scm_blk.c 11.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
/*
 * Block driver for s390 storage class memory.
 *
 * Copyright IBM Corp. 2012
 * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com>
 */

#define KMSG_COMPONENT "scm_block"
#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt

#include <linux/interrupt.h>
#include <linux/spinlock.h>
#include <linux/module.h>
#include <linux/blkdev.h>
#include <linux/genhd.h>
#include <linux/slab.h>
#include <linux/list.h>
#include <asm/eadm.h>
#include "scm_blk.h"

debug_info_t *scm_debug;
static int scm_major;
static DEFINE_SPINLOCK(list_lock);
static LIST_HEAD(inactive_requests);
static unsigned int nr_requests = 64;
static atomic_t nr_devices = ATOMIC_INIT(0);
module_param(nr_requests, uint, S_IRUGO);
MODULE_PARM_DESC(nr_requests, "Number of parallel requests.");

MODULE_DESCRIPTION("Block driver for s390 storage class memory.");
MODULE_LICENSE("GPL");
MODULE_ALIAS("scm:scmdev*");

static void __scm_free_rq(struct scm_request *scmrq)
{
	struct aob_rq_header *aobrq = to_aobrq(scmrq);

	free_page((unsigned long) scmrq->aob);
	free_page((unsigned long) scmrq->aidaw);
40
	__scm_free_rq_cluster(scmrq);
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
	kfree(aobrq);
}

static void scm_free_rqs(void)
{
	struct list_head *iter, *safe;
	struct scm_request *scmrq;

	spin_lock_irq(&list_lock);
	list_for_each_safe(iter, safe, &inactive_requests) {
		scmrq = list_entry(iter, struct scm_request, list);
		list_del(&scmrq->list);
		__scm_free_rq(scmrq);
	}
	spin_unlock_irq(&list_lock);
}

static int __scm_alloc_rq(void)
{
	struct aob_rq_header *aobrq;
	struct scm_request *scmrq;

	aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL);
	if (!aobrq)
		return -ENOMEM;

	scmrq = (void *) aobrq->data;
	scmrq->aidaw = (void *) get_zeroed_page(GFP_DMA);
	scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
	if (!scmrq->aob || !scmrq->aidaw) {
		__scm_free_rq(scmrq);
		return -ENOMEM;
	}
74 75 76 77 78 79

	if (__scm_alloc_rq_cluster(scmrq)) {
		__scm_free_rq(scmrq);
		return -ENOMEM;
	}

80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137
	INIT_LIST_HEAD(&scmrq->list);
	spin_lock_irq(&list_lock);
	list_add(&scmrq->list, &inactive_requests);
	spin_unlock_irq(&list_lock);

	return 0;
}

static int scm_alloc_rqs(unsigned int nrqs)
{
	int ret = 0;

	while (nrqs-- && !ret)
		ret = __scm_alloc_rq();

	return ret;
}

static struct scm_request *scm_request_fetch(void)
{
	struct scm_request *scmrq = NULL;

	spin_lock(&list_lock);
	if (list_empty(&inactive_requests))
		goto out;
	scmrq = list_first_entry(&inactive_requests, struct scm_request, list);
	list_del(&scmrq->list);
out:
	spin_unlock(&list_lock);
	return scmrq;
}

static void scm_request_done(struct scm_request *scmrq)
{
	unsigned long flags;

	spin_lock_irqsave(&list_lock, flags);
	list_add(&scmrq->list, &inactive_requests);
	spin_unlock_irqrestore(&list_lock, flags);
}

static int scm_open(struct block_device *blkdev, fmode_t mode)
{
	return scm_get_ref();
}

static int scm_release(struct gendisk *gendisk, fmode_t mode)
{
	scm_put_ref();
	return 0;
}

static const struct block_device_operations scm_blk_devops = {
	.owner = THIS_MODULE,
	.open = scm_open,
	.release = scm_release,
};

138 139 140 141 142
static bool scm_permit_request(struct scm_blk_dev *bdev, struct request *req)
{
	return rq_data_dir(req) != WRITE || bdev->state != SCM_WR_PROHIBIT;
}

143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184
static void scm_request_prepare(struct scm_request *scmrq)
{
	struct scm_blk_dev *bdev = scmrq->bdev;
	struct scm_device *scmdev = bdev->gendisk->private_data;
	struct aidaw *aidaw = scmrq->aidaw;
	struct msb *msb = &scmrq->aob->msb[0];
	struct req_iterator iter;
	struct bio_vec *bv;

	msb->bs = MSB_BS_4K;
	scmrq->aob->request.msb_count = 1;
	msb->scm_addr = scmdev->address +
		((u64) blk_rq_pos(scmrq->request) << 9);
	msb->oc = (rq_data_dir(scmrq->request) == READ) ?
		MSB_OC_READ : MSB_OC_WRITE;
	msb->flags |= MSB_FLAG_IDA;
	msb->data_addr = (u64) aidaw;

	rq_for_each_segment(bv, scmrq->request, iter) {
		WARN_ON(bv->bv_offset);
		msb->blk_count += bv->bv_len >> 12;
		aidaw->data_addr = (u64) page_address(bv->bv_page);
		aidaw++;
	}
}

static inline void scm_request_init(struct scm_blk_dev *bdev,
				    struct scm_request *scmrq,
				    struct request *req)
{
	struct aob_rq_header *aobrq = to_aobrq(scmrq);
	struct aob *aob = scmrq->aob;

	memset(aob, 0, sizeof(*aob));
	memset(scmrq->aidaw, 0, PAGE_SIZE);
	aobrq->scmdev = bdev->scmdev;
	aob->request.cmd_code = ARQB_CMD_MOVE;
	aob->request.data = (u64) aobrq;
	scmrq->request = req;
	scmrq->bdev = bdev;
	scmrq->retries = 4;
	scmrq->error = 0;
185
	scm_request_cluster_init(scmrq);
186 187 188 189 190 191 192 193 194 195 196
}

static void scm_ensure_queue_restart(struct scm_blk_dev *bdev)
{
	if (atomic_read(&bdev->queued_reqs)) {
		/* Queue restart is triggered by the next interrupt. */
		return;
	}
	blk_delay_queue(bdev->rq, SCM_QUEUE_DELAY);
}

197
void scm_request_requeue(struct scm_request *scmrq)
198 199 200
{
	struct scm_blk_dev *bdev = scmrq->bdev;

201
	scm_release_cluster(scmrq);
202
	blk_requeue_request(bdev->rq, scmrq->request);
203
	atomic_dec(&bdev->queued_reqs);
204 205 206 207
	scm_request_done(scmrq);
	scm_ensure_queue_restart(bdev);
}

208
void scm_request_finish(struct scm_request *scmrq)
209
{
210 211
	struct scm_blk_dev *bdev = scmrq->bdev;

212
	scm_release_cluster(scmrq);
213
	blk_end_request_all(scmrq->request, scmrq->error);
214
	atomic_dec(&bdev->queued_reqs);
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229
	scm_request_done(scmrq);
}

static void scm_blk_request(struct request_queue *rq)
{
	struct scm_device *scmdev = rq->queuedata;
	struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev);
	struct scm_request *scmrq;
	struct request *req;
	int ret;

	while ((req = blk_peek_request(rq))) {
		if (req->cmd_type != REQ_TYPE_FS)
			continue;

230 231 232 233
		if (!scm_permit_request(bdev, req)) {
			scm_ensure_queue_restart(bdev);
			return;
		}
234 235 236 237 238 239 240
		scmrq = scm_request_fetch();
		if (!scmrq) {
			SCM_LOG(5, "no request");
			scm_ensure_queue_restart(bdev);
			return;
		}
		scm_request_init(bdev, scmrq, req);
241 242 243 244 245 246
		if (!scm_reserve_cluster(scmrq)) {
			SCM_LOG(5, "cluster busy");
			scm_request_done(scmrq);
			return;
		}
		if (scm_need_cluster_request(scmrq)) {
247
			atomic_inc(&bdev->queued_reqs);
248 249 250 251
			blk_start_request(req);
			scm_initiate_cluster_request(scmrq);
			return;
		}
252
		scm_request_prepare(scmrq);
253
		atomic_inc(&bdev->queued_reqs);
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296
		blk_start_request(req);

		ret = scm_start_aob(scmrq->aob);
		if (ret) {
			SCM_LOG(5, "no subchannel");
			scm_request_requeue(scmrq);
			return;
		}
	}
}

static void __scmrq_log_error(struct scm_request *scmrq)
{
	struct aob *aob = scmrq->aob;

	if (scmrq->error == -ETIMEDOUT)
		SCM_LOG(1, "Request timeout");
	else {
		SCM_LOG(1, "Request error");
		SCM_LOG_HEX(1, &aob->response, sizeof(aob->response));
	}
	if (scmrq->retries)
		SCM_LOG(1, "Retry request");
	else
		pr_err("An I/O operation to SCM failed with rc=%d\n",
		       scmrq->error);
}

void scm_blk_irq(struct scm_device *scmdev, void *data, int error)
{
	struct scm_request *scmrq = data;
	struct scm_blk_dev *bdev = scmrq->bdev;

	scmrq->error = error;
	if (error)
		__scmrq_log_error(scmrq);

	spin_lock(&bdev->lock);
	list_add_tail(&scmrq->list, &bdev->finished_requests);
	spin_unlock(&bdev->lock);
	tasklet_hi_schedule(&bdev->tasklet);
}

297 298 299 300 301 302 303 304 305 306 307 308 309
static void scm_blk_handle_error(struct scm_request *scmrq)
{
	struct scm_blk_dev *bdev = scmrq->bdev;
	unsigned long flags;

	if (scmrq->error != -EIO)
		goto restart;

	/* For -EIO the response block is valid. */
	switch (scmrq->aob->response.eqc) {
	case EQC_WR_PROHIBIT:
		spin_lock_irqsave(&bdev->lock, flags);
		if (bdev->state != SCM_WR_PROHIBIT)
310
			pr_info("%lx: Write access to the SCM increment is suspended\n",
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
				(unsigned long) bdev->scmdev->address);
		bdev->state = SCM_WR_PROHIBIT;
		spin_unlock_irqrestore(&bdev->lock, flags);
		goto requeue;
	default:
		break;
	}

restart:
	if (!scm_start_aob(scmrq->aob))
		return;

requeue:
	spin_lock_irqsave(&bdev->rq_lock, flags);
	scm_request_requeue(scmrq);
	spin_unlock_irqrestore(&bdev->rq_lock, flags);
}

329 330 331 332 333 334 335 336 337 338 339 340 341
static void scm_blk_tasklet(struct scm_blk_dev *bdev)
{
	struct scm_request *scmrq;
	unsigned long flags;

	spin_lock_irqsave(&bdev->lock, flags);
	while (!list_empty(&bdev->finished_requests)) {
		scmrq = list_first_entry(&bdev->finished_requests,
					 struct scm_request, list);
		list_del(&scmrq->list);
		spin_unlock_irqrestore(&bdev->lock, flags);

		if (scmrq->error && scmrq->retries-- > 0) {
342 343
			scm_blk_handle_error(scmrq);

344 345 346 347
			/* Request restarted or requeued, handle next. */
			spin_lock_irqsave(&bdev->lock, flags);
			continue;
		}
348 349 350 351 352 353 354

		if (scm_test_cluster_request(scmrq)) {
			scm_cluster_request_irq(scmrq);
			spin_lock_irqsave(&bdev->lock, flags);
			continue;
		}

355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
		scm_request_finish(scmrq);
		spin_lock_irqsave(&bdev->lock, flags);
	}
	spin_unlock_irqrestore(&bdev->lock, flags);
	/* Look out for more requests. */
	blk_run_queue(bdev->rq);
}

int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev)
{
	struct request_queue *rq;
	int len, ret = -ENOMEM;
	unsigned int devindex, nr_max_blk;

	devindex = atomic_inc_return(&nr_devices) - 1;
	/* scma..scmz + scmaa..scmzz */
	if (devindex > 701) {
		ret = -ENODEV;
		goto out;
	}

	bdev->scmdev = scmdev;
377
	bdev->state = SCM_OPER;
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
	spin_lock_init(&bdev->rq_lock);
	spin_lock_init(&bdev->lock);
	INIT_LIST_HEAD(&bdev->finished_requests);
	atomic_set(&bdev->queued_reqs, 0);
	tasklet_init(&bdev->tasklet,
		     (void (*)(unsigned long)) scm_blk_tasklet,
		     (unsigned long) bdev);

	rq = blk_init_queue(scm_blk_request, &bdev->rq_lock);
	if (!rq)
		goto out;

	bdev->rq = rq;
	nr_max_blk = min(scmdev->nr_max_block,
			 (unsigned int) (PAGE_SIZE / sizeof(struct aidaw)));

	blk_queue_logical_block_size(rq, 1 << 12);
	blk_queue_max_hw_sectors(rq, nr_max_blk << 3); /* 8 * 512 = blk_size */
	blk_queue_max_segments(rq, nr_max_blk);
	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rq);
398
	scm_blk_dev_cluster_setup(bdev);
399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441

	bdev->gendisk = alloc_disk(SCM_NR_PARTS);
	if (!bdev->gendisk)
		goto out_queue;

	rq->queuedata = scmdev;
	bdev->gendisk->driverfs_dev = &scmdev->dev;
	bdev->gendisk->private_data = scmdev;
	bdev->gendisk->fops = &scm_blk_devops;
	bdev->gendisk->queue = rq;
	bdev->gendisk->major = scm_major;
	bdev->gendisk->first_minor = devindex * SCM_NR_PARTS;

	len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm");
	if (devindex > 25) {
		len += snprintf(bdev->gendisk->disk_name + len,
				DISK_NAME_LEN - len, "%c",
				'a' + (devindex / 26) - 1);
		devindex = devindex % 26;
	}
	snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c",
		 'a' + devindex);

	/* 512 byte sectors */
	set_capacity(bdev->gendisk, scmdev->size >> 9);
	add_disk(bdev->gendisk);
	return 0;

out_queue:
	blk_cleanup_queue(rq);
out:
	atomic_dec(&nr_devices);
	return ret;
}

void scm_blk_dev_cleanup(struct scm_blk_dev *bdev)
{
	tasklet_kill(&bdev->tasklet);
	del_gendisk(bdev->gendisk);
	blk_cleanup_queue(bdev->gendisk->queue);
	put_disk(bdev->gendisk);
}

442 443 444 445 446 447
void scm_blk_set_available(struct scm_blk_dev *bdev)
{
	unsigned long flags;

	spin_lock_irqsave(&bdev->lock, flags);
	if (bdev->state == SCM_WR_PROHIBIT)
448
		pr_info("%lx: Write access to the SCM increment is restored\n",
449 450 451 452 453
			(unsigned long) bdev->scmdev->address);
	bdev->state = SCM_OPER;
	spin_unlock_irqrestore(&bdev->lock, flags);
}

454 455
static int __init scm_blk_init(void)
{
456 457 458 459
	int ret = -EINVAL;

	if (!scm_cluster_size_valid())
		goto out;
460 461 462 463 464 465

	ret = register_blkdev(0, "scm");
	if (ret < 0)
		goto out;

	scm_major = ret;
466 467
	ret = scm_alloc_rqs(nr_requests);
	if (ret)
468
		goto out_free;
469 470

	scm_debug = debug_register("scm_log", 16, 1, 16);
471 472
	if (!scm_debug) {
		ret = -ENOMEM;
473
		goto out_free;
474
	}
475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502

	debug_register_view(scm_debug, &debug_hex_ascii_view);
	debug_set_level(scm_debug, 2);

	ret = scm_drv_init();
	if (ret)
		goto out_dbf;

	return ret;

out_dbf:
	debug_unregister(scm_debug);
out_free:
	scm_free_rqs();
	unregister_blkdev(scm_major, "scm");
out:
	return ret;
}
module_init(scm_blk_init);

static void __exit scm_blk_cleanup(void)
{
	scm_drv_cleanup();
	debug_unregister(scm_debug);
	scm_free_rqs();
	unregister_blkdev(scm_major, "scm");
}
module_exit(scm_blk_cleanup);
新手
引导
客服 返回
顶部