3 * Authors: Dave Boutcher <boutcher@us.ibm.com>
4 * Ryan Arnold <ryanarn@us.ibm.com>
5 * Colin Devilbiss <devilbis@us.ibm.com>
8 * (C) Copyright 2000-2004 IBM Corporation
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License as
12 * published by the Free Software Foundation; either version 2 of the
13 * License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 * This routine provides access to disk space (termed "DASD" in historical
25 * IBM terms) owned and managed by an OS/400 partition running on the
26 * same box as this Linux partition.
28 * All disk operations are performed by sending messages back and forth to
29 * the OS/400 partition.
32 #define pr_fmt(fmt) "viod: " fmt
34 #include <linux/major.h>
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/blkdev.h>
39 #include <linux/genhd.h>
40 #include <linux/hdreg.h>
41 #include <linux/errno.h>
42 #include <linux/init.h>
43 #include <linux/string.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/completion.h>
46 #include <linux/device.h>
47 #include <linux/scatterlist.h>
49 #include <asm/uaccess.h>
51 #include <asm/iseries/hv_types.h>
52 #include <asm/iseries/hv_lp_event.h>
53 #include <asm/iseries/hv_lp_config.h>
54 #include <asm/iseries/vio.h>
55 #include <asm/firmware.h>
57 MODULE_DESCRIPTION("iSeries Virtual DASD");
58 MODULE_AUTHOR("Dave Boutcher");
59 MODULE_LICENSE("GPL");
62 * We only support 7 partitions per physical disk....so with minor
63 * numbers 0-255 we get a maximum of 32 disks.
65 #define VIOD_GENHD_NAME "iseries/vd"
67 #define VIOD_VERS "1.64"
71 MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS,
72 MAX_DISK_NAME = FIELD_SIZEOF(struct gendisk, disk_name)
75 static DEFINE_SPINLOCK(viodasd_spinlock);
79 #define DEVICE_NO(cell) ((struct viodasd_device *)(cell) - &viodasd_devices[0])
81 struct viodasd_waitevent {
82 struct completion com;
85 int max_disk; /* open */
88 static const struct vio_error_entry viodasd_err_table[] = {
89 { 0x0201, EINVAL, "Invalid Range" },
90 { 0x0202, EINVAL, "Invalid Token" },
91 { 0x0203, EIO, "DMA Error" },
92 { 0x0204, EIO, "Use Error" },
93 { 0x0205, EIO, "Release Error" },
94 { 0x0206, EINVAL, "Invalid Disk" },
95 { 0x0207, EBUSY, "Cant Lock" },
96 { 0x0208, EIO, "Already Locked" },
97 { 0x0209, EIO, "Already Unlocked" },
98 { 0x020A, EIO, "Invalid Arg" },
99 { 0x020B, EIO, "Bad IFS File" },
100 { 0x020C, EROFS, "Read Only Device" },
101 { 0x02FF, EIO, "Internal Error" },
106 * Figure out the biggest I/O request (in sectors) we can accept
108 #define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA)
111 * Number of disk I/O requests we've sent to OS/400
113 static int num_req_outstanding;
116 * This is our internal structure for keeping track of disk devices
118 struct viodasd_device {
122 u16 bytes_per_sector;
126 struct gendisk *disk;
128 } viodasd_devices[MAX_DISKNO];
131 * External open entry point.
133 static int viodasd_open(struct block_device *bdev, fmode_t mode)
135 struct viodasd_device *d = bdev->bd_disk->private_data;
137 struct viodasd_waitevent we;
141 if (mode & FMODE_WRITE)
143 flags = vioblockflags_ro;
146 init_completion(&we.com);
148 /* Send the open event to OS/400 */
149 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
150 HvLpEvent_Type_VirtualIo,
151 viomajorsubtype_blockio | vioblockopen,
152 HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
153 viopath_sourceinst(viopath_hostLp),
154 viopath_targetinst(viopath_hostLp),
155 (u64)(unsigned long)&we, VIOVERSION << 16,
156 ((u64)DEVICE_NO(d) << 48) | ((u64)flags << 32),
159 pr_warning("HV open failed %d\n", (int)hvrc);
163 wait_for_completion(&we.com);
165 /* Check the return code */
167 const struct vio_error_entry *err =
168 vio_lookup_rc(viodasd_err_table, we.sub_result);
170 pr_warning("bad rc opening disk: %d:0x%04x (%s)\n",
171 (int)we.rc, we.sub_result, err->msg);
179 * External release entry point.
181 static int viodasd_release(struct gendisk *disk, fmode_t mode)
183 struct viodasd_device *d = disk->private_data;
186 /* Send the event to OS/400. We DON'T expect a response */
187 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
188 HvLpEvent_Type_VirtualIo,
189 viomajorsubtype_blockio | vioblockclose,
190 HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
191 viopath_sourceinst(viopath_hostLp),
192 viopath_targetinst(viopath_hostLp),
194 ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */,
197 pr_warning("HV close call failed %d\n", (int)hvrc);
202 /* External ioctl entry point.
204 static int viodasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
206 struct gendisk *disk = bdev->bd_disk;
207 struct viodasd_device *d = disk->private_data;
209 geo->sectors = d->sectors ? d->sectors : 32;
210 geo->heads = d->tracks ? d->tracks : 64;
211 geo->cylinders = d->cylinders ? d->cylinders :
212 get_capacity(disk) / (geo->sectors * geo->heads);
218 * Our file operations table
220 static const struct block_device_operations viodasd_fops = {
221 .owner = THIS_MODULE,
222 .open = viodasd_open,
223 .release = viodasd_release,
224 .getgeo = viodasd_getgeo,
230 static void viodasd_end_request(struct request *req, int error,
233 __blk_end_request(req, error, num_sectors << 9);
237 * Send an actual I/O request to OS/400
239 static int send_request(struct request *req)
246 struct vioblocklpevent *bevent;
247 struct HvLpEvent *hev;
248 struct scatterlist sg[VIOMAXBLOCKDMA];
250 struct viodasd_device *d;
253 start = (u64)blk_rq_pos(req) << 9;
255 if (rq_data_dir(req) == READ) {
256 direction = DMA_FROM_DEVICE;
257 viocmd = viomajorsubtype_blockio | vioblockread;
259 direction = DMA_TO_DEVICE;
260 viocmd = viomajorsubtype_blockio | vioblockwrite;
263 d = req->rq_disk->private_data;
265 /* Now build the scatter-gather list */
266 sg_init_table(sg, VIOMAXBLOCKDMA);
267 nsg = blk_rq_map_sg(req->q, req, sg);
268 nsg = dma_map_sg(d->dev, sg, nsg, direction);
270 spin_lock_irqsave(&viodasd_spinlock, flags);
271 num_req_outstanding++;
273 /* This optimization handles a single DMA block */
275 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
276 HvLpEvent_Type_VirtualIo, viocmd,
277 HvLpEvent_AckInd_DoAck,
278 HvLpEvent_AckType_ImmediateAck,
279 viopath_sourceinst(viopath_hostLp),
280 viopath_targetinst(viopath_hostLp),
281 (u64)(unsigned long)req, VIOVERSION << 16,
282 ((u64)DEVICE_NO(d) << 48), start,
283 ((u64)sg_dma_address(&sg[0])) << 32,
286 bevent = (struct vioblocklpevent *)
287 vio_get_event_buffer(viomajorsubtype_blockio);
288 if (bevent == NULL) {
289 pr_warning("error allocating disk event buffer\n");
294 * Now build up the actual request. Note that we store
295 * the pointer to the request in the correlation
296 * token so we can match the response up later
298 memset(bevent, 0, sizeof(struct vioblocklpevent));
299 hev = &bevent->event;
300 hev->flags = HV_LP_EVENT_VALID | HV_LP_EVENT_DO_ACK |
302 hev->xType = HvLpEvent_Type_VirtualIo;
303 hev->xSubtype = viocmd;
304 hev->xSourceLp = HvLpConfig_getLpIndex();
305 hev->xTargetLp = viopath_hostLp;
307 offsetof(struct vioblocklpevent, u.rw_data.dma_info) +
308 (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1;
309 hev->xSourceInstanceId = viopath_sourceinst(viopath_hostLp);
310 hev->xTargetInstanceId = viopath_targetinst(viopath_hostLp);
311 hev->xCorrelationToken = (u64)req;
312 bevent->version = VIOVERSION;
313 bevent->disk = DEVICE_NO(d);
314 bevent->u.rw_data.offset = start;
317 * Copy just the dma information from the sg list
320 for (sgindex = 0; sgindex < nsg; sgindex++) {
321 bevent->u.rw_data.dma_info[sgindex].token =
322 sg_dma_address(&sg[sgindex]);
323 bevent->u.rw_data.dma_info[sgindex].len =
324 sg_dma_len(&sg[sgindex]);
327 /* Send the request */
328 hvrc = HvCallEvent_signalLpEvent(&bevent->event);
329 vio_free_event_buffer(viomajorsubtype_blockio, bevent);
332 if (hvrc != HvLpEvent_Rc_Good) {
333 pr_warning("error sending disk event to OS/400 (rc %d)\n",
337 spin_unlock_irqrestore(&viodasd_spinlock, flags);
341 num_req_outstanding--;
342 spin_unlock_irqrestore(&viodasd_spinlock, flags);
343 dma_unmap_sg(d->dev, sg, nsg, direction);
348 * This is the external request processing routine
350 static void do_viodasd_request(struct request_queue *q)
355 * If we already have the maximum number of requests
356 * outstanding to OS/400 just bail out. We'll come
359 while (num_req_outstanding < VIOMAXREQ) {
360 req = blk_fetch_request(q);
363 /* check that request contains a valid command */
364 if (!blk_fs_request(req)) {
365 viodasd_end_request(req, -EIO, blk_rq_sectors(req));
368 /* Try sending the request */
369 if (send_request(req) != 0)
370 viodasd_end_request(req, -EIO, blk_rq_sectors(req));
375 * Probe a single disk and fill in the viodasd_device structure
378 static int probe_disk(struct viodasd_device *d)
381 struct viodasd_waitevent we;
382 int dev_no = DEVICE_NO(d);
384 struct request_queue *q;
388 init_completion(&we.com);
390 /* Send the open event to OS/400 */
391 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
392 HvLpEvent_Type_VirtualIo,
393 viomajorsubtype_blockio | vioblockopen,
394 HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
395 viopath_sourceinst(viopath_hostLp),
396 viopath_targetinst(viopath_hostLp),
397 (u64)(unsigned long)&we, VIOVERSION << 16,
398 ((u64)dev_no << 48) | ((u64)flags<< 32),
401 pr_warning("bad rc on HV open %d\n", (int)hvrc);
405 wait_for_completion(&we.com);
410 /* try again with read only flag set */
411 flags = vioblockflags_ro;
414 if (we.max_disk > (MAX_DISKNO - 1)) {
415 printk_once(KERN_INFO pr_fmt("Only examining the first %d of %d disks connected\n"),
416 MAX_DISKNO, we.max_disk + 1);
419 /* Send the close event to OS/400. We DON'T expect a response */
420 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
421 HvLpEvent_Type_VirtualIo,
422 viomajorsubtype_blockio | vioblockclose,
423 HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
424 viopath_sourceinst(viopath_hostLp),
425 viopath_targetinst(viopath_hostLp),
427 ((u64)dev_no << 48) | ((u64)flags << 32),
430 pr_warning("bad rc sending event to OS/400 %d\n", (int)hvrc);
434 if (d->dev == NULL) {
435 /* this is when we reprobe for new disks */
436 if (vio_create_viodasd(dev_no) == NULL) {
437 pr_warning("cannot allocate virtual device for disk %d\n",
442 * The vio_create_viodasd will have recursed into this
443 * routine with d->dev set to the new vio device and
444 * will finish the setup of the disk below.
449 /* create the request queue for the disk */
450 spin_lock_init(&d->q_lock);
451 q = blk_init_queue(do_viodasd_request, &d->q_lock);
453 pr_warning("cannot allocate queue for disk %d\n", dev_no);
456 g = alloc_disk(1 << PARTITION_SHIFT);
458 pr_warning("cannot allocate disk structure for disk %d\n",
460 blk_cleanup_queue(q);
465 blk_queue_max_segments(q, VIOMAXBLOCKDMA);
466 blk_queue_max_hw_sectors(q, VIODASD_MAXSECTORS);
467 g->major = VIODASD_MAJOR;
468 g->first_minor = dev_no << PARTITION_SHIFT;
470 snprintf(g->disk_name, sizeof(g->disk_name),
471 VIOD_GENHD_NAME "%c%c",
472 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
474 snprintf(g->disk_name, sizeof(g->disk_name),
475 VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
476 g->fops = &viodasd_fops;
479 g->driverfs_dev = d->dev;
480 set_capacity(g, d->size >> 9);
482 pr_info("disk %d: %lu sectors (%lu MB) CHS=%d/%d/%d sector size %d%s\n",
483 dev_no, (unsigned long)(d->size >> 9),
484 (unsigned long)(d->size >> 20),
485 (int)d->cylinders, (int)d->tracks,
486 (int)d->sectors, (int)d->bytes_per_sector,
487 d->read_only ? " (RO)" : "");
489 /* register us in the global list */
494 /* returns the total number of scatterlist elements converted */
495 static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
496 struct scatterlist *sg, int *total_len)
499 const struct rw_data *rw_data = &bevent->u.rw_data;
500 static const int offset =
501 offsetof(struct vioblocklpevent, u.rw_data.dma_info);
502 static const int element_size = sizeof(rw_data->dma_info[0]);
504 numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
505 if (numsg > VIOMAXBLOCKDMA)
506 numsg = VIOMAXBLOCKDMA;
509 sg_init_table(sg, VIOMAXBLOCKDMA);
510 for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
511 sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
512 sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
513 *total_len += rw_data->dma_info[i].len;
519 * Restart all queues, starting with the one _after_ the disk given,
520 * thus reducing the chance of starvation of higher numbered disks.
522 static void viodasd_restart_all_queues_starting_from(int first_index)
526 for (i = first_index + 1; i < MAX_DISKNO; ++i)
527 if (viodasd_devices[i].disk)
528 blk_run_queue(viodasd_devices[i].disk->queue);
529 for (i = 0; i <= first_index; ++i)
530 if (viodasd_devices[i].disk)
531 blk_run_queue(viodasd_devices[i].disk->queue);
535 * For read and write requests, decrement the number of outstanding requests,
536 * Free the DMA buffers we allocated.
538 static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
540 int num_sg, num_sect, pci_direction, total_len;
542 struct scatterlist sg[VIOMAXBLOCKDMA];
543 struct HvLpEvent *event = &bevent->event;
544 unsigned long irq_flags;
545 struct viodasd_device *d;
549 num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
550 num_sect = total_len >> 9;
551 if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
552 pci_direction = DMA_FROM_DEVICE;
554 pci_direction = DMA_TO_DEVICE;
555 req = (struct request *)bevent->event.xCorrelationToken;
556 d = req->rq_disk->private_data;
558 dma_unmap_sg(d->dev, sg, num_sg, pci_direction);
561 * Since this is running in interrupt mode, we need to make sure
562 * we're not stepping on any global I/O operations
564 spin_lock_irqsave(&viodasd_spinlock, irq_flags);
565 num_req_outstanding--;
566 spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);
568 error = (event->xRc == HvLpEvent_Rc_Good) ? 0 : -EIO;
570 const struct vio_error_entry *err;
571 err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
572 pr_warning("read/write error %d:0x%04x (%s)\n",
573 event->xRc, bevent->sub_result, err->msg);
574 num_sect = blk_rq_sectors(req);
576 qlock = req->q->queue_lock;
577 spin_lock_irqsave(qlock, irq_flags);
578 viodasd_end_request(req, error, num_sect);
579 spin_unlock_irqrestore(qlock, irq_flags);
581 /* Finally, try to get more requests off of this device's queue */
582 viodasd_restart_all_queues_starting_from(DEVICE_NO(d));
587 /* This routine handles incoming block LP events */
588 static void handle_block_event(struct HvLpEvent *event)
590 struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
591 struct viodasd_waitevent *pwe;
594 /* Notification that a partition went away! */
596 /* First, we should NEVER get an int here...only acks */
597 if (hvlpevent_is_int(event)) {
598 pr_warning("Yikes! got an int in viodasd event handler!\n");
599 if (hvlpevent_need_ack(event)) {
600 event->xRc = HvLpEvent_Rc_InvalidSubtype;
601 HvCallEvent_ackLpEvent(event);
605 switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
608 * Handle a response to an open request. We get all the
609 * disk information in the response, so update it. The
610 * correlation token contains a pointer to a waitevent
611 * structure that has a completion in it. update the
612 * return code in the waitevent structure and post the
613 * completion to wake up the guy who sent the request
615 pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
616 pwe->rc = event->xRc;
617 pwe->sub_result = bevent->sub_result;
618 if (event->xRc == HvLpEvent_Rc_Good) {
619 const struct open_data *data = &bevent->u.open_data;
620 struct viodasd_device *device =
621 &viodasd_devices[bevent->disk];
623 bevent->flags & vioblockflags_ro;
624 device->size = data->disk_size;
625 device->cylinders = data->cylinders;
626 device->tracks = data->tracks;
627 device->sectors = data->sectors;
628 device->bytes_per_sector = data->bytes_per_sector;
629 pwe->max_disk = data->max_disk;
637 viodasd_handle_read_write(bevent);
641 pr_warning("invalid subtype!");
642 if (hvlpevent_need_ack(event)) {
643 event->xRc = HvLpEvent_Rc_InvalidSubtype;
644 HvCallEvent_ackLpEvent(event);
650 * Get the driver to reprobe for more disks.
652 static ssize_t probe_disks(struct device_driver *drv, const char *buf,
655 struct viodasd_device *d;
657 for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
663 static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);
665 static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
667 struct viodasd_device *d = &viodasd_devices[vdev->unit_address];
675 static int viodasd_remove(struct vio_dev *vdev)
677 struct viodasd_device *d;
679 d = &viodasd_devices[vdev->unit_address];
681 del_gendisk(d->disk);
682 blk_cleanup_queue(d->disk->queue);
691 * viodasd_device_table: Used by vio.c to match devices that we
694 static struct vio_device_id viodasd_device_table[] __devinitdata = {
695 { "block", "IBM,iSeries-viodasd" },
698 MODULE_DEVICE_TABLE(vio, viodasd_device_table);
700 static struct vio_driver viodasd_driver = {
701 .id_table = viodasd_device_table,
702 .probe = viodasd_probe,
703 .remove = viodasd_remove,
706 .owner = THIS_MODULE,
710 static int need_delete_probe;
713 * Initialize the whole device driver. Handle module and non-module
716 static int __init viodasd_init(void)
720 if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
725 /* Try to open to our host lp */
726 if (viopath_hostLp == HvLpIndexInvalid)
729 if (viopath_hostLp == HvLpIndexInvalid) {
730 pr_warning("invalid hosting partition\n");
735 pr_info("vers " VIOD_VERS ", hosting partition %d\n", viopath_hostLp);
737 /* register the block device */
738 rc = register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
740 pr_warning("Unable to get major number %d for %s\n",
741 VIODASD_MAJOR, VIOD_GENHD_NAME);
744 /* Actually open the path to the hosting partition */
745 rc = viopath_open(viopath_hostLp, viomajorsubtype_blockio,
748 pr_warning("error opening path to host partition %d\n",
753 /* Initialize our request handler */
754 vio_setHandler(viomajorsubtype_blockio, handle_block_event);
756 rc = vio_register_driver(&viodasd_driver);
758 pr_warning("vio_register_driver failed\n");
763 * If this call fails, it just means that we cannot dynamically
764 * add virtual disks, but the driver will still work fine for
765 * all existing disk, so ignore the failure.
767 if (!driver_create_file(&viodasd_driver.driver, &driver_attr_probe))
768 need_delete_probe = 1;
773 vio_clearHandler(viomajorsubtype_blockio);
774 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
776 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
780 module_init(viodasd_init);
782 void __exit viodasd_exit(void)
784 if (need_delete_probe)
785 driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
786 vio_unregister_driver(&viodasd_driver);
787 vio_clearHandler(viomajorsubtype_blockio);
788 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
789 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
791 module_exit(viodasd_exit);