/**
* uvd_v6_0_ring_test_ring - register write test
*
* @ring: amdgpu_ring pointer
*
* Test if we can successfully write to the context register
*/
static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
uint32_t tmp = 0;
unsigned i;
int r;
WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
r = amdgpu_ring_alloc(ring, 3);
if (r) {
DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
ring->idx, r);
return r;
}
amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
amdgpu_ring_write(ring, 0xDEADBEEF);
amdgpu_ring_commit(ring);
for (i = 0; i < adev->usec_timeout; i++) {
tmp = RREG32(mmUVD_CONTEXT_ID);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i < adev->usec_timeout) {
DRM_INFO("ring test on %d succeeded in %d usecs\n",
ring->idx, i);
} else {
DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
ring->idx, tmp);
r = -EINVAL;
}
return r;
}
/**
* cik_sdma_ring_emit_ib - Schedule an IB on the DMA engine
*
* @ring: amdgpu ring pointer
* @ib: IB object to schedule
*
* Schedule an IB in the DMA ring (CIK).
*/
static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_ib *ib)
{
u32 extra_bits = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf;
u32 next_rptr = ring->wptr + 5;
while ((next_rptr & 7) != 4)
next_rptr++;
next_rptr += 4;
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
amdgpu_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
amdgpu_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
amdgpu_ring_write(ring, 1); /* number of DWs to follow */
amdgpu_ring_write(ring, next_rptr);
/* IB packet must end on a 8 DW boundary */
while ((ring->wptr & 7) != 4)
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff);
amdgpu_ring_write(ring, ib->length_dw);
}
/**
* uvd_v6_0_ring_test_ring - register write test
*
* @ring: amdgpu_ring pointer
*
* Test if we can successfully write to the context register
*/
static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
uint32_t tmp = 0;
unsigned i;
int r;
WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
r = amdgpu_ring_alloc(ring, 3);
if (r)
return r;
amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
amdgpu_ring_write(ring, 0xDEADBEEF);
amdgpu_ring_commit(ring);
for (i = 0; i < adev->usec_timeout; i++) {
tmp = RREG32(mmUVD_CONTEXT_ID);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i >= adev->usec_timeout)
r = -ETIMEDOUT;
return r;
}
/**
* uvd_v5_0_hw_init - start and test UVD block
*
* @adev: amdgpu_device pointer
*
* Initialize the hardware, boot up the VCPU and do some testing
*/
static int uvd_v5_0_hw_init(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
struct amdgpu_ring *ring = &adev->uvd.ring;
uint32_t tmp;
int r;
/* raise clocks while booting up the VCPU */
amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
r = uvd_v5_0_start(adev);
if (r)
goto done;
ring->ready = true;
r = amdgpu_ring_test_ring(ring);
if (r) {
ring->ready = false;
goto done;
}
r = amdgpu_ring_alloc(ring, 10);
if (r) {
DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
goto done;
}
tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
amdgpu_ring_write(ring, tmp);
amdgpu_ring_write(ring, 0xFFFFF);
tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
amdgpu_ring_write(ring, tmp);
amdgpu_ring_write(ring, 0xFFFFF);
tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
amdgpu_ring_write(ring, tmp);
amdgpu_ring_write(ring, 0xFFFFF);
/* Clear timeout status bits */
amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
amdgpu_ring_write(ring, 0x8);
amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
amdgpu_ring_write(ring, 3);
amdgpu_ring_commit(ring);
done:
/* lower clocks again */
amdgpu_asic_set_uvd_clocks(adev, 0, 0);
if (!r)
DRM_INFO("UVD initialized successfully.\n");
return r;
}
static void uvd_v6_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
unsigned int vmid, uint64_t pd_addr)
{
amdgpu_ring_write(ring, HEVC_ENC_CMD_UPDATE_PTB);
amdgpu_ring_write(ring, vmid);
amdgpu_ring_write(ring, pd_addr >> 12);
amdgpu_ring_write(ring, HEVC_ENC_CMD_FLUSH_TLB);
amdgpu_ring_write(ring, vmid);
}
static void uvd_v6_0_enc_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
{
uint32_t seq = ring->fence_drv.sync_seq;
uint64_t addr = ring->fence_drv.gpu_addr;
amdgpu_ring_write(ring, HEVC_ENC_CMD_WAIT_GE);
amdgpu_ring_write(ring, lower_32_bits(addr));
amdgpu_ring_write(ring, upper_32_bits(addr));
amdgpu_ring_write(ring, seq);
}
/**
* cik_sdma_emit_fill_buffer - fill buffer using the sDMA engine
*
* @ring: amdgpu_ring structure holding ring information
* @src_data: value to write to buffer
* @dst_offset: dst GPU address
* @byte_count: number of bytes to xfer
*
* Fill GPU buffers using the DMA engine (CIK).
*/
static void cik_sdma_emit_fill_buffer(struct amdgpu_ring *ring,
uint32_t src_data,
uint64_t dst_offset,
uint32_t byte_count)
{
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_CONSTANT_FILL, 0, 0));
amdgpu_ring_write(ring, lower_32_bits(dst_offset));
amdgpu_ring_write(ring, upper_32_bits(dst_offset));
amdgpu_ring_write(ring, src_data);
amdgpu_ring_write(ring, byte_count);
}
/**
* uvd_v6_0_enc_ring_emit_fence - emit an enc fence & trap command
*
* @ring: amdgpu_ring pointer
* @fence: fence to emit
*
* Write enc a fence and a trap command to the ring.
*/
static void uvd_v6_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
u64 seq, unsigned flags)
{
WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
amdgpu_ring_write(ring, HEVC_ENC_CMD_FENCE);
amdgpu_ring_write(ring, addr);
amdgpu_ring_write(ring, upper_32_bits(addr));
amdgpu_ring_write(ring, seq);
amdgpu_ring_write(ring, HEVC_ENC_CMD_TRAP);
}
static void uvd_v6_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
{
int i;
WARN_ON(ring->wptr % 2 || count % 2);
for (i = 0; i < count / 2; i++) {
amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
amdgpu_ring_write(ring, 0);
}
}
static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
{
struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
int i;
for (i = 0; i < count; i++)
if (sdma && sdma->burst_nop && (i == 0))
amdgpu_ring_write(ring, ring->funcs->nop |
SDMA_NOP_COUNT(count - 1));
else
amdgpu_ring_write(ring, ring->funcs->nop);
}
/**
* uvd_v5_0_hw_init - start and test UVD block
*
* @adev: amdgpu_device pointer
*
* Initialize the hardware, boot up the VCPU and do some testing
*/
static int uvd_v5_0_hw_init(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
struct amdgpu_ring *ring = &adev->uvd.ring;
uint32_t tmp;
int r;
amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE);
uvd_v5_0_enable_mgcg(adev, true);
ring->ready = true;
r = amdgpu_ring_test_ring(ring);
if (r) {
ring->ready = false;
goto done;
}
r = amdgpu_ring_alloc(ring, 10);
if (r) {
DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
goto done;
}
tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
amdgpu_ring_write(ring, tmp);
amdgpu_ring_write(ring, 0xFFFFF);
tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
amdgpu_ring_write(ring, tmp);
amdgpu_ring_write(ring, 0xFFFFF);
tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
amdgpu_ring_write(ring, tmp);
amdgpu_ring_write(ring, 0xFFFFF);
/* Clear timeout status bits */
amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
amdgpu_ring_write(ring, 0x8);
amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
amdgpu_ring_write(ring, 3);
amdgpu_ring_commit(ring);
done:
if (!r)
DRM_INFO("UVD initialized successfully.\n");
return r;
}
/**
* uvd_v6_0_enc_ring_emit_ib - enc execute indirect buffer
*
* @ring: amdgpu_ring pointer
* @ib: indirect buffer to execute
*
* Write enc ring commands to execute the indirect buffer
*/
static void uvd_v6_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_job *job,
struct amdgpu_ib *ib,
uint32_t flags)
{
unsigned vmid = AMDGPU_JOB_GET_VMID(job);
amdgpu_ring_write(ring, HEVC_ENC_CMD_IB_VM);
amdgpu_ring_write(ring, vmid);
amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, ib->length_dw);
}
/**
* cik_sdma_ring_emit_semaphore - emit a semaphore on the dma ring
*
* @ring: amdgpu_ring structure holding ring information
* @semaphore: amdgpu semaphore object
* @emit_wait: wait or signal semaphore
*
* Add a DMA semaphore packet to the ring wait on or signal
* other rings (CIK).
*/
static bool cik_sdma_ring_emit_semaphore(struct amdgpu_ring *ring,
struct amdgpu_semaphore *semaphore,
bool emit_wait)
{
u64 addr = semaphore->gpu_addr;
u32 extra_bits = emit_wait ? 0 : SDMA_SEMAPHORE_EXTRA_S;
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits));
amdgpu_ring_write(ring, addr & 0xfffffff8);
amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
return true;
}
static void si_dma_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_ib *ib,
unsigned vm_id, bool ctx_switch)
{
/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
* Pad as necessary with NOPs.
*/
while ((ring->wptr & 7) != 5)
amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
amdgpu_ring_write(ring, DMA_IB_PACKET(DMA_PACKET_INDIRECT_BUFFER, vm_id, 0));
amdgpu_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
amdgpu_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF));
}
/**
* cik_sdma_ring_test_ring - simple async dma engine test
*
* @ring: amdgpu_ring structure holding ring information
*
* Test the DMA engine by writing using it to write an
* value to memory. (CIK).
* Returns 0 for success, error for failure.
*/
static int cik_sdma_ring_test_ring(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
unsigned i;
unsigned index;
int r;
u32 tmp;
u64 gpu_addr;
r = amdgpu_wb_get(adev, &index);
if (r) {
dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r);
return r;
}
gpu_addr = adev->wb.gpu_addr + (index * 4);
tmp = 0xCAFEDEAD;
adev->wb.wb[index] = cpu_to_le32(tmp);
r = amdgpu_ring_alloc(ring, 5);
if (r) {
DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r);
amdgpu_wb_free(adev, index);
return r;
}
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
amdgpu_ring_write(ring, lower_32_bits(gpu_addr));
amdgpu_ring_write(ring, upper_32_bits(gpu_addr));
amdgpu_ring_write(ring, 1); /* number of DWs to follow */
amdgpu_ring_write(ring, 0xDEADBEEF);
amdgpu_ring_commit(ring);
for (i = 0; i < adev->usec_timeout; i++) {
tmp = le32_to_cpu(adev->wb.wb[index]);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i < adev->usec_timeout) {
DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
} else {
DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
ring->idx, tmp);
r = -EINVAL;
}
amdgpu_wb_free(adev, index);
return r;
}
/**
* cik_sdma_ring_emit_ib - Schedule an IB on the DMA engine
*
* @ring: amdgpu ring pointer
* @ib: IB object to schedule
*
* Schedule an IB in the DMA ring (CIK).
*/
static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_ib *ib,
unsigned vm_id, bool ctx_switch)
{
u32 extra_bits = vm_id & 0xf;
/* IB packet must end on a 8 DW boundary */
cik_sdma_ring_insert_nop(ring, (12 - (lower_32_bits(ring->wptr) & 7)) % 8);
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
amdgpu_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff);
amdgpu_ring_write(ring, ib->length_dw);
}
/** amdgpu_ring_insert_nop - insert NOP packets
*
* @ring: amdgpu_ring structure holding ring information
* @count: the number of NOP packets to insert
*
* This is the generic insert_nop function for rings except SDMA
*/
void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
{
int i;
for (i = 0; i < count; i++)
amdgpu_ring_write(ring, ring->nop);
}
static void uvd_v6_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
{
uint32_t seq = ring->fence_drv.sync_seq;
uint64_t addr = ring->fence_drv.gpu_addr;
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
amdgpu_ring_write(ring, lower_32_bits(addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
amdgpu_ring_write(ring, upper_32_bits(addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_GP_SCRATCH8, 0));
amdgpu_ring_write(ring, 0xffffffff); /* mask */
amdgpu_ring_write(ring, PACKET0(mmUVD_GP_SCRATCH9, 0));
amdgpu_ring_write(ring, seq);
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
amdgpu_ring_write(ring, 0xE);
}
static void uvd_v6_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
unsigned vmid, uint64_t pd_addr)
{
amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST << 2);
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
amdgpu_ring_write(ring, 0);
amdgpu_ring_write(ring, PACKET0(mmUVD_GP_SCRATCH8, 0));
amdgpu_ring_write(ring, 1 << vmid); /* mask */
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
amdgpu_ring_write(ring, 0xC);
}
/**
* uvd_v6_0_ring_emit_ib - execute indirect buffer
*
* @ring: amdgpu_ring pointer
* @ib: indirect buffer to execute
*
* Write ring commands to execute the indirect buffer
*/
static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_ib *ib,
unsigned vm_id, bool ctx_switch)
{
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_VMID, 0));
amdgpu_ring_write(ring, vm_id);
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
amdgpu_ring_write(ring, ib->length_dw);
}
/**
* cik_sdma_emit_copy_buffer - copy buffer using the sDMA engine
*
* @ring: amdgpu_ring structure holding ring information
* @src_offset: src GPU address
* @dst_offset: dst GPU address
* @byte_count: number of bytes to xfer
*
* Copy GPU buffers using the DMA engine (CIK).
* Used by the amdgpu ttm implementation to move pages if
* registered as the asic copy callback.
*/
static void cik_sdma_emit_copy_buffer(struct amdgpu_ring *ring,
uint64_t src_offset,
uint64_t dst_offset,
uint32_t byte_count)
{
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0));
amdgpu_ring_write(ring, byte_count);
amdgpu_ring_write(ring, 0); /* src/dst endian swap */
amdgpu_ring_write(ring, lower_32_bits(src_offset));
amdgpu_ring_write(ring, upper_32_bits(src_offset));
amdgpu_ring_write(ring, lower_32_bits(dst_offset));
amdgpu_ring_write(ring, upper_32_bits(dst_offset));
}
/**
* uvd_v6_0_ring_emit_ib - execute indirect buffer
*
* @ring: amdgpu_ring pointer
* @ib: indirect buffer to execute
*
* Write ring commands to execute the indirect buffer
*/
static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_job *job,
struct amdgpu_ib *ib,
uint32_t flags)
{
unsigned vmid = AMDGPU_JOB_GET_VMID(job);
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_VMID, 0));
amdgpu_ring_write(ring, vmid);
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
amdgpu_ring_write(ring, ib->length_dw);
}
/**
* uvd_v6_0_ring_emit_ib - execute indirect buffer
*
* @ring: amdgpu_ring pointer
* @ib: indirect buffer to execute
*
* Write ring commands to execute the indirect buffer
*/
static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_ib *ib)
{
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
amdgpu_ring_write(ring, ib->length_dw);
}
/**
* si_dma_ring_emit_fence - emit a fence on the DMA ring
*
* @ring: amdgpu ring pointer
* @fence: amdgpu fence object
*
* Add a DMA fence packet to the ring to write
* the fence seq number and DMA trap packet to generate
* an interrupt if needed (VI).
*/
static void si_dma_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
unsigned flags)
{
bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
/* write the fence */
amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0, 0));
amdgpu_ring_write(ring, addr & 0xfffffffc);
amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xff));
amdgpu_ring_write(ring, seq);
/* optionally write high bits as well */
if (write64bit) {
addr += 4;
amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0, 0));
amdgpu_ring_write(ring, addr & 0xfffffffc);
amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xff));
amdgpu_ring_write(ring, upper_32_bits(seq));
}
/* generate an interrupt */
amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0, 0, 0));
}
static void uvd_v6_0_ring_emit_wreg(struct amdgpu_ring *ring,
uint32_t reg, uint32_t val)
{
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
amdgpu_ring_write(ring, reg << 2);
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
amdgpu_ring_write(ring, val);
amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
amdgpu_ring_write(ring, 0x8);
}
/**
* cik_sdma_ring_emit_pipeline_sync - sync the pipeline
*
* @ring: amdgpu_ring pointer
*
* Make sure all previous operations are completed (CIK).
*/
static void si_dma_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
{
uint32_t seq = ring->fence_drv.sync_seq;
uint64_t addr = ring->fence_drv.gpu_addr;
/* wait for idle */
amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_POLL_REG_MEM, 0, 0, 0, 0) |
(1 << 27)); /* Poll memory */
amdgpu_ring_write(ring, lower_32_bits(addr));
amdgpu_ring_write(ring, (0xff << 16) | upper_32_bits(addr)); /* retry, addr_hi */
amdgpu_ring_write(ring, 0xffffffff); /* mask */
amdgpu_ring_write(ring, seq); /* value */
amdgpu_ring_write(ring, (3 << 28) | 0x20); /* func(equal) | poll interval */
}
/**
* amdgpu_ring_restore - append saved commands to the ring again
*
* @ring: ring to append commands to
* @size: number of dwords we want to write
* @data: saved commands
*
* Allocates space on the ring and restore the previously saved commands.
*/
int amdgpu_ring_restore(struct amdgpu_ring *ring,
unsigned size, uint32_t *data)
{
int i, r;
if (!size || !data)
return 0;
/* restore the saved ring content */
r = amdgpu_ring_alloc(ring, size);
if (r)
return r;
for (i = 0; i < size; ++i) {
amdgpu_ring_write(ring, data[i]);
}
amdgpu_ring_commit(ring);
kfree(data);
return 0;
}
/**
* cik_sdma_ring_emit_pipeline_sync - sync the pipeline
*
* @ring: amdgpu_ring pointer
*
* Make sure all previous operations are completed (CIK).
*/
static void cik_sdma_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
{
uint32_t seq = ring->fence_drv.sync_seq;
uint64_t addr = ring->fence_drv.gpu_addr;
/* wait for idle */
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0,
SDMA_POLL_REG_MEM_EXTRA_OP(0) |
SDMA_POLL_REG_MEM_EXTRA_FUNC(3) | /* equal */
SDMA_POLL_REG_MEM_EXTRA_M));
amdgpu_ring_write(ring, addr & 0xfffffffc);
amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
amdgpu_ring_write(ring, seq); /* reference */
amdgpu_ring_write(ring, 0xfffffff); /* mask */
amdgpu_ring_write(ring, (0xfff << 16) | 4); /* retry count, poll interval */
}
/**
* cik_sdma_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
*
* @ring: amdgpu ring pointer
*
* Emit an hdp flush packet on the requested DMA ring.
*/
static void cik_sdma_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
u32 ref_and_mask;
if (ring == &ring->adev->sdma.instance[0].ring)
ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA0_MASK;
else
ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA1_MASK;
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE << 2);
amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ << 2);
amdgpu_ring_write(ring, ref_and_mask); /* reference */
amdgpu_ring_write(ring, ref_and_mask); /* mask */
amdgpu_ring_write(ring, (0xfff << 16) | 10); /* retry count, poll interval */
}
/**
* uvd_v6_0_enc_ring_test_ring - test if UVD ENC ring is working
*
* @ring: the engine to test on
*
*/
static int uvd_v6_0_enc_ring_test_ring(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
uint32_t rptr = amdgpu_ring_get_rptr(ring);
unsigned i;
int r;
r = amdgpu_ring_alloc(ring, 16);
if (r)
return r;
amdgpu_ring_write(ring, HEVC_ENC_CMD_END);
amdgpu_ring_commit(ring);
for (i = 0; i < adev->usec_timeout; i++) {
if (amdgpu_ring_get_rptr(ring) != rptr)
break;
DRM_UDELAY(1);
}
if (i >= adev->usec_timeout)
r = -ETIMEDOUT;
return r;
}
