arch/arc/include/asm/delay.h

   1 /* SPDX-License-Identifier: GPL-2.0-only */
   2 /*
   3  * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
   4  *
   5  * Delay routines using pre computed loops_per_jiffy value.
   6  *
   7  * vineetg: Feb 2012
   8  *  -Rewrote in "C" to avoid dealing with availability of H/w MPY
   9  *  -Also reduced the num of MPY operations from 3 to 2
  10  *
  11  * Amit Bhor: Codito Technologies 2004
  12  */
  13
  14 #ifndef __ASM_ARC_UDELAY_H
  15 #define __ASM_ARC_UDELAY_H
  16
  17 #include <asm-generic/types.h>
  18 #include <asm/param.h>          /* HZ */
  19
  20 extern unsigned long loops_per_jiffy;
  21
  22 static inline void __delay(unsigned long loops)
  23 {
  24         __asm__ __volatile__(
  25         "       mov lp_count, %0        \n"
  26         "       lp  1f                  \n"
  27         "       nop                     \n"
  28         "1:                             \n"
  29         :
  30         : "r"(loops)
  31         : "lp_count");
  32 }
  33
  34 extern void __bad_udelay(void);
  35
  36 /*
  37  * Normal Math for computing loops in "N" usecs
  38  *  -we have precomputed @loops_per_jiffy
  39  *  -1 sec has HZ jiffies
  40  * loops per "N" usecs = ((loops_per_jiffy * HZ / 1000000) * N)
  41  *
  42  * Approximate Division by multiplication:
  43  *  -Mathematically if we multiply and divide a number by same value the
  44  *   result remains unchanged:  In this case, we use 2^32
  45  *  -> (loops_per_N_usec * 2^32 ) / 2^32
  46  *  -> (((loops_per_jiffy * HZ / 1000000) * N) * 2^32) / 2^32
  47  *  -> (loops_per_jiffy * HZ * N * 4295) / 2^32
  48  *
  49  *  -Divide by 2^32 is very simply right shift by 32
  50  *  -We simply need to ensure that the multiply per above eqn happens in
  51  *   64-bit precision (if CPU doesn't support it - gcc can emaulate it)
  52  */
  53
  54 static inline void __udelay(unsigned long usecs)
  55 {
  56         unsigned long loops;
  57
  58         /* (u64) cast ensures 64 bit MPY - real or emulated
  59          * HZ * 4295 is pre-evaluated by gcc - hence only 2 mpy ops
  60          */
  61         loops = ((u64) usecs * 4295 * HZ * loops_per_jiffy) >> 32;
  62
  63         __delay(loops);
  64 }
  65
  66 #define udelay(n) (__builtin_constant_p(n) ? ((n) > 20000 ? __bad_udelay() \
  67                                 : __udelay(n)) : __udelay(n))
  68
  69 #endif /* __ASM_ARC_UDELAY_H */