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