/**
* cik_sdma_ring_emit_vm_flush - cik vm flush using sDMA
*
* @ring: amdgpu_ring pointer
* @vm: amdgpu_vm pointer
*
* Update the page table base and flush the VM TLB
* using sDMA (CIK).
*/
static void cik_sdma_ring_emit_vm_flush(struct amdgpu_ring *ring,
unsigned vm_id, uint64_t pd_addr)
{
u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(0) |
SDMA_POLL_REG_MEM_EXTRA_FUNC(0)); /* always */
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
if (vm_id < 8) {
amdgpu_ring_write(ring, (mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vm_id));
} else {
amdgpu_ring_write(ring, (mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vm_id - 8));
}
amdgpu_ring_write(ring, pd_addr >> 12);
/* flush TLB */
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
amdgpu_ring_write(ring, 1 << vm_id);
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits));
amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST << 2);
amdgpu_ring_write(ring, 0);
amdgpu_ring_write(ring, 0); /* reference */
amdgpu_ring_write(ring, 0); /* mask */
amdgpu_ring_write(ring, (0xfff << 16) | 10); /* retry count, poll interval */
}
/**
* 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 */
cik_sdma_ring_insert_nop(ring, (12 - (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);
}
/**
* cik_sdma_ring_ib_execute - Schedule an IB on the DMA engine
*
* @rdev: radeon_device pointer
* @ib: IB object to schedule
*
* Schedule an IB in the DMA ring (CIK).
*/
void cik_sdma_ring_ib_execute(struct radeon_device *rdev,
struct radeon_ib *ib)
{
struct radeon_ring *ring = &rdev->ring[ib->ring];
u32 extra_bits = (ib->vm ? ib->vm->id : 0) & 0xf;
if (rdev->wb.enabled) {
u32 next_rptr = ring->wptr + 5;
while ((next_rptr & 7) != 4)
next_rptr++;
next_rptr += 4;
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0));
radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr));
radeon_ring_write(ring, 1); /* number of DWs to follow */
radeon_ring_write(ring, next_rptr);
}
/* IB packet must end on a 8 DW boundary */
while ((ring->wptr & 7) != 4)
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits));
radeon_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */
radeon_ring_write(ring, upper_32_bits(ib->gpu_addr));
radeon_ring_write(ring, ib->length_dw);
}
/**
* cik_sdma_vm_set_page - update the page tables using sDMA
*
* @rdev: radeon_device pointer
* @ib: indirect buffer to fill with commands
* @pe: addr of the page entry
* @addr: dst addr to write into pe
* @count: number of page entries to update
* @incr: increase next addr by incr bytes
* @flags: access flags
*
* Update the page tables using sDMA (CIK).
*/
void cik_sdma_vm_set_page(struct radeon_device *rdev,
struct radeon_ib *ib,
uint64_t pe,
uint64_t addr, unsigned count,
uint32_t incr, uint32_t flags)
{
uint64_t value;
unsigned ndw;
trace_radeon_vm_set_page(pe, addr, count, incr, flags);
if (flags & R600_PTE_SYSTEM) {
while (count) {
ndw = count * 2;
if (ndw > 0xFFFFE)
ndw = 0xFFFFE;
/* for non-physically contiguous pages (system) */
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib->ptr[ib->length_dw++] = pe;
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
ib->ptr[ib->length_dw++] = ndw;
for (; ndw > 0; ndw -= 2, --count, pe += 8) {
value = radeon_vm_map_gart(rdev, addr);
value &= 0xFFFFFFFFFFFFF000ULL;
addr += incr;
value |= flags;
ib->ptr[ib->length_dw++] = value;
ib->ptr[ib->length_dw++] = upper_32_bits(value);
}
}
} else {
while (count) {
ndw = count;
if (ndw > 0x7FFFF)
ndw = 0x7FFFF;
if (flags & R600_PTE_VALID)
value = addr;
else
value = 0;
/* for physically contiguous pages (vram) */
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_GENERATE_PTE_PDE, 0, 0);
ib->ptr[ib->length_dw++] = pe; /* dst addr */
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
ib->ptr[ib->length_dw++] = flags; /* mask */
ib->ptr[ib->length_dw++] = 0;
ib->ptr[ib->length_dw++] = value; /* value */
ib->ptr[ib->length_dw++] = upper_32_bits(value);
ib->ptr[ib->length_dw++] = incr; /* increment size */
ib->ptr[ib->length_dw++] = 0;
ib->ptr[ib->length_dw++] = ndw; /* number of entries */
pe += ndw * 8;
addr += ndw * incr;
count -= ndw;
}
}
while (ib->length_dw & 0x7)
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
}
static int cik_sdma_sw_init(void *handle)
{
struct amdgpu_ring *ring;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
int r;
r = cik_sdma_init_microcode(adev);
if (r) {
DRM_ERROR("Failed to load sdma firmware!\n");
return r;
}
/* SDMA trap event */
r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq);
if (r)
return r;
/* SDMA Privileged inst */
r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq);
if (r)
return r;
/* SDMA Privileged inst */
r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq);
if (r)
return r;
ring = &adev->sdma[0].ring;
ring->ring_obj = NULL;
ring = &adev->sdma[1].ring;
ring->ring_obj = NULL;
ring = &adev->sdma[0].ring;
sprintf(ring->name, "sdma0");
r = amdgpu_ring_init(adev, ring, 256 * 1024,
SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf,
&adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0,
AMDGPU_RING_TYPE_SDMA);
if (r)
return r;
ring = &adev->sdma[1].ring;
sprintf(ring->name, "sdma1");
r = amdgpu_ring_init(adev, ring, 256 * 1024,
SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf,
&adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1,
AMDGPU_RING_TYPE_SDMA);
if (r)
return r;
return r;
}
/**
* cik_sdma_fence_ring_emit - emit a fence on the DMA ring
*
* @rdev: radeon_device pointer
* @fence: radeon 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 (CIK).
*/
void cik_sdma_fence_ring_emit(struct radeon_device *rdev,
struct radeon_fence *fence)
{
struct radeon_ring *ring = &rdev->ring[fence->ring];
u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
/* write the fence */
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, 0));
radeon_ring_write(ring, lower_32_bits(addr));
radeon_ring_write(ring, upper_32_bits(addr));
radeon_ring_write(ring, fence->seq);
/* generate an interrupt */
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_TRAP, 0, 0));
/* flush HDP */
cik_sdma_hdp_flush_ring_emit(rdev, fence->ring);
}
/**
* cik_sdma_vm_write_pages - update PTEs by writing them manually
*
* @ib: indirect buffer to fill with commands
* @pe: addr of the page entry
* @addr: dst addr to write into pe
* @count: number of page entries to update
* @incr: increase next addr by incr bytes
* @flags: access flags
*
* Update PTEs by writing them manually using sDMA (CIK).
*/
static void cik_sdma_vm_write_pte(struct amdgpu_ib *ib,
uint64_t pe,
uint64_t addr, unsigned count,
uint32_t incr, uint32_t flags)
{
uint64_t value;
unsigned ndw;
while (count) {
ndw = count * 2;
if (ndw > 0xFFFFE)
ndw = 0xFFFFE;
/* for non-physically contiguous pages (system) */
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_WRITE,
SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib->ptr[ib->length_dw++] = pe;
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
ib->ptr[ib->length_dw++] = ndw;
for (; ndw > 0; ndw -= 2, --count, pe += 8) {
if (flags & AMDGPU_PTE_SYSTEM) {
value = amdgpu_vm_map_gart(ib->ring->adev, addr);
value &= 0xFFFFFFFFFFFFF000ULL;
} else if (flags & AMDGPU_PTE_VALID) {
value = addr;
} else {
value = 0;
}
addr += incr;
value |= flags;
ib->ptr[ib->length_dw++] = value;
ib->ptr[ib->length_dw++] = upper_32_bits(value);
}
}
}
/**
* cik_sdma_vm_set_pages - update the page tables using sDMA
*
* @ib: indirect buffer to fill with commands
* @pe: addr of the page entry
* @addr: dst addr to write into pe
* @count: number of page entries to update
* @incr: increase next addr by incr bytes
* @flags: access flags
*
* Update the page tables using sDMA (CIK).
*/
static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib,
uint64_t pe,
uint64_t addr, unsigned count,
uint32_t incr, uint32_t flags)
{
uint64_t value;
unsigned ndw;
while (count) {
ndw = count;
if (ndw > 0x7FFFF)
ndw = 0x7FFFF;
if (flags & AMDGPU_PTE_VALID)
value = addr;
else
value = 0;
/* for physically contiguous pages (vram) */
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_GENERATE_PTE_PDE, 0, 0);
ib->ptr[ib->length_dw++] = pe; /* dst addr */
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
ib->ptr[ib->length_dw++] = flags; /* mask */
ib->ptr[ib->length_dw++] = 0;
ib->ptr[ib->length_dw++] = value; /* value */
ib->ptr[ib->length_dw++] = upper_32_bits(value);
ib->ptr[ib->length_dw++] = incr; /* increment size */
ib->ptr[ib->length_dw++] = 0;
ib->ptr[ib->length_dw++] = ndw; /* number of entries */
pe += ndw * 8;
addr += ndw * incr;
count -= ndw;
}
}
/**
* cik_sdma_vm_pad_ib - pad the IB to the required number of dw
*
* @ib: indirect buffer to fill with padding
*
*/
static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
{
struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
u32 pad_count;
int i;
pad_count = (8 - (ib->length_dw & 0x7)) % 8;
for (i = 0; i < pad_count; i++)
if (sdma && sdma->burst_nop && (i == 0))
ib->ptr[ib->length_dw++] =
SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0) |
SDMA_NOP_COUNT(pad_count - 1);
else
ib->ptr[ib->length_dw++] =
SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0);
}
/**
* cik_sdma_vm_set_page - update the page tables using sDMA
*
* @rdev: radeon_device pointer
* @ib: indirect buffer to fill with commands
* @pe: addr of the page entry
* @addr: dst addr to write into pe
* @count: number of page entries to update
* @incr: increase next addr by incr bytes
* @flags: access flags
*
* Update the page tables using sDMA (CIK).
*/
void cik_sdma_vm_set_page(struct radeon_device *rdev,
struct radeon_ib *ib,
uint64_t pe,
uint64_t addr, unsigned count,
uint32_t incr, uint32_t flags)
{
uint64_t value;
unsigned ndw;
trace_radeon_vm_set_page(pe, addr, count, incr, flags);
if (flags == R600_PTE_GART) {
uint64_t src = rdev->gart.table_addr + (addr >> 12) * 8;
while (count) {
unsigned bytes = count * 8;
if (bytes > 0x1FFFF8)
bytes = 0x1FFFF8;
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib->ptr[ib->length_dw++] = bytes;
ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
ib->ptr[ib->length_dw++] = lower_32_bits(src);
ib->ptr[ib->length_dw++] = upper_32_bits(src);
ib->ptr[ib->length_dw++] = lower_32_bits(pe);
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
pe += bytes;
src += bytes;
count -= bytes / 8;
}
} else if (flags & R600_PTE_SYSTEM) {
/**
* cik_sdma_vm_write_pages - update PTEs by writing them manually
*
* @ib: indirect buffer to fill with commands
* @pe: addr of the page entry
* @addr: dst addr to write into pe
* @count: number of page entries to update
* @incr: increase next addr by incr bytes
* @flags: access flags
*
* Update PTEs by writing them manually using sDMA (CIK).
*/
static void cik_sdma_vm_write_pte(struct amdgpu_ib *ib,
const dma_addr_t *pages_addr, uint64_t pe,
uint64_t addr, unsigned count,
uint32_t incr, uint32_t flags)
{
uint64_t value;
unsigned ndw;
while (count) {
ndw = count * 2;
if (ndw > 0xFFFFE)
ndw = 0xFFFFE;
/* for non-physically contiguous pages (system) */
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_WRITE,
SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib->ptr[ib->length_dw++] = pe;
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
ib->ptr[ib->length_dw++] = ndw;
for (; ndw > 0; ndw -= 2, --count, pe += 8) {
value = amdgpu_vm_map_gart(pages_addr, addr);
addr += incr;
value |= flags;
ib->ptr[ib->length_dw++] = value;
ib->ptr[ib->length_dw++] = upper_32_bits(value);
}
}
}
/**
* 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;
}
static void amdgpu_command_submission_write_linear_helper(unsigned ip_type)
{
const int sdma_write_length = 128;
const int pm4_dw = 256;
amdgpu_context_handle context_handle;
amdgpu_bo_handle bo;
amdgpu_bo_handle *resources;
uint32_t *pm4;
struct amdgpu_cs_ib_info *ib_info;
struct amdgpu_cs_request *ibs_request;
uint64_t bo_mc;
volatile uint32_t *bo_cpu;
int i, j, r, loop;
uint64_t gtt_flags[2] = {0, AMDGPU_GEM_CREATE_CPU_GTT_USWC};
amdgpu_va_handle va_handle;
pm4 = calloc(pm4_dw, sizeof(*pm4));
CU_ASSERT_NOT_EQUAL(pm4, NULL);
ib_info = calloc(1, sizeof(*ib_info));
CU_ASSERT_NOT_EQUAL(ib_info, NULL);
ibs_request = calloc(1, sizeof(*ibs_request));
CU_ASSERT_NOT_EQUAL(ibs_request, NULL);
r = amdgpu_cs_ctx_create(device_handle, &context_handle);
CU_ASSERT_EQUAL(r, 0);
/* prepare resource */
resources = calloc(1, sizeof(amdgpu_bo_handle));
CU_ASSERT_NOT_EQUAL(resources, NULL);
loop = 0;
while(loop < 2) {
/* allocate UC bo for sDMA use */
r = amdgpu_bo_alloc_and_map(device_handle,
sdma_write_length * sizeof(uint32_t),
4096, AMDGPU_GEM_DOMAIN_GTT,
gtt_flags[loop], &bo, (void**)&bo_cpu,
&bo_mc, &va_handle);
CU_ASSERT_EQUAL(r, 0);
/* clear bo */
memset((void*)bo_cpu, 0, sdma_write_length * sizeof(uint32_t));
resources[0] = bo;
/* fullfill PM4: test DMA write-linear */
i = j = 0;
if (ip_type == AMDGPU_HW_IP_DMA) {
pm4[i++] = SDMA_PACKET(SDMA_OPCODE_WRITE,
SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
pm4[i++] = 0xffffffff & bo_mc;
pm4[i++] = (0xffffffff00000000 & bo_mc) >> 32;
pm4[i++] = sdma_write_length;
while(j++ < sdma_write_length)
pm4[i++] = 0xdeadbeaf;
} else if (ip_type == AMDGPU_HW_IP_GFX) {
/**
* cik_sdma_ib_test - test an IB on the DMA engine
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
*
* Test a simple IB in the DMA ring (CIK).
* Returns 0 on success, error on failure.
*/
int cik_sdma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
{
struct radeon_ib ib;
unsigned i;
unsigned index;
int r;
u32 tmp = 0;
u64 gpu_addr;
if (ring->idx == R600_RING_TYPE_DMA_INDEX)
index = R600_WB_DMA_RING_TEST_OFFSET;
else
index = CAYMAN_WB_DMA1_RING_TEST_OFFSET;
gpu_addr = rdev->wb.gpu_addr + index;
tmp = 0xCAFEDEAD;
rdev->wb.wb[index/4] = cpu_to_le32(tmp);
r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
if (r) {
DRM_ERROR("radeon: failed to get ib (%d).\n", r);
return r;
}
ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib.ptr[1] = lower_32_bits(gpu_addr);
ib.ptr[2] = upper_32_bits(gpu_addr);
ib.ptr[3] = 1;
ib.ptr[4] = 0xDEADBEEF;
ib.length_dw = 5;
r = radeon_ib_schedule(rdev, &ib, NULL, false);
if (r) {
radeon_ib_free(rdev, &ib);
DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
return r;
}
r = radeon_fence_wait(ib.fence, false);
if (r) {
DRM_ERROR("radeon: fence wait failed (%d).\n", r);
return r;
}
for (i = 0; i < rdev->usec_timeout; i++) {
tmp = le32_to_cpu(rdev->wb.wb[index/4]);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i < rdev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
} else {
DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
radeon_ib_free(rdev, &ib);
return r;
}
/**
* cik_dma_vm_flush - cik vm flush using sDMA
*
* @rdev: radeon_device pointer
*
* Update the page table base and flush the VM TLB
* using sDMA (CIK).
*/
void cik_dma_vm_flush(struct radeon_device *rdev, struct radeon_ring *ring,
unsigned vm_id, uint64_t pd_addr)
{
u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(0) |
SDMA_POLL_REG_MEM_EXTRA_FUNC(0)); /* always */
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
if (vm_id < 8) {
radeon_ring_write(ring, (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm_id << 2)) >> 2);
} else {
/**
* cik_sdma_ib_test - test an IB on the DMA engine
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
*
* Test a simple IB in the DMA ring (CIK).
* Returns 0 on success, error on failure.
*/
int cik_sdma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
{
struct radeon_ib ib;
unsigned i;
int r;
void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
u32 tmp = 0;
if (!ptr) {
DRM_ERROR("invalid vram scratch pointer\n");
return -EINVAL;
}
tmp = 0xCAFEDEAD;
writel(tmp, ptr);
r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
if (r) {
DRM_ERROR("radeon: failed to get ib (%d).\n", r);
return r;
}
ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc;
ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr);
ib.ptr[3] = 1;
ib.ptr[4] = 0xDEADBEEF;
ib.length_dw = 5;
r = radeon_ib_schedule(rdev, &ib, NULL);
if (r) {
radeon_ib_free(rdev, &ib);
DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
return r;
}
r = radeon_fence_wait(ib.fence, false);
if (r) {
DRM_ERROR("radeon: fence wait failed (%d).\n", r);
return r;
}
for (i = 0; i < rdev->usec_timeout; i++) {
tmp = readl(ptr);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i < rdev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
} else {
DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
radeon_ib_free(rdev, &ib);
return r;
}
/**
* 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);
}
/**
* 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_ib *ib,
uint32_t src_data,
uint64_t dst_offset,
uint32_t byte_count)
{
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_CONSTANT_FILL, 0, 0);
ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
ib->ptr[ib->length_dw++] = src_data;
ib->ptr[ib->length_dw++] = byte_count;
}
/**
* cik_dma_vm_flush - cik vm flush using sDMA
*
* @rdev: radeon_device pointer
*
* Update the page table base and flush the VM TLB
* using sDMA (CIK).
*/
void cik_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
{
struct radeon_ring *ring = &rdev->ring[ridx];
if (vm == NULL)
return;
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
if (vm->id < 8) {
radeon_ring_write(ring, (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
} else {
/**
* cik_dma_vm_flush - cik vm flush using sDMA
*
* @rdev: radeon_device pointer
*
* Update the page table base and flush the VM TLB
* using sDMA (CIK).
*/
void cik_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
{
u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(0) |
SDMA_POLL_REG_MEM_EXTRA_FUNC(0)); /* always */
struct radeon_ring *ring = &rdev->ring[ridx];
if (vm == NULL)
return;
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000));
if (vm->id < 8) {
radeon_ring_write(ring, (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
} else {
/**
* 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;
}
/**
* cik_sdma_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 (CIK).
*/
static void cik_sdma_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, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, 0));
amdgpu_ring_write(ring, lower_32_bits(addr));
amdgpu_ring_write(ring, upper_32_bits(addr));
amdgpu_ring_write(ring, lower_32_bits(seq));
/* optionally write high bits as well */
if (write64bit) {
addr += 4;
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, 0));
amdgpu_ring_write(ring, lower_32_bits(addr));
amdgpu_ring_write(ring, upper_32_bits(addr));
amdgpu_ring_write(ring, upper_32_bits(seq));
}
/* generate an interrupt */
amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_TRAP, 0, 0));
}
/**
* 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_ib *ib,
uint64_t src_offset,
uint64_t dst_offset,
uint32_t byte_count)
{
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0);
ib->ptr[ib->length_dw++] = byte_count;
ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
}
/**
* 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));
}
/**
* cik_sdma_semaphore_ring_emit - emit a semaphore on the dma ring
*
* @rdev: radeon_device pointer
* @ring: radeon_ring structure holding ring information
* @semaphore: radeon semaphore object
* @emit_wait: wait or signal semaphore
*
* Add a DMA semaphore packet to the ring wait on or signal
* other rings (CIK).
*/
bool cik_sdma_semaphore_ring_emit(struct radeon_device *rdev,
struct radeon_ring *ring,
struct radeon_semaphore *semaphore,
bool emit_wait)
{
u64 addr = semaphore->gpu_addr;
u32 extra_bits = emit_wait ? 0 : SDMA_SEMAPHORE_EXTRA_S;
radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits));
radeon_ring_write(ring, addr & 0xfffffff8);
radeon_ring_write(ring, upper_32_bits(addr));
return true;
}
/**
* 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_vm_copy_pages - update PTEs by copying them from the GART
*
* @ib: indirect buffer to fill with commands
* @pe: addr of the page entry
* @src: src addr to copy from
* @count: number of page entries to update
*
* Update PTEs by copying them from the GART using sDMA (CIK).
*/
static void cik_sdma_vm_copy_pte(struct amdgpu_ib *ib,
uint64_t pe, uint64_t src,
unsigned count)
{
unsigned bytes = count * 8;
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_COPY,
SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib->ptr[ib->length_dw++] = bytes;
ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
ib->ptr[ib->length_dw++] = lower_32_bits(src);
ib->ptr[ib->length_dw++] = upper_32_bits(src);
ib->ptr[ib->length_dw++] = lower_32_bits(pe);
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
}
/**
* 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);
}
/**
* cik_sdma_vm_set_pages - update the page tables using sDMA
*
* @ib: indirect buffer to fill with commands
* @pe: addr of the page entry
* @addr: dst addr to write into pe
* @count: number of page entries to update
* @incr: increase next addr by incr bytes
* @flags: access flags
*
* Update the page tables using sDMA (CIK).
*/
static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib, uint64_t pe,
uint64_t addr, unsigned count,
uint32_t incr, uint64_t flags)
{
/* for physically contiguous pages (vram) */
ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_GENERATE_PTE_PDE, 0, 0);
ib->ptr[ib->length_dw++] = lower_32_bits(pe); /* dst addr */
ib->ptr[ib->length_dw++] = upper_32_bits(pe);
ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */
ib->ptr[ib->length_dw++] = upper_32_bits(flags);
ib->ptr[ib->length_dw++] = lower_32_bits(addr); /* value */
ib->ptr[ib->length_dw++] = upper_32_bits(addr);
ib->ptr[ib->length_dw++] = incr; /* increment size */
ib->ptr[ib->length_dw++] = 0;
ib->ptr[ib->length_dw++] = count; /* number of entries */
}
/**
* 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 */
}
