1 /**************************************************************************
3 * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 * USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
28 * Authors: Thomas Hellstrom <thellstrom-at-vmware-dot-com>
31 #define pr_fmt(fmt) "[TTM] " fmt
33 #include <ttm/ttm_module.h>
34 #include <ttm/ttm_bo_driver.h>
35 #include <ttm/ttm_placement.h>
37 #include <linux/rbtree.h>
38 #include <linux/module.h>
39 #include <linux/uaccess.h>
41 #define TTM_BO_VM_NUM_PREFAULT 16
43 static struct ttm_buffer_object *ttm_bo_vm_lookup_rb(struct ttm_bo_device *bdev,
44 unsigned long page_start,
45 unsigned long num_pages)
47 struct rb_node *cur = bdev->addr_space_rb.rb_node;
48 unsigned long cur_offset;
49 struct ttm_buffer_object *bo;
50 struct ttm_buffer_object *best_bo = NULL;
52 while (likely(cur != NULL)) {
53 bo = rb_entry(cur, struct ttm_buffer_object, vm_rb);
54 cur_offset = bo->vm_node->start;
55 if (page_start >= cur_offset) {
58 if (page_start == cur_offset)
64 if (unlikely(best_bo == NULL))
67 if (unlikely((best_bo->vm_node->start + best_bo->num_pages) <
68 (page_start + num_pages)))
74 static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
76 struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
78 struct ttm_bo_device *bdev = bo->bdev;
79 unsigned long page_offset;
80 unsigned long page_last;
82 struct ttm_tt *ttm = NULL;
86 unsigned long address = (unsigned long)vmf->virtual_address;
87 int retval = VM_FAULT_NOPAGE;
88 struct ttm_mem_type_manager *man =
89 &bdev->man[bo->mem.mem_type];
92 * Work around locking order reversal in fault / nopfn
93 * between mmap_sem and bo_reserve: Perform a trylock operation
94 * for reserve, and if it fails, retry the fault after scheduling.
97 ret = ttm_bo_reserve(bo, true, true, false, 0);
98 if (unlikely(ret != 0)) {
101 return VM_FAULT_NOPAGE;
104 if (bdev->driver->fault_reserve_notify) {
105 ret = bdev->driver->fault_reserve_notify(bo);
112 retval = VM_FAULT_NOPAGE;
115 retval = VM_FAULT_SIGBUS;
121 * Wait for buffer data in transit, due to a pipelined
125 spin_lock(&bdev->fence_lock);
126 if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) {
127 ret = ttm_bo_wait(bo, false, true, false);
128 spin_unlock(&bdev->fence_lock);
129 if (unlikely(ret != 0)) {
130 retval = (ret != -ERESTARTSYS) ?
131 VM_FAULT_SIGBUS : VM_FAULT_NOPAGE;
135 spin_unlock(&bdev->fence_lock);
137 ret = ttm_mem_io_lock(man, true);
138 if (unlikely(ret != 0)) {
139 retval = VM_FAULT_NOPAGE;
142 ret = ttm_mem_io_reserve_vm(bo);
143 if (unlikely(ret != 0)) {
144 retval = VM_FAULT_SIGBUS;
148 page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) +
149 bo->vm_node->start - vma->vm_pgoff;
150 page_last = ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) +
151 bo->vm_node->start - vma->vm_pgoff;
153 if (unlikely(page_offset >= bo->num_pages)) {
154 retval = VM_FAULT_SIGBUS;
159 * Strictly, we're not allowed to modify vma->vm_page_prot here,
160 * since the mmap_sem is only held in read mode. However, we
161 * modify only the caching bits of vma->vm_page_prot and
162 * consider those bits protected by
163 * the bo->mutex, as we should be the only writers.
164 * There shouldn't really be any readers of these bits except
165 * within vm_insert_mixed()? fork?
167 * TODO: Add a list of vmas to the bo, and change the
168 * vma->vm_page_prot when the object changes caching policy, with
169 * the correct locks held.
171 if (bo->mem.bus.is_iomem) {
172 vma->vm_page_prot = ttm_io_prot(bo->mem.placement,
174 #if defined(CONFIG_XEN) && defined(_PAGE_IOMAP)
175 pgprot_val(vma->vm_page_prot) |= _PAGE_IOMAP;
178 #if defined(CONFIG_XEN) && defined(_PAGE_IOMAP)
179 pgprot_val(vma->vm_page_prot) &= ~_PAGE_IOMAP;
182 vma->vm_page_prot = (bo->mem.placement & TTM_PL_FLAG_CACHED) ?
183 vm_get_page_prot(vma->vm_flags) :
184 ttm_io_prot(bo->mem.placement, vma->vm_page_prot);
186 /* Allocate all page at once, most common usage */
187 if (ttm->bdev->driver->ttm_tt_populate(ttm)) {
188 retval = VM_FAULT_OOM;
194 * Speculatively prefault a number of pages. Only error on
197 for (i = 0; i < TTM_BO_VM_NUM_PREFAULT; ++i) {
198 if (bo->mem.bus.is_iomem)
199 pfn = ((bo->mem.bus.base + bo->mem.bus.offset) >> PAGE_SHIFT) + page_offset;
201 page = ttm->pages[page_offset];
202 if (unlikely(!page && i == 0)) {
203 retval = VM_FAULT_OOM;
205 } else if (unlikely(!page)) {
208 pfn = page_to_pfn(page);
211 ret = vm_insert_mixed(vma, address, pfn);
213 * Somebody beat us to this PTE or prefaulting to
214 * an already populated PTE, or prefaulting error.
217 if (unlikely((ret == -EBUSY) || (ret != 0 && i > 0)))
219 else if (unlikely(ret != 0)) {
221 (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
225 address += PAGE_SIZE;
226 if (unlikely(++page_offset >= page_last))
230 ttm_mem_io_unlock(man);
232 ttm_bo_unreserve(bo);
236 static void ttm_bo_vm_open(struct vm_area_struct *vma)
238 struct ttm_buffer_object *bo =
239 (struct ttm_buffer_object *)vma->vm_private_data;
241 (void)ttm_bo_reference(bo);
244 static void ttm_bo_vm_close(struct vm_area_struct *vma)
246 struct ttm_buffer_object *bo = (struct ttm_buffer_object *)vma->vm_private_data;
249 vma->vm_private_data = NULL;
252 static const struct vm_operations_struct ttm_bo_vm_ops = {
253 .fault = ttm_bo_vm_fault,
254 .open = ttm_bo_vm_open,
255 .close = ttm_bo_vm_close
258 int ttm_bo_mmap(struct file *filp, struct vm_area_struct *vma,
259 struct ttm_bo_device *bdev)
261 struct ttm_bo_driver *driver;
262 struct ttm_buffer_object *bo;
265 read_lock(&bdev->vm_lock);
266 bo = ttm_bo_vm_lookup_rb(bdev, vma->vm_pgoff,
267 (vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
268 if (likely(bo != NULL))
269 ttm_bo_reference(bo);
270 read_unlock(&bdev->vm_lock);
272 if (unlikely(bo == NULL)) {
273 pr_err("Could not find buffer object to map\n");
277 driver = bo->bdev->driver;
278 if (unlikely(!driver->verify_access)) {
282 ret = driver->verify_access(bo, filp);
283 if (unlikely(ret != 0))
286 vma->vm_ops = &ttm_bo_vm_ops;
289 * Note: We're transferring the bo reference to
290 * vma->vm_private_data here.
293 vma->vm_private_data = bo;
294 vma->vm_flags |= VM_RESERVED | VM_IO | VM_MIXEDMAP | VM_DONTEXPAND;
300 EXPORT_SYMBOL(ttm_bo_mmap);
302 int ttm_fbdev_mmap(struct vm_area_struct *vma, struct ttm_buffer_object *bo)
304 if (vma->vm_pgoff != 0)
307 vma->vm_ops = &ttm_bo_vm_ops;
308 vma->vm_private_data = ttm_bo_reference(bo);
309 vma->vm_flags |= VM_RESERVED | VM_IO | VM_MIXEDMAP | VM_DONTEXPAND;
312 EXPORT_SYMBOL(ttm_fbdev_mmap);
315 ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
316 const char __user *wbuf, char __user *rbuf, size_t count,
317 loff_t *f_pos, bool write)
319 struct ttm_buffer_object *bo;
320 struct ttm_bo_driver *driver;
321 struct ttm_bo_kmap_obj map;
322 unsigned long dev_offset = (*f_pos >> PAGE_SHIFT);
323 unsigned long kmap_offset;
324 unsigned long kmap_end;
325 unsigned long kmap_num;
327 unsigned int page_offset;
330 bool no_wait = false;
333 read_lock(&bdev->vm_lock);
334 bo = ttm_bo_vm_lookup_rb(bdev, dev_offset, 1);
335 if (likely(bo != NULL))
336 ttm_bo_reference(bo);
337 read_unlock(&bdev->vm_lock);
339 if (unlikely(bo == NULL))
342 driver = bo->bdev->driver;
343 if (unlikely(!driver->verify_access)) {
348 ret = driver->verify_access(bo, filp);
349 if (unlikely(ret != 0))
352 kmap_offset = dev_offset - bo->vm_node->start;
353 if (unlikely(kmap_offset >= bo->num_pages)) {
358 page_offset = *f_pos & ~PAGE_MASK;
359 io_size = bo->num_pages - kmap_offset;
360 io_size = (io_size << PAGE_SHIFT) - page_offset;
364 kmap_end = (*f_pos + count - 1) >> PAGE_SHIFT;
365 kmap_num = kmap_end - kmap_offset + 1;
367 ret = ttm_bo_reserve(bo, true, no_wait, false, 0);
379 ret = ttm_bo_kmap(bo, kmap_offset, kmap_num, &map);
380 if (unlikely(ret != 0)) {
381 ttm_bo_unreserve(bo);
385 virtual = ttm_kmap_obj_virtual(&map, &dummy);
386 virtual += page_offset;
389 ret = copy_from_user(virtual, wbuf, io_size);
391 ret = copy_to_user(rbuf, virtual, io_size);
394 ttm_bo_unreserve(bo);
397 if (unlikely(ret != 0))
408 ssize_t ttm_bo_fbdev_io(struct ttm_buffer_object *bo, const char __user *wbuf,
409 char __user *rbuf, size_t count, loff_t *f_pos,
412 struct ttm_bo_kmap_obj map;
413 unsigned long kmap_offset;
414 unsigned long kmap_end;
415 unsigned long kmap_num;
417 unsigned int page_offset;
420 bool no_wait = false;
423 kmap_offset = (*f_pos >> PAGE_SHIFT);
424 if (unlikely(kmap_offset >= bo->num_pages))
427 page_offset = *f_pos & ~PAGE_MASK;
428 io_size = bo->num_pages - kmap_offset;
429 io_size = (io_size << PAGE_SHIFT) - page_offset;
433 kmap_end = (*f_pos + count - 1) >> PAGE_SHIFT;
434 kmap_num = kmap_end - kmap_offset + 1;
436 ret = ttm_bo_reserve(bo, true, no_wait, false, 0);
447 ret = ttm_bo_kmap(bo, kmap_offset, kmap_num, &map);
448 if (unlikely(ret != 0)) {
449 ttm_bo_unreserve(bo);
453 virtual = ttm_kmap_obj_virtual(&map, &dummy);
454 virtual += page_offset;
457 ret = copy_from_user(virtual, wbuf, io_size);
459 ret = copy_to_user(rbuf, virtual, io_size);
462 ttm_bo_unreserve(bo);
465 if (unlikely(ret != 0))