xref: /openbmc/linux/drivers/scsi/vmw_pvscsi.c (revision 33ac9dba)
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 
41 MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC);
42 MODULE_AUTHOR("VMware, Inc.");
43 MODULE_LICENSE("GPL");
44 MODULE_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 
51 struct pvscsi_sg_list {
52 	struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT];
53 };
54 
55 struct 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 
69 struct 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 */
108 static int pvscsi_ring_pages;
109 static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
110 static int pvscsi_cmd_per_lun    = PVSCSI_DEFAULT_QUEUE_DEPTH;
111 static bool pvscsi_disable_msi;
112 static bool pvscsi_disable_msix;
113 static bool pvscsi_use_msg       = true;
114 static bool pvscsi_use_req_threshold = true;
115 
116 #define PVSCSI_RW (S_IRUSR | S_IWUSR)
117 
118 module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
119 MODULE_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 
125 module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
126 MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
127 		 __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
128 
129 module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
130 MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
131 		 __stringify(PVSCSI_DEFAULT_QUEUE_DEPTH) ")");
132 
133 module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
134 MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
135 
136 module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
137 MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
138 
139 module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
140 MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
141 
142 module_param_named(use_req_threshold, pvscsi_use_req_threshold,
143 		   bool, PVSCSI_RW);
144 MODULE_PARM_DESC(use_req_threshold, "Use driver-based request coalescing if configured - (default=1)");
145 
146 static const struct pci_device_id pvscsi_pci_tbl[] = {
147 	{ PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
148 	{ 0 }
149 };
150 
151 MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
152 
153 static struct device *
154 pvscsi_dev(const struct pvscsi_adapter *adapter)
155 {
156 	return &(adapter->dev->dev);
157 }
158 
159 static struct pvscsi_ctx *
160 pvscsi_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 
172 static struct pvscsi_ctx *
173 pvscsi_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 
187 static 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  */
200 static 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 
206 static struct pvscsi_ctx *
207 pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
208 {
209 	return &adapter->cmd_map[context - 1];
210 }
211 
212 static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
213 			     u32 offset, u32 val)
214 {
215 	writel(val, adapter->mmioBase + offset);
216 }
217 
218 static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
219 {
220 	return readl(adapter->mmioBase + offset);
221 }
222 
223 static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
224 {
225 	return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
226 }
227 
228 static 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 
234 static 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 
245 static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
246 {
247 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
248 }
249 
250 static 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 
263 static 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 
274 static 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 
279 static 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 
284 static 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 
292 static 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 
306 static 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 
313 static 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 
320 static 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 
332 static 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  */
352 static void 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;
364 
365 	sg = scsi_sglist(cmd);
366 	count = scsi_sg_count(cmd);
367 	if (count != 0) {
368 		int segs = scsi_dma_map(cmd);
369 		if (segs > 1) {
370 			pvscsi_create_sg(ctx, sg, segs);
371 
372 			e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
373 			ctx->sglPA = pci_map_single(adapter->dev, ctx->sgl,
374 						    SGL_SIZE, PCI_DMA_TODEVICE);
375 			e->dataAddr = ctx->sglPA;
376 		} else
377 			e->dataAddr = sg_dma_address(sg);
378 	} else {
379 		/*
380 		 * In case there is no S/G list, scsi_sglist points
381 		 * directly to the buffer.
382 		 */
383 		ctx->dataPA = pci_map_single(adapter->dev, sg, bufflen,
384 					     cmd->sc_data_direction);
385 		e->dataAddr = ctx->dataPA;
386 	}
387 }
388 
389 static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
390 				 struct pvscsi_ctx *ctx)
391 {
392 	struct scsi_cmnd *cmd;
393 	unsigned bufflen;
394 
395 	cmd = ctx->cmd;
396 	bufflen = scsi_bufflen(cmd);
397 
398 	if (bufflen != 0) {
399 		unsigned count = scsi_sg_count(cmd);
400 
401 		if (count != 0) {
402 			scsi_dma_unmap(cmd);
403 			if (ctx->sglPA) {
404 				pci_unmap_single(adapter->dev, ctx->sglPA,
405 						 SGL_SIZE, PCI_DMA_TODEVICE);
406 				ctx->sglPA = 0;
407 			}
408 		} else
409 			pci_unmap_single(adapter->dev, ctx->dataPA, bufflen,
410 					 cmd->sc_data_direction);
411 	}
412 	if (cmd->sense_buffer)
413 		pci_unmap_single(adapter->dev, ctx->sensePA,
414 				 SCSI_SENSE_BUFFERSIZE, PCI_DMA_FROMDEVICE);
415 }
416 
417 static int pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
418 {
419 	adapter->rings_state = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
420 						    &adapter->ringStatePA);
421 	if (!adapter->rings_state)
422 		return -ENOMEM;
423 
424 	adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
425 				 pvscsi_ring_pages);
426 	adapter->req_depth = adapter->req_pages
427 					* PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
428 	adapter->req_ring = pci_alloc_consistent(adapter->dev,
429 						 adapter->req_pages * PAGE_SIZE,
430 						 &adapter->reqRingPA);
431 	if (!adapter->req_ring)
432 		return -ENOMEM;
433 
434 	adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
435 				 pvscsi_ring_pages);
436 	adapter->cmp_ring = pci_alloc_consistent(adapter->dev,
437 						 adapter->cmp_pages * PAGE_SIZE,
438 						 &adapter->cmpRingPA);
439 	if (!adapter->cmp_ring)
440 		return -ENOMEM;
441 
442 	BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
443 	BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
444 	BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
445 
446 	if (!adapter->use_msg)
447 		return 0;
448 
449 	adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
450 				 pvscsi_msg_ring_pages);
451 	adapter->msg_ring = pci_alloc_consistent(adapter->dev,
452 						 adapter->msg_pages * PAGE_SIZE,
453 						 &adapter->msgRingPA);
454 	if (!adapter->msg_ring)
455 		return -ENOMEM;
456 	BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
457 
458 	return 0;
459 }
460 
461 static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
462 {
463 	struct PVSCSICmdDescSetupRings cmd = { 0 };
464 	dma_addr_t base;
465 	unsigned i;
466 
467 	cmd.ringsStatePPN   = adapter->ringStatePA >> PAGE_SHIFT;
468 	cmd.reqRingNumPages = adapter->req_pages;
469 	cmd.cmpRingNumPages = adapter->cmp_pages;
470 
471 	base = adapter->reqRingPA;
472 	for (i = 0; i < adapter->req_pages; i++) {
473 		cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
474 		base += PAGE_SIZE;
475 	}
476 
477 	base = adapter->cmpRingPA;
478 	for (i = 0; i < adapter->cmp_pages; i++) {
479 		cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
480 		base += PAGE_SIZE;
481 	}
482 
483 	memset(adapter->rings_state, 0, PAGE_SIZE);
484 	memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
485 	memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
486 
487 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
488 			      &cmd, sizeof(cmd));
489 
490 	if (adapter->use_msg) {
491 		struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
492 
493 		cmd_msg.numPages = adapter->msg_pages;
494 
495 		base = adapter->msgRingPA;
496 		for (i = 0; i < adapter->msg_pages; i++) {
497 			cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
498 			base += PAGE_SIZE;
499 		}
500 		memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
501 
502 		pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
503 				      &cmd_msg, sizeof(cmd_msg));
504 	}
505 }
506 
507 static int pvscsi_change_queue_depth(struct scsi_device *sdev,
508 				     int qdepth,
509 				     int reason)
510 {
511 	int max_depth;
512 	struct Scsi_Host *shost = sdev->host;
513 
514 	if (reason != SCSI_QDEPTH_DEFAULT)
515 		/*
516 		 * We support only changing default.
517 		 */
518 		return -EOPNOTSUPP;
519 
520 	max_depth = shost->can_queue;
521 	if (!sdev->tagged_supported)
522 		max_depth = 1;
523 	if (qdepth > max_depth)
524 		qdepth = max_depth;
525 	scsi_adjust_queue_depth(sdev, scsi_get_tag_type(sdev), qdepth);
526 
527 	if (sdev->inquiry_len > 7)
528 		sdev_printk(KERN_INFO, sdev,
529 			    "qdepth(%d), tagged(%d), simple(%d), ordered(%d), scsi_level(%d), cmd_que(%d)\n",
530 			    sdev->queue_depth, sdev->tagged_supported,
531 			    sdev->simple_tags, sdev->ordered_tags,
532 			    sdev->scsi_level, (sdev->inquiry[7] & 2) >> 1);
533 	return sdev->queue_depth;
534 }
535 
536 /*
537  * Pull a completion descriptor off and pass the completion back
538  * to the SCSI mid layer.
539  */
540 static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
541 				    const struct PVSCSIRingCmpDesc *e)
542 {
543 	struct pvscsi_ctx *ctx;
544 	struct scsi_cmnd *cmd;
545 	struct completion *abort_cmp;
546 	u32 btstat = e->hostStatus;
547 	u32 sdstat = e->scsiStatus;
548 
549 	ctx = pvscsi_get_context(adapter, e->context);
550 	cmd = ctx->cmd;
551 	abort_cmp = ctx->abort_cmp;
552 	pvscsi_unmap_buffers(adapter, ctx);
553 	pvscsi_release_context(adapter, ctx);
554 	if (abort_cmp) {
555 		/*
556 		 * The command was requested to be aborted. Just signal that
557 		 * the request completed and swallow the actual cmd completion
558 		 * here. The abort handler will post a completion for this
559 		 * command indicating that it got successfully aborted.
560 		 */
561 		complete(abort_cmp);
562 		return;
563 	}
564 
565 	cmd->result = 0;
566 	if (sdstat != SAM_STAT_GOOD &&
567 	    (btstat == BTSTAT_SUCCESS ||
568 	     btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
569 	     btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
570 		cmd->result = (DID_OK << 16) | sdstat;
571 		if (sdstat == SAM_STAT_CHECK_CONDITION && cmd->sense_buffer)
572 			cmd->result |= (DRIVER_SENSE << 24);
573 	} else
574 		switch (btstat) {
575 		case BTSTAT_SUCCESS:
576 		case BTSTAT_LINKED_COMMAND_COMPLETED:
577 		case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
578 			/* If everything went fine, let's move on..  */
579 			cmd->result = (DID_OK << 16);
580 			break;
581 
582 		case BTSTAT_DATARUN:
583 		case BTSTAT_DATA_UNDERRUN:
584 			/* Report residual data in underruns */
585 			scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
586 			cmd->result = (DID_ERROR << 16);
587 			break;
588 
589 		case BTSTAT_SELTIMEO:
590 			/* Our emulation returns this for non-connected devs */
591 			cmd->result = (DID_BAD_TARGET << 16);
592 			break;
593 
594 		case BTSTAT_LUNMISMATCH:
595 		case BTSTAT_TAGREJECT:
596 		case BTSTAT_BADMSG:
597 			cmd->result = (DRIVER_INVALID << 24);
598 			/* fall through */
599 
600 		case BTSTAT_HAHARDWARE:
601 		case BTSTAT_INVPHASE:
602 		case BTSTAT_HATIMEOUT:
603 		case BTSTAT_NORESPONSE:
604 		case BTSTAT_DISCONNECT:
605 		case BTSTAT_HASOFTWARE:
606 		case BTSTAT_BUSFREE:
607 		case BTSTAT_SENSFAILED:
608 			cmd->result |= (DID_ERROR << 16);
609 			break;
610 
611 		case BTSTAT_SENTRST:
612 		case BTSTAT_RECVRST:
613 		case BTSTAT_BUSRESET:
614 			cmd->result = (DID_RESET << 16);
615 			break;
616 
617 		case BTSTAT_ABORTQUEUE:
618 			cmd->result = (DID_ABORT << 16);
619 			break;
620 
621 		case BTSTAT_SCSIPARITY:
622 			cmd->result = (DID_PARITY << 16);
623 			break;
624 
625 		default:
626 			cmd->result = (DID_ERROR << 16);
627 			scmd_printk(KERN_DEBUG, cmd,
628 				    "Unknown completion status: 0x%x\n",
629 				    btstat);
630 	}
631 
632 	dev_dbg(&cmd->device->sdev_gendev,
633 		"cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
634 		cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
635 
636 	cmd->scsi_done(cmd);
637 }
638 
639 /*
640  * barrier usage : Since the PVSCSI device is emulated, there could be cases
641  * where we may want to serialize some accesses between the driver and the
642  * emulation layer. We use compiler barriers instead of the more expensive
643  * memory barriers because PVSCSI is only supported on X86 which has strong
644  * memory access ordering.
645  */
646 static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
647 {
648 	struct PVSCSIRingsState *s = adapter->rings_state;
649 	struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
650 	u32 cmp_entries = s->cmpNumEntriesLog2;
651 
652 	while (s->cmpConsIdx != s->cmpProdIdx) {
653 		struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
654 						      MASK(cmp_entries));
655 		/*
656 		 * This barrier() ensures that *e is not dereferenced while
657 		 * the device emulation still writes data into the slot.
658 		 * Since the device emulation advances s->cmpProdIdx only after
659 		 * updating the slot we want to check it first.
660 		 */
661 		barrier();
662 		pvscsi_complete_request(adapter, e);
663 		/*
664 		 * This barrier() ensures that compiler doesn't reorder write
665 		 * to s->cmpConsIdx before the read of (*e) inside
666 		 * pvscsi_complete_request. Otherwise, device emulation may
667 		 * overwrite *e before we had a chance to read it.
668 		 */
669 		barrier();
670 		s->cmpConsIdx++;
671 	}
672 }
673 
674 /*
675  * Translate a Linux SCSI request into a request ring entry.
676  */
677 static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
678 			     struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
679 {
680 	struct PVSCSIRingsState *s;
681 	struct PVSCSIRingReqDesc *e;
682 	struct scsi_device *sdev;
683 	u32 req_entries;
684 
685 	s = adapter->rings_state;
686 	sdev = cmd->device;
687 	req_entries = s->reqNumEntriesLog2;
688 
689 	/*
690 	 * If this condition holds, we might have room on the request ring, but
691 	 * we might not have room on the completion ring for the response.
692 	 * However, we have already ruled out this possibility - we would not
693 	 * have successfully allocated a context if it were true, since we only
694 	 * have one context per request entry.  Check for it anyway, since it
695 	 * would be a serious bug.
696 	 */
697 	if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
698 		scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
699 			    "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
700 			    s->reqProdIdx, s->cmpConsIdx);
701 		return -1;
702 	}
703 
704 	e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
705 
706 	e->bus    = sdev->channel;
707 	e->target = sdev->id;
708 	memset(e->lun, 0, sizeof(e->lun));
709 	e->lun[1] = sdev->lun;
710 
711 	if (cmd->sense_buffer) {
712 		ctx->sensePA = pci_map_single(adapter->dev, cmd->sense_buffer,
713 					      SCSI_SENSE_BUFFERSIZE,
714 					      PCI_DMA_FROMDEVICE);
715 		e->senseAddr = ctx->sensePA;
716 		e->senseLen = SCSI_SENSE_BUFFERSIZE;
717 	} else {
718 		e->senseLen  = 0;
719 		e->senseAddr = 0;
720 	}
721 	e->cdbLen   = cmd->cmd_len;
722 	e->vcpuHint = smp_processor_id();
723 	memcpy(e->cdb, cmd->cmnd, e->cdbLen);
724 
725 	e->tag = SIMPLE_QUEUE_TAG;
726 	if (sdev->tagged_supported &&
727 	    (cmd->tag == HEAD_OF_QUEUE_TAG ||
728 	     cmd->tag == ORDERED_QUEUE_TAG))
729 		e->tag = cmd->tag;
730 
731 	if (cmd->sc_data_direction == DMA_FROM_DEVICE)
732 		e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
733 	else if (cmd->sc_data_direction == DMA_TO_DEVICE)
734 		e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
735 	else if (cmd->sc_data_direction == DMA_NONE)
736 		e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
737 	else
738 		e->flags = 0;
739 
740 	pvscsi_map_buffers(adapter, ctx, cmd, e);
741 
742 	e->context = pvscsi_map_context(adapter, ctx);
743 
744 	barrier();
745 
746 	s->reqProdIdx++;
747 
748 	return 0;
749 }
750 
751 static int pvscsi_queue_lck(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
752 {
753 	struct Scsi_Host *host = cmd->device->host;
754 	struct pvscsi_adapter *adapter = shost_priv(host);
755 	struct pvscsi_ctx *ctx;
756 	unsigned long flags;
757 
758 	spin_lock_irqsave(&adapter->hw_lock, flags);
759 
760 	ctx = pvscsi_acquire_context(adapter, cmd);
761 	if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
762 		if (ctx)
763 			pvscsi_release_context(adapter, ctx);
764 		spin_unlock_irqrestore(&adapter->hw_lock, flags);
765 		return SCSI_MLQUEUE_HOST_BUSY;
766 	}
767 
768 	cmd->scsi_done = done;
769 
770 	dev_dbg(&cmd->device->sdev_gendev,
771 		"queued cmd %p, ctx %p, op=%x\n", cmd, ctx, cmd->cmnd[0]);
772 
773 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
774 
775 	pvscsi_kick_io(adapter, cmd->cmnd[0]);
776 
777 	return 0;
778 }
779 
780 static DEF_SCSI_QCMD(pvscsi_queue)
781 
782 static int pvscsi_abort(struct scsi_cmnd *cmd)
783 {
784 	struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
785 	struct pvscsi_ctx *ctx;
786 	unsigned long flags;
787 	int result = SUCCESS;
788 	DECLARE_COMPLETION_ONSTACK(abort_cmp);
789 
790 	scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
791 		    adapter->host->host_no, cmd);
792 
793 	spin_lock_irqsave(&adapter->hw_lock, flags);
794 
795 	/*
796 	 * Poll the completion ring first - we might be trying to abort
797 	 * a command that is waiting to be dispatched in the completion ring.
798 	 */
799 	pvscsi_process_completion_ring(adapter);
800 
801 	/*
802 	 * If there is no context for the command, it either already succeeded
803 	 * or else was never properly issued.  Not our problem.
804 	 */
805 	ctx = pvscsi_find_context(adapter, cmd);
806 	if (!ctx) {
807 		scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
808 		goto out;
809 	}
810 
811 	/*
812 	 * Mark that the command has been requested to be aborted and issue
813 	 * the abort.
814 	 */
815 	ctx->abort_cmp = &abort_cmp;
816 
817 	pvscsi_abort_cmd(adapter, ctx);
818 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
819 	/* Wait for 2 secs for the completion. */
820 	wait_for_completion_timeout(&abort_cmp, msecs_to_jiffies(2000));
821 	spin_lock_irqsave(&adapter->hw_lock, flags);
822 
823 	if (!completion_done(&abort_cmp)) {
824 		/*
825 		 * Failed to abort the command, unmark the fact that it
826 		 * was requested to be aborted.
827 		 */
828 		ctx->abort_cmp = NULL;
829 		result = FAILED;
830 		scmd_printk(KERN_DEBUG, cmd,
831 			    "Failed to get completion for aborted cmd %p\n",
832 			    cmd);
833 		goto out;
834 	}
835 
836 	/*
837 	 * Successfully aborted the command.
838 	 */
839 	cmd->result = (DID_ABORT << 16);
840 	cmd->scsi_done(cmd);
841 
842 out:
843 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
844 	return result;
845 }
846 
847 /*
848  * Abort all outstanding requests.  This is only safe to use if the completion
849  * ring will never be walked again or the device has been reset, because it
850  * destroys the 1-1 mapping between context field passed to emulation and our
851  * request structure.
852  */
853 static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
854 {
855 	unsigned i;
856 
857 	for (i = 0; i < adapter->req_depth; i++) {
858 		struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
859 		struct scsi_cmnd *cmd = ctx->cmd;
860 		if (cmd) {
861 			scmd_printk(KERN_ERR, cmd,
862 				    "Forced reset on cmd %p\n", cmd);
863 			pvscsi_unmap_buffers(adapter, ctx);
864 			pvscsi_release_context(adapter, ctx);
865 			cmd->result = (DID_RESET << 16);
866 			cmd->scsi_done(cmd);
867 		}
868 	}
869 }
870 
871 static int pvscsi_host_reset(struct scsi_cmnd *cmd)
872 {
873 	struct Scsi_Host *host = cmd->device->host;
874 	struct pvscsi_adapter *adapter = shost_priv(host);
875 	unsigned long flags;
876 	bool use_msg;
877 
878 	scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
879 
880 	spin_lock_irqsave(&adapter->hw_lock, flags);
881 
882 	use_msg = adapter->use_msg;
883 
884 	if (use_msg) {
885 		adapter->use_msg = 0;
886 		spin_unlock_irqrestore(&adapter->hw_lock, flags);
887 
888 		/*
889 		 * Now that we know that the ISR won't add more work on the
890 		 * workqueue we can safely flush any outstanding work.
891 		 */
892 		flush_workqueue(adapter->workqueue);
893 		spin_lock_irqsave(&adapter->hw_lock, flags);
894 	}
895 
896 	/*
897 	 * We're going to tear down the entire ring structure and set it back
898 	 * up, so stalling new requests until all completions are flushed and
899 	 * the rings are back in place.
900 	 */
901 
902 	pvscsi_process_request_ring(adapter);
903 
904 	ll_adapter_reset(adapter);
905 
906 	/*
907 	 * Now process any completions.  Note we do this AFTER adapter reset,
908 	 * which is strange, but stops races where completions get posted
909 	 * between processing the ring and issuing the reset.  The backend will
910 	 * not touch the ring memory after reset, so the immediately pre-reset
911 	 * completion ring state is still valid.
912 	 */
913 	pvscsi_process_completion_ring(adapter);
914 
915 	pvscsi_reset_all(adapter);
916 	adapter->use_msg = use_msg;
917 	pvscsi_setup_all_rings(adapter);
918 	pvscsi_unmask_intr(adapter);
919 
920 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
921 
922 	return SUCCESS;
923 }
924 
925 static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
926 {
927 	struct Scsi_Host *host = cmd->device->host;
928 	struct pvscsi_adapter *adapter = shost_priv(host);
929 	unsigned long flags;
930 
931 	scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
932 
933 	/*
934 	 * We don't want to queue new requests for this bus after
935 	 * flushing all pending requests to emulation, since new
936 	 * requests could then sneak in during this bus reset phase,
937 	 * so take the lock now.
938 	 */
939 	spin_lock_irqsave(&adapter->hw_lock, flags);
940 
941 	pvscsi_process_request_ring(adapter);
942 	ll_bus_reset(adapter);
943 	pvscsi_process_completion_ring(adapter);
944 
945 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
946 
947 	return SUCCESS;
948 }
949 
950 static int pvscsi_device_reset(struct scsi_cmnd *cmd)
951 {
952 	struct Scsi_Host *host = cmd->device->host;
953 	struct pvscsi_adapter *adapter = shost_priv(host);
954 	unsigned long flags;
955 
956 	scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
957 		    host->host_no, cmd->device->id);
958 
959 	/*
960 	 * We don't want to queue new requests for this device after flushing
961 	 * all pending requests to emulation, since new requests could then
962 	 * sneak in during this device reset phase, so take the lock now.
963 	 */
964 	spin_lock_irqsave(&adapter->hw_lock, flags);
965 
966 	pvscsi_process_request_ring(adapter);
967 	ll_device_reset(adapter, cmd->device->id);
968 	pvscsi_process_completion_ring(adapter);
969 
970 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
971 
972 	return SUCCESS;
973 }
974 
975 static struct scsi_host_template pvscsi_template;
976 
977 static const char *pvscsi_info(struct Scsi_Host *host)
978 {
979 	struct pvscsi_adapter *adapter = shost_priv(host);
980 	static char buf[256];
981 
982 	sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
983 		"%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
984 		adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
985 		pvscsi_template.cmd_per_lun);
986 
987 	return buf;
988 }
989 
990 static struct scsi_host_template pvscsi_template = {
991 	.module				= THIS_MODULE,
992 	.name				= "VMware PVSCSI Host Adapter",
993 	.proc_name			= "vmw_pvscsi",
994 	.info				= pvscsi_info,
995 	.queuecommand			= pvscsi_queue,
996 	.this_id			= -1,
997 	.sg_tablesize			= PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
998 	.dma_boundary			= UINT_MAX,
999 	.max_sectors			= 0xffff,
1000 	.use_clustering			= ENABLE_CLUSTERING,
1001 	.change_queue_depth		= pvscsi_change_queue_depth,
1002 	.eh_abort_handler		= pvscsi_abort,
1003 	.eh_device_reset_handler	= pvscsi_device_reset,
1004 	.eh_bus_reset_handler		= pvscsi_bus_reset,
1005 	.eh_host_reset_handler		= pvscsi_host_reset,
1006 };
1007 
1008 static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
1009 			       const struct PVSCSIRingMsgDesc *e)
1010 {
1011 	struct PVSCSIRingsState *s = adapter->rings_state;
1012 	struct Scsi_Host *host = adapter->host;
1013 	struct scsi_device *sdev;
1014 
1015 	printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
1016 	       e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
1017 
1018 	BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
1019 
1020 	if (e->type == PVSCSI_MSG_DEV_ADDED) {
1021 		struct PVSCSIMsgDescDevStatusChanged *desc;
1022 		desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1023 
1024 		printk(KERN_INFO
1025 		       "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
1026 		       desc->bus, desc->target, desc->lun[1]);
1027 
1028 		if (!scsi_host_get(host))
1029 			return;
1030 
1031 		sdev = scsi_device_lookup(host, desc->bus, desc->target,
1032 					  desc->lun[1]);
1033 		if (sdev) {
1034 			printk(KERN_INFO "vmw_pvscsi: device already exists\n");
1035 			scsi_device_put(sdev);
1036 		} else
1037 			scsi_add_device(adapter->host, desc->bus,
1038 					desc->target, desc->lun[1]);
1039 
1040 		scsi_host_put(host);
1041 	} else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
1042 		struct PVSCSIMsgDescDevStatusChanged *desc;
1043 		desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1044 
1045 		printk(KERN_INFO
1046 		       "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
1047 		       desc->bus, desc->target, desc->lun[1]);
1048 
1049 		if (!scsi_host_get(host))
1050 			return;
1051 
1052 		sdev = scsi_device_lookup(host, desc->bus, desc->target,
1053 					  desc->lun[1]);
1054 		if (sdev) {
1055 			scsi_remove_device(sdev);
1056 			scsi_device_put(sdev);
1057 		} else
1058 			printk(KERN_INFO
1059 			       "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
1060 			       desc->bus, desc->target, desc->lun[1]);
1061 
1062 		scsi_host_put(host);
1063 	}
1064 }
1065 
1066 static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
1067 {
1068 	struct PVSCSIRingsState *s = adapter->rings_state;
1069 
1070 	return s->msgProdIdx != s->msgConsIdx;
1071 }
1072 
1073 static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
1074 {
1075 	struct PVSCSIRingsState *s = adapter->rings_state;
1076 	struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
1077 	u32 msg_entries = s->msgNumEntriesLog2;
1078 
1079 	while (pvscsi_msg_pending(adapter)) {
1080 		struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
1081 						      MASK(msg_entries));
1082 
1083 		barrier();
1084 		pvscsi_process_msg(adapter, e);
1085 		barrier();
1086 		s->msgConsIdx++;
1087 	}
1088 }
1089 
1090 static void pvscsi_msg_workqueue_handler(struct work_struct *data)
1091 {
1092 	struct pvscsi_adapter *adapter;
1093 
1094 	adapter = container_of(data, struct pvscsi_adapter, work);
1095 
1096 	pvscsi_process_msg_ring(adapter);
1097 }
1098 
1099 static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
1100 {
1101 	char name[32];
1102 
1103 	if (!pvscsi_use_msg)
1104 		return 0;
1105 
1106 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1107 			 PVSCSI_CMD_SETUP_MSG_RING);
1108 
1109 	if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
1110 		return 0;
1111 
1112 	snprintf(name, sizeof(name),
1113 		 "vmw_pvscsi_wq_%u", adapter->host->host_no);
1114 
1115 	adapter->workqueue = create_singlethread_workqueue(name);
1116 	if (!adapter->workqueue) {
1117 		printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
1118 		return 0;
1119 	}
1120 	INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
1121 
1122 	return 1;
1123 }
1124 
1125 static bool pvscsi_setup_req_threshold(struct pvscsi_adapter *adapter,
1126 				      bool enable)
1127 {
1128 	u32 val;
1129 
1130 	if (!pvscsi_use_req_threshold)
1131 		return false;
1132 
1133 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1134 			 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD);
1135 	val = pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS);
1136 	if (val == -1) {
1137 		printk(KERN_INFO "vmw_pvscsi: device does not support req_threshold\n");
1138 		return false;
1139 	} else {
1140 		struct PVSCSICmdDescSetupReqCall cmd_msg = { 0 };
1141 		cmd_msg.enable = enable;
1142 		printk(KERN_INFO
1143 		       "vmw_pvscsi: %sabling reqCallThreshold\n",
1144 			enable ? "en" : "dis");
1145 		pvscsi_write_cmd_desc(adapter,
1146 				      PVSCSI_CMD_SETUP_REQCALLTHRESHOLD,
1147 				      &cmd_msg, sizeof(cmd_msg));
1148 		return pvscsi_reg_read(adapter,
1149 				       PVSCSI_REG_OFFSET_COMMAND_STATUS) != 0;
1150 	}
1151 }
1152 
1153 static irqreturn_t pvscsi_isr(int irq, void *devp)
1154 {
1155 	struct pvscsi_adapter *adapter = devp;
1156 	int handled;
1157 
1158 	if (adapter->use_msi || adapter->use_msix)
1159 		handled = true;
1160 	else {
1161 		u32 val = pvscsi_read_intr_status(adapter);
1162 		handled = (val & PVSCSI_INTR_ALL_SUPPORTED) != 0;
1163 		if (handled)
1164 			pvscsi_write_intr_status(devp, val);
1165 	}
1166 
1167 	if (handled) {
1168 		unsigned long flags;
1169 
1170 		spin_lock_irqsave(&adapter->hw_lock, flags);
1171 
1172 		pvscsi_process_completion_ring(adapter);
1173 		if (adapter->use_msg && pvscsi_msg_pending(adapter))
1174 			queue_work(adapter->workqueue, &adapter->work);
1175 
1176 		spin_unlock_irqrestore(&adapter->hw_lock, flags);
1177 	}
1178 
1179 	return IRQ_RETVAL(handled);
1180 }
1181 
1182 static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
1183 {
1184 	struct pvscsi_ctx *ctx = adapter->cmd_map;
1185 	unsigned i;
1186 
1187 	for (i = 0; i < adapter->req_depth; ++i, ++ctx)
1188 		free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
1189 }
1190 
1191 static int pvscsi_setup_msix(const struct pvscsi_adapter *adapter,
1192 			     unsigned int *irq)
1193 {
1194 	struct msix_entry entry = { 0, PVSCSI_VECTOR_COMPLETION };
1195 	int ret;
1196 
1197 	ret = pci_enable_msix_exact(adapter->dev, &entry, 1);
1198 	if (ret)
1199 		return ret;
1200 
1201 	*irq = entry.vector;
1202 
1203 	return 0;
1204 }
1205 
1206 static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
1207 {
1208 	if (adapter->irq) {
1209 		free_irq(adapter->irq, adapter);
1210 		adapter->irq = 0;
1211 	}
1212 	if (adapter->use_msi) {
1213 		pci_disable_msi(adapter->dev);
1214 		adapter->use_msi = 0;
1215 	} else if (adapter->use_msix) {
1216 		pci_disable_msix(adapter->dev);
1217 		adapter->use_msix = 0;
1218 	}
1219 }
1220 
1221 static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
1222 {
1223 	pvscsi_shutdown_intr(adapter);
1224 
1225 	if (adapter->workqueue)
1226 		destroy_workqueue(adapter->workqueue);
1227 
1228 	if (adapter->mmioBase)
1229 		pci_iounmap(adapter->dev, adapter->mmioBase);
1230 
1231 	pci_release_regions(adapter->dev);
1232 
1233 	if (adapter->cmd_map) {
1234 		pvscsi_free_sgls(adapter);
1235 		kfree(adapter->cmd_map);
1236 	}
1237 
1238 	if (adapter->rings_state)
1239 		pci_free_consistent(adapter->dev, PAGE_SIZE,
1240 				    adapter->rings_state, adapter->ringStatePA);
1241 
1242 	if (adapter->req_ring)
1243 		pci_free_consistent(adapter->dev,
1244 				    adapter->req_pages * PAGE_SIZE,
1245 				    adapter->req_ring, adapter->reqRingPA);
1246 
1247 	if (adapter->cmp_ring)
1248 		pci_free_consistent(adapter->dev,
1249 				    adapter->cmp_pages * PAGE_SIZE,
1250 				    adapter->cmp_ring, adapter->cmpRingPA);
1251 
1252 	if (adapter->msg_ring)
1253 		pci_free_consistent(adapter->dev,
1254 				    adapter->msg_pages * PAGE_SIZE,
1255 				    adapter->msg_ring, adapter->msgRingPA);
1256 }
1257 
1258 /*
1259  * Allocate scatter gather lists.
1260  *
1261  * These are statically allocated.  Trying to be clever was not worth it.
1262  *
1263  * Dynamic allocation can fail, and we can't go deep into the memory
1264  * allocator, since we're a SCSI driver, and trying too hard to allocate
1265  * memory might generate disk I/O.  We also don't want to fail disk I/O
1266  * in that case because we can't get an allocation - the I/O could be
1267  * trying to swap out data to free memory.  Since that is pathological,
1268  * just use a statically allocated scatter list.
1269  *
1270  */
1271 static int pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
1272 {
1273 	struct pvscsi_ctx *ctx;
1274 	int i;
1275 
1276 	ctx = adapter->cmd_map;
1277 	BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
1278 
1279 	for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
1280 		ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
1281 						    get_order(SGL_SIZE));
1282 		ctx->sglPA = 0;
1283 		BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
1284 		if (!ctx->sgl) {
1285 			for (; i >= 0; --i, --ctx) {
1286 				free_pages((unsigned long)ctx->sgl,
1287 					   get_order(SGL_SIZE));
1288 				ctx->sgl = NULL;
1289 			}
1290 			return -ENOMEM;
1291 		}
1292 	}
1293 
1294 	return 0;
1295 }
1296 
1297 /*
1298  * Query the device, fetch the config info and return the
1299  * maximum number of targets on the adapter. In case of
1300  * failure due to any reason return default i.e. 16.
1301  */
1302 static u32 pvscsi_get_max_targets(struct pvscsi_adapter *adapter)
1303 {
1304 	struct PVSCSICmdDescConfigCmd cmd;
1305 	struct PVSCSIConfigPageHeader *header;
1306 	struct device *dev;
1307 	dma_addr_t configPagePA;
1308 	void *config_page;
1309 	u32 numPhys = 16;
1310 
1311 	dev = pvscsi_dev(adapter);
1312 	config_page = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
1313 					   &configPagePA);
1314 	if (!config_page) {
1315 		dev_warn(dev, "vmw_pvscsi: failed to allocate memory for config page\n");
1316 		goto exit;
1317 	}
1318 	BUG_ON(configPagePA & ~PAGE_MASK);
1319 
1320 	/* Fetch config info from the device. */
1321 	cmd.configPageAddress = ((u64)PVSCSI_CONFIG_CONTROLLER_ADDRESS) << 32;
1322 	cmd.configPageNum = PVSCSI_CONFIG_PAGE_CONTROLLER;
1323 	cmd.cmpAddr = configPagePA;
1324 	cmd._pad = 0;
1325 
1326 	/*
1327 	 * Mark the completion page header with error values. If the device
1328 	 * completes the command successfully, it sets the status values to
1329 	 * indicate success.
1330 	 */
1331 	header = config_page;
1332 	memset(header, 0, sizeof *header);
1333 	header->hostStatus = BTSTAT_INVPARAM;
1334 	header->scsiStatus = SDSTAT_CHECK;
1335 
1336 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_CONFIG, &cmd, sizeof cmd);
1337 
1338 	if (header->hostStatus == BTSTAT_SUCCESS &&
1339 	    header->scsiStatus == SDSTAT_GOOD) {
1340 		struct PVSCSIConfigPageController *config;
1341 
1342 		config = config_page;
1343 		numPhys = config->numPhys;
1344 	} else
1345 		dev_warn(dev, "vmw_pvscsi: PVSCSI_CMD_CONFIG failed. hostStatus = 0x%x, scsiStatus = 0x%x\n",
1346 			 header->hostStatus, header->scsiStatus);
1347 	pci_free_consistent(adapter->dev, PAGE_SIZE, config_page, configPagePA);
1348 exit:
1349 	return numPhys;
1350 }
1351 
1352 static int pvscsi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1353 {
1354 	struct pvscsi_adapter *adapter;
1355 	struct pvscsi_adapter adapter_temp;
1356 	struct Scsi_Host *host = NULL;
1357 	unsigned int i;
1358 	unsigned long flags = 0;
1359 	int error;
1360 	u32 max_id;
1361 
1362 	error = -ENODEV;
1363 
1364 	if (pci_enable_device(pdev))
1365 		return error;
1366 
1367 	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0 &&
1368 	    pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1369 		printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
1370 	} else if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) == 0 &&
1371 		   pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) == 0) {
1372 		printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
1373 	} else {
1374 		printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
1375 		goto out_disable_device;
1376 	}
1377 
1378 	/*
1379 	 * Let's use a temp pvscsi_adapter struct until we find the number of
1380 	 * targets on the adapter, after that we will switch to the real
1381 	 * allocated struct.
1382 	 */
1383 	adapter = &adapter_temp;
1384 	memset(adapter, 0, sizeof(*adapter));
1385 	adapter->dev  = pdev;
1386 	adapter->rev = pdev->revision;
1387 
1388 	if (pci_request_regions(pdev, "vmw_pvscsi")) {
1389 		printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
1390 		goto out_disable_device;
1391 	}
1392 
1393 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1394 		if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
1395 			continue;
1396 
1397 		if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
1398 			continue;
1399 
1400 		break;
1401 	}
1402 
1403 	if (i == DEVICE_COUNT_RESOURCE) {
1404 		printk(KERN_ERR
1405 		       "vmw_pvscsi: adapter has no suitable MMIO region\n");
1406 		goto out_release_resources_and_disable;
1407 	}
1408 
1409 	adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
1410 
1411 	if (!adapter->mmioBase) {
1412 		printk(KERN_ERR
1413 		       "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
1414 		       i, PVSCSI_MEM_SPACE_SIZE);
1415 		goto out_release_resources_and_disable;
1416 	}
1417 
1418 	pci_set_master(pdev);
1419 
1420 	/*
1421 	 * Ask the device for max number of targets before deciding the
1422 	 * default pvscsi_ring_pages value.
1423 	 */
1424 	max_id = pvscsi_get_max_targets(adapter);
1425 	printk(KERN_INFO "vmw_pvscsi: max_id: %u\n", max_id);
1426 
1427 	if (pvscsi_ring_pages == 0)
1428 		/*
1429 		 * Set the right default value. Up to 16 it is 8, above it is
1430 		 * max.
1431 		 */
1432 		pvscsi_ring_pages = (max_id > 16) ?
1433 			PVSCSI_SETUP_RINGS_MAX_NUM_PAGES :
1434 			PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
1435 	printk(KERN_INFO
1436 	       "vmw_pvscsi: setting ring_pages to %d\n",
1437 	       pvscsi_ring_pages);
1438 
1439 	pvscsi_template.can_queue =
1440 		min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
1441 		PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
1442 	pvscsi_template.cmd_per_lun =
1443 		min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
1444 	host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
1445 	if (!host) {
1446 		printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
1447 		goto out_release_resources_and_disable;
1448 	}
1449 
1450 	/*
1451 	 * Let's use the real pvscsi_adapter struct here onwards.
1452 	 */
1453 	adapter = shost_priv(host);
1454 	memset(adapter, 0, sizeof(*adapter));
1455 	adapter->dev  = pdev;
1456 	adapter->host = host;
1457 	/*
1458 	 * Copy back what we already have to the allocated adapter struct.
1459 	 */
1460 	adapter->rev = adapter_temp.rev;
1461 	adapter->mmioBase = adapter_temp.mmioBase;
1462 
1463 	spin_lock_init(&adapter->hw_lock);
1464 	host->max_channel = 0;
1465 	host->max_lun     = 1;
1466 	host->max_cmd_len = 16;
1467 	host->max_id      = max_id;
1468 
1469 	pci_set_drvdata(pdev, host);
1470 
1471 	ll_adapter_reset(adapter);
1472 
1473 	adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
1474 
1475 	error = pvscsi_allocate_rings(adapter);
1476 	if (error) {
1477 		printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
1478 		goto out_release_resources;
1479 	}
1480 
1481 	/*
1482 	 * From this point on we should reset the adapter if anything goes
1483 	 * wrong.
1484 	 */
1485 	pvscsi_setup_all_rings(adapter);
1486 
1487 	adapter->cmd_map = kcalloc(adapter->req_depth,
1488 				   sizeof(struct pvscsi_ctx), GFP_KERNEL);
1489 	if (!adapter->cmd_map) {
1490 		printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
1491 		error = -ENOMEM;
1492 		goto out_reset_adapter;
1493 	}
1494 
1495 	INIT_LIST_HEAD(&adapter->cmd_pool);
1496 	for (i = 0; i < adapter->req_depth; i++) {
1497 		struct pvscsi_ctx *ctx = adapter->cmd_map + i;
1498 		list_add(&ctx->list, &adapter->cmd_pool);
1499 	}
1500 
1501 	error = pvscsi_allocate_sg(adapter);
1502 	if (error) {
1503 		printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
1504 		goto out_reset_adapter;
1505 	}
1506 
1507 	if (!pvscsi_disable_msix &&
1508 	    pvscsi_setup_msix(adapter, &adapter->irq) == 0) {
1509 		printk(KERN_INFO "vmw_pvscsi: using MSI-X\n");
1510 		adapter->use_msix = 1;
1511 	} else if (!pvscsi_disable_msi && pci_enable_msi(pdev) == 0) {
1512 		printk(KERN_INFO "vmw_pvscsi: using MSI\n");
1513 		adapter->use_msi = 1;
1514 		adapter->irq = pdev->irq;
1515 	} else {
1516 		printk(KERN_INFO "vmw_pvscsi: using INTx\n");
1517 		adapter->irq = pdev->irq;
1518 		flags = IRQF_SHARED;
1519 	}
1520 
1521 	adapter->use_req_threshold = pvscsi_setup_req_threshold(adapter, true);
1522 	printk(KERN_DEBUG "vmw_pvscsi: driver-based request coalescing %sabled\n",
1523 	       adapter->use_req_threshold ? "en" : "dis");
1524 
1525 	error = request_irq(adapter->irq, pvscsi_isr, flags,
1526 			    "vmw_pvscsi", adapter);
1527 	if (error) {
1528 		printk(KERN_ERR
1529 		       "vmw_pvscsi: unable to request IRQ: %d\n", error);
1530 		adapter->irq = 0;
1531 		goto out_reset_adapter;
1532 	}
1533 
1534 	error = scsi_add_host(host, &pdev->dev);
1535 	if (error) {
1536 		printk(KERN_ERR
1537 		       "vmw_pvscsi: scsi_add_host failed: %d\n", error);
1538 		goto out_reset_adapter;
1539 	}
1540 
1541 	dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
1542 		 adapter->rev, host->host_no);
1543 
1544 	pvscsi_unmask_intr(adapter);
1545 
1546 	scsi_scan_host(host);
1547 
1548 	return 0;
1549 
1550 out_reset_adapter:
1551 	ll_adapter_reset(adapter);
1552 out_release_resources:
1553 	pvscsi_release_resources(adapter);
1554 	scsi_host_put(host);
1555 out_disable_device:
1556 	pci_disable_device(pdev);
1557 
1558 	return error;
1559 
1560 out_release_resources_and_disable:
1561 	pvscsi_release_resources(adapter);
1562 	goto out_disable_device;
1563 }
1564 
1565 static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
1566 {
1567 	pvscsi_mask_intr(adapter);
1568 
1569 	if (adapter->workqueue)
1570 		flush_workqueue(adapter->workqueue);
1571 
1572 	pvscsi_shutdown_intr(adapter);
1573 
1574 	pvscsi_process_request_ring(adapter);
1575 	pvscsi_process_completion_ring(adapter);
1576 	ll_adapter_reset(adapter);
1577 }
1578 
1579 static void pvscsi_shutdown(struct pci_dev *dev)
1580 {
1581 	struct Scsi_Host *host = pci_get_drvdata(dev);
1582 	struct pvscsi_adapter *adapter = shost_priv(host);
1583 
1584 	__pvscsi_shutdown(adapter);
1585 }
1586 
1587 static void pvscsi_remove(struct pci_dev *pdev)
1588 {
1589 	struct Scsi_Host *host = pci_get_drvdata(pdev);
1590 	struct pvscsi_adapter *adapter = shost_priv(host);
1591 
1592 	scsi_remove_host(host);
1593 
1594 	__pvscsi_shutdown(adapter);
1595 	pvscsi_release_resources(adapter);
1596 
1597 	scsi_host_put(host);
1598 
1599 	pci_disable_device(pdev);
1600 }
1601 
1602 static struct pci_driver pvscsi_pci_driver = {
1603 	.name		= "vmw_pvscsi",
1604 	.id_table	= pvscsi_pci_tbl,
1605 	.probe		= pvscsi_probe,
1606 	.remove		= pvscsi_remove,
1607 	.shutdown       = pvscsi_shutdown,
1608 };
1609 
1610 static int __init pvscsi_init(void)
1611 {
1612 	pr_info("%s - version %s\n",
1613 		PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
1614 	return pci_register_driver(&pvscsi_pci_driver);
1615 }
1616 
1617 static void __exit pvscsi_exit(void)
1618 {
1619 	pci_unregister_driver(&pvscsi_pci_driver);
1620 }
1621 
1622 module_init(pvscsi_init);
1623 module_exit(pvscsi_exit);
1624