/**
* cik_copy_dma - copy pages using the DMA engine
*
* @rdev: radeon_device pointer
* @src_offset: src GPU address
* @dst_offset: dst GPU address
* @num_gpu_pages: number of GPU pages to xfer
* @resv: reservation object to sync to
*
* Copy GPU paging using the DMA engine (CIK).
* Used by the radeon ttm implementation to move pages if
* registered as the asic copy callback.
*/
struct radeon_fence *cik_copy_dma(struct radeon_device *rdev,
uint64_t src_offset, uint64_t dst_offset,
unsigned num_gpu_pages,
struct reservation_object *resv)
{
struct radeon_semaphore *sem = NULL;
struct radeon_fence *fence;
int ring_index = rdev->asic->copy.dma_ring_index;
struct radeon_ring *ring = &rdev->ring[ring_index];
u32 size_in_bytes, cur_size_in_bytes;
int i, num_loops;
int r = 0;
r = radeon_semaphore_create(rdev, &sem);
if (r) {
DRM_ERROR("radeon: moving bo (%d).\n", r);
return ERR_PTR(r);
}
size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
num_loops = DIV_ROUND_UP(size_in_bytes, 0x1fffff);
r = radeon_ring_lock(rdev, ring, num_loops * 7 + 14);
if (r) {
DRM_ERROR("radeon: moving bo (%d).\n", r);
radeon_semaphore_free(rdev, &sem, NULL);
return ERR_PTR(r);
}
radeon_semaphore_sync_resv(rdev, sem, resv, false);
radeon_semaphore_sync_rings(rdev, sem, ring->idx);
for (i = 0; i < num_loops; i++) {
cur_size_in_bytes = size_in_bytes;
if (cur_size_in_bytes > 0x1fffff)
cur_size_in_bytes = 0x1fffff;
size_in_bytes -= cur_size_in_bytes;
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0));
radeon_ring_write(ring, cur_size_in_bytes);
radeon_ring_write(ring, 0); /* src/dst endian swap */
radeon_ring_write(ring, lower_32_bits(src_offset));
radeon_ring_write(ring, upper_32_bits(src_offset));
radeon_ring_write(ring, lower_32_bits(dst_offset));
radeon_ring_write(ring, upper_32_bits(dst_offset));
src_offset += cur_size_in_bytes;
dst_offset += cur_size_in_bytes;
}
r = radeon_fence_emit(rdev, &fence, ring->idx);
if (r) {
radeon_ring_unlock_undo(rdev, ring);
radeon_semaphore_free(rdev, &sem, NULL);
return ERR_PTR(r);
}
radeon_ring_unlock_commit(rdev, ring, false);
radeon_semaphore_free(rdev, &sem, fence);
return fence;
}
/**
* radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring
*
* @rdev: radeon_device pointer
* @ib: IB object to schedule
* @const_ib: Const IB to schedule (SI only)
*
* Schedule an IB on the associated ring (all asics).
* Returns 0 on success, error on failure.
*
* On SI, there are two parallel engines fed from the primary ring,
* the CE (Constant Engine) and the DE (Drawing Engine). Since
* resource descriptors have moved to memory, the CE allows you to
* prime the caches while the DE is updating register state so that
* the resource descriptors will be already in cache when the draw is
* processed. To accomplish this, the userspace driver submits two
* IBs, one for the CE and one for the DE. If there is a CE IB (called
* a CONST_IB), it will be put on the ring prior to the DE IB. Prior
* to SI there was just a DE IB.
*/
int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
struct radeon_ib *const_ib)
{
struct radeon_ring *ring = &rdev->ring[ib->ring];
int r = 0;
if (!ib->length_dw || !ring->ready) {
/* TODO: Nothings in the ib we should report. */
dev_err(rdev->dev, "couldn't schedule ib\n");
return -EINVAL;
}
/* 64 dwords should be enough for fence too */
r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_SYNCS * 8);
if (r) {
dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
return r;
}
/* grab a vm id if necessary */
if (ib->vm) {
struct radeon_fence *vm_id_fence;
vm_id_fence = radeon_vm_grab_id(rdev, ib->vm, ib->ring);
radeon_semaphore_sync_to(ib->semaphore, vm_id_fence);
}
/* sync with other rings */
r = radeon_semaphore_sync_rings(rdev, ib->semaphore, ib->ring);
if (r) {
dev_err(rdev->dev, "failed to sync rings (%d)\n", r);
radeon_ring_unlock_undo(rdev, ring);
return r;
}
if (ib->vm)
radeon_vm_flush(rdev, ib->vm, ib->ring);
if (const_ib) {
radeon_ring_ib_execute(rdev, const_ib->ring, const_ib);
radeon_semaphore_free(rdev, &const_ib->semaphore, NULL);
}
radeon_ring_ib_execute(rdev, ib->ring, ib);
r = radeon_fence_emit(rdev, &ib->fence, ib->ring);
if (r) {
dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r);
radeon_ring_unlock_undo(rdev, ring);
return r;
}
if (const_ib) {
const_ib->fence = radeon_fence_ref(ib->fence);
}
if (ib->vm)
radeon_vm_fence(rdev, ib->vm, ib->fence);
radeon_ring_unlock_commit(rdev, ring);
return 0;
}
/**
* radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring
*
* @rdev: radeon_device pointer
* @ib: IB object to schedule
* @const_ib: Const IB to schedule (SI only)
*
* Schedule an IB on the associated ring (all asics).
* Returns 0 on success, error on failure.
*
* On SI, there are two parallel engines fed from the primary ring,
* the CE (Constant Engine) and the DE (Drawing Engine). Since
* resource descriptors have moved to memory, the CE allows you to
* prime the caches while the DE is updating register state so that
* the resource descriptors will be already in cache when the draw is
* processed. To accomplish this, the userspace driver submits two
* IBs, one for the CE and one for the DE. If there is a CE IB (called
* a CONST_IB), it will be put on the ring prior to the DE IB. Prior
* to SI there was just a DE IB.
*/
int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
struct radeon_ib *const_ib)
{
struct radeon_ring *ring = &rdev->ring[ib->ring];
bool need_sync = false;
int i, r = 0;
if (!ib->length_dw || !ring->ready) {
/* TODO: Nothings in the ib we should report. */
dev_err(rdev->dev, "couldn't schedule ib\n");
return -EINVAL;
}
/* 64 dwords should be enough for fence too */
r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_RINGS * 8);
if (r) {
dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
return r;
}
for (i = 0; i < RADEON_NUM_RINGS; ++i) {
struct radeon_fence *fence = ib->sync_to[i];
if (radeon_fence_need_sync(fence, ib->ring)) {
need_sync = true;
radeon_semaphore_sync_rings(rdev, ib->semaphore,
fence->ring, ib->ring);
radeon_fence_note_sync(fence, ib->ring);
}
}
/* immediately free semaphore when we don't need to sync */
if (!need_sync) {
radeon_semaphore_free(rdev, &ib->semaphore, NULL);
}
/* if we can't remember our last VM flush then flush now! */
/* XXX figure out why we have to flush for every IB */
if (ib->vm /*&& !ib->vm->last_flush*/) {
radeon_ring_vm_flush(rdev, ib->ring, ib->vm);
}
if (const_ib) {
radeon_ring_ib_execute(rdev, const_ib->ring, const_ib);
radeon_semaphore_free(rdev, &const_ib->semaphore, NULL);
}
radeon_ring_ib_execute(rdev, ib->ring, ib);
r = radeon_fence_emit(rdev, &ib->fence, ib->ring);
if (r) {
dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r);
radeon_ring_unlock_undo(rdev, ring);
return r;
}
if (const_ib) {
const_ib->fence = radeon_fence_ref(ib->fence);
}
/* we just flushed the VM, remember that */
if (ib->vm && !ib->vm->last_flush) {
ib->vm->last_flush = radeon_fence_ref(ib->fence);
}
radeon_ring_unlock_commit(rdev, ring);
return 0;
}
/**
* r600_copy_dma - copy pages using the DMA engine
*
* @rdev: radeon_device pointer
* @src_offset: src GPU address
* @dst_offset: dst GPU address
* @num_gpu_pages: number of GPU pages to xfer
* @resv: reservation object to sync to
*
* Copy GPU paging using the DMA engine (r6xx).
* Used by the radeon ttm implementation to move pages if
* registered as the asic copy callback.
*/
struct radeon_fence *r600_copy_dma(struct radeon_device *rdev,
uint64_t src_offset, uint64_t dst_offset,
unsigned num_gpu_pages,
struct reservation_object *resv)
{
struct radeon_semaphore *sem = NULL;
struct radeon_fence *fence;
int ring_index = rdev->asic->copy.dma_ring_index;
struct radeon_ring *ring = &rdev->ring[ring_index];
u32 size_in_dw, cur_size_in_dw;
int i, num_loops;
int r = 0;
r = radeon_semaphore_create(rdev, &sem);
if (r) {
DRM_ERROR("radeon: moving bo (%d).\n", r);
return ERR_PTR(r);
}
size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
num_loops = DIV_ROUND_UP(size_in_dw, 0xFFFE);
r = radeon_ring_lock(rdev, ring, num_loops * 4 + 8);
if (r) {
DRM_ERROR("radeon: moving bo (%d).\n", r);
radeon_semaphore_free(rdev, &sem, NULL);
return ERR_PTR(r);
}
radeon_semaphore_sync_resv(rdev, sem, resv, false);
radeon_semaphore_sync_rings(rdev, sem, ring->idx);
for (i = 0; i < num_loops; i++) {
cur_size_in_dw = size_in_dw;
if (cur_size_in_dw > 0xFFFE)
cur_size_in_dw = 0xFFFE;
size_in_dw -= cur_size_in_dw;
radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw));
radeon_ring_write(ring, dst_offset & 0xfffffffc);
radeon_ring_write(ring, src_offset & 0xfffffffc);
radeon_ring_write(ring, (((upper_32_bits(dst_offset) & 0xff) << 16) |
(upper_32_bits(src_offset) & 0xff)));
src_offset += cur_size_in_dw * 4;
dst_offset += cur_size_in_dw * 4;
}
r = radeon_fence_emit(rdev, &fence, ring->idx);
if (r) {
radeon_ring_unlock_undo(rdev, ring);
radeon_semaphore_free(rdev, &sem, NULL);
return ERR_PTR(r);
}
radeon_ring_unlock_commit(rdev, ring, false);
radeon_semaphore_free(rdev, &sem, fence);
return fence;
}
