Linux Audio

Check our new training course

Linux kernel drivers training

Mar 31-Apr 9, 2025, special US time zones
Register
Loading...
Note: File does not exist in v3.1.
  1// SPDX-License-Identifier: GPL-2.0 OR MIT
  2/**************************************************************************
  3 *
  4 * Copyright 2019 VMware, Inc., Palo Alto, CA., USA
  5 *
  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:
 13 *
 14 * The above copyright notice and this permission notice (including the
 15 * next paragraph) shall be included in all copies or substantial portions
 16 * of the Software.
 17 *
 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.
 25 *
 26 **************************************************************************/
 27#include "vmwgfx_drv.h"
 28
 29/*
 30 * Different methods for tracking dirty:
 31 * VMW_BO_DIRTY_PAGETABLE - Scan the pagetable for hardware dirty bits
 32 * VMW_BO_DIRTY_MKWRITE - Write-protect page table entries and record write-
 33 * accesses in the VM mkwrite() callback
 34 */
 35enum vmw_bo_dirty_method {
 36	VMW_BO_DIRTY_PAGETABLE,
 37	VMW_BO_DIRTY_MKWRITE,
 38};
 39
 40/*
 41 * No dirtied pages at scan trigger a transition to the _MKWRITE method,
 42 * similarly a certain percentage of dirty pages trigger a transition to
 43 * the _PAGETABLE method. How many triggers should we wait for before
 44 * changing method?
 45 */
 46#define VMW_DIRTY_NUM_CHANGE_TRIGGERS 2
 47
 48/* Percentage to trigger a transition to the _PAGETABLE method */
 49#define VMW_DIRTY_PERCENTAGE 10
 50
 51/**
 52 * struct vmw_bo_dirty - Dirty information for buffer objects
 53 * @start: First currently dirty bit
 54 * @end: Last currently dirty bit + 1
 55 * @method: The currently used dirty method
 56 * @change_count: Number of consecutive method change triggers
 57 * @ref_count: Reference count for this structure
 58 * @bitmap_size: The size of the bitmap in bits. Typically equal to the
 59 * nuber of pages in the bo.
 60 * @bitmap: A bitmap where each bit represents a page. A set bit means a
 61 * dirty page.
 62 */
 63struct vmw_bo_dirty {
 64	unsigned long start;
 65	unsigned long end;
 66	enum vmw_bo_dirty_method method;
 67	unsigned int change_count;
 68	unsigned int ref_count;
 69	unsigned long bitmap_size;
 70	unsigned long bitmap[];
 71};
 72
 73/**
 74 * vmw_bo_dirty_scan_pagetable - Perform a pagetable scan for dirty bits
 75 * @vbo: The buffer object to scan
 76 *
 77 * Scans the pagetable for dirty bits. Clear those bits and modify the
 78 * dirty structure with the results. This function may change the
 79 * dirty-tracking method.
 80 */
 81static void vmw_bo_dirty_scan_pagetable(struct vmw_buffer_object *vbo)
 82{
 83	struct vmw_bo_dirty *dirty = vbo->dirty;
 84	pgoff_t offset = drm_vma_node_start(&vbo->base.base.vma_node);
 85	struct address_space *mapping = vbo->base.bdev->dev_mapping;
 86	pgoff_t num_marked;
 87
 88	num_marked = clean_record_shared_mapping_range
 89		(mapping,
 90		 offset, dirty->bitmap_size,
 91		 offset, &dirty->bitmap[0],
 92		 &dirty->start, &dirty->end);
 93	if (num_marked == 0)
 94		dirty->change_count++;
 95	else
 96		dirty->change_count = 0;
 97
 98	if (dirty->change_count > VMW_DIRTY_NUM_CHANGE_TRIGGERS) {
 99		dirty->change_count = 0;
100		dirty->method = VMW_BO_DIRTY_MKWRITE;
101		wp_shared_mapping_range(mapping,
102					offset, dirty->bitmap_size);
103		clean_record_shared_mapping_range(mapping,
104						  offset, dirty->bitmap_size,
105						  offset, &dirty->bitmap[0],
106						  &dirty->start, &dirty->end);
107	}
108}
109
110/**
111 * vmw_bo_dirty_scan_mkwrite - Reset the mkwrite dirty-tracking method
112 * @vbo: The buffer object to scan
113 *
114 * Write-protect pages written to so that consecutive write accesses will
115 * trigger a call to mkwrite.
116 *
117 * This function may change the dirty-tracking method.
118 */
119static void vmw_bo_dirty_scan_mkwrite(struct vmw_buffer_object *vbo)
120{
121	struct vmw_bo_dirty *dirty = vbo->dirty;
122	unsigned long offset = drm_vma_node_start(&vbo->base.base.vma_node);
123	struct address_space *mapping = vbo->base.bdev->dev_mapping;
124	pgoff_t num_marked;
125
126	if (dirty->end <= dirty->start)
127		return;
128
129	num_marked = wp_shared_mapping_range(vbo->base.bdev->dev_mapping,
130					dirty->start + offset,
131					dirty->end - dirty->start);
132
133	if (100UL * num_marked / dirty->bitmap_size >
134	    VMW_DIRTY_PERCENTAGE) {
135		dirty->change_count++;
136	} else {
137		dirty->change_count = 0;
138	}
139
140	if (dirty->change_count > VMW_DIRTY_NUM_CHANGE_TRIGGERS) {
141		pgoff_t start = 0;
142		pgoff_t end = dirty->bitmap_size;
143
144		dirty->method = VMW_BO_DIRTY_PAGETABLE;
145		clean_record_shared_mapping_range(mapping, offset, end, offset,
146						  &dirty->bitmap[0],
147						  &start, &end);
148		bitmap_clear(&dirty->bitmap[0], 0, dirty->bitmap_size);
149		if (dirty->start < dirty->end)
150			bitmap_set(&dirty->bitmap[0], dirty->start,
151				   dirty->end - dirty->start);
152		dirty->change_count = 0;
153	}
154}
155
156/**
157 * vmw_bo_dirty_scan - Scan for dirty pages and add them to the dirty
158 * tracking structure
159 * @vbo: The buffer object to scan
160 *
161 * This function may change the dirty tracking method.
162 */
163void vmw_bo_dirty_scan(struct vmw_buffer_object *vbo)
164{
165	struct vmw_bo_dirty *dirty = vbo->dirty;
166
167	if (dirty->method == VMW_BO_DIRTY_PAGETABLE)
168		vmw_bo_dirty_scan_pagetable(vbo);
169	else
170		vmw_bo_dirty_scan_mkwrite(vbo);
171}
172
173/**
174 * vmw_bo_dirty_pre_unmap - write-protect and pick up dirty pages before
175 * an unmap_mapping_range operation.
176 * @vbo: The buffer object,
177 * @start: First page of the range within the buffer object.
178 * @end: Last page of the range within the buffer object + 1.
179 *
180 * If we're using the _PAGETABLE scan method, we may leak dirty pages
181 * when calling unmap_mapping_range(). This function makes sure we pick
182 * up all dirty pages.
183 */
184static void vmw_bo_dirty_pre_unmap(struct vmw_buffer_object *vbo,
185				   pgoff_t start, pgoff_t end)
186{
187	struct vmw_bo_dirty *dirty = vbo->dirty;
188	unsigned long offset = drm_vma_node_start(&vbo->base.base.vma_node);
189	struct address_space *mapping = vbo->base.bdev->dev_mapping;
190
191	if (dirty->method != VMW_BO_DIRTY_PAGETABLE || start >= end)
192		return;
193
194	wp_shared_mapping_range(mapping, start + offset, end - start);
195	clean_record_shared_mapping_range(mapping, start + offset,
196					  end - start, offset,
197					  &dirty->bitmap[0], &dirty->start,
198					  &dirty->end);
199}
200
201/**
202 * vmw_bo_dirty_unmap - Clear all ptes pointing to a range within a bo
203 * @vbo: The buffer object,
204 * @start: First page of the range within the buffer object.
205 * @end: Last page of the range within the buffer object + 1.
206 *
207 * This is similar to ttm_bo_unmap_virtual() except it takes a subrange.
208 */
209void vmw_bo_dirty_unmap(struct vmw_buffer_object *vbo,
210			pgoff_t start, pgoff_t end)
211{
212	unsigned long offset = drm_vma_node_start(&vbo->base.base.vma_node);
213	struct address_space *mapping = vbo->base.bdev->dev_mapping;
214
215	vmw_bo_dirty_pre_unmap(vbo, start, end);
216	unmap_shared_mapping_range(mapping, (offset + start) << PAGE_SHIFT,
217				   (loff_t) (end - start) << PAGE_SHIFT);
218}
219
220/**
221 * vmw_bo_dirty_add - Add a dirty-tracking user to a buffer object
222 * @vbo: The buffer object
223 *
224 * This function registers a dirty-tracking user to a buffer object.
225 * A user can be for example a resource or a vma in a special user-space
226 * mapping.
227 *
228 * Return: Zero on success, -ENOMEM on memory allocation failure.
229 */
230int vmw_bo_dirty_add(struct vmw_buffer_object *vbo)
231{
232	struct vmw_bo_dirty *dirty = vbo->dirty;
233	pgoff_t num_pages = PFN_UP(vbo->base.resource->size);
234	size_t size;
235	int ret;
236
237	if (dirty) {
238		dirty->ref_count++;
239		return 0;
240	}
241
242	size = sizeof(*dirty) + BITS_TO_LONGS(num_pages) * sizeof(long);
243	dirty = kvzalloc(size, GFP_KERNEL);
244	if (!dirty) {
245		ret = -ENOMEM;
246		goto out_no_dirty;
247	}
248
249	dirty->bitmap_size = num_pages;
250	dirty->start = dirty->bitmap_size;
251	dirty->end = 0;
252	dirty->ref_count = 1;
253	if (num_pages < PAGE_SIZE / sizeof(pte_t)) {
254		dirty->method = VMW_BO_DIRTY_PAGETABLE;
255	} else {
256		struct address_space *mapping = vbo->base.bdev->dev_mapping;
257		pgoff_t offset = drm_vma_node_start(&vbo->base.base.vma_node);
258
259		dirty->method = VMW_BO_DIRTY_MKWRITE;
260
261		/* Write-protect and then pick up already dirty bits */
262		wp_shared_mapping_range(mapping, offset, num_pages);
263		clean_record_shared_mapping_range(mapping, offset, num_pages,
264						  offset,
265						  &dirty->bitmap[0],
266						  &dirty->start, &dirty->end);
267	}
268
269	vbo->dirty = dirty;
270
271	return 0;
272
273out_no_dirty:
274	return ret;
275}
276
277/**
278 * vmw_bo_dirty_release - Release a dirty-tracking user from a buffer object
279 * @vbo: The buffer object
280 *
281 * This function releases a dirty-tracking user from a buffer object.
282 * If the reference count reaches zero, then the dirty-tracking object is
283 * freed and the pointer to it cleared.
284 *
285 * Return: Zero on success, -ENOMEM on memory allocation failure.
286 */
287void vmw_bo_dirty_release(struct vmw_buffer_object *vbo)
288{
289	struct vmw_bo_dirty *dirty = vbo->dirty;
290
291	if (dirty && --dirty->ref_count == 0) {
292		kvfree(dirty);
293		vbo->dirty = NULL;
294	}
295}
296
297/**
298 * vmw_bo_dirty_transfer_to_res - Pick up a resource's dirty region from
299 * its backing mob.
300 * @res: The resource
301 *
302 * This function will pick up all dirty ranges affecting the resource from
303 * it's backup mob, and call vmw_resource_dirty_update() once for each
304 * range. The transferred ranges will be cleared from the backing mob's
305 * dirty tracking.
306 */
307void vmw_bo_dirty_transfer_to_res(struct vmw_resource *res)
308{
309	struct vmw_buffer_object *vbo = res->backup;
310	struct vmw_bo_dirty *dirty = vbo->dirty;
311	pgoff_t start, cur, end;
312	unsigned long res_start = res->backup_offset;
313	unsigned long res_end = res->backup_offset + res->backup_size;
314
315	WARN_ON_ONCE(res_start & ~PAGE_MASK);
316	res_start >>= PAGE_SHIFT;
317	res_end = DIV_ROUND_UP(res_end, PAGE_SIZE);
318
319	if (res_start >= dirty->end || res_end <= dirty->start)
320		return;
321
322	cur = max(res_start, dirty->start);
323	res_end = max(res_end, dirty->end);
324	while (cur < res_end) {
325		unsigned long num;
326
327		start = find_next_bit(&dirty->bitmap[0], res_end, cur);
328		if (start >= res_end)
329			break;
330
331		end = find_next_zero_bit(&dirty->bitmap[0], res_end, start + 1);
332		cur = end + 1;
333		num = end - start;
334		bitmap_clear(&dirty->bitmap[0], start, num);
335		vmw_resource_dirty_update(res, start, end);
336	}
337
338	if (res_start <= dirty->start && res_end > dirty->start)
339		dirty->start = res_end;
340	if (res_start < dirty->end && res_end >= dirty->end)
341		dirty->end = res_start;
342}
343
344/**
345 * vmw_bo_dirty_clear_res - Clear a resource's dirty region from
346 * its backing mob.
347 * @res: The resource
348 *
349 * This function will clear all dirty ranges affecting the resource from
350 * it's backup mob's dirty tracking.
351 */
352void vmw_bo_dirty_clear_res(struct vmw_resource *res)
353{
354	unsigned long res_start = res->backup_offset;
355	unsigned long res_end = res->backup_offset + res->backup_size;
356	struct vmw_buffer_object *vbo = res->backup;
357	struct vmw_bo_dirty *dirty = vbo->dirty;
358
359	res_start >>= PAGE_SHIFT;
360	res_end = DIV_ROUND_UP(res_end, PAGE_SIZE);
361
362	if (res_start >= dirty->end || res_end <= dirty->start)
363		return;
364
365	res_start = max(res_start, dirty->start);
366	res_end = min(res_end, dirty->end);
367	bitmap_clear(&dirty->bitmap[0], res_start, res_end - res_start);
368
369	if (res_start <= dirty->start && res_end > dirty->start)
370		dirty->start = res_end;
371	if (res_start < dirty->end && res_end >= dirty->end)
372		dirty->end = res_start;
373}
374
375vm_fault_t vmw_bo_vm_mkwrite(struct vm_fault *vmf)
376{
377	struct vm_area_struct *vma = vmf->vma;
378	struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
379	    vma->vm_private_data;
380	vm_fault_t ret;
381	unsigned long page_offset;
382	unsigned int save_flags;
383	struct vmw_buffer_object *vbo =
384		container_of(bo, typeof(*vbo), base);
385
386	/*
387	 * mkwrite() doesn't handle the VM_FAULT_RETRY return value correctly.
388	 * So make sure the TTM helpers are aware.
389	 */
390	save_flags = vmf->flags;
391	vmf->flags &= ~FAULT_FLAG_ALLOW_RETRY;
392	ret = ttm_bo_vm_reserve(bo, vmf);
393	vmf->flags = save_flags;
394	if (ret)
395		return ret;
396
397	page_offset = vmf->pgoff - drm_vma_node_start(&bo->base.vma_node);
398	if (unlikely(page_offset >= PFN_UP(bo->resource->size))) {
399		ret = VM_FAULT_SIGBUS;
400		goto out_unlock;
401	}
402
403	if (vbo->dirty && vbo->dirty->method == VMW_BO_DIRTY_MKWRITE &&
404	    !test_bit(page_offset, &vbo->dirty->bitmap[0])) {
405		struct vmw_bo_dirty *dirty = vbo->dirty;
406
407		__set_bit(page_offset, &dirty->bitmap[0]);
408		dirty->start = min(dirty->start, page_offset);
409		dirty->end = max(dirty->end, page_offset + 1);
410	}
411
412out_unlock:
413	dma_resv_unlock(bo->base.resv);
414	return ret;
415}
416
417vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf)
418{
419	struct vm_area_struct *vma = vmf->vma;
420	struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
421	    vma->vm_private_data;
422	struct vmw_buffer_object *vbo =
423		container_of(bo, struct vmw_buffer_object, base);
424	pgoff_t num_prefault;
425	pgprot_t prot;
426	vm_fault_t ret;
427
428	ret = ttm_bo_vm_reserve(bo, vmf);
429	if (ret)
430		return ret;
431
432	num_prefault = (vma->vm_flags & VM_RAND_READ) ? 1 :
433		TTM_BO_VM_NUM_PREFAULT;
434
435	if (vbo->dirty) {
436		pgoff_t allowed_prefault;
437		unsigned long page_offset;
438
439		page_offset = vmf->pgoff -
440			drm_vma_node_start(&bo->base.vma_node);
441		if (page_offset >= PFN_UP(bo->resource->size) ||
442		    vmw_resources_clean(vbo, page_offset,
443					page_offset + PAGE_SIZE,
444					&allowed_prefault)) {
445			ret = VM_FAULT_SIGBUS;
446			goto out_unlock;
447		}
448
449		num_prefault = min(num_prefault, allowed_prefault);
450	}
451
452	/*
453	 * If we don't track dirty using the MKWRITE method, make sure
454	 * sure the page protection is write-enabled so we don't get
455	 * a lot of unnecessary write faults.
456	 */
457	if (vbo->dirty && vbo->dirty->method == VMW_BO_DIRTY_MKWRITE)
458		prot = vm_get_page_prot(vma->vm_flags & ~VM_SHARED);
459	else
460		prot = vm_get_page_prot(vma->vm_flags);
461
462	ret = ttm_bo_vm_fault_reserved(vmf, prot, num_prefault);
463	if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
464		return ret;
465
466out_unlock:
467	dma_resv_unlock(bo->base.resv);
468
469	return ret;
470}