[S390] Calibrate delay and bogomips.

Preset the bogomips number to the cpu capacity value reported by
store system information in SYSIB 1.2.2. This value is constant
for a particular machine model and can be used to determine
relative performance differences between machines.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/Kconfig

@@ -34,10 +34,6 @@ config GENERIC_HWEIGHT
 	bool
 	default y
 
-config GENERIC_CALIBRATE_DELAY
-	bool
-	default y
-
 config GENERIC_TIME
 	def_bool y
 

arch/s390/kernel/setup.c

@@ -938,6 +938,7 @@ static int show_cpuinfo(struct seq_file *m, void *v)
         struct cpuinfo_S390 *cpuinfo;
 	unsigned long n = (unsigned long) v - 1;
 
+	s390_adjust_jiffies();
 	preempt_disable();
 	if (!n) {
 		seq_printf(m, "vendor_id       : IBM/S390\n"

arch/s390/lib/Makefile

@@ -4,7 +4,7 @@
 
 EXTRA_AFLAGS := -traditional
 
-lib-y += delay.o string.o uaccess_std.o uaccess_pt.o
+lib-y += delay.o string.o uaccess_std.o uaccess_pt.o qrnnd.o
 lib-$(CONFIG_32BIT) += div64.o
 lib-$(CONFIG_64BIT) += uaccess_mvcos.o
 lib-$(CONFIG_SMP) += spinlock.o

arch/s390/lib/qrnnd.S

@@ -0,0 +1,77 @@
+# S/390 __udiv_qrnnd
+
+# r2 : &__r
+# r3 : upper half of 64 bit word n
+# r4 : lower half of 64 bit word n
+# r5 : divisor d
+# the reminder r of the division is to be stored to &__r and
+# the quotient q is to be returned
+
+	.text
+	.globl __udiv_qrnnd
+__udiv_qrnnd:
+	st    %r2,24(%r15)	  # store pointer to reminder for later
+	lr    %r0,%r3		  # reload n
+	lr    %r1,%r4
+	ltr   %r2,%r5		  # reload and test divisor
+	jp    5f
+	# divisor >= 0x80000000
+	srdl  %r0,2		  # n/4
+	srl   %r2,1		  # d/2
+	slr   %r1,%r2		  # special case if last bit of d is set
+	brc   3,0f		  #  (n/4) div (n/2) can overflow by 1
+	ahi   %r0,-1		  #  trick: subtract n/2, then divide
+0:	dr    %r0,%r2		  # signed division
+	ahi   %r1,1		  #  trick part 2: add 1 to the quotient
+	# now (n >> 2) = (d >> 1) * %r1 + %r0
+	lhi   %r3,1
+	nr    %r3,%r1		  # test last bit of q
+	jz    1f
+	alr   %r0,%r2		  # add (d>>1) to r
+1:	srl   %r1,1		  # q >>= 1
+	# now (n >> 2) = (d&-2) * %r1 + %r0
+	lhi   %r3,1
+	nr    %r3,%r5		  # test last bit of d
+	jz    2f
+	slr   %r0,%r1		  # r -= q
+	brc   3,2f		  # borrow ?
+	alr   %r0,%r5		  # r += d
+	ahi   %r1,-1
+2:	# now (n >> 2) = d * %r1 + %r0
+	alr   %r1,%r1		  # q <<= 1
+	alr   %r0,%r0		  # r <<= 1
+	brc   12,3f		  # overflow on r ?
+	slr   %r0,%r5		  # r -= d
+	ahi   %r1,1		  # q += 1
+3:	lhi   %r3,2
+	nr    %r3,%r4		  # test next to last bit of n
+	jz    4f
+	ahi   %r0,1		  # r += 1
+4:	clr   %r0,%r5		  # r >= d ?
+	jl    6f
+	slr   %r0,%r5		  # r -= d
+	ahi   %r1,1		  # q += 1
+	# now (n >> 1) = d * %r1 + %r0
+	j     6f
+5:	# divisor < 0x80000000
+	srdl  %r0,1
+	dr    %r0,%r2		  # signed division
+	# now (n >> 1) = d * %r1 + %r0
+6:	alr   %r1,%r1		  # q <<= 1
+	alr   %r0,%r0		  # r <<= 1
+	brc   12,7f		  # overflow on r ?
+	slr   %r0,%r5		  # r -= d
+	ahi   %r1,1		  # q += 1
+7:	lhi   %r3,1
+	nr    %r3,%r4		  # isolate last bit of n
+	alr   %r0,%r3		  # r += (n & 1)
+	clr   %r0,%r5		  # r >= d ?
+	jl    8f
+	slr   %r0,%r5		  # r -= d
+	ahi   %r1,1		  # q += 1
+8:	# now n = d * %r1 + %r0
+	l     %r2,24(%r15)
+	st    %r0,0(%r2)
+	lr    %r2,%r1
+	br    %r14
+	.end	__udiv_qrnnd

arch/s390/math-emu/Makefile

@@ -2,7 +2,7 @@
 # Makefile for the FPU instruction emulation.
 #
 
-obj-$(CONFIG_MATHEMU) := math.o qrnnd.o
+obj-$(CONFIG_MATHEMU) := math.o
 
 EXTRA_CFLAGS := -I$(src) -Iinclude/math-emu -w
 EXTRA_AFLAGS := -traditional

arch/s390/math-emu/math.c

@@ -15,7 +15,7 @@
 #include <asm/uaccess.h>
 #include <asm/lowcore.h>
 
-#include "sfp-util.h"
+#include <asm/sfp-util.h>
 #include <math-emu/soft-fp.h>
 #include <math-emu/single.h>
 #include <math-emu/double.h>

drivers/s390/Makefile

@@ -2,6 +2,8 @@
 # Makefile for the S/390 specific device drivers
 #
 
+CFLAGS_sysinfo.o += -Iinclude/math-emu -Iarch/s390/math-emu -w
+
 obj-y += s390mach.o sysinfo.o s390_rdev.o
 obj-y += cio/ block/ char/ crypto/ net/ scsi/
 

drivers/s390/sysinfo.c

@@ -9,8 +9,14 @@
 #include <linux/mm.h>
 #include <linux/proc_fs.h>
 #include <linux/init.h>
+#include <linux/delay.h>
 #include <asm/ebcdic.h>
 
+/* Sigh, math-emu. Don't ask. */
+#include <asm/sfp-util.h>
+#include <math-emu/soft-fp.h>
+#include <math-emu/single.h>
+
 struct sysinfo_1_1_1 {
 	char reserved_0[32];
 	char manufacturer[16];
@@ -198,7 +204,7 @@ static int stsi_1_2_2(struct sysinfo_1_2_2 *info, char *page, int len)
 		 * if the higher order 8 bits are not zero. Printing
 		 * a floating point number in the kernel is a no-no,
 		 * always print the number as 32 bit unsigned integer.
-		 * The user-space needs to know about the stange
+		 * The user-space needs to know about the strange
 		 * encoding of the alternate cpu capability.
 		 */
 		len += sprintf(page + len, "Capability:           %u %u\n",
@@ -351,3 +357,58 @@ static __init int create_proc_sysinfo(void)
 
 __initcall(create_proc_sysinfo);
 
+/*
+ * CPU capability might have changed. Therefore recalculate loops_per_jiffy.
+ */
+void s390_adjust_jiffies(void)
+{
+	struct sysinfo_1_2_2 *info;
+	const unsigned int fmil = 0x4b189680;	/* 1e7 as 32-bit float. */
+	FP_DECL_S(SA); FP_DECL_S(SB); FP_DECL_S(SR);
+	FP_DECL_EX;
+	unsigned int capability;
+
+	info = (void *) get_zeroed_page(GFP_KERNEL);
+	if (!info)
+		return;
+
+	if (stsi(info, 1, 2, 2) != -ENOSYS) {
+		/*
+		 * Major sigh. The cpu capability encoding is "special".
+		 * If the first 9 bits of info->capability are 0 then it
+		 * is a 32 bit unsigned integer in the range 0 .. 2^23.
+		 * If the first 9 bits are != 0 then it is a 32 bit float.
+		 * In addition a lower value indicates a proportionally
+		 * higher cpu capacity. Bogomips are the other way round.
+		 * To get to a halfway suitable number we divide 1e7
+		 * by the cpu capability number. Yes, that means a floating
+		 * point division .. math-emu here we come :-)
+		 */
+		FP_UNPACK_SP(SA, &fmil);
+		if ((info->capability >> 23) == 0)
+			FP_FROM_INT_S(SB, info->capability, 32, int);
+		else
+			FP_UNPACK_SP(SB, &info->capability);
+		FP_DIV_S(SR, SA, SB);
+		FP_TO_INT_S(capability, SR, 32, 0);
+	} else
+		/*
+		 * Really old machine without stsi block for basic
+		 * cpu information. Report 42.0 bogomips.
+		 */
+		capability = 42;
+	loops_per_jiffy = capability * (500000/HZ);
+	free_page((unsigned long) info);
+}
+
+/*
+ * calibrate the delay loop
+ */
+void __init calibrate_delay(void)
+{
+	s390_adjust_jiffies();
+	/* Print the good old Bogomips line .. */
+	printk(KERN_DEBUG "Calibrating delay loop (skipped)... "
+	       "%lu.%02lu BogoMIPS preset\n", loops_per_jiffy/(500000/HZ),
+	       (loops_per_jiffy/(5000/HZ)) % 100);
+}

include/asm-s390/processor.h

@@ -50,6 +50,7 @@ struct cpuinfo_S390
         unsigned long pgtable_cache_sz;
 };
 
+extern void s390_adjust_jiffies(void);
 extern void print_cpu_info(struct cpuinfo_S390 *);
 
 /* Lazy FPU handling on uni-processor */

arch/s390/math-emu/sfp-util.h → include/asm-s390/sfp-util.h

@@ -52,12 +52,12 @@
 })
 
 #define udiv_qrnnd(q, r, n1, n0, d)			\
-  do { unsigned long __r;				\
+  do { unsigned int __r;				\
     (q) = __udiv_qrnnd (&__r, (n1), (n0), (d));		\
     (r) = __r;						\
   } while (0)
-extern unsigned long __udiv_qrnnd (unsigned long *, unsigned long,
-				   unsigned long , unsigned long);
+extern unsigned long __udiv_qrnnd (unsigned int *, unsigned int,
+				   unsigned int , unsigned int);
 
 #define UDIV_NEEDS_NORMALIZATION 0