_flush_msg(mgr->filled_msg_ptr);
340 
341     RS_SET_FIELD(mgr, msgProdIdx, mgr->filled_msg_ptr);
342 }
343 
344 static void
345 pvscsi_reset_state(PVSCSIState *s)
346 {
347     s->curr_cmd = PVSCSI_CMD_FIRST;
348     s->curr_cmd_data_cntr = 0;
349     s->reg_command_status = PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
350     s->reg_interrupt_status = 0;
351     pvscsi_ring_cleanup(&s->rings);
352     s->rings_info_valid = FALSE;
353     s->msg_ring_info_valid = FALSE;
354     QTAILQ_INIT(&s->pending_queue);
355     QTAILQ_INIT(&s->completion_queue);
356 }
357 
358 static void
359 pvscsi_update_irq_status(PVSCSIState *s)
360 {
361     PCIDevice *d = PCI_DEVICE(s);
362     bool should_raise = s->reg_interrupt_enabled & s->reg_interrupt_status;
363 
364     trace_pvscsi_update_irq_level(should_raise, s->reg_interrupt_enabled,
365                                   s->reg_interrupt_status);
366 
367     if (msi_enabled(d)) {
368         if (should_raise) {
369             trace_pvscsi_update_irq_msi();
370             msi_notify(d, PVSCSI_VECTOR_COMPLETION);
371         }
372         return;
373     }
374 
375     pci_set_irq(d, !!should_raise);
376 }
377 
378 static void
379 pvscsi_raise_completion_interrupt(PVSCSIState *s)
380 {
381     s->reg_interrupt_status |= PVSCSI_INTR_CMPL_0;
382 
383     /* Memory barrier to flush interrupt status register changes*/
384     smp_wmb();
385 
386     pvscsi_update_irq_status(s);
387 }
388 
389 static void
390 pvscsi_raise_message_interrupt(PVSCSIState *s)
391 {
392     s->reg_interrupt_status |= PVSCSI_INTR_MSG_0;
393 
394     /* Memory barrier to flush interrupt status register changes*/
395     smp_wmb();
396 
397     pvscsi_update_irq_status(s);
398 }
399 
400 static void
401 pvscsi_cmp_ring_put(PVSCSIState *s, struct PVSCSIRingCmpDesc *cmp_desc)
402 {
403     hwaddr cmp_descr_pa;
404 
405     cmp_descr_pa = pvscsi_ring_pop_cmp_descr(&s->rings);
406     trace_pvscsi_cmp_ring_put(cmp_descr_pa);
407     cpu_physical_memory_write(cmp_descr_pa, cmp_desc, sizeof(*cmp_desc));
408 }
409 
410 static void
411 pvscsi_msg_ring_put(PVSCSIState *s, struct PVSCSIRingMsgDesc *msg_desc)
412 {
413     hwaddr msg_descr_pa;
414 
415     msg_descr_pa = pvscsi_ring_pop_msg_descr(&s->rings);
416     trace_pvscsi_msg_ring_put(msg_descr_pa);
417     cpu_physical_memory_write(msg_descr_pa, msg_desc, sizeof(*msg_desc));
418 }
419 
420 static void
421 pvscsi_process_completion_queue(void *opaque)
422 {
423     PVSCSIState *s = opaque;
424     PVSCSIRequest *pvscsi_req;
425     bool has_completed = false;
426 
427     while (!QTAILQ_EMPTY(&s->completion_queue)) {
428         pvscsi_req = QTAILQ_FIRST(&s->completion_queue);
429         QTAILQ_REMOVE(&s->completion_queue, pvscsi_req, next);
430         pvscsi_cmp_ring_put(s, &pvscsi_req->cmp);
431         g_free(pvscsi_req);
432         has_completed = true;
433     }
434 
435     if (has_completed) {
436         pvscsi_ring_flush_cmp(&s->rings);
437         pvscsi_raise_completion_interrupt(s);
438     }
439 }
440 
441 static void
442 pvscsi_reset_adapter(PVSCSIState *s)
443 {
444     s->resetting++;
445     qbus_reset_all(BUS(&s->bus));
446     s->resetting--;
447     pvscsi_process_completion_queue(s);
448     assert(QTAILQ_EMPTY(&s->pending_queue));
449     pvscsi_reset_state(s);
450 }
451 
452 static void
453 pvscsi_schedule_completion_processing(PVSCSIState *s)
454 {
455     /* Try putting more complete requests on the ring. */
456     if (!QTAILQ_EMPTY(&s->completion_queue)) {
457         qemu_bh_schedule(s->completion_worker);
458     }
459 }
460 
461 static void
462 pvscsi_complete_request(PVSCSIState *s, PVSCSIRequest *r)
463 {
464     assert(!r->completed);
465 
466     trace_pvscsi_complete_request(r->cmp.context, r->cmp.dataLen,
467                                   r->sense_key);
468     if (r->sreq != NULL) {
469         scsi_req_unref(r->sreq);
470         r->sreq = NULL;
471     }
472     r->completed = 1;
473     QTAILQ_REMOVE(&s->pending_queue, r, next);
474     QTAILQ_INSERT_TAIL(&s->completion_queue, r, next);
475     pvscsi_schedule_completion_processing(s);
476 }
477 
478 static QEMUSGList *pvscsi_get_sg_list(SCSIRequest *r)
479 {
480     PVSCSIRequest *req = r->hba_private;
481 
482     trace_pvscsi_get_sg_list(req->sgl.nsg, req->sgl.size);
483 
484     return &req->sgl;
485 }
486 
487 static void
488 pvscsi_get_next_sg_elem(PVSCSISGState *sg)
489 {
490     struct PVSCSISGElement elem;
491 
492     cpu_physical_memory_read(sg->elemAddr, &elem, sizeof(elem));
493     if ((elem.flags & ~PVSCSI_KNOWN_FLAGS) != 0) {
494         /*
495             * There is PVSCSI_SGE_FLAG_CHAIN_ELEMENT flag described in
496             * header file but its value is unknown. This flag requires
497             * additional processing, so we put warning here to catch it
498             * some day and make proper implementation
499             */
500         trace_pvscsi_get_next_sg_elem(elem.flags);
501     }
502 
503     sg->elemAddr += sizeof(elem);
504     sg->dataAddr = elem.addr;
505     sg->resid = elem.length;
506 }
507 
508 static void
509 pvscsi_write_sense(PVSCSIRequest *r, uint8_t *sense, int len)
510 {
511     r->cmp.senseLen = MIN(r->req.senseLen, len);
512     r->sense_key = sense[(sense[0] & 2) ? 1 : 2];
513     cpu_physical_memory_write(r->req.senseAddr, sense, r->cmp.senseLen);
514 }
515 
516 static void
517 pvscsi_command_complete(SCSIRequest *req, uint32_t status, size_t resid)
518 {
519     PVSCSIRequest *pvscsi_req = req->hba_private;
520     PVSCSIState *s;
521 
522     if (!pvscsi_req) {
523         trace_pvscsi_command_complete_not_found(req->tag);
524         return;
525     }
526     s = pvscsi_req->dev;
527 
528     if (resid) {
529         /* Short transfer.  */
530         trace_pvscsi_command_complete_data_run();
531         pvscsi_req->cmp.hostStatus = BTSTAT_DATARUN;
532     }
533 
534     pvscsi_req->cmp.scsiStatus = status;
535     if (pvscsi_req->cmp.scsiStatus == CHECK_CONDITION) {
536         uint8_t sense[SCSI_SENSE_BUF_SIZE];
537         int sense_len =
538             scsi_req_get_sense(pvscsi_req->sreq, sense, sizeof(sense));
539 
540         trace_pvscsi_command_complete_sense_len(sense_len);
541         pvscsi_write_sense(pvscsi_req, sense, sense_len);
542     }
543     qemu_sglist_destroy(&pvscsi_req->sgl);
544     pvscsi_complete_request(s, pvscsi_req);
545 }
546 
547 static void
548 pvscsi_send_msg(PVSCSIState *s, SCSIDevice *dev, uint32_t msg_type)
549 {
550     if (s->msg_ring_info_valid && pvscsi_ring_msg_has_room(&s->rings)) {
551         PVSCSIMsgDescDevStatusChanged msg = {0};
552 
553         msg.type = msg_type;
554         msg.bus = dev->channel;
555         msg.target = dev->id;
556         msg.lun[1] = dev->lun;
557 
558         pvscsi_msg_ring_put(s, (PVSCSIRingMsgDesc *)&msg);
559         pvscsi_ring_flush_msg(&s->rings);
560         pvscsi_raise_message_interrupt(s);
561     }
562 }
563 
564 static void
565 pvscsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
566 {
567     PVSCSIState *s = PVSCSI(hotplug_dev);
568 
569     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_ADDED);
570 }
571 
572 static void
573 pvscsi_hot_unplug(HotplugHandler *hotplug_dev, DeviceState *dev, Error **errp)
574 {
575     PVSCSIState *s = PVSCSI(hotplug_dev);
576 
577     pvscsi_send_msg(s, SCSI_DEVICE(dev), PVSCSI_MSG_DEV_REMOVED);
578     qdev_simple_device_unplug_cb(hotplug_dev, dev, errp);
579 }
580 
581 static void
582 pvscsi_request_cancelled(SCSIRequest *req)
583 {
584     PVSCSIRequest *pvscsi_req = req->hba_private;
585     PVSCSIState *s = pvscsi_req->dev;
586 
587     if (pvscsi_req->completed) {
588         return;
589     }
590 
591    if (pvscsi_req->dev->resetting) {
592        pvscsi_req->cmp.hostStatus = BTSTAT_BUSRESET;
593     } else {
594        pvscsi_req->cmp.hostStatus = BTSTAT_ABORTQUEUE;
595     }
596 
597     pvscsi_complete_request(s, pvscsi_req);
598 }
599 
600 static SCSIDevice*
601 pvscsi_device_find(PVSCSIState *s, int channel, int target,
602                    uint8_t *requested_lun, uint8_t *target_lun)
603 {
604     if (requested_lun[0] || requested_lun[2] || requested_lun[3] ||
605         requested_lun[4] || requested_lun[5] || requested_lun[6] ||
606         requested_lun[7] || (target > PVSCSI_MAX_DEVS)) {
607         return NULL;
608     } else {
609         *target_lun = requested_lun[1];
610         return scsi_device_find(&s->bus, channel, target, *target_lun);
611     }
612 }
613 
614 static PVSCSIRequest *
615 pvscsi_queue_pending_descriptor(PVSCSIState *s, SCSIDevice **d,
616                                 struct PVSCSIRingReqDesc *descr)
617 {
618     PVSCSIRequest *pvscsi_req;
619     uint8_t lun;
620 
621     pvscsi_req = g_malloc0(sizeof(*pvscsi_req));
622     pvscsi_req->dev = s;
623     pvscsi_req->req = *descr;
624     pvscsi_req->cmp.context = pvscsi_req->req.context;
625     QTAILQ_INSERT_TAIL(&s->pending_queue, pvscsi_req, next);
626 
627     *d = pvscsi_device_find(s, descr->bus, descr->target, descr->lun, &lun);
628     if (*d) {
629         pvscsi_req->lun = lun;
630     }
631 
632     return pvscsi_req;
633 }
634 
635 static void
636 pvscsi_convert_sglist(PVSCSIRequest *r)
637 {
638     uint32_t chunk_size, elmcnt = 0;
639     uint64_t data_length = r->req.dataLen;
640     PVSCSISGState sg = r->sg;
641     while (data_length && elmcnt < PVSCSI_MAX_SG_ELEM) {
642         while (!sg.resid && elmcnt++ < PVSCSI_MAX_SG_ELEM) {
643             pvscsi_get_next_sg_elem(&sg);
644             trace_pvscsi_convert_sglist(r->req.context, r->sg.dataAddr,
645                                         r->sg.resid);
646         }
647         chunk_size = MIN(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     if (!s->rings_info_valid) {
723         return;
724     }
725 
726     while ((next_descr_pa = pvscsi_ring_pop_req_descr(&s->rings)) != 0) {
727 
728         /* Only read after production index verification */
729         smp_rmb();
730 
731         trace_pvscsi_process_io(next_descr_pa);
732         cpu_physical_memory_read(next_descr_pa, &descr, sizeof(descr));
733         pvscsi_process_request_descriptor(s, &descr);
734     }
735 
736     pvscsi_ring_flush_req(&s->rings);
737 }
738 
739 static void
740 pvscsi_dbg_dump_tx_rings_config(PVSCSICmdDescSetupRings *rc)
741 {
742     int i;
743     trace_pvscsi_tx_rings_ppn("Rings State", rc->ringsStatePPN);
744 
745     trace_pvscsi_tx_rings_num_pages("Request Ring", rc->reqRingNumPages);
746     for (i = 0; i < rc->reqRingNumPages; i++) {
747         trace_pvscsi_tx_rings_ppn("Request Ring", rc->reqRingPPNs[i]);
748     }
749 
750     trace_pvscsi_tx_rings_num_pages("Confirm Ring", rc->cmpRingNumPages);
751     for (i = 0; i < rc->cmpRingNumPages; i++) {
752         trace_pvscsi_tx_rings_ppn("Confirm Ring", rc->cmpRingPPNs[i]);
753     }
754 }
755 
756 static uint64_t
757 pvscsi_on_cmd_config(PVSCSIState *s)
758 {
759     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_CONFIG");
760     return PVSCSI_COMMAND_PROCESSING_FAILED;
761 }
762 
763 static uint64_t
764 pvscsi_on_cmd_unplug(PVSCSIState *s)
765 {
766     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_DEVICE_UNPLUG");
767     return PVSCSI_COMMAND_PROCESSING_FAILED;
768 }
769 
770 static uint64_t
771 pvscsi_on_issue_scsi(PVSCSIState *s)
772 {
773     trace_pvscsi_on_cmd_noimpl("PVSCSI_CMD_ISSUE_SCSI");
774     return PVSCSI_COMMAND_PROCESSING_FAILED;
775 }
776 
777 static uint64_t
778 pvscsi_on_cmd_setup_rings(PVSCSIState *s)
779 {
780     PVSCSICmdDescSetupRings *rc =
781         (PVSCSICmdDescSetupRings *) s->curr_cmd_data;
782 
783     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_RINGS");
784 
785     if (!rc->reqRingNumPages
786         || rc->reqRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES
787         || !rc->cmpRingNumPages
788         || rc->cmpRingNumPages > PVSCSI_SETUP_RINGS_MAX_NUM_PAGES) {
789         return PVSCSI_COMMAND_PROCESSING_FAILED;
790     }
791 
792     pvscsi_dbg_dump_tx_rings_config(rc);
793     pvscsi_ring_init_data(&s->rings, rc);
794 
795     s->rings_info_valid = TRUE;
796     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
797 }
798 
799 static uint64_t
800 pvscsi_on_cmd_abort(PVSCSIState *s)
801 {
802     PVSCSICmdDescAbortCmd *cmd = (PVSCSICmdDescAbortCmd *) s->curr_cmd_data;
803     PVSCSIRequest *r, *next;
804 
805     trace_pvscsi_on_cmd_abort(cmd->context, cmd->target);
806 
807     QTAILQ_FOREACH_SAFE(r, &s->pending_queue, next, next) {
808         if (r->req.context == cmd->context) {
809             break;
810         }
811     }
812     if (r) {
813         assert(!r->completed);
814         r->cmp.hostStatus = BTSTAT_ABORTQUEUE;
815         scsi_req_cancel(r->sreq);
816     }
817 
818     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
819 }
820 
821 static uint64_t
822 pvscsi_on_cmd_unknown(PVSCSIState *s)
823 {
824     trace_pvscsi_on_cmd_unknown_data(s->curr_cmd_data[0]);
825     return PVSCSI_COMMAND_PROCESSING_FAILED;
826 }
827 
828 static uint64_t
829 pvscsi_on_cmd_reset_device(PVSCSIState *s)
830 {
831     uint8_t target_lun = 0;
832     struct PVSCSICmdDescResetDevice *cmd =
833         (struct PVSCSICmdDescResetDevice *) s->curr_cmd_data;
834     SCSIDevice *sdev;
835 
836     sdev = pvscsi_device_find(s, 0, cmd->target, cmd->lun, &target_lun);
837 
838     trace_pvscsi_on_cmd_reset_dev(cmd->target, (int) target_lun, sdev);
839 
840     if (sdev != NULL) {
841         s->resetting++;
842         device_legacy_reset(&sdev->qdev);
843         s->resetting--;
844         return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
845     }
846 
847     return PVSCSI_COMMAND_PROCESSING_FAILED;
848 }
849 
850 static uint64_t
851 pvscsi_on_cmd_reset_bus(PVSCSIState *s)
852 {
853     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_RESET_BUS");
854 
855     s->resetting++;
856     qbus_reset_all(BUS(&s->bus));
857     s->resetting--;
858     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
859 }
860 
861 static uint64_t
862 pvscsi_on_cmd_setup_msg_ring(PVSCSIState *s)
863 {
864     PVSCSICmdDescSetupMsgRing *rc =
865         (PVSCSICmdDescSetupMsgRing *) s->curr_cmd_data;
866 
867     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_SETUP_MSG_RING");
868 
869     if (!s->use_msg) {
870         return PVSCSI_COMMAND_PROCESSING_FAILED;
871     }
872 
873     if (s->rings_info_valid) {
874         if (pvscsi_ring_init_msg(&s->rings, rc) < 0) {
875             return PVSCSI_COMMAND_PROCESSING_FAILED;
876         }
877         s->msg_ring_info_valid = TRUE;
878     }
879     return sizeof(PVSCSICmdDescSetupMsgRing) / sizeof(uint32_t);
880 }
881 
882 static uint64_t
883 pvscsi_on_cmd_adapter_reset(PVSCSIState *s)
884 {
885     trace_pvscsi_on_cmd_arrived("PVSCSI_CMD_ADAPTER_RESET");
886 
887     pvscsi_reset_adapter(s);
888     return PVSCSI_COMMAND_PROCESSING_SUCCEEDED;
889 }
890 
891 static const struct {
892     int       data_size;
893     uint64_t  (*handler_fn)(PVSCSIState *s);
894 } pvscsi_commands[] = {
895     [PVSCSI_CMD_FIRST] = {
896         .data_size = 0,
897         .handler_fn = pvscsi_on_cmd_unknown,
898     },
899 
900     /* Not implemented, data size defined based on what arrives on windows */
901     [PVSCSI_CMD_CONFIG] = {
902         .data_size = 6 * sizeof(uint32_t),
903         .handler_fn = pvscsi_on_cmd_config,
904     },
905 
906     /* Command not implemented, data size is unknown */
907     [PVSCSI_CMD_ISSUE_SCSI] = {
908         .data_size = 0,
909         .handler_fn = pvscsi_on_issue_scsi,
910     },
911 
912     /* Command not implemented, data size is unknown */
913     [PVSCSI_CMD_DEVICE_UNPLUG] = {
914         .data_size = 0,
915         .handler_fn = pvscsi_on_cmd_unplug,
916     },
917 
918     [PVSCSI_CMD_SETUP_RINGS] = {
919         .data_size = sizeof(PVSCSICmdDescSetupRings),
920         .handler_fn = pvscsi_on_cmd_setup_rings,
921     },
922 
923     [PVSCSI_CMD_RESET_DEVICE] = {
924         .data_size = sizeof(struct PVSCSICmdDescResetDevice),
925         .handler_fn = pvscsi_on_cmd_reset_device,
926     },
927 
928     [PVSCSI_CMD_RESET_BUS] = {
929         .data_size = 0,
930         .handler_fn = pvscsi_on_cmd_reset_bus,
931     },
932 
933     [PVSCSI_CMD_SETUP_MSG_RING] = {
934         .data_size = sizeof(PVSCSICmdDescSetupMsgRing),
935         .handler_fn = pvscsi_on_cmd_setup_msg_ring,
936     },
937 
938     [PVSCSI_CMD_ADAPTER_RESET] = {
939         .data_size = 0,
940         .handler_fn = pvscsi_on_cmd_adapter_reset,
941     },
942 
943     [PVSCSI_CMD_ABORT_CMD] = {
944         .data_size = sizeof(struct PVSCSICmdDescAbortCmd),
945         .handler_fn = pvscsi_on_cmd_abort,
946     },
947 };
948 
949 static void
950 pvscsi_do_command_processing(PVSCSIState *s)
951 {
952     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
953 
954     assert(s->curr_cmd < PVSCSI_CMD_LAST);
955     if (bytes_arrived >= pvscsi_commands[s->curr_cmd].data_size) {
956         s->reg_command_status = pvscsi_commands[s->curr_cmd].handler_fn(s);
957         s->curr_cmd = PVSCSI_CMD_FIRST;
958         s->curr_cmd_data_cntr   = 0;
959     }
960 }
961 
962 static void
963 pvscsi_on_command_data(PVSCSIState *s, uint32_t value)
964 {
965     size_t bytes_arrived = s->curr_cmd_data_cntr * sizeof(uint32_t);
966 
967     assert(bytes_arrived < sizeof(s->curr_cmd_data));
968     s->curr_cmd_data[s->curr_cmd_data_cntr++] = value;
969 
970     pvscsi_do_command_processing(s);
971 }
972 
973 static void
974 pvscsi_on_command(PVSCSIState *s, uint64_t cmd_id)
975 {
976     if ((cmd_id > PVSCSI_CMD_FIRST) && (cmd_id < PVSCSI_CMD_LAST)) {
977         s->curr_cmd = cmd_id;
978     } else {
979         s->curr_cmd = PVSCSI_CMD_FIRST;
980         trace_pvscsi_on_cmd_unknown(cmd_id);
981     }
982 
983     s->curr_cmd_data_cntr = 0;
984     s->reg_command_status = PVSCSI_COMMAND_NOT_ENOUGH_DATA;
985 
986     pvscsi_do_command_processing(s);
987 }
988 
989 static void
990 pvscsi_io_write(void *opaque, hwaddr addr,
991                 uint64_t val, unsigned size)
992 {
993     PVSCSIState *s = opaque;
994 
995     switch (addr) {
996     case PVSCSI_REG_OFFSET_COMMAND:
997         pvscsi_on_command(s, val);
998         break;
999 
1000     case PVSCSI_REG_OFFSET_COMMAND_DATA:
1001         pvscsi_on_command_data(s, (uint32_t) val);
1002         break;
1003 
1004     case PVSCSI_REG_OFFSET_INTR_STATUS:
1005         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_STATUS", val);
1006         s->reg_interrupt_status &= ~val;
1007         pvscsi_update_irq_status(s);
1008         pvscsi_schedule_completion_processing(s);
1009         break;
1010 
1011     case PVSCSI_REG_OFFSET_INTR_MASK:
1012         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_INTR_MASK", val);
1013         s->reg_interrupt_enabled = val;
1014         pvscsi_update_irq_status(s);
1015         break;
1016 
1017     case PVSCSI_REG_OFFSET_KICK_NON_RW_IO:
1018         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_NON_RW_IO", val);
1019         pvscsi_process_io(s);
1020         break;
1021 
1022     case PVSCSI_REG_OFFSET_KICK_RW_IO:
1023         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_KICK_RW_IO", val);
1024         pvscsi_process_io(s);
1025         break;
1026 
1027     case PVSCSI_REG_OFFSET_DEBUG:
1028         trace_pvscsi_io_write("PVSCSI_REG_OFFSET_DEBUG", val);
1029         break;
1030 
1031     default:
1032         trace_pvscsi_io_write_unknown(addr, size, val);
1033         break;
1034     }
1035 
1036 }
1037 
1038 static uint64_t
1039 pvscsi_io_read(void *opaque, hwaddr addr, unsigned size)
1040 {
1041     PVSCSIState *s = opaque;
1042 
1043     switch (addr) {
1044     case PVSCSI_REG_OFFSET_INTR_STATUS:
1045         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_STATUS",
1046                              s->reg_interrupt_status);
1047         return s->reg_interrupt_status;
1048 
1049     case PVSCSI_REG_OFFSET_INTR_MASK:
1050         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_INTR_MASK",
1051                              s->reg_interrupt_status);
1052         return s->reg_interrupt_enabled;
1053 
1054     case PVSCSI_REG_OFFSET_COMMAND_STATUS:
1055         trace_pvscsi_io_read("PVSCSI_REG_OFFSET_COMMAND_STATUS",
1056                              s->reg_interrupt_status);
1057         return s->reg_command_status;
1058 
1059     default:
1060         trace_pvscsi_io_read_unknown(addr, size);
1061         return 0;
1062     }
1063 }
1064 
1065 
1066 static void
1067 pvscsi_init_msi(PVSCSIState *s)
1068 {
1069     int res;
1070     PCIDevice *d = PCI_DEVICE(s);
1071 
1072     res = msi_init(d, PVSCSI_MSI_OFFSET(s), PVSCSI_MSIX_NUM_VECTORS,
1073                    PVSCSI_USE_64BIT, PVSCSI_PER_VECTOR_MASK, NULL);
1074     if (res < 0) {
1075         trace_pvscsi_init_msi_fail(res);
1076         s->msi_used = false;
1077     } else {
1078         s->msi_used = true;
1079     }
1080 }
1081 
1082 static void
1083 pvscsi_cleanup_msi(PVSCSIState *s)
1084 {
1085     PCIDevice *d = PCI_DEVICE(s);
1086 
1087     msi_uninit(d);
1088 }
1089 
1090 static const MemoryRegionOps pvscsi_ops = {
1091         .read = pvscsi_io_read,
1092         .write = pvscsi_io_write,
1093         .endianness = DEVICE_LITTLE_ENDIAN,
1094         .impl = {
1095                 .min_access_size = 4,
1096                 .max_access_size = 4,
1097         },
1098 };
1099 
1100 static const struct SCSIBusInfo pvscsi_scsi_info = {
1101         .tcq = true,
1102         .max_target = PVSCSI_MAX_DEVS,
1103         .max_channel = 0,
1104         .max_lun = 0,
1105 
1106         .get_sg_list = pvscsi_get_sg_list,
1107         .complete = pvscsi_command_complete,
1108         .cancel = pvscsi_request_cancelled,
1109 };
1110 
1111 static void
1112 pvscsi_realizefn(PCIDevice *pci_dev, Error **errp)
1113 {
1114     PVSCSIState *s = PVSCSI(pci_dev);
1115 
1116     trace_pvscsi_state("init");
1117 
1118     /* PCI subsystem ID, subsystem vendor ID, revision */
1119     if (PVSCSI_USE_OLD_PCI_CONFIGURATION(s)) {
1120         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID, 0x1000);
1121     } else {
1122         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1123                      PCI_VENDOR_ID_VMWARE);
1124         pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1125                      PCI_DEVICE_ID_VMWARE_PVSCSI);
1126         pci_config_set_revision(pci_dev->config, 0x2);
1127     }
1128 
1129     /* PCI latency timer = 255 */
1130     pci_dev->config[PCI_LATENCY_TIMER] = 0xff;
1131 
1132     /* Interrupt pin A */
1133     pci_config_set_interrupt_pin(pci_dev->config, 1);
1134 
1135     memory_region_init_io(&s->io_space, OBJECT(s), &pvscsi_ops, s,
1136                           "pvscsi-io", PVSCSI_MEM_SPACE_SIZE);
1137     pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->io_space);
1138 
1139     pvscsi_init_msi(s);
1140 
1141     if (pci_is_express(pci_dev) && pci_bus_is_express(pci_get_bus(pci_dev))) {
1142         pcie_endpoint_cap_init(pci_dev, PVSCSI_EXP_EP_OFFSET);
1143     }
1144 
1145     s->completion_worker = qemu_bh_new(pvscsi_process_completion_queue, s);
1146 
1147     scsi_bus_new(&s->bus, sizeof(s->bus), DEVICE(pci_dev),
1148                  &pvscsi_scsi_info, NULL);
1149     /* override default SCSI bus hotplug-handler, with pvscsi's one */
1150     qbus_set_hotplug_handler(BUS(&s->bus), OBJECT(s));
1151     pvscsi_reset_state(s);
1152 }
1153 
1154 static void
1155 pvscsi_uninit(PCIDevice *pci_dev)
1156 {
1157     PVSCSIState *s = PVSCSI(pci_dev);
1158 
1159     trace_pvscsi_state("uninit");
1160     qemu_bh_delete(s->completion_worker);
1161 
1162     pvscsi_cleanup_msi(s);
1163 }
1164 
1165 static void
1166 pvscsi_reset(DeviceState *dev)
1167 {
1168     PCIDevice *d = PCI_DEVICE(dev);
1169     PVSCSIState *s = PVSCSI(d);
1170 
1171     trace_pvscsi_state("reset");
1172     pvscsi_reset_adapter(s);
1173 }
1174 
1175 static int
1176 pvscsi_pre_save(void *opaque)
1177 {
1178     PVSCSIState *s = (PVSCSIState *) opaque;
1179 
1180     trace_pvscsi_state("presave");
1181 
1182     assert(QTAILQ_EMPTY(&s->pending_queue));
1183     assert(QTAILQ_EMPTY(&s->completion_queue));
1184 
1185     return 0;
1186 }
1187 
1188 static int
1189 pvscsi_post_load(void *opaque, int version_id)
1190 {
1191     trace_pvscsi_state("postload");
1192     return 0;
1193 }
1194 
1195 static bool pvscsi_vmstate_need_pcie_device(void *opaque)
1196 {
1197     PVSCSIState *s = PVSCSI(opaque);
1198 
1199     return !(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE);
1200 }
1201 
1202 static bool pvscsi_vmstate_test_pci_device(void *opaque, int version_id)
1203 {
1204     return !pvscsi_vmstate_need_pcie_device(opaque);
1205 }
1206 
1207 static const VMStateDescription vmstate_pvscsi_pcie_device = {
1208     .name = "pvscsi/pcie",
1209     .needed = pvscsi_vmstate_need_pcie_device,
1210     .fields = (VMStateField[]) {
1211         VMSTATE_PCI_DEVICE(parent_obj, PVSCSIState),
1212         VMSTATE_END_OF_LIST()
1213     }
1214 };
1215 
1216 static const VMStateDescription vmstate_pvscsi = {
1217     .name = "pvscsi",
1218     .version_id = 0,
1219     .minimum_version_id = 0,
1220     .pre_save = pvscsi_pre_save,
1221     .post_load = pvscsi_post_load,
1222     .fields = (VMStateField[]) {
1223         VMSTATE_STRUCT_TEST(parent_obj, PVSCSIState,
1224                             pvscsi_vmstate_test_pci_device, 0,
1225                             vmstate_pci_device, PCIDevice),
1226         VMSTATE_UINT8(msi_used, PVSCSIState),
1227         VMSTATE_UINT32(resetting, PVSCSIState),
1228         VMSTATE_UINT64(reg_interrupt_status, PVSCSIState),
1229         VMSTATE_UINT64(reg_interrupt_enabled, PVSCSIState),
1230         VMSTATE_UINT64(reg_command_status, PVSCSIState),
1231         VMSTATE_UINT64(curr_cmd, PVSCSIState),
1232         VMSTATE_UINT32(curr_cmd_data_cntr, PVSCSIState),
1233         VMSTATE_UINT32_ARRAY(curr_cmd_data, PVSCSIState,
1234                              ARRAY_SIZE(((PVSCSIState *)NULL)->curr_cmd_data)),
1235         VMSTATE_UINT8(rings_info_valid, PVSCSIState),
1236         VMSTATE_UINT8(msg_ring_info_valid, PVSCSIState),
1237         VMSTATE_UINT8(use_msg, PVSCSIState),
1238 
1239         VMSTATE_UINT64(rings.rs_pa, PVSCSIState),
1240         VMSTATE_UINT32(rings.txr_len_mask, PVSCSIState),
1241         VMSTATE_UINT32(rings.rxr_len_mask, PVSCSIState),
1242         VMSTATE_UINT64_ARRAY(rings.req_ring_pages_pa, PVSCSIState,
1243                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1244         VMSTATE_UINT64_ARRAY(rings.cmp_ring_pages_pa, PVSCSIState,
1245                              PVSCSI_SETUP_RINGS_MAX_NUM_PAGES),
1246         VMSTATE_UINT64(rings.consumed_ptr, PVSCSIState),
1247         VMSTATE_UINT64(rings.filled_cmp_ptr, PVSCSIState),
1248 
1249         VMSTATE_END_OF_LIST()
1250     },
1251     .subsections = (const VMStateDescription*[]) {
1252         &vmstate_pvscsi_pcie_device,
1253         NULL
1254     }
1255 };
1256 
1257 static Property pvscsi_properties[] = {
1258     DEFINE_PROP_UINT8("use_msg", PVSCSIState, use_msg, 1),
1259     DEFINE_PROP_BIT("x-old-pci-configuration", PVSCSIState, compat_flags,
1260                     PVSCSI_COMPAT_OLD_PCI_CONFIGURATION_BIT, false),
1261     DEFINE_PROP_BIT("x-disable-pcie", PVSCSIState, compat_flags,
1262                     PVSCSI_COMPAT_DISABLE_PCIE_BIT, false),
1263     DEFINE_PROP_END_OF_LIST(),
1264 };
1265 
1266 static void pvscsi_realize(DeviceState *qdev, Error **errp)
1267 {
1268     PVSCSIClass *pvs_c = PVSCSI_DEVICE_GET_CLASS(qdev);
1269     PCIDevice *pci_dev = PCI_DEVICE(qdev);
1270     PVSCSIState *s = PVSCSI(qdev);
1271 
1272     if (!(s->compat_flags & PVSCSI_COMPAT_DISABLE_PCIE)) {
1273         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1274     }
1275 
1276     pvs_c->parent_dc_realize(qdev, errp);
1277 }
1278 
1279 static void pvscsi_class_init(ObjectClass *klass, void *data)
1280 {
1281     DeviceClass *dc = DEVICE_CLASS(klass);
1282     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1283     PVSCSIClass *pvs_k = PVSCSI_DEVICE_CLASS(klass);
1284     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(klass);
1285 
1286     k->realize = pvscsi_realizefn;
1287     k->exit = pvscsi_uninit;
1288     k->vendor_id = PCI_VENDOR_ID_VMWARE;
1289     k->device_id = PCI_DEVICE_ID_VMWARE_PVSCSI;
1290     k->class_id = PCI_CLASS_STORAGE_SCSI;
1291     k->subsystem_id = 0x1000;
1292     device_class_set_parent_realize(dc, pvscsi_realize,
1293                                     &pvs_k->parent_dc_realize);
1294     dc->reset = pvscsi_reset;
1295     dc->vmsd = &vmstate_pvscsi;
1296     device_class_set_props(dc, pvscsi_properties);
1297     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1298     hc->unplug = pvscsi_hot_unplug;
1299     hc->plug = pvscsi_hotplug;
1300 }
1301 
1302 static const TypeInfo pvscsi_info = {
1303     .name          = TYPE_PVSCSI,
1304     .parent        = TYPE_PCI_DEVICE,
1305     .class_size    = sizeof(PVSCSIClass),
1306     .instance_size = sizeof(PVSCSIState),
1307     .class_init    = pvscsi_class_init,
1308     .interfaces = (InterfaceInfo[]) {
1309         { TYPE_HOTPLUG_HANDLER },
1310         { INTERFACE_PCIE_DEVICE },
1311         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
1312         { }
1313     }
1314 };
1315 
1316 static void
1317 pvscsi_register_types(void)
1318 {
1319     type_register_static(&pvscsi_info);
1320 }
1321 
1322 type_init(pvscsi_register_types);
1323