History

drm/radeon: add userptr support v8

This patch adds an IOCTL for turning a pointer supplied by userspace into a buffer object. It imposes several restrictions upon the memory being mapped: 1. It must be page aligned (both start/end addresses, i.e ptr and size). 2. It must be normal system memory, not a pointer into another map of IO space (e.g. it must not be a GTT mmapping of another object). 3. The BO is mapped into GTT, so the maximum amount of memory mapped at all times is still the GTT limit. 4. The BO is only mapped readonly for now, so no write support. 5. List of backing pages is only acquired once, so they represent a snapshot of the first use. Exporting and sharing as well as mapping of buffer objects created by this function is forbidden and results in an -EPERM. v2: squash all previous changes into first public version v3: fix tabs, map readonly, don't use MM callback any more v4: set TTM_PAGE_FLAG_SG so that TTM never messes with the pages, pin/unpin pages on bind/unbind instead of populate/unpopulate v5: rebased on 3.17-wip, IOCTL renamed to userptr, reject any unknown flags, better handle READONLY flag, improve permission check v6: fix ptr cast warning, use set_page_dirty/mark_page_accessed on unpin v7: add warning about it's availability in the API definition v8: drop access_ok check, fix VM mapping bits

Christian König committed 12 years ago

Commit f72a1131 parent 7963e9d10 files · +279 −3Browse files

Unified · Split

10 files
  1. drivers/gpu/drm/radeon/radeon.h +6 −0
  2. drivers/gpu/drm/radeon/radeon_cs.c +23 −2
  3. drivers/gpu/drm/radeon/radeon_drv.c +4 −1
  4. drivers/gpu/drm/radeon/radeon_gem.c +68 −0
  5. drivers/gpu/drm/radeon/radeon_kms.c +1 −0
  6. drivers/gpu/drm/radeon/radeon_object.c +3 −0
  7. drivers/gpu/drm/radeon/radeon_prime.c +10 −0
  8. drivers/gpu/drm/radeon/radeon_ttm.c +145 −0
  9. drivers/gpu/drm/radeon/radeon_vm.c +3 −0
  10. include/uapi/drm/radeon_drm.h +16 −0
drivers/gpu/drm/radeon/radeon.h+6−0
@@ -2138,6 +2138,8 @@ int radeon_gem_info_ioctl(struct drm_device *dev, void *data, · 2137 unmodified lines
21382138 struct drm_file *filp);
21392139 int radeon_gem_create_ioctl(struct drm_device *dev, void *data,
21402140 struct drm_file *filp);
2141+int radeon_gem_userptr_ioctl(struct drm_device *dev, void *data,
2142+ struct drm_file *filp);
21412143 int radeon_gem_pin_ioctl(struct drm_device *dev, void *data,
21422144 struct drm_file *file_priv);
21432145 int radeon_gem_unpin_ioctl(struct drm_device *dev, void *data,
@@ -2871,6 +2873,10 @@ extern void radeon_legacy_set_clock_gating(struct radeon_device *rdev, int enabl · 727 unmodified lines
28712873 extern void radeon_atom_set_clock_gating(struct radeon_device *rdev, int enable);
28722874 extern void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain);
28732875 extern bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo);
2876+extern int radeon_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
2877+ uint32_t flags);
2878+extern bool radeon_ttm_tt_has_userptr(struct ttm_tt *ttm);
2879+extern bool radeon_ttm_tt_is_readonly(struct ttm_tt *ttm);
28742880 extern void radeon_vram_location(struct radeon_device *rdev, struct radeon_mc *mc, u64 base);
28752881 extern void radeon_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
28762882 extern int radeon_resume_kms(struct drm_device *dev, bool resume, bool fbcon);
drivers/gpu/drm/radeon/radeon_cs.c+23−2
@@ -78,7 +78,8 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p) · 77 unmodified lines
7878 struct radeon_cs_chunk *chunk;
7979 struct radeon_cs_buckets buckets;
8080 unsigned i, j;
81− bool duplicate;
81+ bool duplicate, need_mmap_lock = false;
82+ int r;
8283
8384 if (p->chunk_relocs_idx == -1) {
8485 return 0;
@@ -164,6 +165,19 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p) · 79 unmodified lines
164165 p->relocs[i].allowed_domains = domain;
165166 }
166167
168+ if (radeon_ttm_tt_has_userptr(p->relocs[i].robj->tbo.ttm)) {
169+ uint32_t domain = p->relocs[i].prefered_domains;
170+ if (!(domain & RADEON_GEM_DOMAIN_GTT)) {
171+ DRM_ERROR("Only RADEON_GEM_DOMAIN_GTT is "
172+ "allowed for userptr BOs\n");
173+ return -EINVAL;
174+ }
175+ need_mmap_lock = true;
176+ domain = RADEON_GEM_DOMAIN_GTT;
177+ p->relocs[i].prefered_domains = domain;
178+ p->relocs[i].allowed_domains = domain;
179+ }
180+
167181 p->relocs[i].tv.bo = &p->relocs[i].robj->tbo;
168182 p->relocs[i].handle = r->handle;
169183
@@ -176,8 +190,15 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p) · 6 unmodified lines
176190 if (p->cs_flags & RADEON_CS_USE_VM)
177191 p->vm_bos = radeon_vm_get_bos(p->rdev, p->ib.vm,
178192 &p->validated);
193+ if (need_mmap_lock)
194+ down_read(&current->mm->mmap_sem);
195+
196+ r = radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
179197
180− return radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
198+ if (need_mmap_lock)
199+ up_read(&current->mm->mmap_sem);
200+
201+ return r;
181202 }
182203
183204 static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)
drivers/gpu/drm/radeon/radeon_drv.c+4−1
@@ -114,6 +114,9 @@ int radeon_gem_object_open(struct drm_gem_object *obj, · 113 unmodified lines
114114 struct drm_file *file_priv);
115115 void radeon_gem_object_close(struct drm_gem_object *obj,
116116 struct drm_file *file_priv);
117+struct dma_buf *radeon_gem_prime_export(struct drm_device *dev,
118+ struct drm_gem_object *gobj,
119+ int flags);
117120 extern int radeon_get_crtc_scanoutpos(struct drm_device *dev, int crtc,
118121 unsigned int flags,
119122 int *vpos, int *hpos, ktime_t *stime,
@@ -568,7 +571,7 @@ static struct drm_driver kms_driver = { · 448 unmodified lines
568571
569572 .prime_handle_to_fd = drm_gem_prime_handle_to_fd,
570573 .prime_fd_to_handle = drm_gem_prime_fd_to_handle,
571− .gem_prime_export = drm_gem_prime_export,
574+ .gem_prime_export = radeon_gem_prime_export,
572575 .gem_prime_import = drm_gem_prime_import,
573576 .gem_prime_pin = radeon_gem_prime_pin,
574577 .gem_prime_unpin = radeon_gem_prime_unpin,
drivers/gpu/drm/radeon/radeon_gem.c+68−0
@@ -272,6 +272,65 @@ int radeon_gem_create_ioctl(struct drm_device *dev, void *data, · 271 unmodified lines
272272 return 0;
273273 }
274274
275+int radeon_gem_userptr_ioctl(struct drm_device *dev, void *data,
276+ struct drm_file *filp)
277+{
278+ struct radeon_device *rdev = dev->dev_private;
279+ struct drm_radeon_gem_userptr *args = data;
280+ struct drm_gem_object *gobj;
281+ struct radeon_bo *bo;
282+ uint32_t handle;
283+ int r;
284+
285+ if (offset_in_page(args->addr | args->size))
286+ return -EINVAL;
287+
288+ /* we only support read only mappings for now */
289+ if (!(args->flags & RADEON_GEM_USERPTR_READONLY))
290+ return -EACCES;
291+
292+ /* reject unknown flag values */
293+ if (args->flags & ~RADEON_GEM_USERPTR_READONLY)
294+ return -EINVAL;
295+
296+ /* readonly pages not tested on older hardware */
297+ if (rdev->family < CHIP_R600)
298+ return -EINVAL;
299+
300+ down_read(&rdev->exclusive_lock);
301+
302+ /* create a gem object to contain this object in */
303+ r = radeon_gem_object_create(rdev, args->size, 0,
304+ RADEON_GEM_DOMAIN_CPU, 0,
305+ false, &gobj);
306+ if (r)
307+ goto handle_lockup;
308+
309+ bo = gem_to_radeon_bo(gobj);
310+ r = radeon_ttm_tt_set_userptr(bo->tbo.ttm, args->addr, args->flags);
311+ if (r)
312+ goto release_object;
313+
314+ r = drm_gem_handle_create(filp, gobj, &handle);
315+ /* drop reference from allocate - handle holds it now */
316+ drm_gem_object_unreference_unlocked(gobj);
317+ if (r)
318+ goto handle_lockup;
319+
320+ args->handle = handle;
321+ up_read(&rdev->exclusive_lock);
322+ return 0;
323+
324+release_object:
325+ drm_gem_object_unreference_unlocked(gobj);
326+
327+handle_lockup:
328+ up_read(&rdev->exclusive_lock);
329+ r = radeon_gem_handle_lockup(rdev, r);
330+
331+ return r;
332+}
333+
275334 int radeon_gem_set_domain_ioctl(struct drm_device *dev, void *data,
276335 struct drm_file *filp)
277336 {
@@ -315,6 +374,10 @@ int radeon_mode_dumb_mmap(struct drm_file *filp, · 37 unmodified lines
315374 return -ENOENT;
316375 }
317376 robj = gem_to_radeon_bo(gobj);
377+ if (radeon_ttm_tt_has_userptr(robj->tbo.ttm)) {
378+ drm_gem_object_unreference_unlocked(gobj);
379+ return -EPERM;
380+ }
318381 *offset_p = radeon_bo_mmap_offset(robj);
319382 drm_gem_object_unreference_unlocked(gobj);
320383 return 0;
@@ -532,6 +595,11 @@ int radeon_gem_op_ioctl(struct drm_device *dev, void *data, · 211 unmodified lines
532595 return -ENOENT;
533596 }
534597 robj = gem_to_radeon_bo(gobj);
598+
599+ r = -EPERM;
600+ if (radeon_ttm_tt_has_userptr(robj->tbo.ttm))
601+ goto out;
602+
535603 r = radeon_bo_reserve(robj, false);
536604 if (unlikely(r))
537605 goto out;
drivers/gpu/drm/radeon/radeon_kms.c+1−0
@@ -885,5 +885,6 @@ const struct drm_ioctl_desc radeon_ioctls_kms[] = { · 884 unmodified lines
885885 DRM_IOCTL_DEF_DRV(RADEON_GEM_BUSY, radeon_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
886886 DRM_IOCTL_DEF_DRV(RADEON_GEM_VA, radeon_gem_va_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
887887 DRM_IOCTL_DEF_DRV(RADEON_GEM_OP, radeon_gem_op_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
888+ DRM_IOCTL_DEF_DRV(RADEON_GEM_USERPTR, radeon_gem_userptr_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
888889 };
889890 int radeon_max_kms_ioctl = ARRAY_SIZE(radeon_ioctls_kms);
drivers/gpu/drm/radeon/radeon_object.c+3−0
@@ -264,6 +264,9 @@ int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset, · 263 unmodified lines
264264 {
265265 int r, i;
266266
267+ if (radeon_ttm_tt_has_userptr(bo->tbo.ttm))
268+ return -EPERM;
269+
267270 if (bo->pin_count) {
268271 bo->pin_count++;
269272 if (gpu_addr)
drivers/gpu/drm/radeon/radeon_prime.c+10−0
@@ -103,3 +103,13 @@ void radeon_gem_prime_unpin(struct drm_gem_object *obj) · 102 unmodified lines
103103 radeon_bo_unpin(bo);
104104 radeon_bo_unreserve(bo);
105105 }
106+
107+struct dma_buf *radeon_gem_prime_export(struct drm_device *dev,
108+ struct drm_gem_object *gobj,
109+ int flags)
110+{
111+ struct radeon_bo *bo = gem_to_radeon_bo(gobj);
112+ if (radeon_ttm_tt_has_userptr(bo->tbo.ttm))
113+ return ERR_PTR(-EPERM);
114+ return drm_gem_prime_export(dev, gobj, flags);
115+}
drivers/gpu/drm/radeon/radeon_ttm.c+145−0
@@ -39,6 +39,8 @@ · 38 unmodified lines
3939 #include <linux/seq_file.h>
4040 #include <linux/slab.h>
4141 #include <linux/swiotlb.h>
42+#include <linux/swap.h>
43+#include <linux/pagemap.h>
4244 #include <linux/debugfs.h>
4345 #include "radeon_reg.h"
4446 #include "radeon.h"
@@ -515,8 +517,92 @@ struct radeon_ttm_tt { · 470 unmodified lines
515517 struct ttm_dma_tt ttm;
516518 struct radeon_device *rdev;
517519 u64 offset;
520+
521+ uint64_t userptr;
522+ struct mm_struct *usermm;
523+ uint32_t userflags;
518524 };
519525
526+/* prepare the sg table with the user pages */
527+static int radeon_ttm_tt_pin_userptr(struct ttm_tt *ttm)
528+{
529+ struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
530+ struct radeon_ttm_tt *gtt = (void *)ttm;
531+ unsigned pinned = 0, nents;
532+ int r;
533+
534+ int write = !(gtt->userflags & RADEON_GEM_USERPTR_READONLY);
535+ enum dma_data_direction direction = write ?
536+ DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
537+
538+ if (current->mm != gtt->usermm)
539+ return -EPERM;
540+
541+ do {
542+ unsigned num_pages = ttm->num_pages - pinned;
543+ uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
544+ struct page **pages = ttm->pages + pinned;
545+
546+ r = get_user_pages(current, current->mm, userptr, num_pages,
547+ write, 0, pages, NULL);
548+ if (r < 0)
549+ goto release_pages;
550+
551+ pinned += r;
552+
553+ } while (pinned < ttm->num_pages);
554+
555+ r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
556+ ttm->num_pages << PAGE_SHIFT,
557+ GFP_KERNEL);
558+ if (r)
559+ goto release_sg;
560+
561+ r = -ENOMEM;
562+ nents = dma_map_sg(rdev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
563+ if (nents != ttm->sg->nents)
564+ goto release_sg;
565+
566+ drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
567+ gtt->ttm.dma_address, ttm->num_pages);
568+
569+ return 0;
570+
571+release_sg:
572+ kfree(ttm->sg);
573+
574+release_pages:
575+ release_pages(ttm->pages, pinned, 0);
576+ return r;
577+}
578+
579+static void radeon_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
580+{
581+ struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
582+ struct radeon_ttm_tt *gtt = (void *)ttm;
583+ struct scatterlist *sg;
584+ int i;
585+
586+ int write = !(gtt->userflags & RADEON_GEM_USERPTR_READONLY);
587+ enum dma_data_direction direction = write ?
588+ DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
589+
590+ /* free the sg table and pages again */
591+ dma_unmap_sg(rdev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
592+
593+ for_each_sg(ttm->sg->sgl, sg, ttm->sg->nents, i) {
594+ struct page *page = sg_page(sg);
595+
596+ if (!(gtt->userflags & RADEON_GEM_USERPTR_READONLY))
597+ set_page_dirty(page);
598+
599+ mark_page_accessed(page);
600+ page_cache_release(page);
601+ }
602+
603+ sg_free_table(ttm->sg);
604+}
605+
520606 static int radeon_ttm_backend_bind(struct ttm_tt *ttm,
521607 struct ttm_mem_reg *bo_mem)
522608 {
@@ -525,6 +611,11 @@ static int radeon_ttm_backend_bind(struct ttm_tt *ttm, · 2 unmodified lines
525611 RADEON_GART_PAGE_WRITE;
526612 int r;
527613
614+ if (gtt->userptr) {
615+ radeon_ttm_tt_pin_userptr(ttm);
616+ flags &= ~RADEON_GART_PAGE_WRITE;
617+ }
618+
528619 gtt->offset = (unsigned long)(bo_mem->start << PAGE_SHIFT);
529620 if (!ttm->num_pages) {
530621 WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n",
@@ -547,6 +638,10 @@ static int radeon_ttm_backend_unbind(struct ttm_tt *ttm) · 16 unmodified lines
547638 struct radeon_ttm_tt *gtt = (void *)ttm;
548639
549640 radeon_gart_unbind(gtt->rdev, gtt->offset, ttm->num_pages);
641+
642+ if (gtt->userptr)
643+ radeon_ttm_tt_unpin_userptr(ttm);
644+
550645 return 0;
551646 }
552647
@@ -603,6 +698,16 @@ static int radeon_ttm_tt_populate(struct ttm_tt *ttm) · 50 unmodified lines
603698 if (ttm->state != tt_unpopulated)
604699 return 0;
605700
701+ if (gtt->userptr) {
702+ ttm->sg = kcalloc(1, sizeof(struct sg_table), GFP_KERNEL);
703+ if (!ttm->sg)
704+ return -ENOMEM;
705+
706+ ttm->page_flags |= TTM_PAGE_FLAG_SG;
707+ ttm->state = tt_unbound;
708+ return 0;
709+ }
710+
606711 if (slave && ttm->sg) {
607712 drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
608713 gtt->ttm.dma_address, ttm->num_pages);
@@ -652,6 +757,12 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm) · 43 unmodified lines
652757 unsigned i;
653758 bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);
654759
760+ if (gtt->userptr) {
761+ kfree(ttm->sg);
762+ ttm->page_flags &= ~TTM_PAGE_FLAG_SG;
763+ return;
764+ }
765+
655766 if (slave)
656767 return;
657768
@@ -680,6 +791,40 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm) · 22 unmodified lines
680791 ttm_pool_unpopulate(ttm);
681792 }
682793
794+int radeon_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
795+ uint32_t flags)
796+{
797+ struct radeon_ttm_tt *gtt = (void *)ttm;
798+
799+ if (gtt == NULL)
800+ return -EINVAL;
801+
802+ gtt->userptr = addr;
803+ gtt->usermm = current->mm;
804+ gtt->userflags = flags;
805+ return 0;
806+}
807+
808+bool radeon_ttm_tt_has_userptr(struct ttm_tt *ttm)
809+{
810+ struct radeon_ttm_tt *gtt = (void *)ttm;
811+
812+ if (gtt == NULL)
813+ return false;
814+
815+ return !!gtt->userptr;
816+}
817+
818+bool radeon_ttm_tt_is_readonly(struct ttm_tt *ttm)
819+{
820+ struct radeon_ttm_tt *gtt = (void *)ttm;
821+
822+ if (gtt == NULL)
823+ return false;
824+
825+ return !!(gtt->userflags & RADEON_GEM_USERPTR_READONLY);
826+}
827+
683828 static struct ttm_bo_driver radeon_bo_driver = {
684829 .ttm_tt_create = &radeon_ttm_tt_create,
685830 .ttm_tt_populate = &radeon_ttm_tt_populate,
drivers/gpu/drm/radeon/radeon_vm.c+3−0
@@ -888,6 +888,9 @@ int radeon_vm_bo_update(struct radeon_device *rdev, · 887 unmodified lines
888888 bo_va->flags &= ~RADEON_VM_PAGE_VALID;
889889 bo_va->flags &= ~RADEON_VM_PAGE_SYSTEM;
890890 bo_va->flags &= ~RADEON_VM_PAGE_SNOOPED;
891+ if (bo_va->bo && radeon_ttm_tt_is_readonly(bo_va->bo->tbo.ttm))
892+ bo_va->flags &= ~RADEON_VM_PAGE_WRITEABLE;
893+
891894 if (mem) {
892895 addr = mem->start << PAGE_SHIFT;
893896 if (mem->mem_type != TTM_PL_SYSTEM) {
include/uapi/drm/radeon_drm.h+16−0
@@ -511,6 +511,7 @@ typedef struct { · 510 unmodified lines
511511 #define DRM_RADEON_GEM_BUSY 0x2a
512512 #define DRM_RADEON_GEM_VA 0x2b
513513 #define DRM_RADEON_GEM_OP 0x2c
514+#define DRM_RADEON_GEM_USERPTR 0x2d
514515
515516 #define DRM_IOCTL_RADEON_CP_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_RADEON_CP_INIT, drm_radeon_init_t)
516517 #define DRM_IOCTL_RADEON_CP_START DRM_IO( DRM_COMMAND_BASE + DRM_RADEON_CP_START)
@@ -554,6 +555,7 @@ typedef struct { · 37 unmodified lines
554555 #define DRM_IOCTL_RADEON_GEM_BUSY DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_BUSY, struct drm_radeon_gem_busy)
555556 #define DRM_IOCTL_RADEON_GEM_VA DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_VA, struct drm_radeon_gem_va)
556557 #define DRM_IOCTL_RADEON_GEM_OP DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_OP, struct drm_radeon_gem_op)
558+#define DRM_IOCTL_RADEON_GEM_USERPTR DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_USERPTR, struct drm_radeon_gem_userptr)
557559
558560 typedef struct drm_radeon_init {
559561 enum {
@@ -808,6 +810,20 @@ struct drm_radeon_gem_create { · 248 unmodified lines
808810 uint32_t flags;
809811 };
810812
813+/*
814+ * This is not a reliable API and you should expect it to fail for any
815+ * number of reasons and have fallback path that do not use userptr to
816+ * perform any operation.
817+ */
818+#define RADEON_GEM_USERPTR_READONLY (1 << 0)
819+
820+struct drm_radeon_gem_userptr {
821+ uint64_t addr;
822+ uint64_t size;
823+ uint32_t flags;
824+ uint32_t handle;
825+};
826+
811827 #define RADEON_TILING_MACRO 0x1
812828 #define RADEON_TILING_MICRO 0x2
813829 #define RADEON_TILING_SWAP_16BIT 0x4

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t