Merge amd-staging into amd-master 20220617

Signed-off-by: Hao Zhou <Hao.Zhou@amd.com>
Change-Id: I1bfa6a012b2bfb7f744e018f129539a495c2c5db


[ROCm/rocm_smi_lib commit: 0635134df4]

projects/rocm-smi-lib/CMakeLists.txt

@@ -53,6 +53,9 @@ set(ROCM_SMI_PACKAGE rocm-smi-lib)
 project(${AMD_SMI_LIBS_TARGET})
 
 set(COMMON_PROJ_ROOT ${PROJECT_SOURCE_DIR})
+# Set default libdir to be "lib" for ROCm, distros/users can override this
+set(CMAKE_INSTALL_LIBDIR "lib" CACHE STRING "Library install directory")
+include(GNUInstallDirs)
 
 ## Verbose output.
 set(CMAKE_VERBOSE_MAKEFILE on)
@@ -149,19 +152,17 @@ option(FILE_REORG_BACKWARD_COMPATIBILITY "Enable File Reorg with backward compat
 if(FILE_REORG_BACKWARD_COMPATIBILITY)
   include(rocm_smi-backward-compat.cmake)
 endif()
-#TODO: Should use GNUInstallDirs to match distro standards
-#This need fix in subdirectory CMakefile as well
-#include(GNUInstallDirs)
-set(LIB_INSTALL_DIR  "lib")
-set(INCLUDE_INSTALL_DIR  "include")
-set(BIN_INSTALL_DIR  "bin")
+
+set(LIB_INSTALL_DIR  "${CMAKE_INSTALL_LIBDIR}")
+set(INCLUDE_INSTALL_DIR  "${CMAKE_INSTALL_INCLUDEDIR}")
+set(BIN_INSTALL_DIR  "${CMAKE_INSTALL_BINDIR}")
 
 include(CMakePackageConfigHelpers)
 
 configure_package_config_file(
     rocm_smi-config.cmake.in
     ${CMAKE_CURRENT_BINARY_DIR}/rocm_smi-config.cmake
-    INSTALL_DESTINATION ${LIB_INSTALL_DIR}/cmake/rocm_smi
+    INSTALL_DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/rocm_smi
     PATH_VARS LIB_INSTALL_DIR INCLUDE_INSTALL_DIR BIN_INSTALL_DIR
     )
 
@@ -175,7 +176,7 @@ install(
     FILES
     ${CMAKE_CURRENT_BINARY_DIR}/rocm_smi-config.cmake ${CMAKE_CURRENT_BINARY_DIR}/rocm_smi-config-version.cmake
     DESTINATION
-    lib/cmake/${ROCM_SMI}
+    ${CMAKE_INSTALL_LIBDIR}/cmake/${ROCM_SMI}
     COMPONENT dev
     )
 
@@ -190,11 +191,11 @@ export(PACKAGE rocm_smi)
 # Create the rocm_smiConfig.cmake and rocm_smiConfigVersion files
 # ... for the build tree
 install(EXPORT rocm_smiTargets DESTINATION
-  "lib/cmake/${ROCM_SMI}" COMPONENT dev)
+  "${CMAKE_INSTALL_LIBDIR}/cmake/${ROCM_SMI}" COMPONENT dev)
 
 #License file
 set(CPACK_RPM_PACKAGE_LICENSE "NCSA")
-install( FILES ${CPACK_RESOURCE_FILE_LICENSE} DESTINATION share/doc/${ROCM_SMI} RENAME LICENSE.txt)
+install( FILES ${CPACK_RESOURCE_FILE_LICENSE} DESTINATION ${CMAKE_INSTALL_DATADIR}/doc/${ROCM_SMI} RENAME LICENSE.txt)
 
 ###########################
 # Packaging directives

projects/rocm-smi-lib/DEBIAN/postinst.in

@@ -5,7 +5,7 @@ set -e
 do_ldconfig() {
   # left-hand term originates from ENABLE_LDCONFIG = ON/OFF at package build
   if [ "@ENABLE_LDCONFIG@" == "ON" ]; then
-    echo @CPACK_PACKAGING_INSTALL_PREFIX@/lib > /etc/ld.so.conf.d/x86_64-librocm_smi_lib.conf
+    echo @CPACK_PACKAGING_INSTALL_PREFIX@/@CMAKE_INSTALL_LIBDIR@ > /etc/ld.so.conf.d/x86_64-librocm_smi_lib.conf
     ldconfig
   fi
 }

projects/rocm-smi-lib/DEBIAN/prerm.in

@@ -12,7 +12,7 @@ rm_ldconfig() {
 
 rm_pyc() {
  # remove pyc file generated by python
- rm -rf @CPACK_PACKAGING_INSTALL_PREFIX@/libexec/rocm_smi/__pycache__
+ rm -rf @CPACK_PACKAGING_INSTALL_PREFIX@/@CMAKE_INSTALL_LIBEXECDIR@/rocm_smi/__pycache__
 }
 
 case "$1" in

projects/rocm-smi-lib/RPM/post.in

@@ -1,5 +1,5 @@
 # left-hand term originates from ENABLE_LDCONFIG = ON/OFF at package build
 if [ "@ENABLE_LDCONFIG@" == "ON" ]; then
-  echo -e "@CPACK_PACKAGING_INSTALL_PREFIX@/lib\n@CPACK_PACKAGING_INSTALL_PREFIX@/lib64" > /etc/ld.so.conf.d/x86_64-librocm_smi_lib.conf
+  echo -e "@CPACK_PACKAGING_INSTALL_PREFIX@/@CMAKE_INSTALL_LIBDIR@" > /etc/ld.so.conf.d/x86_64-librocm_smi_lib.conf
   ldconfig
 fi

projects/rocm-smi-lib/RPM/postun.in

@@ -4,3 +4,6 @@ if [ $1 -le 1 ] && [ "@ENABLE_LDCONFIG@" == "ON" ]; then
     rm -f /etc/ld.so.conf.d/x86_64-librocm_smi_lib.conf
     ldconfig
 fi
+
+# remove pyc file generated by python
+rm -rf @CPACK_PACKAGING_INSTALL_PREFIX@/@CMAKE_INSTALL_LIBEXECDIR@/rocm_smi/__pycache__

projects/rocm-smi-lib/oam/CMakeLists.txt

@@ -101,12 +101,12 @@ target_include_directories(${OAM_TARGET}
 ## Add the install directives for the runtime library.
 install(TARGETS ${OAM_TARGET}
           EXPORT rocm_smiTargets
-          LIBRARY DESTINATION lib
-          ARCHIVE DESTINATION lib
+          LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+          ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
           COMPONENT ${OAM_COMPONENT})
 install(FILES ${COMMON_SRC_ROOT}/oam/include/oam/oam_mapi.h
               ${COMMON_SRC_ROOT}/oam/include/oam/amd_oam.h
-                                        DESTINATION include/oam)
+                                        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/oam)
 
 # Generate Doxygen documentation
 if (DOXYGEN_FOUND)

projects/rocm-smi-lib/python_smi_tools/rocm_smi.py

@@ -509,7 +509,6 @@ def printLog(device, metricName, value):
         logstr = 'GPU[%s]\t\t: %s' % (device, metricName)
     if device is None:
         logstr = logstr[13:]
-    logging.debug(logstr)
     # Force thread safe printing
     print(logstr + '\n', end='')
 
@@ -791,11 +790,13 @@ def setClockRange(deviceList, clkType, minvalue, maxvalue, autoRespond):
     printLogSpacer(' Set Valid %s Range ' % (clkType))
     for device in deviceList:
         ret = rocmsmi.rsmi_dev_clk_range_set(device, int(minvalue), int(maxvalue), rsmi_clk_names_dict[clkType])
-        if rsmi_ret_ok(ret, device):
+        if rsmi_ret_ok(ret, device, silent=True):
             printLog(device, 'Successfully set %s from %s(MHz) to %s(MHz)' % (clkType, minvalue, maxvalue), None)
         else:
             printErrLog(device, 'Unable to set %s from %s(MHz) to %s(MHz)' % (clkType, minvalue, maxvalue))
             RETCODE = 1
+            if ret == rsmi_status_t.RSMI_STATUS_NOT_SUPPORTED:
+                printLog(device, 'Setting %s range is not supported for this device.' % (clkType), None)
 
 
 def setVoltageCurve(deviceList, point, clk, volt, autoRespond):
@@ -2043,7 +2044,7 @@ def showProfile(deviceList):
     status = rsmi_power_profile_status_t()
     for device in deviceList:
         ret = rocmsmi.rsmi_dev_power_profile_presets_get(device, 0, byref(status))
-        if rsmi_ret_ok(ret, device, 'profiles'):
+        if rsmi_ret_ok(ret, device, 'profiles', silent=False):
             binaryMaskString = str(format(status.available_profiles, '07b'))[::-1]
             bitMaskPosition = 0
             profileNumber = 0
@@ -2075,7 +2076,7 @@ def showRange(deviceList, rangeType):
     odvf = rsmi_od_volt_freq_data_t()
     for device in deviceList:
         ret = rocmsmi.rsmi_dev_od_volt_info_get(device, byref(odvf))
-        if rsmi_ret_ok(ret, device, 'od volt'):
+        if rsmi_ret_ok(ret, device, 'od volt', silent=False):
             if rangeType == 'sclk':
                 printLog(device, 'Valid sclk range: %sMhz - %sMhz' % (
                 int(odvf.curr_sclk_range.lower_bound / 1000000), int(odvf.curr_sclk_range.upper_bound / 1000000)), None)
@@ -2097,8 +2098,6 @@ def showRange(deviceList, rangeType):
                                  None)
                 else:
                     printLog(device, 'Unable to display %s range' % (rangeType), None)
-        else:
-            printLog(device, 'Unable to display %s range' % (rangeType), None)
     printLogSpacer()
 
 
@@ -2181,6 +2180,10 @@ def showSerialNumber(deviceList):
     for device in deviceList:
         sn = create_string_buffer(256)
         ret = rocmsmi.rsmi_dev_serial_number_get(device, sn, 256)
+        if not str(sn).isalnum():
+            printErrLog(device, "FRU Serial Number contains non-alphanumeric characters. FRU is likely corrupted")
+            continue
+
         if rsmi_ret_ok(ret, device) and sn.value.decode():
             printLog(device, 'Serial Number', sn.value.decode())
         else:
@@ -2311,13 +2314,11 @@ def showVoltageCurve(deviceList):
     odvf = rsmi_od_volt_freq_data_t()
     for device in deviceList:
         ret = rocmsmi.rsmi_dev_od_volt_info_get(device, byref(odvf))
-        if rsmi_ret_ok(ret, device, 'od volt'):
+        if rsmi_ret_ok(ret, device, 'od volt', silent=False):
             for position in range(3):
                 printLog(device, 'Voltage point %d: %sMhz %smV' % (
                 position, int(list(odvf.curve.vc_points)[position].frequency / 1000000),
                 int(list(odvf.curve.vc_points)[position].voltage)), None)
-        else:
-            printLog(device, 'Voltage Curve is not supported', None)
     printLogSpacer()
 
 
@@ -2807,10 +2808,10 @@ def rsmi_ret_ok(my_ret, device=None, metric=None, silent=False):
             returnString += ' %s: ' % (metric)
         returnString += '%s\t' % (err_str.value.decode())
         if not PRINT_JSON:
-            if silent:
-                logging.info('%s', returnString)
-            else:
-                logging.error('%s', returnString)
+            logging.debug('%s', returnString)
+            if not silent:
+                if my_ret in rsmi_status_verbose_err_out:
+                    printLog(device, rsmi_status_verbose_err_out[my_ret], None)
         RETCODE = my_ret
         return False
     return True

projects/rocm-smi-lib/python_smi_tools/rsmiBindings.py

@@ -69,6 +69,27 @@ class rsmi_status_t(c_int):
     RSMI_STATUS_UNKNOWN_ERROR = 0xFFFFFFFF
 
 
+#Dictionary of rsmi ret codes and it's verbose output
+rsmi_status_verbose_err_out = {
+    rsmi_status_t.RSMI_STATUS_SUCCESS: 'Operation was successful',
+    rsmi_status_t.RSMI_STATUS_INVALID_ARGS: 'Invalid arguments provided',
+    rsmi_status_t.RSMI_STATUS_NOT_SUPPORTED: 'Not supported on the given system',
+    rsmi_status_t.RSMI_STATUS_FILE_ERROR: 'Problem accessing a file',
+    rsmi_status_t.RSMI_STATUS_PERMISSION: 'Permission denied',
+    rsmi_status_t.RSMI_STATUS_OUT_OF_RESOURCES: 'Unable to acquire memory or other resource',
+    rsmi_status_t.RSMI_STATUS_INTERNAL_EXCEPTION: 'An internal exception was caught',
+    rsmi_status_t.RSMI_STATUS_INPUT_OUT_OF_BOUNDS: 'Provided input is out of allowable or safe range',
+    rsmi_status_t.RSMI_INITIALIZATION_ERROR: 'Error occured during rsmi initialization',
+    rsmi_status_t.RSMI_STATUS_NOT_YET_IMPLEMENTED: 'Requested function is not implemented on this setup',
+    rsmi_status_t.RSMI_STATUS_NOT_FOUND: 'Item searched for but not found',
+    rsmi_status_t.RSMI_STATUS_INSUFFICIENT_SIZE: 'Insufficient resources available',
+    rsmi_status_t.RSMI_STATUS_INTERRUPT: 'Interrupt occured during execution',
+    rsmi_status_t.RSMI_STATUS_UNEXPECTED_SIZE: 'Unexpected amount of data read',
+    rsmi_status_t.RSMI_STATUS_NO_DATA: 'No data found for the given input',
+    rsmi_status_t.RSMI_STATUS_UNKNOWN_ERROR: 'Unknown error occured'
+}
+
+
 class rsmi_init_flags_t(c_int):
     RSMI_INIT_FLAG_ALL_GPUS = 0x1
 

projects/rocm-smi-lib/rocm_smi-backward-compat.cmake

@@ -72,7 +72,7 @@ function(generate_wrapper_header)
     set(include_guard "${include_guard}COMGR_WRAPPER_INCLUDE_${INC_GAURD_NAME}_H")
     #set #include statement
     get_filename_component(file_name ${header_file} NAME)
-    set(include_statements "${include_statements}#include \"../../../include/${ROCM_SMI}/${file_name}\"\n")
+    set(include_statements "${include_statements}#include \"../../../${CMAKE_INSTALL_INCLUDEDIR}/${ROCM_SMI}/${file_name}\"\n")
     configure_file(${RSMI_WRAPPER_DIR}/header.hpp.in ${RSMI_WRAPPER_INC_DIR}/${file_name})
     unset(include_guard)
     unset(include_statements)
@@ -90,7 +90,7 @@ function(generate_wrapper_header)
     set(include_guard "${include_guard}COMGR_WRAPPER_INCLUDE_${INC_GAURD_NAME}_H")
     #set #include statement
     get_filename_component(file_name ${header_file} NAME)
-    set(include_statements "${include_statements}#include \"../../../include/${OAM_TARGET_NAME}/${file_name}\"\n")
+    set(include_statements "${include_statements}#include \"../../../${CMAKE_INSTALL_INCLUDEDIR}/${OAM_TARGET_NAME}/${file_name}\"\n")
     configure_file(${RSMI_WRAPPER_DIR}/header.hpp.in ${OAM_WRAPPER_INC_DIR}/${file_name})
     unset(include_guard)
     unset(include_statements)
@@ -127,7 +127,7 @@ function(create_library_symlink)
      add_custom_target(link_${file_name} ALL
                   WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
                   COMMAND ${CMAKE_COMMAND} -E create_symlink
-                  ../../lib/${file_name} ${RSMI_WRAPPER_LIB_DIR}/${file_name})
+                  ../../${CMAKE_INSTALL_LIBDIR}/${file_name} ${RSMI_WRAPPER_LIB_DIR}/${file_name})
   endforeach()
 
   file(MAKE_DIRECTORY ${OAM_WRAPPER_LIB_DIR})
@@ -155,7 +155,7 @@ function(create_library_symlink)
      add_custom_target(link_${file_name} ALL
                   WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
                   COMMAND ${CMAKE_COMMAND} -E create_symlink
-                  ../../lib/${file_name} ${OAM_WRAPPER_LIB_DIR}/${file_name})
+                  ../../${CMAKE_INSTALL_LIBDIR}/${file_name} ${OAM_WRAPPER_LIB_DIR}/${file_name})
   endforeach()
 
 endfunction()

projects/rocm-smi-lib/rocm_smi/CMakeLists.txt

@@ -108,29 +108,29 @@ file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/bin)
 add_custom_target(link-rocm-smi ALL
                   WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
                   COMMAND ${CMAKE_COMMAND} -E create_symlink
-                  ../libexec/${ROCM_SMI}/rocm_smi.py ${CMAKE_CURRENT_BINARY_DIR}/bin/rocm-smi)
+                  ../${CMAKE_INSTALL_LIBEXECDIR}/${ROCM_SMI}/rocm_smi.py ${CMAKE_CURRENT_BINARY_DIR}/bin/rocm-smi)
 
 ## Add the install directives for the runtime library.
 install(TARGETS ${ROCM_SMI_TARGET}
           EXPORT rocm_smiTargets
-          LIBRARY DESTINATION lib
-          ARCHIVE DESTINATION lib
+          LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+          ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
           COMPONENT ${ROCM_SMI_COMPONENT})
 
 install(FILES ${COMMON_SRC_ROOT}/include/rocm_smi/rocm_smi.h
-                                        DESTINATION include/rocm_smi)
+                                        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/rocm_smi)
 install(FILES ${COMMON_SRC_ROOT}/include/rocm_smi/${ROCM_SMI_TARGET}Config.h
-                                        DESTINATION include/rocm_smi)
+                                        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/rocm_smi)
 install(FILES ${COMMON_SRC_ROOT}/include/rocm_smi/kfd_ioctl.h
-                                        DESTINATION include/rocm_smi)
+                                        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/rocm_smi)
 install(FILES ${COMMON_SRC_ROOT}/python_smi_tools/rsmiBindings.py
-                                        DESTINATION libexec/${ROCM_SMI})
+                                        DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${ROCM_SMI})
 install(FILES ${COMMON_SRC_ROOT}/python_smi_tools/rocm_smi.py
        PERMISSIONS OWNER_READ OWNER_WRITE OWNER_EXECUTE GROUP_READ
                          GROUP_WRITE GROUP_EXECUTE WORLD_READ WORLD_EXECUTE
-                                        DESTINATION libexec/${ROCM_SMI})
+                                        DESTINATION ${CMAKE_INSTALL_LIBEXECDIR}/${ROCM_SMI})
 install(FILES ${CMAKE_CURRENT_BINARY_DIR}/bin/rocm-smi
-                                        DESTINATION bin)
+                                        DESTINATION ${CMAKE_INSTALL_BINDIR})
 
 # Generate Doxygen documentation
 find_package(Doxygen)
@@ -157,9 +157,9 @@ if (DOXYGEN_FOUND AND LATEX_FOUND)
 
   add_dependencies(${ROCM_SMI_TARGET} docs)
   install(FILES ${CMAKE_CURRENT_BINARY_DIR}/latex/refman.pdf
-                         DESTINATION share/doc/${ROCM_SMI} RENAME ${RSMI_MANUAL_NAME}.pdf)
+                         DESTINATION ${CMAKE_INSTALL_DATADIR}/doc/${ROCM_SMI} RENAME ${RSMI_MANUAL_NAME}.pdf)
   install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/../README.md
-                          DESTINATION share/doc/${ROCM_SMI}/)
+                          DESTINATION ${CMAKE_INSTALL_DATADIR}/doc/${ROCM_SMI}/)
 else()
   message("Doxygen or Latex is not found. Will not generate documents.")
 endif(DOXYGEN_FOUND AND LATEX_FOUND)