void radeon_ttm_fini(struct radeon_device *rdev)
{
int r;
if (!rdev->mman.initialized)
return;
if (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);
ttm_bo_device_release(&rdev->mman.bdev);
radeon_gart_fini(rdev);
radeon_ttm_global_fini(rdev);
rdev->mman.initialized = false;
DRM_INFO("radeon: ttm finalized\n");
}
int radeon_gem_get_tiling_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp)
{
struct drm_radeon_gem_get_tiling *args = data;
struct drm_gem_object *gobj;
struct radeon_bo *rbo;
int r = 0;
DRM_DEBUG("\n");
gobj = drm_gem_object_lookup(dev, filp, args->handle);
if (gobj == NULL)
return -ENOENT;
rbo = gem_to_radeon_bo(gobj);
r = radeon_bo_reserve(rbo, false);
if (unlikely(r != 0))
goto out;
radeon_bo_get_tiling_flags(rbo, &args->tiling_flags, &args->pitch);
radeon_bo_unreserve(rbo);
out:
drm_gem_object_unreference_unlocked(gobj);
return r;
}
/**
* radeon_ring_fini - tear down the driver ring struct.
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
*
* Tear down the driver information for the selected ring (all asics).
*/
void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
{
int r;
struct radeon_bo *ring_obj;
mutex_lock(&rdev->ring_lock);
ring_obj = ring->ring_obj;
ring->ready = false;
ring->ring = NULL;
ring->ring_obj = NULL;
mutex_unlock(&rdev->ring_lock);
if (ring_obj) {
r = radeon_bo_reserve(ring_obj, false);
if (likely(r == 0)) {
radeon_bo_kunmap(ring_obj);
radeon_bo_unpin(ring_obj);
radeon_bo_unreserve(ring_obj);
}
radeon_bo_unref(&ring_obj);
}
}
/**
* radeon_driver_postclose_kms - drm callback for post close
*
* @dev: drm dev pointer
* @file_priv: drm file
*
* On device close, tear down hyperz and cmask filps on r1xx-r5xx
* (all asics). And tear down vm on cayman+ (all asics).
*/
void radeon_driver_postclose_kms(struct drm_device *dev,
struct drm_file *file_priv)
{
struct radeon_device *rdev = dev->dev_private;
pm_runtime_get_sync(dev->dev);
mutex_lock(&rdev->gem.mutex);
if (rdev->hyperz_filp == file_priv)
rdev->hyperz_filp = NULL;
if (rdev->cmask_filp == file_priv)
rdev->cmask_filp = NULL;
mutex_unlock(&rdev->gem.mutex);
radeon_uvd_free_handles(rdev, file_priv);
radeon_vce_free_handles(rdev, file_priv);
/* new gpu have virtual address space support */
if (rdev->family >= CHIP_CAYMAN && file_priv->driver_priv) {
struct radeon_fpriv *fpriv = file_priv->driver_priv;
struct radeon_vm *vm = &fpriv->vm;
int r;
if (rdev->accel_working) {
r = radeon_bo_reserve(rdev->ring_tmp_bo.bo, false);
if (!r) {
if (vm->ib_bo_va)
radeon_vm_bo_rmv(rdev, vm->ib_bo_va);
radeon_bo_unreserve(rdev->ring_tmp_bo.bo);
}
radeon_vm_fini(rdev, vm);
}
kfree(fpriv);
file_priv->driver_priv = NULL;
}
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
}
int radeon_uvd_resume(struct radeon_device *rdev)
{
int r;
if (rdev->uvd.vcpu_bo == NULL)
return -EINVAL;
r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
if (r) {
radeon_bo_unref(&rdev->uvd.vcpu_bo);
dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r);
return r;
}
/* Have been pin in cpu unmap unpin */
radeon_bo_kunmap(rdev->uvd.vcpu_bo);
radeon_bo_unpin(rdev->uvd.vcpu_bo);
r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
&rdev->uvd.gpu_addr);
if (r) {
radeon_bo_unreserve(rdev->uvd.vcpu_bo);
radeon_bo_unref(&rdev->uvd.vcpu_bo);
dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r);
return r;
}
r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr);
if (r) {
dev_err(rdev->dev, "(%d) UVD map failed\n", r);
return r;
}
radeon_bo_unreserve(rdev->uvd.vcpu_bo);
return 0;
}
/* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */
void radeon_test_moves(struct radeon_device *rdev)
{
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;
int r;
size = 1024 * 1024;
/* Number of tests =
* (Total GTT - IB pool - writeback page - ring buffers) / test size
*/
n = rdev->mc.gtt_size - RADEON_IB_POOL_SIZE*64*1024;
for (i = 0; i < RADEON_NUM_RINGS; ++i)
n -= rdev->ring[i].ring_size;
if (rdev->wb.wb_obj)
n -= RADEON_GPU_PAGE_SIZE;
if (rdev->ih.ring_obj)
n -= rdev->ih.ring_size;
n /= size;
gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
if (!gtt_obj) {
DRM_ERROR("Failed to allocate %d pointers\n", n);
r = 1;
goto out_cleanup;
}
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
&vram_obj);
if (r) {
DRM_ERROR("Failed to create VRAM object\n");
goto out_cleanup;
}
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_bo_create(rdev, size, PAGE_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_bo_reserve(gtt_obj[i], false);
if (unlikely(r != 0))
goto out_cleanup;
r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr);
if (r) {
DRM_ERROR("Failed to pin GTT object %d\n", i);
goto out_cleanup;
}
r = radeon_bo_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size;
gtt_start < gtt_end;
gtt_start++)
*gtt_start = gtt_start;
radeon_bo_kunmap(gtt_obj[i]);
r = radeon_fence_create(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
if (r) {
DRM_ERROR("Failed to create GTT->VRAM fence %d\n", i);
goto out_cleanup;
}
r = radeon_copy(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
goto out_cleanup;
}
radeon_fence_unref(&fence);
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;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
vram_start < vram_end;
gtt_start++, vram_start++) {
if (*vram_start != gtt_start) {
DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, "
"expected 0x%p (GTT/VRAM offset "
"0x%16llx/0x%16llx)\n",
i, *vram_start, gtt_start,
(unsigned long long)
(gtt_addr - rdev->mc.gtt_start +
(void*)gtt_start - gtt_map),
(unsigned long long)
(vram_addr - rdev->mc.vram_start +
(void*)gtt_start - gtt_map));
radeon_bo_kunmap(vram_obj);
goto out_cleanup;
}
*vram_start = vram_start;
}
radeon_bo_kunmap(vram_obj);
r = radeon_fence_create(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
if (r) {
DRM_ERROR("Failed to create VRAM->GTT fence %d\n", i);
goto out_cleanup;
}
r = radeon_copy(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
goto out_cleanup;
}
radeon_fence_unref(&fence);
r = radeon_bo_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object after copy %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
gtt_start < gtt_end;
gtt_start++, vram_start++) {
if (*gtt_start != vram_start) {
DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, "
"expected 0x%p (VRAM/GTT offset "
"0x%16llx/0x%16llx)\n",
i, *gtt_start, vram_start,
(unsigned long long)
(vram_addr - rdev->mc.vram_start +
(void*)vram_start - vram_map),
(unsigned long long)
(gtt_addr - rdev->mc.gtt_start +
(void*)vram_start - vram_map));
radeon_bo_kunmap(gtt_obj[i]);
goto out_cleanup;
}
}
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_start);
}
out_cleanup:
if (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]) {
if (radeon_bo_is_reserved(gtt_obj[i])) {
radeon_bo_unpin(gtt_obj[i]);
radeon_bo_unreserve(gtt_obj[i]);
}
radeon_bo_unref(>t_obj[i]);
}
}
kfree(gtt_obj);
}
if (fence) {
radeon_fence_unref(&fence);
}
if (r) {
printk(KERN_WARNING "Error while testing BO move.\n");
}
}
/**
* radeon_vce_init - allocate memory, load vce firmware
*
* @rdev: radeon_device pointer
*
* First step to get VCE online, allocate memory and load the firmware
*/
int radeon_vce_init(struct radeon_device *rdev)
{
static const char *fw_version = "[ATI LIB=VCEFW,";
static const char *fb_version = "[ATI LIB=VCEFWSTATS,";
unsigned long size;
const char *fw_name, *c;
uint8_t start, mid, end;
int i, r;
INIT_DELAYED_WORK(&rdev->vce.idle_work, radeon_vce_idle_work_handler);
switch (rdev->family) {
case CHIP_BONAIRE:
case CHIP_KAVERI:
case CHIP_KABINI:
fw_name = FIRMWARE_BONAIRE;
break;
default:
return -EINVAL;
}
r = request_firmware(&rdev->vce_fw, fw_name, rdev->dev);
if (r) {
dev_err(rdev->dev, "radeon_vce: Can't load firmware \"%s\"\n",
fw_name);
return r;
}
/* search for firmware version */
size = rdev->vce_fw->size - strlen(fw_version) - 9;
c = rdev->vce_fw->data;
for (;size > 0; --size, ++c)
if (strncmp(c, fw_version, strlen(fw_version)) == 0)
break;
if (size == 0)
return -EINVAL;
c += strlen(fw_version);
if (sscanf(c, "%2hhd.%2hhd.%2hhd]", &start, &mid, &end) != 3)
return -EINVAL;
/* search for feedback version */
size = rdev->vce_fw->size - strlen(fb_version) - 3;
c = rdev->vce_fw->data;
for (;size > 0; --size, ++c)
if (strncmp(c, fb_version, strlen(fb_version)) == 0)
break;
if (size == 0)
return -EINVAL;
c += strlen(fb_version);
if (sscanf(c, "%2u]", &rdev->vce.fb_version) != 1)
return -EINVAL;
DRM_INFO("Found VCE firmware/feedback version %hhd.%hhd.%hhd / %d!\n",
start, mid, end, rdev->vce.fb_version);
rdev->vce.fw_version = (start << 24) | (mid << 16) | (end << 8);
/* we can only work with this fw version for now */
if (rdev->vce.fw_version != ((40 << 24) | (2 << 16) | (2 << 8)))
return -EINVAL;
/* allocate firmware, stack and heap BO */
size = RADEON_GPU_PAGE_ALIGN(rdev->vce_fw->size) +
RADEON_VCE_STACK_SIZE + RADEON_VCE_HEAP_SIZE;
r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_VRAM, NULL, &rdev->vce.vcpu_bo);
if (r) {
dev_err(rdev->dev, "(%d) failed to allocate VCE bo\n", r);
return r;
}
r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
if (r) {
radeon_bo_unref(&rdev->vce.vcpu_bo);
dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
return r;
}
r = radeon_bo_pin(rdev->vce.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
&rdev->vce.gpu_addr);
radeon_bo_unreserve(rdev->vce.vcpu_bo);
if (r) {
radeon_bo_unref(&rdev->vce.vcpu_bo);
dev_err(rdev->dev, "(%d) VCE bo pin failed\n", r);
return r;
}
for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
atomic_set(&rdev->vce.handles[i], 0);
rdev->vce.filp[i] = NULL;
}
return 0;
}
/* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */
static void radeon_do_test_moves(struct radeon_device *rdev, int flag)
{
struct radeon_bo *vram_obj = NULL;
struct radeon_bo **gtt_obj = NULL;
uint64_t gtt_addr, vram_addr;
unsigned n, size;
int i, r, ring;
switch (flag) {
case RADEON_TEST_COPY_DMA:
ring = radeon_copy_dma_ring_index(rdev);
break;
case RADEON_TEST_COPY_BLIT:
ring = radeon_copy_blit_ring_index(rdev);
break;
default:
DRM_ERROR("Unknown copy method\n");
return;
}
size = 1024 * 1024;
/* Number of tests =
* (Total GTT - IB pool - writeback page - ring buffers) / test size
*/
n = rdev->mc.gtt_size - RADEON_IB_POOL_SIZE*64*1024;
for (i = 0; i < RADEON_NUM_RINGS; ++i)
n -= rdev->ring[i].ring_size;
if (rdev->wb.wb_obj)
n -= RADEON_GPU_PAGE_SIZE;
if (rdev->ih.ring_obj)
n -= rdev->ih.ring_size;
n /= size;
gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
if (!gtt_obj) {
DRM_ERROR("Failed to allocate %d pointers\n", n);
r = 1;
goto out_cleanup;
}
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
NULL, &vram_obj);
if (r) {
DRM_ERROR("Failed to create VRAM object\n");
goto out_cleanup;
}
r = radeon_bo_reserve(vram_obj, false);
if (unlikely(r != 0))
goto out_unref;
r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
if (r) {
DRM_ERROR("Failed to pin VRAM object\n");
goto out_unres;
}
for (i = 0; i < n; i++) {
void *gtt_map, *vram_map;
void **gtt_start, **gtt_end;
void **vram_start, **vram_end;
struct radeon_fence *fence = NULL;
r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_GTT, NULL, gtt_obj + i);
if (r) {
DRM_ERROR("Failed to create GTT object %d\n", i);
goto out_lclean;
}
r = radeon_bo_reserve(gtt_obj[i], false);
if (unlikely(r != 0))
goto out_lclean_unref;
r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr);
if (r) {
DRM_ERROR("Failed to pin GTT object %d\n", i);
goto out_lclean_unres;
}
r = radeon_bo_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object %d\n", i);
goto out_lclean_unpin;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size;
gtt_start < gtt_end;
gtt_start++)
*gtt_start = gtt_start;
radeon_bo_kunmap(gtt_obj[i]);
if (ring == R600_RING_TYPE_DMA_INDEX)
r = radeon_copy_dma(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
else
r = radeon_copy_blit(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
if (r) {
DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
goto out_lclean_unpin;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
goto out_lclean_unpin;
}
radeon_fence_unref(&fence);
r = radeon_bo_kmap(vram_obj, &vram_map);
if (r) {
DRM_ERROR("Failed to map VRAM object after copy %d\n", i);
goto out_lclean_unpin;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
vram_start < vram_end;
gtt_start++, vram_start++) {
if (*vram_start != gtt_start) {
DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, "
"expected 0x%p (GTT/VRAM offset "
"0x%16llx/0x%16llx)\n",
i, *vram_start, gtt_start,
(unsigned long long)
(gtt_addr - rdev->mc.gtt_start +
(void*)gtt_start - gtt_map),
(unsigned long long)
(vram_addr - rdev->mc.vram_start +
(void*)gtt_start - gtt_map));
radeon_bo_kunmap(vram_obj);
goto out_lclean_unpin;
}
*vram_start = vram_start;
}
radeon_bo_kunmap(vram_obj);
if (ring == R600_RING_TYPE_DMA_INDEX)
r = radeon_copy_dma(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
else
r = radeon_copy_blit(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
if (r) {
DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
goto out_lclean_unpin;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
goto out_lclean_unpin;
}
radeon_fence_unref(&fence);
r = radeon_bo_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object after copy %d\n", i);
goto out_lclean_unpin;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
gtt_start < gtt_end;
gtt_start++, vram_start++) {
if (*gtt_start != vram_start) {
DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, "
"expected 0x%p (VRAM/GTT offset "
"0x%16llx/0x%16llx)\n",
i, *gtt_start, vram_start,
(unsigned long long)
(vram_addr - rdev->mc.vram_start +
(void*)vram_start - vram_map),
(unsigned long long)
(gtt_addr - rdev->mc.gtt_start +
(void*)vram_start - vram_map));
radeon_bo_kunmap(gtt_obj[i]);
goto out_lclean_unpin;
}
}
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_start);
continue;
out_lclean_unpin:
radeon_bo_unpin(gtt_obj[i]);
out_lclean_unres:
radeon_bo_unreserve(gtt_obj[i]);
out_lclean_unref:
radeon_bo_unref(>t_obj[i]);
out_lclean:
for (--i; i >= 0; --i) {
radeon_bo_unpin(gtt_obj[i]);
radeon_bo_unreserve(gtt_obj[i]);
radeon_bo_unref(>t_obj[i]);
}
if (fence)
radeon_fence_unref(&fence);
break;
}
radeon_bo_unpin(vram_obj);
out_unres:
radeon_bo_unreserve(vram_obj);
out_unref:
radeon_bo_unref(&vram_obj);
out_cleanup:
kfree(gtt_obj);
if (r) {
printk(KERN_WARNING "Error while testing BO move.\n");
}
}
int radeon_uvd_init(struct radeon_device *rdev)
{
unsigned long bo_size;
const char *fw_name;
int i, r;
INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler);
switch (rdev->family) {
case CHIP_RV710:
case CHIP_RV730:
case CHIP_RV740:
fw_name = FIRMWARE_RV710;
break;
case CHIP_CYPRESS:
case CHIP_HEMLOCK:
case CHIP_JUNIPER:
case CHIP_REDWOOD:
case CHIP_CEDAR:
fw_name = FIRMWARE_CYPRESS;
break;
case CHIP_SUMO:
case CHIP_SUMO2:
case CHIP_PALM:
case CHIP_CAYMAN:
case CHIP_BARTS:
case CHIP_TURKS:
case CHIP_CAICOS:
fw_name = FIRMWARE_SUMO;
break;
case CHIP_TAHITI:
case CHIP_VERDE:
case CHIP_PITCAIRN:
case CHIP_ARUBA:
fw_name = FIRMWARE_TAHITI;
break;
case CHIP_BONAIRE:
case CHIP_KABINI:
case CHIP_KAVERI:
fw_name = FIRMWARE_BONAIRE;
break;
default:
return -EINVAL;
}
r = request_firmware(&rdev->uvd_fw, fw_name, rdev->dev);
if (r) {
dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n",
fw_name);
return r;
}
bo_size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 8) +
RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE;
r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_VRAM, NULL, &rdev->uvd.vcpu_bo);
if (r) {
dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r);
return r;
}
r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
if (r) {
radeon_bo_unref(&rdev->uvd.vcpu_bo);
dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r);
return r;
}
r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
&rdev->uvd.gpu_addr);
if (r) {
radeon_bo_unreserve(rdev->uvd.vcpu_bo);
radeon_bo_unref(&rdev->uvd.vcpu_bo);
dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r);
return r;
}
r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr);
if (r) {
dev_err(rdev->dev, "(%d) UVD map failed\n", r);
return r;
}
radeon_bo_unreserve(rdev->uvd.vcpu_bo);
for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
atomic_set(&rdev->uvd.handles[i], 0);
rdev->uvd.filp[i] = NULL;
}
return 0;
}
static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size,
unsigned sdomain, unsigned ddomain)
{
struct radeon_bo *dobj = NULL;
struct radeon_bo *sobj = NULL;
uint64_t saddr, daddr;
int r, n;
int time;
n = RADEON_BENCHMARK_ITERATIONS;
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, NULL, &sobj);
if (r) {
goto out_cleanup;
}
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_bo_create(rdev, size, PAGE_SIZE, true, ddomain, NULL, &dobj);
if (r) {
goto out_cleanup;
}
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;
}
/* r100 doesn't have dma engine so skip the test */
/* also, VRAM-to-VRAM test doesn't make much sense for DMA */
/* skip it as well if domains are the same */
if ((rdev->asic->copy.dma) && (sdomain != ddomain)) {
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_DMA, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "dma");
}
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_BLIT, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "blit");
out_cleanup:
if (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) {
r = radeon_bo_reserve(dobj, false);
if (likely(r == 0)) {
radeon_bo_unpin(dobj);
radeon_bo_unreserve(dobj);
}
radeon_bo_unref(&dobj);
}
if (r) {
DRM_ERROR("Error while benchmarking BO move.\n");
}
}
int radeon_ttm_init(struct radeon_device *rdev)
{
int r;
r = radeon_ttm_global_init(rdev);
if (r) {
return r;
}
/* No others user of address space so set it to 0 */
r = ttm_bo_device_init(&rdev->mman.bdev,
rdev->mman.bo_global_ref.ref.object,
&radeon_bo_driver, DRM_FILE_PAGE_OFFSET,
rdev->need_dma32);
if (r) {
DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
return r;
}
rdev->mman.bdev.iot = rdev->iot;
rdev->mman.bdev.memt = rdev->memt;
rdev->mman.bdev.dmat = rdev->dmat;
rdev->mman.initialized = true;
r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM,
rdev->mc.real_vram_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing VRAM heap.\n");
return r;
}
/* Change the size here instead of the init above so only lpfn is affected */
radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
#ifdef __sparc64__
r = radeon_bo_create(rdev, rdev->fb_offset, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_VRAM,
NULL, &rdev->stollen_vga_memory);
#else
r = radeon_bo_create(rdev, 256 * 1024, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_VRAM,
NULL, &rdev->stollen_vga_memory);
#endif
if (r) {
return r;
}
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_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,
rdev->mc.gtt_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing GTT heap.\n");
return r;
}
DRM_INFO("radeon: %uM of GTT memory ready.\n",
(unsigned)(rdev->mc.gtt_size / (1024 * 1024)));
#ifdef notyet
rdev->mman.bdev.dev_mapping = rdev->ddev->dev_mapping;
#endif
r = radeon_ttm_debugfs_init(rdev);
if (r) {
DRM_ERROR("Failed to init debugfs\n");
return r;
}
return 0;
}
static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size,
unsigned sdomain, unsigned ddomain)
{
struct radeon_bo *dobj = NULL;
struct radeon_bo *sobj = NULL;
uint64_t saddr, daddr;
int r, n;
int time;
n = RADEON_BENCHMARK_ITERATIONS;
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, 0, NULL, &sobj);
if (r) {
goto out_cleanup;
}
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_bo_create(rdev, size, PAGE_SIZE, true, ddomain, 0, NULL, &dobj);
if (r) {
goto out_cleanup;
}
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;
}
if (rdev->asic->copy.dma) {
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_DMA, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "dma");
}
if (rdev->asic->copy.blit) {
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_BLIT, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "blit");
}
out_cleanup:
if (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) {
r = radeon_bo_reserve(dobj, false);
if (likely(r == 0)) {
radeon_bo_unpin(dobj);
radeon_bo_unreserve(dobj);
}
radeon_bo_unref(&dobj);
}
if (r) {
DRM_ERROR("Error while benchmarking BO move.\n");
}
}
static int radeonfb_create_pinned_object(struct radeon_fbdev *rfbdev,
struct drm_mode_fb_cmd2 *mode_cmd,
struct drm_gem_object **gobj_p)
{
struct radeon_device *rdev = rfbdev->rdev;
struct drm_gem_object *gobj = NULL;
struct radeon_bo *rbo = NULL;
bool fb_tiled = false; /* useful for testing */
u32 tiling_flags = 0;
int ret;
int aligned_size, size;
int height = mode_cmd->height;
u32 bpp, depth;
drm_fb_get_bpp_depth(mode_cmd->pixel_format, &depth, &bpp);
/* need to align pitch with crtc limits */
mode_cmd->pitches[0] = radeon_align_pitch(rdev, mode_cmd->width, bpp,
fb_tiled) * ((bpp + 1) / 8);
if (rdev->family >= CHIP_R600)
height = ALIGN(mode_cmd->height, 8);
size = mode_cmd->pitches[0] * height;
aligned_size = ALIGN(size, PAGE_SIZE);
ret = radeon_gem_object_create(rdev, aligned_size, 0,
RADEON_GEM_DOMAIN_VRAM,
0, true, &gobj);
if (ret) {
printk(KERN_ERR "failed to allocate framebuffer (%d)\n",
aligned_size);
return -ENOMEM;
}
rbo = gem_to_radeon_bo(gobj);
if (fb_tiled)
tiling_flags = RADEON_TILING_MACRO;
#ifdef __BIG_ENDIAN
switch (bpp) {
case 32:
tiling_flags |= RADEON_TILING_SWAP_32BIT;
break;
case 16:
tiling_flags |= RADEON_TILING_SWAP_16BIT;
default:
break;
}
#endif
if (tiling_flags) {
ret = radeon_bo_set_tiling_flags(rbo,
tiling_flags | RADEON_TILING_SURFACE,
mode_cmd->pitches[0]);
if (ret)
dev_err(rdev->dev, "FB failed to set tiling flags\n");
}
ret = radeon_bo_reserve(rbo, false);
if (unlikely(ret != 0))
goto out_unref;
/* Only 27 bit offset for legacy CRTC */
ret = radeon_bo_pin_restricted(rbo, RADEON_GEM_DOMAIN_VRAM,
ASIC_IS_AVIVO(rdev) ? 0 : 1 << 27,
NULL);
if (ret) {
radeon_bo_unreserve(rbo);
goto out_unref;
}
if (fb_tiled)
radeon_bo_check_tiling(rbo, 0, 0);
ret = radeon_bo_kmap(rbo, NULL);
radeon_bo_unreserve(rbo);
if (ret) {
goto out_unref;
}
*gobj_p = gobj;
return 0;
out_unref:
radeonfb_destroy_pinned_object(gobj);
*gobj_p = NULL;
return ret;
}
void radeon_benchmark_move(struct radeon_device *rdev, unsigned bsize,
unsigned sdomain, unsigned ddomain)
{
struct radeon_bo *dobj = NULL;
struct radeon_bo *sobj = NULL;
struct radeon_fence *fence = NULL;
uint64_t saddr, daddr;
unsigned long start_jiffies;
unsigned long end_jiffies;
unsigned long time;
unsigned i, n, size;
int r;
size = bsize;
n = 1024;
r = radeon_bo_create(rdev, NULL, size, PAGE_SIZE, true, sdomain, &sobj);
if (r) {
goto out_cleanup;
}
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_bo_create(rdev, NULL, size, PAGE_SIZE, true, ddomain, &dobj);
if (r) {
goto out_cleanup;
}
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;
}
/* r100 doesn't have dma engine so skip the test */
if (rdev->asic->copy_dma) {
start_jiffies = jiffies;
for (i = 0; i < n; i++) {
r = radeon_fence_create(rdev, &fence);
if (r) {
goto out_cleanup;
}
r = radeon_copy_dma(rdev, saddr, daddr,
size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
goto out_cleanup;
}
radeon_fence_unref(&fence);
}
end_jiffies = jiffies;
time = end_jiffies - start_jiffies;
time = jiffies_to_msecs(time);
if (time > 0) {
i = ((n * size) >> 10) / time;
printk(KERN_INFO "radeon: dma %u bo moves of %ukb from"
" %d to %d in %lums (%ukb/ms %ukb/s %uM/s)\n",
n, size >> 10,
sdomain, ddomain, time,
i, i * 1000, (i * 1000) / 1024);
}
int radeon_gem_va_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp)
{
struct drm_radeon_gem_va *args = data;
struct drm_gem_object *gobj;
struct radeon_device *rdev = dev->dev_private;
struct radeon_fpriv *fpriv = filp->driver_priv;
struct radeon_bo *rbo;
struct radeon_bo_va *bo_va;
u32 invalid_flags;
int r = 0;
if (!rdev->vm_manager.enabled) {
args->operation = RADEON_VA_RESULT_ERROR;
return -ENOTTY;
}
/* !! DONT REMOVE !!
* We don't support vm_id yet, to be sure we don't have have broken
* userspace, reject anyone trying to use non 0 value thus moving
* forward we can use those fields without breaking existant userspace
*/
if (args->vm_id) {
args->operation = RADEON_VA_RESULT_ERROR;
return -EINVAL;
}
if (args->offset < RADEON_VA_RESERVED_SIZE) {
dev_err(dev->device,
"offset 0x%lX is in reserved area 0x%X\n",
(unsigned long)args->offset,
RADEON_VA_RESERVED_SIZE);
args->operation = RADEON_VA_RESULT_ERROR;
return -EINVAL;
}
/* don't remove, we need to enforce userspace to set the snooped flag
* otherwise we will endup with broken userspace and we won't be able
* to enable this feature without adding new interface
*/
invalid_flags = RADEON_VM_PAGE_VALID | RADEON_VM_PAGE_SYSTEM;
if ((args->flags & invalid_flags)) {
dev_err(dev->device, "invalid flags 0x%08X vs 0x%08X\n",
args->flags, invalid_flags);
args->operation = RADEON_VA_RESULT_ERROR;
return -EINVAL;
}
if (!(args->flags & RADEON_VM_PAGE_SNOOPED)) {
dev_err(dev->device, "only supported snooped mapping for now\n");
args->operation = RADEON_VA_RESULT_ERROR;
return -EINVAL;
}
switch (args->operation) {
case RADEON_VA_MAP:
case RADEON_VA_UNMAP:
break;
default:
dev_err(dev->device, "unsupported operation %d\n",
args->operation);
args->operation = RADEON_VA_RESULT_ERROR;
return -EINVAL;
}
gobj = drm_gem_object_lookup(dev, filp, args->handle);
if (gobj == NULL) {
args->operation = RADEON_VA_RESULT_ERROR;
return -ENOENT;
}
rbo = gem_to_radeon_bo(gobj);
r = radeon_bo_reserve(rbo, false);
if (r) {
args->operation = RADEON_VA_RESULT_ERROR;
drm_gem_object_unreference_unlocked(gobj);
return r;
}
bo_va = radeon_vm_bo_find(&fpriv->vm, rbo);
if (!bo_va) {
args->operation = RADEON_VA_RESULT_ERROR;
drm_gem_object_unreference_unlocked(gobj);
return -ENOENT;
}
switch (args->operation) {
case RADEON_VA_MAP:
if (bo_va->soffset) {
args->operation = RADEON_VA_RESULT_VA_EXIST;
args->offset = bo_va->soffset;
goto out;
}
r = radeon_vm_bo_set_addr(rdev, bo_va, args->offset, args->flags);
break;
case RADEON_VA_UNMAP:
r = radeon_vm_bo_set_addr(rdev, bo_va, 0, 0);
break;
default:
break;
}
args->operation = RADEON_VA_RESULT_OK;
if (r) {
args->operation = RADEON_VA_RESULT_ERROR;
}
out:
radeon_bo_unreserve(rbo);
drm_gem_object_unreference_unlocked(gobj);
return r;
}
/**
* radeon_driver_open_kms - drm callback for open
*
* @dev: drm dev pointer
* @file_priv: drm file
*
* On device open, init vm on cayman+ (all asics).
* Returns 0 on success, error on failure.
*/
int radeon_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
{
struct radeon_device *rdev = dev->dev_private;
file_priv->driver_priv = NULL;
#ifdef PM_TODO
r = pm_runtime_get_sync(dev->dev);
if (r < 0)
return r;
#endif
/* new gpu have virtual address space support */
if (rdev->family >= CHIP_CAYMAN) {
struct radeon_fpriv *fpriv;
struct radeon_vm *vm;
int r;
fpriv = kzalloc(sizeof(*fpriv), GFP_KERNEL);
if (unlikely(!fpriv)) {
return -ENOMEM;
}
vm = &fpriv->vm;
r = radeon_vm_init(rdev, vm);
if (r) {
kfree(fpriv);
return r;
}
if (rdev->accel_working) {
r = radeon_bo_reserve(rdev->ring_tmp_bo.bo, false);
if (r) {
radeon_vm_fini(rdev, vm);
kfree(fpriv);
return r;
}
/* map the ib pool buffer read only into
* virtual address space */
vm->ib_bo_va = radeon_vm_bo_add(rdev, vm,
rdev->ring_tmp_bo.bo);
r = radeon_vm_bo_set_addr(rdev, vm->ib_bo_va,
RADEON_VA_IB_OFFSET,
RADEON_VM_PAGE_READABLE |
RADEON_VM_PAGE_SNOOPED);
radeon_bo_unreserve(rdev->ring_tmp_bo.bo);
if (r) {
radeon_vm_fini(rdev, vm);
kfree(fpriv);
return r;
}
}
file_priv->driver_priv = fpriv;
}
#ifdef PM_TODO
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
#endif
return 0;
}
int radeon_ttm_init(struct radeon_device *rdev)
{
int r, r2;
r = radeon_ttm_global_init(rdev);
if (r) {
return r;
}
/* No others user of address space so set it to 0 */
r = ttm_bo_device_init(&rdev->mman.bdev,
rdev->mman.bo_global_ref.ref.object,
&radeon_bo_driver, DRM_FILE_PAGE_OFFSET,
rdev->need_dma32);
if (r) {
DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
return r;
}
rdev->mman.initialized = true;
rdev->ddev->drm_ttm_bdev = &rdev->mman.bdev;
r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM,
rdev->mc.real_vram_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing VRAM heap.\n");
return r;
}
r = radeon_bo_create(rdev, 256 * 1024, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_VRAM,
NULL, &rdev->stollen_vga_memory);
if (r) {
return r;
}
r = radeon_bo_reserve(rdev->stollen_vga_memory, false);
if (r) {
radeon_bo_unref(&rdev->stollen_vga_memory);
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_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,
rdev->mc.gtt_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing GTT heap.\n");
r2 = radeon_bo_reserve(rdev->stollen_vga_memory, false);
if (likely(r2 == 0)) {
radeon_bo_unpin(rdev->stollen_vga_memory);
radeon_bo_unreserve(rdev->stollen_vga_memory);
}
radeon_bo_unref(&rdev->stollen_vga_memory);
return r;
}
DRM_INFO("radeon: %uM of GTT memory ready.\n",
(unsigned)(rdev->mc.gtt_size / (1024 * 1024)));
r = radeon_ttm_debugfs_init(rdev);
if (r) {
DRM_ERROR("Failed to init debugfs\n");
r2 = radeon_bo_reserve(rdev->stollen_vga_memory, false);
if (likely(r2 == 0)) {
radeon_bo_unpin(rdev->stollen_vga_memory);
radeon_bo_unreserve(rdev->stollen_vga_memory);
}
radeon_bo_unref(&rdev->stollen_vga_memory);
return r;
}
return 0;
}
