[ARM] pxa: update defconfig for Verdex Pro
[linux-2.6/verdex.git] / drivers / net / mlx4 / mr.c
blobca7ab8e7b4cc3917cc94599315f6d16f8e930150
1 /*
2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
4 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
35 #include <linux/errno.h>
37 #include <linux/mlx4/cmd.h>
39 #include "mlx4.h"
40 #include "icm.h"
43 * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits.
45 struct mlx4_mpt_entry {
46 __be32 flags;
47 __be32 qpn;
48 __be32 key;
49 __be32 pd_flags;
50 __be64 start;
51 __be64 length;
52 __be32 lkey;
53 __be32 win_cnt;
54 u8 reserved1[3];
55 u8 mtt_rep;
56 __be64 mtt_seg;
57 __be32 mtt_sz;
58 __be32 entity_size;
59 __be32 first_byte_offset;
60 } __attribute__((packed));
62 #define MLX4_MPT_FLAG_SW_OWNS (0xfUL << 28)
63 #define MLX4_MPT_FLAG_FREE (0x3UL << 28)
64 #define MLX4_MPT_FLAG_MIO (1 << 17)
65 #define MLX4_MPT_FLAG_BIND_ENABLE (1 << 15)
66 #define MLX4_MPT_FLAG_PHYSICAL (1 << 9)
67 #define MLX4_MPT_FLAG_REGION (1 << 8)
69 #define MLX4_MPT_PD_FLAG_FAST_REG (1 << 27)
70 #define MLX4_MPT_PD_FLAG_RAE (1 << 28)
71 #define MLX4_MPT_PD_FLAG_EN_INV (3 << 24)
73 #define MLX4_MPT_STATUS_SW 0xF0
74 #define MLX4_MPT_STATUS_HW 0x00
76 static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order)
78 int o;
79 int m;
80 u32 seg;
82 spin_lock(&buddy->lock);
84 for (o = order; o <= buddy->max_order; ++o)
85 if (buddy->num_free[o]) {
86 m = 1 << (buddy->max_order - o);
87 seg = find_first_bit(buddy->bits[o], m);
88 if (seg < m)
89 goto found;
92 spin_unlock(&buddy->lock);
93 return -1;
95 found:
96 clear_bit(seg, buddy->bits[o]);
97 --buddy->num_free[o];
99 while (o > order) {
100 --o;
101 seg <<= 1;
102 set_bit(seg ^ 1, buddy->bits[o]);
103 ++buddy->num_free[o];
106 spin_unlock(&buddy->lock);
108 seg <<= order;
110 return seg;
113 static void mlx4_buddy_free(struct mlx4_buddy *buddy, u32 seg, int order)
115 seg >>= order;
117 spin_lock(&buddy->lock);
119 while (test_bit(seg ^ 1, buddy->bits[order])) {
120 clear_bit(seg ^ 1, buddy->bits[order]);
121 --buddy->num_free[order];
122 seg >>= 1;
123 ++order;
126 set_bit(seg, buddy->bits[order]);
127 ++buddy->num_free[order];
129 spin_unlock(&buddy->lock);
132 static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order)
134 int i, s;
136 buddy->max_order = max_order;
137 spin_lock_init(&buddy->lock);
139 buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *),
140 GFP_KERNEL);
141 buddy->num_free = kzalloc((buddy->max_order + 1) * sizeof (int *),
142 GFP_KERNEL);
143 if (!buddy->bits || !buddy->num_free)
144 goto err_out;
146 for (i = 0; i <= buddy->max_order; ++i) {
147 s = BITS_TO_LONGS(1 << (buddy->max_order - i));
148 buddy->bits[i] = kmalloc(s * sizeof (long), GFP_KERNEL);
149 if (!buddy->bits[i])
150 goto err_out_free;
151 bitmap_zero(buddy->bits[i], 1 << (buddy->max_order - i));
154 set_bit(0, buddy->bits[buddy->max_order]);
155 buddy->num_free[buddy->max_order] = 1;
157 return 0;
159 err_out_free:
160 for (i = 0; i <= buddy->max_order; ++i)
161 kfree(buddy->bits[i]);
163 err_out:
164 kfree(buddy->bits);
165 kfree(buddy->num_free);
167 return -ENOMEM;
170 static void mlx4_buddy_cleanup(struct mlx4_buddy *buddy)
172 int i;
174 for (i = 0; i <= buddy->max_order; ++i)
175 kfree(buddy->bits[i]);
177 kfree(buddy->bits);
178 kfree(buddy->num_free);
181 static u32 mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order)
183 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
184 u32 seg;
186 seg = mlx4_buddy_alloc(&mr_table->mtt_buddy, order);
187 if (seg == -1)
188 return -1;
190 if (mlx4_table_get_range(dev, &mr_table->mtt_table, seg,
191 seg + (1 << order) - 1)) {
192 mlx4_buddy_free(&mr_table->mtt_buddy, seg, order);
193 return -1;
196 return seg;
199 int mlx4_mtt_init(struct mlx4_dev *dev, int npages, int page_shift,
200 struct mlx4_mtt *mtt)
202 int i;
204 if (!npages) {
205 mtt->order = -1;
206 mtt->page_shift = MLX4_ICM_PAGE_SHIFT;
207 return 0;
208 } else
209 mtt->page_shift = page_shift;
211 for (mtt->order = 0, i = dev->caps.mtts_per_seg; i < npages; i <<= 1)
212 ++mtt->order;
214 mtt->first_seg = mlx4_alloc_mtt_range(dev, mtt->order);
215 if (mtt->first_seg == -1)
216 return -ENOMEM;
218 return 0;
220 EXPORT_SYMBOL_GPL(mlx4_mtt_init);
222 void mlx4_mtt_cleanup(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
224 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
226 if (mtt->order < 0)
227 return;
229 mlx4_buddy_free(&mr_table->mtt_buddy, mtt->first_seg, mtt->order);
230 mlx4_table_put_range(dev, &mr_table->mtt_table, mtt->first_seg,
231 mtt->first_seg + (1 << mtt->order) - 1);
233 EXPORT_SYMBOL_GPL(mlx4_mtt_cleanup);
235 u64 mlx4_mtt_addr(struct mlx4_dev *dev, struct mlx4_mtt *mtt)
237 return (u64) mtt->first_seg * dev->caps.mtt_entry_sz;
239 EXPORT_SYMBOL_GPL(mlx4_mtt_addr);
241 static u32 hw_index_to_key(u32 ind)
243 return (ind >> 24) | (ind << 8);
246 static u32 key_to_hw_index(u32 key)
248 return (key << 24) | (key >> 8);
251 static int mlx4_SW2HW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
252 int mpt_index)
254 return mlx4_cmd(dev, mailbox->dma, mpt_index, 0, MLX4_CMD_SW2HW_MPT,
255 MLX4_CMD_TIME_CLASS_B);
258 static int mlx4_HW2SW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
259 int mpt_index)
261 return mlx4_cmd_box(dev, 0, mailbox ? mailbox->dma : 0, mpt_index,
262 !mailbox, MLX4_CMD_HW2SW_MPT, MLX4_CMD_TIME_CLASS_B);
265 int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access,
266 int npages, int page_shift, struct mlx4_mr *mr)
268 struct mlx4_priv *priv = mlx4_priv(dev);
269 u32 index;
270 int err;
272 index = mlx4_bitmap_alloc(&priv->mr_table.mpt_bitmap);
273 if (index == -1)
274 return -ENOMEM;
276 mr->iova = iova;
277 mr->size = size;
278 mr->pd = pd;
279 mr->access = access;
280 mr->enabled = 0;
281 mr->key = hw_index_to_key(index);
283 err = mlx4_mtt_init(dev, npages, page_shift, &mr->mtt);
284 if (err)
285 mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, index);
287 return err;
289 EXPORT_SYMBOL_GPL(mlx4_mr_alloc);
291 void mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr)
293 struct mlx4_priv *priv = mlx4_priv(dev);
294 int err;
296 if (mr->enabled) {
297 err = mlx4_HW2SW_MPT(dev, NULL,
298 key_to_hw_index(mr->key) &
299 (dev->caps.num_mpts - 1));
300 if (err)
301 mlx4_warn(dev, "HW2SW_MPT failed (%d)\n", err);
304 mlx4_mtt_cleanup(dev, &mr->mtt);
305 mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, key_to_hw_index(mr->key));
307 EXPORT_SYMBOL_GPL(mlx4_mr_free);
309 int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr)
311 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
312 struct mlx4_cmd_mailbox *mailbox;
313 struct mlx4_mpt_entry *mpt_entry;
314 int err;
316 err = mlx4_table_get(dev, &mr_table->dmpt_table, key_to_hw_index(mr->key));
317 if (err)
318 return err;
320 mailbox = mlx4_alloc_cmd_mailbox(dev);
321 if (IS_ERR(mailbox)) {
322 err = PTR_ERR(mailbox);
323 goto err_table;
325 mpt_entry = mailbox->buf;
327 memset(mpt_entry, 0, sizeof *mpt_entry);
329 mpt_entry->flags = cpu_to_be32(MLX4_MPT_FLAG_MIO |
330 MLX4_MPT_FLAG_REGION |
331 mr->access);
333 mpt_entry->key = cpu_to_be32(key_to_hw_index(mr->key));
334 mpt_entry->pd_flags = cpu_to_be32(mr->pd | MLX4_MPT_PD_FLAG_EN_INV);
335 mpt_entry->start = cpu_to_be64(mr->iova);
336 mpt_entry->length = cpu_to_be64(mr->size);
337 mpt_entry->entity_size = cpu_to_be32(mr->mtt.page_shift);
339 if (mr->mtt.order < 0) {
340 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL);
341 mpt_entry->mtt_seg = 0;
342 } else {
343 mpt_entry->mtt_seg = cpu_to_be64(mlx4_mtt_addr(dev, &mr->mtt));
346 if (mr->mtt.order >= 0 && mr->mtt.page_shift == 0) {
347 /* fast register MR in free state */
348 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_FREE);
349 mpt_entry->pd_flags |= cpu_to_be32(MLX4_MPT_PD_FLAG_FAST_REG |
350 MLX4_MPT_PD_FLAG_RAE);
351 mpt_entry->mtt_sz = cpu_to_be32((1 << mr->mtt.order) *
352 dev->caps.mtts_per_seg);
353 } else {
354 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_SW_OWNS);
357 err = mlx4_SW2HW_MPT(dev, mailbox,
358 key_to_hw_index(mr->key) & (dev->caps.num_mpts - 1));
359 if (err) {
360 mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err);
361 goto err_cmd;
364 mr->enabled = 1;
366 mlx4_free_cmd_mailbox(dev, mailbox);
368 return 0;
370 err_cmd:
371 mlx4_free_cmd_mailbox(dev, mailbox);
373 err_table:
374 mlx4_table_put(dev, &mr_table->dmpt_table, key_to_hw_index(mr->key));
375 return err;
377 EXPORT_SYMBOL_GPL(mlx4_mr_enable);
379 static int mlx4_write_mtt_chunk(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
380 int start_index, int npages, u64 *page_list)
382 struct mlx4_priv *priv = mlx4_priv(dev);
383 __be64 *mtts;
384 dma_addr_t dma_handle;
385 int i;
386 int s = start_index * sizeof (u64);
388 /* All MTTs must fit in the same page */
389 if (start_index / (PAGE_SIZE / sizeof (u64)) !=
390 (start_index + npages - 1) / (PAGE_SIZE / sizeof (u64)))
391 return -EINVAL;
393 if (start_index & (dev->caps.mtts_per_seg - 1))
394 return -EINVAL;
396 mtts = mlx4_table_find(&priv->mr_table.mtt_table, mtt->first_seg +
397 s / dev->caps.mtt_entry_sz, &dma_handle);
398 if (!mtts)
399 return -ENOMEM;
401 dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle,
402 npages * sizeof (u64), DMA_TO_DEVICE);
404 for (i = 0; i < npages; ++i)
405 mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
407 dma_sync_single_for_device(&dev->pdev->dev, dma_handle,
408 npages * sizeof (u64), DMA_TO_DEVICE);
410 return 0;
413 int mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
414 int start_index, int npages, u64 *page_list)
416 int chunk;
417 int err;
419 if (mtt->order < 0)
420 return -EINVAL;
422 while (npages > 0) {
423 chunk = min_t(int, PAGE_SIZE / sizeof(u64), npages);
424 err = mlx4_write_mtt_chunk(dev, mtt, start_index, chunk, page_list);
425 if (err)
426 return err;
428 npages -= chunk;
429 start_index += chunk;
430 page_list += chunk;
433 return 0;
435 EXPORT_SYMBOL_GPL(mlx4_write_mtt);
437 int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
438 struct mlx4_buf *buf)
440 u64 *page_list;
441 int err;
442 int i;
444 page_list = kmalloc(buf->npages * sizeof *page_list, GFP_KERNEL);
445 if (!page_list)
446 return -ENOMEM;
448 for (i = 0; i < buf->npages; ++i)
449 if (buf->nbufs == 1)
450 page_list[i] = buf->direct.map + (i << buf->page_shift);
451 else
452 page_list[i] = buf->page_list[i].map;
454 err = mlx4_write_mtt(dev, mtt, 0, buf->npages, page_list);
456 kfree(page_list);
457 return err;
459 EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt);
461 int mlx4_init_mr_table(struct mlx4_dev *dev)
463 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
464 int err;
466 err = mlx4_bitmap_init(&mr_table->mpt_bitmap, dev->caps.num_mpts,
467 ~0, dev->caps.reserved_mrws, 0);
468 if (err)
469 return err;
471 err = mlx4_buddy_init(&mr_table->mtt_buddy,
472 ilog2(dev->caps.num_mtt_segs));
473 if (err)
474 goto err_buddy;
476 if (dev->caps.reserved_mtts) {
477 if (mlx4_alloc_mtt_range(dev, fls(dev->caps.reserved_mtts - 1)) == -1) {
478 mlx4_warn(dev, "MTT table of order %d is too small.\n",
479 mr_table->mtt_buddy.max_order);
480 err = -ENOMEM;
481 goto err_reserve_mtts;
485 return 0;
487 err_reserve_mtts:
488 mlx4_buddy_cleanup(&mr_table->mtt_buddy);
490 err_buddy:
491 mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
493 return err;
496 void mlx4_cleanup_mr_table(struct mlx4_dev *dev)
498 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
500 mlx4_buddy_cleanup(&mr_table->mtt_buddy);
501 mlx4_bitmap_cleanup(&mr_table->mpt_bitmap);
504 static inline int mlx4_check_fmr(struct mlx4_fmr *fmr, u64 *page_list,
505 int npages, u64 iova)
507 int i, page_mask;
509 if (npages > fmr->max_pages)
510 return -EINVAL;
512 page_mask = (1 << fmr->page_shift) - 1;
514 /* We are getting page lists, so va must be page aligned. */
515 if (iova & page_mask)
516 return -EINVAL;
518 /* Trust the user not to pass misaligned data in page_list */
519 if (0)
520 for (i = 0; i < npages; ++i) {
521 if (page_list[i] & ~page_mask)
522 return -EINVAL;
525 if (fmr->maps >= fmr->max_maps)
526 return -EINVAL;
528 return 0;
531 int mlx4_map_phys_fmr(struct mlx4_dev *dev, struct mlx4_fmr *fmr, u64 *page_list,
532 int npages, u64 iova, u32 *lkey, u32 *rkey)
534 u32 key;
535 int i, err;
537 err = mlx4_check_fmr(fmr, page_list, npages, iova);
538 if (err)
539 return err;
541 ++fmr->maps;
543 key = key_to_hw_index(fmr->mr.key);
544 key += dev->caps.num_mpts;
545 *lkey = *rkey = fmr->mr.key = hw_index_to_key(key);
547 *(u8 *) fmr->mpt = MLX4_MPT_STATUS_SW;
549 /* Make sure MPT status is visible before writing MTT entries */
550 wmb();
552 dma_sync_single_for_cpu(&dev->pdev->dev, fmr->dma_handle,
553 npages * sizeof(u64), DMA_TO_DEVICE);
555 for (i = 0; i < npages; ++i)
556 fmr->mtts[i] = cpu_to_be64(page_list[i] | MLX4_MTT_FLAG_PRESENT);
558 dma_sync_single_for_device(&dev->pdev->dev, fmr->dma_handle,
559 npages * sizeof(u64), DMA_TO_DEVICE);
561 fmr->mpt->key = cpu_to_be32(key);
562 fmr->mpt->lkey = cpu_to_be32(key);
563 fmr->mpt->length = cpu_to_be64(npages * (1ull << fmr->page_shift));
564 fmr->mpt->start = cpu_to_be64(iova);
566 /* Make MTT entries are visible before setting MPT status */
567 wmb();
569 *(u8 *) fmr->mpt = MLX4_MPT_STATUS_HW;
571 /* Make sure MPT status is visible before consumer can use FMR */
572 wmb();
574 return 0;
576 EXPORT_SYMBOL_GPL(mlx4_map_phys_fmr);
578 int mlx4_fmr_alloc(struct mlx4_dev *dev, u32 pd, u32 access, int max_pages,
579 int max_maps, u8 page_shift, struct mlx4_fmr *fmr)
581 struct mlx4_priv *priv = mlx4_priv(dev);
582 u64 mtt_seg;
583 int err = -ENOMEM;
585 if (page_shift < (ffs(dev->caps.page_size_cap) - 1) || page_shift >= 32)
586 return -EINVAL;
588 /* All MTTs must fit in the same page */
589 if (max_pages * sizeof *fmr->mtts > PAGE_SIZE)
590 return -EINVAL;
592 fmr->page_shift = page_shift;
593 fmr->max_pages = max_pages;
594 fmr->max_maps = max_maps;
595 fmr->maps = 0;
597 err = mlx4_mr_alloc(dev, pd, 0, 0, access, max_pages,
598 page_shift, &fmr->mr);
599 if (err)
600 return err;
602 mtt_seg = fmr->mr.mtt.first_seg * dev->caps.mtt_entry_sz;
604 fmr->mtts = mlx4_table_find(&priv->mr_table.mtt_table,
605 fmr->mr.mtt.first_seg,
606 &fmr->dma_handle);
607 if (!fmr->mtts) {
608 err = -ENOMEM;
609 goto err_free;
612 return 0;
614 err_free:
615 mlx4_mr_free(dev, &fmr->mr);
616 return err;
618 EXPORT_SYMBOL_GPL(mlx4_fmr_alloc);
620 int mlx4_fmr_enable(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
622 struct mlx4_priv *priv = mlx4_priv(dev);
623 int err;
625 err = mlx4_mr_enable(dev, &fmr->mr);
626 if (err)
627 return err;
629 fmr->mpt = mlx4_table_find(&priv->mr_table.dmpt_table,
630 key_to_hw_index(fmr->mr.key), NULL);
631 if (!fmr->mpt)
632 return -ENOMEM;
634 return 0;
636 EXPORT_SYMBOL_GPL(mlx4_fmr_enable);
638 void mlx4_fmr_unmap(struct mlx4_dev *dev, struct mlx4_fmr *fmr,
639 u32 *lkey, u32 *rkey)
641 if (!fmr->maps)
642 return;
644 fmr->maps = 0;
646 *(u8 *) fmr->mpt = MLX4_MPT_STATUS_SW;
648 EXPORT_SYMBOL_GPL(mlx4_fmr_unmap);
650 int mlx4_fmr_free(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
652 if (fmr->maps)
653 return -EBUSY;
655 fmr->mr.enabled = 0;
656 mlx4_mr_free(dev, &fmr->mr);
658 return 0;
660 EXPORT_SYMBOL_GPL(mlx4_fmr_free);
662 int mlx4_SYNC_TPT(struct mlx4_dev *dev)
664 return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_SYNC_TPT, 1000);
666 EXPORT_SYMBOL_GPL(mlx4_SYNC_TPT);