rooty/open-gpu-kernel-modules
1
0
Fork 0
mirror of https://github.com/NVIDIA/open-gpu-kernel-modules.git synced 2024-12-11 09:24:15 +01:00
Code Issues Projects Releases Wiki Activity
94eaea9726
open-gpu-kernel-modules/kernel-open/nvidia-uvm/uvm_maxwell_ce.c
Andy Ritger 965db98552
515.48.07
2022-05-27 16:40:24 -07:00

378 lines
15 KiB
C
Raw Blame History

/*******************************************************************************
Copyright (c) 2021 NVIDIA Corporation
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to
deal in the Software without restriction, including without limitation the
rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
sell copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.
*******************************************************************************/
#include "uvm_hal.h"
#include "uvm_push.h"
#include "clb06f.h"
#include "clb0b5.h"
void uvm_hal_maxwell_ce_init(uvm_push_t *push)
{
// Notably this sends SET_OBJECT with the CE class on subchannel 0 instead
// of the recommended by HW subchannel 4 (subchannel 4 is recommended to
// match CE usage on GRCE). For the UVM driver using subchannel 0 has the
// benefit of also verifying that we ended up on the right PBDMA though as
// SET_OBJECT with CE class on subchannel 0 would fail on GRCE.
NV_PUSH_1U(B06F, SET_OBJECT, uvm_push_get_gpu(push)->parent->rm_info.ceClass);
}
void uvm_hal_maxwell_ce_offset_out(uvm_push_t *push, NvU64 offset_out)
{
NV_PUSH_2U(B0B5, OFFSET_OUT_UPPER, HWVALUE(B0B5, OFFSET_OUT_UPPER, UPPER, NvOffset_HI32(offset_out)),
OFFSET_OUT_LOWER, HWVALUE(B0B5, OFFSET_OUT_LOWER, VALUE, NvOffset_LO32(offset_out)));
}
void uvm_hal_maxwell_ce_offset_in_out(uvm_push_t *push, NvU64 offset_in, NvU64 offset_out)
{
NV_PUSH_4U(B0B5, OFFSET_IN_UPPER, HWVALUE(B0B5, OFFSET_IN_UPPER, UPPER, NvOffset_HI32(offset_in)),
OFFSET_IN_LOWER, HWVALUE(B0B5, OFFSET_IN_LOWER, VALUE, NvOffset_LO32(offset_in)),
OFFSET_OUT_UPPER, HWVALUE(B0B5, OFFSET_OUT_UPPER, UPPER, NvOffset_HI32(offset_out)),
OFFSET_OUT_LOWER, HWVALUE(B0B5, OFFSET_OUT_LOWER, VALUE, NvOffset_LO32(offset_out)));
}
// Perform an appropriate membar before a semaphore operation. Returns whether
// the semaphore operation should include a flush.
static bool maxwell_membar_before_semaphore(uvm_push_t *push)
{
uvm_gpu_t *gpu;
if (uvm_push_get_and_reset_flag(push, UVM_PUSH_FLAG_NEXT_MEMBAR_NONE)) {
// No MEMBAR requested, don't use a flush.
return false;
}
if (!uvm_push_get_and_reset_flag(push, UVM_PUSH_FLAG_NEXT_MEMBAR_GPU)) {
// By default do a MEMBAR SYS and for that we can just use flush on the
// semaphore operation.
return true;
}
// MEMBAR GPU requested, do it on the HOST and skip the CE flush as CE
// doesn't have this capability.
gpu = uvm_push_get_gpu(push);
gpu->parent->host_hal->wait_for_idle(push);
gpu->parent->host_hal->membar_gpu(push);
return false;
}
void uvm_hal_maxwell_ce_semaphore_release(uvm_push_t *push, NvU64 gpu_va, NvU32 payload)
{
NvU32 flush_value;
bool use_flush;
use_flush = maxwell_membar_before_semaphore(push);
if (use_flush)
flush_value = HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, TRUE);
else
flush_value = HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, FALSE);
NV_PUSH_3U(B0B5, SET_SEMAPHORE_A, HWVALUE(B0B5, SET_SEMAPHORE_A, UPPER, NvOffset_HI32(gpu_va)),
SET_SEMAPHORE_B, HWVALUE(B0B5, SET_SEMAPHORE_B, LOWER, NvOffset_LO32(gpu_va)),
SET_SEMAPHORE_PAYLOAD, payload);
NV_PUSH_1U(B0B5, LAUNCH_DMA, flush_value |
HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, NONE) |
HWCONST(B0B5, LAUNCH_DMA, SEMAPHORE_TYPE, RELEASE_ONE_WORD_SEMAPHORE));
}
void uvm_hal_maxwell_ce_semaphore_reduction_inc(uvm_push_t *push, NvU64 gpu_va, NvU32 payload)
{
NvU32 flush_value;
bool use_flush;
use_flush = maxwell_membar_before_semaphore(push);
if (use_flush)
flush_value = HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, TRUE);
else
flush_value = HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, FALSE);
NV_PUSH_3U(B0B5, SET_SEMAPHORE_A, HWVALUE(B0B5, SET_SEMAPHORE_A, UPPER, NvOffset_HI32(gpu_va)),
SET_SEMAPHORE_B, HWVALUE(B0B5, SET_SEMAPHORE_B, LOWER, NvOffset_LO32(gpu_va)),
SET_SEMAPHORE_PAYLOAD, payload);
NV_PUSH_1U(B0B5, LAUNCH_DMA, flush_value |
HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, NONE) |
HWCONST(B0B5, LAUNCH_DMA, SEMAPHORE_TYPE, RELEASE_ONE_WORD_SEMAPHORE) |
HWCONST(B0B5, LAUNCH_DMA, SEMAPHORE_REDUCTION, INC) |
HWCONST(B0B5, LAUNCH_DMA, SEMAPHORE_REDUCTION_SIGN, UNSIGNED) |
HWCONST(B0B5, LAUNCH_DMA, SEMAPHORE_REDUCTION_ENABLE, TRUE));
}
void uvm_hal_maxwell_ce_semaphore_timestamp(uvm_push_t *push, NvU64 gpu_va)
{
NvU32 flush_value;
bool use_flush;
use_flush = maxwell_membar_before_semaphore(push);
if (use_flush)
flush_value = HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, TRUE);
else
flush_value = HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, FALSE);
NV_PUSH_3U(B0B5, SET_SEMAPHORE_A, HWVALUE(B0B5, SET_SEMAPHORE_A, UPPER, NvOffset_HI32(gpu_va)),
SET_SEMAPHORE_B, HWVALUE(B0B5, SET_SEMAPHORE_B, LOWER, NvOffset_LO32(gpu_va)),
SET_SEMAPHORE_PAYLOAD, 0xdeadbeef);
NV_PUSH_1U(B0B5, LAUNCH_DMA, flush_value |
HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, NONE) |
HWCONST(B0B5, LAUNCH_DMA, SEMAPHORE_TYPE, RELEASE_FOUR_WORD_SEMAPHORE));
}
static void maxwell_membar_after_transfer(uvm_push_t *push)
{
uvm_gpu_t *gpu = uvm_push_get_gpu(push);
if (uvm_push_get_and_reset_flag(push, UVM_PUSH_FLAG_NEXT_MEMBAR_NONE))
return;
// Flush on transfers only works when paired with a semaphore release. Use a
// host WFI + MEMBAR.
// Bug 1709888
gpu->parent->host_hal->wait_for_idle(push);
if (uvm_push_get_and_reset_flag(push, UVM_PUSH_FLAG_NEXT_MEMBAR_GPU))
gpu->parent->host_hal->membar_gpu(push);
else
gpu->parent->host_hal->membar_sys(push);
}
static NvU32 ce_aperture(uvm_aperture_t aperture)
{
BUILD_BUG_ON(HWCONST(B0B5, SET_SRC_PHYS_MODE, TARGET, LOCAL_FB) !=
HWCONST(B0B5, SET_DST_PHYS_MODE, TARGET, LOCAL_FB));
BUILD_BUG_ON(HWCONST(B0B5, SET_SRC_PHYS_MODE, TARGET, COHERENT_SYSMEM) !=
HWCONST(B0B5, SET_DST_PHYS_MODE, TARGET, COHERENT_SYSMEM));
UVM_ASSERT_MSG(aperture == UVM_APERTURE_VID || aperture == UVM_APERTURE_SYS, "aperture 0x%x\n", aperture);
if (aperture == UVM_APERTURE_SYS)
return HWCONST(B0B5, SET_SRC_PHYS_MODE, TARGET, COHERENT_SYSMEM);
else
return HWCONST(B0B5, SET_SRC_PHYS_MODE, TARGET, LOCAL_FB);
}
// Push SET_{SRC,DST}_PHYS mode if needed and return LAUNCH_DMA_{SRC,DST}_TYPE
// flags
NvU32 uvm_hal_maxwell_ce_phys_mode(uvm_push_t *push, uvm_gpu_address_t dst, uvm_gpu_address_t src)
{
NvU32 launch_dma_src_dst_type = 0;
if (src.is_virtual)
launch_dma_src_dst_type |= HWCONST(B0B5, LAUNCH_DMA, SRC_TYPE, VIRTUAL);
else
launch_dma_src_dst_type |= HWCONST(B0B5, LAUNCH_DMA, SRC_TYPE, PHYSICAL);
if (dst.is_virtual)
launch_dma_src_dst_type |= HWCONST(B0B5, LAUNCH_DMA, DST_TYPE, VIRTUAL);
else
launch_dma_src_dst_type |= HWCONST(B0B5, LAUNCH_DMA, DST_TYPE, PHYSICAL);
if (!src.is_virtual && !dst.is_virtual) {
NV_PUSH_2U(B0B5, SET_SRC_PHYS_MODE, ce_aperture(src.aperture),
SET_DST_PHYS_MODE, ce_aperture(dst.aperture));
}
else if (!src.is_virtual) {
NV_PUSH_1U(B0B5, SET_SRC_PHYS_MODE, ce_aperture(src.aperture));
}
else if (!dst.is_virtual) {
NV_PUSH_1U(B0B5, SET_DST_PHYS_MODE, ce_aperture(dst.aperture));
}
return launch_dma_src_dst_type;
}
// Noop, since DISABLE_PLC doesn't exist in Maxwell.
NvU32 uvm_hal_maxwell_ce_plc_mode(void)
{
return 0;
}
void uvm_hal_maxwell_ce_memcopy(uvm_push_t *push, uvm_gpu_address_t dst, uvm_gpu_address_t src, size_t size)
{
// If >4GB copies ever become an important use case, this function should
// use multi-line transfers so we don't have to iterate (bug 1766588).
static const size_t max_single_copy_size = 0xFFFFFFFF;
uvm_gpu_t *gpu = uvm_push_get_gpu(push);
NvU32 pipelined_value;
NvU32 launch_dma_src_dst_type;
NvU32 launch_dma_plc_mode;
bool first_operation = true;
UVM_ASSERT_MSG(gpu->parent->ce_hal->memcopy_validate(push, dst, src),
"Memcopy validation failed in channel %s, GPU %s",
push->channel->name,
uvm_gpu_name(gpu));
gpu->parent->ce_hal->memcopy_patch_src(push, &src);
launch_dma_src_dst_type = gpu->parent->ce_hal->phys_mode(push, dst, src);
launch_dma_plc_mode = gpu->parent->ce_hal->plc_mode();
do {
NvU32 copy_this_time = (NvU32)min(size, max_single_copy_size);
if (first_operation && uvm_push_get_and_reset_flag(push, UVM_PUSH_FLAG_CE_NEXT_PIPELINED))
pipelined_value = HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, PIPELINED);
else
pipelined_value = HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, NON_PIPELINED);
gpu->parent->ce_hal->offset_in_out(push, src.address, dst.address);
NV_PUSH_1U(B0B5, LINE_LENGTH_IN, copy_this_time);
NV_PUSH_1U(B0B5, LAUNCH_DMA,
HWCONST(B0B5, LAUNCH_DMA, SRC_MEMORY_LAYOUT, PITCH) |
HWCONST(B0B5, LAUNCH_DMA, DST_MEMORY_LAYOUT, PITCH) |
HWCONST(B0B5, LAUNCH_DMA, MULTI_LINE_ENABLE, FALSE) |
HWCONST(B0B5, LAUNCH_DMA, REMAP_ENABLE, FALSE) |
HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, FALSE) |
launch_dma_src_dst_type |
launch_dma_plc_mode |
pipelined_value);
dst.address += copy_this_time;
src.address += copy_this_time;
size -= copy_this_time;
first_operation = false;
} while (size > 0);
maxwell_membar_after_transfer(push);
}
void uvm_hal_maxwell_ce_memcopy_v_to_v(uvm_push_t *push, NvU64 dst_va, NvU64 src_va, size_t size)
{
uvm_hal_maxwell_ce_memcopy(push, uvm_gpu_address_virtual(dst_va), uvm_gpu_address_virtual(src_va), size);
}
// Push SET_DST_PHYS mode if needed and return LAUNCH_DMA_DST_TYPE flags
static NvU32 memset_push_phys_mode(uvm_push_t *push, uvm_gpu_address_t dst)
{
if (dst.is_virtual)
return HWCONST(B0B5, LAUNCH_DMA, DST_TYPE, VIRTUAL);
NV_PUSH_1U(B0B5, SET_DST_PHYS_MODE, ce_aperture(dst.aperture));
return HWCONST(B0B5, LAUNCH_DMA, DST_TYPE, PHYSICAL);
}
static void memset_common(uvm_push_t *push, uvm_gpu_address_t dst, size_t size, size_t memset_element_size)
{
// If >4GB memsets ever become an important use case, this function should
// use multi-line transfers so we don't have to iterate (bug 1766588).
static const size_t max_single_memset_size = 0xFFFFFFFF;
uvm_gpu_t *gpu = uvm_push_get_gpu(push);
NvU32 pipelined_value;
NvU32 launch_dma_dst_type;
NvU32 launch_dma_plc_mode;
UVM_ASSERT_MSG(gpu->parent->ce_hal->memset_validate(push, dst, memset_element_size),
"Memset validation failed in channel %s, GPU %s",
push->channel->name,
uvm_gpu_name(gpu));
launch_dma_dst_type = memset_push_phys_mode(push, dst);
launch_dma_plc_mode = gpu->parent->ce_hal->plc_mode();
if (uvm_push_get_and_reset_flag(push, UVM_PUSH_FLAG_CE_NEXT_PIPELINED))
pipelined_value = HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, PIPELINED);
else
pipelined_value = HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, NON_PIPELINED);
do {
NvU32 memset_this_time = (NvU32)min(size, max_single_memset_size);
gpu->parent->ce_hal->offset_out(push, dst.address);
NV_PUSH_1U(B0B5, LINE_LENGTH_IN, memset_this_time);
NV_PUSH_1U(B0B5, LAUNCH_DMA,
HWCONST(B0B5, LAUNCH_DMA, SRC_MEMORY_LAYOUT, PITCH) |
HWCONST(B0B5, LAUNCH_DMA, DST_MEMORY_LAYOUT, PITCH) |
HWCONST(B0B5, LAUNCH_DMA, MULTI_LINE_ENABLE, FALSE) |
HWCONST(B0B5, LAUNCH_DMA, REMAP_ENABLE, TRUE) |
HWCONST(B0B5, LAUNCH_DMA, FLUSH_ENABLE, FALSE) |
launch_dma_dst_type |
launch_dma_plc_mode |
pipelined_value);
dst.address += memset_this_time * memset_element_size;
size -= memset_this_time;
pipelined_value = HWCONST(B0B5, LAUNCH_DMA, DATA_TRANSFER_TYPE, NON_PIPELINED);
} while (size > 0);
maxwell_membar_after_transfer(push);
}
void uvm_hal_maxwell_ce_memset_1(uvm_push_t *push, uvm_gpu_address_t dst, NvU8 value, size_t size)
{
NV_PUSH_2U(B0B5, SET_REMAP_CONST_B, (NvU32)value,
SET_REMAP_COMPONENTS,
HWCONST(B0B5, SET_REMAP_COMPONENTS, DST_X, CONST_B) |
HWCONST(B0B5, SET_REMAP_COMPONENTS, COMPONENT_SIZE, ONE) |
HWCONST(B0B5, SET_REMAP_COMPONENTS, NUM_DST_COMPONENTS, ONE));
memset_common(push, dst, size, 1);
}
void uvm_hal_maxwell_ce_memset_4(uvm_push_t *push, uvm_gpu_address_t dst, NvU32 value, size_t size)
{
UVM_ASSERT_MSG(size % 4 == 0, "size: %zd\n", size);
size /= 4;
NV_PUSH_2U(B0B5, SET_REMAP_CONST_B, value,
SET_REMAP_COMPONENTS,
HWCONST(B0B5, SET_REMAP_COMPONENTS, DST_X, CONST_B) |
HWCONST(B0B5, SET_REMAP_COMPONENTS, COMPONENT_SIZE, FOUR) |
HWCONST(B0B5, SET_REMAP_COMPONENTS, NUM_DST_COMPONENTS, ONE));
memset_common(push, dst, size, 4);
}
void uvm_hal_maxwell_ce_memset_8(uvm_push_t *push, uvm_gpu_address_t dst, NvU64 value, size_t size)
{
UVM_ASSERT_MSG(size % 8 == 0, "size: %zd\n", size);
size /= 8;
NV_PUSH_3U(B0B5, SET_REMAP_CONST_A, (NvU32)value,
SET_REMAP_CONST_B, (NvU32)(value >> 32),
SET_REMAP_COMPONENTS,
HWCONST(B0B5, SET_REMAP_COMPONENTS, DST_X, CONST_A) |
HWCONST(B0B5, SET_REMAP_COMPONENTS, DST_Y, CONST_B) |
HWCONST(B0B5, SET_REMAP_COMPONENTS, COMPONENT_SIZE, FOUR) |
HWCONST(B0B5, SET_REMAP_COMPONENTS, NUM_DST_COMPONENTS, TWO));
memset_common(push, dst, size, 8);
}
void uvm_hal_maxwell_ce_memset_v_4(uvm_push_t *push, NvU64 dst_va, NvU32 value, size_t size)
{
uvm_hal_maxwell_ce_memset_4(push, uvm_gpu_address_virtual(dst_va), value, size);
}
Reference in New Issue View Git Blame Copy Permalink