mirror of
https://github.com/hanzoai/visor.git
synced 2026-09-24 17:04:14 +00:00
Ayush RanjanandgVisor bot
c652514e4f
The objsMu sychronizes 2 things: - nvproxy object book-keeping - host ioctl(2) invocations Since this mutex was on nvproxy device level, nvproxy was synchronizing host ioctl invocations across clients even though the NVIDIA kernel driver allows ioctl commands across clients to proceed in parallel. Move the lock to rootClient and implement locking on a per-client basis. This is similar to what the NVIDIA kernel driver does; see usages of _serverLockClientWithLockInfo() in the driver. As far as we can tell, CUDA programs issue one command to each client at a time so synchronizing on a per-client basis shouldn't hurt concurrency. Investigated-by: Jamie Liu <[email protected]> Suggested-by: Jamie Liu <[email protected]> PiperOrigin-RevId: 740647187
27 lines
953 B
Go
27 lines
953 B
Go
// Copyright 2023 The gVisor Authors.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
package nvgpu
|
|
|
|
// Status codes, from src/common/sdk/nvidia/inc/nvstatuscodes.h.
|
|
const (
|
|
NV_OK = 0x00000000
|
|
NV_ERR_INVALID_ADDRESS = 0x0000001e
|
|
NV_ERR_INVALID_ARGUMENT = 0x0000001f
|
|
NV_ERR_INVALID_CLASS = 0x00000022
|
|
NV_ERR_INVALID_CLIENT = 0x00000023
|
|
NV_ERR_INVALID_LIMIT = 0x0000002e
|
|
NV_ERR_NOT_SUPPORTED = 0x00000056
|
|
)
|