fix: failed on hygon detection

Signed-off-by: thxCode <thxcode0824@gmail.com>

Author: thxCode

gpustack_runtime/detector/hygon.py

@@ -2,6 +2,7 @@
 
 import logging
 from functools import lru_cache
+from pathlib import Path
 
 from .. import envs
 from . import pyrocmsmi
@@ -77,7 +78,11 @@ def detect(self) -> Devices | None:
         try:
             pyrocmsmi.rsmi_init()
 
-            sys_driver_ver = pyrocmsmi.rsmi_driver_version_get()
+            sys_driver_ver = None
+            sys_driver_ver_path = Path("/sys/module/hydcu/version")
+            if sys_driver_ver_path.exists():
+                with sys_driver_ver_path.open(encoding="utf-8") as f:
+                    sys_driver_ver = f.read().strip()
 
             devs_count = pyrocmsmi.rsmi_num_monitor_devices()
             for dev_idx in range(devs_count):

gpustack_runtime/detector/pyamdsmi/__init__.py

@@ -33,9 +33,9 @@ def amdsmi_get_rocm_original_version() -> str | None:
     locs = [
         "librocm-core.so",
     ]
-    rocm_path = Path(os.getenv("ROCM_PATH", os.getenv("ROCM_HOME") or "/opt/rocm"))
+    rocm_path = Path(os.getenv("ROCM_HOME", os.getenv("ROCM_PATH") or "/opt/rocm"))
     if rocm_path.exists():
-        locs.append(str(rocm_path / "lib/librocm-core.so"))
+        locs.append(str(rocm_path / "lib" / "librocm-core.so"))
     for loc in locs:
         try:
             clib = CDLL(loc)

gpustack_runtime/detector/pyrocmsmi/__init__.py

@@ -20,15 +20,31 @@
 libLoadLock = threading.Lock()
 
 if rocmsmiLib is None:
-    rocm_path = Path(os.getenv("ROCM_PATH", os.getenv("ROCM_HOME") or "/opt/rocm"))
-
+    # Example ROCM_SMI_LIB_PATH
+    # - /opt/dtk-24.04.3/rocm_smi/lib
+    # - /opt/rocm/rocm_smi/lib
     rocmsmi_lib_path = os.getenv("ROCM_SMI_LIB_PATH")
     if not rocmsmi_lib_path:
-        rocmsmi_lib_path = str(rocm_path / "lib")
+        # Example ROCM_PATH/ROCM_HOME
+        # - /opt/dtk-24.04.3
+        # - /opt/rocm
+        rocm_path = Path(os.getenv("ROCM_HOME", os.getenv("ROCM_PATH") or "/opt/rocm"))
+        rocmsmi_lib_path = str(rocm_path / "rocm_smi" / "lib")
+    else:
+        rocm_path = Path(rocmsmi_lib_path).parent.parent
 
     rocmsmi_lib_loc = Path(rocmsmi_lib_path) / "librocm_smi64.so"
     if rocmsmi_lib_loc.exists():
-        rocmsmi_bindings_path = rocm_path / "libexec" / "rocm_smi"
+        rocmsmi_bindings_paths = [
+            (rocm_path / "rocm_smi" / "bindings"),
+            (rocm_path / "libexec" / "rocm_smi"),
+        ]
+        rocmsmi_bindings_path = None
+        for p in rocmsmi_bindings_paths:
+            if p.exists():
+                rocmsmi_bindings_path = p
+                break
+
         if rocmsmi_bindings_path.exists():
             if str(rocmsmi_bindings_path) not in sys.path:
                 sys.path.append(str(rocmsmi_bindings_path))
@@ -197,18 +213,21 @@ def rsmi_dev_target_graphics_version_get(device=0):
     if not rocmsmiLib:
         raise ROCMSMIError(ROCMSMI_ERROR_UNINITIALIZED)
 
-    c_version = c_uint64()
-    ret = rocmsmiLib.rsmi_dev_target_graphics_version_get(device, byref(c_version))
-    _rocmsmiCheckReturn(ret)
-    version = str(c_version.value)
-    if len(version) == 4:
-        dev_name = rsmi_dev_name_get(device)
-        if "Instinct MI2" in dev_name:
-            hex_part = str(hex(int(version[2:]))).replace("0x", "")
-            version = version[:2] + hex_part
-    else:
-        version = str(c_version.value // 10 + c_version.value % 10)
-    return "gfx" + version
+    try:
+        c_version = c_uint64()
+        ret = rocmsmiLib.rsmi_dev_target_graphics_version_get(device, byref(c_version))
+        _rocmsmiCheckReturn(ret)
+        version = str(c_version.value)
+        if len(version) == 4:
+            dev_name = rsmi_dev_name_get(device)
+            if "Instinct MI2" in dev_name:
+                hex_part = str(hex(int(version[2:]))).replace("0x", "")
+                version = version[:2] + hex_part
+        else:
+            version = str(c_version.value // 10 + c_version.value % 10)
+        return "gfx" + version
+    except AttributeError:
+        return None
 
 
 def rsmi_dev_temp_metric_get(device=0, sensor=None, metric=None):
@@ -240,17 +259,32 @@ def rsmi_dev_power_cap_get(device=0):
     return c_power_cap.value // 1000000
 
 
-def rsmi_dev_power_get(device=0):
+def rsmi_dev_power_ave_get(device=0):
     if not rocmsmiLib:
         raise ROCMSMIError(ROCMSMI_ERROR_UNINITIALIZED)
 
+    c_device_chip = c_uint32(0)
     c_power = c_uint64()
-    c_power_type = rsmi_power_type_t()
-    ret = rocmsmiLib.rsmi_dev_power_get(device, byref(c_power), byref(c_power_type))
+    ret = rocmsmiLib.rsmi_dev_power_ave_get(device, c_device_chip, byref(c_power))
     _rocmsmiCheckReturn(ret)
     return c_power.value // 1000000
 
 
+def rsmi_dev_power_get(device=0):
+    if not rocmsmiLib:
+        raise ROCMSMIError(ROCMSMI_ERROR_UNINITIALIZED)
+
+    try:
+        c_power = c_uint64()
+        c_power_type = rsmi_power_type_t()
+        ret = rocmsmiLib.rsmi_dev_power_get(device, byref(c_power), byref(c_power_type))
+        _rocmsmiCheckReturn(ret)
+        return c_power.value // 1000000
+    except NameError:
+        # Fallback for older versions without rsmi_dev_power_get
+        return rsmi_dev_power_ave_get(device)
+
+
 def rsmi_dev_node_id_get(device=0):
     if not rocmsmiLib:
         raise ROCMSMIError(ROCMSMI_ERROR_UNINITIALIZED)

pack/Dockerfile

@@ -244,14 +244,6 @@ ARG TARGETARCH
 VOLUME /var/lib/gpustack
 ENV GPUSTACK_RUNTIME_DOCKER_EPHEMERAL_FILES_DIR="/var/lib/gpustack/cache/gpustack-runtime"
 
-## Update
-RUN <<EOF
-    # Update
-
-    # Update PCI IDs database for better device name detection.
-    curl -o /usr/share/misc/pci.ids https://pci-ids.ucw.cz/v2.2/pci.ids
-EOF
-
 ## Install
 
 RUN --mount=type=cache,target=/root/.cache \
@@ -269,6 +261,9 @@ RUN --mount=type=cache,target=/root/.cache \
     uv pip install \
         /workspace/runtime
     uv pip tree
+
+    # Update PCI IDs
+    update-pciids || true
 EOF
 
 ## Entrypoint
@@ -279,7 +274,6 @@ EOF
 ## Options:
 ## - Mount /sys from the host to detect the correct devices' PCI info.
 ## - Mount /opt/rocm from the host to detect the correct ROCm version.
-## - Mount /usr/share/misc/pci.ids from the host to detect the correct device name.
 ## E.g. docker run --rm -it --privileged -v /var/run/docker.sock:/var/run/docker.sock gpustack/runtime:main gpustack-runtime detect --format json
 RUN --mount=type=bind,from=rocm-base,source=/opt/rocm/share,target=/opt/rocm/share,rw <<EOF
     # Reinstall amd-smi
@@ -291,7 +285,6 @@ RUN --mount=type=bind,from=rocm-base,source=/opt/rocm/share,target=/opt/rocm/sha
     uv pip tree
 EOF
 ENV AMD_VISIBLE_DEVICES="0" \
-    ROCM_HOME="/opt/rocm" \
     GPUSTACK_RUNTIME_DEPLOY_MIRRORED_DEPLOYMENT_IGNORE_VOLUMES="/opt/rocm"
 
 ## Active all Ascend devices detection,
@@ -321,7 +314,8 @@ ENV CAMBRICON_VISIBLE_DEVICES="0"
 ## See https://github.com/Project-HAMi/dcu-dcgm/blob/master/pkg/dcgm/include/rocm_smi.h.
 ## Options:
 ## - Mount /sys from the host to detect the correct devices' PCI info.
-## E.g. docker run --rm -it --privileged -v /var/run/docker.sock:/var/run/docker.sock gpustack/runtime:main gpustack-runtime detect --format json
+## - Mount /usr/share/hwdata and /usr/share/misc from the host to detect the correct device name.
+## E.g. docker run --rm -it --privileged -v /var/run/docker.sock:/var/run/docker.sock -v /opt/hyhal:/opt/hyhal -v /opt/dtk:/opt/dtk -e ROCM_PATH=/opt/dtk gpustack/runtime:main gpustack-runtime detect --format json
 ENV HYGON_VISIBLE_DEVICES="0" \
     GPUSTACK_RUNTIME_DEPLOY_MIRRORED_DEPLOYMENT_IGNORE_VOLUMES="/opt/dtk;${GPUSTACK_RUNTIME_DEPLOY_MIRRORED_DEPLOYMENT_IGNORE_VOLUMES}"