From: Alex Deucher alexander.deucher@amd.com
Async page table updates using the sDMA engine. sDMA has a special packet for updating entries for contiguous pages that reduces overhead.
v2: add support for and use the CP for now. v3: update for 2 level PTs v4: rebase, fix DMA packet v5: switch to using an IB
Signed-off-by: Alex Deucher alexander.deucher@amd.com --- drivers/gpu/drm/radeon/cik.c | 109 ++++++++++++++++++++++++++++++++++++++++++ 1 files changed, 109 insertions(+), 0 deletions(-)
diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c index 3c18a63..cf1e0b1 100644 --- a/drivers/gpu/drm/radeon/cik.c +++ b/drivers/gpu/drm/radeon/cik.c @@ -3408,6 +3408,115 @@ void cik_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm) }
/** + * cik_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 CP or sDMA (CIK). + */ +void cik_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) +{ + uint32_t r600_flags = cayman_vm_page_flags(rdev, flags); + uint64_t value; + unsigned ndw; + + if (rdev->asic->vm.pt_ring_index == RADEON_RING_TYPE_GFX_INDEX) { + /* CP */ + while (count) { + ndw = 2 + count * 2; + if (ndw > 0x3FFE) + ndw = 0x3FFE; + + ib->ptr[ib->length_dw++] = PACKET3(PACKET3_WRITE_DATA, ndw); + ib->ptr[ib->length_dw++] = (WRITE_DATA_ENGINE_SEL(0) | + WRITE_DATA_DST_SEL(1)); + ib->ptr[ib->length_dw++] = pe; + ib->ptr[ib->length_dw++] = upper_32_bits(pe); + for (; ndw > 2; ndw -= 2, --count, pe += 8) { + if (flags & RADEON_VM_PAGE_SYSTEM) { + value = radeon_vm_map_gart(rdev, addr); + value &= 0xFFFFFFFFFFFFF000ULL; + } else if (flags & RADEON_VM_PAGE_VALID) { + value = addr; + } else { + value = 0; + } + addr += incr; + value |= r600_flags; + ib->ptr[ib->length_dw++] = value; + ib->ptr[ib->length_dw++] = upper_32_bits(value); + } + } + } else { + /* DMA */ + if (flags & RADEON_VM_PAGE_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) { + if (flags & RADEON_VM_PAGE_SYSTEM) { + value = radeon_vm_map_gart(rdev, addr); + value &= 0xFFFFFFFFFFFFF000ULL; + } else if (flags & RADEON_VM_PAGE_VALID) { + value = addr; + } else { + value = 0; + } + addr += incr; + value |= r600_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 & RADEON_VM_PAGE_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++] = r600_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); + } +} + +/** * cik_dma_vm_flush - cik vm flush using sDMA * * @rdev: radeon_device pointer