tyler.durden/bottom
1
0
Fork 0
mirror of https://github.com/ClementTsang/bottom.git synced 2025-04-08 17:05:59 +02:00
Code Issues Packages Projects Releases Wiki Activity

other: handle systems with only libnvidia-ml.so.1 (#1655)

Browse Source 
Recently, NVIDIA CUDA repository packages started shipping only
`libnvidia-ml.so.1` file, without `libnvidia-ml.so`. The upstream
`nvml-wrapper` package has a fix proposed
(https://github.com/Cldfire/nvml-wrapper/pull/63), yet the package is
in search of a maintainer at the moment.

To allow `bottom` to correctly detect NVIDIA GPUs on Ubuntu with
official NVIDIA packages, add a wrapper around `Nvml::init` to be more
persistent in its search for the NVML library.
This commit is contained in:
Andrey Alekseenko 2025-01-07 05:36:20 +01:00 committed by GitHub
parent dbda1ee56f
commit 915c25a50e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 24 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
src/data_collection/nvidia.rs
View File

@ -21,12 +21,35 @@ pub struct GpusData {
pub procs: Option<(u64, Vec<HashMap<u32, (u64, u32)>>)>,
}
/// Wrapper around Nvml::init
///
/// On Linux, if `Nvml::init()` fails, this function attempts to explicitly load
/// the library from `libnvidia-ml.so.1`. On other platforms, it simply calls `Nvml::init`.
///
/// This is a workaround until https://github.com/Cldfire/nvml-wrapper/pull/63 is accepted.
/// Then, we can go back to calling `Nvml::init` directly on all platforms.
fn init_nvml() -> Result<Nvml, NvmlError> {
#[cfg(not(target_os = "linux"))]
{
Nvml::init()
}
#[cfg(target_os = "linux")]
{
match Nvml::init() {
Ok(nvml) => Ok(nvml),
Err(_) => Nvml::builder()
.lib_path(std::ffi::OsStr::new("libnvidia-ml.so.1"))
.init(),
}
}
}
/// Returns the GPU data from NVIDIA cards.
#[inline]
pub fn get_nvidia_vecs(
temp_type: &TemperatureType, filter: &Option<Filter>, widgets_to_harvest: &UsedWidgets,
) -> Option<GpusData> {
if let Ok(nvml) = NVML_DATA.get_or_init(Nvml::init) {
if let Ok(nvml) = NVML_DATA.get_or_init(init_nvml) {
if let Ok(num_gpu) = nvml.device_count() {
let mut temp_vec = Vec::with_capacity(num_gpu as usize);
let mut mem_vec = Vec::with_capacity(num_gpu as usize);