[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>

1 files changed, 10 insertions, 0 deletions

diff --git a/arch/mips/lib/memcpy.S b/arch/mips/lib/memcpy.S
index aded7b159052..01e450b1ebc9 100644
--- a/arch/mips/lib/memcpy.S
+++ b/arch/mips/lib/memcpy.S
@@ -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)