[MIPS] IP28: added cache barrier to assembly routines

IP28 needs special treatment to avoid speculative accesses. gcc
takes care for .c code, but for assembly code we need to do it
manually.

This is taken from Peter Fuersts IP28 patches.

Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This commit is contained in:
Thomas Bogendoerfer 2007-11-25 11:47:56 +01:00 committed by Ralf Baechle
parent 2064ba23e5
commit 930bff8822
4 changed files with 24 additions and 0 deletions

View file

@ -199,6 +199,7 @@ FEXPORT(__copy_user)
*/
#define rem t8
R10KCBARRIER(0(ra))
/*
* The "issue break"s below are very approximate.
* Issue delays for dcache fills will perturb the schedule, as will
@ -231,6 +232,7 @@ both_aligned:
PREF( 1, 3*32(dst) )
.align 4
1:
R10KCBARRIER(0(ra))
EXC( LOAD t0, UNIT(0)(src), l_exc)
EXC( LOAD t1, UNIT(1)(src), l_exc_copy)
EXC( LOAD t2, UNIT(2)(src), l_exc_copy)
@ -272,6 +274,7 @@ EXC( LOAD t2, UNIT(2)(src), l_exc_copy)
EXC( LOAD t3, UNIT(3)(src), l_exc_copy)
SUB len, len, 4*NBYTES
ADD src, src, 4*NBYTES
R10KCBARRIER(0(ra))
EXC( STORE t0, UNIT(0)(dst), s_exc_p4u)
EXC( STORE t1, UNIT(1)(dst), s_exc_p3u)
EXC( STORE t2, UNIT(2)(dst), s_exc_p2u)
@ -287,6 +290,7 @@ less_than_4units:
beq rem, len, copy_bytes
nop
1:
R10KCBARRIER(0(ra))
EXC( LOAD t0, 0(src), l_exc)
ADD src, src, NBYTES
SUB len, len, NBYTES
@ -334,6 +338,7 @@ EXC( LDFIRST t3, FIRST(0)(src), l_exc)
EXC( LDREST t3, REST(0)(src), l_exc_copy)
SUB t2, t2, t1 # t2 = number of bytes copied
xor match, t0, t1
R10KCBARRIER(0(ra))
EXC( STFIRST t3, FIRST(0)(dst), s_exc)
beq len, t2, done
SUB len, len, t2
@ -354,6 +359,7 @@ src_unaligned_dst_aligned:
* It's OK to load FIRST(N+1) before REST(N) because the two addresses
* are to the same unit (unless src is aligned, but it's not).
*/
R10KCBARRIER(0(ra))
EXC( LDFIRST t0, FIRST(0)(src), l_exc)
EXC( LDFIRST t1, FIRST(1)(src), l_exc_copy)
SUB len, len, 4*NBYTES
@ -384,6 +390,7 @@ cleanup_src_unaligned:
beq rem, len, copy_bytes
nop
1:
R10KCBARRIER(0(ra))
EXC( LDFIRST t0, FIRST(0)(src), l_exc)
EXC( LDREST t0, REST(0)(src), l_exc_copy)
ADD src, src, NBYTES
@ -399,6 +406,7 @@ copy_bytes_checklen:
nop
copy_bytes:
/* 0 < len < NBYTES */
R10KCBARRIER(0(ra))
#define COPY_BYTE(N) \
EXC( lb t0, N(src), l_exc); \
SUB len, len, 1; \
@ -528,6 +536,7 @@ LEAF(__rmemcpy) /* a0=dst a1=src a2=len */
ADD a1, a2 # src = src + len
r_end_bytes:
R10KCBARRIER(0(ra))
lb t0, -1(a1)
SUB a2, a2, 0x1
sb t0, -1(a0)
@ -542,6 +551,7 @@ r_out:
move a2, zero
r_end_bytes_up:
R10KCBARRIER(0(ra))
lb t0, (a1)
SUB a2, a2, 0x1
sb t0, (a0)

View file

@ -86,6 +86,7 @@ FEXPORT(__bzero)
.set at
#endif
R10KCBARRIER(0(ra))
#ifdef __MIPSEB__
EX(LONG_S_L, a1, (a0), first_fixup) /* make word/dword aligned */
#endif
@ -103,11 +104,13 @@ FEXPORT(__bzero)
PTR_ADDU t1, a0 /* end address */
.set reorder
1: PTR_ADDIU a0, 64
R10KCBARRIER(0(ra))
f_fill64 a0, -64, a1, fwd_fixup
bne t1, a0, 1b
.set noreorder
memset_partial:
R10KCBARRIER(0(ra))
PTR_LA t1, 2f /* where to start */
#if LONGSIZE == 4
PTR_SUBU t1, t0
@ -129,6 +132,7 @@ memset_partial:
beqz a2, 1f
PTR_ADDU a0, a2 /* What's left */
R10KCBARRIER(0(ra))
#ifdef __MIPSEB__
EX(LONG_S_R, a1, -1(a0), last_fixup)
#endif
@ -143,6 +147,7 @@ small_memset:
PTR_ADDU t1, a0, a2
1: PTR_ADDIU a0, 1 /* fill bytewise */
R10KCBARRIER(0(ra))
bne t1, a0, 1b
sb a1, -1(a0)

View file

@ -38,6 +38,7 @@ FEXPORT(__strncpy_from_user_nocheck_asm)
.set noreorder
1: EX(lbu, t0, (v1), fault)
PTR_ADDIU v1, 1
R10KCBARRIER(0(ra))
beqz t0, 2f
sb t0, (a0)
PTR_ADDIU v0, 1

View file

@ -398,4 +398,12 @@ symbol = value
#define SSNOP sll zero, zero, 1
#ifdef CONFIG_SGI_IP28
/* Inhibit speculative stores to volatile (e.g.DMA) or invalid addresses. */
#include <asm/cacheops.h>
#define R10KCBARRIER(addr) cache Cache_Barrier, addr;
#else
#define R10KCBARRIER(addr)
#endif
#endif /* __ASM_ASM_H */