NVIDIA/open-gpu-kernel-modules
1
0
Fork 0
mirror of https://github.com/NVIDIA/open-gpu-kernel-modules.git synced 2026-02-07 08:39:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

575.51.02

Browse Source
This commit is contained in:
Bernhard Stoeckner
2025-04-17 19:35:38 +02:00
parent e8113f665d
commit 4159579888
1142 changed files with 309085 additions and 272273 deletions
Download Patch File Download Diff File Expand all files Collapse all files

86
kernel-open/nvidia-uvm/uvm_hal.c
View File

@@ -407,6 +407,32 @@ static uvm_hal_class_ops_t arch_table[] =
},
};
// chip_table[] is different from the other class op tables - it is used to
// apply chip specific overrides to arch ops. This means unlike the other class
// op tables, parent_id does not refer to a preceding entry within the table
// itself. parent_id is an architecture (not a chip id) and instead refers to an
// entry in arch_table[]. This means that arch_table[] must be initialized
// before chip_table[]. chip_table[] must be initialized using
// ops_init_from_table(arch_table) instead of ops_init_from_parent().
// TODO: BUG 5044266: the chip ops should be separated from the arch ops.
static uvm_hal_class_ops_t chip_table[] =
{
{
.id = NV2080_CTRL_MC_ARCH_INFO_ARCHITECTURE_GB100 | NV2080_CTRL_MC_ARCH_INFO_IMPLEMENTATION_GB10B,
.parent_id = NV2080_CTRL_MC_ARCH_INFO_ARCHITECTURE_GB100,
.u.arch_ops = {
.mmu_mode_hal = uvm_hal_mmu_mode_blackwell_integrated,
}
},
{
.id = NV2080_CTRL_MC_ARCH_INFO_ARCHITECTURE_GB200 | NV2080_CTRL_MC_ARCH_INFO_IMPLEMENTATION_GB20B,
.parent_id = NV2080_CTRL_MC_ARCH_INFO_ARCHITECTURE_GB200,
.u.arch_ops = {
.mmu_mode_hal = uvm_hal_mmu_mode_blackwell_integrated,
}
},
};
static uvm_hal_class_ops_t fault_buffer_table[] =
{
{
@@ -670,33 +696,35 @@ static inline void op_copy(uvm_hal_class_ops_t *dst, uvm_hal_class_ops_t *src, N
memcpy(m_dst, m_src, sizeof(void *));
}
static inline NV_STATUS ops_init_from_parent(uvm_hal_class_ops_t *table,
NvU32 row_count,
NvLength op_count,
NvLength op_offset)
static inline NV_STATUS ops_init_from_table(uvm_hal_class_ops_t *dest_table,
NvU32 dest_row_count,
uvm_hal_class_ops_t *src_table,
NvU32 src_row_count,
NvLength op_count,
NvLength op_offset)
{
NvLength i;
for (i = 0; i < row_count; i++) {
for (i = 0; i < dest_row_count; i++) {
NvLength j;
uvm_hal_class_ops_t *parent = NULL;
if (table[i].parent_id != 0) {
parent = ops_find_by_id(table, i, table[i].parent_id);
if (dest_table[i].parent_id != 0) {
parent = ops_find_by_id(src_table, src_row_count, dest_table[i].parent_id);
if (parent == NULL)
return NV_ERR_INVALID_CLASS;
// Go through all the ops and assign from parent's corresponding op
// if NULL
for (j = 0; j < op_count; j++) {
if (op_is_null(table + i, j, op_offset))
op_copy(table + i, parent, j, op_offset);
if (op_is_null(dest_table + i, j, op_offset))
op_copy(dest_table + i, parent, j, op_offset);
}
}
// At this point, it is an error to have missing HAL operations
for (j = 0; j < op_count; j++) {
if (op_is_null(table + i, j, op_offset))
if (op_is_null(dest_table + i, j, op_offset))
return NV_ERR_INVALID_STATE;
}
}
@@ -704,6 +732,19 @@ static inline NV_STATUS ops_init_from_parent(uvm_hal_class_ops_t *table,
return NV_OK;
}
static inline NV_STATUS ops_init_from_parent(uvm_hal_class_ops_t *table,
NvU32 row_count,
NvLength op_count,
NvLength op_offset)
{
return ops_init_from_table(table,
row_count,
table,
row_count,
op_count,
op_offset);
}
NV_STATUS uvm_hal_init_table(void)
{
NV_STATUS status;
@@ -732,6 +773,18 @@ NV_STATUS uvm_hal_init_table(void)
return status;
}
// chip_table[] must be initialized after arch_table[].
status = ops_init_from_table(chip_table,
ARRAY_SIZE(chip_table),
arch_table,
ARRAY_SIZE(arch_table),
ARCH_OP_COUNT,
offsetof(uvm_hal_class_ops_t, u.arch_ops));
if (status != NV_OK) {
UVM_ERR_PRINT("ops_init_from_table(chip_table) failed: %s\n", nvstatusToString(status));
return status;
}
status = ops_init_from_parent(fault_buffer_table,
ARRAY_SIZE(fault_buffer_table),
FAULT_BUFFER_OP_COUNT,
@@ -797,6 +850,13 @@ NV_STATUS uvm_hal_init_gpu(uvm_parent_gpu_t *parent_gpu)
parent_gpu->arch_hal = &class_ops->u.arch_ops;
// Apply per chip overrides if required
class_ops = ops_find_by_id(chip_table,
ARRAY_SIZE(chip_table),
gpu_info->gpuArch | gpu_info->gpuImplementation);
if (class_ops)
parent_gpu->arch_hal = &class_ops->u.arch_ops;
class_ops = ops_find_by_id(fault_buffer_table, ARRAY_SIZE(fault_buffer_table), gpu_info->gpuArch);
if (class_ops == NULL) {
UVM_ERR_PRINT("Fault buffer HAL not found, GPU %s, arch: 0x%X\n",
@@ -840,6 +900,12 @@ static void hal_override_properties(uvm_parent_gpu_t *parent_gpu)
// TODO: Bug 200692962: Add support for access counters in vGPU
if ((parent_gpu->virt_mode != UVM_VIRT_MODE_NONE) || g_uvm_global.conf_computing_enabled)
parent_gpu->access_counters_supported = false;
// TODO: Bug 4637114: [UVM] Remove support for physical access counter
// notifications. Always set to false, until we remove the PMM reverse
// mapping code.
parent_gpu->access_counters_can_use_physical_addresses = false;
}
void uvm_hal_init_properties(uvm_parent_gpu_t *parent_gpu)