rocr: Correct gpu dumped core contents

[cfreeamd]

Includes several tests (rocrtst) for this capability.

Motivation

Technical Details

JIRA ID

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[Copilot]

Pull request overview

Updates ROCr GPU core dump generation (especially for regular files) and adds rocrtst coverage for configurable GPU core dump patterns and content validation.

Changes:

• Update core dump writer to use pwrite for regular files and adjust size-limit handling/truncation behavior.
• Add new rocrtst functional tests for GPU core dump patterns, disable flag, pipe patterns, and basic ELF/content integrity checks.
• Add a faulting-kernel test case (disabled by default) and wire new tests into the rocrtst test runner.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
projects/rocr-runtime/runtime/hsa-runtime/libamdhsacode/lnx/amd_core_dump.cpp	Adds pwrite-based emission for regular files and changes size-limit truncation logic for core dump writing.
projects/rocr-runtime/runtime/hsa-runtime/core/runtime/runtime.cpp	Adds a VM fault handler stderr print.
projects/rocr-runtime/rocrtst/suites/test_common/main.cc	Registers new GPU core dump tests and replaces the prior interrupt-disabled example test with a disabled faulting test.
projects/rocr-runtime/rocrtst/suites/functional/test_fault_example.h	Declares a new fault-inducing test case (disabled by default).
projects/rocr-runtime/rocrtst/suites/functional/test_fault_example.cc	Implements a kernel dispatch that intentionally passes null pointers to trigger a GPU fault.
projects/rocr-runtime/rocrtst/suites/functional/gpu_coredump.h	Declares a new test fixture for core dump pattern/content validation.
projects/rocr-runtime/rocrtst/suites/functional/gpu_coredump.cc	Implements multiple GPU core dump tests, including pattern matching, ELF validation, and segment checks.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The HSAResourceGuard RAII cleanup won’t run because the child uses _exit(...) on most paths, which bypasses destructors. Either avoid _exit when you want RAII cleanup (return from the function / call exit), or explicitly perform the required cleanup before _exit so resources/HSA shutdown behavior is deterministic.

[kentrussell]

No issues from me aside from style things. Will let David chime in though

[cfreeamd]

FYI, the tests pass when I run locally. But in CI, the forked processes fail to generate a fault and core, so the tests end up failing because a core isn't found.
So I put in more prints to see what is failing in the child processes, and adjusted the tests to not fail if no fault occurred (this is an error with the test, or a test machine config issue vs with ROCr). My hope is that the prints I put in will explain why no fault occurs and I can adjust.

[cfreeamd]

The new rocrtstFunc.GpuCoreDump_* tests pass in the PSDB tests, but other non-related tests are failing.
http://rocm-ci.amd.com//job/rocm-tests/38626/testReport/junit/tests/TestSuite/test_rocrtst/