linux-kernel - [PATCH v3 0/3] ACPI: APEI: GHES: Performance improvements for error notification handlers

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <20260112032239.30023-1-xueshuai@linux.alibaba.com>
Date: Mon, 12 Jan 2026 11:22:36 +0800
From: Shuai Xue <xueshuai@...ux.alibaba.com>
To: tony.luck@...el.com,
	guohanjun@...wei.com,
	mchehab@...nel.org,
	dolinux.peng@...il.com,
	yazen.ghannam@....com,
	rafael@...nel.org
Cc: dave.jiang@...el.com,
	Smita.KoralahalliChannabasappa@....com,
	leitao@...ian.org,
	pengdonglin@...omi.com,
	xueshuai@...ux.alibaba.com,
	baolin.wang@...ux.alibaba.com,
	benjamin.cheatham@....com,
	bp@...en8.de,
	dan.j.williams@...el.com,
	james.morse@....com,
	lenb@...nel.org,
	linux-acpi@...r.kernel.org,
	linux-kernel@...r.kernel.org,
	zhuo.song@...ux.alibaba.com
Subject: [PATCH v3 0/3] ACPI: APEI: GHES: Performance improvements for error notification handlers

changes since v2:
- Use `guard(rcu)()` instead of explicit `rcu_read_lock()`/`rcu_read_unlock()` per Donglin Peng

changes since v1:
- add Tested-by and Reviewed-by tags from Tony
- change return value from AE_BAD_ADDRESS to -EINVAL ghes_map_error_status per Hanjun
- remove unnecessary blank lines per Hanjun

This patch series improves the performance of GHES error notification handlers
(NMI and SEA) by optimizing how they check for active error conditions.

Currently, both ghes_notify_nmi() and ghes_notify_sea() perform expensive
operations on each invocation to determine if there are actual error records
to process. This includes mapping/unmapping physical addresses and accessing
hardware registers, which causes significant overhead especially on systems
with many cores.

The optimizations introduced in this series:
1. Pre-map error status registers during initialization
2. Directly check for active errors using mapped virtual addresses
3. Extract common functionality into reusable helper functions
4. Apply the same optimization to both NMI and SEA handlers

These changes significantly reduce the overhead of error checking:
- NMI handler: From ~15,000 TSC cycles to ~900 cycles
- SEA handler: From 8,138.3 ns to a much faster check

The initial idea for this optimization came from Tony Luck [1], who identified
and implemented the approach for the NMI handler. This series extends the
same concept to the SEA handler and refactors common code into shared helpers.

Patch 1 (Tony Luck): Improves ghes_notify_nmi() status check by pre-mapping
                     error status registers and avoiding repeated mappings.

Patch 2 (Shuai Xue): Extracts common helper functions for error status handling
                     to eliminate code duplication.

Patch 3 (Shuai Xue): Applies the same optimization to ghes_notify_sea() to improve
                     ARMv8 system performance.

https://lore.kernel.org/lkml/20251103230547.8715-1-tony.luck@intel.com/

Shuai Xue (2):
  ACPI: APEI: GHES: Extract helper functions for error status handling
  ACPI: APEI: GHES: Improve ghes_notify_sea() status check

Tony Luck (1):
  ACPI: APEI: GHES: Improve ghes_notify_nmi() status check

 drivers/acpi/apei/ghes.c | 107 ++++++++++++++++++++++++++++++++++++---
 include/acpi/ghes.h      |   1 +
 2 files changed, 102 insertions(+), 6 deletions(-)

-- 
2.39.3