ROCm: hsa api call failure at line 900, file: /home/jenkins/jenkins-root/workspace/compute-rocm-rel-1.8/rocminfo/rocminfo.cc. Call returned 4104
I’m use Debian 9 with 4.16. kernel with Nitro+ RX570
wget -qO - http://repo.radeon.com/rocm/apt/debian/rocm.gpg.key | sudo apt-key add -
sudo sh -c 'echo deb [arch=amd64] http://repo.radeon.com/rocm/apt/debian/ xenial main > /etc/apt/sources.list.d/rocm.list'
sudo apt update
sudo apt-get install -y rocm-opencl-dev rocm-dkms rocminfo
sudo usermod -a -G video $LOGNAME
GRUB_CMDLINE_LINUX_DEFAULT="selinux=0 amdgpu.vm_fragment_size=9 nmi_watchdog=0 pti=off 3 spectre_v2=off nospectre_v2 nopti retp=0 ibrs=0 ibpb=0"
update-initramfs -u
update-grub
root@z820:~# lspci | grep -i AMD
05:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Ellesmere [Radeon RX 470/480] (rev ef)
05:00.1 Audio device: Advanced Micro Devices, Inc. [AMD/ATI] Ellesmere [Radeon RX 580]
root@z820:~#
All installed, but need make correct PATHs and etc. What should i do also?
root@z820:/opt/rocm/bin# ./rocminfo
hsa api call failure at line 900, file: /home/jenkins/jenkins-root/workspace/compute-rocm-rel-1.8/rocminfo/rocminfo.cc. Call returned 4104
root@z820:/opt/rocm/bin#
root@z820:/opt/rocm/opencl/bin/x86_64# ./clinfo
terminate called after throwing an instance of 'cl::Error'
what(): clGetPlatformIDs
Aborted
root@z820:/opt/rocm# tree -d
.
├── bin
├── hcc
│ ├── bin
│ ├── include
│ │ ├── clang-c
│ │ ├── experimental
│ │ │ └── impl
│ │ ├── llvm
│ │ │ └── Target
│ │ │ └── AMDGPU
│ │ │ └── Disassembler
│ │ └── llvm-c
│ ├── lib
│ │ ├── clang
│ │ │ └── 7.0.0
│ │ │ ├── include
│ │ │ │ ├── cuda_wrappers
│ │ │ │ ├── sanitizer
│ │ │ │ └── xray
│ │ │ ├── lib
│ │ │ │ └── linux
│ │ │ └── share
│ │ └── cmake
│ │ └── hcc
│ ├── libexec
│ ├── rocdl
│ │ ├── hc
│ │ ├── irif
│ │ ├── lib
│ │ ├── ockl
│ │ ├── oclc
│ │ ├── ocml
│ │ └── opencl
│ └── share
│ ├── clang
│ ├── man
│ │ └── man1
│ ├── opt-viewer
│ ├── scan-build
│ └── scan-view
├── hip
│ ├── bin
│ ├── cmake
│ │ └── FindHIP
│ ├── docs
│ │ └── docs
│ │ └── RuntimeAPI
│ │ └── html
│ │ └── search
│ ├── include
│ │ └── hip
│ │ ├── hcc_detail
│ │ │ └── cuda
│ │ └── nvcc_detail
│ ├── lib
│ │ └── cmake
│ │ └── hip
│ └── samples
│ ├── 0_Intro
│ │ ├── bit_extract
│ │ ├── hcc_dialects
│ │ ├── module_api
│ │ ├── module_api_global
│ │ └── square
│ ├── 1_Utils
│ │ ├── hipBusBandwidth
│ │ ├── hipCommander
│ │ │ └── perf
│ │ │ └── scripts
│ │ ├── hipDispatchLatency
│ │ └── hipInfo
│ └── 2_Cookbook
│ ├── 0_MatrixTranspose
│ ├── 10_inline_asm
│ ├── 11_texture_driver
│ ├── 12_cmake_hip_add_executable
│ ├── 1_hipEvent
│ ├── 2_Profiler
│ ├── 3_shared_memory
│ ├── 4_shfl
│ ├── 5_2dshfl
│ ├── 6_dynamic_shared
│ ├── 7_streams
│ ├── 8_peer2peer
│ └── 9_unroll
├── hsa
│ ├── bin
│ ├── include
│ │ └── hsa
│ ├── lib
│ └── sample
├── hsa-amd-aqlprofile
│ └── lib
├── include
│ ├── hcc -> /opt/rocm/hcc/include
│ ├── hip -> /opt/rocm/hip/include/hip
│ ├── hsa -> ../hsa/include/hsa
│ └── libhsakmt -> ../libhsakmt/include/libhsakmt
├── lib
│ └── cmake
│ ├── hcc -> /opt/rocm/hcc/lib/cmake/hcc
│ └── hip -> /opt/rocm/hip/lib/cmake/hip
├── libhsakmt
│ ├── include
│ │ └── libhsakmt
│ │ └── linux
│ └── lib
└── opencl
├── bin
│ └── x86_64
├── include
│ └── CL
└── lib
└── x86_64
└── bitcode
115 directories
root@z820:/opt/rocm#
About this issue
- Original URL
- State: closed
- Created 6 years ago
- Reactions: 1
- Comments: 29 (3 by maintainers)
@pradeepisro if you just did an install today, you may be seeing problem #510. If you’re running Ubuntu, please ensure you’re not running kernel 4.15.0-33 at the moment, as the ROCk kernel drivers are currently not working properly with it. We are working to fix this.
I would like to add to this on how I solved my problem since I’ve been here recently. Just in case somebody might come across this issue.
Using
dmesg |grep kfdI get:So the kernel driver is loaded but I’m getting error when I do
rocminfoThis one fixed my problem (it was actually my fault for using a custom kernel) I had in my kernel
HMM_MIRRORandDRM_AMDGPU_USERPTRboth disabled so I recompiled both enabled in my kernel then rebooted.HMM_MIRRORis required to enableDRM_AMDGPU_USERPTRNot sure if this one is required but I have specifically set GEN3 for PCIE in my
BIOS/UEFI. Hope it helps.I’m using kernel v5.3.7 btw