Roland Dreier | eb8ffbf | 2005-07-07 17:57:14 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2005 Topspin Communications. All rights reserved. |
| 3 | * Copyright (c) 2005 Cisco Systems. All rights reserved. |
Roland Dreier | 2a1d9b7 | 2005-08-10 23:03:10 -0700 | [diff] [blame] | 4 | * Copyright (c) 2005 Mellanox Technologies. All rights reserved. |
Roland Dreier | eb8ffbf | 2005-07-07 17:57:14 -0700 | [diff] [blame] | 5 | * |
| 6 | * This software is available to you under a choice of one of two |
| 7 | * licenses. You may choose to be licensed under the terms of the GNU |
| 8 | * General Public License (GPL) Version 2, available from the file |
| 9 | * COPYING in the main directory of this source tree, or the |
| 10 | * OpenIB.org BSD license below: |
| 11 | * |
| 12 | * Redistribution and use in source and binary forms, with or |
| 13 | * without modification, are permitted provided that the following |
| 14 | * conditions are met: |
| 15 | * |
| 16 | * - Redistributions of source code must retain the above |
| 17 | * copyright notice, this list of conditions and the following |
| 18 | * disclaimer. |
| 19 | * |
| 20 | * - Redistributions in binary form must reproduce the above |
| 21 | * copyright notice, this list of conditions and the following |
| 22 | * disclaimer in the documentation and/or other materials |
| 23 | * provided with the distribution. |
| 24 | * |
| 25 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
| 26 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
| 27 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
| 28 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS |
| 29 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN |
| 30 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN |
| 31 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 32 | * SOFTWARE. |
| 33 | * |
| 34 | * $Id: uverbs_mem.c 2743 2005-06-28 22:27:59Z roland $ |
| 35 | */ |
| 36 | |
| 37 | #include <linux/mm.h> |
| 38 | #include <linux/dma-mapping.h> |
| 39 | |
| 40 | #include "uverbs.h" |
| 41 | |
| 42 | struct ib_umem_account_work { |
| 43 | struct work_struct work; |
| 44 | struct mm_struct *mm; |
| 45 | unsigned long diff; |
| 46 | }; |
| 47 | |
| 48 | |
| 49 | static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty) |
| 50 | { |
| 51 | struct ib_umem_chunk *chunk, *tmp; |
| 52 | int i; |
| 53 | |
| 54 | list_for_each_entry_safe(chunk, tmp, &umem->chunk_list, list) { |
| 55 | dma_unmap_sg(dev->dma_device, chunk->page_list, |
| 56 | chunk->nents, DMA_BIDIRECTIONAL); |
| 57 | for (i = 0; i < chunk->nents; ++i) { |
| 58 | if (umem->writable && dirty) |
| 59 | set_page_dirty_lock(chunk->page_list[i].page); |
| 60 | put_page(chunk->page_list[i].page); |
| 61 | } |
| 62 | |
| 63 | kfree(chunk); |
| 64 | } |
| 65 | } |
| 66 | |
| 67 | int ib_umem_get(struct ib_device *dev, struct ib_umem *mem, |
| 68 | void *addr, size_t size, int write) |
| 69 | { |
| 70 | struct page **page_list; |
| 71 | struct ib_umem_chunk *chunk; |
| 72 | unsigned long locked; |
| 73 | unsigned long lock_limit; |
| 74 | unsigned long cur_base; |
| 75 | unsigned long npages; |
| 76 | int ret = 0; |
| 77 | int off; |
| 78 | int i; |
| 79 | |
| 80 | if (!can_do_mlock()) |
| 81 | return -EPERM; |
| 82 | |
| 83 | page_list = (struct page **) __get_free_page(GFP_KERNEL); |
| 84 | if (!page_list) |
| 85 | return -ENOMEM; |
| 86 | |
| 87 | mem->user_base = (unsigned long) addr; |
| 88 | mem->length = size; |
| 89 | mem->offset = (unsigned long) addr & ~PAGE_MASK; |
| 90 | mem->page_size = PAGE_SIZE; |
| 91 | mem->writable = write; |
| 92 | |
| 93 | INIT_LIST_HEAD(&mem->chunk_list); |
| 94 | |
| 95 | npages = PAGE_ALIGN(size + mem->offset) >> PAGE_SHIFT; |
| 96 | |
| 97 | down_write(¤t->mm->mmap_sem); |
| 98 | |
| 99 | locked = npages + current->mm->locked_vm; |
| 100 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; |
| 101 | |
| 102 | if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { |
| 103 | ret = -ENOMEM; |
| 104 | goto out; |
| 105 | } |
| 106 | |
| 107 | cur_base = (unsigned long) addr & PAGE_MASK; |
| 108 | |
| 109 | while (npages) { |
| 110 | ret = get_user_pages(current, current->mm, cur_base, |
| 111 | min_t(int, npages, |
| 112 | PAGE_SIZE / sizeof (struct page *)), |
| 113 | 1, !write, page_list, NULL); |
| 114 | |
| 115 | if (ret < 0) |
| 116 | goto out; |
| 117 | |
| 118 | cur_base += ret * PAGE_SIZE; |
| 119 | npages -= ret; |
| 120 | |
| 121 | off = 0; |
| 122 | |
| 123 | while (ret) { |
| 124 | chunk = kmalloc(sizeof *chunk + sizeof (struct scatterlist) * |
| 125 | min_t(int, ret, IB_UMEM_MAX_PAGE_CHUNK), |
| 126 | GFP_KERNEL); |
| 127 | if (!chunk) { |
| 128 | ret = -ENOMEM; |
| 129 | goto out; |
| 130 | } |
| 131 | |
| 132 | chunk->nents = min_t(int, ret, IB_UMEM_MAX_PAGE_CHUNK); |
| 133 | for (i = 0; i < chunk->nents; ++i) { |
| 134 | chunk->page_list[i].page = page_list[i + off]; |
| 135 | chunk->page_list[i].offset = 0; |
| 136 | chunk->page_list[i].length = PAGE_SIZE; |
| 137 | } |
| 138 | |
| 139 | chunk->nmap = dma_map_sg(dev->dma_device, |
| 140 | &chunk->page_list[0], |
| 141 | chunk->nents, |
| 142 | DMA_BIDIRECTIONAL); |
| 143 | if (chunk->nmap <= 0) { |
| 144 | for (i = 0; i < chunk->nents; ++i) |
| 145 | put_page(chunk->page_list[i].page); |
| 146 | kfree(chunk); |
| 147 | |
| 148 | ret = -ENOMEM; |
| 149 | goto out; |
| 150 | } |
| 151 | |
| 152 | ret -= chunk->nents; |
| 153 | off += chunk->nents; |
| 154 | list_add_tail(&chunk->list, &mem->chunk_list); |
| 155 | } |
| 156 | |
| 157 | ret = 0; |
| 158 | } |
| 159 | |
| 160 | out: |
| 161 | if (ret < 0) |
| 162 | __ib_umem_release(dev, mem, 0); |
| 163 | else |
| 164 | current->mm->locked_vm = locked; |
| 165 | |
| 166 | up_write(¤t->mm->mmap_sem); |
| 167 | free_page((unsigned long) page_list); |
| 168 | |
| 169 | return ret; |
| 170 | } |
| 171 | |
| 172 | void ib_umem_release(struct ib_device *dev, struct ib_umem *umem) |
| 173 | { |
| 174 | __ib_umem_release(dev, umem, 1); |
| 175 | |
| 176 | down_write(¤t->mm->mmap_sem); |
| 177 | current->mm->locked_vm -= |
| 178 | PAGE_ALIGN(umem->length + umem->offset) >> PAGE_SHIFT; |
| 179 | up_write(¤t->mm->mmap_sem); |
| 180 | } |
| 181 | |
| 182 | static void ib_umem_account(void *work_ptr) |
| 183 | { |
| 184 | struct ib_umem_account_work *work = work_ptr; |
| 185 | |
| 186 | down_write(&work->mm->mmap_sem); |
| 187 | work->mm->locked_vm -= work->diff; |
| 188 | up_write(&work->mm->mmap_sem); |
| 189 | mmput(work->mm); |
| 190 | kfree(work); |
| 191 | } |
| 192 | |
| 193 | void ib_umem_release_on_close(struct ib_device *dev, struct ib_umem *umem) |
| 194 | { |
| 195 | struct ib_umem_account_work *work; |
| 196 | struct mm_struct *mm; |
| 197 | |
| 198 | __ib_umem_release(dev, umem, 1); |
| 199 | |
| 200 | mm = get_task_mm(current); |
| 201 | if (!mm) |
| 202 | return; |
| 203 | |
| 204 | /* |
| 205 | * We may be called with the mm's mmap_sem already held. This |
| 206 | * can happen when a userspace munmap() is the call that drops |
| 207 | * the last reference to our file and calls our release |
| 208 | * method. If there are memory regions to destroy, we'll end |
| 209 | * up here and not be able to take the mmap_sem. Therefore we |
| 210 | * defer the vm_locked accounting to the system workqueue. |
| 211 | */ |
| 212 | |
| 213 | work = kmalloc(sizeof *work, GFP_KERNEL); |
| 214 | if (!work) |
| 215 | return; |
| 216 | |
| 217 | INIT_WORK(&work->work, ib_umem_account, work); |
| 218 | work->mm = mm; |
| 219 | work->diff = PAGE_ALIGN(umem->length + umem->offset) >> PAGE_SHIFT; |
| 220 | |
| 221 | schedule_work(&work->work); |
| 222 | } |