ACPI APEI Firmware-First Platform Error Reporting (GHES, EINJ)
ACPI's Platform Error Interface, including GHES (error source consumer) and EINJ (error injection), lets firmware on modern x86 and ARM servers forward hardware faults — memory errors, PCIe and CXL link issues, CPU machine checks — to Linux for logging and recovery. It underpins reliability features on current enterprise hardware including NVIDIA DGX and CXL servers.
recommendation
It should stay because APEI is the standard mechanism modern servers use to deliver hardware error reports (memory, PCIe, CXL) from firmware to the OS, and it is actively maintained: fixes and new features were still landing in 2026, including a dedicated NVIDIA CPER handler for DGX/HGX systems. It is core RAS infrastructure with no replacement on the horizon.
repository signals
sources
- lore.kernel.org
APEI core code is still receiving upstream fixes in 2026 ('ACPI: APEI: check return value of task_work_add to prevent memory leaks').
- lore.kernel.org
APEI/GHES is seeing new feature work in 2026, including a v3 series for a device-managed GHES notifier and NVIDIA CPER handler.
- docs.kernel.org
Current kernel documentation still documents APEI/EINJ and references modern ACPI 6.5 CXL error injection support, showing the subsystem remains relevant to current server RAS flows.
- docs.nvidia.com
NVIDIA maintains current DGX H100/H200 firmware documentation in 2026, supporting the inference that server platforms using firmware-mediated RAS paths remain in active deployment.
codex reasoning notes (technical)
`keep` because lore activity is strong and recent: lore_file_timeline on `drivers/acpi/apei/ghes.c` showed heavy churn through 2026-04-17, and lore_regex on linux-acpi subjects surfaced multiple 2026 APEI/GHES/EINJ fix/feature threads, not a removal series. docs.kernel.org was obtained via web search and shows ongoing APEI/EINJ documentation with ACPI 6.5/CXL support. NVIDIA DGX firmware docs were obtained via web search; combined with current-tree Kconfig text read via shell (GHES NVIDIA handler for DGX/HGX), that supports continued use on new server platforms. This is platform firmware/RAS infrastructure rather than a legacy single-device driver, so there is no natural replacement driver.