xref: /openbmc/linux/drivers/scsi/vmw_pvscsi.c (revision 9be08a27)
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: Jim Gill <jgill@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 	u8				rev;
72 	bool				use_msg;
73 	bool				use_req_threshold;
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 */
105 static int pvscsi_ring_pages;
106 static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
107 static int pvscsi_cmd_per_lun    = PVSCSI_DEFAULT_QUEUE_DEPTH;
108 static bool pvscsi_disable_msi;
109 static bool pvscsi_disable_msix;
110 static bool pvscsi_use_msg       = true;
111 static bool pvscsi_use_req_threshold = true;
112 
113 #define PVSCSI_RW (S_IRUSR | S_IWUSR)
114 
115 module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
116 MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default="
117 		 __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING)
118 		 "[up to 16 targets],"
119 		 __stringify(PVSCSI_SETUP_RINGS_MAX_NUM_PAGES)
120 		 "[for 16+ targets])");
121 
122 module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
123 MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
124 		 __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
125 
126 module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
127 MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
128 		 __stringify(PVSCSI_DEFAULT_QUEUE_DEPTH) ")");
129 
130 module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
131 MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
132 
133 module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
134 MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
135 
136 module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
137 MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
138 
139 module_param_named(use_req_threshold, pvscsi_use_req_threshold,
140 		   bool, PVSCSI_RW);
141 MODULE_PARM_DESC(use_req_threshold, "Use driver-based request coalescing if configured - (default=1)");
142 
143 static const struct pci_device_id pvscsi_pci_tbl[] = {
144 	{ PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
145 	{ 0 }
146 };
147 
148 MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
149 
150 static struct device *
151 pvscsi_dev(const struct pvscsi_adapter *adapter)
152 {
153 	return &(adapter->dev->dev);
154 }
155 
156 static struct pvscsi_ctx *
157 pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
158 {
159 	struct pvscsi_ctx *ctx, *end;
160 
161 	end = &adapter->cmd_map[adapter->req_depth];
162 	for (ctx = adapter->cmd_map; ctx < end; ctx++)
163 		if (ctx->cmd == cmd)
164 			return ctx;
165 
166 	return NULL;
167 }
168 
169 static struct pvscsi_ctx *
170 pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
171 {
172 	struct pvscsi_ctx *ctx;
173 
174 	if (list_empty(&adapter->cmd_pool))
175 		return NULL;
176 
177 	ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list);
178 	ctx->cmd = cmd;
179 	list_del(&ctx->list);
180 
181 	return ctx;
182 }
183 
184 static void pvscsi_release_context(struct pvscsi_adapter *adapter,
185 				   struct pvscsi_ctx *ctx)
186 {
187 	ctx->cmd = NULL;
188 	ctx->abort_cmp = NULL;
189 	list_add(&ctx->list, &adapter->cmd_pool);
190 }
191 
192 /*
193  * Map a pvscsi_ctx struct to a context ID field value; we map to a simple
194  * non-zero integer. ctx always points to an entry in cmd_map array, hence
195  * the return value is always >=1.
196  */
197 static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter,
198 			      const struct pvscsi_ctx *ctx)
199 {
200 	return ctx - adapter->cmd_map + 1;
201 }
202 
203 static struct pvscsi_ctx *
204 pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
205 {
206 	return &adapter->cmd_map[context - 1];
207 }
208 
209 static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
210 			     u32 offset, u32 val)
211 {
212 	writel(val, adapter->mmioBase + offset);
213 }
214 
215 static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
216 {
217 	return readl(adapter->mmioBase + offset);
218 }
219 
220 static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
221 {
222 	return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
223 }
224 
225 static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter,
226 				     u32 val)
227 {
228 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val);
229 }
230 
231 static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter)
232 {
233 	u32 intr_bits;
234 
235 	intr_bits = PVSCSI_INTR_CMPL_MASK;
236 	if (adapter->use_msg)
237 		intr_bits |= PVSCSI_INTR_MSG_MASK;
238 
239 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits);
240 }
241 
242 static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
243 {
244 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
245 }
246 
247 static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter,
248 				  u32 cmd, const void *desc, size_t len)
249 {
250 	const u32 *ptr = desc;
251 	size_t i;
252 
253 	len /= sizeof(*ptr);
254 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd);
255 	for (i = 0; i < len; i++)
256 		pvscsi_reg_write(adapter,
257 				 PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
258 }
259 
260 static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter,
261 			     const struct pvscsi_ctx *ctx)
262 {
263 	struct PVSCSICmdDescAbortCmd cmd = { 0 };
264 
265 	cmd.target = ctx->cmd->device->id;
266 	cmd.context = pvscsi_map_context(adapter, ctx);
267 
268 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd));
269 }
270 
271 static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter)
272 {
273 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
274 }
275 
276 static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter)
277 {
278 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0);
279 }
280 
281 static int scsi_is_rw(unsigned char op)
282 {
283 	return op == READ_6  || op == WRITE_6 ||
284 	       op == READ_10 || op == WRITE_10 ||
285 	       op == READ_12 || op == WRITE_12 ||
286 	       op == READ_16 || op == WRITE_16;
287 }
288 
289 static void pvscsi_kick_io(const struct pvscsi_adapter *adapter,
290 			   unsigned char op)
291 {
292 	if (scsi_is_rw(op)) {
293 		struct PVSCSIRingsState *s = adapter->rings_state;
294 
295 		if (!adapter->use_req_threshold ||
296 		    s->reqProdIdx - s->reqConsIdx >= s->reqCallThreshold)
297 			pvscsi_kick_rw_io(adapter);
298 	} else {
299 		pvscsi_process_request_ring(adapter);
300 	}
301 }
302 
303 static void ll_adapter_reset(const struct pvscsi_adapter *adapter)
304 {
305 	dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter);
306 
307 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0);
308 }
309 
310 static void ll_bus_reset(const struct pvscsi_adapter *adapter)
311 {
312 	dev_dbg(pvscsi_dev(adapter), "Resetting bus on %p\n", adapter);
313 
314 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0);
315 }
316 
317 static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target)
318 {
319 	struct PVSCSICmdDescResetDevice cmd = { 0 };
320 
321 	dev_dbg(pvscsi_dev(adapter), "Resetting device: target=%u\n", target);
322 
323 	cmd.target = target;
324 
325 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE,
326 			      &cmd, sizeof(cmd));
327 }
328 
329 static void pvscsi_create_sg(struct pvscsi_ctx *ctx,
330 			     struct scatterlist *sg, unsigned count)
331 {
332 	unsigned i;
333 	struct PVSCSISGElement *sge;
334 
335 	BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT);
336 
337 	sge = &ctx->sgl->sge[0];
338 	for (i = 0; i < count; i++, sg++) {
339 		sge[i].addr   = sg_dma_address(sg);
340 		sge[i].length = sg_dma_len(sg);
341 		sge[i].flags  = 0;
342 	}
343 }
344 
345 /*
346  * Map all data buffers for a command into PCI space and
347  * setup the scatter/gather list if needed.
348  */
349 static int pvscsi_map_buffers(struct pvscsi_adapter *adapter,
350 			      struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd,
351 			      struct PVSCSIRingReqDesc *e)
352 {
353 	unsigned count;
354 	unsigned bufflen = scsi_bufflen(cmd);
355 	struct scatterlist *sg;
356 
357 	e->dataLen = bufflen;
358 	e->dataAddr = 0;
359 	if (bufflen == 0)
360 		return 0;
361 
362 	sg = scsi_sglist(cmd);
363 	count = scsi_sg_count(cmd);
364 	if (count != 0) {
365 		int segs = scsi_dma_map(cmd);
366 
367 		if (segs == -ENOMEM) {
368 			scmd_printk(KERN_ERR, cmd,
369 				    "vmw_pvscsi: Failed to map cmd sglist for DMA.\n");
370 			return -ENOMEM;
371 		} else if (segs > 1) {
372 			pvscsi_create_sg(ctx, sg, segs);
373 
374 			e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
375 			ctx->sglPA = pci_map_single(adapter->dev, ctx->sgl,
376 						    SGL_SIZE, PCI_DMA_TODEVICE);
377 			if (pci_dma_mapping_error(adapter->dev, ctx->sglPA)) {
378 				scmd_printk(KERN_ERR, cmd,
379 					    "vmw_pvscsi: Failed to map ctx sglist for DMA.\n");
380 				scsi_dma_unmap(cmd);
381 				ctx->sglPA = 0;
382 				return -ENOMEM;
383 			}
384 			e->dataAddr = ctx->sglPA;
385 		} else
386 			e->dataAddr = sg_dma_address(sg);
387 	} else {
388 		/*
389 		 * In case there is no S/G list, scsi_sglist points
390 		 * directly to the buffer.
391 		 */
392 		ctx->dataPA = pci_map_single(adapter->dev, sg, bufflen,
393 					     cmd->sc_data_direction);
394 		if (pci_dma_mapping_error(adapter->dev, ctx->dataPA)) {
395 			scmd_printk(KERN_ERR, cmd,
396 				    "vmw_pvscsi: Failed to map direct data buffer for DMA.\n");
397 			return -ENOMEM;
398 		}
399 		e->dataAddr = ctx->dataPA;
400 	}
401 
402 	return 0;
403 }
404 
405 static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
406 				 struct pvscsi_ctx *ctx)
407 {
408 	struct scsi_cmnd *cmd;
409 	unsigned bufflen;
410 
411 	cmd = ctx->cmd;
412 	bufflen = scsi_bufflen(cmd);
413 
414 	if (bufflen != 0) {
415 		unsigned count = scsi_sg_count(cmd);
416 
417 		if (count != 0) {
418 			scsi_dma_unmap(cmd);
419 			if (ctx->sglPA) {
420 				pci_unmap_single(adapter->dev, ctx->sglPA,
421 						 SGL_SIZE, PCI_DMA_TODEVICE);
422 				ctx->sglPA = 0;
423 			}
424 		} else
425 			pci_unmap_single(adapter->dev, ctx->dataPA, bufflen,
426 					 cmd->sc_data_direction);
427 	}
428 	if (cmd->sense_buffer)
429 		pci_unmap_single(adapter->dev, ctx->sensePA,
430 				 SCSI_SENSE_BUFFERSIZE, PCI_DMA_FROMDEVICE);
431 }
432 
433 static int pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
434 {
435 	adapter->rings_state = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
436 						    &adapter->ringStatePA);
437 	if (!adapter->rings_state)
438 		return -ENOMEM;
439 
440 	adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
441 				 pvscsi_ring_pages);
442 	adapter->req_depth = adapter->req_pages
443 					* PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
444 	adapter->req_ring = pci_alloc_consistent(adapter->dev,
445 						 adapter->req_pages * PAGE_SIZE,
446 						 &adapter->reqRingPA);
447 	if (!adapter->req_ring)
448 		return -ENOMEM;
449 
450 	adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
451 				 pvscsi_ring_pages);
452 	adapter->cmp_ring = pci_alloc_consistent(adapter->dev,
453 						 adapter->cmp_pages * PAGE_SIZE,
454 						 &adapter->cmpRingPA);
455 	if (!adapter->cmp_ring)
456 		return -ENOMEM;
457 
458 	BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
459 	BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
460 	BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
461 
462 	if (!adapter->use_msg)
463 		return 0;
464 
465 	adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
466 				 pvscsi_msg_ring_pages);
467 	adapter->msg_ring = pci_alloc_consistent(adapter->dev,
468 						 adapter->msg_pages * PAGE_SIZE,
469 						 &adapter->msgRingPA);
470 	if (!adapter->msg_ring)
471 		return -ENOMEM;
472 	BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
473 
474 	return 0;
475 }
476 
477 static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
478 {
479 	struct PVSCSICmdDescSetupRings cmd = { 0 };
480 	dma_addr_t base;
481 	unsigned i;
482 
483 	cmd.ringsStatePPN   = adapter->ringStatePA >> PAGE_SHIFT;
484 	cmd.reqRingNumPages = adapter->req_pages;
485 	cmd.cmpRingNumPages = adapter->cmp_pages;
486 
487 	base = adapter->reqRingPA;
488 	for (i = 0; i < adapter->req_pages; i++) {
489 		cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
490 		base += PAGE_SIZE;
491 	}
492 
493 	base = adapter->cmpRingPA;
494 	for (i = 0; i < adapter->cmp_pages; i++) {
495 		cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
496 		base += PAGE_SIZE;
497 	}
498 
499 	memset(adapter->rings_state, 0, PAGE_SIZE);
500 	memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
501 	memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
502 
503 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
504 			      &cmd, sizeof(cmd));
505 
506 	if (adapter->use_msg) {
507 		struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
508 
509 		cmd_msg.numPages = adapter->msg_pages;
510 
511 		base = adapter->msgRingPA;
512 		for (i = 0; i < adapter->msg_pages; i++) {
513 			cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
514 			base += PAGE_SIZE;
515 		}
516 		memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
517 
518 		pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
519 				      &cmd_msg, sizeof(cmd_msg));
520 	}
521 }
522 
523 static int pvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
524 {
525 	if (!sdev->tagged_supported)
526 		qdepth = 1;
527 	return scsi_change_queue_depth(sdev, qdepth);
528 }
529 
530 /*
531  * Pull a completion descriptor off and pass the completion back
532  * to the SCSI mid layer.
533  */
534 static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
535 				    const struct PVSCSIRingCmpDesc *e)
536 {
537 	struct pvscsi_ctx *ctx;
538 	struct scsi_cmnd *cmd;
539 	struct completion *abort_cmp;
540 	u32 btstat = e->hostStatus;
541 	u32 sdstat = e->scsiStatus;
542 
543 	ctx = pvscsi_get_context(adapter, e->context);
544 	cmd = ctx->cmd;
545 	abort_cmp = ctx->abort_cmp;
546 	pvscsi_unmap_buffers(adapter, ctx);
547 	pvscsi_release_context(adapter, ctx);
548 	if (abort_cmp) {
549 		/*
550 		 * The command was requested to be aborted. Just signal that
551 		 * the request completed and swallow the actual cmd completion
552 		 * here. The abort handler will post a completion for this
553 		 * command indicating that it got successfully aborted.
554 		 */
555 		complete(abort_cmp);
556 		return;
557 	}
558 
559 	cmd->result = 0;
560 	if (sdstat != SAM_STAT_GOOD &&
561 	    (btstat == BTSTAT_SUCCESS ||
562 	     btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
563 	     btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
564 		if (sdstat == SAM_STAT_COMMAND_TERMINATED) {
565 			cmd->result = (DID_RESET << 16);
566 		} else {
567 			cmd->result = (DID_OK << 16) | sdstat;
568 			if (sdstat == SAM_STAT_CHECK_CONDITION &&
569 			    cmd->sense_buffer)
570 				cmd->result |= (DRIVER_SENSE << 24);
571 		}
572 	} else
573 		switch (btstat) {
574 		case BTSTAT_SUCCESS:
575 		case BTSTAT_LINKED_COMMAND_COMPLETED:
576 		case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
577 			/* If everything went fine, let's move on..  */
578 			cmd->result = (DID_OK << 16);
579 			break;
580 
581 		case BTSTAT_DATARUN:
582 		case BTSTAT_DATA_UNDERRUN:
583 			/* Report residual data in underruns */
584 			scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
585 			cmd->result = (DID_ERROR << 16);
586 			break;
587 
588 		case BTSTAT_SELTIMEO:
589 			/* Our emulation returns this for non-connected devs */
590 			cmd->result = (DID_BAD_TARGET << 16);
591 			break;
592 
593 		case BTSTAT_LUNMISMATCH:
594 		case BTSTAT_TAGREJECT:
595 		case BTSTAT_BADMSG:
596 			cmd->result = (DRIVER_INVALID << 24);
597 			/* fall through */
598 
599 		case BTSTAT_HAHARDWARE:
600 		case BTSTAT_INVPHASE:
601 		case BTSTAT_HATIMEOUT:
602 		case BTSTAT_NORESPONSE:
603 		case BTSTAT_DISCONNECT:
604 		case BTSTAT_HASOFTWARE:
605 		case BTSTAT_BUSFREE:
606 		case BTSTAT_SENSFAILED:
607 			cmd->result |= (DID_ERROR << 16);
608 			break;
609 
610 		case BTSTAT_SENTRST:
611 		case BTSTAT_RECVRST:
612 		case BTSTAT_BUSRESET:
613 			cmd->result = (DID_RESET << 16);
614 			break;
615 
616 		case BTSTAT_ABORTQUEUE:
617 			cmd->result = (DID_BUS_BUSY << 16);
618 			break;
619 
620 		case BTSTAT_SCSIPARITY:
621 			cmd->result = (DID_PARITY << 16);
622 			break;
623 
624 		default:
625 			cmd->result = (DID_ERROR << 16);
626 			scmd_printk(KERN_DEBUG, cmd,
627 				    "Unknown completion status: 0x%x\n",
628 				    btstat);
629 	}
630 
631 	dev_dbg(&cmd->device->sdev_gendev,
632 		"cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
633 		cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
634 
635 	cmd->scsi_done(cmd);
636 }
637 
638 /*
639  * barrier usage : Since the PVSCSI device is emulated, there could be cases
640  * where we may want to serialize some accesses between the driver and the
641  * emulation layer. We use compiler barriers instead of the more expensive
642  * memory barriers because PVSCSI is only supported on X86 which has strong
643  * memory access ordering.
644  */
645 static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
646 {
647 	struct PVSCSIRingsState *s = adapter->rings_state;
648 	struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
649 	u32 cmp_entries = s->cmpNumEntriesLog2;
650 
651 	while (s->cmpConsIdx != s->cmpProdIdx) {
652 		struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
653 						      MASK(cmp_entries));
654 		/*
655 		 * This barrier() ensures that *e is not dereferenced while
656 		 * the device emulation still writes data into the slot.
657 		 * Since the device emulation advances s->cmpProdIdx only after
658 		 * updating the slot we want to check it first.
659 		 */
660 		barrier();
661 		pvscsi_complete_request(adapter, e);
662 		/*
663 		 * This barrier() ensures that compiler doesn't reorder write
664 		 * to s->cmpConsIdx before the read of (*e) inside
665 		 * pvscsi_complete_request. Otherwise, device emulation may
666 		 * overwrite *e before we had a chance to read it.
667 		 */
668 		barrier();
669 		s->cmpConsIdx++;
670 	}
671 }
672 
673 /*
674  * Translate a Linux SCSI request into a request ring entry.
675  */
676 static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
677 			     struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
678 {
679 	struct PVSCSIRingsState *s;
680 	struct PVSCSIRingReqDesc *e;
681 	struct scsi_device *sdev;
682 	u32 req_entries;
683 
684 	s = adapter->rings_state;
685 	sdev = cmd->device;
686 	req_entries = s->reqNumEntriesLog2;
687 
688 	/*
689 	 * If this condition holds, we might have room on the request ring, but
690 	 * we might not have room on the completion ring for the response.
691 	 * However, we have already ruled out this possibility - we would not
692 	 * have successfully allocated a context if it were true, since we only
693 	 * have one context per request entry.  Check for it anyway, since it
694 	 * would be a serious bug.
695 	 */
696 	if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
697 		scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
698 			    "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
699 			    s->reqProdIdx, s->cmpConsIdx);
700 		return -1;
701 	}
702 
703 	e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
704 
705 	e->bus    = sdev->channel;
706 	e->target = sdev->id;
707 	memset(e->lun, 0, sizeof(e->lun));
708 	e->lun[1] = sdev->lun;
709 
710 	if (cmd->sense_buffer) {
711 		ctx->sensePA = pci_map_single(adapter->dev, cmd->sense_buffer,
712 					      SCSI_SENSE_BUFFERSIZE,
713 					      PCI_DMA_FROMDEVICE);
714 		if (pci_dma_mapping_error(adapter->dev, ctx->sensePA)) {
715 			scmd_printk(KERN_ERR, cmd,
716 				    "vmw_pvscsi: Failed to map sense buffer for DMA.\n");
717 			ctx->sensePA = 0;
718 			return -ENOMEM;
719 		}
720 		e->senseAddr = ctx->sensePA;
721 		e->senseLen = SCSI_SENSE_BUFFERSIZE;
722 	} else {
723 		e->senseLen  = 0;
724 		e->senseAddr = 0;
725 	}
726 	e->cdbLen   = cmd->cmd_len;
727 	e->vcpuHint = smp_processor_id();
728 	memcpy(e->cdb, cmd->cmnd, e->cdbLen);
729 
730 	e->tag = SIMPLE_QUEUE_TAG;
731 
732 	if (cmd->sc_data_direction == DMA_FROM_DEVICE)
733 		e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
734 	else if (cmd->sc_data_direction == DMA_TO_DEVICE)
735 		e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
736 	else if (cmd->sc_data_direction == DMA_NONE)
737 		e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
738 	else
739 		e->flags = 0;
740 
741 	if (pvscsi_map_buffers(adapter, ctx, cmd, e) != 0) {
742 		if (cmd->sense_buffer) {
743 			pci_unmap_single(adapter->dev, ctx->sensePA,
744 					 SCSI_SENSE_BUFFERSIZE,
745 					 PCI_DMA_FROMDEVICE);
746 			ctx->sensePA = 0;
747 		}
748 		return -ENOMEM;
749 	}
750 
751 	e->context = pvscsi_map_context(adapter, ctx);
752 
753 	barrier();
754 
755 	s->reqProdIdx++;
756 
757 	return 0;
758 }
759 
760 static int pvscsi_queue_lck(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
761 {
762 	struct Scsi_Host *host = cmd->device->host;
763 	struct pvscsi_adapter *adapter = shost_priv(host);
764 	struct pvscsi_ctx *ctx;
765 	unsigned long flags;
766 
767 	spin_lock_irqsave(&adapter->hw_lock, flags);
768 
769 	ctx = pvscsi_acquire_context(adapter, cmd);
770 	if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
771 		if (ctx)
772 			pvscsi_release_context(adapter, ctx);
773 		spin_unlock_irqrestore(&adapter->hw_lock, flags);
774 		return SCSI_MLQUEUE_HOST_BUSY;
775 	}
776 
777 	cmd->scsi_done = done;
778 
779 	dev_dbg(&cmd->device->sdev_gendev,
780 		"queued cmd %p, ctx %p, op=%x\n", cmd, ctx, cmd->cmnd[0]);
781 
782 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
783 
784 	pvscsi_kick_io(adapter, cmd->cmnd[0]);
785 
786 	return 0;
787 }
788 
789 static DEF_SCSI_QCMD(pvscsi_queue)
790 
791 static int pvscsi_abort(struct scsi_cmnd *cmd)
792 {
793 	struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
794 	struct pvscsi_ctx *ctx;
795 	unsigned long flags;
796 	int result = SUCCESS;
797 	DECLARE_COMPLETION_ONSTACK(abort_cmp);
798 	int done;
799 
800 	scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
801 		    adapter->host->host_no, cmd);
802 
803 	spin_lock_irqsave(&adapter->hw_lock, flags);
804 
805 	/*
806 	 * Poll the completion ring first - we might be trying to abort
807 	 * a command that is waiting to be dispatched in the completion ring.
808 	 */
809 	pvscsi_process_completion_ring(adapter);
810 
811 	/*
812 	 * If there is no context for the command, it either already succeeded
813 	 * or else was never properly issued.  Not our problem.
814 	 */
815 	ctx = pvscsi_find_context(adapter, cmd);
816 	if (!ctx) {
817 		scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
818 		goto out;
819 	}
820 
821 	/*
822 	 * Mark that the command has been requested to be aborted and issue
823 	 * the abort.
824 	 */
825 	ctx->abort_cmp = &abort_cmp;
826 
827 	pvscsi_abort_cmd(adapter, ctx);
828 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
829 	/* Wait for 2 secs for the completion. */
830 	done = wait_for_completion_timeout(&abort_cmp, msecs_to_jiffies(2000));
831 	spin_lock_irqsave(&adapter->hw_lock, flags);
832 
833 	if (!done) {
834 		/*
835 		 * Failed to abort the command, unmark the fact that it
836 		 * was requested to be aborted.
837 		 */
838 		ctx->abort_cmp = NULL;
839 		result = FAILED;
840 		scmd_printk(KERN_DEBUG, cmd,
841 			    "Failed to get completion for aborted cmd %p\n",
842 			    cmd);
843 		goto out;
844 	}
845 
846 	/*
847 	 * Successfully aborted the command.
848 	 */
849 	cmd->result = (DID_ABORT << 16);
850 	cmd->scsi_done(cmd);
851 
852 out:
853 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
854 	return result;
855 }
856 
857 /*
858  * Abort all outstanding requests.  This is only safe to use if the completion
859  * ring will never be walked again or the device has been reset, because it
860  * destroys the 1-1 mapping between context field passed to emulation and our
861  * request structure.
862  */
863 static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
864 {
865 	unsigned i;
866 
867 	for (i = 0; i < adapter->req_depth; i++) {
868 		struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
869 		struct scsi_cmnd *cmd = ctx->cmd;
870 		if (cmd) {
871 			scmd_printk(KERN_ERR, cmd,
872 				    "Forced reset on cmd %p\n", cmd);
873 			pvscsi_unmap_buffers(adapter, ctx);
874 			pvscsi_release_context(adapter, ctx);
875 			cmd->result = (DID_RESET << 16);
876 			cmd->scsi_done(cmd);
877 		}
878 	}
879 }
880 
881 static int pvscsi_host_reset(struct scsi_cmnd *cmd)
882 {
883 	struct Scsi_Host *host = cmd->device->host;
884 	struct pvscsi_adapter *adapter = shost_priv(host);
885 	unsigned long flags;
886 	bool use_msg;
887 
888 	scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
889 
890 	spin_lock_irqsave(&adapter->hw_lock, flags);
891 
892 	use_msg = adapter->use_msg;
893 
894 	if (use_msg) {
895 		adapter->use_msg = 0;
896 		spin_unlock_irqrestore(&adapter->hw_lock, flags);
897 
898 		/*
899 		 * Now that we know that the ISR won't add more work on the
900 		 * workqueue we can safely flush any outstanding work.
901 		 */
902 		flush_workqueue(adapter->workqueue);
903 		spin_lock_irqsave(&adapter->hw_lock, flags);
904 	}
905 
906 	/*
907 	 * We're going to tear down the entire ring structure and set it back
908 	 * up, so stalling new requests until all completions are flushed and
909 	 * the rings are back in place.
910 	 */
911 
912 	pvscsi_process_request_ring(adapter);
913 
914 	ll_adapter_reset(adapter);
915 
916 	/*
917 	 * Now process any completions.  Note we do this AFTER adapter reset,
918 	 * which is strange, but stops races where completions get posted
919 	 * between processing the ring and issuing the reset.  The backend will
920 	 * not touch the ring memory after reset, so the immediately pre-reset
921 	 * completion ring state is still valid.
922 	 */
923 	pvscsi_process_completion_ring(adapter);
924 
925 	pvscsi_reset_all(adapter);
926 	adapter->use_msg = use_msg;
927 	pvscsi_setup_all_rings(adapter);
928 	pvscsi_unmask_intr(adapter);
929 
930 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
931 
932 	return SUCCESS;
933 }
934 
935 static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
936 {
937 	struct Scsi_Host *host = cmd->device->host;
938 	struct pvscsi_adapter *adapter = shost_priv(host);
939 	unsigned long flags;
940 
941 	scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
942 
943 	/*
944 	 * We don't want to queue new requests for this bus after
945 	 * flushing all pending requests to emulation, since new
946 	 * requests could then sneak in during this bus reset phase,
947 	 * so take the lock now.
948 	 */
949 	spin_lock_irqsave(&adapter->hw_lock, flags);
950 
951 	pvscsi_process_request_ring(adapter);
952 	ll_bus_reset(adapter);
953 	pvscsi_process_completion_ring(adapter);
954 
955 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
956 
957 	return SUCCESS;
958 }
959 
960 static int pvscsi_device_reset(struct scsi_cmnd *cmd)
961 {
962 	struct Scsi_Host *host = cmd->device->host;
963 	struct pvscsi_adapter *adapter = shost_priv(host);
964 	unsigned long flags;
965 
966 	scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
967 		    host->host_no, cmd->device->id);
968 
969 	/*
970 	 * We don't want to queue new requests for this device after flushing
971 	 * all pending requests to emulation, since new requests could then
972 	 * sneak in during this device reset phase, so take the lock now.
973 	 */
974 	spin_lock_irqsave(&adapter->hw_lock, flags);
975 
976 	pvscsi_process_request_ring(adapter);
977 	ll_device_reset(adapter, cmd->device->id);
978 	pvscsi_process_completion_ring(adapter);
979 
980 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
981 
982 	return SUCCESS;
983 }
984 
985 static struct scsi_host_template pvscsi_template;
986 
987 static const char *pvscsi_info(struct Scsi_Host *host)
988 {
989 	struct pvscsi_adapter *adapter = shost_priv(host);
990 	static char buf[256];
991 
992 	sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
993 		"%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
994 		adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
995 		pvscsi_template.cmd_per_lun);
996 
997 	return buf;
998 }
999 
1000 static struct scsi_host_template pvscsi_template = {
1001 	.module				= THIS_MODULE,
1002 	.name				= "VMware PVSCSI Host Adapter",
1003 	.proc_name			= "vmw_pvscsi",
1004 	.info				= pvscsi_info,
1005 	.queuecommand			= pvscsi_queue,
1006 	.this_id			= -1,
1007 	.sg_tablesize			= PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
1008 	.dma_boundary			= UINT_MAX,
1009 	.max_sectors			= 0xffff,
1010 	.use_clustering			= ENABLE_CLUSTERING,
1011 	.change_queue_depth		= pvscsi_change_queue_depth,
1012 	.eh_abort_handler		= pvscsi_abort,
1013 	.eh_device_reset_handler	= pvscsi_device_reset,
1014 	.eh_bus_reset_handler		= pvscsi_bus_reset,
1015 	.eh_host_reset_handler		= pvscsi_host_reset,
1016 };
1017 
1018 static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
1019 			       const struct PVSCSIRingMsgDesc *e)
1020 {
1021 	struct PVSCSIRingsState *s = adapter->rings_state;
1022 	struct Scsi_Host *host = adapter->host;
1023 	struct scsi_device *sdev;
1024 
1025 	printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
1026 	       e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
1027 
1028 	BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
1029 
1030 	if (e->type == PVSCSI_MSG_DEV_ADDED) {
1031 		struct PVSCSIMsgDescDevStatusChanged *desc;
1032 		desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1033 
1034 		printk(KERN_INFO
1035 		       "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
1036 		       desc->bus, desc->target, desc->lun[1]);
1037 
1038 		if (!scsi_host_get(host))
1039 			return;
1040 
1041 		sdev = scsi_device_lookup(host, desc->bus, desc->target,
1042 					  desc->lun[1]);
1043 		if (sdev) {
1044 			printk(KERN_INFO "vmw_pvscsi: device already exists\n");
1045 			scsi_device_put(sdev);
1046 		} else
1047 			scsi_add_device(adapter->host, desc->bus,
1048 					desc->target, desc->lun[1]);
1049 
1050 		scsi_host_put(host);
1051 	} else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
1052 		struct PVSCSIMsgDescDevStatusChanged *desc;
1053 		desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1054 
1055 		printk(KERN_INFO
1056 		       "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
1057 		       desc->bus, desc->target, desc->lun[1]);
1058 
1059 		if (!scsi_host_get(host))
1060 			return;
1061 
1062 		sdev = scsi_device_lookup(host, desc->bus, desc->target,
1063 					  desc->lun[1]);
1064 		if (sdev) {
1065 			scsi_remove_device(sdev);
1066 			scsi_device_put(sdev);
1067 		} else
1068 			printk(KERN_INFO
1069 			       "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
1070 			       desc->bus, desc->target, desc->lun[1]);
1071 
1072 		scsi_host_put(host);
1073 	}
1074 }
1075 
1076 static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
1077 {
1078 	struct PVSCSIRingsState *s = adapter->rings_state;
1079 
1080 	return s->msgProdIdx != s->msgConsIdx;
1081 }
1082 
1083 static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
1084 {
1085 	struct PVSCSIRingsState *s = adapter->rings_state;
1086 	struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
1087 	u32 msg_entries = s->msgNumEntriesLog2;
1088 
1089 	while (pvscsi_msg_pending(adapter)) {
1090 		struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
1091 						      MASK(msg_entries));
1092 
1093 		barrier();
1094 		pvscsi_process_msg(adapter, e);
1095 		barrier();
1096 		s->msgConsIdx++;
1097 	}
1098 }
1099 
1100 static void pvscsi_msg_workqueue_handler(struct work_struct *data)
1101 {
1102 	struct pvscsi_adapter *adapter;
1103 
1104 	adapter = container_of(data, struct pvscsi_adapter, work);
1105 
1106 	pvscsi_process_msg_ring(adapter);
1107 }
1108 
1109 static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
1110 {
1111 	char name[32];
1112 
1113 	if (!pvscsi_use_msg)
1114 		return 0;
1115 
1116 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1117 			 PVSCSI_CMD_SETUP_MSG_RING);
1118 
1119 	if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
1120 		return 0;
1121 
1122 	snprintf(name, sizeof(name),
1123 		 "vmw_pvscsi_wq_%u", adapter->host->host_no);
1124 
1125 	adapter->workqueue = create_singlethread_workqueue(name);
1126 	if (!adapter->workqueue) {
1127 		printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
1128 		return 0;
1129 	}
1130 	INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
1131 
1132 	return 1;
1133 }
1134 
1135 static bool pvscsi_setup_req_threshold(struct pvscsi_adapter *adapter,
1136 				      bool enable)
1137 {
1138 	u32 val;
1139 
1140 	if (!pvscsi_use_req_threshold)
1141 		return false;
1142 
1143 	pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1144 			 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD);
1145 	val = pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS);
1146 	if (val == -1) {
1147 		printk(KERN_INFO "vmw_pvscsi: device does not support req_threshold\n");
1148 		return false;
1149 	} else {
1150 		struct PVSCSICmdDescSetupReqCall cmd_msg = { 0 };
1151 		cmd_msg.enable = enable;
1152 		printk(KERN_INFO
1153 		       "vmw_pvscsi: %sabling reqCallThreshold\n",
1154 			enable ? "en" : "dis");
1155 		pvscsi_write_cmd_desc(adapter,
1156 				      PVSCSI_CMD_SETUP_REQCALLTHRESHOLD,
1157 				      &cmd_msg, sizeof(cmd_msg));
1158 		return pvscsi_reg_read(adapter,
1159 				       PVSCSI_REG_OFFSET_COMMAND_STATUS) != 0;
1160 	}
1161 }
1162 
1163 static irqreturn_t pvscsi_isr(int irq, void *devp)
1164 {
1165 	struct pvscsi_adapter *adapter = devp;
1166 	unsigned long flags;
1167 
1168 	spin_lock_irqsave(&adapter->hw_lock, flags);
1169 	pvscsi_process_completion_ring(adapter);
1170 	if (adapter->use_msg && pvscsi_msg_pending(adapter))
1171 		queue_work(adapter->workqueue, &adapter->work);
1172 	spin_unlock_irqrestore(&adapter->hw_lock, flags);
1173 
1174 	return IRQ_HANDLED;
1175 }
1176 
1177 static irqreturn_t pvscsi_shared_isr(int irq, void *devp)
1178 {
1179 	struct pvscsi_adapter *adapter = devp;
1180 	u32 val = pvscsi_read_intr_status(adapter);
1181 
1182 	if (!(val & PVSCSI_INTR_ALL_SUPPORTED))
1183 		return IRQ_NONE;
1184 	pvscsi_write_intr_status(devp, val);
1185 	return pvscsi_isr(irq, devp);
1186 }
1187 
1188 static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
1189 {
1190 	struct pvscsi_ctx *ctx = adapter->cmd_map;
1191 	unsigned i;
1192 
1193 	for (i = 0; i < adapter->req_depth; ++i, ++ctx)
1194 		free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
1195 }
1196 
1197 static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
1198 {
1199 	free_irq(pci_irq_vector(adapter->dev, 0), adapter);
1200 	pci_free_irq_vectors(adapter->dev);
1201 }
1202 
1203 static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
1204 {
1205 	pvscsi_shutdown_intr(adapter);
1206 
1207 	if (adapter->workqueue)
1208 		destroy_workqueue(adapter->workqueue);
1209 
1210 	if (adapter->mmioBase)
1211 		pci_iounmap(adapter->dev, adapter->mmioBase);
1212 
1213 	pci_release_regions(adapter->dev);
1214 
1215 	if (adapter->cmd_map) {
1216 		pvscsi_free_sgls(adapter);
1217 		kfree(adapter->cmd_map);
1218 	}
1219 
1220 	if (adapter->rings_state)
1221 		pci_free_consistent(adapter->dev, PAGE_SIZE,
1222 				    adapter->rings_state, adapter->ringStatePA);
1223 
1224 	if (adapter->req_ring)
1225 		pci_free_consistent(adapter->dev,
1226 				    adapter->req_pages * PAGE_SIZE,
1227 				    adapter->req_ring, adapter->reqRingPA);
1228 
1229 	if (adapter->cmp_ring)
1230 		pci_free_consistent(adapter->dev,
1231 				    adapter->cmp_pages * PAGE_SIZE,
1232 				    adapter->cmp_ring, adapter->cmpRingPA);
1233 
1234 	if (adapter->msg_ring)
1235 		pci_free_consistent(adapter->dev,
1236 				    adapter->msg_pages * PAGE_SIZE,
1237 				    adapter->msg_ring, adapter->msgRingPA);
1238 }
1239 
1240 /*
1241  * Allocate scatter gather lists.
1242  *
1243  * These are statically allocated.  Trying to be clever was not worth it.
1244  *
1245  * Dynamic allocation can fail, and we can't go deep into the memory
1246  * allocator, since we're a SCSI driver, and trying too hard to allocate
1247  * memory might generate disk I/O.  We also don't want to fail disk I/O
1248  * in that case because we can't get an allocation - the I/O could be
1249  * trying to swap out data to free memory.  Since that is pathological,
1250  * just use a statically allocated scatter list.
1251  *
1252  */
1253 static int pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
1254 {
1255 	struct pvscsi_ctx *ctx;
1256 	int i;
1257 
1258 	ctx = adapter->cmd_map;
1259 	BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
1260 
1261 	for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
1262 		ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
1263 						    get_order(SGL_SIZE));
1264 		ctx->sglPA = 0;
1265 		BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
1266 		if (!ctx->sgl) {
1267 			for (; i >= 0; --i, --ctx) {
1268 				free_pages((unsigned long)ctx->sgl,
1269 					   get_order(SGL_SIZE));
1270 				ctx->sgl = NULL;
1271 			}
1272 			return -ENOMEM;
1273 		}
1274 	}
1275 
1276 	return 0;
1277 }
1278 
1279 /*
1280  * Query the device, fetch the config info and return the
1281  * maximum number of targets on the adapter. In case of
1282  * failure due to any reason return default i.e. 16.
1283  */
1284 static u32 pvscsi_get_max_targets(struct pvscsi_adapter *adapter)
1285 {
1286 	struct PVSCSICmdDescConfigCmd cmd;
1287 	struct PVSCSIConfigPageHeader *header;
1288 	struct device *dev;
1289 	dma_addr_t configPagePA;
1290 	void *config_page;
1291 	u32 numPhys = 16;
1292 
1293 	dev = pvscsi_dev(adapter);
1294 	config_page = pci_alloc_consistent(adapter->dev, PAGE_SIZE,
1295 					   &configPagePA);
1296 	if (!config_page) {
1297 		dev_warn(dev, "vmw_pvscsi: failed to allocate memory for config page\n");
1298 		goto exit;
1299 	}
1300 	BUG_ON(configPagePA & ~PAGE_MASK);
1301 
1302 	/* Fetch config info from the device. */
1303 	cmd.configPageAddress = ((u64)PVSCSI_CONFIG_CONTROLLER_ADDRESS) << 32;
1304 	cmd.configPageNum = PVSCSI_CONFIG_PAGE_CONTROLLER;
1305 	cmd.cmpAddr = configPagePA;
1306 	cmd._pad = 0;
1307 
1308 	/*
1309 	 * Mark the completion page header with error values. If the device
1310 	 * completes the command successfully, it sets the status values to
1311 	 * indicate success.
1312 	 */
1313 	header = config_page;
1314 	memset(header, 0, sizeof *header);
1315 	header->hostStatus = BTSTAT_INVPARAM;
1316 	header->scsiStatus = SDSTAT_CHECK;
1317 
1318 	pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_CONFIG, &cmd, sizeof cmd);
1319 
1320 	if (header->hostStatus == BTSTAT_SUCCESS &&
1321 	    header->scsiStatus == SDSTAT_GOOD) {
1322 		struct PVSCSIConfigPageController *config;
1323 
1324 		config = config_page;
1325 		numPhys = config->numPhys;
1326 	} else
1327 		dev_warn(dev, "vmw_pvscsi: PVSCSI_CMD_CONFIG failed. hostStatus = 0x%x, scsiStatus = 0x%x\n",
1328 			 header->hostStatus, header->scsiStatus);
1329 	pci_free_consistent(adapter->dev, PAGE_SIZE, config_page, configPagePA);
1330 exit:
1331 	return numPhys;
1332 }
1333 
1334 static int pvscsi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1335 {
1336 	unsigned int irq_flag = PCI_IRQ_MSIX | PCI_IRQ_MSI | PCI_IRQ_LEGACY;
1337 	struct pvscsi_adapter *adapter;
1338 	struct pvscsi_adapter adapter_temp;
1339 	struct Scsi_Host *host = NULL;
1340 	unsigned int i;
1341 	int error;
1342 	u32 max_id;
1343 
1344 	error = -ENODEV;
1345 
1346 	if (pci_enable_device(pdev))
1347 		return error;
1348 
1349 	if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0 &&
1350 	    pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1351 		printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
1352 	} else if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) == 0 &&
1353 		   pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) == 0) {
1354 		printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
1355 	} else {
1356 		printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
1357 		goto out_disable_device;
1358 	}
1359 
1360 	/*
1361 	 * Let's use a temp pvscsi_adapter struct until we find the number of
1362 	 * targets on the adapter, after that we will switch to the real
1363 	 * allocated struct.
1364 	 */
1365 	adapter = &adapter_temp;
1366 	memset(adapter, 0, sizeof(*adapter));
1367 	adapter->dev  = pdev;
1368 	adapter->rev = pdev->revision;
1369 
1370 	if (pci_request_regions(pdev, "vmw_pvscsi")) {
1371 		printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
1372 		goto out_disable_device;
1373 	}
1374 
1375 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1376 		if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
1377 			continue;
1378 
1379 		if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
1380 			continue;
1381 
1382 		break;
1383 	}
1384 
1385 	if (i == DEVICE_COUNT_RESOURCE) {
1386 		printk(KERN_ERR
1387 		       "vmw_pvscsi: adapter has no suitable MMIO region\n");
1388 		goto out_release_resources_and_disable;
1389 	}
1390 
1391 	adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
1392 
1393 	if (!adapter->mmioBase) {
1394 		printk(KERN_ERR
1395 		       "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
1396 		       i, PVSCSI_MEM_SPACE_SIZE);
1397 		goto out_release_resources_and_disable;
1398 	}
1399 
1400 	pci_set_master(pdev);
1401 
1402 	/*
1403 	 * Ask the device for max number of targets before deciding the
1404 	 * default pvscsi_ring_pages value.
1405 	 */
1406 	max_id = pvscsi_get_max_targets(adapter);
1407 	printk(KERN_INFO "vmw_pvscsi: max_id: %u\n", max_id);
1408 
1409 	if (pvscsi_ring_pages == 0)
1410 		/*
1411 		 * Set the right default value. Up to 16 it is 8, above it is
1412 		 * max.
1413 		 */
1414 		pvscsi_ring_pages = (max_id > 16) ?
1415 			PVSCSI_SETUP_RINGS_MAX_NUM_PAGES :
1416 			PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
1417 	printk(KERN_INFO
1418 	       "vmw_pvscsi: setting ring_pages to %d\n",
1419 	       pvscsi_ring_pages);
1420 
1421 	pvscsi_template.can_queue =
1422 		min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
1423 		PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
1424 	pvscsi_template.cmd_per_lun =
1425 		min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
1426 	host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
1427 	if (!host) {
1428 		printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
1429 		goto out_release_resources_and_disable;
1430 	}
1431 
1432 	/*
1433 	 * Let's use the real pvscsi_adapter struct here onwards.
1434 	 */
1435 	adapter = shost_priv(host);
1436 	memset(adapter, 0, sizeof(*adapter));
1437 	adapter->dev  = pdev;
1438 	adapter->host = host;
1439 	/*
1440 	 * Copy back what we already have to the allocated adapter struct.
1441 	 */
1442 	adapter->rev = adapter_temp.rev;
1443 	adapter->mmioBase = adapter_temp.mmioBase;
1444 
1445 	spin_lock_init(&adapter->hw_lock);
1446 	host->max_channel = 0;
1447 	host->max_lun     = 1;
1448 	host->max_cmd_len = 16;
1449 	host->max_id      = max_id;
1450 
1451 	pci_set_drvdata(pdev, host);
1452 
1453 	ll_adapter_reset(adapter);
1454 
1455 	adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
1456 
1457 	error = pvscsi_allocate_rings(adapter);
1458 	if (error) {
1459 		printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
1460 		goto out_release_resources;
1461 	}
1462 
1463 	/*
1464 	 * From this point on we should reset the adapter if anything goes
1465 	 * wrong.
1466 	 */
1467 	pvscsi_setup_all_rings(adapter);
1468 
1469 	adapter->cmd_map = kcalloc(adapter->req_depth,
1470 				   sizeof(struct pvscsi_ctx), GFP_KERNEL);
1471 	if (!adapter->cmd_map) {
1472 		printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
1473 		error = -ENOMEM;
1474 		goto out_reset_adapter;
1475 	}
1476 
1477 	INIT_LIST_HEAD(&adapter->cmd_pool);
1478 	for (i = 0; i < adapter->req_depth; i++) {
1479 		struct pvscsi_ctx *ctx = adapter->cmd_map + i;
1480 		list_add(&ctx->list, &adapter->cmd_pool);
1481 	}
1482 
1483 	error = pvscsi_allocate_sg(adapter);
1484 	if (error) {
1485 		printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
1486 		goto out_reset_adapter;
1487 	}
1488 
1489 	if (pvscsi_disable_msix)
1490 		irq_flag &= ~PCI_IRQ_MSIX;
1491 	if (pvscsi_disable_msi)
1492 		irq_flag &= ~PCI_IRQ_MSI;
1493 
1494 	error = pci_alloc_irq_vectors(adapter->dev, 1, 1, irq_flag);
1495 	if (error < 0)
1496 		goto out_reset_adapter;
1497 
1498 	adapter->use_req_threshold = pvscsi_setup_req_threshold(adapter, true);
1499 	printk(KERN_DEBUG "vmw_pvscsi: driver-based request coalescing %sabled\n",
1500 	       adapter->use_req_threshold ? "en" : "dis");
1501 
1502 	if (adapter->dev->msix_enabled || adapter->dev->msi_enabled) {
1503 		printk(KERN_INFO "vmw_pvscsi: using MSI%s\n",
1504 			adapter->dev->msix_enabled ? "-X" : "");
1505 		error = request_irq(pci_irq_vector(pdev, 0), pvscsi_isr,
1506 				0, "vmw_pvscsi", adapter);
1507 	} else {
1508 		printk(KERN_INFO "vmw_pvscsi: using INTx\n");
1509 		error = request_irq(pci_irq_vector(pdev, 0), pvscsi_shared_isr,
1510 				IRQF_SHARED, "vmw_pvscsi", adapter);
1511 	}
1512 
1513 	if (error) {
1514 		printk(KERN_ERR
1515 		       "vmw_pvscsi: unable to request IRQ: %d\n", error);
1516 		goto out_reset_adapter;
1517 	}
1518 
1519 	error = scsi_add_host(host, &pdev->dev);
1520 	if (error) {
1521 		printk(KERN_ERR
1522 		       "vmw_pvscsi: scsi_add_host failed: %d\n", error);
1523 		goto out_reset_adapter;
1524 	}
1525 
1526 	dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
1527 		 adapter->rev, host->host_no);
1528 
1529 	pvscsi_unmask_intr(adapter);
1530 
1531 	scsi_scan_host(host);
1532 
1533 	return 0;
1534 
1535 out_reset_adapter:
1536 	ll_adapter_reset(adapter);
1537 out_release_resources:
1538 	pvscsi_release_resources(adapter);
1539 	scsi_host_put(host);
1540 out_disable_device:
1541 	pci_disable_device(pdev);
1542 
1543 	return error;
1544 
1545 out_release_resources_and_disable:
1546 	pvscsi_release_resources(adapter);
1547 	goto out_disable_device;
1548 }
1549 
1550 static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
1551 {
1552 	pvscsi_mask_intr(adapter);
1553 
1554 	if (adapter->workqueue)
1555 		flush_workqueue(adapter->workqueue);
1556 
1557 	pvscsi_shutdown_intr(adapter);
1558 
1559 	pvscsi_process_request_ring(adapter);
1560 	pvscsi_process_completion_ring(adapter);
1561 	ll_adapter_reset(adapter);
1562 }
1563 
1564 static void pvscsi_shutdown(struct pci_dev *dev)
1565 {
1566 	struct Scsi_Host *host = pci_get_drvdata(dev);
1567 	struct pvscsi_adapter *adapter = shost_priv(host);
1568 
1569 	__pvscsi_shutdown(adapter);
1570 }
1571 
1572 static void pvscsi_remove(struct pci_dev *pdev)
1573 {
1574 	struct Scsi_Host *host = pci_get_drvdata(pdev);
1575 	struct pvscsi_adapter *adapter = shost_priv(host);
1576 
1577 	scsi_remove_host(host);
1578 
1579 	__pvscsi_shutdown(adapter);
1580 	pvscsi_release_resources(adapter);
1581 
1582 	scsi_host_put(host);
1583 
1584 	pci_disable_device(pdev);
1585 }
1586 
1587 static struct pci_driver pvscsi_pci_driver = {
1588 	.name		= "vmw_pvscsi",
1589 	.id_table	= pvscsi_pci_tbl,
1590 	.probe		= pvscsi_probe,
1591 	.remove		= pvscsi_remove,
1592 	.shutdown       = pvscsi_shutdown,
1593 };
1594 
1595 static int __init pvscsi_init(void)
1596 {
1597 	pr_info("%s - version %s\n",
1598 		PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
1599 	return pci_register_driver(&pvscsi_pci_driver);
1600 }
1601 
1602 static void __exit pvscsi_exit(void)
1603 {
1604 	pci_unregister_driver(&pvscsi_pci_driver);
1605 }
1606 
1607 module_init(pvscsi_init);
1608 module_exit(pvscsi_exit);
1609