[PATCH] drm/radeon: add user pointer support v2
Christian König
deathsimple at vodafone.de
Mon Jun 30 06:04:16 PDT 2014
From: Christian König <christian.koenig at amd.com>
This patch adds an IOCTL for turning a pointer supplied by
userspace into a buffer object.
It works similar to the recently added userptr support for i915,
so it also 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.
Exporting and sharing as well as mapping of buffer objects created by
this function is forbidden and results in an -EPERM.
In difference to the i915 implementation we don't use an interval tree to
manage all the buffers objects created and instead just register a separate
MMU notifier callback for each BO so that buffer objects are invalidated
whenever any modification is made to the processes page table. This probably
needs further optimization.
Please review and / or comment.
v2: squash all previous changes into first public version
Signed-off-by: Christian König <christian.koenig at amd.com>
---
drivers/gpu/drm/Kconfig | 1 +
drivers/gpu/drm/radeon/radeon.h | 8 +++
drivers/gpu/drm/radeon/radeon_cs.c | 16 ++++-
drivers/gpu/drm/radeon/radeon_drv.c | 5 +-
drivers/gpu/drm/radeon/radeon_gem.c | 58 ++++++++++++++++++
drivers/gpu/drm/radeon/radeon_kms.c | 1 +
drivers/gpu/drm/radeon/radeon_object.c | 51 ++++++++++++++++
drivers/gpu/drm/radeon/radeon_object.h | 1 +
drivers/gpu/drm/radeon/radeon_prime.c | 10 +++
drivers/gpu/drm/radeon/radeon_ttm.c | 108 +++++++++++++++++++++++++++++++++
include/uapi/drm/radeon_drm.h | 9 +++
11 files changed, 265 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index f512004..1b3c162 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -114,6 +114,7 @@ config DRM_RADEON
select POWER_SUPPLY
select HWMON
select BACKLIGHT_CLASS_DEVICE
+ select MMU_NOTIFIER
help
Choose this option if you have an ATI Radeon graphics card. There
are both PCI and AGP versions. You don't need to choose this to
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 4b0bbf8..09a7be1 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -64,6 +64,7 @@
#include <linux/wait.h>
#include <linux/list.h>
#include <linux/kref.h>
+#include <linux/mmu_notifier.h>
#include <ttm/ttm_bo_api.h>
#include <ttm/ttm_bo_driver.h>
@@ -478,6 +479,9 @@ struct radeon_bo {
struct ttm_bo_kmap_obj dma_buf_vmap;
pid_t pid;
+
+ /* MM callback for user pointer support */
+ struct mmu_notifier mn;
};
#define gem_to_radeon_bo(gobj) container_of((gobj), struct radeon_bo, gem_base)
@@ -2115,6 +2119,8 @@ int radeon_gem_info_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp);
int radeon_gem_create_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp);
+int radeon_gem_import_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *filp);
int radeon_gem_pin_ioctl(struct drm_device *dev, void *data,
struct drm_file *file_priv);
int radeon_gem_unpin_ioctl(struct drm_device *dev, void *data,
@@ -2806,6 +2812,8 @@ extern void radeon_legacy_set_clock_gating(struct radeon_device *rdev, int enabl
extern void radeon_atom_set_clock_gating(struct radeon_device *rdev, int enable);
extern void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain);
extern bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo);
+extern int radeon_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t userptr);
+extern bool radeon_ttm_tt_has_userptr(struct ttm_tt *ttm);
extern void radeon_vram_location(struct radeon_device *rdev, struct radeon_mc *mc, u64 base);
extern void radeon_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
extern int radeon_resume_kms(struct drm_device *dev, bool resume, bool fbcon);
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 71a1434..0a195b9 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -78,7 +78,8 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
struct radeon_cs_chunk *chunk;
struct radeon_cs_buckets buckets;
unsigned i, j;
- bool duplicate;
+ bool duplicate, need_mmap_lock = false;
+ int r;
if (p->chunk_relocs_idx == -1) {
return 0;
@@ -100,6 +101,7 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
for (i = 0; i < p->nrelocs; i++) {
struct drm_radeon_cs_reloc *r;
+ struct ttm_tt *ttm;
unsigned priority;
duplicate = false;
@@ -126,6 +128,9 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
p->relocs_ptr[i] = &p->relocs[i];
p->relocs[i].robj = gem_to_radeon_bo(p->relocs[i].gobj);
+ ttm = p->relocs[i].robj->tbo.ttm;
+ need_mmap_lock |= radeon_ttm_tt_has_userptr(ttm);
+
/* The userspace buffer priorities are from 0 to 15. A higher
* number means the buffer is more important.
* Also, the buffers used for write have a higher priority than
@@ -176,8 +181,15 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
if (p->cs_flags & RADEON_CS_USE_VM)
p->vm_bos = radeon_vm_get_bos(p->rdev, p->ib.vm,
&p->validated);
+ if (need_mmap_lock)
+ down_read(¤t->mm->mmap_sem);
+
+ r = radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
- return radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
+ if (need_mmap_lock)
+ up_read(¤t->mm->mmap_sem);
+
+ return r;
}
static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)
diff --git a/drivers/gpu/drm/radeon/radeon_drv.c b/drivers/gpu/drm/radeon/radeon_drv.c
index 6e30174..355aa67 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.c
+++ b/drivers/gpu/drm/radeon/radeon_drv.c
@@ -112,6 +112,9 @@ int radeon_gem_object_open(struct drm_gem_object *obj,
struct drm_file *file_priv);
void radeon_gem_object_close(struct drm_gem_object *obj,
struct drm_file *file_priv);
+struct dma_buf *radeon_gem_prime_export(struct drm_device *dev,
+ struct drm_gem_object *gobj,
+ int flags);
extern int radeon_get_crtc_scanoutpos(struct drm_device *dev, int crtc,
unsigned int flags,
int *vpos, int *hpos, ktime_t *stime,
@@ -558,7 +561,7 @@ static struct drm_driver kms_driver = {
.prime_handle_to_fd = drm_gem_prime_handle_to_fd,
.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
- .gem_prime_export = drm_gem_prime_export,
+ .gem_prime_export = radeon_gem_prime_export,
.gem_prime_import = drm_gem_prime_import,
.gem_prime_pin = radeon_gem_prime_pin,
.gem_prime_unpin = radeon_gem_prime_unpin,
diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c
index d09650c..ac939a8 100644
--- a/drivers/gpu/drm/radeon/radeon_gem.c
+++ b/drivers/gpu/drm/radeon/radeon_gem.c
@@ -272,6 +272,60 @@ int radeon_gem_create_ioctl(struct drm_device *dev, void *data,
return 0;
}
+int radeon_gem_import_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *filp)
+{
+ struct radeon_device *rdev = dev->dev_private;
+ struct drm_radeon_gem_import *args = data;
+ struct drm_gem_object *gobj;
+ struct radeon_bo *bo;
+ uint32_t handle;
+ int r;
+
+ if (offset_in_page(args->addr | args->size))
+ return -EINVAL;
+
+ if (!access_ok(VERIFY_WRITE, (char __user *)args->addr, args->size))
+ return -EFAULT;
+
+ down_read(&rdev->exclusive_lock);
+
+ /* create a gem object to contain this object in */
+ r = radeon_gem_object_create(rdev, args->size, 0,
+ RADEON_GEM_DOMAIN_CPU,
+ false, false, &gobj);
+ if (r)
+ goto handle_lockup;
+
+ bo = gem_to_radeon_bo(gobj);
+ r = radeon_ttm_tt_set_userptr(bo->tbo.ttm, args->addr);
+ if (r)
+ goto release_object;
+
+ r = radeon_bo_register_mm_callback(bo);
+ if (r)
+ goto release_object;
+
+ r = drm_gem_handle_create(filp, gobj, &handle);
+ /* drop reference from allocate - handle holds it now */
+ drm_gem_object_unreference_unlocked(gobj);
+ if (r)
+ goto handle_lockup;
+
+ args->handle = handle;
+ up_read(&rdev->exclusive_lock);
+ return 0;
+
+release_object:
+ drm_gem_object_unreference_unlocked(gobj);
+
+handle_lockup:
+ up_read(&rdev->exclusive_lock);
+ r = radeon_gem_handle_lockup(rdev, r);
+
+ return r;
+}
+
int radeon_gem_set_domain_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp)
{
@@ -315,6 +369,8 @@ int radeon_mode_dumb_mmap(struct drm_file *filp,
return -ENOENT;
}
robj = gem_to_radeon_bo(gobj);
+ if (radeon_ttm_tt_has_userptr(robj->tbo.ttm))
+ return -EPERM;
*offset_p = radeon_bo_mmap_offset(robj);
drm_gem_object_unreference_unlocked(gobj);
return 0;
@@ -535,6 +591,8 @@ int radeon_gem_op_ioctl(struct drm_device *dev, void *data,
return -ENOENT;
}
robj = gem_to_radeon_bo(gobj);
+ if (radeon_ttm_tt_has_userptr(robj->tbo.ttm))
+ return -EPERM;
r = radeon_bo_reserve(robj, false);
if (unlikely(r))
goto out;
diff --git a/drivers/gpu/drm/radeon/radeon_kms.c b/drivers/gpu/drm/radeon/radeon_kms.c
index 35d9318..39e8a5c 100644
--- a/drivers/gpu/drm/radeon/radeon_kms.c
+++ b/drivers/gpu/drm/radeon/radeon_kms.c
@@ -874,5 +874,6 @@ const struct drm_ioctl_desc radeon_ioctls_kms[] = {
DRM_IOCTL_DEF_DRV(RADEON_GEM_BUSY, radeon_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_VA, radeon_gem_va_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
DRM_IOCTL_DEF_DRV(RADEON_GEM_OP, radeon_gem_op_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(RADEON_GEM_IMPORT, radeon_gem_import_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
};
int radeon_max_kms_ioctl = ARRAY_SIZE(radeon_ioctls_kms);
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index 6c717b2..4bf5335 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -86,6 +86,9 @@ static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo)
radeon_update_memory_usage(bo, bo->tbo.mem.mem_type, -1);
+ if (bo->mn.ops)
+ mmu_notifier_unregister(&bo->mn, current->mm);
+
mutex_lock(&bo->rdev->gem.mutex);
list_del_init(&bo->list);
mutex_unlock(&bo->rdev->gem.mutex);
@@ -253,6 +256,9 @@ int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset,
{
int r, i;
+ if (radeon_ttm_tt_has_userptr(bo->tbo.ttm))
+ return -EPERM;
+
if (bo->pin_count) {
bo->pin_count++;
if (gpu_addr)
@@ -743,3 +749,48 @@ int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type, bool no_wait)
ttm_bo_unreserve(&bo->tbo);
return r;
}
+
+static void radeon_bo_mn_release(struct mmu_notifier *mn,
+ struct mm_struct *mm)
+{
+ struct radeon_bo *bo = container_of(mn, struct radeon_bo, mn);
+
+ bo->mn.ops = NULL;
+}
+
+static void radeon_bo_mn_invalidate_range_start(struct mmu_notifier *mn,
+ struct mm_struct *mm,
+ unsigned long start,
+ unsigned long end)
+{
+ struct radeon_bo *bo = container_of(mn, struct radeon_bo, mn);
+ int r;
+
+ /* TODO: optimize! */
+
+ r = radeon_bo_reserve(bo, true);
+ if (r) {
+ dev_err(bo->rdev->dev, "(%d) failed to reserve user bo\n", r);
+ return;
+ }
+
+ radeon_ttm_placement_from_domain(bo, RADEON_GEM_DOMAIN_CPU);
+ r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
+ if (r) {
+ dev_err(bo->rdev->dev, "(%d) failed to validate user bo\n", r);
+ return;
+ }
+
+ radeon_bo_unreserve(bo);
+}
+
+static const struct mmu_notifier_ops radeon_bo_notifier = {
+ .release = radeon_bo_mn_release,
+ .invalidate_range_start = radeon_bo_mn_invalidate_range_start,
+};
+
+int radeon_bo_register_mm_callback(struct radeon_bo *bo)
+{
+ bo->mn.ops = &radeon_bo_notifier;
+ return mmu_notifier_register(&bo->mn, current->mm);
+}
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index 5a873f3..b10368a 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -153,6 +153,7 @@ extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
struct ttm_mem_reg *new_mem);
extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
+extern int radeon_bo_register_mm_callback(struct radeon_bo *bo);
/*
* sub allocation
diff --git a/drivers/gpu/drm/radeon/radeon_prime.c b/drivers/gpu/drm/radeon/radeon_prime.c
index 2007456..1f0d8f7 100644
--- a/drivers/gpu/drm/radeon/radeon_prime.c
+++ b/drivers/gpu/drm/radeon/radeon_prime.c
@@ -103,3 +103,13 @@ void radeon_gem_prime_unpin(struct drm_gem_object *obj)
radeon_bo_unpin(bo);
radeon_bo_unreserve(bo);
}
+
+struct dma_buf *radeon_gem_prime_export(struct drm_device *dev,
+ struct drm_gem_object *gobj,
+ int flags)
+{
+ struct radeon_bo *bo = gem_to_radeon_bo(gobj);
+ if (radeon_ttm_tt_has_userptr(bo->tbo.ttm))
+ return ERR_PTR(-EPERM);
+ return drm_gem_prime_export(dev, gobj, flags);
+}
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index c8a8a51..37c5e19 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -515,6 +515,9 @@ struct radeon_ttm_tt {
struct ttm_dma_tt ttm;
struct radeon_device *rdev;
u64 offset;
+
+ uint64_t userptr;
+ struct mm_struct *usermm;
};
static int radeon_ttm_backend_bind(struct ttm_tt *ttm,
@@ -588,6 +591,77 @@ static struct ttm_tt *radeon_ttm_tt_create(struct ttm_bo_device *bdev,
return >t->ttm.ttm;
}
+static int radeon_ttm_tt_pin_userptr(struct ttm_tt *ttm)
+{
+ struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
+ struct radeon_ttm_tt *gtt = (void *)ttm;
+ unsigned pinned = 0, nents;
+ int r;
+
+ /* prepare the sg table with the user pages */
+ if (current->mm != gtt->usermm)
+ return -EPERM;
+
+ do {
+ unsigned num_pages = ttm->num_pages - pinned;
+ uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
+ struct page **pages = ttm->pages + pinned;
+
+ r = get_user_pages(current, current->mm, userptr, num_pages,
+ 1, 0, pages, NULL);
+ if (r < 0) {
+ release_pages(ttm->pages, pinned, 0);
+ return r;
+ }
+ pinned += r;
+
+ } while (pinned < ttm->num_pages);
+
+ r = -ENOMEM;
+ ttm->sg = kcalloc(1, sizeof(struct sg_table), GFP_KERNEL);
+ if (!ttm->sg)
+ goto release_pages;
+
+ r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
+ ttm->num_pages << PAGE_SHIFT,
+ GFP_KERNEL);
+ if (r)
+ goto release_sg;
+
+ r = -ENOMEM;
+ nents = dma_map_sg(rdev->dev, ttm->sg->sgl, ttm->sg->nents,
+ DMA_BIDIRECTIONAL);
+ if (nents != ttm->sg->nents)
+ goto release_sg;
+
+ return 0;
+
+release_sg:
+ kfree(ttm->sg);
+
+release_pages:
+ release_pages(ttm->pages, pinned, 0);
+ return r;
+}
+
+static void radeon_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
+{
+ struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
+ unsigned i;
+
+ /* free the sg table and pages again */
+ dma_unmap_sg(rdev->dev, ttm->sg->sgl,
+ ttm->sg->nents, DMA_BIDIRECTIONAL);
+
+ sg_free_table(ttm->sg);
+ kfree(ttm->sg);
+
+ for (i = 0; i < ttm->num_pages; ++i)
+ set_page_dirty(ttm->pages[i]);
+
+ release_pages(ttm->pages, ttm->num_pages, 0);
+}
+
static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
{
struct radeon_device *rdev;
@@ -599,6 +673,13 @@ static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
if (ttm->state != tt_unpopulated)
return 0;
+ if (gtt->userptr) {
+ r = radeon_ttm_tt_pin_userptr(ttm);
+ if (r)
+ return r;
+ slave = true;
+ }
+
if (slave && ttm->sg) {
drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
gtt->ttm.dma_address, ttm->num_pages);
@@ -648,6 +729,11 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
unsigned i;
bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);
+ if (gtt->userptr) {
+ radeon_ttm_tt_unpin_userptr(ttm);
+ return;
+ }
+
if (slave)
return;
@@ -676,6 +762,28 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
ttm_pool_unpopulate(ttm);
}
+int radeon_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t userptr)
+{
+ struct radeon_ttm_tt *gtt = (void *)ttm;
+
+ if (gtt == NULL)
+ return -EINVAL;
+
+ gtt->userptr = userptr;
+ gtt->usermm = current->mm;
+ return 0;
+}
+
+bool radeon_ttm_tt_has_userptr(struct ttm_tt *ttm)
+{
+ struct radeon_ttm_tt *gtt = (void *)ttm;
+
+ if (gtt == NULL)
+ return false;
+
+ return !!gtt->userptr;
+}
+
static struct ttm_bo_driver radeon_bo_driver = {
.ttm_tt_create = &radeon_ttm_tt_create,
.ttm_tt_populate = &radeon_ttm_tt_populate,
diff --git a/include/uapi/drm/radeon_drm.h b/include/uapi/drm/radeon_drm.h
index 1cc0b61..54e7421 100644
--- a/include/uapi/drm/radeon_drm.h
+++ b/include/uapi/drm/radeon_drm.h
@@ -511,6 +511,7 @@ typedef struct {
#define DRM_RADEON_GEM_BUSY 0x2a
#define DRM_RADEON_GEM_VA 0x2b
#define DRM_RADEON_GEM_OP 0x2c
+#define DRM_RADEON_GEM_IMPORT 0x2d
#define DRM_IOCTL_RADEON_CP_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_RADEON_CP_INIT, drm_radeon_init_t)
#define DRM_IOCTL_RADEON_CP_START DRM_IO( DRM_COMMAND_BASE + DRM_RADEON_CP_START)
@@ -554,6 +555,7 @@ typedef struct {
#define DRM_IOCTL_RADEON_GEM_BUSY DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_BUSY, struct drm_radeon_gem_busy)
#define DRM_IOCTL_RADEON_GEM_VA DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_VA, struct drm_radeon_gem_va)
#define DRM_IOCTL_RADEON_GEM_OP DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_OP, struct drm_radeon_gem_op)
+#define DRM_IOCTL_RADEON_GEM_IMPORT DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_IMPORT, struct drm_radeon_gem_import)
typedef struct drm_radeon_init {
enum {
@@ -806,6 +808,13 @@ struct drm_radeon_gem_create {
uint32_t flags;
};
+struct drm_radeon_gem_import {
+ uint64_t addr;
+ uint64_t size;
+ uint32_t flags;
+ uint32_t handle;
+};
+
#define RADEON_TILING_MACRO 0x1
#define RADEON_TILING_MICRO 0x2
#define RADEON_TILING_SWAP_16BIT 0x4
--
1.9.1
More information about the dri-devel
mailing list