powerpc/perf: Fix ABIv2 kernel backtraces
authorAnton Blanchard <anton@samba.org>
Tue, 26 Aug 2014 02:44:15 +0000 (12:44 +1000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 5 Oct 2014 20:41:08 +0000 (13:41 -0700)
commitcc8dcb6944fd20dffaca942a1235c0f54220232c
tree33b28b34f096a9f9aa8d3850265ecbe9c9d7136e
parent981b26114255f6908ca0570550b834f9ab71503e
powerpc/perf: Fix ABIv2 kernel backtraces

commit 85101af13bb854a6572fa540df7c7201958624b9 upstream.

ABIv2 kernels are failing to backtrace through the kernel. An example:

39.30%  readseek2_proce  [kernel.kallsyms]    [k] find_get_entry
            |
            --- find_get_entry
               __GI___libc_read

The problem is in valid_next_sp() where we check that the new stack
pointer is at least STACK_FRAME_OVERHEAD below the previous one.

ABIv1 has a minimum stack frame size of 112 bytes consisting of 48 bytes
and 64 bytes of parameter save area. ABIv2 changes that to 32 bytes
with no paramter save area.

STACK_FRAME_OVERHEAD is in theory the minimum stack frame size,
but we over 240 uses of it, some of which assume that it includes
space for the parameter area.

We need to work through all our stack defines and rationalise them
but let's fix perf now by creating STACK_FRAME_MIN_SIZE and using
in valid_next_sp(). This fixes the issue:

30.64%  readseek2_proce  [kernel.kallsyms]    [k] find_get_entry
            |
            --- find_get_entry
               pagecache_get_page
               generic_file_read_iter
               new_sync_read
               vfs_read
               sys_read
               syscall_exit
               __GI___libc_read

Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/powerpc/include/asm/ptrace.h
arch/powerpc/perf/callchain.c