2 * Copyright (c) 2006 QLogic, Inc. All rights reserved.
3 * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
35 #include <linux/device.h>
36 #include <linux/vs_memory.h>
38 #include "ipath_kernel.h"
40 static void __ipath_release_user_pages(struct page **p, size_t num_pages,
45 for (i = 0; i < num_pages; i++) {
46 ipath_cdbg(MM, "%lu/%lu put_page %p\n", (unsigned long) i,
47 (unsigned long) num_pages, p[i]);
49 set_page_dirty_lock(p[i]);
54 /* call with current->mm->mmap_sem held */
55 static int __get_user_pages(unsigned long start_page, size_t num_pages,
56 struct page **p, struct vm_area_struct **vma)
58 unsigned long lock_limit;
62 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >>
65 if (num_pages > lock_limit ||
66 !vx_vmlocked_avail(current->mm, num_pages)) {
71 ipath_cdbg(VERBOSE, "pin %lx pages from vaddr %lx\n",
72 (unsigned long) num_pages, start_page);
74 for (got = 0; got < num_pages; got += ret) {
75 ret = get_user_pages(current, current->mm,
76 start_page + got * PAGE_SIZE,
77 num_pages - got, 1, 1,
83 vx_vmlocked_add(current->mm, num_pages);
89 __ipath_release_user_pages(p, got, 0);
95 * ipath_map_page - a safety wrapper around pci_map_page()
97 * A dma_addr of all 0's is interpreted by the chip as "disabled".
98 * Unfortunately, it can also be a valid dma_addr returned on some
101 * The powerpc iommu assigns dma_addrs in ascending order, so we don't
102 * have to bother with retries or mapping a dummy page to insure we
103 * don't just get the same mapping again.
105 * I'm sure we won't be so lucky with other iommu's, so FIXME.
107 dma_addr_t ipath_map_page(struct pci_dev *hwdev, struct page *page,
108 unsigned long offset, size_t size, int direction)
112 phys = pci_map_page(hwdev, page, offset, size, direction);
115 pci_unmap_page(hwdev, phys, size, direction);
116 phys = pci_map_page(hwdev, page, offset, size, direction);
118 * FIXME: If we get 0 again, we should keep this page,
119 * map another, then free the 0 page.
127 * ipath_map_single - a safety wrapper around pci_map_single()
129 * Same idea as ipath_map_page().
131 dma_addr_t ipath_map_single(struct pci_dev *hwdev, void *ptr, size_t size,
136 phys = pci_map_single(hwdev, ptr, size, direction);
139 pci_unmap_single(hwdev, phys, size, direction);
140 phys = pci_map_single(hwdev, ptr, size, direction);
142 * FIXME: If we get 0 again, we should keep this page,
143 * map another, then free the 0 page.
151 * ipath_get_user_pages - lock user pages into memory
152 * @start_page: the start page
153 * @num_pages: the number of pages
154 * @p: the output page structures
156 * This function takes a given start page (page aligned user virtual
157 * address) and pins it and the following specified number of pages. For
158 * now, num_pages is always 1, but that will probably change at some point
159 * (because caller is doing expected sends on a single virtually contiguous
160 * buffer, so we can do all pages at once).
162 int ipath_get_user_pages(unsigned long start_page, size_t num_pages,
167 down_write(¤t->mm->mmap_sem);
169 ret = __get_user_pages(start_page, num_pages, p, NULL);
171 up_write(¤t->mm->mmap_sem);
177 * ipath_get_user_pages_nocopy - lock a single page for I/O and mark shared
178 * @start_page: the page to lock
179 * @p: the output page structure
181 * This is similar to ipath_get_user_pages, but it's always one page, and we
182 * mark the page as locked for I/O, and shared. This is used for the user
183 * process page that contains the destination address for the rcvhdrq tail
184 * update, so we need to have the vma. If we don't do this, the page can be
185 * taken away from us on fork, even if the child never touches it, and then
186 * the user process never sees the tail register updates.
188 int ipath_get_user_pages_nocopy(unsigned long page, struct page **p)
190 struct vm_area_struct *vma;
193 down_write(¤t->mm->mmap_sem);
195 ret = __get_user_pages(page, 1, p, &vma);
197 up_write(¤t->mm->mmap_sem);
202 void ipath_release_user_pages(struct page **p, size_t num_pages)
204 down_write(¤t->mm->mmap_sem);
206 __ipath_release_user_pages(p, num_pages, 1);
208 vx_vmlocked_sub(current->mm, num_pages);
210 up_write(¤t->mm->mmap_sem);
213 struct ipath_user_pages_work {
214 struct work_struct work;
215 struct mm_struct *mm;
216 unsigned long num_pages;
219 static void user_pages_account(struct work_struct *_work)
221 struct ipath_user_pages_work *work =
222 container_of(_work, struct ipath_user_pages_work, work);
224 down_write(&work->mm->mmap_sem);
225 vx_vmlocked_sub(work->mm, work->num_pages);
226 up_write(&work->mm->mmap_sem);
231 void ipath_release_user_pages_on_close(struct page **p, size_t num_pages)
233 struct ipath_user_pages_work *work;
234 struct mm_struct *mm;
236 __ipath_release_user_pages(p, num_pages, 1);
238 mm = get_task_mm(current);
242 work = kmalloc(sizeof(*work), GFP_KERNEL);
248 INIT_WORK(&work->work, user_pages_account);
250 work->num_pages = num_pages;