[PATCH 12/14] m68k: Implement ndelay() based on the existing udelay() logic

[Michael Schmitz <schmitzmic@gmail.com>]

Add ndelay macro, modeled after the CF udelay macro. The ISP1160
USB HCD chip on EtherNAT and NetUSBee needs a 150ns delay after
register access.

Signed-off-by: Michael Schmitz <schmitz@debian.org>
---
 arch/m68k/include/asm/delay.h |   20 ++++++++++++++++++++
 1 files changed, 20 insertions(+), 0 deletions(-)

diff --git a/arch/m68k/include/asm/delay.h b/arch/m68k/include/asm/delay.h
index 12d8fe4..8a9971d 100644
--- a/arch/m68k/include/asm/delay.h
+++ b/arch/m68k/include/asm/delay.h
@@ -92,5 +92,25 @@ static inline void __udelay(unsigned long usecs)
 #define udelay(n) (__builtin_constant_p(n) ? \
 	((n) > 20000 ? __bad_udelay() : __const_udelay(n)) : __udelay(n))
 
+/*
+ * nanosecond delay:
+ *
+ * ((((HZSCALE) >> 11) * (loops_per_jiffy >> 11)) >> 6) is the number of loops per microsecond
+ *
+ * 1000 / ((((HZSCALE) >> 11) * (loops_per_jiffy >> 11)) >> 6) is the number of nanoseconds per loop
+ *
+ * So n / ( 1000 / ((((HZSCALE) >> 11) * (loops_per_jiffy >> 11)) >> 6) ) would be the number of loops for n nanoseconds
+ */
+
+/*
+ * The simpler m68k and ColdFire processors do not have a 32*32->64
+ * multiply instruction. So we need to handle them a little differently.
+ * We use a bit of shifting and a single 32*32->32 multiply to get close.
+ * This is a macro so that the const version can factor out the first
+ * multiply and shift.
+ */
+#define	HZSCALE		(268435456 / (1000000 / HZ))
+
+#define ndelay(n) __delay(DIV_ROUND_UP((n) * ((((HZSCALE) >> 11) * (loops_per_jiffy >> 11)) >> 6), 1000));
 
 #endif /* defined(_M68K_DELAY_H) */
-- 
1.7.0.4