diff options
| author | Paul Mundt <lethal@linux-sh.org> | 2009-08-04 17:17:00 +0900 |
|---|---|---|
| committer | Paul Mundt <lethal@linux-sh.org> | 2009-08-04 17:17:00 +0900 |
| commit | d14d751ff9234595639a16e53b3cf0c575946bde (patch) | |
| tree | 7179ac038a54d781abc985263705d2ebc2d87919 /arch/sh/lib64 | |
| parent | c7914834ef3b8a396b7e82ea34ac07cdcfe6f868 (diff) | |
| download | kernel_samsung_crespo-d14d751ff9234595639a16e53b3cf0c575946bde.zip kernel_samsung_crespo-d14d751ff9234595639a16e53b3cf0c575946bde.tar.gz kernel_samsung_crespo-d14d751ff9234595639a16e53b3cf0c575946bde.tar.bz2 | |
sh64: Kill off special clear_page() implementation.
This can use the now generic clear_page() implementation, which is backed
by the sh64 optimized memset routine. This also fixes up the case where
PAGE_SIZE != 4kB.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Diffstat (limited to 'arch/sh/lib64')
| -rw-r--r-- | arch/sh/lib64/Makefile | 2 | ||||
| -rw-r--r-- | arch/sh/lib64/clear_page.S | 54 |
2 files changed, 1 insertions, 55 deletions
diff --git a/arch/sh/lib64/Makefile b/arch/sh/lib64/Makefile index 334bb2d..1fee75a 100644 --- a/arch/sh/lib64/Makefile +++ b/arch/sh/lib64/Makefile @@ -11,7 +11,7 @@ # Panic should really be compiled as PIC lib-y := udelay.o dbg.o panic.o memcpy.o memset.o \ - copy_user_memcpy.o copy_page.o clear_page.o strcpy.o strlen.o + copy_user_memcpy.o copy_page.o strcpy.o strlen.o # Extracted from libgcc lib-y += udivsi3.o udivdi3.o sdivsi3.o diff --git a/arch/sh/lib64/clear_page.S b/arch/sh/lib64/clear_page.S deleted file mode 100644 index 007ab48..0000000 --- a/arch/sh/lib64/clear_page.S +++ /dev/null @@ -1,54 +0,0 @@ -/* - Copyright 2003 Richard Curnow, SuperH (UK) Ltd. - - This file is subject to the terms and conditions of the GNU General Public - License. See the file "COPYING" in the main directory of this archive - for more details. - - Tight version of memset for the case of just clearing a page. It turns out - that having the alloco's spaced out slightly due to the increment/branch - pair causes them to contend less for access to the cache. Similarly, - keeping the stores apart from the allocos causes less contention. => Do two - separate loops. Do multiple stores per loop to amortise the - increment/branch cost a little. - - Parameters: - r2 : source effective address (start of page) - - Always clears 4096 bytes. - - Note : alloco guarded by synco to avoid TAKum03020 erratum - -*/ - - .section .text..SHmedia32,"ax" - .little - - .balign 8 - .global clear_page -clear_page: - pta/l 1f, tr1 - pta/l 2f, tr2 - ptabs/l r18, tr0 - - movi 4096, r7 - add r2, r7, r7 - add r2, r63, r6 -1: - alloco r6, 0 - synco ! TAKum03020 - addi r6, 32, r6 - bgt/l r7, r6, tr1 - - add r2, r63, r6 -2: - st.q r6, 0, r63 - st.q r6, 8, r63 - st.q r6, 16, r63 - st.q r6, 24, r63 - addi r6, 32, r6 - bgt/l r7, r6, tr2 - - blink tr0, r63 - - |