听说使用ICC编译Linux内核，性能可以提升40%，我的第一感觉是，应该有提升，也没这么夸张吧，应该是有人估计夸大，玩数字游戏。 从4月1日起，开始尝试编译。从Intel官网下载ICC 10.0.022，在虚拟机的Fedora10上安装(安装需要licence)。然后从LinuxDNA.com下载2.6.22内核代码，下载最新的封装脚本(http://linuxdna.com/wrapper.html)，使用POE提供给我的一个补丁(针对2.6.18)，下载linuxdna配置文件(http://www.linuxdna.com/configdna)放到内核根目录 手工对内核代码进行了修改，按照LinuxDNA网页的简单说明编译内核。 刚刚把内核安装成功，发现启动速度比原来(2.6.22)快了一半，我都怀疑是不是有其他问题，但目前的使用都正常，特别是响应速度明显提升，也不知道是不是心理作用。下面需要一些具体的测试，才能说明问题。 不过进入系统发现lsmod的结果为空，没有加载任何驱动，网络也不能用，这可能是快的原因吧，到底性能有没有提升，还是得测试。 备注：LinuxDNA是一个使用ICC编译Linux内核的项目，旨在提高内核性能。
1. First get icc from Intel.com and install it.
2. Download the source code of kernel. You may need to fix a patch. Or change the source code refer to some patch. You can find them on the LinuxDNA’s GoogleGroup.
3. Get the config file from (http://www.linuxdna.com/configdna) and copy it to the root directory of kernel source code.
4. Next get the shell script file from (http://linuxdna.com/wrapper.html) and copy it to /usr/bin, set the filename as intelwrapper.
5. Next run make menuconfig / oldconfig to configure your kernel again
6. Compile kernel with “make HOSTCC=intelwrapper CC=intelwrapper Ar=xiar LD=xild”
(note:This is the patch for 2.6.22. My icc version is 10.0.022, So please change the icc version number in this patch when you use it )
diff -ur linux-2.6.22.old/arch/i386/boot/compressed/relocs.c linux-2.6.22/arch/i386/boot/compressed/relocs.c
--- linux-2.6.22.old/arch/i386/boot/compressed/relocs.c 2009-04-06 19:04:22.000000000 +0800
+++ linux-2.6.22/arch/i386/boot/compressed/relocs.c 2009-04-06 19:10:28.000000000 +0800
@@ -10,7 +10,7 @@
#define USE_BSD
#include
-#define MAX_SHDRS 100
+#define MAX_SHDRS 1000
#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
static Elf32_Ehdr ehdr;
static Elf32_Shdr shdr[MAX_SHDRS];
diff -ur linux-2.6.22.old/Makefile linux-2.6.22/Makefile
--- linux-2.6.22.old/Makefile 2009-04-06 19:04:03.000000000 +0800
+++ linux-2.6.22/Makefile 2009-04-06 19:11:51.000000000 +0800
@@ -610,7 +610,7 @@
quiet_cmd_vmlinux__ ?= LD $@
cmd_vmlinux__ ?= $(LD) $(LDFLAGS) $(LDFLAGS_vmlinux) -o $@
-T $(vmlinux-lds) $(vmlinux-init)
- --start-group $(vmlinux-main) /opt/intel/cc/10.1.018/lib/libirc_s.a --end-group
+ --start-group $(vmlinux-main) /opt/intel/cc/10.1.022/lib/libirc_s.a --end-group
$(filter-out $(vmlinux-lds) $(vmlinux-init) $(vmlinux-main) FORCE ,$^)
# Generate new vmlinux version
diff -ur linux-2.6.22.old/mm/slab.c linux-2.6.22/mm/slab.c
--- linux-2.6.22.old/mm/slab.c 2009-04-06 19:04:03.000000000 +0800
+++ linux-2.6.22/mm/slab.c 2009-04-06 19:12:04.000000000 +0800
@@ -326,7 +326,7 @@
{
extern void __bad_size(void);
- if (__builtin_constant_p(size)) {
+// if (__builtin_constant_p(size)) {
int i = 0;
#define CACHE(x)
@@ -336,9 +336,9 @@
i++;
#include "linux/kmalloc_sizes.h"
#undef CACHE
- __bad_size();
- } else
- __bad_size();
+// __bad_size();
+// } else
+// __bad_size();
return 0;
}
diff -ur linux-2.6.22.old/scripts/Makefile.build linux-2.6.22/scripts/Makefile.build
--- linux-2.6.22.old/scripts/Makefile.build 2009-04-06 19:04:02.000000000 +0800
+++ linux-2.6.22/scripts/Makefile.build 2009-04-06 19:12:58.000000000 +0800
@@ -299,7 +299,7 @@
cmd_link_multi-y = $(LD) $(ld_flags) -r -o $@ $(link_multi_deps)
quiet_cmd_link_multi-m = LD [M] $@
-cmd_link_multi-m = $(LD) $(ld_flags) $(LDFLAGS_MODULE) -o $@ $(link_multi_deps)
+cmd_link_multi-m = $(LD) $(ld_flags) $(LDFLAGS_MODULE) -o $@ $(link_multi_deps) /opt/intel/cc/10.1.022/lib/libirc_s.a
# We would rather have a list of rules like
# foo.o: $(foo-objs)
diff -ur linux-2.6.22.old/scripts/Makefile.modpost linux-2.6.22/scripts/Makefile.modpost
--- linux-2.6.22.old/scripts/Makefile.modpost 2009-04-06 19:04:02.000000000 +0800
+++ linux-2.6.22/scripts/Makefile.modpost 2009-04-06 19:13:10.000000000 +0800
@@ -98,7 +98,7 @@
# Step 6), final link of the modules
quiet_cmd_ld_ko_o = LD [M] $@
cmd_ld_ko_o = $(LD) $(LDFLAGS) $(LDFLAGS_MODULE) -o $@
- $(filter-out FORCE,$^)
+ $(filter-out FORCE,$^) /opt/intel/cc/10.1.022/lib/libirc_s.a
$(modules): %.ko :%.o %.mod.o FORCE
$(call if_changed,ld_ko_o)
