also apply gcc4.4.0 specific patches to 2.6.27 (#5318)
authorFlorian Fainelli <florian@openwrt.org>
Sat, 13 Jun 2009 20:36:48 +0000 (20:36 +0000)
committerFlorian Fainelli <florian@openwrt.org>
Sat, 13 Jun 2009 20:36:48 +0000 (20:36 +0000)
SVN-Revision: 16439

target/linux/generic-2.6/patches-2.6.27/022-mips_div64_gcc4.4.0.patch [new file with mode: 0644]
target/linux/generic-2.6/patches-2.6.27/023-mips_delay_gcc4.4.0.patch [new file with mode: 0644]

diff --git a/target/linux/generic-2.6/patches-2.6.27/022-mips_div64_gcc4.4.0.patch b/target/linux/generic-2.6/patches-2.6.27/022-mips_div64_gcc4.4.0.patch
new file mode 100644 (file)
index 0000000..c01765d
--- /dev/null
@@ -0,0 +1,171 @@
+From: Ralf Baechle <ralf@linux-mips.org>
+Date: Thu, 30 Apr 2009 16:14:56 +0000 (+0200)
+Subject: MIPS: Rewrite <asm/div64.h> to work with gcc 4.4.0.
+X-Git-Url: http://www.linux-mips.org/git?p=linux.git;a=commitdiff_plain;h=a1b68289997030df64cba8478d5767fe10e42a58
+
+MIPS: Rewrite <asm/div64.h> to work with gcc 4.4.0.
+
+The inline assembler used on 32-bit kernels was using the "h" constraint
+which was considered dangerous and removed for gcc 4.4.0.
+
+Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
+---
+
+--- a/include/asm-mips/div64.h
++++ b/include/asm-mips/div64.h
+@@ -6,105 +6,63 @@
+  * License.  See the file "COPYING" in the main directory of this archive
+  * for more details.
+  */
+-#ifndef _ASM_DIV64_H
+-#define _ASM_DIV64_H
++#ifndef __ASM_DIV64_H
++#define __ASM_DIV64_H
+-#include <linux/types.h>
++#include <asm-generic/div64.h>
+-#if (_MIPS_SZLONG == 32)
++#if BITS_PER_LONG == 64
+-#include <asm/compiler.h>
++#include <linux/types.h>
+ /*
+  * No traps on overflows for any of these...
+  */
+-#define do_div64_32(res, high, low, base) ({ \
+-      unsigned long __quot32, __mod32; \
+-      unsigned long __cf, __tmp, __tmp2, __i; \
+-      \
+-      __asm__(".set   push\n\t" \
+-              ".set   noat\n\t" \
+-              ".set   noreorder\n\t" \
+-              "move   %2, $0\n\t" \
+-              "move   %3, $0\n\t" \
+-              "b      1f\n\t" \
+-              " li    %4, 0x21\n" \
+-              "0:\n\t" \
+-              "sll    $1, %0, 0x1\n\t" \
+-              "srl    %3, %0, 0x1f\n\t" \
+-              "or     %0, $1, %5\n\t" \
+-              "sll    %1, %1, 0x1\n\t" \
+-              "sll    %2, %2, 0x1\n" \
+-              "1:\n\t" \
+-              "bnez   %3, 2f\n\t" \
+-              " sltu  %5, %0, %z6\n\t" \
+-              "bnez   %5, 3f\n" \
+-              "2:\n\t" \
+-              " addiu %4, %4, -1\n\t" \
+-              "subu   %0, %0, %z6\n\t" \
+-              "addiu  %2, %2, 1\n" \
+-              "3:\n\t" \
+-              "bnez   %4, 0b\n\t" \
+-              " srl   %5, %1, 0x1f\n\t" \
+-              ".set   pop" \
+-              : "=&r" (__mod32), "=&r" (__tmp), \
+-                "=&r" (__quot32), "=&r" (__cf), \
+-                "=&r" (__i), "=&r" (__tmp2) \
+-              : "Jr" (base), "0" (high), "1" (low)); \
+-      \
+-      (res) = __quot32; \
+-      __mod32; })
+-
+-#define do_div(n, base) ({ \
+-      unsigned long long __quot; \
+-      unsigned long __mod; \
+-      unsigned long long __div; \
+-      unsigned long __upper, __low, __high, __base; \
+-      \
+-      __div = (n); \
+-      __base = (base); \
+-      \
+-      __high = __div >> 32; \
+-      __low = __div; \
+-      __upper = __high; \
+-      \
+-      if (__high) \
+-              __asm__("divu   $0, %z2, %z3" \
+-                      : "=h" (__upper), "=l" (__high) \
+-                      : "Jr" (__high), "Jr" (__base) \
+-                      : GCC_REG_ACCUM); \
+-      \
+-      __mod = do_div64_32(__low, __upper, __low, __base); \
+-      \
+-      __quot = __high; \
+-      __quot = __quot << 32 | __low; \
+-      (n) = __quot; \
+-      __mod; })
+-
+-#endif /* (_MIPS_SZLONG == 32) */
+-
+-#if (_MIPS_SZLONG == 64)
+-
+-/*
+- * Hey, we're already 64-bit, no
+- * need to play games..
+- */
+-#define do_div(n, base) ({ \
+-      unsigned long __quot; \
+-      unsigned int __mod; \
+-      unsigned long __div; \
+-      unsigned int __base; \
+-      \
+-      __div = (n); \
+-      __base = (base); \
+-      \
+-      __mod = __div % __base; \
+-      __quot = __div / __base; \
+-      \
+-      (n) = __quot; \
+-      __mod; })
++#define __div64_32(n, base)                                           \
++({                                                                    \
++      unsigned long __cf, __tmp, __tmp2, __i;                         \
++      unsigned long __quot32, __mod32;                                \
++      unsigned long __high, __low;                                    \
++      unsigned long long __n;                                         \
++                                                                      \
++      __high = *__n >> 32;                                            \
++      __low = __n;                                                    \
++      __asm__(                                                        \
++      "       .set    push                                    \n"     \
++      "       .set    noat                                    \n"     \
++      "       .set    noreorder                               \n"     \
++      "       move    %2, $0                                  \n"     \
++      "       move    %3, $0                                  \n"     \
++      "       b       1f                                      \n"     \
++      "        li     %4, 0x21                                \n"     \
++      "0:                                                     \n"     \
++      "       sll     $1, %0, 0x1                             \n"     \
++      "       srl     %3, %0, 0x1f                            \n"     \
++      "       or      %0, $1, %5                              \n"     \
++      "       sll     %1, %1, 0x1                             \n"     \
++      "       sll     %2, %2, 0x1                             \n"     \
++      "1:                                                     \n"     \
++      "       bnez    %3, 2f                                  \n"     \
++      "        sltu   %5, %0, %z6                             \n"     \
++      "       bnez    %5, 3f                                  \n"     \
++      "2:                                                     \n"     \
++      "        addiu  %4, %4, -1                              \n"     \
++      "       subu    %0, %0, %z6                             \n"     \
++      "       addiu   %2, %2, 1                               \n"     \
++      "3:                                                     \n"     \
++      "       bnez    %4, 0b\n\t"                                     \
++      "        srl    %5, %1, 0x1f\n\t"                               \
++      "       .set    pop"                                            \
++      : "=&r" (__mod32), "=&r" (__tmp),                               \
++        "=&r" (__quot32), "=&r" (__cf),                               \
++        "=&r" (__i), "=&r" (__tmp2)                                   \
++      : "Jr" (base), "0" (__high), "1" (__low));                      \
++                                                                      \
++      (__n) = __quot32;                                               \
++      __mod32;                                                        \
++})
+-#endif /* (_MIPS_SZLONG == 64) */
++#endif /* BITS_PER_LONG == 64 */
+-#endif /* _ASM_DIV64_H */
++#endif /* __ASM_DIV64_H */
diff --git a/target/linux/generic-2.6/patches-2.6.27/023-mips_delay_gcc4.4.0.patch b/target/linux/generic-2.6/patches-2.6.27/023-mips_delay_gcc4.4.0.patch
new file mode 100644 (file)
index 0000000..f001b81
--- /dev/null
@@ -0,0 +1,163 @@
+From: Wu Zhangjin <wuzj@lemote.com>
+
+the gcc 4.4 support for MIPS mostly refer to this PATCH:
+http://www.nabble.com/-PATCH--MIPS:-Handle-removal-of-%27h%27-constraint-in-GCC-4.4-td22192768.html
+but have been tuned a little.
+
+because only gcc 4.4 have loongson-specific support, so, we need to
+choose the suitable -march argument for gcc <= 4.3 and gcc >= 4.4, and
+we also need to consider use -march=loongson2e and -march=loongson2f for
+loongson2e and loongson2f respectively. this is handled by adding two
+new kernel options: CPU_LOONGSON2E and CPU_LOONGSON2F(thanks for the
+solutin provided by ZhangLe).
+
+I have tested it on FuLoong(2f) in 32bit and 64bit with gcc-4.4 and
+gcc-4.3. so, basically, it works.
+
+Signed-off-by: Wu Zhangjin <wuzhangjin@gmail.com>
+---
+ arch/mips/Makefile               |    9 +++++-
+ arch/mips/include/asm/compiler.h |   10 ++++++
+ arch/mips/include/asm/delay.h    |   58 +++++++++++++++++++++++++------------
+ 3 files changed, 57 insertions(+), 20 deletions(-)
+
+diff --git a/arch/mips/Makefile b/arch/mips/Makefile
+index a25c2e5..1ee5504 100644
+--- a/arch/mips/Makefile
++++ b/arch/mips/Makefile
+@@ -120,7 +120,14 @@ cflags-$(CONFIG_CPU_R4300)        += -march=r4300 -Wa,--trap
+ cflags-$(CONFIG_CPU_VR41XX)   += -march=r4100 -Wa,--trap
+ cflags-$(CONFIG_CPU_R4X00)    += -march=r4600 -Wa,--trap
+ cflags-$(CONFIG_CPU_TX49XX)   += -march=r4600 -Wa,--trap
+-cflags-$(CONFIG_CPU_LOONGSON2)        += -march=r4600 -Wa,--trap
++
++# only gcc >= 4.4 have the loongson-specific support
++cflags-$(CONFIG_CPU_LOONGSON2)        += -Wa,--trap
++cflags-$(CONFIG_CPU_LOONGSON2E)       += $(shell if [ $(call cc-version) -lt 0440 ] ; then \
++      echo $(call cc-option,-march=r4600); else echo $(call cc-option,-march=loongson2e); fi ;)
++cflags-$(CONFIG_CPU_LOONGSON2F)       += $(shell if [ $(call cc-version) -lt 0440 ] ; then \
++      echo $(call cc-option,-march=r4600); else echo $(call cc-option,-march=loongson2f); fi ;)
++
+ cflags-$(CONFIG_CPU_MIPS32_R1)        += $(call cc-option,-march=mips32,-mips32 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS32) \
+                       -Wa,-mips32 -Wa,--trap
+ cflags-$(CONFIG_CPU_MIPS32_R2)        += $(call cc-option,-march=mips32r2,-mips32r2 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS32) \
+diff --git a/include/asm-mips/compiler.h b/include/asm-mips/compiler.h
+index 71f5c5c..95256a8 100644
+--- a/include/asm-mips/compiler.h
++++ b/include/asm-mips/compiler.h
+@@ -1,5 +1,6 @@
+ /*
+  * Copyright (C) 2004, 2007  Maciej W. Rozycki
++ * Copyright (C) 2009  Wu Zhangjin, wuzj@lemote.com
+  *
+  * This file is subject to the terms and conditions of the GNU General Public
+  * License.  See the file "COPYING" in the main directory of this archive
+@@ -16,4 +17,13 @@
+ #define GCC_REG_ACCUM "accum"
+ #endif
++#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4)
++#define GCC_NO_H_CONSTRAINT
++#ifdef CONFIG_64BIT
++typedef unsigned int uintx_t __attribute__((mode(TI)));
++#else
++typedef u64 uintx_t;
++#endif
++#endif
++
+ #endif /* _ASM_COMPILER_H */
+diff --git a/include/asm-mips/delay.h b/include/asm-mips/delay.h
+index b0bccd2..00d7969 100644
+--- a/include/asm-mips/delay.h
++++ b/include/asm-mips/delay.h
+@@ -7,6 +7,7 @@
+  * Copyright (C) 1995 - 2000, 01, 03 by Ralf Baechle
+  * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
+  * Copyright (C) 2007  Maciej W. Rozycki
++ * Copyright (C) 2009  Wu Zhangjin, wuzj@lemote.com
+  */
+ #ifndef _ASM_DELAY_H
+ #define _ASM_DELAY_H
+@@ -48,6 +49,43 @@ static inline void __delay(unsigned long loops)
+               : "0" (loops), "r" (1));
+ }
++/*
++ * convert usecs to loops
++ *
++ * handle removal of 'h' constraint in GCC 4.4
++ */
++
++#ifndef GCC_NO_H_CONSTRAINT   /* gcc <= 4.3 */
++static inline unsigned long __usecs_to_loops(unsigned long usecs,
++              unsigned long lpj)
++{
++      unsigned long hi, lo;
++
++      if (sizeof(long) == 4)
++              __asm__("multu\t%2, %3"
++              : "=h" (usecs), "=l" (lo)
++              : "r" (usecs), "r" (lpj)
++              : GCC_REG_ACCUM);
++      else if (sizeof(long) == 8 && !R4000_WAR)
++              __asm__("dmultu\t%2, %3"
++              : "=h" (usecs), "=l" (lo)
++              : "r" (usecs), "r" (lpj)
++              : GCC_REG_ACCUM);
++      else if (sizeof(long) == 8 && R4000_WAR)
++              __asm__("dmultu\t%3, %4\n\tmfhi\t%0"
++              : "=r" (usecs), "=h" (hi), "=l" (lo)
++              : "r" (usecs), "r" (lpj)
++              : GCC_REG_ACCUM);
++
++      return usecs;
++}
++#else /* GCC_NO_H_CONSTRAINT, gcc >= 4.4 */
++static inline unsigned long __usecs_to_loops(unsigned long usecs,
++              unsigned long lpj)
++{
++      return ((uintx_t)usecs * lpj) >> BITS_PER_LONG;
++}
++#endif
+ /*
+  * Division by multiplication: you don't have to worry about
+@@ -62,8 +100,6 @@ static inline void __delay(unsigned long loops)
+ static inline void __udelay(unsigned long usecs, unsigned long lpj)
+ {
+-      unsigned long hi, lo;
+-
+       /*
+        * The rates of 128 is rounded wrongly by the catchall case
+        * for 64-bit.  Excessive precission?  Probably ...
+@@ -77,23 +113,7 @@ static inline void __udelay(unsigned long usecs, unsigned long lpj)
+                                  0x80000000ULL) >> 32);
+ #endif
+-      if (sizeof(long) == 4)
+-              __asm__("multu\t%2, %3"
+-              : "=h" (usecs), "=l" (lo)
+-              : "r" (usecs), "r" (lpj)
+-              : GCC_REG_ACCUM);
+-      else if (sizeof(long) == 8 && !R4000_WAR)
+-              __asm__("dmultu\t%2, %3"
+-              : "=h" (usecs), "=l" (lo)
+-              : "r" (usecs), "r" (lpj)
+-              : GCC_REG_ACCUM);
+-      else if (sizeof(long) == 8 && R4000_WAR)
+-              __asm__("dmultu\t%3, %4\n\tmfhi\t%0"
+-              : "=r" (usecs), "=h" (hi), "=l" (lo)
+-              : "r" (usecs), "r" (lpj)
+-              : GCC_REG_ACCUM);
+-
+-      __delay(usecs);
++      __delay(__usecs_to_loops(usecs, lpj));
+ }
+ #define __udelay_val cpu_data[raw_smp_processor_id()].udelay_val
+-- 
+1.6.0.4
+
+
+