2 * Copyright 2016 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
22 * Authors: Christian König
25 #include <linux/dma-mapping.h>
27 #include "amdgpu_vm.h"
28 #include "amdgpu_atomfirmware.h"
31 static inline struct amdgpu_vram_mgr *to_vram_mgr(struct ttm_resource_manager *man)
33 return container_of(man, struct amdgpu_vram_mgr, manager);
36 static inline struct amdgpu_device *to_amdgpu_device(struct amdgpu_vram_mgr *mgr)
38 return container_of(mgr, struct amdgpu_device, mman.vram_mgr);
42 * DOC: mem_info_vram_total
44 * The amdgpu driver provides a sysfs API for reporting current total VRAM
45 * available on the device
46 * The file mem_info_vram_total is used for this and returns the total
47 * amount of VRAM in bytes
49 static ssize_t amdgpu_mem_info_vram_total_show(struct device *dev,
50 struct device_attribute *attr, char *buf)
52 struct drm_device *ddev = dev_get_drvdata(dev);
53 struct amdgpu_device *adev = drm_to_adev(ddev);
55 return sysfs_emit(buf, "%llu\n", adev->gmc.real_vram_size);
59 * DOC: mem_info_vis_vram_total
61 * The amdgpu driver provides a sysfs API for reporting current total
62 * visible VRAM available on the device
63 * The file mem_info_vis_vram_total is used for this and returns the total
64 * amount of visible VRAM in bytes
66 static ssize_t amdgpu_mem_info_vis_vram_total_show(struct device *dev,
67 struct device_attribute *attr, char *buf)
69 struct drm_device *ddev = dev_get_drvdata(dev);
70 struct amdgpu_device *adev = drm_to_adev(ddev);
72 return sysfs_emit(buf, "%llu\n", adev->gmc.visible_vram_size);
76 * DOC: mem_info_vram_used
78 * The amdgpu driver provides a sysfs API for reporting current total VRAM
79 * available on the device
80 * The file mem_info_vram_used is used for this and returns the total
81 * amount of currently used VRAM in bytes
83 static ssize_t amdgpu_mem_info_vram_used_show(struct device *dev,
84 struct device_attribute *attr, char *buf)
86 struct drm_device *ddev = dev_get_drvdata(dev);
87 struct amdgpu_device *adev = drm_to_adev(ddev);
88 struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
90 return sysfs_emit(buf, "%llu\n", amdgpu_vram_mgr_usage(man));
94 * DOC: mem_info_vis_vram_used
96 * The amdgpu driver provides a sysfs API for reporting current total of
98 * The file mem_info_vis_vram_used is used for this and returns the total
99 * amount of currently used visible VRAM in bytes
101 static ssize_t amdgpu_mem_info_vis_vram_used_show(struct device *dev,
102 struct device_attribute *attr, char *buf)
104 struct drm_device *ddev = dev_get_drvdata(dev);
105 struct amdgpu_device *adev = drm_to_adev(ddev);
106 struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
108 return sysfs_emit(buf, "%llu\n", amdgpu_vram_mgr_vis_usage(man));
111 static ssize_t amdgpu_mem_info_vram_vendor(struct device *dev,
112 struct device_attribute *attr,
115 struct drm_device *ddev = dev_get_drvdata(dev);
116 struct amdgpu_device *adev = drm_to_adev(ddev);
118 switch (adev->gmc.vram_vendor) {
120 return sysfs_emit(buf, "samsung\n");
122 return sysfs_emit(buf, "infineon\n");
124 return sysfs_emit(buf, "elpida\n");
126 return sysfs_emit(buf, "etron\n");
128 return sysfs_emit(buf, "nanya\n");
130 return sysfs_emit(buf, "hynix\n");
132 return sysfs_emit(buf, "mosel\n");
134 return sysfs_emit(buf, "winbond\n");
136 return sysfs_emit(buf, "esmt\n");
138 return sysfs_emit(buf, "micron\n");
140 return sysfs_emit(buf, "unknown\n");
144 static DEVICE_ATTR(mem_info_vram_total, S_IRUGO,
145 amdgpu_mem_info_vram_total_show, NULL);
146 static DEVICE_ATTR(mem_info_vis_vram_total, S_IRUGO,
147 amdgpu_mem_info_vis_vram_total_show,NULL);
148 static DEVICE_ATTR(mem_info_vram_used, S_IRUGO,
149 amdgpu_mem_info_vram_used_show, NULL);
150 static DEVICE_ATTR(mem_info_vis_vram_used, S_IRUGO,
151 amdgpu_mem_info_vis_vram_used_show, NULL);
152 static DEVICE_ATTR(mem_info_vram_vendor, S_IRUGO,
153 amdgpu_mem_info_vram_vendor, NULL);
155 static const struct attribute *amdgpu_vram_mgr_attributes[] = {
156 &dev_attr_mem_info_vram_total.attr,
157 &dev_attr_mem_info_vis_vram_total.attr,
158 &dev_attr_mem_info_vram_used.attr,
159 &dev_attr_mem_info_vis_vram_used.attr,
160 &dev_attr_mem_info_vram_vendor.attr,
164 static const struct ttm_resource_manager_func amdgpu_vram_mgr_func;
167 * amdgpu_vram_mgr_init - init VRAM manager and DRM MM
169 * @adev: amdgpu_device pointer
171 * Allocate and initialize the VRAM manager.
173 int amdgpu_vram_mgr_init(struct amdgpu_device *adev)
175 struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr;
176 struct ttm_resource_manager *man = &mgr->manager;
179 ttm_resource_manager_init(man, adev->gmc.real_vram_size >> PAGE_SHIFT);
181 man->func = &amdgpu_vram_mgr_func;
183 drm_mm_init(&mgr->mm, 0, man->size);
184 spin_lock_init(&mgr->lock);
185 INIT_LIST_HEAD(&mgr->reservations_pending);
186 INIT_LIST_HEAD(&mgr->reserved_pages);
188 /* Add the two VRAM-related sysfs files */
189 ret = sysfs_create_files(&adev->dev->kobj, amdgpu_vram_mgr_attributes);
191 DRM_ERROR("Failed to register sysfs\n");
193 ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM, &mgr->manager);
194 ttm_resource_manager_set_used(man, true);
199 * amdgpu_vram_mgr_fini - free and destroy VRAM manager
201 * @adev: amdgpu_device pointer
203 * Destroy and free the VRAM manager, returns -EBUSY if ranges are still
204 * allocated inside it.
206 void amdgpu_vram_mgr_fini(struct amdgpu_device *adev)
208 struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr;
209 struct ttm_resource_manager *man = &mgr->manager;
211 struct amdgpu_vram_reservation *rsv, *temp;
213 ttm_resource_manager_set_used(man, false);
215 ret = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
219 spin_lock(&mgr->lock);
220 list_for_each_entry_safe(rsv, temp, &mgr->reservations_pending, node)
223 list_for_each_entry_safe(rsv, temp, &mgr->reserved_pages, node) {
224 drm_mm_remove_node(&rsv->mm_node);
227 drm_mm_takedown(&mgr->mm);
228 spin_unlock(&mgr->lock);
230 sysfs_remove_files(&adev->dev->kobj, amdgpu_vram_mgr_attributes);
232 ttm_resource_manager_cleanup(man);
233 ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_VRAM, NULL);
237 * amdgpu_vram_mgr_vis_size - Calculate visible node size
239 * @adev: amdgpu_device pointer
240 * @node: MM node structure
242 * Calculate how many bytes of the MM node are inside visible VRAM
244 static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev,
245 struct drm_mm_node *node)
247 uint64_t start = node->start << PAGE_SHIFT;
248 uint64_t end = (node->size + node->start) << PAGE_SHIFT;
250 if (start >= adev->gmc.visible_vram_size)
253 return (end > adev->gmc.visible_vram_size ?
254 adev->gmc.visible_vram_size : end) - start;
258 * amdgpu_vram_mgr_bo_visible_size - CPU visible BO size
260 * @bo: &amdgpu_bo buffer object (must be in VRAM)
263 * How much of the given &amdgpu_bo buffer object lies in CPU visible VRAM.
265 u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo)
267 struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
268 struct ttm_resource *mem = &bo->tbo.mem;
269 struct drm_mm_node *nodes = mem->mm_node;
270 unsigned pages = mem->num_pages;
273 if (amdgpu_gmc_vram_full_visible(&adev->gmc))
274 return amdgpu_bo_size(bo);
276 if (mem->start >= adev->gmc.visible_vram_size >> PAGE_SHIFT)
279 for (usage = 0; nodes && pages; pages -= nodes->size, nodes++)
280 usage += amdgpu_vram_mgr_vis_size(adev, nodes);
285 static void amdgpu_vram_mgr_do_reserve(struct ttm_resource_manager *man)
287 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
288 struct amdgpu_device *adev = to_amdgpu_device(mgr);
289 struct drm_mm *mm = &mgr->mm;
290 struct amdgpu_vram_reservation *rsv, *temp;
293 list_for_each_entry_safe(rsv, temp, &mgr->reservations_pending, node) {
294 if (drm_mm_reserve_node(mm, &rsv->mm_node))
297 dev_dbg(adev->dev, "Reservation 0x%llx - %lld, Succeeded\n",
298 rsv->mm_node.start, rsv->mm_node.size);
300 vis_usage = amdgpu_vram_mgr_vis_size(adev, &rsv->mm_node);
301 atomic64_add(vis_usage, &mgr->vis_usage);
302 atomic64_add(rsv->mm_node.size << PAGE_SHIFT, &mgr->usage);
303 list_move(&rsv->node, &mgr->reserved_pages);
308 * amdgpu_vram_mgr_reserve_range - Reserve a range from VRAM
310 * @man: TTM memory type manager
311 * @start: start address of the range in VRAM
312 * @size: size of the range
314 * Reserve memory from start addess with the specified size in VRAM
316 int amdgpu_vram_mgr_reserve_range(struct ttm_resource_manager *man,
317 uint64_t start, uint64_t size)
319 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
320 struct amdgpu_vram_reservation *rsv;
322 rsv = kzalloc(sizeof(*rsv), GFP_KERNEL);
326 INIT_LIST_HEAD(&rsv->node);
327 rsv->mm_node.start = start >> PAGE_SHIFT;
328 rsv->mm_node.size = size >> PAGE_SHIFT;
330 spin_lock(&mgr->lock);
331 list_add_tail(&mgr->reservations_pending, &rsv->node);
332 amdgpu_vram_mgr_do_reserve(man);
333 spin_unlock(&mgr->lock);
339 * amdgpu_vram_mgr_query_page_status - query the reservation status
341 * @man: TTM memory type manager
342 * @start: start address of a page in VRAM
345 * -EBUSY: the page is still hold and in pending list
346 * 0: the page has been reserved
347 * -ENOENT: the input page is not a reservation
349 int amdgpu_vram_mgr_query_page_status(struct ttm_resource_manager *man,
352 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
353 struct amdgpu_vram_reservation *rsv;
356 spin_lock(&mgr->lock);
358 list_for_each_entry(rsv, &mgr->reservations_pending, node) {
359 if ((rsv->mm_node.start <= start) &&
360 (start < (rsv->mm_node.start + rsv->mm_node.size))) {
366 list_for_each_entry(rsv, &mgr->reserved_pages, node) {
367 if ((rsv->mm_node.start <= start) &&
368 (start < (rsv->mm_node.start + rsv->mm_node.size))) {
376 spin_unlock(&mgr->lock);
381 * amdgpu_vram_mgr_virt_start - update virtual start address
383 * @mem: ttm_resource to update
384 * @node: just allocated node
386 * Calculate a virtual BO start address to easily check if everything is CPU
389 static void amdgpu_vram_mgr_virt_start(struct ttm_resource *mem,
390 struct drm_mm_node *node)
394 start = node->start + node->size;
395 if (start > mem->num_pages)
396 start -= mem->num_pages;
399 mem->start = max(mem->start, start);
403 * amdgpu_vram_mgr_new - allocate new ranges
405 * @man: TTM memory type manager
406 * @tbo: TTM BO we need this range for
407 * @place: placement flags and restrictions
408 * @mem: the resulting mem object
410 * Allocate VRAM for the given BO.
412 static int amdgpu_vram_mgr_new(struct ttm_resource_manager *man,
413 struct ttm_buffer_object *tbo,
414 const struct ttm_place *place,
415 struct ttm_resource *mem)
417 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
418 struct amdgpu_device *adev = to_amdgpu_device(mgr);
419 struct drm_mm *mm = &mgr->mm;
420 struct drm_mm_node *nodes;
421 enum drm_mm_insert_mode mode;
422 unsigned long lpfn, num_nodes, pages_per_node, pages_left;
423 uint64_t vis_usage = 0, mem_bytes, max_bytes;
431 max_bytes = adev->gmc.mc_vram_size;
432 if (tbo->type != ttm_bo_type_kernel)
433 max_bytes -= AMDGPU_VM_RESERVED_VRAM;
435 /* bail out quickly if there's likely not enough VRAM for this BO */
436 mem_bytes = (u64)mem->num_pages << PAGE_SHIFT;
437 if (atomic64_add_return(mem_bytes, &mgr->usage) > max_bytes) {
438 atomic64_sub(mem_bytes, &mgr->usage);
442 if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
443 pages_per_node = ~0ul;
446 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
447 pages_per_node = HPAGE_PMD_NR;
450 pages_per_node = (2UL << (20UL - PAGE_SHIFT));
452 pages_per_node = max((uint32_t)pages_per_node, mem->page_alignment);
453 num_nodes = DIV_ROUND_UP(mem->num_pages, pages_per_node);
456 nodes = kvmalloc_array((uint32_t)num_nodes, sizeof(*nodes),
457 GFP_KERNEL | __GFP_ZERO);
459 atomic64_sub(mem_bytes, &mgr->usage);
463 mode = DRM_MM_INSERT_BEST;
464 if (place->flags & TTM_PL_FLAG_TOPDOWN)
465 mode = DRM_MM_INSERT_HIGH;
468 pages_left = mem->num_pages;
470 spin_lock(&mgr->lock);
471 for (i = 0; pages_left >= pages_per_node; ++i) {
472 unsigned long pages = rounddown_pow_of_two(pages_left);
474 /* Limit maximum size to 2GB due to SG table limitations */
475 pages = min(pages, (2UL << (30 - PAGE_SHIFT)));
477 r = drm_mm_insert_node_in_range(mm, &nodes[i], pages,
484 vis_usage += amdgpu_vram_mgr_vis_size(adev, &nodes[i]);
485 amdgpu_vram_mgr_virt_start(mem, &nodes[i]);
489 for (; pages_left; ++i) {
490 unsigned long pages = min(pages_left, pages_per_node);
491 uint32_t alignment = mem->page_alignment;
493 if (pages == pages_per_node)
494 alignment = pages_per_node;
496 r = drm_mm_insert_node_in_range(mm, &nodes[i],
503 vis_usage += amdgpu_vram_mgr_vis_size(adev, &nodes[i]);
504 amdgpu_vram_mgr_virt_start(mem, &nodes[i]);
507 spin_unlock(&mgr->lock);
509 atomic64_add(vis_usage, &mgr->vis_usage);
511 mem->mm_node = nodes;
517 drm_mm_remove_node(&nodes[i]);
518 spin_unlock(&mgr->lock);
519 atomic64_sub(mem->num_pages << PAGE_SHIFT, &mgr->usage);
526 * amdgpu_vram_mgr_del - free ranges
528 * @man: TTM memory type manager
529 * @mem: TTM memory object
531 * Free the allocated VRAM again.
533 static void amdgpu_vram_mgr_del(struct ttm_resource_manager *man,
534 struct ttm_resource *mem)
536 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
537 struct amdgpu_device *adev = to_amdgpu_device(mgr);
538 struct drm_mm_node *nodes = mem->mm_node;
539 uint64_t usage = 0, vis_usage = 0;
540 unsigned pages = mem->num_pages;
545 spin_lock(&mgr->lock);
547 pages -= nodes->size;
548 drm_mm_remove_node(nodes);
549 usage += nodes->size << PAGE_SHIFT;
550 vis_usage += amdgpu_vram_mgr_vis_size(adev, nodes);
553 amdgpu_vram_mgr_do_reserve(man);
554 spin_unlock(&mgr->lock);
556 atomic64_sub(usage, &mgr->usage);
557 atomic64_sub(vis_usage, &mgr->vis_usage);
559 kvfree(mem->mm_node);
564 * amdgpu_vram_mgr_alloc_sgt - allocate and fill a sg table
566 * @adev: amdgpu device pointer
567 * @mem: TTM memory object
568 * @dev: the other device
569 * @dir: dma direction
570 * @sgt: resulting sg table
572 * Allocate and fill a sg table from a VRAM allocation.
574 int amdgpu_vram_mgr_alloc_sgt(struct amdgpu_device *adev,
575 struct ttm_resource *mem,
577 enum dma_data_direction dir,
578 struct sg_table **sgt)
580 struct drm_mm_node *node;
581 struct scatterlist *sg;
586 *sgt = kmalloc(sizeof(**sgt), GFP_KERNEL);
590 for (pages = mem->num_pages, node = mem->mm_node;
591 pages; pages -= node->size, ++node)
594 r = sg_alloc_table(*sgt, num_entries, GFP_KERNEL);
598 for_each_sgtable_sg((*sgt), sg, i)
602 for_each_sgtable_sg((*sgt), sg, i) {
603 phys_addr_t phys = (node->start << PAGE_SHIFT) +
605 size_t size = node->size << PAGE_SHIFT;
609 addr = dma_map_resource(dev, phys, size, dir,
610 DMA_ATTR_SKIP_CPU_SYNC);
611 r = dma_mapping_error(dev, addr);
615 sg_set_page(sg, NULL, size, 0);
616 sg_dma_address(sg) = addr;
617 sg_dma_len(sg) = size;
622 for_each_sgtable_sg((*sgt), sg, i) {
626 dma_unmap_resource(dev, sg->dma_address,
628 DMA_ATTR_SKIP_CPU_SYNC);
638 * amdgpu_vram_mgr_free_sgt - allocate and fill a sg table
640 * @dev: device pointer
641 * @dir: data direction of resource to unmap
642 * @sgt: sg table to free
644 * Free a previously allocate sg table.
646 void amdgpu_vram_mgr_free_sgt(struct device *dev,
647 enum dma_data_direction dir,
648 struct sg_table *sgt)
650 struct scatterlist *sg;
653 for_each_sgtable_sg(sgt, sg, i)
654 dma_unmap_resource(dev, sg->dma_address,
656 DMA_ATTR_SKIP_CPU_SYNC);
662 * amdgpu_vram_mgr_usage - how many bytes are used in this domain
664 * @man: TTM memory type manager
666 * Returns how many bytes are used in this domain.
668 uint64_t amdgpu_vram_mgr_usage(struct ttm_resource_manager *man)
670 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
672 return atomic64_read(&mgr->usage);
676 * amdgpu_vram_mgr_vis_usage - how many bytes are used in the visible part
678 * @man: TTM memory type manager
680 * Returns how many bytes are used in the visible part of VRAM
682 uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_resource_manager *man)
684 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
686 return atomic64_read(&mgr->vis_usage);
690 * amdgpu_vram_mgr_debug - dump VRAM table
692 * @man: TTM memory type manager
693 * @printer: DRM printer to use
695 * Dump the table content using printk.
697 static void amdgpu_vram_mgr_debug(struct ttm_resource_manager *man,
698 struct drm_printer *printer)
700 struct amdgpu_vram_mgr *mgr = to_vram_mgr(man);
702 spin_lock(&mgr->lock);
703 drm_mm_print(&mgr->mm, printer);
704 spin_unlock(&mgr->lock);
706 drm_printf(printer, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
707 man->size, amdgpu_vram_mgr_usage(man) >> 20,
708 amdgpu_vram_mgr_vis_usage(man) >> 20);
711 static const struct ttm_resource_manager_func amdgpu_vram_mgr_func = {
712 .alloc = amdgpu_vram_mgr_new,
713 .free = amdgpu_vram_mgr_del,
714 .debug = amdgpu_vram_mgr_debug