arch/arc/include/asm/delay.h

   1 /*
   2  * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * Delay routines using pre computed loops_per_jiffy value.
   9  *
  10  * vineetg: Feb 2012
  11  *  -Rewrote in "C" to avoid dealing with availability of H/w MPY
  12  *  -Also reduced the num of MPY operations from 3 to 2
  13  *
  14  * Amit Bhor: Codito Technologies 2004
  15  */
  16
  17 #ifndef __ASM_ARC_UDELAY_H
  18 #define __ASM_ARC_UDELAY_H
  19
  20 #include <asm/param.h>          /* HZ */
  21
  22 static inline void __delay(unsigned long loops)
  23 {
  24         __asm__ __volatile__(
  25         "1:     sub.f %0, %0, 1 \n"
  26         "       jpnz 1b         \n"
  27         : "+r"(loops)
  28         :
  29         : "cc");
  30 }
  31
  32 extern void __bad_udelay(void);
  33
  34 /*
  35  * Normal Math for computing loops in "N" usecs
  36  *  -we have precomputed @loops_per_jiffy
  37  *  -1 sec has HZ jiffies
  38  * loops per "N" usecs = ((loops_per_jiffy * HZ / 1000000) * N)
  39  *
  40  * Approximate Division by multiplication:
  41  *  -Mathematically if we multiply and divide a number by same value the
  42  *   result remains unchanged:  In this case, we use 2^32
  43  *  -> (loops_per_N_usec * 2^32 ) / 2^32
  44  *  -> (((loops_per_jiffy * HZ / 1000000) * N) * 2^32) / 2^32
  45  *  -> (loops_per_jiffy * HZ * N * 4295) / 2^32
  46  *
  47  *  -Divide by 2^32 is very simply right shift by 32
  48  *  -We simply need to ensure that the multiply per above eqn happens in
  49  *   64-bit precision (if CPU doesn't support it - gcc can emaulate it)
  50  */
  51
  52 static inline void __udelay(unsigned long usecs)
  53 {
  54         unsigned long loops;
  55
  56         /* (u64) cast ensures 64 bit MPY - real or emulated
  57          * HZ * 4295 is pre-evaluated by gcc - hence only 2 mpy ops
  58          */
  59         loops = ((u64) usecs * 4295 * HZ * loops_per_jiffy) >> 32;
  60
  61         __delay(loops);
  62 }
  63
  64 #define udelay(n) (__builtin_constant_p(n) ? ((n) > 20000 ? __bad_udelay() \
  65                                 : __udelay(n)) : __udelay(n))
  66
  67 #endif /* __ASM_ARC_UDELAY_H */