2 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/slab.h>
38 static u32 convert_access(int acc)
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) |
48 static enum mlx4_mw_type to_mlx4_type(enum ib_mw_type type)
51 case IB_MW_TYPE_1: return MLX4_MW_TYPE_1;
52 case IB_MW_TYPE_2: return MLX4_MW_TYPE_2;
57 struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc)
59 struct mlx4_ib_mr *mr;
62 mr = kzalloc(sizeof(*mr), GFP_KERNEL);
64 return ERR_PTR(-ENOMEM);
66 err = mlx4_mr_alloc(to_mdev(pd->device)->dev, to_mpd(pd)->pdn, 0,
67 ~0ull, convert_access(acc), 0, 0, &mr->mmr);
71 err = mlx4_mr_enable(to_mdev(pd->device)->dev, &mr->mmr);
75 mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key;
81 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
89 int mlx4_ib_umem_write_mtt(struct mlx4_ib_dev *dev, struct mlx4_mtt *mtt,
97 struct scatterlist *sg;
99 pages = (u64 *) __get_free_page(GFP_KERNEL);
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) +
111 * Be friendly to mlx4_write_mtt() and
112 * pass it chunks of appropriate size.
114 if (i == PAGE_SIZE / sizeof (u64)) {
115 err = mlx4_write_mtt(dev->dev, mtt, n,
126 err = mlx4_write_mtt(dev->dev, mtt, n, i, pages);
129 free_page((unsigned long) pages);
133 struct 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)
137 struct mlx4_ib_dev *dev = to_mdev(pd->device);
138 struct mlx4_ib_mr *mr;
143 mr = kzalloc(sizeof(*mr), GFP_KERNEL);
145 return ERR_PTR(-ENOMEM);
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);
156 n = ib_umem_page_count(mr->umem);
157 shift = ilog2(mr->umem->page_size);
159 err = mlx4_mr_alloc(dev->dev, to_mpd(pd)->pdn, virt_addr, length,
160 convert_access(access_flags), n, shift, &mr->mmr);
164 err = mlx4_ib_umem_write_mtt(dev, &mr->mmr.mtt, mr->umem);
168 err = mlx4_mr_enable(dev->dev, &mr->mmr);
172 mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key;
177 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
180 ib_umem_release(mr->umem);
188 int 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)
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;
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
203 err = mlx4_mr_hw_get_mpt(dev->dev, &mmr->mmr, &pmpt_entry);
208 if (flags & IB_MR_REREG_PD) {
209 err = mlx4_mr_hw_change_pd(dev->dev, *pmpt_entry,
213 goto release_mpt_entry;
216 if (flags & IB_MR_REREG_ACCESS) {
217 err = mlx4_mr_hw_change_access(dev->dev, *pmpt_entry,
218 convert_access(mr_access_flags));
221 goto release_mpt_entry;
224 if (flags & IB_MR_REREG_TRANS) {
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,
232 IB_ACCESS_LOCAL_WRITE,
234 if (IS_ERR(mmr->umem)) {
235 err = PTR_ERR(mmr->umem);
236 /* Prevent mlx4_ib_dereg_mr from free'ing invalid pointer */
238 goto release_mpt_entry;
240 n = ib_umem_page_count(mmr->umem);
241 shift = ilog2(mmr->umem->page_size);
243 err = mlx4_mr_rereg_mem_write(dev->dev, &mmr->mmr,
244 virt_addr, length, n, shift,
247 ib_umem_release(mmr->umem);
248 goto release_mpt_entry;
250 mmr->mmr.iova = virt_addr;
251 mmr->mmr.size = length;
253 err = mlx4_ib_umem_write_mtt(dev, &mmr->mmr.mtt, mmr->umem);
255 mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr);
256 ib_umem_release(mmr->umem);
257 goto release_mpt_entry;
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.
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;
269 mlx4_mr_hw_put_mpt(dev->dev, pmpt_entry);
275 mlx4_alloc_priv_pages(struct ib_device *device,
276 struct mlx4_ib_mr *mr,
279 int size = max_pages * sizeof(u64);
283 add_size = max_t(int, MLX4_MR_PAGES_ALIGN - ARCH_KMALLOC_MINALIGN, 0);
285 mr->pages_alloc = kzalloc(size + add_size, GFP_KERNEL);
286 if (!mr->pages_alloc)
289 mr->pages = PTR_ALIGN(mr->pages_alloc, MLX4_MR_PAGES_ALIGN);
291 mr->page_map = dma_map_single(device->dma_device, mr->pages,
292 size, DMA_TO_DEVICE);
294 if (dma_mapping_error(device->dma_device, mr->page_map)) {
301 kfree(mr->pages_alloc);
307 mlx4_free_priv_pages(struct mlx4_ib_mr *mr)
310 struct ib_device *device = mr->ibmr.device;
311 int size = mr->max_pages * sizeof(u64);
313 dma_unmap_single(device->dma_device, mr->page_map,
314 size, DMA_TO_DEVICE);
315 kfree(mr->pages_alloc);
320 int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
322 struct mlx4_ib_mr *mr = to_mmr(ibmr);
325 mlx4_free_priv_pages(mr);
327 ret = mlx4_mr_free(to_mdev(ibmr->device)->dev, &mr->mmr);
331 ib_umem_release(mr->umem);
337 struct ib_mw *mlx4_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
339 struct mlx4_ib_dev *dev = to_mdev(pd->device);
340 struct mlx4_ib_mw *mw;
343 mw = kmalloc(sizeof(*mw), GFP_KERNEL);
345 return ERR_PTR(-ENOMEM);
347 err = mlx4_mw_alloc(dev->dev, to_mpd(pd)->pdn,
348 to_mlx4_type(type), &mw->mmw);
352 err = mlx4_mw_enable(dev->dev, &mw->mmw);
356 mw->ibmw.rkey = mw->mmw.key;
361 mlx4_mw_free(dev->dev, &mw->mmw);
369 int mlx4_ib_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
370 struct ib_mw_bind *mw_bind)
372 struct ib_bind_mw_wr wr;
373 struct ib_send_wr *bad_wr;
376 memset(&wr, 0, sizeof(wr));
377 wr.wr.opcode = IB_WR_BIND_MW;
378 wr.wr.wr_id = mw_bind->wr_id;
379 wr.wr.send_flags = mw_bind->send_flags;
381 wr.bind_info = mw_bind->bind_info;
382 wr.rkey = ib_inc_rkey(mw->rkey);
384 ret = mlx4_ib_post_send(qp, &wr.wr, &bad_wr);
391 int mlx4_ib_dealloc_mw(struct ib_mw *ibmw)
393 struct mlx4_ib_mw *mw = to_mmw(ibmw);
395 mlx4_mw_free(to_mdev(ibmw->device)->dev, &mw->mmw);
401 struct ib_mr *mlx4_ib_alloc_mr(struct ib_pd *pd,
402 enum ib_mr_type mr_type,
405 struct mlx4_ib_dev *dev = to_mdev(pd->device);
406 struct mlx4_ib_mr *mr;
409 if (mr_type != IB_MR_TYPE_MEM_REG ||
410 max_num_sg > MLX4_MAX_FAST_REG_PAGES)
411 return ERR_PTR(-EINVAL);
413 mr = kzalloc(sizeof(*mr), GFP_KERNEL);
415 return ERR_PTR(-ENOMEM);
417 err = mlx4_mr_alloc(dev->dev, to_mpd(pd)->pdn, 0, 0, 0,
418 max_num_sg, 0, &mr->mmr);
422 err = mlx4_alloc_priv_pages(pd->device, mr, max_num_sg);
426 mr->max_pages = max_num_sg;
428 err = mlx4_mr_enable(dev->dev, &mr->mmr);
432 mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key;
438 mlx4_free_priv_pages(mr);
440 (void) mlx4_mr_free(dev->dev, &mr->mmr);
446 struct ib_fmr *mlx4_ib_fmr_alloc(struct ib_pd *pd, int acc,
447 struct ib_fmr_attr *fmr_attr)
449 struct mlx4_ib_dev *dev = to_mdev(pd->device);
450 struct mlx4_ib_fmr *fmr;
453 fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
455 return ERR_PTR(-ENOMEM);
457 err = mlx4_fmr_alloc(dev->dev, to_mpd(pd)->pdn, convert_access(acc),
458 fmr_attr->max_pages, fmr_attr->max_maps,
459 fmr_attr->page_shift, &fmr->mfmr);
463 err = mlx4_fmr_enable(to_mdev(pd->device)->dev, &fmr->mfmr);
467 fmr->ibfmr.rkey = fmr->ibfmr.lkey = fmr->mfmr.mr.key;
472 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &fmr->mfmr.mr);
480 int mlx4_ib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
481 int npages, u64 iova)
483 struct mlx4_ib_fmr *ifmr = to_mfmr(ibfmr);
484 struct mlx4_ib_dev *dev = to_mdev(ifmr->ibfmr.device);
486 return mlx4_map_phys_fmr(dev->dev, &ifmr->mfmr, page_list, npages, iova,
487 &ifmr->ibfmr.lkey, &ifmr->ibfmr.rkey);
490 int mlx4_ib_unmap_fmr(struct list_head *fmr_list)
492 struct ib_fmr *ibfmr;
494 struct mlx4_dev *mdev = NULL;
496 list_for_each_entry(ibfmr, fmr_list, list) {
497 if (mdev && to_mdev(ibfmr->device)->dev != mdev)
499 mdev = to_mdev(ibfmr->device)->dev;
505 list_for_each_entry(ibfmr, fmr_list, list) {
506 struct mlx4_ib_fmr *ifmr = to_mfmr(ibfmr);
508 mlx4_fmr_unmap(mdev, &ifmr->mfmr, &ifmr->ibfmr.lkey, &ifmr->ibfmr.rkey);
512 * Make sure all MPT status updates are visible before issuing
513 * SYNC_TPT firmware command.
517 err = mlx4_SYNC_TPT(mdev);
519 pr_warn("SYNC_TPT error %d when "
520 "unmapping FMRs\n", err);
525 int mlx4_ib_fmr_dealloc(struct ib_fmr *ibfmr)
527 struct mlx4_ib_fmr *ifmr = to_mfmr(ibfmr);
528 struct mlx4_ib_dev *dev = to_mdev(ibfmr->device);
531 err = mlx4_fmr_free(dev->dev, &ifmr->mfmr);
539 static int mlx4_set_page(struct ib_mr *ibmr, u64 addr)
541 struct mlx4_ib_mr *mr = to_mmr(ibmr);
543 if (unlikely(mr->npages == mr->max_pages))
546 mr->pages[mr->npages++] = cpu_to_be64(addr | MLX4_MTT_FLAG_PRESENT);
551 int mlx4_ib_map_mr_sg(struct ib_mr *ibmr,
552 struct scatterlist *sg,
555 struct mlx4_ib_mr *mr = to_mmr(ibmr);
560 ib_dma_sync_single_for_cpu(ibmr->device, mr->page_map,
561 sizeof(u64) * mr->max_pages,
564 rc = ib_sg_to_pages(ibmr, sg, sg_nents, mlx4_set_page);
566 ib_dma_sync_single_for_device(ibmr->device, mr->page_map,
567 sizeof(u64) * mr->max_pages,