target/linux/generic-2.6/patches-2.6.28/023-mips_delay_gcc4.4.0.patch

   1 From: Wu Zhangjin <wuzj@lemote.com>
   2
   3 the gcc 4.4 support for MIPS mostly refer to this PATCH:
   4 http://www.nabble.com/-PATCH--MIPS:-Handle-removal-of-%27h%27-constraint-in-GCC-4.4-td22192768.html
   5 but have been tuned a little.
   6
   7 because only gcc 4.4 have loongson-specific support, so, we need to
   8 choose the suitable -march argument for gcc <= 4.3 and gcc >= 4.4, and
   9 we also need to consider use -march=loongson2e and -march=loongson2f for
  10 loongson2e and loongson2f respectively. this is handled by adding two
  11 new kernel options: CPU_LOONGSON2E and CPU_LOONGSON2F(thanks for the
  12 solutin provided by ZhangLe).
  13
  14 I have tested it on FuLoong(2f) in 32bit and 64bit with gcc-4.4 and
  15 gcc-4.3. so, basically, it works.
  16
  17 Signed-off-by: Wu Zhangjin <wuzhangjin@gmail.com>
  18 ---
  19  arch/mips/Makefile               |    9 +++++-
  20  arch/mips/include/asm/compiler.h |   10 ++++++
  21  arch/mips/include/asm/delay.h    |   58 +++++++++++++++++++++++++------------
  22  3 files changed, 57 insertions(+), 20 deletions(-)
  23
  24 --- a/arch/mips/Makefile
  25 +++ b/arch/mips/Makefile
  26 @@ -119,7 +119,14 @@ cflags-$(CONFIG_CPU_R4300) += -march=r43
  27  cflags-$(CONFIG_CPU_VR41XX)    += -march=r4100 -Wa,--trap
  28  cflags-$(CONFIG_CPU_R4X00)     += -march=r4600 -Wa,--trap
  29  cflags-$(CONFIG_CPU_TX49XX)    += -march=r4600 -Wa,--trap
  30 -cflags-$(CONFIG_CPU_LOONGSON2) += -march=r4600 -Wa,--trap
  31 +
  32 +# only gcc >= 4.4 have the loongson-specific support
  33 +cflags-$(CONFIG_CPU_LOONGSON2) += -Wa,--trap
  34 +cflags-$(CONFIG_CPU_LOONGSON2E)        += $(shell if [ $(call cc-version) -lt 0440 ] ; then \
  35 +       echo $(call cc-option,-march=r4600); else echo $(call cc-option,-march=loongson2e); fi ;)
  36 +cflags-$(CONFIG_CPU_LOONGSON2F)        += $(shell if [ $(call cc-version) -lt 0440 ] ; then \
  37 +       echo $(call cc-option,-march=r4600); else echo $(call cc-option,-march=loongson2f); fi ;)
  38 +
  39  cflags-$(CONFIG_CPU_MIPS32_R1) += $(call cc-option,-march=mips32,-mips32 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS32) \
  40                         -Wa,-mips32 -Wa,--trap
  41  cflags-$(CONFIG_CPU_MIPS32_R2) += $(call cc-option,-march=mips32r2,-mips32r2 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS32) \
  42 --- a/arch/mips/include/asm/compiler.h
  43 +++ b/arch/mips/include/asm/compiler.h
  44 @@ -1,5 +1,6 @@
  45  /*
  46   * Copyright (C) 2004, 2007  Maciej W. Rozycki
  47 + * Copyright (C) 2009  Wu Zhangjin, wuzj@lemote.com
  48   *
  49   * This file is subject to the terms and conditions of the GNU General Public
  50   * License.  See the file "COPYING" in the main directory of this archive
  51 @@ -16,4 +17,13 @@
  52  #define GCC_REG_ACCUM "accum"
  53  #endif
  54
  55 +#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4)
  56 +#define GCC_NO_H_CONSTRAINT
  57 +#ifdef CONFIG_64BIT
  58 +typedef unsigned int uintx_t __attribute__((mode(TI)));
  59 +#else
  60 +typedef u64 uintx_t;
  61 +#endif
  62 +#endif
  63 +
  64  #endif /* _ASM_COMPILER_H */
  65 --- a/arch/mips/include/asm/delay.h
  66 +++ b/arch/mips/include/asm/delay.h
  67 @@ -7,6 +7,7 @@
  68   * Copyright (C) 1995 - 2000, 01, 03 by Ralf Baechle
  69   * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
  70   * Copyright (C) 2007  Maciej W. Rozycki
  71 + * Copyright (C) 2009  Wu Zhangjin, wuzj@lemote.com
  72   */
  73  #ifndef _ASM_DELAY_H
  74  #define _ASM_DELAY_H
  75 @@ -48,6 +49,43 @@ static inline void __delay(unsigned long
  76                 : "0" (loops), "r" (1));
  77  }
  78
  79 +/*
  80 + * convert usecs to loops
  81 + *
  82 + * handle removal of 'h' constraint in GCC 4.4
  83 + */
  84 +
  85 +#ifndef GCC_NO_H_CONSTRAINT    /* gcc <= 4.3 */
  86 +static inline unsigned long __usecs_to_loops(unsigned long usecs,
  87 +               unsigned long lpj)
  88 +{
  89 +       unsigned long hi, lo;
  90 +
  91 +       if (sizeof(long) == 4)
  92 +               __asm__("multu\t%2, %3"
  93 +               : "=h" (usecs), "=l" (lo)
  94 +               : "r" (usecs), "r" (lpj)
  95 +               : GCC_REG_ACCUM);
  96 +       else if (sizeof(long) == 8 && !R4000_WAR)
  97 +               __asm__("dmultu\t%2, %3"
  98 +               : "=h" (usecs), "=l" (lo)
  99 +               : "r" (usecs), "r" (lpj)
 100 +               : GCC_REG_ACCUM);
 101 +       else if (sizeof(long) == 8 && R4000_WAR)
 102 +               __asm__("dmultu\t%3, %4\n\tmfhi\t%0"
 103 +               : "=r" (usecs), "=h" (hi), "=l" (lo)
 104 +               : "r" (usecs), "r" (lpj)
 105 +               : GCC_REG_ACCUM);
 106 +
 107 +       return usecs;
 108 +}
 109 +#else  /* GCC_NO_H_CONSTRAINT, gcc >= 4.4 */
 110 +static inline unsigned long __usecs_to_loops(unsigned long usecs,
 111 +               unsigned long lpj)
 112 +{
 113 +       return ((uintx_t)usecs * lpj) >> BITS_PER_LONG;
 114 +}
 115 +#endif
 116
 117  /*
 118   * Division by multiplication: you don't have to worry about
 119 @@ -62,8 +100,6 @@ static inline void __delay(unsigned long
 120
 121  static inline void __udelay(unsigned long usecs, unsigned long lpj)
 122  {
 123 -       unsigned long hi, lo;
 124 -
 125         /*
 126          * The rates of 128 is rounded wrongly by the catchall case
 127          * for 64-bit.  Excessive precission?  Probably ...
 128 @@ -77,23 +113,7 @@ static inline void __udelay(unsigned lon
 129                                    0x80000000ULL) >> 32);
 130  #endif
 131
 132 -       if (sizeof(long) == 4)
 133 -               __asm__("multu\t%2, %3"
 134 -               : "=h" (usecs), "=l" (lo)
 135 -               : "r" (usecs), "r" (lpj)
 136 -               : GCC_REG_ACCUM);
 137 -       else if (sizeof(long) == 8 && !R4000_WAR)
 138 -               __asm__("dmultu\t%2, %3"
 139 -               : "=h" (usecs), "=l" (lo)
 140 -               : "r" (usecs), "r" (lpj)
 141 -               : GCC_REG_ACCUM);
 142 -       else if (sizeof(long) == 8 && R4000_WAR)
 143 -               __asm__("dmultu\t%3, %4\n\tmfhi\t%0"
 144 -               : "=r" (usecs), "=h" (hi), "=l" (lo)
 145 -               : "r" (usecs), "r" (lpj)
 146 -               : GCC_REG_ACCUM);
 147 -
 148 -       __delay(usecs);
 149 +       __delay(__usecs_to_loops(usecs, lpj));
 150  }
 151
 152  #define __udelay_val cpu_data[raw_smp_processor_id()].udelay_val