Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* -*- 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> | |
8962cadb | 6 | * Stephen Rothwell |
1da177e4 LT |
7 | * |
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 | */ | |
dc942cee JP |
31 | |
32 | #define pr_fmt(fmt) "viod: " fmt | |
33 | ||
1da177e4 LT |
34 | #include <linux/major.h> |
35 | #include <linux/fs.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> | |
2a48fc0a | 44 | #include <linux/mutex.h> |
1da177e4 LT |
45 | #include <linux/dma-mapping.h> |
46 | #include <linux/completion.h> | |
47 | #include <linux/device.h> | |
45711f1a | 48 | #include <linux/scatterlist.h> |
1da177e4 LT |
49 | |
50 | #include <asm/uaccess.h> | |
51 | #include <asm/vio.h> | |
1ec65d76 | 52 | #include <asm/iseries/hv_types.h> |
e45423ea | 53 | #include <asm/iseries/hv_lp_event.h> |
15b17189 | 54 | #include <asm/iseries/hv_lp_config.h> |
b4206778 | 55 | #include <asm/iseries/vio.h> |
fb8b5007 | 56 | #include <asm/firmware.h> |
1da177e4 LT |
57 | |
58 | MODULE_DESCRIPTION("iSeries Virtual DASD"); | |
59 | MODULE_AUTHOR("Dave Boutcher"); | |
60 | MODULE_LICENSE("GPL"); | |
61 | ||
62 | /* | |
63 | * We only support 7 partitions per physical disk....so with minor | |
64 | * numbers 0-255 we get a maximum of 32 disks. | |
65 | */ | |
66 | #define VIOD_GENHD_NAME "iseries/vd" | |
1da177e4 LT |
67 | |
68 | #define VIOD_VERS "1.64" | |
69 | ||
1da177e4 LT |
70 | enum { |
71 | PARTITION_SHIFT = 3, | |
72 | MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS, | |
ea6728c1 | 73 | MAX_DISK_NAME = FIELD_SIZEOF(struct gendisk, disk_name) |
1da177e4 LT |
74 | }; |
75 | ||
2a48fc0a | 76 | static DEFINE_MUTEX(viodasd_mutex); |
1da177e4 LT |
77 | static DEFINE_SPINLOCK(viodasd_spinlock); |
78 | ||
79 | #define VIOMAXREQ 16 | |
1da177e4 LT |
80 | |
81 | #define DEVICE_NO(cell) ((struct viodasd_device *)(cell) - &viodasd_devices[0]) | |
82 | ||
1da177e4 LT |
83 | struct viodasd_waitevent { |
84 | struct completion com; | |
85 | int rc; | |
86 | u16 sub_result; | |
87 | int max_disk; /* open */ | |
88 | }; | |
89 | ||
90 | static const struct vio_error_entry viodasd_err_table[] = { | |
91 | { 0x0201, EINVAL, "Invalid Range" }, | |
92 | { 0x0202, EINVAL, "Invalid Token" }, | |
93 | { 0x0203, EIO, "DMA Error" }, | |
94 | { 0x0204, EIO, "Use Error" }, | |
95 | { 0x0205, EIO, "Release Error" }, | |
96 | { 0x0206, EINVAL, "Invalid Disk" }, | |
97 | { 0x0207, EBUSY, "Cant Lock" }, | |
98 | { 0x0208, EIO, "Already Locked" }, | |
99 | { 0x0209, EIO, "Already Unlocked" }, | |
100 | { 0x020A, EIO, "Invalid Arg" }, | |
101 | { 0x020B, EIO, "Bad IFS File" }, | |
102 | { 0x020C, EROFS, "Read Only Device" }, | |
103 | { 0x02FF, EIO, "Internal Error" }, | |
104 | { 0x0000, 0, NULL }, | |
105 | }; | |
106 | ||
107 | /* | |
108 | * Figure out the biggest I/O request (in sectors) we can accept | |
109 | */ | |
110 | #define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA) | |
111 | ||
112 | /* | |
113 | * Number of disk I/O requests we've sent to OS/400 | |
114 | */ | |
115 | static int num_req_outstanding; | |
116 | ||
117 | /* | |
118 | * This is our internal structure for keeping track of disk devices | |
119 | */ | |
120 | struct viodasd_device { | |
121 | u16 cylinders; | |
122 | u16 tracks; | |
123 | u16 sectors; | |
124 | u16 bytes_per_sector; | |
125 | u64 size; | |
126 | int read_only; | |
127 | spinlock_t q_lock; | |
128 | struct gendisk *disk; | |
129 | struct device *dev; | |
130 | } viodasd_devices[MAX_DISKNO]; | |
131 | ||
132 | /* | |
133 | * External open entry point. | |
134 | */ | |
f115a14a | 135 | static int viodasd_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 136 | { |
f115a14a | 137 | struct viodasd_device *d = bdev->bd_disk->private_data; |
1da177e4 LT |
138 | HvLpEvent_Rc hvrc; |
139 | struct viodasd_waitevent we; | |
140 | u16 flags = 0; | |
141 | ||
142 | if (d->read_only) { | |
f115a14a | 143 | if (mode & FMODE_WRITE) |
1da177e4 LT |
144 | return -EROFS; |
145 | flags = vioblockflags_ro; | |
146 | } | |
147 | ||
148 | init_completion(&we.com); | |
149 | ||
150 | /* Send the open event to OS/400 */ | |
151 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, | |
152 | HvLpEvent_Type_VirtualIo, | |
153 | viomajorsubtype_blockio | vioblockopen, | |
154 | HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck, | |
155 | viopath_sourceinst(viopath_hostLp), | |
156 | viopath_targetinst(viopath_hostLp), | |
157 | (u64)(unsigned long)&we, VIOVERSION << 16, | |
158 | ((u64)DEVICE_NO(d) << 48) | ((u64)flags << 32), | |
159 | 0, 0, 0); | |
160 | if (hvrc != 0) { | |
dc942cee | 161 | pr_warning("HV open failed %d\n", (int)hvrc); |
1da177e4 LT |
162 | return -EIO; |
163 | } | |
164 | ||
165 | wait_for_completion(&we.com); | |
166 | ||
167 | /* Check the return code */ | |
168 | if (we.rc != 0) { | |
169 | const struct vio_error_entry *err = | |
170 | vio_lookup_rc(viodasd_err_table, we.sub_result); | |
171 | ||
dc942cee JP |
172 | pr_warning("bad rc opening disk: %d:0x%04x (%s)\n", |
173 | (int)we.rc, we.sub_result, err->msg); | |
1da177e4 LT |
174 | return -EIO; |
175 | } | |
176 | ||
177 | return 0; | |
178 | } | |
179 | ||
6e9624b8 AB |
180 | static int viodasd_unlocked_open(struct block_device *bdev, fmode_t mode) |
181 | { | |
182 | int ret; | |
183 | ||
2a48fc0a | 184 | mutex_lock(&viodasd_mutex); |
6e9624b8 | 185 | ret = viodasd_open(bdev, mode); |
2a48fc0a | 186 | mutex_unlock(&viodasd_mutex); |
6e9624b8 AB |
187 | |
188 | return ret; | |
189 | } | |
190 | ||
191 | ||
1da177e4 LT |
192 | /* |
193 | * External release entry point. | |
194 | */ | |
f115a14a | 195 | static int viodasd_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 196 | { |
f115a14a | 197 | struct viodasd_device *d = disk->private_data; |
1da177e4 LT |
198 | HvLpEvent_Rc hvrc; |
199 | ||
2a48fc0a | 200 | mutex_lock(&viodasd_mutex); |
1da177e4 LT |
201 | /* Send the event to OS/400. We DON'T expect a response */ |
202 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, | |
203 | HvLpEvent_Type_VirtualIo, | |
204 | viomajorsubtype_blockio | vioblockclose, | |
205 | HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck, | |
206 | viopath_sourceinst(viopath_hostLp), | |
207 | viopath_targetinst(viopath_hostLp), | |
208 | 0, VIOVERSION << 16, | |
209 | ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */, | |
210 | 0, 0, 0); | |
211 | if (hvrc != 0) | |
dc942cee | 212 | pr_warning("HV close call failed %d\n", (int)hvrc); |
6e9624b8 | 213 | |
2a48fc0a | 214 | mutex_unlock(&viodasd_mutex); |
6e9624b8 | 215 | |
1da177e4 LT |
216 | return 0; |
217 | } | |
218 | ||
219 | ||
220 | /* External ioctl entry point. | |
221 | */ | |
a885c8c4 | 222 | static int viodasd_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
1da177e4 | 223 | { |
a885c8c4 CH |
224 | struct gendisk *disk = bdev->bd_disk; |
225 | struct viodasd_device *d = disk->private_data; | |
226 | ||
36a70030 | 227 | geo->sectors = d->sectors ? d->sectors : 32; |
a885c8c4 CH |
228 | geo->heads = d->tracks ? d->tracks : 64; |
229 | geo->cylinders = d->cylinders ? d->cylinders : | |
36a70030 | 230 | get_capacity(disk) / (geo->sectors * geo->heads); |
1da177e4 | 231 | |
a885c8c4 | 232 | return 0; |
1da177e4 LT |
233 | } |
234 | ||
235 | /* | |
236 | * Our file operations table | |
237 | */ | |
83d5cde4 | 238 | static const struct block_device_operations viodasd_fops = { |
1da177e4 | 239 | .owner = THIS_MODULE, |
6e9624b8 | 240 | .open = viodasd_unlocked_open, |
f115a14a | 241 | .release = viodasd_release, |
a885c8c4 | 242 | .getgeo = viodasd_getgeo, |
1da177e4 LT |
243 | }; |
244 | ||
245 | /* | |
246 | * End a request | |
247 | */ | |
b2aec24e | 248 | static void viodasd_end_request(struct request *req, int error, |
1da177e4 LT |
249 | int num_sectors) |
250 | { | |
b2aec24e | 251 | __blk_end_request(req, error, num_sectors << 9); |
1da177e4 LT |
252 | } |
253 | ||
254 | /* | |
255 | * Send an actual I/O request to OS/400 | |
256 | */ | |
257 | static int send_request(struct request *req) | |
258 | { | |
259 | u64 start; | |
260 | int direction; | |
261 | int nsg; | |
262 | u16 viocmd; | |
263 | HvLpEvent_Rc hvrc; | |
264 | struct vioblocklpevent *bevent; | |
677f8c0d | 265 | struct HvLpEvent *hev; |
1da177e4 LT |
266 | struct scatterlist sg[VIOMAXBLOCKDMA]; |
267 | int sgindex; | |
1da177e4 LT |
268 | struct viodasd_device *d; |
269 | unsigned long flags; | |
270 | ||
83096ebf | 271 | start = (u64)blk_rq_pos(req) << 9; |
1da177e4 LT |
272 | |
273 | if (rq_data_dir(req) == READ) { | |
274 | direction = DMA_FROM_DEVICE; | |
275 | viocmd = viomajorsubtype_blockio | vioblockread; | |
1da177e4 LT |
276 | } else { |
277 | direction = DMA_TO_DEVICE; | |
278 | viocmd = viomajorsubtype_blockio | vioblockwrite; | |
1da177e4 LT |
279 | } |
280 | ||
281 | d = req->rq_disk->private_data; | |
282 | ||
283 | /* Now build the scatter-gather list */ | |
45711f1a | 284 | sg_init_table(sg, VIOMAXBLOCKDMA); |
1da177e4 LT |
285 | nsg = blk_rq_map_sg(req->q, req, sg); |
286 | nsg = dma_map_sg(d->dev, sg, nsg, direction); | |
287 | ||
288 | spin_lock_irqsave(&viodasd_spinlock, flags); | |
289 | num_req_outstanding++; | |
290 | ||
291 | /* This optimization handles a single DMA block */ | |
292 | if (nsg == 1) | |
293 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, | |
294 | HvLpEvent_Type_VirtualIo, viocmd, | |
295 | HvLpEvent_AckInd_DoAck, | |
296 | HvLpEvent_AckType_ImmediateAck, | |
297 | viopath_sourceinst(viopath_hostLp), | |
298 | viopath_targetinst(viopath_hostLp), | |
299 | (u64)(unsigned long)req, VIOVERSION << 16, | |
300 | ((u64)DEVICE_NO(d) << 48), start, | |
301 | ((u64)sg_dma_address(&sg[0])) << 32, | |
302 | sg_dma_len(&sg[0])); | |
303 | else { | |
304 | bevent = (struct vioblocklpevent *) | |
305 | vio_get_event_buffer(viomajorsubtype_blockio); | |
306 | if (bevent == NULL) { | |
dc942cee | 307 | pr_warning("error allocating disk event buffer\n"); |
1da177e4 LT |
308 | goto error_ret; |
309 | } | |
310 | ||
311 | /* | |
312 | * Now build up the actual request. Note that we store | |
313 | * the pointer to the request in the correlation | |
314 | * token so we can match the response up later | |
315 | */ | |
316 | memset(bevent, 0, sizeof(struct vioblocklpevent)); | |
677f8c0d SR |
317 | hev = &bevent->event; |
318 | hev->flags = HV_LP_EVENT_VALID | HV_LP_EVENT_DO_ACK | | |
319 | HV_LP_EVENT_INT; | |
320 | hev->xType = HvLpEvent_Type_VirtualIo; | |
321 | hev->xSubtype = viocmd; | |
322 | hev->xSourceLp = HvLpConfig_getLpIndex(); | |
323 | hev->xTargetLp = viopath_hostLp; | |
324 | hev->xSizeMinus1 = | |
1da177e4 LT |
325 | offsetof(struct vioblocklpevent, u.rw_data.dma_info) + |
326 | (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1; | |
677f8c0d SR |
327 | hev->xSourceInstanceId = viopath_sourceinst(viopath_hostLp); |
328 | hev->xTargetInstanceId = viopath_targetinst(viopath_hostLp); | |
329 | hev->xCorrelationToken = (u64)req; | |
1da177e4 LT |
330 | bevent->version = VIOVERSION; |
331 | bevent->disk = DEVICE_NO(d); | |
332 | bevent->u.rw_data.offset = start; | |
333 | ||
334 | /* | |
335 | * Copy just the dma information from the sg list | |
336 | * into the request | |
337 | */ | |
338 | for (sgindex = 0; sgindex < nsg; sgindex++) { | |
339 | bevent->u.rw_data.dma_info[sgindex].token = | |
340 | sg_dma_address(&sg[sgindex]); | |
341 | bevent->u.rw_data.dma_info[sgindex].len = | |
342 | sg_dma_len(&sg[sgindex]); | |
343 | } | |
344 | ||
345 | /* Send the request */ | |
346 | hvrc = HvCallEvent_signalLpEvent(&bevent->event); | |
347 | vio_free_event_buffer(viomajorsubtype_blockio, bevent); | |
348 | } | |
349 | ||
350 | if (hvrc != HvLpEvent_Rc_Good) { | |
dc942cee JP |
351 | pr_warning("error sending disk event to OS/400 (rc %d)\n", |
352 | (int)hvrc); | |
1da177e4 LT |
353 | goto error_ret; |
354 | } | |
355 | spin_unlock_irqrestore(&viodasd_spinlock, flags); | |
356 | return 0; | |
357 | ||
358 | error_ret: | |
359 | num_req_outstanding--; | |
360 | spin_unlock_irqrestore(&viodasd_spinlock, flags); | |
361 | dma_unmap_sg(d->dev, sg, nsg, direction); | |
362 | return -1; | |
363 | } | |
364 | ||
365 | /* | |
366 | * This is the external request processing routine | |
367 | */ | |
165125e1 | 368 | static void do_viodasd_request(struct request_queue *q) |
1da177e4 LT |
369 | { |
370 | struct request *req; | |
371 | ||
372 | /* | |
373 | * If we already have the maximum number of requests | |
374 | * outstanding to OS/400 just bail out. We'll come | |
375 | * back later. | |
376 | */ | |
377 | while (num_req_outstanding < VIOMAXREQ) { | |
9934c8c0 | 378 | req = blk_fetch_request(q); |
1da177e4 LT |
379 | if (req == NULL) |
380 | return; | |
1da177e4 | 381 | /* check that request contains a valid command */ |
33659ebb | 382 | if (req->cmd_type != REQ_TYPE_FS) { |
5b93629b | 383 | viodasd_end_request(req, -EIO, blk_rq_sectors(req)); |
1da177e4 LT |
384 | continue; |
385 | } | |
386 | /* Try sending the request */ | |
387 | if (send_request(req) != 0) | |
5b93629b | 388 | viodasd_end_request(req, -EIO, blk_rq_sectors(req)); |
1da177e4 LT |
389 | } |
390 | } | |
391 | ||
392 | /* | |
393 | * Probe a single disk and fill in the viodasd_device structure | |
394 | * for it. | |
395 | */ | |
8251b4c4 | 396 | static int probe_disk(struct viodasd_device *d) |
1da177e4 LT |
397 | { |
398 | HvLpEvent_Rc hvrc; | |
399 | struct viodasd_waitevent we; | |
400 | int dev_no = DEVICE_NO(d); | |
401 | struct gendisk *g; | |
402 | struct request_queue *q; | |
403 | u16 flags = 0; | |
404 | ||
405 | retry: | |
406 | init_completion(&we.com); | |
407 | ||
408 | /* Send the open event to OS/400 */ | |
409 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, | |
410 | HvLpEvent_Type_VirtualIo, | |
411 | viomajorsubtype_blockio | vioblockopen, | |
412 | HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck, | |
413 | viopath_sourceinst(viopath_hostLp), | |
414 | viopath_targetinst(viopath_hostLp), | |
415 | (u64)(unsigned long)&we, VIOVERSION << 16, | |
416 | ((u64)dev_no << 48) | ((u64)flags<< 32), | |
417 | 0, 0, 0); | |
418 | if (hvrc != 0) { | |
dc942cee | 419 | pr_warning("bad rc on HV open %d\n", (int)hvrc); |
8251b4c4 | 420 | return 0; |
1da177e4 LT |
421 | } |
422 | ||
423 | wait_for_completion(&we.com); | |
424 | ||
425 | if (we.rc != 0) { | |
426 | if (flags != 0) | |
8251b4c4 | 427 | return 0; |
1da177e4 LT |
428 | /* try again with read only flag set */ |
429 | flags = vioblockflags_ro; | |
430 | goto retry; | |
431 | } | |
432 | if (we.max_disk > (MAX_DISKNO - 1)) { | |
dc942cee JP |
433 | printk_once(KERN_INFO pr_fmt("Only examining the first %d of %d disks connected\n"), |
434 | MAX_DISKNO, we.max_disk + 1); | |
1da177e4 LT |
435 | } |
436 | ||
437 | /* Send the close event to OS/400. We DON'T expect a response */ | |
438 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, | |
439 | HvLpEvent_Type_VirtualIo, | |
440 | viomajorsubtype_blockio | vioblockclose, | |
441 | HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck, | |
442 | viopath_sourceinst(viopath_hostLp), | |
443 | viopath_targetinst(viopath_hostLp), | |
444 | 0, VIOVERSION << 16, | |
445 | ((u64)dev_no << 48) | ((u64)flags << 32), | |
446 | 0, 0, 0); | |
447 | if (hvrc != 0) { | |
dc942cee | 448 | pr_warning("bad rc sending event to OS/400 %d\n", (int)hvrc); |
8251b4c4 | 449 | return 0; |
1da177e4 | 450 | } |
8251b4c4 SR |
451 | |
452 | if (d->dev == NULL) { | |
453 | /* this is when we reprobe for new disks */ | |
454 | if (vio_create_viodasd(dev_no) == NULL) { | |
dc942cee JP |
455 | pr_warning("cannot allocate virtual device for disk %d\n", |
456 | dev_no); | |
8251b4c4 SR |
457 | return 0; |
458 | } | |
459 | /* | |
460 | * The vio_create_viodasd will have recursed into this | |
461 | * routine with d->dev set to the new vio device and | |
462 | * will finish the setup of the disk below. | |
463 | */ | |
464 | return 1; | |
465 | } | |
466 | ||
1da177e4 LT |
467 | /* create the request queue for the disk */ |
468 | spin_lock_init(&d->q_lock); | |
469 | q = blk_init_queue(do_viodasd_request, &d->q_lock); | |
470 | if (q == NULL) { | |
dc942cee | 471 | pr_warning("cannot allocate queue for disk %d\n", dev_no); |
8251b4c4 | 472 | return 0; |
1da177e4 LT |
473 | } |
474 | g = alloc_disk(1 << PARTITION_SHIFT); | |
475 | if (g == NULL) { | |
dc942cee JP |
476 | pr_warning("cannot allocate disk structure for disk %d\n", |
477 | dev_no); | |
1da177e4 | 478 | blk_cleanup_queue(q); |
8251b4c4 | 479 | return 0; |
1da177e4 LT |
480 | } |
481 | ||
482 | d->disk = g; | |
8a78362c | 483 | blk_queue_max_segments(q, VIOMAXBLOCKDMA); |
086fa5ff | 484 | blk_queue_max_hw_sectors(q, VIODASD_MAXSECTORS); |
1da177e4 LT |
485 | g->major = VIODASD_MAJOR; |
486 | g->first_minor = dev_no << PARTITION_SHIFT; | |
487 | if (dev_no >= 26) | |
488 | snprintf(g->disk_name, sizeof(g->disk_name), | |
489 | VIOD_GENHD_NAME "%c%c", | |
490 | 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26)); | |
491 | else | |
492 | snprintf(g->disk_name, sizeof(g->disk_name), | |
493 | VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26)); | |
1da177e4 LT |
494 | g->fops = &viodasd_fops; |
495 | g->queue = q; | |
496 | g->private_data = d; | |
497 | g->driverfs_dev = d->dev; | |
498 | set_capacity(g, d->size >> 9); | |
499 | ||
dc942cee JP |
500 | pr_info("disk %d: %lu sectors (%lu MB) CHS=%d/%d/%d sector size %d%s\n", |
501 | dev_no, (unsigned long)(d->size >> 9), | |
502 | (unsigned long)(d->size >> 20), | |
503 | (int)d->cylinders, (int)d->tracks, | |
504 | (int)d->sectors, (int)d->bytes_per_sector, | |
505 | d->read_only ? " (RO)" : ""); | |
1da177e4 LT |
506 | |
507 | /* register us in the global list */ | |
508 | add_disk(g); | |
8251b4c4 | 509 | return 1; |
1da177e4 LT |
510 | } |
511 | ||
512 | /* returns the total number of scatterlist elements converted */ | |
513 | static int block_event_to_scatterlist(const struct vioblocklpevent *bevent, | |
514 | struct scatterlist *sg, int *total_len) | |
515 | { | |
516 | int i, numsg; | |
517 | const struct rw_data *rw_data = &bevent->u.rw_data; | |
518 | static const int offset = | |
519 | offsetof(struct vioblocklpevent, u.rw_data.dma_info); | |
520 | static const int element_size = sizeof(rw_data->dma_info[0]); | |
521 | ||
522 | numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size; | |
523 | if (numsg > VIOMAXBLOCKDMA) | |
524 | numsg = VIOMAXBLOCKDMA; | |
525 | ||
526 | *total_len = 0; | |
25c0a7b8 | 527 | sg_init_table(sg, VIOMAXBLOCKDMA); |
1da177e4 LT |
528 | for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) { |
529 | sg_dma_address(&sg[i]) = rw_data->dma_info[i].token; | |
530 | sg_dma_len(&sg[i]) = rw_data->dma_info[i].len; | |
531 | *total_len += rw_data->dma_info[i].len; | |
532 | } | |
533 | return i; | |
534 | } | |
535 | ||
536 | /* | |
537 | * Restart all queues, starting with the one _after_ the disk given, | |
538 | * thus reducing the chance of starvation of higher numbered disks. | |
539 | */ | |
540 | static void viodasd_restart_all_queues_starting_from(int first_index) | |
541 | { | |
542 | int i; | |
543 | ||
544 | for (i = first_index + 1; i < MAX_DISKNO; ++i) | |
545 | if (viodasd_devices[i].disk) | |
546 | blk_run_queue(viodasd_devices[i].disk->queue); | |
547 | for (i = 0; i <= first_index; ++i) | |
548 | if (viodasd_devices[i].disk) | |
549 | blk_run_queue(viodasd_devices[i].disk->queue); | |
550 | } | |
551 | ||
552 | /* | |
553 | * For read and write requests, decrement the number of outstanding requests, | |
554 | * Free the DMA buffers we allocated. | |
555 | */ | |
556 | static int viodasd_handle_read_write(struct vioblocklpevent *bevent) | |
557 | { | |
558 | int num_sg, num_sect, pci_direction, total_len; | |
559 | struct request *req; | |
560 | struct scatterlist sg[VIOMAXBLOCKDMA]; | |
561 | struct HvLpEvent *event = &bevent->event; | |
562 | unsigned long irq_flags; | |
563 | struct viodasd_device *d; | |
564 | int error; | |
565 | spinlock_t *qlock; | |
566 | ||
567 | num_sg = block_event_to_scatterlist(bevent, sg, &total_len); | |
568 | num_sect = total_len >> 9; | |
569 | if (event->xSubtype == (viomajorsubtype_blockio | vioblockread)) | |
570 | pci_direction = DMA_FROM_DEVICE; | |
571 | else | |
572 | pci_direction = DMA_TO_DEVICE; | |
573 | req = (struct request *)bevent->event.xCorrelationToken; | |
574 | d = req->rq_disk->private_data; | |
575 | ||
576 | dma_unmap_sg(d->dev, sg, num_sg, pci_direction); | |
577 | ||
578 | /* | |
579 | * Since this is running in interrupt mode, we need to make sure | |
580 | * we're not stepping on any global I/O operations | |
581 | */ | |
582 | spin_lock_irqsave(&viodasd_spinlock, irq_flags); | |
583 | num_req_outstanding--; | |
584 | spin_unlock_irqrestore(&viodasd_spinlock, irq_flags); | |
585 | ||
b2aec24e | 586 | error = (event->xRc == HvLpEvent_Rc_Good) ? 0 : -EIO; |
1da177e4 LT |
587 | if (error) { |
588 | const struct vio_error_entry *err; | |
589 | err = vio_lookup_rc(viodasd_err_table, bevent->sub_result); | |
dc942cee JP |
590 | pr_warning("read/write error %d:0x%04x (%s)\n", |
591 | event->xRc, bevent->sub_result, err->msg); | |
5b93629b | 592 | num_sect = blk_rq_sectors(req); |
1da177e4 LT |
593 | } |
594 | qlock = req->q->queue_lock; | |
595 | spin_lock_irqsave(qlock, irq_flags); | |
b2aec24e | 596 | viodasd_end_request(req, error, num_sect); |
1da177e4 LT |
597 | spin_unlock_irqrestore(qlock, irq_flags); |
598 | ||
599 | /* Finally, try to get more requests off of this device's queue */ | |
600 | viodasd_restart_all_queues_starting_from(DEVICE_NO(d)); | |
601 | ||
602 | return 0; | |
603 | } | |
604 | ||
605 | /* This routine handles incoming block LP events */ | |
606 | static void handle_block_event(struct HvLpEvent *event) | |
607 | { | |
608 | struct vioblocklpevent *bevent = (struct vioblocklpevent *)event; | |
609 | struct viodasd_waitevent *pwe; | |
610 | ||
611 | if (event == NULL) | |
612 | /* Notification that a partition went away! */ | |
613 | return; | |
614 | /* First, we should NEVER get an int here...only acks */ | |
677f8c0d | 615 | if (hvlpevent_is_int(event)) { |
dc942cee | 616 | pr_warning("Yikes! got an int in viodasd event handler!\n"); |
677f8c0d | 617 | if (hvlpevent_need_ack(event)) { |
1da177e4 LT |
618 | 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: | |
dc942cee | 659 | pr_warning("invalid subtype!"); |
677f8c0d | 660 | if (hvlpevent_need_ack(event)) { |
1da177e4 LT |
661 | event->xRc = HvLpEvent_Rc_InvalidSubtype; |
662 | HvCallEvent_ackLpEvent(event); | |
663 | } | |
664 | } | |
665 | } | |
666 | ||
667 | /* | |
668 | * Get the driver to reprobe for more disks. | |
669 | */ | |
670 | static 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 | } | |
681 | static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks); | |
682 | ||
683 | static 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; | |
8251b4c4 | 688 | if (!probe_disk(d)) |
1da177e4 LT |
689 | return -ENODEV; |
690 | return 0; | |
691 | } | |
692 | ||
693 | static 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 | */ | |
712 | static struct vio_device_id viodasd_device_table[] __devinitdata = { | |
fbabeb60 | 713 | { "block", "IBM,iSeries-viodasd" }, |
fb120da6 | 714 | { "", "" } |
1da177e4 | 715 | }; |
1da177e4 | 716 | MODULE_DEVICE_TABLE(vio, viodasd_device_table); |
915124d8 | 717 | |
1da177e4 | 718 | static struct vio_driver viodasd_driver = { |
1da177e4 LT |
719 | .id_table = viodasd_device_table, |
720 | .probe = viodasd_probe, | |
6fdf5392 SR |
721 | .remove = viodasd_remove, |
722 | .driver = { | |
723 | .name = "viodasd", | |
915124d8 | 724 | .owner = THIS_MODULE, |
6fdf5392 | 725 | } |
1da177e4 LT |
726 | }; |
727 | ||
f9df68ec SR |
728 | static int need_delete_probe; |
729 | ||
1da177e4 LT |
730 | /* |
731 | * Initialize the whole device driver. Handle module and non-module | |
732 | * versions | |
733 | */ | |
734 | static int __init viodasd_init(void) | |
735 | { | |
736 | int rc; | |
737 | ||
fb8b5007 SR |
738 | if (!firmware_has_feature(FW_FEATURE_ISERIES)) { |
739 | rc = -ENODEV; | |
740 | goto early_fail; | |
741 | } | |
742 | ||
1da177e4 LT |
743 | /* Try to open to our host lp */ |
744 | if (viopath_hostLp == HvLpIndexInvalid) | |
745 | vio_set_hostlp(); | |
746 | ||
747 | if (viopath_hostLp == HvLpIndexInvalid) { | |
dc942cee | 748 | pr_warning("invalid hosting partition\n"); |
f9df68ec SR |
749 | rc = -EIO; |
750 | goto early_fail; | |
1da177e4 LT |
751 | } |
752 | ||
dc942cee | 753 | pr_info("vers " VIOD_VERS ", hosting partition %d\n", viopath_hostLp); |
1da177e4 LT |
754 | |
755 | /* register the block device */ | |
f9df68ec SR |
756 | rc = register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME); |
757 | if (rc) { | |
dc942cee JP |
758 | pr_warning("Unable to get major number %d for %s\n", |
759 | VIODASD_MAJOR, VIOD_GENHD_NAME); | |
f9df68ec | 760 | goto early_fail; |
1da177e4 LT |
761 | } |
762 | /* Actually open the path to the hosting partition */ | |
f9df68ec SR |
763 | rc = viopath_open(viopath_hostLp, viomajorsubtype_blockio, |
764 | VIOMAXREQ + 2); | |
765 | if (rc) { | |
dc942cee JP |
766 | pr_warning("error opening path to host partition %d\n", |
767 | viopath_hostLp); | |
f9df68ec | 768 | goto unregister_blk; |
1da177e4 LT |
769 | } |
770 | ||
771 | /* Initialize our request handler */ | |
772 | vio_setHandler(viomajorsubtype_blockio, handle_block_event); | |
773 | ||
774 | rc = vio_register_driver(&viodasd_driver); | |
f9df68ec | 775 | if (rc) { |
dc942cee | 776 | pr_warning("vio_register_driver failed\n"); |
f9df68ec SR |
777 | goto unset_handler; |
778 | } | |
779 | ||
780 | /* | |
781 | * If this call fails, it just means that we cannot dynamically | |
782 | * add virtual disks, but the driver will still work fine for | |
783 | * all existing disk, so ignore the failure. | |
784 | */ | |
785 | if (!driver_create_file(&viodasd_driver.driver, &driver_attr_probe)) | |
786 | need_delete_probe = 1; | |
787 | ||
788 | return 0; | |
789 | ||
790 | unset_handler: | |
791 | vio_clearHandler(viomajorsubtype_blockio); | |
792 | viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2); | |
793 | unregister_blk: | |
794 | unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME); | |
795 | early_fail: | |
1da177e4 LT |
796 | return rc; |
797 | } | |
798 | module_init(viodasd_init); | |
799 | ||
f9df68ec | 800 | void __exit viodasd_exit(void) |
1da177e4 | 801 | { |
f9df68ec SR |
802 | if (need_delete_probe) |
803 | driver_remove_file(&viodasd_driver.driver, &driver_attr_probe); | |
1da177e4 LT |
804 | vio_unregister_driver(&viodasd_driver); |
805 | vio_clearHandler(viomajorsubtype_blockio); | |
1da177e4 | 806 | viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2); |
f9df68ec | 807 | unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME); |
1da177e4 | 808 | } |
1da177e4 | 809 | module_exit(viodasd_exit); |