535.183.01

This commit is contained in:
Bernhard Stoeckner
2024-06-04 10:45:14 +02:00
parent f4bdce9a0a
commit 4459285b60
43 changed files with 2447 additions and 1544 deletions

View File

@@ -340,6 +340,7 @@
#define NV_MSGBOX_CMD_ARG1_ECC_V6_ERROR_TYPE 15:8
#define NV_MSGBOX_CMD_ARG1_ECC_V6_ERROR_TYPE_CORRECTABLE_ERROR 0
#define NV_MSGBOX_CMD_ARG1_ECC_V6_ERROR_TYPE_UNCORRECTABLE_ERROR 1
#define NV_MSGBOX_CMD_ARG1_ECC_V6_ERROR_TYPE_ECC_STATE_FLAGS 2
#define NV_MSGBOX_CMD_ARG1_ENERGY_COUNTER_GPU 0x00000000
#define NV_MSGBOX_CMD_ARG1_ENERGY_COUNTER_MODULE 0x00000003
@@ -633,7 +634,7 @@
/* MSGBOX data, capability dword structure */
#define NV_MSGBOX_DATA_REG 31:0
#define NV_MSGBOX_DATA_CAP_COUNT 5
#define NV_MSGBOX_DATA_CAP_COUNT 6
#define NV_MSGBOX_DATA_CAP_0 0
#define NV_MSGBOX_DATA_CAP_0_TEMP_GPU_0 0:0
@@ -950,6 +951,10 @@
#define NV_MSGBOX_DATA_CAP_4_CONFIGURE_PROGRAMMABLE_EDPP_NOT_AVAILABLE 0x00000000
#define NV_MSGBOX_DATA_CAP_4_CONFIGURE_PROGRAMMABLE_EDPP_AVAILABLE 0x00000001
#define NV_MSGBOX_DATA_CAP_5_SRAM_ERROR_THRESHOLD_EXCEEDED 9:9
#define NV_MSGBOX_DATA_CAP_5_SRAM_ERROR_THRESHOLD_EXCEEDED_NOT_AVAILABLE 0x00000000
#define NV_MSGBOX_DATA_CAP_5_SRAM_ERROR_THRESHOLD_EXCEEDED_AVAILABLE 0x00000001
/* ECC counters */
#define NV_MSGBOX_DATA_ECC_CNT_16BIT_DBE 31:16
#define NV_MSGBOX_DATA_ECC_CNT_16BIT_SBE 16:0
@@ -984,6 +989,13 @@
#define NV_MSGBOX_DATA_ECC_V5_METADATA_LOCATION_ID 26:22
#define NV_MSGBOX_DATA_ECC_V5_METADATA_SUBLOCATION_ID 31:27
/* ECC state flags */
#define NV_MSGBOX_DATA_ECC_V6_STATE_FLAGS 31:0
#define NV_MSGBOX_DATA_ECC_V6_STATE_FLAGS_SRAM_ERROR_THRESHOLD_EXCEEDED 0:0
#define NV_MSGBOX_DATA_ECC_V6_STATE_FLAGS_SRAM_ERROR_THRESHOLD_EXCEEDED_FALSE 0
#define NV_MSGBOX_DATA_ECC_V6_STATE_FLAGS_SRAM_ERROR_THRESHOLD_EXCEEDED_TRUE 1
/* NV_MSGBOX_CMD_OPCODE_SCRATCH_COPY src offset argument */
#define NV_MSGBOX_DATA_COPY_SRC_OFFSET 7:0

View File

@@ -124,6 +124,16 @@ static void __nvoc_init_funcTable_CrashCatReport_1(CrashCatReport *pThis) {
PORT_UNREFERENCED_VARIABLE(reportHal);
PORT_UNREFERENCED_VARIABLE(reportHal_HalVarIdx);
// Hal function -- crashcatReportSourceContainment
if (( ((reportHal_HalVarIdx >> 5) == 0UL) && ((1UL << (reportHal_HalVarIdx & 0x1f)) & 0x00000004UL) )) /* CrashCatReportHal: V1_LIBOS3 */
{
pThis->__crashcatReportSourceContainment__ = &crashcatReportSourceContainment_V1_LIBOS3;
}
else
{
pThis->__crashcatReportSourceContainment__ = &crashcatReportSourceContainment_3e9f29;
}
// Hal function -- crashcatReportLogReporter
if (( ((reportHal_HalVarIdx >> 5) == 0UL) && ((1UL << (reportHal_HalVarIdx & 0x1f)) & 0x00000002UL) )) /* CrashCatReportHal: V1_LIBOS2 */
{

View File

@@ -82,6 +82,7 @@ struct CrashCatReport {
struct Object __nvoc_base_Object;
struct Object *__nvoc_pbase_Object;
struct CrashCatReport *__nvoc_pbase_CrashCatReport;
NV_CRASHCAT_CONTAINMENT (*__crashcatReportSourceContainment__)(struct CrashCatReport *);
void (*__crashcatReportLogReporter__)(struct CrashCatReport *);
void (*__crashcatReportLogSource__)(struct CrashCatReport *);
struct CrashCatReportHal reportHal;
@@ -119,6 +120,8 @@ NV_STATUS __nvoc_objCreate_CrashCatReport(CrashCatReport**, Dynamic*, NvU32,
#define __objCreate_CrashCatReport(ppNewObj, pParent, createFlags, CrashCatReportHal_version, CrashCatReportHal_implementer, arg_ppReportBytes, arg_bytesRemaining) \
__nvoc_objCreate_CrashCatReport((ppNewObj), staticCast((pParent), Dynamic), (createFlags), CrashCatReportHal_version, CrashCatReportHal_implementer, arg_ppReportBytes, arg_bytesRemaining)
#define crashcatReportSourceContainment(arg0) crashcatReportSourceContainment_DISPATCH(arg0)
#define crashcatReportSourceContainment_HAL(arg0) crashcatReportSourceContainment_DISPATCH(arg0)
#define crashcatReportLogReporter(arg0) crashcatReportLogReporter_DISPATCH(arg0)
#define crashcatReportLogReporter_HAL(arg0) crashcatReportLogReporter_DISPATCH(arg0)
#define crashcatReportLogSource(arg0) crashcatReportLogSource_DISPATCH(arg0)
@@ -263,6 +266,16 @@ static inline void crashcatReportLogIo32State(struct CrashCatReport *arg0) {
#define crashcatReportLogIo32State_HAL(arg0) crashcatReportLogIo32State(arg0)
static inline NV_CRASHCAT_CONTAINMENT crashcatReportSourceContainment_3e9f29(struct CrashCatReport *arg0) {
return NV_CRASHCAT_CONTAINMENT_UNSPECIFIED;
}
NV_CRASHCAT_CONTAINMENT crashcatReportSourceContainment_V1_LIBOS3(struct CrashCatReport *arg0);
static inline NV_CRASHCAT_CONTAINMENT crashcatReportSourceContainment_DISPATCH(struct CrashCatReport *arg0) {
return arg0->__crashcatReportSourceContainment__(arg0);
}
void crashcatReportLogReporter_V1_GENERIC(struct CrashCatReport *arg0);
void crashcatReportLogReporter_V1_LIBOS2(struct CrashCatReport *arg0);

View File

@@ -973,6 +973,7 @@ struct OBJGPU {
NvBool PDB_PROP_GPU_SKIP_CE_MAPPINGS_NO_NVLINK;
NvBool PDB_PROP_GPU_C2C_SYSMEM;
NvBool PDB_PROP_GPU_IN_TCC_MODE;
NvBool PDB_PROP_GPU_SUPPORTS_TDR_EVENT;
NvBool PDB_PROP_GPU_MSHYBRID_GC6_ACTIVE;
NvBool PDB_PROP_GPU_VGPU_BIG_PAGE_SIZE_64K;
NvBool PDB_PROP_GPU_OPTIMIZE_SPARSE_TEXTURE_BY_DEFAULT;
@@ -1306,6 +1307,8 @@ extern const struct NVOC_CLASS_DEF __nvoc_class_def_OBJGPU;
#define PDB_PROP_GPU_IN_PM_CODEPATH_BASE_NAME PDB_PROP_GPU_IN_PM_CODEPATH
#define PDB_PROP_GPU_ALLOW_PAGE_RETIREMENT_BASE_CAST
#define PDB_PROP_GPU_ALLOW_PAGE_RETIREMENT_BASE_NAME PDB_PROP_GPU_ALLOW_PAGE_RETIREMENT
#define PDB_PROP_GPU_SUPPORTS_TDR_EVENT_BASE_CAST
#define PDB_PROP_GPU_SUPPORTS_TDR_EVENT_BASE_NAME PDB_PROP_GPU_SUPPORTS_TDR_EVENT
#define PDB_PROP_GPU_UPSTREAM_PORT_L1_UNSUPPORTED_BASE_CAST
#define PDB_PROP_GPU_UPSTREAM_PORT_L1_UNSUPPORTED_BASE_NAME PDB_PROP_GPU_UPSTREAM_PORT_L1_UNSUPPORTED
#define PDB_PROP_GPU_BEHIND_BR03_BASE_CAST

View File

@@ -349,6 +349,28 @@ static void __nvoc_init_funcTable_KernelFifo_1(KernelFifo *pThis, RmHalspecOwner
pThis->__kfifoGetMaxCeChannelGroups__ = &kfifoGetMaxCeChannelGroups_GA100;
}
// Hal function -- kfifoStartChannelHalt
if (( ((chipHal_HalVarIdx >> 5) == 1UL) && ((1UL << (chipHal_HalVarIdx & 0x1f)) & 0x11f0fc00UL) )) /* ChipHal: GA100 | GA102 | GA103 | GA104 | GA106 | GA107 | AD102 | AD103 | AD104 | AD106 | AD107 | GH100 */
{
pThis->__kfifoStartChannelHalt__ = &kfifoStartChannelHalt_GA100;
}
// default
else
{
pThis->__kfifoStartChannelHalt__ = &kfifoStartChannelHalt_b3696a;
}
// Hal function -- kfifoCompleteChannelHalt
if (( ((chipHal_HalVarIdx >> 5) == 1UL) && ((1UL << (chipHal_HalVarIdx & 0x1f)) & 0x11f0fc00UL) )) /* ChipHal: GA100 | GA102 | GA103 | GA104 | GA106 | GA107 | AD102 | AD103 | AD104 | AD106 | AD107 | GH100 */
{
pThis->__kfifoCompleteChannelHalt__ = &kfifoCompleteChannelHalt_GA100;
}
// default
else
{
pThis->__kfifoCompleteChannelHalt__ = &kfifoCompleteChannelHalt_b3696a;
}
pThis->__nvoc_base_OBJENGSTATE.__engstateConstructEngine__ = &__nvoc_thunk_KernelFifo_engstateConstructEngine;
pThis->__nvoc_base_OBJENGSTATE.__engstateStateInitLocked__ = &__nvoc_thunk_KernelFifo_engstateStateInitLocked;

View File

@@ -41,6 +41,7 @@ extern "C" {
\***************************************************************************/
#include "kernel/gpu/eng_state.h"
#include "kernel/gpu/gpu_timeout.h"
#include "kernel/gpu/gpu_halspec.h"
#include "kernel/gpu/fifo/channel_descendant.h"
#include "kernel/gpu/gpu_engine_type.h"
@@ -493,6 +494,8 @@ struct KernelFifo {
NV_STATUS (*__kfifoUpdateUsermodeDoorbell__)(struct OBJGPU *, struct KernelFifo *, NvU32, NvU32);
NvU32 (*__kfifoRunlistGetBaseShift__)(struct KernelFifo *);
NvU32 (*__kfifoGetMaxCeChannelGroups__)(struct OBJGPU *, struct KernelFifo *);
void (*__kfifoStartChannelHalt__)(struct OBJGPU *, struct KernelFifo *, struct KernelChannel *);
void (*__kfifoCompleteChannelHalt__)(struct OBJGPU *, struct KernelFifo *, struct KernelChannel *, RMTIMEOUT *);
NV_STATUS (*__kfifoStateLoad__)(POBJGPU, struct KernelFifo *, NvU32);
NV_STATUS (*__kfifoStateUnload__)(POBJGPU, struct KernelFifo *, NvU32);
NV_STATUS (*__kfifoStatePreLoad__)(POBJGPU, struct KernelFifo *, NvU32);
@@ -595,6 +598,10 @@ NV_STATUS __nvoc_objCreate_KernelFifo(KernelFifo**, Dynamic*, NvU32);
#define kfifoRunlistGetBaseShift_HAL(pKernelFifo) kfifoRunlistGetBaseShift_DISPATCH(pKernelFifo)
#define kfifoGetMaxCeChannelGroups(pGpu, pKernelFifo) kfifoGetMaxCeChannelGroups_DISPATCH(pGpu, pKernelFifo)
#define kfifoGetMaxCeChannelGroups_HAL(pGpu, pKernelFifo) kfifoGetMaxCeChannelGroups_DISPATCH(pGpu, pKernelFifo)
#define kfifoStartChannelHalt(pGpu, pKernelFifo, pKernelChannel) kfifoStartChannelHalt_DISPATCH(pGpu, pKernelFifo, pKernelChannel)
#define kfifoStartChannelHalt_HAL(pGpu, pKernelFifo, pKernelChannel) kfifoStartChannelHalt_DISPATCH(pGpu, pKernelFifo, pKernelChannel)
#define kfifoCompleteChannelHalt(pGpu, pKernelFifo, pKernelChannel, pTimeout) kfifoCompleteChannelHalt_DISPATCH(pGpu, pKernelFifo, pKernelChannel, pTimeout)
#define kfifoCompleteChannelHalt_HAL(pGpu, pKernelFifo, pKernelChannel, pTimeout) kfifoCompleteChannelHalt_DISPATCH(pGpu, pKernelFifo, pKernelChannel, pTimeout)
#define kfifoStateLoad(pGpu, pEngstate, arg0) kfifoStateLoad_DISPATCH(pGpu, pEngstate, arg0)
#define kfifoStateUnload(pGpu, pEngstate, arg0) kfifoStateUnload_DISPATCH(pGpu, pEngstate, arg0)
#define kfifoStatePreLoad(pGpu, pEngstate, arg0) kfifoStatePreLoad_DISPATCH(pGpu, pEngstate, arg0)
@@ -1474,6 +1481,26 @@ static inline NvU32 kfifoGetMaxCeChannelGroups_DISPATCH(struct OBJGPU *pGpu, str
return pKernelFifo->__kfifoGetMaxCeChannelGroups__(pGpu, pKernelFifo);
}
void kfifoStartChannelHalt_GA100(struct OBJGPU *pGpu, struct KernelFifo *pKernelFifo, struct KernelChannel *pKernelChannel);
static inline void kfifoStartChannelHalt_b3696a(struct OBJGPU *pGpu, struct KernelFifo *pKernelFifo, struct KernelChannel *pKernelChannel) {
return;
}
static inline void kfifoStartChannelHalt_DISPATCH(struct OBJGPU *pGpu, struct KernelFifo *pKernelFifo, struct KernelChannel *pKernelChannel) {
pKernelFifo->__kfifoStartChannelHalt__(pGpu, pKernelFifo, pKernelChannel);
}
void kfifoCompleteChannelHalt_GA100(struct OBJGPU *pGpu, struct KernelFifo *pKernelFifo, struct KernelChannel *pKernelChannel, RMTIMEOUT *pTimeout);
static inline void kfifoCompleteChannelHalt_b3696a(struct OBJGPU *pGpu, struct KernelFifo *pKernelFifo, struct KernelChannel *pKernelChannel, RMTIMEOUT *pTimeout) {
return;
}
static inline void kfifoCompleteChannelHalt_DISPATCH(struct OBJGPU *pGpu, struct KernelFifo *pKernelFifo, struct KernelChannel *pKernelChannel, RMTIMEOUT *pTimeout) {
pKernelFifo->__kfifoCompleteChannelHalt__(pGpu, pKernelFifo, pKernelChannel, pTimeout);
}
static inline NV_STATUS kfifoStateLoad_DISPATCH(POBJGPU pGpu, struct KernelFifo *pEngstate, NvU32 arg0) {
return pEngstate->__kfifoStateLoad__(pGpu, pEngstate, arg0);
}

View File

@@ -7,7 +7,7 @@ extern "C" {
#endif
/*
* SPDX-FileCopyrightText: Copyright (c) 2020-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-FileCopyrightText: Copyright (c) 2020-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a
@@ -343,7 +343,7 @@ static inline void kgrctxDecObjectCount(struct OBJGPU *arg0, struct KernelGraphi
#define kgrctxDecObjectCount_HAL(arg0, arg1, classNum) kgrctxDecObjectCount(arg0, arg1, classNum)
GR_GLOBALCTX_BUFFER kgrctxGetRegisterAccessMapId_PF(struct OBJGPU *arg0, struct KernelGraphicsContext *arg1, struct KernelChannel *arg2);
GR_GLOBALCTX_BUFFER kgrctxGetRegisterAccessMapId_IMPL(struct OBJGPU *arg0, struct KernelGraphicsContext *arg1, struct KernelChannel *arg2);
#ifdef __nvoc_kernel_graphics_context_h_disabled
@@ -354,7 +354,7 @@ static inline GR_GLOBALCTX_BUFFER kgrctxGetRegisterAccessMapId(struct OBJGPU *ar
return ret;
}
#else //__nvoc_kernel_graphics_context_h_disabled
#define kgrctxGetRegisterAccessMapId(arg0, arg1, arg2) kgrctxGetRegisterAccessMapId_PF(arg0, arg1, arg2)
#define kgrctxGetRegisterAccessMapId(arg0, arg1, arg2) kgrctxGetRegisterAccessMapId_IMPL(arg0, arg1, arg2)
#endif //__nvoc_kernel_graphics_context_h_disabled
#define kgrctxGetRegisterAccessMapId_HAL(arg0, arg1, arg2) kgrctxGetRegisterAccessMapId(arg0, arg1, arg2)

View File

@@ -223,6 +223,20 @@ void __nvoc_init_dataField_KernelGraphics(KernelGraphics *pThis, RmHalspecOwner
pThis->bFecsRecordUcodeSeqnoSupported = ((NvBool)(0 != 0));
}
}
// Hal field -- bBug4208224WAREnabled
if (( ((rmVariantHal_HalVarIdx >> 5) == 0UL) && ((1UL << (rmVariantHal_HalVarIdx & 0x1f)) & 0x00000002UL) )) /* RmVariantHal: PF_KERNEL_ONLY */
{
if (( ((chipHal_HalVarIdx >> 5) == 1UL) && ((1UL << (chipHal_HalVarIdx & 0x1f)) & 0x000000e0UL) )) /* ChipHal: TU102 | TU104 | TU106 */
{
pThis->bBug4208224WAREnabled = ((NvBool)(0 == 0));
}
// default
else
{
pThis->bBug4208224WAREnabled = ((NvBool)(0 != 0));
}
}
}
NV_STATUS __nvoc_ctor_OBJENGSTATE(OBJENGSTATE* );
@@ -276,6 +290,39 @@ static void __nvoc_init_funcTable_KernelGraphics_1(KernelGraphics *pThis, RmHals
pThis->__kgraphicsServiceNotificationInterrupt__ = &kgraphicsServiceNotificationInterrupt_IMPL;
// Hal function -- kgraphicsCreateBug4208224Channel
if (( ((chipHal_HalVarIdx >> 5) == 1UL) && ((1UL << (chipHal_HalVarIdx & 0x1f)) & 0x000000e0UL) )) /* ChipHal: TU102 | TU104 | TU106 */
{
pThis->__kgraphicsCreateBug4208224Channel__ = &kgraphicsCreateBug4208224Channel_TU102;
}
// default
else
{
pThis->__kgraphicsCreateBug4208224Channel__ = &kgraphicsCreateBug4208224Channel_56cd7a;
}
// Hal function -- kgraphicsInitializeBug4208224WAR
if (( ((chipHal_HalVarIdx >> 5) == 1UL) && ((1UL << (chipHal_HalVarIdx & 0x1f)) & 0x000000e0UL) )) /* ChipHal: TU102 | TU104 | TU106 */
{
pThis->__kgraphicsInitializeBug4208224WAR__ = &kgraphicsInitializeBug4208224WAR_TU102;
}
// default
else
{
pThis->__kgraphicsInitializeBug4208224WAR__ = &kgraphicsInitializeBug4208224WAR_56cd7a;
}
// Hal function -- kgraphicsIsBug4208224WARNeeded
if (( ((chipHal_HalVarIdx >> 5) == 1UL) && ((1UL << (chipHal_HalVarIdx & 0x1f)) & 0x000000e0UL) )) /* ChipHal: TU102 | TU104 | TU106 */
{
pThis->__kgraphicsIsBug4208224WARNeeded__ = &kgraphicsIsBug4208224WARNeeded_TU102;
}
// default
else
{
pThis->__kgraphicsIsBug4208224WARNeeded__ = &kgraphicsIsBug4208224WARNeeded_491d52;
}
// Hal function -- kgraphicsClearInterrupt
pThis->__kgraphicsClearInterrupt__ = &kgraphicsClearInterrupt_GP100;

View File

@@ -53,6 +53,7 @@ struct KGRAPHICS_STATIC_INFO;
typedef struct KGRAPHICS_STATIC_INFO KGRAPHICS_STATIC_INFO;
typedef struct KGRAPHICS_FECS_TRACE_INFO KGRAPHICS_FECS_TRACE_INFO;
typedef struct KGRAPHICS_GLOBAL_CTX_BUFFERS_INFO KGRAPHICS_GLOBAL_CTX_BUFFERS_INFO;
typedef struct KGRAPHICS_BUG4208224_CONTEXT_INFO KGRAPHICS_BUG4208224_CONTEXT_INFO;
/*!
* Static info retrieved from Physical RM detailing the configuration of the
@@ -135,6 +136,24 @@ struct KGRAPHICS_GLOBAL_CTX_BUFFERS_INFO
GR_BUFFER_ATTR vfGlobalCtxAttr[GR_GLOBALCTX_BUFFER_COUNT];
};
#define KGRAPHICS_SCRUBBER_HANDLE_VAS 0xdada0042
#define KGRAPHICS_SCRUBBER_HANDLE_PBVIRT (KGRAPHICS_SCRUBBER_HANDLE_VAS + 1)
#define KGRAPHICS_SCRUBBER_HANDLE_PBPHYS (KGRAPHICS_SCRUBBER_HANDLE_VAS + 2)
#define KGRAPHICS_SCRUBBER_HANDLE_CHANNEL (KGRAPHICS_SCRUBBER_HANDLE_VAS + 3)
#define KGRAPHICS_SCRUBBER_HANDLE_3DOBJ (KGRAPHICS_SCRUBBER_HANDLE_VAS + 4)
#define KGRAPHICS_SCRUBBER_HANDLE_USERD (KGRAPHICS_SCRUBBER_HANDLE_VAS + 5)
struct KGRAPHICS_BUG4208224_CONTEXT_INFO
{
/* Dynamically allocated client handles */
NvHandle hClient;
NvHandle hDeviceId;
NvHandle hSubdeviceId;
// Have resources been setup
NvBool bConstructed;
};
// Opaque forward declarations
typedef struct KGRAPHICS_PRIVATE_DATA KGRAPHICS_PRIVATE_DATA;
typedef struct KGRAPHICS_FECS_TRACE_INFO KGRAPHICS_FECS_TRACE_INFO;
@@ -168,6 +187,9 @@ struct KernelGraphics {
NV_STATUS (*__kgraphicsStatePostLoad__)(OBJGPU *, struct KernelGraphics *, NvU32);
void (*__kgraphicsRegisterIntrService__)(OBJGPU *, struct KernelGraphics *, IntrServiceRecord *);
NV_STATUS (*__kgraphicsServiceNotificationInterrupt__)(OBJGPU *, struct KernelGraphics *, IntrServiceServiceNotificationInterruptArguments *);
NV_STATUS (*__kgraphicsCreateBug4208224Channel__)(OBJGPU *, struct KernelGraphics *);
NV_STATUS (*__kgraphicsInitializeBug4208224WAR__)(OBJGPU *, struct KernelGraphics *);
NvBool (*__kgraphicsIsBug4208224WARNeeded__)(OBJGPU *, struct KernelGraphics *);
NvBool (*__kgraphicsClearInterrupt__)(OBJGPU *, struct KernelGraphics *, IntrServiceClearInterruptArguments *);
NvU32 (*__kgraphicsServiceInterrupt__)(OBJGPU *, struct KernelGraphics *, IntrServiceServiceInterruptArguments *);
NV_STATUS (*__kgraphicsStatePreLoad__)(POBJGPU, struct KernelGraphics *, NvU32);
@@ -185,6 +207,7 @@ struct KernelGraphics {
NvBool PRIVATE_FIELD(bUcodeSupportsPrivAccessMap);
NvBool PRIVATE_FIELD(bRtvCbSupported);
NvBool PRIVATE_FIELD(bFecsRecordUcodeSeqnoSupported);
NvBool PRIVATE_FIELD(bBug4208224WAREnabled);
NvU32 PRIVATE_FIELD(instance);
KGRAPHICS_PRIVATE_DATA *PRIVATE_FIELD(pPrivate);
NvBool PRIVATE_FIELD(bCollectingDeferredStaticData);
@@ -193,6 +216,7 @@ struct KernelGraphics {
struct CTX_BUF_POOL_INFO *PRIVATE_FIELD(pCtxBufPool);
CTX_BUF_INFO PRIVATE_FIELD(maxCtxBufSize)[9];
GR_BUFFER_ATTR PRIVATE_FIELD(ctxAttr)[9];
struct KGRAPHICS_BUG4208224_CONTEXT_INFO PRIVATE_FIELD(bug4208224Info);
};
struct KernelGraphics_PRIVATE {
const struct NVOC_RTTI *__nvoc_rtti;
@@ -212,6 +236,9 @@ struct KernelGraphics_PRIVATE {
NV_STATUS (*__kgraphicsStatePostLoad__)(OBJGPU *, struct KernelGraphics *, NvU32);
void (*__kgraphicsRegisterIntrService__)(OBJGPU *, struct KernelGraphics *, IntrServiceRecord *);
NV_STATUS (*__kgraphicsServiceNotificationInterrupt__)(OBJGPU *, struct KernelGraphics *, IntrServiceServiceNotificationInterruptArguments *);
NV_STATUS (*__kgraphicsCreateBug4208224Channel__)(OBJGPU *, struct KernelGraphics *);
NV_STATUS (*__kgraphicsInitializeBug4208224WAR__)(OBJGPU *, struct KernelGraphics *);
NvBool (*__kgraphicsIsBug4208224WARNeeded__)(OBJGPU *, struct KernelGraphics *);
NvBool (*__kgraphicsClearInterrupt__)(OBJGPU *, struct KernelGraphics *, IntrServiceClearInterruptArguments *);
NvU32 (*__kgraphicsServiceInterrupt__)(OBJGPU *, struct KernelGraphics *, IntrServiceServiceInterruptArguments *);
NV_STATUS (*__kgraphicsStatePreLoad__)(POBJGPU, struct KernelGraphics *, NvU32);
@@ -229,6 +256,7 @@ struct KernelGraphics_PRIVATE {
NvBool bUcodeSupportsPrivAccessMap;
NvBool bRtvCbSupported;
NvBool bFecsRecordUcodeSeqnoSupported;
NvBool bBug4208224WAREnabled;
NvU32 instance;
KGRAPHICS_PRIVATE_DATA *pPrivate;
NvBool bCollectingDeferredStaticData;
@@ -237,6 +265,7 @@ struct KernelGraphics_PRIVATE {
struct CTX_BUF_POOL_INFO *pCtxBufPool;
CTX_BUF_INFO maxCtxBufSize[9];
GR_BUFFER_ATTR ctxAttr[9];
struct KGRAPHICS_BUG4208224_CONTEXT_INFO bug4208224Info;
};
#ifndef __NVOC_CLASS_KernelGraphics_TYPEDEF__
@@ -279,6 +308,12 @@ NV_STATUS __nvoc_objCreate_KernelGraphics(KernelGraphics**, Dynamic*, NvU32);
#define kgraphicsStatePostLoad(arg0, arg1, flags) kgraphicsStatePostLoad_DISPATCH(arg0, arg1, flags)
#define kgraphicsRegisterIntrService(arg0, arg1, arg2) kgraphicsRegisterIntrService_DISPATCH(arg0, arg1, arg2)
#define kgraphicsServiceNotificationInterrupt(arg0, arg1, arg2) kgraphicsServiceNotificationInterrupt_DISPATCH(arg0, arg1, arg2)
#define kgraphicsCreateBug4208224Channel(arg0, arg1) kgraphicsCreateBug4208224Channel_DISPATCH(arg0, arg1)
#define kgraphicsCreateBug4208224Channel_HAL(arg0, arg1) kgraphicsCreateBug4208224Channel_DISPATCH(arg0, arg1)
#define kgraphicsInitializeBug4208224WAR(arg0, arg1) kgraphicsInitializeBug4208224WAR_DISPATCH(arg0, arg1)
#define kgraphicsInitializeBug4208224WAR_HAL(arg0, arg1) kgraphicsInitializeBug4208224WAR_DISPATCH(arg0, arg1)
#define kgraphicsIsBug4208224WARNeeded(arg0, arg1) kgraphicsIsBug4208224WARNeeded_DISPATCH(arg0, arg1)
#define kgraphicsIsBug4208224WARNeeded_HAL(arg0, arg1) kgraphicsIsBug4208224WARNeeded_DISPATCH(arg0, arg1)
#define kgraphicsClearInterrupt(arg0, arg1, arg2) kgraphicsClearInterrupt_DISPATCH(arg0, arg1, arg2)
#define kgraphicsClearInterrupt_HAL(arg0, arg1, arg2) kgraphicsClearInterrupt_DISPATCH(arg0, arg1, arg2)
#define kgraphicsServiceInterrupt(arg0, arg1, arg2) kgraphicsServiceInterrupt_DISPATCH(arg0, arg1, arg2)
@@ -463,6 +498,36 @@ static inline NV_STATUS kgraphicsServiceNotificationInterrupt_DISPATCH(OBJGPU *a
return arg1->__kgraphicsServiceNotificationInterrupt__(arg0, arg1, arg2);
}
NV_STATUS kgraphicsCreateBug4208224Channel_TU102(OBJGPU *arg0, struct KernelGraphics *arg1);
static inline NV_STATUS kgraphicsCreateBug4208224Channel_56cd7a(OBJGPU *arg0, struct KernelGraphics *arg1) {
return NV_OK;
}
static inline NV_STATUS kgraphicsCreateBug4208224Channel_DISPATCH(OBJGPU *arg0, struct KernelGraphics *arg1) {
return arg1->__kgraphicsCreateBug4208224Channel__(arg0, arg1);
}
static inline NV_STATUS kgraphicsInitializeBug4208224WAR_56cd7a(OBJGPU *arg0, struct KernelGraphics *arg1) {
return NV_OK;
}
NV_STATUS kgraphicsInitializeBug4208224WAR_TU102(OBJGPU *arg0, struct KernelGraphics *arg1);
static inline NV_STATUS kgraphicsInitializeBug4208224WAR_DISPATCH(OBJGPU *arg0, struct KernelGraphics *arg1) {
return arg1->__kgraphicsInitializeBug4208224WAR__(arg0, arg1);
}
static inline NvBool kgraphicsIsBug4208224WARNeeded_491d52(OBJGPU *arg0, struct KernelGraphics *arg1) {
return ((NvBool)(0 != 0));
}
NvBool kgraphicsIsBug4208224WARNeeded_TU102(OBJGPU *arg0, struct KernelGraphics *arg1);
static inline NvBool kgraphicsIsBug4208224WARNeeded_DISPATCH(OBJGPU *arg0, struct KernelGraphics *arg1) {
return arg1->__kgraphicsIsBug4208224WARNeeded__(arg0, arg1);
}
NvBool kgraphicsClearInterrupt_GP100(OBJGPU *arg0, struct KernelGraphics *arg1, IntrServiceClearInterruptArguments *arg2);
static inline NvBool kgraphicsClearInterrupt_DISPATCH(OBJGPU *arg0, struct KernelGraphics *arg1, IntrServiceClearInterruptArguments *arg2) {
@@ -574,6 +639,16 @@ static inline NvBool kgraphicsIsFecsRecordUcodeSeqnoSupported(OBJGPU *pGpu, stru
return pKernelGraphics_PRIVATE->bFecsRecordUcodeSeqnoSupported;
}
static inline NvBool kgraphicsGetBug4208224WAREnabled(OBJGPU *pGpu, struct KernelGraphics *pKernelGraphics) {
struct KernelGraphics_PRIVATE *pKernelGraphics_PRIVATE = (struct KernelGraphics_PRIVATE *)pKernelGraphics;
return pKernelGraphics_PRIVATE->bBug4208224WAREnabled;
}
static inline void kgraphicsSetBug4208224WAREnabled(OBJGPU *pGpu, struct KernelGraphics *pKernelGraphics, NvBool bProp) {
struct KernelGraphics_PRIVATE *pKernelGraphics_PRIVATE = (struct KernelGraphics_PRIVATE *)pKernelGraphics;
pKernelGraphics_PRIVATE->bBug4208224WAREnabled = bProp;
}
void kgraphicsDestruct_IMPL(struct KernelGraphics *arg0);
#define __nvoc_kgraphicsDestruct(arg0) kgraphicsDestruct_IMPL(arg0)

View File

@@ -1339,6 +1339,16 @@ static inline NV_STATUS kgspAllocateBooterUnloadUcodeImage(struct OBJGPU *pGpu,
#define kgspAllocateBooterUnloadUcodeImage(pGpu, pKernelGsp, ppBooterUnloadUcode) kgspAllocateBooterUnloadUcodeImage_IMPL(pGpu, pKernelGsp, ppBooterUnloadUcode)
#endif //__nvoc_kernel_gsp_h_disabled
void kgspRcAndNotifyAllUserChannels_IMPL(struct OBJGPU *pGpu, struct KernelGsp *pKernelGsp, NvU32 exceptType);
#ifdef __nvoc_kernel_gsp_h_disabled
static inline void kgspRcAndNotifyAllUserChannels(struct OBJGPU *pGpu, struct KernelGsp *pKernelGsp, NvU32 exceptType) {
NV_ASSERT_FAILED_PRECOMP("KernelGsp was disabled!");
}
#else //__nvoc_kernel_gsp_h_disabled
#define kgspRcAndNotifyAllUserChannels(pGpu, pKernelGsp, exceptType) kgspRcAndNotifyAllUserChannels_IMPL(pGpu, pKernelGsp, exceptType)
#endif //__nvoc_kernel_gsp_h_disabled
#undef PRIVATE_FIELD

View File

@@ -1044,6 +1044,7 @@ static const CHIPS_RELEASED sChipsReleased[] = {
{ 0x28A1, 0x0000, 0x0000, "NVIDIA GeForce RTX 4050 Laptop GPU" },
{ 0x28B8, 0x0000, 0x0000, "NVIDIA RTX 2000 Ada Generation Laptop GPU" },
{ 0x28B9, 0x0000, 0x0000, "NVIDIA RTX 1000 Ada Generation Laptop GPU" },
{ 0x28BA, 0x0000, 0x0000, "NVIDIA RTX 500 Ada Generation Laptop GPU" },
{ 0x28BB, 0x0000, 0x0000, "NVIDIA RTX 500 Ada Generation Laptop GPU" },
{ 0x28E0, 0x0000, 0x0000, "NVIDIA GeForce RTX 4060 Laptop GPU" },
{ 0x28E1, 0x0000, 0x0000, "NVIDIA GeForce RTX 4050 Laptop GPU" },

File diff suppressed because it is too large Load Diff

View File

@@ -346,6 +346,7 @@ struct Subdevice {
NV_STATUS (*__subdeviceCtrlCmdKGrInternalStaticGetFecsRecordSize__)(struct Subdevice *, NV2080_CTRL_INTERNAL_STATIC_GR_GET_FECS_RECORD_SIZE_PARAMS *);
NV_STATUS (*__subdeviceCtrlCmdKGrInternalStaticGetFecsTraceDefines__)(struct Subdevice *, NV2080_CTRL_INTERNAL_STATIC_GR_GET_FECS_TRACE_DEFINES_PARAMS *);
NV_STATUS (*__subdeviceCtrlCmdKGrInternalStaticGetPdbProperties__)(struct Subdevice *, NV2080_CTRL_INTERNAL_STATIC_GR_GET_PDB_PROPERTIES_PARAMS *);
NV_STATUS (*__subdeviceCtrlCmdKGrInternalInitBug4208224War__)(struct Subdevice *, NV2080_CTRL_INTERNAL_KGR_INIT_BUG4208224_WAR_PARAMS *);
NV_STATUS (*__subdeviceCtrlCmdGpuGetCachedInfo__)(struct Subdevice *, NV2080_CTRL_GPU_GET_INFO_V2_PARAMS *);
NV_STATUS (*__subdeviceCtrlCmdGpuGetInfoV2__)(struct Subdevice *, NV2080_CTRL_GPU_GET_INFO_V2_PARAMS *);
NV_STATUS (*__subdeviceCtrlCmdGpuGetIpVersion__)(struct Subdevice *, NV2080_CTRL_GPU_GET_IP_VERSION_PARAMS *);
@@ -944,6 +945,7 @@ NV_STATUS __nvoc_objCreate_Subdevice(Subdevice**, Dynamic*, NvU32, struct CALL_C
#define subdeviceCtrlCmdKGrInternalStaticGetFecsRecordSize(pSubdevice, pParams) subdeviceCtrlCmdKGrInternalStaticGetFecsRecordSize_DISPATCH(pSubdevice, pParams)
#define subdeviceCtrlCmdKGrInternalStaticGetFecsTraceDefines(pSubdevice, pParams) subdeviceCtrlCmdKGrInternalStaticGetFecsTraceDefines_DISPATCH(pSubdevice, pParams)
#define subdeviceCtrlCmdKGrInternalStaticGetPdbProperties(pSubdevice, pParams) subdeviceCtrlCmdKGrInternalStaticGetPdbProperties_DISPATCH(pSubdevice, pParams)
#define subdeviceCtrlCmdKGrInternalInitBug4208224War(pSubdevice, pParams) subdeviceCtrlCmdKGrInternalInitBug4208224War_DISPATCH(pSubdevice, pParams)
#define subdeviceCtrlCmdGpuGetCachedInfo(pSubdevice, pGpuInfoParams) subdeviceCtrlCmdGpuGetCachedInfo_DISPATCH(pSubdevice, pGpuInfoParams)
#define subdeviceCtrlCmdGpuGetInfoV2(pSubdevice, pGpuInfoParams) subdeviceCtrlCmdGpuGetInfoV2_DISPATCH(pSubdevice, pGpuInfoParams)
#define subdeviceCtrlCmdGpuGetIpVersion(pSubdevice, pGpuIpVersionParams) subdeviceCtrlCmdGpuGetIpVersion_DISPATCH(pSubdevice, pGpuIpVersionParams)
@@ -2634,6 +2636,12 @@ static inline NV_STATUS subdeviceCtrlCmdKGrInternalStaticGetPdbProperties_DISPAT
return pSubdevice->__subdeviceCtrlCmdKGrInternalStaticGetPdbProperties__(pSubdevice, pParams);
}
NV_STATUS subdeviceCtrlCmdKGrInternalInitBug4208224War_IMPL(struct Subdevice *pSubdevice, NV2080_CTRL_INTERNAL_KGR_INIT_BUG4208224_WAR_PARAMS *pParams);
static inline NV_STATUS subdeviceCtrlCmdKGrInternalInitBug4208224War_DISPATCH(struct Subdevice *pSubdevice, NV2080_CTRL_INTERNAL_KGR_INIT_BUG4208224_WAR_PARAMS *pParams) {
return pSubdevice->__subdeviceCtrlCmdKGrInternalInitBug4208224War__(pSubdevice, pParams);
}
NV_STATUS subdeviceCtrlCmdGpuGetCachedInfo_IMPL(struct Subdevice *pSubdevice, NV2080_CTRL_GPU_GET_INFO_V2_PARAMS *pGpuInfoParams);
static inline NV_STATUS subdeviceCtrlCmdGpuGetCachedInfo_DISPATCH(struct Subdevice *pSubdevice, NV2080_CTRL_GPU_GET_INFO_V2_PARAMS *pGpuInfoParams) {

View File

@@ -238,12 +238,12 @@ static NvBool __nvoc_thunk_RmResource_vgpuconfigapiAccessCallback(struct VgpuCon
static const struct NVOC_EXPORTED_METHOD_DEF __nvoc_exported_method_def_VgpuConfigApi[] =
{
{ /* [0] */
#if NVOC_EXPORTED_METHOD_DISABLED_BY_FLAG(0x10u)
#if NVOC_EXPORTED_METHOD_DISABLED_BY_FLAG(0x4u)
/*pFunc=*/ (void (*)(void)) NULL,
#else
/*pFunc=*/ (void (*)(void)) vgpuconfigapiCtrlCmdVgpuConfigSetInfo_IMPL,
#endif // NVOC_EXPORTED_METHOD_DISABLED_BY_FLAG(0x10u)
/*flags=*/ 0x10u,
#endif // NVOC_EXPORTED_METHOD_DISABLED_BY_FLAG(0x4u)
/*flags=*/ 0x4u,
/*accessRight=*/0x0u,
/*methodId=*/ 0xa0810101u,
/*paramSize=*/ sizeof(NVA081_CTRL_VGPU_CONFIG_INFO_PARAMS),
@@ -571,7 +571,7 @@ __nvoc_ctor_VgpuConfigApi_exit:
static void __nvoc_init_funcTable_VgpuConfigApi_1(VgpuConfigApi *pThis) {
PORT_UNREFERENCED_VARIABLE(pThis);
#if !NVOC_EXPORTED_METHOD_DISABLED_BY_FLAG(0x10u)
#if !NVOC_EXPORTED_METHOD_DISABLED_BY_FLAG(0x4u)
pThis->__vgpuconfigapiCtrlCmdVgpuConfigSetInfo__ = &vgpuconfigapiCtrlCmdVgpuConfigSetInfo_IMPL;
#endif

View File

@@ -177,6 +177,7 @@ typedef struct GspSystemInfo
NvBool bIsPassthru;
NvU64 sysTimerOffsetNs;
GSP_VF_INFO gspVFInfo;
NvBool bTdrEventSupported;
} GspSystemInfo;

View File

@@ -1352,6 +1352,10 @@ NV_STATUS rpcGspSetSystemInfo_v17_00
OBJTMR *pTmr = GPU_GET_TIMER(pGpu);
rpcInfo->sysTimerOffsetNs = pTmr->sysTimerOffsetNs;
// Indicate whether the driver supports NV2080_NOTIFIERS_UCODE_RESET event.
rpcInfo->bTdrEventSupported = pGpu->getProperty(pGpu, PDB_PROP_GPU_SUPPORTS_TDR_EVENT);
status = _issueRpcAsync(pGpu, pRpc);
}

View File

@@ -35,6 +35,7 @@
#include "published/ampere/ga100/dev_ram.h"
#include "published/ampere/ga100/dev_ctrl.h"
#include "published/ampere/ga100/dev_runlist.h"
NV_STATUS
@@ -317,3 +318,93 @@ kfifoGetMaxCeChannelGroups_GA100
return maxCeChannels;
}
/**
* @brief Starts halting a channel. A start operation must be matched with a
* complete operation later to wait for the channel to be preempted.
*
* @param[in] pGpu GPU object pointer
* @param[in] pKernelFifo Kernel FIFO object pointer
* @param[in] pKernelChannel Pointer to the channel to be halted.
*/
void
kfifoStartChannelHalt_GA100
(
OBJGPU *pGpu,
KernelFifo *pKernelFifo,
KernelChannel *pKernelChannel
)
{
NvU32 chramPriBase;
NvU32 channelVal;
NvU32 runlistId;
NvU32 runlistPriBase;
NvU32 runlistVal = 0;
runlistId = kchannelGetRunlistId(pKernelChannel);
if (kfifoEngineInfoXlate_HAL(pGpu, pKernelFifo,
ENGINE_INFO_TYPE_RUNLIST, runlistId,
ENGINE_INFO_TYPE_CHRAM_PRI_BASE, &chramPriBase) != NV_OK)
{
return;
}
if (kfifoEngineInfoXlate_HAL(pGpu, pKernelFifo,
ENGINE_INFO_TYPE_RUNLIST, runlistId,
ENGINE_INFO_TYPE_RUNLIST_PRI_BASE, &runlistPriBase) != NV_OK)
{
return;
}
// Disable this channel.
channelVal = FLD_SET_DRF(_CHRAM, _CHANNEL, _WRITE_CONTROL, _ONES_CLEAR_BITS, 0);
channelVal = FLD_SET_DRF(_CHRAM, _CHANNEL, _ENABLE, _IN_USE, channelVal);
GPU_REG_WR32(pGpu, chramPriBase + NV_CHRAM_CHANNEL(pKernelChannel->ChID), channelVal);
// Preempt the channel.
runlistVal = FLD_SET_DRF(_RUNLIST, _PREEMPT, _TYPE, _RUNLIST, 0);
GPU_REG_WR32(pGpu, runlistPriBase + NV_RUNLIST_PREEMPT, runlistVal);
}
/**
* @brief Completes halting a channel, waiting the channel preemption to
* complete, up to the specified timeout.
*
* @param[in] pGpu GPU object pointer
* @param[in] pKernelFifo Kernel FIFO object pointer
* @param[in] pKernelChannel Pointer to the channel in process of being halted.
* @param[in] pTimeout Specifies the timeout to wait for the channel
* preemption.
*/
void
kfifoCompleteChannelHalt_GA100
(
OBJGPU *pGpu,
KernelFifo *pKernelFifo,
KernelChannel *pKernelChannel,
RMTIMEOUT *pTimeout
)
{
NvU32 runlistId;
NvU32 runlistPriBase;
NvU32 runlistVal = 0;
runlistId = kchannelGetRunlistId(pKernelChannel);
if (kfifoEngineInfoXlate_HAL(pGpu, pKernelFifo,
ENGINE_INFO_TYPE_RUNLIST, runlistId,
ENGINE_INFO_TYPE_RUNLIST_PRI_BASE, &runlistPriBase) != NV_OK)
{
return;
}
// Wait for the preemption to complete.
do
{
if (gpuCheckTimeout(pGpu, pTimeout) == NV_ERR_TIMEOUT)
{
break;
}
runlistVal = GPU_REG_RD32(pGpu, runlistPriBase + NV_RUNLIST_PREEMPT);
} while (FLD_TEST_DRF(_RUNLIST, _PREEMPT, _RUNLIST_PREEMPT_PENDING, _TRUE, runlistVal));
}

View File

@@ -1,5 +1,5 @@
/*
* SPDX-FileCopyrightText: Copyright (c) 2020-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-FileCopyrightText: Copyright (c) 2020-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a
@@ -1674,6 +1674,9 @@ void kchannelNotifyGeneric_IMPL
// validate notifyIndex
NV_CHECK_OR_RETURN_VOID(LEVEL_INFO, notifyIndex < classInfo.notifiersMaxCount);
// Check if we have allocated the channel notifier action table
NV_CHECK_OR_RETURN_VOID(LEVEL_ERROR, pKernelChannel->pNotifyActions != NULL);
// handle notification if client wants it
if (pKernelChannel->pNotifyActions[notifyIndex] != classInfo.eventActionDisable)
{

View File

@@ -179,3 +179,310 @@ kgraphicsAllocGrGlobalCtxBuffers_TU102
return status;
}
/**
* @brief Initializes Bug 4208224 by performing the following actions
* 1.) Sets up static handles inside an info struct to be referenced later
* 2.) Creates a channel tied to VEID0 on GR0
* 3.) Sends an RPC to physical RM for the physical side initialization
*/
NV_STATUS
kgraphicsInitializeBug4208224WAR_TU102
(
OBJGPU *pGpu,
KernelGraphics *pKernelGraphics
)
{
NV_STATUS status = NV_OK;
RM_API *pRmApi = rmapiGetInterface(RMAPI_GPU_LOCK_INTERNAL);
NV2080_CTRL_INTERNAL_KGR_INIT_BUG4208224_WAR_PARAMS params = {0};
NV_CHECK_OK_OR_RETURN(LEVEL_ERROR,
kgraphicsCreateBug4208224Channel_HAL(pGpu, pKernelGraphics));
params.bTeardown = NV_FALSE;
status = pRmApi->Control(pRmApi,
pKernelGraphics->bug4208224Info.hClient,
pKernelGraphics->bug4208224Info.hSubdeviceId,
NV2080_CTRL_CMD_INTERNAL_KGR_INIT_BUG4208224_WAR,
&params,
sizeof(params));
if (status != NV_OK)
{
NV_ASSERT_OK(pRmApi->Free(pRmApi,
pKernelGraphics->bug4208224Info.hClient,
pKernelGraphics->bug4208224Info.hClient));
}
return status;
}
/*!
* @brief Creates a VEID0 channel for Bug 4208224 WAR
*
* @return NV_OK if channel created successfully
*/
NV_STATUS
kgraphicsCreateBug4208224Channel_TU102
(
OBJGPU *pGpu,
KernelGraphics *pKernelGraphics
)
{
NV_STATUS status = NV_OK;
NvHandle hClientId = NV01_NULL_OBJECT;
NvHandle hDeviceId;
NvHandle hSubdeviceId;
NvHandle hVASpace = KGRAPHICS_SCRUBBER_HANDLE_VAS;
NvHandle hPBVirtMemId = KGRAPHICS_SCRUBBER_HANDLE_PBVIRT;
NvHandle hPBPhysMemId = KGRAPHICS_SCRUBBER_HANDLE_PBPHYS;
NvHandle hChannelId = KGRAPHICS_SCRUBBER_HANDLE_CHANNEL;
NvHandle hObj3D = KGRAPHICS_SCRUBBER_HANDLE_3DOBJ;
NvHandle hUserdId = KGRAPHICS_SCRUBBER_HANDLE_USERD;
NvU32 gpFifoEntries = 32; // power-of-2 random choice
NvU64 gpFifoSize = NVA06F_GP_ENTRY__SIZE * gpFifoEntries;
NvU64 chSize = gpFifoSize;
RM_API *pRmApi = rmapiGetInterface(RMAPI_GPU_LOCK_INTERNAL);
RsClient *pClientId;
NvBool bBcStatus;
NvBool bAcquireLock = NV_FALSE;
NvU32 sliLoopReentrancy;
NV_VASPACE_ALLOCATION_PARAMETERS vaParams;
NV_MEMORY_ALLOCATION_PARAMS memAllocParams;
NV_CHANNEL_ALLOC_PARAMS channelGPFIFOAllocParams;
NvU32 classNum;
NvU32 primarySliSubDeviceInstance;
// XXX This should be removed when broadcast SLI support is deprecated
if (!gpumgrIsParentGPU(pGpu))
{
return NV_OK;
}
bBcStatus = gpumgrGetBcEnabledStatus(pGpu);
// FIXME these allocations corrupt BC state
NV_ASSERT_OK_OR_RETURN(
rmapiutilAllocClientAndDeviceHandles(pRmApi, pGpu, &hClientId, &hDeviceId, &hSubdeviceId));
pKernelGraphics->bug4208224Info.hClient = hClientId;
pKernelGraphics->bug4208224Info.hDeviceId = hDeviceId;
pKernelGraphics->bug4208224Info.hSubdeviceId = hSubdeviceId;
// rmapiutilAllocClientAndDeviceHandles allocates a subdevice object for this subDeviceInstance
primarySliSubDeviceInstance = gpumgrGetSubDeviceInstanceFromGpu(pGpu);
NV_ASSERT_OK_OR_RETURN(serverGetClientUnderLock(&g_resServ, hClientId, &pClientId));
gpumgrSetBcEnabledStatus(pGpu, NV_TRUE);
// As we have forced here SLI broadcast mode, temporarily reset the reentrancy count
sliLoopReentrancy = gpumgrSLILoopReentrancyPop(pGpu);
// Allocate subdevices for secondary GPUs
SLI_LOOP_START(SLI_LOOP_FLAGS_BC_ONLY)
{
NvHandle hSecondary;
NV2080_ALLOC_PARAMETERS nv2080AllocParams;
NvU32 thisSubDeviceInstance = gpumgrGetSubDeviceInstanceFromGpu(pGpu);
// Skip if already allocated by rmapiutilAllocClientAndDeviceHandles()
if (thisSubDeviceInstance == primarySliSubDeviceInstance)
SLI_LOOP_CONTINUE;
// Allocate a subDevice
NV_CHECK_OK_OR_GOTO(status, LEVEL_ERROR,
clientGenResourceHandle(pClientId, &hSecondary),
cleanup);
portMemSet(&nv2080AllocParams, 0, sizeof(nv2080AllocParams));
nv2080AllocParams.subDeviceId = thisSubDeviceInstance;
NV_CHECK_OK(status, LEVEL_SILENT,
pRmApi->AllocWithHandle(pRmApi,
hClientId,
hDeviceId,
hSecondary,
NV20_SUBDEVICE_0,
&nv2080AllocParams,
sizeof(nv2080AllocParams)));
}
SLI_LOOP_END;
//
// VidHeapControl and vaspace creation calls should happen outside GPU locks
// UVM/CUDA may be holding the GPU locks here and the allocation may subsequently fail
// So explicitly release GPU locks before RmVidHeapControl
//
rmGpuLocksRelease(GPUS_LOCK_FLAGS_NONE, NULL);
bAcquireLock = NV_TRUE;
pRmApi = rmapiGetInterface(RMAPI_API_LOCK_INTERNAL);
// Create a new VAspace for channel
portMemSet(&vaParams, 0, sizeof(NV_VASPACE_ALLOCATION_PARAMETERS));
vaParams.flags = NV_VASPACE_ALLOCATION_FLAGS_PTETABLE_HEAP_MANAGED;
NV_ASSERT_OK_OR_GOTO(status,
pRmApi->AllocWithHandle(pRmApi, hClientId, hDeviceId, hVASpace, FERMI_VASPACE_A, &vaParams, sizeof(vaParams)),
cleanup);
// Allocate gpfifo entries
portMemSet(&memAllocParams, 0, sizeof(NV_MEMORY_ALLOCATION_PARAMS));
memAllocParams.owner = HEAP_OWNER_RM_CLIENT_GENERIC;
memAllocParams.type = NVOS32_TYPE_IMAGE;
memAllocParams.size = chSize;
memAllocParams.attr = DRF_DEF(OS32, _ATTR, _LOCATION, _PCI);
memAllocParams.hVASpace = 0; // Physical allocations don't expect vaSpace handles
NV_ASSERT_OK_OR_GOTO(status,
pRmApi->AllocWithHandle(pRmApi, hClientId, hDeviceId, hPBPhysMemId, NV01_MEMORY_SYSTEM, &memAllocParams, sizeof(memAllocParams)),
cleanup);
portMemSet(&memAllocParams, 0, sizeof(NV_MEMORY_ALLOCATION_PARAMS));
memAllocParams.owner = HEAP_OWNER_RM_CLIENT_GENERIC;
memAllocParams.type = NVOS32_TYPE_IMAGE;
memAllocParams.size = chSize;
memAllocParams.attr = DRF_DEF(OS32, _ATTR, _LOCATION, _PCI);
memAllocParams.flags = NVOS32_ALLOC_FLAGS_VIRTUAL;
memAllocParams.hVASpace = hVASpace; // Virtual allocation expect vaSpace handles
// 0 handle = allocations on gpu default vaSpace
NV_ASSERT_OK_OR_GOTO(status,
pRmApi->AllocWithHandle(pRmApi, hClientId, hDeviceId, hPBVirtMemId, NV50_MEMORY_VIRTUAL, &memAllocParams, sizeof(memAllocParams)),
cleanup);
// Allocate Userd
NvU32 userdMemClass = NV01_MEMORY_LOCAL_USER;
NvU32 ctrlSize;
if (gpuIsClassSupported(pGpu, VOLTA_CHANNEL_GPFIFO_A))
{
ctrlSize = sizeof(Nvc36fControl);
}
else if (gpuIsClassSupported(pGpu, TURING_CHANNEL_GPFIFO_A))
{
ctrlSize = sizeof(Nvc46fControl);
}
else
{
status = NV_ERR_NOT_SUPPORTED;
goto cleanup;
}
portMemSet(&memAllocParams, 0, sizeof(NV_MEMORY_ALLOCATION_PARAMS));
memAllocParams.owner = HEAP_OWNER_RM_CLIENT_GENERIC;
memAllocParams.size = ctrlSize;
memAllocParams.type = NVOS32_TYPE_IMAGE;
// Apply registry overrides to USERD.
switch (DRF_VAL(_REG_STR_RM, _INST_LOC, _USERD, pGpu->instLocOverrides))
{
case NV_REG_STR_RM_INST_LOC_USERD_NCOH:
case NV_REG_STR_RM_INST_LOC_USERD_COH:
userdMemClass = NV01_MEMORY_SYSTEM;
memAllocParams.attr = DRF_DEF(OS32, _ATTR, _LOCATION, _PCI);
break;
case NV_REG_STR_RM_INST_LOC_USERD_VID:
case NV_REG_STR_RM_INST_LOC_USERD_DEFAULT:
memAllocParams.attr = DRF_DEF(OS32, _ATTR, _LOCATION, _VIDMEM);
break;
}
NV_ASSERT_OK_OR_GOTO(status,
pRmApi->AllocWithHandle(pRmApi, hClientId, hDeviceId, hUserdId,
userdMemClass, &memAllocParams, sizeof(memAllocParams)),
cleanup);
// Get fifo channel class Id
classNum = kfifoGetChannelClassId(pGpu, GPU_GET_KERNEL_FIFO(pGpu));
NV_ASSERT_OR_GOTO(classNum != 0, cleanup);
// Allocate a bare channel
portMemSet(&channelGPFIFOAllocParams, 0, sizeof(NV_CHANNEL_ALLOC_PARAMS));
channelGPFIFOAllocParams.hVASpace = hVASpace;
channelGPFIFOAllocParams.hObjectBuffer = hPBVirtMemId;
channelGPFIFOAllocParams.gpFifoEntries = gpFifoEntries;
//
// Set the gpFifoOffset to zero intentionally since we only need this channel
// to be created, but will not submit any work to it. So it's fine not to
// provide a valid offset here.
//
channelGPFIFOAllocParams.gpFifoOffset = 0;
channelGPFIFOAllocParams.hUserdMemory[0] = hUserdId;
channelGPFIFOAllocParams.engineType = gpuGetNv2080EngineType(RM_ENGINE_TYPE_GR0);
NV_ASSERT_OK_OR_GOTO(status,
pRmApi->AllocWithHandle(pRmApi, hClientId, hDeviceId, hChannelId,
classNum, &channelGPFIFOAllocParams, sizeof(channelGPFIFOAllocParams)),
cleanup);
// Free userD handle as it shouldn't be needed for this channel
NV_ASSERT_OK_OR_GOTO(status,
pRmApi->Free(pRmApi, hClientId, hUserdId),
cleanup);
// Reaquire the GPU locks
NV_ASSERT_OK_OR_GOTO(status,
rmGpuLocksAcquire(GPUS_LOCK_FLAGS_NONE, RM_LOCK_MODULES_GR),
cleanup);
bAcquireLock = NV_FALSE;
pRmApi = rmapiGetInterface(RMAPI_GPU_LOCK_INTERNAL);
// Get KernelGraphicsObject class Id
NV_ASSERT_OK_OR_GOTO(status,
kgraphicsGetClassByType(pGpu, pKernelGraphics, GR_OBJECT_TYPE_3D, &classNum),
cleanup);
NV_ASSERT_OR_GOTO(classNum != 0, cleanup);
// Allocate a GR object on the channel
NV_ASSERT_OK_OR_GOTO(status,
pRmApi->AllocWithHandle(pRmApi, hClientId, hChannelId, hObj3D, classNum, NULL, 0),
cleanup);
cleanup:
if (bAcquireLock)
{
NV_ASSERT_OK_OR_CAPTURE_FIRST_ERROR(status,
rmGpuLocksAcquire(GPUS_LOCK_FLAGS_NONE, RM_LOCK_MODULES_GR));
pRmApi = rmapiGetInterface(RMAPI_GPU_LOCK_INTERNAL);
}
if (status != NV_OK)
{
// Drop GPU lock while freeing memory and channel handles
// Free all handles
NV_ASSERT_OK_OR_CAPTURE_FIRST_ERROR(status,
pRmApi->Free(pRmApi, hClientId, hClientId));
}
pKernelGraphics->bug4208224Info.bConstructed = (status == NV_OK);
// Restore the reentrancy count
gpumgrSLILoopReentrancyPush(pGpu, sliLoopReentrancy);
gpumgrSetBcEnabledStatus(pGpu, bBcStatus);
return status;
}
/*!
* @brief Determines if a channel for Bug 4208224 is needed
*/
NvBool
kgraphicsIsBug4208224WARNeeded_TU102
(
OBJGPU *pGpu,
KernelGraphics *pKernelGraphics
)
{
if (pGpu->getProperty(pGpu, PDB_PROP_GPU_IS_ALL_INST_IN_SYSMEM))
{
return NV_FALSE;
}
return kgraphicsGetBug4208224WAREnabled(pGpu, pKernelGraphics);
}

View File

@@ -98,6 +98,21 @@ static NV_STATUS _kgraphicsPostSchedulingEnableHandler(OBJGPU *, void *);
static void
_kgraphicsInitRegistryOverrides(OBJGPU *pGpu, KernelGraphics *pKernelGraphics)
{
{
NvU32 data;
if (osReadRegistryDword(pGpu, NV_REG_STR_RM_FORCE_GR_SCRUBBER_CHANNEL, &data) == NV_OK)
{
if (data == NV_REG_STR_RM_FORCE_GR_SCRUBBER_CHANNEL_DISABLE)
{
kgraphicsSetBug4208224WAREnabled(pGpu, pKernelGraphics, NV_FALSE);
}
else if (data == NV_REG_STR_RM_FORCE_GR_SCRUBBER_CHANNEL_ENABLE)
{
kgraphicsSetBug4208224WAREnabled(pGpu, pKernelGraphics, NV_TRUE);
}
}
}
return;
}
@@ -308,6 +323,10 @@ kgraphicsStateInitLocked_IMPL
NULL, NULL);
}
pKernelGraphics->bug4208224Info.hClient = NV01_NULL_OBJECT;
pKernelGraphics->bug4208224Info.hDeviceId = NV01_NULL_OBJECT;
pKernelGraphics->bug4208224Info.hSubdeviceId = NV01_NULL_OBJECT;
pKernelGraphics->bug4208224Info.bConstructed = NV_FALSE;
return NV_OK;
}
@@ -356,6 +375,21 @@ kgraphicsStatePreUnload_IMPL
NvU32 flags
)
{
if (pKernelGraphics->bug4208224Info.bConstructed)
{
RM_API *pRmApi = rmapiGetInterface(RMAPI_GPU_LOCK_INTERNAL);
NV2080_CTRL_INTERNAL_KGR_INIT_BUG4208224_WAR_PARAMS params = {0};
params.bTeardown = NV_TRUE;
NV_ASSERT_OK(pRmApi->Control(pRmApi,
pKernelGraphics->bug4208224Info.hClient,
pKernelGraphics->bug4208224Info.hSubdeviceId,
NV2080_CTRL_CMD_INTERNAL_KGR_INIT_BUG4208224_WAR,
&params,
sizeof(params)));
NV_ASSERT_OK(pRmApi->Free(pRmApi, pKernelGraphics->bug4208224Info.hClient, pKernelGraphics->bug4208224Info.hClient));
pKernelGraphics->bug4208224Info.bConstructed = NV_FALSE;
}
fecsBufferUnmap(pGpu, pKernelGraphics);
@@ -440,7 +474,7 @@ _kgraphicsPostSchedulingEnableHandler
const KGRAPHICS_STATIC_INFO *pKernelGraphicsStaticInfo = kgraphicsGetStaticInfo(pGpu, pKernelGraphics);
// Nothing to do for non-GSPCLIENT
if (!IS_GSP_CLIENT(pGpu))
if (!IS_GSP_CLIENT(pGpu) && !kgraphicsIsBug4208224WARNeeded_HAL(pGpu, pKernelGraphics))
return NV_OK;
// Defer golden context channel creation to GPU instance configuration
@@ -471,7 +505,13 @@ _kgraphicsPostSchedulingEnableHandler
}
}
return kgraphicsCreateGoldenImageChannel(pGpu, pKernelGraphics);
NV_CHECK_OK_OR_RETURN(LEVEL_ERROR, kgraphicsCreateGoldenImageChannel(pGpu, pKernelGraphics));
if (kgraphicsIsBug4208224WARNeeded_HAL(pGpu, pKernelGraphics))
{
return kgraphicsInitializeBug4208224WAR_HAL(pGpu, pKernelGraphics);
}
return NV_OK;
}
void

View File

@@ -3201,19 +3201,16 @@ kgrctxDecObjectCount_IMPL
* one VGPU configuration.
*/
GR_GLOBALCTX_BUFFER
kgrctxGetRegisterAccessMapId_PF
kgrctxGetRegisterAccessMapId_IMPL
(
OBJGPU *pGpu,
KernelGraphicsContext *pKernelGraphicsContext,
KernelChannel *pKernelChannel
)
{
RmClient *pRmClient = dynamicCast(RES_GET_CLIENT(pKernelChannel), RmClient);
RS_PRIV_LEVEL privLevel = rmclientGetCachedPrivilege(pRmClient);
// Using cached privilege because this function is called at a raised IRQL.
if ((privLevel >= RS_PRIV_LEVEL_USER_ROOT)
&& !hypervisorIsVgxHyper() && IS_GFID_PF(kchannelGetGfid(pKernelChannel)))
if (kchannelCheckIsAdmin(pKernelChannel)
&& !hypervisorIsVgxHyper() && IS_GFID_PF(kchannelGetGfid(pKernelChannel)))
{
return GR_GLOBALCTX_BUFFER_UNRESTRICTED_PRIV_ACCESS_MAP;
}

View File

@@ -773,6 +773,22 @@ kgspResetHw_TU102
return NV_OK;
}
static NvBool kgspCrashCatReportImpactsGspRm(CrashCatReport *pReport)
{
NV_CRASHCAT_CONTAINMENT containment;
containment = crashcatReportSourceContainment_HAL(pReport);
switch (containment)
{
case NV_CRASHCAT_CONTAINMENT_RISCV_MODE_M:
case NV_CRASHCAT_CONTAINMENT_RISCV_HART:
case NV_CRASHCAT_CONTAINMENT_UNCONTAINED:
return NV_TRUE;
default:
return NV_FALSE;
}
}
NvBool
kgspHealthCheck_TU102
(
@@ -791,7 +807,8 @@ kgspHealthCheck_TU102
while ((pReport = crashcatEngineGetNextCrashReport(pCrashCatEng)) != NULL)
{
bHealthy = NV_FALSE;
if (kgspCrashCatReportImpactsGspRm(pReport))
bHealthy = NV_FALSE;
NV_PRINTF(LEVEL_ERROR,
"****************************** GSP-CrashCat Report *******************************\n");
@@ -839,10 +856,19 @@ kgspHealthCheck_TU102
exit_health_check:
if (!bHealthy)
{
NvBool bFirstFatal = !pKernelGsp->bFatalError;
pKernelGsp->bFatalError = NV_TRUE;
if (pKernelGsp->pRpc)
{
kgspLogRpcDebugInfo(pGpu, pKernelGsp->pRpc, GSP_ERROR, pKernelGsp->bPollingForRpcResponse);
}
if (bFirstFatal)
{
kgspRcAndNotifyAllUserChannels(pGpu, pKernelGsp, GSP_ERROR);
}
gpuCheckEccCounts_HAL(pGpu);

View File

@@ -583,6 +583,81 @@ _kgspRpcRCTriggered
rpc_params->partitionAttributionId);
}
/*!
* This function is called on critical FW crash to RC and notify an error code to
* all user mode channels, allowing the user mode apps to fail deterministically.
*
* @param[in] pGpu GPU object pointer
* @param[in] pKernelGsp KernelGsp object pointer
* @param[in] exceptType Error code to send to the RC notifiers
*
*/
void
kgspRcAndNotifyAllUserChannels
(
OBJGPU *pGpu,
KernelGsp *pKernelGsp,
NvU32 exceptType
)
{
KernelRc *pKernelRc = GPU_GET_KERNEL_RC(pGpu);
KernelChannel *pKernelChannel;
KernelFifo *pKernelFifo = GPU_GET_KERNEL_FIFO(pGpu);
CHANNEL_ITERATOR chanIt;
RMTIMEOUT timeout;
NV_PRINTF(LEVEL_ERROR, "RC all user channels for critical error %d.\n", exceptType);
// Pass 1: halt all user channels.
kfifoGetChannelIterator(pGpu, pKernelFifo, &chanIt);
while (kfifoGetNextKernelChannel(pGpu, pKernelFifo, &chanIt, &pKernelChannel) == NV_OK)
{
//
// Kernel (uvm) channels are skipped to workaround nvbug 4503046, where
// uvm attributes all errors as global and fails operations on all GPUs,
// in addition to the current failing GPU.
//
if (kchannelCheckIsKernel(pKernelChannel))
{
continue;
}
kfifoStartChannelHalt(pGpu, pKernelFifo, pKernelChannel);
}
//
// Pass 2: Wait for the halts to complete, and RC notify the user channels.
// The channel halts require a preemption, which may not be able to complete
// since the GSP is no longer servicing interrupts. Wait for up to the
// default GPU timeout value for the preemptions to complete.
//
gpuSetTimeout(pGpu, GPU_TIMEOUT_DEFAULT, &timeout, 0);
kfifoGetChannelIterator(pGpu, pKernelFifo, &chanIt);
while (kfifoGetNextKernelChannel(pGpu, pKernelFifo, &chanIt, &pKernelChannel) == NV_OK)
{
// Skip kernel (uvm) channels as only user channel halts are initiated above.
if (kchannelCheckIsKernel(pKernelChannel))
{
continue;
}
kfifoCompleteChannelHalt(pGpu, pKernelFifo, pKernelChannel, &timeout);
NV_ASSERT_OK(krcErrorSetNotifier(pGpu, pKernelRc,
pKernelChannel,
exceptType,
kchannelGetEngineType(pKernelChannel),
RC_NOTIFIER_SCOPE_CHANNEL));
NV_ASSERT_OK(krcErrorSendEventNotifications_HAL(pGpu, pKernelRc,
pKernelChannel,
kchannelGetEngineType(pKernelChannel),
exceptType,
RC_NOTIFIER_SCOPE_CHANNEL,
0));
}
}
/*!
* Receive Xid notification from GSP-RM
*

View File

@@ -125,7 +125,7 @@ _vidmemPmaAllocate
MemoryManager *pMemoryManager = GPU_GET_MEMORY_MANAGER(pGpu);
PMA *pPma = &pHeap->pmaObject;
NvU64 size = 0;
NvU32 pageCount;
NvU32 pageCount = 0;
NvU32 pmaInfoSize;
NvU64 pageSize;
NV_STATUS status;
@@ -290,6 +290,9 @@ _vidmemPmaAllocate
// Get the number of pages to be allocated by PMA
NV_CHECK_OR_RETURN(LEVEL_ERROR,
(NV_DIV_AND_CEIL(size, pageSize) <= NV_U32_MAX),
NV_ERR_NO_MEMORY);
pageCount = (NvU32) NV_DIV_AND_CEIL(size, pageSize);
retry_alloc:
@@ -301,7 +304,12 @@ retry_alloc:
}
else
{
pmaInfoSize = sizeof(PMA_ALLOC_INFO) + ((pageCount - 1) * sizeof(NvU64));
NV_CHECK_OR_RETURN(LEVEL_ERROR,
portSafeMulU32((pageCount - 1), (sizeof(NvU64)), &pmaInfoSize),
NV_ERR_NO_MEMORY);
NV_CHECK_OR_RETURN(LEVEL_ERROR,
portSafeAddU32(pmaInfoSize, (sizeof(PMA_ALLOC_INFO)), &pmaInfoSize),
NV_ERR_NO_MEMORY);
}
// Alloc the tracking structure and store the values in it.

View File

@@ -115,6 +115,11 @@ void crashcatReportLogSource_V1_LIBOS3(CrashCatReport *pReport)
}
}
NV_CRASHCAT_CONTAINMENT crashcatReportSourceContainment_V1_LIBOS3(CrashCatReport *pReport)
{
return crashcatReportV1SourceCauseContainment(&pReport->v1.report);
}
void crashcatReportLogReporter_V1_LIBOS3(CrashCatReport *pReport)
{
NvCrashCatReport_V1 *pReportV1 = &pReport->v1.report;