Merge with Linux 2.4.0-test10.

4 files changed, 146 insertions, 54 deletions

diff --git a/arch/ia64/lib/Makefile b/arch/ia64/lib/Makefile
index 318e314cc..90e697179 100644
--- a/arch/ia64/lib/Makefile
+++ b/arch/ia64/lib/Makefile
@@ -3,30 +3,49 @@
 #
 
 .S.o:
-	$(CC) $(AFLAGS) -c $< -o $@
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c $< -o $@
 
 L_TARGET = lib.a
 
-L_OBJS  = __divdi3.o __udivdi3.o __moddi3.o __umoddi3.o			\
-	checksum.o clear_page.o csum_partial_copy.o copy_page.o		\
-	copy_user.o clear_user.o memcpy.o memset.o strncpy_from_user.o	\
-	strlen.o strlen_user.o strnlen_user.o				\
+L_OBJS  = __divsi3.o __udivsi3.o __modsi3.o __umodsi3.o					\
+	__divdi3.o __udivdi3.o __moddi3.o __umoddi3.o					\
+	checksum.o clear_page.o csum_partial_copy.o copy_page.o				\
+	copy_user.o clear_user.o strncpy_from_user.o strlen_user.o strnlen_user.o	\
 	flush.o do_csum.o
 
+ifneq ($(CONFIG_ITANIUM_ASTEP_SPECIFIC),y)
+  L_OBJS += memcpy.o memset.o strlen.o
+endif
+
 LX_OBJS = io.o
 
-IGNORE_FLAGS_OBJS = __divdi3.o __udivdi3.o __moddi3.o __umoddi3.o
+IGNORE_FLAGS_OBJS =	__divsi3.o __udivsi3.o __modsi3.o __umodsi3.o \
+			__divdi3.o __udivdi3.o __moddi3.o __umoddi3.o
 
-include $(TOPDIR)/Rules.make
+$(L_TARGET):
+
+__divdi3.o: idiv64.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -o $@ $<
+
+__udivdi3.o: idiv64.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -DUNSIGNED -c -o $@ $<
 
-__divdi3.o: idiv.S
-	$(CC) $(AFLAGS) -c -o $@ $<
+__moddi3.o: idiv64.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -DMODULO -c -o $@ $<
 
-__udivdi3.o: idiv.S
-	$(CC) $(AFLAGS) -c -DUNSIGNED -c -o $@ $<
+__umoddi3.o: idiv64.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -DMODULO -DUNSIGNED -c -o $@ $<
 
-__moddi3.o: idiv.S
-	$(CC) $(AFLAGS) -c -DMODULO -c -o $@ $<
+__divsi3.o: idiv32.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -o $@ $<
 
-__umoddi3.o: idiv.S
-	$(CC) $(AFLAGS) -c -DMODULO -DUNSIGNED -c -o $@ $<
+__udivsi3.o: idiv32.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -DUNSIGNED -c -o $@ $<
+
+__modsi3.o: idiv32.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -DMODULO -c -o $@ $<
+
+__umodsi3.o: idiv32.S
+	$(CC) $(AFLAGS) $(AFLAGS_KERNEL) -c -DMODULO -DUNSIGNED -c -o $@ $<
+
+include $(TOPDIR)/Rules.make
diff --git a/arch/ia64/lib/idiv32.S b/arch/ia64/lib/idiv32.S
new file mode 100644
index 000000000..13f4608c3
--- /dev/null
+++ b/arch/ia64/lib/idiv32.S
@@ -0,0 +1,83 @@
+/*
+ * Copyright (C) 2000 Hewlett-Packard Co
+ * Copyright (C) 2000 David Mosberger-Tang <davidm@hpl.hp.com>
+ *
+ * 32-bit integer division.
+ *
+ * This code is based on the application note entitled "Divide, Square Root
+ * and Remainder Algorithms for the IA-64 Architecture".  This document
+ * is available as Intel document number 248725-002 or via the web at
+ * http://developer.intel.com/software/opensource/numerics/
+ *
+ * For more details on the theory behind these algorithms, see "IA-64
+ * and Elementary Functions" by Peter Markstein; HP Professional Books
+ * (http://www.hp.com/go/retailbooks/)
+ */
+
+#include <asm/asmmacro.h>
+
+#ifdef MODULO
+# define OP	mod
+#else
+# define OP	div
+#endif
+
+#ifdef UNSIGNED
+# define SGN	u
+# define EXTEND	zxt4
+# define INT_TO_FP(a,b)	fcvt.xuf.s1 a=b
+# define FP_TO_INT(a,b)	fcvt.fxu.trunc.s1 a=b
+#else
+# define SGN
+# define EXTEND	sxt4
+# define INT_TO_FP(a,b)	fcvt.xf a=b
+# define FP_TO_INT(a,b)	fcvt.fx.trunc.s1 a=b
+#endif
+
+#define PASTE1(a,b)	a##b
+#define PASTE(a,b)	PASTE1(a,b)
+#define NAME		PASTE(PASTE(__,SGN),PASTE(OP,si3))
+
+GLOBAL_ENTRY(NAME)
+	.regstk 2,0,0,0
+	// Transfer inputs to FP registers.
+	mov r2 = 0xffdd			// r2 = -34 + 65535 (fp reg format bias)
+	EXTEND in0 = in0		// in0 = a
+	EXTEND in1 = in1		// in1 = b
+	;;
+	setf.sig f8 = in0
+	setf.sig f9 = in1
+#ifdef MODULO
+	sub in1 = r0, in1		// in1 = -b
+#endif
+	;;
+	// Convert the inputs to FP, to avoid FP software-assist faults.
+	INT_TO_FP(f8, f8)
+	INT_TO_FP(f9, f9)
+	;;
+	setf.exp f7 = r2		// f7 = 2^-34
+	frcpa.s1 f6, p6 = f8, f9	// y0 = frcpa(b)
+	;;
+(p6)	fmpy.s1 f8 = f8, f6		// q0 = a*y0
+(p6)	fnma.s1 f6 = f9, f6, f1		// e0 = -b*y0 + 1 
+	;;
+#ifdef MODULO
+	setf.sig f9 = in1		// f9 = -b
+#endif
+(p6)	fma.s1 f8 = f6, f8, f8		// q1 = e0*q0 + q0
+(p6)	fma.s1 f6 = f6, f6, f7		// e1 = e0*e0 + 2^-34
+	;;
+#ifdef MODULO
+	setf.sig f7 = in0
+#endif
+(p6)	fma.s1 f6 = f6, f8, f8		// q2 = e1*q1 + q1
+	;;
+	FP_TO_INT(f6, f6)		// q = trunc(q2)
+	;;
+#ifdef MODULO
+	xma.l f6 = f6, f9, f7		// r = q*(-b) + a
+	;;
+#endif
+	getf.sig r8 = f6		// transfer result to result register
+	br.ret.sptk rp
+END(NAME)
diff --git a/arch/ia64/lib/idiv.S b/arch/ia64/lib/idiv64.S
index da96863d3..db7d2284e 100644
--- a/arch/ia64/lib/idiv.S
+++ b/arch/ia64/lib/idiv64.S
@@ -1,25 +1,21 @@
 /*
- * Integer division routine.
- *
  * Copyright (C) 1999-2000 Hewlett-Packard Co
  * Copyright (C) 1999-2000 David Mosberger-Tang <davidm@hpl.hp.com>
- */
-
-#include <asm/asmmacro.h>
-
-/*
- * Compute a 64-bit unsigned integer quotient.
  *
- * Use reciprocal approximation and Newton-Raphson iteration to compute the
- * quotient.  frcpa gives 8.6 significant bits, so we need 3 iterations
- * to get more than the 64 bits of precision that we need for DImode.
+ * 64-bit integer division.
  *
- * Must use max precision for the reciprocal computations to get 64 bits of
- * precision.
+ * This code is based on the application note entitled "Divide, Square Root
+ * and Remainder Algorithms for the IA-64 Architecture".  This document
+ * is available as Intel document number 248725-002 or via the web at
+ * http://developer.intel.com/software/opensource/numerics/
  *
- * r32 holds the dividend.  r33 holds the divisor.
+ * For more details on the theory behind these algorithms, see "IA-64
+ * and Elementary Functions" by Peter Markstein; HP Professional Books
+ * (http://www.hp.com/go/retailbooks/)
  */
 
+#include <asm/asmmacro.h>
+
 #ifdef MODULO
 # define OP	mod
 #else
@@ -59,40 +55,38 @@ GLOBAL_ENTRY(NAME)
 	UNW(.body)
 	INT_TO_FP(f9, f9)
 	;;
-	frcpa.s1 f17, p6 = f8, f9	// y = frcpa(b)
+	frcpa.s1 f17, p6 = f8, f9	// y0 = frcpa(b)
 	;;
-	/*
-	 * This is the magic algorithm described in Section 8.6.2 of "IA-64
-	 * and Elementary Functions" by Peter Markstein; HP Professional Books
-	 * (http://www.hp.com/go/retailbooks/)
-	 */
-(p6)	fmpy.s1 f7 = f8, f17		// q = a*y
-(p6)	fnma.s1 f6 = f9, f17, f1	// e = -b*y + 1 
+(p6)	fmpy.s1 f7 = f8, f17		// q0 = a*y0
+(p6)	fnma.s1 f6 = f9, f17, f1	// e0 = -b*y0 + 1 
 	;;
-(p6)	fma.s1 f16 = f7, f6, f7		// q1 = q*e + q
-(p6)	fmpy.s1 f7 = f6, f6		// e1 = e*e
+(p6)	fma.s1 f16 = f7, f6, f7		// q1 = q0*e0 + q0
+(p6)	fmpy.s1 f7 = f6, f6		// e1 = e0*e0
 	;;
+#ifdef MODULO
+	sub in1 = r0, in1		// in1 = -b
+#endif
 (p6)	fma.s1 f16 = f16, f7, f16	// q2 = q1*e1 + q1
-(p6)	fma.s1 f6 = f17, f6, f17	// y1 = y*e + y 
+(p6)	fma.s1 f6 = f17, f6, f17	// y1 = y0*e0 + y0
 	;;
 (p6)	fma.s1 f6 = f6, f7, f6		// y2 = y1*e1 + y1
 (p6)	fnma.s1 f7 = f9, f16, f8	// r = -b*q2 + a 
 	;;
-(p6)	fma.s1 f17 = f7, f6, f16	// q3 = r*y2 + q2
-	;;
 #ifdef MODULO
-	FP_TO_INT(f17, f17)		// round quotient to an unsigned integer
-	;;
-	INT_TO_FP(f17, f17)		// renormalize
-	;;
-	fnma.s1 f17 = f17, f9, f8	// compute remainder
-	;;
+	setf.sig f8 = in0		// f8 = a
+	setf.sig f9 = in1		// f9 = -b
 #endif
+(p6)	fma.s1 f17 = f7, f6, f16	// q3 = r*y2 + q2
+	;;
 	UNW(.restore sp)
 	ldf.fill f16 = [sp], 16
-	FP_TO_INT(f8, f17)		// round result to an (unsigned) integer
+	FP_TO_INT(f17, f17)		// q = trunc(q3)
 	;;
+#ifdef MODULO
+	xma.l f17 = f17, f9, f8		// r = q*(-b) + a
+	;;
+#endif
+	getf.sig r8 = f17		// transfer result to result register
 	ldf.fill f17 = [sp]
-	getf.sig r8 = f8		// transfer result to result register
 	br.ret.sptk rp
 END(NAME)
diff --git a/arch/ia64/lib/io.c b/arch/ia64/lib/io.c
index 466335172..baa408e01 100644
--- a/arch/ia64/lib/io.c
+++ b/arch/ia64/lib/io.c
@@ -1,4 +1,3 @@
-#include <linux/module.h>
 #include <linux/types.h>
 
 #include <asm/io.h>
@@ -49,6 +48,3 @@ __ia64_memset_c_io (unsigned long dst, unsigned long c, long count)
 	}
 }
 
-EXPORT_SYMBOL(__ia64_memcpy_fromio);
-EXPORT_SYMBOL(__ia64_memcpy_toio);
-EXPORT_SYMBOL(__ia64_memset_c_io);