1 /*
2 * Copyright (c) 2006 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/device.h>
35 #include <linux/netdevice.h>
36 #include <linux/etherdevice.h>
37 #include <linux/delay.h>
38 #include <linux/errno.h>
39 #include <linux/list.h>
40 #include <linux/sched.h>
41 #include <linux/spinlock.h>
42 #include <linux/ethtool.h>
43 #include <linux/rtnetlink.h>
44 #include <linux/inetdevice.h>
45 #include <linux/slab.h>
46
47 #include <asm/io.h>
48 #include <asm/irq.h>
49 #include <asm/byteorder.h>
50
51 #include <rdma/iw_cm.h>
52 #include <rdma/ib_verbs.h>
53 #include <rdma/ib_smi.h>
54 #include <rdma/ib_umem.h>
55 #include <rdma/ib_user_verbs.h>
56
57 #include "cxio_hal.h"
58 #include "iwch.h"
59 #include "iwch_provider.h"
60 #include "iwch_cm.h"
61 #include "iwch_user.h"
62 #include "common.h"
63
iwch_ah_create(struct ib_pd * pd,struct ib_ah_attr * ah_attr)64 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
65 struct ib_ah_attr *ah_attr)
66 {
67 return ERR_PTR(-ENOSYS);
68 }
69
iwch_ah_destroy(struct ib_ah * ah)70 static int iwch_ah_destroy(struct ib_ah *ah)
71 {
72 return -ENOSYS;
73 }
74
iwch_multicast_attach(struct ib_qp * ibqp,union ib_gid * gid,u16 lid)75 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
76 {
77 return -ENOSYS;
78 }
79
iwch_multicast_detach(struct ib_qp * ibqp,union ib_gid * gid,u16 lid)80 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
81 {
82 return -ENOSYS;
83 }
84
iwch_process_mad(struct ib_device * ibdev,int mad_flags,u8 port_num,const struct ib_wc * in_wc,const struct ib_grh * in_grh,const struct ib_mad_hdr * in_mad,size_t in_mad_size,struct ib_mad_hdr * out_mad,size_t * out_mad_size,u16 * out_mad_pkey_index)85 static int iwch_process_mad(struct ib_device *ibdev,
86 int mad_flags,
87 u8 port_num,
88 const struct ib_wc *in_wc,
89 const struct ib_grh *in_grh,
90 const struct ib_mad_hdr *in_mad,
91 size_t in_mad_size,
92 struct ib_mad_hdr *out_mad,
93 size_t *out_mad_size,
94 u16 *out_mad_pkey_index)
95 {
96 return -ENOSYS;
97 }
98
iwch_dealloc_ucontext(struct ib_ucontext * context)99 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
100 {
101 struct iwch_dev *rhp = to_iwch_dev(context->device);
102 struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
103 struct iwch_mm_entry *mm, *tmp;
104
105 PDBG("%s context %p\n", __func__, context);
106 list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
107 kfree(mm);
108 cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
109 kfree(ucontext);
110 return 0;
111 }
112
iwch_alloc_ucontext(struct ib_device * ibdev,struct ib_udata * udata)113 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
114 struct ib_udata *udata)
115 {
116 struct iwch_ucontext *context;
117 struct iwch_dev *rhp = to_iwch_dev(ibdev);
118
119 PDBG("%s ibdev %p\n", __func__, ibdev);
120 context = kzalloc(sizeof(*context), GFP_KERNEL);
121 if (!context)
122 return ERR_PTR(-ENOMEM);
123 cxio_init_ucontext(&rhp->rdev, &context->uctx);
124 INIT_LIST_HEAD(&context->mmaps);
125 spin_lock_init(&context->mmap_lock);
126 return &context->ibucontext;
127 }
128
iwch_destroy_cq(struct ib_cq * ib_cq)129 static int iwch_destroy_cq(struct ib_cq *ib_cq)
130 {
131 struct iwch_cq *chp;
132
133 PDBG("%s ib_cq %p\n", __func__, ib_cq);
134 chp = to_iwch_cq(ib_cq);
135
136 remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
137 atomic_dec(&chp->refcnt);
138 wait_event(chp->wait, !atomic_read(&chp->refcnt));
139
140 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
141 kfree(chp);
142 return 0;
143 }
144
iwch_create_cq(struct ib_device * ibdev,const struct ib_cq_init_attr * attr,struct ib_ucontext * ib_context,struct ib_udata * udata)145 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev,
146 const struct ib_cq_init_attr *attr,
147 struct ib_ucontext *ib_context,
148 struct ib_udata *udata)
149 {
150 int entries = attr->cqe;
151 struct iwch_dev *rhp;
152 struct iwch_cq *chp;
153 struct iwch_create_cq_resp uresp;
154 struct iwch_create_cq_req ureq;
155 struct iwch_ucontext *ucontext = NULL;
156 static int warned;
157 size_t resplen;
158
159 PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
160 if (attr->flags)
161 return ERR_PTR(-EINVAL);
162
163 rhp = to_iwch_dev(ibdev);
164 chp = kzalloc(sizeof(*chp), GFP_KERNEL);
165 if (!chp)
166 return ERR_PTR(-ENOMEM);
167
168 if (ib_context) {
169 ucontext = to_iwch_ucontext(ib_context);
170 if (!t3a_device(rhp)) {
171 if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
172 kfree(chp);
173 return ERR_PTR(-EFAULT);
174 }
175 chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
176 }
177 }
178
179 if (t3a_device(rhp)) {
180
181 /*
182 * T3A: Add some fluff to handle extra CQEs inserted
183 * for various errors.
184 * Additional CQE possibilities:
185 * TERMINATE,
186 * incoming RDMA WRITE Failures
187 * incoming RDMA READ REQUEST FAILUREs
188 * NOTE: We cannot ensure the CQ won't overflow.
189 */
190 entries += 16;
191 }
192 entries = roundup_pow_of_two(entries);
193 chp->cq.size_log2 = ilog2(entries);
194
195 if (cxio_create_cq(&rhp->rdev, &chp->cq, !ucontext)) {
196 kfree(chp);
197 return ERR_PTR(-ENOMEM);
198 }
199 chp->rhp = rhp;
200 chp->ibcq.cqe = 1 << chp->cq.size_log2;
201 spin_lock_init(&chp->lock);
202 spin_lock_init(&chp->comp_handler_lock);
203 atomic_set(&chp->refcnt, 1);
204 init_waitqueue_head(&chp->wait);
205 if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
206 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
207 kfree(chp);
208 return ERR_PTR(-ENOMEM);
209 }
210
211 if (ucontext) {
212 struct iwch_mm_entry *mm;
213
214 mm = kmalloc(sizeof *mm, GFP_KERNEL);
215 if (!mm) {
216 iwch_destroy_cq(&chp->ibcq);
217 return ERR_PTR(-ENOMEM);
218 }
219 uresp.cqid = chp->cq.cqid;
220 uresp.size_log2 = chp->cq.size_log2;
221 spin_lock(&ucontext->mmap_lock);
222 uresp.key = ucontext->key;
223 ucontext->key += PAGE_SIZE;
224 spin_unlock(&ucontext->mmap_lock);
225 mm->key = uresp.key;
226 mm->addr = virt_to_phys(chp->cq.queue);
227 if (udata->outlen < sizeof uresp) {
228 if (!warned++)
229 printk(KERN_WARNING MOD "Warning - "
230 "downlevel libcxgb3 (non-fatal).\n");
231 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
232 sizeof(struct t3_cqe));
233 resplen = sizeof(struct iwch_create_cq_resp_v0);
234 } else {
235 mm->len = PAGE_ALIGN(((1UL << uresp.size_log2) + 1) *
236 sizeof(struct t3_cqe));
237 uresp.memsize = mm->len;
238 uresp.reserved = 0;
239 resplen = sizeof uresp;
240 }
241 if (ib_copy_to_udata(udata, &uresp, resplen)) {
242 kfree(mm);
243 iwch_destroy_cq(&chp->ibcq);
244 return ERR_PTR(-EFAULT);
245 }
246 insert_mmap(ucontext, mm);
247 }
248 PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
249 chp->cq.cqid, chp, (1 << chp->cq.size_log2),
250 (unsigned long long) chp->cq.dma_addr);
251 return &chp->ibcq;
252 }
253
iwch_resize_cq(struct ib_cq * cq,int cqe,struct ib_udata * udata)254 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
255 {
256 #ifdef notyet
257 struct iwch_cq *chp = to_iwch_cq(cq);
258 struct t3_cq oldcq, newcq;
259 int ret;
260
261 PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
262
263 /* We don't downsize... */
264 if (cqe <= cq->cqe)
265 return 0;
266
267 /* create new t3_cq with new size */
268 cqe = roundup_pow_of_two(cqe+1);
269 newcq.size_log2 = ilog2(cqe);
270
271 /* Dont allow resize to less than the current wce count */
272 if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
273 return -ENOMEM;
274 }
275
276 /* Quiesce all QPs using this CQ */
277 ret = iwch_quiesce_qps(chp);
278 if (ret) {
279 return ret;
280 }
281
282 ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
283 if (ret) {
284 return ret;
285 }
286
287 /* copy CQEs */
288 memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
289 sizeof(struct t3_cqe));
290
291 /* old iwch_qp gets new t3_cq but keeps old cqid */
292 oldcq = chp->cq;
293 chp->cq = newcq;
294 chp->cq.cqid = oldcq.cqid;
295
296 /* resize new t3_cq to update the HW context */
297 ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
298 if (ret) {
299 chp->cq = oldcq;
300 return ret;
301 }
302 chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
303
304 /* destroy old t3_cq */
305 oldcq.cqid = newcq.cqid;
306 ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
307 if (ret) {
308 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
309 __func__, ret);
310 }
311
312 /* add user hooks here */
313
314 /* resume qps */
315 ret = iwch_resume_qps(chp);
316 return ret;
317 #else
318 return -ENOSYS;
319 #endif
320 }
321
iwch_arm_cq(struct ib_cq * ibcq,enum ib_cq_notify_flags flags)322 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
323 {
324 struct iwch_dev *rhp;
325 struct iwch_cq *chp;
326 enum t3_cq_opcode cq_op;
327 int err;
328 unsigned long flag;
329 u32 rptr;
330
331 chp = to_iwch_cq(ibcq);
332 rhp = chp->rhp;
333 if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
334 cq_op = CQ_ARM_SE;
335 else
336 cq_op = CQ_ARM_AN;
337 if (chp->user_rptr_addr) {
338 if (get_user(rptr, chp->user_rptr_addr))
339 return -EFAULT;
340 spin_lock_irqsave(&chp->lock, flag);
341 chp->cq.rptr = rptr;
342 } else
343 spin_lock_irqsave(&chp->lock, flag);
344 PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
345 err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
346 spin_unlock_irqrestore(&chp->lock, flag);
347 if (err < 0)
348 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
349 chp->cq.cqid);
350 if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
351 err = 0;
352 return err;
353 }
354
iwch_mmap(struct ib_ucontext * context,struct vm_area_struct * vma)355 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
356 {
357 int len = vma->vm_end - vma->vm_start;
358 u32 key = vma->vm_pgoff << PAGE_SHIFT;
359 struct cxio_rdev *rdev_p;
360 int ret = 0;
361 struct iwch_mm_entry *mm;
362 struct iwch_ucontext *ucontext;
363 u64 addr;
364
365 PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
366 key, len);
367
368 if (vma->vm_start & (PAGE_SIZE-1)) {
369 return -EINVAL;
370 }
371
372 rdev_p = &(to_iwch_dev(context->device)->rdev);
373 ucontext = to_iwch_ucontext(context);
374
375 mm = remove_mmap(ucontext, key, len);
376 if (!mm)
377 return -EINVAL;
378 addr = mm->addr;
379 kfree(mm);
380
381 if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
382 (addr < (rdev_p->rnic_info.udbell_physbase +
383 rdev_p->rnic_info.udbell_len))) {
384
385 /*
386 * Map T3 DB register.
387 */
388 if (vma->vm_flags & VM_READ) {
389 return -EPERM;
390 }
391
392 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
393 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
394 vma->vm_flags &= ~VM_MAYREAD;
395 ret = io_remap_pfn_range(vma, vma->vm_start,
396 addr >> PAGE_SHIFT,
397 len, vma->vm_page_prot);
398 } else {
399
400 /*
401 * Map WQ or CQ contig dma memory...
402 */
403 ret = remap_pfn_range(vma, vma->vm_start,
404 addr >> PAGE_SHIFT,
405 len, vma->vm_page_prot);
406 }
407
408 return ret;
409 }
410
iwch_deallocate_pd(struct ib_pd * pd)411 static int iwch_deallocate_pd(struct ib_pd *pd)
412 {
413 struct iwch_dev *rhp;
414 struct iwch_pd *php;
415
416 php = to_iwch_pd(pd);
417 rhp = php->rhp;
418 PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
419 cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
420 kfree(php);
421 return 0;
422 }
423
iwch_allocate_pd(struct ib_device * ibdev,struct ib_ucontext * context,struct ib_udata * udata)424 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
425 struct ib_ucontext *context,
426 struct ib_udata *udata)
427 {
428 struct iwch_pd *php;
429 u32 pdid;
430 struct iwch_dev *rhp;
431
432 PDBG("%s ibdev %p\n", __func__, ibdev);
433 rhp = (struct iwch_dev *) ibdev;
434 pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
435 if (!pdid)
436 return ERR_PTR(-EINVAL);
437 php = kzalloc(sizeof(*php), GFP_KERNEL);
438 if (!php) {
439 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
440 return ERR_PTR(-ENOMEM);
441 }
442 php->pdid = pdid;
443 php->rhp = rhp;
444 if (context) {
445 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
446 iwch_deallocate_pd(&php->ibpd);
447 return ERR_PTR(-EFAULT);
448 }
449 }
450 PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
451 return &php->ibpd;
452 }
453
iwch_dereg_mr(struct ib_mr * ib_mr)454 static int iwch_dereg_mr(struct ib_mr *ib_mr)
455 {
456 struct iwch_dev *rhp;
457 struct iwch_mr *mhp;
458 u32 mmid;
459
460 PDBG("%s ib_mr %p\n", __func__, ib_mr);
461 /* There can be no memory windows */
462 if (atomic_read(&ib_mr->usecnt))
463 return -EINVAL;
464
465 mhp = to_iwch_mr(ib_mr);
466 kfree(mhp->pages);
467 rhp = mhp->rhp;
468 mmid = mhp->attr.stag >> 8;
469 cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
470 mhp->attr.pbl_addr);
471 iwch_free_pbl(mhp);
472 remove_handle(rhp, &rhp->mmidr, mmid);
473 if (mhp->kva)
474 kfree((void *) (unsigned long) mhp->kva);
475 if (mhp->umem)
476 ib_umem_release(mhp->umem);
477 PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
478 kfree(mhp);
479 return 0;
480 }
481
iwch_register_phys_mem(struct ib_pd * pd,struct ib_phys_buf * buffer_list,int num_phys_buf,int acc,u64 * iova_start)482 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
483 struct ib_phys_buf *buffer_list,
484 int num_phys_buf,
485 int acc,
486 u64 *iova_start)
487 {
488 __be64 *page_list;
489 int shift;
490 u64 total_size;
491 int npages;
492 struct iwch_dev *rhp;
493 struct iwch_pd *php;
494 struct iwch_mr *mhp;
495 int ret;
496
497 PDBG("%s ib_pd %p\n", __func__, pd);
498 php = to_iwch_pd(pd);
499 rhp = php->rhp;
500
501 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
502 if (!mhp)
503 return ERR_PTR(-ENOMEM);
504
505 mhp->rhp = rhp;
506
507 /* First check that we have enough alignment */
508 if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
509 ret = -EINVAL;
510 goto err;
511 }
512
513 if (num_phys_buf > 1 &&
514 ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
515 ret = -EINVAL;
516 goto err;
517 }
518
519 ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
520 &total_size, &npages, &shift, &page_list);
521 if (ret)
522 goto err;
523
524 ret = iwch_alloc_pbl(mhp, npages);
525 if (ret) {
526 kfree(page_list);
527 goto err_pbl;
528 }
529
530 ret = iwch_write_pbl(mhp, page_list, npages, 0);
531 kfree(page_list);
532 if (ret)
533 goto err_pbl;
534
535 mhp->attr.pdid = php->pdid;
536 mhp->attr.zbva = 0;
537
538 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
539 mhp->attr.va_fbo = *iova_start;
540 mhp->attr.page_size = shift - 12;
541
542 mhp->attr.len = (u32) total_size;
543 mhp->attr.pbl_size = npages;
544 ret = iwch_register_mem(rhp, php, mhp, shift);
545 if (ret)
546 goto err_pbl;
547
548 return &mhp->ibmr;
549
550 err_pbl:
551 iwch_free_pbl(mhp);
552
553 err:
554 kfree(mhp);
555 return ERR_PTR(ret);
556
557 }
558
iwch_reregister_phys_mem(struct ib_mr * mr,int mr_rereg_mask,struct ib_pd * pd,struct ib_phys_buf * buffer_list,int num_phys_buf,int acc,u64 * iova_start)559 static int iwch_reregister_phys_mem(struct ib_mr *mr,
560 int mr_rereg_mask,
561 struct ib_pd *pd,
562 struct ib_phys_buf *buffer_list,
563 int num_phys_buf,
564 int acc, u64 * iova_start)
565 {
566
567 struct iwch_mr mh, *mhp;
568 struct iwch_pd *php;
569 struct iwch_dev *rhp;
570 __be64 *page_list = NULL;
571 int shift = 0;
572 u64 total_size;
573 int npages = 0;
574 int ret;
575
576 PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
577
578 /* There can be no memory windows */
579 if (atomic_read(&mr->usecnt))
580 return -EINVAL;
581
582 mhp = to_iwch_mr(mr);
583 rhp = mhp->rhp;
584 php = to_iwch_pd(mr->pd);
585
586 /* make sure we are on the same adapter */
587 if (rhp != php->rhp)
588 return -EINVAL;
589
590 memcpy(&mh, mhp, sizeof *mhp);
591
592 if (mr_rereg_mask & IB_MR_REREG_PD)
593 php = to_iwch_pd(pd);
594 if (mr_rereg_mask & IB_MR_REREG_ACCESS)
595 mh.attr.perms = iwch_ib_to_tpt_access(acc);
596 if (mr_rereg_mask & IB_MR_REREG_TRANS) {
597 ret = build_phys_page_list(buffer_list, num_phys_buf,
598 iova_start,
599 &total_size, &npages,
600 &shift, &page_list);
601 if (ret)
602 return ret;
603 }
604
605 ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
606 kfree(page_list);
607 if (ret) {
608 return ret;
609 }
610 if (mr_rereg_mask & IB_MR_REREG_PD)
611 mhp->attr.pdid = php->pdid;
612 if (mr_rereg_mask & IB_MR_REREG_ACCESS)
613 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
614 if (mr_rereg_mask & IB_MR_REREG_TRANS) {
615 mhp->attr.zbva = 0;
616 mhp->attr.va_fbo = *iova_start;
617 mhp->attr.page_size = shift - 12;
618 mhp->attr.len = (u32) total_size;
619 mhp->attr.pbl_size = npages;
620 }
621
622 return 0;
623 }
624
625
iwch_reg_user_mr(struct ib_pd * pd,u64 start,u64 length,u64 virt,int acc,struct ib_udata * udata)626 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
627 u64 virt, int acc, struct ib_udata *udata)
628 {
629 __be64 *pages;
630 int shift, n, len;
631 int i, k, entry;
632 int err = 0;
633 struct iwch_dev *rhp;
634 struct iwch_pd *php;
635 struct iwch_mr *mhp;
636 struct iwch_reg_user_mr_resp uresp;
637 struct scatterlist *sg;
638 PDBG("%s ib_pd %p\n", __func__, pd);
639
640 php = to_iwch_pd(pd);
641 rhp = php->rhp;
642 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
643 if (!mhp)
644 return ERR_PTR(-ENOMEM);
645
646 mhp->rhp = rhp;
647
648 mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
649 if (IS_ERR(mhp->umem)) {
650 err = PTR_ERR(mhp->umem);
651 kfree(mhp);
652 return ERR_PTR(err);
653 }
654
655 shift = ffs(mhp->umem->page_size) - 1;
656
657 n = mhp->umem->nmap;
658
659 err = iwch_alloc_pbl(mhp, n);
660 if (err)
661 goto err;
662
663 pages = (__be64 *) __get_free_page(GFP_KERNEL);
664 if (!pages) {
665 err = -ENOMEM;
666 goto err_pbl;
667 }
668
669 i = n = 0;
670
671 for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
672 len = sg_dma_len(sg) >> shift;
673 for (k = 0; k < len; ++k) {
674 pages[i++] = cpu_to_be64(sg_dma_address(sg) +
675 mhp->umem->page_size * k);
676 if (i == PAGE_SIZE / sizeof *pages) {
677 err = iwch_write_pbl(mhp, pages, i, n);
678 if (err)
679 goto pbl_done;
680 n += i;
681 i = 0;
682 }
683 }
684 }
685
686 if (i)
687 err = iwch_write_pbl(mhp, pages, i, n);
688
689 pbl_done:
690 free_page((unsigned long) pages);
691 if (err)
692 goto err_pbl;
693
694 mhp->attr.pdid = php->pdid;
695 mhp->attr.zbva = 0;
696 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
697 mhp->attr.va_fbo = virt;
698 mhp->attr.page_size = shift - 12;
699 mhp->attr.len = (u32) length;
700
701 err = iwch_register_mem(rhp, php, mhp, shift);
702 if (err)
703 goto err_pbl;
704
705 if (udata && !t3a_device(rhp)) {
706 uresp.pbl_addr = (mhp->attr.pbl_addr -
707 rhp->rdev.rnic_info.pbl_base) >> 3;
708 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
709 uresp.pbl_addr);
710
711 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
712 iwch_dereg_mr(&mhp->ibmr);
713 err = -EFAULT;
714 goto err;
715 }
716 }
717
718 return &mhp->ibmr;
719
720 err_pbl:
721 iwch_free_pbl(mhp);
722
723 err:
724 ib_umem_release(mhp->umem);
725 kfree(mhp);
726 return ERR_PTR(err);
727 }
728
iwch_get_dma_mr(struct ib_pd * pd,int acc)729 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
730 {
731 struct ib_phys_buf bl;
732 u64 kva;
733 struct ib_mr *ibmr;
734
735 PDBG("%s ib_pd %p\n", __func__, pd);
736
737 /*
738 * T3 only supports 32 bits of size.
739 */
740 if (sizeof(phys_addr_t) > 4) {
741 pr_warn_once(MOD "Cannot support dma_mrs on this platform.\n");
742 return ERR_PTR(-ENOTSUPP);
743 }
744 bl.size = 0xffffffff;
745 bl.addr = 0;
746 kva = 0;
747 ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
748 return ibmr;
749 }
750
iwch_alloc_mw(struct ib_pd * pd,enum ib_mw_type type)751 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
752 {
753 struct iwch_dev *rhp;
754 struct iwch_pd *php;
755 struct iwch_mw *mhp;
756 u32 mmid;
757 u32 stag = 0;
758 int ret;
759
760 if (type != IB_MW_TYPE_1)
761 return ERR_PTR(-EINVAL);
762
763 php = to_iwch_pd(pd);
764 rhp = php->rhp;
765 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
766 if (!mhp)
767 return ERR_PTR(-ENOMEM);
768 ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
769 if (ret) {
770 kfree(mhp);
771 return ERR_PTR(ret);
772 }
773 mhp->rhp = rhp;
774 mhp->attr.pdid = php->pdid;
775 mhp->attr.type = TPT_MW;
776 mhp->attr.stag = stag;
777 mmid = (stag) >> 8;
778 mhp->ibmw.rkey = stag;
779 if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
780 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
781 kfree(mhp);
782 return ERR_PTR(-ENOMEM);
783 }
784 PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
785 return &(mhp->ibmw);
786 }
787
iwch_dealloc_mw(struct ib_mw * mw)788 static int iwch_dealloc_mw(struct ib_mw *mw)
789 {
790 struct iwch_dev *rhp;
791 struct iwch_mw *mhp;
792 u32 mmid;
793
794 mhp = to_iwch_mw(mw);
795 rhp = mhp->rhp;
796 mmid = (mw->rkey) >> 8;
797 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
798 remove_handle(rhp, &rhp->mmidr, mmid);
799 PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
800 kfree(mhp);
801 return 0;
802 }
803
iwch_alloc_mr(struct ib_pd * pd,enum ib_mr_type mr_type,u32 max_num_sg)804 static struct ib_mr *iwch_alloc_mr(struct ib_pd *pd,
805 enum ib_mr_type mr_type,
806 u32 max_num_sg)
807 {
808 struct iwch_dev *rhp;
809 struct iwch_pd *php;
810 struct iwch_mr *mhp;
811 u32 mmid;
812 u32 stag = 0;
813 int ret = 0;
814
815 if (mr_type != IB_MR_TYPE_MEM_REG ||
816 max_num_sg > T3_MAX_FASTREG_DEPTH)
817 return ERR_PTR(-EINVAL);
818
819 php = to_iwch_pd(pd);
820 rhp = php->rhp;
821 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
822 if (!mhp)
823 goto err;
824
825 mhp->pages = kcalloc(max_num_sg, sizeof(u64), GFP_KERNEL);
826 if (!mhp->pages) {
827 ret = -ENOMEM;
828 goto pl_err;
829 }
830
831 mhp->rhp = rhp;
832 ret = iwch_alloc_pbl(mhp, max_num_sg);
833 if (ret)
834 goto err1;
835 mhp->attr.pbl_size = max_num_sg;
836 ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
837 mhp->attr.pbl_size, mhp->attr.pbl_addr);
838 if (ret)
839 goto err2;
840 mhp->attr.pdid = php->pdid;
841 mhp->attr.type = TPT_NON_SHARED_MR;
842 mhp->attr.stag = stag;
843 mhp->attr.state = 1;
844 mmid = (stag) >> 8;
845 mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
846 if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
847 goto err3;
848
849 PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
850 return &(mhp->ibmr);
851 err3:
852 cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
853 mhp->attr.pbl_addr);
854 err2:
855 iwch_free_pbl(mhp);
856 err1:
857 kfree(mhp->pages);
858 pl_err:
859 kfree(mhp);
860 err:
861 return ERR_PTR(ret);
862 }
863
iwch_set_page(struct ib_mr * ibmr,u64 addr)864 static int iwch_set_page(struct ib_mr *ibmr, u64 addr)
865 {
866 struct iwch_mr *mhp = to_iwch_mr(ibmr);
867
868 if (unlikely(mhp->npages == mhp->attr.pbl_size))
869 return -ENOMEM;
870
871 mhp->pages[mhp->npages++] = addr;
872
873 return 0;
874 }
875
iwch_map_mr_sg(struct ib_mr * ibmr,struct scatterlist * sg,int sg_nents)876 static int iwch_map_mr_sg(struct ib_mr *ibmr,
877 struct scatterlist *sg,
878 int sg_nents)
879 {
880 struct iwch_mr *mhp = to_iwch_mr(ibmr);
881
882 mhp->npages = 0;
883
884 return ib_sg_to_pages(ibmr, sg, sg_nents, iwch_set_page);
885 }
886
iwch_destroy_qp(struct ib_qp * ib_qp)887 static int iwch_destroy_qp(struct ib_qp *ib_qp)
888 {
889 struct iwch_dev *rhp;
890 struct iwch_qp *qhp;
891 struct iwch_qp_attributes attrs;
892 struct iwch_ucontext *ucontext;
893
894 qhp = to_iwch_qp(ib_qp);
895 rhp = qhp->rhp;
896
897 attrs.next_state = IWCH_QP_STATE_ERROR;
898 iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
899 wait_event(qhp->wait, !qhp->ep);
900
901 remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
902
903 atomic_dec(&qhp->refcnt);
904 wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
905
906 ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
907 : NULL;
908 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
909 ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
910
911 PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
912 ib_qp, qhp->wq.qpid, qhp);
913 kfree(qhp);
914 return 0;
915 }
916
iwch_create_qp(struct ib_pd * pd,struct ib_qp_init_attr * attrs,struct ib_udata * udata)917 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
918 struct ib_qp_init_attr *attrs,
919 struct ib_udata *udata)
920 {
921 struct iwch_dev *rhp;
922 struct iwch_qp *qhp;
923 struct iwch_pd *php;
924 struct iwch_cq *schp;
925 struct iwch_cq *rchp;
926 struct iwch_create_qp_resp uresp;
927 int wqsize, sqsize, rqsize;
928 struct iwch_ucontext *ucontext;
929
930 PDBG("%s ib_pd %p\n", __func__, pd);
931 if (attrs->qp_type != IB_QPT_RC)
932 return ERR_PTR(-EINVAL);
933 php = to_iwch_pd(pd);
934 rhp = php->rhp;
935 schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
936 rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
937 if (!schp || !rchp)
938 return ERR_PTR(-EINVAL);
939
940 /* The RQT size must be # of entries + 1 rounded up to a power of two */
941 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
942 if (rqsize == attrs->cap.max_recv_wr)
943 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
944
945 /* T3 doesn't support RQT depth < 16 */
946 if (rqsize < 16)
947 rqsize = 16;
948
949 if (rqsize > T3_MAX_RQ_SIZE)
950 return ERR_PTR(-EINVAL);
951
952 if (attrs->cap.max_inline_data > T3_MAX_INLINE)
953 return ERR_PTR(-EINVAL);
954
955 /*
956 * NOTE: The SQ and total WQ sizes don't need to be
957 * a power of two. However, all the code assumes
958 * they are. EG: Q_FREECNT() and friends.
959 */
960 sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
961 wqsize = roundup_pow_of_two(rqsize + sqsize);
962
963 /*
964 * Kernel users need more wq space for fastreg WRs which can take
965 * 2 WR fragments.
966 */
967 ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
968 if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
969 wqsize = roundup_pow_of_two(rqsize +
970 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
971 PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
972 wqsize, sqsize, rqsize);
973 qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
974 if (!qhp)
975 return ERR_PTR(-ENOMEM);
976 qhp->wq.size_log2 = ilog2(wqsize);
977 qhp->wq.rq_size_log2 = ilog2(rqsize);
978 qhp->wq.sq_size_log2 = ilog2(sqsize);
979 if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
980 ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
981 kfree(qhp);
982 return ERR_PTR(-ENOMEM);
983 }
984
985 attrs->cap.max_recv_wr = rqsize - 1;
986 attrs->cap.max_send_wr = sqsize;
987 attrs->cap.max_inline_data = T3_MAX_INLINE;
988
989 qhp->rhp = rhp;
990 qhp->attr.pd = php->pdid;
991 qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
992 qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
993 qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
994 qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
995 qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
996 qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
997 qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
998 qhp->attr.state = IWCH_QP_STATE_IDLE;
999 qhp->attr.next_state = IWCH_QP_STATE_IDLE;
1000
1001 /*
1002 * XXX - These don't get passed in from the openib user
1003 * at create time. The CM sets them via a QP modify.
1004 * Need to fix... I think the CM should
1005 */
1006 qhp->attr.enable_rdma_read = 1;
1007 qhp->attr.enable_rdma_write = 1;
1008 qhp->attr.enable_bind = 1;
1009 qhp->attr.max_ord = 1;
1010 qhp->attr.max_ird = 1;
1011
1012 spin_lock_init(&qhp->lock);
1013 init_waitqueue_head(&qhp->wait);
1014 atomic_set(&qhp->refcnt, 1);
1015
1016 if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
1017 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
1018 ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
1019 kfree(qhp);
1020 return ERR_PTR(-ENOMEM);
1021 }
1022
1023 if (udata) {
1024
1025 struct iwch_mm_entry *mm1, *mm2;
1026
1027 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
1028 if (!mm1) {
1029 iwch_destroy_qp(&qhp->ibqp);
1030 return ERR_PTR(-ENOMEM);
1031 }
1032
1033 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
1034 if (!mm2) {
1035 kfree(mm1);
1036 iwch_destroy_qp(&qhp->ibqp);
1037 return ERR_PTR(-ENOMEM);
1038 }
1039
1040 uresp.qpid = qhp->wq.qpid;
1041 uresp.size_log2 = qhp->wq.size_log2;
1042 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
1043 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
1044 spin_lock(&ucontext->mmap_lock);
1045 uresp.key = ucontext->key;
1046 ucontext->key += PAGE_SIZE;
1047 uresp.db_key = ucontext->key;
1048 ucontext->key += PAGE_SIZE;
1049 spin_unlock(&ucontext->mmap_lock);
1050 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
1051 kfree(mm1);
1052 kfree(mm2);
1053 iwch_destroy_qp(&qhp->ibqp);
1054 return ERR_PTR(-EFAULT);
1055 }
1056 mm1->key = uresp.key;
1057 mm1->addr = virt_to_phys(qhp->wq.queue);
1058 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1059 insert_mmap(ucontext, mm1);
1060 mm2->key = uresp.db_key;
1061 mm2->addr = qhp->wq.udb & PAGE_MASK;
1062 mm2->len = PAGE_SIZE;
1063 insert_mmap(ucontext, mm2);
1064 }
1065 qhp->ibqp.qp_num = qhp->wq.qpid;
1066 init_timer(&(qhp->timer));
1067 PDBG("%s sq_num_entries %d, rq_num_entries %d "
1068 "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1069 __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1070 qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1071 1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1072 return &qhp->ibqp;
1073 }
1074
iwch_ib_modify_qp(struct ib_qp * ibqp,struct ib_qp_attr * attr,int attr_mask,struct ib_udata * udata)1075 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1076 int attr_mask, struct ib_udata *udata)
1077 {
1078 struct iwch_dev *rhp;
1079 struct iwch_qp *qhp;
1080 enum iwch_qp_attr_mask mask = 0;
1081 struct iwch_qp_attributes attrs;
1082
1083 PDBG("%s ib_qp %p\n", __func__, ibqp);
1084
1085 /* iwarp does not support the RTR state */
1086 if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1087 attr_mask &= ~IB_QP_STATE;
1088
1089 /* Make sure we still have something left to do */
1090 if (!attr_mask)
1091 return 0;
1092
1093 memset(&attrs, 0, sizeof attrs);
1094 qhp = to_iwch_qp(ibqp);
1095 rhp = qhp->rhp;
1096
1097 attrs.next_state = iwch_convert_state(attr->qp_state);
1098 attrs.enable_rdma_read = (attr->qp_access_flags &
1099 IB_ACCESS_REMOTE_READ) ? 1 : 0;
1100 attrs.enable_rdma_write = (attr->qp_access_flags &
1101 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1102 attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1103
1104
1105 mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1106 mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1107 (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1108 IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1109 IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1110
1111 return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1112 }
1113
iwch_qp_add_ref(struct ib_qp * qp)1114 void iwch_qp_add_ref(struct ib_qp *qp)
1115 {
1116 PDBG("%s ib_qp %p\n", __func__, qp);
1117 atomic_inc(&(to_iwch_qp(qp)->refcnt));
1118 }
1119
iwch_qp_rem_ref(struct ib_qp * qp)1120 void iwch_qp_rem_ref(struct ib_qp *qp)
1121 {
1122 PDBG("%s ib_qp %p\n", __func__, qp);
1123 if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1124 wake_up(&(to_iwch_qp(qp)->wait));
1125 }
1126
iwch_get_qp(struct ib_device * dev,int qpn)1127 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1128 {
1129 PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1130 return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1131 }
1132
1133
iwch_query_pkey(struct ib_device * ibdev,u8 port,u16 index,u16 * pkey)1134 static int iwch_query_pkey(struct ib_device *ibdev,
1135 u8 port, u16 index, u16 * pkey)
1136 {
1137 PDBG("%s ibdev %p\n", __func__, ibdev);
1138 *pkey = 0;
1139 return 0;
1140 }
1141
iwch_query_gid(struct ib_device * ibdev,u8 port,int index,union ib_gid * gid)1142 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1143 int index, union ib_gid *gid)
1144 {
1145 struct iwch_dev *dev;
1146
1147 PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1148 __func__, ibdev, port, index, gid);
1149 dev = to_iwch_dev(ibdev);
1150 BUG_ON(port == 0 || port > 2);
1151 memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1152 memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1153 return 0;
1154 }
1155
fw_vers_string_to_u64(struct iwch_dev * iwch_dev)1156 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1157 {
1158 struct ethtool_drvinfo info;
1159 struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1160 char *cp, *next;
1161 unsigned fw_maj, fw_min, fw_mic;
1162
1163 lldev->ethtool_ops->get_drvinfo(lldev, &info);
1164
1165 next = info.fw_version + 1;
1166 cp = strsep(&next, ".");
1167 sscanf(cp, "%i", &fw_maj);
1168 cp = strsep(&next, ".");
1169 sscanf(cp, "%i", &fw_min);
1170 cp = strsep(&next, ".");
1171 sscanf(cp, "%i", &fw_mic);
1172
1173 return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1174 (fw_mic & 0xffff);
1175 }
1176
iwch_query_device(struct ib_device * ibdev,struct ib_device_attr * props,struct ib_udata * uhw)1177 static int iwch_query_device(struct ib_device *ibdev, struct ib_device_attr *props,
1178 struct ib_udata *uhw)
1179 {
1180
1181 struct iwch_dev *dev;
1182
1183 PDBG("%s ibdev %p\n", __func__, ibdev);
1184
1185 if (uhw->inlen || uhw->outlen)
1186 return -EINVAL;
1187
1188 dev = to_iwch_dev(ibdev);
1189 memset(props, 0, sizeof *props);
1190 memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1191 props->hw_ver = dev->rdev.t3cdev_p->type;
1192 props->fw_ver = fw_vers_string_to_u64(dev);
1193 props->device_cap_flags = dev->device_cap_flags;
1194 props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1195 props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1196 props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1197 props->max_mr_size = dev->attr.max_mr_size;
1198 props->max_qp = dev->attr.max_qps;
1199 props->max_qp_wr = dev->attr.max_wrs;
1200 props->max_sge = dev->attr.max_sge_per_wr;
1201 props->max_sge_rd = 1;
1202 props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1203 props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1204 props->max_cq = dev->attr.max_cqs;
1205 props->max_cqe = dev->attr.max_cqes_per_cq;
1206 props->max_mr = dev->attr.max_mem_regs;
1207 props->max_pd = dev->attr.max_pds;
1208 props->local_ca_ack_delay = 0;
1209 props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1210
1211 return 0;
1212 }
1213
iwch_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props)1214 static int iwch_query_port(struct ib_device *ibdev,
1215 u8 port, struct ib_port_attr *props)
1216 {
1217 struct iwch_dev *dev;
1218 struct net_device *netdev;
1219 struct in_device *inetdev;
1220
1221 PDBG("%s ibdev %p\n", __func__, ibdev);
1222
1223 dev = to_iwch_dev(ibdev);
1224 netdev = dev->rdev.port_info.lldevs[port-1];
1225
1226 memset(props, 0, sizeof(struct ib_port_attr));
1227 props->max_mtu = IB_MTU_4096;
1228 if (netdev->mtu >= 4096)
1229 props->active_mtu = IB_MTU_4096;
1230 else if (netdev->mtu >= 2048)
1231 props->active_mtu = IB_MTU_2048;
1232 else if (netdev->mtu >= 1024)
1233 props->active_mtu = IB_MTU_1024;
1234 else if (netdev->mtu >= 512)
1235 props->active_mtu = IB_MTU_512;
1236 else
1237 props->active_mtu = IB_MTU_256;
1238
1239 if (!netif_carrier_ok(netdev))
1240 props->state = IB_PORT_DOWN;
1241 else {
1242 inetdev = in_dev_get(netdev);
1243 if (inetdev) {
1244 if (inetdev->ifa_list)
1245 props->state = IB_PORT_ACTIVE;
1246 else
1247 props->state = IB_PORT_INIT;
1248 in_dev_put(inetdev);
1249 } else
1250 props->state = IB_PORT_INIT;
1251 }
1252
1253 props->port_cap_flags =
1254 IB_PORT_CM_SUP |
1255 IB_PORT_SNMP_TUNNEL_SUP |
1256 IB_PORT_REINIT_SUP |
1257 IB_PORT_DEVICE_MGMT_SUP |
1258 IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1259 props->gid_tbl_len = 1;
1260 props->pkey_tbl_len = 1;
1261 props->active_width = 2;
1262 props->active_speed = IB_SPEED_DDR;
1263 props->max_msg_sz = -1;
1264
1265 return 0;
1266 }
1267
show_rev(struct device * dev,struct device_attribute * attr,char * buf)1268 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1269 char *buf)
1270 {
1271 struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1272 ibdev.dev);
1273 PDBG("%s dev 0x%p\n", __func__, dev);
1274 return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1275 }
1276
show_fw_ver(struct device * dev,struct device_attribute * attr,char * buf)1277 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1278 {
1279 struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1280 ibdev.dev);
1281 struct ethtool_drvinfo info;
1282 struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1283
1284 PDBG("%s dev 0x%p\n", __func__, dev);
1285 lldev->ethtool_ops->get_drvinfo(lldev, &info);
1286 return sprintf(buf, "%s\n", info.fw_version);
1287 }
1288
show_hca(struct device * dev,struct device_attribute * attr,char * buf)1289 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1290 char *buf)
1291 {
1292 struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1293 ibdev.dev);
1294 struct ethtool_drvinfo info;
1295 struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1296
1297 PDBG("%s dev 0x%p\n", __func__, dev);
1298 lldev->ethtool_ops->get_drvinfo(lldev, &info);
1299 return sprintf(buf, "%s\n", info.driver);
1300 }
1301
show_board(struct device * dev,struct device_attribute * attr,char * buf)1302 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1303 char *buf)
1304 {
1305 struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1306 ibdev.dev);
1307 PDBG("%s dev 0x%p\n", __func__, dev);
1308 return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1309 iwch_dev->rdev.rnic_info.pdev->device);
1310 }
1311
iwch_get_mib(struct ib_device * ibdev,union rdma_protocol_stats * stats)1312 static int iwch_get_mib(struct ib_device *ibdev,
1313 union rdma_protocol_stats *stats)
1314 {
1315 struct iwch_dev *dev;
1316 struct tp_mib_stats m;
1317 int ret;
1318
1319 PDBG("%s ibdev %p\n", __func__, ibdev);
1320 dev = to_iwch_dev(ibdev);
1321 ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1322 if (ret)
1323 return -ENOSYS;
1324
1325 memset(stats, 0, sizeof *stats);
1326 stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1327 m.ipInReceive_lo;
1328 stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1329 m.ipInHdrErrors_lo;
1330 stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1331 m.ipInAddrErrors_lo;
1332 stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1333 m.ipInUnknownProtos_lo;
1334 stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1335 m.ipInDiscards_lo;
1336 stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1337 m.ipInDelivers_lo;
1338 stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1339 m.ipOutRequests_lo;
1340 stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1341 m.ipOutDiscards_lo;
1342 stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1343 m.ipOutNoRoutes_lo;
1344 stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1345 stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1346 stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1347 stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1348 stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1349 stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1350 stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1351 stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1352 stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1353 stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1354 stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1355 m.tcpInSegs_lo;
1356 stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1357 m.tcpOutSegs_lo;
1358 stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1359 m.tcpRetransSeg_lo;
1360 stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1361 m.tcpInErrs_lo;
1362 stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1363 stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1364 return 0;
1365 }
1366
1367 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1368 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1369 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1370 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1371
1372 static struct device_attribute *iwch_class_attributes[] = {
1373 &dev_attr_hw_rev,
1374 &dev_attr_fw_ver,
1375 &dev_attr_hca_type,
1376 &dev_attr_board_id,
1377 };
1378
iwch_port_immutable(struct ib_device * ibdev,u8 port_num,struct ib_port_immutable * immutable)1379 static int iwch_port_immutable(struct ib_device *ibdev, u8 port_num,
1380 struct ib_port_immutable *immutable)
1381 {
1382 struct ib_port_attr attr;
1383 int err;
1384
1385 err = iwch_query_port(ibdev, port_num, &attr);
1386 if (err)
1387 return err;
1388
1389 immutable->pkey_tbl_len = attr.pkey_tbl_len;
1390 immutable->gid_tbl_len = attr.gid_tbl_len;
1391 immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
1392
1393 return 0;
1394 }
1395
iwch_register_device(struct iwch_dev * dev)1396 int iwch_register_device(struct iwch_dev *dev)
1397 {
1398 int ret;
1399 int i;
1400
1401 PDBG("%s iwch_dev %p\n", __func__, dev);
1402 strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1403 memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1404 memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1405 dev->ibdev.owner = THIS_MODULE;
1406 dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1407 IB_DEVICE_MEM_WINDOW |
1408 IB_DEVICE_MEM_MGT_EXTENSIONS;
1409
1410 /* cxgb3 supports STag 0. */
1411 dev->ibdev.local_dma_lkey = 0;
1412
1413 dev->ibdev.uverbs_cmd_mask =
1414 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1415 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1416 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1417 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1418 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1419 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1420 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1421 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1422 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1423 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1424 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1425 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1426 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1427 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1428 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1429 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1430 (1ull << IB_USER_VERBS_CMD_POST_RECV);
1431 dev->ibdev.node_type = RDMA_NODE_RNIC;
1432 memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1433 dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1434 dev->ibdev.num_comp_vectors = 1;
1435 dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1436 dev->ibdev.query_device = iwch_query_device;
1437 dev->ibdev.query_port = iwch_query_port;
1438 dev->ibdev.query_pkey = iwch_query_pkey;
1439 dev->ibdev.query_gid = iwch_query_gid;
1440 dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1441 dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1442 dev->ibdev.mmap = iwch_mmap;
1443 dev->ibdev.alloc_pd = iwch_allocate_pd;
1444 dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1445 dev->ibdev.create_ah = iwch_ah_create;
1446 dev->ibdev.destroy_ah = iwch_ah_destroy;
1447 dev->ibdev.create_qp = iwch_create_qp;
1448 dev->ibdev.modify_qp = iwch_ib_modify_qp;
1449 dev->ibdev.destroy_qp = iwch_destroy_qp;
1450 dev->ibdev.create_cq = iwch_create_cq;
1451 dev->ibdev.destroy_cq = iwch_destroy_cq;
1452 dev->ibdev.resize_cq = iwch_resize_cq;
1453 dev->ibdev.poll_cq = iwch_poll_cq;
1454 dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1455 dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1456 dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1457 dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1458 dev->ibdev.dereg_mr = iwch_dereg_mr;
1459 dev->ibdev.alloc_mw = iwch_alloc_mw;
1460 dev->ibdev.bind_mw = iwch_bind_mw;
1461 dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1462 dev->ibdev.alloc_mr = iwch_alloc_mr;
1463 dev->ibdev.map_mr_sg = iwch_map_mr_sg;
1464 dev->ibdev.attach_mcast = iwch_multicast_attach;
1465 dev->ibdev.detach_mcast = iwch_multicast_detach;
1466 dev->ibdev.process_mad = iwch_process_mad;
1467 dev->ibdev.req_notify_cq = iwch_arm_cq;
1468 dev->ibdev.post_send = iwch_post_send;
1469 dev->ibdev.post_recv = iwch_post_receive;
1470 dev->ibdev.get_protocol_stats = iwch_get_mib;
1471 dev->ibdev.uverbs_abi_ver = IWCH_UVERBS_ABI_VERSION;
1472 dev->ibdev.get_port_immutable = iwch_port_immutable;
1473
1474 dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1475 if (!dev->ibdev.iwcm)
1476 return -ENOMEM;
1477
1478 dev->ibdev.iwcm->connect = iwch_connect;
1479 dev->ibdev.iwcm->accept = iwch_accept_cr;
1480 dev->ibdev.iwcm->reject = iwch_reject_cr;
1481 dev->ibdev.iwcm->create_listen = iwch_create_listen;
1482 dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1483 dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1484 dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1485 dev->ibdev.iwcm->get_qp = iwch_get_qp;
1486
1487 ret = ib_register_device(&dev->ibdev, NULL);
1488 if (ret)
1489 goto bail1;
1490
1491 for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1492 ret = device_create_file(&dev->ibdev.dev,
1493 iwch_class_attributes[i]);
1494 if (ret) {
1495 goto bail2;
1496 }
1497 }
1498 return 0;
1499 bail2:
1500 ib_unregister_device(&dev->ibdev);
1501 bail1:
1502 kfree(dev->ibdev.iwcm);
1503 return ret;
1504 }
1505
iwch_unregister_device(struct iwch_dev * dev)1506 void iwch_unregister_device(struct iwch_dev *dev)
1507 {
1508 int i;
1509
1510 PDBG("%s iwch_dev %p\n", __func__, dev);
1511 for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1512 device_remove_file(&dev->ibdev.dev,
1513 iwch_class_attributes[i]);
1514 ib_unregister_device(&dev->ibdev);
1515 kfree(dev->ibdev.iwcm);
1516 return;
1517 }
1518