Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
[cascardo/linux.git] / drivers / infiniband / hw / cxgb4 / device.c
1 /*
2  * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/moduleparam.h>
34 #include <linux/debugfs.h>
35
36 #include <rdma/ib_verbs.h>
37
38 #include "iw_cxgb4.h"
39
40 #define DRV_VERSION "0.1"
41
42 MODULE_AUTHOR("Steve Wise");
43 MODULE_DESCRIPTION("Chelsio T4 RDMA Driver");
44 MODULE_LICENSE("Dual BSD/GPL");
45 MODULE_VERSION(DRV_VERSION);
46
47 static LIST_HEAD(uld_ctx_list);
48 static DEFINE_MUTEX(dev_mutex);
49
50 static struct dentry *c4iw_debugfs_root;
51
52 struct c4iw_debugfs_data {
53         struct c4iw_dev *devp;
54         char *buf;
55         int bufsize;
56         int pos;
57 };
58
59 static int count_idrs(int id, void *p, void *data)
60 {
61         int *countp = data;
62
63         *countp = *countp + 1;
64         return 0;
65 }
66
67 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
68                             loff_t *ppos)
69 {
70         struct c4iw_debugfs_data *d = file->private_data;
71
72         return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
73 }
74
75 static int dump_qp(int id, void *p, void *data)
76 {
77         struct c4iw_qp *qp = p;
78         struct c4iw_debugfs_data *qpd = data;
79         int space;
80         int cc;
81
82         if (id != qp->wq.sq.qid)
83                 return 0;
84
85         space = qpd->bufsize - qpd->pos - 1;
86         if (space == 0)
87                 return 1;
88
89         if (qp->ep)
90                 cc = snprintf(qpd->buf + qpd->pos, space,
91                              "qp sq id %u rq id %u state %u onchip %u "
92                              "ep tid %u state %u %pI4:%u->%pI4:%u\n",
93                              qp->wq.sq.qid, qp->wq.rq.qid, (int)qp->attr.state,
94                              qp->wq.sq.flags & T4_SQ_ONCHIP,
95                              qp->ep->hwtid, (int)qp->ep->com.state,
96                              &qp->ep->com.local_addr.sin_addr.s_addr,
97                              ntohs(qp->ep->com.local_addr.sin_port),
98                              &qp->ep->com.remote_addr.sin_addr.s_addr,
99                              ntohs(qp->ep->com.remote_addr.sin_port));
100         else
101                 cc = snprintf(qpd->buf + qpd->pos, space,
102                              "qp sq id %u rq id %u state %u onchip %u\n",
103                               qp->wq.sq.qid, qp->wq.rq.qid,
104                               (int)qp->attr.state,
105                               qp->wq.sq.flags & T4_SQ_ONCHIP);
106         if (cc < space)
107                 qpd->pos += cc;
108         return 0;
109 }
110
111 static int qp_release(struct inode *inode, struct file *file)
112 {
113         struct c4iw_debugfs_data *qpd = file->private_data;
114         if (!qpd) {
115                 printk(KERN_INFO "%s null qpd?\n", __func__);
116                 return 0;
117         }
118         kfree(qpd->buf);
119         kfree(qpd);
120         return 0;
121 }
122
123 static int qp_open(struct inode *inode, struct file *file)
124 {
125         struct c4iw_debugfs_data *qpd;
126         int ret = 0;
127         int count = 1;
128
129         qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
130         if (!qpd) {
131                 ret = -ENOMEM;
132                 goto out;
133         }
134         qpd->devp = inode->i_private;
135         qpd->pos = 0;
136
137         spin_lock_irq(&qpd->devp->lock);
138         idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
139         spin_unlock_irq(&qpd->devp->lock);
140
141         qpd->bufsize = count * 128;
142         qpd->buf = kmalloc(qpd->bufsize, GFP_KERNEL);
143         if (!qpd->buf) {
144                 ret = -ENOMEM;
145                 goto err1;
146         }
147
148         spin_lock_irq(&qpd->devp->lock);
149         idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
150         spin_unlock_irq(&qpd->devp->lock);
151
152         qpd->buf[qpd->pos++] = 0;
153         file->private_data = qpd;
154         goto out;
155 err1:
156         kfree(qpd);
157 out:
158         return ret;
159 }
160
161 static const struct file_operations qp_debugfs_fops = {
162         .owner   = THIS_MODULE,
163         .open    = qp_open,
164         .release = qp_release,
165         .read    = debugfs_read,
166         .llseek  = default_llseek,
167 };
168
169 static int dump_stag(int id, void *p, void *data)
170 {
171         struct c4iw_debugfs_data *stagd = data;
172         int space;
173         int cc;
174
175         space = stagd->bufsize - stagd->pos - 1;
176         if (space == 0)
177                 return 1;
178
179         cc = snprintf(stagd->buf + stagd->pos, space, "0x%x\n", id<<8);
180         if (cc < space)
181                 stagd->pos += cc;
182         return 0;
183 }
184
185 static int stag_release(struct inode *inode, struct file *file)
186 {
187         struct c4iw_debugfs_data *stagd = file->private_data;
188         if (!stagd) {
189                 printk(KERN_INFO "%s null stagd?\n", __func__);
190                 return 0;
191         }
192         kfree(stagd->buf);
193         kfree(stagd);
194         return 0;
195 }
196
197 static int stag_open(struct inode *inode, struct file *file)
198 {
199         struct c4iw_debugfs_data *stagd;
200         int ret = 0;
201         int count = 1;
202
203         stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
204         if (!stagd) {
205                 ret = -ENOMEM;
206                 goto out;
207         }
208         stagd->devp = inode->i_private;
209         stagd->pos = 0;
210
211         spin_lock_irq(&stagd->devp->lock);
212         idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
213         spin_unlock_irq(&stagd->devp->lock);
214
215         stagd->bufsize = count * sizeof("0x12345678\n");
216         stagd->buf = kmalloc(stagd->bufsize, GFP_KERNEL);
217         if (!stagd->buf) {
218                 ret = -ENOMEM;
219                 goto err1;
220         }
221
222         spin_lock_irq(&stagd->devp->lock);
223         idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
224         spin_unlock_irq(&stagd->devp->lock);
225
226         stagd->buf[stagd->pos++] = 0;
227         file->private_data = stagd;
228         goto out;
229 err1:
230         kfree(stagd);
231 out:
232         return ret;
233 }
234
235 static const struct file_operations stag_debugfs_fops = {
236         .owner   = THIS_MODULE,
237         .open    = stag_open,
238         .release = stag_release,
239         .read    = debugfs_read,
240         .llseek  = default_llseek,
241 };
242
243 static int setup_debugfs(struct c4iw_dev *devp)
244 {
245         struct dentry *de;
246
247         if (!devp->debugfs_root)
248                 return -1;
249
250         de = debugfs_create_file("qps", S_IWUSR, devp->debugfs_root,
251                                  (void *)devp, &qp_debugfs_fops);
252         if (de && de->d_inode)
253                 de->d_inode->i_size = 4096;
254
255         de = debugfs_create_file("stags", S_IWUSR, devp->debugfs_root,
256                                  (void *)devp, &stag_debugfs_fops);
257         if (de && de->d_inode)
258                 de->d_inode->i_size = 4096;
259         return 0;
260 }
261
262 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
263                                struct c4iw_dev_ucontext *uctx)
264 {
265         struct list_head *pos, *nxt;
266         struct c4iw_qid_list *entry;
267
268         mutex_lock(&uctx->lock);
269         list_for_each_safe(pos, nxt, &uctx->qpids) {
270                 entry = list_entry(pos, struct c4iw_qid_list, entry);
271                 list_del_init(&entry->entry);
272                 if (!(entry->qid & rdev->qpmask))
273                         c4iw_put_resource(&rdev->resource.qid_fifo, entry->qid,
274                                           &rdev->resource.qid_fifo_lock);
275                 kfree(entry);
276         }
277
278         list_for_each_safe(pos, nxt, &uctx->qpids) {
279                 entry = list_entry(pos, struct c4iw_qid_list, entry);
280                 list_del_init(&entry->entry);
281                 kfree(entry);
282         }
283         mutex_unlock(&uctx->lock);
284 }
285
286 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
287                             struct c4iw_dev_ucontext *uctx)
288 {
289         INIT_LIST_HEAD(&uctx->qpids);
290         INIT_LIST_HEAD(&uctx->cqids);
291         mutex_init(&uctx->lock);
292 }
293
294 /* Caller takes care of locking if needed */
295 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
296 {
297         int err;
298
299         c4iw_init_dev_ucontext(rdev, &rdev->uctx);
300
301         /*
302          * qpshift is the number of bits to shift the qpid left in order
303          * to get the correct address of the doorbell for that qp.
304          */
305         rdev->qpshift = PAGE_SHIFT - ilog2(rdev->lldi.udb_density);
306         rdev->qpmask = rdev->lldi.udb_density - 1;
307         rdev->cqshift = PAGE_SHIFT - ilog2(rdev->lldi.ucq_density);
308         rdev->cqmask = rdev->lldi.ucq_density - 1;
309         PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
310              "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
311              "qp qid start %u size %u cq qid start %u size %u\n",
312              __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
313              rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
314              rdev->lldi.vr->pbl.start,
315              rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
316              rdev->lldi.vr->rq.size,
317              rdev->lldi.vr->qp.start,
318              rdev->lldi.vr->qp.size,
319              rdev->lldi.vr->cq.start,
320              rdev->lldi.vr->cq.size);
321         PDBG("udb len 0x%x udb base %p db_reg %p gts_reg %p qpshift %lu "
322              "qpmask 0x%x cqshift %lu cqmask 0x%x\n",
323              (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
324              (void *)pci_resource_start(rdev->lldi.pdev, 2),
325              rdev->lldi.db_reg,
326              rdev->lldi.gts_reg,
327              rdev->qpshift, rdev->qpmask,
328              rdev->cqshift, rdev->cqmask);
329
330         if (c4iw_num_stags(rdev) == 0) {
331                 err = -EINVAL;
332                 goto err1;
333         }
334
335         err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
336         if (err) {
337                 printk(KERN_ERR MOD "error %d initializing resources\n", err);
338                 goto err1;
339         }
340         err = c4iw_pblpool_create(rdev);
341         if (err) {
342                 printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
343                 goto err2;
344         }
345         err = c4iw_rqtpool_create(rdev);
346         if (err) {
347                 printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
348                 goto err3;
349         }
350         err = c4iw_ocqp_pool_create(rdev);
351         if (err) {
352                 printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
353                 goto err4;
354         }
355         return 0;
356 err4:
357         c4iw_rqtpool_destroy(rdev);
358 err3:
359         c4iw_pblpool_destroy(rdev);
360 err2:
361         c4iw_destroy_resource(&rdev->resource);
362 err1:
363         return err;
364 }
365
366 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
367 {
368         c4iw_pblpool_destroy(rdev);
369         c4iw_rqtpool_destroy(rdev);
370         c4iw_destroy_resource(&rdev->resource);
371 }
372
373 struct uld_ctx {
374         struct list_head entry;
375         struct cxgb4_lld_info lldi;
376         struct c4iw_dev *dev;
377 };
378
379 static void c4iw_dealloc(struct uld_ctx *ctx)
380 {
381         c4iw_rdev_close(&ctx->dev->rdev);
382         idr_destroy(&ctx->dev->cqidr);
383         idr_destroy(&ctx->dev->qpidr);
384         idr_destroy(&ctx->dev->mmidr);
385         iounmap(ctx->dev->rdev.oc_mw_kva);
386         ib_dealloc_device(&ctx->dev->ibdev);
387         ctx->dev = NULL;
388 }
389
390 static void c4iw_remove(struct uld_ctx *ctx)
391 {
392         PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
393         c4iw_unregister_device(ctx->dev);
394         c4iw_dealloc(ctx);
395 }
396
397 static int rdma_supported(const struct cxgb4_lld_info *infop)
398 {
399         return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
400                infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
401                infop->vr->cq.size > 0 && infop->vr->ocq.size > 0;
402 }
403
404 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
405 {
406         struct c4iw_dev *devp;
407         int ret;
408
409         if (!rdma_supported(infop)) {
410                 printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
411                        pci_name(infop->pdev));
412                 return ERR_PTR(-ENOSYS);
413         }
414         devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
415         if (!devp) {
416                 printk(KERN_ERR MOD "Cannot allocate ib device\n");
417                 return ERR_PTR(-ENOMEM);
418         }
419         devp->rdev.lldi = *infop;
420
421         devp->rdev.oc_mw_pa = pci_resource_start(devp->rdev.lldi.pdev, 2) +
422                 (pci_resource_len(devp->rdev.lldi.pdev, 2) -
423                  roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size));
424         devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
425                                                devp->rdev.lldi.vr->ocq.size);
426
427         PDBG(KERN_INFO MOD "ocq memory: "
428                "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
429                devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
430                devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
431
432         ret = c4iw_rdev_open(&devp->rdev);
433         if (ret) {
434                 printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
435                 ib_dealloc_device(&devp->ibdev);
436                 return ERR_PTR(ret);
437         }
438
439         idr_init(&devp->cqidr);
440         idr_init(&devp->qpidr);
441         idr_init(&devp->mmidr);
442         spin_lock_init(&devp->lock);
443
444         if (c4iw_debugfs_root) {
445                 devp->debugfs_root = debugfs_create_dir(
446                                         pci_name(devp->rdev.lldi.pdev),
447                                         c4iw_debugfs_root);
448                 setup_debugfs(devp);
449         }
450         return devp;
451 }
452
453 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
454 {
455         struct uld_ctx *ctx;
456         static int vers_printed;
457         int i;
458
459         if (!vers_printed++)
460                 printk(KERN_INFO MOD "Chelsio T4 RDMA Driver - version %s\n",
461                        DRV_VERSION);
462
463         ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
464         if (!ctx) {
465                 ctx = ERR_PTR(-ENOMEM);
466                 goto out;
467         }
468         ctx->lldi = *infop;
469
470         PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
471              __func__, pci_name(ctx->lldi.pdev),
472              ctx->lldi.nchan, ctx->lldi.nrxq,
473              ctx->lldi.ntxq, ctx->lldi.nports);
474
475         mutex_lock(&dev_mutex);
476         list_add_tail(&ctx->entry, &uld_ctx_list);
477         mutex_unlock(&dev_mutex);
478
479         for (i = 0; i < ctx->lldi.nrxq; i++)
480                 PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
481 out:
482         return ctx;
483 }
484
485 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
486                         const struct pkt_gl *gl)
487 {
488         struct uld_ctx *ctx = handle;
489         struct c4iw_dev *dev = ctx->dev;
490         struct sk_buff *skb;
491         const struct cpl_act_establish *rpl;
492         unsigned int opcode;
493
494         if (gl == NULL) {
495                 /* omit RSS and rsp_ctrl at end of descriptor */
496                 unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
497
498                 skb = alloc_skb(256, GFP_ATOMIC);
499                 if (!skb)
500                         goto nomem;
501                 __skb_put(skb, len);
502                 skb_copy_to_linear_data(skb, &rsp[1], len);
503         } else if (gl == CXGB4_MSG_AN) {
504                 const struct rsp_ctrl *rc = (void *)rsp;
505
506                 u32 qid = be32_to_cpu(rc->pldbuflen_qid);
507                 c4iw_ev_handler(dev, qid);
508                 return 0;
509         } else {
510                 skb = cxgb4_pktgl_to_skb(gl, 128, 128);
511                 if (unlikely(!skb))
512                         goto nomem;
513         }
514
515         rpl = cplhdr(skb);
516         opcode = rpl->ot.opcode;
517
518         if (c4iw_handlers[opcode])
519                 c4iw_handlers[opcode](dev, skb);
520         else
521                 printk(KERN_INFO "%s no handler opcode 0x%x...\n", __func__,
522                        opcode);
523
524         return 0;
525 nomem:
526         return -1;
527 }
528
529 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
530 {
531         struct uld_ctx *ctx = handle;
532
533         PDBG("%s new_state %u\n", __func__, new_state);
534         switch (new_state) {
535         case CXGB4_STATE_UP:
536                 printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
537                 if (!ctx->dev) {
538                         int ret;
539
540                         ctx->dev = c4iw_alloc(&ctx->lldi);
541                         if (IS_ERR(ctx->dev)) {
542                                 printk(KERN_ERR MOD
543                                        "%s: initialization failed: %ld\n",
544                                        pci_name(ctx->lldi.pdev),
545                                        PTR_ERR(ctx->dev));
546                                 ctx->dev = NULL;
547                                 break;
548                         }
549                         ret = c4iw_register_device(ctx->dev);
550                         if (ret) {
551                                 printk(KERN_ERR MOD
552                                        "%s: RDMA registration failed: %d\n",
553                                        pci_name(ctx->lldi.pdev), ret);
554                                 c4iw_dealloc(ctx);
555                         }
556                 }
557                 break;
558         case CXGB4_STATE_DOWN:
559                 printk(KERN_INFO MOD "%s: Down\n",
560                        pci_name(ctx->lldi.pdev));
561                 if (ctx->dev)
562                         c4iw_remove(ctx);
563                 break;
564         case CXGB4_STATE_START_RECOVERY:
565                 printk(KERN_INFO MOD "%s: Fatal Error\n",
566                        pci_name(ctx->lldi.pdev));
567                 if (ctx->dev) {
568                         struct ib_event event;
569
570                         ctx->dev->rdev.flags |= T4_FATAL_ERROR;
571                         memset(&event, 0, sizeof event);
572                         event.event  = IB_EVENT_DEVICE_FATAL;
573                         event.device = &ctx->dev->ibdev;
574                         ib_dispatch_event(&event);
575                         c4iw_remove(ctx);
576                 }
577                 break;
578         case CXGB4_STATE_DETACH:
579                 printk(KERN_INFO MOD "%s: Detach\n",
580                        pci_name(ctx->lldi.pdev));
581                 if (ctx->dev)
582                         c4iw_remove(ctx);
583                 break;
584         }
585         return 0;
586 }
587
588 static struct cxgb4_uld_info c4iw_uld_info = {
589         .name = DRV_NAME,
590         .add = c4iw_uld_add,
591         .rx_handler = c4iw_uld_rx_handler,
592         .state_change = c4iw_uld_state_change,
593 };
594
595 static int __init c4iw_init_module(void)
596 {
597         int err;
598
599         err = c4iw_cm_init();
600         if (err)
601                 return err;
602
603         c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
604         if (!c4iw_debugfs_root)
605                 printk(KERN_WARNING MOD
606                        "could not create debugfs entry, continuing\n");
607
608         cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
609
610         return 0;
611 }
612
613 static void __exit c4iw_exit_module(void)
614 {
615         struct uld_ctx *ctx, *tmp;
616
617         mutex_lock(&dev_mutex);
618         list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
619                 if (ctx->dev)
620                         c4iw_remove(ctx);
621                 kfree(ctx);
622         }
623         mutex_unlock(&dev_mutex);
624         cxgb4_unregister_uld(CXGB4_ULD_RDMA);
625         c4iw_cm_term();
626         debugfs_remove_recursive(c4iw_debugfs_root);
627 }
628
629 module_init(c4iw_init_module);
630 module_exit(c4iw_exit_module);