2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
35 #include "mthca_memfree.h"
36 #include "mthca_dev.h"
37 #include "mthca_cmd.h"
40 * We allocate in as big chunks as we can, up to a maximum of 256 KB
44 MTHCA_ICM_ALLOC_SIZE = 1 << 18,
45 MTHCA_TABLE_CHUNK_SIZE = 1 << 18
48 void mthca_free_icm(struct mthca_dev *dev, struct mthca_icm *icm)
50 struct mthca_icm_chunk *chunk, *tmp;
56 list_for_each_entry_safe(chunk, tmp, &icm->chunk_list, list) {
58 pci_unmap_sg(dev->pdev, chunk->mem, chunk->npages,
59 PCI_DMA_BIDIRECTIONAL);
61 for (i = 0; i < chunk->npages; ++i)
62 __free_pages(chunk->mem[i].page,
63 get_order(chunk->mem[i].length));
71 struct mthca_icm *mthca_alloc_icm(struct mthca_dev *dev, int npages,
72 unsigned int gfp_mask)
74 struct mthca_icm *icm;
75 struct mthca_icm_chunk *chunk = NULL;
78 icm = kmalloc(sizeof *icm, gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN));
83 INIT_LIST_HEAD(&icm->chunk_list);
85 cur_order = get_order(MTHCA_ICM_ALLOC_SIZE);
89 chunk = kmalloc(sizeof *chunk,
90 gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN));
96 list_add_tail(&chunk->list, &icm->chunk_list);
99 while (1 << cur_order > npages)
102 chunk->mem[chunk->npages].page = alloc_pages(gfp_mask, cur_order);
103 if (chunk->mem[chunk->npages].page) {
104 chunk->mem[chunk->npages].length = PAGE_SIZE << cur_order;
105 chunk->mem[chunk->npages].offset = 0;
107 if (++chunk->npages == MTHCA_ICM_CHUNK_LEN) {
108 chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
110 PCI_DMA_BIDIRECTIONAL);
118 npages -= 1 << cur_order;
127 chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
129 PCI_DMA_BIDIRECTIONAL);
138 mthca_free_icm(dev, icm);
142 int mthca_table_get(struct mthca_dev *dev, struct mthca_icm_table *table, int obj)
144 int i = (obj & (table->num_obj - 1)) * table->obj_size / MTHCA_TABLE_CHUNK_SIZE;
151 ++table->icm[i]->refcount;
155 table->icm[i] = mthca_alloc_icm(dev, MTHCA_TABLE_CHUNK_SIZE >> PAGE_SHIFT,
156 (table->lowmem ? GFP_KERNEL : GFP_HIGHUSER) |
158 if (!table->icm[i]) {
163 if (mthca_MAP_ICM(dev, table->icm[i], table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
164 &status) || status) {
165 mthca_free_icm(dev, table->icm[i]);
166 table->icm[i] = NULL;
171 ++table->icm[i]->refcount;
178 void mthca_table_put(struct mthca_dev *dev, struct mthca_icm_table *table, int obj)
180 int i = (obj & (table->num_obj - 1)) * table->obj_size / MTHCA_TABLE_CHUNK_SIZE;
185 if (--table->icm[i]->refcount == 0) {
186 mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
187 MTHCA_TABLE_CHUNK_SIZE >> 12, &status);
188 mthca_free_icm(dev, table->icm[i]);
189 table->icm[i] = NULL;
195 void *mthca_table_find(struct mthca_icm_table *table, int obj)
198 struct mthca_icm_chunk *chunk;
199 struct mthca_icm *icm;
200 struct page *page = NULL;
207 idx = (obj & (table->num_obj - 1)) * table->obj_size;
208 icm = table->icm[idx / MTHCA_TABLE_CHUNK_SIZE];
209 offset = idx % MTHCA_TABLE_CHUNK_SIZE;
214 list_for_each_entry(chunk, &icm->chunk_list, list) {
215 for (i = 0; i < chunk->npages; ++i) {
216 if (chunk->mem[i].length >= offset) {
217 page = chunk->mem[i].page;
220 offset -= chunk->mem[i].length;
226 return page ? lowmem_page_address(page) + offset : NULL;
229 int mthca_table_get_range(struct mthca_dev *dev, struct mthca_icm_table *table,
232 int inc = MTHCA_TABLE_CHUNK_SIZE / table->obj_size;
235 for (i = start; i <= end; i += inc) {
236 err = mthca_table_get(dev, table, i);
246 mthca_table_put(dev, table, i);
252 void mthca_table_put_range(struct mthca_dev *dev, struct mthca_icm_table *table,
257 for (i = start; i <= end; i += MTHCA_TABLE_CHUNK_SIZE / table->obj_size)
258 mthca_table_put(dev, table, i);
261 struct mthca_icm_table *mthca_alloc_icm_table(struct mthca_dev *dev,
262 u64 virt, int obj_size,
263 int nobj, int reserved,
266 struct mthca_icm_table *table;
271 num_icm = obj_size * nobj / MTHCA_TABLE_CHUNK_SIZE;
273 table = kmalloc(sizeof *table + num_icm * sizeof *table->icm, GFP_KERNEL);
278 table->num_icm = num_icm;
279 table->num_obj = nobj;
280 table->obj_size = obj_size;
281 table->lowmem = use_lowmem;
282 init_MUTEX(&table->mutex);
284 for (i = 0; i < num_icm; ++i)
285 table->icm[i] = NULL;
287 for (i = 0; i * MTHCA_TABLE_CHUNK_SIZE < reserved * obj_size; ++i) {
288 table->icm[i] = mthca_alloc_icm(dev, MTHCA_TABLE_CHUNK_SIZE >> PAGE_SHIFT,
289 (use_lowmem ? GFP_KERNEL : GFP_HIGHUSER) |
293 if (mthca_MAP_ICM(dev, table->icm[i], virt + i * MTHCA_TABLE_CHUNK_SIZE,
294 &status) || status) {
295 mthca_free_icm(dev, table->icm[i]);
296 table->icm[i] = NULL;
301 * Add a reference to this ICM chunk so that it never
302 * gets freed (since it contains reserved firmware objects).
304 ++table->icm[i]->refcount;
310 for (i = 0; i < num_icm; ++i)
312 mthca_UNMAP_ICM(dev, virt + i * MTHCA_TABLE_CHUNK_SIZE,
313 MTHCA_TABLE_CHUNK_SIZE >> 12, &status);
314 mthca_free_icm(dev, table->icm[i]);
322 void mthca_free_icm_table(struct mthca_dev *dev, struct mthca_icm_table *table)
327 for (i = 0; i < table->num_icm; ++i)
329 mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
330 MTHCA_TABLE_CHUNK_SIZE >> 12, &status);
331 mthca_free_icm(dev, table->icm[i]);
337 static u64 mthca_uarc_virt(struct mthca_dev *dev, int page)
339 return dev->uar_table.uarc_base +
340 dev->driver_uar.index * dev->uar_table.uarc_size +
344 int mthca_alloc_db(struct mthca_dev *dev, int type, u32 qn, u32 **db)
349 struct mthca_db_page *page;
353 down(&dev->db_tab->mutex);
356 case MTHCA_DB_TYPE_CQ_ARM:
357 case MTHCA_DB_TYPE_SQ:
360 end = dev->db_tab->max_group1;
364 case MTHCA_DB_TYPE_CQ_SET_CI:
365 case MTHCA_DB_TYPE_RQ:
366 case MTHCA_DB_TYPE_SRQ:
368 start = dev->db_tab->npages - 1;
369 end = dev->db_tab->min_group2;
378 for (i = start; i != end; i += dir)
379 if (dev->db_tab->page[i].db_rec &&
380 !bitmap_full(dev->db_tab->page[i].used,
381 MTHCA_DB_REC_PER_PAGE)) {
382 page = dev->db_tab->page + i;
386 if (dev->db_tab->max_group1 >= dev->db_tab->min_group2 - 1) {
391 page = dev->db_tab->page + end;
392 page->db_rec = dma_alloc_coherent(&dev->pdev->dev, 4096,
393 &page->mapping, GFP_KERNEL);
398 memset(page->db_rec, 0, 4096);
400 ret = mthca_MAP_ICM_page(dev, page->mapping, mthca_uarc_virt(dev, i), &status);
404 dma_free_coherent(&dev->pdev->dev, 4096,
405 page->db_rec, page->mapping);
409 bitmap_zero(page->used, MTHCA_DB_REC_PER_PAGE);
411 ++dev->db_tab->max_group1;
413 --dev->db_tab->min_group2;
416 j = find_first_zero_bit(page->used, MTHCA_DB_REC_PER_PAGE);
417 set_bit(j, page->used);
420 j = MTHCA_DB_REC_PER_PAGE - 1 - j;
422 ret = i * MTHCA_DB_REC_PER_PAGE + j;
424 page->db_rec[j] = cpu_to_be64((qn << 8) | (type << 5));
426 *db = (u32 *) &page->db_rec[j];
429 up(&dev->db_tab->mutex);
434 void mthca_free_db(struct mthca_dev *dev, int type, int db_index)
437 struct mthca_db_page *page;
440 i = db_index / MTHCA_DB_REC_PER_PAGE;
441 j = db_index % MTHCA_DB_REC_PER_PAGE;
443 page = dev->db_tab->page + i;
445 down(&dev->db_tab->mutex);
448 if (i >= dev->db_tab->min_group2)
449 j = MTHCA_DB_REC_PER_PAGE - 1 - j;
450 clear_bit(j, page->used);
452 if (bitmap_empty(page->used, MTHCA_DB_REC_PER_PAGE) &&
453 i >= dev->db_tab->max_group1 - 1) {
454 mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, i), 1, &status);
456 dma_free_coherent(&dev->pdev->dev, 4096,
457 page->db_rec, page->mapping);
460 if (i == dev->db_tab->max_group1) {
461 --dev->db_tab->max_group1;
462 /* XXX may be able to unmap more pages now */
464 if (i == dev->db_tab->min_group2)
465 ++dev->db_tab->min_group2;
468 up(&dev->db_tab->mutex);
471 int mthca_init_db_tab(struct mthca_dev *dev)
475 if (dev->hca_type != ARBEL_NATIVE)
478 dev->db_tab = kmalloc(sizeof *dev->db_tab, GFP_KERNEL);
482 init_MUTEX(&dev->db_tab->mutex);
484 dev->db_tab->npages = dev->uar_table.uarc_size / 4096;
485 dev->db_tab->max_group1 = 0;
486 dev->db_tab->min_group2 = dev->db_tab->npages - 1;
488 dev->db_tab->page = kmalloc(dev->db_tab->npages *
489 sizeof *dev->db_tab->page,
491 if (!dev->db_tab->page) {
496 for (i = 0; i < dev->db_tab->npages; ++i)
497 dev->db_tab->page[i].db_rec = NULL;
502 void mthca_cleanup_db_tab(struct mthca_dev *dev)
507 if (dev->hca_type != ARBEL_NATIVE)
511 * Because we don't always free our UARC pages when they
512 * become empty to make mthca_free_db() simpler we need to
513 * make a sweep through the doorbell pages and free any
514 * leftover pages now.
516 for (i = 0; i < dev->db_tab->npages; ++i) {
517 if (!dev->db_tab->page[i].db_rec)
520 if (!bitmap_empty(dev->db_tab->page[i].used, MTHCA_DB_REC_PER_PAGE))
521 mthca_warn(dev, "Kernel UARC page %d not empty\n", i);
523 mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, i), 1, &status);
525 dma_free_coherent(&dev->pdev->dev, 4096,
526 dev->db_tab->page[i].db_rec,
527 dev->db_tab->page[i].mapping);
530 kfree(dev->db_tab->page);