skyeye下面 insmod 模块的错误

peterpanpan

加载RTAI模块的时候出现错误
但是在pxa255板子上就没有错误
skyeye1.20
busybox1.1.2
#ash
麻烦大家告知 ：是skyeye insmod 的 bug么？


# insmod rtai_hal.ko
Unable to handle kernel paging request at virtual address bf000000
pgd = c3740000
[bf000000] *pgd=00000000
Internal error: Oops: 817 [#1]
Modules linked in:
CPU: 0
PC is at __memzero+0x24/0x80
LR is at 0x0
pc : [<c00d8164>]    lr : [<00000000>]    Not tainted
sp : c375de74  ip : 00000000  fp : c375dfa4
r10: 000004d8  r9 : c4851000  r8 : c4869f50
r7 : 00000004  r6 : c4852f68  r5 : c486a400  r4 : 0000001f
r3 : 00000000  r2 : 00000000  r1 : 00005cd4  r0 : bf000000
Flags: nzCv  IRQs on  FIQs on  Mode SVC_32  Segment user
Control: 3937  Table: A3740000  DAC: 00000015
Process insmod (pid: 681, stack limit = 0xc375c190)
Stack: (0xc375de74 to 0xc375e000)
de60:                                              c004bbc4 c00555ec c018573c
de80: c018573c 00000100 000000b8 00000000 c30d35dc c0019124 c30d35dc c375dec8
dea0: c375deac c00d46c4 c00d45ec 00000000 c1f6bbe0 bf000000 00000000 00000000
dec0: 00000000 00000007 00000000 0000000b 00000004 00000000 00000000 0000001e
dee0: 0000001d c4872bc8 c4869e33 c375c000 c375df54 c375df00 c00606c4 c005eb40
df00: c0019334 00000000 c02a6620 00000000 00000000 00000023 00000000 00000000
df20: c0019334 c001933c c001931c 00000002 c375c000 00000002 c375c000 40000000
df40: c02a6620 00000000 c375dfa4 c375df58 c0022864 c006bd94 00000002 00000000
df60: 00000000 00000000 00022a2f 00000001 00000002 00000003 00000000 00000003
df80: 00000000 00000100 00000080 c001eaf0 c375c000 00000002 00000000 c375dfa8
dfa0: c001e900 c004b4f4 00000000 00000100 00900080 40000000 00022a2f 000ce2e8
dfc0: 00000003 00000000 00000100 befffec4 beffff83 00008340 00000002 00000000
dfe0: befffc38 befffc2c 00014dcc 000660a0 00000010 00900080 00000000 00000000
Backtrace:
[<c004b4e8>] (sys_init_module+0x0/0x1460) from [<c001e900>] (ret_fast_syscall+0x0/0xc)
Code: e52de004 e1a0c002 e1a0e002 e2511040 (a8a0500c)
Segmentation fault

peterpanpan

2.6.10的内核
arm-linux-gcc3.4
编译最简单的 printk个hello的内核模块hello.ko
insmod hello.ko
同样出现这些调试信息错误
是不是skyeye不支持内核模块加载～～

chyyuu

你在哪里，如果在北京，我们可以好好交流一下。
我对你把rtai移植到skyeye上很感兴趣！

peterpanpan

陈渝老师 您没有看到我的论坛内留言么？
因为我不知道您邮箱 您如果要联系我 麻烦send a mail to [email protected] for contact
谢谢

arm-linux_Simon

请问skyeye模拟arm linux支持insmod吗?

ksh

目前在SkyEye上的arm linux运行insmod有些问题，有可能是SkyEye的一个bug

21cnbao

是的，确实是skyeye的一个bug，我跟了一下，找到了一个解决的方法。分析过程如下：
我们使用strace去寻找死在哪个系统调用：
# strace modprobe hello
execve("/sbin/modprobe", ["modprobe", "hello"], [/* 7 vars */]) = 0
uname({sys="Linux", node="lihacker", ...}) = 0
brk(0)                                  = 0x15d000
brk(0x15dd02)                           = 0x15dd02
set_tls(0x15d4a0, 0x1573c0, 0, 0x1, 0x15d4a0) = 0
brk(0x17ed02)                           = 0x17ed02
brk(0x17f000)                           = 0x17f000
getuid32()                              = 0
chdir("/lib/modules")                   = 0
uname({sys="Linux", node="lihacker", ...}) = 0
chdir("2.6.31.6-svn79")                 = 0
open("/proc/modules", O_RDONLY)         = 3
fstat64(3, {st_mode=S_IFREG|0444, st_size=0, ...}) = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x40000000
read(3, ""..., 1024)                    = 0
close(3)                                = 0
munmap(0x40000000, 4096)                = 0
lstat64("/etc/modprobe.conf", 0xbe846880) = -1 ENOENT (No such file or directory)
lstat64("/etc/modprobe.d", 0xbe846880)  = -1 ENOENT (No such file or directory)
open("modules.dep", O_RDONLY)           = 3
fstat64(3, {st_mode=S_IFREG|0644, st_size=17141, ...}) = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x40001000
read(3, "/lib/modules/2.6.31.6-svn79/kerne"..., 1024) = 1024
read(3, "nel/net/netfilter/x_tables.kon/li"..., 1024) = 1024
read(3, "/lib/modules/2.6.31.6-svn79/kerne"..., 1024) = 1024
read(3, "/net/netfilter/xt_esp.ko: /lib/mo"..., 1024) = 1024
read(3, "ables.ko /lib/modules/2.6.31.6-sv"..., 1024) = 1024
read(3, "es/2.6.31.6-svn79/kernel/net/netf"..., 1024) = 1024
read(3, ".6-svn79/kernel/net/netfilter/nf_"..., 1024) = 1024
read(3, "b/modules/2.6.31.6-svn79/kernel/n"..., 1024) = 1024
read(3, "etfilter/nf_conntrack_ipv4.ko /li"..., 1024) = 1024
read(3, ".6.31.6-svn79/kernel/net/netfilte"..., 1024) = 1024
read(3, "defrag_ipv4.kon/lib/modules/2.6.3"..., 1024) = 1024
read(3, "tfilter/nf_conntrack_ipv4.ko /lib"..., 1024) = 1024
read(3, ".6-svn79/kernel/net/ipv4/netfilte"..., 1024) = 1024
read(3, "1.6-svn79/kernel/net/netfilter/x_"..., 1024) = 1024
read(3, "ib/modules/2.6.31.6-svn79/kernel/"..., 1024) = 1024
read(3, "ib/modules/2.6.31.6-svn79/kernel/"..., 1024) = 1024
read(3, "o:n/lib/modules/2.6.31.6-svn79/ke"..., 1024) = 757
close(3)                                = 0
munmap(0x40001000, 4096)                = 0
open("/lib/modules/2.6.31.6-svn79/kernel/drivers/char/driver_examples/hello.ko", O_RDONLY) = 3
fstat64(3, {st_mode=S_IFREG|0644, st_size=2425, ...}) = 0
read(3, "177ELF111���������1�(�1�����������210"..., 3072) = 2425
read(3, ""..., 647)                     = 0
close(3)                                = 0
init_module(0x15e300, 2425, ""Unable to handle kernel paging request at virtual address bf000000
pgd = c0620000
[bf000000] *pgd=00000000
Internal error: Oops: 807 [#1]
Modules linked in:
CPU: 0    Not tainted  (2.6.31.6-svn79 #21)
PC is at __memzero+0x24/0x80
LR is at 0x0
pc : [<c0107404>]    lr : [<00000000>]    psr: 20000013
sp : c069ff0c  ip : 00000000  fp : 00000005
r10: c2807230  r9 : c2807000  r8 : c2807388
r7 : 00000004  r6 : c2807168  r5 : 00000012  r4 : bf000000
r3 : 00000000  r2 : 00000000  r1 : 00000464  r0 : bf000000
Flags: nzCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment user
Control: 0000317f  Table: c0620000  DAC: 00000015
Process modprobe (pid: 796, stack limit = 0xc069e270)
Stack: (0xc069ff0c to 0xc06a0000)
ff00:                            c005fba0 00000000 00000000 c2807518 c2807608
ff20: c28072dc c1139240 c28078a0 00000010 00000000 c2807388 00000000 c2807630
ff40: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
ff60: 00000000 00000979 0015e2f0 00000000 0015e300 c002bf48 c069e000 00000000
ff80: 00000000 c0060878 be846958 c002dbe0 00000000 00000979 0015e2f0 0015dfa8
ffa0: 00000080 c002bf0c 00000979 0015e2f0 0015e300 00000979 0015e2f0 00000000
ffc0: 00000979 0015e2f0 0015dfa8 00000080 0015e2a0 0015dfbc 00000001 00000000
ffe0: be846958 be846948 000177f8 00009350 20000010 0015e300 ebff1fc9 e3500000
[<c0107404>] (__memzero+0x24/0x80) from [<00000000>] (0x0)
Code: e52de004 e1a0c002 e1a0e002 e2511040 (a8a0500c)
---[ end trace 1e19dcbb1e19dcba ]---
<unfinished ...>
+++ killed by SIGSEGV +++

很明显，死在内核的linux-2.6.31/kernel/module.c的init_module()函数中,加上几个printk:

2453 SYSCALL_DEFINE3(init_module, void __user *, umod,
2454         unsigned long, len, const char __user *, uargs)
2455 {
2456     struct module *mod;
2457     int ret = 0;
2458
2459     printk("%s %dn", __func__, __LINE__);
2460
2461     /* Must have permission */
2462     if (!capable(CAP_SYS_MODULE) || modules_disabled)
2463         return -EPERM;
2464
2465     +printk("%s %dn", __func__, __LINE__);
2466     
2467     /* Only one module load at a time, please */
2468     if (mutex_lock_interruptible(&module_mutex) != 0)
2469         return -EINTR;
2470
2471     +printk("%s %dn", __func__, __LINE__);
2472     
2473     /* Do all the hard work */
2474     mod = load_module(umod, len, uargs);
2475     if (IS_ERR(mod)) {
2476         mutex_unlock(&module_mutex);
2477         return PTR_ERR(mod);
2478     }
2479
2480     +printk("%s %dn", __func__, __LINE__);
2481     

再次运行:

# modprobe hello
sys_init_module 2459
sys_init_module 2465
sys_init_module 2471
Unable to handle kernel paging request at virtual address bf000000
pgd = c05f0000
[bf000000] *pgd=00000000
Internal error: Oops: 807 [#1]
Modules linked in:

从打印信息可以看出，panic发生在load_module函数，既然"sys_init_module 2471"被打印而"sys_init_module 2480"未打印。 继续跟进load_module()，直到最后发现内核崩溃在memset(ptr, 0, mod->core_size);：

static noinline struct module *load_module(void __user *umod,
                   unsigned long len,
                   const char __user *uargs)
{
...
     ptr = module_alloc_update_bounds(mod->core_size);
...
     kmemleak_not_leak(ptr);
...
     memset(ptr, 0, mod->core_size);

一路跟踪源代码，发现ARM的Kernel会将模块的内存申请在3G-16M到3G的区域，而skyeye在内核空间处理3G以下地址的时候，会发生错误:

"Unable to handle kernel paging request at virtual address bf000000"

因此，workaround掉这个bug:

Index: arch/arm/kernel/module.c
===================================================================
--- arch/arm/kernel/module.c    (revision 87)
+++ arch/arm/kernel/module.c    (working copy)
@@ -38,7 +38,7 @@
#ifdef CONFIG_MMU
void *module_alloc(unsigned long size)
{
-       struct vm_struct *area;
+       /*struct vm_struct *area;

        size = PAGE_ALIGN(size);
        if (!size)
@@ -49,6 +49,9 @@
                return NULL;

        return __vmalloc_area(area, GFP_KERNEL, PAGE_KERNEL_EXEC);
+*/
+
+       return size == 0 ? NULL : kmalloc(size, GFP_KERNEL);
}
#else /* CONFIG_MMU */
void *module_alloc(unsigned long size)
@@ -59,7 +62,7 @@

void module_free(struct module *module, void *region)
{
-       vfree(region);
+       kfree(region);
}

int module_frob_arch_sections(Elf_Ehdr *hdr,

由于此workaround使用了kmalloc，如果内核模块很大，kmalloc申请的内存不够存放，则可以修改： include/linux/slab.h中的KMALLOC_SHIFT_HIGH宏定义。另外，此workaround成立的条件是mem=32M或小于32M。 现在我们可以自由地加载模块了，看看结果：

加载模块：
# modprobe hello
Hello World enter
# modprobe globalmem  globalmem_major=250
# modprobe globalfifo globalfifo_major=251

查看加载的模块:
# lsmod
    Not tainted
globalmem 3356 0 - Live 0xc0f66000
hello 1188 0 - Live 0xc1128000
globalfifo 4292 0 - Live 0xc182a000

查看设备：
# cat /proc/devices
Character devices:
  1 mem
  2 pty
  3 ttyp
  4 /dev/vc/0
  4 tty
  4 ttyS
  5 /dev/tty
  5 /dev/console
  5 /dev/ptmx
  7 vcs
10 misc
13 input
21 sg
29 fb
128 ptm
136 pts
204 s3c2410_serial
250 globalmem
251 globalfifo

建立结点：
# mknod /dev/globalmem c 250 0
# mknod /dev/globalfifo c 251 0
#
读写设备文件：
# echo "hello, cisco" > /dev/globalmem
written 13 bytes(s) from 0
# cat /dev/globalmem
read 4096 bytes(s) from 0
hello, cisco

# cat /dev/globalfifo &
# echo "hello,linux, I love you" > /dev/globalfifo
written 24 bytes(s),current_len:24
# read 24 bytes(s),current_len:0
hello,linux, I love you