Bug in v7_coherent_kern_range() ?

[b32955@freescale.com (Huang Shijie)]

[1] Platform:
freescale's IMX6Q(4 cores) , ARM CORTEX-A9

[2] kernel:
3.0.15(I have cherry-picked many patches, and the arch/arm/mm/cache-v7.S
is same code with the latest kernel v3.4-rc1)
enable SMP, VIPT,

[3] application:
I use our our application which will clone many threads,
two threads (assume as A and B) may do the same thing at the same time
as the following code:

In most of the time, it's ok.
But in some unknown situation, cacheflush() failed and one threads
(assume A) may hung up in the following code:
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

open("/usr/lib/lib_mp3_dec_arm12_elinux.so.2.10.0", O_RDONLY) = 8
read(8,
"\177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0(\0\1\0\0\0\20\35\0\0004\0\0\0"...,
512) = 512
fstat64(8, {st_mode=S_IFREG|0644, st_size=56232, ...}) = 0
mmap2(NULL, 88032, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 8, 0)
= 0x2ff0a000
mprotect(0x2ff18000, 28672, PROT_NONE) = 0
mmap2(0x2ff1f000, 4096, PROT_READ|PROT_WRITE,
MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 8, 0xd) = 0x2ff1f000
close(8) = 0
mprotect(0x2ff0a000, 57344, PROT_READ|PROT_WRITE) = 0
mprotect(0x2ff0a000, 57344, PROT_READ|PROT_EXEC) = 0
cacheflush(0x2ff0a000, 0x2ff18000, 0, 0x6, 0x2cd03420) = 0 // System
hung up here!!!
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++


[4] kernel log
I use "echo t > /proc/sysrq-trigger" to show the tasks's information:
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

multiqueue0:src D 804cd678 0 7328 5963 0x00000001
[<804cd678>] (__schedule+0x228/0x760) from [<804d0564>]
(__down_read+0xa8/0xe0)
[<804d0564>] (__down_read+0xa8/0xe0) from [<800478c4>]
(do_page_fault+0xbc/0x480)
[<800478c4>] (do_page_fault+0xbc/0x480) from [<8003841c>]
(do_DataAbort+0x34/0x98)
[<8003841c>] (do_DataAbort+0x34/0x98) from [<8003df10>]
(__dabt_svc+0x70/0xa0)
Exception stack(0xbae37ea8 to 0xbae37ef0)
7ea0: 31e05000 31e1d000 00000020 0000001f 31e05000 31e1d000
7ec0: bfac86b8 31e05000 31e1d000 bae36000 08100075 31e056fc 31e08000
bae37ef0
7ee0: 800424a8 8004a1fc 800f0013 ffffffff
[<8003df10>] (__dabt_svc+0x70/0xa0) from [<8004a1fc>]
(v7_coherent_kern_range+0x20/0x80)
[<8004a1fc>] (v7_coherent_kern_range+0x20/0x80) from [<800424a8>]
(arm_syscall+0x2a0/0x2c4)
[<800424a8>] (arm_syscall+0x2a0/0x2c4) from [<8003e500>]
(ret_fast_syscall+0x0/0x3c)
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++


The do_cache_op() has already held the mm->mmap_sem, but
v7_coherent_kern_range()
cause one page fault during it flush the cache. deadlock! So it hung up
in the do_page_fault().

[5] questions:
Why the v7_coherent_kern_range() can caused the data abort?
Is there something wrong about the v7_coherent_kern_range()?


thanks
Huang Shijie