[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250214010607.7067-1-chang.seok.bae@intel.com>
Date: Thu, 13 Feb 2025 17:06:05 -0800
From: "Chang S. Bae" <chang.seok.bae@...el.com>
To: linux-kernel@...r.kernel.org
Cc: x86@...nel.org,
tglx@...utronix.de,
mingo@...hat.com,
bp@...en8.de,
dave.hansen@...ux.intel.com,
aruna.ramakrishna@...cle.com,
TonyWWang-oc@...oxin.com,
chang.seok.bae@...el.com
Subject: [PATCH 0/2] x86/pkeys: Remove unnecessary XGETBV1 dependency in signal frame exposition
Hi all,
This series removes an unnecessary dependency on XGETBV1 which was
brought by commit
ae6012d72fa6 ("x86/pkeys: Ensure updated PKRU value is XRSTOR'd").
Currently, the PKRU exposure logic overwrites xregs_state->header
->xfeatures in the signal frame, seemingly to emulate XSAVE behavior:
combining the requested feature bitmap with a processor-tracked status
bitmap derived from XGETBV(1).
A previous posting [1] attempted to handle the case where XGETBV(1) is
disabled or unexposed. However, that approach assumed this dependency was
necessary and further modified the logic to skip PKRU writes in such
cases, which appears problematic.
Instead, the XSAVE-written value can be read directly from userspace
memory, as is already done for legacy states [2]. Thus, tying PKRU
exposure in the ABI to XGETBV1 is excessive. This series refactors
reusable code into a helper and applies it to PKRU handling, fully
eliminating the XGETBV1 dependency.
Thanks,
Chang
[1]: https://lore.kernel.org/lkml/20250102075419.2559-1-TonyWWang-oc@zhaoxin.com/
[2]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/kernel/fpu/signal.c#n139
Chang S. Bae (2):
x86/fpu: Refactor xfeature bitmask update code for sigframe XSAVE
x86/pkeys: Simplify PKRU update in signal frame
arch/x86/kernel/fpu/signal.c | 11 +----------
arch/x86/kernel/fpu/xstate.h | 21 +++++++++++++++------
2 files changed, 16 insertions(+), 16 deletions(-)
base-commit: b36de8b904b8ff2095ece7af6b3cfff8c73c2fb1
--
2.45.2
Powered by blists - more mailing lists