x86: don't destroy %rbp on kernel-mode faults
commit9d8ad5d6c7fce31fd2c0fd4fe9977bda3e92e340
authorVegard Nossum <vegard.nossum@gmail.com>
Fri, 27 Jun 2008 15:22:17 +0000 (27 17:22 +0200)
committerIngo Molnar <mingo@elte.hu>
Fri, 27 Jun 2008 15:45:59 +0000 (27 17:45 +0200)
tree1760e6576072b880a9b1f9c07879a33cce5306f7
parent543cf4cb3fe6f6cae3651ba918b9c56200b257d0
x86: don't destroy %rbp on kernel-mode faults

From the code:

    "B stepping K8s sometimes report an truncated RIP for IRET exceptions
    returning to compat mode. Check for these here too."

The code then proceeds to truncate the upper 32 bits of %rbp. This means
that when do_page_fault() is finally called, its prologue,

    do_page_fault:
        push %rbp
        movl %rsp, %rbp

will put the truncated base pointer on the stack. This means that the
stack tracer will not be able to follow the base-pointer changes and
will see all subsequent stack frames as unreliable.

This patch changes the code to use a different register (%rcx) for the
checking and leaves %rbp untouched.

Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Acked-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/kernel/entry_64.S