1/*
2 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses.  You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 *     Redistribution and use in source and binary forms, with or
12 *     without modification, are permitted provided that the following
13 *     conditions are met:
14 *
15 *      - Redistributions of source code must retain the above
16 *        copyright notice, this list of conditions and the following
17 *        disclaimer.
18 *
19 *      - Redistributions in binary form must reproduce the above
20 *        copyright notice, this list of conditions and the following
21 *        disclaimer in the documentation and/or other materials
22 *        provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34#include <linux/slab.h>
35
36#include "mlx4_ib.h"
37
38static u32 convert_access(int acc)
39{
40	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX4_PERM_ATOMIC       : 0) |
41	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX4_PERM_REMOTE_WRITE : 0) |
42	       (acc & IB_ACCESS_REMOTE_READ   ? MLX4_PERM_REMOTE_READ  : 0) |
43	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX4_PERM_LOCAL_WRITE  : 0) |
44	       (acc & IB_ACCESS_MW_BIND	      ? MLX4_PERM_BIND_MW      : 0) |
45	       MLX4_PERM_LOCAL_READ;
46}
47
48static enum mlx4_mw_type to_mlx4_type(enum ib_mw_type type)
49{
50	switch (type) {
51	case IB_MW_TYPE_1:	return MLX4_MW_TYPE_1;
52	case IB_MW_TYPE_2:	return MLX4_MW_TYPE_2;
53	default:		return -1;
54	}
55}
56
57struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc)
58{
59	struct mlx4_ib_mr *mr;
60	int err;
61
62	mr = kmalloc(sizeof *mr, GFP_KERNEL);
63	if (!mr)
64		return ERR_PTR(-ENOMEM);
65
66	err = mlx4_mr_alloc(to_mdev(pd->device)->dev, to_mpd(pd)->pdn, 0,
67			    ~0ull, convert_access(acc), 0, 0, &mr->mmr);
68	if (err)
69		goto err_free;
70
71	err = mlx4_mr_enable(to_mdev(pd->device)->dev, &mr->mmr);
72	if (err)
73		goto err_mr;
74
75	mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key;
76	mr->umem = NULL;
77
78	return &mr->ibmr;
79
80err_mr:
81	(void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
82
83err_free:
84	kfree(mr);
85
86	return ERR_PTR(err);
87}
88
89int mlx4_ib_umem_write_mtt(struct mlx4_ib_dev *dev, struct mlx4_mtt *mtt,
90			   struct ib_umem *umem)
91{
92	u64 *pages;
93	int i, k, entry;
94	int n;
95	int len;
96	int err = 0;
97	struct scatterlist *sg;
98
99	pages = (u64 *) __get_free_page(GFP_KERNEL);
100	if (!pages)
101		return -ENOMEM;
102
103	i = n = 0;
104
105	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
106		len = sg_dma_len(sg) >> mtt->page_shift;
107		for (k = 0; k < len; ++k) {
108			pages[i++] = sg_dma_address(sg) +
109				umem->page_size * k;
110			/*
111			 * Be friendly to mlx4_write_mtt() and
112			 * pass it chunks of appropriate size.
113			 */
114			if (i == PAGE_SIZE / sizeof (u64)) {
115				err = mlx4_write_mtt(dev->dev, mtt, n,
116						     i, pages);
117				if (err)
118					goto out;
119				n += i;
120				i = 0;
121			}
122		}
123	}
124
125	if (i)
126		err = mlx4_write_mtt(dev->dev, mtt, n, i, pages);
127
128out:
129	free_page((unsigned long) pages);
130	return err;
131}
132
133struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
134				  u64 virt_addr, int access_flags,
135				  struct ib_udata *udata)
136{
137	struct mlx4_ib_dev *dev = to_mdev(pd->device);
138	struct mlx4_ib_mr *mr;
139	int shift;
140	int err;
141	int n;
142
143	mr = kmalloc(sizeof *mr, GFP_KERNEL);
144	if (!mr)
145		return ERR_PTR(-ENOMEM);
146
147	/* Force registering the memory as writable. */
148	/* Used for memory re-registeration. HCA protects the access */
149	mr->umem = ib_umem_get(pd->uobject->context, start, length,
150			       access_flags | IB_ACCESS_LOCAL_WRITE, 0);
151	if (IS_ERR(mr->umem)) {
152		err = PTR_ERR(mr->umem);
153		goto err_free;
154	}
155
156	n = ib_umem_page_count(mr->umem);
157	shift = ilog2(mr->umem->page_size);
158
159	err = mlx4_mr_alloc(dev->dev, to_mpd(pd)->pdn, virt_addr, length,
160			    convert_access(access_flags), n, shift, &mr->mmr);
161	if (err)
162		goto err_umem;
163
164	err = mlx4_ib_umem_write_mtt(dev, &mr->mmr.mtt, mr->umem);
165	if (err)
166		goto err_mr;
167
168	err = mlx4_mr_enable(dev->dev, &mr->mmr);
169	if (err)
170		goto err_mr;
171
172	mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key;
173
174	return &mr->ibmr;
175
176err_mr:
177	(void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
178
179err_umem:
180	ib_umem_release(mr->umem);
181
182err_free:
183	kfree(mr);
184
185	return ERR_PTR(err);
186}
187
188int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags,
189			  u64 start, u64 length, u64 virt_addr,
190			  int mr_access_flags, struct ib_pd *pd,
191			  struct ib_udata *udata)
192{
193	struct mlx4_ib_dev *dev = to_mdev(mr->device);
194	struct mlx4_ib_mr *mmr = to_mmr(mr);
195	struct mlx4_mpt_entry *mpt_entry;
196	struct mlx4_mpt_entry **pmpt_entry = &mpt_entry;
197	int err;
198
199	/* Since we synchronize this call and mlx4_ib_dereg_mr via uverbs,
200	 * we assume that the calls can't run concurrently. Otherwise, a
201	 * race exists.
202	 */
203	err =  mlx4_mr_hw_get_mpt(dev->dev, &mmr->mmr, &pmpt_entry);
204
205	if (err)
206		return err;
207
208	if (flags & IB_MR_REREG_PD) {
209		err = mlx4_mr_hw_change_pd(dev->dev, *pmpt_entry,
210					   to_mpd(pd)->pdn);
211
212		if (err)
213			goto release_mpt_entry;
214	}
215
216	if (flags & IB_MR_REREG_ACCESS) {
217		err = mlx4_mr_hw_change_access(dev->dev, *pmpt_entry,
218					       convert_access(mr_access_flags));
219
220		if (err)
221			goto release_mpt_entry;
222	}
223
224	if (flags & IB_MR_REREG_TRANS) {
225		int shift;
226		int n;
227
228		mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr);
229		ib_umem_release(mmr->umem);
230		mmr->umem = ib_umem_get(mr->uobject->context, start, length,
231					mr_access_flags |
232					IB_ACCESS_LOCAL_WRITE,
233					0);
234		if (IS_ERR(mmr->umem)) {
235			err = PTR_ERR(mmr->umem);
236			/* Prevent mlx4_ib_dereg_mr from free'ing invalid pointer */
237			mmr->umem = NULL;
238			goto release_mpt_entry;
239		}
240		n = ib_umem_page_count(mmr->umem);
241		shift = ilog2(mmr->umem->page_size);
242
243		err = mlx4_mr_rereg_mem_write(dev->dev, &mmr->mmr,
244					      virt_addr, length, n, shift,
245					      *pmpt_entry);
246		if (err) {
247			ib_umem_release(mmr->umem);
248			goto release_mpt_entry;
249		}
250		mmr->mmr.iova       = virt_addr;
251		mmr->mmr.size       = length;
252
253		err = mlx4_ib_umem_write_mtt(dev, &mmr->mmr.mtt, mmr->umem);
254		if (err) {
255			mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr);
256			ib_umem_release(mmr->umem);
257			goto release_mpt_entry;
258		}
259	}
260
261	/* If we couldn't transfer the MR to the HCA, just remember to
262	 * return a failure. But dereg_mr will free the resources.
263	 */
264	err = mlx4_mr_hw_write_mpt(dev->dev, &mmr->mmr, pmpt_entry);
265	if (!err && flags & IB_MR_REREG_ACCESS)
266		mmr->mmr.access = mr_access_flags;
267
268release_mpt_entry:
269	mlx4_mr_hw_put_mpt(dev->dev, pmpt_entry);
270
271	return err;
272}
273
274int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
275{
276	struct mlx4_ib_mr *mr = to_mmr(ibmr);
277	int ret;
278
279	ret = mlx4_mr_free(to_mdev(ibmr->device)->dev, &mr->mmr);
280	if (ret)
281		return ret;
282	if (mr->umem)
283		ib_umem_release(mr->umem);
284	kfree(mr);
285
286	return 0;
287}
288
289struct ib_mw *mlx4_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
290{
291	struct mlx4_ib_dev *dev = to_mdev(pd->device);
292	struct mlx4_ib_mw *mw;
293	int err;
294
295	mw = kmalloc(sizeof(*mw), GFP_KERNEL);
296	if (!mw)
297		return ERR_PTR(-ENOMEM);
298
299	err = mlx4_mw_alloc(dev->dev, to_mpd(pd)->pdn,
300			    to_mlx4_type(type), &mw->mmw);
301	if (err)
302		goto err_free;
303
304	err = mlx4_mw_enable(dev->dev, &mw->mmw);
305	if (err)
306		goto err_mw;
307
308	mw->ibmw.rkey = mw->mmw.key;
309
310	return &mw->ibmw;
311
312err_mw:
313	mlx4_mw_free(dev->dev, &mw->mmw);
314
315err_free:
316	kfree(mw);
317
318	return ERR_PTR(err);
319}
320
321int mlx4_ib_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
322		    struct ib_mw_bind *mw_bind)
323{
324	struct ib_send_wr  wr;
325	struct ib_send_wr *bad_wr;
326	int ret;
327
328	memset(&wr, 0, sizeof(wr));
329	wr.opcode               = IB_WR_BIND_MW;
330	wr.wr_id                = mw_bind->wr_id;
331	wr.send_flags           = mw_bind->send_flags;
332	wr.wr.bind_mw.mw        = mw;
333	wr.wr.bind_mw.bind_info = mw_bind->bind_info;
334	wr.wr.bind_mw.rkey      = ib_inc_rkey(mw->rkey);
335
336	ret = mlx4_ib_post_send(qp, &wr, &bad_wr);
337	if (!ret)
338		mw->rkey = wr.wr.bind_mw.rkey;
339
340	return ret;
341}
342
343int mlx4_ib_dealloc_mw(struct ib_mw *ibmw)
344{
345	struct mlx4_ib_mw *mw = to_mmw(ibmw);
346
347	mlx4_mw_free(to_mdev(ibmw->device)->dev, &mw->mmw);
348	kfree(mw);
349
350	return 0;
351}
352
353struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd,
354					int max_page_list_len)
355{
356	struct mlx4_ib_dev *dev = to_mdev(pd->device);
357	struct mlx4_ib_mr *mr;
358	int err;
359
360	mr = kmalloc(sizeof *mr, GFP_KERNEL);
361	if (!mr)
362		return ERR_PTR(-ENOMEM);
363
364	err = mlx4_mr_alloc(dev->dev, to_mpd(pd)->pdn, 0, 0, 0,
365			    max_page_list_len, 0, &mr->mmr);
366	if (err)
367		goto err_free;
368
369	err = mlx4_mr_enable(dev->dev, &mr->mmr);
370	if (err)
371		goto err_mr;
372
373	mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key;
374	mr->umem = NULL;
375
376	return &mr->ibmr;
377
378err_mr:
379	(void) mlx4_mr_free(dev->dev, &mr->mmr);
380
381err_free:
382	kfree(mr);
383	return ERR_PTR(err);
384}
385
386struct ib_fast_reg_page_list *mlx4_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
387							       int page_list_len)
388{
389	struct mlx4_ib_dev *dev = to_mdev(ibdev);
390	struct mlx4_ib_fast_reg_page_list *mfrpl;
391	int size = page_list_len * sizeof (u64);
392
393	if (page_list_len > MLX4_MAX_FAST_REG_PAGES)
394		return ERR_PTR(-EINVAL);
395
396	mfrpl = kmalloc(sizeof *mfrpl, GFP_KERNEL);
397	if (!mfrpl)
398		return ERR_PTR(-ENOMEM);
399
400	mfrpl->ibfrpl.page_list = kmalloc(size, GFP_KERNEL);
401	if (!mfrpl->ibfrpl.page_list)
402		goto err_free;
403
404	mfrpl->mapped_page_list = dma_alloc_coherent(&dev->dev->persist->
405						     pdev->dev,
406						     size, &mfrpl->map,
407						     GFP_KERNEL);
408	if (!mfrpl->mapped_page_list)
409		goto err_free;
410
411	WARN_ON(mfrpl->map & 0x3f);
412
413	return &mfrpl->ibfrpl;
414
415err_free:
416	kfree(mfrpl->ibfrpl.page_list);
417	kfree(mfrpl);
418	return ERR_PTR(-ENOMEM);
419}
420
421void mlx4_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list)
422{
423	struct mlx4_ib_dev *dev = to_mdev(page_list->device);
424	struct mlx4_ib_fast_reg_page_list *mfrpl = to_mfrpl(page_list);
425	int size = page_list->max_page_list_len * sizeof (u64);
426
427	dma_free_coherent(&dev->dev->persist->pdev->dev, size,
428			  mfrpl->mapped_page_list,
429			  mfrpl->map);
430	kfree(mfrpl->ibfrpl.page_list);
431	kfree(mfrpl);
432}
433
434struct ib_fmr *mlx4_ib_fmr_alloc(struct ib_pd *pd, int acc,
435				 struct ib_fmr_attr *fmr_attr)
436{
437	struct mlx4_ib_dev *dev = to_mdev(pd->device);
438	struct mlx4_ib_fmr *fmr;
439	int err = -ENOMEM;
440
441	fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
442	if (!fmr)
443		return ERR_PTR(-ENOMEM);
444
445	err = mlx4_fmr_alloc(dev->dev, to_mpd(pd)->pdn, convert_access(acc),
446			     fmr_attr->max_pages, fmr_attr->max_maps,
447			     fmr_attr->page_shift, &fmr->mfmr);
448	if (err)
449		goto err_free;
450
451	err = mlx4_fmr_enable(to_mdev(pd->device)->dev, &fmr->mfmr);
452	if (err)
453		goto err_mr;
454
455	fmr->ibfmr.rkey = fmr->ibfmr.lkey = fmr->mfmr.mr.key;
456
457	return &fmr->ibfmr;
458
459err_mr:
460	(void) mlx4_mr_free(to_mdev(pd->device)->dev, &fmr->mfmr.mr);
461
462err_free:
463	kfree(fmr);
464
465	return ERR_PTR(err);
466}
467
468int mlx4_ib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
469		      int npages, u64 iova)
470{
471	struct mlx4_ib_fmr *ifmr = to_mfmr(ibfmr);
472	struct mlx4_ib_dev *dev = to_mdev(ifmr->ibfmr.device);
473
474	return mlx4_map_phys_fmr(dev->dev, &ifmr->mfmr, page_list, npages, iova,
475				 &ifmr->ibfmr.lkey, &ifmr->ibfmr.rkey);
476}
477
478int mlx4_ib_unmap_fmr(struct list_head *fmr_list)
479{
480	struct ib_fmr *ibfmr;
481	int err;
482	struct mlx4_dev *mdev = NULL;
483
484	list_for_each_entry(ibfmr, fmr_list, list) {
485		if (mdev && to_mdev(ibfmr->device)->dev != mdev)
486			return -EINVAL;
487		mdev = to_mdev(ibfmr->device)->dev;
488	}
489
490	if (!mdev)
491		return 0;
492
493	list_for_each_entry(ibfmr, fmr_list, list) {
494		struct mlx4_ib_fmr *ifmr = to_mfmr(ibfmr);
495
496		mlx4_fmr_unmap(mdev, &ifmr->mfmr, &ifmr->ibfmr.lkey, &ifmr->ibfmr.rkey);
497	}
498
499	/*
500	 * Make sure all MPT status updates are visible before issuing
501	 * SYNC_TPT firmware command.
502	 */
503	wmb();
504
505	err = mlx4_SYNC_TPT(mdev);
506	if (err)
507		pr_warn("SYNC_TPT error %d when "
508		       "unmapping FMRs\n", err);
509
510	return 0;
511}
512
513int mlx4_ib_fmr_dealloc(struct ib_fmr *ibfmr)
514{
515	struct mlx4_ib_fmr *ifmr = to_mfmr(ibfmr);
516	struct mlx4_ib_dev *dev = to_mdev(ibfmr->device);
517	int err;
518
519	err = mlx4_fmr_free(dev->dev, &ifmr->mfmr);
520
521	if (!err)
522		kfree(ifmr);
523
524	return err;
525}
526