ipl: Enable FPU and remove workarounds

It turns out, that the FPU isn't initialized by firmware, which is
why the floating point instructions in libgcc's __udivdi3, __umoddi3
and __muldi3 routines simply hangs the machine.

This patch enables the FPU at bootup and thus we can drop the
current work-arounds.

Signed-off-by: Helge Deller <deller@gmx.de>

6 files changed, 32 insertions, 41 deletions

diff --git a/ipl/Makefile b/ipl/Makefile
index 3409868..2254887 100644
--- a/ipl/Makefile
+++ b/ipl/Makefile
@@ -28,7 +28,6 @@ CROSS_COMPILE := $(call cc-cross-prefix, \
 CC	= ${CROSS_COMPILE}gcc
 AR	= ar
 LD	= ${CROSS_COMPILE}ld
-OBJDUMP = ${CROSS_COMPILE}objdump
 
 ifneq ("$(wildcard /etc/debian_version)","")
 BLDINFO := $(shell echo https://parisc.wiki.kernel.org - `dpkg-parsechangelog -l../debian/changelog -SDate`)
@@ -40,7 +39,7 @@ endif
 VPATH=../lib:.
 
 AFLAGS	= -I../lib
-CFLAGS	= -DIPL_LOADER -I. -I../lib -I../include -O2 -mdisable-fpregs -Wall -fno-delete-null-pointer-checks -fno-builtin -mpa-risc-1-0
+CFLAGS	= -DIPL_LOADER -I. -I../lib -I../include -O2 -mdisable-fpregs -Wall -fno-delete-null-pointer-checks -fno-builtin
 LDFLAGS  = -N --section-start .init=0x60000 -e '$$START$$'
 
 all:	iplelf
@@ -50,7 +49,6 @@ iplelf:	crt0.o ipl.a
 	$(CC) $(CFLAGS) -c -o build.o build.c
 	-rm -f build.c
 	$(LD) $(LDFLAGS) -o iplelf crt0.o ipl.a build.o `$(CC) -print-libgcc-file-name`
-	$(OBJDUMP) -t iplelf | grep -E '__udivdi3|__umoddi3|__muldi3' && (echo "ERROR: Can not use __udivdi3 or __umoddi3 in palo ipl code" && false) || true
 
 ipl.a:	$(OFILES)
 	$(AR) rv ipl.a $?
diff --git a/ipl/bootloader.h b/ipl/bootloader.h
index af4733f..c6a9f18 100644
--- a/ipl/bootloader.h
+++ b/ipl/bootloader.h
@@ -27,6 +27,7 @@ int pdc_os_bits();
 int pdc_iodc_bootin(__u64 devaddr, char *memaddr, unsigned size);
 int pdc_read_conspath(unsigned char *memaddr);
 int pdc_do_reset(void);
+int pdc_coproc_cfg();
 int pdc_model_sysmodel(char *name);
 
 typedef void (*describe_t)(int fd, int *bufalign, int *blocksize);
diff --git a/ipl/ext2.c b/ipl/ext2.c
index 6add1a0..8d3ac42 100644
--- a/ipl/ext2.c
+++ b/ipl/ext2.c
@@ -64,27 +64,15 @@ static struct inode_table_entry {
 #undef DEBUG
 #define Debug 0
 
-/* div_blocksize() and mul_blocksize() are trivial functions to avoid 64-bit
-   __udivdi3 and __muldi3 millicode routines. They only work because the ext2
-   blocksize is a multiple of 2 */
 static __u64 div_blocksize(__u64 devaddr, unsigned blocksize)
 {
-	while (blocksize > 1) {
-		devaddr >>= 1;
-		blocksize >>= 1;
-	}
-	return devaddr;
+	return devaddr / blocksize;
 }
 static __u64 mul_blocksize(__u64 devaddr, unsigned blocksize)
 {
-	while (blocksize > 1) {
-		devaddr <<= 1;
-		blocksize >>= 1;
-	}
-	return devaddr;
+	return devaddr * blocksize;
 }
 
-
 static struct ext3_extent_header *ext3_extent_header(struct ext2_inode *i)
 {
 	return (struct ext3_extent_header *)&i->i_block[0];
diff --git a/ipl/ipl.c b/ipl/ipl.c
index 22abca6..f6fe251 100644
--- a/ipl/ipl.c
+++ b/ipl/ipl.c
@@ -519,6 +519,7 @@ iplmain(int is_interactive, char *initialstackptr, int started_wide)
     struct firstblock f;
     int blocked_bootdev;
     int wide;
+    int ccr_functional;
     int kern_part, rd_part;
     char kern_name[128], rd_name[128];
     char kern_fullname[128+10];
@@ -543,6 +544,14 @@ iplmain(int is_interactive, char *initialstackptr, int started_wide)
 	printf("palo loaded at %p-%p.\n", &_start, &_end);
     }
 
+    ccr_functional = pdc_coproc_cfg();
+    if (ccr_functional)
+	asm("mtctl %0,%1" : : "r" (ccr_functional), "i" (10));
+    else {
+	printf("No functional FPUs found - stopping.\n");
+	while (1) ;
+    }
+
     disk_2gb_limit = pdc_bootdisk_2GB_limit();
     printf("Boot limited to sectors below 2GB: %s\n", disk_2gb_limit ? "YES":"NO");
 
diff --git a/ipl/pdc_misc.c b/ipl/pdc_misc.c
index b2296da..12c4613 100644
--- a/ipl/pdc_misc.c
+++ b/ipl/pdc_misc.c
@@ -221,6 +221,20 @@ pdc_os_bits()
 }
 
 int
+pdc_coproc_cfg(void)
+{
+    int r;
+
+    r = firmware_call(mem_pdc, PDC_COPROC, PDC_COPROC_CFG, pdc_result);
+    convert_from_wide(pdc_result);
+    if (r == PDC_OK) {
+		return pdc_result[0]; /* ccr_functional */
+        } else {
+		return 0;
+        }
+}
+
+int
 pdc_model_sysmodel(char *name)
 {
     int r;
diff --git a/ipl/vsprintf.c b/ipl/vsprintf.c
index 052511a..4a7a060 100644
--- a/ipl/vsprintf.c
+++ b/ipl/vsprintf.c
@@ -36,17 +36,10 @@ static int skip_atoi(const char **s)
 #define SPECIAL	32		/* 0x */
 #define LARGE	64		/* use 'ABCDEF' instead of 'abcdef' */
 
-/* inline on "unsigned long", not "long long" to avoid using __umoddi3 and __udivdi3 */
-#define do_div_long(n,base) ({ \
+#define do_div(n,base) ({ \
 int __res; \
-__res = ((unsigned long) n) % (unsigned) base; \
-n = ((unsigned long) n) / (unsigned) base; \
-__res; })
-
-#define do_div_hex(n) ({ \
-unsigned long __res; \
-__res = ((unsigned long) n) & 0x0f; \
-n >>= 4; \
+__res = ((unsigned long long) n) % (unsigned) base; \
+n = ((unsigned long long) n) / (unsigned) base; \
 __res; })
 
 static char * number(char * str, unsigned long long num, int base, int size, int precision
@@ -86,12 +79,8 @@ static char * number(char * str, unsigned long long num, int base, int size, int
 	i = 0;
 	if (num == 0)
 		tmp[i++]='0';
-	else while (num != 0) {
-		if (base == 16)
-			tmp[i++] = digits[do_div_hex(num)];
-		else
-			tmp[i++] = digits[do_div_long(num,base)];
-	}
+	else while (num != 0)
+		tmp[i++] = digits[do_div(num,base)];
 	if (i > precision)
 		precision = i;
 	size -= precision;
@@ -288,14 +277,6 @@ again:
 			num = va_arg(args, int);
 		else
 			num = va_arg(args, unsigned int);
-		/* workaround to output all integers > 32bit as hex. Avoids millicode routines */
-		if (num >> 32) {
-			if (base != 16) { // append hex symbol "0x"
-				*str++ = '0';
-				*str++ = 'x';
-			}
-			base = 16;
-		}
 		str = number(str, num, base, field_width, precision, flags);
 	}
 	*str = '\0';