2 * Copyright (c) 2004 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
32 * $Id: mthca_mr.c 1349 2004-12-16 21:09:43Z roland $
35 #include <linux/slab.h>
36 #include <linux/init.h>
37 #include <linux/errno.h>
39 #include "mthca_dev.h"
40 #include "mthca_cmd.h"
41 #include "mthca_memfree.h"
44 * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits.
46 struct mthca_mpt_entry {
55 u32 window_count_limit;
57 u32 mtt_sz; /* Arbel only */
59 } __attribute__((packed));
61 #define MTHCA_MPT_FLAG_SW_OWNS (0xfUL << 28)
62 #define MTHCA_MPT_FLAG_MIO (1 << 17)
63 #define MTHCA_MPT_FLAG_BIND_ENABLE (1 << 15)
64 #define MTHCA_MPT_FLAG_PHYSICAL (1 << 9)
65 #define MTHCA_MPT_FLAG_REGION (1 << 8)
67 #define MTHCA_MTT_FLAG_PRESENT 1
70 * Buddy allocator for MTT segments (currently not very efficient
71 * since it doesn't keep a free list and just searches linearly
72 * through the bitmaps)
75 static u32 __mthca_alloc_mtt(struct mthca_dev *dev, int order)
81 spin_lock(&dev->mr_table.mpt_alloc.lock);
83 for (o = order; o <= dev->mr_table.max_mtt_order; ++o) {
84 m = 1 << (dev->mr_table.max_mtt_order - o);
85 seg = find_first_bit(dev->mr_table.mtt_buddy[o], m);
90 spin_unlock(&dev->mr_table.mpt_alloc.lock);
94 clear_bit(seg, dev->mr_table.mtt_buddy[o]);
99 set_bit(seg ^ 1, dev->mr_table.mtt_buddy[o]);
102 spin_unlock(&dev->mr_table.mpt_alloc.lock);
109 static void __mthca_free_mtt(struct mthca_dev *dev, u32 seg, int order)
113 spin_lock(&dev->mr_table.mpt_alloc.lock);
115 while (test_bit(seg ^ 1, dev->mr_table.mtt_buddy[order])) {
116 clear_bit(seg ^ 1, dev->mr_table.mtt_buddy[order]);
121 set_bit(seg, dev->mr_table.mtt_buddy[order]);
123 spin_unlock(&dev->mr_table.mpt_alloc.lock);
126 static u32 mthca_alloc_mtt(struct mthca_dev *dev, int order)
128 u32 seg = __mthca_alloc_mtt(dev, order);
133 if (dev->hca_type == ARBEL_NATIVE)
134 if (mthca_table_get_range(dev, dev->mr_table.mtt_table, seg,
135 seg + (1 << order) - 1)) {
136 __mthca_free_mtt(dev, seg, order);
143 static void mthca_free_mtt(struct mthca_dev *dev, u32 seg, int order)
145 __mthca_free_mtt(dev, seg, order);
147 if (dev->hca_type == ARBEL_NATIVE)
148 mthca_table_put_range(dev, dev->mr_table.mtt_table, seg,
149 seg + (1 << order) - 1);
152 static inline u32 hw_index_to_key(struct mthca_dev *dev, u32 ind)
154 if (dev->hca_type == ARBEL_NATIVE)
155 return (ind >> 24) | (ind << 8);
160 static inline u32 key_to_hw_index(struct mthca_dev *dev, u32 key)
162 if (dev->hca_type == ARBEL_NATIVE)
163 return (key << 24) | (key >> 8);
168 int mthca_mr_alloc_notrans(struct mthca_dev *dev, u32 pd,
169 u32 access, struct mthca_mr *mr)
171 void *mailbox = NULL;
172 struct mthca_mpt_entry *mpt_entry;
180 key = mthca_alloc(&dev->mr_table.mpt_alloc);
183 mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key);
185 if (dev->hca_type == ARBEL_NATIVE) {
186 err = mthca_table_get(dev, dev->mr_table.mpt_table, key);
188 goto err_out_mpt_free;
191 mailbox = kmalloc(sizeof *mpt_entry + MTHCA_CMD_MAILBOX_EXTRA,
197 mpt_entry = MAILBOX_ALIGN(mailbox);
199 mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS |
201 MTHCA_MPT_FLAG_PHYSICAL |
202 MTHCA_MPT_FLAG_REGION |
204 mpt_entry->page_size = 0;
205 mpt_entry->key = cpu_to_be32(key);
206 mpt_entry->pd = cpu_to_be32(pd);
207 mpt_entry->start = 0;
208 mpt_entry->length = ~0ULL;
210 memset(&mpt_entry->lkey, 0,
211 sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, lkey));
213 err = mthca_SW2HW_MPT(dev, mpt_entry,
214 key & (dev->limits.num_mpts - 1),
217 mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
220 mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
230 if (dev->hca_type == ARBEL_NATIVE)
231 mthca_table_put(dev, dev->mr_table.mpt_table, key);
234 mthca_free(&dev->mr_table.mpt_alloc, mr->ibmr.lkey);
239 int mthca_mr_alloc_phys(struct mthca_dev *dev, u32 pd,
240 u64 *buffer_list, int buffer_size_shift,
241 int list_len, u64 iova, u64 total_size,
242 u32 access, struct mthca_mr *mr)
246 struct mthca_mpt_entry *mpt_entry;
253 WARN_ON(buffer_size_shift >= 32);
255 key = mthca_alloc(&dev->mr_table.mpt_alloc);
258 mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key);
260 if (dev->hca_type == ARBEL_NATIVE) {
261 err = mthca_table_get(dev, dev->mr_table.mpt_table, key);
263 goto err_out_mpt_free;
266 for (i = dev->limits.mtt_seg_size / 8, mr->order = 0;
268 i <<= 1, ++mr->order)
271 mr->first_seg = mthca_alloc_mtt(dev, mr->order);
272 if (mr->first_seg == -1)
276 * If list_len is odd, we add one more dummy entry for
277 * firmware efficiency.
279 mailbox = kmalloc(max(sizeof *mpt_entry,
280 (size_t) 8 * (list_len + (list_len & 1) + 2)) +
281 MTHCA_CMD_MAILBOX_EXTRA,
284 goto err_out_free_mtt;
286 mtt_entry = MAILBOX_ALIGN(mailbox);
288 mtt_entry[0] = cpu_to_be64(dev->mr_table.mtt_base +
289 mr->first_seg * dev->limits.mtt_seg_size);
291 for (i = 0; i < list_len; ++i)
292 mtt_entry[i + 2] = cpu_to_be64(buffer_list[i] |
293 MTHCA_MTT_FLAG_PRESENT);
295 mtt_entry[i + 2] = 0;
300 mthca_dbg(dev, "Dumping MPT entry\n");
301 for (i = 0; i < list_len + 2; ++i)
302 printk(KERN_ERR "[%2d] %016llx\n",
303 i, (unsigned long long) be64_to_cpu(mtt_entry[i]));
306 err = mthca_WRITE_MTT(dev, mtt_entry, list_len, &status);
308 mthca_warn(dev, "WRITE_MTT failed (%d)\n", err);
309 goto err_out_mailbox_free;
312 mthca_warn(dev, "WRITE_MTT returned status 0x%02x\n",
315 goto err_out_mailbox_free;
318 mpt_entry = MAILBOX_ALIGN(mailbox);
320 mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS |
322 MTHCA_MPT_FLAG_REGION |
325 mpt_entry->page_size = cpu_to_be32(buffer_size_shift - 12);
326 mpt_entry->key = cpu_to_be32(key);
327 mpt_entry->pd = cpu_to_be32(pd);
328 mpt_entry->start = cpu_to_be64(iova);
329 mpt_entry->length = cpu_to_be64(total_size);
330 memset(&mpt_entry->lkey, 0,
331 sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, lkey));
332 mpt_entry->mtt_seg = cpu_to_be64(dev->mr_table.mtt_base +
333 mr->first_seg * dev->limits.mtt_seg_size);
336 mthca_dbg(dev, "Dumping MPT entry %08x:\n", mr->ibmr.lkey);
337 for (i = 0; i < sizeof (struct mthca_mpt_entry) / 4; ++i) {
339 printk("[%02x] ", i * 4);
340 printk(" %08x", be32_to_cpu(((u32 *) mpt_entry)[i]));
341 if ((i + 1) % 4 == 0)
346 err = mthca_SW2HW_MPT(dev, mpt_entry,
347 key & (dev->limits.num_mpts - 1),
350 mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
352 mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
360 err_out_mailbox_free:
364 mthca_free_mtt(dev, mr->first_seg, mr->order);
367 if (dev->hca_type == ARBEL_NATIVE)
368 mthca_table_put(dev, dev->mr_table.mpt_table, key);
371 mthca_free(&dev->mr_table.mpt_alloc, mr->ibmr.lkey);
375 void mthca_free_mr(struct mthca_dev *dev, struct mthca_mr *mr)
382 err = mthca_HW2SW_MPT(dev, NULL,
383 key_to_hw_index(dev, mr->ibmr.lkey) &
384 (dev->limits.num_mpts - 1),
387 mthca_warn(dev, "HW2SW_MPT failed (%d)\n", err);
389 mthca_warn(dev, "HW2SW_MPT returned status 0x%02x\n",
393 mthca_free_mtt(dev, mr->first_seg, mr->order);
395 if (dev->hca_type == ARBEL_NATIVE)
396 mthca_table_put(dev, dev->mr_table.mpt_table,
397 key_to_hw_index(dev, mr->ibmr.lkey));
398 mthca_free(&dev->mr_table.mpt_alloc, key_to_hw_index(dev, mr->ibmr.lkey));
401 int __devinit mthca_init_mr_table(struct mthca_dev *dev)
406 err = mthca_alloc_init(&dev->mr_table.mpt_alloc,
407 dev->limits.num_mpts,
408 ~0, dev->limits.reserved_mrws);
414 for (i = 1, dev->mr_table.max_mtt_order = 0;
415 i < dev->limits.num_mtt_segs;
416 i <<= 1, ++dev->mr_table.max_mtt_order)
419 dev->mr_table.mtt_buddy = kmalloc((dev->mr_table.max_mtt_order + 1) *
422 if (!dev->mr_table.mtt_buddy)
425 for (i = 0; i <= dev->mr_table.max_mtt_order; ++i)
426 dev->mr_table.mtt_buddy[i] = NULL;
428 for (i = 0; i <= dev->mr_table.max_mtt_order; ++i) {
429 s = BITS_TO_LONGS(1 << (dev->mr_table.max_mtt_order - i));
430 dev->mr_table.mtt_buddy[i] = kmalloc(s * sizeof (long),
432 if (!dev->mr_table.mtt_buddy[i])
434 bitmap_zero(dev->mr_table.mtt_buddy[i],
435 1 << (dev->mr_table.max_mtt_order - i));
438 set_bit(0, dev->mr_table.mtt_buddy[dev->mr_table.max_mtt_order]);
440 for (i = 0; i < dev->mr_table.max_mtt_order; ++i)
441 if (1 << i >= dev->limits.reserved_mtts)
444 if (i == dev->mr_table.max_mtt_order) {
445 mthca_err(dev, "MTT table of order %d is "
450 (void) mthca_alloc_mtt(dev, i);
455 for (i = 0; i <= dev->mr_table.max_mtt_order; ++i)
456 kfree(dev->mr_table.mtt_buddy[i]);
459 mthca_alloc_cleanup(&dev->mr_table.mpt_alloc);
464 void __devexit mthca_cleanup_mr_table(struct mthca_dev *dev)
468 /* XXX check if any MRs are still allocated? */
469 for (i = 0; i <= dev->mr_table.max_mtt_order; ++i)
470 kfree(dev->mr_table.mtt_buddy[i]);
471 kfree(dev->mr_table.mtt_buddy);
472 mthca_alloc_cleanup(&dev->mr_table.mpt_alloc);