alpha: fix alignment problem in csum_ipv6_magic()
Hopefully this fixes http://bugzilla.kernel.org/show_bug.cgi?id=8635
The struct in6_addr passed to csum_ipv6_magic() is 4 byte aligned, so we
can't use the regular 64-bit loads. Since the cost of handling of 4 byte
and 1 byte aligned 64-bit data is roughly the same, this code can cope with
any src/dst [mis]alignment.
Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Dustin Marquess <jailbird@alcatraz.fdf.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/arch/alpha/lib/csum_ipv6_magic.S b/arch/alpha/lib/csum_ipv6_magic.S
index e09748d..2c2acb9 100644
--- a/arch/alpha/lib/csum_ipv6_magic.S
+++ b/arch/alpha/lib/csum_ipv6_magic.S
@@ -7,6 +7,9 @@
* __u32 len,
* unsigned short proto,
* unsigned int csum);
+ *
+ * Misalignment handling (which costs 16 instructions / 8 cycles)
+ * added by Ivan Kokshaysky <ink@jurassic.park.msu.ru>
*/
.globl csum_ipv6_magic
@@ -16,37 +19,57 @@
csum_ipv6_magic:
.prologue 0
- ldq $0,0($16) # e0 : load src & dst addr words
+ ldq_u $0,0($16) # e0 : load src & dst addr words
zapnot $20,15,$20 # .. e1 : zero extend incoming csum
extqh $18,1,$4 # e0 : byte swap len & proto while we wait
- ldq $1,8($16) # .. e1 :
+ ldq_u $21,7($16) # .. e1 : handle misalignment
extbl $18,1,$5 # e0 :
- ldq $2,0($17) # .. e1 :
+ ldq_u $1,8($16) # .. e1 :
extbl $18,2,$6 # e0 :
- ldq $3,8($17) # .. e1 :
+ ldq_u $22,15($16) # .. e1 :
extbl $18,3,$18 # e0 :
+ ldq_u $2,0($17) # .. e1 :
sra $4,32,$4 # e0 :
+ ldq_u $23,7($17) # .. e1 :
+
+ extql $0,$16,$0 # e0 :
+ ldq_u $3,8($17) # .. e1 :
+ extqh $21,$16,$21 # e0 :
+ ldq_u $24,15($17) # .. e1 :
+
sll $5,16,$5 # e0 :
+ or $0,$21,$0 # .. e1 : 1st src word complete
+ extql $1,$16,$1 # e0 :
addq $20,$0,$20 # .. e1 : begin summing the words
- sll $6,8,$6 # e0 :
+ extqh $22,$16,$22 # e0 :
cmpult $20,$0,$0 # .. e1 :
- extwh $19,7,$7 # e0 :
+ sll $6,8,$6 # e0 :
+ or $1,$22,$1 # .. e1 : 2nd src word complete
+
+ extql $2,$17,$2 # e0 :
or $4,$18,$18 # .. e1 :
-
- extbl $19,1,$19 # e0 :
+ extqh $23,$17,$23 # e0 :
or $5,$6,$5 # .. e1 :
- or $18,$5,$18 # e0 : len complete
- or $19,$7,$19 # .. e1 :
- sll $19,48,$19 # e0 :
+ extql $3,$17,$3 # e0 :
+ or $2,$23,$2 # .. e1 : 1st dst word complete
+ extqh $24,$17,$24 # e0 :
+ or $18,$5,$18 # .. e1 : len complete
+
+ extwh $19,7,$7 # e0 :
+ or $3,$24,$3 # .. e1 : 2nd dst word complete
+ extbl $19,1,$19 # e0 :
addq $20,$1,$20 # .. e1 :
- sra $19,32,$19 # e0 : proto complete
- cmpult $20,$1,$1 # .. e1 :
- nop # e0 :
+ or $19,$7,$19 # e0 :
+ cmpult $20,$1,$1 # .. e1 :
+ sll $19,48,$19 # e0 :
+ nop # .. e0 :
+
+ sra $19,32,$19 # e0 : proto complete
addq $20,$2,$20 # .. e1 :
cmpult $20,$2,$2 # e0 :
addq $20,$3,$20 # .. e1 :
@@ -84,7 +107,7 @@
extwl $0,2,$1 # e0 : fold 17-bit value
zapnot $0,3,$0 # .. e1 :
addq $0,$1,$0 # e0 :
- not $0,$0 # e1 : and complement.
+ not $0,$0 # .. e1 : and complement.
zapnot $0,3,$0 # e0 :
ret # .. e1 :
