2 * Linux driver for VMware's para-virtualized SCSI HBA.
4 * Copyright (C) 2008-2014, VMware, Inc. All Rights Reserved.
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.
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
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.
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/slab.h>
26 #include <linux/workqueue.h>
27 #include <linux/pci.h>
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_host.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_tcq.h>
35 #include "vmw_pvscsi.h"
37 #define PVSCSI_LINUX_DRIVER_DESC "VMware PVSCSI driver"
39 MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC);
40 MODULE_AUTHOR("VMware, Inc.");
41 MODULE_LICENSE("GPL");
42 MODULE_VERSION(PVSCSI_DRIVER_VERSION_STRING);
44 #define PVSCSI_DEFAULT_NUM_PAGES_PER_RING 8
45 #define PVSCSI_DEFAULT_NUM_PAGES_MSG_RING 1
46 #define PVSCSI_DEFAULT_QUEUE_DEPTH 254
47 #define SGL_SIZE PAGE_SIZE
49 struct pvscsi_sg_list {
50 struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT];
55 * The index of the context in cmd_map serves as the context ID for a
56 * 1-to-1 mapping completions back to requests.
58 struct scsi_cmnd *cmd;
59 struct pvscsi_sg_list *sgl;
60 struct list_head list;
64 struct completion *abort_cmp;
67 struct pvscsi_adapter {
71 bool use_req_threshold;
75 struct workqueue_struct *workqueue;
76 struct work_struct work;
78 struct PVSCSIRingReqDesc *req_ring;
83 struct PVSCSIRingCmpDesc *cmp_ring;
87 struct PVSCSIRingMsgDesc *msg_ring;
91 struct PVSCSIRingsState *rings_state;
92 dma_addr_t ringStatePA;
95 struct Scsi_Host *host;
97 struct list_head cmd_pool;
98 struct pvscsi_ctx *cmd_map;
102 /* Command line parameters */
103 static int pvscsi_ring_pages;
104 static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING;
105 static int pvscsi_cmd_per_lun = PVSCSI_DEFAULT_QUEUE_DEPTH;
106 static bool pvscsi_disable_msi;
107 static bool pvscsi_disable_msix;
108 static bool pvscsi_use_msg = true;
109 static bool pvscsi_use_req_threshold = true;
111 #define PVSCSI_RW (S_IRUSR | S_IWUSR)
113 module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW);
114 MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default="
115 __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING)
116 "[up to 16 targets],"
117 __stringify(PVSCSI_SETUP_RINGS_MAX_NUM_PAGES)
118 "[for 16+ targets])");
120 module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW);
121 MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default="
122 __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")");
124 module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW);
125 MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default="
126 __stringify(PVSCSI_DEFAULT_QUEUE_DEPTH) ")");
128 module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW);
129 MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)");
131 module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW);
132 MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)");
134 module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW);
135 MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)");
137 module_param_named(use_req_threshold, pvscsi_use_req_threshold,
139 MODULE_PARM_DESC(use_req_threshold, "Use driver-based request coalescing if configured - (default=1)");
141 static const struct pci_device_id pvscsi_pci_tbl[] = {
142 { PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) },
146 MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl);
148 static struct device *
149 pvscsi_dev(const struct pvscsi_adapter *adapter)
151 return &(adapter->dev->dev);
154 static struct pvscsi_ctx *
155 pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
157 struct pvscsi_ctx *ctx, *end;
159 end = &adapter->cmd_map[adapter->req_depth];
160 for (ctx = adapter->cmd_map; ctx < end; ctx++)
167 static struct pvscsi_ctx *
168 pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd)
170 struct pvscsi_ctx *ctx;
172 if (list_empty(&adapter->cmd_pool))
175 ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list);
177 list_del(&ctx->list);
182 static void pvscsi_release_context(struct pvscsi_adapter *adapter,
183 struct pvscsi_ctx *ctx)
186 ctx->abort_cmp = NULL;
187 list_add(&ctx->list, &adapter->cmd_pool);
191 * Map a pvscsi_ctx struct to a context ID field value; we map to a simple
192 * non-zero integer. ctx always points to an entry in cmd_map array, hence
193 * the return value is always >=1.
195 static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter,
196 const struct pvscsi_ctx *ctx)
198 return ctx - adapter->cmd_map + 1;
201 static struct pvscsi_ctx *
202 pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context)
204 return &adapter->cmd_map[context - 1];
207 static void pvscsi_reg_write(const struct pvscsi_adapter *adapter,
210 writel(val, adapter->mmioBase + offset);
213 static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset)
215 return readl(adapter->mmioBase + offset);
218 static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter)
220 return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS);
223 static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter,
226 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val);
229 static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter)
233 intr_bits = PVSCSI_INTR_CMPL_MASK;
234 if (adapter->use_msg)
235 intr_bits |= PVSCSI_INTR_MSG_MASK;
237 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits);
240 static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter)
242 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0);
245 static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter,
246 u32 cmd, const void *desc, size_t len)
248 const u32 *ptr = desc;
252 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd);
253 for (i = 0; i < len; i++)
254 pvscsi_reg_write(adapter,
255 PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
258 static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter,
259 const struct pvscsi_ctx *ctx)
261 struct PVSCSICmdDescAbortCmd cmd = { 0 };
263 cmd.target = ctx->cmd->device->id;
264 cmd.context = pvscsi_map_context(adapter, ctx);
266 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd));
269 static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter)
271 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
274 static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter)
276 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0);
279 static int scsi_is_rw(unsigned char op)
281 return op == READ_6 || op == WRITE_6 ||
282 op == READ_10 || op == WRITE_10 ||
283 op == READ_12 || op == WRITE_12 ||
284 op == READ_16 || op == WRITE_16;
287 static void pvscsi_kick_io(const struct pvscsi_adapter *adapter,
290 if (scsi_is_rw(op)) {
291 struct PVSCSIRingsState *s = adapter->rings_state;
293 if (!adapter->use_req_threshold ||
294 s->reqProdIdx - s->reqConsIdx >= s->reqCallThreshold)
295 pvscsi_kick_rw_io(adapter);
297 pvscsi_process_request_ring(adapter);
301 static void ll_adapter_reset(const struct pvscsi_adapter *adapter)
303 dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter);
305 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0);
308 static void ll_bus_reset(const struct pvscsi_adapter *adapter)
310 dev_dbg(pvscsi_dev(adapter), "Resetting bus on %p\n", adapter);
312 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0);
315 static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target)
317 struct PVSCSICmdDescResetDevice cmd = { 0 };
319 dev_dbg(pvscsi_dev(adapter), "Resetting device: target=%u\n", target);
323 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE,
327 static void pvscsi_create_sg(struct pvscsi_ctx *ctx,
328 struct scatterlist *sg, unsigned count)
331 struct PVSCSISGElement *sge;
333 BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT);
335 sge = &ctx->sgl->sge[0];
336 for (i = 0; i < count; i++, sg = sg_next(sg)) {
337 sge[i].addr = sg_dma_address(sg);
338 sge[i].length = sg_dma_len(sg);
344 * Map all data buffers for a command into PCI space and
345 * setup the scatter/gather list if needed.
347 static int pvscsi_map_buffers(struct pvscsi_adapter *adapter,
348 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd,
349 struct PVSCSIRingReqDesc *e)
352 unsigned bufflen = scsi_bufflen(cmd);
353 struct scatterlist *sg;
355 e->dataLen = bufflen;
360 sg = scsi_sglist(cmd);
361 count = scsi_sg_count(cmd);
363 int segs = scsi_dma_map(cmd);
365 if (segs == -ENOMEM) {
366 scmd_printk(KERN_DEBUG, cmd,
367 "vmw_pvscsi: Failed to map cmd sglist for DMA.\n");
369 } else if (segs > 1) {
370 pvscsi_create_sg(ctx, sg, segs);
372 e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST;
373 ctx->sglPA = dma_map_single(&adapter->dev->dev,
374 ctx->sgl, SGL_SIZE, DMA_TO_DEVICE);
375 if (dma_mapping_error(&adapter->dev->dev, ctx->sglPA)) {
376 scmd_printk(KERN_ERR, cmd,
377 "vmw_pvscsi: Failed to map ctx sglist for DMA.\n");
382 e->dataAddr = ctx->sglPA;
384 e->dataAddr = sg_dma_address(sg);
387 * In case there is no S/G list, scsi_sglist points
388 * directly to the buffer.
390 ctx->dataPA = dma_map_single(&adapter->dev->dev, sg, bufflen,
391 cmd->sc_data_direction);
392 if (dma_mapping_error(&adapter->dev->dev, ctx->dataPA)) {
393 scmd_printk(KERN_DEBUG, cmd,
394 "vmw_pvscsi: Failed to map direct data buffer for DMA.\n");
397 e->dataAddr = ctx->dataPA;
404 * The device incorrectly doesn't clear the first byte of the sense
405 * buffer in some cases. We have to do it ourselves.
406 * Otherwise we run into trouble when SWIOTLB is forced.
408 static void pvscsi_patch_sense(struct scsi_cmnd *cmd)
410 if (cmd->sense_buffer)
411 cmd->sense_buffer[0] = 0;
414 static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
415 struct pvscsi_ctx *ctx)
417 struct scsi_cmnd *cmd;
421 bufflen = scsi_bufflen(cmd);
424 unsigned count = scsi_sg_count(cmd);
429 dma_unmap_single(&adapter->dev->dev, ctx->sglPA,
430 SGL_SIZE, DMA_TO_DEVICE);
434 dma_unmap_single(&adapter->dev->dev, ctx->dataPA,
435 bufflen, cmd->sc_data_direction);
437 if (cmd->sense_buffer)
438 dma_unmap_single(&adapter->dev->dev, ctx->sensePA,
439 SCSI_SENSE_BUFFERSIZE, DMA_FROM_DEVICE);
442 static int pvscsi_allocate_rings(struct pvscsi_adapter *adapter)
444 adapter->rings_state = dma_alloc_coherent(&adapter->dev->dev, PAGE_SIZE,
445 &adapter->ringStatePA, GFP_KERNEL);
446 if (!adapter->rings_state)
449 adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING,
451 adapter->req_depth = adapter->req_pages
452 * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
453 adapter->req_ring = dma_alloc_coherent(&adapter->dev->dev,
454 adapter->req_pages * PAGE_SIZE, &adapter->reqRingPA,
456 if (!adapter->req_ring)
459 adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING,
461 adapter->cmp_ring = dma_alloc_coherent(&adapter->dev->dev,
462 adapter->cmp_pages * PAGE_SIZE, &adapter->cmpRingPA,
464 if (!adapter->cmp_ring)
467 BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE));
468 BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE));
469 BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE));
471 if (!adapter->use_msg)
474 adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING,
475 pvscsi_msg_ring_pages);
476 adapter->msg_ring = dma_alloc_coherent(&adapter->dev->dev,
477 adapter->msg_pages * PAGE_SIZE, &adapter->msgRingPA,
479 if (!adapter->msg_ring)
481 BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE));
486 static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter)
488 struct PVSCSICmdDescSetupRings cmd = { 0 };
492 cmd.ringsStatePPN = adapter->ringStatePA >> PAGE_SHIFT;
493 cmd.reqRingNumPages = adapter->req_pages;
494 cmd.cmpRingNumPages = adapter->cmp_pages;
496 base = adapter->reqRingPA;
497 for (i = 0; i < adapter->req_pages; i++) {
498 cmd.reqRingPPNs[i] = base >> PAGE_SHIFT;
502 base = adapter->cmpRingPA;
503 for (i = 0; i < adapter->cmp_pages; i++) {
504 cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT;
508 memset(adapter->rings_state, 0, PAGE_SIZE);
509 memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE);
510 memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE);
512 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS,
515 if (adapter->use_msg) {
516 struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 };
518 cmd_msg.numPages = adapter->msg_pages;
520 base = adapter->msgRingPA;
521 for (i = 0; i < adapter->msg_pages; i++) {
522 cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT;
525 memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE);
527 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING,
528 &cmd_msg, sizeof(cmd_msg));
532 static int pvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
534 if (!sdev->tagged_supported)
536 return scsi_change_queue_depth(sdev, qdepth);
540 * Pull a completion descriptor off and pass the completion back
541 * to the SCSI mid layer.
543 static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
544 const struct PVSCSIRingCmpDesc *e)
546 struct pvscsi_ctx *ctx;
547 struct scsi_cmnd *cmd;
548 struct completion *abort_cmp;
549 u32 btstat = e->hostStatus;
550 u32 sdstat = e->scsiStatus;
552 ctx = pvscsi_get_context(adapter, e->context);
554 abort_cmp = ctx->abort_cmp;
555 pvscsi_unmap_buffers(adapter, ctx);
556 if (sdstat != SAM_STAT_CHECK_CONDITION)
557 pvscsi_patch_sense(cmd);
558 pvscsi_release_context(adapter, ctx);
561 * The command was requested to be aborted. Just signal that
562 * the request completed and swallow the actual cmd completion
563 * here. The abort handler will post a completion for this
564 * command indicating that it got successfully aborted.
571 if (sdstat != SAM_STAT_GOOD &&
572 (btstat == BTSTAT_SUCCESS ||
573 btstat == BTSTAT_LINKED_COMMAND_COMPLETED ||
574 btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) {
575 if (sdstat == SAM_STAT_COMMAND_TERMINATED) {
576 cmd->result = (DID_RESET << 16);
578 cmd->result = (DID_OK << 16) | sdstat;
579 if (sdstat == SAM_STAT_CHECK_CONDITION &&
581 cmd->result |= (DRIVER_SENSE << 24);
586 case BTSTAT_LINKED_COMMAND_COMPLETED:
587 case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG:
589 * Commands like INQUIRY may transfer less data than
590 * requested by the initiator via bufflen. Set residual
591 * count to make upper layer aware of the actual amount
594 scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
595 cmd->result = (DID_OK << 16);
599 case BTSTAT_DATA_UNDERRUN:
600 /* Report residual data in underruns */
601 scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen);
602 cmd->result = (DID_ERROR << 16);
605 case BTSTAT_SELTIMEO:
606 /* Our emulation returns this for non-connected devs */
607 cmd->result = (DID_BAD_TARGET << 16);
610 case BTSTAT_LUNMISMATCH:
611 case BTSTAT_TAGREJECT:
613 cmd->result = (DRIVER_INVALID << 24);
616 case BTSTAT_HAHARDWARE:
617 case BTSTAT_INVPHASE:
618 case BTSTAT_HATIMEOUT:
619 case BTSTAT_NORESPONSE:
620 case BTSTAT_DISCONNECT:
621 case BTSTAT_HASOFTWARE:
623 case BTSTAT_SENSFAILED:
624 cmd->result |= (DID_ERROR << 16);
629 case BTSTAT_BUSRESET:
630 cmd->result = (DID_RESET << 16);
633 case BTSTAT_ABORTQUEUE:
634 cmd->result = (DID_BUS_BUSY << 16);
637 case BTSTAT_SCSIPARITY:
638 cmd->result = (DID_PARITY << 16);
642 cmd->result = (DID_ERROR << 16);
643 scmd_printk(KERN_DEBUG, cmd,
644 "Unknown completion status: 0x%x\n",
648 dev_dbg(&cmd->device->sdev_gendev,
649 "cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n",
650 cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat);
656 * barrier usage : Since the PVSCSI device is emulated, there could be cases
657 * where we may want to serialize some accesses between the driver and the
658 * emulation layer. We use compiler barriers instead of the more expensive
659 * memory barriers because PVSCSI is only supported on X86 which has strong
660 * memory access ordering.
662 static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter)
664 struct PVSCSIRingsState *s = adapter->rings_state;
665 struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring;
666 u32 cmp_entries = s->cmpNumEntriesLog2;
668 while (s->cmpConsIdx != s->cmpProdIdx) {
669 struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx &
672 * This barrier() ensures that *e is not dereferenced while
673 * the device emulation still writes data into the slot.
674 * Since the device emulation advances s->cmpProdIdx only after
675 * updating the slot we want to check it first.
678 pvscsi_complete_request(adapter, e);
680 * This barrier() ensures that compiler doesn't reorder write
681 * to s->cmpConsIdx before the read of (*e) inside
682 * pvscsi_complete_request. Otherwise, device emulation may
683 * overwrite *e before we had a chance to read it.
691 * Translate a Linux SCSI request into a request ring entry.
693 static int pvscsi_queue_ring(struct pvscsi_adapter *adapter,
694 struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd)
696 struct PVSCSIRingsState *s;
697 struct PVSCSIRingReqDesc *e;
698 struct scsi_device *sdev;
701 s = adapter->rings_state;
703 req_entries = s->reqNumEntriesLog2;
706 * If this condition holds, we might have room on the request ring, but
707 * we might not have room on the completion ring for the response.
708 * However, we have already ruled out this possibility - we would not
709 * have successfully allocated a context if it were true, since we only
710 * have one context per request entry. Check for it anyway, since it
711 * would be a serious bug.
713 if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) {
714 scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: "
715 "ring full: reqProdIdx=%d cmpConsIdx=%d\n",
716 s->reqProdIdx, s->cmpConsIdx);
720 e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries));
722 e->bus = sdev->channel;
723 e->target = sdev->id;
724 memset(e->lun, 0, sizeof(e->lun));
725 e->lun[1] = sdev->lun;
727 if (cmd->sense_buffer) {
728 ctx->sensePA = dma_map_single(&adapter->dev->dev,
729 cmd->sense_buffer, SCSI_SENSE_BUFFERSIZE,
731 if (dma_mapping_error(&adapter->dev->dev, ctx->sensePA)) {
732 scmd_printk(KERN_DEBUG, cmd,
733 "vmw_pvscsi: Failed to map sense buffer for DMA.\n");
737 e->senseAddr = ctx->sensePA;
738 e->senseLen = SCSI_SENSE_BUFFERSIZE;
743 e->cdbLen = cmd->cmd_len;
744 e->vcpuHint = smp_processor_id();
745 memcpy(e->cdb, cmd->cmnd, e->cdbLen);
747 e->tag = SIMPLE_QUEUE_TAG;
749 if (cmd->sc_data_direction == DMA_FROM_DEVICE)
750 e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST;
751 else if (cmd->sc_data_direction == DMA_TO_DEVICE)
752 e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE;
753 else if (cmd->sc_data_direction == DMA_NONE)
754 e->flags = PVSCSI_FLAG_CMD_DIR_NONE;
758 if (pvscsi_map_buffers(adapter, ctx, cmd, e) != 0) {
759 if (cmd->sense_buffer) {
760 dma_unmap_single(&adapter->dev->dev, ctx->sensePA,
761 SCSI_SENSE_BUFFERSIZE,
768 e->context = pvscsi_map_context(adapter, ctx);
777 static int pvscsi_queue_lck(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
779 struct Scsi_Host *host = cmd->device->host;
780 struct pvscsi_adapter *adapter = shost_priv(host);
781 struct pvscsi_ctx *ctx;
785 spin_lock_irqsave(&adapter->hw_lock, flags);
787 ctx = pvscsi_acquire_context(adapter, cmd);
788 if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) {
790 pvscsi_release_context(adapter, ctx);
791 spin_unlock_irqrestore(&adapter->hw_lock, flags);
792 return SCSI_MLQUEUE_HOST_BUSY;
795 cmd->scsi_done = done;
798 dev_dbg(&cmd->device->sdev_gendev,
799 "queued cmd %p, ctx %p, op=%x\n", cmd, ctx, op);
801 spin_unlock_irqrestore(&adapter->hw_lock, flags);
803 pvscsi_kick_io(adapter, op);
808 static DEF_SCSI_QCMD(pvscsi_queue)
810 static int pvscsi_abort(struct scsi_cmnd *cmd)
812 struct pvscsi_adapter *adapter = shost_priv(cmd->device->host);
813 struct pvscsi_ctx *ctx;
815 int result = SUCCESS;
816 DECLARE_COMPLETION_ONSTACK(abort_cmp);
819 scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n",
820 adapter->host->host_no, cmd);
822 spin_lock_irqsave(&adapter->hw_lock, flags);
825 * Poll the completion ring first - we might be trying to abort
826 * a command that is waiting to be dispatched in the completion ring.
828 pvscsi_process_completion_ring(adapter);
831 * If there is no context for the command, it either already succeeded
832 * or else was never properly issued. Not our problem.
834 ctx = pvscsi_find_context(adapter, cmd);
836 scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd);
841 * Mark that the command has been requested to be aborted and issue
844 ctx->abort_cmp = &abort_cmp;
846 pvscsi_abort_cmd(adapter, ctx);
847 spin_unlock_irqrestore(&adapter->hw_lock, flags);
848 /* Wait for 2 secs for the completion. */
849 done = wait_for_completion_timeout(&abort_cmp, msecs_to_jiffies(2000));
850 spin_lock_irqsave(&adapter->hw_lock, flags);
854 * Failed to abort the command, unmark the fact that it
855 * was requested to be aborted.
857 ctx->abort_cmp = NULL;
859 scmd_printk(KERN_DEBUG, cmd,
860 "Failed to get completion for aborted cmd %p\n",
866 * Successfully aborted the command.
868 cmd->result = (DID_ABORT << 16);
872 spin_unlock_irqrestore(&adapter->hw_lock, flags);
877 * Abort all outstanding requests. This is only safe to use if the completion
878 * ring will never be walked again or the device has been reset, because it
879 * destroys the 1-1 mapping between context field passed to emulation and our
882 static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
886 for (i = 0; i < adapter->req_depth; i++) {
887 struct pvscsi_ctx *ctx = &adapter->cmd_map[i];
888 struct scsi_cmnd *cmd = ctx->cmd;
890 scmd_printk(KERN_ERR, cmd,
891 "Forced reset on cmd %p\n", cmd);
892 pvscsi_unmap_buffers(adapter, ctx);
893 pvscsi_patch_sense(cmd);
894 pvscsi_release_context(adapter, ctx);
895 cmd->result = (DID_RESET << 16);
901 static int pvscsi_host_reset(struct scsi_cmnd *cmd)
903 struct Scsi_Host *host = cmd->device->host;
904 struct pvscsi_adapter *adapter = shost_priv(host);
908 scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n");
910 spin_lock_irqsave(&adapter->hw_lock, flags);
912 use_msg = adapter->use_msg;
915 adapter->use_msg = false;
916 spin_unlock_irqrestore(&adapter->hw_lock, flags);
919 * Now that we know that the ISR won't add more work on the
920 * workqueue we can safely flush any outstanding work.
922 flush_workqueue(adapter->workqueue);
923 spin_lock_irqsave(&adapter->hw_lock, flags);
927 * We're going to tear down the entire ring structure and set it back
928 * up, so stalling new requests until all completions are flushed and
929 * the rings are back in place.
932 pvscsi_process_request_ring(adapter);
934 ll_adapter_reset(adapter);
937 * Now process any completions. Note we do this AFTER adapter reset,
938 * which is strange, but stops races where completions get posted
939 * between processing the ring and issuing the reset. The backend will
940 * not touch the ring memory after reset, so the immediately pre-reset
941 * completion ring state is still valid.
943 pvscsi_process_completion_ring(adapter);
945 pvscsi_reset_all(adapter);
946 adapter->use_msg = use_msg;
947 pvscsi_setup_all_rings(adapter);
948 pvscsi_unmask_intr(adapter);
950 spin_unlock_irqrestore(&adapter->hw_lock, flags);
955 static int pvscsi_bus_reset(struct scsi_cmnd *cmd)
957 struct Scsi_Host *host = cmd->device->host;
958 struct pvscsi_adapter *adapter = shost_priv(host);
961 scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n");
964 * We don't want to queue new requests for this bus after
965 * flushing all pending requests to emulation, since new
966 * requests could then sneak in during this bus reset phase,
967 * so take the lock now.
969 spin_lock_irqsave(&adapter->hw_lock, flags);
971 pvscsi_process_request_ring(adapter);
972 ll_bus_reset(adapter);
973 pvscsi_process_completion_ring(adapter);
975 spin_unlock_irqrestore(&adapter->hw_lock, flags);
980 static int pvscsi_device_reset(struct scsi_cmnd *cmd)
982 struct Scsi_Host *host = cmd->device->host;
983 struct pvscsi_adapter *adapter = shost_priv(host);
986 scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n",
987 host->host_no, cmd->device->id);
990 * We don't want to queue new requests for this device after flushing
991 * all pending requests to emulation, since new requests could then
992 * sneak in during this device reset phase, so take the lock now.
994 spin_lock_irqsave(&adapter->hw_lock, flags);
996 pvscsi_process_request_ring(adapter);
997 ll_device_reset(adapter, cmd->device->id);
998 pvscsi_process_completion_ring(adapter);
1000 spin_unlock_irqrestore(&adapter->hw_lock, flags);
1005 static struct scsi_host_template pvscsi_template;
1007 static const char *pvscsi_info(struct Scsi_Host *host)
1009 struct pvscsi_adapter *adapter = shost_priv(host);
1010 static char buf[256];
1012 sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: "
1013 "%u/%u/%u pages, cmd_per_lun=%u", adapter->rev,
1014 adapter->req_pages, adapter->cmp_pages, adapter->msg_pages,
1015 pvscsi_template.cmd_per_lun);
1020 static struct scsi_host_template pvscsi_template = {
1021 .module = THIS_MODULE,
1022 .name = "VMware PVSCSI Host Adapter",
1023 .proc_name = "vmw_pvscsi",
1024 .info = pvscsi_info,
1025 .queuecommand = pvscsi_queue,
1027 .sg_tablesize = PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT,
1028 .dma_boundary = UINT_MAX,
1029 .max_sectors = 0xffff,
1030 .change_queue_depth = pvscsi_change_queue_depth,
1031 .eh_abort_handler = pvscsi_abort,
1032 .eh_device_reset_handler = pvscsi_device_reset,
1033 .eh_bus_reset_handler = pvscsi_bus_reset,
1034 .eh_host_reset_handler = pvscsi_host_reset,
1037 static void pvscsi_process_msg(const struct pvscsi_adapter *adapter,
1038 const struct PVSCSIRingMsgDesc *e)
1040 struct PVSCSIRingsState *s = adapter->rings_state;
1041 struct Scsi_Host *host = adapter->host;
1042 struct scsi_device *sdev;
1044 printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n",
1045 e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2);
1047 BUILD_BUG_ON(PVSCSI_MSG_LAST != 2);
1049 if (e->type == PVSCSI_MSG_DEV_ADDED) {
1050 struct PVSCSIMsgDescDevStatusChanged *desc;
1051 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1054 "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n",
1055 desc->bus, desc->target, desc->lun[1]);
1057 if (!scsi_host_get(host))
1060 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1063 printk(KERN_INFO "vmw_pvscsi: device already exists\n");
1064 scsi_device_put(sdev);
1066 scsi_add_device(adapter->host, desc->bus,
1067 desc->target, desc->lun[1]);
1069 scsi_host_put(host);
1070 } else if (e->type == PVSCSI_MSG_DEV_REMOVED) {
1071 struct PVSCSIMsgDescDevStatusChanged *desc;
1072 desc = (struct PVSCSIMsgDescDevStatusChanged *)e;
1075 "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n",
1076 desc->bus, desc->target, desc->lun[1]);
1078 if (!scsi_host_get(host))
1081 sdev = scsi_device_lookup(host, desc->bus, desc->target,
1084 scsi_remove_device(sdev);
1085 scsi_device_put(sdev);
1088 "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n",
1089 desc->bus, desc->target, desc->lun[1]);
1091 scsi_host_put(host);
1095 static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter)
1097 struct PVSCSIRingsState *s = adapter->rings_state;
1099 return s->msgProdIdx != s->msgConsIdx;
1102 static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter)
1104 struct PVSCSIRingsState *s = adapter->rings_state;
1105 struct PVSCSIRingMsgDesc *ring = adapter->msg_ring;
1106 u32 msg_entries = s->msgNumEntriesLog2;
1108 while (pvscsi_msg_pending(adapter)) {
1109 struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx &
1113 pvscsi_process_msg(adapter, e);
1119 static void pvscsi_msg_workqueue_handler(struct work_struct *data)
1121 struct pvscsi_adapter *adapter;
1123 adapter = container_of(data, struct pvscsi_adapter, work);
1125 pvscsi_process_msg_ring(adapter);
1128 static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter)
1132 if (!pvscsi_use_msg)
1135 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1136 PVSCSI_CMD_SETUP_MSG_RING);
1138 if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1)
1141 snprintf(name, sizeof(name),
1142 "vmw_pvscsi_wq_%u", adapter->host->host_no);
1144 adapter->workqueue = create_singlethread_workqueue(name);
1145 if (!adapter->workqueue) {
1146 printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n");
1149 INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler);
1154 static bool pvscsi_setup_req_threshold(struct pvscsi_adapter *adapter,
1159 if (!pvscsi_use_req_threshold)
1162 pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND,
1163 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD);
1164 val = pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS);
1166 printk(KERN_INFO "vmw_pvscsi: device does not support req_threshold\n");
1169 struct PVSCSICmdDescSetupReqCall cmd_msg = { 0 };
1170 cmd_msg.enable = enable;
1172 "vmw_pvscsi: %sabling reqCallThreshold\n",
1173 enable ? "en" : "dis");
1174 pvscsi_write_cmd_desc(adapter,
1175 PVSCSI_CMD_SETUP_REQCALLTHRESHOLD,
1176 &cmd_msg, sizeof(cmd_msg));
1177 return pvscsi_reg_read(adapter,
1178 PVSCSI_REG_OFFSET_COMMAND_STATUS) != 0;
1182 static irqreturn_t pvscsi_isr(int irq, void *devp)
1184 struct pvscsi_adapter *adapter = devp;
1185 unsigned long flags;
1187 spin_lock_irqsave(&adapter->hw_lock, flags);
1188 pvscsi_process_completion_ring(adapter);
1189 if (adapter->use_msg && pvscsi_msg_pending(adapter))
1190 queue_work(adapter->workqueue, &adapter->work);
1191 spin_unlock_irqrestore(&adapter->hw_lock, flags);
1196 static irqreturn_t pvscsi_shared_isr(int irq, void *devp)
1198 struct pvscsi_adapter *adapter = devp;
1199 u32 val = pvscsi_read_intr_status(adapter);
1201 if (!(val & PVSCSI_INTR_ALL_SUPPORTED))
1203 pvscsi_write_intr_status(devp, val);
1204 return pvscsi_isr(irq, devp);
1207 static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter)
1209 struct pvscsi_ctx *ctx = adapter->cmd_map;
1212 for (i = 0; i < adapter->req_depth; ++i, ++ctx)
1213 free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE));
1216 static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter)
1218 free_irq(pci_irq_vector(adapter->dev, 0), adapter);
1219 pci_free_irq_vectors(adapter->dev);
1222 static void pvscsi_release_resources(struct pvscsi_adapter *adapter)
1224 if (adapter->workqueue)
1225 destroy_workqueue(adapter->workqueue);
1227 if (adapter->mmioBase)
1228 pci_iounmap(adapter->dev, adapter->mmioBase);
1230 pci_release_regions(adapter->dev);
1232 if (adapter->cmd_map) {
1233 pvscsi_free_sgls(adapter);
1234 kfree(adapter->cmd_map);
1237 if (adapter->rings_state)
1238 dma_free_coherent(&adapter->dev->dev, PAGE_SIZE,
1239 adapter->rings_state, adapter->ringStatePA);
1241 if (adapter->req_ring)
1242 dma_free_coherent(&adapter->dev->dev,
1243 adapter->req_pages * PAGE_SIZE,
1244 adapter->req_ring, adapter->reqRingPA);
1246 if (adapter->cmp_ring)
1247 dma_free_coherent(&adapter->dev->dev,
1248 adapter->cmp_pages * PAGE_SIZE,
1249 adapter->cmp_ring, adapter->cmpRingPA);
1251 if (adapter->msg_ring)
1252 dma_free_coherent(&adapter->dev->dev,
1253 adapter->msg_pages * PAGE_SIZE,
1254 adapter->msg_ring, adapter->msgRingPA);
1258 * Allocate scatter gather lists.
1260 * These are statically allocated. Trying to be clever was not worth it.
1262 * Dynamic allocation can fail, and we can't go deep into the memory
1263 * allocator, since we're a SCSI driver, and trying too hard to allocate
1264 * memory might generate disk I/O. We also don't want to fail disk I/O
1265 * in that case because we can't get an allocation - the I/O could be
1266 * trying to swap out data to free memory. Since that is pathological,
1267 * just use a statically allocated scatter list.
1270 static int pvscsi_allocate_sg(struct pvscsi_adapter *adapter)
1272 struct pvscsi_ctx *ctx;
1275 ctx = adapter->cmd_map;
1276 BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE);
1278 for (i = 0; i < adapter->req_depth; ++i, ++ctx) {
1279 ctx->sgl = (void *)__get_free_pages(GFP_KERNEL,
1280 get_order(SGL_SIZE));
1282 BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE));
1284 for (; i >= 0; --i, --ctx) {
1285 free_pages((unsigned long)ctx->sgl,
1286 get_order(SGL_SIZE));
1297 * Query the device, fetch the config info and return the
1298 * maximum number of targets on the adapter. In case of
1299 * failure due to any reason return default i.e. 16.
1301 static u32 pvscsi_get_max_targets(struct pvscsi_adapter *adapter)
1303 struct PVSCSICmdDescConfigCmd cmd;
1304 struct PVSCSIConfigPageHeader *header;
1306 dma_addr_t configPagePA;
1310 dev = pvscsi_dev(adapter);
1311 config_page = dma_alloc_coherent(&adapter->dev->dev, PAGE_SIZE,
1312 &configPagePA, GFP_KERNEL);
1314 dev_warn(dev, "vmw_pvscsi: failed to allocate memory for config page\n");
1317 BUG_ON(configPagePA & ~PAGE_MASK);
1319 /* Fetch config info from the device. */
1320 cmd.configPageAddress = ((u64)PVSCSI_CONFIG_CONTROLLER_ADDRESS) << 32;
1321 cmd.configPageNum = PVSCSI_CONFIG_PAGE_CONTROLLER;
1322 cmd.cmpAddr = configPagePA;
1326 * Mark the completion page header with error values. If the device
1327 * completes the command successfully, it sets the status values to
1330 header = config_page;
1331 memset(header, 0, sizeof *header);
1332 header->hostStatus = BTSTAT_INVPARAM;
1333 header->scsiStatus = SDSTAT_CHECK;
1335 pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_CONFIG, &cmd, sizeof cmd);
1337 if (header->hostStatus == BTSTAT_SUCCESS &&
1338 header->scsiStatus == SDSTAT_GOOD) {
1339 struct PVSCSIConfigPageController *config;
1341 config = config_page;
1342 numPhys = config->numPhys;
1344 dev_warn(dev, "vmw_pvscsi: PVSCSI_CMD_CONFIG failed. hostStatus = 0x%x, scsiStatus = 0x%x\n",
1345 header->hostStatus, header->scsiStatus);
1346 dma_free_coherent(&adapter->dev->dev, PAGE_SIZE, config_page,
1352 static int pvscsi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1354 unsigned int irq_flag = PCI_IRQ_MSIX | PCI_IRQ_MSI | PCI_IRQ_LEGACY;
1355 struct pvscsi_adapter *adapter;
1356 struct pvscsi_adapter adapter_temp;
1357 struct Scsi_Host *host = NULL;
1364 if (pci_enable_device(pdev))
1367 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
1368 printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n");
1369 } else if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32))) {
1370 printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n");
1372 printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n");
1373 goto out_disable_device;
1377 * Let's use a temp pvscsi_adapter struct until we find the number of
1378 * targets on the adapter, after that we will switch to the real
1381 adapter = &adapter_temp;
1382 memset(adapter, 0, sizeof(*adapter));
1383 adapter->dev = pdev;
1384 adapter->rev = pdev->revision;
1386 if (pci_request_regions(pdev, "vmw_pvscsi")) {
1387 printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n");
1388 goto out_disable_device;
1391 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
1392 if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO))
1395 if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE)
1401 if (i == DEVICE_COUNT_RESOURCE) {
1403 "vmw_pvscsi: adapter has no suitable MMIO region\n");
1404 goto out_release_resources_and_disable;
1407 adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE);
1409 if (!adapter->mmioBase) {
1411 "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n",
1412 i, PVSCSI_MEM_SPACE_SIZE);
1413 goto out_release_resources_and_disable;
1416 pci_set_master(pdev);
1419 * Ask the device for max number of targets before deciding the
1420 * default pvscsi_ring_pages value.
1422 max_id = pvscsi_get_max_targets(adapter);
1423 printk(KERN_INFO "vmw_pvscsi: max_id: %u\n", max_id);
1425 if (pvscsi_ring_pages == 0)
1427 * Set the right default value. Up to 16 it is 8, above it is
1430 pvscsi_ring_pages = (max_id > 16) ?
1431 PVSCSI_SETUP_RINGS_MAX_NUM_PAGES :
1432 PVSCSI_DEFAULT_NUM_PAGES_PER_RING;
1434 "vmw_pvscsi: setting ring_pages to %d\n",
1437 pvscsi_template.can_queue =
1438 min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) *
1439 PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE;
1440 pvscsi_template.cmd_per_lun =
1441 min(pvscsi_template.can_queue, pvscsi_cmd_per_lun);
1442 host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter));
1444 printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n");
1445 goto out_release_resources_and_disable;
1449 * Let's use the real pvscsi_adapter struct here onwards.
1451 adapter = shost_priv(host);
1452 memset(adapter, 0, sizeof(*adapter));
1453 adapter->dev = pdev;
1454 adapter->host = host;
1456 * Copy back what we already have to the allocated adapter struct.
1458 adapter->rev = adapter_temp.rev;
1459 adapter->mmioBase = adapter_temp.mmioBase;
1461 spin_lock_init(&adapter->hw_lock);
1462 host->max_channel = 0;
1464 host->max_cmd_len = 16;
1465 host->max_id = max_id;
1467 pci_set_drvdata(pdev, host);
1469 ll_adapter_reset(adapter);
1471 adapter->use_msg = pvscsi_setup_msg_workqueue(adapter);
1473 error = pvscsi_allocate_rings(adapter);
1475 printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n");
1476 goto out_release_resources;
1480 * From this point on we should reset the adapter if anything goes
1483 pvscsi_setup_all_rings(adapter);
1485 adapter->cmd_map = kcalloc(adapter->req_depth,
1486 sizeof(struct pvscsi_ctx), GFP_KERNEL);
1487 if (!adapter->cmd_map) {
1488 printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n");
1490 goto out_reset_adapter;
1493 INIT_LIST_HEAD(&adapter->cmd_pool);
1494 for (i = 0; i < adapter->req_depth; i++) {
1495 struct pvscsi_ctx *ctx = adapter->cmd_map + i;
1496 list_add(&ctx->list, &adapter->cmd_pool);
1499 error = pvscsi_allocate_sg(adapter);
1501 printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n");
1502 goto out_reset_adapter;
1505 if (pvscsi_disable_msix)
1506 irq_flag &= ~PCI_IRQ_MSIX;
1507 if (pvscsi_disable_msi)
1508 irq_flag &= ~PCI_IRQ_MSI;
1510 error = pci_alloc_irq_vectors(adapter->dev, 1, 1, irq_flag);
1512 goto out_reset_adapter;
1514 adapter->use_req_threshold = pvscsi_setup_req_threshold(adapter, true);
1515 printk(KERN_DEBUG "vmw_pvscsi: driver-based request coalescing %sabled\n",
1516 adapter->use_req_threshold ? "en" : "dis");
1518 if (adapter->dev->msix_enabled || adapter->dev->msi_enabled) {
1519 printk(KERN_INFO "vmw_pvscsi: using MSI%s\n",
1520 adapter->dev->msix_enabled ? "-X" : "");
1521 error = request_irq(pci_irq_vector(pdev, 0), pvscsi_isr,
1522 0, "vmw_pvscsi", adapter);
1524 printk(KERN_INFO "vmw_pvscsi: using INTx\n");
1525 error = request_irq(pci_irq_vector(pdev, 0), pvscsi_shared_isr,
1526 IRQF_SHARED, "vmw_pvscsi", adapter);
1531 "vmw_pvscsi: unable to request IRQ: %d\n", error);
1532 goto out_reset_adapter;
1535 error = scsi_add_host(host, &pdev->dev);
1538 "vmw_pvscsi: scsi_add_host failed: %d\n", error);
1539 goto out_reset_adapter;
1542 dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n",
1543 adapter->rev, host->host_no);
1545 pvscsi_unmask_intr(adapter);
1547 scsi_scan_host(host);
1552 ll_adapter_reset(adapter);
1553 out_release_resources:
1554 pvscsi_shutdown_intr(adapter);
1555 pvscsi_release_resources(adapter);
1556 scsi_host_put(host);
1558 pci_disable_device(pdev);
1562 out_release_resources_and_disable:
1563 pvscsi_shutdown_intr(adapter);
1564 pvscsi_release_resources(adapter);
1565 goto out_disable_device;
1568 static void __pvscsi_shutdown(struct pvscsi_adapter *adapter)
1570 pvscsi_mask_intr(adapter);
1572 if (adapter->workqueue)
1573 flush_workqueue(adapter->workqueue);
1575 pvscsi_shutdown_intr(adapter);
1577 pvscsi_process_request_ring(adapter);
1578 pvscsi_process_completion_ring(adapter);
1579 ll_adapter_reset(adapter);
1582 static void pvscsi_shutdown(struct pci_dev *dev)
1584 struct Scsi_Host *host = pci_get_drvdata(dev);
1585 struct pvscsi_adapter *adapter = shost_priv(host);
1587 __pvscsi_shutdown(adapter);
1590 static void pvscsi_remove(struct pci_dev *pdev)
1592 struct Scsi_Host *host = pci_get_drvdata(pdev);
1593 struct pvscsi_adapter *adapter = shost_priv(host);
1595 scsi_remove_host(host);
1597 __pvscsi_shutdown(adapter);
1598 pvscsi_release_resources(adapter);
1600 scsi_host_put(host);
1602 pci_disable_device(pdev);
1605 static struct pci_driver pvscsi_pci_driver = {
1606 .name = "vmw_pvscsi",
1607 .id_table = pvscsi_pci_tbl,
1608 .probe = pvscsi_probe,
1609 .remove = pvscsi_remove,
1610 .shutdown = pvscsi_shutdown,
1613 static int __init pvscsi_init(void)
1615 pr_info("%s - version %s\n",
1616 PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING);
1617 return pci_register_driver(&pvscsi_pci_driver);
1620 static void __exit pvscsi_exit(void)
1622 pci_unregister_driver(&pvscsi_pci_driver);
1625 module_init(pvscsi_init);
1626 module_exit(pvscsi_exit);