summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/ipath/ipath_user_pages.c
blob: 27034d38b3ddcadcdcfed50c066081533e8915ac (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
/*
 * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
 * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/mm.h>
#include <linux/device.h>

#include "ipath_kernel.h"

static void __ipath_release_user_pages(struct page **p, size_t num_pages,
				   int dirty)
{
	size_t i;

	for (i = 0; i < num_pages; i++) {
		ipath_cdbg(MM, "%lu/%lu put_page %p\n", (unsigned long) i,
			   (unsigned long) num_pages, p[i]);
		if (dirty)
			set_page_dirty_lock(p[i]);
		put_page(p[i]);
	}
}

/* call with current->mm->mmap_sem held */
static int __get_user_pages(unsigned long start_page, size_t num_pages,
			struct page **p, struct vm_area_struct **vma)
{
	unsigned long lock_limit;
	size_t got;
	int ret;

	lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >>
		PAGE_SHIFT;

	if (num_pages > lock_limit) {
		ret = -ENOMEM;
		goto bail;
	}

	ipath_cdbg(VERBOSE, "pin %lx pages from vaddr %lx\n",
		   (unsigned long) num_pages, start_page);

	for (got = 0; got < num_pages; got += ret) {
		ret = get_user_pages(current, current->mm,
				     start_page + got * PAGE_SIZE,
				     num_pages - got, 1, 1,
				     p + got, vma);
		if (ret < 0)
			goto bail_release;
	}

	current->mm->locked_vm += num_pages;

	ret = 0;
	goto bail;

bail_release:
	__ipath_release_user_pages(p, got, 0);
bail:
	return ret;
}

/**
 * ipath_map_page - a safety wrapper around pci_map_page()
 *
 * A dma_addr of all 0's is interpreted by the chip as "disabled".
 * Unfortunately, it can also be a valid dma_addr returned on some
 * architectures.
 *
 * The powerpc iommu assigns dma_addrs in ascending order, so we don't
 * have to bother with retries or mapping a dummy page to insure we
 * don't just get the same mapping again.
 *
 * I'm sure we won't be so lucky with other iommu's, so FIXME.
 */
dma_addr_t ipath_map_page(struct pci_dev *hwdev, struct page *page,
	unsigned long offset, size_t size, int direction)
{
	dma_addr_t phys;

	phys = pci_map_page(hwdev, page, offset, size, direction);

	if (phys == 0) {
		pci_unmap_page(hwdev, phys, size, direction);
		phys = pci_map_page(hwdev, page, offset, size, direction);
		/*
		 * FIXME: If we get 0 again, we should keep this page,
		 * map another, then free the 0 page.
		 */
	}

	return phys;
}

/**
 * ipath_map_single - a safety wrapper around pci_map_single()
 *
 * Same idea as ipath_map_page().
 */
dma_addr_t ipath_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
	int direction)
{
	dma_addr_t phys;

	phys = pci_map_single(hwdev, ptr, size, direction);

	if (phys == 0) {
		pci_unmap_single(hwdev, phys, size, direction);
		phys = pci_map_single(hwdev, ptr, size, direction);
		/*
		 * FIXME: If we get 0 again, we should keep this page,
		 * map another, then free the 0 page.
		 */
	}

	return phys;
}

/**
 * ipath_get_user_pages - lock user pages into memory
 * @start_page: the start page
 * @num_pages: the number of pages
 * @p: the output page structures
 *
 * This function takes a given start page (page aligned user virtual
 * address) and pins it and the following specified number of pages.  For
 * now, num_pages is always 1, but that will probably change at some point
 * (because caller is doing expected sends on a single virtually contiguous
 * buffer, so we can do all pages at once).
 */
int ipath_get_user_pages(unsigned long start_page, size_t num_pages,
			 struct page **p)
{
	int ret;

	down_write(&current->mm->mmap_sem);

	ret = __get_user_pages(start_page, num_pages, p, NULL);

	up_write(&current->mm->mmap_sem);

	return ret;
}

/**
 * ipath_get_user_pages_nocopy - lock a single page for I/O and mark shared
 * @start_page: the page to lock
 * @p: the output page structure
 *
 * This is similar to ipath_get_user_pages, but it's always one page, and we
 * mark the page as locked for I/O, and shared.  This is used for the user
 * process page that contains the destination address for the rcvhdrq tail
 * update, so we need to have the vma. If we don't do this, the page can be
 * taken away from us on fork, even if the child never touches it, and then
 * the user process never sees the tail register updates.
 */
int ipath_get_user_pages_nocopy(unsigned long page, struct page **p)
{
	struct vm_area_struct *vma;
	int ret;

	down_write(&current->mm->mmap_sem);

	ret = __get_user_pages(page, 1, p, &vma);

	up_write(&current->mm->mmap_sem);

	return ret;
}

void ipath_release_user_pages(struct page **p, size_t num_pages)
{
	down_write(&current->mm->mmap_sem);

	__ipath_release_user_pages(p, num_pages, 1);

	current->mm->locked_vm -= num_pages;

	up_write(&current->mm->mmap_sem);
}

struct ipath_user_pages_work {
	struct work_struct work;
	struct mm_struct *mm;
	unsigned long num_pages;
};

static void user_pages_account(struct work_struct *_work)
{
	struct ipath_user_pages_work *work =
		container_of(_work, struct ipath_user_pages_work, work);

	down_write(&work->mm->mmap_sem);
	work->mm->locked_vm -= work->num_pages;
	up_write(&work->mm->mmap_sem);
	mmput(work->mm);
	kfree(work);
}

void ipath_release_user_pages_on_close(struct page **p, size_t num_pages)
{
	struct ipath_user_pages_work *work;
	struct mm_struct *mm;

	__ipath_release_user_pages(p, num_pages, 1);

	mm = get_task_mm(current);
	if (!mm)
		goto bail;

	work = kmalloc(sizeof(*work), GFP_KERNEL);
	if (!work)
		goto bail_mm;

	goto bail;

	INIT_WORK(&work->work, user_pages_account);
	work->mm = mm;
	work->num_pages = num_pages;

bail_mm:
	mmput(mm);
bail:
	return;
}