1 /**************************************************************************
3 * Copyright © 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 #include "vmwgfx_drv.h"
29 #include <drm/ttm/ttm_bo_driver.h>
30 #include <drm/ttm/ttm_placement.h>
31 #include <drm/ttm/ttm_page_alloc.h>
33 static uint32_t vram_placement_flags = TTM_PL_FLAG_VRAM |
36 static uint32_t vram_ne_placement_flags = TTM_PL_FLAG_VRAM |
40 static uint32_t sys_placement_flags = TTM_PL_FLAG_SYSTEM |
43 static uint32_t sys_ne_placement_flags = TTM_PL_FLAG_SYSTEM |
47 static uint32_t gmr_placement_flags = VMW_PL_FLAG_GMR |
50 static uint32_t gmr_ne_placement_flags = VMW_PL_FLAG_GMR |
54 static uint32_t mob_placement_flags = VMW_PL_FLAG_MOB |
57 struct ttm_placement vmw_vram_placement = {
61 .placement = &vram_placement_flags,
62 .num_busy_placement = 1,
63 .busy_placement = &vram_placement_flags
66 static uint32_t vram_gmr_placement_flags[] = {
67 TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED,
68 VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED
71 static uint32_t gmr_vram_placement_flags[] = {
72 VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED,
73 TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED
76 struct ttm_placement vmw_vram_gmr_placement = {
80 .placement = vram_gmr_placement_flags,
81 .num_busy_placement = 1,
82 .busy_placement = &gmr_placement_flags
85 static uint32_t vram_gmr_ne_placement_flags[] = {
86 TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED | TTM_PL_FLAG_NO_EVICT,
87 VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED | TTM_PL_FLAG_NO_EVICT
90 struct ttm_placement vmw_vram_gmr_ne_placement = {
94 .placement = vram_gmr_ne_placement_flags,
95 .num_busy_placement = 1,
96 .busy_placement = &gmr_ne_placement_flags
99 struct ttm_placement vmw_vram_sys_placement = {
103 .placement = &vram_placement_flags,
104 .num_busy_placement = 1,
105 .busy_placement = &sys_placement_flags
108 struct ttm_placement vmw_vram_ne_placement = {
112 .placement = &vram_ne_placement_flags,
113 .num_busy_placement = 1,
114 .busy_placement = &vram_ne_placement_flags
117 struct ttm_placement vmw_sys_placement = {
121 .placement = &sys_placement_flags,
122 .num_busy_placement = 1,
123 .busy_placement = &sys_placement_flags
126 struct ttm_placement vmw_sys_ne_placement = {
130 .placement = &sys_ne_placement_flags,
131 .num_busy_placement = 1,
132 .busy_placement = &sys_ne_placement_flags
135 static uint32_t evictable_placement_flags[] = {
136 TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED,
137 TTM_PL_FLAG_VRAM | TTM_PL_FLAG_CACHED,
138 VMW_PL_FLAG_GMR | TTM_PL_FLAG_CACHED,
139 VMW_PL_FLAG_MOB | TTM_PL_FLAG_CACHED
142 struct ttm_placement vmw_evictable_placement = {
146 .placement = evictable_placement_flags,
147 .num_busy_placement = 1,
148 .busy_placement = &sys_placement_flags
151 struct ttm_placement vmw_srf_placement = {
155 .num_busy_placement = 2,
156 .placement = &gmr_placement_flags,
157 .busy_placement = gmr_vram_placement_flags
160 struct ttm_placement vmw_mob_placement = {
164 .num_busy_placement = 1,
165 .placement = &mob_placement_flags,
166 .busy_placement = &mob_placement_flags
170 struct ttm_dma_tt dma_ttm;
171 struct vmw_private *dev_priv;
176 struct vmw_sg_table vsgt;
177 uint64_t sg_alloc_size;
181 const size_t vmw_tt_size = sizeof(struct vmw_ttm_tt);
184 * Helper functions to advance a struct vmw_piter iterator.
186 * @viter: Pointer to the iterator.
188 * These functions return false if past the end of the list,
189 * true otherwise. Functions are selected depending on the current
192 static bool __vmw_piter_non_sg_next(struct vmw_piter *viter)
194 return ++(viter->i) < viter->num_pages;
197 static bool __vmw_piter_sg_next(struct vmw_piter *viter)
199 return __sg_page_iter_next(&viter->iter);
204 * Helper functions to return a pointer to the current page.
206 * @viter: Pointer to the iterator
208 * These functions return a pointer to the page currently
209 * pointed to by @viter. Functions are selected depending on the
210 * current mapping mode.
212 static struct page *__vmw_piter_non_sg_page(struct vmw_piter *viter)
214 return viter->pages[viter->i];
217 static struct page *__vmw_piter_sg_page(struct vmw_piter *viter)
219 return sg_page_iter_page(&viter->iter);
224 * Helper functions to return the DMA address of the current page.
226 * @viter: Pointer to the iterator
228 * These functions return the DMA address of the page currently
229 * pointed to by @viter. Functions are selected depending on the
230 * current mapping mode.
232 static dma_addr_t __vmw_piter_phys_addr(struct vmw_piter *viter)
234 return page_to_phys(viter->pages[viter->i]);
237 static dma_addr_t __vmw_piter_dma_addr(struct vmw_piter *viter)
239 return viter->addrs[viter->i];
242 static dma_addr_t __vmw_piter_sg_addr(struct vmw_piter *viter)
244 return sg_page_iter_dma_address(&viter->iter);
249 * vmw_piter_start - Initialize a struct vmw_piter.
251 * @viter: Pointer to the iterator to initialize
252 * @vsgt: Pointer to a struct vmw_sg_table to initialize from
254 * Note that we're following the convention of __sg_page_iter_start, so that
255 * the iterator doesn't point to a valid page after initialization; it has
256 * to be advanced one step first.
258 void vmw_piter_start(struct vmw_piter *viter, const struct vmw_sg_table *vsgt,
259 unsigned long p_offset)
261 viter->i = p_offset - 1;
262 viter->num_pages = vsgt->num_pages;
263 switch (vsgt->mode) {
265 viter->next = &__vmw_piter_non_sg_next;
266 viter->dma_address = &__vmw_piter_phys_addr;
267 viter->page = &__vmw_piter_non_sg_page;
268 viter->pages = vsgt->pages;
270 case vmw_dma_alloc_coherent:
271 viter->next = &__vmw_piter_non_sg_next;
272 viter->dma_address = &__vmw_piter_dma_addr;
273 viter->page = &__vmw_piter_non_sg_page;
274 viter->addrs = vsgt->addrs;
276 case vmw_dma_map_populate:
277 case vmw_dma_map_bind:
278 viter->next = &__vmw_piter_sg_next;
279 viter->dma_address = &__vmw_piter_sg_addr;
280 viter->page = &__vmw_piter_sg_page;
281 __sg_page_iter_start(&viter->iter, vsgt->sgt->sgl,
282 vsgt->sgt->orig_nents, p_offset);
290 * vmw_ttm_unmap_from_dma - unmap device addresses previsouly mapped for
293 * @vmw_tt: Pointer to a struct vmw_ttm_backend
295 * Used to free dma mappings previously mapped by vmw_ttm_map_for_dma.
297 static void vmw_ttm_unmap_from_dma(struct vmw_ttm_tt *vmw_tt)
299 struct device *dev = vmw_tt->dev_priv->dev->dev;
301 dma_unmap_sg(dev, vmw_tt->sgt.sgl, vmw_tt->sgt.nents,
303 vmw_tt->sgt.nents = vmw_tt->sgt.orig_nents;
307 * vmw_ttm_map_for_dma - map TTM pages to get device addresses
309 * @vmw_tt: Pointer to a struct vmw_ttm_backend
311 * This function is used to get device addresses from the kernel DMA layer.
312 * However, it's violating the DMA API in that when this operation has been
313 * performed, it's illegal for the CPU to write to the pages without first
314 * unmapping the DMA mappings, or calling dma_sync_sg_for_cpu(). It is
315 * therefore only legal to call this function if we know that the function
316 * dma_sync_sg_for_cpu() is a NOP, and dma_sync_sg_for_device() is at most
317 * a CPU write buffer flush.
319 static int vmw_ttm_map_for_dma(struct vmw_ttm_tt *vmw_tt)
321 struct device *dev = vmw_tt->dev_priv->dev->dev;
324 ret = dma_map_sg(dev, vmw_tt->sgt.sgl, vmw_tt->sgt.orig_nents,
326 if (unlikely(ret == 0))
329 vmw_tt->sgt.nents = ret;
335 * vmw_ttm_map_dma - Make sure TTM pages are visible to the device
337 * @vmw_tt: Pointer to a struct vmw_ttm_tt
339 * Select the correct function for and make sure the TTM pages are
340 * visible to the device. Allocate storage for the device mappings.
341 * If a mapping has already been performed, indicated by the storage
342 * pointer being non NULL, the function returns success.
344 static int vmw_ttm_map_dma(struct vmw_ttm_tt *vmw_tt)
346 struct vmw_private *dev_priv = vmw_tt->dev_priv;
347 struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
348 struct vmw_sg_table *vsgt = &vmw_tt->vsgt;
349 struct vmw_piter iter;
352 static size_t sgl_size;
353 static size_t sgt_size;
358 vsgt->mode = dev_priv->map_mode;
359 vsgt->pages = vmw_tt->dma_ttm.ttm.pages;
360 vsgt->num_pages = vmw_tt->dma_ttm.ttm.num_pages;
361 vsgt->addrs = vmw_tt->dma_ttm.dma_address;
362 vsgt->sgt = &vmw_tt->sgt;
364 switch (dev_priv->map_mode) {
365 case vmw_dma_map_bind:
366 case vmw_dma_map_populate:
367 if (unlikely(!sgl_size)) {
368 sgl_size = ttm_round_pot(sizeof(struct scatterlist));
369 sgt_size = ttm_round_pot(sizeof(struct sg_table));
371 vmw_tt->sg_alloc_size = sgt_size + sgl_size * vsgt->num_pages;
372 ret = ttm_mem_global_alloc(glob, vmw_tt->sg_alloc_size, false,
374 if (unlikely(ret != 0))
377 ret = sg_alloc_table_from_pages(&vmw_tt->sgt, vsgt->pages,
380 vsgt->num_pages << PAGE_SHIFT,
382 if (unlikely(ret != 0))
383 goto out_sg_alloc_fail;
385 if (vsgt->num_pages > vmw_tt->sgt.nents) {
386 uint64_t over_alloc =
387 sgl_size * (vsgt->num_pages -
390 ttm_mem_global_free(glob, over_alloc);
391 vmw_tt->sg_alloc_size -= over_alloc;
394 ret = vmw_ttm_map_for_dma(vmw_tt);
395 if (unlikely(ret != 0))
403 old = ~((dma_addr_t) 0);
404 vmw_tt->vsgt.num_regions = 0;
405 for (vmw_piter_start(&iter, vsgt, 0); vmw_piter_next(&iter);) {
406 dma_addr_t cur = vmw_piter_dma_addr(&iter);
408 if (cur != old + PAGE_SIZE)
409 vmw_tt->vsgt.num_regions++;
413 vmw_tt->mapped = true;
417 sg_free_table(vmw_tt->vsgt.sgt);
418 vmw_tt->vsgt.sgt = NULL;
420 ttm_mem_global_free(glob, vmw_tt->sg_alloc_size);
425 * vmw_ttm_unmap_dma - Tear down any TTM page device mappings
427 * @vmw_tt: Pointer to a struct vmw_ttm_tt
429 * Tear down any previously set up device DMA mappings and free
430 * any storage space allocated for them. If there are no mappings set up,
431 * this function is a NOP.
433 static void vmw_ttm_unmap_dma(struct vmw_ttm_tt *vmw_tt)
435 struct vmw_private *dev_priv = vmw_tt->dev_priv;
437 if (!vmw_tt->vsgt.sgt)
440 switch (dev_priv->map_mode) {
441 case vmw_dma_map_bind:
442 case vmw_dma_map_populate:
443 vmw_ttm_unmap_from_dma(vmw_tt);
444 sg_free_table(vmw_tt->vsgt.sgt);
445 vmw_tt->vsgt.sgt = NULL;
446 ttm_mem_global_free(vmw_mem_glob(dev_priv),
447 vmw_tt->sg_alloc_size);
452 vmw_tt->mapped = false;
455 static int vmw_ttm_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem)
457 struct vmw_ttm_tt *vmw_be =
458 container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
461 ret = vmw_ttm_map_dma(vmw_be);
462 if (unlikely(ret != 0))
465 vmw_be->gmr_id = bo_mem->start;
466 vmw_be->mem_type = bo_mem->mem_type;
468 switch (bo_mem->mem_type) {
470 return vmw_gmr_bind(vmw_be->dev_priv, &vmw_be->vsgt,
471 ttm->num_pages, vmw_be->gmr_id);
473 if (unlikely(vmw_be->mob == NULL)) {
475 vmw_mob_create(ttm->num_pages);
476 if (unlikely(vmw_be->mob == NULL))
480 return vmw_mob_bind(vmw_be->dev_priv, vmw_be->mob,
481 ttm->pages, ttm->num_pages,
489 static int vmw_ttm_unbind(struct ttm_tt *ttm)
491 struct vmw_ttm_tt *vmw_be =
492 container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
494 switch (vmw_be->mem_type) {
496 vmw_gmr_unbind(vmw_be->dev_priv, vmw_be->gmr_id);
499 vmw_mob_unbind(vmw_be->dev_priv, vmw_be->mob);
505 if (vmw_be->dev_priv->map_mode == vmw_dma_map_bind)
506 vmw_ttm_unmap_dma(vmw_be);
512 static void vmw_ttm_destroy(struct ttm_tt *ttm)
514 struct vmw_ttm_tt *vmw_be =
515 container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
517 vmw_ttm_unmap_dma(vmw_be);
518 if (vmw_be->dev_priv->map_mode == vmw_dma_alloc_coherent)
519 ttm_dma_tt_fini(&vmw_be->dma_ttm);
524 vmw_mob_destroy(vmw_be->mob);
529 static int vmw_ttm_populate(struct ttm_tt *ttm)
531 struct vmw_ttm_tt *vmw_tt =
532 container_of(ttm, struct vmw_ttm_tt, dma_ttm.ttm);
533 struct vmw_private *dev_priv = vmw_tt->dev_priv;
534 struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
537 if (ttm->state != tt_unpopulated)
540 if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
542 ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
543 ret = ttm_mem_global_alloc(glob, size, false, true);
544 if (unlikely(ret != 0))
547 ret = ttm_dma_populate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
548 if (unlikely(ret != 0))
549 ttm_mem_global_free(glob, size);
551 ret = ttm_pool_populate(ttm);
556 static void vmw_ttm_unpopulate(struct ttm_tt *ttm)
558 struct vmw_ttm_tt *vmw_tt = container_of(ttm, struct vmw_ttm_tt,
560 struct vmw_private *dev_priv = vmw_tt->dev_priv;
561 struct ttm_mem_global *glob = vmw_mem_glob(dev_priv);
565 vmw_mob_destroy(vmw_tt->mob);
569 vmw_ttm_unmap_dma(vmw_tt);
570 if (dev_priv->map_mode == vmw_dma_alloc_coherent) {
572 ttm_round_pot(ttm->num_pages * sizeof(dma_addr_t));
574 ttm_dma_unpopulate(&vmw_tt->dma_ttm, dev_priv->dev->dev);
575 ttm_mem_global_free(glob, size);
577 ttm_pool_unpopulate(ttm);
580 static struct ttm_backend_func vmw_ttm_func = {
581 .bind = vmw_ttm_bind,
582 .unbind = vmw_ttm_unbind,
583 .destroy = vmw_ttm_destroy,
586 static struct ttm_tt *vmw_ttm_tt_create(struct ttm_bo_device *bdev,
587 unsigned long size, uint32_t page_flags,
588 struct page *dummy_read_page)
590 struct vmw_ttm_tt *vmw_be;
593 vmw_be = kzalloc(sizeof(*vmw_be), GFP_KERNEL);
597 vmw_be->dma_ttm.ttm.func = &vmw_ttm_func;
598 vmw_be->dev_priv = container_of(bdev, struct vmw_private, bdev);
601 if (vmw_be->dev_priv->map_mode == vmw_dma_alloc_coherent)
602 ret = ttm_dma_tt_init(&vmw_be->dma_ttm, bdev, size, page_flags,
605 ret = ttm_tt_init(&vmw_be->dma_ttm.ttm, bdev, size, page_flags,
607 if (unlikely(ret != 0))
610 return &vmw_be->dma_ttm.ttm;
616 static int vmw_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
621 static int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
622 struct ttm_mem_type_manager *man)
628 man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
629 man->available_caching = TTM_PL_FLAG_CACHED;
630 man->default_caching = TTM_PL_FLAG_CACHED;
633 /* "On-card" video ram */
634 man->func = &ttm_bo_manager_func;
636 man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_MAPPABLE;
637 man->available_caching = TTM_PL_FLAG_CACHED;
638 man->default_caching = TTM_PL_FLAG_CACHED;
643 * "Guest Memory Regions" is an aperture like feature with
644 * one slot per bo. There is an upper limit of the number of
645 * slots as well as the bo size.
647 man->func = &vmw_gmrid_manager_func;
649 man->flags = TTM_MEMTYPE_FLAG_CMA | TTM_MEMTYPE_FLAG_MAPPABLE;
650 man->available_caching = TTM_PL_FLAG_CACHED;
651 man->default_caching = TTM_PL_FLAG_CACHED;
654 DRM_ERROR("Unsupported memory type %u\n", (unsigned)type);
660 static void vmw_evict_flags(struct ttm_buffer_object *bo,
661 struct ttm_placement *placement)
663 *placement = vmw_sys_placement;
666 static int vmw_verify_access(struct ttm_buffer_object *bo, struct file *filp)
668 struct ttm_object_file *tfile =
669 vmw_fpriv((struct drm_file *)filp->private_data)->tfile;
671 return vmw_user_dmabuf_verify_access(bo, tfile);
674 static int vmw_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
676 struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
677 struct vmw_private *dev_priv = container_of(bdev, struct vmw_private, bdev);
679 mem->bus.addr = NULL;
680 mem->bus.is_iomem = false;
682 mem->bus.size = mem->num_pages << PAGE_SHIFT;
684 if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
686 switch (mem->mem_type) {
692 mem->bus.offset = mem->start << PAGE_SHIFT;
693 mem->bus.base = dev_priv->vram_start;
694 mem->bus.is_iomem = true;
702 static void vmw_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
706 static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
712 * FIXME: We're using the old vmware polling method to sync.
713 * Do this with fences instead.
716 static void *vmw_sync_obj_ref(void *sync_obj)
720 vmw_fence_obj_reference((struct vmw_fence_obj *) sync_obj);
723 static void vmw_sync_obj_unref(void **sync_obj)
725 vmw_fence_obj_unreference((struct vmw_fence_obj **) sync_obj);
728 static int vmw_sync_obj_flush(void *sync_obj)
730 vmw_fence_obj_flush((struct vmw_fence_obj *) sync_obj);
734 static bool vmw_sync_obj_signaled(void *sync_obj)
736 return vmw_fence_obj_signaled((struct vmw_fence_obj *) sync_obj,
737 DRM_VMW_FENCE_FLAG_EXEC);
741 static int vmw_sync_obj_wait(void *sync_obj, bool lazy, bool interruptible)
743 return vmw_fence_obj_wait((struct vmw_fence_obj *) sync_obj,
744 DRM_VMW_FENCE_FLAG_EXEC,
746 VMW_FENCE_WAIT_TIMEOUT);
750 * vmw_move_notify - TTM move_notify_callback
752 * @bo: The TTM buffer object about to move.
753 * @mem: The truct ttm_mem_reg indicating to what memory
754 * region the move is taking place.
756 * Calls move_notify for all subsystems needing it.
757 * (currently only resources).
759 static void vmw_move_notify(struct ttm_buffer_object *bo,
760 struct ttm_mem_reg *mem)
762 vmw_resource_move_notify(bo, mem);
767 * vmw_swap_notify - TTM move_notify_callback
769 * @bo: The TTM buffer object about to be swapped out.
771 static void vmw_swap_notify(struct ttm_buffer_object *bo)
773 struct ttm_bo_device *bdev = bo->bdev;
775 spin_lock(&bdev->fence_lock);
776 ttm_bo_wait(bo, false, false, false);
777 spin_unlock(&bdev->fence_lock);
781 struct ttm_bo_driver vmw_bo_driver = {
782 .ttm_tt_create = &vmw_ttm_tt_create,
783 .ttm_tt_populate = &vmw_ttm_populate,
784 .ttm_tt_unpopulate = &vmw_ttm_unpopulate,
785 .invalidate_caches = vmw_invalidate_caches,
786 .init_mem_type = vmw_init_mem_type,
787 .evict_flags = vmw_evict_flags,
789 .verify_access = vmw_verify_access,
790 .sync_obj_signaled = vmw_sync_obj_signaled,
791 .sync_obj_wait = vmw_sync_obj_wait,
792 .sync_obj_flush = vmw_sync_obj_flush,
793 .sync_obj_unref = vmw_sync_obj_unref,
794 .sync_obj_ref = vmw_sync_obj_ref,
795 .move_notify = vmw_move_notify,
796 .swap_notify = vmw_swap_notify,
797 .fault_reserve_notify = &vmw_ttm_fault_reserve_notify,
798 .io_mem_reserve = &vmw_ttm_io_mem_reserve,
799 .io_mem_free = &vmw_ttm_io_mem_free,