1/*
2 * Copyright (c) 2006, 2007 Cisco Systems, Inc.  All rights reserved.
3 * Copyright (c) 2007, 2008, 2014 Mellanox Technologies. All rights reserved.
4 *
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:
10 *
11 *     Redistribution and use in source and binary forms, with or
12 *     without modification, are permitted provided that the following
13 *     conditions are met:
14 *
15 *      - Redistributions of source code must retain the above
16 *        copyright notice, this list of conditions and the following
17 *        disclaimer.
18 *
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.
23 *
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
31 * SOFTWARE.
32 */
33
34#include <linux/errno.h>
35#include <linux/slab.h>
36#include <linux/mm.h>
37#include <linux/module.h>
38#include <linux/dma-mapping.h>
39#include <linux/vmalloc.h>
40
41#include "mlx4.h"
42
43u32 mlx4_bitmap_alloc(struct mlx4_bitmap *bitmap)
44{
45	u32 obj;
46
47	spin_lock(&bitmap->lock);
48
49	obj = find_next_zero_bit(bitmap->table, bitmap->max, bitmap->last);
50	if (obj >= bitmap->max) {
51		bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
52				& bitmap->mask;
53		obj = find_first_zero_bit(bitmap->table, bitmap->max);
54	}
55
56	if (obj < bitmap->max) {
57		set_bit(obj, bitmap->table);
58		bitmap->last = (obj + 1);
59		if (bitmap->last == bitmap->max)
60			bitmap->last = 0;
61		obj |= bitmap->top;
62	} else
63		obj = -1;
64
65	if (obj != -1)
66		--bitmap->avail;
67
68	spin_unlock(&bitmap->lock);
69
70	return obj;
71}
72
73void mlx4_bitmap_free(struct mlx4_bitmap *bitmap, u32 obj, int use_rr)
74{
75	mlx4_bitmap_free_range(bitmap, obj, 1, use_rr);
76}
77
78static unsigned long find_aligned_range(unsigned long *bitmap,
79					u32 start, u32 nbits,
80					int len, int align, u32 skip_mask)
81{
82	unsigned long end, i;
83
84again:
85	start = ALIGN(start, align);
86
87	while ((start < nbits) && (test_bit(start, bitmap) ||
88				   (start & skip_mask)))
89		start += align;
90
91	if (start >= nbits)
92		return -1;
93
94	end = start+len;
95	if (end > nbits)
96		return -1;
97
98	for (i = start + 1; i < end; i++) {
99		if (test_bit(i, bitmap) || ((u32)i & skip_mask)) {
100			start = i + 1;
101			goto again;
102		}
103	}
104
105	return start;
106}
107
108u32 mlx4_bitmap_alloc_range(struct mlx4_bitmap *bitmap, int cnt,
109			    int align, u32 skip_mask)
110{
111	u32 obj;
112
113	if (likely(cnt == 1 && align == 1 && !skip_mask))
114		return mlx4_bitmap_alloc(bitmap);
115
116	spin_lock(&bitmap->lock);
117
118	obj = find_aligned_range(bitmap->table, bitmap->last,
119				bitmap->max, cnt, align, skip_mask);
120	if (obj >= bitmap->max) {
121		bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
122				& bitmap->mask;
123		obj = find_aligned_range(bitmap->table, 0, bitmap->max,
124						cnt, align, skip_mask);
125	}
126
127	if (obj < bitmap->max) {
128		bitmap_set(bitmap->table, obj, cnt);
129		if (obj == bitmap->last) {
130			bitmap->last = (obj + cnt);
131			if (bitmap->last >= bitmap->max)
132				bitmap->last = 0;
133		}
134		obj |= bitmap->top;
135	} else
136		obj = -1;
137
138	if (obj != -1)
139		bitmap->avail -= cnt;
140
141	spin_unlock(&bitmap->lock);
142
143	return obj;
144}
145
146u32 mlx4_bitmap_avail(struct mlx4_bitmap *bitmap)
147{
148	return bitmap->avail;
149}
150
151void mlx4_bitmap_free_range(struct mlx4_bitmap *bitmap, u32 obj, int cnt,
152			    int use_rr)
153{
154	obj &= bitmap->max + bitmap->reserved_top - 1;
155
156	spin_lock(&bitmap->lock);
157	if (!use_rr) {
158		bitmap->last = min(bitmap->last, obj);
159		bitmap->top = (bitmap->top + bitmap->max + bitmap->reserved_top)
160				& bitmap->mask;
161	}
162	bitmap_clear(bitmap->table, obj, cnt);
163	bitmap->avail += cnt;
164	spin_unlock(&bitmap->lock);
165}
166
167int mlx4_bitmap_init(struct mlx4_bitmap *bitmap, u32 num, u32 mask,
168		     u32 reserved_bot, u32 reserved_top)
169{
170	/* sanity check */
171	if (num <= (u64)reserved_top + reserved_bot)
172		return -EINVAL;
173
174	/* num must be a power of 2 */
175	if (num != roundup_pow_of_two(num))
176		return -EINVAL;
177
178	if (reserved_bot + reserved_top >= num)
179		return -EINVAL;
180
181	bitmap->last = 0;
182	bitmap->top  = 0;
183	bitmap->max  = num - reserved_top;
184	bitmap->mask = mask;
185	bitmap->reserved_top = reserved_top;
186	bitmap->avail = num - reserved_top - reserved_bot;
187	spin_lock_init(&bitmap->lock);
188	bitmap->table = kzalloc(BITS_TO_LONGS(bitmap->max) *
189				sizeof (long), GFP_KERNEL);
190	if (!bitmap->table)
191		return -ENOMEM;
192
193	bitmap_set(bitmap->table, 0, reserved_bot);
194
195	return 0;
196}
197
198void mlx4_bitmap_cleanup(struct mlx4_bitmap *bitmap)
199{
200	kfree(bitmap->table);
201}
202
203/*
204 * Handling for queue buffers -- we allocate a bunch of memory and
205 * register it in a memory region at HCA virtual address 0.  If the
206 * requested size is > max_direct, we split the allocation into
207 * multiple pages, so we don't require too much contiguous memory.
208 */
209
210int mlx4_buf_alloc(struct mlx4_dev *dev, int size, int max_direct,
211		   struct mlx4_buf *buf)
212{
213	dma_addr_t t;
214
215	if (size <= max_direct) {
216		buf->nbufs        = 1;
217		buf->npages       = 1;
218		buf->page_shift   = get_order(size) + PAGE_SHIFT;
219		buf->direct.buf   = dma_alloc_coherent(&dev->pdev->dev,
220						       size, &t, GFP_KERNEL);
221		if (!buf->direct.buf)
222			return -ENOMEM;
223
224		buf->direct.map = t;
225
226		while (t & ((1 << buf->page_shift) - 1)) {
227			--buf->page_shift;
228			buf->npages *= 2;
229		}
230
231		memset(buf->direct.buf, 0, size);
232	} else {
233		int i;
234
235		buf->direct.buf  = NULL;
236		buf->nbufs       = (size + PAGE_SIZE - 1) / PAGE_SIZE;
237		buf->npages      = buf->nbufs;
238		buf->page_shift  = PAGE_SHIFT;
239		buf->page_list   = kcalloc(buf->nbufs, sizeof(*buf->page_list),
240					   GFP_KERNEL);
241		if (!buf->page_list)
242			return -ENOMEM;
243
244		for (i = 0; i < buf->nbufs; ++i) {
245			buf->page_list[i].buf =
246				dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
247						   &t, GFP_KERNEL);
248			if (!buf->page_list[i].buf)
249				goto err_free;
250
251			buf->page_list[i].map = t;
252
253			memset(buf->page_list[i].buf, 0, PAGE_SIZE);
254		}
255
256		if (BITS_PER_LONG == 64) {
257			struct page **pages;
258			pages = kmalloc(sizeof *pages * buf->nbufs, GFP_KERNEL);
259			if (!pages)
260				goto err_free;
261			for (i = 0; i < buf->nbufs; ++i)
262				pages[i] = virt_to_page(buf->page_list[i].buf);
263			buf->direct.buf = vmap(pages, buf->nbufs, VM_MAP, PAGE_KERNEL);
264			kfree(pages);
265			if (!buf->direct.buf)
266				goto err_free;
267		}
268	}
269
270	return 0;
271
272err_free:
273	mlx4_buf_free(dev, size, buf);
274
275	return -ENOMEM;
276}
277EXPORT_SYMBOL_GPL(mlx4_buf_alloc);
278
279void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf)
280{
281	int i;
282
283	if (buf->nbufs == 1)
284		dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
285				  buf->direct.map);
286	else {
287		if (BITS_PER_LONG == 64 && buf->direct.buf)
288			vunmap(buf->direct.buf);
289
290		for (i = 0; i < buf->nbufs; ++i)
291			if (buf->page_list[i].buf)
292				dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
293						  buf->page_list[i].buf,
294						  buf->page_list[i].map);
295		kfree(buf->page_list);
296	}
297}
298EXPORT_SYMBOL_GPL(mlx4_buf_free);
299
300static struct mlx4_db_pgdir *mlx4_alloc_db_pgdir(struct device *dma_device)
301{
302	struct mlx4_db_pgdir *pgdir;
303
304	pgdir = kzalloc(sizeof *pgdir, GFP_KERNEL);
305	if (!pgdir)
306		return NULL;
307
308	bitmap_fill(pgdir->order1, MLX4_DB_PER_PAGE / 2);
309	pgdir->bits[0] = pgdir->order0;
310	pgdir->bits[1] = pgdir->order1;
311	pgdir->db_page = dma_alloc_coherent(dma_device, PAGE_SIZE,
312					    &pgdir->db_dma, GFP_KERNEL);
313	if (!pgdir->db_page) {
314		kfree(pgdir);
315		return NULL;
316	}
317
318	return pgdir;
319}
320
321static int mlx4_alloc_db_from_pgdir(struct mlx4_db_pgdir *pgdir,
322				    struct mlx4_db *db, int order)
323{
324	int o;
325	int i;
326
327	for (o = order; o <= 1; ++o) {
328		i = find_first_bit(pgdir->bits[o], MLX4_DB_PER_PAGE >> o);
329		if (i < MLX4_DB_PER_PAGE >> o)
330			goto found;
331	}
332
333	return -ENOMEM;
334
335found:
336	clear_bit(i, pgdir->bits[o]);
337
338	i <<= o;
339
340	if (o > order)
341		set_bit(i ^ 1, pgdir->bits[order]);
342
343	db->u.pgdir = pgdir;
344	db->index   = i;
345	db->db      = pgdir->db_page + db->index;
346	db->dma     = pgdir->db_dma  + db->index * 4;
347	db->order   = order;
348
349	return 0;
350}
351
352int mlx4_db_alloc(struct mlx4_dev *dev, struct mlx4_db *db, int order)
353{
354	struct mlx4_priv *priv = mlx4_priv(dev);
355	struct mlx4_db_pgdir *pgdir;
356	int ret = 0;
357
358	mutex_lock(&priv->pgdir_mutex);
359
360	list_for_each_entry(pgdir, &priv->pgdir_list, list)
361		if (!mlx4_alloc_db_from_pgdir(pgdir, db, order))
362			goto out;
363
364	pgdir = mlx4_alloc_db_pgdir(&(dev->pdev->dev));
365	if (!pgdir) {
366		ret = -ENOMEM;
367		goto out;
368	}
369
370	list_add(&pgdir->list, &priv->pgdir_list);
371
372	/* This should never fail -- we just allocated an empty page: */
373	WARN_ON(mlx4_alloc_db_from_pgdir(pgdir, db, order));
374
375out:
376	mutex_unlock(&priv->pgdir_mutex);
377
378	return ret;
379}
380EXPORT_SYMBOL_GPL(mlx4_db_alloc);
381
382void mlx4_db_free(struct mlx4_dev *dev, struct mlx4_db *db)
383{
384	struct mlx4_priv *priv = mlx4_priv(dev);
385	int o;
386	int i;
387
388	mutex_lock(&priv->pgdir_mutex);
389
390	o = db->order;
391	i = db->index;
392
393	if (db->order == 0 && test_bit(i ^ 1, db->u.pgdir->order0)) {
394		clear_bit(i ^ 1, db->u.pgdir->order0);
395		++o;
396	}
397	i >>= o;
398	set_bit(i, db->u.pgdir->bits[o]);
399
400	if (bitmap_full(db->u.pgdir->order1, MLX4_DB_PER_PAGE / 2)) {
401		dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE,
402				  db->u.pgdir->db_page, db->u.pgdir->db_dma);
403		list_del(&db->u.pgdir->list);
404		kfree(db->u.pgdir);
405	}
406
407	mutex_unlock(&priv->pgdir_mutex);
408}
409EXPORT_SYMBOL_GPL(mlx4_db_free);
410
411int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
412		       int size, int max_direct)
413{
414	int err;
415
416	err = mlx4_db_alloc(dev, &wqres->db, 1);
417	if (err)
418		return err;
419
420	*wqres->db.db = 0;
421
422	err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf);
423	if (err)
424		goto err_db;
425
426	err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
427			    &wqres->mtt);
428	if (err)
429		goto err_buf;
430
431	err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf);
432	if (err)
433		goto err_mtt;
434
435	return 0;
436
437err_mtt:
438	mlx4_mtt_cleanup(dev, &wqres->mtt);
439err_buf:
440	mlx4_buf_free(dev, size, &wqres->buf);
441err_db:
442	mlx4_db_free(dev, &wqres->db);
443
444	return err;
445}
446EXPORT_SYMBOL_GPL(mlx4_alloc_hwq_res);
447
448void mlx4_free_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
449		       int size)
450{
451	mlx4_mtt_cleanup(dev, &wqres->mtt);
452	mlx4_buf_free(dev, size, &wqres->buf);
453	mlx4_db_free(dev, &wqres->db);
454}
455EXPORT_SYMBOL_GPL(mlx4_free_hwq_res);
456