2 * Block driver for s390 storage class memory.
4 * Copyright IBM Corp. 2012
5 * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com>
8 #define KMSG_COMPONENT "scm_block"
9 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
11 #include <linux/interrupt.h>
12 #include <linux/spinlock.h>
13 #include <linux/module.h>
14 #include <linux/blkdev.h>
15 #include <linux/genhd.h>
16 #include <linux/slab.h>
17 #include <linux/list.h>
21 debug_info_t *scm_debug;
23 static DEFINE_SPINLOCK(list_lock);
24 static LIST_HEAD(inactive_requests);
25 static unsigned int nr_requests = 64;
26 static atomic_t nr_devices = ATOMIC_INIT(0);
27 module_param(nr_requests, uint, S_IRUGO);
28 MODULE_PARM_DESC(nr_requests, "Number of parallel requests.");
30 MODULE_DESCRIPTION("Block driver for s390 storage class memory.");
31 MODULE_LICENSE("GPL");
32 MODULE_ALIAS("scm:scmdev*");
34 static void __scm_free_rq(struct scm_request *scmrq)
36 struct aob_rq_header *aobrq = to_aobrq(scmrq);
38 free_page((unsigned long) scmrq->aob);
39 free_page((unsigned long) scmrq->aidaw);
40 __scm_free_rq_cluster(scmrq);
44 static void scm_free_rqs(void)
46 struct list_head *iter, *safe;
47 struct scm_request *scmrq;
49 spin_lock_irq(&list_lock);
50 list_for_each_safe(iter, safe, &inactive_requests) {
51 scmrq = list_entry(iter, struct scm_request, list);
52 list_del(&scmrq->list);
55 spin_unlock_irq(&list_lock);
58 static int __scm_alloc_rq(void)
60 struct aob_rq_header *aobrq;
61 struct scm_request *scmrq;
63 aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL);
67 scmrq = (void *) aobrq->data;
68 scmrq->aidaw = (void *) get_zeroed_page(GFP_DMA);
69 scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
70 if (!scmrq->aob || !scmrq->aidaw) {
75 if (__scm_alloc_rq_cluster(scmrq)) {
80 INIT_LIST_HEAD(&scmrq->list);
81 spin_lock_irq(&list_lock);
82 list_add(&scmrq->list, &inactive_requests);
83 spin_unlock_irq(&list_lock);
88 static int scm_alloc_rqs(unsigned int nrqs)
92 while (nrqs-- && !ret)
93 ret = __scm_alloc_rq();
98 static struct scm_request *scm_request_fetch(void)
100 struct scm_request *scmrq = NULL;
102 spin_lock(&list_lock);
103 if (list_empty(&inactive_requests))
105 scmrq = list_first_entry(&inactive_requests, struct scm_request, list);
106 list_del(&scmrq->list);
108 spin_unlock(&list_lock);
112 static void scm_request_done(struct scm_request *scmrq)
116 spin_lock_irqsave(&list_lock, flags);
117 list_add(&scmrq->list, &inactive_requests);
118 spin_unlock_irqrestore(&list_lock, flags);
121 static int scm_open(struct block_device *blkdev, fmode_t mode)
123 return scm_get_ref();
126 static int scm_release(struct gendisk *gendisk, fmode_t mode)
132 static const struct block_device_operations scm_blk_devops = {
133 .owner = THIS_MODULE,
135 .release = scm_release,
138 static void scm_request_prepare(struct scm_request *scmrq)
140 struct scm_blk_dev *bdev = scmrq->bdev;
141 struct scm_device *scmdev = bdev->gendisk->private_data;
142 struct aidaw *aidaw = scmrq->aidaw;
143 struct msb *msb = &scmrq->aob->msb[0];
144 struct req_iterator iter;
148 scmrq->aob->request.msb_count = 1;
149 msb->scm_addr = scmdev->address +
150 ((u64) blk_rq_pos(scmrq->request) << 9);
151 msb->oc = (rq_data_dir(scmrq->request) == READ) ?
152 MSB_OC_READ : MSB_OC_WRITE;
153 msb->flags |= MSB_FLAG_IDA;
154 msb->data_addr = (u64) aidaw;
156 rq_for_each_segment(bv, scmrq->request, iter) {
157 WARN_ON(bv->bv_offset);
158 msb->blk_count += bv->bv_len >> 12;
159 aidaw->data_addr = (u64) page_address(bv->bv_page);
164 static inline void scm_request_init(struct scm_blk_dev *bdev,
165 struct scm_request *scmrq,
168 struct aob_rq_header *aobrq = to_aobrq(scmrq);
169 struct aob *aob = scmrq->aob;
171 memset(aob, 0, sizeof(*aob));
172 memset(scmrq->aidaw, 0, PAGE_SIZE);
173 aobrq->scmdev = bdev->scmdev;
174 aob->request.cmd_code = ARQB_CMD_MOVE;
175 aob->request.data = (u64) aobrq;
176 scmrq->request = req;
180 scm_request_cluster_init(scmrq);
183 static void scm_ensure_queue_restart(struct scm_blk_dev *bdev)
185 if (atomic_read(&bdev->queued_reqs)) {
186 /* Queue restart is triggered by the next interrupt. */
189 blk_delay_queue(bdev->rq, SCM_QUEUE_DELAY);
192 void scm_request_requeue(struct scm_request *scmrq)
194 struct scm_blk_dev *bdev = scmrq->bdev;
196 scm_release_cluster(scmrq);
197 blk_requeue_request(bdev->rq, scmrq->request);
198 scm_request_done(scmrq);
199 scm_ensure_queue_restart(bdev);
202 void scm_request_finish(struct scm_request *scmrq)
204 scm_release_cluster(scmrq);
205 blk_end_request_all(scmrq->request, scmrq->error);
206 scm_request_done(scmrq);
209 static void scm_blk_request(struct request_queue *rq)
211 struct scm_device *scmdev = rq->queuedata;
212 struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev);
213 struct scm_request *scmrq;
217 while ((req = blk_peek_request(rq))) {
218 if (req->cmd_type != REQ_TYPE_FS)
221 scmrq = scm_request_fetch();
223 SCM_LOG(5, "no request");
224 scm_ensure_queue_restart(bdev);
227 scm_request_init(bdev, scmrq, req);
228 if (!scm_reserve_cluster(scmrq)) {
229 SCM_LOG(5, "cluster busy");
230 scm_request_done(scmrq);
233 if (scm_need_cluster_request(scmrq)) {
234 blk_start_request(req);
235 scm_initiate_cluster_request(scmrq);
238 scm_request_prepare(scmrq);
239 blk_start_request(req);
241 ret = scm_start_aob(scmrq->aob);
243 SCM_LOG(5, "no subchannel");
244 scm_request_requeue(scmrq);
247 atomic_inc(&bdev->queued_reqs);
251 static void __scmrq_log_error(struct scm_request *scmrq)
253 struct aob *aob = scmrq->aob;
255 if (scmrq->error == -ETIMEDOUT)
256 SCM_LOG(1, "Request timeout");
258 SCM_LOG(1, "Request error");
259 SCM_LOG_HEX(1, &aob->response, sizeof(aob->response));
262 SCM_LOG(1, "Retry request");
264 pr_err("An I/O operation to SCM failed with rc=%d\n",
268 void scm_blk_irq(struct scm_device *scmdev, void *data, int error)
270 struct scm_request *scmrq = data;
271 struct scm_blk_dev *bdev = scmrq->bdev;
273 scmrq->error = error;
275 __scmrq_log_error(scmrq);
277 spin_lock(&bdev->lock);
278 list_add_tail(&scmrq->list, &bdev->finished_requests);
279 spin_unlock(&bdev->lock);
280 tasklet_hi_schedule(&bdev->tasklet);
283 static void scm_blk_tasklet(struct scm_blk_dev *bdev)
285 struct scm_request *scmrq;
288 spin_lock_irqsave(&bdev->lock, flags);
289 while (!list_empty(&bdev->finished_requests)) {
290 scmrq = list_first_entry(&bdev->finished_requests,
291 struct scm_request, list);
292 list_del(&scmrq->list);
293 spin_unlock_irqrestore(&bdev->lock, flags);
295 if (scmrq->error && scmrq->retries-- > 0) {
296 if (scm_start_aob(scmrq->aob)) {
297 spin_lock_irqsave(&bdev->rq_lock, flags);
298 scm_request_requeue(scmrq);
299 spin_unlock_irqrestore(&bdev->rq_lock, flags);
301 /* Request restarted or requeued, handle next. */
302 spin_lock_irqsave(&bdev->lock, flags);
306 if (scm_test_cluster_request(scmrq)) {
307 scm_cluster_request_irq(scmrq);
308 spin_lock_irqsave(&bdev->lock, flags);
312 scm_request_finish(scmrq);
313 atomic_dec(&bdev->queued_reqs);
314 spin_lock_irqsave(&bdev->lock, flags);
316 spin_unlock_irqrestore(&bdev->lock, flags);
317 /* Look out for more requests. */
318 blk_run_queue(bdev->rq);
321 int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev)
323 struct request_queue *rq;
324 int len, ret = -ENOMEM;
325 unsigned int devindex, nr_max_blk;
327 devindex = atomic_inc_return(&nr_devices) - 1;
328 /* scma..scmz + scmaa..scmzz */
329 if (devindex > 701) {
334 bdev->scmdev = scmdev;
335 spin_lock_init(&bdev->rq_lock);
336 spin_lock_init(&bdev->lock);
337 INIT_LIST_HEAD(&bdev->finished_requests);
338 atomic_set(&bdev->queued_reqs, 0);
339 tasklet_init(&bdev->tasklet,
340 (void (*)(unsigned long)) scm_blk_tasklet,
341 (unsigned long) bdev);
343 rq = blk_init_queue(scm_blk_request, &bdev->rq_lock);
348 nr_max_blk = min(scmdev->nr_max_block,
349 (unsigned int) (PAGE_SIZE / sizeof(struct aidaw)));
351 blk_queue_logical_block_size(rq, 1 << 12);
352 blk_queue_max_hw_sectors(rq, nr_max_blk << 3); /* 8 * 512 = blk_size */
353 blk_queue_max_segments(rq, nr_max_blk);
354 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rq);
355 scm_blk_dev_cluster_setup(bdev);
357 bdev->gendisk = alloc_disk(SCM_NR_PARTS);
361 rq->queuedata = scmdev;
362 bdev->gendisk->driverfs_dev = &scmdev->dev;
363 bdev->gendisk->private_data = scmdev;
364 bdev->gendisk->fops = &scm_blk_devops;
365 bdev->gendisk->queue = rq;
366 bdev->gendisk->major = scm_major;
367 bdev->gendisk->first_minor = devindex * SCM_NR_PARTS;
369 len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm");
371 len += snprintf(bdev->gendisk->disk_name + len,
372 DISK_NAME_LEN - len, "%c",
373 'a' + (devindex / 26) - 1);
374 devindex = devindex % 26;
376 snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c",
379 /* 512 byte sectors */
380 set_capacity(bdev->gendisk, scmdev->size >> 9);
381 add_disk(bdev->gendisk);
385 blk_cleanup_queue(rq);
387 atomic_dec(&nr_devices);
391 void scm_blk_dev_cleanup(struct scm_blk_dev *bdev)
393 tasklet_kill(&bdev->tasklet);
394 del_gendisk(bdev->gendisk);
395 blk_cleanup_queue(bdev->gendisk->queue);
396 put_disk(bdev->gendisk);
399 static int __init scm_blk_init(void)
403 if (!scm_cluster_size_valid())
406 ret = register_blkdev(0, "scm");
411 if (scm_alloc_rqs(nr_requests))
414 scm_debug = debug_register("scm_log", 16, 1, 16);
418 debug_register_view(scm_debug, &debug_hex_ascii_view);
419 debug_set_level(scm_debug, 2);
421 ret = scm_drv_init();
428 debug_unregister(scm_debug);
432 unregister_blkdev(scm_major, "scm");
436 module_init(scm_blk_init);
438 static void __exit scm_blk_cleanup(void)
441 debug_unregister(scm_debug);
443 unregister_blkdev(scm_major, "scm");
445 module_exit(scm_blk_cleanup);