microblaze: Add libgcc function directly to kernel

Replaced libgcc functions with asm optimized implementation.

Signed-off-by: Michal Simek <[email protected]>

arch/microblaze/Makefile

@@ -42,11 +42,8 @@ KBUILD_CFLAGS += -ffixed-r31 $(CPUFLAGS-1) $(CPUFLAGS-2)
 LDFLAGS		:=
 LDFLAGS_vmlinux	:=
 
-LIBGCC := $(shell $(CC) $(KBUILD_CFLAGS) -print-libgcc-file-name)
-
 head-y := arch/microblaze/kernel/head.o
 libs-y += arch/microblaze/lib/
-libs-y += $(LIBGCC)
 core-y += arch/microblaze/kernel/
 core-y += arch/microblaze/mm/
 core-y += arch/microblaze/platform/

arch/microblaze/kernel/microblaze_ksyms.c

@@ -21,31 +21,6 @@
 #include <linux/ftrace.h>
 #include <linux/uaccess.h>
 
-/*
- * libgcc functions - functions that are used internally by the
- * compiler... (prototypes are not correct though, but that
- * doesn't really matter since they're not versioned).
- */
-extern void __ashldi3(void);
-EXPORT_SYMBOL(__ashldi3);
-extern void __ashrdi3(void);
-EXPORT_SYMBOL(__ashrdi3);
-extern void __divsi3(void);
-EXPORT_SYMBOL(__divsi3);
-extern void __lshrdi3(void);
-EXPORT_SYMBOL(__lshrdi3);
-extern void __modsi3(void);
-EXPORT_SYMBOL(__modsi3);
-extern void __mulsi3(void);
-EXPORT_SYMBOL(__mulsi3);
-extern void __muldi3(void);
-EXPORT_SYMBOL(__muldi3);
-extern void __ucmpdi2(void);
-EXPORT_SYMBOL(__ucmpdi2);
-extern void __udivsi3(void);
-EXPORT_SYMBOL(__udivsi3);
-extern void __umodsi3(void);
-EXPORT_SYMBOL(__umodsi3);
 extern char *_ebss;
 EXPORT_SYMBOL_GPL(_ebss);
 #ifdef CONFIG_FUNCTION_TRACER

arch/microblaze/lib/Makefile

@@ -11,3 +11,13 @@ lib-y += memcpy.o memmove.o
 endif
 
 lib-y += uaccess_old.o
+
+lib-y += ashldi3.o
+lib-y += ashrdi3.o
+lib-y += divsi3.o
+lib-y += lshrdi3.o
+lib-y += modsi3.o
+lib-y += muldi3.o
+lib-y += mulsi3.o
+lib-y += udivsi3.o
+lib-y += umodsi3.o

arch/microblaze/lib/ashldi3.c

@@ -0,0 +1,29 @@
+#include <linux/module.h>
+
+#include "libgcc.h"
+
+long long __ashldi3(long long u, word_type b)
+{
+	DWunion uu, w;
+	word_type bm;
+
+	if (b == 0)
+		return u;
+
+	uu.ll = u;
+	bm = 32 - b;
+
+	if (bm <= 0) {
+		w.s.low = 0;
+		w.s.high = (unsigned int) uu.s.low << -bm;
+	} else {
+		const unsigned int carries = (unsigned int) uu.s.low >> bm;
+
+		w.s.low = (unsigned int) uu.s.low << b;
+		w.s.high = ((unsigned int) uu.s.high << b) | carries;
+	}
+
+	return w.ll;
+}
+
+EXPORT_SYMBOL(__ashldi3);

arch/microblaze/lib/ashrdi3.c

@@ -0,0 +1,31 @@
+#include <linux/module.h>
+
+#include "libgcc.h"
+
+long long __ashrdi3(long long u, word_type b)
+{
+	DWunion uu, w;
+	word_type bm;
+
+	if (b == 0)
+		return u;
+
+	uu.ll = u;
+	bm = 32 - b;
+
+	if (bm <= 0) {
+		/* w.s.high = 1..1 or 0..0 */
+		w.s.high =
+		    uu.s.high >> 31;
+		w.s.low = uu.s.high >> -bm;
+	} else {
+		const unsigned int carries = (unsigned int) uu.s.high << bm;
+
+		w.s.high = uu.s.high >> b;
+		w.s.low = ((unsigned int) uu.s.low >> b) | carries;
+	}
+
+	return w.ll;
+}
+
+EXPORT_SYMBOL(__ashrdi3);

arch/microblaze/lib/divsi3.S

@@ -0,0 +1,73 @@
+#include <linux/linkage.h>
+
+/*
+* Divide operation for 32 bit integers.
+*	Input :	Dividend in Reg r5
+*		Divisor in Reg r6
+*	Output: Result in Reg r3
+*/
+	.text
+	.globl	__divsi3
+	.type __divsi3, @function
+	.ent __divsi3
+__divsi3:
+	.frame	r1, 0, r15
+
+	addik	r1, r1, -16
+	swi	r28, r1, 0
+	swi	r29, r1, 4
+	swi	r30, r1, 8
+	swi	r31, r1, 12
+
+	beqi	r6, div_by_zero /* div_by_zero - division error */
+	beqi	r5, result_is_zero /* result is zero */
+	bgeid	r5, r5_pos
+	xor	r28, r5, r6 /* get the sign of the result */
+	rsubi	r5, r5, 0 /* make r5 positive */
+r5_pos:
+	bgei	r6, r6_pos
+	rsubi	r6, r6, 0 /* make r6 positive */
+r6_pos:
+	addik	r30, r0, 0 /* clear mod */
+	addik	r3, r0, 0 /* clear div */
+	addik	r29, r0, 32 /* initialize the loop count */
+
+	/* first part try to find the first '1' in the r5 */
+div0:
+	blti	r5, div2 /* this traps r5 == 0x80000000 */
+div1:
+	add	r5, r5, r5 /* left shift logical r5 */
+	bgtid	r5, div1
+	addik	r29, r29, -1
+div2:
+	/* left shift logical r5 get the '1' into the carry */
+	add	r5, r5, r5
+	addc	r30, r30, r30 /* move that bit into the mod register */
+	rsub	r31, r6, r30 /* try to subtract (r30 a r6) */
+	blti	r31, mod_too_small
+	/* move the r31 to mod since the result was positive */
+	or	r30, r0, r31
+	addik	r3, r3, 1
+mod_too_small:
+	addik	r29, r29, -1
+	beqi	r29, loop_end
+	add	r3, r3, r3 /* shift in the '1' into div */
+	bri	div2 /* div2 */
+loop_end:
+	bgei	r28, return_here
+	brid	return_here
+	rsubi	r3, r3, 0 /* negate the result */
+div_by_zero:
+result_is_zero:
+	or	r3, r0, r0 /* set result to 0 */
+return_here:
+/* restore values of csrs and that of r3 and the divisor and the dividend */
+	lwi	r28, r1, 0
+	lwi	r29, r1, 4
+	lwi	r30, r1, 8
+	lwi	r31, r1, 12
+	rtsd	r15, 8
+	addik	r1, r1, 16
+
+.size __divsi3, . - __divsi3
+.end __divsi3

arch/microblaze/lib/libgcc.h

@@ -0,0 +1,25 @@
+#ifndef __ASM_LIBGCC_H
+#define __ASM_LIBGCC_H
+
+#include <asm/byteorder.h>
+
+typedef int word_type __attribute__ ((mode (__word__)));
+
+#ifdef __BIG_ENDIAN
+struct DWstruct {
+	int high, low;
+};
+#elif defined(__LITTLE_ENDIAN)
+struct DWstruct {
+	int low, high;
+};
+#else
+#error I feel sick.
+#endif
+
+typedef union {
+	struct DWstruct s;
+	long long ll;
+} DWunion;
+
+#endif /* __ASM_LIBGCC_H */

arch/microblaze/lib/lshrdi3.c

@@ -0,0 +1,29 @@
+#include <linux/module.h>
+
+#include "libgcc.h"
+
+long long __lshrdi3(long long u, word_type b)
+{
+	DWunion uu, w;
+	word_type bm;
+
+	if (b == 0)
+		return u;
+
+	uu.ll = u;
+	bm = 32 - b;
+
+	if (bm <= 0) {
+		w.s.high = 0;
+		w.s.low = (unsigned int) uu.s.high >> -bm;
+	} else {
+		const unsigned int carries = (unsigned int) uu.s.high << bm;
+
+		w.s.high = (unsigned int) uu.s.high >> b;
+		w.s.low = ((unsigned int) uu.s.low >> b) | carries;
+	}
+
+	return w.ll;
+}
+
+EXPORT_SYMBOL(__lshrdi3);

arch/microblaze/lib/modsi3.S

@@ -0,0 +1,73 @@
+#include <linux/linkage.h>
+
+/*
+* modulo operation for 32 bit integers.
+*	Input :	op1 in Reg r5
+*		op2 in Reg r6
+*	Output: op1 mod op2 in Reg r3
+*/
+
+	.text
+	.globl	__modsi3
+	.type __modsi3,  @function
+	.ent __modsi3
+
+__modsi3:
+	.frame	r1, 0, r15
+
+	addik	r1, r1, -16
+	swi	r28, r1, 0
+	swi	r29, r1, 4
+	swi	r30, r1, 8
+	swi	r31, r1, 12
+
+	beqi	r6, div_by_zero /* div_by_zero division error */
+	beqi	r5, result_is_zero /* result is zero */
+	bgeid	r5, r5_pos
+	/* get the sign of the result [ depends only on the first arg] */
+	add	r28, r5, r0
+	rsubi	r5, r5, 0	 /* make r5 positive */
+r5_pos:
+	bgei	r6, r6_pos
+	rsubi	r6, r6, 0	 /* make r6 positive */
+r6_pos:
+	addik	r3, r0, 0 /* clear mod */
+	addik	r30, r0, 0 /* clear div */
+	addik	r29, r0, 32 /* initialize the loop count */
+/* first part try to find the first '1' in the r5 */
+div1:
+	add	r5, r5, r5 /* left shift logical r5 */
+	bgeid	r5, div1
+	addik	r29, r29, -1
+div2:
+	/* left shift logical r5 get the '1' into the carry */
+	add	r5, r5, r5
+	addc	r3, r3, r3 /* move that bit into the mod register */
+	rsub	r31, r6, r3 /* try to subtract (r30 a r6) */
+	blti	r31, mod_too_small
+	/* move the r31 to mod since the result was positive */
+	or	r3, r0, r31
+	addik	r30, r30, 1
+mod_too_small:
+	addik	r29, r29, -1
+	beqi	r29, loop_end
+	add	r30, r30, r30 /* shift in the '1' into div */
+	bri	div2 /* div2 */
+loop_end:
+	bgei	r28, return_here
+	brid	return_here
+	rsubi	r3, r3, 0 /* negate the result */
+div_by_zero:
+result_is_zero:
+	or	r3, r0, r0 /* set result to 0 [both mod as well as div are 0] */
+return_here:
+/* restore values of csrs and that of r3 and the divisor and the dividend */
+	lwi	r28, r1, 0
+	lwi	r29, r1, 4
+	lwi	r30, r1, 8
+	lwi	r31, r1, 12
+	rtsd	r15, 8
+	addik	r1, r1, 16
+
+.size __modsi3,  . - __modsi3
+.end __modsi3