[PATCH] RAID6 Altivec fix

This patch fixes a signedness bug with RAID6 for Altivec, and makes the Altivec code testable in userspace. Signed-off-by: H. Peter Anvin <hpa@zytor.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] RAID6 Altivec fix
This patch fixes a signedness bug with RAID6 for Altivec, and makes the Altivec code testable in userspace. Signed-off-by: H. Peter Anvin <hpa@zytor.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
H. Peter Anvin · Linus Torvalds
1 parent 0d0fc3a2d6
Showing 4 changed files with 40 additions and 10 deletions Side-by-side Diff
drivers/md/raid6.h
drivers/md/raid6algos.c
drivers/md/raid6altivec.uc
drivers/md/raid6test/Makefile
@@ -69,9 +69,13 @@
 #define __init
 #define __exit
 #define __attribute_const__ __attribute__((const))
+#define noinline __attribute__((noinline))
  
 #define preempt_enable()
 #define preempt_disable()
+#define cpu_has_feature(x) 1
+#define enable_kernel_altivec()
+#define disable_kernel_altivec()
  
 #endif /* __KERNEL__ */
  
@@ -19,6 +19,7 @@
 #include "raid6.h"
 #ifndef __KERNEL__
 #include <sys/mman.h>
+#include <stdio.h>
 #endif
  
 struct raid6_calls raid6_call;
@@ -27,16 +27,20 @@
 #ifdef CONFIG_ALTIVEC
  
 #include <altivec.h>
-#include <asm/system.h>
-#include <asm/cputable.h>
+#ifdef __KERNEL__
+# include <asm/system.h>
+# include <asm/cputable.h>
+#endif
  
 /*
- * This is the C data type to use
+ * This is the C data type to use.  We use a vector of
+ * signed char so vec_cmpgt() will generate the right
+ * instruction.
  */
  
-typedef vector unsigned char unative_t;
+typedef vector signed char unative_t;
  
-#define NBYTES(x) ((vector unsigned char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
+#define NBYTES(x) ((vector signed char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
 #define NSIZE	sizeof(unative_t)
  
 /*
  
@@ -108,7 +112,11 @@
 int raid6_have_altivec(void)
 {
 	/* This assumes either all CPUs have Altivec or none does */
+# ifdef __KERNEL__
 	return cpu_has_feature(CPU_FTR_ALTIVEC);
+# else
+	return 1;
+# endif
 }
 #endif
  
@@ -8,6 +8,8 @@
 CFLAGS	 = -I.. -g $(OPTFLAGS)
 LD	 = ld
 PERL	 = perl
+AR	 = ar
+RANLIB	 = ranlib
  
 .c.o:
 	$(CC) $(CFLAGS) -c -o $@ $<
  
  
  
  
  
@@ -18,18 +20,33 @@
 %.uc: ../%.uc
 	cp -f $< $@
  
-all:	raid6.o raid6test
+all:	raid6.a raid6test
  
-raid6.o: raid6int1.o raid6int2.o raid6int4.o raid6int8.o raid6int16.o \
+raid6.a: raid6int1.o raid6int2.o raid6int4.o raid6int8.o raid6int16.o \
 	 raid6int32.o \
 	 raid6mmx.o raid6sse1.o raid6sse2.o \
+	 raid6altivec1.o raid6altivec2.o raid6altivec4.o raid6altivec8.o \
 	 raid6recov.o raid6algos.o \
 	 raid6tables.o
-	$(LD) -r -o $@ $^
+	 rm -f $@
+	 $(AR) cq $@ $^
+	 $(RANLIB) $@
  
-raid6test: raid6.o test.c
+raid6test: test.c raid6.a
 	$(CC) $(CFLAGS) -o raid6test $^
  
+raid6altivec1.c: raid6altivec.uc ../unroll.pl
+	$(PERL) ../unroll.pl 1 < raid6altivec.uc > $@
+
+raid6altivec2.c: raid6altivec.uc ../unroll.pl
+	$(PERL) ../unroll.pl 2 < raid6altivec.uc > $@
+
+raid6altivec4.c: raid6altivec.uc ../unroll.pl
+	$(PERL) ../unroll.pl 4 < raid6altivec.uc > $@
+
+raid6altivec8.c: raid6altivec.uc ../unroll.pl
+	$(PERL) ../unroll.pl 8 < raid6altivec.uc > $@
+
 raid6int1.c: raid6int.uc ../unroll.pl
 	$(PERL) ../unroll.pl 1 < raid6int.uc > $@
  
@@ -52,7 +69,7 @@
 	./mktables > raid6tables.c
  
 clean:
-	rm -f *.o mktables mktables.c raid6int.uc raid6*.c raid6test
+	rm -f *.o *.a mktables mktables.c raid6int.uc raid6*.c raid6test
  
 spotless: clean
 	rm -f *~
...	...	@@ -69,9 +69,13 @@
69	69	#define __init
70	70	#define __exit
71	71	#define __attribute_const__ __attribute__((const))
	72	+#define noinline __attribute__((noinline))
72	73
73	74	#define preempt_enable()
74	75	#define preempt_disable()
	76	+#define cpu_has_feature(x) 1
	77	+#define enable_kernel_altivec()
	78	+#define disable_kernel_altivec()
75	79
76	80	#endif /* __KERNEL__ */
77	81
...	...	@@ -19,6 +19,7 @@
19	19	#include "raid6.h"
20	20	#ifndef __KERNEL__
21	21	#include <sys/mman.h>
	22	+#include <stdio.h>
22	23	#endif
23	24
24	25	struct raid6_calls raid6_call;
...	...	@@ -27,16 +27,20 @@
27	27	#ifdef CONFIG_ALTIVEC
28	28
29	29	#include <altivec.h>
30		-#include <asm/system.h>
31		-#include <asm/cputable.h>
	30	+#ifdef __KERNEL__
	31	+# include <asm/system.h>
	32	+# include <asm/cputable.h>
	33	+#endif
32	34
33	35	/*
34		- * This is the C data type to use
	36	+ * This is the C data type to use. We use a vector of
	37	+ * signed char so vec_cmpgt() will generate the right
	38	+ * instruction.
35	39	*/
36	40
37		-typedef vector unsigned char unative_t;
	41	+typedef vector signed char unative_t;
38	42
39		-#define NBYTES(x) ((vector unsigned char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
	43	+#define NBYTES(x) ((vector signed char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
40	44	#define NSIZE sizeof(unative_t)
41	45
42	46	/*
43	47
...	...	@@ -108,7 +112,11 @@
108	112	int raid6_have_altivec(void)
109	113	{
110	114	/* This assumes either all CPUs have Altivec or none does */
	115	+# ifdef __KERNEL__
111	116	return cpu_has_feature(CPU_FTR_ALTIVEC);
	117	+# else
	118	+ return 1;
	119	+# endif
112	120	}
113	121	#endif
114	122
...	...	@@ -8,6 +8,8 @@
8	8	CFLAGS = -I.. -g $(OPTFLAGS)
9	9	LD = ld
10	10	PERL = perl
	11	+AR = ar
	12	+RANLIB = ranlib
11	13
12	14	.c.o:
13	15	$(CC) $(CFLAGS) -c -o $@ $<
14	16
15	17
16	18
17	19
18	20
...	...	@@ -18,18 +20,33 @@
18	20	%.uc: ../%.uc
19	21	cp -f $< $@
20	22
21		-all: raid6.o raid6test
	23	+all: raid6.a raid6test
22	24
23		-raid6.o: raid6int1.o raid6int2.o raid6int4.o raid6int8.o raid6int16.o \
	25	+raid6.a: raid6int1.o raid6int2.o raid6int4.o raid6int8.o raid6int16.o \
24	26	raid6int32.o \
25	27	raid6mmx.o raid6sse1.o raid6sse2.o \
	28	+ raid6altivec1.o raid6altivec2.o raid6altivec4.o raid6altivec8.o \
26	29	raid6recov.o raid6algos.o \
27	30	raid6tables.o
28		- $(LD) -r -o $@ $^
	31	+ rm -f $@
	32	+ $(AR) cq $@ $^
	33	+ $(RANLIB) $@
29	34
30		-raid6test: raid6.o test.c
	35	+raid6test: test.c raid6.a
31	36	$(CC) $(CFLAGS) -o raid6test $^
32	37
	38	+raid6altivec1.c: raid6altivec.uc ../unroll.pl
	39	+ $(PERL) ../unroll.pl 1 < raid6altivec.uc > $@
	40	+
	41	+raid6altivec2.c: raid6altivec.uc ../unroll.pl
	42	+ $(PERL) ../unroll.pl 2 < raid6altivec.uc > $@
	43	+
	44	+raid6altivec4.c: raid6altivec.uc ../unroll.pl
	45	+ $(PERL) ../unroll.pl 4 < raid6altivec.uc > $@
	46	+
	47	+raid6altivec8.c: raid6altivec.uc ../unroll.pl
	48	+ $(PERL) ../unroll.pl 8 < raid6altivec.uc > $@
	49	+
33	50	raid6int1.c: raid6int.uc ../unroll.pl
34	51	$(PERL) ../unroll.pl 1 < raid6int.uc > $@
35	52
...	...	@@ -52,7 +69,7 @@
52	69	./mktables > raid6tables.c
53	70
54	71	clean:
55		- rm -f .o mktables mktables.c raid6int.uc raid6.c raid6test
	72	+ rm -f .o .a mktables mktables.c raid6int.uc raid6*.c raid6test
56	73
57	74	spotless: clean
58	75	rm -f *~