2009-11-20 13:29:29

by Jerome Glisse

Subject:

This patch series add ttm range validation function. Aim is to
include this in 2.6.33 so i have time to iron out issue, comments.

ttm:
I duplicated a bunch of ttm functions but now i think, best would
be to add range to all function and use free list if range cover
all the manager space. Doing so we might also be able to simplify
mem_space alocation into a simpler function like ttm_bo_mem_space_range

radeon:
The second patch is a rework/cleanup of radeon object, it solves
few issues along the way (i can't remember them now after fews
days testing the patches). Biggest change is that we now rely
on BO being validated before doing any change to radeon bo structure.
As with any big patch i might introduce regressions, so far after
testing on AGP:R1XX,R2XX,R3XX,R6XX PCIE:R3XX,R4XX,R5XX,R6XX,R7XX
and RS480,RS690 i didn't found anythings obvious (test being X +
glxgears + compiz(on hw which support it) + suspend/resume).

Last patch is smaller, it just use the interface introduced by
the first patch.

2009-11-20 13:29:44

by Jerome Glisse

Subject: [PATCH 1/3] drm/ttm: Add range validation function

Add a function to validate buffer in a given range of
memory. This is needed by some hw for some of their
buffer (scanout buffer, cursor ...).

Signed-off-by: Jerome Glisse <[email protected]>
---
drivers/gpu/drm/ttm/ttm_bo.c | 305 ++++++++++++++++++++++++++++++++++++++-
include/drm/ttm/ttm_bo_api.h | 5 +
include/drm/ttm/ttm_bo_driver.h | 1 +
3 files changed, 310 insertions(+), 1 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 87c0625..6b0e7e8 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -247,7 +247,6 @@ EXPORT_SYMBOL(ttm_bo_unreserve);
/*
* Call bo->mutex locked.
*/
-
static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
{
struct ttm_bo_device *bdev = bo->bdev;
@@ -418,6 +417,7 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo, bool remove_all)
kref_put(&bo->list_kref, ttm_bo_ref_bug);
}
if (bo->mem.mm_node) {
+ bo->mem.mm_node->private = NULL;
drm_mm_put_block(bo->mem.mm_node);
bo->mem.mm_node = NULL;
}
@@ -610,6 +610,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, unsigned mem_type,

spin_lock(&glob->lru_lock);
if (evict_mem.mm_node) {
+ evict_mem.mm_node->private = NULL;
drm_mm_put_block(evict_mem.mm_node);
evict_mem.mm_node = NULL;
}
@@ -826,6 +827,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,
mem->mm_node = node;
mem->mem_type = mem_type;
mem->placement = cur_flags;
+ node->private = bo;
return 0;
}

@@ -856,6 +858,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,

if (ret == 0 && mem->mm_node) {
mem->placement = cur_flags;
+ mem->mm_node->private = bo;
return 0;
}

@@ -868,6 +871,173 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,
}
EXPORT_SYMBOL(ttm_bo_mem_space);

+static unsigned long ttm_bo_free_size_if_evicted(struct ttm_buffer_object *bo)
+{
+ struct ttm_mem_type_manager *man = &bo->bdev->man[bo->mem.mem_type];
+ struct drm_mm_node *node;
+ unsigned long size;
+
+ size = bo->mem.mm_node->size;
+ if (bo->mem.mm_node->ml_entry.prev != &man->manager.ml_entry) {
+ node = list_entry(bo->mem.mm_node->ml_entry.prev,
+ struct drm_mm_node, ml_entry);
+ if (node->free)
+ size += node->size;
+ }
+ if (bo->mem.mm_node->ml_entry.next != &man->manager.ml_entry) {
+ node = list_entry(bo->mem.mm_node->ml_entry.next,
+ struct drm_mm_node, ml_entry);
+ if (node->free)
+ size += node->size;
+ }
+ return size;
+}
+
+static void ttm_manager_evict_first(struct ttm_buffer_object *bo)
+{
+ struct ttm_mem_type_manager *man = &bo->bdev->man[bo->mem.mem_type];
+ unsigned long free_size_bo, free_size_bo_first;
+ struct ttm_buffer_object *bo_first;
+
+ /* BO is not on lru list, don't add it */
+ if (!list_empty(&bo->lru))
+ return;
+ bo_first = list_first_entry(&man->lru, struct ttm_buffer_object, lru);
+ free_size_bo = ttm_bo_free_size_if_evicted(bo);
+ free_size_bo_first = ttm_bo_free_size_if_evicted(bo_first);
+ if (free_size_bo > free_size_bo_first) {
+ list_del_init(&bo->lru);
+ list_add(&bo->lru, &man->lru);
+ }
+}
+
+static int ttm_manager_space_range(struct ttm_buffer_object *bo,
+ uint32_t mem_type,
+ struct ttm_mem_reg *mem,
+ unsigned long start, unsigned long end,
+ bool interruptible, bool no_wait)
+{
+ struct ttm_bo_global *glob = bo->glob;
+ struct drm_mm_node *entry;
+ struct ttm_mem_type_manager *man = &bo->bdev->man[mem_type];
+ struct drm_mm *mm = &man->manager;
+ unsigned size = end - start;
+ struct ttm_buffer_object *tbo;
+ unsigned wasted;
+ int ret;
+
+ mem->mm_node = NULL;
+ ret = drm_mm_pre_get(&man->manager);
+ if (unlikely(ret))
+ return ret;
+ spin_lock(&glob->lru_lock);
+ list_for_each_entry(entry, &mm->ml_entry, ml_entry) {
+ wasted = 0;
+ if (mem->page_alignment) {
+ unsigned tmp = entry->start % mem->page_alignment;
+ if (tmp)
+ wasted += mem->page_alignment - tmp;
+ }
+ if (entry->start < end && (entry->start+entry->size) > start) {
+ if (!entry->free) {
+ tbo = (struct ttm_buffer_object*)entry->private;
+ ttm_manager_evict_first(tbo);
+ } else {
+ if ((entry->size - wasted) <= size) {
+ /* Found a place */
+ entry = drm_mm_get_block_atomic(entry,
+ mem->num_pages,
+ mem->page_alignment);
+ mem->mm_node = entry;
+ mem->mem_type = mem_type;
+ entry->private = bo;
+ break;
+ }
+ }
+ }
+ }
+ spin_unlock(&glob->lru_lock);
+ return 0;
+}
+
+static int ttm_bo_mem_space_range(struct ttm_buffer_object *bo,
+ uint32_t proposed_placement,
+ struct ttm_mem_reg *mem,
+ unsigned long start, unsigned long end,
+ bool interruptible, bool no_wait)
+{
+ struct ttm_bo_device *bdev = bo->bdev;
+ struct ttm_bo_global *glob = bo->glob;
+ struct ttm_mem_type_manager *man;
+ uint32_t num_prios = bdev->driver->num_mem_type_prio;
+ const uint32_t *prios = bdev->driver->mem_type_prio;
+ uint32_t i;
+ uint32_t mem_type = TTM_PL_SYSTEM;
+ uint32_t cur_flags = 0;
+ bool type_found = false;
+ bool type_ok = false;
+ bool evicted;
+ int ret;
+
+ mem->mm_node = NULL;
+retry:
+ for (i = 0; i < num_prios; ++i) {
+ mem_type = prios[i];
+ man = &bdev->man[mem_type];
+ type_ok = ttm_bo_mt_compatible(man,
+ bo->type == ttm_bo_type_user,
+ mem_type, proposed_placement,
+ &cur_flags);
+ if (!type_ok)
+ continue;
+ if (start > (man->offset + man->size) || end < man->offset)
+ continue;
+ cur_flags = ttm_bo_select_caching(man, bo->mem.placement,
+ cur_flags);
+ if (mem_type == TTM_PL_SYSTEM)
+ break;
+ if (man->has_type && man->use_type) {
+ type_found = true;
+ mem->placement = cur_flags;
+ ret = ttm_manager_space_range(bo, mem_type, mem,
+ start, end,
+ interruptible, no_wait);
+ if (ret)
+ return ret;
+ }
+ if (mem->mm_node)
+ break;
+ }
+ if ((type_ok && (mem_type == TTM_PL_SYSTEM)) || mem->mm_node) {
+ mem->mem_type = mem_type;
+ mem->placement = cur_flags;
+ return 0;
+ }
+ for (i = 0, evicted = false; i < num_prios; ++i) {
+ struct ttm_buffer_object *tmp;
+
+ mem_type = prios[i];
+ man = &bdev->man[mem_type];
+ type_ok = ttm_bo_mt_compatible(man,
+ bo->type == ttm_bo_type_user,
+ mem_type, proposed_placement,
+ &cur_flags);
+ if (!type_ok)
+ continue;
+ if (start > (man->offset + man->size) || end < man->offset)
+ continue;
+ spin_lock(&glob->lru_lock);
+ tmp = list_first_entry(&man->lru, struct ttm_buffer_object, lru);
+ spin_unlock(&glob->lru_lock);
+ ret = ttm_bo_evict(tmp, mem_type, interruptible, no_wait);
+ if (unlikely(ret != 0))
+ return ret;
+ }
+ if (!evicted)
+ return -ENOMEM;
+ goto retry;
+}
+
int ttm_bo_wait_cpu(struct ttm_buffer_object *bo, bool no_wait)
{
int ret = 0;
@@ -925,6 +1095,7 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
out_unlock:
if (ret && mem.mm_node) {
spin_lock(&glob->lru_lock);
+ mem.mm_node->private = NULL;
drm_mm_put_block(mem.mm_node);
spin_unlock(&glob->lru_lock);
}
@@ -998,6 +1169,137 @@ int ttm_buffer_object_validate(struct ttm_buffer_object *bo,
}
EXPORT_SYMBOL(ttm_buffer_object_validate);

+int ttm_bo_move_buffer_range(struct ttm_buffer_object *bo,
+ uint32_t proposed_placement,
+ unsigned long start, unsigned long end,
+ bool interruptible, bool no_wait)
+{
+ struct ttm_bo_global *glob = bo->glob;
+ int ret = 0;
+ struct ttm_mem_reg mem;
+
+ BUG_ON(!atomic_read(&bo->reserved));
+
+ /*
+ * FIXME: It's possible to pipeline buffer moves.
+ * Have the driver move function wait for idle when necessary,
+ * instead of doing it here.
+ */
+
+ spin_lock(&bo->lock);
+ ret = ttm_bo_wait(bo, false, interruptible, no_wait);
+ spin_unlock(&bo->lock);
+
+ if (ret)
+ return ret;
+
+ mem.num_pages = bo->num_pages;
+ mem.size = mem.num_pages << PAGE_SHIFT;
+ mem.page_alignment = bo->mem.page_alignment;
+
+ /*
+ * Determine where to move the buffer.
+ */
+
+ ret = ttm_bo_mem_space_range(bo, proposed_placement, &mem,
+ start, end, interruptible, no_wait);
+ if (ret)
+ goto out_unlock;
+
+ ret = ttm_bo_handle_move_mem(bo, &mem, false, interruptible, no_wait);
+
+out_unlock:
+ if (ret && mem.mm_node) {
+ spin_lock(&glob->lru_lock);
+ mem.mm_node->private = NULL;
+ drm_mm_put_block(mem.mm_node);
+ spin_unlock(&glob->lru_lock);
+ }
+ return ret;
+}
+
+static int ttm_bo_mem_compat_range(uint32_t proposed_placement,
+ struct ttm_mem_reg *mem,
+ unsigned long start, unsigned long end)
+{
+ if ((proposed_placement & mem->placement & TTM_PL_MASK_MEM) == 0)
+ return 0;
+ if ((proposed_placement & mem->placement & TTM_PL_MASK_CACHING) == 0)
+ return 0;
+ if (mem->mm_node) {
+ unsigned long eend = mem->mm_node->start + mem->mm_node->size;
+ if (mem->mm_node->start < start || eend > end)
+ return 0;
+ }
+
+ return 1;
+}
+
+int ttm_buffer_object_validate_range(struct ttm_buffer_object *bo,
+ uint32_t proposed_placement,
+ unsigned long start, unsigned long end,
+ bool interruptible, bool no_wait)
+{
+ int ret;
+
+ /* Convert to page */
+ start = start >> PAGE_SHIFT;
+ end = end >> PAGE_SHIFT;
+ /* Check that range is valid */
+ if (start > end || (end - start) < bo->num_pages)
+ return -EINVAL;
+
+ BUG_ON(!atomic_read(&bo->reserved));
+ bo->proposed_placement = proposed_placement;
+
+ TTM_DEBUG("Proposed placement 0x%08lx, Old flags 0x%08lx\n",
+ (unsigned long)proposed_placement,
+ (unsigned long)bo->mem.placement);
+
+ /*
+ * Check whether we need to move buffer.
+ */
+
+ if (!ttm_bo_mem_compat_range(bo->proposed_placement, &bo->mem,
+ start, end)) {
+ ret = ttm_bo_move_buffer_range(bo, bo->proposed_placement,
+ start, end,
+ interruptible, no_wait);
+ if (ret) {
+ if (ret != -ERESTART)
+ printk(KERN_ERR TTM_PFX
+ "Failed moving buffer. "
+ "Proposed placement 0x%08x\n",
+ bo->proposed_placement);
+ if (ret == -ENOMEM)
+ printk(KERN_ERR TTM_PFX
+ "Out of aperture space or "
+ "DRM memory quota.\n");
+ return ret;
+ }
+ }
+
+ /*
+ * We might need to add a TTM.
+ */
+ if (bo->mem.mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
+ ret = ttm_bo_add_ttm(bo, true);
+ if (ret)
+ return ret;
+ }
+ /*
+ * Validation has succeeded, move the access and other
+ * non-mapping-related flag bits from the proposed flags to
+ * the active flags
+ */
+
+ ttm_flag_masked(&bo->mem.placement, bo->proposed_placement,
+ ~TTM_PL_MASK_MEMTYPE);
+
+ return 0;
+}
+EXPORT_SYMBOL(ttm_buffer_object_validate_range);
+
int
ttm_bo_check_placement(struct ttm_buffer_object *bo,
uint32_t set_flags, uint32_t clr_flags)
@@ -1321,6 +1623,7 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
man->has_type = true;
man->use_type = true;
man->size = p_size;
+ man->offset = p_offset;

INIT_LIST_HEAD(&man->lru);

diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 4911461..3756970 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -308,6 +308,11 @@ extern int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy,
extern int ttm_buffer_object_validate(struct ttm_buffer_object *bo,
uint32_t proposed_placement,
bool interruptible, bool no_wait);
+extern int ttm_buffer_object_validate_range(struct ttm_buffer_object *bo,
+ uint32_t proposed_placement,
+ unsigned long start, unsigned long end,
+ bool interruptible, bool no_wait);
+
/**
* ttm_bo_unref
*
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index e8cd6d2..f00ba12 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -224,6 +224,7 @@ struct ttm_mem_type_manager {
unsigned long gpu_offset;
unsigned long io_offset;
unsigned long io_size;
+ unsigned long offset;
void *io_addr;
uint64_t size;
uint32_t available_caching;
--
1.6.5.2

2009-11-20 13:30:24

by Jerome Glisse

Subject: [PATCH 2/3] drm/radeon/kms: Rework radeon object handling

The locking & protection of radeon object was somewhat messy.
This patch completely rework it to now use ttm reserve as a
protection for the radeon object structure member. It also
shrink down the various radeon object structure by removing
field which were redondant with the ttm information. Last it
converts few simple functions to inline which should with
performances.

Signed-off-by: Jerome Glisse <[email protected]>
---
drivers/gpu/drm/radeon/atombios_crtc.c | 30 +-
drivers/gpu/drm/radeon/r100.c | 90 +++--
drivers/gpu/drm/radeon/r100_track.h | 10 +-
drivers/gpu/drm/radeon/r300.c | 15 +-
drivers/gpu/drm/radeon/r420.c | 4 +-
drivers/gpu/drm/radeon/r520.c | 2 +-
drivers/gpu/drm/radeon/r600.c | 65 +++-
drivers/gpu/drm/radeon/r600_blit_kms.c | 30 +-
drivers/gpu/drm/radeon/radeon.h | 111 +++---
drivers/gpu/drm/radeon/radeon_benchmark.c | 36 ++-
drivers/gpu/drm/radeon/radeon_cs.c | 13 +-
drivers/gpu/drm/radeon/radeon_device.c | 16 +-
drivers/gpu/drm/radeon/radeon_fb.c | 63 ++--
drivers/gpu/drm/radeon/radeon_gart.c | 42 ++-
drivers/gpu/drm/radeon/radeon_gem.c | 90 +++---
drivers/gpu/drm/radeon/radeon_legacy_crtc.c | 27 +-
drivers/gpu/drm/radeon/radeon_object.c | 538 +++++++++++----------------
drivers/gpu/drm/radeon/radeon_object.h | 157 +++++++-
drivers/gpu/drm/radeon/radeon_ring.c | 67 ++--
drivers/gpu/drm/radeon/radeon_test.c | 55 ++--
drivers/gpu/drm/radeon/radeon_ttm.c | 36 ++-
drivers/gpu/drm/radeon/rs400.c | 4 +-
drivers/gpu/drm/radeon/rs600.c | 15 +-
drivers/gpu/drm/radeon/rs690.c | 4 +-
drivers/gpu/drm/radeon/rv515.c | 6 +-
drivers/gpu/drm/radeon/rv770.c | 30 ++-
26 files changed, 873 insertions(+), 683 deletions(-)

diff --git a/drivers/gpu/drm/radeon/atombios_crtc.c b/drivers/gpu/drm/radeon/atombios_crtc.c
index c15287a..80d7f46 100644
--- a/drivers/gpu/drm/radeon/atombios_crtc.c
+++ b/drivers/gpu/drm/radeon/atombios_crtc.c
@@ -574,21 +574,31 @@ int atombios_crtc_set_base(struct drm_crtc *crtc, int x, int y,
struct radeon_device *rdev = dev->dev_private;
struct radeon_framebuffer *radeon_fb;
struct drm_gem_object *obj;
- struct drm_radeon_gem_object *obj_priv;
+ struct radeon_bo *rbo;
uint64_t fb_location;
uint32_t fb_format, fb_pitch_pixels, tiling_flags;
+ int r;

if (!crtc->fb)
return -EINVAL;

radeon_fb = to_radeon_framebuffer(crtc->fb);

+ /* Pin framebuffer & get tilling informations */
obj = radeon_fb->obj;
- obj_priv = obj->driver_private;
-
- if (radeon_gem_object_pin(obj, RADEON_GEM_DOMAIN_VRAM, &fb_location)) {
+ rbo = obj->driver_private;
+ r = radeon_bo_reserve(rbo, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(rbo, RADEON_GEM_DOMAIN_VRAM, &fb_location);
+ if (unlikely(r != 0)) {
+ radeon_bo_unreserve(rbo);
return -EINVAL;
}
+ radeon_bo_get_tiling_flags(rbo, &tiling_flags, NULL);
+ radeon_bo_unreserve(rbo);
+ if (tiling_flags & RADEON_TILING_MACRO)
+ fb_format |= AVIVO_D1GRPH_MACRO_ADDRESS_MODE;

switch (crtc->fb->bits_per_pixel) {
case 8:
@@ -618,11 +628,6 @@ int atombios_crtc_set_base(struct drm_crtc *crtc, int x, int y,
return -EINVAL;
}

- radeon_object_get_tiling_flags(obj->driver_private,
- &tiling_flags, NULL);
- if (tiling_flags & RADEON_TILING_MACRO)
- fb_format |= AVIVO_D1GRPH_MACRO_ADDRESS_MODE;
-
if (tiling_flags & RADEON_TILING_MICRO)
fb_format |= AVIVO_D1GRPH_TILED;

@@ -674,7 +679,12 @@ int atombios_crtc_set_base(struct drm_crtc *crtc, int x, int y,

if (old_fb && old_fb != crtc->fb) {
radeon_fb = to_radeon_framebuffer(old_fb);
- radeon_gem_object_unpin(radeon_fb->obj);
+ rbo = radeon_fb->obj->driver_private;
+ r = radeon_bo_reserve(rbo, false);
+ if (unlikely(r != 0))
+ return r;
+ radeon_bo_unpin(rbo);
+ radeon_bo_unreserve(rbo);
}

/* Bytes per pixel may have changed */
diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index 5e821a3..5b83583 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -255,24 +255,27 @@ int r100_wb_init(struct radeon_device *rdev)
int r;

if (rdev->wb.wb_obj == NULL) {
- r = radeon_object_create(rdev, NULL, RADEON_GPU_PAGE_SIZE,
- true,
- RADEON_GEM_DOMAIN_GTT,
- false, &rdev->wb.wb_obj);
+ r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
+ RADEON_GEM_DOMAIN_GTT,
+ &rdev->wb.wb_obj);
if (r) {
- DRM_ERROR("radeon: failed to create WB buffer (%d).\n", r);
+ dev_err(rdev->dev, "(%d) create WB buffer failed\n", r);
return r;
}
- r = radeon_object_pin(rdev->wb.wb_obj,
- RADEON_GEM_DOMAIN_GTT,
- &rdev->wb.gpu_addr);
+ r = radeon_bo_reserve(rdev->wb.wb_obj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
+ &rdev->wb.gpu_addr);
if (r) {
- DRM_ERROR("radeon: failed to pin WB buffer (%d).\n", r);
+ dev_err(rdev->dev, "(%d) pin WB buffer failed\n", r);
+ radeon_bo_unreserve(rdev->wb.wb_obj);
return r;
}
- r = radeon_object_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
+ r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
+ radeon_bo_unreserve(rdev->wb.wb_obj);
if (r) {
- DRM_ERROR("radeon: failed to map WB buffer (%d).\n", r);
+ dev_err(rdev->dev, "(%d) map WB buffer failed\n", r);
return r;
}
}
@@ -290,11 +293,19 @@ void r100_wb_disable(struct radeon_device *rdev)

void r100_wb_fini(struct radeon_device *rdev)
{
+ int r;
+
r100_wb_disable(rdev);
if (rdev->wb.wb_obj) {
- radeon_object_kunmap(rdev->wb.wb_obj);
- radeon_object_unpin(rdev->wb.wb_obj);
- radeon_object_unref(&rdev->wb.wb_obj);
+ r = radeon_bo_reserve(rdev->wb.wb_obj, false);
+ if (unlikely(r != 0)) {
+ dev_err(rdev->dev, "(%d) can't finish WB\n", r);
+ return;
+ }
+ radeon_bo_kunmap(rdev->wb.wb_obj);
+ radeon_bo_unpin(rdev->wb.wb_obj);
+ radeon_bo_unreserve(rdev->wb.wb_obj);
+ radeon_bo_unref(&rdev->wb.wb_obj);
rdev->wb.wb = NULL;
rdev->wb.wb_obj = NULL;
}
@@ -1288,17 +1299,17 @@ static int r100_packet0_check(struct radeon_cs_parser *p,

int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
struct radeon_cs_packet *pkt,
- struct radeon_object *robj)
+ struct radeon_bo *robj)
{
unsigned idx;
u32 value;
idx = pkt->idx + 1;
value = radeon_get_ib_value(p, idx + 2);
- if ((value + 1) > radeon_object_size(robj)) {
+ if ((value + 1) > radeon_bo_size(robj)) {
DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
"(need %u have %lu) !\n",
value + 1,
- radeon_object_size(robj));
+ radeon_bo_size(robj));
return -EINVAL;
}
return 0;
@@ -2580,7 +2591,7 @@ static int r100_cs_track_cube(struct radeon_device *rdev,
struct r100_cs_track *track, unsigned idx)
{
unsigned face, w, h;
- struct radeon_object *cube_robj;
+ struct radeon_bo *cube_robj;
unsigned long size;

for (face = 0; face < 5; face++) {
@@ -2593,9 +2604,9 @@ static int r100_cs_track_cube(struct radeon_device *rdev,

size += track->textures[idx].cube_info[face].offset;

- if (size > radeon_object_size(cube_robj)) {
+ if (size > radeon_bo_size(cube_robj)) {
DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
- size, radeon_object_size(cube_robj));
+ size, radeon_bo_size(cube_robj));
r100_cs_track_texture_print(&track->textures[idx]);
return -1;
}
@@ -2606,7 +2617,7 @@ static int r100_cs_track_cube(struct radeon_device *rdev,
static int r100_cs_track_texture_check(struct radeon_device *rdev,
struct r100_cs_track *track)
{
- struct radeon_object *robj;
+ struct radeon_bo *robj;
unsigned long size;
unsigned u, i, w, h;
int ret;
@@ -2662,9 +2673,9 @@ static int r100_cs_track_texture_check(struct radeon_device *rdev,
"%u\n", track->textures[u].tex_coord_type, u);
return -EINVAL;
}
- if (size > radeon_object_size(robj)) {
+ if (size > radeon_bo_size(robj)) {
DRM_ERROR("Texture of unit %u needs %lu bytes but is "
- "%lu\n", u, size, radeon_object_size(robj));
+ "%lu\n", u, size, radeon_bo_size(robj));
r100_cs_track_texture_print(&track->textures[u]);
return -EINVAL;
}
@@ -2686,10 +2697,10 @@ int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
}
size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
size += track->cb[i].offset;
- if (size > radeon_object_size(track->cb[i].robj)) {
+ if (size > radeon_bo_size(track->cb[i].robj)) {
DRM_ERROR("[drm] Buffer too small for color buffer %d "
"(need %lu have %lu) !\n", i, size,
- radeon_object_size(track->cb[i].robj));
+ radeon_bo_size(track->cb[i].robj));
DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
i, track->cb[i].pitch, track->cb[i].cpp,
track->cb[i].offset, track->maxy);
@@ -2703,10 +2714,10 @@ int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
}
size = track->zb.pitch * track->zb.cpp * track->maxy;
size += track->zb.offset;
- if (size > radeon_object_size(track->zb.robj)) {
+ if (size > radeon_bo_size(track->zb.robj)) {
DRM_ERROR("[drm] Buffer too small for z buffer "
"(need %lu have %lu) !\n", size,
- radeon_object_size(track->zb.robj));
+ radeon_bo_size(track->zb.robj));
DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
track->zb.pitch, track->zb.cpp,
track->zb.offset, track->maxy);
@@ -2724,11 +2735,12 @@ int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
"bound\n", prim_walk, i);
return -EINVAL;
}
- if (size > radeon_object_size(track->arrays[i].robj)) {
- DRM_ERROR("(PW %u) Vertex array %u need %lu dwords "
- "have %lu dwords\n", prim_walk, i,
- size >> 2,
- radeon_object_size(track->arrays[i].robj) >> 2);
+ if (size > radeon_bo_size(track->arrays[i].robj)) {
+ dev_err(rdev->dev, "(PW %u) Vertex array %u "
+ "need %lu dwords have %lu dwords\n",
+ prim_walk, i, size >> 2,
+ radeon_bo_size(track->arrays[i].robj)
+ >> 2);
DRM_ERROR("Max indices %u\n", track->max_indx);
return -EINVAL;
}
@@ -2742,10 +2754,12 @@ int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
"bound\n", prim_walk, i);
return -EINVAL;
}
- if (size > radeon_object_size(track->arrays[i].robj)) {
- DRM_ERROR("(PW %u) Vertex array %u need %lu dwords "
- "have %lu dwords\n", prim_walk, i, size >> 2,
- radeon_object_size(track->arrays[i].robj) >> 2);
+ if (size > radeon_bo_size(track->arrays[i].robj)) {
+ dev_err(rdev->dev, "(PW %u) Vertex array %u "
+ "need %lu dwords have %lu dwords\n",
+ prim_walk, i, size >> 2,
+ radeon_bo_size(track->arrays[i].robj)
+ >> 2);
return -EINVAL;
}
}
@@ -3160,7 +3174,7 @@ void r100_fini(struct radeon_device *rdev)
r100_pci_gart_fini(rdev);
radeon_irq_kms_fini(rdev);
radeon_fence_driver_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
@@ -3250,7 +3264,7 @@ int r100_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
if (rdev->flags & RADEON_IS_PCI) {
diff --git a/drivers/gpu/drm/radeon/r100_track.h b/drivers/gpu/drm/radeon/r100_track.h
index 0daf0d7..ca50903 100644
--- a/drivers/gpu/drm/radeon/r100_track.h
+++ b/drivers/gpu/drm/radeon/r100_track.h
@@ -10,26 +10,26 @@
* CS functions
*/
struct r100_cs_track_cb {
- struct radeon_object *robj;
+ struct radeon_bo *robj;
unsigned pitch;
unsigned cpp;
unsigned offset;
};

struct r100_cs_track_array {
- struct radeon_object *robj;
+ struct radeon_bo *robj;
unsigned esize;
};

struct r100_cs_cube_info {
- struct radeon_object *robj;
- unsigned offset;
+ struct radeon_bo *robj;
+ unsigned offset;
unsigned width;
unsigned height;
};

struct r100_cs_track_texture {
- struct radeon_object *robj;
+ struct radeon_bo *robj;
struct r100_cs_cube_info cube_info[5]; /* info for 5 non-primary faces */
unsigned pitch;
unsigned width;
diff --git a/drivers/gpu/drm/radeon/r300.c b/drivers/gpu/drm/radeon/r300.c
index 2f43ee8..b76fe28 100644
--- a/drivers/gpu/drm/radeon/r300.c
+++ b/drivers/gpu/drm/radeon/r300.c
@@ -137,14 +137,19 @@ int rv370_pcie_gart_enable(struct radeon_device *rdev)

void rv370_pcie_gart_disable(struct radeon_device *rdev)
{
- uint32_t tmp;
+ u32 tmp;
+ int r;

tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp & ~RADEON_PCIE_TX_GART_EN);
if (rdev->gart.table.vram.robj) {
- radeon_object_kunmap(rdev->gart.table.vram.robj);
- radeon_object_unpin(rdev->gart.table.vram.robj);
+ r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
+ if (likely(r == 0)) {
+ radeon_bo_kunmap(rdev->gart.table.vram.robj);
+ radeon_bo_unpin(rdev->gart.table.vram.robj);
+ radeon_bo_unreserve(rdev->gart.table.vram.robj);
+ }
}
}

@@ -1265,7 +1270,7 @@ void r300_fini(struct radeon_device *rdev)
r100_pci_gart_fini(rdev);
radeon_irq_kms_fini(rdev);
radeon_fence_driver_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
@@ -1325,7 +1330,7 @@ int r300_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
if (rdev->flags & RADEON_IS_PCIE) {
diff --git a/drivers/gpu/drm/radeon/r420.c b/drivers/gpu/drm/radeon/r420.c
index 1cefdbc..dc5e93a 100644
--- a/drivers/gpu/drm/radeon/r420.c
+++ b/drivers/gpu/drm/radeon/r420.c
@@ -258,7 +258,7 @@ void r420_fini(struct radeon_device *rdev)
radeon_agp_fini(rdev);
radeon_irq_kms_fini(rdev);
radeon_fence_driver_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
if (rdev->is_atom_bios) {
radeon_atombios_fini(rdev);
} else {
@@ -331,7 +331,7 @@ int r420_init(struct radeon_device *rdev)
return r;
}
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r) {
return r;
}
diff --git a/drivers/gpu/drm/radeon/r520.c b/drivers/gpu/drm/radeon/r520.c
index f743518..a41e9c7 100644
--- a/drivers/gpu/drm/radeon/r520.c
+++ b/drivers/gpu/drm/radeon/r520.c
@@ -277,7 +277,7 @@ int r520_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
r = rv370_pcie_gart_init(rdev);
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index 8d6bc12..9f1f005 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -180,7 +180,7 @@ int r600_pcie_gart_enable(struct radeon_device *rdev)
void r600_pcie_gart_disable(struct radeon_device *rdev)
{
u32 tmp;
- int i;
+ int i, r;

/* Disable all tables */
for (i = 0; i < 7; i++)
@@ -208,8 +208,12 @@ void r600_pcie_gart_disable(struct radeon_device *rdev)
WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp);
WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
if (rdev->gart.table.vram.robj) {
- radeon_object_kunmap(rdev->gart.table.vram.robj);
- radeon_object_unpin(rdev->gart.table.vram.robj);
+ r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
+ if (likely(r == 0)) {
+ radeon_bo_kunmap(rdev->gart.table.vram.robj);
+ radeon_bo_unpin(rdev->gart.table.vram.robj);
+ radeon_bo_unreserve(rdev->gart.table.vram.robj);
+ }
}
}

@@ -1381,10 +1385,16 @@ int r600_ring_test(struct radeon_device *rdev)

void r600_wb_disable(struct radeon_device *rdev)
{
+ int r;
+
WREG32(SCRATCH_UMSK, 0);
if (rdev->wb.wb_obj) {
- radeon_object_kunmap(rdev->wb.wb_obj);
- radeon_object_unpin(rdev->wb.wb_obj);
+ r = radeon_bo_reserve(rdev->wb.wb_obj, false);
+ if (unlikely(r != 0))
+ return;
+ radeon_bo_kunmap(rdev->wb.wb_obj);
+ radeon_bo_unpin(rdev->wb.wb_obj);
+ radeon_bo_unreserve(rdev->wb.wb_obj);
}
}

@@ -1392,7 +1402,7 @@ void r600_wb_fini(struct radeon_device *rdev)
{
r600_wb_disable(rdev);
if (rdev->wb.wb_obj) {
- radeon_object_unref(&rdev->wb.wb_obj);
+ radeon_bo_unref(&rdev->wb.wb_obj);
rdev->wb.wb = NULL;
rdev->wb.wb_obj = NULL;
}
@@ -1403,22 +1413,29 @@ int r600_wb_enable(struct radeon_device *rdev)
int r;

if (rdev->wb.wb_obj == NULL) {
- r = radeon_object_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
- RADEON_GEM_DOMAIN_GTT, false, &rdev->wb.wb_obj);
+ r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
+ RADEON_GEM_DOMAIN_GTT, &rdev->wb.wb_obj);
if (r) {
- dev_warn(rdev->dev, "failed to create WB buffer (%d).\n", r);
+ dev_warn(rdev->dev, "(%d) create WB bo failed\n", r);
+ return r;
+ }
+ r = radeon_bo_reserve(rdev->wb.wb_obj, false);
+ if (unlikely(r != 0)) {
+ r600_wb_fini(rdev);
return r;
}
- r = radeon_object_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
+ r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
&rdev->wb.gpu_addr);
if (r) {
- dev_warn(rdev->dev, "failed to pin WB buffer (%d).\n", r);
+ radeon_bo_unreserve(rdev->wb.wb_obj);
+ dev_warn(rdev->dev, "(%d) pin WB bo failed\n", r);
r600_wb_fini(rdev);
return r;
}
- r = radeon_object_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
+ r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
+ radeon_bo_unreserve(rdev->wb.wb_obj);
if (r) {
- dev_warn(rdev->dev, "failed to map WB buffer (%d).\n", r);
+ dev_warn(rdev->dev, "(%d) map WB bo failed\n", r);
r600_wb_fini(rdev);
return r;
}
@@ -1516,10 +1533,14 @@ int r600_startup(struct radeon_device *rdev)
}
r600_gpu_init(rdev);

- r = radeon_object_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
- &rdev->r600_blit.shader_gpu_addr);
+ r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
+ &rdev->r600_blit.shader_gpu_addr);
+ radeon_bo_unreserve(rdev->r600_blit.shader_obj);
if (r) {
- DRM_ERROR("failed to pin blit object %d\n", r);
+ dev_err(rdev->dev, "(%d) pin blit object failed\n", r);
return r;
}

@@ -1569,13 +1590,19 @@ int r600_resume(struct radeon_device *rdev)

int r600_suspend(struct radeon_device *rdev)
{
+ int r;
+
/* FIXME: we should wait for ring to be empty */
r600_cp_stop(rdev);
rdev->cp.ready = false;
r600_wb_disable(rdev);
r600_pcie_gart_disable(rdev);
/* unpin shaders bo */
- radeon_object_unpin(rdev->r600_blit.shader_obj);
+ r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
+ if (unlikely(r != 0))
+ return r;
+ radeon_bo_unpin(rdev->r600_blit.shader_obj);
+ radeon_bo_unreserve(rdev->r600_blit.shader_obj);
return 0;
}

@@ -1636,7 +1663,7 @@ int r600_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
rdev->cp.ring_obj = NULL;
@@ -1698,7 +1725,7 @@ void r600_fini(struct radeon_device *rdev)
radeon_clocks_fini(rdev);
if (rdev->flags & RADEON_IS_AGP)
radeon_agp_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
diff --git a/drivers/gpu/drm/radeon/r600_blit_kms.c b/drivers/gpu/drm/radeon/r600_blit_kms.c
index dbf716e..f42943a 100644
--- a/drivers/gpu/drm/radeon/r600_blit_kms.c
+++ b/drivers/gpu/drm/radeon/r600_blit_kms.c
@@ -473,9 +473,8 @@ int r600_blit_init(struct radeon_device *rdev)
obj_size += r6xx_ps_size * 4;
obj_size = ALIGN(obj_size, 256);

- r = radeon_object_create(rdev, NULL, obj_size,
- true, RADEON_GEM_DOMAIN_VRAM,
- false, &rdev->r600_blit.shader_obj);
+ r = radeon_bo_create(rdev, NULL, obj_size, true, RADEON_GEM_DOMAIN_VRAM,
+ &rdev->r600_blit.shader_obj);
if (r) {
DRM_ERROR("r600 failed to allocate shader\n");
return r;
@@ -485,12 +484,14 @@ int r600_blit_init(struct radeon_device *rdev)
obj_size,
rdev->r600_blit.vs_offset, rdev->r600_blit.ps_offset);

- r = radeon_object_kmap(rdev->r600_blit.shader_obj, &ptr);
+ r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_kmap(rdev->r600_blit.shader_obj, &ptr);
if (r) {
DRM_ERROR("failed to map blit object %d\n", r);
return r;
}
-
if (rdev->family >= CHIP_RV770)
memcpy_toio(ptr + rdev->r600_blit.state_offset,
r7xx_default_state, rdev->r600_blit.state_len * 4);
@@ -500,19 +501,26 @@ int r600_blit_init(struct radeon_device *rdev)
if (num_packet2s)
memcpy_toio(ptr + rdev->r600_blit.state_offset + (rdev->r600_blit.state_len * 4),
packet2s, num_packet2s * 4);
-
-
memcpy(ptr + rdev->r600_blit.vs_offset, r6xx_vs, r6xx_vs_size * 4);
memcpy(ptr + rdev->r600_blit.ps_offset, r6xx_ps, r6xx_ps_size * 4);
-
- radeon_object_kunmap(rdev->r600_blit.shader_obj);
+ radeon_bo_kunmap(rdev->r600_blit.shader_obj);
+ radeon_bo_unreserve(rdev->r600_blit.shader_obj);
return 0;
}

void r600_blit_fini(struct radeon_device *rdev)
{
- radeon_object_unpin(rdev->r600_blit.shader_obj);
- radeon_object_unref(&rdev->r600_blit.shader_obj);
+ int r;
+
+ r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
+ if (unlikely(r != 0)) {
+ dev_err(rdev->dev, "(%d) can't finish r600 blit\n", r);
+ goto out_unref;
+ }
+ radeon_bo_unpin(rdev->r600_blit.shader_obj);
+ radeon_bo_unreserve(rdev->r600_blit.shader_obj);
+out_unref:
+ radeon_bo_unref(&rdev->r600_blit.shader_obj);
}

int r600_vb_ib_get(struct radeon_device *rdev)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 9f0bd98..4b5d86b 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -28,8 +28,6 @@
#ifndef __RADEON_H__
#define __RADEON_H__

-#include "radeon_object.h"
-
/* TODO: Here are things that needs to be done :
* - surface allocator & initializer : (bit like scratch reg) should
* initialize HDP_ stuff on RS600, R600, R700 hw, well anythings
@@ -67,6 +65,11 @@
#include <linux/list.h>
#include <linux/kref.h>

+#include <ttm/ttm_bo_api.h>
+#include <ttm/ttm_bo_driver.h>
+#include <ttm/ttm_placement.h>
+#include <ttm/ttm_module.h>
+
#include "radeon_family.h"
#include "radeon_mode.h"
#include "radeon_reg.h"
@@ -186,76 +189,60 @@ void radeon_fence_unref(struct radeon_fence **fence);
* Tiling registers
*/
struct radeon_surface_reg {
- struct radeon_object *robj;
+ struct radeon_bo *bo;
};

#define RADEON_GEM_MAX_SURFACES 8

/*
- * Radeon buffer.
+ * TTM.
*/
-struct radeon_object;
+struct radeon_mman {
+ struct ttm_bo_global_ref bo_global_ref;
+ struct ttm_global_reference mem_global_ref;
+ bool mem_global_referenced;
+ struct ttm_bo_device bdev;
+};
+
+struct radeon_bo {
+ /* Protected by gem.mutex */
+ struct list_head list;
+ /* Protected by tbo.reserved */
+ struct ttm_buffer_object tbo;
+ struct ttm_bo_kmap_obj kmap;
+ unsigned pin_count;
+ void *kptr;
+ u32 tiling_flags;
+ u32 pitch;
+ int surface_reg;
+ /* Constant after initialization */
+ struct radeon_device *rdev;
+ struct drm_gem_object *gobj;
+};

-struct radeon_object_list {
+struct radeon_bo_list {
struct list_head list;
- struct radeon_object *robj;
+ struct radeon_bo *bo;
uint64_t gpu_offset;
unsigned rdomain;
unsigned wdomain;
- uint32_t tiling_flags;
+ u32 tiling_flags;
};

-int radeon_object_init(struct radeon_device *rdev);
-void radeon_object_fini(struct radeon_device *rdev);
-int radeon_object_create(struct radeon_device *rdev,
- struct drm_gem_object *gobj,
- unsigned long size,
- bool kernel,
- uint32_t domain,
- bool interruptible,
- struct radeon_object **robj_ptr);
-int radeon_object_kmap(struct radeon_object *robj, void **ptr);
-void radeon_object_kunmap(struct radeon_object *robj);
-void radeon_object_unref(struct radeon_object **robj);
-int radeon_object_pin(struct radeon_object *robj, uint32_t domain,
- uint64_t *gpu_addr);
-void radeon_object_unpin(struct radeon_object *robj);
-int radeon_object_wait(struct radeon_object *robj);
-int radeon_object_busy_domain(struct radeon_object *robj, uint32_t *cur_placement);
-int radeon_object_evict_vram(struct radeon_device *rdev);
-int radeon_object_mmap(struct radeon_object *robj, uint64_t *offset);
-void radeon_object_force_delete(struct radeon_device *rdev);
-void radeon_object_list_add_object(struct radeon_object_list *lobj,
- struct list_head *head);
-int radeon_object_list_validate(struct list_head *head, void *fence);
-void radeon_object_list_unvalidate(struct list_head *head);
-void radeon_object_list_clean(struct list_head *head);
-int radeon_object_fbdev_mmap(struct radeon_object *robj,
- struct vm_area_struct *vma);
-unsigned long radeon_object_size(struct radeon_object *robj);
-void radeon_object_clear_surface_reg(struct radeon_object *robj);
-int radeon_object_check_tiling(struct radeon_object *robj, bool has_moved,
- bool force_drop);
-void radeon_object_set_tiling_flags(struct radeon_object *robj,
- uint32_t tiling_flags, uint32_t pitch);
-void radeon_object_get_tiling_flags(struct radeon_object *robj, uint32_t *tiling_flags, uint32_t *pitch);
-void radeon_bo_move_notify(struct ttm_buffer_object *bo,
- struct ttm_mem_reg *mem);
-void radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
/*
* GEM objects.
*/
struct radeon_gem {
+ struct mutex mutex;
struct list_head objects;
};

int radeon_gem_init(struct radeon_device *rdev);
void radeon_gem_fini(struct radeon_device *rdev);
int radeon_gem_object_create(struct radeon_device *rdev, int size,
- int alignment, int initial_domain,
- bool discardable, bool kernel,
- bool interruptible,
- struct drm_gem_object **obj);
+ int alignment, int initial_domain,
+ bool discardable, bool kernel,
+ struct drm_gem_object **obj);
int radeon_gem_object_pin(struct drm_gem_object *obj, uint32_t pin_domain,
uint64_t *gpu_addr);
void radeon_gem_object_unpin(struct drm_gem_object *obj);
@@ -271,7 +258,7 @@ struct radeon_gart_table_ram {
};

struct radeon_gart_table_vram {
- struct radeon_object *robj;
+ struct radeon_bo *robj;
volatile uint32_t *ptr;
};

@@ -376,7 +363,7 @@ struct radeon_ib {
*/
struct radeon_ib_pool {
struct mutex mutex;
- struct radeon_object *robj;
+ struct radeon_bo *robj;
struct list_head scheduled_ibs;
struct radeon_ib ibs[RADEON_IB_POOL_SIZE];
bool ready;
@@ -384,7 +371,7 @@ struct radeon_ib_pool {
};

struct radeon_cp {
- struct radeon_object *ring_obj;
+ struct radeon_bo *ring_obj;
volatile uint32_t *ring;
unsigned rptr;
unsigned wptr;
@@ -400,7 +387,7 @@ struct radeon_cp {
};

struct r600_blit {
- struct radeon_object *shader_obj;
+ struct radeon_bo *shader_obj;
u64 shader_gpu_addr;
u32 vs_offset, ps_offset;
u32 state_offset;
@@ -430,8 +417,8 @@ void radeon_ring_fini(struct radeon_device *rdev);
*/
struct radeon_cs_reloc {
struct drm_gem_object *gobj;
- struct radeon_object *robj;
- struct radeon_object_list lobj;
+ struct radeon_bo *robj;
+ struct radeon_bo_list lobj;
uint32_t handle;
uint32_t flags;
};
@@ -526,7 +513,7 @@ void radeon_agp_fini(struct radeon_device *rdev);
* Writeback
*/
struct radeon_wb {
- struct radeon_object *wb_obj;
+ struct radeon_bo *wb_obj;
volatile uint32_t *wb;
uint64_t gpu_addr;
};
@@ -749,9 +736,9 @@ struct radeon_device {
uint8_t *bios;
bool is_atom_bios;
uint16_t bios_header_start;
- struct radeon_object *stollen_vga_memory;
+ struct radeon_bo *stollen_vga_memory;
struct fb_info *fbdev_info;
- struct radeon_object *fbdev_robj;
+ struct radeon_bo *fbdev_rbo;
struct radeon_framebuffer *fbdev_rfb;
/* Register mmio */
resource_size_t rmmio_base;
@@ -827,6 +814,10 @@ static inline void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32
}
}

+/*
+ * Cast helper
+ */
+#define to_radeon_fence(p) ((struct radeon_fence *)(p))

/*
* Registers read & write functions.
@@ -1018,7 +1009,7 @@ extern int r100_cp_reset(struct radeon_device *rdev);
extern void r100_vga_render_disable(struct radeon_device *rdev);
extern int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
struct radeon_cs_packet *pkt,
- struct radeon_object *robj);
+ struct radeon_bo *robj);
extern int r100_cs_parse_packet0(struct radeon_cs_parser *p,
struct radeon_cs_packet *pkt,
const unsigned *auth, unsigned n,
@@ -1104,4 +1095,6 @@ extern void r600_blit_fini(struct radeon_device *rdev);
extern int r600_cp_init_microcode(struct radeon_device *rdev);
extern int r600_gpu_reset(struct radeon_device *rdev);

+#include "radeon_object.h"
+
#endif
diff --git a/drivers/gpu/drm/radeon/radeon_benchmark.c b/drivers/gpu/drm/radeon/radeon_benchmark.c
index 10bd50a..4ddfd4b 100644
--- a/drivers/gpu/drm/radeon/radeon_benchmark.c
+++ b/drivers/gpu/drm/radeon/radeon_benchmark.c
@@ -29,8 +29,8 @@
void radeon_benchmark_move(struct radeon_device *rdev, unsigned bsize,
unsigned sdomain, unsigned ddomain)
{
- struct radeon_object *dobj = NULL;
- struct radeon_object *sobj = NULL;
+ struct radeon_bo *dobj = NULL;
+ struct radeon_bo *sobj = NULL;
struct radeon_fence *fence = NULL;
uint64_t saddr, daddr;
unsigned long start_jiffies;
@@ -41,19 +41,27 @@ void radeon_benchmark_move(struct radeon_device *rdev, unsigned bsize,

size = bsize;
n = 1024;
- r = radeon_object_create(rdev, NULL, size, true, sdomain, false, &sobj);
+ r = radeon_bo_create(rdev, NULL, size, true, sdomain, &sobj);
if (r) {
goto out_cleanup;
}
- r = radeon_object_pin(sobj, sdomain, &saddr);
+ r = radeon_bo_reserve(sobj, false);
+ if (unlikely(r != 0))
+ goto out_cleanup;
+ r = radeon_bo_pin(sobj, sdomain, &saddr);
+ radeon_bo_unreserve(sobj);
if (r) {
goto out_cleanup;
}
- r = radeon_object_create(rdev, NULL, size, true, ddomain, false, &dobj);
+ r = radeon_bo_create(rdev, NULL, size, true, ddomain, &dobj);
if (r) {
goto out_cleanup;
}
- r = radeon_object_pin(dobj, ddomain, &daddr);
+ r = radeon_bo_reserve(dobj, false);
+ if (unlikely(r != 0))
+ goto out_cleanup;
+ r = radeon_bo_pin(dobj, ddomain, &daddr);
+ radeon_bo_unreserve(dobj);
if (r) {
goto out_cleanup;
}
@@ -109,12 +117,20 @@ void radeon_benchmark_move(struct radeon_device *rdev, unsigned bsize,
}
out_cleanup:
if (sobj) {
- radeon_object_unpin(sobj);
- radeon_object_unref(&sobj);
+ r = radeon_bo_reserve(sobj, false);
+ if (likely(r == 0)) {
+ radeon_bo_unpin(sobj);
+ radeon_bo_unreserve(sobj);
+ }
+ radeon_bo_unref(&sobj);
}
if (dobj) {
- radeon_object_unpin(dobj);
- radeon_object_unref(&dobj);
+ r = radeon_bo_reserve(dobj, false);
+ if (likely(r == 0)) {
+ radeon_bo_unpin(dobj);
+ radeon_bo_unreserve(dobj);
+ }
+ radeon_bo_unref(&dobj);
}
if (fence) {
radeon_fence_unref(&fence);
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 5ab2cf9..65590a0 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -76,17 +76,17 @@ int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
}
p->relocs_ptr[i] = &p->relocs[i];
p->relocs[i].robj = p->relocs[i].gobj->driver_private;
- p->relocs[i].lobj.robj = p->relocs[i].robj;
+ p->relocs[i].lobj.bo = p->relocs[i].robj;
p->relocs[i].lobj.rdomain = r->read_domains;
p->relocs[i].lobj.wdomain = r->write_domain;
p->relocs[i].handle = r->handle;
p->relocs[i].flags = r->flags;
INIT_LIST_HEAD(&p->relocs[i].lobj.list);
- radeon_object_list_add_object(&p->relocs[i].lobj,
- &p->validated);
+ radeon_bo_list_add_object(&p->relocs[i].lobj,
+ &p->validated);
}
}
- return radeon_object_list_validate(&p->validated, p->ib->fence);
+ return radeon_bo_list_validate(&p->validated, p->ib->fence);
}

int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
@@ -190,9 +190,10 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
unsigned i;

if (error) {
- radeon_object_list_unvalidate(&parser->validated);
+ radeon_bo_list_unvalidate(&parser->validated,
+ parser->ib->fence);
} else {
- radeon_object_list_clean(&parser->validated);
+ radeon_bo_list_unreserve(&parser->validated);
}
for (i = 0; i < parser->nrelocs; i++) {
if (parser->relocs[i].gobj) {
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index 88c1907..e13f723 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -531,6 +531,7 @@ int radeon_device_init(struct radeon_device *rdev,
mutex_init(&rdev->cs_mutex);
mutex_init(&rdev->ib_pool.mutex);
mutex_init(&rdev->cp.mutex);
+ mutex_init(&rdev->gem.mutex);
rwlock_init(&rdev->fence_drv.lock);
INIT_LIST_HEAD(&rdev->gem.objects);

@@ -614,6 +615,7 @@ int radeon_suspend_kms(struct drm_device *dev, pm_message_t state)
{
struct radeon_device *rdev = dev->dev_private;
struct drm_crtc *crtc;
+ int r;

if (dev == NULL || rdev == NULL) {
return -ENODEV;
@@ -624,18 +626,22 @@ int radeon_suspend_kms(struct drm_device *dev, pm_message_t state)
/* unpin the front buffers */
list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
struct radeon_framebuffer *rfb = to_radeon_framebuffer(crtc->fb);
- struct radeon_object *robj;
+ struct radeon_bo *robj;

if (rfb == NULL || rfb->obj == NULL) {
continue;
}
robj = rfb->obj->driver_private;
- if (robj != rdev->fbdev_robj) {
- radeon_object_unpin(robj);
+ if (robj != rdev->fbdev_rbo) {
+ r = radeon_bo_reserve(robj, false);
+ if (unlikely(r == 0)) {
+ radeon_bo_unpin(robj);
+ radeon_bo_unreserve(robj);
+ }
}
}
/* evict vram memory */
- radeon_object_evict_vram(rdev);
+ radeon_bo_evict_vram(rdev);
/* wait for gpu to finish processing current batch */
radeon_fence_wait_last(rdev);

@@ -643,7 +649,7 @@ int radeon_suspend_kms(struct drm_device *dev, pm_message_t state)

radeon_suspend(rdev);
/* evict remaining vram memory */
- radeon_object_evict_vram(rdev);
+ radeon_bo_evict_vram(rdev);

pci_save_state(dev->pdev);
if (state.event == PM_EVENT_SUSPEND) {
diff --git a/drivers/gpu/drm/radeon/radeon_fb.c b/drivers/gpu/drm/radeon/radeon_fb.c
index b38c4c8..f510058 100644
--- a/drivers/gpu/drm/radeon/radeon_fb.c
+++ b/drivers/gpu/drm/radeon/radeon_fb.c
@@ -140,7 +140,7 @@ int radeonfb_create(struct drm_device *dev,
struct radeon_framebuffer *rfb;
struct drm_mode_fb_cmd mode_cmd;
struct drm_gem_object *gobj = NULL;
- struct radeon_object *robj = NULL;
+ struct radeon_bo *rbo = NULL;
struct device *device = &rdev->pdev->dev;
int size, aligned_size, ret;
u64 fb_gpuaddr;
@@ -168,14 +168,14 @@ int radeonfb_create(struct drm_device *dev,
ret = radeon_gem_object_create(rdev, aligned_size, 0,
RADEON_GEM_DOMAIN_VRAM,
false, ttm_bo_type_kernel,
- false, &gobj);
+ &gobj);
if (ret) {
printk(KERN_ERR "failed to allocate framebuffer (%d %d)\n",
surface_width, surface_height);
ret = -ENOMEM;
goto out;
}
- robj = gobj->driver_private;
+ rbo = gobj->driver_private;

if (fb_tiled)
tiling_flags = RADEON_TILING_MACRO;
@@ -192,8 +192,13 @@ int radeonfb_create(struct drm_device *dev,
}
#endif

- if (tiling_flags)
- radeon_object_set_tiling_flags(robj, tiling_flags | RADEON_TILING_SURFACE, mode_cmd.pitch);
+ if (tiling_flags) {
+ ret = radeon_bo_set_tiling_flags(rbo,
+ tiling_flags | RADEON_TILING_SURFACE,
+ mode_cmd.pitch);
+ if (ret)
+ dev_err(rdev->dev, "FB failed to set tiling flags\n");
+ }
mutex_lock(&rdev->ddev->struct_mutex);
fb = radeon_framebuffer_create(rdev->ddev, &mode_cmd, gobj);
if (fb == NULL) {
@@ -201,10 +206,19 @@ int radeonfb_create(struct drm_device *dev,
ret = -ENOMEM;
goto out_unref;
}
- ret = radeon_object_pin(robj, RADEON_GEM_DOMAIN_VRAM, &fb_gpuaddr);
+ ret = radeon_bo_reserve(rbo, false);
+ if (unlikely(ret != 0))
+ goto out_unref;
+ ret = radeon_bo_pin(rbo, RADEON_GEM_DOMAIN_VRAM, &fb_gpuaddr);
+ if (ret) {
+ radeon_bo_unreserve(rbo);
+ goto out_unref;
+ }
+ if (fb_tiled)
+ radeon_bo_check_tiling(rbo, 0, 0);
+ ret = radeon_bo_kmap(rbo, &fbptr);
+ radeon_bo_unreserve(rbo);
if (ret) {
- printk(KERN_ERR "failed to pin framebuffer\n");
- ret = -ENOMEM;
goto out_unref;
}

@@ -213,7 +227,7 @@ int radeonfb_create(struct drm_device *dev,
*fb_p = fb;
rfb = to_radeon_framebuffer(fb);
rdev->fbdev_rfb = rfb;
- rdev->fbdev_robj = robj;
+ rdev->fbdev_rbo = rbo;

info = framebuffer_alloc(sizeof(struct radeon_fb_device), device);
if (info == NULL) {
@@ -234,15 +248,7 @@ int radeonfb_create(struct drm_device *dev,
if (ret)
goto out_unref;

- if (fb_tiled)
- radeon_object_check_tiling(robj, 0, 0);
-
- ret = radeon_object_kmap(robj, &fbptr);
- if (ret) {
- goto out_unref;
- }
-
- memset_io(fbptr, 0, aligned_size);
+ memset_io(fbptr, 0xff, aligned_size);

strcpy(info->fix.id, "radeondrmfb");

@@ -288,8 +294,12 @@ int radeonfb_create(struct drm_device *dev,
return 0;

out_unref:
- if (robj) {
- radeon_object_kunmap(robj);
+ if (rbo) {
+ ret = radeon_bo_reserve(rbo, false);
+ if (likely(ret == 0)) {
+ radeon_bo_kunmap(rbo);
+ radeon_bo_unreserve(rbo);
+ }
}
if (fb && ret) {
list_del(&fb->filp_head);
@@ -328,7 +338,8 @@ int radeonfb_remove(struct drm_device *dev, struct drm_framebuffer *fb)
{
struct fb_info *info;
struct radeon_framebuffer *rfb = to_radeon_framebuffer(fb);
- struct radeon_object *robj;
+ struct radeon_bo *rbo;
+ int r;

if (!fb) {
return -EINVAL;
@@ -336,10 +347,14 @@ int radeonfb_remove(struct drm_device *dev, struct drm_framebuffer *fb)
info = fb->fbdev;
if (info) {
struct radeon_fb_device *rfbdev = info->par;
- robj = rfb->obj->driver_private;
+ rbo = rfb->obj->driver_private;
unregister_framebuffer(info);
- radeon_object_kunmap(robj);
- radeon_object_unpin(robj);
+ r = radeon_bo_reserve(rbo, false);
+ if (likely(r == 0)) {
+ radeon_bo_kunmap(rbo);
+ radeon_bo_unpin(rbo);
+ radeon_bo_unreserve(rbo);
+ }
drm_fb_helper_free(&rfbdev->helper);
framebuffer_release(info);
}
diff --git a/drivers/gpu/drm/radeon/radeon_gart.c b/drivers/gpu/drm/radeon/radeon_gart.c
index a68d756..e73d56e 100644
--- a/drivers/gpu/drm/radeon/radeon_gart.c
+++ b/drivers/gpu/drm/radeon/radeon_gart.c
@@ -78,11 +78,9 @@ int radeon_gart_table_vram_alloc(struct radeon_device *rdev)
int r;

if (rdev->gart.table.vram.robj == NULL) {
- r = radeon_object_create(rdev, NULL,
- rdev->gart.table_size,
- true,
- RADEON_GEM_DOMAIN_VRAM,
- false, &rdev->gart.table.vram.robj);
+ r = radeon_bo_create(rdev, NULL, rdev->gart.table_size,
+ true, RADEON_GEM_DOMAIN_VRAM,
+ &rdev->gart.table.vram.robj);
if (r) {
return r;
}
@@ -95,32 +93,38 @@ int radeon_gart_table_vram_pin(struct radeon_device *rdev)
uint64_t gpu_addr;
int r;

- r = radeon_object_pin(rdev->gart.table.vram.robj,
- RADEON_GEM_DOMAIN_VRAM, &gpu_addr);
- if (r) {
- radeon_object_unref(&rdev->gart.table.vram.robj);
+ r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
+ if (unlikely(r != 0))
return r;
- }
- r = radeon_object_kmap(rdev->gart.table.vram.robj,
- (void **)&rdev->gart.table.vram.ptr);
+ r = radeon_bo_pin(rdev->gart.table.vram.robj,
+ RADEON_GEM_DOMAIN_VRAM, &gpu_addr);
if (r) {
- radeon_object_unpin(rdev->gart.table.vram.robj);
- radeon_object_unref(&rdev->gart.table.vram.robj);
- DRM_ERROR("radeon: failed to map gart vram table.\n");
+ radeon_bo_unreserve(rdev->gart.table.vram.robj);
return r;
}
+ r = radeon_bo_kmap(rdev->gart.table.vram.robj,
+ (void **)&rdev->gart.table.vram.ptr);
+ if (r)
+ radeon_bo_unpin(rdev->gart.table.vram.robj);
+ radeon_bo_unreserve(rdev->gart.table.vram.robj);
rdev->gart.table_addr = gpu_addr;
- return 0;
+ return r;
}

void radeon_gart_table_vram_free(struct radeon_device *rdev)
{
+ int r;
+
if (rdev->gart.table.vram.robj == NULL) {
return;
}
- radeon_object_kunmap(rdev->gart.table.vram.robj);
- radeon_object_unpin(rdev->gart.table.vram.robj);
- radeon_object_unref(&rdev->gart.table.vram.robj);
+ r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
+ if (likely(r == 0)) {
+ radeon_bo_kunmap(rdev->gart.table.vram.robj);
+ radeon_bo_unpin(rdev->gart.table.vram.robj);
+ radeon_bo_unreserve(rdev->gart.table.vram.robj);
+ }
+ radeon_bo_unref(&rdev->gart.table.vram.robj);
}


diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c
index d880edf..f3fb2bf 100644
--- a/drivers/gpu/drm/radeon/radeon_gem.c
+++ b/drivers/gpu/drm/radeon/radeon_gem.c
@@ -38,22 +38,21 @@ int radeon_gem_object_init(struct drm_gem_object *obj)

void radeon_gem_object_free(struct drm_gem_object *gobj)
{
- struct radeon_object *robj = gobj->driver_private;
+ struct radeon_bo *robj = gobj->driver_private;

gobj->driver_private = NULL;
if (robj) {
- radeon_object_unref(&robj);
+ radeon_bo_unref(&robj);
}
}

int radeon_gem_object_create(struct radeon_device *rdev, int size,
- int alignment, int initial_domain,
- bool discardable, bool kernel,
- bool interruptible,
- struct drm_gem_object **obj)
+ int alignment, int initial_domain,
+ bool discardable, bool kernel,
+ struct drm_gem_object **obj)
{
struct drm_gem_object *gobj;
- struct radeon_object *robj;
+ struct radeon_bo *robj;
int r;

*obj = NULL;
@@ -65,8 +64,7 @@ int radeon_gem_object_create(struct radeon_device *rdev, int size,
if (alignment < PAGE_SIZE) {
alignment = PAGE_SIZE;
}
- r = radeon_object_create(rdev, gobj, size, kernel, initial_domain,
- interruptible, &robj);
+ r = radeon_bo_create(rdev, gobj, size, kernel, initial_domain, &robj);
if (r) {
DRM_ERROR("Failed to allocate GEM object (%d, %d, %u)\n",
size, initial_domain, alignment);
@@ -83,33 +81,33 @@ int radeon_gem_object_create(struct radeon_device *rdev, int size,
int radeon_gem_object_pin(struct drm_gem_object *obj, uint32_t pin_domain,
uint64_t *gpu_addr)
{
- struct radeon_object *robj = obj->driver_private;
- uint32_t flags;
+ struct radeon_bo *robj = obj->driver_private;
+ int r;

- switch (pin_domain) {
- case RADEON_GEM_DOMAIN_VRAM:
- flags = TTM_PL_FLAG_VRAM;
- break;
- case RADEON_GEM_DOMAIN_GTT:
- flags = TTM_PL_FLAG_TT;
- break;
- default:
- flags = TTM_PL_FLAG_SYSTEM;
- break;
- }
- return radeon_object_pin(robj, flags, gpu_addr);
+ r = radeon_bo_reserve(robj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(robj, pin_domain, gpu_addr);
+ radeon_bo_unreserve(robj);
+ return r;
}

void radeon_gem_object_unpin(struct drm_gem_object *obj)
{
- struct radeon_object *robj = obj->driver_private;
- radeon_object_unpin(robj);
+ struct radeon_bo *robj = obj->driver_private;
+ int r;
+
+ r = radeon_bo_reserve(robj, false);
+ if (likely(r == 0)) {
+ radeon_bo_unpin(robj);
+ radeon_bo_unreserve(robj);
+ }
}

int radeon_gem_set_domain(struct drm_gem_object *gobj,
uint32_t rdomain, uint32_t wdomain)
{
- struct radeon_object *robj;
+ struct radeon_bo *robj;
uint32_t domain;
int r;

@@ -127,7 +125,7 @@ int radeon_gem_set_domain(struct drm_gem_object *gobj,
}
if (domain == RADEON_GEM_DOMAIN_CPU) {
/* Asking for cpu access wait for object idle */
- r = radeon_object_wait(robj);
+ r = radeon_bo_wait(robj, NULL, false);
if (r) {
printk(KERN_ERR "Failed to wait for object !\n");
return r;
@@ -144,7 +142,7 @@ int radeon_gem_init(struct radeon_device *rdev)

void radeon_gem_fini(struct radeon_device *rdev)
{
- radeon_object_force_delete(rdev);
+ radeon_bo_force_delete(rdev);
}


@@ -192,8 +190,8 @@ int radeon_gem_create_ioctl(struct drm_device *dev, void *data,
/* create a gem object to contain this object in */
args->size = roundup(args->size, PAGE_SIZE);
r = radeon_gem_object_create(rdev, args->size, args->alignment,
- args->initial_domain, false,
- false, true, &gobj);
+ args->initial_domain, false,
+ false, &gobj);
if (r) {
return r;
}
@@ -218,7 +216,7 @@ int radeon_gem_set_domain_ioctl(struct drm_device *dev, void *data,
* just validate the BO into a certain domain */
struct drm_radeon_gem_set_domain *args = data;
struct drm_gem_object *gobj;
- struct radeon_object *robj;
+ struct radeon_bo *robj;
int r;

/* for now if someone requests domain CPU -
@@ -244,19 +242,18 @@ int radeon_gem_mmap_ioctl(struct drm_device *dev, void *data,
{
struct drm_radeon_gem_mmap *args = data;
struct drm_gem_object *gobj;
- struct radeon_object *robj;
- int r;
+ struct radeon_bo *robj;

gobj = drm_gem_object_lookup(dev, filp, args->handle);
if (gobj == NULL) {
return -EINVAL;
}
robj = gobj->driver_private;
- r = radeon_object_mmap(robj, &args->addr_ptr);
+ args->addr_ptr = radeon_bo_mmap_offset(robj);
mutex_lock(&dev->struct_mutex);
drm_gem_object_unreference(gobj);
mutex_unlock(&dev->struct_mutex);
- return r;
+ return 0;
}

int radeon_gem_busy_ioctl(struct drm_device *dev, void *data,
@@ -264,7 +261,7 @@ int radeon_gem_busy_ioctl(struct drm_device *dev, void *data,
{
struct drm_radeon_gem_busy *args = data;
struct drm_gem_object *gobj;
- struct radeon_object *robj;
+ struct radeon_bo *robj;
int r;
uint32_t cur_placement;

@@ -273,7 +270,7 @@ int radeon_gem_busy_ioctl(struct drm_device *dev, void *data,
return -EINVAL;
}
robj = gobj->driver_private;
- r = radeon_object_busy_domain(robj, &cur_placement);
+ r = radeon_bo_wait(robj, &cur_placement, true);
switch (cur_placement) {
case TTM_PL_VRAM:
args->domain = RADEON_GEM_DOMAIN_VRAM;
@@ -297,7 +294,7 @@ int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
{
struct drm_radeon_gem_wait_idle *args = data;
struct drm_gem_object *gobj;
- struct radeon_object *robj;
+ struct radeon_bo *robj;
int r;

gobj = drm_gem_object_lookup(dev, filp, args->handle);
@@ -305,7 +302,7 @@ int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
return -EINVAL;
}
robj = gobj->driver_private;
- r = radeon_object_wait(robj);
+ r = radeon_bo_wait(robj, NULL, false);
mutex_lock(&dev->struct_mutex);
drm_gem_object_unreference(gobj);
mutex_unlock(&dev->struct_mutex);
@@ -317,7 +314,7 @@ int radeon_gem_set_tiling_ioctl(struct drm_device *dev, void *data,
{
struct drm_radeon_gem_set_tiling *args = data;
struct drm_gem_object *gobj;
- struct radeon_object *robj;
+ struct radeon_bo *robj;
int r = 0;

DRM_DEBUG("%d \n", args->handle);
@@ -325,7 +322,7 @@ int radeon_gem_set_tiling_ioctl(struct drm_device *dev, void *data,
if (gobj == NULL)
return -EINVAL;
robj = gobj->driver_private;
- radeon_object_set_tiling_flags(robj, args->tiling_flags, args->pitch);
+ r = radeon_bo_set_tiling_flags(robj, args->tiling_flags, args->pitch);
mutex_lock(&dev->struct_mutex);
drm_gem_object_unreference(gobj);
mutex_unlock(&dev->struct_mutex);
@@ -337,16 +334,19 @@ int radeon_gem_get_tiling_ioctl(struct drm_device *dev, void *data,
{
struct drm_radeon_gem_get_tiling *args = data;
struct drm_gem_object *gobj;
- struct radeon_object *robj;
+ struct radeon_bo *rbo;
int r = 0;

DRM_DEBUG("\n");
gobj = drm_gem_object_lookup(dev, filp, args->handle);
if (gobj == NULL)
return -EINVAL;
- robj = gobj->driver_private;
- radeon_object_get_tiling_flags(robj, &args->tiling_flags,
- &args->pitch);
+ rbo = gobj->driver_private;
+ r = radeon_bo_reserve(rbo, false);
+ if (unlikely(r != 0))
+ return r;
+ radeon_bo_get_tiling_flags(rbo, &args->tiling_flags, &args->pitch);
+ radeon_bo_unreserve(rbo);
mutex_lock(&dev->struct_mutex);
drm_gem_object_unreference(gobj);
mutex_unlock(&dev->struct_mutex);
diff --git a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
index 8d0b7aa..e8a984d 100644
--- a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
+++ b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
@@ -400,12 +400,14 @@ int radeon_crtc_set_base(struct drm_crtc *crtc, int x, int y,
struct radeon_crtc *radeon_crtc = to_radeon_crtc(crtc);
struct radeon_framebuffer *radeon_fb;
struct drm_gem_object *obj;
+ struct radeon_bo *rbo;
uint64_t base;
uint32_t crtc_offset, crtc_offset_cntl, crtc_tile_x0_y0 = 0;
uint32_t crtc_pitch, pitch_pixels;
uint32_t tiling_flags;
int format;
uint32_t gen_cntl_reg, gen_cntl_val;
+ int r;

DRM_DEBUG("\n");

@@ -431,10 +433,22 @@ int radeon_crtc_set_base(struct drm_crtc *crtc, int x, int y,
return false;
}

+ /* Pin framebuffer & get tilling informations */
obj = radeon_fb->obj;
- if (radeon_gem_object_pin(obj, RADEON_GEM_DOMAIN_VRAM, &base)) {
+ rbo = obj->driver_private;
+ r = radeon_bo_reserve(rbo, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(rbo, RADEON_GEM_DOMAIN_VRAM, &base);
+ if (unlikely(r != 0)) {
+ radeon_bo_unreserve(rbo);
return -EINVAL;
}
+ radeon_bo_get_tiling_flags(rbo, &tiling_flags, NULL);
+ radeon_bo_unreserve(rbo);
+ if (tiling_flags & RADEON_TILING_MICRO)
+ DRM_ERROR("trying to scanout microtiled buffer\n");
+
/* if scanout was in GTT this really wouldn't work */
/* crtc offset is from display base addr not FB location */
radeon_crtc->legacy_display_base_addr = rdev->mc.vram_location;
@@ -449,10 +463,6 @@ int radeon_crtc_set_base(struct drm_crtc *crtc, int x, int y,
(crtc->fb->bits_per_pixel * 8));
crtc_pitch |= crtc_pitch << 16;

- radeon_object_get_tiling_flags(obj->driver_private,
- &tiling_flags, NULL);
- if (tiling_flags & RADEON_TILING_MICRO)
- DRM_ERROR("trying to scanout microtiled buffer\n");

if (tiling_flags & RADEON_TILING_MACRO) {
if (ASIC_IS_R300(rdev))
@@ -530,7 +540,12 @@ int radeon_crtc_set_base(struct drm_crtc *crtc, int x, int y,

if (old_fb && old_fb != crtc->fb) {
radeon_fb = to_radeon_framebuffer(old_fb);
- radeon_gem_object_unpin(radeon_fb->obj);
+ rbo = radeon_fb->obj->driver_private;
+ r = radeon_bo_reserve(rbo, false);
+ if (unlikely(r != 0))
+ return r;
+ radeon_bo_unpin(rbo);
+ radeon_bo_unreserve(rbo);
}

/* Bytes per pixel may have changed */
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index 1f056da..bec4943 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -34,74 +34,32 @@
#include "radeon_drm.h"
#include "radeon.h"

-struct radeon_object {
- struct ttm_buffer_object tobj;
- struct list_head list;
- struct radeon_device *rdev;
- struct drm_gem_object *gobj;
- struct ttm_bo_kmap_obj kmap;
- unsigned pin_count;
- uint64_t gpu_addr;
- void *kptr;
- bool is_iomem;
- uint32_t tiling_flags;
- uint32_t pitch;
- int surface_reg;
-};

int radeon_ttm_init(struct radeon_device *rdev);
void radeon_ttm_fini(struct radeon_device *rdev);
+static void radeon_bo_clear_surface_reg(struct radeon_bo *bo);

/*
* To exclude mutual BO access we rely on bo_reserve exclusion, as all
* function are calling it.
*/

-static int radeon_object_reserve(struct radeon_object *robj, bool interruptible)
+static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo)
{
- return ttm_bo_reserve(&robj->tobj, interruptible, false, false, 0);
-}
+ struct radeon_bo *bo;

-static void radeon_object_unreserve(struct radeon_object *robj)
-{
- ttm_bo_unreserve(&robj->tobj);
+ bo = container_of(tbo, struct radeon_bo, tbo);
+ mutex_lock(&bo->rdev->gem.mutex);
+ list_del_init(&bo->list);
+ mutex_unlock(&bo->rdev->gem.mutex);
+ radeon_bo_clear_surface_reg(bo);
+ kfree(bo);
}

-static void radeon_ttm_object_object_destroy(struct ttm_buffer_object *tobj)
+static inline u32 radeon_ttm_flags_from_domain(u32 domain)
{
- struct radeon_object *robj;
-
- robj = container_of(tobj, struct radeon_object, tobj);
- list_del_init(&robj->list);
- radeon_object_clear_surface_reg(robj);
- kfree(robj);
-}
-
-static inline void radeon_object_gpu_addr(struct radeon_object *robj)
-{
- /* Default gpu address */
- robj->gpu_addr = 0xFFFFFFFFFFFFFFFFULL;
- if (robj->tobj.mem.mm_node == NULL) {
- return;
- }
- robj->gpu_addr = ((u64)robj->tobj.mem.mm_node->start) << PAGE_SHIFT;
- switch (robj->tobj.mem.mem_type) {
- case TTM_PL_VRAM:
- robj->gpu_addr += (u64)robj->rdev->mc.vram_location;
- break;
- case TTM_PL_TT:
- robj->gpu_addr += (u64)robj->rdev->mc.gtt_location;
- break;
- default:
- DRM_ERROR("Unknown placement %d\n", robj->tobj.mem.mem_type);
- robj->gpu_addr = 0xFFFFFFFFFFFFFFFFULL;
- return;
- }
-}
+ u32 flags = 0;

-static inline uint32_t radeon_object_flags_from_domain(uint32_t domain)
-{
- uint32_t flags = 0;
if (domain & RADEON_GEM_DOMAIN_VRAM) {
flags |= TTM_PL_FLAG_VRAM | TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED;
}
@@ -117,17 +75,13 @@ static inline uint32_t radeon_object_flags_from_domain(uint32_t domain)
return flags;
}

-int radeon_object_create(struct radeon_device *rdev,
- struct drm_gem_object *gobj,
- unsigned long size,
- bool kernel,
- uint32_t domain,
- bool interruptible,
- struct radeon_object **robj_ptr)
+int radeon_bo_create(struct radeon_device *rdev, struct drm_gem_object *gobj,
+ unsigned long size, bool kernel, u32 domain,
+ struct radeon_bo **bo_ptr)
{
- struct radeon_object *robj;
+ struct radeon_bo *bo;
enum ttm_bo_type type;
- uint32_t flags;
+ u32 flags;
int r;

if (unlikely(rdev->mman.bdev.dev_mapping == NULL)) {
@@ -138,206 +92,140 @@ int radeon_object_create(struct radeon_device *rdev,
} else {
type = ttm_bo_type_device;
}
- *robj_ptr = NULL;
- robj = kzalloc(sizeof(struct radeon_object), GFP_KERNEL);
- if (robj == NULL) {
+ *bo_ptr = NULL;
+ bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL);
+ if (bo == NULL)
return -ENOMEM;
- }
- robj->rdev = rdev;
- robj->gobj = gobj;
- robj->surface_reg = -1;
- INIT_LIST_HEAD(&robj->list);
-
- flags = radeon_object_flags_from_domain(domain);
- r = ttm_buffer_object_init(&rdev->mman.bdev, &robj->tobj, size, type, flags,
- 0, 0, false, NULL, size,
- &radeon_ttm_object_object_destroy);
+ bo->rdev = rdev;
+ bo->gobj = gobj;
+ bo->surface_reg = -1;
+ INIT_LIST_HEAD(&bo->list);
+
+ flags = radeon_ttm_flags_from_domain(domain);
+retry:
+ r = ttm_buffer_object_init(&rdev->mman.bdev, &bo->tbo, size, type,
+ flags, 0, 0, true, NULL, size,
+ &radeon_ttm_bo_destroy);
if (unlikely(r != 0)) {
+ if (r == -ERESTART)
+ goto retry;
/* ttm call radeon_ttm_object_object_destroy if error happen */
- DRM_ERROR("Failed to allocate TTM object (%ld, 0x%08X, %u)\n",
- size, flags, 0);
+ dev_err(rdev->dev, "object_init failed for (%ld, 0x%08X)\n",
+ size, flags);
return r;
}
- *robj_ptr = robj;
+ *bo_ptr = bo;
if (gobj) {
- list_add_tail(&robj->list, &rdev->gem.objects);
+ mutex_lock(&bo->rdev->gem.mutex);
+ list_add_tail(&bo->list, &rdev->gem.objects);
+ mutex_unlock(&bo->rdev->gem.mutex);
}
return 0;
}

-int radeon_object_kmap(struct radeon_object *robj, void **ptr)
+int radeon_bo_kmap(struct radeon_bo *bo, void **ptr)
{
+ bool is_iomem;
int r;

- spin_lock(&robj->tobj.lock);
- if (robj->kptr) {
+ if (bo->kptr) {
if (ptr) {
- *ptr = robj->kptr;
+ *ptr = bo->kptr;
}
- spin_unlock(&robj->tobj.lock);
return 0;
}
- spin_unlock(&robj->tobj.lock);
- r = ttm_bo_kmap(&robj->tobj, 0, robj->tobj.num_pages, &robj->kmap);
+ r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap);
if (r) {
return r;
}
- spin_lock(&robj->tobj.lock);
- robj->kptr = ttm_kmap_obj_virtual(&robj->kmap, &robj->is_iomem);
- spin_unlock(&robj->tobj.lock);
+ bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
if (ptr) {
- *ptr = robj->kptr;
+ *ptr = bo->kptr;
}
- radeon_object_check_tiling(robj, 0, 0);
+ radeon_bo_check_tiling(bo, 0, 0);
return 0;
}

-void radeon_object_kunmap(struct radeon_object *robj)
+void radeon_bo_kunmap(struct radeon_bo *bo)
{
- spin_lock(&robj->tobj.lock);
- if (robj->kptr == NULL) {
- spin_unlock(&robj->tobj.lock);
+ if (bo->kptr == NULL)
return;
- }
- robj->kptr = NULL;
- spin_unlock(&robj->tobj.lock);
- radeon_object_check_tiling(robj, 0, 0);
- ttm_bo_kunmap(&robj->kmap);
+ bo->kptr = NULL;
+ radeon_bo_check_tiling(bo, 0, 0);
+ ttm_bo_kunmap(&bo->kmap);
}

-void radeon_object_unref(struct radeon_object **robj)
+void radeon_bo_unref(struct radeon_bo **bo)
{
- struct ttm_buffer_object *tobj;
+ struct ttm_buffer_object *tbo;

- if ((*robj) == NULL) {
+ if ((*bo) == NULL)
return;
- }
- tobj = &((*robj)->tobj);
- ttm_bo_unref(&tobj);
- if (tobj == NULL) {
- *robj = NULL;
- }
-}
-
-int radeon_object_mmap(struct radeon_object *robj, uint64_t *offset)
-{
- *offset = robj->tobj.addr_space_offset;
- return 0;
+ tbo = &((*bo)->tbo);
+ ttm_bo_unref(&tbo);
+ if (tbo == NULL)
+ *bo = NULL;
}

-int radeon_object_pin(struct radeon_object *robj, uint32_t domain,
- uint64_t *gpu_addr)
+int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
{
- uint32_t flags;
- uint32_t tmp;
+ u32 flags;
+ u32 tmp;
int r;

- flags = radeon_object_flags_from_domain(domain);
- spin_lock(&robj->tobj.lock);
- if (robj->pin_count) {
- robj->pin_count++;
- if (gpu_addr != NULL) {
- *gpu_addr = robj->gpu_addr;
- }
- spin_unlock(&robj->tobj.lock);
+ flags = radeon_ttm_flags_from_domain(domain);
+ if (bo->pin_count) {
+ bo->pin_count++;
+ if (gpu_addr)
+ *gpu_addr = radeon_bo_gpu_offset(bo);
return 0;
}
- spin_unlock(&robj->tobj.lock);
- r = radeon_object_reserve(robj, false);
- if (unlikely(r != 0)) {
- DRM_ERROR("radeon: failed to reserve object for pinning it.\n");
- return r;
- }
- tmp = robj->tobj.mem.placement;
+ tmp = bo->tbo.mem.placement;
ttm_flag_masked(&tmp, flags, TTM_PL_MASK_MEM);
- robj->tobj.proposed_placement = tmp | TTM_PL_FLAG_NO_EVICT | TTM_PL_MASK_CACHING;
- r = ttm_buffer_object_validate(&robj->tobj,
- robj->tobj.proposed_placement,
- false, false);
- radeon_object_gpu_addr(robj);
- if (gpu_addr != NULL) {
- *gpu_addr = robj->gpu_addr;
+ bo->tbo.proposed_placement = tmp | TTM_PL_FLAG_NO_EVICT |
+ TTM_PL_MASK_CACHING;
+retry:
+ r = ttm_buffer_object_validate(&bo->tbo, bo->tbo.proposed_placement,
+ true, false);
+ if (likely(r == 0)) {
+ bo->pin_count = 1;
+ if (gpu_addr != NULL)
+ *gpu_addr = radeon_bo_gpu_offset(bo);
}
- robj->pin_count = 1;
if (unlikely(r != 0)) {
- DRM_ERROR("radeon: failed to pin object.\n");
+ if (r == -ERESTART)
+ goto retry;
+ dev_err(bo->rdev->dev, "%p pin failed\n", bo);
}
- radeon_object_unreserve(robj);
return r;
}

-void radeon_object_unpin(struct radeon_object *robj)
+int radeon_bo_unpin(struct radeon_bo *bo)
{
- uint32_t flags;
int r;

- spin_lock(&robj->tobj.lock);
- if (!robj->pin_count) {
- spin_unlock(&robj->tobj.lock);
- printk(KERN_WARNING "Unpin not necessary for %p !\n", robj);
- return;
- }
- robj->pin_count--;
- if (robj->pin_count) {
- spin_unlock(&robj->tobj.lock);
- return;
- }
- spin_unlock(&robj->tobj.lock);
- r = radeon_object_reserve(robj, false);
- if (unlikely(r != 0)) {
- DRM_ERROR("radeon: failed to reserve object for unpinning it.\n");
- return;
- }
- flags = robj->tobj.mem.placement;
- robj->tobj.proposed_placement = flags & ~TTM_PL_FLAG_NO_EVICT;
- r = ttm_buffer_object_validate(&robj->tobj,
- robj->tobj.proposed_placement,
- false, false);
- if (unlikely(r != 0)) {
- DRM_ERROR("radeon: failed to unpin buffer.\n");
- }
- radeon_object_unreserve(robj);
-}
-
-int radeon_object_wait(struct radeon_object *robj)
-{
- int r = 0;
-
- /* FIXME: should use block reservation instead */
- r = radeon_object_reserve(robj, true);
- if (unlikely(r != 0)) {
- DRM_ERROR("radeon: failed to reserve object for waiting.\n");
- return r;
- }
- spin_lock(&robj->tobj.lock);
- if (robj->tobj.sync_obj) {
- r = ttm_bo_wait(&robj->tobj, true, true, false);
+ if (!bo->pin_count) {
+ dev_warn(bo->rdev->dev, "%p unpin not necessary\n", bo);
+ return 0;
}
- spin_unlock(&robj->tobj.lock);
- radeon_object_unreserve(robj);
- return r;
-}
-
-int radeon_object_busy_domain(struct radeon_object *robj, uint32_t *cur_placement)
-{
- int r = 0;
-
- r = radeon_object_reserve(robj, true);
+ bo->pin_count--;
+ if (bo->pin_count)
+ return 0;
+ bo->tbo.proposed_placement = bo->tbo.mem.placement &
+ ~TTM_PL_FLAG_NO_EVICT;
+retry:
+ r = ttm_buffer_object_validate(&bo->tbo, bo->tbo.proposed_placement,
+ true, false);
if (unlikely(r != 0)) {
- DRM_ERROR("radeon: failed to reserve object for waiting.\n");
+ if (r == -ERESTART)
+ goto retry;
+ dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo);
return r;
}
- spin_lock(&robj->tobj.lock);
- *cur_placement = robj->tobj.mem.mem_type;
- if (robj->tobj.sync_obj) {
- r = ttm_bo_wait(&robj->tobj, true, true, true);
- }
- spin_unlock(&robj->tobj.lock);
- radeon_object_unreserve(robj);
- return r;
+ return 0;
}

-int radeon_object_evict_vram(struct radeon_device *rdev)
+int radeon_bo_evict_vram(struct radeon_device *rdev)
{
if (rdev->flags & RADEON_IS_IGP) {
/* Useless to evict on IGP chips */
@@ -346,30 +234,32 @@ int radeon_object_evict_vram(struct radeon_device *rdev)
return ttm_bo_evict_mm(&rdev->mman.bdev, TTM_PL_VRAM);
}

-void radeon_object_force_delete(struct radeon_device *rdev)
+void radeon_bo_force_delete(struct radeon_device *rdev)
{
- struct radeon_object *robj, *n;
+ struct radeon_bo *bo, *n;
struct drm_gem_object *gobj;

if (list_empty(&rdev->gem.objects)) {
return;
}
- DRM_ERROR("Userspace still has active objects !\n");
- list_for_each_entry_safe(robj, n, &rdev->gem.objects, list) {
+ dev_err(rdev->dev, "Userspace still has active objects !\n");
+ list_for_each_entry_safe(bo, n, &rdev->gem.objects, list) {
mutex_lock(&rdev->ddev->struct_mutex);
- gobj = robj->gobj;
- DRM_ERROR("Force free for (%p,%p,%lu,%lu)\n",
- gobj, robj, (unsigned long)gobj->size,
- *((unsigned long *)&gobj->refcount));
- list_del_init(&robj->list);
- radeon_object_unref(&robj);
+ gobj = bo->gobj;
+ dev_err(rdev->dev, "%p %p %lu %lu force free\n",
+ gobj, bo, (unsigned long)gobj->size,
+ *((unsigned long *)&gobj->refcount));
+ mutex_lock(&bo->rdev->gem.mutex);
+ list_del_init(&bo->list);
+ mutex_unlock(&bo->rdev->gem.mutex);
+ radeon_bo_unref(&bo);
gobj->driver_private = NULL;
drm_gem_object_unreference(gobj);
mutex_unlock(&rdev->ddev->struct_mutex);
}
}

-int radeon_object_init(struct radeon_device *rdev)
+int radeon_bo_init(struct radeon_device *rdev)
{
/* Add an MTRR for the VRAM */
rdev->mc.vram_mtrr = mtrr_add(rdev->mc.aper_base, rdev->mc.aper_size,
@@ -382,13 +272,13 @@ int radeon_object_init(struct radeon_device *rdev)
return radeon_ttm_init(rdev);
}

-void radeon_object_fini(struct radeon_device *rdev)
+void radeon_bo_fini(struct radeon_device *rdev)
{
radeon_ttm_fini(rdev);
}

-void radeon_object_list_add_object(struct radeon_object_list *lobj,
- struct list_head *head)
+void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
+ struct list_head *head)
{
if (lobj->wdomain) {
list_add(&lobj->list, head);
@@ -397,72 +287,67 @@ void radeon_object_list_add_object(struct radeon_object_list *lobj,
}
}

-int radeon_object_list_reserve(struct list_head *head)
+int radeon_bo_list_reserve(struct list_head *head)
{
- struct radeon_object_list *lobj;
+ struct radeon_bo_list *lobj;
int r;

list_for_each_entry(lobj, head, list){
- if (!lobj->robj->pin_count) {
- r = radeon_object_reserve(lobj->robj, true);
- if (unlikely(r != 0)) {
- DRM_ERROR("radeon: failed to reserve object.\n");
- return r;
- }
- } else {
- }
+ r = radeon_bo_reserve(lobj->bo, false);
+ if (unlikely(r != 0))
+ return r;
}
return 0;
}

-void radeon_object_list_unreserve(struct list_head *head)
+void radeon_bo_list_unreserve(struct list_head *head)
{
- struct radeon_object_list *lobj;
+ struct radeon_bo_list *lobj;

list_for_each_entry(lobj, head, list) {
- if (!lobj->robj->pin_count) {
- radeon_object_unreserve(lobj->robj);
- }
+ /* only unreserve object we successfully reserved */
+ if (radeon_bo_is_reserved(lobj->bo))
+ radeon_bo_unreserve(lobj->bo);
}
}

-int radeon_object_list_validate(struct list_head *head, void *fence)
+int radeon_bo_list_validate(struct list_head *head, void *fence)
{
- struct radeon_object_list *lobj;
- struct radeon_object *robj;
+ struct radeon_bo_list *lobj;
+ struct radeon_bo *bo;
struct radeon_fence *old_fence = NULL;
int r;

- r = radeon_object_list_reserve(head);
+ r = radeon_bo_list_reserve(head);
if (unlikely(r != 0)) {
- radeon_object_list_unreserve(head);
return r;
}
list_for_each_entry(lobj, head, list) {
- robj = lobj->robj;
- if (!robj->pin_count) {
+ bo = lobj->bo;
+ if (!bo->pin_count) {
if (lobj->wdomain) {
- robj->tobj.proposed_placement =
- radeon_object_flags_from_domain(lobj->wdomain);
+ bo->tbo.proposed_placement =
+ radeon_ttm_flags_from_domain(lobj->wdomain);
} else {
- robj->tobj.proposed_placement =
- radeon_object_flags_from_domain(lobj->rdomain);
+ bo->tbo.proposed_placement =
+ radeon_ttm_flags_from_domain(lobj->rdomain);
}
- r = ttm_buffer_object_validate(&robj->tobj,
- robj->tobj.proposed_placement,
- true, false);
+retry:
+ r = ttm_buffer_object_validate(&bo->tbo,
+ bo->tbo.proposed_placement,
+ true, false);
if (unlikely(r)) {
- DRM_ERROR("radeon: failed to validate.\n");
+ if (r == -ERESTART)
+ goto retry;
return r;
}
- radeon_object_gpu_addr(robj);
}
- lobj->gpu_offset = robj->gpu_addr;
- lobj->tiling_flags = robj->tiling_flags;
+ lobj->gpu_offset = radeon_bo_gpu_offset(bo);
+ lobj->tiling_flags = bo->tiling_flags;
if (fence) {
- old_fence = (struct radeon_fence *)robj->tobj.sync_obj;
- robj->tobj.sync_obj = radeon_fence_ref(fence);
- robj->tobj.sync_obj_arg = NULL;
+ old_fence = (struct radeon_fence *)bo->tbo.sync_obj;
+ bo->tbo.sync_obj = radeon_fence_ref(fence);
+ bo->tbo.sync_obj_arg = NULL;
}
if (old_fence) {
radeon_fence_unref(&old_fence);
@@ -471,51 +356,44 @@ int radeon_object_list_validate(struct list_head *head, void *fence)
return 0;
}

-void radeon_object_list_unvalidate(struct list_head *head)
+void radeon_bo_list_unvalidate(struct list_head *head, void *fence)
{
- struct radeon_object_list *lobj;
- struct radeon_fence *old_fence = NULL;
+ struct radeon_bo_list *lobj;
+ struct radeon_fence *old_fence;

- list_for_each_entry(lobj, head, list) {
- old_fence = (struct radeon_fence *)lobj->robj->tobj.sync_obj;
- lobj->robj->tobj.sync_obj = NULL;
- if (old_fence) {
- radeon_fence_unref(&old_fence);
+ if (fence)
+ list_for_each_entry(lobj, head, list) {
+ old_fence = to_radeon_fence(lobj->bo->tbo.sync_obj);
+ if (old_fence == fence) {
+ lobj->bo->tbo.sync_obj = NULL;
+ radeon_fence_unref(&old_fence);
+ }
}
- }
- radeon_object_list_unreserve(head);
-}
-
-void radeon_object_list_clean(struct list_head *head)
-{
- radeon_object_list_unreserve(head);
+ radeon_bo_list_unreserve(head);
}

-int radeon_object_fbdev_mmap(struct radeon_object *robj,
+int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
struct vm_area_struct *vma)
{
- return ttm_fbdev_mmap(vma, &robj->tobj);
+ return ttm_fbdev_mmap(vma, &bo->tbo);
}

-unsigned long radeon_object_size(struct radeon_object *robj)
+static int radeon_bo_get_surface_reg(struct radeon_bo *bo)
{
- return robj->tobj.num_pages << PAGE_SHIFT;
-}
-
-int radeon_object_get_surface_reg(struct radeon_object *robj)
-{
- struct radeon_device *rdev = robj->rdev;
+ struct radeon_device *rdev = bo->rdev;
struct radeon_surface_reg *reg;
- struct radeon_object *old_object;
+ struct radeon_bo *old_object;
int steal;
int i;

- if (!robj->tiling_flags)
+ BUG_ON(!atomic_read(&bo->tbo.reserved));
+
+ if (!bo->tiling_flags)
return 0;

- if (robj->surface_reg >= 0) {
- reg = &rdev->surface_regs[robj->surface_reg];
- i = robj->surface_reg;
+ if (bo->surface_reg >= 0) {
+ reg = &rdev->surface_regs[bo->surface_reg];
+ i = bo->surface_reg;
goto out;
}

@@ -523,10 +401,10 @@ int radeon_object_get_surface_reg(struct radeon_object *robj)
for (i = 0; i < RADEON_GEM_MAX_SURFACES; i++) {

reg = &rdev->surface_regs[i];
- if (!reg->robj)
+ if (!reg->bo)
break;

- old_object = reg->robj;
+ old_object = reg->bo;
if (old_object->pin_count == 0)
steal = i;
}
@@ -537,91 +415,101 @@ int radeon_object_get_surface_reg(struct radeon_object *robj)
return -ENOMEM;
/* find someone with a surface reg and nuke their BO */
reg = &rdev->surface_regs[steal];
- old_object = reg->robj;
+ old_object = reg->bo;
/* blow away the mapping */
DRM_DEBUG("stealing surface reg %d from %p\n", steal, old_object);
- ttm_bo_unmap_virtual(&old_object->tobj);
+ ttm_bo_unmap_virtual(&old_object->tbo);
old_object->surface_reg = -1;
i = steal;
}

- robj->surface_reg = i;
- reg->robj = robj;
+ bo->surface_reg = i;
+ reg->bo = bo;

out:
- radeon_set_surface_reg(rdev, i, robj->tiling_flags, robj->pitch,
- robj->tobj.mem.mm_node->start << PAGE_SHIFT,
- robj->tobj.num_pages << PAGE_SHIFT);
+ radeon_set_surface_reg(rdev, i, bo->tiling_flags, bo->pitch,
+ bo->tbo.mem.mm_node->start << PAGE_SHIFT,
+ bo->tbo.num_pages << PAGE_SHIFT);
return 0;
}

-void radeon_object_clear_surface_reg(struct radeon_object *robj)
+static void radeon_bo_clear_surface_reg(struct radeon_bo *bo)
{
- struct radeon_device *rdev = robj->rdev;
+ struct radeon_device *rdev = bo->rdev;
struct radeon_surface_reg *reg;

- if (robj->surface_reg == -1)
+ if (bo->surface_reg == -1)
return;

- reg = &rdev->surface_regs[robj->surface_reg];
- radeon_clear_surface_reg(rdev, robj->surface_reg);
+ reg = &rdev->surface_regs[bo->surface_reg];
+ radeon_clear_surface_reg(rdev, bo->surface_reg);

- reg->robj = NULL;
- robj->surface_reg = -1;
+ reg->bo = NULL;
+ bo->surface_reg = -1;
}

-void radeon_object_set_tiling_flags(struct radeon_object *robj,
- uint32_t tiling_flags, uint32_t pitch)
+int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
+ uint32_t tiling_flags, uint32_t pitch)
{
- robj->tiling_flags = tiling_flags;
- robj->pitch = pitch;
+ int r;
+
+ r = radeon_bo_reserve(bo, false);
+ if (unlikely(r != 0))
+ return r;
+ bo->tiling_flags = tiling_flags;
+ bo->pitch = pitch;
+ radeon_bo_unreserve(bo);
+ return 0;
}

-void radeon_object_get_tiling_flags(struct radeon_object *robj,
- uint32_t *tiling_flags,
- uint32_t *pitch)
+void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
+ uint32_t *tiling_flags,
+ uint32_t *pitch)
{
+ BUG_ON(!atomic_read(&bo->tbo.reserved));
if (tiling_flags)
- *tiling_flags = robj->tiling_flags;
+ *tiling_flags = bo->tiling_flags;
if (pitch)
- *pitch = robj->pitch;
+ *pitch = bo->pitch;
}

-int radeon_object_check_tiling(struct radeon_object *robj, bool has_moved,
- bool force_drop)
+int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
+ bool force_drop)
{
- if (!(robj->tiling_flags & RADEON_TILING_SURFACE))
+ BUG_ON(!atomic_read(&bo->tbo.reserved));
+
+ if (!(bo->tiling_flags & RADEON_TILING_SURFACE))
return 0;

if (force_drop) {
- radeon_object_clear_surface_reg(robj);
+ radeon_bo_clear_surface_reg(bo);
return 0;
}

- if (robj->tobj.mem.mem_type != TTM_PL_VRAM) {
+ if (bo->tbo.mem.mem_type != TTM_PL_VRAM) {
if (!has_moved)
return 0;

- if (robj->surface_reg >= 0)
- radeon_object_clear_surface_reg(robj);
+ if (bo->surface_reg >= 0)
+ radeon_bo_clear_surface_reg(bo);
return 0;
}

- if ((robj->surface_reg >= 0) && !has_moved)
+ if ((bo->surface_reg >= 0) && !has_moved)
return 0;

- return radeon_object_get_surface_reg(robj);
+ return radeon_bo_get_surface_reg(bo);
}

void radeon_bo_move_notify(struct ttm_buffer_object *bo,
- struct ttm_mem_reg *mem)
+ struct ttm_mem_reg *mem)
{
- struct radeon_object *robj = container_of(bo, struct radeon_object, tobj);
- radeon_object_check_tiling(robj, 0, 1);
+ struct radeon_bo *rbo = container_of(bo, struct radeon_bo, tbo);
+ radeon_bo_check_tiling(rbo, 0, 1);
}

void radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
{
- struct radeon_object *robj = container_of(bo, struct radeon_object, tobj);
- radeon_object_check_tiling(robj, 0, 0);
+ struct radeon_bo *rbo = container_of(bo, struct radeon_bo, tbo);
+ radeon_bo_check_tiling(rbo, 0, 0);
}
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index 10e8af6..e9da130 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -28,19 +28,152 @@
#ifndef __RADEON_OBJECT_H__
#define __RADEON_OBJECT_H__

-#include <ttm/ttm_bo_api.h>
-#include <ttm/ttm_bo_driver.h>
-#include <ttm/ttm_placement.h>
-#include <ttm/ttm_module.h>
+#include <drm/radeon_drm.h>
+#include "radeon.h"

-/*
- * TTM.
+/**
+ * radeon_mem_type_to_domain - return domain corresponding to mem_type
+ * @mem_type: ttm memory type
+ *
+ * Returns corresponding domain of the ttm mem_type
+ */
+static inline unsigned radeon_mem_type_to_domain(u32 mem_type)
+{
+ switch (mem_type) {
+ case TTM_PL_VRAM:
+ return RADEON_GEM_DOMAIN_VRAM;
+ case TTM_PL_TT:
+ return RADEON_GEM_DOMAIN_GTT;
+ case TTM_PL_SYSTEM:
+ return RADEON_GEM_DOMAIN_CPU;
+ default:
+ break;
+ }
+ return 0;
+}
+
+/**
+ * radeon_bo_reserve - reserve bo
+ * @bo: bo structure
+ * @no_wait: don't sleep while trying to reserve (return -EBUSY)
+ *
+ * Returns:
+ * -EBUSY: buffer is busy and @no_wait is true
+ * -ERESTART: A wait for the buffer to become unreserved was interrupted by
+ * a signal. Release all buffer reservations and return to user-space.
+ */
+static inline int radeon_bo_reserve(struct radeon_bo *bo, bool no_wait)
+{
+ int r;
+
+retry:
+ r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
+ if (unlikely(r != 0)) {
+ if (r == -ERESTART)
+ goto retry;
+ dev_err(bo->rdev->dev, "%p reserve failed\n", bo);
+ return r;
+ }
+ return 0;
+}
+
+static inline void radeon_bo_unreserve(struct radeon_bo *bo)
+{
+ ttm_bo_unreserve(&bo->tbo);
+}
+
+/**
+ * radeon_bo_gpu_offset - return GPU offset of bo
+ * @bo: radeon object for which we query the offset
+ *
+ * Returns current GPU offset of the object.
+ *
+ * Note: object should either be pinned or reserved when calling this
+ * function, it might be usefull to add check for this for debugging.
+ */
+static inline u64 radeon_bo_gpu_offset(struct radeon_bo *bo)
+{
+ return bo->tbo.offset;
+}
+
+static inline unsigned long radeon_bo_size(struct radeon_bo *bo)
+{
+ return bo->tbo.num_pages << PAGE_SHIFT;
+}
+
+static inline bool radeon_bo_is_reserved(struct radeon_bo *bo)
+{
+ return !!atomic_read(&bo->tbo.reserved);
+}
+
+/**
+ * radeon_bo_mmap_offset - return mmap offset of bo
+ * @bo: radeon object for which we query the offset
+ *
+ * Returns mmap offset of the object.
+ *
+ * Note: addr_space_offset is constant after ttm bo init thus isn't protected
+ * by any lock.
*/
-struct radeon_mman {
- struct ttm_bo_global_ref bo_global_ref;
- struct ttm_global_reference mem_global_ref;
- bool mem_global_referenced;
- struct ttm_bo_device bdev;
-};
+static inline u64 radeon_bo_mmap_offset(struct radeon_bo *bo)
+{
+ return bo->tbo.addr_space_offset;
+}
+
+static inline int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type,
+ bool no_wait)
+{
+ int r;
+
+retry:
+ r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
+ if (unlikely(r != 0)) {
+ if (r == -ERESTART)
+ goto retry;
+ dev_err(bo->rdev->dev, "%p reserve failed for wait\n", bo);
+ return r;
+ }
+ spin_lock(&bo->tbo.lock);
+ if (mem_type)
+ *mem_type = bo->tbo.mem.mem_type;
+ if (bo->tbo.sync_obj)
+ r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
+ spin_unlock(&bo->tbo.lock);
+ ttm_bo_unreserve(&bo->tbo);
+ if (unlikely(r == -ERESTART))
+ goto retry;
+ return r;
+}
+
+extern int radeon_bo_create(struct radeon_device *rdev,
+ struct drm_gem_object *gobj, unsigned long size,
+ bool kernel, u32 domain,
+ struct radeon_bo **bo_ptr);
+extern int radeon_bo_kmap(struct radeon_bo *bo, void **ptr);
+extern void radeon_bo_kunmap(struct radeon_bo *bo);
+extern void radeon_bo_unref(struct radeon_bo **bo);
+extern int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr);
+extern int radeon_bo_unpin(struct radeon_bo *bo);
+extern int radeon_bo_evict_vram(struct radeon_device *rdev);
+extern void radeon_bo_force_delete(struct radeon_device *rdev);
+extern int radeon_bo_init(struct radeon_device *rdev);
+extern void radeon_bo_fini(struct radeon_device *rdev);
+extern void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
+ struct list_head *head);
+extern int radeon_bo_list_reserve(struct list_head *head);
+extern void radeon_bo_list_unreserve(struct list_head *head);
+extern int radeon_bo_list_validate(struct list_head *head, void *fence);
+extern void radeon_bo_list_unvalidate(struct list_head *head, void *fence);
+extern int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
+ struct vm_area_struct *vma);
+extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
+ u32 tiling_flags, u32 pitch);
+extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
+ u32 *tiling_flags, u32 *pitch);
+extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
+ bool force_drop);
+extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
+ struct ttm_mem_reg *mem);
+extern void radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);

#endif
diff --git a/drivers/gpu/drm/radeon/radeon_ring.c b/drivers/gpu/drm/radeon/radeon_ring.c
index 747b4bf..4d12b2d 100644
--- a/drivers/gpu/drm/radeon/radeon_ring.c
+++ b/drivers/gpu/drm/radeon/radeon_ring.c
@@ -165,19 +165,24 @@ int radeon_ib_pool_init(struct radeon_device *rdev)
return 0;
/* Allocate 1M object buffer */
INIT_LIST_HEAD(&rdev->ib_pool.scheduled_ibs);
- r = radeon_object_create(rdev, NULL, RADEON_IB_POOL_SIZE*64*1024,
- true, RADEON_GEM_DOMAIN_GTT,
- false, &rdev->ib_pool.robj);
+ r = radeon_bo_create(rdev, NULL, RADEON_IB_POOL_SIZE*64*1024,
+ true, RADEON_GEM_DOMAIN_GTT,
+ &rdev->ib_pool.robj);
if (r) {
DRM_ERROR("radeon: failed to ib pool (%d).\n", r);
return r;
}
- r = radeon_object_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr);
+ r = radeon_bo_reserve(rdev->ib_pool.robj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr);
if (r) {
+ radeon_bo_unreserve(rdev->ib_pool.robj);
DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r);
return r;
}
- r = radeon_object_kmap(rdev->ib_pool.robj, &ptr);
+ r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr);
+ radeon_bo_unreserve(rdev->ib_pool.robj);
if (r) {
DRM_ERROR("radeon: failed to map ib poll (%d).\n", r);
return r;
@@ -203,14 +208,21 @@ int radeon_ib_pool_init(struct radeon_device *rdev)

void radeon_ib_pool_fini(struct radeon_device *rdev)
{
+ int r;
+
if (!rdev->ib_pool.ready) {
return;
}
mutex_lock(&rdev->ib_pool.mutex);
bitmap_zero(rdev->ib_pool.alloc_bm, RADEON_IB_POOL_SIZE);
if (rdev->ib_pool.robj) {
- radeon_object_kunmap(rdev->ib_pool.robj);
- radeon_object_unref(&rdev->ib_pool.robj);
+ r = radeon_bo_reserve(rdev->ib_pool.robj, false);
+ if (likely(r == 0)) {
+ radeon_bo_kunmap(rdev->ib_pool.robj);
+ radeon_bo_unpin(rdev->ib_pool.robj);
+ radeon_bo_unreserve(rdev->ib_pool.robj);
+ }
+ radeon_bo_unref(&rdev->ib_pool.robj);
rdev->ib_pool.robj = NULL;
}
mutex_unlock(&rdev->ib_pool.mutex);
@@ -288,29 +300,28 @@ int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size)
rdev->cp.ring_size = ring_size;
/* Allocate ring buffer */
if (rdev->cp.ring_obj == NULL) {
- r = radeon_object_create(rdev, NULL, rdev->cp.ring_size,
- true,
- RADEON_GEM_DOMAIN_GTT,
- false,
- &rdev->cp.ring_obj);
+ r = radeon_bo_create(rdev, NULL, rdev->cp.ring_size, true,
+ RADEON_GEM_DOMAIN_GTT,
+ &rdev->cp.ring_obj);
if (r) {
- DRM_ERROR("radeon: failed to create ring buffer (%d).\n", r);
- mutex_unlock(&rdev->cp.mutex);
+ dev_err(rdev->dev, "(%d) ring create failed\n", r);
return r;
}
- r = radeon_object_pin(rdev->cp.ring_obj,
- RADEON_GEM_DOMAIN_GTT,
- &rdev->cp.gpu_addr);
+ r = radeon_bo_reserve(rdev->cp.ring_obj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(rdev->cp.ring_obj, RADEON_GEM_DOMAIN_GTT,
+ &rdev->cp.gpu_addr);
if (r) {
- DRM_ERROR("radeon: failed to pin ring buffer (%d).\n", r);
- mutex_unlock(&rdev->cp.mutex);
+ radeon_bo_unreserve(rdev->cp.ring_obj);
+ dev_err(rdev->dev, "(%d) ring pin failed\n", r);
return r;
}
- r = radeon_object_kmap(rdev->cp.ring_obj,
+ r = radeon_bo_kmap(rdev->cp.ring_obj,
(void **)&rdev->cp.ring);
+ radeon_bo_unreserve(rdev->cp.ring_obj);
if (r) {
- DRM_ERROR("radeon: failed to map ring buffer (%d).\n", r);
- mutex_unlock(&rdev->cp.mutex);
+ dev_err(rdev->dev, "(%d) ring map failed\n", r);
return r;
}
}
@@ -321,11 +332,17 @@ int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size)

void radeon_ring_fini(struct radeon_device *rdev)
{
+ int r;
+
mutex_lock(&rdev->cp.mutex);
if (rdev->cp.ring_obj) {
- radeon_object_kunmap(rdev->cp.ring_obj);
- radeon_object_unpin(rdev->cp.ring_obj);
- radeon_object_unref(&rdev->cp.ring_obj);
+ r = radeon_bo_reserve(rdev->cp.ring_obj, false);
+ if (likely(r == 0)) {
+ radeon_bo_kunmap(rdev->cp.ring_obj);
+ radeon_bo_unpin(rdev->cp.ring_obj);
+ radeon_bo_unreserve(rdev->cp.ring_obj);
+ }
+ radeon_bo_unref(&rdev->cp.ring_obj);
rdev->cp.ring = NULL;
rdev->cp.ring_obj = NULL;
}
diff --git a/drivers/gpu/drm/radeon/radeon_test.c b/drivers/gpu/drm/radeon/radeon_test.c
index f8a465d..391c973 100644
--- a/drivers/gpu/drm/radeon/radeon_test.c
+++ b/drivers/gpu/drm/radeon/radeon_test.c
@@ -30,8 +30,8 @@
/* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */
void radeon_test_moves(struct radeon_device *rdev)
{
- struct radeon_object *vram_obj = NULL;
- struct radeon_object **gtt_obj = NULL;
+ struct radeon_bo *vram_obj = NULL;
+ struct radeon_bo **gtt_obj = NULL;
struct radeon_fence *fence = NULL;
uint64_t gtt_addr, vram_addr;
unsigned i, n, size;
@@ -52,38 +52,42 @@ void radeon_test_moves(struct radeon_device *rdev)
goto out_cleanup;
}

- r = radeon_object_create(rdev, NULL, size, true, RADEON_GEM_DOMAIN_VRAM,
- false, &vram_obj);
+ r = radeon_bo_create(rdev, NULL, size, true, RADEON_GEM_DOMAIN_VRAM,
+ &vram_obj);
if (r) {
DRM_ERROR("Failed to create VRAM object\n");
goto out_cleanup;
}
-
- r = radeon_object_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
+ r = radeon_bo_reserve(vram_obj, false);
+ if (unlikely(r != 0))
+ goto out_cleanup;
+ r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
if (r) {
DRM_ERROR("Failed to pin VRAM object\n");
goto out_cleanup;
}
-
for (i = 0; i < n; i++) {
void *gtt_map, *vram_map;
void **gtt_start, **gtt_end;
void **vram_start, **vram_end;

- r = radeon_object_create(rdev, NULL, size, true,
- RADEON_GEM_DOMAIN_GTT, false, gtt_obj + i);
+ r = radeon_bo_create(rdev, NULL, size, true,
+ RADEON_GEM_DOMAIN_GTT, gtt_obj + i);
if (r) {
DRM_ERROR("Failed to create GTT object %d\n", i);
goto out_cleanup;
}

- r = radeon_object_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, &gtt_addr);
+ r = radeon_bo_reserve(gtt_obj[i], false);
+ if (unlikely(r != 0))
+ goto out_cleanup;
+ r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, &gtt_addr);
if (r) {
DRM_ERROR("Failed to pin GTT object %d\n", i);
goto out_cleanup;
}

- r = radeon_object_kmap(gtt_obj[i], &gtt_map);
+ r = radeon_bo_kmap(gtt_obj[i], &gtt_map);
if (r) {
DRM_ERROR("Failed to map GTT object %d\n", i);
goto out_cleanup;
@@ -94,7 +98,7 @@ void radeon_test_moves(struct radeon_device *rdev)
gtt_start++)
*gtt_start = gtt_start;

- radeon_object_kunmap(gtt_obj[i]);
+ radeon_bo_kunmap(gtt_obj[i]);

r = radeon_fence_create(rdev, &fence);
if (r) {
@@ -116,7 +120,7 @@ void radeon_test_moves(struct radeon_device *rdev)

radeon_fence_unref(&fence);

- r = radeon_object_kmap(vram_obj, &vram_map);
+ r = radeon_bo_kmap(vram_obj, &vram_map);
if (r) {
DRM_ERROR("Failed to map VRAM object after copy %d\n", i);
goto out_cleanup;
@@ -131,13 +135,13 @@ void radeon_test_moves(struct radeon_device *rdev)
"expected 0x%p (GTT map 0x%p-0x%p)\n",
i, *vram_start, gtt_start, gtt_map,
gtt_end);
- radeon_object_kunmap(vram_obj);
+ radeon_bo_kunmap(vram_obj);
goto out_cleanup;
}
*vram_start = vram_start;
}

- radeon_object_kunmap(vram_obj);
+ radeon_bo_kunmap(vram_obj);

r = radeon_fence_create(rdev, &fence);
if (r) {
@@ -159,7 +163,7 @@ void radeon_test_moves(struct radeon_device *rdev)

radeon_fence_unref(&fence);

- r = radeon_object_kmap(gtt_obj[i], &gtt_map);
+ r = radeon_bo_kmap(gtt_obj[i], &gtt_map);
if (r) {
DRM_ERROR("Failed to map GTT object after copy %d\n", i);
goto out_cleanup;
@@ -174,12 +178,12 @@ void radeon_test_moves(struct radeon_device *rdev)
"expected 0x%p (VRAM map 0x%p-0x%p)\n",
i, *gtt_start, vram_start, vram_map,
vram_end);
- radeon_object_kunmap(gtt_obj[i]);
+ radeon_bo_kunmap(gtt_obj[i]);
goto out_cleanup;
}
}

- radeon_object_kunmap(gtt_obj[i]);
+ radeon_bo_kunmap(gtt_obj[i]);

DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
gtt_addr - rdev->mc.gtt_location);
@@ -187,14 +191,20 @@ void radeon_test_moves(struct radeon_device *rdev)

out_cleanup:
if (vram_obj) {
- radeon_object_unpin(vram_obj);
- radeon_object_unref(&vram_obj);
+ if (radeon_bo_is_reserved(vram_obj)) {
+ radeon_bo_unpin(vram_obj);
+ radeon_bo_unreserve(vram_obj);
+ }
+ radeon_bo_unref(&vram_obj);
}
if (gtt_obj) {
for (i = 0; i < n; i++) {
if (gtt_obj[i]) {
- radeon_object_unpin(gtt_obj[i]);
- radeon_object_unref(&gtt_obj[i]);
+ if (radeon_bo_is_reserved(gtt_obj[i])) {
+ radeon_bo_unpin(gtt_obj[i]);
+ radeon_bo_unreserve(gtt_obj[i]);
+ }
+ radeon_bo_unref(&gtt_obj[i]);
}
}
kfree(gtt_obj);
@@ -206,4 +216,3 @@ out_cleanup:
printk(KERN_WARNING "Error while testing BO move.\n");
}
}
-
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index f489c0d..6b6b459 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -150,7 +150,7 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
man->default_caching = TTM_PL_FLAG_CACHED;
break;
case TTM_PL_TT:
- man->gpu_offset = 0;
+ man->gpu_offset = rdev->mc.gtt_location;
man->available_caching = TTM_PL_MASK_CACHING;
man->default_caching = TTM_PL_FLAG_CACHED;
man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | TTM_MEMTYPE_FLAG_CMA;
@@ -180,7 +180,7 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
break;
case TTM_PL_VRAM:
/* "On-card" video ram */
- man->gpu_offset = 0;
+ man->gpu_offset = rdev->mc.vram_location;
man->flags = TTM_MEMTYPE_FLAG_FIXED |
TTM_MEMTYPE_FLAG_NEEDS_IOREMAP |
TTM_MEMTYPE_FLAG_MAPPABLE;
@@ -482,27 +482,31 @@ int radeon_ttm_init(struct radeon_device *rdev)
DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
return r;
}
- r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM, 0,
- ((rdev->mc.real_vram_size) >> PAGE_SHIFT));
+ r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM,
+ 0, rdev->mc.real_vram_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing VRAM heap.\n");
return r;
}
- r = radeon_object_create(rdev, NULL, 256 * 1024, true,
- RADEON_GEM_DOMAIN_VRAM, false,
- &rdev->stollen_vga_memory);
+ r = radeon_bo_create(rdev, NULL, 256 * 1024, true,
+ RADEON_GEM_DOMAIN_VRAM,
+ &rdev->stollen_vga_memory);
if (r) {
return r;
}
- r = radeon_object_pin(rdev->stollen_vga_memory, RADEON_GEM_DOMAIN_VRAM, NULL);
+ r = radeon_bo_reserve(rdev->stollen_vga_memory, false);
+ if (r)
+ return r;
+ r = radeon_bo_pin(rdev->stollen_vga_memory, RADEON_GEM_DOMAIN_VRAM, NULL);
+ radeon_bo_unreserve(rdev->stollen_vga_memory);
if (r) {
- radeon_object_unref(&rdev->stollen_vga_memory);
+ radeon_bo_unref(&rdev->stollen_vga_memory);
return r;
}
DRM_INFO("radeon: %uM of VRAM memory ready\n",
(unsigned)rdev->mc.real_vram_size / (1024 * 1024));
- r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_TT, 0,
- ((rdev->mc.gtt_size) >> PAGE_SHIFT));
+ r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_TT,
+ 0, rdev->mc.gtt_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing GTT heap.\n");
return r;
@@ -523,9 +527,15 @@ int radeon_ttm_init(struct radeon_device *rdev)

void radeon_ttm_fini(struct radeon_device *rdev)
{
+ int r;
+
if (rdev->stollen_vga_memory) {
- radeon_object_unpin(rdev->stollen_vga_memory);
- radeon_object_unref(&rdev->stollen_vga_memory);
+ r = radeon_bo_reserve(rdev->stollen_vga_memory, false);
+ if (r == 0) {
+ radeon_bo_unpin(rdev->stollen_vga_memory);
+ radeon_bo_unreserve(rdev->stollen_vga_memory);
+ }
+ radeon_bo_unref(&rdev->stollen_vga_memory);
}
ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_VRAM);
ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_TT);
diff --git a/drivers/gpu/drm/radeon/rs400.c b/drivers/gpu/drm/radeon/rs400.c
index ca03716..968016b 100644
--- a/drivers/gpu/drm/radeon/rs400.c
+++ b/drivers/gpu/drm/radeon/rs400.c
@@ -452,7 +452,7 @@ void rs400_fini(struct radeon_device *rdev)
rs400_gart_fini(rdev);
radeon_irq_kms_fini(rdev);
radeon_fence_driver_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
@@ -510,7 +510,7 @@ int rs400_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
r = rs400_gart_init(rdev);
diff --git a/drivers/gpu/drm/radeon/rs600.c b/drivers/gpu/drm/radeon/rs600.c
index 5f117cd..30dc65d 100644
--- a/drivers/gpu/drm/radeon/rs600.c
+++ b/drivers/gpu/drm/radeon/rs600.c
@@ -146,15 +146,20 @@ int rs600_gart_enable(struct radeon_device *rdev)

void rs600_gart_disable(struct radeon_device *rdev)
{
- uint32_t tmp;
+ u32 tmp;
+ int r;

/* FIXME: disable out of gart access */
WREG32_MC(R_000100_MC_PT0_CNTL, 0);
tmp = RREG32_MC(R_000009_MC_CNTL1);
WREG32_MC(R_000009_MC_CNTL1, tmp & C_000009_ENABLE_PAGE_TABLES);
if (rdev->gart.table.vram.robj) {
- radeon_object_kunmap(rdev->gart.table.vram.robj);
- radeon_object_unpin(rdev->gart.table.vram.robj);
+ r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
+ if (r == 0) {
+ radeon_bo_kunmap(rdev->gart.table.vram.robj);
+ radeon_bo_unpin(rdev->gart.table.vram.robj);
+ radeon_bo_unreserve(rdev->gart.table.vram.robj);
+ }
}
}

@@ -445,7 +450,7 @@ void rs600_fini(struct radeon_device *rdev)
rs600_gart_fini(rdev);
radeon_irq_kms_fini(rdev);
radeon_fence_driver_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
@@ -505,7 +510,7 @@ int rs600_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
r = rs600_gart_init(rdev);
diff --git a/drivers/gpu/drm/radeon/rs690.c b/drivers/gpu/drm/radeon/rs690.c
index 2754717..2d576af 100644
--- a/drivers/gpu/drm/radeon/rs690.c
+++ b/drivers/gpu/drm/radeon/rs690.c
@@ -662,7 +662,7 @@ void rs690_fini(struct radeon_device *rdev)
rs400_gart_fini(rdev);
radeon_irq_kms_fini(rdev);
radeon_fence_driver_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
@@ -723,7 +723,7 @@ int rs690_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
r = rs400_gart_init(rdev);
diff --git a/drivers/gpu/drm/radeon/rv515.c b/drivers/gpu/drm/radeon/rv515.c
index 7935f79..10b10c1 100644
--- a/drivers/gpu/drm/radeon/rv515.c
+++ b/drivers/gpu/drm/radeon/rv515.c
@@ -541,11 +541,11 @@ void rv515_fini(struct radeon_device *rdev)
r100_wb_fini(rdev);
r100_ib_fini(rdev);
radeon_gem_fini(rdev);
- rv370_pcie_gart_fini(rdev);
+ rv370_pcie_gart_fini(rdev);
radeon_agp_fini(rdev);
radeon_irq_kms_fini(rdev);
radeon_fence_driver_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
@@ -604,7 +604,7 @@ int rv515_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
r = rv370_pcie_gart_init(rdev);
diff --git a/drivers/gpu/drm/radeon/rv770.c b/drivers/gpu/drm/radeon/rv770.c
index b0efd0d..c90c7c3 100644
--- a/drivers/gpu/drm/radeon/rv770.c
+++ b/drivers/gpu/drm/radeon/rv770.c
@@ -92,7 +92,7 @@ int rv770_pcie_gart_enable(struct radeon_device *rdev)
void rv770_pcie_gart_disable(struct radeon_device *rdev)
{
u32 tmp;
- int i;
+ int i, r;

/* Disable all tables */
for (i = 0; i < 7; i++)
@@ -113,8 +113,12 @@ void rv770_pcie_gart_disable(struct radeon_device *rdev)
WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
if (rdev->gart.table.vram.robj) {
- radeon_object_kunmap(rdev->gart.table.vram.robj);
- radeon_object_unpin(rdev->gart.table.vram.robj);
+ r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
+ if (likely(r == 0)) {
+ radeon_bo_kunmap(rdev->gart.table.vram.robj);
+ radeon_bo_unpin(rdev->gart.table.vram.robj);
+ radeon_bo_unreserve(rdev->gart.table.vram.robj);
+ }
}
}

@@ -880,8 +884,12 @@ static int rv770_startup(struct radeon_device *rdev)
}
rv770_gpu_init(rdev);

- r = radeon_object_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
- &rdev->r600_blit.shader_gpu_addr);
+ r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
+ &rdev->r600_blit.shader_gpu_addr);
+ radeon_bo_unreserve(rdev->r600_blit.shader_obj);
if (r) {
DRM_ERROR("failed to pin blit object %d\n", r);
return r;
@@ -934,13 +942,19 @@ int rv770_resume(struct radeon_device *rdev)

int rv770_suspend(struct radeon_device *rdev)
{
+ int r;
+
/* FIXME: we should wait for ring to be empty */
r700_cp_stop(rdev);
rdev->cp.ready = false;
r600_wb_disable(rdev);
rv770_pcie_gart_disable(rdev);
/* unpin shaders bo */
- radeon_object_unpin(rdev->r600_blit.shader_obj);
+ r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
+ if (likely(r == 0)) {
+ radeon_bo_unpin(rdev->r600_blit.shader_obj);
+ radeon_bo_unreserve(rdev->r600_blit.shader_obj);
+ }
return 0;
}

@@ -998,7 +1012,7 @@ int rv770_init(struct radeon_device *rdev)
if (r)
return r;
/* Memory manager */
- r = radeon_object_init(rdev);
+ r = radeon_bo_init(rdev);
if (r)
return r;
rdev->cp.ring_obj = NULL;
@@ -1059,7 +1073,7 @@ void rv770_fini(struct radeon_device *rdev)
radeon_clocks_fini(rdev);
if (rdev->flags & RADEON_IS_AGP)
radeon_agp_fini(rdev);
- radeon_object_fini(rdev);
+ radeon_bo_fini(rdev);
radeon_atombios_fini(rdev);
kfree(rdev->bios);
rdev->bios = NULL;
--
1.6.5.2

2009-11-20 13:30:11

by Jerome Glisse

Subject: [PATCH 3/3] drm/radeon/kms: Add range pinning to crtc/cursor bo

We force the crtc & cursor bo to be in the first 64M, this
will help on legacy modesetting hw where the offset of
scanout buffer and cursor are relative to a base address.

Signed-off-by: Jerome Glisse <[email protected]>
---
drivers/gpu/drm/radeon/radeon.h | 3 ++
drivers/gpu/drm/radeon/radeon_cursor.c | 3 +-
drivers/gpu/drm/radeon/radeon_gem.c | 15 +++++++++++
drivers/gpu/drm/radeon/radeon_legacy_crtc.c | 6 ++++-
drivers/gpu/drm/radeon/radeon_object.c | 35 +++++++++++++++++++++++++++
drivers/gpu/drm/radeon/radeon_object.h | 2 +
6 files changed, 62 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 4b5d86b..5cb1b79 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -245,6 +245,9 @@ int radeon_gem_object_create(struct radeon_device *rdev, int size,
struct drm_gem_object **obj);
int radeon_gem_object_pin(struct drm_gem_object *obj, uint32_t pin_domain,
uint64_t *gpu_addr);
+extern int radeon_gem_object_pin_range(struct drm_gem_object *obj,
+ uint32_t pin_domain, uint64_t *gpu_addr,
+ unsigned long start, unsigned long end);
void radeon_gem_object_unpin(struct drm_gem_object *obj);


diff --git a/drivers/gpu/drm/radeon/radeon_cursor.c b/drivers/gpu/drm/radeon/radeon_cursor.c
index 28772a3..205fca1 100644
--- a/drivers/gpu/drm/radeon/radeon_cursor.c
+++ b/drivers/gpu/drm/radeon/radeon_cursor.c
@@ -156,7 +156,8 @@ int radeon_crtc_cursor_set(struct drm_crtc *crtc,
return -EINVAL;
}

- ret = radeon_gem_object_pin(obj, RADEON_GEM_DOMAIN_VRAM, &gpu_addr);
+ ret = radeon_gem_object_pin_range(obj, RADEON_GEM_DOMAIN_VRAM,
+ &gpu_addr, 0, 64 * 1024 * 1024);
if (ret)
goto fail;

diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c
index f3fb2bf..7db8fcb 100644
--- a/drivers/gpu/drm/radeon/radeon_gem.c
+++ b/drivers/gpu/drm/radeon/radeon_gem.c
@@ -92,6 +92,21 @@ int radeon_gem_object_pin(struct drm_gem_object *obj, uint32_t pin_domain,
return r;
}

+int radeon_gem_object_pin_range(struct drm_gem_object *obj,
+ uint32_t pin_domain, uint64_t *gpu_addr,
+ unsigned long start, unsigned long end)
+{
+ struct radeon_bo *robj = obj->driver_private;
+ int r;
+
+ r = radeon_bo_reserve(robj, false);
+ if (unlikely(r != 0))
+ return r;
+ r = radeon_bo_pin_range(robj, pin_domain, gpu_addr, start, end);
+ radeon_bo_unreserve(robj);
+ return r;
+}
+
void radeon_gem_object_unpin(struct drm_gem_object *obj)
{
struct radeon_bo *robj = obj->driver_private;
diff --git a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
index e8a984d..3a2fb68 100644
--- a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
+++ b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
@@ -439,7 +439,11 @@ int radeon_crtc_set_base(struct drm_crtc *crtc, int x, int y,
r = radeon_bo_reserve(rbo, false);
if (unlikely(r != 0))
return r;
- r = radeon_bo_pin(rbo, RADEON_GEM_DOMAIN_VRAM, &base);
+ /* Pin buffer in the first 64M and always set crtc_base to start of
+ * vram
+ */
+ r = radeon_bo_pin_range(rbo, RADEON_GEM_DOMAIN_VRAM, &base,
+ 0, 64 * 1024 * 1024);
if (unlikely(r != 0)) {
radeon_bo_unreserve(rbo);
return -EINVAL;
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index bec4943..15399e6 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -200,6 +200,41 @@ retry:
return r;
}

+int radeon_bo_pin_range(struct radeon_bo *bo, u32 domain, u64 *gpu_addr,
+ unsigned long start, unsigned long end)
+{
+ u32 flags;
+ u32 tmp;
+ int r;
+
+ flags = radeon_ttm_flags_from_domain(domain);
+ if (bo->pin_count) {
+ bo->pin_count++;
+ if (gpu_addr)
+ *gpu_addr = radeon_bo_gpu_offset(bo);
+ return 0;
+ }
+ tmp = bo->tbo.mem.placement;
+ ttm_flag_masked(&tmp, flags, TTM_PL_MASK_MEM);
+ bo->tbo.proposed_placement = tmp | TTM_PL_FLAG_NO_EVICT |
+ TTM_PL_MASK_CACHING;
+retry:
+ r = ttm_buffer_object_validate_range(&bo->tbo,
+ bo->tbo.proposed_placement,
+ start, end, true, false);
+ if (likely(r == 0)) {
+ bo->pin_count = 1;
+ if (gpu_addr != NULL)
+ *gpu_addr = radeon_bo_gpu_offset(bo);
+ }
+ if (unlikely(r != 0)) {
+ if (r == -ERESTART)
+ goto retry;
+ dev_err(bo->rdev->dev, "%p pin failed\n", bo);
+ }
+ return r;
+}
+
int radeon_bo_unpin(struct radeon_bo *bo)
{
int r;
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index e9da130..a40bf67 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -153,6 +153,8 @@ extern int radeon_bo_kmap(struct radeon_bo *bo, void **ptr);
extern void radeon_bo_kunmap(struct radeon_bo *bo);
extern void radeon_bo_unref(struct radeon_bo **bo);
extern int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr);
+extern int radeon_bo_pin_range(struct radeon_bo *bo, u32 domain, u64 *gpu_addr,
+ unsigned long start, unsigned long end);
extern int radeon_bo_unpin(struct radeon_bo *bo);
extern int radeon_bo_evict_vram(struct radeon_device *rdev);
extern void radeon_bo_force_delete(struct radeon_device *rdev);
--
1.6.5.2

2009-11-20 16:47:20

by Alex Deucher

Subject: Re: [PATCH 3/3] drm/radeon/kms: Add range pinning to crtc/cursor bo

On Fri, Nov 20, 2009 at 8:29 AM, Jerome Glisse <[email protected]> wrote:
> We force the crtc & cursor bo to be in the first 64M, this
> will help on legacy modesetting hw where the offset of
> scanout buffer and cursor are relative to a base address.

This limitation only applies to pre-avivo (r1xx-r4xx) chips, there's
no need on newer hardware. Also the cursor and crtc have to be within
128 MB of each other, so you could bump the limits to 128 MB. Another
thing to consider down the road might be to treat memory over 128 MB
as "invisible" on these pre-avivo chips (once we support that in
general) then we no longer need this limit.

Alex

>
> Signed-off-by: Jerome Glisse <[email protected]>
> ---
> ?drivers/gpu/drm/radeon/radeon.h ? ? ? ? ? ? | ? ?3 ++
> ?drivers/gpu/drm/radeon/radeon_cursor.c ? ? ?| ? ?3 +-
> ?drivers/gpu/drm/radeon/radeon_gem.c ? ? ? ? | ? 15 +++++++++++
> ?drivers/gpu/drm/radeon/radeon_legacy_crtc.c | ? ?6 ++++-
> ?drivers/gpu/drm/radeon/radeon_object.c ? ? ?| ? 35 +++++++++++++++++++++++++++
> ?drivers/gpu/drm/radeon/radeon_object.h ? ? ?| ? ?2 +
> ?6 files changed, 62 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
> index 4b5d86b..5cb1b79 100644
> --- a/drivers/gpu/drm/radeon/radeon.h
> +++ b/drivers/gpu/drm/radeon/radeon.h
> @@ -245,6 +245,9 @@ int radeon_gem_object_create(struct radeon_device *rdev, int size,
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?struct drm_gem_object **obj);
> ?int radeon_gem_object_pin(struct drm_gem_object *obj, uint32_t pin_domain,
> ? ? ? ? ? ? ? ? ? ? ? ? ?uint64_t *gpu_addr);
> +extern int radeon_gem_object_pin_range(struct drm_gem_object *obj,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? uint32_t pin_domain, uint64_t *gpu_addr,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? unsigned long start, unsigned long end);
> ?void radeon_gem_object_unpin(struct drm_gem_object *obj);
>
>
> diff --git a/drivers/gpu/drm/radeon/radeon_cursor.c b/drivers/gpu/drm/radeon/radeon_cursor.c
> index 28772a3..205fca1 100644
> --- a/drivers/gpu/drm/radeon/radeon_cursor.c
> +++ b/drivers/gpu/drm/radeon/radeon_cursor.c
> @@ -156,7 +156,8 @@ int radeon_crtc_cursor_set(struct drm_crtc *crtc,
> ? ? ? ? ? ? ? ?return -EINVAL;
> ? ? ? ?}
>
> - ? ? ? ret = radeon_gem_object_pin(obj, RADEON_GEM_DOMAIN_VRAM, &gpu_addr);
> + ? ? ? ret = radeon_gem_object_pin_range(obj, RADEON_GEM_DOMAIN_VRAM,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? &gpu_addr, 0, 64 * 1024 * 1024);
> ? ? ? ?if (ret)
> ? ? ? ? ? ? ? ?goto fail;
>
> diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c
> index f3fb2bf..7db8fcb 100644
> --- a/drivers/gpu/drm/radeon/radeon_gem.c
> +++ b/drivers/gpu/drm/radeon/radeon_gem.c
> @@ -92,6 +92,21 @@ int radeon_gem_object_pin(struct drm_gem_object *obj, uint32_t pin_domain,
> ? ? ? ?return r;
> ?}
>
> +int radeon_gem_object_pin_range(struct drm_gem_object *obj,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? uint32_t pin_domain, uint64_t *gpu_addr,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? unsigned long start, unsigned long end)
> +{
> + ? ? ? struct radeon_bo *robj = obj->driver_private;
> + ? ? ? int r;
> +
> + ? ? ? r = radeon_bo_reserve(robj, false);
> + ? ? ? if (unlikely(r != 0))
> + ? ? ? ? ? ? ? return r;
> + ? ? ? r = radeon_bo_pin_range(robj, pin_domain, gpu_addr, start, end);
> + ? ? ? radeon_bo_unreserve(robj);
> + ? ? ? return r;
> +}
> +
> ?void radeon_gem_object_unpin(struct drm_gem_object *obj)
> ?{
> ? ? ? ?struct radeon_bo *robj = obj->driver_private;
> diff --git a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
> index e8a984d..3a2fb68 100644
> --- a/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
> +++ b/drivers/gpu/drm/radeon/radeon_legacy_crtc.c
> @@ -439,7 +439,11 @@ int radeon_crtc_set_base(struct drm_crtc *crtc, int x, int y,
> ? ? ? ?r = radeon_bo_reserve(rbo, false);
> ? ? ? ?if (unlikely(r != 0))
> ? ? ? ? ? ? ? ?return r;
> - ? ? ? r = radeon_bo_pin(rbo, RADEON_GEM_DOMAIN_VRAM, &base);
> + ? ? ? /* Pin buffer in the first 64M and always set crtc_base to start of
> + ? ? ? ?* vram
> + ? ? ? ?*/
> + ? ? ? r = radeon_bo_pin_range(rbo, RADEON_GEM_DOMAIN_VRAM, &base,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 0, 64 * 1024 * 1024);
> ? ? ? ?if (unlikely(r != 0)) {
> ? ? ? ? ? ? ? ?radeon_bo_unreserve(rbo);
> ? ? ? ? ? ? ? ?return -EINVAL;
> diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
> index bec4943..15399e6 100644
> --- a/drivers/gpu/drm/radeon/radeon_object.c
> +++ b/drivers/gpu/drm/radeon/radeon_object.c
> @@ -200,6 +200,41 @@ retry:
> ? ? ? ?return r;
> ?}
>
> +int radeon_bo_pin_range(struct radeon_bo *bo, u32 domain, u64 *gpu_addr,
> + ? ? ? ? ? ? ? ? ? ? ? unsigned long start, unsigned long end)
> +{
> + ? ? ? u32 flags;
> + ? ? ? u32 tmp;
> + ? ? ? int r;
> +
> + ? ? ? flags = radeon_ttm_flags_from_domain(domain);
> + ? ? ? if (bo->pin_count) {
> + ? ? ? ? ? ? ? bo->pin_count++;
> + ? ? ? ? ? ? ? if (gpu_addr)
> + ? ? ? ? ? ? ? ? ? ? ? *gpu_addr = radeon_bo_gpu_offset(bo);
> + ? ? ? ? ? ? ? return 0;
> + ? ? ? }
> + ? ? ? tmp = bo->tbo.mem.placement;
> + ? ? ? ttm_flag_masked(&tmp, flags, TTM_PL_MASK_MEM);
> + ? ? ? bo->tbo.proposed_placement = tmp | TTM_PL_FLAG_NO_EVICT |
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? TTM_PL_MASK_CACHING;
> +retry:
> + ? ? ? r = ttm_buffer_object_validate_range(&bo->tbo,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? bo->tbo.proposed_placement,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? start, end, true, false);
> + ? ? ? if (likely(r == 0)) {
> + ? ? ? ? ? ? ? bo->pin_count = 1;
> + ? ? ? ? ? ? ? if (gpu_addr != NULL)
> + ? ? ? ? ? ? ? ? ? ? ? *gpu_addr = radeon_bo_gpu_offset(bo);
> + ? ? ? }
> + ? ? ? if (unlikely(r != 0)) {
> + ? ? ? ? ? ? ? if (r == -ERESTART)
> + ? ? ? ? ? ? ? ? ? ? ? goto retry;
> + ? ? ? ? ? ? ? dev_err(bo->rdev->dev, "%p pin failed\n", bo);
> + ? ? ? }
> + ? ? ? return r;
> +}
> +
> ?int radeon_bo_unpin(struct radeon_bo *bo)
> ?{
> ? ? ? ?int r;
> diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
> index e9da130..a40bf67 100644
> --- a/drivers/gpu/drm/radeon/radeon_object.h
> +++ b/drivers/gpu/drm/radeon/radeon_object.h
> @@ -153,6 +153,8 @@ extern int radeon_bo_kmap(struct radeon_bo *bo, void **ptr);
> ?extern void radeon_bo_kunmap(struct radeon_bo *bo);
> ?extern void radeon_bo_unref(struct radeon_bo **bo);
> ?extern int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr);
> +extern int radeon_bo_pin_range(struct radeon_bo *bo, u32 domain, u64 *gpu_addr,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? unsigned long start, unsigned long end);
> ?extern int radeon_bo_unpin(struct radeon_bo *bo);
> ?extern int radeon_bo_evict_vram(struct radeon_device *rdev);
> ?extern void radeon_bo_force_delete(struct radeon_device *rdev);
> --
> 1.6.5.2
>
>
> ------------------------------------------------------------------------------
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
> trial. Simplify your report design, integration and deployment - and focus on
> what you do best, core application coding. Discover what's new with
> Crystal Reports now. ?http://p.sf.net/sfu/bobj-july
> --
> _______________________________________________
> Dri-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/dri-devel
>

2009-11-20 18:07:25

by Jerome Glisse

Subject: Re: [PATCH 3/3] drm/radeon/kms: Add range pinning to crtc/cursor bo

On Fri, Nov 20, 2009 at 11:47:22AM -0500, Alex Deucher wrote:
> On Fri, Nov 20, 2009 at 8:29 AM, Jerome Glisse <[email protected]> wrote:
> > We force the crtc & cursor bo to be in the first 64M, this
> > will help on legacy modesetting hw where the offset of
> > scanout buffer and cursor are relative to a base address.
>
> This limitation only applies to pre-avivo (r1xx-r4xx) chips, there's
> no need on newer hardware. Also the cursor and crtc have to be within
> 128 MB of each other, so you could bump the limits to 128 MB. Another
> thing to consider down the road might be to treat memory over 128 MB
> as "invisible" on these pre-avivo chips (once we support that in
> general) then we no longer need this limit.
>
> Alex
>

I choosed 64M in hope to disminish VRAM fragmentation a little,
this is also why i did not bother to make this conditional for
legacy modesetting hw. For supporting non visible vram see my
other ttm thread. I am working on adding this knowledge to ttm.

Cheers,
Jerome

2009-12-01 23:53:11

by Dave Airlie

Subject: Re:

On Fri, Nov 20, 2009 at 11:29 PM, Jerome Glisse <[email protected]> wrote:
> This patch series add ttm range validation function. Aim is to
> include this in 2.6.33 so i have time to iron out issue, comments.

I missed these first time around,

Thomas if you have any opinions on the TTM stuff please see if you
can take a look.

> ttm:
> I duplicated a bunch of ttm functions but now i think, best would
> be to add range to all function and use free list if range cover
> all the manager space. Doing so we might also be able to simplify
> mem_space alocation into a simpler function like ttm_bo_mem_space_range
>
> radeon:
> The second patch is a rework/cleanup of radeon object, it solves
> few issues along the way (i can't remember them now after fews
> days testing the patches). Biggest change is that we now rely
> on BO being validated before doing any change to radeon bo structure.
> As with any big patch i might introduce regressions, so far after
> testing on AGP:R1XX,R2XX,R3XX,R6XX PCIE:R3XX,R4XX,R5XX,R6XX,R7XX
> and RS480,RS690 i didn't found anythings obvious (test being X +
> glxgears + compiz(on hw which support it) + suspend/resume).

I'll try and take a look at this here it doesn't seem to depend on the first
patch so I can push it separate if needed.

Dave.

>
> Last patch is smaller, it just use the interface introduced by
> the first patch.
>
>

2009-12-02 07:17:54

by Thomas Hellstrom

Subject: Re:

Dave Airlie wrote:
> On Fri, Nov 20, 2009 at 11:29 PM, Jerome Glisse <[email protected]> wrote:
>
>> This patch series add ttm range validation function. Aim is to
>> include this in 2.6.33 so i have time to iron out issue, comments.
>>
>
> I missed these first time around,
>
> Thomas if you have any opinions on the TTM stuff please see if you
> can take a look.
>
>
Sure, I'll take a look.

/Thomas

2009-12-02 10:05:12

by Thomas Hellstrom

Subject: Re: [PATCH 1/3] drm/ttm: Add range validation function

Jerome,
Please see comments inline.


Jerome Glisse wrote:
> Add a function to validate buffer in a given range of
> memory. This is needed by some hw for some of their
> buffer (scanout buffer, cursor ...).
>
> Signed-off-by: Jerome Glisse <[email protected]>
> ---
> drivers/gpu/drm/ttm/ttm_bo.c | 305 ++++++++++++++++++++++++++++++++++++++-
> include/drm/ttm/ttm_bo_api.h | 5 +
> include/drm/ttm/ttm_bo_driver.h | 1 +
> 3 files changed, 310 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 87c0625..6b0e7e8 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -247,7 +247,6 @@ EXPORT_SYMBOL(ttm_bo_unreserve);
> /*
> * Call bo->mutex locked.
> */
> -
>

Whitespace.

> static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
> {
> struct ttm_bo_device *bdev = bo->bdev;
> @@ -418,6 +417,7 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo, bool remove_all)
> kref_put(&bo->list_kref, ttm_bo_ref_bug);
> }
> if (bo->mem.mm_node) {
> + bo->mem.mm_node->private = NULL;
> drm_mm_put_block(bo->mem.mm_node);
> bo->mem.mm_node = NULL;
> }
> @@ -610,6 +610,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, unsigned mem_type,
>
> spin_lock(&glob->lru_lock);
> if (evict_mem.mm_node) {
> + evict_mem.mm_node->private = NULL;
> drm_mm_put_block(evict_mem.mm_node);
> evict_mem.mm_node = NULL;
> }
>
> @@ -826,6 +827,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,
> mem->mm_node = node;
> mem->mem_type = mem_type;
> mem->placement = cur_flags;
> + node->private = bo;
> return 0;
> }
>
>

I'd like a big warning here somewhere, because since the pointer to the
bo is not refcounted, it may
never be dereferenced outside of the lru spinlock, because as soon as
you release the lru spinlock, someone
else may destroy the buffer. The current usage is valid AFAICT. If you
choose to refcount it, use bo::list_kref.


> @@ -856,6 +858,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,
>
> if (ret == 0 && mem->mm_node) {
> mem->placement = cur_flags;
> + mem->mm_node->private = bo;
> return 0;
> }
>
> @@ -868,6 +871,173 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,
> }
> EXPORT_SYMBOL(ttm_bo_mem_space);
>
> +static unsigned long ttm_bo_free_size_if_evicted(struct ttm_buffer_object *bo)
> +{
> + struct ttm_mem_type_manager *man = &bo->bdev->man[bo->mem.mem_type];
> + struct drm_mm_node *node;
> + unsigned long size;
> +
> + size = bo->mem.mm_node->size;
> + if (bo->mem.mm_node->ml_entry.prev != &man->manager.ml_entry) {
> + node = list_entry(bo->mem.mm_node->ml_entry.prev,
> + struct drm_mm_node, ml_entry);
> + if (node->free)
> + size += node->size;
> + }
> + if (bo->mem.mm_node->ml_entry.next != &man->manager.ml_entry) {
> + node = list_entry(bo->mem.mm_node->ml_entry.next,
> + struct drm_mm_node, ml_entry);
> + if (node->free)
> + size += node->size;
> + }
> + return size;
> +}
> +
> +static void ttm_manager_evict_first(struct ttm_buffer_object *bo)
> +{
> + struct ttm_mem_type_manager *man = &bo->bdev->man[bo->mem.mem_type];
> + unsigned long free_size_bo, free_size_bo_first;
> + struct ttm_buffer_object *bo_first;
> +
> + /* BO is not on lru list, don't add it */
> + if (!list_empty(&bo->lru))
> + return;
> + bo_first = list_first_entry(&man->lru, struct ttm_buffer_object, lru);
> + free_size_bo = ttm_bo_free_size_if_evicted(bo);
> + free_size_bo_first = ttm_bo_free_size_if_evicted(bo_first);
> + if (free_size_bo > free_size_bo_first) {
> + list_del_init(&bo->lru);
> + list_add(&bo->lru, &man->lru);
> + }
> +}
> +
>

Hmm, Can you explain why we'd ever *not* want to put bo first on the lru
list?
I mean , bo_first might not even be in the correct range? That would
also save the size computations.


> +static int ttm_manager_space_range(struct ttm_buffer_object *bo,
> + uint32_t mem_type,
> + struct ttm_mem_reg *mem,
> + unsigned long start, unsigned long end,
> + bool interruptible, bool no_wait)
> +{
> + struct ttm_bo_global *glob = bo->glob;
> + struct drm_mm_node *entry;
> + struct ttm_mem_type_manager *man = &bo->bdev->man[mem_type];
> + struct drm_mm *mm = &man->manager;
> + unsigned size = end - start;
> + struct ttm_buffer_object *tbo;
> + unsigned wasted;
> + int ret;
> +
> + mem->mm_node = NULL;
> + ret = drm_mm_pre_get(&man->manager);
> + if (unlikely(ret))
> + return ret;
> + spin_lock(&glob->lru_lock);
> + list_for_each_entry(entry, &mm->ml_entry, ml_entry) {
> + wasted = 0;
> + if (mem->page_alignment) {
> + unsigned tmp = entry->start % mem->page_alignment;
> + if (tmp)
> + wasted += mem->page_alignment - tmp;
> + }
> + if (entry->start < end && (entry->start+entry->size) > start) {
> + if (!entry->free) {
> + tbo = (struct ttm_buffer_object*)entry->private;
> + ttm_manager_evict_first(tbo);
> + } else {
> + if ((entry->size - wasted) <= size) {
> + /* Found a place */
> + entry = drm_mm_get_block_atomic(entry,
> + mem->num_pages,
> + mem->page_alignment);
> + mem->mm_node = entry;
> + mem->mem_type = mem_type;
> + entry->private = bo;
> + break;
> + }
> + }
> + }
> + }
> + spin_unlock(&glob->lru_lock);
> + return 0;
> +}
>

There's stuff here that belongs in drm_mm.c rather in the ttm code!
What about
drm_mm_search_free_range() // Returns a free block in the given range if
any.
and perhaps functions to return the first non-free entry?


> +
> +static int ttm_bo_mem_space_range(struct ttm_buffer_object *bo,
> + uint32_t proposed_placement,
> + struct ttm_mem_reg *mem,
> + unsigned long start, unsigned long end,
> + bool interruptible, bool no_wait)
> +{
> + struct ttm_bo_device *bdev = bo->bdev;
> + struct ttm_bo_global *glob = bo->glob;
> + struct ttm_mem_type_manager *man;
> + uint32_t num_prios = bdev->driver->num_mem_type_prio;
> + const uint32_t *prios = bdev->driver->mem_type_prio;
> + uint32_t i;
> + uint32_t mem_type = TTM_PL_SYSTEM;
> + uint32_t cur_flags = 0;
> + bool type_found = false;
> + bool type_ok = false;
> + bool evicted;
> + int ret;
> +
> + mem->mm_node = NULL;
> +retry:
> + for (i = 0; i < num_prios; ++i) {
> + mem_type = prios[i];
> + man = &bdev->man[mem_type];
> + type_ok = ttm_bo_mt_compatible(man,
> + bo->type == ttm_bo_type_user,
> + mem_type, proposed_placement,
> + &cur_flags);
> + if (!type_ok)
> + continue;
> + if (start > (man->offset + man->size) || end < man->offset)
> + continue;
> + cur_flags = ttm_bo_select_caching(man, bo->mem.placement,
> + cur_flags);
> + if (mem_type == TTM_PL_SYSTEM)
> + break;
> + if (man->has_type && man->use_type) {
> + type_found = true;
> + mem->placement = cur_flags;
> + ret = ttm_manager_space_range(bo, mem_type, mem,
> + start, end,
> + interruptible, no_wait);
> + if (ret)
> + return ret;
> + }
> + if (mem->mm_node)
> + break;
> + }
> + if ((type_ok && (mem_type == TTM_PL_SYSTEM)) || mem->mm_node) {
> + mem->mem_type = mem_type;
> + mem->placement = cur_flags;
> + return 0;
> + }
> + for (i = 0, evicted = false; i < num_prios; ++i) {
> + struct ttm_buffer_object *tmp;
> +
> + mem_type = prios[i];
> + man = &bdev->man[mem_type];
> + type_ok = ttm_bo_mt_compatible(man,
> + bo->type == ttm_bo_type_user,
> + mem_type, proposed_placement,
> + &cur_flags);
> + if (!type_ok)
> + continue;
> + if (start > (man->offset + man->size) || end < man->offset)
> + continue;
>

> + spin_lock(&glob->lru_lock);
> + tmp = list_first_entry(&man->lru, struct ttm_buffer_object, lru);
> + spin_unlock(&glob->lru_lock);
> + ret = ttm_bo_evict(tmp, mem_type, interruptible, no_wait);
>

^^^^This is incorrect. As soon as you release the lru lock, tmp may be
destroyed by another process.
Furthermore, tmp must be reserved before calling ttm_bo_evict. Please
see how this is done in
ttm_bo_mem_force_space.


> + if (unlikely(ret != 0))
> + return ret;
> + }
> + if (!evicted)
> + return -ENOMEM;
> + goto retry;
> +}
> +
> int ttm_bo_wait_cpu(struct ttm_buffer_object *bo, bool no_wait)
> {
> int ret = 0;
> @@ -925,6 +1095,7 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
> out_unlock:
> if (ret && mem.mm_node) {
> spin_lock(&glob->lru_lock);
> + mem.mm_node->private = NULL;
> drm_mm_put_block(mem.mm_node);
> spin_unlock(&glob->lru_lock);
> }
> @@ -998,6 +1169,137 @@ int ttm_buffer_object_validate(struct ttm_buffer_object *bo,
> }
> EXPORT_SYMBOL(ttm_buffer_object_validate);
>
> +int ttm_bo_move_buffer_range(struct ttm_buffer_object *bo,
> + uint32_t proposed_placement,
> + unsigned long start, unsigned long end,
> + bool interruptible, bool no_wait)
> +{
> + struct ttm_bo_global *glob = bo->glob;
> + int ret = 0;
> + struct ttm_mem_reg mem;
> +
> + BUG_ON(!atomic_read(&bo->reserved));
> +
> + /*
> + * FIXME: It's possible to pipeline buffer moves.
> + * Have the driver move function wait for idle when necessary,
> + * instead of doing it here.
> + */
> +
> + spin_lock(&bo->lock);
> + ret = ttm_bo_wait(bo, false, interruptible, no_wait);
> + spin_unlock(&bo->lock);
> +
> + if (ret)
> + return ret;
> +
> + mem.num_pages = bo->num_pages;
> + mem.size = mem.num_pages << PAGE_SHIFT;
> + mem.page_alignment = bo->mem.page_alignment;
> +
> + /*
> + * Determine where to move the buffer.
> + */
> +
> + ret = ttm_bo_mem_space_range(bo, proposed_placement, &mem,
> + start, end, interruptible, no_wait);
> + if (ret)
> + goto out_unlock;
> +
> + ret = ttm_bo_handle_move_mem(bo, &mem, false, interruptible, no_wait);
> +
> +out_unlock:
> + if (ret && mem.mm_node) {
> + spin_lock(&glob->lru_lock);
> + mem.mm_node->private = NULL;
> + drm_mm_put_block(mem.mm_node);
> + spin_unlock(&glob->lru_lock);
> + }
> + return ret;
> +}
> +
> +static int ttm_bo_mem_compat_range(uint32_t proposed_placement,
> + struct ttm_mem_reg *mem,
> + unsigned long start, unsigned long end)
> +{
> + if ((proposed_placement & mem->placement & TTM_PL_MASK_MEM) == 0)
> + return 0;
> + if ((proposed_placement & mem->placement & TTM_PL_MASK_CACHING) == 0)
> + return 0;
> + if (mem->mm_node) {
> + unsigned long eend = mem->mm_node->start + mem->mm_node->size;
> + if (mem->mm_node->start < start || eend > end)
> + return 0;
> + }
> +
> + return 1;
> +}
> +
> +int ttm_buffer_object_validate_range(struct ttm_buffer_object *bo,
> + uint32_t proposed_placement,
> + unsigned long start, unsigned long end,
> + bool interruptible, bool no_wait)
> +{
> + int ret;
> +
> + /* Convert to page */
> + start = start >> PAGE_SHIFT;
> + end = end >> PAGE_SHIFT;
> + /* Check that range is valid */
> + if (start > end || (end - start) < bo->num_pages)
> + return -EINVAL;
> +
> + BUG_ON(!atomic_read(&bo->reserved));
> + bo->proposed_placement = proposed_placement;
> +
> + TTM_DEBUG("Proposed placement 0x%08lx, Old flags 0x%08lx\n",
> + (unsigned long)proposed_placement,
> + (unsigned long)bo->mem.placement);
> +
> + /*
> + * Check whether we need to move buffer.
> + */
> +
> + if (!ttm_bo_mem_compat_range(bo->proposed_placement, &bo->mem,
> + start, end)) {
> + ret = ttm_bo_move_buffer_range(bo, bo->proposed_placement,
> + start, end,
> + interruptible, no_wait);
> + if (ret) {
> + if (ret != -ERESTART)
> + printk(KERN_ERR TTM_PFX
> + "Failed moving buffer. "
> + "Proposed placement 0x%08x\n",
> + bo->proposed_placement);
> + if (ret == -ENOMEM)
> + printk(KERN_ERR TTM_PFX
> + "Out of aperture space or "
> + "DRM memory quota.\n");
> + return ret;
> + }
> + }
> +
> + /*
> + * We might need to add a TTM.
> + */
> + if (bo->mem.mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
> + ret = ttm_bo_add_ttm(bo, true);
> + if (ret)
> + return ret;
> + }
> + /*
> + * Validation has succeeded, move the access and other
> + * non-mapping-related flag bits from the proposed flags to
> + * the active flags
> + */
> +
> + ttm_flag_masked(&bo->mem.placement, bo->proposed_placement,
> + ~TTM_PL_MASK_MEMTYPE);
> +
> + return 0;
> +}
> +EXPORT_SYMBOL(ttm_buffer_object_validate_range);
> +
> int
> ttm_bo_check_placement(struct ttm_buffer_object *bo,
> uint32_t set_flags, uint32_t clr_flags)
> @@ -1321,6 +1623,7 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
> man->has_type = true;
> man->use_type = true;
> man->size = p_size;
> + man->offset = p_offset;
>
> INIT_LIST_HEAD(&man->lru);
>
> diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
> index 4911461..3756970 100644
> --- a/include/drm/ttm/ttm_bo_api.h
> +++ b/include/drm/ttm/ttm_bo_api.h
> @@ -308,6 +308,11 @@ extern int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy,
> extern int ttm_buffer_object_validate(struct ttm_buffer_object *bo,
> uint32_t proposed_placement,
> bool interruptible, bool no_wait);
> +extern int ttm_buffer_object_validate_range(struct ttm_buffer_object *bo,
> + uint32_t proposed_placement,
> + unsigned long start, unsigned long end,
> + bool interruptible, bool no_wait);
>

Please add a documentation entry in the header.


> +
> /**
> * ttm_bo_unref
> *
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index e8cd6d2..f00ba12 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -224,6 +224,7 @@ struct ttm_mem_type_manager {
> unsigned long gpu_offset;
> unsigned long io_offset;
> unsigned long io_size;
> + unsigned long offset;
> void *io_addr;
> uint64_t size;
> uint32_t available_caching;
>

Please add a doc entry in the header also here.
@offset is a bit confusing given all other offset members. What about
managed_offset or something similar?

2009-12-04 00:10:19

by Dave Airlie

Subject: Re: [PATCH 2/3] drm/radeon/kms: Rework radeon object handling

On Fri, Nov 20, 2009 at 11:29 PM, Jerome Glisse <[email protected]> wrote:
> The locking & protection of radeon object was somewhat messy.
> This patch completely rework it to now use ttm reserve as a
> protection for the radeon object structure member. It also
> shrink down the various radeon object structure by removing
> field which were redondant with the ttm information. Last it
> converts few simple functions to inline which should with
> performances.
>
> Signed-off-by: Jerome Glisse <[email protected]>



>
> ?int radeon_gem_busy_ioctl(struct drm_device *dev, void *data,
> @@ -264,7 +261,7 @@ int radeon_gem_busy_ioctl(struct drm_device *dev, void *data,
> ?{
> ? ? ? ?struct drm_radeon_gem_busy *args = data;
> ? ? ? ?struct drm_gem_object *gobj;
> - ? ? ? struct radeon_object *robj;
> + ? ? ? struct radeon_bo *robj;
> ? ? ? ?int r;
> ? ? ? ?uint32_t cur_placement;
>
> @@ -273,7 +270,7 @@ int radeon_gem_busy_ioctl(struct drm_device *dev, void *data,
> ? ? ? ? ? ? ? ?return -EINVAL;
> ? ? ? ?}
> ? ? ? ?robj = gobj->driver_private;
> - ? ? ? r = radeon_object_busy_domain(robj, &cur_placement);
> + ? ? ? r = radeon_bo_wait(robj, &cur_placement, true);

Here we pass an uninitialised cur_placement to busy ioctl.

> +static inline int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? bool no_wait)
> +{
> + ? ? ? int r;
> +
> +retry:
> + ? ? ? r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
> + ? ? ? if (unlikely(r != 0)) {
> + ? ? ? ? ? ? ? if (r == -ERESTART)
> + ? ? ? ? ? ? ? ? ? ? ? goto retry;
> + ? ? ? ? ? ? ? dev_err(bo->rdev->dev, "%p reserve failed for wait\n", bo);
> + ? ? ? ? ? ? ? return r;
> + ? ? ? }
> + ? ? ? spin_lock(&bo->tbo.lock);
> + ? ? ? if (mem_type)
> + ? ? ? ? ? ? ? *mem_type = bo->tbo.mem.mem_type;
> + ? ? ? if (bo->tbo.sync_obj)
> + ? ? ? ? ? ? ? r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
> + ? ? ? spin_unlock(&bo->tbo.lock);
> + ? ? ? ttm_bo_unreserve(&bo->tbo);
> + ? ? ? if (unlikely(r == -ERESTART))
> + ? ? ? ? ? ? ? goto retry;
> + ? ? ? return r;
> +}

If this return early mem_type never gets set to anything useful.

The compiler warns on this here.

Can you send a fix up patch as I've merged this to drm-radeon-next already.

Dave.