Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
| 6 | * Stephen Rothwell <sfr@au1.ibm.com> |
| 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 | */ |
| 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> |
| 44 | #include <linux/kernel.h> |
| 45 | |
| 46 | #include <asm/uaccess.h> |
| 47 | #include <asm/vio.h> |
| 48 | #include <asm/iSeries/HvTypes.h> |
| 49 | #include <asm/iSeries/HvLpEvent.h> |
| 50 | #include <asm/iSeries/HvLpConfig.h> |
| 51 | #include <asm/iSeries/vio.h> |
| 52 | |
| 53 | MODULE_DESCRIPTION("iSeries Virtual DASD"); |
| 54 | MODULE_AUTHOR("Dave Boutcher"); |
| 55 | MODULE_LICENSE("GPL"); |
| 56 | |
| 57 | /* |
| 58 | * We only support 7 partitions per physical disk....so with minor |
| 59 | * numbers 0-255 we get a maximum of 32 disks. |
| 60 | */ |
| 61 | #define VIOD_GENHD_NAME "iseries/vd" |
| 62 | #define VIOD_GENHD_DEVFS_NAME "iseries/disc" |
| 63 | |
| 64 | #define VIOD_VERS "1.64" |
| 65 | |
| 66 | #define VIOD_KERN_WARNING KERN_WARNING "viod: " |
| 67 | #define VIOD_KERN_INFO KERN_INFO "viod: " |
| 68 | |
| 69 | enum { |
| 70 | PARTITION_SHIFT = 3, |
| 71 | MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS, |
| 72 | MAX_DISK_NAME = sizeof(((struct gendisk *)0)->disk_name) |
| 73 | }; |
| 74 | |
| 75 | static DEFINE_SPINLOCK(viodasd_spinlock); |
| 76 | |
| 77 | #define VIOMAXREQ 16 |
| 78 | #define VIOMAXBLOCKDMA 12 |
| 79 | |
| 80 | #define DEVICE_NO(cell) ((struct viodasd_device *)(cell) - &viodasd_devices[0]) |
| 81 | |
| 82 | struct open_data { |
| 83 | u64 disk_size; |
| 84 | u16 max_disk; |
| 85 | u16 cylinders; |
| 86 | u16 tracks; |
| 87 | u16 sectors; |
| 88 | u16 bytes_per_sector; |
| 89 | }; |
| 90 | |
| 91 | struct rw_data { |
| 92 | u64 offset; |
| 93 | struct { |
| 94 | u32 token; |
| 95 | u32 reserved; |
| 96 | u64 len; |
| 97 | } dma_info[VIOMAXBLOCKDMA]; |
| 98 | }; |
| 99 | |
| 100 | struct vioblocklpevent { |
| 101 | struct HvLpEvent event; |
| 102 | u32 reserved; |
| 103 | u16 version; |
| 104 | u16 sub_result; |
| 105 | u16 disk; |
| 106 | u16 flags; |
| 107 | union { |
| 108 | struct open_data open_data; |
| 109 | struct rw_data rw_data; |
| 110 | u64 changed; |
| 111 | } u; |
| 112 | }; |
| 113 | |
| 114 | #define vioblockflags_ro 0x0001 |
| 115 | |
| 116 | enum vioblocksubtype { |
| 117 | vioblockopen = 0x0001, |
| 118 | vioblockclose = 0x0002, |
| 119 | vioblockread = 0x0003, |
| 120 | vioblockwrite = 0x0004, |
| 121 | vioblockflush = 0x0005, |
| 122 | vioblockcheck = 0x0007 |
| 123 | }; |
| 124 | |
| 125 | struct viodasd_waitevent { |
| 126 | struct completion com; |
| 127 | int rc; |
| 128 | u16 sub_result; |
| 129 | int max_disk; /* open */ |
| 130 | }; |
| 131 | |
| 132 | static const struct vio_error_entry viodasd_err_table[] = { |
| 133 | { 0x0201, EINVAL, "Invalid Range" }, |
| 134 | { 0x0202, EINVAL, "Invalid Token" }, |
| 135 | { 0x0203, EIO, "DMA Error" }, |
| 136 | { 0x0204, EIO, "Use Error" }, |
| 137 | { 0x0205, EIO, "Release Error" }, |
| 138 | { 0x0206, EINVAL, "Invalid Disk" }, |
| 139 | { 0x0207, EBUSY, "Cant Lock" }, |
| 140 | { 0x0208, EIO, "Already Locked" }, |
| 141 | { 0x0209, EIO, "Already Unlocked" }, |
| 142 | { 0x020A, EIO, "Invalid Arg" }, |
| 143 | { 0x020B, EIO, "Bad IFS File" }, |
| 144 | { 0x020C, EROFS, "Read Only Device" }, |
| 145 | { 0x02FF, EIO, "Internal Error" }, |
| 146 | { 0x0000, 0, NULL }, |
| 147 | }; |
| 148 | |
| 149 | /* |
| 150 | * Figure out the biggest I/O request (in sectors) we can accept |
| 151 | */ |
| 152 | #define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA) |
| 153 | |
| 154 | /* |
| 155 | * Number of disk I/O requests we've sent to OS/400 |
| 156 | */ |
| 157 | static int num_req_outstanding; |
| 158 | |
| 159 | /* |
| 160 | * This is our internal structure for keeping track of disk devices |
| 161 | */ |
| 162 | struct viodasd_device { |
| 163 | u16 cylinders; |
| 164 | u16 tracks; |
| 165 | u16 sectors; |
| 166 | u16 bytes_per_sector; |
| 167 | u64 size; |
| 168 | int read_only; |
| 169 | spinlock_t q_lock; |
| 170 | struct gendisk *disk; |
| 171 | struct device *dev; |
| 172 | } viodasd_devices[MAX_DISKNO]; |
| 173 | |
| 174 | /* |
| 175 | * External open entry point. |
| 176 | */ |
| 177 | static int viodasd_open(struct inode *ino, struct file *fil) |
| 178 | { |
| 179 | struct viodasd_device *d = ino->i_bdev->bd_disk->private_data; |
| 180 | HvLpEvent_Rc hvrc; |
| 181 | struct viodasd_waitevent we; |
| 182 | u16 flags = 0; |
| 183 | |
| 184 | if (d->read_only) { |
| 185 | if ((fil != NULL) && (fil->f_mode & FMODE_WRITE)) |
| 186 | return -EROFS; |
| 187 | flags = vioblockflags_ro; |
| 188 | } |
| 189 | |
| 190 | init_completion(&we.com); |
| 191 | |
| 192 | /* Send the open event to OS/400 */ |
| 193 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, |
| 194 | HvLpEvent_Type_VirtualIo, |
| 195 | viomajorsubtype_blockio | vioblockopen, |
| 196 | HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck, |
| 197 | viopath_sourceinst(viopath_hostLp), |
| 198 | viopath_targetinst(viopath_hostLp), |
| 199 | (u64)(unsigned long)&we, VIOVERSION << 16, |
| 200 | ((u64)DEVICE_NO(d) << 48) | ((u64)flags << 32), |
| 201 | 0, 0, 0); |
| 202 | if (hvrc != 0) { |
| 203 | printk(VIOD_KERN_WARNING "HV open failed %d\n", (int)hvrc); |
| 204 | return -EIO; |
| 205 | } |
| 206 | |
| 207 | wait_for_completion(&we.com); |
| 208 | |
| 209 | /* Check the return code */ |
| 210 | if (we.rc != 0) { |
| 211 | const struct vio_error_entry *err = |
| 212 | vio_lookup_rc(viodasd_err_table, we.sub_result); |
| 213 | |
| 214 | printk(VIOD_KERN_WARNING |
| 215 | "bad rc opening disk: %d:0x%04x (%s)\n", |
| 216 | (int)we.rc, we.sub_result, err->msg); |
| 217 | return -EIO; |
| 218 | } |
| 219 | |
| 220 | return 0; |
| 221 | } |
| 222 | |
| 223 | /* |
| 224 | * External release entry point. |
| 225 | */ |
| 226 | static int viodasd_release(struct inode *ino, struct file *fil) |
| 227 | { |
| 228 | struct viodasd_device *d = ino->i_bdev->bd_disk->private_data; |
| 229 | HvLpEvent_Rc hvrc; |
| 230 | |
| 231 | /* Send the event to OS/400. We DON'T expect a response */ |
| 232 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, |
| 233 | HvLpEvent_Type_VirtualIo, |
| 234 | viomajorsubtype_blockio | vioblockclose, |
| 235 | HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck, |
| 236 | viopath_sourceinst(viopath_hostLp), |
| 237 | viopath_targetinst(viopath_hostLp), |
| 238 | 0, VIOVERSION << 16, |
| 239 | ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */, |
| 240 | 0, 0, 0); |
| 241 | if (hvrc != 0) |
| 242 | printk(VIOD_KERN_WARNING "HV close call failed %d\n", |
| 243 | (int)hvrc); |
| 244 | return 0; |
| 245 | } |
| 246 | |
| 247 | |
| 248 | /* External ioctl entry point. |
| 249 | */ |
| 250 | static int viodasd_ioctl(struct inode *ino, struct file *fil, |
| 251 | unsigned int cmd, unsigned long arg) |
| 252 | { |
| 253 | unsigned char sectors; |
| 254 | unsigned char heads; |
| 255 | unsigned short cylinders; |
| 256 | struct hd_geometry *geo; |
| 257 | struct gendisk *gendisk; |
| 258 | struct viodasd_device *d; |
| 259 | |
| 260 | switch (cmd) { |
| 261 | case HDIO_GETGEO: |
| 262 | geo = (struct hd_geometry *)arg; |
| 263 | if (geo == NULL) |
| 264 | return -EINVAL; |
| 265 | if (!access_ok(VERIFY_WRITE, geo, sizeof(*geo))) |
| 266 | return -EFAULT; |
| 267 | gendisk = ino->i_bdev->bd_disk; |
| 268 | d = gendisk->private_data; |
| 269 | sectors = d->sectors; |
| 270 | if (sectors == 0) |
| 271 | sectors = 32; |
| 272 | heads = d->tracks; |
| 273 | if (heads == 0) |
| 274 | heads = 64; |
| 275 | cylinders = d->cylinders; |
| 276 | if (cylinders == 0) |
| 277 | cylinders = get_capacity(gendisk) / (sectors * heads); |
| 278 | if (__put_user(sectors, &geo->sectors) || |
| 279 | __put_user(heads, &geo->heads) || |
| 280 | __put_user(cylinders, &geo->cylinders) || |
| 281 | __put_user(get_start_sect(ino->i_bdev), &geo->start)) |
| 282 | return -EFAULT; |
| 283 | return 0; |
| 284 | } |
| 285 | |
| 286 | return -EINVAL; |
| 287 | } |
| 288 | |
| 289 | /* |
| 290 | * Our file operations table |
| 291 | */ |
| 292 | static struct block_device_operations viodasd_fops = { |
| 293 | .owner = THIS_MODULE, |
| 294 | .open = viodasd_open, |
| 295 | .release = viodasd_release, |
| 296 | .ioctl = viodasd_ioctl, |
| 297 | }; |
| 298 | |
| 299 | /* |
| 300 | * End a request |
| 301 | */ |
| 302 | static void viodasd_end_request(struct request *req, int uptodate, |
| 303 | int num_sectors) |
| 304 | { |
| 305 | if (end_that_request_first(req, uptodate, num_sectors)) |
| 306 | return; |
| 307 | add_disk_randomness(req->rq_disk); |
| 308 | end_that_request_last(req); |
| 309 | } |
| 310 | |
| 311 | /* |
| 312 | * Send an actual I/O request to OS/400 |
| 313 | */ |
| 314 | static int send_request(struct request *req) |
| 315 | { |
| 316 | u64 start; |
| 317 | int direction; |
| 318 | int nsg; |
| 319 | u16 viocmd; |
| 320 | HvLpEvent_Rc hvrc; |
| 321 | struct vioblocklpevent *bevent; |
| 322 | struct scatterlist sg[VIOMAXBLOCKDMA]; |
| 323 | int sgindex; |
| 324 | int statindex; |
| 325 | struct viodasd_device *d; |
| 326 | unsigned long flags; |
| 327 | |
| 328 | start = (u64)req->sector << 9; |
| 329 | |
| 330 | if (rq_data_dir(req) == READ) { |
| 331 | direction = DMA_FROM_DEVICE; |
| 332 | viocmd = viomajorsubtype_blockio | vioblockread; |
| 333 | statindex = 0; |
| 334 | } else { |
| 335 | direction = DMA_TO_DEVICE; |
| 336 | viocmd = viomajorsubtype_blockio | vioblockwrite; |
| 337 | statindex = 1; |
| 338 | } |
| 339 | |
| 340 | d = req->rq_disk->private_data; |
| 341 | |
| 342 | /* Now build the scatter-gather list */ |
| 343 | nsg = blk_rq_map_sg(req->q, req, sg); |
| 344 | nsg = dma_map_sg(d->dev, sg, nsg, direction); |
| 345 | |
| 346 | spin_lock_irqsave(&viodasd_spinlock, flags); |
| 347 | num_req_outstanding++; |
| 348 | |
| 349 | /* This optimization handles a single DMA block */ |
| 350 | if (nsg == 1) |
| 351 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, |
| 352 | HvLpEvent_Type_VirtualIo, viocmd, |
| 353 | HvLpEvent_AckInd_DoAck, |
| 354 | HvLpEvent_AckType_ImmediateAck, |
| 355 | viopath_sourceinst(viopath_hostLp), |
| 356 | viopath_targetinst(viopath_hostLp), |
| 357 | (u64)(unsigned long)req, VIOVERSION << 16, |
| 358 | ((u64)DEVICE_NO(d) << 48), start, |
| 359 | ((u64)sg_dma_address(&sg[0])) << 32, |
| 360 | sg_dma_len(&sg[0])); |
| 361 | else { |
| 362 | bevent = (struct vioblocklpevent *) |
| 363 | vio_get_event_buffer(viomajorsubtype_blockio); |
| 364 | if (bevent == NULL) { |
| 365 | printk(VIOD_KERN_WARNING |
| 366 | "error allocating disk event buffer\n"); |
| 367 | goto error_ret; |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * Now build up the actual request. Note that we store |
| 372 | * the pointer to the request in the correlation |
| 373 | * token so we can match the response up later |
| 374 | */ |
| 375 | memset(bevent, 0, sizeof(struct vioblocklpevent)); |
| 376 | bevent->event.xFlags.xValid = 1; |
| 377 | bevent->event.xFlags.xFunction = HvLpEvent_Function_Int; |
| 378 | bevent->event.xFlags.xAckInd = HvLpEvent_AckInd_DoAck; |
| 379 | bevent->event.xFlags.xAckType = HvLpEvent_AckType_ImmediateAck; |
| 380 | bevent->event.xType = HvLpEvent_Type_VirtualIo; |
| 381 | bevent->event.xSubtype = viocmd; |
| 382 | bevent->event.xSourceLp = HvLpConfig_getLpIndex(); |
| 383 | bevent->event.xTargetLp = viopath_hostLp; |
| 384 | bevent->event.xSizeMinus1 = |
| 385 | offsetof(struct vioblocklpevent, u.rw_data.dma_info) + |
| 386 | (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1; |
| 387 | bevent->event.xSourceInstanceId = |
| 388 | viopath_sourceinst(viopath_hostLp); |
| 389 | bevent->event.xTargetInstanceId = |
| 390 | viopath_targetinst(viopath_hostLp); |
| 391 | bevent->event.xCorrelationToken = (u64)req; |
| 392 | bevent->version = VIOVERSION; |
| 393 | bevent->disk = DEVICE_NO(d); |
| 394 | bevent->u.rw_data.offset = start; |
| 395 | |
| 396 | /* |
| 397 | * Copy just the dma information from the sg list |
| 398 | * into the request |
| 399 | */ |
| 400 | for (sgindex = 0; sgindex < nsg; sgindex++) { |
| 401 | bevent->u.rw_data.dma_info[sgindex].token = |
| 402 | sg_dma_address(&sg[sgindex]); |
| 403 | bevent->u.rw_data.dma_info[sgindex].len = |
| 404 | sg_dma_len(&sg[sgindex]); |
| 405 | } |
| 406 | |
| 407 | /* Send the request */ |
| 408 | hvrc = HvCallEvent_signalLpEvent(&bevent->event); |
| 409 | vio_free_event_buffer(viomajorsubtype_blockio, bevent); |
| 410 | } |
| 411 | |
| 412 | if (hvrc != HvLpEvent_Rc_Good) { |
| 413 | printk(VIOD_KERN_WARNING |
| 414 | "error sending disk event to OS/400 (rc %d)\n", |
| 415 | (int)hvrc); |
| 416 | goto error_ret; |
| 417 | } |
| 418 | spin_unlock_irqrestore(&viodasd_spinlock, flags); |
| 419 | return 0; |
| 420 | |
| 421 | error_ret: |
| 422 | num_req_outstanding--; |
| 423 | spin_unlock_irqrestore(&viodasd_spinlock, flags); |
| 424 | dma_unmap_sg(d->dev, sg, nsg, direction); |
| 425 | return -1; |
| 426 | } |
| 427 | |
| 428 | /* |
| 429 | * This is the external request processing routine |
| 430 | */ |
| 431 | static void do_viodasd_request(request_queue_t *q) |
| 432 | { |
| 433 | struct request *req; |
| 434 | |
| 435 | /* |
| 436 | * If we already have the maximum number of requests |
| 437 | * outstanding to OS/400 just bail out. We'll come |
| 438 | * back later. |
| 439 | */ |
| 440 | while (num_req_outstanding < VIOMAXREQ) { |
| 441 | req = elv_next_request(q); |
| 442 | if (req == NULL) |
| 443 | return; |
| 444 | /* dequeue the current request from the queue */ |
| 445 | blkdev_dequeue_request(req); |
| 446 | /* check that request contains a valid command */ |
| 447 | if (!blk_fs_request(req)) { |
| 448 | viodasd_end_request(req, 0, req->hard_nr_sectors); |
| 449 | continue; |
| 450 | } |
| 451 | /* Try sending the request */ |
| 452 | if (send_request(req) != 0) |
| 453 | viodasd_end_request(req, 0, req->hard_nr_sectors); |
| 454 | } |
| 455 | } |
| 456 | |
| 457 | /* |
| 458 | * Probe a single disk and fill in the viodasd_device structure |
| 459 | * for it. |
| 460 | */ |
| 461 | static void probe_disk(struct viodasd_device *d) |
| 462 | { |
| 463 | HvLpEvent_Rc hvrc; |
| 464 | struct viodasd_waitevent we; |
| 465 | int dev_no = DEVICE_NO(d); |
| 466 | struct gendisk *g; |
| 467 | struct request_queue *q; |
| 468 | u16 flags = 0; |
| 469 | |
| 470 | retry: |
| 471 | init_completion(&we.com); |
| 472 | |
| 473 | /* Send the open event to OS/400 */ |
| 474 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, |
| 475 | HvLpEvent_Type_VirtualIo, |
| 476 | viomajorsubtype_blockio | vioblockopen, |
| 477 | HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck, |
| 478 | viopath_sourceinst(viopath_hostLp), |
| 479 | viopath_targetinst(viopath_hostLp), |
| 480 | (u64)(unsigned long)&we, VIOVERSION << 16, |
| 481 | ((u64)dev_no << 48) | ((u64)flags<< 32), |
| 482 | 0, 0, 0); |
| 483 | if (hvrc != 0) { |
| 484 | printk(VIOD_KERN_WARNING "bad rc on HV open %d\n", (int)hvrc); |
| 485 | return; |
| 486 | } |
| 487 | |
| 488 | wait_for_completion(&we.com); |
| 489 | |
| 490 | if (we.rc != 0) { |
| 491 | if (flags != 0) |
| 492 | return; |
| 493 | /* try again with read only flag set */ |
| 494 | flags = vioblockflags_ro; |
| 495 | goto retry; |
| 496 | } |
| 497 | if (we.max_disk > (MAX_DISKNO - 1)) { |
| 498 | static int warned; |
| 499 | |
| 500 | if (warned == 0) { |
| 501 | warned++; |
| 502 | printk(VIOD_KERN_INFO |
| 503 | "Only examining the first %d " |
| 504 | "of %d disks connected\n", |
| 505 | MAX_DISKNO, we.max_disk + 1); |
| 506 | } |
| 507 | } |
| 508 | |
| 509 | /* Send the close event to OS/400. We DON'T expect a response */ |
| 510 | hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp, |
| 511 | HvLpEvent_Type_VirtualIo, |
| 512 | viomajorsubtype_blockio | vioblockclose, |
| 513 | HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck, |
| 514 | viopath_sourceinst(viopath_hostLp), |
| 515 | viopath_targetinst(viopath_hostLp), |
| 516 | 0, VIOVERSION << 16, |
| 517 | ((u64)dev_no << 48) | ((u64)flags << 32), |
| 518 | 0, 0, 0); |
| 519 | if (hvrc != 0) { |
| 520 | printk(VIOD_KERN_WARNING |
| 521 | "bad rc sending event to OS/400 %d\n", (int)hvrc); |
| 522 | return; |
| 523 | } |
| 524 | /* create the request queue for the disk */ |
| 525 | spin_lock_init(&d->q_lock); |
| 526 | q = blk_init_queue(do_viodasd_request, &d->q_lock); |
| 527 | if (q == NULL) { |
| 528 | printk(VIOD_KERN_WARNING "cannot allocate queue for disk %d\n", |
| 529 | dev_no); |
| 530 | return; |
| 531 | } |
| 532 | g = alloc_disk(1 << PARTITION_SHIFT); |
| 533 | if (g == NULL) { |
| 534 | printk(VIOD_KERN_WARNING |
| 535 | "cannot allocate disk structure for disk %d\n", |
| 536 | dev_no); |
| 537 | blk_cleanup_queue(q); |
| 538 | return; |
| 539 | } |
| 540 | |
| 541 | d->disk = g; |
| 542 | blk_queue_max_hw_segments(q, VIOMAXBLOCKDMA); |
| 543 | blk_queue_max_phys_segments(q, VIOMAXBLOCKDMA); |
| 544 | blk_queue_max_sectors(q, VIODASD_MAXSECTORS); |
| 545 | g->major = VIODASD_MAJOR; |
| 546 | g->first_minor = dev_no << PARTITION_SHIFT; |
| 547 | if (dev_no >= 26) |
| 548 | snprintf(g->disk_name, sizeof(g->disk_name), |
| 549 | VIOD_GENHD_NAME "%c%c", |
| 550 | 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26)); |
| 551 | else |
| 552 | snprintf(g->disk_name, sizeof(g->disk_name), |
| 553 | VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26)); |
| 554 | snprintf(g->devfs_name, sizeof(g->devfs_name), |
| 555 | "%s%d", VIOD_GENHD_DEVFS_NAME, dev_no); |
| 556 | g->fops = &viodasd_fops; |
| 557 | g->queue = q; |
| 558 | g->private_data = d; |
| 559 | g->driverfs_dev = d->dev; |
| 560 | set_capacity(g, d->size >> 9); |
| 561 | |
| 562 | printk(VIOD_KERN_INFO "disk %d: %lu sectors (%lu MB) " |
| 563 | "CHS=%d/%d/%d sector size %d%s\n", |
| 564 | dev_no, (unsigned long)(d->size >> 9), |
| 565 | (unsigned long)(d->size >> 20), |
| 566 | (int)d->cylinders, (int)d->tracks, |
| 567 | (int)d->sectors, (int)d->bytes_per_sector, |
| 568 | d->read_only ? " (RO)" : ""); |
| 569 | |
| 570 | /* register us in the global list */ |
| 571 | add_disk(g); |
| 572 | } |
| 573 | |
| 574 | /* returns the total number of scatterlist elements converted */ |
| 575 | static int block_event_to_scatterlist(const struct vioblocklpevent *bevent, |
| 576 | struct scatterlist *sg, int *total_len) |
| 577 | { |
| 578 | int i, numsg; |
| 579 | const struct rw_data *rw_data = &bevent->u.rw_data; |
| 580 | static const int offset = |
| 581 | offsetof(struct vioblocklpevent, u.rw_data.dma_info); |
| 582 | static const int element_size = sizeof(rw_data->dma_info[0]); |
| 583 | |
| 584 | numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size; |
| 585 | if (numsg > VIOMAXBLOCKDMA) |
| 586 | numsg = VIOMAXBLOCKDMA; |
| 587 | |
| 588 | *total_len = 0; |
| 589 | memset(sg, 0, sizeof(sg[0]) * VIOMAXBLOCKDMA); |
| 590 | |
| 591 | for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) { |
| 592 | sg_dma_address(&sg[i]) = rw_data->dma_info[i].token; |
| 593 | sg_dma_len(&sg[i]) = rw_data->dma_info[i].len; |
| 594 | *total_len += rw_data->dma_info[i].len; |
| 595 | } |
| 596 | return i; |
| 597 | } |
| 598 | |
| 599 | /* |
| 600 | * Restart all queues, starting with the one _after_ the disk given, |
| 601 | * thus reducing the chance of starvation of higher numbered disks. |
| 602 | */ |
| 603 | static void viodasd_restart_all_queues_starting_from(int first_index) |
| 604 | { |
| 605 | int i; |
| 606 | |
| 607 | for (i = first_index + 1; i < MAX_DISKNO; ++i) |
| 608 | if (viodasd_devices[i].disk) |
| 609 | blk_run_queue(viodasd_devices[i].disk->queue); |
| 610 | for (i = 0; i <= first_index; ++i) |
| 611 | if (viodasd_devices[i].disk) |
| 612 | blk_run_queue(viodasd_devices[i].disk->queue); |
| 613 | } |
| 614 | |
| 615 | /* |
| 616 | * For read and write requests, decrement the number of outstanding requests, |
| 617 | * Free the DMA buffers we allocated. |
| 618 | */ |
| 619 | static int viodasd_handle_read_write(struct vioblocklpevent *bevent) |
| 620 | { |
| 621 | int num_sg, num_sect, pci_direction, total_len; |
| 622 | struct request *req; |
| 623 | struct scatterlist sg[VIOMAXBLOCKDMA]; |
| 624 | struct HvLpEvent *event = &bevent->event; |
| 625 | unsigned long irq_flags; |
| 626 | struct viodasd_device *d; |
| 627 | int error; |
| 628 | spinlock_t *qlock; |
| 629 | |
| 630 | num_sg = block_event_to_scatterlist(bevent, sg, &total_len); |
| 631 | num_sect = total_len >> 9; |
| 632 | if (event->xSubtype == (viomajorsubtype_blockio | vioblockread)) |
| 633 | pci_direction = DMA_FROM_DEVICE; |
| 634 | else |
| 635 | pci_direction = DMA_TO_DEVICE; |
| 636 | req = (struct request *)bevent->event.xCorrelationToken; |
| 637 | d = req->rq_disk->private_data; |
| 638 | |
| 639 | dma_unmap_sg(d->dev, sg, num_sg, pci_direction); |
| 640 | |
| 641 | /* |
| 642 | * Since this is running in interrupt mode, we need to make sure |
| 643 | * we're not stepping on any global I/O operations |
| 644 | */ |
| 645 | spin_lock_irqsave(&viodasd_spinlock, irq_flags); |
| 646 | num_req_outstanding--; |
| 647 | spin_unlock_irqrestore(&viodasd_spinlock, irq_flags); |
| 648 | |
| 649 | error = event->xRc != HvLpEvent_Rc_Good; |
| 650 | if (error) { |
| 651 | const struct vio_error_entry *err; |
| 652 | err = vio_lookup_rc(viodasd_err_table, bevent->sub_result); |
| 653 | printk(VIOD_KERN_WARNING "read/write error %d:0x%04x (%s)\n", |
| 654 | event->xRc, bevent->sub_result, err->msg); |
| 655 | num_sect = req->hard_nr_sectors; |
| 656 | } |
| 657 | qlock = req->q->queue_lock; |
| 658 | spin_lock_irqsave(qlock, irq_flags); |
| 659 | viodasd_end_request(req, !error, num_sect); |
| 660 | spin_unlock_irqrestore(qlock, irq_flags); |
| 661 | |
| 662 | /* Finally, try to get more requests off of this device's queue */ |
| 663 | viodasd_restart_all_queues_starting_from(DEVICE_NO(d)); |
| 664 | |
| 665 | return 0; |
| 666 | } |
| 667 | |
| 668 | /* This routine handles incoming block LP events */ |
| 669 | static void handle_block_event(struct HvLpEvent *event) |
| 670 | { |
| 671 | struct vioblocklpevent *bevent = (struct vioblocklpevent *)event; |
| 672 | struct viodasd_waitevent *pwe; |
| 673 | |
| 674 | if (event == NULL) |
| 675 | /* Notification that a partition went away! */ |
| 676 | return; |
| 677 | /* First, we should NEVER get an int here...only acks */ |
| 678 | if (event->xFlags.xFunction == HvLpEvent_Function_Int) { |
| 679 | printk(VIOD_KERN_WARNING |
| 680 | "Yikes! got an int in viodasd event handler!\n"); |
| 681 | if (event->xFlags.xAckInd == HvLpEvent_AckInd_DoAck) { |
| 682 | event->xRc = HvLpEvent_Rc_InvalidSubtype; |
| 683 | HvCallEvent_ackLpEvent(event); |
| 684 | } |
| 685 | } |
| 686 | |
| 687 | switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) { |
| 688 | case vioblockopen: |
| 689 | /* |
| 690 | * Handle a response to an open request. We get all the |
| 691 | * disk information in the response, so update it. The |
| 692 | * correlation token contains a pointer to a waitevent |
| 693 | * structure that has a completion in it. update the |
| 694 | * return code in the waitevent structure and post the |
| 695 | * completion to wake up the guy who sent the request |
| 696 | */ |
| 697 | pwe = (struct viodasd_waitevent *)event->xCorrelationToken; |
| 698 | pwe->rc = event->xRc; |
| 699 | pwe->sub_result = bevent->sub_result; |
| 700 | if (event->xRc == HvLpEvent_Rc_Good) { |
| 701 | const struct open_data *data = &bevent->u.open_data; |
| 702 | struct viodasd_device *device = |
| 703 | &viodasd_devices[bevent->disk]; |
| 704 | device->read_only = |
| 705 | bevent->flags & vioblockflags_ro; |
| 706 | device->size = data->disk_size; |
| 707 | device->cylinders = data->cylinders; |
| 708 | device->tracks = data->tracks; |
| 709 | device->sectors = data->sectors; |
| 710 | device->bytes_per_sector = data->bytes_per_sector; |
| 711 | pwe->max_disk = data->max_disk; |
| 712 | } |
| 713 | complete(&pwe->com); |
| 714 | break; |
| 715 | case vioblockclose: |
| 716 | break; |
| 717 | case vioblockread: |
| 718 | case vioblockwrite: |
| 719 | viodasd_handle_read_write(bevent); |
| 720 | break; |
| 721 | |
| 722 | default: |
| 723 | printk(VIOD_KERN_WARNING "invalid subtype!"); |
| 724 | if (event->xFlags.xAckInd == HvLpEvent_AckInd_DoAck) { |
| 725 | event->xRc = HvLpEvent_Rc_InvalidSubtype; |
| 726 | HvCallEvent_ackLpEvent(event); |
| 727 | } |
| 728 | } |
| 729 | } |
| 730 | |
| 731 | /* |
| 732 | * Get the driver to reprobe for more disks. |
| 733 | */ |
| 734 | static ssize_t probe_disks(struct device_driver *drv, const char *buf, |
| 735 | size_t count) |
| 736 | { |
| 737 | struct viodasd_device *d; |
| 738 | |
| 739 | for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) { |
| 740 | if (d->disk == NULL) |
| 741 | probe_disk(d); |
| 742 | } |
| 743 | return count; |
| 744 | } |
| 745 | static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks); |
| 746 | |
| 747 | static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id) |
| 748 | { |
| 749 | struct viodasd_device *d = &viodasd_devices[vdev->unit_address]; |
| 750 | |
| 751 | d->dev = &vdev->dev; |
| 752 | probe_disk(d); |
| 753 | if (d->disk == NULL) |
| 754 | return -ENODEV; |
| 755 | return 0; |
| 756 | } |
| 757 | |
| 758 | static int viodasd_remove(struct vio_dev *vdev) |
| 759 | { |
| 760 | struct viodasd_device *d; |
| 761 | |
| 762 | d = &viodasd_devices[vdev->unit_address]; |
| 763 | if (d->disk) { |
| 764 | del_gendisk(d->disk); |
| 765 | blk_cleanup_queue(d->disk->queue); |
| 766 | put_disk(d->disk); |
| 767 | d->disk = NULL; |
| 768 | } |
| 769 | d->dev = NULL; |
| 770 | return 0; |
| 771 | } |
| 772 | |
| 773 | /** |
| 774 | * viodasd_device_table: Used by vio.c to match devices that we |
| 775 | * support. |
| 776 | */ |
| 777 | static struct vio_device_id viodasd_device_table[] __devinitdata = { |
| 778 | { "viodasd", "" }, |
Stephen Rothwell | fb120da | 2005-08-17 16:42:59 +1000 | [diff] [blame] | 779 | { "", "" } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | }; |
| 781 | |
| 782 | MODULE_DEVICE_TABLE(vio, viodasd_device_table); |
| 783 | static struct vio_driver viodasd_driver = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 784 | .id_table = viodasd_device_table, |
| 785 | .probe = viodasd_probe, |
Stephen Rothwell | 6fdf539 | 2005-10-24 14:53:21 +1000 | [diff] [blame] | 786 | .remove = viodasd_remove, |
| 787 | .driver = { |
| 788 | .name = "viodasd", |
| 789 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 790 | }; |
| 791 | |
| 792 | /* |
| 793 | * Initialize the whole device driver. Handle module and non-module |
| 794 | * versions |
| 795 | */ |
| 796 | static int __init viodasd_init(void) |
| 797 | { |
| 798 | int rc; |
| 799 | |
| 800 | /* Try to open to our host lp */ |
| 801 | if (viopath_hostLp == HvLpIndexInvalid) |
| 802 | vio_set_hostlp(); |
| 803 | |
| 804 | if (viopath_hostLp == HvLpIndexInvalid) { |
| 805 | printk(VIOD_KERN_WARNING "invalid hosting partition\n"); |
| 806 | return -EIO; |
| 807 | } |
| 808 | |
| 809 | printk(VIOD_KERN_INFO "vers " VIOD_VERS ", hosting partition %d\n", |
| 810 | viopath_hostLp); |
| 811 | |
| 812 | /* register the block device */ |
| 813 | if (register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME)) { |
| 814 | printk(VIOD_KERN_WARNING |
| 815 | "Unable to get major number %d for %s\n", |
| 816 | VIODASD_MAJOR, VIOD_GENHD_NAME); |
| 817 | return -EIO; |
| 818 | } |
| 819 | /* Actually open the path to the hosting partition */ |
| 820 | if (viopath_open(viopath_hostLp, viomajorsubtype_blockio, |
| 821 | VIOMAXREQ + 2)) { |
| 822 | printk(VIOD_KERN_WARNING |
| 823 | "error opening path to host partition %d\n", |
| 824 | viopath_hostLp); |
| 825 | unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME); |
| 826 | return -EIO; |
| 827 | } |
| 828 | |
| 829 | /* Initialize our request handler */ |
| 830 | vio_setHandler(viomajorsubtype_blockio, handle_block_event); |
| 831 | |
| 832 | rc = vio_register_driver(&viodasd_driver); |
| 833 | if (rc == 0) |
| 834 | driver_create_file(&viodasd_driver.driver, &driver_attr_probe); |
| 835 | return rc; |
| 836 | } |
| 837 | module_init(viodasd_init); |
| 838 | |
| 839 | void viodasd_exit(void) |
| 840 | { |
| 841 | driver_remove_file(&viodasd_driver.driver, &driver_attr_probe); |
| 842 | vio_unregister_driver(&viodasd_driver); |
| 843 | vio_clearHandler(viomajorsubtype_blockio); |
| 844 | unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME); |
| 845 | viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2); |
| 846 | } |
| 847 | |
| 848 | module_exit(viodasd_exit); |