2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
20 * K. Y. Srinivasan <kys@microsoft.com>
22 #include <linux/init.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <linux/device.h>
26 #include <linux/blkdev.h>
27 #include <linux/dmi.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_host.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_tcq.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_devinfo.h>
35 #include <scsi/scsi_dbg.h>
38 #include "hyperv_storage.h"
40 static int storvsc_ringbuffer_size = STORVSC_RING_BUFFER_SIZE;
42 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
43 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
45 static const char *driver_name = "storvsc";
47 /* {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f} */
48 static const struct hv_guid stor_vsci_device_type = {
50 0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
51 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f
55 struct hv_host_device {
56 struct hv_device *dev;
57 struct kmem_cache *request_pool;
63 struct storvsc_cmd_request {
64 struct list_head entry;
65 struct scsi_cmnd *cmd;
67 unsigned int bounce_sgl_count;
68 struct scatterlist *bounce_sgl;
70 struct hv_storvsc_request request;
74 static int storvsc_device_alloc(struct scsi_device *sdevice)
77 * This enables luns to be located sparsely. Otherwise, we may not
80 sdevice->sdev_bflags |= BLIST_SPARSELUN | BLIST_LARGELUN;
84 static int storvsc_merge_bvec(struct request_queue *q,
85 struct bvec_merge_data *bmd, struct bio_vec *bvec)
87 /* checking done by caller. */
91 static int storvsc_device_configure(struct scsi_device *sdevice)
93 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
94 STORVSC_MAX_IO_REQUESTS);
96 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
98 blk_queue_merge_bvec(sdevice->request_queue, storvsc_merge_bvec);
100 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
105 static void destroy_bounce_buffer(struct scatterlist *sgl,
106 unsigned int sg_count)
109 struct page *page_buf;
111 for (i = 0; i < sg_count; i++) {
112 page_buf = sg_page((&sgl[i]));
113 if (page_buf != NULL)
114 __free_page(page_buf);
120 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
124 /* No need to check */
128 /* We have at least 2 sg entries */
129 for (i = 0; i < sg_count; i++) {
131 /* make sure 1st one does not have hole */
132 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
134 } else if (i == sg_count - 1) {
135 /* make sure last one does not have hole */
136 if (sgl[i].offset != 0)
139 /* make sure no hole in the middle */
140 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
147 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
148 unsigned int sg_count,
153 struct scatterlist *bounce_sgl;
154 struct page *page_buf;
156 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
158 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
162 for (i = 0; i < num_pages; i++) {
163 page_buf = alloc_page(GFP_ATOMIC);
166 sg_set_page(&bounce_sgl[i], page_buf, 0, 0);
172 destroy_bounce_buffer(bounce_sgl, num_pages);
177 /* Assume the original sgl has enough room */
178 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
179 struct scatterlist *bounce_sgl,
180 unsigned int orig_sgl_count)
184 unsigned long src, dest;
185 unsigned int srclen, destlen, copylen;
186 unsigned int total_copied = 0;
187 unsigned long bounce_addr = 0;
188 unsigned long dest_addr = 0;
191 local_irq_save(flags);
193 for (i = 0; i < orig_sgl_count; i++) {
194 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
195 KM_IRQ0) + orig_sgl[i].offset;
197 destlen = orig_sgl[i].length;
199 if (bounce_addr == 0)
201 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
205 src = bounce_addr + bounce_sgl[j].offset;
206 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
208 copylen = min(srclen, destlen);
209 memcpy((void *)dest, (void *)src, copylen);
211 total_copied += copylen;
212 bounce_sgl[j].offset += copylen;
216 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
218 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
221 /* if we need to use another bounce buffer */
222 if (destlen || i != orig_sgl_count - 1)
224 (unsigned long)kmap_atomic(
225 sg_page((&bounce_sgl[j])), KM_IRQ0);
226 } else if (destlen == 0 && i == orig_sgl_count - 1) {
227 /* unmap the last bounce that is < PAGE_SIZE */
228 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
232 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
236 local_irq_restore(flags);
242 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
243 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
244 struct scatterlist *bounce_sgl,
245 unsigned int orig_sgl_count)
249 unsigned long src, dest;
250 unsigned int srclen, destlen, copylen;
251 unsigned int total_copied = 0;
252 unsigned long bounce_addr = 0;
253 unsigned long src_addr = 0;
256 local_irq_save(flags);
258 for (i = 0; i < orig_sgl_count; i++) {
259 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
260 KM_IRQ0) + orig_sgl[i].offset;
262 srclen = orig_sgl[i].length;
264 if (bounce_addr == 0)
266 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
270 /* assume bounce offset always == 0 */
271 dest = bounce_addr + bounce_sgl[j].length;
272 destlen = PAGE_SIZE - bounce_sgl[j].length;
274 copylen = min(srclen, destlen);
275 memcpy((void *)dest, (void *)src, copylen);
277 total_copied += copylen;
278 bounce_sgl[j].length += copylen;
282 if (bounce_sgl[j].length == PAGE_SIZE) {
283 /* full..move to next entry */
284 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
287 /* if we need to use another bounce buffer */
288 if (srclen || i != orig_sgl_count - 1)
290 (unsigned long)kmap_atomic(
291 sg_page((&bounce_sgl[j])), KM_IRQ0);
293 } else if (srclen == 0 && i == orig_sgl_count - 1) {
294 /* unmap the last bounce that is < PAGE_SIZE */
295 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
299 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
302 local_irq_restore(flags);
308 static int storvsc_remove(struct hv_device *dev)
310 struct Scsi_Host *host = dev_get_drvdata(&dev->device);
311 struct hv_host_device *host_dev =
312 (struct hv_host_device *)host->hostdata;
314 scsi_remove_host(host);
318 storvsc_dev_remove(dev);
319 if (host_dev->request_pool) {
320 kmem_cache_destroy(host_dev->request_pool);
321 host_dev->request_pool = NULL;
327 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
328 sector_t capacity, int *info)
330 sector_t nsect = capacity;
331 sector_t cylinders = nsect;
332 int heads, sectors_pt;
335 * We are making up these values; let us keep it simple.
338 sectors_pt = 0x3f; /* Sectors per track */
339 sector_div(cylinders, heads * sectors_pt);
340 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
344 info[1] = sectors_pt;
345 info[2] = (int)cylinders;
350 static int storvsc_host_reset(struct hv_device *device)
352 struct storvsc_device *stor_device;
353 struct hv_storvsc_request *request;
354 struct vstor_packet *vstor_packet;
358 stor_device = get_stor_device(device);
362 request = &stor_device->reset_request;
363 vstor_packet = &request->vstor_packet;
365 init_completion(&request->wait_event);
367 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
368 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
369 vstor_packet->vm_srb.path_id = stor_device->path_id;
371 ret = vmbus_sendpacket(device->channel, vstor_packet,
372 sizeof(struct vstor_packet),
373 (unsigned long)&stor_device->reset_request,
375 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
379 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
387 * At this point, all outstanding requests in the adapter
388 * should have been flushed out and return to us
392 put_stor_device(device);
398 * storvsc_host_reset_handler - Reset the scsi HBA
400 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
403 struct hv_host_device *host_dev =
404 (struct hv_host_device *)scmnd->device->host->hostdata;
405 struct hv_device *dev = host_dev->dev;
407 ret = storvsc_host_reset(dev);
416 * storvsc_commmand_completion - Command completion processing
418 static void storvsc_commmand_completion(struct hv_storvsc_request *request)
420 struct storvsc_cmd_request *cmd_request =
421 (struct storvsc_cmd_request *)request->context;
422 struct scsi_cmnd *scmnd = cmd_request->cmd;
423 struct hv_host_device *host_dev =
424 (struct hv_host_device *)scmnd->device->host->hostdata;
425 void (*scsi_done_fn)(struct scsi_cmnd *);
426 struct scsi_sense_hdr sense_hdr;
427 struct vmscsi_request *vm_srb;
429 if (cmd_request->bounce_sgl_count) {
431 /* FIXME: We can optimize on writes by just skipping this */
432 copy_from_bounce_buffer(scsi_sglist(scmnd),
433 cmd_request->bounce_sgl,
434 scsi_sg_count(scmnd));
435 destroy_bounce_buffer(cmd_request->bounce_sgl,
436 cmd_request->bounce_sgl_count);
439 vm_srb = &request->vstor_packet.vm_srb;
440 scmnd->result = vm_srb->scsi_status;
443 if (scsi_normalize_sense(scmnd->sense_buffer,
444 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
445 scsi_print_sense_hdr("storvsc", &sense_hdr);
448 scsi_set_resid(scmnd,
449 request->data_buffer.len -
450 vm_srb->data_transfer_length);
452 scsi_done_fn = scmnd->scsi_done;
454 scmnd->host_scribble = NULL;
455 scmnd->scsi_done = NULL;
459 kmem_cache_free(host_dev->request_pool, cmd_request);
464 * storvsc_queuecommand - Initiate command processing
466 static int storvsc_queuecommand_lck(struct scsi_cmnd *scmnd,
467 void (*done)(struct scsi_cmnd *))
470 struct hv_host_device *host_dev =
471 (struct hv_host_device *)scmnd->device->host->hostdata;
472 struct hv_device *dev = host_dev->dev;
473 struct hv_storvsc_request *request;
474 struct storvsc_cmd_request *cmd_request;
475 unsigned int request_size = 0;
477 struct scatterlist *sgl;
478 unsigned int sg_count = 0;
479 struct vmscsi_request *vm_srb;
482 /* If retrying, no need to prep the cmd */
483 if (scmnd->host_scribble) {
486 (struct storvsc_cmd_request *)scmnd->host_scribble;
491 scmnd->scsi_done = done;
493 request_size = sizeof(struct storvsc_cmd_request);
495 cmd_request = kmem_cache_zalloc(host_dev->request_pool,
498 scmnd->scsi_done = NULL;
499 return SCSI_MLQUEUE_DEVICE_BUSY;
502 /* Setup the cmd request */
503 cmd_request->bounce_sgl_count = 0;
504 cmd_request->bounce_sgl = NULL;
505 cmd_request->cmd = scmnd;
507 scmnd->host_scribble = (unsigned char *)cmd_request;
509 request = &cmd_request->request;
510 vm_srb = &request->vstor_packet.vm_srb;
514 switch (scmnd->sc_data_direction) {
516 vm_srb->data_in = WRITE_TYPE;
518 case DMA_FROM_DEVICE:
519 vm_srb->data_in = READ_TYPE;
522 vm_srb->data_in = UNKNOWN_TYPE;
526 request->on_io_completion = storvsc_commmand_completion;
527 request->context = cmd_request;/* scmnd; */
529 vm_srb->port_number = host_dev->port;
530 vm_srb->path_id = scmnd->device->channel;
531 vm_srb->target_id = scmnd->device->id;
532 vm_srb->lun = scmnd->device->lun;
534 vm_srb->cdb_length = scmnd->cmd_len;
536 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
538 request->sense_buffer = scmnd->sense_buffer;
541 request->data_buffer.len = scsi_bufflen(scmnd);
542 if (scsi_sg_count(scmnd)) {
543 sgl = (struct scatterlist *)scsi_sglist(scmnd);
544 sg_count = scsi_sg_count(scmnd);
546 /* check if we need to bounce the sgl */
547 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
548 cmd_request->bounce_sgl =
549 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
550 scsi_bufflen(scmnd));
551 if (!cmd_request->bounce_sgl) {
552 scmnd->scsi_done = NULL;
553 scmnd->host_scribble = NULL;
554 kmem_cache_free(host_dev->request_pool,
557 return SCSI_MLQUEUE_HOST_BUSY;
560 cmd_request->bounce_sgl_count =
561 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
565 * FIXME: We can optimize on reads by just skipping
568 copy_to_bounce_buffer(sgl, cmd_request->bounce_sgl,
569 scsi_sg_count(scmnd));
571 sgl = cmd_request->bounce_sgl;
572 sg_count = cmd_request->bounce_sgl_count;
575 request->data_buffer.offset = sgl[0].offset;
577 for (i = 0; i < sg_count; i++)
578 request->data_buffer.pfn_array[i] =
579 page_to_pfn(sg_page((&sgl[i])));
581 } else if (scsi_sglist(scmnd)) {
582 request->data_buffer.offset =
583 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
584 request->data_buffer.pfn_array[0] =
585 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
589 /* Invokes the vsc to start an IO */
590 ret = storvsc_do_io(dev, &cmd_request->request);
595 if (cmd_request->bounce_sgl_count) {
597 * FIXME: We can optimize on writes by just skipping
600 copy_from_bounce_buffer(scsi_sglist(scmnd),
601 cmd_request->bounce_sgl,
602 scsi_sg_count(scmnd));
603 destroy_bounce_buffer(cmd_request->bounce_sgl,
604 cmd_request->bounce_sgl_count);
607 kmem_cache_free(host_dev->request_pool, cmd_request);
609 scmnd->scsi_done = NULL;
610 scmnd->host_scribble = NULL;
612 ret = SCSI_MLQUEUE_DEVICE_BUSY;
618 static DEF_SCSI_QCMD(storvsc_queuecommand)
622 static struct scsi_host_template scsi_driver = {
623 .module = THIS_MODULE,
624 .name = "storvsc_host_t",
625 .bios_param = storvsc_get_chs,
626 .queuecommand = storvsc_queuecommand,
627 .eh_host_reset_handler = storvsc_host_reset_handler,
628 .slave_alloc = storvsc_device_alloc,
629 .slave_configure = storvsc_device_configure,
631 /* 64 max_queue * 1 target */
632 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
634 /* no use setting to 0 since ll_blk_rw reset it to 1 */
636 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
638 * ENABLE_CLUSTERING allows mutiple physically contig bio_vecs to merge
639 * into 1 sg element. If set, we must limit the max_segment_size to
640 * PAGE_SIZE, otherwise we may get 1 sg element that represents
643 /* physically contig pfns (ie sg[x].length > PAGE_SIZE). */
644 .use_clustering = ENABLE_CLUSTERING,
645 /* Make sure we dont get a sg segment crosses a page boundary */
646 .dma_boundary = PAGE_SIZE-1,
651 * storvsc_probe - Add a new device for this driver
654 static int storvsc_probe(struct hv_device *device)
657 struct Scsi_Host *host;
658 struct hv_host_device *host_dev;
659 struct storvsc_device_info device_info;
661 host = scsi_host_alloc(&scsi_driver,
662 sizeof(struct hv_host_device));
666 dev_set_drvdata(&device->device, host);
668 host_dev = (struct hv_host_device *)host->hostdata;
669 memset(host_dev, 0, sizeof(struct hv_host_device));
671 host_dev->port = host->host_no;
672 host_dev->dev = device;
674 host_dev->request_pool =
675 kmem_cache_create(dev_name(&device->device),
676 sizeof(struct storvsc_cmd_request), 0,
677 SLAB_HWCACHE_ALIGN, NULL);
679 if (!host_dev->request_pool) {
684 device_info.port_number = host->host_no;
685 device_info.ring_buffer_size = storvsc_ringbuffer_size;
686 /* Call to the vsc driver to add the device */
687 ret = storvsc_dev_add(device, (void *)&device_info);
690 kmem_cache_destroy(host_dev->request_pool);
695 host_dev->path = device_info.path_id;
696 host_dev->target = device_info.target_id;
698 /* max # of devices per target */
699 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
700 /* max # of targets per channel */
701 host->max_id = STORVSC_MAX_TARGETS;
702 /* max # of channels */
703 host->max_channel = STORVSC_MAX_CHANNELS - 1;
705 /* Register the HBA and start the scsi bus scan */
706 ret = scsi_add_host(host, &device->device);
709 storvsc_dev_remove(device);
711 kmem_cache_destroy(host_dev->request_pool);
716 scsi_scan_host(host);
720 /* The one and only one */
722 static struct hv_driver storvsc_drv = {
723 .probe = storvsc_probe,
724 .remove = storvsc_remove,
728 * We use a DMI table to determine if we should autoload this driver This is
729 * needed by distro tools to determine if the hyperv drivers should be
730 * installed and/or configured. We don't do anything else with the table, but
731 * it needs to be present.
734 static const struct dmi_system_id __initconst
735 hv_stor_dmi_table[] __maybe_unused = {
739 DMI_MATCH(DMI_SYS_VENDOR, "Microsoft Corporation"),
740 DMI_MATCH(DMI_PRODUCT_NAME, "Virtual Machine"),
741 DMI_MATCH(DMI_BOARD_NAME, "Virtual Machine"),
746 MODULE_DEVICE_TABLE(dmi, hv_stor_dmi_table);
748 static int __init storvsc_drv_init(void)
751 struct hv_driver *drv = &storvsc_drv;
752 u32 max_outstanding_req_per_channel;
755 * Divide the ring buffer data size (which is 1 page less
756 * than the ring buffer size since that page is reserved for
757 * the ring buffer indices) by the max request size (which is
758 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
761 max_outstanding_req_per_channel =
762 ((storvsc_ringbuffer_size - PAGE_SIZE) /
763 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
764 sizeof(struct vstor_packet) + sizeof(u64),
767 memcpy(&drv->dev_type, &stor_vsci_device_type,
768 sizeof(struct hv_guid));
770 if (max_outstanding_req_per_channel <
771 STORVSC_MAX_IO_REQUESTS)
774 drv->driver.name = driver_name;
777 /* The driver belongs to vmbus */
778 ret = vmbus_child_driver_register(&drv->driver);
783 static void __exit storvsc_drv_exit(void)
785 vmbus_child_driver_unregister(&storvsc_drv.driver);
788 MODULE_LICENSE("GPL");
789 MODULE_VERSION(HV_DRV_VERSION);
790 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
791 module_init(storvsc_drv_init);
792 module_exit(storvsc_drv_exit);