3 * Authors: Dave Boutcher <boutcher@us.ibm.com>
4 * Ryan Arnold <ryanarn@us.ibm.com>
5 * Colin Devilbiss <devilbis@us.ibm.com>
6 * Stephen Rothwell <sfr@au1.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.
31 #include <linux/major.h>
33 #include <linux/module.h>
34 #include <linux/kernel.h>
35 #include <linux/blkdev.h>
36 #include <linux/genhd.h>
37 #include <linux/hdreg.h>
38 #include <linux/errno.h>
39 #include <linux/init.h>
40 #include <linux/string.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/completion.h>
43 #include <linux/device.h>
44 #include <linux/scatterlist.h>
46 #include <asm/uaccess.h>
48 #include <asm/iseries/hv_types.h>
49 #include <asm/iseries/hv_lp_event.h>
50 #include <asm/iseries/hv_lp_config.h>
51 #include <asm/iseries/vio.h>
52 #include <asm/firmware.h>
54 MODULE_DESCRIPTION("iSeries Virtual DASD");
55 MODULE_AUTHOR("Dave Boutcher");
56 MODULE_LICENSE("GPL");
59 * We only support 7 partitions per physical disk....so with minor
60 * numbers 0-255 we get a maximum of 32 disks.
62 #define VIOD_GENHD_NAME "iseries/vd"
64 #define VIOD_VERS "1.64"
66 #define VIOD_KERN_WARNING KERN_WARNING "viod: "
67 #define VIOD_KERN_INFO KERN_INFO "viod: "
71 MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS,
72 MAX_DISK_NAME = sizeof(((struct gendisk *)0)->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 inode *ino, struct file *fil)
135 struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
137 struct viodasd_waitevent we;
141 if ((fil != NULL) && (fil->f_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 printk(VIOD_KERN_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 printk(VIOD_KERN_WARNING
171 "bad rc opening disk: %d:0x%04x (%s)\n",
172 (int)we.rc, we.sub_result, err->msg);
180 * External release entry point.
182 static int viodasd_release(struct inode *ino, struct file *fil)
184 struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
187 /* Send the event to OS/400. We DON'T expect a response */
188 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
189 HvLpEvent_Type_VirtualIo,
190 viomajorsubtype_blockio | vioblockclose,
191 HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
192 viopath_sourceinst(viopath_hostLp),
193 viopath_targetinst(viopath_hostLp),
195 ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */,
198 printk(VIOD_KERN_WARNING "HV close call failed %d\n",
204 /* External ioctl entry point.
206 static int viodasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
208 struct gendisk *disk = bdev->bd_disk;
209 struct viodasd_device *d = disk->private_data;
211 geo->sectors = d->sectors ? d->sectors : 32;
212 geo->heads = d->tracks ? d->tracks : 64;
213 geo->cylinders = d->cylinders ? d->cylinders :
214 get_capacity(disk) / (geo->sectors * geo->heads);
220 * Our file operations table
222 static struct block_device_operations viodasd_fops = {
223 .owner = THIS_MODULE,
224 .open = viodasd_open,
225 .release = viodasd_release,
226 .getgeo = viodasd_getgeo,
232 static void viodasd_end_request(struct request *req, int uptodate,
235 if (end_that_request_first(req, uptodate, num_sectors))
237 add_disk_randomness(req->rq_disk);
238 end_that_request_last(req, uptodate);
242 * Send an actual I/O request to OS/400
244 static int send_request(struct request *req)
251 struct vioblocklpevent *bevent;
252 struct HvLpEvent *hev;
253 struct scatterlist sg[VIOMAXBLOCKDMA];
256 struct viodasd_device *d;
259 start = (u64)req->sector << 9;
261 if (rq_data_dir(req) == READ) {
262 direction = DMA_FROM_DEVICE;
263 viocmd = viomajorsubtype_blockio | vioblockread;
266 direction = DMA_TO_DEVICE;
267 viocmd = viomajorsubtype_blockio | vioblockwrite;
271 d = req->rq_disk->private_data;
273 /* Now build the scatter-gather list */
274 sg_init_table(sg, VIOMAXBLOCKDMA);
275 nsg = blk_rq_map_sg(req->q, req, sg);
276 nsg = dma_map_sg(d->dev, sg, nsg, direction);
278 spin_lock_irqsave(&viodasd_spinlock, flags);
279 num_req_outstanding++;
281 /* This optimization handles a single DMA block */
283 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
284 HvLpEvent_Type_VirtualIo, viocmd,
285 HvLpEvent_AckInd_DoAck,
286 HvLpEvent_AckType_ImmediateAck,
287 viopath_sourceinst(viopath_hostLp),
288 viopath_targetinst(viopath_hostLp),
289 (u64)(unsigned long)req, VIOVERSION << 16,
290 ((u64)DEVICE_NO(d) << 48), start,
291 ((u64)sg_dma_address(&sg[0])) << 32,
294 bevent = (struct vioblocklpevent *)
295 vio_get_event_buffer(viomajorsubtype_blockio);
296 if (bevent == NULL) {
297 printk(VIOD_KERN_WARNING
298 "error allocating disk event buffer\n");
303 * Now build up the actual request. Note that we store
304 * the pointer to the request in the correlation
305 * token so we can match the response up later
307 memset(bevent, 0, sizeof(struct vioblocklpevent));
308 hev = &bevent->event;
309 hev->flags = HV_LP_EVENT_VALID | HV_LP_EVENT_DO_ACK |
311 hev->xType = HvLpEvent_Type_VirtualIo;
312 hev->xSubtype = viocmd;
313 hev->xSourceLp = HvLpConfig_getLpIndex();
314 hev->xTargetLp = viopath_hostLp;
316 offsetof(struct vioblocklpevent, u.rw_data.dma_info) +
317 (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1;
318 hev->xSourceInstanceId = viopath_sourceinst(viopath_hostLp);
319 hev->xTargetInstanceId = viopath_targetinst(viopath_hostLp);
320 hev->xCorrelationToken = (u64)req;
321 bevent->version = VIOVERSION;
322 bevent->disk = DEVICE_NO(d);
323 bevent->u.rw_data.offset = start;
326 * Copy just the dma information from the sg list
329 for (sgindex = 0; sgindex < nsg; sgindex++) {
330 bevent->u.rw_data.dma_info[sgindex].token =
331 sg_dma_address(&sg[sgindex]);
332 bevent->u.rw_data.dma_info[sgindex].len =
333 sg_dma_len(&sg[sgindex]);
336 /* Send the request */
337 hvrc = HvCallEvent_signalLpEvent(&bevent->event);
338 vio_free_event_buffer(viomajorsubtype_blockio, bevent);
341 if (hvrc != HvLpEvent_Rc_Good) {
342 printk(VIOD_KERN_WARNING
343 "error sending disk event to OS/400 (rc %d)\n",
347 spin_unlock_irqrestore(&viodasd_spinlock, flags);
351 num_req_outstanding--;
352 spin_unlock_irqrestore(&viodasd_spinlock, flags);
353 dma_unmap_sg(d->dev, sg, nsg, direction);
358 * This is the external request processing routine
360 static void do_viodasd_request(struct request_queue *q)
365 * If we already have the maximum number of requests
366 * outstanding to OS/400 just bail out. We'll come
369 while (num_req_outstanding < VIOMAXREQ) {
370 req = elv_next_request(q);
373 /* dequeue the current request from the queue */
374 blkdev_dequeue_request(req);
375 /* check that request contains a valid command */
376 if (!blk_fs_request(req)) {
377 viodasd_end_request(req, 0, req->hard_nr_sectors);
380 /* Try sending the request */
381 if (send_request(req) != 0)
382 viodasd_end_request(req, 0, req->hard_nr_sectors);
387 * Probe a single disk and fill in the viodasd_device structure
390 static int probe_disk(struct viodasd_device *d)
393 struct viodasd_waitevent we;
394 int dev_no = DEVICE_NO(d);
396 struct request_queue *q;
400 init_completion(&we.com);
402 /* Send the open event to OS/400 */
403 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
404 HvLpEvent_Type_VirtualIo,
405 viomajorsubtype_blockio | vioblockopen,
406 HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
407 viopath_sourceinst(viopath_hostLp),
408 viopath_targetinst(viopath_hostLp),
409 (u64)(unsigned long)&we, VIOVERSION << 16,
410 ((u64)dev_no << 48) | ((u64)flags<< 32),
413 printk(VIOD_KERN_WARNING "bad rc on HV open %d\n", (int)hvrc);
417 wait_for_completion(&we.com);
422 /* try again with read only flag set */
423 flags = vioblockflags_ro;
426 if (we.max_disk > (MAX_DISKNO - 1)) {
431 printk(VIOD_KERN_INFO
432 "Only examining the first %d "
433 "of %d disks connected\n",
434 MAX_DISKNO, we.max_disk + 1);
438 /* Send the close event to OS/400. We DON'T expect a response */
439 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
440 HvLpEvent_Type_VirtualIo,
441 viomajorsubtype_blockio | vioblockclose,
442 HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
443 viopath_sourceinst(viopath_hostLp),
444 viopath_targetinst(viopath_hostLp),
446 ((u64)dev_no << 48) | ((u64)flags << 32),
449 printk(VIOD_KERN_WARNING
450 "bad rc sending event to OS/400 %d\n", (int)hvrc);
454 if (d->dev == NULL) {
455 /* this is when we reprobe for new disks */
456 if (vio_create_viodasd(dev_no) == NULL) {
457 printk(VIOD_KERN_WARNING
458 "cannot allocate virtual device for disk %d\n",
463 * The vio_create_viodasd will have recursed into this
464 * routine with d->dev set to the new vio device and
465 * will finish the setup of the disk below.
470 /* create the request queue for the disk */
471 spin_lock_init(&d->q_lock);
472 q = blk_init_queue(do_viodasd_request, &d->q_lock);
474 printk(VIOD_KERN_WARNING "cannot allocate queue for disk %d\n",
478 g = alloc_disk(1 << PARTITION_SHIFT);
480 printk(VIOD_KERN_WARNING
481 "cannot allocate disk structure for disk %d\n",
483 blk_cleanup_queue(q);
488 blk_queue_max_hw_segments(q, VIOMAXBLOCKDMA);
489 blk_queue_max_phys_segments(q, VIOMAXBLOCKDMA);
490 blk_queue_max_sectors(q, VIODASD_MAXSECTORS);
491 g->major = VIODASD_MAJOR;
492 g->first_minor = dev_no << PARTITION_SHIFT;
494 snprintf(g->disk_name, sizeof(g->disk_name),
495 VIOD_GENHD_NAME "%c%c",
496 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
498 snprintf(g->disk_name, sizeof(g->disk_name),
499 VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
500 g->fops = &viodasd_fops;
503 g->driverfs_dev = d->dev;
504 set_capacity(g, d->size >> 9);
506 printk(VIOD_KERN_INFO "disk %d: %lu sectors (%lu MB) "
507 "CHS=%d/%d/%d sector size %d%s\n",
508 dev_no, (unsigned long)(d->size >> 9),
509 (unsigned long)(d->size >> 20),
510 (int)d->cylinders, (int)d->tracks,
511 (int)d->sectors, (int)d->bytes_per_sector,
512 d->read_only ? " (RO)" : "");
514 /* register us in the global list */
519 /* returns the total number of scatterlist elements converted */
520 static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
521 struct scatterlist *sg, int *total_len)
524 const struct rw_data *rw_data = &bevent->u.rw_data;
525 static const int offset =
526 offsetof(struct vioblocklpevent, u.rw_data.dma_info);
527 static const int element_size = sizeof(rw_data->dma_info[0]);
529 numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
530 if (numsg > VIOMAXBLOCKDMA)
531 numsg = VIOMAXBLOCKDMA;
534 memset(sg, 0, sizeof(sg[0]) * VIOMAXBLOCKDMA);
536 for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
537 sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
538 sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
539 *total_len += rw_data->dma_info[i].len;
545 * Restart all queues, starting with the one _after_ the disk given,
546 * thus reducing the chance of starvation of higher numbered disks.
548 static void viodasd_restart_all_queues_starting_from(int first_index)
552 for (i = first_index + 1; i < MAX_DISKNO; ++i)
553 if (viodasd_devices[i].disk)
554 blk_run_queue(viodasd_devices[i].disk->queue);
555 for (i = 0; i <= first_index; ++i)
556 if (viodasd_devices[i].disk)
557 blk_run_queue(viodasd_devices[i].disk->queue);
561 * For read and write requests, decrement the number of outstanding requests,
562 * Free the DMA buffers we allocated.
564 static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
566 int num_sg, num_sect, pci_direction, total_len;
568 struct scatterlist sg[VIOMAXBLOCKDMA];
569 struct HvLpEvent *event = &bevent->event;
570 unsigned long irq_flags;
571 struct viodasd_device *d;
575 num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
576 num_sect = total_len >> 9;
577 if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
578 pci_direction = DMA_FROM_DEVICE;
580 pci_direction = DMA_TO_DEVICE;
581 req = (struct request *)bevent->event.xCorrelationToken;
582 d = req->rq_disk->private_data;
584 dma_unmap_sg(d->dev, sg, num_sg, pci_direction);
587 * Since this is running in interrupt mode, we need to make sure
588 * we're not stepping on any global I/O operations
590 spin_lock_irqsave(&viodasd_spinlock, irq_flags);
591 num_req_outstanding--;
592 spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);
594 error = event->xRc != HvLpEvent_Rc_Good;
596 const struct vio_error_entry *err;
597 err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
598 printk(VIOD_KERN_WARNING "read/write error %d:0x%04x (%s)\n",
599 event->xRc, bevent->sub_result, err->msg);
600 num_sect = req->hard_nr_sectors;
602 qlock = req->q->queue_lock;
603 spin_lock_irqsave(qlock, irq_flags);
604 viodasd_end_request(req, !error, num_sect);
605 spin_unlock_irqrestore(qlock, irq_flags);
607 /* Finally, try to get more requests off of this device's queue */
608 viodasd_restart_all_queues_starting_from(DEVICE_NO(d));
613 /* This routine handles incoming block LP events */
614 static void handle_block_event(struct HvLpEvent *event)
616 struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
617 struct viodasd_waitevent *pwe;
620 /* Notification that a partition went away! */
622 /* First, we should NEVER get an int here...only acks */
623 if (hvlpevent_is_int(event)) {
624 printk(VIOD_KERN_WARNING
625 "Yikes! got an int in viodasd event handler!\n");
626 if (hvlpevent_need_ack(event)) {
627 event->xRc = HvLpEvent_Rc_InvalidSubtype;
628 HvCallEvent_ackLpEvent(event);
632 switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
635 * Handle a response to an open request. We get all the
636 * disk information in the response, so update it. The
637 * correlation token contains a pointer to a waitevent
638 * structure that has a completion in it. update the
639 * return code in the waitevent structure and post the
640 * completion to wake up the guy who sent the request
642 pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
643 pwe->rc = event->xRc;
644 pwe->sub_result = bevent->sub_result;
645 if (event->xRc == HvLpEvent_Rc_Good) {
646 const struct open_data *data = &bevent->u.open_data;
647 struct viodasd_device *device =
648 &viodasd_devices[bevent->disk];
650 bevent->flags & vioblockflags_ro;
651 device->size = data->disk_size;
652 device->cylinders = data->cylinders;
653 device->tracks = data->tracks;
654 device->sectors = data->sectors;
655 device->bytes_per_sector = data->bytes_per_sector;
656 pwe->max_disk = data->max_disk;
664 viodasd_handle_read_write(bevent);
668 printk(VIOD_KERN_WARNING "invalid subtype!");
669 if (hvlpevent_need_ack(event)) {
670 event->xRc = HvLpEvent_Rc_InvalidSubtype;
671 HvCallEvent_ackLpEvent(event);
677 * Get the driver to reprobe for more disks.
679 static ssize_t probe_disks(struct device_driver *drv, const char *buf,
682 struct viodasd_device *d;
684 for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
690 static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);
692 static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
694 struct viodasd_device *d = &viodasd_devices[vdev->unit_address];
702 static int viodasd_remove(struct vio_dev *vdev)
704 struct viodasd_device *d;
706 d = &viodasd_devices[vdev->unit_address];
708 del_gendisk(d->disk);
709 blk_cleanup_queue(d->disk->queue);
718 * viodasd_device_table: Used by vio.c to match devices that we
721 static struct vio_device_id viodasd_device_table[] __devinitdata = {
722 { "block", "IBM,iSeries-viodasd" },
725 MODULE_DEVICE_TABLE(vio, viodasd_device_table);
727 static struct vio_driver viodasd_driver = {
728 .id_table = viodasd_device_table,
729 .probe = viodasd_probe,
730 .remove = viodasd_remove,
733 .owner = THIS_MODULE,
737 static int need_delete_probe;
740 * Initialize the whole device driver. Handle module and non-module
743 static int __init viodasd_init(void)
747 if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
752 /* Try to open to our host lp */
753 if (viopath_hostLp == HvLpIndexInvalid)
756 if (viopath_hostLp == HvLpIndexInvalid) {
757 printk(VIOD_KERN_WARNING "invalid hosting partition\n");
762 printk(VIOD_KERN_INFO "vers " VIOD_VERS ", hosting partition %d\n",
765 /* register the block device */
766 rc = register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
768 printk(VIOD_KERN_WARNING
769 "Unable to get major number %d for %s\n",
770 VIODASD_MAJOR, VIOD_GENHD_NAME);
773 /* Actually open the path to the hosting partition */
774 rc = viopath_open(viopath_hostLp, viomajorsubtype_blockio,
777 printk(VIOD_KERN_WARNING
778 "error opening path to host partition %d\n",
783 /* Initialize our request handler */
784 vio_setHandler(viomajorsubtype_blockio, handle_block_event);
786 rc = vio_register_driver(&viodasd_driver);
788 printk(VIOD_KERN_WARNING "vio_register_driver failed\n");
793 * If this call fails, it just means that we cannot dynamically
794 * add virtual disks, but the driver will still work fine for
795 * all existing disk, so ignore the failure.
797 if (!driver_create_file(&viodasd_driver.driver, &driver_attr_probe))
798 need_delete_probe = 1;
803 vio_clearHandler(viomajorsubtype_blockio);
804 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
806 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
810 module_init(viodasd_init);
812 void __exit viodasd_exit(void)
814 if (need_delete_probe)
815 driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
816 vio_unregister_driver(&viodasd_driver);
817 vio_clearHandler(viomajorsubtype_blockio);
818 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
819 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
821 module_exit(viodasd_exit);