diff options
| author | Heiko Carstens <heiko.carstens@de.ibm.com> | 2013-11-19 14:25:17 +0100 |
|---|---|---|
| committer | Martin Schwidefsky <schwidefsky@de.ibm.com> | 2013-11-20 09:04:55 +0100 |
| commit | dba6bb60043ed73abca8990f237db63a8cea6c50 (patch) | |
| tree | cdd10045cafcd15567c00faaa66bad74164aa507 | |
| parent | 26a35f373fbe6f21e8ad5ca4de1c01021e38fe2f (diff) | |
| download | linux-dba6bb60043ed73abca8990f237db63a8cea6c50.tar.bz2 | |
s390/mm: optimize copy_page
Always use the mvcl instruction to copy a page instead of mvpg or a
couple of mvc instructions.
Copying a huge page is 25% faster this way. Also bypass caches when
copying pages since only parts of a page will be used afterwards.
Especially when copying a huge page this would kick everything out
of the L1 and L2 data caches on a zEC12 machine.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
| -rw-r--r-- | arch/s390/include/asm/page.h | 38 |
1 files changed, 13 insertions, 25 deletions
diff --git a/arch/s390/include/asm/page.h b/arch/s390/include/asm/page.h index 316c8503a3b4..114258eeaacd 100644 --- a/arch/s390/include/asm/page.h +++ b/arch/s390/include/asm/page.h @@ -48,33 +48,21 @@ static inline void clear_page(void *page) : "memory", "cc"); } +/* + * copy_page uses the mvcl instruction with 0xb0 padding byte in order to + * bypass caches when copying a page. Especially when copying huge pages + * this keeps L1 and L2 data caches alive. + */ static inline void copy_page(void *to, void *from) { - if (MACHINE_HAS_MVPG) { - register unsigned long reg0 asm ("0") = 0; - asm volatile( - " mvpg %0,%1" - : : "a" (to), "a" (from), "d" (reg0) - : "memory", "cc"); - } else - asm volatile( - " mvc 0(256,%0),0(%1)\n" - " mvc 256(256,%0),256(%1)\n" - " mvc 512(256,%0),512(%1)\n" - " mvc 768(256,%0),768(%1)\n" - " mvc 1024(256,%0),1024(%1)\n" - " mvc 1280(256,%0),1280(%1)\n" - " mvc 1536(256,%0),1536(%1)\n" - " mvc 1792(256,%0),1792(%1)\n" - " mvc 2048(256,%0),2048(%1)\n" - " mvc 2304(256,%0),2304(%1)\n" - " mvc 2560(256,%0),2560(%1)\n" - " mvc 2816(256,%0),2816(%1)\n" - " mvc 3072(256,%0),3072(%1)\n" - " mvc 3328(256,%0),3328(%1)\n" - " mvc 3584(256,%0),3584(%1)\n" - " mvc 3840(256,%0),3840(%1)\n" - : : "a" (to), "a" (from) : "memory"); + register void *reg2 asm ("2") = to; + register unsigned long reg3 asm ("3") = 0x1000; + register void *reg4 asm ("4") = from; + register unsigned long reg5 asm ("5") = 0xb0001000; + asm volatile( + " mvcl 2,4" + : "+d" (reg2), "+d" (reg3), "+d" (reg4), "+d" (reg5) + : : "memory", "cc"); } #define clear_user_page(page, vaddr, pg) clear_page(page) |