- **Added evicted_time metric for kfd processes**.
- Time that queues are evicted on a GPU in milliseconds
- Added to CLI in `amd-smi monitor -q` and `amd-smi process`
- Added to C API and Python API:
- amdsmi_get_gpu_process_list()
- amdsmi_get_gpu_compute_process_info()
- amdsmi_get_gpu_compute_process_info_by_pid()
---------
Signed-off-by: Pryor, Adam <Adam.Pryor@amd.com>
* Updates:
- [ASAN] GCC does not support `-shared-libsan flags`, so removed this one
- [Clang] Fixed refernces to local binding errors (name collision)
& other strict scope/structure/lamda binding errors
- [Clang] Fix rsmi_wrapper error: \"error: missing default argument on parameter \'args\'\"
- [ASAN] Fixed stack-buffer-overflow found in
`amdsmi_get_gpu_accelerator_partition_profile()`
Change-Id: I854007efb75d828dbb8088c0d56dbc125081f0f2
Signed-off-by: Charis Poag <Charis.Poag@amd.com>
* [SWDEV-542718] Correct socket_affinity
Updated Socket affinity to show bitmask and expanded cpu list.
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
* Update per-device local_cpulist for socket_affinity
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
* Added amdsmi_get_cpu_affinity_from_local_cpulist API.
Updated the wrapper.
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
* Revert "Added amdsmi_get_cpu_affinity_from_local_cpulist API."
This reverts commit 9a2ef934b1787f8aa09d3e4efe02f897b4295215.
* Moved the changes to C API.
In case of SOCKET_SCOPE, use local_cpulist first.
If it is unavailable or not readable, fallback to
numa.
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
* Addressed review comments
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
---------
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
* Clarified comments regarding power limit retrieval and its support on virtualized systems.
* Change unsupported comment to UINT32_MAX
---------
Signed-off-by: josnarlo <Joseph.Narlo@amd.com>
Signed-off-by: Arif, Maisam <Maisam.Arif@amd.com>
Added the following API's to amdsmi_interface.py.
amdsmi_get_cpu_handle()
amdsmi_get_esmi_err_msg()
amdsmi_get_gpu_event_notification()
amdsmi_get_processor_count_from_handles()
amdsmi_get_processor_handles_by_type()
amdsmi_gpu_validate_ras_eeprom()
amdsmi_init_gpu_event_notification()
amdsmi_set_gpu_event_notification_mask()
amdsmi_stop_gpu_event_notification()
amdsmi_get_gpu_busy_percent()
Added additional return value to API amdsmi_get_xgmi_plpd().
The entry policies is added to the end of the dictionary to match API definition.
The entry plpds is marked for deprecation as it has the same information as policies.
---------
Signed-off-by: josnarlo <Joseph.Narlo@amd.com>
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
* Used KFD to determine linking between GPUs and PIDs rather than depend on fdinfo's per pid single gpu bdf info that we were getting.
Signed-off-by: adapryor <Adam.pryor@amd.com>
---------
Signed-off-by: adapryor <Adam.pryor@amd.com>
Signed-off-by: Arif, Maisam <Maisam.Arif@amd.com>
Having the SOVERSION derived from the git tags doesn't scale well
for distributions that don't have the git history while building
(such as a tarball).
As part of e7d6590 the strings are parsed from a header. Re-use
those.
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
- Changed amd-smi static --vbios to accept ifwi
- Change population logic for vbios version API
- Added IFWI boot_firmware to the CLI, C++, Rust, and Python API
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Change-Id: I4ea504d40a43cfb011ab38fc9a664ecf12d39c8a
Increased the AMDSMI_MAX_DEVICES to 64 to accomodate all
devices in CPX mode. The link type has been modified in
amd-smi to match with rocm-smi types, updated the same
for drm tests.
---------
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
When running on a system that doesn't support HSMP (such as an APU)
then the following is observed:
```
/usr/include/c++/15.1.1/bits/stl_vector.h:1263: std::vector<_Tp, _Alloc>::reference std::vector<_Tp, _Alloc>::operator[](size_type) [with _Tp = void*; _Alloc = std::allocator<void*>; reference = void*&; size_type = long unsigned int]: Assertion '__n < this->size()' failed.
```
This is because no "CPU" are detected on the SOC, which really means
no CPUs that support HSMP. Catch this case so that a clean return
can be passed up.
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
* Cmake fix updates
* Next fix will be addressing libdrm further
---------
Signed-off-by: adapryor <Adam.pryor@amd.com>
Signed-off-by: Justin Williams <juwillia@amd.com>
* Remove vm checks in rocm-smi
* Move virtualization checks up the stack into amd-smi
---------
Signed-off-by: adapryor <Adam.pryor@amd.com>
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Co-authored-by: Maisam Arif <Maisam.Arif@amd.com>
Changes:
- This aligns back to original struct naming for ROCm 7.0. This removes
any Major ABI breakages for updates for 7.0 release.
- Minor ABI breakage is required since there were additions to the
header. Refer to changelog for these updates.
Change-Id: If35af74eac6beac8c267d05ce789b7761ed24bff
Signed-off-by: Charis Poag <Charis.Poag@amd.com>
Changes:
- Update violation status logic and metric naming for XCP/XCC metrics (thrm/thm consistency)
- Added XCP identifier in monitor to allow partition metrics to be shown with applicable APIs
(Violation Status is the first example of this in monitor)
- Improve CLI monitor output:
support multiple GPU lines per GPU, add new columns, and better formatting
- Refactor helpers and logger for flexible unit formatting and table rendering
- Add examples for amdsmi_get_gpu_pm_metrics_info()/amdsmi_get_gpu_reg_table_info()
new metrics APIs in C++ example
- Sync Python/C++ interface and structures for new metrics fields and naming
- Remove deprecated/unused RSMI activity APIs, documentation not needed since
the APIs no longer exist in ROCm SMI either.
- Cleanup metric violations + fix handle watch arguments
- Provide better handling/doc for average_flattened_ints()
- Group xcp metrics with brackets in human readable + adjust output size
Signed-off-by: Poag, Charis <Charis.Poag@amd.com>
Description:
- Added a new API `amdsmi_gpu_driver_reload()` to reload the AMD GPU driver independently.
- Updated CLI (`sudo amd-smi reset -r`) and Python bindings to support driver reload functionality.
- Removed automatic driver reload from `amdsmi_set_gpu_memory_partition()` and `amdsmi_set_gpu_memory_partition_mode()`.
- Enhanced CLI and test cases to allow users to control when the driver reload occurs.
- Updated documentation and changelog to reflect the new driver reload process.
- Improved error handling and logging for driver reload operations.
- Added progress bar and user confirmation prompts for driver reload commands.
* Update build/test strategy to only allow one test execution at a time
* Modify API verbage + modify systemctl error output
- Systemctl is typically not enabled on docker.
- And is an edge case for gpu being active process/etc for display devices.
* Remove AMDSMI_STATUS_AMDGPU_RESTART_ERR from the return values
* Move driver reload to after we save original compute partitions
---------
Signed-off-by: Charis Poag <Charis.Poag@amd.com>
Add support to Query UBB/OAM temperature.
* Updated Python API with new temperature metrics enum
---------
Co-authored-by: Bill Liu <shuzhliu@amd.com>
Co-authored-by: gabrpham_amdeng <Gabriel.Pham@amd.com>
Moved the bit_rate and max_bandwidth back into links in the
amdsmi_link_metrics_t struct as this change was impacting
other teams. Modified the C and python API's, wrapper, and
CLI accordingly.
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
Modified the file used to fetch process name so that complete name with path can be displayed.
Changes:
amd-smi monitor -q
- human readable format will output only the process name
- csv and json formats will print the full path
amd-smi process
- name will always be the full path to the process
amd-smi (default output)
- name will always be truncated.
---------
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
The bug was reproduced like this.
In terminal #1, run command:
sudo amd-smi ras --cper --gpu 6 --severity all --folder /tmp/cper_dump --follow
In terminal #2, inject errors:
while true; do sudo amdgpuras -b 7 -s 1 -m 6 -t 2; sleep 2; done
The terminal #1 starts dumping cper entry information that it captures. After 20 entries have been captured, open terminal #3 and run same command as terminal #1:
sudo amd-smi ras --cper --gpu 6 --severity all --folder /tmp/cper_dump --follow
From terminal #3, there will be no output, even when terminal #1 continues capturing and printing information.
The fix:
Since we already have more than 20 CPER entries available in the GPU buffer, when we run the command from terminal #3 to start capturing from the beginning and pass 20 buffers to copy entries to, the C++ API returns a code saying there is more data available.
The Python CLI should not treat this as an error, but should continue to print what the API returned.
---------
Signed-off-by: Oosman Saeed <oossaeed@amd.com>
Pryor, Adam
Charis Poag
Saeed, Oosman
Kanangot Balakrishnan, Bindhiya
Narlo, Joseph
Maisam Arif
Oosman Saeed
Mario Limonciello
Stella Laurenzo
Mario Limonciello (AMD)
Bill Liu
Liu, Shuzhou (Bill)
gabrpham_amdeng