[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <alpine.LFD.2.21.1810020209310.20762@eddie.linux-mips.org>
Date: Tue, 2 Oct 2018 12:50:00 +0100 (BST)
From: "Maciej W. Rozycki" <macro@...ux-mips.org>
To: Ralf Baechle <ralf@...ux-mips.org>,
Paul Burton <paul.burton@...s.com>
cc: linux-mips@...ux-mips.org, linux-kernel@...r.kernel.org
Subject: [PATCH 0/2] MIPS: memset: Fix `noreorder' issues
Hi,
A recent change broke CPU_DADDI_WORKAROUNDS support in memset.S, due to a
delay-slot instruction expanding to multiple hardware operations for the
affected configurations.
The underlying cause is the excessive use of the `noreorder' assembly
mode, while it is only needed in couple of places where either there is a
data dependency between a branch and its delay slot instruction, or there
is a section switch involved that would prevent automatic delay slot
scheduling.
These changes address both problems and for clarity, not to mix multiple
conceptually separate changes and to make backporting easier I made them a
small patch series. See individual change descriptions for details.
This has been build-time and run-time verified with 32-bit and 64-bit
DECstation configurations, build-time verified with big-endian and
little-endian 64-bit SWARM configurations. Build-time verification was
made by running `objdump -d arch/mips/lib/memset.o' with a pristine and
and a patched build to make sure there has been no change in machine code
generation, except for the delay-slot multiple instruction with the 64-bit
CPU_DADDI_WORKAROUNDS DECstation configuration.
Please apply.
Maciej
Powered by blists - more mailing lists