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 * Maintained by: Arvind Kumar <arvindkumar@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#include <scsi/scsi_tcq.h>
36
37#include "vmw_pvscsi.h"
38
39#define PVSCSI_LINUX_DRIVER_DESC "VMware PVSCSI driver"
40
41MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC);
42MODULE_AUTHOR("VMware, Inc.");
43MODULE_LICENSE("GPL");
44MODULE_VERSION(PVSCSI_DRIVER_VERSION_STRING);
45
46#define PVSCSI_DEFAULT_NUM_PAGES_PER_RING 8
47#define PVSCSI_DEFAULT_NUM_PAGES_MSG_RING 1
48#define PVSCSI_DEFAULT_QUEUE_DEPTH 254
49#define SGL_SIZE PAGE_SIZE
50
51struct pvscsi_sg_list {
52 struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT];
53};
54
55struct pvscsi_ctx {
56 /*
57 * The index of the context in cmd_map serves as the context ID for a
58 * 1-to-1 mapping completions back to requests.
59 */
60 struct scsi_cmnd *cmd;
61 struct pvscsi_sg_list *sgl;
62 struct list_head list;
63 dma_addr_t dataPA;
64 dma_addr_t sensePA;
65 dma_addr_t sglPA;
66 struct completion *abort_cmp;
67};
68
69struct pvscsi_adapter {
70 char *mmioBase;
71 unsigned int irq;
72 u8 rev;
73 bool use_msi;
74 bool use_msix;
75 bool use_msg;
76 bool use_req_threshold;
77
78 spinlock_t hw_lock;
79
80 struct workqueue_struct *workqueue;
81 struct work_struct work;
82
83 struct PVSCSIRingReqDesc *req_ring;
84 unsigned req_pages;
85 unsigned req_depth;
86 dma_addr_t reqRingPA;
87
88 struct PVSCSIRingCmpDesc *cmp_ring;
89 unsigned cmp_pages;
90 dma_addr_t cmpRingPA;
91
92 struct PVSCSIRingMsgDesc *msg_ring;
93 unsigned msg_pages;
94 dma_addr_t msgRingPA;
95
96 struct PVSCSIRingsState *rings_state;
97 dma_addr_t ringStatePA;
98
99 struct pci_dev *dev;
100 struct Scsi_Host *host;
101
102 struct list_head cmd_pool;
103 struct pvscsi_ctx *cmd_map;
104};
105
106
107/* Command line parameters */
108static int pvscsi_ring_pages;
109static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
110static int pvscsi_cmd_per_lun = PVSCSI_DEFAULT_QUEUE_DEPTH;
111static bool pvscsi_disable_msi;
112static bool pvscsi_disable_msix;
113static bool pvscsi_use_msg = true;
114static bool pvscsi_use_req_threshold = true;
115
116#define PVSCSI_RW (S_IRUSR | S_IWUSR)
117
118module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
119MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default="
120 __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING)
121 "[up to 16 targets],"
122 __stringify(PVSCSI_SETUP_RINGS_MAX_NUM_PAGES)
123 "[for 16+ targets])");
124
125module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
126MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
127 __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
128
129module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
130MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
131 __stringify(PVSCSI_DEFAULT_QUEUE_DEPTH) ")");
132
133module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
134MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
135
136module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
137MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
138
139module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
140MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
141
142module_param_named(use_req_threshold, pvscsi_use_req_threshold,
143 bool, PVSCSI_RW);
144MODULE_PARM_DESC(use_req_threshold, "Use driver-based request coalescing if configured - (default=1)");
145
146static const struct pci_device_id pvscsi_pci_tbl[] = {
147 { PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
148 { 0 }
149};
150
151MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
152
153static struct device *
154pvscsi_dev(const struct pvscsi_adapter *adapter)
155{
156 return &(adapter->dev->dev);
157}
158
159static struct pvscsi_ctx *
160pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
161{
162 struct pvscsi_ctx *ctx, *end;
163
164 end = &adapter->cmd_map[adapter->req_depth];
165 for (ctx = adapter->cmd_map; ctx < end; ctx++)
166 if (ctx->cmd == cmd)
167 return ctx;
168
169 return NULL;
170}
171
172static struct pvscsi_ctx *
173pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
174{
175 struct pvscsi_ctx *ctx;
176
177 if (list_empty(&adapter->cmd_pool))
178 return NULL;
179
180 ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list);
181 ctx->cmd = cmd;
182 list_del(&ctx->list);
183
184 return ctx;
185}
186
187static void pvscsi_release_context(struct pvscsi_adapter *adapter,
188 struct pvscsi_ctx *ctx)
189{
190 ctx->cmd = NULL;
191 ctx->abort_cmp = NULL;
192 list_add(&ctx->list, &adapter->cmd_pool);
193}
194
195/*
196 * Map a pvscsi_ctx struct to a context ID field value; we map to a simple
197 * non-zero integer. ctx always points to an entry in cmd_map array, hence
198 * the return value is always >=1.
199 */
200static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter,
201 const struct pvscsi_ctx *ctx)
202{
203 return ctx - adapter->cmd_map + 1;
204}
205
206static struct pvscsi_ctx *
207pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
208{
209 return &adapter->cmd_map[context - 1];
210}
211
212static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
213 u32 offset, u32 val)
214{
215 writel(val, adapter->mmioBase + offset);
216}
217
218static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
219{
220 return readl(adapter->mmioBase + offset);
221}
222
223static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
224{
225 return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
226}
227
228static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter,
229 u32 val)
230{
231 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val);
232}
233
234static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter)
235{
236 u32 intr_bits;
237
238 intr_bits = PVSCSI_INTR_CMPL_MASK;
239 if (adapter->use_msg)
240 intr_bits |= PVSCSI_INTR_MSG_MASK;
241
242 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits);
243}
244
245static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
246{
247 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
248}
249
250static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter,
251 u32 cmd, const void *desc, size_t len)
252{
253 const u32 *ptr = desc;
254 size_t i;
255
256 len /= sizeof(*ptr);
257 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd);
258 for (i = 0; i < len; i++)
259 pvscsi_reg_write(adapter,
260 PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
261}
262
263static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter,
264 const struct pvscsi_ctx *ctx)
265{
266 struct PVSCSICmdDescAbortCmd cmd = { 0 };
267
268 cmd.target = ctx->cmd->device->id;
269 cmd.context = pvscsi_map_context(adapter, ctx);
270
271 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd));
272}
273
274static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter)
275{
276 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
277}
278
279static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter)
280{
281 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0);
282}
283
284static int scsi_is_rw(unsigned char op)
285{
286 return op == READ_6 || op == WRITE_6 ||
287 op == READ_10 || op == WRITE_10 ||
288 op == READ_12 || op == WRITE_12 ||
289 op == READ_16 || op == WRITE_16;
290}
291
292static void pvscsi_kick_io(const struct pvscsi_adapter *adapter,
293 unsigned char op)
294{
295 if (scsi_is_rw(op)) {
296 struct PVSCSIRingsState *s = adapter->rings_state;
297
298 if (!adapter->use_req_threshold ||
299 s->reqProdIdx - s->reqConsIdx >= s->reqCallThreshold)
300 pvscsi_kick_rw_io(adapter);
301 } else {
302 pvscsi_process_request_ring(adapter);
303 }
304}
305
306static void ll_adapter_reset(const struct pvscsi_adapter *adapter)
307{
308 dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter);
309
310 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0);
311}
312
313static void ll_bus_reset(const struct pvscsi_adapter *adapter)
314{
315 dev_dbg(pvscsi_dev(adapter), "Resetting bus on %p\n", adapter);
316
317 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0);
318}
319
320static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target)
321{
322 struct PVSCSICmdDescResetDevice cmd = { 0 };
323
324 dev_dbg(pvscsi_dev(adapter), "Resetting device: target=%u\n", target);
325
326 cmd.target = target;
327
328 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE,
329 &cmd, sizeof(cmd));
330}
331
332static void pvscsi_create_sg(struct pvscsi_ctx *ctx,
333 struct scatterlist *sg, unsigned count)
334{
335 unsigned i;
336 struct PVSCSISGElement *sge;
337
338 BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT);
339
340 sge = &ctx->sgl->sge[0];
341 for (i = 0; i < count; i++, sg++) {
342 sge[i].addr = sg_dma_address(sg);
343 sge[i].length = sg_dma_len(sg);
344 sge[i].flags = 0;
345 }
346}
347
348/*
349 * Map all data buffers for a command into PCI space and
350 * setup the scatter/gather list if needed.
351 */
352static int pvscsi_map_buffers(struct pvscsi_adapter *adapter,
353 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd,
354 struct PVSCSIRingReqDesc *e)
355{
356 unsigned count;
357 unsigned bufflen = scsi_bufflen(cmd);
358 struct scatterlist *sg;
359
360 e->dataLen = bufflen;
361 e->dataAddr = 0;
362 if (bufflen == 0)
363 return 0;
364
365 sg = scsi_sglist(cmd);
366 count = scsi_sg_count(cmd);
367 if (count != 0) {
368 int segs = scsi_dma_map(cmd);
369
370 if (segs == -ENOMEM) {
371 scmd_printk(KERN_ERR, cmd,
372 "vmw_pvscsi: Failed to map cmd sglist for DMA.\n");
373 return -ENOMEM;
374 } else if (segs > 1) {
375 pvscsi_create_sg(ctx, sg, segs);
376
377 e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
378 ctx->sglPA = pci_map_single(adapter->dev, ctx->sgl,
379 SGL_SIZE, PCI_DMA_TODEVICE);
380 if (pci_dma_mapping_error(adapter->dev, ctx->sglPA)) {
381 scmd_printk(KERN_ERR, cmd,
382 "vmw_pvscsi: Failed to map ctx sglist for DMA.\n");
383 scsi_dma_unmap(cmd);
384 ctx->sglPA = 0;
385 return -ENOMEM;
386 }
387 e->dataAddr = ctx->sglPA;
388 } else
389 e->dataAddr = sg_dma_address(sg);
390 } else {
391 /*
392 * In case there is no S/G list, scsi_sglist points
393 * directly to the buffer.
394 */
395 ctx->dataPA = pci_map_single(adapter->dev, sg, bufflen,
396 cmd->sc_data_direction);
397 if (pci_dma_mapping_error(adapter->dev, ctx->dataPA)) {
398 scmd_printk(KERN_ERR, cmd,
399 "vmw_pvscsi: Failed to map direct data buffer for DMA.\n");
400 return -ENOMEM;
401 }
402 e->dataAddr = ctx->dataPA;
403 }
404
405 return 0;
406}
407
408static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
409 struct pvscsi_ctx *ctx)
410{
411 struct scsi_cmnd *cmd;
412 unsigned bufflen;
413
414 cmd = ctx->cmd;
415 bufflen = scsi_bufflen(cmd);
416
417 if (bufflen != 0) {
418 unsigned count = scsi_sg_count(cmd);
419
420 if (count != 0) {
421 scsi_dma_unmap(cmd);
422 if (ctx->sglPA) {
423 pci_unmap_single(adapter->dev, ctx->sglPA,
424 SGL_SIZE, PCI_DMA_TODEVICE);
425 ctx->sglPA = 0;
426 }
427 } else
428 pci_unmap_single(adapter->dev, ctx->dataPA, bufflen,
429 cmd->sc_data_direction);
430 }
431 if (cmd->sense_buffer)
432 pci_unmap_single(adapter->dev, ctx->sensePA,
433 SCSI_SENSE_BUFFERSIZE, PCI_DMA_FROMDEVICE);
434}
435
436static int pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
437{
438 adapter->rings_state = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
439 &adapter->ringStatePA);
440 if (!adapter->rings_state)
441 return -ENOMEM;
442
443 adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
444 pvscsi_ring_pages);
445 adapter->req_depth = adapter->req_pages
446 * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
447 adapter->req_ring = pci_alloc_consistent(adapter->dev,
448 adapter->req_pages * PAGE_SIZE,
449 &adapter->reqRingPA);
450 if (!adapter->req_ring)
451 return -ENOMEM;
452
453 adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
454 pvscsi_ring_pages);
455 adapter->cmp_ring = pci_alloc_consistent(adapter->dev,
456 adapter->cmp_pages * PAGE_SIZE,
457 &adapter->cmpRingPA);
458 if (!adapter->cmp_ring)
459 return -ENOMEM;
460
461 BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
462 BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
463 BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
464
465 if (!adapter->use_msg)
466 return 0;
467
468 adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
469 pvscsi_msg_ring_pages);
470 adapter->msg_ring = pci_alloc_consistent(adapter->dev,
471 adapter->msg_pages * PAGE_SIZE,
472 &adapter->msgRingPA);
473 if (!adapter->msg_ring)
474 return -ENOMEM;
475 BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
476
477 return 0;
478}
479
480static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
481{
482 struct PVSCSICmdDescSetupRings cmd = { 0 };
483 dma_addr_t base;
484 unsigned i;
485
486 cmd.ringsStatePPN = adapter->ringStatePA >> PAGE_SHIFT;
487 cmd.reqRingNumPages = adapter->req_pages;
488 cmd.cmpRingNumPages = adapter->cmp_pages;
489
490 base = adapter->reqRingPA;
491 for (i = 0; i < adapter->req_pages; i++) {
492 cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
493 base += PAGE_SIZE;
494 }
495
496 base = adapter->cmpRingPA;
497 for (i = 0; i < adapter->cmp_pages; i++) {
498 cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
499 base += PAGE_SIZE;
500 }
501
502 memset(adapter->rings_state, 0, PAGE_SIZE);
503 memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
504 memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
505
506 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
507 &cmd, sizeof(cmd));
508
509 if (adapter->use_msg) {
510 struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
511
512 cmd_msg.numPages = adapter->msg_pages;
513
514 base = adapter->msgRingPA;
515 for (i = 0; i < adapter->msg_pages; i++) {
516 cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
517 base += PAGE_SIZE;
518 }
519 memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
520
521 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
522 &cmd_msg, sizeof(cmd_msg));
523 }
524}
525
526static int pvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
527{
528 if (!sdev->tagged_supported)
529 qdepth = 1;
530 return scsi_change_queue_depth(sdev, qdepth);
531}
532
533/*
534 * Pull a completion descriptor off and pass the completion back
535 * to the SCSI mid layer.
536 */
537static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
538 const struct PVSCSIRingCmpDesc *e)
539{
540 struct pvscsi_ctx *ctx;
541 struct scsi_cmnd *cmd;
542 struct completion *abort_cmp;
543 u32 btstat = e->hostStatus;
544 u32 sdstat = e->scsiStatus;
545
546 ctx = pvscsi_get_context(adapter, e->context);
547 cmd = ctx->cmd;
548 abort_cmp = ctx->abort_cmp;
549 pvscsi_unmap_buffers(adapter, ctx);
550 pvscsi_release_context(adapter, ctx);
551 if (abort_cmp) {
552 /*
553 * The command was requested to be aborted. Just signal that
554 * the request completed and swallow the actual cmd completion
555 * here. The abort handler will post a completion for this
556 * command indicating that it got successfully aborted.
557 */
558 complete(abort_cmp);
559 return;
560 }
561
562 cmd->result = 0;
563 if (sdstat != SAM_STAT_GOOD &&
564 (btstat == BTSTAT_SUCCESS ||
565 btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
566 btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
567 cmd->result = (DID_OK << 16) | sdstat;
568 if (sdstat == SAM_STAT_CHECK_CONDITION && cmd->sense_buffer)
569 cmd->result |= (DRIVER_SENSE << 24);
570 } else
571 switch (btstat) {
572 case BTSTAT_SUCCESS:
573 case BTSTAT_LINKED_COMMAND_COMPLETED:
574 case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
575 /* If everything went fine, let's move on.. */
576 cmd->result = (DID_OK << 16);
577 break;
578
579 case BTSTAT_DATARUN:
580 case BTSTAT_DATA_UNDERRUN:
581 /* Report residual data in underruns */
582 scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
583 cmd->result = (DID_ERROR << 16);
584 break;
585
586 case BTSTAT_SELTIMEO:
587 /* Our emulation returns this for non-connected devs */
588 cmd->result = (DID_BAD_TARGET << 16);
589 break;
590
591 case BTSTAT_LUNMISMATCH:
592 case BTSTAT_TAGREJECT:
593 case BTSTAT_BADMSG:
594 cmd->result = (DRIVER_INVALID << 24);
595 /* fall through */
596
597 case BTSTAT_HAHARDWARE:
598 case BTSTAT_INVPHASE:
599 case BTSTAT_HATIMEOUT:
600 case BTSTAT_NORESPONSE:
601 case BTSTAT_DISCONNECT:
602 case BTSTAT_HASOFTWARE:
603 case BTSTAT_BUSFREE:
604 case BTSTAT_SENSFAILED:
605 cmd->result |= (DID_ERROR << 16);
606 break;
607
608 case BTSTAT_SENTRST:
609 case BTSTAT_RECVRST:
610 case BTSTAT_BUSRESET:
611 cmd->result = (DID_RESET << 16);
612 break;
613
614 case BTSTAT_ABORTQUEUE:
615 cmd->result = (DID_ABORT << 16);
616 break;
617
618 case BTSTAT_SCSIPARITY:
619 cmd->result = (DID_PARITY << 16);
620 break;
621
622 default:
623 cmd->result = (DID_ERROR << 16);
624 scmd_printk(KERN_DEBUG, cmd,
625 "Unknown completion status: 0x%x\n",
626 btstat);
627 }
628
629 dev_dbg(&cmd->device->sdev_gendev,
630 "cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
631 cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
632
633 cmd->scsi_done(cmd);
634}
635
636/*
637 * barrier usage : Since the PVSCSI device is emulated, there could be cases
638 * where we may want to serialize some accesses between the driver and the
639 * emulation layer. We use compiler barriers instead of the more expensive
640 * memory barriers because PVSCSI is only supported on X86 which has strong
641 * memory access ordering.
642 */
643static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
644{
645 struct PVSCSIRingsState *s = adapter->rings_state;
646 struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
647 u32 cmp_entries = s->cmpNumEntriesLog2;
648
649 while (s->cmpConsIdx != s->cmpProdIdx) {
650 struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
651 MASK(cmp_entries));
652 /*
653 * This barrier() ensures that *e is not dereferenced while
654 * the device emulation still writes data into the slot.
655 * Since the device emulation advances s->cmpProdIdx only after
656 * updating the slot we want to check it first.
657 */
658 barrier();
659 pvscsi_complete_request(adapter, e);
660 /*
661 * This barrier() ensures that compiler doesn't reorder write
662 * to s->cmpConsIdx before the read of (*e) inside
663 * pvscsi_complete_request. Otherwise, device emulation may
664 * overwrite *e before we had a chance to read it.
665 */
666 barrier();
667 s->cmpConsIdx++;
668 }
669}
670
671/*
672 * Translate a Linux SCSI request into a request ring entry.
673 */
674static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
675 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
676{
677 struct PVSCSIRingsState *s;
678 struct PVSCSIRingReqDesc *e;
679 struct scsi_device *sdev;
680 u32 req_entries;
681
682 s = adapter->rings_state;
683 sdev = cmd->device;
684 req_entries = s->reqNumEntriesLog2;
685
686 /*
687 * If this condition holds, we might have room on the request ring, but
688 * we might not have room on the completion ring for the response.
689 * However, we have already ruled out this possibility - we would not
690 * have successfully allocated a context if it were true, since we only
691 * have one context per request entry. Check for it anyway, since it
692 * would be a serious bug.
693 */
694 if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
695 scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
696 "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
697 s->reqProdIdx, s->cmpConsIdx);
698 return -1;
699 }
700
701 e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
702
703 e->bus = sdev->channel;
704 e->target = sdev->id;
705 memset(e->lun, 0, sizeof(e->lun));
706 e->lun[1] = sdev->lun;
707
708 if (cmd->sense_buffer) {
709 ctx->sensePA = pci_map_single(adapter->dev, cmd->sense_buffer,
710 SCSI_SENSE_BUFFERSIZE,
711 PCI_DMA_FROMDEVICE);
712 if (pci_dma_mapping_error(adapter->dev, ctx->sensePA)) {
713 scmd_printk(KERN_ERR, cmd,
714 "vmw_pvscsi: Failed to map sense buffer for DMA.\n");
715 ctx->sensePA = 0;
716 return -ENOMEM;
717 }
718 e->senseAddr = ctx->sensePA;
719 e->senseLen = SCSI_SENSE_BUFFERSIZE;
720 } else {
721 e->senseLen = 0;
722 e->senseAddr = 0;
723 }
724 e->cdbLen = cmd->cmd_len;
725 e->vcpuHint = smp_processor_id();
726 memcpy(e->cdb, cmd->cmnd, e->cdbLen);
727
728 e->tag = SIMPLE_QUEUE_TAG;
729
730 if (cmd->sc_data_direction == DMA_FROM_DEVICE)
731 e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
732 else if (cmd->sc_data_direction == DMA_TO_DEVICE)
733 e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
734 else if (cmd->sc_data_direction == DMA_NONE)
735 e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
736 else
737 e->flags = 0;
738
739 if (pvscsi_map_buffers(adapter, ctx, cmd, e) != 0) {
740 if (cmd->sense_buffer) {
741 pci_unmap_single(adapter->dev, ctx->sensePA,
742 SCSI_SENSE_BUFFERSIZE,
743 PCI_DMA_FROMDEVICE);
744 ctx->sensePA = 0;
745 }
746 return -ENOMEM;
747 }
748
749 e->context = pvscsi_map_context(adapter, ctx);
750
751 barrier();
752
753 s->reqProdIdx++;
754
755 return 0;
756}
757
758static int pvscsi_queue_lck(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
759{
760 struct Scsi_Host *host = cmd->device->host;
761 struct pvscsi_adapter *adapter = shost_priv(host);
762 struct pvscsi_ctx *ctx;
763 unsigned long flags;
764
765 spin_lock_irqsave(&adapter->hw_lock, flags);
766
767 ctx = pvscsi_acquire_context(adapter, cmd);
768 if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
769 if (ctx)
770 pvscsi_release_context(adapter, ctx);
771 spin_unlock_irqrestore(&adapter->hw_lock, flags);
772 return SCSI_MLQUEUE_HOST_BUSY;
773 }
774
775 cmd->scsi_done = done;
776
777 dev_dbg(&cmd->device->sdev_gendev,
778 "queued cmd %p, ctx %p, op=%x\n", cmd, ctx, cmd->cmnd[0]);
779
780 spin_unlock_irqrestore(&adapter->hw_lock, flags);
781
782 pvscsi_kick_io(adapter, cmd->cmnd[0]);
783
784 return 0;
785}
786
787static DEF_SCSI_QCMD(pvscsi_queue)
788
789static int pvscsi_abort(struct scsi_cmnd *cmd)
790{
791 struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
792 struct pvscsi_ctx *ctx;
793 unsigned long flags;
794 int result = SUCCESS;
795 DECLARE_COMPLETION_ONSTACK(abort_cmp);
796
797 scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
798 adapter->host->host_no, cmd);
799
800 spin_lock_irqsave(&adapter->hw_lock, flags);
801
802 /*
803 * Poll the completion ring first - we might be trying to abort
804 * a command that is waiting to be dispatched in the completion ring.
805 */
806 pvscsi_process_completion_ring(adapter);
807
808 /*
809 * If there is no context for the command, it either already succeeded
810 * or else was never properly issued. Not our problem.
811 */
812 ctx = pvscsi_find_context(adapter, cmd);
813 if (!ctx) {
814 scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
815 goto out;
816 }
817
818 /*
819 * Mark that the command has been requested to be aborted and issue
820 * the abort.
821 */
822 ctx->abort_cmp = &abort_cmp;
823
824 pvscsi_abort_cmd(adapter, ctx);
825 spin_unlock_irqrestore(&adapter->hw_lock, flags);
826 /* Wait for 2 secs for the completion. */
827 wait_for_completion_timeout(&abort_cmp, msecs_to_jiffies(2000));
828 spin_lock_irqsave(&adapter->hw_lock, flags);
829
830 if (!completion_done(&abort_cmp)) {
831 /*
832 * Failed to abort the command, unmark the fact that it
833 * was requested to be aborted.
834 */
835 ctx->abort_cmp = NULL;
836 result = FAILED;
837 scmd_printk(KERN_DEBUG, cmd,
838 "Failed to get completion for aborted cmd %p\n",
839 cmd);
840 goto out;
841 }
842
843 /*
844 * Successfully aborted the command.
845 */
846 cmd->result = (DID_ABORT << 16);
847 cmd->scsi_done(cmd);
848
849out:
850 spin_unlock_irqrestore(&adapter->hw_lock, flags);
851 return result;
852}
853
854/*
855 * Abort all outstanding requests. This is only safe to use if the completion
856 * ring will never be walked again or the device has been reset, because it
857 * destroys the 1-1 mapping between context field passed to emulation and our
858 * request structure.
859 */
860static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
861{
862 unsigned i;
863
864 for (i = 0; i < adapter->req_depth; i++) {
865 struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
866 struct scsi_cmnd *cmd = ctx->cmd;
867 if (cmd) {
868 scmd_printk(KERN_ERR, cmd,
869 "Forced reset on cmd %p\n", cmd);
870 pvscsi_unmap_buffers(adapter, ctx);
871 pvscsi_release_context(adapter, ctx);
872 cmd->result = (DID_RESET << 16);
873 cmd->scsi_done(cmd);
874 }
875 }
876}
877
878static int pvscsi_host_reset(struct scsi_cmnd *cmd)
879{
880 struct Scsi_Host *host = cmd->device->host;
881 struct pvscsi_adapter *adapter = shost_priv(host);
882 unsigned long flags;
883 bool use_msg;
884
885 scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
886
887 spin_lock_irqsave(&adapter->hw_lock, flags);
888
889 use_msg = adapter->use_msg;
890
891 if (use_msg) {
892 adapter->use_msg = 0;
893 spin_unlock_irqrestore(&adapter->hw_lock, flags);
894
895 /*
896 * Now that we know that the ISR won't add more work on the
897 * workqueue we can safely flush any outstanding work.
898 */
899 flush_workqueue(adapter->workqueue);
900 spin_lock_irqsave(&adapter->hw_lock, flags);
901 }
902
903 /*
904 * We're going to tear down the entire ring structure and set it back
905 * up, so stalling new requests until all completions are flushed and
906 * the rings are back in place.
907 */
908
909 pvscsi_process_request_ring(adapter);
910
911 ll_adapter_reset(adapter);
912
913 /*
914 * Now process any completions. Note we do this AFTER adapter reset,
915 * which is strange, but stops races where completions get posted
916 * between processing the ring and issuing the reset. The backend will
917 * not touch the ring memory after reset, so the immediately pre-reset
918 * completion ring state is still valid.
919 */
920 pvscsi_process_completion_ring(adapter);
921
922 pvscsi_reset_all(adapter);
923 adapter->use_msg = use_msg;
924 pvscsi_setup_all_rings(adapter);
925 pvscsi_unmask_intr(adapter);
926
927 spin_unlock_irqrestore(&adapter->hw_lock, flags);
928
929 return SUCCESS;
930}
931
932static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
933{
934 struct Scsi_Host *host = cmd->device->host;
935 struct pvscsi_adapter *adapter = shost_priv(host);
936 unsigned long flags;
937
938 scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
939
940 /*
941 * We don't want to queue new requests for this bus after
942 * flushing all pending requests to emulation, since new
943 * requests could then sneak in during this bus reset phase,
944 * so take the lock now.
945 */
946 spin_lock_irqsave(&adapter->hw_lock, flags);
947
948 pvscsi_process_request_ring(adapter);
949 ll_bus_reset(adapter);
950 pvscsi_process_completion_ring(adapter);
951
952 spin_unlock_irqrestore(&adapter->hw_lock, flags);
953
954 return SUCCESS;
955}
956
957static int pvscsi_device_reset(struct scsi_cmnd *cmd)
958{
959 struct Scsi_Host *host = cmd->device->host;
960 struct pvscsi_adapter *adapter = shost_priv(host);
961 unsigned long flags;
962
963 scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
964 host->host_no, cmd->device->id);
965
966 /*
967 * We don't want to queue new requests for this device after flushing
968 * all pending requests to emulation, since new requests could then
969 * sneak in during this device reset phase, so take the lock now.
970 */
971 spin_lock_irqsave(&adapter->hw_lock, flags);
972
973 pvscsi_process_request_ring(adapter);
974 ll_device_reset(adapter, cmd->device->id);
975 pvscsi_process_completion_ring(adapter);
976
977 spin_unlock_irqrestore(&adapter->hw_lock, flags);
978
979 return SUCCESS;
980}
981
982static struct scsi_host_template pvscsi_template;
983
984static const char *pvscsi_info(struct Scsi_Host *host)
985{
986 struct pvscsi_adapter *adapter = shost_priv(host);
987 static char buf[256];
988
989 sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
990 "%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
991 adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
992 pvscsi_template.cmd_per_lun);
993
994 return buf;
995}
996
997static struct scsi_host_template pvscsi_template = {
998 .module = THIS_MODULE,
999 .name = "VMware PVSCSI Host Adapter",
1000 .proc_name = "vmw_pvscsi",
1001 .info = pvscsi_info,
1002 .queuecommand = pvscsi_queue,
1003 .this_id = -1,
1004 .sg_tablesize = PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
1005 .dma_boundary = UINT_MAX,
1006 .max_sectors = 0xffff,
1007 .use_clustering = ENABLE_CLUSTERING,
1008 .change_queue_depth = pvscsi_change_queue_depth,
1009 .eh_abort_handler = pvscsi_abort,
1010 .eh_device_reset_handler = pvscsi_device_reset,
1011 .eh_bus_reset_handler = pvscsi_bus_reset,
1012 .eh_host_reset_handler = pvscsi_host_reset,
1013};
1014
1015static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
1016 const struct PVSCSIRingMsgDesc *e)
1017{
1018 struct PVSCSIRingsState *s = adapter->rings_state;
1019 struct Scsi_Host *host = adapter->host;
1020 struct scsi_device *sdev;
1021
1022 printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
1023 e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
1024
1025 BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
1026
1027 if (e->type == PVSCSI_MSG_DEV_ADDED) {
1028 struct PVSCSIMsgDescDevStatusChanged *desc;
1029 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1030
1031 printk(KERN_INFO
1032 "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
1033 desc->bus, desc->target, desc->lun[1]);
1034
1035 if (!scsi_host_get(host))
1036 return;
1037
1038 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1039 desc->lun[1]);
1040 if (sdev) {
1041 printk(KERN_INFO "vmw_pvscsi: device already exists\n");
1042 scsi_device_put(sdev);
1043 } else
1044 scsi_add_device(adapter->host, desc->bus,
1045 desc->target, desc->lun[1]);
1046
1047 scsi_host_put(host);
1048 } else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
1049 struct PVSCSIMsgDescDevStatusChanged *desc;
1050 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1051
1052 printk(KERN_INFO
1053 "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
1054 desc->bus, desc->target, desc->lun[1]);
1055
1056 if (!scsi_host_get(host))
1057 return;
1058
1059 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1060 desc->lun[1]);
1061 if (sdev) {
1062 scsi_remove_device(sdev);
1063 scsi_device_put(sdev);
1064 } else
1065 printk(KERN_INFO
1066 "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
1067 desc->bus, desc->target, desc->lun[1]);
1068
1069 scsi_host_put(host);
1070 }
1071}
1072
1073static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
1074{
1075 struct PVSCSIRingsState *s = adapter->rings_state;
1076
1077 return s->msgProdIdx != s->msgConsIdx;
1078}
1079
1080static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
1081{
1082 struct PVSCSIRingsState *s = adapter->rings_state;
1083 struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
1084 u32 msg_entries = s->msgNumEntriesLog2;
1085
1086 while (pvscsi_msg_pending(adapter)) {
1087 struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
1088 MASK(msg_entries));
1089
1090 barrier();
1091 pvscsi_process_msg(adapter, e);
1092 barrier();
1093 s->msgConsIdx++;
1094 }
1095}
1096
1097static void pvscsi_msg_workqueue_handler(struct work_struct *data)
1098{
1099 struct pvscsi_adapter *adapter;
1100
1101 adapter = container_of(data, struct pvscsi_adapter, work);
1102
1103 pvscsi_process_msg_ring(adapter);
1104}
1105
1106static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
1107{
1108 char name[32];
1109
1110 if (!pvscsi_use_msg)
1111 return 0;
1112
1113 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1114 PVSCSI_CMD_SETUP_MSG_RING);
1115
1116 if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
1117 return 0;
1118
1119 snprintf(name, sizeof(name),
1120 "vmw_pvscsi_wq_%u", adapter->host->host_no);
1121
1122 adapter->workqueue = create_singlethread_workqueue(name);
1123 if (!adapter->workqueue) {
1124 printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
1125 return 0;
1126 }
1127 INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
1128
1129 return 1;
1130}
1131
1132static bool pvscsi_setup_req_threshold(struct pvscsi_adapter *adapter,
1133 bool enable)
1134{
1135 u32 val;
1136
1137 if (!pvscsi_use_req_threshold)
1138 return false;
1139
1140 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1141 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD);
1142 val = pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS);
1143 if (val == -1) {
1144 printk(KERN_INFO "vmw_pvscsi: device does not support req_threshold\n");
1145 return false;
1146 } else {
1147 struct PVSCSICmdDescSetupReqCall cmd_msg = { 0 };
1148 cmd_msg.enable = enable;
1149 printk(KERN_INFO
1150 "vmw_pvscsi: %sabling reqCallThreshold\n",
1151 enable ? "en" : "dis");
1152 pvscsi_write_cmd_desc(adapter,
1153 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD,
1154 &cmd_msg, sizeof(cmd_msg));
1155 return pvscsi_reg_read(adapter,
1156 PVSCSI_REG_OFFSET_COMMAND_STATUS) != 0;
1157 }
1158}
1159
1160static irqreturn_t pvscsi_isr(int irq, void *devp)
1161{
1162 struct pvscsi_adapter *adapter = devp;
1163 int handled;
1164
1165 if (adapter->use_msi || adapter->use_msix)
1166 handled = true;
1167 else {
1168 u32 val = pvscsi_read_intr_status(adapter);
1169 handled = (val & PVSCSI_INTR_ALL_SUPPORTED) != 0;
1170 if (handled)
1171 pvscsi_write_intr_status(devp, val);
1172 }
1173
1174 if (handled) {
1175 unsigned long flags;
1176
1177 spin_lock_irqsave(&adapter->hw_lock, flags);
1178
1179 pvscsi_process_completion_ring(adapter);
1180 if (adapter->use_msg && pvscsi_msg_pending(adapter))
1181 queue_work(adapter->workqueue, &adapter->work);
1182
1183 spin_unlock_irqrestore(&adapter->hw_lock, flags);
1184 }
1185
1186 return IRQ_RETVAL(handled);
1187}
1188
1189static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
1190{
1191 struct pvscsi_ctx *ctx = adapter->cmd_map;
1192 unsigned i;
1193
1194 for (i = 0; i < adapter->req_depth; ++i, ++ctx)
1195 free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
1196}
1197
1198static int pvscsi_setup_msix(const struct pvscsi_adapter *adapter,
1199 unsigned int *irq)
1200{
1201 struct msix_entry entry = { 0, PVSCSI_VECTOR_COMPLETION };
1202 int ret;
1203
1204 ret = pci_enable_msix_exact(adapter->dev, &entry, 1);
1205 if (ret)
1206 return ret;
1207
1208 *irq = entry.vector;
1209
1210 return 0;
1211}
1212
1213static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
1214{
1215 if (adapter->irq) {
1216 free_irq(adapter->irq, adapter);
1217 adapter->irq = 0;
1218 }
1219 if (adapter->use_msi) {
1220 pci_disable_msi(adapter->dev);
1221 adapter->use_msi = 0;
1222 } else if (adapter->use_msix) {
1223 pci_disable_msix(adapter->dev);
1224 adapter->use_msix = 0;
1225 }
1226}
1227
1228static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
1229{
1230 pvscsi_shutdown_intr(adapter);
1231
1232 if (adapter->workqueue)
1233 destroy_workqueue(adapter->workqueue);
1234
1235 if (adapter->mmioBase)
1236 pci_iounmap(adapter->dev, adapter->mmioBase);
1237
1238 pci_release_regions(adapter->dev);
1239
1240 if (adapter->cmd_map) {
1241 pvscsi_free_sgls(adapter);
1242 kfree(adapter->cmd_map);
1243 }
1244
1245 if (adapter->rings_state)
1246 pci_free_consistent(adapter->dev, PAGE_SIZE,
1247 adapter->rings_state, adapter->ringStatePA);
1248
1249 if (adapter->req_ring)
1250 pci_free_consistent(adapter->dev,
1251 adapter->req_pages * PAGE_SIZE,
1252 adapter->req_ring, adapter->reqRingPA);
1253
1254 if (adapter->cmp_ring)
1255 pci_free_consistent(adapter->dev,
1256 adapter->cmp_pages * PAGE_SIZE,
1257 adapter->cmp_ring, adapter->cmpRingPA);
1258
1259 if (adapter->msg_ring)
1260 pci_free_consistent(adapter->dev,
1261 adapter->msg_pages * PAGE_SIZE,
1262 adapter->msg_ring, adapter->msgRingPA);
1263}
1264
1265/*
1266 * Allocate scatter gather lists.
1267 *
1268 * These are statically allocated. Trying to be clever was not worth it.
1269 *
1270 * Dynamic allocation can fail, and we can't go deep into the memory
1271 * allocator, since we're a SCSI driver, and trying too hard to allocate
1272 * memory might generate disk I/O. We also don't want to fail disk I/O
1273 * in that case because we can't get an allocation - the I/O could be
1274 * trying to swap out data to free memory. Since that is pathological,
1275 * just use a statically allocated scatter list.
1276 *
1277 */
1278static int pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
1279{
1280 struct pvscsi_ctx *ctx;
1281 int i;
1282
1283 ctx = adapter->cmd_map;
1284 BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
1285
1286 for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
1287 ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
1288 get_order(SGL_SIZE));
1289 ctx->sglPA = 0;
1290 BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
1291 if (!ctx->sgl) {
1292 for (; i >= 0; --i, --ctx) {
1293 free_pages((unsigned long)ctx->sgl,
1294 get_order(SGL_SIZE));
1295 ctx->sgl = NULL;
1296 }
1297 return -ENOMEM;
1298 }
1299 }
1300
1301 return 0;
1302}
1303
1304/*
1305 * Query the device, fetch the config info and return the
1306 * maximum number of targets on the adapter. In case of
1307 * failure due to any reason return default i.e. 16.
1308 */
1309static u32 pvscsi_get_max_targets(struct pvscsi_adapter *adapter)
1310{
1311 struct PVSCSICmdDescConfigCmd cmd;
1312 struct PVSCSIConfigPageHeader *header;
1313 struct device *dev;
1314 dma_addr_t configPagePA;
1315 void *config_page;
1316 u32 numPhys = 16;
1317
1318 dev = pvscsi_dev(adapter);
1319 config_page = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
1320 &configPagePA);
1321 if (!config_page) {
1322 dev_warn(dev, "vmw_pvscsi: failed to allocate memory for config page\n");
1323 goto exit;
1324 }
1325 BUG_ON(configPagePA & ~PAGE_MASK);
1326
1327 /* Fetch config info from the device. */
1328 cmd.configPageAddress = ((u64)PVSCSI_CONFIG_CONTROLLER_ADDRESS) << 32;
1329 cmd.configPageNum = PVSCSI_CONFIG_PAGE_CONTROLLER;
1330 cmd.cmpAddr = configPagePA;
1331 cmd._pad = 0;
1332
1333 /*
1334 * Mark the completion page header with error values. If the device
1335 * completes the command successfully, it sets the status values to
1336 * indicate success.
1337 */
1338 header = config_page;
1339 memset(header, 0, sizeof *header);
1340 header->hostStatus = BTSTAT_INVPARAM;
1341 header->scsiStatus = SDSTAT_CHECK;
1342
1343 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_CONFIG, &cmd, sizeof cmd);
1344
1345 if (header->hostStatus == BTSTAT_SUCCESS &&
1346 header->scsiStatus == SDSTAT_GOOD) {
1347 struct PVSCSIConfigPageController *config;
1348
1349 config = config_page;
1350 numPhys = config->numPhys;
1351 } else
1352 dev_warn(dev, "vmw_pvscsi: PVSCSI_CMD_CONFIG failed. hostStatus = 0x%x, scsiStatus = 0x%x\n",
1353 header->hostStatus, header->scsiStatus);
1354 pci_free_consistent(adapter->dev, PAGE_SIZE, config_page, configPagePA);
1355exit:
1356 return numPhys;
1357}
1358
1359static int pvscsi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1360{
1361 struct pvscsi_adapter *adapter;
1362 struct pvscsi_adapter adapter_temp;
1363 struct Scsi_Host *host = NULL;
1364 unsigned int i;
1365 unsigned long flags = 0;
1366 int error;
1367 u32 max_id;
1368
1369 error = -ENODEV;
1370
1371 if (pci_enable_device(pdev))
1372 return error;
1373
1374 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0 &&
1375 pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1376 printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
1377 } else if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) == 0 &&
1378 pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) == 0) {
1379 printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
1380 } else {
1381 printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
1382 goto out_disable_device;
1383 }
1384
1385 /*
1386 * Let's use a temp pvscsi_adapter struct until we find the number of
1387 * targets on the adapter, after that we will switch to the real
1388 * allocated struct.
1389 */
1390 adapter = &adapter_temp;
1391 memset(adapter, 0, sizeof(*adapter));
1392 adapter->dev = pdev;
1393 adapter->rev = pdev->revision;
1394
1395 if (pci_request_regions(pdev, "vmw_pvscsi")) {
1396 printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
1397 goto out_disable_device;
1398 }
1399
1400 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1401 if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
1402 continue;
1403
1404 if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
1405 continue;
1406
1407 break;
1408 }
1409
1410 if (i == DEVICE_COUNT_RESOURCE) {
1411 printk(KERN_ERR
1412 "vmw_pvscsi: adapter has no suitable MMIO region\n");
1413 goto out_release_resources_and_disable;
1414 }
1415
1416 adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
1417
1418 if (!adapter->mmioBase) {
1419 printk(KERN_ERR
1420 "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
1421 i, PVSCSI_MEM_SPACE_SIZE);
1422 goto out_release_resources_and_disable;
1423 }
1424
1425 pci_set_master(pdev);
1426
1427 /*
1428 * Ask the device for max number of targets before deciding the
1429 * default pvscsi_ring_pages value.
1430 */
1431 max_id = pvscsi_get_max_targets(adapter);
1432 printk(KERN_INFO "vmw_pvscsi: max_id: %u\n", max_id);
1433
1434 if (pvscsi_ring_pages == 0)
1435 /*
1436 * Set the right default value. Up to 16 it is 8, above it is
1437 * max.
1438 */
1439 pvscsi_ring_pages = (max_id > 16) ?
1440 PVSCSI_SETUP_RINGS_MAX_NUM_PAGES :
1441 PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
1442 printk(KERN_INFO
1443 "vmw_pvscsi: setting ring_pages to %d\n",
1444 pvscsi_ring_pages);
1445
1446 pvscsi_template.can_queue =
1447 min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
1448 PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
1449 pvscsi_template.cmd_per_lun =
1450 min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
1451 host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
1452 if (!host) {
1453 printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
1454 goto out_release_resources_and_disable;
1455 }
1456
1457 /*
1458 * Let's use the real pvscsi_adapter struct here onwards.
1459 */
1460 adapter = shost_priv(host);
1461 memset(adapter, 0, sizeof(*adapter));
1462 adapter->dev = pdev;
1463 adapter->host = host;
1464 /*
1465 * Copy back what we already have to the allocated adapter struct.
1466 */
1467 adapter->rev = adapter_temp.rev;
1468 adapter->mmioBase = adapter_temp.mmioBase;
1469
1470 spin_lock_init(&adapter->hw_lock);
1471 host->max_channel = 0;
1472 host->max_lun = 1;
1473 host->max_cmd_len = 16;
1474 host->max_id = max_id;
1475
1476 pci_set_drvdata(pdev, host);
1477
1478 ll_adapter_reset(adapter);
1479
1480 adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
1481
1482 error = pvscsi_allocate_rings(adapter);
1483 if (error) {
1484 printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
1485 goto out_release_resources;
1486 }
1487
1488 /*
1489 * From this point on we should reset the adapter if anything goes
1490 * wrong.
1491 */
1492 pvscsi_setup_all_rings(adapter);
1493
1494 adapter->cmd_map = kcalloc(adapter->req_depth,
1495 sizeof(struct pvscsi_ctx), GFP_KERNEL);
1496 if (!adapter->cmd_map) {
1497 printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
1498 error = -ENOMEM;
1499 goto out_reset_adapter;
1500 }
1501
1502 INIT_LIST_HEAD(&adapter->cmd_pool);
1503 for (i = 0; i < adapter->req_depth; i++) {
1504 struct pvscsi_ctx *ctx = adapter->cmd_map + i;
1505 list_add(&ctx->list, &adapter->cmd_pool);
1506 }
1507
1508 error = pvscsi_allocate_sg(adapter);
1509 if (error) {
1510 printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
1511 goto out_reset_adapter;
1512 }
1513
1514 if (!pvscsi_disable_msix &&
1515 pvscsi_setup_msix(adapter, &adapter->irq) == 0) {
1516 printk(KERN_INFO "vmw_pvscsi: using MSI-X\n");
1517 adapter->use_msix = 1;
1518 } else if (!pvscsi_disable_msi && pci_enable_msi(pdev) == 0) {
1519 printk(KERN_INFO "vmw_pvscsi: using MSI\n");
1520 adapter->use_msi = 1;
1521 adapter->irq = pdev->irq;
1522 } else {
1523 printk(KERN_INFO "vmw_pvscsi: using INTx\n");
1524 adapter->irq = pdev->irq;
1525 flags = IRQF_SHARED;
1526 }
1527
1528 adapter->use_req_threshold = pvscsi_setup_req_threshold(adapter, true);
1529 printk(KERN_DEBUG "vmw_pvscsi: driver-based request coalescing %sabled\n",
1530 adapter->use_req_threshold ? "en" : "dis");
1531
1532 error = request_irq(adapter->irq, pvscsi_isr, flags,
1533 "vmw_pvscsi", adapter);
1534 if (error) {
1535 printk(KERN_ERR
1536 "vmw_pvscsi: unable to request IRQ: %d\n", error);
1537 adapter->irq = 0;
1538 goto out_reset_adapter;
1539 }
1540
1541 error = scsi_add_host(host, &pdev->dev);
1542 if (error) {
1543 printk(KERN_ERR
1544 "vmw_pvscsi: scsi_add_host failed: %d\n", error);
1545 goto out_reset_adapter;
1546 }
1547
1548 dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
1549 adapter->rev, host->host_no);
1550
1551 pvscsi_unmask_intr(adapter);
1552
1553 scsi_scan_host(host);
1554
1555 return 0;
1556
1557out_reset_adapter:
1558 ll_adapter_reset(adapter);
1559out_release_resources:
1560 pvscsi_release_resources(adapter);
1561 scsi_host_put(host);
1562out_disable_device:
1563 pci_disable_device(pdev);
1564
1565 return error;
1566
1567out_release_resources_and_disable:
1568 pvscsi_release_resources(adapter);
1569 goto out_disable_device;
1570}
1571
1572static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
1573{
1574 pvscsi_mask_intr(adapter);
1575
1576 if (adapter->workqueue)
1577 flush_workqueue(adapter->workqueue);
1578
1579 pvscsi_shutdown_intr(adapter);
1580
1581 pvscsi_process_request_ring(adapter);
1582 pvscsi_process_completion_ring(adapter);
1583 ll_adapter_reset(adapter);
1584}
1585
1586static void pvscsi_shutdown(struct pci_dev *dev)
1587{
1588 struct Scsi_Host *host = pci_get_drvdata(dev);
1589 struct pvscsi_adapter *adapter = shost_priv(host);
1590
1591 __pvscsi_shutdown(adapter);
1592}
1593
1594static void pvscsi_remove(struct pci_dev *pdev)
1595{
1596 struct Scsi_Host *host = pci_get_drvdata(pdev);
1597 struct pvscsi_adapter *adapter = shost_priv(host);
1598
1599 scsi_remove_host(host);
1600
1601 __pvscsi_shutdown(adapter);
1602 pvscsi_release_resources(adapter);
1603
1604 scsi_host_put(host);
1605
1606 pci_disable_device(pdev);
1607}
1608
1609static struct pci_driver pvscsi_pci_driver = {
1610 .name = "vmw_pvscsi",
1611 .id_table = pvscsi_pci_tbl,
1612 .probe = pvscsi_probe,
1613 .remove = pvscsi_remove,
1614 .shutdown = pvscsi_shutdown,
1615};
1616
1617static int __init pvscsi_init(void)
1618{
1619 pr_info("%s - version %s\n",
1620 PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
1621 return pci_register_driver(&pvscsi_pci_driver);
1622}
1623
1624static void __exit pvscsi_exit(void)
1625{
1626 pci_unregister_driver(&pvscsi_pci_driver);
1627}
1628
1629module_init(pvscsi_init);
1630module_exit(pvscsi_exit);