blob: 390d69bb7c482ad1a79a09a42192575b692b4980 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* -*- linux-c -*-
2 * viodasd.c
3 * Authors: Dave Boutcher <boutcher@us.ibm.com>
4 * Ryan Arnold <ryanarn@us.ibm.com>
5 * Colin Devilbiss <devilbis@us.ibm.com>
Stephen Rothwell8962cad2008-05-23 11:41:46 +10006 * Stephen Rothwell
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
8 * (C) Copyright 2000-2004 IBM Corporation
9 *
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.
14 *
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.
19 *
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
23 *
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.
27 *
28 * All disk operations are performed by sending messages back and forth to
29 * the OS/400 partition.
30 */
31#include <linux/major.h>
32#include <linux/fs.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>
Jens Axboe45711f12007-10-22 21:19:53 +020044#include <linux/scatterlist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46#include <asm/uaccess.h>
47#include <asm/vio.h>
Kelly Daly1ec65d72005-11-02 13:46:07 +110048#include <asm/iseries/hv_types.h>
Kelly Dalye45423e2005-11-02 12:08:31 +110049#include <asm/iseries/hv_lp_event.h>
Kelly Daly15b17182005-11-02 11:55:28 +110050#include <asm/iseries/hv_lp_config.h>
Kelly Dalyb4206772005-11-02 15:13:57 +110051#include <asm/iseries/vio.h>
Stephen Rothwellfb8b5002006-12-15 15:40:08 +110052#include <asm/firmware.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54MODULE_DESCRIPTION("iSeries Virtual DASD");
55MODULE_AUTHOR("Dave Boutcher");
56MODULE_LICENSE("GPL");
57
58/*
59 * We only support 7 partitions per physical disk....so with minor
60 * numbers 0-255 we get a maximum of 32 disks.
61 */
62#define VIOD_GENHD_NAME "iseries/vd"
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64#define VIOD_VERS "1.64"
65
66#define VIOD_KERN_WARNING KERN_WARNING "viod: "
67#define VIOD_KERN_INFO KERN_INFO "viod: "
68
69enum {
70 PARTITION_SHIFT = 3,
71 MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS,
Julia Lawallea6728c2008-03-29 08:21:08 +110072 MAX_DISK_NAME = FIELD_SIZEOF(struct gendisk, disk_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073};
74
75static DEFINE_SPINLOCK(viodasd_spinlock);
76
77#define VIOMAXREQ 16
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79#define DEVICE_NO(cell) ((struct viodasd_device *)(cell) - &viodasd_devices[0])
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081struct viodasd_waitevent {
82 struct completion com;
83 int rc;
84 u16 sub_result;
85 int max_disk; /* open */
86};
87
88static 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" },
102 { 0x0000, 0, NULL },
103};
104
105/*
106 * Figure out the biggest I/O request (in sectors) we can accept
107 */
108#define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA)
109
110/*
111 * Number of disk I/O requests we've sent to OS/400
112 */
113static int num_req_outstanding;
114
115/*
116 * This is our internal structure for keeping track of disk devices
117 */
118struct viodasd_device {
119 u16 cylinders;
120 u16 tracks;
121 u16 sectors;
122 u16 bytes_per_sector;
123 u64 size;
124 int read_only;
125 spinlock_t q_lock;
126 struct gendisk *disk;
127 struct device *dev;
128} viodasd_devices[MAX_DISKNO];
129
130/*
131 * External open entry point.
132 */
Al Virof115a142008-03-02 10:22:07 -0500133static int viodasd_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134{
Al Virof115a142008-03-02 10:22:07 -0500135 struct viodasd_device *d = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 HvLpEvent_Rc hvrc;
137 struct viodasd_waitevent we;
138 u16 flags = 0;
139
140 if (d->read_only) {
Al Virof115a142008-03-02 10:22:07 -0500141 if (mode & FMODE_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 return -EROFS;
143 flags = vioblockflags_ro;
144 }
145
146 init_completion(&we.com);
147
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),
157 0, 0, 0);
158 if (hvrc != 0) {
159 printk(VIOD_KERN_WARNING "HV open failed %d\n", (int)hvrc);
160 return -EIO;
161 }
162
163 wait_for_completion(&we.com);
164
165 /* Check the return code */
166 if (we.rc != 0) {
167 const struct vio_error_entry *err =
168 vio_lookup_rc(viodasd_err_table, we.sub_result);
169
170 printk(VIOD_KERN_WARNING
171 "bad rc opening disk: %d:0x%04x (%s)\n",
172 (int)we.rc, we.sub_result, err->msg);
173 return -EIO;
174 }
175
176 return 0;
177}
178
179/*
180 * External release entry point.
181 */
Al Virof115a142008-03-02 10:22:07 -0500182static int viodasd_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Al Virof115a142008-03-02 10:22:07 -0500184 struct viodasd_device *d = disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 HvLpEvent_Rc hvrc;
186
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),
194 0, VIOVERSION << 16,
195 ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */,
196 0, 0, 0);
197 if (hvrc != 0)
198 printk(VIOD_KERN_WARNING "HV close call failed %d\n",
199 (int)hvrc);
200 return 0;
201}
202
203
204/* External ioctl entry point.
205 */
Christoph Hellwiga885c8c2006-01-08 01:02:50 -0800206static int viodasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
Christoph Hellwiga885c8c2006-01-08 01:02:50 -0800208 struct gendisk *disk = bdev->bd_disk;
209 struct viodasd_device *d = disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Stephen Rothwell36a70032007-05-29 16:06:54 +1000211 geo->sectors = d->sectors ? d->sectors : 32;
Christoph Hellwiga885c8c2006-01-08 01:02:50 -0800212 geo->heads = d->tracks ? d->tracks : 64;
213 geo->cylinders = d->cylinders ? d->cylinders :
Stephen Rothwell36a70032007-05-29 16:06:54 +1000214 get_capacity(disk) / (geo->sectors * geo->heads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Christoph Hellwiga885c8c2006-01-08 01:02:50 -0800216 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
219/*
220 * Our file operations table
221 */
222static struct block_device_operations viodasd_fops = {
223 .owner = THIS_MODULE,
Al Virof115a142008-03-02 10:22:07 -0500224 .open = viodasd_open,
225 .release = viodasd_release,
Christoph Hellwiga885c8c2006-01-08 01:02:50 -0800226 .getgeo = viodasd_getgeo,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227};
228
229/*
230 * End a request
231 */
Kiyoshi Uedab2aec242007-12-11 17:47:14 -0500232static void viodasd_end_request(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 int num_sectors)
234{
Kiyoshi Uedab2aec242007-12-11 17:47:14 -0500235 __blk_end_request(req, error, num_sectors << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
237
238/*
239 * Send an actual I/O request to OS/400
240 */
241static int send_request(struct request *req)
242{
243 u64 start;
244 int direction;
245 int nsg;
246 u16 viocmd;
247 HvLpEvent_Rc hvrc;
248 struct vioblocklpevent *bevent;
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100249 struct HvLpEvent *hev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 struct scatterlist sg[VIOMAXBLOCKDMA];
251 int sgindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 struct viodasd_device *d;
253 unsigned long flags;
254
Tejun Heo83096eb2009-05-07 22:24:39 +0900255 start = (u64)blk_rq_pos(req) << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257 if (rq_data_dir(req) == READ) {
258 direction = DMA_FROM_DEVICE;
259 viocmd = viomajorsubtype_blockio | vioblockread;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 } else {
261 direction = DMA_TO_DEVICE;
262 viocmd = viomajorsubtype_blockio | vioblockwrite;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 }
264
265 d = req->rq_disk->private_data;
266
267 /* Now build the scatter-gather list */
Jens Axboe45711f12007-10-22 21:19:53 +0200268 sg_init_table(sg, VIOMAXBLOCKDMA);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 nsg = blk_rq_map_sg(req->q, req, sg);
270 nsg = dma_map_sg(d->dev, sg, nsg, direction);
271
272 spin_lock_irqsave(&viodasd_spinlock, flags);
273 num_req_outstanding++;
274
275 /* This optimization handles a single DMA block */
276 if (nsg == 1)
277 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
278 HvLpEvent_Type_VirtualIo, viocmd,
279 HvLpEvent_AckInd_DoAck,
280 HvLpEvent_AckType_ImmediateAck,
281 viopath_sourceinst(viopath_hostLp),
282 viopath_targetinst(viopath_hostLp),
283 (u64)(unsigned long)req, VIOVERSION << 16,
284 ((u64)DEVICE_NO(d) << 48), start,
285 ((u64)sg_dma_address(&sg[0])) << 32,
286 sg_dma_len(&sg[0]));
287 else {
288 bevent = (struct vioblocklpevent *)
289 vio_get_event_buffer(viomajorsubtype_blockio);
290 if (bevent == NULL) {
291 printk(VIOD_KERN_WARNING
292 "error allocating disk event buffer\n");
293 goto error_ret;
294 }
295
296 /*
297 * Now build up the actual request. Note that we store
298 * the pointer to the request in the correlation
299 * token so we can match the response up later
300 */
301 memset(bevent, 0, sizeof(struct vioblocklpevent));
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100302 hev = &bevent->event;
303 hev->flags = HV_LP_EVENT_VALID | HV_LP_EVENT_DO_ACK |
304 HV_LP_EVENT_INT;
305 hev->xType = HvLpEvent_Type_VirtualIo;
306 hev->xSubtype = viocmd;
307 hev->xSourceLp = HvLpConfig_getLpIndex();
308 hev->xTargetLp = viopath_hostLp;
309 hev->xSizeMinus1 =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 offsetof(struct vioblocklpevent, u.rw_data.dma_info) +
311 (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1;
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100312 hev->xSourceInstanceId = viopath_sourceinst(viopath_hostLp);
313 hev->xTargetInstanceId = viopath_targetinst(viopath_hostLp);
314 hev->xCorrelationToken = (u64)req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 bevent->version = VIOVERSION;
316 bevent->disk = DEVICE_NO(d);
317 bevent->u.rw_data.offset = start;
318
319 /*
320 * Copy just the dma information from the sg list
321 * into the request
322 */
323 for (sgindex = 0; sgindex < nsg; sgindex++) {
324 bevent->u.rw_data.dma_info[sgindex].token =
325 sg_dma_address(&sg[sgindex]);
326 bevent->u.rw_data.dma_info[sgindex].len =
327 sg_dma_len(&sg[sgindex]);
328 }
329
330 /* Send the request */
331 hvrc = HvCallEvent_signalLpEvent(&bevent->event);
332 vio_free_event_buffer(viomajorsubtype_blockio, bevent);
333 }
334
335 if (hvrc != HvLpEvent_Rc_Good) {
336 printk(VIOD_KERN_WARNING
337 "error sending disk event to OS/400 (rc %d)\n",
338 (int)hvrc);
339 goto error_ret;
340 }
341 spin_unlock_irqrestore(&viodasd_spinlock, flags);
342 return 0;
343
344error_ret:
345 num_req_outstanding--;
346 spin_unlock_irqrestore(&viodasd_spinlock, flags);
347 dma_unmap_sg(d->dev, sg, nsg, direction);
348 return -1;
349}
350
351/*
352 * This is the external request processing routine
353 */
Jens Axboe165125e2007-07-24 09:28:11 +0200354static void do_viodasd_request(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
356 struct request *req;
357
358 /*
359 * If we already have the maximum number of requests
360 * outstanding to OS/400 just bail out. We'll come
361 * back later.
362 */
363 while (num_req_outstanding < VIOMAXREQ) {
Tejun Heo9934c8c2009-05-08 11:54:16 +0900364 req = blk_fetch_request(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (req == NULL)
366 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 /* check that request contains a valid command */
368 if (!blk_fs_request(req)) {
Tejun Heo5b936292009-05-07 22:24:38 +0900369 viodasd_end_request(req, -EIO, blk_rq_sectors(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 continue;
371 }
372 /* Try sending the request */
373 if (send_request(req) != 0)
Tejun Heo5b936292009-05-07 22:24:38 +0900374 viodasd_end_request(req, -EIO, blk_rq_sectors(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 }
376}
377
378/*
379 * Probe a single disk and fill in the viodasd_device structure
380 * for it.
381 */
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000382static int probe_disk(struct viodasd_device *d)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383{
384 HvLpEvent_Rc hvrc;
385 struct viodasd_waitevent we;
386 int dev_no = DEVICE_NO(d);
387 struct gendisk *g;
388 struct request_queue *q;
389 u16 flags = 0;
390
391retry:
392 init_completion(&we.com);
393
394 /* Send the open event to OS/400 */
395 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
396 HvLpEvent_Type_VirtualIo,
397 viomajorsubtype_blockio | vioblockopen,
398 HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
399 viopath_sourceinst(viopath_hostLp),
400 viopath_targetinst(viopath_hostLp),
401 (u64)(unsigned long)&we, VIOVERSION << 16,
402 ((u64)dev_no << 48) | ((u64)flags<< 32),
403 0, 0, 0);
404 if (hvrc != 0) {
405 printk(VIOD_KERN_WARNING "bad rc on HV open %d\n", (int)hvrc);
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000406 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 }
408
409 wait_for_completion(&we.com);
410
411 if (we.rc != 0) {
412 if (flags != 0)
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000413 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 /* try again with read only flag set */
415 flags = vioblockflags_ro;
416 goto retry;
417 }
418 if (we.max_disk > (MAX_DISKNO - 1)) {
419 static int warned;
420
421 if (warned == 0) {
422 warned++;
423 printk(VIOD_KERN_INFO
424 "Only examining the first %d "
425 "of %d disks connected\n",
426 MAX_DISKNO, we.max_disk + 1);
427 }
428 }
429
430 /* Send the close event to OS/400. We DON'T expect a response */
431 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
432 HvLpEvent_Type_VirtualIo,
433 viomajorsubtype_blockio | vioblockclose,
434 HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
435 viopath_sourceinst(viopath_hostLp),
436 viopath_targetinst(viopath_hostLp),
437 0, VIOVERSION << 16,
438 ((u64)dev_no << 48) | ((u64)flags << 32),
439 0, 0, 0);
440 if (hvrc != 0) {
441 printk(VIOD_KERN_WARNING
442 "bad rc sending event to OS/400 %d\n", (int)hvrc);
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000443 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 }
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000445
446 if (d->dev == NULL) {
447 /* this is when we reprobe for new disks */
448 if (vio_create_viodasd(dev_no) == NULL) {
449 printk(VIOD_KERN_WARNING
450 "cannot allocate virtual device for disk %d\n",
451 dev_no);
452 return 0;
453 }
454 /*
455 * The vio_create_viodasd will have recursed into this
456 * routine with d->dev set to the new vio device and
457 * will finish the setup of the disk below.
458 */
459 return 1;
460 }
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 /* create the request queue for the disk */
463 spin_lock_init(&d->q_lock);
464 q = blk_init_queue(do_viodasd_request, &d->q_lock);
465 if (q == NULL) {
466 printk(VIOD_KERN_WARNING "cannot allocate queue for disk %d\n",
467 dev_no);
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000468 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
470 g = alloc_disk(1 << PARTITION_SHIFT);
471 if (g == NULL) {
472 printk(VIOD_KERN_WARNING
473 "cannot allocate disk structure for disk %d\n",
474 dev_no);
475 blk_cleanup_queue(q);
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000476 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
478
479 d->disk = g;
480 blk_queue_max_hw_segments(q, VIOMAXBLOCKDMA);
481 blk_queue_max_phys_segments(q, VIOMAXBLOCKDMA);
482 blk_queue_max_sectors(q, VIODASD_MAXSECTORS);
483 g->major = VIODASD_MAJOR;
484 g->first_minor = dev_no << PARTITION_SHIFT;
485 if (dev_no >= 26)
486 snprintf(g->disk_name, sizeof(g->disk_name),
487 VIOD_GENHD_NAME "%c%c",
488 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
489 else
490 snprintf(g->disk_name, sizeof(g->disk_name),
491 VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 g->fops = &viodasd_fops;
493 g->queue = q;
494 g->private_data = d;
495 g->driverfs_dev = d->dev;
496 set_capacity(g, d->size >> 9);
497
498 printk(VIOD_KERN_INFO "disk %d: %lu sectors (%lu MB) "
499 "CHS=%d/%d/%d sector size %d%s\n",
500 dev_no, (unsigned long)(d->size >> 9),
501 (unsigned long)(d->size >> 20),
502 (int)d->cylinders, (int)d->tracks,
503 (int)d->sectors, (int)d->bytes_per_sector,
504 d->read_only ? " (RO)" : "");
505
506 /* register us in the global list */
507 add_disk(g);
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000508 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509}
510
511/* returns the total number of scatterlist elements converted */
512static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
513 struct scatterlist *sg, int *total_len)
514{
515 int i, numsg;
516 const struct rw_data *rw_data = &bevent->u.rw_data;
517 static const int offset =
518 offsetof(struct vioblocklpevent, u.rw_data.dma_info);
519 static const int element_size = sizeof(rw_data->dma_info[0]);
520
521 numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
522 if (numsg > VIOMAXBLOCKDMA)
523 numsg = VIOMAXBLOCKDMA;
524
525 *total_len = 0;
Benjamin Herrenschmidt25c0a7b2008-03-12 17:23:56 +1100526 sg_init_table(sg, VIOMAXBLOCKDMA);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
528 sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
529 sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
530 *total_len += rw_data->dma_info[i].len;
531 }
532 return i;
533}
534
535/*
536 * Restart all queues, starting with the one _after_ the disk given,
537 * thus reducing the chance of starvation of higher numbered disks.
538 */
539static void viodasd_restart_all_queues_starting_from(int first_index)
540{
541 int i;
542
543 for (i = first_index + 1; i < MAX_DISKNO; ++i)
544 if (viodasd_devices[i].disk)
545 blk_run_queue(viodasd_devices[i].disk->queue);
546 for (i = 0; i <= first_index; ++i)
547 if (viodasd_devices[i].disk)
548 blk_run_queue(viodasd_devices[i].disk->queue);
549}
550
551/*
552 * For read and write requests, decrement the number of outstanding requests,
553 * Free the DMA buffers we allocated.
554 */
555static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
556{
557 int num_sg, num_sect, pci_direction, total_len;
558 struct request *req;
559 struct scatterlist sg[VIOMAXBLOCKDMA];
560 struct HvLpEvent *event = &bevent->event;
561 unsigned long irq_flags;
562 struct viodasd_device *d;
563 int error;
564 spinlock_t *qlock;
565
566 num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
567 num_sect = total_len >> 9;
568 if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
569 pci_direction = DMA_FROM_DEVICE;
570 else
571 pci_direction = DMA_TO_DEVICE;
572 req = (struct request *)bevent->event.xCorrelationToken;
573 d = req->rq_disk->private_data;
574
575 dma_unmap_sg(d->dev, sg, num_sg, pci_direction);
576
577 /*
578 * Since this is running in interrupt mode, we need to make sure
579 * we're not stepping on any global I/O operations
580 */
581 spin_lock_irqsave(&viodasd_spinlock, irq_flags);
582 num_req_outstanding--;
583 spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);
584
Kiyoshi Uedab2aec242007-12-11 17:47:14 -0500585 error = (event->xRc == HvLpEvent_Rc_Good) ? 0 : -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (error) {
587 const struct vio_error_entry *err;
588 err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
589 printk(VIOD_KERN_WARNING "read/write error %d:0x%04x (%s)\n",
590 event->xRc, bevent->sub_result, err->msg);
Tejun Heo5b936292009-05-07 22:24:38 +0900591 num_sect = blk_rq_sectors(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
593 qlock = req->q->queue_lock;
594 spin_lock_irqsave(qlock, irq_flags);
Kiyoshi Uedab2aec242007-12-11 17:47:14 -0500595 viodasd_end_request(req, error, num_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 spin_unlock_irqrestore(qlock, irq_flags);
597
598 /* Finally, try to get more requests off of this device's queue */
599 viodasd_restart_all_queues_starting_from(DEVICE_NO(d));
600
601 return 0;
602}
603
604/* This routine handles incoming block LP events */
605static void handle_block_event(struct HvLpEvent *event)
606{
607 struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
608 struct viodasd_waitevent *pwe;
609
610 if (event == NULL)
611 /* Notification that a partition went away! */
612 return;
613 /* First, we should NEVER get an int here...only acks */
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100614 if (hvlpevent_is_int(event)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 printk(VIOD_KERN_WARNING
616 "Yikes! got an int in viodasd event handler!\n");
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100617 if (hvlpevent_need_ack(event)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 event->xRc = HvLpEvent_Rc_InvalidSubtype;
619 HvCallEvent_ackLpEvent(event);
620 }
621 }
622
623 switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
624 case vioblockopen:
625 /*
626 * Handle a response to an open request. We get all the
627 * disk information in the response, so update it. The
628 * correlation token contains a pointer to a waitevent
629 * structure that has a completion in it. update the
630 * return code in the waitevent structure and post the
631 * completion to wake up the guy who sent the request
632 */
633 pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
634 pwe->rc = event->xRc;
635 pwe->sub_result = bevent->sub_result;
636 if (event->xRc == HvLpEvent_Rc_Good) {
637 const struct open_data *data = &bevent->u.open_data;
638 struct viodasd_device *device =
639 &viodasd_devices[bevent->disk];
640 device->read_only =
641 bevent->flags & vioblockflags_ro;
642 device->size = data->disk_size;
643 device->cylinders = data->cylinders;
644 device->tracks = data->tracks;
645 device->sectors = data->sectors;
646 device->bytes_per_sector = data->bytes_per_sector;
647 pwe->max_disk = data->max_disk;
648 }
649 complete(&pwe->com);
650 break;
651 case vioblockclose:
652 break;
653 case vioblockread:
654 case vioblockwrite:
655 viodasd_handle_read_write(bevent);
656 break;
657
658 default:
659 printk(VIOD_KERN_WARNING "invalid subtype!");
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100660 if (hvlpevent_need_ack(event)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 event->xRc = HvLpEvent_Rc_InvalidSubtype;
662 HvCallEvent_ackLpEvent(event);
663 }
664 }
665}
666
667/*
668 * Get the driver to reprobe for more disks.
669 */
670static ssize_t probe_disks(struct device_driver *drv, const char *buf,
671 size_t count)
672{
673 struct viodasd_device *d;
674
675 for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
676 if (d->disk == NULL)
677 probe_disk(d);
678 }
679 return count;
680}
681static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);
682
683static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
684{
685 struct viodasd_device *d = &viodasd_devices[vdev->unit_address];
686
687 d->dev = &vdev->dev;
Stephen Rothwell8251b4c2007-10-11 14:59:54 +1000688 if (!probe_disk(d))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 return -ENODEV;
690 return 0;
691}
692
693static int viodasd_remove(struct vio_dev *vdev)
694{
695 struct viodasd_device *d;
696
697 d = &viodasd_devices[vdev->unit_address];
698 if (d->disk) {
699 del_gendisk(d->disk);
700 blk_cleanup_queue(d->disk->queue);
701 put_disk(d->disk);
702 d->disk = NULL;
703 }
704 d->dev = NULL;
705 return 0;
706}
707
708/**
709 * viodasd_device_table: Used by vio.c to match devices that we
710 * support.
711 */
712static struct vio_device_id viodasd_device_table[] __devinitdata = {
Stephen Rothwellfbabeb62006-05-15 13:41:22 +1000713 { "block", "IBM,iSeries-viodasd" },
Stephen Rothwellfb120da2005-08-17 16:42:59 +1000714 { "", "" }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716MODULE_DEVICE_TABLE(vio, viodasd_device_table);
Stephen Rothwell915124d2005-10-24 15:12:22 +1000717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718static struct vio_driver viodasd_driver = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 .id_table = viodasd_device_table,
720 .probe = viodasd_probe,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +1000721 .remove = viodasd_remove,
722 .driver = {
723 .name = "viodasd",
Stephen Rothwell915124d2005-10-24 15:12:22 +1000724 .owner = THIS_MODULE,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +1000725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726};
727
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100728static int need_delete_probe;
729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730/*
731 * Initialize the whole device driver. Handle module and non-module
732 * versions
733 */
734static int __init viodasd_init(void)
735{
736 int rc;
737
Stephen Rothwellfb8b5002006-12-15 15:40:08 +1100738 if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
739 rc = -ENODEV;
740 goto early_fail;
741 }
742
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 /* Try to open to our host lp */
744 if (viopath_hostLp == HvLpIndexInvalid)
745 vio_set_hostlp();
746
747 if (viopath_hostLp == HvLpIndexInvalid) {
748 printk(VIOD_KERN_WARNING "invalid hosting partition\n");
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100749 rc = -EIO;
750 goto early_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 }
752
753 printk(VIOD_KERN_INFO "vers " VIOD_VERS ", hosting partition %d\n",
754 viopath_hostLp);
755
756 /* register the block device */
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100757 rc = register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
758 if (rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 printk(VIOD_KERN_WARNING
760 "Unable to get major number %d for %s\n",
761 VIODASD_MAJOR, VIOD_GENHD_NAME);
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100762 goto early_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 }
764 /* Actually open the path to the hosting partition */
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100765 rc = viopath_open(viopath_hostLp, viomajorsubtype_blockio,
766 VIOMAXREQ + 2);
767 if (rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 printk(VIOD_KERN_WARNING
769 "error opening path to host partition %d\n",
770 viopath_hostLp);
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100771 goto unregister_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 }
773
774 /* Initialize our request handler */
775 vio_setHandler(viomajorsubtype_blockio, handle_block_event);
776
777 rc = vio_register_driver(&viodasd_driver);
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100778 if (rc) {
779 printk(VIOD_KERN_WARNING "vio_register_driver failed\n");
780 goto unset_handler;
781 }
782
783 /*
784 * If this call fails, it just means that we cannot dynamically
785 * add virtual disks, but the driver will still work fine for
786 * all existing disk, so ignore the failure.
787 */
788 if (!driver_create_file(&viodasd_driver.driver, &driver_attr_probe))
789 need_delete_probe = 1;
790
791 return 0;
792
793unset_handler:
794 vio_clearHandler(viomajorsubtype_blockio);
795 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
796unregister_blk:
797 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
798early_fail:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return rc;
800}
801module_init(viodasd_init);
802
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100803void __exit viodasd_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804{
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100805 if (need_delete_probe)
806 driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 vio_unregister_driver(&viodasd_driver);
808 vio_clearHandler(viomajorsubtype_blockio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
Stephen Rothwellf9df68e2006-11-13 14:43:17 +1100810 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812module_exit(viodasd_exit);