[PATCH i-g-t v13 0/4] lib/intel_device_info: get the xe .graphics_rel from GMD_ID

[Xin Wang <x.wang@intel.com>]

This series enables IGT to retrieve the accurate IP minor version
(graphics_rel) at runtime for xe platforms with GMD_ID support,
instead of relying on hardcoded values in the PCI device table.

Background:
-----------
Current IGT uses static PCI device tables with hardcoded graphics_rel
values. For Xe2+ platforms, devices sharing the same PCI ID may have
different graphics_rel (minor version) values.

The kernel's GMD_ID query provides accurate runtime IP version
information (ip_ver_major.ip_ver_minor) which should be used for
platform-specific workarounds and feature detection.

Limitation:
-----------
The device info cache is keyed by PCI device ID. A system with multiple
GPUs sharing a device ID but reporting different IP versions is not
supported; the last probed device wins. This is considered unlikely in
practice, while the hardcoded devid -> IP version mapping fixed here
affects every such device.

Benefits:
---------
- Accurate IP version for platform-specific workarounds
- Proper feature detection for hardware variants
- Unified device info API between i915 and xe drivers
- No changes needed to existing test code

V2:
- set the graphics version info in xe_device_get() don't copy
  the whole struct of the intel_device_info
- update the graphics_rel when the info is already in the cache
V3:
- Optimize the coding style
V4:
- add braces around the else statement body
V5:
- add new struct xe_device_ipver to hold the ipver info
- separate cache map to eliminate collision (Roper, Matthew D)
- changed function name to xe_ipver_cache_lookup() to avoid
  confusion (Roper, Matthew D)
V6:
- optimize the coding style. (Summers, Stuart)
V7:
- rebased to latest mainline
V8:
- Reworked caching in intel_get_device_info(): replace the per-thread single-entry cache
  with a process-wide cache keyed by devid (igt_map + pthread mutex), with automatic
  init/fini via igt_constructor/igt_destructor.
- Dropped the xe_ipver_cache_lookup() weak-symbol plumbing and the separate xe_query-side
  ipver cache. Instead, xe_query updates the cached intel_device_info directly via
  xe_dev_update_graphics_ver() when MAIN GT exposes ip_ver_{major,minor} via GMD_ID.
V9:
- make xe_dev_update_graphics_ver() return error and guard NULL cache.
- assert in xe query path to avoid silently ignoring update failures.
V10:
- rebased to latest master.
- statically initialize the cache mutex and create the map lazily
  instead of relying on igt_constructor ordering.
- fall back to the static table entry instead of intel_generic_info
  when a cache entry cannot be allocated.
- return -ENOENT from xe_dev_update_graphics_ver() and fix coding style.
- document the devid keying and the lockless readers.
- also drop graphics_rel from WCL and NVL-P.
- reword commit messages.

Xin Wang (4):
  lib/igt_core: add igt_destructor helper macro
  lib/intel_device_info: make device info cache process-wide
  lib/intel_device_info: allow xe_query to override graphics version
  lib/intel_device_info: remove the graphics_rel from xe2+ devices

 lib/igt_core.h          |  12 ++++
 lib/intel_chipset.h     |   1 +
 lib/intel_device_info.c | 156 ++++++++++++++++++++++++++++++++++------
 lib/meson.build         |   1 +
 lib/xe/xe_query.c       |  17 +++++
 5 files changed, 165 insertions(+), 22 deletions(-)

-- 
2.43.0