NVIDIA/open-gpu-kernel-modules
1
0
Fork 0
mirror of https://github.com/NVIDIA/open-gpu-kernel-modules.git synced 2026-01-31 21:49:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

560.28.03

Browse Source
This commit is contained in:
Gaurav Juvekar
2024-07-19 15:45:15 -07:00
parent 5fdf5032fb
commit 448d5cc656
859 changed files with 165424 additions and 91129 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/common/unix/common/inc/nv_amodel_enum.h
View File

@@ -37,6 +37,7 @@ typedef enum _NVAModelConfig {
NV_AMODEL_AMPERE,
NV_AMODEL_ADA,
NV_AMODEL_HOPPER,
NV_AMODEL_BLACKWELL,
} NVAModelConfig;
#endif /* __NV_AMODEL_ENUM_H__ */

1
src/common/unix/nvidia-3d/interface/nvidia-3d-types.h
View File

@@ -362,6 +362,7 @@ enum Nv3dShaderArch {
NV3D_SHADER_ARCH_TURING,
NV3D_SHADER_ARCH_AMPERE,
NV3D_SHADER_ARCH_HOPPER,
NV3D_SHADER_ARCH_BLACKWELL,
NV3D_SHADER_ARCH_COUNT,
};

8
src/common/unix/nvidia-3d/src/nvidia-3d-init.c
View File

@@ -38,6 +38,7 @@
#include "nvidia-push-init.h" // nvPushGetSupportedClassIndex()
#include "nvidia-push-utils.h" // nvPushIsAmodel()
#include <class/clcd97.h> // BLACKWELL_A
#include <class/clcb97.h> // HOPPER_A
#include <class/clc997.h> // ADA_A
#include <class/clc797.h> // AMPERE_B
@@ -58,6 +59,7 @@
#include "g_turing_shader_info.h"
#include "g_ampere_shader_info.h"
#include "g_hopper_shader_info.h"
#include "g_blackwell_shader_info.h"
#define _NV3D_CHANNEL_PROGRAMS_ENTRY(_archLower, _archCamel, _archUpper) \
[NV3D_SHADER_ARCH_ ## _archUpper ] = { \
@@ -92,6 +94,7 @@ static Nv3dChannelProgramsRec PickProgramsRec(
_NV3D_CHANNEL_PROGRAMS_ENTRY(turing, Turing, TURING),
_NV3D_CHANNEL_PROGRAMS_ENTRY(ampere, Ampere, AMPERE),
_NV3D_CHANNEL_PROGRAMS_ENTRY(hopper, Hopper, HOPPER),
_NV3D_CHANNEL_PROGRAMS_ENTRY(blackwell, Blackwell, BLACKWELL),
};
return programsTable[p3dDevice->shaderArch];
@@ -197,6 +200,7 @@ static NvU32 GetSmVersion(
[NV_AMODEL_AMPERE] = NV2080_CTRL_GR_INFO_SM_VERSION_8_2,
[NV_AMODEL_ADA] = NV2080_CTRL_GR_INFO_SM_VERSION_8_9,
[NV_AMODEL_HOPPER] = NV2080_CTRL_GR_INFO_SM_VERSION_9_0,
[NV_AMODEL_BLACKWELL] = NV2080_CTRL_GR_INFO_SM_VERSION_10_0,
};
if (pPushDevice->amodelConfig >= ARRAY_LEN(table)) {
@@ -292,6 +296,9 @@ static NvBool GetSpaVersion(
/* Hopper */
{ NV2080_CTRL_GR_INFO_SM_VERSION_9_0, { 9,0 } },
/* Blackwell */
{ NV2080_CTRL_GR_INFO_SM_VERSION_10_0, { 10,0 } },
{ NV2080_CTRL_GR_INFO_SM_VERSION_10_1, { 10,1 } },
};
const NvU32 smVersion = GetSmVersion(pPushDevice);
@@ -399,6 +406,7 @@ NvBool nv3dAllocDevice(
* classNumber | | | | | |
* | | | | | | |
*/
ENTRY(BLACKWELL_A,BLACKWELL, BLACKWELL, 0, 0, 32768, Hopper),
ENTRY(HOPPER_A, HOPPER, HOPPER, 0, 0, 32768, Hopper),
ENTRY(ADA_A, AMPERE, ADA, 0, 0, 32768, Ampere),
ENTRY(AMPERE_B, AMPERE, AMPERE, 0, 0, 32768, Ampere),

3
src/common/unix/nvidia-push/interface/nvidia-push-types.h
View File

@@ -1,5 +1,5 @@
/*
* SPDX-FileCopyrightText: Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-FileCopyrightText: Copyright (c) 2021-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a
@@ -108,6 +108,7 @@ typedef struct _NvPushDeviceRec {
NvBool hostLBoverflowBug1667921 : 1;
NvBool clientSli : 1; /* Provided by the host driver */
NvBool hasFb : 1; /* Computed from supportedClasses[] */
NvU32 clientHandle; /* Provided by the host driver */
NvU32 numSubDevices; /* Provided by the host driver */

61
src/common/unix/nvidia-push/src/nvidia-push-init.c
View File

@@ -1,5 +1,5 @@
/*
* SPDX-FileCopyrightText: Copyright (c) 1993-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-FileCopyrightText: Copyright (c) 1993-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a
@@ -46,6 +46,7 @@
#include "class/cl50a0.h" // NV50_MEMORY_VIRTUAL
#include "class/clc56f.h" // AMPERE_CHANNEL_GPFIFO_A
#include "class/clc86f.h" // HOPPER_CHANNEL_GPFIFO_A
#include "class/clc96f.h" // BLACKWELL_CHANNEL_GPFIFO_A
#include "class/clc361.h" // VOLTA_USERMODE_A
#include "class/clc661.h" // HOPPER_USERMODE_A
@@ -498,6 +499,7 @@ static NvBool nvDmaAllocUserD(
{
NvPushDevicePtr pDevice = p->pDevice;
int deviceIndex;
NvBool bHasFB = pDevice->hasFb;
if (!pDevice->hal.caps.clientAllocatesUserD) {
return TRUE;
@@ -507,15 +509,28 @@ static NvBool nvDmaAllocUserD(
deviceIndex < __nvPushGetNumDevices(pDevice);
deviceIndex++) {
NV_MEMORY_ALLOCATION_PARAMS memAllocParams = { 0 };
const NvU32 attr =
DRF_DEF(OS32, _ATTR, _LOCATION, _VIDMEM) |
DRF_DEF(OS32, _ATTR, _PAGE_SIZE, _4KB) |
DRF_DEF(OS32, _ATTR, _COHERENCY, _UNCACHED);
const NvU32 flags =
NVOS32_ALLOC_FLAGS_ALIGNMENT_FORCE |
NVOS32_ALLOC_FLAGS_PERSISTENT_VIDMEM;
NvU32 ret;
/* For GPUs which do not have framebuffer memory, use allocation from
* system memory instead.
*/
const NvU32 hClass = bHasFB ? NV01_MEMORY_LOCAL_USER : NV01_MEMORY_SYSTEM;
const NvU32 attr =
bHasFB ?
DRF_DEF(OS32, _ATTR, _LOCATION, _VIDMEM) |
DRF_DEF(OS32, _ATTR, _PAGE_SIZE, _4KB) |
DRF_DEF(OS32, _ATTR, _COHERENCY, _UNCACHED)
:
DRF_DEF(OS32, _ATTR, _LOCATION, _PCI) |
DRF_DEF(OS32, _ATTR, _PAGE_SIZE, _4KB) |
DRF_DEF(OS32, _ATTR, _COHERENCY, _UNCACHED);
const NvU32 flags =
bHasFB ?
NVOS32_ALLOC_FLAGS_ALIGNMENT_FORCE |
NVOS32_ALLOC_FLAGS_PERSISTENT_VIDMEM
:
NVOS32_ALLOC_FLAGS_ALIGNMENT_FORCE;
NvU32 hMemory = GetChannelHandle(pParams, pUsedHandleBitmask);
memAllocParams.owner = pDevice->clientHandle;
@@ -528,7 +543,7 @@ static NvBool nvDmaAllocUserD(
ret = nvPushImportRmApiAlloc(pDevice,
pDevice->subDevice[deviceIndex].deviceHandle,
hMemory,
NV01_MEMORY_LOCAL_USER,
hClass,
&memAllocParams);
if (ret != NV_OK) {
return FALSE;
@@ -540,13 +555,26 @@ static NvBool nvDmaAllocUserD(
return TRUE;
}
static NvBool IsClassSupported(
const NvPushDeviceRec *pDevice,
NvU32 classNumber)
{
unsigned int j;
for (j = 0; j < pDevice->numClasses; j++) {
if (classNumber == pDevice->supportedClasses[j]) {
return TRUE;
}
}
return FALSE;
}
int nvPushGetSupportedClassIndex(
NvPushDevicePtr pDevice,
const void *pClassTable,
size_t classTableStride,
size_t classTableLength)
{
unsigned int i, j;
unsigned int i;
for (i = 0; i < classTableLength; i++) {
@@ -562,10 +590,8 @@ int nvPushGetSupportedClassIndex(
continue;
}
for (j = 0; j < pDevice->numClasses; j++) {
if (pClass->classNumber == pDevice->supportedClasses[j]) {
return i;
}
if (IsClassSupported(pDevice, pClass->classNumber)) {
return i;
}
}
return -1;
@@ -579,6 +605,11 @@ static NvBool GetChannelClassAndUserDSize(
NvPushSupportedClass base;
size_t gpFifoSize;
} gpFifoDmaClasses[] = {
{
{ BLACKWELL_CHANNEL_GPFIFO_A,
NV_AMODEL_BLACKWELL },
sizeof(BlackwellAControlGPFifo)
},
{
{ HOPPER_CHANNEL_GPFIFO_A,
NV_AMODEL_HOPPER },
@@ -1485,6 +1516,8 @@ NvBool nvPushAllocDevice(
pDevice->numClasses = pParams->numClasses;
pDevice->supportedClasses = pParams->supportedClasses;
pDevice->hasFb = IsClassSupported(pDevice, NV01_MEMORY_LOCAL_USER);
pDevice->confidentialComputeMode = pParams->confidentialComputeMode;
for (sd = 0; sd < pParams->numSubDevices; sd++) {

4
src/common/unix/nvidia-push/src/nvidia-push.c
View File

@@ -45,6 +45,7 @@
#include "class/clc56f.h" // AMPERE_CHANNEL_GPFIFO_A
#include "class/clc86f.h" // HOPPER_CHANNEL_GPFIFO_A
#include "class/clc361.h" // VOLTA_USERMODE_A
#include "class/clc96f.h" // BLACKWELL_CHANNEL_GPFIFO_A
#include "ctrl/ctrl906f.h" // NV906F_CTRL_GET_CLASS_ENGINEID
/*
@@ -1097,6 +1098,9 @@ NvBool __nvPushGetHal(
NvPushHal *pHal)
{
switch (channelClass) {
case BLACKWELL_CHANNEL_GPFIFO_A:
// backwards compatible with Hopper
// fall through
case HOPPER_CHANNEL_GPFIFO_A:
pHal->caps.extendedBase = TRUE;
// otherwise backwards compatible with the Volta DMA HAL