Loading...
1/*
2 * Linux driver for VMware's para-virtualized SCSI HBA.
3 *
4 * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; version 2 of the License and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
14 * details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Maintained by: Alok N Kataria <akataria@vmware.com>
21 *
22 */
23
24#include <linux/kernel.h>
25#include <linux/module.h>
26#include <linux/interrupt.h>
27#include <linux/slab.h>
28#include <linux/workqueue.h>
29#include <linux/pci.h>
30
31#include <scsi/scsi.h>
32#include <scsi/scsi_host.h>
33#include <scsi/scsi_cmnd.h>
34#include <scsi/scsi_device.h>
35
36#include "vmw_pvscsi.h"
37
38#define PVSCSI_LINUX_DRIVER_DESC "VMware PVSCSI driver"
39
40MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC);
41MODULE_AUTHOR("VMware, Inc.");
42MODULE_LICENSE("GPL");
43MODULE_VERSION(PVSCSI_DRIVER_VERSION_STRING);
44
45#define PVSCSI_DEFAULT_NUM_PAGES_PER_RING 8
46#define PVSCSI_DEFAULT_NUM_PAGES_MSG_RING 1
47#define PVSCSI_DEFAULT_QUEUE_DEPTH 64
48#define SGL_SIZE PAGE_SIZE
49
50struct pvscsi_sg_list {
51 struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT];
52};
53
54struct pvscsi_ctx {
55 /*
56 * The index of the context in cmd_map serves as the context ID for a
57 * 1-to-1 mapping completions back to requests.
58 */
59 struct scsi_cmnd *cmd;
60 struct pvscsi_sg_list *sgl;
61 struct list_head list;
62 dma_addr_t dataPA;
63 dma_addr_t sensePA;
64 dma_addr_t sglPA;
65};
66
67struct pvscsi_adapter {
68 char *mmioBase;
69 unsigned int irq;
70 u8 rev;
71 bool use_msi;
72 bool use_msix;
73 bool use_msg;
74
75 spinlock_t hw_lock;
76
77 struct workqueue_struct *workqueue;
78 struct work_struct work;
79
80 struct PVSCSIRingReqDesc *req_ring;
81 unsigned req_pages;
82 unsigned req_depth;
83 dma_addr_t reqRingPA;
84
85 struct PVSCSIRingCmpDesc *cmp_ring;
86 unsigned cmp_pages;
87 dma_addr_t cmpRingPA;
88
89 struct PVSCSIRingMsgDesc *msg_ring;
90 unsigned msg_pages;
91 dma_addr_t msgRingPA;
92
93 struct PVSCSIRingsState *rings_state;
94 dma_addr_t ringStatePA;
95
96 struct pci_dev *dev;
97 struct Scsi_Host *host;
98
99 struct list_head cmd_pool;
100 struct pvscsi_ctx *cmd_map;
101};
102
103
104/* Command line parameters */
105static int pvscsi_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
106static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
107static int pvscsi_cmd_per_lun = PVSCSI_DEFAULT_QUEUE_DEPTH;
108static bool pvscsi_disable_msi;
109static bool pvscsi_disable_msix;
110static bool pvscsi_use_msg = true;
111
112#define PVSCSI_RW (S_IRUSR | S_IWUSR)
113
114module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
115MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default="
116 __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING) ")");
117
118module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
119MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
120 __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
121
122module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
123MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
124 __stringify(PVSCSI_MAX_REQ_QUEUE_DEPTH) ")");
125
126module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
127MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
128
129module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
130MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
131
132module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
133MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
134
135static const struct pci_device_id pvscsi_pci_tbl[] = {
136 { PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
137 { 0 }
138};
139
140MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
141
142static struct device *
143pvscsi_dev(const struct pvscsi_adapter *adapter)
144{
145 return &(adapter->dev->dev);
146}
147
148static struct pvscsi_ctx *
149pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
150{
151 struct pvscsi_ctx *ctx, *end;
152
153 end = &adapter->cmd_map[adapter->req_depth];
154 for (ctx = adapter->cmd_map; ctx < end; ctx++)
155 if (ctx->cmd == cmd)
156 return ctx;
157
158 return NULL;
159}
160
161static struct pvscsi_ctx *
162pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
163{
164 struct pvscsi_ctx *ctx;
165
166 if (list_empty(&adapter->cmd_pool))
167 return NULL;
168
169 ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list);
170 ctx->cmd = cmd;
171 list_del(&ctx->list);
172
173 return ctx;
174}
175
176static void pvscsi_release_context(struct pvscsi_adapter *adapter,
177 struct pvscsi_ctx *ctx)
178{
179 ctx->cmd = NULL;
180 list_add(&ctx->list, &adapter->cmd_pool);
181}
182
183/*
184 * Map a pvscsi_ctx struct to a context ID field value; we map to a simple
185 * non-zero integer. ctx always points to an entry in cmd_map array, hence
186 * the return value is always >=1.
187 */
188static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter,
189 const struct pvscsi_ctx *ctx)
190{
191 return ctx - adapter->cmd_map + 1;
192}
193
194static struct pvscsi_ctx *
195pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
196{
197 return &adapter->cmd_map[context - 1];
198}
199
200static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
201 u32 offset, u32 val)
202{
203 writel(val, adapter->mmioBase + offset);
204}
205
206static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
207{
208 return readl(adapter->mmioBase + offset);
209}
210
211static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
212{
213 return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
214}
215
216static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter,
217 u32 val)
218{
219 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val);
220}
221
222static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter)
223{
224 u32 intr_bits;
225
226 intr_bits = PVSCSI_INTR_CMPL_MASK;
227 if (adapter->use_msg)
228 intr_bits |= PVSCSI_INTR_MSG_MASK;
229
230 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits);
231}
232
233static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
234{
235 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
236}
237
238static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter,
239 u32 cmd, const void *desc, size_t len)
240{
241 const u32 *ptr = desc;
242 size_t i;
243
244 len /= sizeof(*ptr);
245 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd);
246 for (i = 0; i < len; i++)
247 pvscsi_reg_write(adapter,
248 PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
249}
250
251static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter,
252 const struct pvscsi_ctx *ctx)
253{
254 struct PVSCSICmdDescAbortCmd cmd = { 0 };
255
256 cmd.target = ctx->cmd->device->id;
257 cmd.context = pvscsi_map_context(adapter, ctx);
258
259 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd));
260}
261
262static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter)
263{
264 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
265}
266
267static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter)
268{
269 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0);
270}
271
272static int scsi_is_rw(unsigned char op)
273{
274 return op == READ_6 || op == WRITE_6 ||
275 op == READ_10 || op == WRITE_10 ||
276 op == READ_12 || op == WRITE_12 ||
277 op == READ_16 || op == WRITE_16;
278}
279
280static void pvscsi_kick_io(const struct pvscsi_adapter *adapter,
281 unsigned char op)
282{
283 if (scsi_is_rw(op))
284 pvscsi_kick_rw_io(adapter);
285 else
286 pvscsi_process_request_ring(adapter);
287}
288
289static void ll_adapter_reset(const struct pvscsi_adapter *adapter)
290{
291 dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter);
292
293 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0);
294}
295
296static void ll_bus_reset(const struct pvscsi_adapter *adapter)
297{
298 dev_dbg(pvscsi_dev(adapter), "Reseting bus on %p\n", adapter);
299
300 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0);
301}
302
303static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target)
304{
305 struct PVSCSICmdDescResetDevice cmd = { 0 };
306
307 dev_dbg(pvscsi_dev(adapter), "Reseting device: target=%u\n", target);
308
309 cmd.target = target;
310
311 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE,
312 &cmd, sizeof(cmd));
313}
314
315static void pvscsi_create_sg(struct pvscsi_ctx *ctx,
316 struct scatterlist *sg, unsigned count)
317{
318 unsigned i;
319 struct PVSCSISGElement *sge;
320
321 BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT);
322
323 sge = &ctx->sgl->sge[0];
324 for (i = 0; i < count; i++, sg++) {
325 sge[i].addr = sg_dma_address(sg);
326 sge[i].length = sg_dma_len(sg);
327 sge[i].flags = 0;
328 }
329}
330
331/*
332 * Map all data buffers for a command into PCI space and
333 * setup the scatter/gather list if needed.
334 */
335static void pvscsi_map_buffers(struct pvscsi_adapter *adapter,
336 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd,
337 struct PVSCSIRingReqDesc *e)
338{
339 unsigned count;
340 unsigned bufflen = scsi_bufflen(cmd);
341 struct scatterlist *sg;
342
343 e->dataLen = bufflen;
344 e->dataAddr = 0;
345 if (bufflen == 0)
346 return;
347
348 sg = scsi_sglist(cmd);
349 count = scsi_sg_count(cmd);
350 if (count != 0) {
351 int segs = scsi_dma_map(cmd);
352 if (segs > 1) {
353 pvscsi_create_sg(ctx, sg, segs);
354
355 e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
356 ctx->sglPA = pci_map_single(adapter->dev, ctx->sgl,
357 SGL_SIZE, PCI_DMA_TODEVICE);
358 e->dataAddr = ctx->sglPA;
359 } else
360 e->dataAddr = sg_dma_address(sg);
361 } else {
362 /*
363 * In case there is no S/G list, scsi_sglist points
364 * directly to the buffer.
365 */
366 ctx->dataPA = pci_map_single(adapter->dev, sg, bufflen,
367 cmd->sc_data_direction);
368 e->dataAddr = ctx->dataPA;
369 }
370}
371
372static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
373 struct pvscsi_ctx *ctx)
374{
375 struct scsi_cmnd *cmd;
376 unsigned bufflen;
377
378 cmd = ctx->cmd;
379 bufflen = scsi_bufflen(cmd);
380
381 if (bufflen != 0) {
382 unsigned count = scsi_sg_count(cmd);
383
384 if (count != 0) {
385 scsi_dma_unmap(cmd);
386 if (ctx->sglPA) {
387 pci_unmap_single(adapter->dev, ctx->sglPA,
388 SGL_SIZE, PCI_DMA_TODEVICE);
389 ctx->sglPA = 0;
390 }
391 } else
392 pci_unmap_single(adapter->dev, ctx->dataPA, bufflen,
393 cmd->sc_data_direction);
394 }
395 if (cmd->sense_buffer)
396 pci_unmap_single(adapter->dev, ctx->sensePA,
397 SCSI_SENSE_BUFFERSIZE, PCI_DMA_FROMDEVICE);
398}
399
400static int __devinit pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
401{
402 adapter->rings_state = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
403 &adapter->ringStatePA);
404 if (!adapter->rings_state)
405 return -ENOMEM;
406
407 adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
408 pvscsi_ring_pages);
409 adapter->req_depth = adapter->req_pages
410 * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
411 adapter->req_ring = pci_alloc_consistent(adapter->dev,
412 adapter->req_pages * PAGE_SIZE,
413 &adapter->reqRingPA);
414 if (!adapter->req_ring)
415 return -ENOMEM;
416
417 adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
418 pvscsi_ring_pages);
419 adapter->cmp_ring = pci_alloc_consistent(adapter->dev,
420 adapter->cmp_pages * PAGE_SIZE,
421 &adapter->cmpRingPA);
422 if (!adapter->cmp_ring)
423 return -ENOMEM;
424
425 BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
426 BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
427 BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
428
429 if (!adapter->use_msg)
430 return 0;
431
432 adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
433 pvscsi_msg_ring_pages);
434 adapter->msg_ring = pci_alloc_consistent(adapter->dev,
435 adapter->msg_pages * PAGE_SIZE,
436 &adapter->msgRingPA);
437 if (!adapter->msg_ring)
438 return -ENOMEM;
439 BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
440
441 return 0;
442}
443
444static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
445{
446 struct PVSCSICmdDescSetupRings cmd = { 0 };
447 dma_addr_t base;
448 unsigned i;
449
450 cmd.ringsStatePPN = adapter->ringStatePA >> PAGE_SHIFT;
451 cmd.reqRingNumPages = adapter->req_pages;
452 cmd.cmpRingNumPages = adapter->cmp_pages;
453
454 base = adapter->reqRingPA;
455 for (i = 0; i < adapter->req_pages; i++) {
456 cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
457 base += PAGE_SIZE;
458 }
459
460 base = adapter->cmpRingPA;
461 for (i = 0; i < adapter->cmp_pages; i++) {
462 cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
463 base += PAGE_SIZE;
464 }
465
466 memset(adapter->rings_state, 0, PAGE_SIZE);
467 memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
468 memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
469
470 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
471 &cmd, sizeof(cmd));
472
473 if (adapter->use_msg) {
474 struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
475
476 cmd_msg.numPages = adapter->msg_pages;
477
478 base = adapter->msgRingPA;
479 for (i = 0; i < adapter->msg_pages; i++) {
480 cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
481 base += PAGE_SIZE;
482 }
483 memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
484
485 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
486 &cmd_msg, sizeof(cmd_msg));
487 }
488}
489
490/*
491 * Pull a completion descriptor off and pass the completion back
492 * to the SCSI mid layer.
493 */
494static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
495 const struct PVSCSIRingCmpDesc *e)
496{
497 struct pvscsi_ctx *ctx;
498 struct scsi_cmnd *cmd;
499 u32 btstat = e->hostStatus;
500 u32 sdstat = e->scsiStatus;
501
502 ctx = pvscsi_get_context(adapter, e->context);
503 cmd = ctx->cmd;
504 pvscsi_unmap_buffers(adapter, ctx);
505 pvscsi_release_context(adapter, ctx);
506 cmd->result = 0;
507
508 if (sdstat != SAM_STAT_GOOD &&
509 (btstat == BTSTAT_SUCCESS ||
510 btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
511 btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
512 cmd->result = (DID_OK << 16) | sdstat;
513 if (sdstat == SAM_STAT_CHECK_CONDITION && cmd->sense_buffer)
514 cmd->result |= (DRIVER_SENSE << 24);
515 } else
516 switch (btstat) {
517 case BTSTAT_SUCCESS:
518 case BTSTAT_LINKED_COMMAND_COMPLETED:
519 case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
520 /* If everything went fine, let's move on.. */
521 cmd->result = (DID_OK << 16);
522 break;
523
524 case BTSTAT_DATARUN:
525 case BTSTAT_DATA_UNDERRUN:
526 /* Report residual data in underruns */
527 scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
528 cmd->result = (DID_ERROR << 16);
529 break;
530
531 case BTSTAT_SELTIMEO:
532 /* Our emulation returns this for non-connected devs */
533 cmd->result = (DID_BAD_TARGET << 16);
534 break;
535
536 case BTSTAT_LUNMISMATCH:
537 case BTSTAT_TAGREJECT:
538 case BTSTAT_BADMSG:
539 cmd->result = (DRIVER_INVALID << 24);
540 /* fall through */
541
542 case BTSTAT_HAHARDWARE:
543 case BTSTAT_INVPHASE:
544 case BTSTAT_HATIMEOUT:
545 case BTSTAT_NORESPONSE:
546 case BTSTAT_DISCONNECT:
547 case BTSTAT_HASOFTWARE:
548 case BTSTAT_BUSFREE:
549 case BTSTAT_SENSFAILED:
550 cmd->result |= (DID_ERROR << 16);
551 break;
552
553 case BTSTAT_SENTRST:
554 case BTSTAT_RECVRST:
555 case BTSTAT_BUSRESET:
556 cmd->result = (DID_RESET << 16);
557 break;
558
559 case BTSTAT_ABORTQUEUE:
560 cmd->result = (DID_ABORT << 16);
561 break;
562
563 case BTSTAT_SCSIPARITY:
564 cmd->result = (DID_PARITY << 16);
565 break;
566
567 default:
568 cmd->result = (DID_ERROR << 16);
569 scmd_printk(KERN_DEBUG, cmd,
570 "Unknown completion status: 0x%x\n",
571 btstat);
572 }
573
574 dev_dbg(&cmd->device->sdev_gendev,
575 "cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
576 cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
577
578 cmd->scsi_done(cmd);
579}
580
581/*
582 * barrier usage : Since the PVSCSI device is emulated, there could be cases
583 * where we may want to serialize some accesses between the driver and the
584 * emulation layer. We use compiler barriers instead of the more expensive
585 * memory barriers because PVSCSI is only supported on X86 which has strong
586 * memory access ordering.
587 */
588static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
589{
590 struct PVSCSIRingsState *s = adapter->rings_state;
591 struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
592 u32 cmp_entries = s->cmpNumEntriesLog2;
593
594 while (s->cmpConsIdx != s->cmpProdIdx) {
595 struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
596 MASK(cmp_entries));
597 /*
598 * This barrier() ensures that *e is not dereferenced while
599 * the device emulation still writes data into the slot.
600 * Since the device emulation advances s->cmpProdIdx only after
601 * updating the slot we want to check it first.
602 */
603 barrier();
604 pvscsi_complete_request(adapter, e);
605 /*
606 * This barrier() ensures that compiler doesn't reorder write
607 * to s->cmpConsIdx before the read of (*e) inside
608 * pvscsi_complete_request. Otherwise, device emulation may
609 * overwrite *e before we had a chance to read it.
610 */
611 barrier();
612 s->cmpConsIdx++;
613 }
614}
615
616/*
617 * Translate a Linux SCSI request into a request ring entry.
618 */
619static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
620 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
621{
622 struct PVSCSIRingsState *s;
623 struct PVSCSIRingReqDesc *e;
624 struct scsi_device *sdev;
625 u32 req_entries;
626
627 s = adapter->rings_state;
628 sdev = cmd->device;
629 req_entries = s->reqNumEntriesLog2;
630
631 /*
632 * If this condition holds, we might have room on the request ring, but
633 * we might not have room on the completion ring for the response.
634 * However, we have already ruled out this possibility - we would not
635 * have successfully allocated a context if it were true, since we only
636 * have one context per request entry. Check for it anyway, since it
637 * would be a serious bug.
638 */
639 if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
640 scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
641 "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
642 s->reqProdIdx, s->cmpConsIdx);
643 return -1;
644 }
645
646 e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
647
648 e->bus = sdev->channel;
649 e->target = sdev->id;
650 memset(e->lun, 0, sizeof(e->lun));
651 e->lun[1] = sdev->lun;
652
653 if (cmd->sense_buffer) {
654 ctx->sensePA = pci_map_single(adapter->dev, cmd->sense_buffer,
655 SCSI_SENSE_BUFFERSIZE,
656 PCI_DMA_FROMDEVICE);
657 e->senseAddr = ctx->sensePA;
658 e->senseLen = SCSI_SENSE_BUFFERSIZE;
659 } else {
660 e->senseLen = 0;
661 e->senseAddr = 0;
662 }
663 e->cdbLen = cmd->cmd_len;
664 e->vcpuHint = smp_processor_id();
665 memcpy(e->cdb, cmd->cmnd, e->cdbLen);
666
667 e->tag = SIMPLE_QUEUE_TAG;
668 if (sdev->tagged_supported &&
669 (cmd->tag == HEAD_OF_QUEUE_TAG ||
670 cmd->tag == ORDERED_QUEUE_TAG))
671 e->tag = cmd->tag;
672
673 if (cmd->sc_data_direction == DMA_FROM_DEVICE)
674 e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
675 else if (cmd->sc_data_direction == DMA_TO_DEVICE)
676 e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
677 else if (cmd->sc_data_direction == DMA_NONE)
678 e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
679 else
680 e->flags = 0;
681
682 pvscsi_map_buffers(adapter, ctx, cmd, e);
683
684 e->context = pvscsi_map_context(adapter, ctx);
685
686 barrier();
687
688 s->reqProdIdx++;
689
690 return 0;
691}
692
693static int pvscsi_queue_lck(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
694{
695 struct Scsi_Host *host = cmd->device->host;
696 struct pvscsi_adapter *adapter = shost_priv(host);
697 struct pvscsi_ctx *ctx;
698 unsigned long flags;
699
700 spin_lock_irqsave(&adapter->hw_lock, flags);
701
702 ctx = pvscsi_acquire_context(adapter, cmd);
703 if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
704 if (ctx)
705 pvscsi_release_context(adapter, ctx);
706 spin_unlock_irqrestore(&adapter->hw_lock, flags);
707 return SCSI_MLQUEUE_HOST_BUSY;
708 }
709
710 cmd->scsi_done = done;
711
712 dev_dbg(&cmd->device->sdev_gendev,
713 "queued cmd %p, ctx %p, op=%x\n", cmd, ctx, cmd->cmnd[0]);
714
715 spin_unlock_irqrestore(&adapter->hw_lock, flags);
716
717 pvscsi_kick_io(adapter, cmd->cmnd[0]);
718
719 return 0;
720}
721
722static DEF_SCSI_QCMD(pvscsi_queue)
723
724static int pvscsi_abort(struct scsi_cmnd *cmd)
725{
726 struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
727 struct pvscsi_ctx *ctx;
728 unsigned long flags;
729
730 scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
731 adapter->host->host_no, cmd);
732
733 spin_lock_irqsave(&adapter->hw_lock, flags);
734
735 /*
736 * Poll the completion ring first - we might be trying to abort
737 * a command that is waiting to be dispatched in the completion ring.
738 */
739 pvscsi_process_completion_ring(adapter);
740
741 /*
742 * If there is no context for the command, it either already succeeded
743 * or else was never properly issued. Not our problem.
744 */
745 ctx = pvscsi_find_context(adapter, cmd);
746 if (!ctx) {
747 scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
748 goto out;
749 }
750
751 pvscsi_abort_cmd(adapter, ctx);
752
753 pvscsi_process_completion_ring(adapter);
754
755out:
756 spin_unlock_irqrestore(&adapter->hw_lock, flags);
757 return SUCCESS;
758}
759
760/*
761 * Abort all outstanding requests. This is only safe to use if the completion
762 * ring will never be walked again or the device has been reset, because it
763 * destroys the 1-1 mapping between context field passed to emulation and our
764 * request structure.
765 */
766static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
767{
768 unsigned i;
769
770 for (i = 0; i < adapter->req_depth; i++) {
771 struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
772 struct scsi_cmnd *cmd = ctx->cmd;
773 if (cmd) {
774 scmd_printk(KERN_ERR, cmd,
775 "Forced reset on cmd %p\n", cmd);
776 pvscsi_unmap_buffers(adapter, ctx);
777 pvscsi_release_context(adapter, ctx);
778 cmd->result = (DID_RESET << 16);
779 cmd->scsi_done(cmd);
780 }
781 }
782}
783
784static int pvscsi_host_reset(struct scsi_cmnd *cmd)
785{
786 struct Scsi_Host *host = cmd->device->host;
787 struct pvscsi_adapter *adapter = shost_priv(host);
788 unsigned long flags;
789 bool use_msg;
790
791 scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
792
793 spin_lock_irqsave(&adapter->hw_lock, flags);
794
795 use_msg = adapter->use_msg;
796
797 if (use_msg) {
798 adapter->use_msg = 0;
799 spin_unlock_irqrestore(&adapter->hw_lock, flags);
800
801 /*
802 * Now that we know that the ISR won't add more work on the
803 * workqueue we can safely flush any outstanding work.
804 */
805 flush_workqueue(adapter->workqueue);
806 spin_lock_irqsave(&adapter->hw_lock, flags);
807 }
808
809 /*
810 * We're going to tear down the entire ring structure and set it back
811 * up, so stalling new requests until all completions are flushed and
812 * the rings are back in place.
813 */
814
815 pvscsi_process_request_ring(adapter);
816
817 ll_adapter_reset(adapter);
818
819 /*
820 * Now process any completions. Note we do this AFTER adapter reset,
821 * which is strange, but stops races where completions get posted
822 * between processing the ring and issuing the reset. The backend will
823 * not touch the ring memory after reset, so the immediately pre-reset
824 * completion ring state is still valid.
825 */
826 pvscsi_process_completion_ring(adapter);
827
828 pvscsi_reset_all(adapter);
829 adapter->use_msg = use_msg;
830 pvscsi_setup_all_rings(adapter);
831 pvscsi_unmask_intr(adapter);
832
833 spin_unlock_irqrestore(&adapter->hw_lock, flags);
834
835 return SUCCESS;
836}
837
838static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
839{
840 struct Scsi_Host *host = cmd->device->host;
841 struct pvscsi_adapter *adapter = shost_priv(host);
842 unsigned long flags;
843
844 scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
845
846 /*
847 * We don't want to queue new requests for this bus after
848 * flushing all pending requests to emulation, since new
849 * requests could then sneak in during this bus reset phase,
850 * so take the lock now.
851 */
852 spin_lock_irqsave(&adapter->hw_lock, flags);
853
854 pvscsi_process_request_ring(adapter);
855 ll_bus_reset(adapter);
856 pvscsi_process_completion_ring(adapter);
857
858 spin_unlock_irqrestore(&adapter->hw_lock, flags);
859
860 return SUCCESS;
861}
862
863static int pvscsi_device_reset(struct scsi_cmnd *cmd)
864{
865 struct Scsi_Host *host = cmd->device->host;
866 struct pvscsi_adapter *adapter = shost_priv(host);
867 unsigned long flags;
868
869 scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
870 host->host_no, cmd->device->id);
871
872 /*
873 * We don't want to queue new requests for this device after flushing
874 * all pending requests to emulation, since new requests could then
875 * sneak in during this device reset phase, so take the lock now.
876 */
877 spin_lock_irqsave(&adapter->hw_lock, flags);
878
879 pvscsi_process_request_ring(adapter);
880 ll_device_reset(adapter, cmd->device->id);
881 pvscsi_process_completion_ring(adapter);
882
883 spin_unlock_irqrestore(&adapter->hw_lock, flags);
884
885 return SUCCESS;
886}
887
888static struct scsi_host_template pvscsi_template;
889
890static const char *pvscsi_info(struct Scsi_Host *host)
891{
892 struct pvscsi_adapter *adapter = shost_priv(host);
893 static char buf[256];
894
895 sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
896 "%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
897 adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
898 pvscsi_template.cmd_per_lun);
899
900 return buf;
901}
902
903static struct scsi_host_template pvscsi_template = {
904 .module = THIS_MODULE,
905 .name = "VMware PVSCSI Host Adapter",
906 .proc_name = "vmw_pvscsi",
907 .info = pvscsi_info,
908 .queuecommand = pvscsi_queue,
909 .this_id = -1,
910 .sg_tablesize = PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
911 .dma_boundary = UINT_MAX,
912 .max_sectors = 0xffff,
913 .use_clustering = ENABLE_CLUSTERING,
914 .eh_abort_handler = pvscsi_abort,
915 .eh_device_reset_handler = pvscsi_device_reset,
916 .eh_bus_reset_handler = pvscsi_bus_reset,
917 .eh_host_reset_handler = pvscsi_host_reset,
918};
919
920static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
921 const struct PVSCSIRingMsgDesc *e)
922{
923 struct PVSCSIRingsState *s = adapter->rings_state;
924 struct Scsi_Host *host = adapter->host;
925 struct scsi_device *sdev;
926
927 printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
928 e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
929
930 BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
931
932 if (e->type == PVSCSI_MSG_DEV_ADDED) {
933 struct PVSCSIMsgDescDevStatusChanged *desc;
934 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
935
936 printk(KERN_INFO
937 "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
938 desc->bus, desc->target, desc->lun[1]);
939
940 if (!scsi_host_get(host))
941 return;
942
943 sdev = scsi_device_lookup(host, desc->bus, desc->target,
944 desc->lun[1]);
945 if (sdev) {
946 printk(KERN_INFO "vmw_pvscsi: device already exists\n");
947 scsi_device_put(sdev);
948 } else
949 scsi_add_device(adapter->host, desc->bus,
950 desc->target, desc->lun[1]);
951
952 scsi_host_put(host);
953 } else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
954 struct PVSCSIMsgDescDevStatusChanged *desc;
955 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
956
957 printk(KERN_INFO
958 "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
959 desc->bus, desc->target, desc->lun[1]);
960
961 if (!scsi_host_get(host))
962 return;
963
964 sdev = scsi_device_lookup(host, desc->bus, desc->target,
965 desc->lun[1]);
966 if (sdev) {
967 scsi_remove_device(sdev);
968 scsi_device_put(sdev);
969 } else
970 printk(KERN_INFO
971 "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
972 desc->bus, desc->target, desc->lun[1]);
973
974 scsi_host_put(host);
975 }
976}
977
978static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
979{
980 struct PVSCSIRingsState *s = adapter->rings_state;
981
982 return s->msgProdIdx != s->msgConsIdx;
983}
984
985static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
986{
987 struct PVSCSIRingsState *s = adapter->rings_state;
988 struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
989 u32 msg_entries = s->msgNumEntriesLog2;
990
991 while (pvscsi_msg_pending(adapter)) {
992 struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
993 MASK(msg_entries));
994
995 barrier();
996 pvscsi_process_msg(adapter, e);
997 barrier();
998 s->msgConsIdx++;
999 }
1000}
1001
1002static void pvscsi_msg_workqueue_handler(struct work_struct *data)
1003{
1004 struct pvscsi_adapter *adapter;
1005
1006 adapter = container_of(data, struct pvscsi_adapter, work);
1007
1008 pvscsi_process_msg_ring(adapter);
1009}
1010
1011static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
1012{
1013 char name[32];
1014
1015 if (!pvscsi_use_msg)
1016 return 0;
1017
1018 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1019 PVSCSI_CMD_SETUP_MSG_RING);
1020
1021 if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
1022 return 0;
1023
1024 snprintf(name, sizeof(name),
1025 "vmw_pvscsi_wq_%u", adapter->host->host_no);
1026
1027 adapter->workqueue = create_singlethread_workqueue(name);
1028 if (!adapter->workqueue) {
1029 printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
1030 return 0;
1031 }
1032 INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
1033
1034 return 1;
1035}
1036
1037static irqreturn_t pvscsi_isr(int irq, void *devp)
1038{
1039 struct pvscsi_adapter *adapter = devp;
1040 int handled;
1041
1042 if (adapter->use_msi || adapter->use_msix)
1043 handled = true;
1044 else {
1045 u32 val = pvscsi_read_intr_status(adapter);
1046 handled = (val & PVSCSI_INTR_ALL_SUPPORTED) != 0;
1047 if (handled)
1048 pvscsi_write_intr_status(devp, val);
1049 }
1050
1051 if (handled) {
1052 unsigned long flags;
1053
1054 spin_lock_irqsave(&adapter->hw_lock, flags);
1055
1056 pvscsi_process_completion_ring(adapter);
1057 if (adapter->use_msg && pvscsi_msg_pending(adapter))
1058 queue_work(adapter->workqueue, &adapter->work);
1059
1060 spin_unlock_irqrestore(&adapter->hw_lock, flags);
1061 }
1062
1063 return IRQ_RETVAL(handled);
1064}
1065
1066static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
1067{
1068 struct pvscsi_ctx *ctx = adapter->cmd_map;
1069 unsigned i;
1070
1071 for (i = 0; i < adapter->req_depth; ++i, ++ctx)
1072 free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
1073}
1074
1075static int pvscsi_setup_msix(const struct pvscsi_adapter *adapter,
1076 unsigned int *irq)
1077{
1078 struct msix_entry entry = { 0, PVSCSI_VECTOR_COMPLETION };
1079 int ret;
1080
1081 ret = pci_enable_msix(adapter->dev, &entry, 1);
1082 if (ret)
1083 return ret;
1084
1085 *irq = entry.vector;
1086
1087 return 0;
1088}
1089
1090static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
1091{
1092 if (adapter->irq) {
1093 free_irq(adapter->irq, adapter);
1094 adapter->irq = 0;
1095 }
1096 if (adapter->use_msi) {
1097 pci_disable_msi(adapter->dev);
1098 adapter->use_msi = 0;
1099 } else if (adapter->use_msix) {
1100 pci_disable_msix(adapter->dev);
1101 adapter->use_msix = 0;
1102 }
1103}
1104
1105static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
1106{
1107 pvscsi_shutdown_intr(adapter);
1108
1109 if (adapter->workqueue)
1110 destroy_workqueue(adapter->workqueue);
1111
1112 if (adapter->mmioBase)
1113 pci_iounmap(adapter->dev, adapter->mmioBase);
1114
1115 pci_release_regions(adapter->dev);
1116
1117 if (adapter->cmd_map) {
1118 pvscsi_free_sgls(adapter);
1119 kfree(adapter->cmd_map);
1120 }
1121
1122 if (adapter->rings_state)
1123 pci_free_consistent(adapter->dev, PAGE_SIZE,
1124 adapter->rings_state, adapter->ringStatePA);
1125
1126 if (adapter->req_ring)
1127 pci_free_consistent(adapter->dev,
1128 adapter->req_pages * PAGE_SIZE,
1129 adapter->req_ring, adapter->reqRingPA);
1130
1131 if (adapter->cmp_ring)
1132 pci_free_consistent(adapter->dev,
1133 adapter->cmp_pages * PAGE_SIZE,
1134 adapter->cmp_ring, adapter->cmpRingPA);
1135
1136 if (adapter->msg_ring)
1137 pci_free_consistent(adapter->dev,
1138 adapter->msg_pages * PAGE_SIZE,
1139 adapter->msg_ring, adapter->msgRingPA);
1140}
1141
1142/*
1143 * Allocate scatter gather lists.
1144 *
1145 * These are statically allocated. Trying to be clever was not worth it.
1146 *
1147 * Dynamic allocation can fail, and we can't go deeep into the memory
1148 * allocator, since we're a SCSI driver, and trying too hard to allocate
1149 * memory might generate disk I/O. We also don't want to fail disk I/O
1150 * in that case because we can't get an allocation - the I/O could be
1151 * trying to swap out data to free memory. Since that is pathological,
1152 * just use a statically allocated scatter list.
1153 *
1154 */
1155static int __devinit pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
1156{
1157 struct pvscsi_ctx *ctx;
1158 int i;
1159
1160 ctx = adapter->cmd_map;
1161 BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
1162
1163 for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
1164 ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
1165 get_order(SGL_SIZE));
1166 ctx->sglPA = 0;
1167 BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
1168 if (!ctx->sgl) {
1169 for (; i >= 0; --i, --ctx) {
1170 free_pages((unsigned long)ctx->sgl,
1171 get_order(SGL_SIZE));
1172 ctx->sgl = NULL;
1173 }
1174 return -ENOMEM;
1175 }
1176 }
1177
1178 return 0;
1179}
1180
1181static int __devinit pvscsi_probe(struct pci_dev *pdev,
1182 const struct pci_device_id *id)
1183{
1184 struct pvscsi_adapter *adapter;
1185 struct Scsi_Host *host;
1186 unsigned int i;
1187 unsigned long flags = 0;
1188 int error;
1189
1190 error = -ENODEV;
1191
1192 if (pci_enable_device(pdev))
1193 return error;
1194
1195 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0 &&
1196 pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1197 printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
1198 } else if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) == 0 &&
1199 pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) == 0) {
1200 printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
1201 } else {
1202 printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
1203 goto out_disable_device;
1204 }
1205
1206 pvscsi_template.can_queue =
1207 min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
1208 PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
1209 pvscsi_template.cmd_per_lun =
1210 min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
1211 host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
1212 if (!host) {
1213 printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
1214 goto out_disable_device;
1215 }
1216
1217 adapter = shost_priv(host);
1218 memset(adapter, 0, sizeof(*adapter));
1219 adapter->dev = pdev;
1220 adapter->host = host;
1221
1222 spin_lock_init(&adapter->hw_lock);
1223
1224 host->max_channel = 0;
1225 host->max_id = 16;
1226 host->max_lun = 1;
1227 host->max_cmd_len = 16;
1228
1229 adapter->rev = pdev->revision;
1230
1231 if (pci_request_regions(pdev, "vmw_pvscsi")) {
1232 printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
1233 goto out_free_host;
1234 }
1235
1236 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1237 if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
1238 continue;
1239
1240 if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
1241 continue;
1242
1243 break;
1244 }
1245
1246 if (i == DEVICE_COUNT_RESOURCE) {
1247 printk(KERN_ERR
1248 "vmw_pvscsi: adapter has no suitable MMIO region\n");
1249 goto out_release_resources;
1250 }
1251
1252 adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
1253
1254 if (!adapter->mmioBase) {
1255 printk(KERN_ERR
1256 "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
1257 i, PVSCSI_MEM_SPACE_SIZE);
1258 goto out_release_resources;
1259 }
1260
1261 pci_set_master(pdev);
1262 pci_set_drvdata(pdev, host);
1263
1264 ll_adapter_reset(adapter);
1265
1266 adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
1267
1268 error = pvscsi_allocate_rings(adapter);
1269 if (error) {
1270 printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
1271 goto out_release_resources;
1272 }
1273
1274 /*
1275 * From this point on we should reset the adapter if anything goes
1276 * wrong.
1277 */
1278 pvscsi_setup_all_rings(adapter);
1279
1280 adapter->cmd_map = kcalloc(adapter->req_depth,
1281 sizeof(struct pvscsi_ctx), GFP_KERNEL);
1282 if (!adapter->cmd_map) {
1283 printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
1284 error = -ENOMEM;
1285 goto out_reset_adapter;
1286 }
1287
1288 INIT_LIST_HEAD(&adapter->cmd_pool);
1289 for (i = 0; i < adapter->req_depth; i++) {
1290 struct pvscsi_ctx *ctx = adapter->cmd_map + i;
1291 list_add(&ctx->list, &adapter->cmd_pool);
1292 }
1293
1294 error = pvscsi_allocate_sg(adapter);
1295 if (error) {
1296 printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
1297 goto out_reset_adapter;
1298 }
1299
1300 if (!pvscsi_disable_msix &&
1301 pvscsi_setup_msix(adapter, &adapter->irq) == 0) {
1302 printk(KERN_INFO "vmw_pvscsi: using MSI-X\n");
1303 adapter->use_msix = 1;
1304 } else if (!pvscsi_disable_msi && pci_enable_msi(pdev) == 0) {
1305 printk(KERN_INFO "vmw_pvscsi: using MSI\n");
1306 adapter->use_msi = 1;
1307 adapter->irq = pdev->irq;
1308 } else {
1309 printk(KERN_INFO "vmw_pvscsi: using INTx\n");
1310 adapter->irq = pdev->irq;
1311 flags = IRQF_SHARED;
1312 }
1313
1314 error = request_irq(adapter->irq, pvscsi_isr, flags,
1315 "vmw_pvscsi", adapter);
1316 if (error) {
1317 printk(KERN_ERR
1318 "vmw_pvscsi: unable to request IRQ: %d\n", error);
1319 adapter->irq = 0;
1320 goto out_reset_adapter;
1321 }
1322
1323 error = scsi_add_host(host, &pdev->dev);
1324 if (error) {
1325 printk(KERN_ERR
1326 "vmw_pvscsi: scsi_add_host failed: %d\n", error);
1327 goto out_reset_adapter;
1328 }
1329
1330 dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
1331 adapter->rev, host->host_no);
1332
1333 pvscsi_unmask_intr(adapter);
1334
1335 scsi_scan_host(host);
1336
1337 return 0;
1338
1339out_reset_adapter:
1340 ll_adapter_reset(adapter);
1341out_release_resources:
1342 pvscsi_release_resources(adapter);
1343out_free_host:
1344 scsi_host_put(host);
1345out_disable_device:
1346 pci_set_drvdata(pdev, NULL);
1347 pci_disable_device(pdev);
1348
1349 return error;
1350}
1351
1352static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
1353{
1354 pvscsi_mask_intr(adapter);
1355
1356 if (adapter->workqueue)
1357 flush_workqueue(adapter->workqueue);
1358
1359 pvscsi_shutdown_intr(adapter);
1360
1361 pvscsi_process_request_ring(adapter);
1362 pvscsi_process_completion_ring(adapter);
1363 ll_adapter_reset(adapter);
1364}
1365
1366static void pvscsi_shutdown(struct pci_dev *dev)
1367{
1368 struct Scsi_Host *host = pci_get_drvdata(dev);
1369 struct pvscsi_adapter *adapter = shost_priv(host);
1370
1371 __pvscsi_shutdown(adapter);
1372}
1373
1374static void pvscsi_remove(struct pci_dev *pdev)
1375{
1376 struct Scsi_Host *host = pci_get_drvdata(pdev);
1377 struct pvscsi_adapter *adapter = shost_priv(host);
1378
1379 scsi_remove_host(host);
1380
1381 __pvscsi_shutdown(adapter);
1382 pvscsi_release_resources(adapter);
1383
1384 scsi_host_put(host);
1385
1386 pci_set_drvdata(pdev, NULL);
1387 pci_disable_device(pdev);
1388}
1389
1390static struct pci_driver pvscsi_pci_driver = {
1391 .name = "vmw_pvscsi",
1392 .id_table = pvscsi_pci_tbl,
1393 .probe = pvscsi_probe,
1394 .remove = __devexit_p(pvscsi_remove),
1395 .shutdown = pvscsi_shutdown,
1396};
1397
1398static int __init pvscsi_init(void)
1399{
1400 pr_info("%s - version %s\n",
1401 PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
1402 return pci_register_driver(&pvscsi_pci_driver);
1403}
1404
1405static void __exit pvscsi_exit(void)
1406{
1407 pci_unregister_driver(&pvscsi_pci_driver);
1408}
1409
1410module_init(pvscsi_init);
1411module_exit(pvscsi_exit);
1/*
2 * Linux driver for VMware's para-virtualized SCSI HBA.
3 *
4 * Copyright (C) 2008-2014, VMware, Inc. All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; version 2 of the License and no later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
14 * details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 */
21
22#include <linux/kernel.h>
23#include <linux/module.h>
24#include <linux/interrupt.h>
25#include <linux/slab.h>
26#include <linux/workqueue.h>
27#include <linux/pci.h>
28
29#include <scsi/scsi.h>
30#include <scsi/scsi_host.h>
31#include <scsi/scsi_cmnd.h>
32#include <scsi/scsi_device.h>
33#include <scsi/scsi_tcq.h>
34
35#include "vmw_pvscsi.h"
36
37#define PVSCSI_LINUX_DRIVER_DESC "VMware PVSCSI driver"
38
39MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC);
40MODULE_AUTHOR("VMware, Inc.");
41MODULE_LICENSE("GPL");
42MODULE_VERSION(PVSCSI_DRIVER_VERSION_STRING);
43
44#define PVSCSI_DEFAULT_NUM_PAGES_PER_RING 8
45#define PVSCSI_DEFAULT_NUM_PAGES_MSG_RING 1
46#define PVSCSI_DEFAULT_QUEUE_DEPTH 254
47#define SGL_SIZE PAGE_SIZE
48
49struct pvscsi_sg_list {
50 struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT];
51};
52
53struct pvscsi_ctx {
54 /*
55 * The index of the context in cmd_map serves as the context ID for a
56 * 1-to-1 mapping completions back to requests.
57 */
58 struct scsi_cmnd *cmd;
59 struct pvscsi_sg_list *sgl;
60 struct list_head list;
61 dma_addr_t dataPA;
62 dma_addr_t sensePA;
63 dma_addr_t sglPA;
64 struct completion *abort_cmp;
65};
66
67struct pvscsi_adapter {
68 char *mmioBase;
69 u8 rev;
70 bool use_msg;
71 bool use_req_threshold;
72
73 spinlock_t hw_lock;
74
75 struct workqueue_struct *workqueue;
76 struct work_struct work;
77
78 struct PVSCSIRingReqDesc *req_ring;
79 unsigned req_pages;
80 unsigned req_depth;
81 dma_addr_t reqRingPA;
82
83 struct PVSCSIRingCmpDesc *cmp_ring;
84 unsigned cmp_pages;
85 dma_addr_t cmpRingPA;
86
87 struct PVSCSIRingMsgDesc *msg_ring;
88 unsigned msg_pages;
89 dma_addr_t msgRingPA;
90
91 struct PVSCSIRingsState *rings_state;
92 dma_addr_t ringStatePA;
93
94 struct pci_dev *dev;
95 struct Scsi_Host *host;
96
97 struct list_head cmd_pool;
98 struct pvscsi_ctx *cmd_map;
99};
100
101
102/* Command line parameters */
103static int pvscsi_ring_pages;
104static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
105static int pvscsi_cmd_per_lun = PVSCSI_DEFAULT_QUEUE_DEPTH;
106static bool pvscsi_disable_msi;
107static bool pvscsi_disable_msix;
108static bool pvscsi_use_msg = true;
109static bool pvscsi_use_req_threshold = true;
110
111#define PVSCSI_RW (S_IRUSR | S_IWUSR)
112
113module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
114MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default="
115 __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING)
116 "[up to 16 targets],"
117 __stringify(PVSCSI_SETUP_RINGS_MAX_NUM_PAGES)
118 "[for 16+ targets])");
119
120module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
121MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
122 __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
123
124module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
125MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
126 __stringify(PVSCSI_DEFAULT_QUEUE_DEPTH) ")");
127
128module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
129MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
130
131module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
132MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
133
134module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
135MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
136
137module_param_named(use_req_threshold, pvscsi_use_req_threshold,
138 bool, PVSCSI_RW);
139MODULE_PARM_DESC(use_req_threshold, "Use driver-based request coalescing if configured - (default=1)");
140
141static const struct pci_device_id pvscsi_pci_tbl[] = {
142 { PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
143 { 0 }
144};
145
146MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
147
148static struct device *
149pvscsi_dev(const struct pvscsi_adapter *adapter)
150{
151 return &(adapter->dev->dev);
152}
153
154static struct pvscsi_ctx *
155pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
156{
157 struct pvscsi_ctx *ctx, *end;
158
159 end = &adapter->cmd_map[adapter->req_depth];
160 for (ctx = adapter->cmd_map; ctx < end; ctx++)
161 if (ctx->cmd == cmd)
162 return ctx;
163
164 return NULL;
165}
166
167static struct pvscsi_ctx *
168pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
169{
170 struct pvscsi_ctx *ctx;
171
172 if (list_empty(&adapter->cmd_pool))
173 return NULL;
174
175 ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list);
176 ctx->cmd = cmd;
177 list_del(&ctx->list);
178
179 return ctx;
180}
181
182static void pvscsi_release_context(struct pvscsi_adapter *adapter,
183 struct pvscsi_ctx *ctx)
184{
185 ctx->cmd = NULL;
186 ctx->abort_cmp = NULL;
187 list_add(&ctx->list, &adapter->cmd_pool);
188}
189
190/*
191 * Map a pvscsi_ctx struct to a context ID field value; we map to a simple
192 * non-zero integer. ctx always points to an entry in cmd_map array, hence
193 * the return value is always >=1.
194 */
195static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter,
196 const struct pvscsi_ctx *ctx)
197{
198 return ctx - adapter->cmd_map + 1;
199}
200
201static struct pvscsi_ctx *
202pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
203{
204 return &adapter->cmd_map[context - 1];
205}
206
207static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
208 u32 offset, u32 val)
209{
210 writel(val, adapter->mmioBase + offset);
211}
212
213static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
214{
215 return readl(adapter->mmioBase + offset);
216}
217
218static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
219{
220 return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
221}
222
223static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter,
224 u32 val)
225{
226 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val);
227}
228
229static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter)
230{
231 u32 intr_bits;
232
233 intr_bits = PVSCSI_INTR_CMPL_MASK;
234 if (adapter->use_msg)
235 intr_bits |= PVSCSI_INTR_MSG_MASK;
236
237 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits);
238}
239
240static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
241{
242 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
243}
244
245static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter,
246 u32 cmd, const void *desc, size_t len)
247{
248 const u32 *ptr = desc;
249 size_t i;
250
251 len /= sizeof(*ptr);
252 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd);
253 for (i = 0; i < len; i++)
254 pvscsi_reg_write(adapter,
255 PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
256}
257
258static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter,
259 const struct pvscsi_ctx *ctx)
260{
261 struct PVSCSICmdDescAbortCmd cmd = { 0 };
262
263 cmd.target = ctx->cmd->device->id;
264 cmd.context = pvscsi_map_context(adapter, ctx);
265
266 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd));
267}
268
269static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter)
270{
271 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
272}
273
274static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter)
275{
276 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0);
277}
278
279static int scsi_is_rw(unsigned char op)
280{
281 return op == READ_6 || op == WRITE_6 ||
282 op == READ_10 || op == WRITE_10 ||
283 op == READ_12 || op == WRITE_12 ||
284 op == READ_16 || op == WRITE_16;
285}
286
287static void pvscsi_kick_io(const struct pvscsi_adapter *adapter,
288 unsigned char op)
289{
290 if (scsi_is_rw(op)) {
291 struct PVSCSIRingsState *s = adapter->rings_state;
292
293 if (!adapter->use_req_threshold ||
294 s->reqProdIdx - s->reqConsIdx >= s->reqCallThreshold)
295 pvscsi_kick_rw_io(adapter);
296 } else {
297 pvscsi_process_request_ring(adapter);
298 }
299}
300
301static void ll_adapter_reset(const struct pvscsi_adapter *adapter)
302{
303 dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter);
304
305 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0);
306}
307
308static void ll_bus_reset(const struct pvscsi_adapter *adapter)
309{
310 dev_dbg(pvscsi_dev(adapter), "Resetting bus on %p\n", adapter);
311
312 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0);
313}
314
315static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target)
316{
317 struct PVSCSICmdDescResetDevice cmd = { 0 };
318
319 dev_dbg(pvscsi_dev(adapter), "Resetting device: target=%u\n", target);
320
321 cmd.target = target;
322
323 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE,
324 &cmd, sizeof(cmd));
325}
326
327static void pvscsi_create_sg(struct pvscsi_ctx *ctx,
328 struct scatterlist *sg, unsigned count)
329{
330 unsigned i;
331 struct PVSCSISGElement *sge;
332
333 BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT);
334
335 sge = &ctx->sgl->sge[0];
336 for (i = 0; i < count; i++, sg = sg_next(sg)) {
337 sge[i].addr = sg_dma_address(sg);
338 sge[i].length = sg_dma_len(sg);
339 sge[i].flags = 0;
340 }
341}
342
343/*
344 * Map all data buffers for a command into PCI space and
345 * setup the scatter/gather list if needed.
346 */
347static int pvscsi_map_buffers(struct pvscsi_adapter *adapter,
348 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd,
349 struct PVSCSIRingReqDesc *e)
350{
351 unsigned count;
352 unsigned bufflen = scsi_bufflen(cmd);
353 struct scatterlist *sg;
354
355 e->dataLen = bufflen;
356 e->dataAddr = 0;
357 if (bufflen == 0)
358 return 0;
359
360 sg = scsi_sglist(cmd);
361 count = scsi_sg_count(cmd);
362 if (count != 0) {
363 int segs = scsi_dma_map(cmd);
364
365 if (segs == -ENOMEM) {
366 scmd_printk(KERN_DEBUG, cmd,
367 "vmw_pvscsi: Failed to map cmd sglist for DMA.\n");
368 return -ENOMEM;
369 } else if (segs > 1) {
370 pvscsi_create_sg(ctx, sg, segs);
371
372 e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
373 ctx->sglPA = dma_map_single(&adapter->dev->dev,
374 ctx->sgl, SGL_SIZE, DMA_TO_DEVICE);
375 if (dma_mapping_error(&adapter->dev->dev, ctx->sglPA)) {
376 scmd_printk(KERN_ERR, cmd,
377 "vmw_pvscsi: Failed to map ctx sglist for DMA.\n");
378 scsi_dma_unmap(cmd);
379 ctx->sglPA = 0;
380 return -ENOMEM;
381 }
382 e->dataAddr = ctx->sglPA;
383 } else
384 e->dataAddr = sg_dma_address(sg);
385 } else {
386 /*
387 * In case there is no S/G list, scsi_sglist points
388 * directly to the buffer.
389 */
390 ctx->dataPA = dma_map_single(&adapter->dev->dev, sg, bufflen,
391 cmd->sc_data_direction);
392 if (dma_mapping_error(&adapter->dev->dev, ctx->dataPA)) {
393 scmd_printk(KERN_DEBUG, cmd,
394 "vmw_pvscsi: Failed to map direct data buffer for DMA.\n");
395 return -ENOMEM;
396 }
397 e->dataAddr = ctx->dataPA;
398 }
399
400 return 0;
401}
402
403/*
404 * The device incorrectly doesn't clear the first byte of the sense
405 * buffer in some cases. We have to do it ourselves.
406 * Otherwise we run into trouble when SWIOTLB is forced.
407 */
408static void pvscsi_patch_sense(struct scsi_cmnd *cmd)
409{
410 if (cmd->sense_buffer)
411 cmd->sense_buffer[0] = 0;
412}
413
414static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
415 struct pvscsi_ctx *ctx)
416{
417 struct scsi_cmnd *cmd;
418 unsigned bufflen;
419
420 cmd = ctx->cmd;
421 bufflen = scsi_bufflen(cmd);
422
423 if (bufflen != 0) {
424 unsigned count = scsi_sg_count(cmd);
425
426 if (count != 0) {
427 scsi_dma_unmap(cmd);
428 if (ctx->sglPA) {
429 dma_unmap_single(&adapter->dev->dev, ctx->sglPA,
430 SGL_SIZE, DMA_TO_DEVICE);
431 ctx->sglPA = 0;
432 }
433 } else
434 dma_unmap_single(&adapter->dev->dev, ctx->dataPA,
435 bufflen, cmd->sc_data_direction);
436 }
437 if (cmd->sense_buffer)
438 dma_unmap_single(&adapter->dev->dev, ctx->sensePA,
439 SCSI_SENSE_BUFFERSIZE, DMA_FROM_DEVICE);
440}
441
442static int pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
443{
444 adapter->rings_state = dma_alloc_coherent(&adapter->dev->dev, PAGE_SIZE,
445 &adapter->ringStatePA, GFP_KERNEL);
446 if (!adapter->rings_state)
447 return -ENOMEM;
448
449 adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
450 pvscsi_ring_pages);
451 adapter->req_depth = adapter->req_pages
452 * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
453 adapter->req_ring = dma_alloc_coherent(&adapter->dev->dev,
454 adapter->req_pages * PAGE_SIZE, &adapter->reqRingPA,
455 GFP_KERNEL);
456 if (!adapter->req_ring)
457 return -ENOMEM;
458
459 adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
460 pvscsi_ring_pages);
461 adapter->cmp_ring = dma_alloc_coherent(&adapter->dev->dev,
462 adapter->cmp_pages * PAGE_SIZE, &adapter->cmpRingPA,
463 GFP_KERNEL);
464 if (!adapter->cmp_ring)
465 return -ENOMEM;
466
467 BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
468 BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
469 BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
470
471 if (!adapter->use_msg)
472 return 0;
473
474 adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
475 pvscsi_msg_ring_pages);
476 adapter->msg_ring = dma_alloc_coherent(&adapter->dev->dev,
477 adapter->msg_pages * PAGE_SIZE, &adapter->msgRingPA,
478 GFP_KERNEL);
479 if (!adapter->msg_ring)
480 return -ENOMEM;
481 BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
482
483 return 0;
484}
485
486static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
487{
488 struct PVSCSICmdDescSetupRings cmd = { 0 };
489 dma_addr_t base;
490 unsigned i;
491
492 cmd.ringsStatePPN = adapter->ringStatePA >> PAGE_SHIFT;
493 cmd.reqRingNumPages = adapter->req_pages;
494 cmd.cmpRingNumPages = adapter->cmp_pages;
495
496 base = adapter->reqRingPA;
497 for (i = 0; i < adapter->req_pages; i++) {
498 cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
499 base += PAGE_SIZE;
500 }
501
502 base = adapter->cmpRingPA;
503 for (i = 0; i < adapter->cmp_pages; i++) {
504 cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
505 base += PAGE_SIZE;
506 }
507
508 memset(adapter->rings_state, 0, PAGE_SIZE);
509 memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
510 memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
511
512 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
513 &cmd, sizeof(cmd));
514
515 if (adapter->use_msg) {
516 struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
517
518 cmd_msg.numPages = adapter->msg_pages;
519
520 base = adapter->msgRingPA;
521 for (i = 0; i < adapter->msg_pages; i++) {
522 cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
523 base += PAGE_SIZE;
524 }
525 memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
526
527 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
528 &cmd_msg, sizeof(cmd_msg));
529 }
530}
531
532static int pvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
533{
534 if (!sdev->tagged_supported)
535 qdepth = 1;
536 return scsi_change_queue_depth(sdev, qdepth);
537}
538
539/*
540 * Pull a completion descriptor off and pass the completion back
541 * to the SCSI mid layer.
542 */
543static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
544 const struct PVSCSIRingCmpDesc *e)
545{
546 struct pvscsi_ctx *ctx;
547 struct scsi_cmnd *cmd;
548 struct completion *abort_cmp;
549 u32 btstat = e->hostStatus;
550 u32 sdstat = e->scsiStatus;
551
552 ctx = pvscsi_get_context(adapter, e->context);
553 cmd = ctx->cmd;
554 abort_cmp = ctx->abort_cmp;
555 pvscsi_unmap_buffers(adapter, ctx);
556 if (sdstat != SAM_STAT_CHECK_CONDITION)
557 pvscsi_patch_sense(cmd);
558 pvscsi_release_context(adapter, ctx);
559 if (abort_cmp) {
560 /*
561 * The command was requested to be aborted. Just signal that
562 * the request completed and swallow the actual cmd completion
563 * here. The abort handler will post a completion for this
564 * command indicating that it got successfully aborted.
565 */
566 complete(abort_cmp);
567 return;
568 }
569
570 cmd->result = 0;
571 if (sdstat != SAM_STAT_GOOD &&
572 (btstat == BTSTAT_SUCCESS ||
573 btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
574 btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
575 if (sdstat == SAM_STAT_COMMAND_TERMINATED) {
576 cmd->result = (DID_RESET << 16);
577 } else {
578 cmd->result = (DID_OK << 16) | sdstat;
579 }
580 } else
581 switch (btstat) {
582 case BTSTAT_SUCCESS:
583 case BTSTAT_LINKED_COMMAND_COMPLETED:
584 case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
585 /*
586 * Commands like INQUIRY may transfer less data than
587 * requested by the initiator via bufflen. Set residual
588 * count to make upper layer aware of the actual amount
589 * of data returned. There are cases when controller
590 * returns zero dataLen with non zero data - do not set
591 * residual count in that case.
592 */
593 if (e->dataLen && (e->dataLen < scsi_bufflen(cmd)))
594 scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
595 cmd->result = (DID_OK << 16);
596 break;
597
598 case BTSTAT_DATARUN:
599 case BTSTAT_DATA_UNDERRUN:
600 /* Report residual data in underruns */
601 scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
602 cmd->result = (DID_ERROR << 16);
603 break;
604
605 case BTSTAT_SELTIMEO:
606 /* Our emulation returns this for non-connected devs */
607 cmd->result = (DID_BAD_TARGET << 16);
608 break;
609
610 case BTSTAT_LUNMISMATCH:
611 case BTSTAT_TAGREJECT:
612 case BTSTAT_BADMSG:
613 case BTSTAT_HAHARDWARE:
614 case BTSTAT_INVPHASE:
615 case BTSTAT_HATIMEOUT:
616 case BTSTAT_NORESPONSE:
617 case BTSTAT_DISCONNECT:
618 case BTSTAT_HASOFTWARE:
619 case BTSTAT_BUSFREE:
620 case BTSTAT_SENSFAILED:
621 cmd->result |= (DID_ERROR << 16);
622 break;
623
624 case BTSTAT_SENTRST:
625 case BTSTAT_RECVRST:
626 case BTSTAT_BUSRESET:
627 cmd->result = (DID_RESET << 16);
628 break;
629
630 case BTSTAT_ABORTQUEUE:
631 cmd->result = (DID_BUS_BUSY << 16);
632 break;
633
634 case BTSTAT_SCSIPARITY:
635 cmd->result = (DID_PARITY << 16);
636 break;
637
638 default:
639 cmd->result = (DID_ERROR << 16);
640 scmd_printk(KERN_DEBUG, cmd,
641 "Unknown completion status: 0x%x\n",
642 btstat);
643 }
644
645 dev_dbg(&cmd->device->sdev_gendev,
646 "cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
647 cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
648
649 scsi_done(cmd);
650}
651
652/*
653 * barrier usage : Since the PVSCSI device is emulated, there could be cases
654 * where we may want to serialize some accesses between the driver and the
655 * emulation layer. We use compiler barriers instead of the more expensive
656 * memory barriers because PVSCSI is only supported on X86 which has strong
657 * memory access ordering.
658 */
659static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
660{
661 struct PVSCSIRingsState *s = adapter->rings_state;
662 struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
663 u32 cmp_entries = s->cmpNumEntriesLog2;
664
665 while (s->cmpConsIdx != s->cmpProdIdx) {
666 struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
667 MASK(cmp_entries));
668 /*
669 * This barrier() ensures that *e is not dereferenced while
670 * the device emulation still writes data into the slot.
671 * Since the device emulation advances s->cmpProdIdx only after
672 * updating the slot we want to check it first.
673 */
674 barrier();
675 pvscsi_complete_request(adapter, e);
676 /*
677 * This barrier() ensures that compiler doesn't reorder write
678 * to s->cmpConsIdx before the read of (*e) inside
679 * pvscsi_complete_request. Otherwise, device emulation may
680 * overwrite *e before we had a chance to read it.
681 */
682 barrier();
683 s->cmpConsIdx++;
684 }
685}
686
687/*
688 * Translate a Linux SCSI request into a request ring entry.
689 */
690static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
691 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
692{
693 struct PVSCSIRingsState *s;
694 struct PVSCSIRingReqDesc *e;
695 struct scsi_device *sdev;
696 u32 req_entries;
697
698 s = adapter->rings_state;
699 sdev = cmd->device;
700 req_entries = s->reqNumEntriesLog2;
701
702 /*
703 * If this condition holds, we might have room on the request ring, but
704 * we might not have room on the completion ring for the response.
705 * However, we have already ruled out this possibility - we would not
706 * have successfully allocated a context if it were true, since we only
707 * have one context per request entry. Check for it anyway, since it
708 * would be a serious bug.
709 */
710 if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
711 scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
712 "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
713 s->reqProdIdx, s->cmpConsIdx);
714 return -1;
715 }
716
717 e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
718
719 e->bus = sdev->channel;
720 e->target = sdev->id;
721 memset(e->lun, 0, sizeof(e->lun));
722 e->lun[1] = sdev->lun;
723
724 if (cmd->sense_buffer) {
725 ctx->sensePA = dma_map_single(&adapter->dev->dev,
726 cmd->sense_buffer, SCSI_SENSE_BUFFERSIZE,
727 DMA_FROM_DEVICE);
728 if (dma_mapping_error(&adapter->dev->dev, ctx->sensePA)) {
729 scmd_printk(KERN_DEBUG, cmd,
730 "vmw_pvscsi: Failed to map sense buffer for DMA.\n");
731 ctx->sensePA = 0;
732 return -ENOMEM;
733 }
734 e->senseAddr = ctx->sensePA;
735 e->senseLen = SCSI_SENSE_BUFFERSIZE;
736 } else {
737 e->senseLen = 0;
738 e->senseAddr = 0;
739 }
740 e->cdbLen = cmd->cmd_len;
741 e->vcpuHint = smp_processor_id();
742 memcpy(e->cdb, cmd->cmnd, e->cdbLen);
743
744 e->tag = SIMPLE_QUEUE_TAG;
745
746 if (cmd->sc_data_direction == DMA_FROM_DEVICE)
747 e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
748 else if (cmd->sc_data_direction == DMA_TO_DEVICE)
749 e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
750 else if (cmd->sc_data_direction == DMA_NONE)
751 e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
752 else
753 e->flags = 0;
754
755 if (pvscsi_map_buffers(adapter, ctx, cmd, e) != 0) {
756 if (cmd->sense_buffer) {
757 dma_unmap_single(&adapter->dev->dev, ctx->sensePA,
758 SCSI_SENSE_BUFFERSIZE,
759 DMA_FROM_DEVICE);
760 ctx->sensePA = 0;
761 }
762 return -ENOMEM;
763 }
764
765 e->context = pvscsi_map_context(adapter, ctx);
766
767 barrier();
768
769 s->reqProdIdx++;
770
771 return 0;
772}
773
774static int pvscsi_queue_lck(struct scsi_cmnd *cmd)
775{
776 struct Scsi_Host *host = cmd->device->host;
777 struct pvscsi_adapter *adapter = shost_priv(host);
778 struct pvscsi_ctx *ctx;
779 unsigned long flags;
780 unsigned char op;
781
782 spin_lock_irqsave(&adapter->hw_lock, flags);
783
784 ctx = pvscsi_acquire_context(adapter, cmd);
785 if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
786 if (ctx)
787 pvscsi_release_context(adapter, ctx);
788 spin_unlock_irqrestore(&adapter->hw_lock, flags);
789 return SCSI_MLQUEUE_HOST_BUSY;
790 }
791
792 op = cmd->cmnd[0];
793
794 dev_dbg(&cmd->device->sdev_gendev,
795 "queued cmd %p, ctx %p, op=%x\n", cmd, ctx, op);
796
797 spin_unlock_irqrestore(&adapter->hw_lock, flags);
798
799 pvscsi_kick_io(adapter, op);
800
801 return 0;
802}
803
804static DEF_SCSI_QCMD(pvscsi_queue)
805
806static int pvscsi_abort(struct scsi_cmnd *cmd)
807{
808 struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
809 struct pvscsi_ctx *ctx;
810 unsigned long flags;
811 int result = SUCCESS;
812 DECLARE_COMPLETION_ONSTACK(abort_cmp);
813 int done;
814
815 scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
816 adapter->host->host_no, cmd);
817
818 spin_lock_irqsave(&adapter->hw_lock, flags);
819
820 /*
821 * Poll the completion ring first - we might be trying to abort
822 * a command that is waiting to be dispatched in the completion ring.
823 */
824 pvscsi_process_completion_ring(adapter);
825
826 /*
827 * If there is no context for the command, it either already succeeded
828 * or else was never properly issued. Not our problem.
829 */
830 ctx = pvscsi_find_context(adapter, cmd);
831 if (!ctx) {
832 scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
833 goto out;
834 }
835
836 /*
837 * Mark that the command has been requested to be aborted and issue
838 * the abort.
839 */
840 ctx->abort_cmp = &abort_cmp;
841
842 pvscsi_abort_cmd(adapter, ctx);
843 spin_unlock_irqrestore(&adapter->hw_lock, flags);
844 /* Wait for 2 secs for the completion. */
845 done = wait_for_completion_timeout(&abort_cmp, msecs_to_jiffies(2000));
846 spin_lock_irqsave(&adapter->hw_lock, flags);
847
848 if (!done) {
849 /*
850 * Failed to abort the command, unmark the fact that it
851 * was requested to be aborted.
852 */
853 ctx->abort_cmp = NULL;
854 result = FAILED;
855 scmd_printk(KERN_DEBUG, cmd,
856 "Failed to get completion for aborted cmd %p\n",
857 cmd);
858 goto out;
859 }
860
861 /*
862 * Successfully aborted the command.
863 */
864 cmd->result = (DID_ABORT << 16);
865 scsi_done(cmd);
866
867out:
868 spin_unlock_irqrestore(&adapter->hw_lock, flags);
869 return result;
870}
871
872/*
873 * Abort all outstanding requests. This is only safe to use if the completion
874 * ring will never be walked again or the device has been reset, because it
875 * destroys the 1-1 mapping between context field passed to emulation and our
876 * request structure.
877 */
878static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
879{
880 unsigned i;
881
882 for (i = 0; i < adapter->req_depth; i++) {
883 struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
884 struct scsi_cmnd *cmd = ctx->cmd;
885 if (cmd) {
886 scmd_printk(KERN_ERR, cmd,
887 "Forced reset on cmd %p\n", cmd);
888 pvscsi_unmap_buffers(adapter, ctx);
889 pvscsi_patch_sense(cmd);
890 pvscsi_release_context(adapter, ctx);
891 cmd->result = (DID_RESET << 16);
892 scsi_done(cmd);
893 }
894 }
895}
896
897static int pvscsi_host_reset(struct scsi_cmnd *cmd)
898{
899 struct Scsi_Host *host = cmd->device->host;
900 struct pvscsi_adapter *adapter = shost_priv(host);
901 unsigned long flags;
902 bool use_msg;
903
904 scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
905
906 spin_lock_irqsave(&adapter->hw_lock, flags);
907
908 use_msg = adapter->use_msg;
909
910 if (use_msg) {
911 adapter->use_msg = false;
912 spin_unlock_irqrestore(&adapter->hw_lock, flags);
913
914 /*
915 * Now that we know that the ISR won't add more work on the
916 * workqueue we can safely flush any outstanding work.
917 */
918 flush_workqueue(adapter->workqueue);
919 spin_lock_irqsave(&adapter->hw_lock, flags);
920 }
921
922 /*
923 * We're going to tear down the entire ring structure and set it back
924 * up, so stalling new requests until all completions are flushed and
925 * the rings are back in place.
926 */
927
928 pvscsi_process_request_ring(adapter);
929
930 ll_adapter_reset(adapter);
931
932 /*
933 * Now process any completions. Note we do this AFTER adapter reset,
934 * which is strange, but stops races where completions get posted
935 * between processing the ring and issuing the reset. The backend will
936 * not touch the ring memory after reset, so the immediately pre-reset
937 * completion ring state is still valid.
938 */
939 pvscsi_process_completion_ring(adapter);
940
941 pvscsi_reset_all(adapter);
942 adapter->use_msg = use_msg;
943 pvscsi_setup_all_rings(adapter);
944 pvscsi_unmask_intr(adapter);
945
946 spin_unlock_irqrestore(&adapter->hw_lock, flags);
947
948 return SUCCESS;
949}
950
951static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
952{
953 struct Scsi_Host *host = cmd->device->host;
954 struct pvscsi_adapter *adapter = shost_priv(host);
955 unsigned long flags;
956
957 scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
958
959 /*
960 * We don't want to queue new requests for this bus after
961 * flushing all pending requests to emulation, since new
962 * requests could then sneak in during this bus reset phase,
963 * so take the lock now.
964 */
965 spin_lock_irqsave(&adapter->hw_lock, flags);
966
967 pvscsi_process_request_ring(adapter);
968 ll_bus_reset(adapter);
969 pvscsi_process_completion_ring(adapter);
970
971 spin_unlock_irqrestore(&adapter->hw_lock, flags);
972
973 return SUCCESS;
974}
975
976static int pvscsi_device_reset(struct scsi_cmnd *cmd)
977{
978 struct Scsi_Host *host = cmd->device->host;
979 struct pvscsi_adapter *adapter = shost_priv(host);
980 unsigned long flags;
981
982 scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
983 host->host_no, cmd->device->id);
984
985 /*
986 * We don't want to queue new requests for this device after flushing
987 * all pending requests to emulation, since new requests could then
988 * sneak in during this device reset phase, so take the lock now.
989 */
990 spin_lock_irqsave(&adapter->hw_lock, flags);
991
992 pvscsi_process_request_ring(adapter);
993 ll_device_reset(adapter, cmd->device->id);
994 pvscsi_process_completion_ring(adapter);
995
996 spin_unlock_irqrestore(&adapter->hw_lock, flags);
997
998 return SUCCESS;
999}
1000
1001static struct scsi_host_template pvscsi_template;
1002
1003static const char *pvscsi_info(struct Scsi_Host *host)
1004{
1005 struct pvscsi_adapter *adapter = shost_priv(host);
1006 static char buf[256];
1007
1008 sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
1009 "%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
1010 adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
1011 pvscsi_template.cmd_per_lun);
1012
1013 return buf;
1014}
1015
1016static struct scsi_host_template pvscsi_template = {
1017 .module = THIS_MODULE,
1018 .name = "VMware PVSCSI Host Adapter",
1019 .proc_name = "vmw_pvscsi",
1020 .info = pvscsi_info,
1021 .queuecommand = pvscsi_queue,
1022 .this_id = -1,
1023 .sg_tablesize = PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
1024 .dma_boundary = UINT_MAX,
1025 .max_sectors = 0xffff,
1026 .change_queue_depth = pvscsi_change_queue_depth,
1027 .eh_abort_handler = pvscsi_abort,
1028 .eh_device_reset_handler = pvscsi_device_reset,
1029 .eh_bus_reset_handler = pvscsi_bus_reset,
1030 .eh_host_reset_handler = pvscsi_host_reset,
1031};
1032
1033static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
1034 const struct PVSCSIRingMsgDesc *e)
1035{
1036 struct PVSCSIRingsState *s = adapter->rings_state;
1037 struct Scsi_Host *host = adapter->host;
1038 struct scsi_device *sdev;
1039
1040 printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
1041 e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
1042
1043 BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
1044
1045 if (e->type == PVSCSI_MSG_DEV_ADDED) {
1046 struct PVSCSIMsgDescDevStatusChanged *desc;
1047 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1048
1049 printk(KERN_INFO
1050 "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
1051 desc->bus, desc->target, desc->lun[1]);
1052
1053 if (!scsi_host_get(host))
1054 return;
1055
1056 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1057 desc->lun[1]);
1058 if (sdev) {
1059 printk(KERN_INFO "vmw_pvscsi: device already exists\n");
1060 scsi_device_put(sdev);
1061 } else
1062 scsi_add_device(adapter->host, desc->bus,
1063 desc->target, desc->lun[1]);
1064
1065 scsi_host_put(host);
1066 } else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
1067 struct PVSCSIMsgDescDevStatusChanged *desc;
1068 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1069
1070 printk(KERN_INFO
1071 "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
1072 desc->bus, desc->target, desc->lun[1]);
1073
1074 if (!scsi_host_get(host))
1075 return;
1076
1077 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1078 desc->lun[1]);
1079 if (sdev) {
1080 scsi_remove_device(sdev);
1081 scsi_device_put(sdev);
1082 } else
1083 printk(KERN_INFO
1084 "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
1085 desc->bus, desc->target, desc->lun[1]);
1086
1087 scsi_host_put(host);
1088 }
1089}
1090
1091static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
1092{
1093 struct PVSCSIRingsState *s = adapter->rings_state;
1094
1095 return s->msgProdIdx != s->msgConsIdx;
1096}
1097
1098static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
1099{
1100 struct PVSCSIRingsState *s = adapter->rings_state;
1101 struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
1102 u32 msg_entries = s->msgNumEntriesLog2;
1103
1104 while (pvscsi_msg_pending(adapter)) {
1105 struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
1106 MASK(msg_entries));
1107
1108 barrier();
1109 pvscsi_process_msg(adapter, e);
1110 barrier();
1111 s->msgConsIdx++;
1112 }
1113}
1114
1115static void pvscsi_msg_workqueue_handler(struct work_struct *data)
1116{
1117 struct pvscsi_adapter *adapter;
1118
1119 adapter = container_of(data, struct pvscsi_adapter, work);
1120
1121 pvscsi_process_msg_ring(adapter);
1122}
1123
1124static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
1125{
1126 char name[32];
1127
1128 if (!pvscsi_use_msg)
1129 return 0;
1130
1131 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1132 PVSCSI_CMD_SETUP_MSG_RING);
1133
1134 if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
1135 return 0;
1136
1137 snprintf(name, sizeof(name),
1138 "vmw_pvscsi_wq_%u", adapter->host->host_no);
1139
1140 adapter->workqueue =
1141 alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM, name);
1142 if (!adapter->workqueue) {
1143 printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
1144 return 0;
1145 }
1146 INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
1147
1148 return 1;
1149}
1150
1151static bool pvscsi_setup_req_threshold(struct pvscsi_adapter *adapter,
1152 bool enable)
1153{
1154 u32 val;
1155
1156 if (!pvscsi_use_req_threshold)
1157 return false;
1158
1159 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1160 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD);
1161 val = pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS);
1162 if (val == -1) {
1163 printk(KERN_INFO "vmw_pvscsi: device does not support req_threshold\n");
1164 return false;
1165 } else {
1166 struct PVSCSICmdDescSetupReqCall cmd_msg = { 0 };
1167 cmd_msg.enable = enable;
1168 printk(KERN_INFO
1169 "vmw_pvscsi: %sabling reqCallThreshold\n",
1170 enable ? "en" : "dis");
1171 pvscsi_write_cmd_desc(adapter,
1172 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD,
1173 &cmd_msg, sizeof(cmd_msg));
1174 return pvscsi_reg_read(adapter,
1175 PVSCSI_REG_OFFSET_COMMAND_STATUS) != 0;
1176 }
1177}
1178
1179static irqreturn_t pvscsi_isr(int irq, void *devp)
1180{
1181 struct pvscsi_adapter *adapter = devp;
1182 unsigned long flags;
1183
1184 spin_lock_irqsave(&adapter->hw_lock, flags);
1185 pvscsi_process_completion_ring(adapter);
1186 if (adapter->use_msg && pvscsi_msg_pending(adapter))
1187 queue_work(adapter->workqueue, &adapter->work);
1188 spin_unlock_irqrestore(&adapter->hw_lock, flags);
1189
1190 return IRQ_HANDLED;
1191}
1192
1193static irqreturn_t pvscsi_shared_isr(int irq, void *devp)
1194{
1195 struct pvscsi_adapter *adapter = devp;
1196 u32 val = pvscsi_read_intr_status(adapter);
1197
1198 if (!(val & PVSCSI_INTR_ALL_SUPPORTED))
1199 return IRQ_NONE;
1200 pvscsi_write_intr_status(devp, val);
1201 return pvscsi_isr(irq, devp);
1202}
1203
1204static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
1205{
1206 struct pvscsi_ctx *ctx = adapter->cmd_map;
1207 unsigned i;
1208
1209 for (i = 0; i < adapter->req_depth; ++i, ++ctx)
1210 free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
1211}
1212
1213static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
1214{
1215 free_irq(pci_irq_vector(adapter->dev, 0), adapter);
1216 pci_free_irq_vectors(adapter->dev);
1217}
1218
1219static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
1220{
1221 if (adapter->workqueue)
1222 destroy_workqueue(adapter->workqueue);
1223
1224 if (adapter->mmioBase)
1225 pci_iounmap(adapter->dev, adapter->mmioBase);
1226
1227 pci_release_regions(adapter->dev);
1228
1229 if (adapter->cmd_map) {
1230 pvscsi_free_sgls(adapter);
1231 kfree(adapter->cmd_map);
1232 }
1233
1234 if (adapter->rings_state)
1235 dma_free_coherent(&adapter->dev->dev, PAGE_SIZE,
1236 adapter->rings_state, adapter->ringStatePA);
1237
1238 if (adapter->req_ring)
1239 dma_free_coherent(&adapter->dev->dev,
1240 adapter->req_pages * PAGE_SIZE,
1241 adapter->req_ring, adapter->reqRingPA);
1242
1243 if (adapter->cmp_ring)
1244 dma_free_coherent(&adapter->dev->dev,
1245 adapter->cmp_pages * PAGE_SIZE,
1246 adapter->cmp_ring, adapter->cmpRingPA);
1247
1248 if (adapter->msg_ring)
1249 dma_free_coherent(&adapter->dev->dev,
1250 adapter->msg_pages * PAGE_SIZE,
1251 adapter->msg_ring, adapter->msgRingPA);
1252}
1253
1254/*
1255 * Allocate scatter gather lists.
1256 *
1257 * These are statically allocated. Trying to be clever was not worth it.
1258 *
1259 * Dynamic allocation can fail, and we can't go deep into the memory
1260 * allocator, since we're a SCSI driver, and trying too hard to allocate
1261 * memory might generate disk I/O. We also don't want to fail disk I/O
1262 * in that case because we can't get an allocation - the I/O could be
1263 * trying to swap out data to free memory. Since that is pathological,
1264 * just use a statically allocated scatter list.
1265 *
1266 */
1267static int pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
1268{
1269 struct pvscsi_ctx *ctx;
1270 int i;
1271
1272 ctx = adapter->cmd_map;
1273 BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
1274
1275 for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
1276 ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
1277 get_order(SGL_SIZE));
1278 ctx->sglPA = 0;
1279 BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
1280 if (!ctx->sgl) {
1281 for (; i >= 0; --i, --ctx) {
1282 free_pages((unsigned long)ctx->sgl,
1283 get_order(SGL_SIZE));
1284 ctx->sgl = NULL;
1285 }
1286 return -ENOMEM;
1287 }
1288 }
1289
1290 return 0;
1291}
1292
1293/*
1294 * Query the device, fetch the config info and return the
1295 * maximum number of targets on the adapter. In case of
1296 * failure due to any reason return default i.e. 16.
1297 */
1298static u32 pvscsi_get_max_targets(struct pvscsi_adapter *adapter)
1299{
1300 struct PVSCSICmdDescConfigCmd cmd;
1301 struct PVSCSIConfigPageHeader *header;
1302 struct device *dev;
1303 dma_addr_t configPagePA;
1304 void *config_page;
1305 u32 numPhys = 16;
1306
1307 dev = pvscsi_dev(adapter);
1308 config_page = dma_alloc_coherent(&adapter->dev->dev, PAGE_SIZE,
1309 &configPagePA, GFP_KERNEL);
1310 if (!config_page) {
1311 dev_warn(dev, "vmw_pvscsi: failed to allocate memory for config page\n");
1312 goto exit;
1313 }
1314 BUG_ON(configPagePA & ~PAGE_MASK);
1315
1316 /* Fetch config info from the device. */
1317 cmd.configPageAddress = ((u64)PVSCSI_CONFIG_CONTROLLER_ADDRESS) << 32;
1318 cmd.configPageNum = PVSCSI_CONFIG_PAGE_CONTROLLER;
1319 cmd.cmpAddr = configPagePA;
1320 cmd._pad = 0;
1321
1322 /*
1323 * Mark the completion page header with error values. If the device
1324 * completes the command successfully, it sets the status values to
1325 * indicate success.
1326 */
1327 header = config_page;
1328 header->hostStatus = BTSTAT_INVPARAM;
1329 header->scsiStatus = SDSTAT_CHECK;
1330
1331 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_CONFIG, &cmd, sizeof cmd);
1332
1333 if (header->hostStatus == BTSTAT_SUCCESS &&
1334 header->scsiStatus == SDSTAT_GOOD) {
1335 struct PVSCSIConfigPageController *config;
1336
1337 config = config_page;
1338 numPhys = config->numPhys;
1339 } else
1340 dev_warn(dev, "vmw_pvscsi: PVSCSI_CMD_CONFIG failed. hostStatus = 0x%x, scsiStatus = 0x%x\n",
1341 header->hostStatus, header->scsiStatus);
1342 dma_free_coherent(&adapter->dev->dev, PAGE_SIZE, config_page,
1343 configPagePA);
1344exit:
1345 return numPhys;
1346}
1347
1348static int pvscsi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1349{
1350 unsigned int irq_flag = PCI_IRQ_ALL_TYPES;
1351 struct pvscsi_adapter *adapter;
1352 struct pvscsi_adapter adapter_temp;
1353 struct Scsi_Host *host = NULL;
1354 unsigned int i;
1355 int error;
1356 u32 max_id;
1357
1358 error = -ENODEV;
1359
1360 if (pci_enable_device(pdev))
1361 return error;
1362
1363 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
1364 printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
1365 } else if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32))) {
1366 printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
1367 } else {
1368 printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
1369 goto out_disable_device;
1370 }
1371
1372 /*
1373 * Let's use a temp pvscsi_adapter struct until we find the number of
1374 * targets on the adapter, after that we will switch to the real
1375 * allocated struct.
1376 */
1377 adapter = &adapter_temp;
1378 memset(adapter, 0, sizeof(*adapter));
1379 adapter->dev = pdev;
1380 adapter->rev = pdev->revision;
1381
1382 if (pci_request_regions(pdev, "vmw_pvscsi")) {
1383 printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
1384 goto out_disable_device;
1385 }
1386
1387 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1388 if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
1389 continue;
1390
1391 if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
1392 continue;
1393
1394 break;
1395 }
1396
1397 if (i == DEVICE_COUNT_RESOURCE) {
1398 printk(KERN_ERR
1399 "vmw_pvscsi: adapter has no suitable MMIO region\n");
1400 goto out_release_resources_and_disable;
1401 }
1402
1403 adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
1404
1405 if (!adapter->mmioBase) {
1406 printk(KERN_ERR
1407 "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
1408 i, PVSCSI_MEM_SPACE_SIZE);
1409 goto out_release_resources_and_disable;
1410 }
1411
1412 pci_set_master(pdev);
1413
1414 /*
1415 * Ask the device for max number of targets before deciding the
1416 * default pvscsi_ring_pages value.
1417 */
1418 max_id = pvscsi_get_max_targets(adapter);
1419 printk(KERN_INFO "vmw_pvscsi: max_id: %u\n", max_id);
1420
1421 if (pvscsi_ring_pages == 0)
1422 /*
1423 * Set the right default value. Up to 16 it is 8, above it is
1424 * max.
1425 */
1426 pvscsi_ring_pages = (max_id > 16) ?
1427 PVSCSI_SETUP_RINGS_MAX_NUM_PAGES :
1428 PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
1429 printk(KERN_INFO
1430 "vmw_pvscsi: setting ring_pages to %d\n",
1431 pvscsi_ring_pages);
1432
1433 pvscsi_template.can_queue =
1434 min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
1435 PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
1436 pvscsi_template.cmd_per_lun =
1437 min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
1438 host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
1439 if (!host) {
1440 printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
1441 goto out_release_resources_and_disable;
1442 }
1443
1444 /*
1445 * Let's use the real pvscsi_adapter struct here onwards.
1446 */
1447 adapter = shost_priv(host);
1448 memset(adapter, 0, sizeof(*adapter));
1449 adapter->dev = pdev;
1450 adapter->host = host;
1451 /*
1452 * Copy back what we already have to the allocated adapter struct.
1453 */
1454 adapter->rev = adapter_temp.rev;
1455 adapter->mmioBase = adapter_temp.mmioBase;
1456
1457 spin_lock_init(&adapter->hw_lock);
1458 host->max_channel = 0;
1459 host->max_lun = 1;
1460 host->max_cmd_len = 16;
1461 host->max_id = max_id;
1462
1463 pci_set_drvdata(pdev, host);
1464
1465 ll_adapter_reset(adapter);
1466
1467 adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
1468
1469 error = pvscsi_allocate_rings(adapter);
1470 if (error) {
1471 printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
1472 goto out_release_resources;
1473 }
1474
1475 /*
1476 * From this point on we should reset the adapter if anything goes
1477 * wrong.
1478 */
1479 pvscsi_setup_all_rings(adapter);
1480
1481 adapter->cmd_map = kcalloc(adapter->req_depth,
1482 sizeof(struct pvscsi_ctx), GFP_KERNEL);
1483 if (!adapter->cmd_map) {
1484 printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
1485 error = -ENOMEM;
1486 goto out_reset_adapter;
1487 }
1488
1489 INIT_LIST_HEAD(&adapter->cmd_pool);
1490 for (i = 0; i < adapter->req_depth; i++) {
1491 struct pvscsi_ctx *ctx = adapter->cmd_map + i;
1492 list_add(&ctx->list, &adapter->cmd_pool);
1493 }
1494
1495 error = pvscsi_allocate_sg(adapter);
1496 if (error) {
1497 printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
1498 goto out_reset_adapter;
1499 }
1500
1501 if (pvscsi_disable_msix)
1502 irq_flag &= ~PCI_IRQ_MSIX;
1503 if (pvscsi_disable_msi)
1504 irq_flag &= ~PCI_IRQ_MSI;
1505
1506 error = pci_alloc_irq_vectors(adapter->dev, 1, 1, irq_flag);
1507 if (error < 0)
1508 goto out_reset_adapter;
1509
1510 adapter->use_req_threshold = pvscsi_setup_req_threshold(adapter, true);
1511 printk(KERN_DEBUG "vmw_pvscsi: driver-based request coalescing %sabled\n",
1512 adapter->use_req_threshold ? "en" : "dis");
1513
1514 if (adapter->dev->msix_enabled || adapter->dev->msi_enabled) {
1515 printk(KERN_INFO "vmw_pvscsi: using MSI%s\n",
1516 adapter->dev->msix_enabled ? "-X" : "");
1517 error = request_irq(pci_irq_vector(pdev, 0), pvscsi_isr,
1518 0, "vmw_pvscsi", adapter);
1519 } else {
1520 printk(KERN_INFO "vmw_pvscsi: using INTx\n");
1521 error = request_irq(pci_irq_vector(pdev, 0), pvscsi_shared_isr,
1522 IRQF_SHARED, "vmw_pvscsi", adapter);
1523 }
1524
1525 if (error) {
1526 printk(KERN_ERR
1527 "vmw_pvscsi: unable to request IRQ: %d\n", error);
1528 goto out_reset_adapter;
1529 }
1530
1531 error = scsi_add_host(host, &pdev->dev);
1532 if (error) {
1533 printk(KERN_ERR
1534 "vmw_pvscsi: scsi_add_host failed: %d\n", error);
1535 goto out_reset_adapter;
1536 }
1537
1538 dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
1539 adapter->rev, host->host_no);
1540
1541 pvscsi_unmask_intr(adapter);
1542
1543 scsi_scan_host(host);
1544
1545 return 0;
1546
1547out_reset_adapter:
1548 ll_adapter_reset(adapter);
1549out_release_resources:
1550 pvscsi_shutdown_intr(adapter);
1551 pvscsi_release_resources(adapter);
1552 scsi_host_put(host);
1553out_disable_device:
1554 pci_disable_device(pdev);
1555
1556 return error;
1557
1558out_release_resources_and_disable:
1559 pvscsi_shutdown_intr(adapter);
1560 pvscsi_release_resources(adapter);
1561 goto out_disable_device;
1562}
1563
1564static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
1565{
1566 pvscsi_mask_intr(adapter);
1567
1568 if (adapter->workqueue)
1569 flush_workqueue(adapter->workqueue);
1570
1571 pvscsi_shutdown_intr(adapter);
1572
1573 pvscsi_process_request_ring(adapter);
1574 pvscsi_process_completion_ring(adapter);
1575 ll_adapter_reset(adapter);
1576}
1577
1578static void pvscsi_shutdown(struct pci_dev *dev)
1579{
1580 struct Scsi_Host *host = pci_get_drvdata(dev);
1581 struct pvscsi_adapter *adapter = shost_priv(host);
1582
1583 __pvscsi_shutdown(adapter);
1584}
1585
1586static void pvscsi_remove(struct pci_dev *pdev)
1587{
1588 struct Scsi_Host *host = pci_get_drvdata(pdev);
1589 struct pvscsi_adapter *adapter = shost_priv(host);
1590
1591 scsi_remove_host(host);
1592
1593 __pvscsi_shutdown(adapter);
1594 pvscsi_release_resources(adapter);
1595
1596 scsi_host_put(host);
1597
1598 pci_disable_device(pdev);
1599}
1600
1601static struct pci_driver pvscsi_pci_driver = {
1602 .name = "vmw_pvscsi",
1603 .id_table = pvscsi_pci_tbl,
1604 .probe = pvscsi_probe,
1605 .remove = pvscsi_remove,
1606 .shutdown = pvscsi_shutdown,
1607};
1608
1609static int __init pvscsi_init(void)
1610{
1611 pr_info("%s - version %s\n",
1612 PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
1613 return pci_register_driver(&pvscsi_pci_driver);
1614}
1615
1616static void __exit pvscsi_exit(void)
1617{
1618 pci_unregister_driver(&pvscsi_pci_driver);
1619}
1620
1621module_init(pvscsi_init);
1622module_exit(pvscsi_exit);