[llvmlinux] [i586] Boot issues

Behan Webster behanw at converseincode.com
Mon Oct 28 20:08:54 UTC 2013


On 10/08/13 11:24, Ghitulete Razvan wrote:
>
> Hi,
>
> I tried to run the kernel with gdb, but due to some gdb issues. It
> runs into a segmentation fault when trying to set a breakpoint on the
> llvm compiled kernel symbols. But I managed to speed up the whole
> compile-boot process by stripping down the config and using ccache so
> I think I found the root of the problem. For whatever reason it seems
> that the clang builtin_memcpy and builtin_memset crash. Although the
> symbols are defined in clang, whenever the kernel attempts to do a
> memcpy (for example) the cpu halts, which is consistent to an
> untreated hardware interrupt (i.e. SIGSEGV,  SIGFPE).
>
> I used the memcpy and memset from the llvmlinux patches which are
> usually used in arch/x86/boot and it seems to do the trick for normal
> mem[cpy/set] calls but, the issue still remains with structure
> operations which are translated by the compiler into the `faulty`
> builtin ones.
>
> Has this problem happened before? Are there any flags used that might
> [lead to/avoid] this behaviour? I wrote a small test program to check
> if indeed there is something general with the builtin memcpy/memset
> and they seem to work just fine for said test program.
>
Whoops. Sorry I didn't reply earlier.

I'm not sure I've heard of anyone else having this issue.  Certainly
this is a big deal if it's happening to others as well.

I'd certainly like to know why this happened.

Behan

-- 
Behan Webster
behanw at converseincode.com



More information about the LLVMLinux mailing list