Eliminate do_div().

This eliminates the use of do_div() in favor of using libgcc
functions.

This was tested by building and booting on Google Snow (ARMv7)
and Qemu (x86). printk()s which use division in vtxprintf() look good.

Change-Id: Icad001d84a3c05bfbf77098f3d644816280b4a4d
Signed-off-by: Gabe Black <gabeblack@chromium.org>
Signed-off-by: David Hendricks <dhendrix@chromium.org>
Reviewed-on: http://review.coreboot.org/2606
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>

6 files changed, 5 insertions, 478 deletions

diff --git a/src/arch/armv7/Makefile.inc b/src/arch/armv7/Makefile.inc
index e708f6d818..39c115ef8f 100644
--- a/src/arch/armv7/Makefile.inc
+++ b/src/arch/armv7/Makefile.inc
@@ -124,7 +124,7 @@ endif
 $(objgenerated)/coreboot_ram.o: $(stages_o) $$(ramstage-objs) $(LIBGCC_FILE_NAME)
 	@printf "    CC         $(subst $(obj)/,,$(@))\n"
 ifeq ($(CONFIG_COMPILER_LLVM_CLANG),y)
-	$(LD) -m -m armelf_linux_eabi -r -o $@ --wrap __divdi3 --wrap __udivdi3 --wrap __moddi3 --wrap __umoddi3 --wrap __uidiv --wrap __do_div64 --start-group $(ramstage-objs) $(LIBGCC_FILE_NAME) --end-group
+	$(LD) -m -m armelf_linux_eabi -r -o $@ --wrap __divdi3 --wrap __udivdi3 --wrap __moddi3 --wrap __umoddi3 --wrap __uidiv --start-group $(ramstage-objs) $(LIBGCC_FILE_NAME) --end-group
 else
 	$(CC) $(CFLAGS) -nostdlib -r -o $@ -Wl,--start-group $(stages_o) $(ramstage-objs) $(LIBGCC_FILE_NAME) -Wl,--end-group
 endif
diff --git a/src/arch/armv7/include/div64.h b/src/arch/armv7/include/div64.h
deleted file mode 100644
index 5baed2b294..0000000000
--- a/src/arch/armv7/include/div64.h
+++ /dev/null
@@ -1,233 +0,0 @@
-/* taken from linux 2.6.31.14 */
-
-#ifndef __ASM_ARM_DIV64
-#define __ASM_ARM_DIV64
-
-//#include <asm/system.h>
-//#include <linux/types.h>
-// FIXME
-
-#define __asmeq(x, y)  ".ifnc " x "," y " ; .err ; .endif\n\t"
-#define __LINUX_ARM_ARCH__  7
-
-/*
- * The semantics of do_div() are:
- *
- * uint32_t do_div(uint64_t *n, uint32_t base)
- * {
- * 	uint32_t remainder = *n % base;
- * 	*n = *n / base;
- * 	return remainder;
- * }
- *
- * In other words, a 64-bit dividend with a 32-bit divisor producing
- * a 64-bit result and a 32-bit remainder.  To accomplish this optimally
- * we call a special __do_div64 helper with completely non standard
- * calling convention for arguments and results (beware).
- */
-
-#ifdef __ARMEB__
-#define __xh "r0"
-#define __xl "r1"
-#else
-#define __xl "r0"
-#define __xh "r1"
-#endif
-
-#define __do_div_asm(n, base)					\
-({								\
-	register unsigned int __base      asm("r4") = base;	\
-	register unsigned long long __n   asm("r0") = n;	\
-	register unsigned long long __res asm("r2");		\
-	register unsigned int __rem       asm(__xh);		\
-	asm(	__asmeq("%0", __xh)				\
-		__asmeq("%1", "r2")				\
-		__asmeq("%2", "r0")				\
-		__asmeq("%3", "r4")				\
-		"bl	__do_div64"				\
-		: "=r" (__rem), "=r" (__res)			\
-		: "r" (__n), "r" (__base)			\
-		: "ip", "lr", "cc");				\
-	n = __res;						\
-	__rem;							\
-})
-
-#if __GNUC__ < 4
-
-/*
- * gcc versions earlier than 4.0 are simply too problematic for the
- * optimized implementation below. First there is gcc PR 15089 that
- * tend to trig on more complex constructs, spurious .global __udivsi3
- * are inserted even if none of those symbols are referenced in the
- * generated code, and those gcc versions are not able to do constant
- * propagation on long long values anyway.
- */
-#define do_div(n, base) __do_div_asm(n, base)
-
-#elif __GNUC__ >= 4
-
-//#include <asm/bug.h>
-
-/*
- * If the divisor happens to be constant, we determine the appropriate
- * inverse at compile time to turn the division into a few inline
- * multiplications instead which is much faster. And yet only if compiling
- * for ARMv4 or higher (we need umull/umlal) and if the gcc version is
- * sufficiently recent to perform proper long long constant propagation.
- * (It is unfortunate that gcc doesn't perform all this internally.)
- */
-#define do_div(n, base)							\
-({									\
-	unsigned int __r, __b = (base);					\
-	if (!__builtin_constant_p(__b) || __b == 0 ||			\
-	    (__LINUX_ARM_ARCH__ < 4 && (__b & (__b - 1)) != 0)) {	\
-		/* non-constant divisor (or zero): slow path */		\
-		__r = __do_div_asm(n, __b);				\
-	} else if ((__b & (__b - 1)) == 0) {				\
-		/* Trivial: __b is constant and a power of 2 */		\
-		/* gcc does the right thing with this code.  */		\
-		__r = n;						\
-		__r &= (__b - 1);					\
-		n /= __b;						\
-	} else {							\
-		/* Multiply by inverse of __b: n/b = n*(p/b)/p       */	\
-		/* We rely on the fact that most of this code gets   */	\
-		/* optimized away at compile time due to constant    */	\
-		/* propagation and only a couple inline assembly     */	\
-		/* instructions should remain. Better avoid any      */	\
-		/* code construct that might prevent that.           */	\
-		unsigned long long __res, __x, __t, __m, __n = n;	\
-		unsigned int __c, __p, __z = 0;				\
-		/* preserve low part of n for reminder computation */	\
-		__r = __n;						\
-		/* determine number of bits to represent __b */		\
-		__p = 1 << __div64_fls(__b);				\
-		/* compute __m = ((__p << 64) + __b - 1) / __b */	\
-		__m = (~0ULL / __b) * __p;				\
-		__m += (((~0ULL % __b + 1) * __p) + __b - 1) / __b;	\
-		/* compute __res = __m*(~0ULL/__b*__b-1)/(__p << 64) */	\
-		__x = ~0ULL / __b * __b - 1;				\
-		__res = (__m & 0xffffffff) * (__x & 0xffffffff);	\
-		__res >>= 32;						\
-		__res += (__m & 0xffffffff) * (__x >> 32);		\
-		__t = __res;						\
-		__res += (__x & 0xffffffff) * (__m >> 32);		\
-		__t = (__res < __t) ? (1ULL << 32) : 0;			\
-		__res = (__res >> 32) + __t;				\
-		__res += (__m >> 32) * (__x >> 32);			\
-		__res /= __p;						\
-		/* Now sanitize and optimize what we've got. */		\
-		if (~0ULL % (__b / (__b & -__b)) == 0) {		\
-			/* those cases can be simplified with: */	\
-			__n /= (__b & -__b);				\
-			__m = ~0ULL / (__b / (__b & -__b));		\
-			__p = 1;					\
-			__c = 1;					\
-		} else if (__res != __x / __b) {			\
-			/* We can't get away without a correction    */	\
-			/* to compensate for bit truncation errors.  */	\
-			/* To avoid it we'd need an additional bit   */	\
-			/* to represent __m which would overflow it. */	\
-			/* Instead we do m=p/b and n/b=(n*m+m)/p.    */	\
-			__c = 1;					\
-			/* Compute __m = (__p << 64) / __b */		\
-			__m = (~0ULL / __b) * __p;			\
-			__m += ((~0ULL % __b + 1) * __p) / __b;		\
-		} else {						\
-			/* Reduce __m/__p, and try to clear bit 31   */	\
-			/* of __m when possible otherwise that'll    */	\
-			/* need extra overflow handling later.       */	\
-			unsigned int __bits = -(__m & -__m);		\
-			__bits |= __m >> 32;				\
-			__bits = (~__bits) << 1;			\
-			/* If __bits == 0 then setting bit 31 is     */	\
-			/* unavoidable.  Simply apply the maximum    */	\
-			/* possible reduction in that case.          */	\
-			/* Otherwise the MSB of __bits indicates the */	\
-			/* best reduction we should apply.           */	\
-			if (!__bits) {					\
-				__p /= (__m & -__m);			\
-				__m /= (__m & -__m);			\
-			} else {					\
-				__p >>= __div64_fls(__bits);		\
-				__m >>= __div64_fls(__bits);		\
-			}						\
-			/* No correction needed. */			\
-			__c = 0;					\
-		}							\
-		/* Now we have a combination of 2 conditions:        */	\
-		/* 1) whether or not we need a correction (__c), and */	\
-		/* 2) whether or not there might be an overflow in   */	\
-		/*    the cross product (__m & ((1<<63) | (1<<31)))  */	\
-		/* Select the best insn combination to perform the   */	\
-		/* actual __m * __n / (__p << 64) operation.         */	\
-		if (!__c) {						\
-			asm (	"umull	%Q0, %R0, %1, %Q2\n\t"		\
-				"mov	%Q0, #0"			\
-				: "=&r" (__res)				\
-				: "r" (__m), "r" (__n)			\
-				: "cc" );				\
-		} else if (!(__m & ((1ULL << 63) | (1ULL << 31)))) {	\
-			__res = __m;					\
-			asm (	"umlal	%Q0, %R0, %Q1, %Q2\n\t"		\
-				"mov	%Q0, #0"			\
-				: "+&r" (__res)				\
-				: "r" (__m), "r" (__n)			\
-				: "cc" );				\
-		} else {						\
-			asm (	"umull	%Q0, %R0, %Q1, %Q2\n\t"		\
-				"cmn	%Q0, %Q1\n\t"			\
-				"adcs	%R0, %R0, %R1\n\t"		\
-				"adc	%Q0, %3, #0"			\
-				: "=&r" (__res)				\
-				: "r" (__m), "r" (__n), "r" (__z)	\
-				: "cc" );				\
-		}							\
-		if (!(__m & ((1ULL << 63) | (1ULL << 31)))) {		\
-			asm (	"umlal	%R0, %Q0, %R1, %Q2\n\t"		\
-				"umlal	%R0, %Q0, %Q1, %R2\n\t"		\
-				"mov	%R0, #0\n\t"			\
-				"umlal	%Q0, %R0, %R1, %R2"		\
-				: "+&r" (__res)				\
-				: "r" (__m), "r" (__n)			\
-				: "cc" );				\
-		} else {						\
-			asm (	"umlal	%R0, %Q0, %R2, %Q3\n\t"		\
-				"umlal	%R0, %1, %Q2, %R3\n\t"		\
-				"mov	%R0, #0\n\t"			\
-				"adds	%Q0, %1, %Q0\n\t"		\
-				"adc	%R0, %R0, #0\n\t"		\
-				"umlal	%Q0, %R0, %R2, %R3"		\
-				: "+&r" (__res), "+&r" (__z)		\
-				: "r" (__m), "r" (__n)			\
-				: "cc" );				\
-		}							\
-		__res /= __p;						\
-		/* The reminder can be computed with 32-bit regs     */	\
-		/* only, and gcc is good at that.                    */	\
-		{							\
-			unsigned int __res0 = __res;			\
-			unsigned int __b0 = __b;			\
-			__r -= __res0 * __b0;				\
-		}							\
-		/* BUG_ON(__r >= __b || __res * __b + __r != n); */	\
-		n = __res;						\
-	}								\
-	__r;								\
-})
-
-/* our own fls implementation to make sure constant propagation is fine */
-#define __div64_fls(bits)						\
-({									\
-	unsigned int __left = (bits), __nr = 0;				\
-	if (__left & 0xffff0000) __nr += 16, __left >>= 16;		\
-	if (__left & 0x0000ff00) __nr +=  8, __left >>=  8;		\
-	if (__left & 0x000000f0) __nr +=  4, __left >>=  4;		\
-	if (__left & 0x0000000c) __nr +=  2, __left >>=  2;		\
-	if (__left & 0x00000002) __nr +=  1;				\
-	__nr;								\
-})
-
-#endif
-
-#endif
diff --git a/src/arch/armv7/lib/Makefile.inc b/src/arch/armv7/lib/Makefile.inc
index 388864aa28..508f7766f7 100644
--- a/src/arch/armv7/lib/Makefile.inc
+++ b/src/arch/armv7/lib/Makefile.inc
@@ -6,12 +6,10 @@ bootblock-y += cache-cp15.c
 romstage-y += cache_v7.c
 romstage-y += cache-cp15.c
 romstage-y += div0.c
-romstage-y += div64.S
 romstage-y += syslib.c
 romstage-$(CONFIG_EARLY_CONSOLE) += early_console.c
 
 ramstage-y += div0.c
-ramstage-y += div64.S
 #ramstage-y += interrupts.c
 #ramstage-y += memcpy.S
 #ramstage-y += memset.S
diff --git a/src/arch/armv7/lib/div64.S b/src/arch/armv7/lib/div64.S
deleted file mode 100644
index 44edf480ca..0000000000
--- a/src/arch/armv7/lib/div64.S
+++ /dev/null
@@ -1,208 +0,0 @@
-/*
- *  linux/arch/arm/lib/div64.S
- *
- *  Optimized computation of 64-bit dividend / 32-bit divisor
- *
- *  Author:	Nicolas Pitre
- *  Created:	Oct 5, 2003
- *  Copyright:	Monta Vista Software, Inc.
- *
- *  This program is free software; you can redistribute it and/or modify
- *  it under the terms of the GNU General Public License version 2 as
- *  published by the Free Software Foundation.
- */
-
-// FIXME
-//#include <linux/linkage.h>
-#define __LINUX_ARM_ARCH__ 7
-
-#ifdef __ARMEB__
-#define xh r0
-#define xl r1
-#define yh r2
-#define yl r3
-#else
-#define xl r0
-#define xh r1
-#define yl r2
-#define yh r3
-#endif
-
-/*
- * __do_div64: perform a division with 64-bit dividend and 32-bit divisor.
- *
- * Note: Calling convention is totally non standard for optimal code.
- *       This is meant to be used by do_div() from include/asm/div64.h only.
- *
- * Input parameters:
- * 	xh-xl	= dividend (clobbered)
- * 	r4	= divisor (preserved)
- *
- * Output values:
- * 	yh-yl	= result
- * 	xh	= remainder
- *
- * Clobbered regs: xl, ip
- */
-
-
-.globl __do_div64;
-.align 4,0x90
-__do_div64:
-
-	@ Test for easy paths first.
-	subs	ip, r4, #1
-	bls	9f			@ divisor is 0 or 1
-	tst	ip, r4
-	beq	8f			@ divisor is power of 2
-
-	@ See if we need to handle upper 32-bit result.
-	cmp	xh, r4
-	mov	yh, #0
-	blo	3f
-
-	@ Align divisor with upper part of dividend.
-	@ The aligned divisor is stored in yl preserving the original.
-	@ The bit position is stored in ip.
-
-#if __LINUX_ARM_ARCH__ >= 5
-
-	clz	yl, r4
-	clz	ip, xh
-	sub	yl, yl, ip
-	mov	ip, #1
-	mov	ip, ip, lsl yl
-	mov	yl, r4, lsl yl
-
-#else
-
-	mov	yl, r4
-	mov	ip, #1
-1:	cmp	yl, #0x80000000
-	cmpcc	yl, xh
-	movcc	yl, yl, lsl #1
-	movcc	ip, ip, lsl #1
-	bcc	1b
-
-#endif
-
-	@ The division loop for needed upper bit positions.
- 	@ Break out early if dividend reaches 0.
-2:	cmp	xh, yl
-	orrcs	yh, yh, ip
-	subcss	xh, xh, yl
-	movnes	ip, ip, lsr #1
-	mov	yl, yl, lsr #1
-	bne	2b
-
-	@ See if we need to handle lower 32-bit result.
-3:	cmp	xh, #0
-	mov	yl, #0
-	cmpeq	xl, r4
-	movlo	xh, xl
-	movlo	pc, lr
-
-	@ The division loop for lower bit positions.
-	@ Here we shift remainer bits leftwards rather than moving the
-	@ divisor for comparisons, considering the carry-out bit as well.
-	mov	ip, #0x80000000
-4:	movs	xl, xl, lsl #1
-	adcs	xh, xh, xh
-	beq	6f
-	cmpcc	xh, r4
-5:	orrcs	yl, yl, ip
-	subcs	xh, xh, r4
-	movs	ip, ip, lsr #1
-	bne	4b
-	mov	pc, lr
-
-	@ The top part of remainder became zero.  If carry is set
-	@ (the 33th bit) this is a false positive so resume the loop.
-	@ Otherwise, if lower part is also null then we are done.
-6:	bcs	5b
-	cmp	xl, #0
-	moveq	pc, lr
-
-	@ We still have remainer bits in the low part.  Bring them up.
-
-#if __LINUX_ARM_ARCH__ >= 5
-
-	clz	xh, xl			@ we know xh is zero here so...
-	add	xh, xh, #1
-	mov	xl, xl, lsl xh
-	mov	ip, ip, lsr xh
-
-#else
-
-7:	movs	xl, xl, lsl #1
-	mov	ip, ip, lsr #1
-	bcc	7b
-
-#endif
-
-	@ Current remainder is now 1.  It is worthless to compare with
-	@ divisor at this point since divisor can not be smaller than 3 here.
-	@ If possible, branch for another shift in the division loop.
-	@ If no bit position left then we are done.
-	movs	ip, ip, lsr #1
-	mov	xh, #1
-	bne	4b
-	mov	pc, lr
-
-8:	@ Division by a power of 2: determine what that divisor order is
-	@ then simply shift values around
-
-#if __LINUX_ARM_ARCH__ >= 5
-
-	clz	ip, r4
-	rsb	ip, ip, #31
-
-#else
-
-	mov	yl, r4
-	cmp	r4, #(1 << 16)
-	mov	ip, #0
-	movhs	yl, yl, lsr #16
-	movhs	ip, #16
-
-	cmp	yl, #(1 << 8)
-	movhs	yl, yl, lsr #8
-	addhs	ip, ip, #8
-
-	cmp	yl, #(1 << 4)
-	movhs	yl, yl, lsr #4
-	addhs	ip, ip, #4
-
-	cmp	yl, #(1 << 2)
-	addhi	ip, ip, #3
-	addls	ip, ip, yl, lsr #1
-
-#endif
-
-	mov	yh, xh, lsr ip
-	mov	yl, xl, lsr ip
-	rsb	ip, ip, #32
-	orr	yl, yl, xh, lsl ip
-	mov	xh, xl, lsl ip
-	mov	xh, xh, lsr ip
-	mov	pc, lr
-
-	@ eq -> division by 1: obvious enough...
-9:	moveq	yl, xl
-	moveq	yh, xh
-	moveq	xh, #0
-	moveq	pc, lr
-
-	@ Division by 0:
-	str	lr, [sp, #-8]!
-	bl	__div0
-
-	@ as wrong as it could be...
-	mov	yl, #0
-	mov	yh, #0
-	mov	xh, #0
-	ldr	pc, [sp], #8
-
-@.type __do_div64, @function;
-@.size __do_div64, .-__do_div64,
-
diff --git a/src/arch/x86/Makefile.inc b/src/arch/x86/Makefile.inc
index 98723f3f1a..0892efd37b 100644
--- a/src/arch/x86/Makefile.inc
+++ b/src/arch/x86/Makefile.inc
@@ -354,9 +354,9 @@ endif
 $(objcbfs)/romstage_null.debug: $$(romstage-objs) $(objgenerated)/romstage_null.ld
 	@printf "    LINK       $(subst $(obj)/,,$(@))\n"
 ifeq ($(CONFIG_COMPILER_LLVM_CLANG),y)
-	$(LD) -nostdlib -nostartfiles -static -o $@ -L$(obj) $(romstage-objs) -T $(objgenerated)/romstage_null.ld
+	$(LD) -nostdlib -nostartfiles -static -o $@ -L$(obj) --wrap __divdi3 --wrap __udivdi3 --wrap __moddi3 --wrap __umoddi3 --start-group $(romstage-objs) $(LIBGCC_FILE_NAME) --end-group -T $(objgenerated)/romstage_null.ld
 else
-	$(CC) -nostdlib -nostartfiles -static -o $@ -L$(obj) -T $(objgenerated)/romstage_null.ld $(romstage-objs)
+	$(CC) -nostdlib -nostartfiles -static -o $@ -L$(obj) -T $(objgenerated)/romstage_null.ld -Wl,--wrap,__divdi3 -Wl,--wrap,__udivdi3 -Wl,--wrap,__moddi3 -Wl,--wrap,__umoddi3 -Wl,--start-group $(romstage-objs) $(LIBGCC_FILE_NAME) -Wl,--end-group
 endif
 	$(NM) $@ | grep -q " [DdBb] "; if [ $$? -eq 0 ]; then \
 		echo "Forbidden global variables in romstage:"; \
@@ -366,9 +366,9 @@ endif
 $(objcbfs)/romstage_xip.debug: $$(romstage-objs) $(objgenerated)/romstage_xip.ld
 	@printf "    LINK       $(subst $(obj)/,,$(@))\n"
 ifeq ($(CONFIG_COMPILER_LLVM_CLANG),y)
-	$(LD) -nostdlib -nostartfiles -static -o $@ -L$(obj) $(romstage-objs) -T $(objgenerated)/romstage_xip.ld
+	$(LD) -nostdlib -nostartfiles -static -o $@ -L$(obj) --wrap __divdi3 --wrap __udivdi3 --wrap __moddi3 --wrap __umoddi3 --start-group $(romstage-objs) $(LIBGCC_FILE_NAME) --end-group -T $(objgenerated)/romstage_xip.ld
 else
-	$(CC) -nostdlib -nostartfiles -static -o $@ -L$(obj) -T $(objgenerated)/romstage_xip.ld $(romstage-objs)
+	$(CC) -nostdlib -nostartfiles -static -o $@ -L$(obj) -T $(objgenerated)/romstage_xip.ld -Wl,--wrap,__divdi3 -Wl,--wrap,__udivdi3 -Wl,--wrap,__moddi3 -Wl,--wrap,__umoddi3 -Wl,--start-group $(romstage-objs) $(LIBGCC_FILE_NAME) -Wl,--end-group
 endif
 
 $(objgenerated)/romstage_null.ld: $$(ldscripts) $(obj)/ldoptions
diff --git a/src/arch/x86/include/div64.h b/src/arch/x86/include/div64.h
deleted file mode 100644
index bbc9921d29..0000000000
--- a/src/arch/x86/include/div64.h
+++ /dev/null
@@ -1,30 +0,0 @@
-#ifndef __I386_DIV64
-#define __I386_DIV64
-
-/*
- * do_div() is NOT a C function. It wants to return
- * two values (the quotient and the remainder), but
- * since that doesn't work very well in C, what it
- * does is:
- *
- * - modifies the 64-bit dividend _in_place_
- * - returns the 32-bit remainder
- *
- * This ends up being the most efficient "calling
- * convention" on x86.
- */
-#define do_div(n,base) ({ \
-	unsigned long __upper, __low, __high, __mod, __base; \
-	__base = (base); \
-	asm("":"=a" (__low), "=d" (__high):"A" (n)); \
-	__upper = __high; \
-	if (__high) { \
-		__upper = __high % (__base); \
-		__high = __high / (__base); \
-	} \
-	asm("divl %2":"=a" (__low), "=d" (__mod):"rm" (__base), "0" (__low), "1" (__upper)); \
-	asm("":"=A" (n):"a" (__low),"d" (__high)); \
-	__mod; \
-})
-
-#endif