mthca_mr.c revision 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 
1/*
2 * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3 *
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:
9 *
10 *     Redistribution and use in source and binary forms, with or
11 *     without modification, are permitted provided that the following
12 *     conditions are met:
13 *
14 *      - Redistributions of source code must retain the above
15 *        copyright notice, this list of conditions and the following
16 *        disclaimer.
17 *
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.
22 *
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
30 * SOFTWARE.
31 *
32 * $Id: mthca_mr.c 1349 2004-12-16 21:09:43Z roland $
33 */
34
35#include <linux/slab.h>
36#include <linux/init.h>
37#include <linux/errno.h>
38
39#include "mthca_dev.h"
40#include "mthca_cmd.h"
41
42/*
43 * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits.
44 */
45struct mthca_mpt_entry {
46	u32 flags;
47	u32 page_size;
48	u32 key;
49	u32 pd;
50	u64 start;
51	u64 length;
52	u32 lkey;
53	u32 window_count;
54	u32 window_count_limit;
55	u64 mtt_seg;
56	u32 mtt_sz;		/* Arbel only */
57	u32 reserved[2];
58} __attribute__((packed));
59
60#define MTHCA_MPT_FLAG_SW_OWNS       (0xfUL << 28)
61#define MTHCA_MPT_FLAG_MIO           (1 << 17)
62#define MTHCA_MPT_FLAG_BIND_ENABLE   (1 << 15)
63#define MTHCA_MPT_FLAG_PHYSICAL      (1 <<  9)
64#define MTHCA_MPT_FLAG_REGION        (1 <<  8)
65
66#define MTHCA_MTT_FLAG_PRESENT       1
67
68/*
69 * Buddy allocator for MTT segments (currently not very efficient
70 * since it doesn't keep a free list and just searches linearly
71 * through the bitmaps)
72 */
73
74static u32 mthca_alloc_mtt(struct mthca_dev *dev, int order)
75{
76	int o;
77	int m;
78	u32 seg;
79
80	spin_lock(&dev->mr_table.mpt_alloc.lock);
81
82	for (o = order; o <= dev->mr_table.max_mtt_order; ++o) {
83		m = 1 << (dev->mr_table.max_mtt_order - o);
84		seg = find_first_bit(dev->mr_table.mtt_buddy[o], m);
85		if (seg < m)
86			goto found;
87	}
88
89	spin_unlock(&dev->mr_table.mpt_alloc.lock);
90	return -1;
91
92 found:
93	clear_bit(seg, dev->mr_table.mtt_buddy[o]);
94
95	while (o > order) {
96		--o;
97		seg <<= 1;
98		set_bit(seg ^ 1, dev->mr_table.mtt_buddy[o]);
99	}
100
101	spin_unlock(&dev->mr_table.mpt_alloc.lock);
102
103	seg <<= order;
104
105	return seg;
106}
107
108static void mthca_free_mtt(struct mthca_dev *dev, u32 seg, int order)
109{
110	seg >>= order;
111
112	spin_lock(&dev->mr_table.mpt_alloc.lock);
113
114	while (test_bit(seg ^ 1, dev->mr_table.mtt_buddy[order])) {
115		clear_bit(seg ^ 1, dev->mr_table.mtt_buddy[order]);
116		seg >>= 1;
117		++order;
118	}
119
120	set_bit(seg, dev->mr_table.mtt_buddy[order]);
121
122	spin_unlock(&dev->mr_table.mpt_alloc.lock);
123}
124
125static inline u32 hw_index_to_key(struct mthca_dev *dev, u32 ind)
126{
127	if (dev->hca_type == ARBEL_NATIVE)
128		return (ind >> 24) | (ind << 8);
129	else
130		return ind;
131}
132
133static inline u32 key_to_hw_index(struct mthca_dev *dev, u32 key)
134{
135	if (dev->hca_type == ARBEL_NATIVE)
136		return (key << 24) | (key >> 8);
137	else
138		return key;
139}
140
141int mthca_mr_alloc_notrans(struct mthca_dev *dev, u32 pd,
142			   u32 access, struct mthca_mr *mr)
143{
144	void *mailbox;
145	struct mthca_mpt_entry *mpt_entry;
146	u32 key;
147	int err;
148	u8 status;
149
150	might_sleep();
151
152	mr->order = -1;
153	key = mthca_alloc(&dev->mr_table.mpt_alloc);
154	if (key == -1)
155		return -ENOMEM;
156	mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key);
157
158	mailbox = kmalloc(sizeof *mpt_entry + MTHCA_CMD_MAILBOX_EXTRA,
159			  GFP_KERNEL);
160	if (!mailbox) {
161		mthca_free(&dev->mr_table.mpt_alloc, mr->ibmr.lkey);
162		return -ENOMEM;
163	}
164	mpt_entry = MAILBOX_ALIGN(mailbox);
165
166	mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS     |
167				       MTHCA_MPT_FLAG_MIO         |
168				       MTHCA_MPT_FLAG_PHYSICAL    |
169				       MTHCA_MPT_FLAG_REGION      |
170				       access);
171	mpt_entry->page_size = 0;
172	mpt_entry->key       = cpu_to_be32(key);
173	mpt_entry->pd        = cpu_to_be32(pd);
174	mpt_entry->start     = 0;
175	mpt_entry->length    = ~0ULL;
176
177	memset(&mpt_entry->lkey, 0,
178	       sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, lkey));
179
180	err = mthca_SW2HW_MPT(dev, mpt_entry,
181			      key & (dev->limits.num_mpts - 1),
182			      &status);
183	if (err)
184		mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
185	else if (status) {
186		mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
187			   status);
188		err = -EINVAL;
189	}
190
191	kfree(mailbox);
192	return err;
193}
194
195int mthca_mr_alloc_phys(struct mthca_dev *dev, u32 pd,
196			u64 *buffer_list, int buffer_size_shift,
197			int list_len, u64 iova, u64 total_size,
198			u32 access, struct mthca_mr *mr)
199{
200	void *mailbox;
201	u64 *mtt_entry;
202	struct mthca_mpt_entry *mpt_entry;
203	u32 key;
204	int err = -ENOMEM;
205	u8 status;
206	int i;
207
208	might_sleep();
209	WARN_ON(buffer_size_shift >= 32);
210
211	key = mthca_alloc(&dev->mr_table.mpt_alloc);
212	if (key == -1)
213		return -ENOMEM;
214	mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key);
215
216	for (i = dev->limits.mtt_seg_size / 8, mr->order = 0;
217	     i < list_len;
218	     i <<= 1, ++mr->order)
219		; /* nothing */
220
221	mr->first_seg = mthca_alloc_mtt(dev, mr->order);
222	if (mr->first_seg == -1)
223		goto err_out_mpt_free;
224
225	/*
226	 * If list_len is odd, we add one more dummy entry for
227	 * firmware efficiency.
228	 */
229	mailbox = kmalloc(max(sizeof *mpt_entry,
230			      (size_t) 8 * (list_len + (list_len & 1) + 2)) +
231			  MTHCA_CMD_MAILBOX_EXTRA,
232			  GFP_KERNEL);
233	if (!mailbox)
234		goto err_out_free_mtt;
235
236	mtt_entry = MAILBOX_ALIGN(mailbox);
237
238	mtt_entry[0] = cpu_to_be64(dev->mr_table.mtt_base +
239				   mr->first_seg * dev->limits.mtt_seg_size);
240	mtt_entry[1] = 0;
241	for (i = 0; i < list_len; ++i)
242		mtt_entry[i + 2] = cpu_to_be64(buffer_list[i] |
243					       MTHCA_MTT_FLAG_PRESENT);
244	if (list_len & 1) {
245		mtt_entry[i + 2] = 0;
246		++list_len;
247	}
248
249	if (0) {
250		mthca_dbg(dev, "Dumping MPT entry\n");
251		for (i = 0; i < list_len + 2; ++i)
252			printk(KERN_ERR "[%2d] %016llx\n",
253			       i, (unsigned long long) be64_to_cpu(mtt_entry[i]));
254	}
255
256	err = mthca_WRITE_MTT(dev, mtt_entry, list_len, &status);
257	if (err) {
258		mthca_warn(dev, "WRITE_MTT failed (%d)\n", err);
259		goto err_out_mailbox_free;
260	}
261	if (status) {
262		mthca_warn(dev, "WRITE_MTT returned status 0x%02x\n",
263			   status);
264		err = -EINVAL;
265		goto err_out_mailbox_free;
266	}
267
268	mpt_entry = MAILBOX_ALIGN(mailbox);
269
270	mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS     |
271				       MTHCA_MPT_FLAG_MIO         |
272				       MTHCA_MPT_FLAG_REGION      |
273				       access);
274
275	mpt_entry->page_size = cpu_to_be32(buffer_size_shift - 12);
276	mpt_entry->key       = cpu_to_be32(key);
277	mpt_entry->pd        = cpu_to_be32(pd);
278	mpt_entry->start     = cpu_to_be64(iova);
279	mpt_entry->length    = cpu_to_be64(total_size);
280	memset(&mpt_entry->lkey, 0,
281	       sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, lkey));
282	mpt_entry->mtt_seg   = cpu_to_be64(dev->mr_table.mtt_base +
283					   mr->first_seg * dev->limits.mtt_seg_size);
284
285	if (0) {
286		mthca_dbg(dev, "Dumping MPT entry %08x:\n", mr->ibmr.lkey);
287		for (i = 0; i < sizeof (struct mthca_mpt_entry) / 4; ++i) {
288			if (i % 4 == 0)
289				printk("[%02x] ", i * 4);
290			printk(" %08x", be32_to_cpu(((u32 *) mpt_entry)[i]));
291			if ((i + 1) % 4 == 0)
292				printk("\n");
293		}
294	}
295
296	err = mthca_SW2HW_MPT(dev, mpt_entry,
297			      key & (dev->limits.num_mpts - 1),
298			      &status);
299	if (err)
300		mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
301	else if (status) {
302		mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
303			   status);
304		err = -EINVAL;
305	}
306
307	kfree(mailbox);
308	return err;
309
310 err_out_mailbox_free:
311	kfree(mailbox);
312
313 err_out_free_mtt:
314	mthca_free_mtt(dev, mr->first_seg, mr->order);
315
316 err_out_mpt_free:
317	mthca_free(&dev->mr_table.mpt_alloc, mr->ibmr.lkey);
318	return err;
319}
320
321void mthca_free_mr(struct mthca_dev *dev, struct mthca_mr *mr)
322{
323	int err;
324	u8 status;
325
326	might_sleep();
327
328	err = mthca_HW2SW_MPT(dev, NULL,
329			      key_to_hw_index(dev, mr->ibmr.lkey) &
330			      (dev->limits.num_mpts - 1),
331			      &status);
332	if (err)
333		mthca_warn(dev, "HW2SW_MPT failed (%d)\n", err);
334	else if (status)
335		mthca_warn(dev, "HW2SW_MPT returned status 0x%02x\n",
336			   status);
337
338	if (mr->order >= 0)
339		mthca_free_mtt(dev, mr->first_seg, mr->order);
340
341	mthca_free(&dev->mr_table.mpt_alloc, key_to_hw_index(dev, mr->ibmr.lkey));
342}
343
344int __devinit mthca_init_mr_table(struct mthca_dev *dev)
345{
346	int err;
347	int i, s;
348
349	err = mthca_alloc_init(&dev->mr_table.mpt_alloc,
350			       dev->limits.num_mpts,
351			       ~0, dev->limits.reserved_mrws);
352	if (err)
353		return err;
354
355	err = -ENOMEM;
356
357	for (i = 1, dev->mr_table.max_mtt_order = 0;
358	     i < dev->limits.num_mtt_segs;
359	     i <<= 1, ++dev->mr_table.max_mtt_order)
360		; /* nothing */
361
362	dev->mr_table.mtt_buddy = kmalloc((dev->mr_table.max_mtt_order + 1) *
363					  sizeof (long *),
364					  GFP_KERNEL);
365	if (!dev->mr_table.mtt_buddy)
366		goto err_out;
367
368	for (i = 0; i <= dev->mr_table.max_mtt_order; ++i)
369		dev->mr_table.mtt_buddy[i] = NULL;
370
371	for (i = 0; i <= dev->mr_table.max_mtt_order; ++i) {
372		s = BITS_TO_LONGS(1 << (dev->mr_table.max_mtt_order - i));
373		dev->mr_table.mtt_buddy[i] = kmalloc(s * sizeof (long),
374						     GFP_KERNEL);
375		if (!dev->mr_table.mtt_buddy[i])
376			goto err_out_free;
377		bitmap_zero(dev->mr_table.mtt_buddy[i],
378			    1 << (dev->mr_table.max_mtt_order - i));
379	}
380
381	set_bit(0, dev->mr_table.mtt_buddy[dev->mr_table.max_mtt_order]);
382
383	for (i = 0; i < dev->mr_table.max_mtt_order; ++i)
384		if (1 << i >= dev->limits.reserved_mtts)
385			break;
386
387	if (i == dev->mr_table.max_mtt_order) {
388		mthca_err(dev, "MTT table of order %d is "
389			  "too small.\n", i);
390		goto err_out_free;
391	}
392
393	(void) mthca_alloc_mtt(dev, i);
394
395	return 0;
396
397 err_out_free:
398	for (i = 0; i <= dev->mr_table.max_mtt_order; ++i)
399		kfree(dev->mr_table.mtt_buddy[i]);
400
401 err_out:
402	mthca_alloc_cleanup(&dev->mr_table.mpt_alloc);
403
404	return err;
405}
406
407void __devexit mthca_cleanup_mr_table(struct mthca_dev *dev)
408{
409	int i;
410
411	/* XXX check if any MRs are still allocated? */
412	for (i = 0; i <= dev->mr_table.max_mtt_order; ++i)
413		kfree(dev->mr_table.mtt_buddy[i]);
414	kfree(dev->mr_table.mtt_buddy);
415	mthca_alloc_cleanup(&dev->mr_table.mpt_alloc);
416}
417