]> asedeno.scripts.mit.edu Git - linux.git/commit
powerpc/modules: Fix crashes by adding CONFIG_RELOCATABLE to vermagic
authorMichael Ellerman <mpe@ellerman.id.au>
Mon, 26 Mar 2018 08:27:01 +0000 (19:27 +1100)
committerMichael Ellerman <mpe@ellerman.id.au>
Mon, 9 Apr 2018 03:21:02 +0000 (13:21 +1000)
commit73aca179d78eaa11604ba0783a6d8b2125fbc332
treededb304b7c231117a49c5946b3e34dc8fb3faf44
parent49a695ba723224875df50e327bd7b0b65dd9a56b
powerpc/modules: Fix crashes by adding CONFIG_RELOCATABLE to vermagic

If you build the kernel with CONFIG_RELOCATABLE=n, then install the
modules, rebuild the kernel with CONFIG_RELOCATABLE=y and leave the
old modules installed, we crash something like:

  Unable to handle kernel paging request for data at address 0xd000000018d66cef
  Faulting instruction address: 0xc0000000021ddd08
  Oops: Kernel access of bad area, sig: 11 [#1]
  Modules linked in: x_tables autofs4
  CPU: 2 PID: 1 Comm: systemd Not tainted 4.16.0-rc6-gcc_ubuntu_le-g99fec39 #1
  ...
  NIP check_version.isra.22+0x118/0x170
  Call Trace:
    __ksymtab_xt_unregister_table+0x58/0xfffffffffffffcb8 [x_tables] (unreliable)
    resolve_symbol+0xb4/0x150
    load_module+0x10e8/0x29a0
    SyS_finit_module+0x110/0x140
    system_call+0x58/0x6c

This happens because since commit 71810db27c1c ("modversions: treat
symbol CRCs as 32 bit quantities"), a relocatable kernel encodes and
handles symbol CRCs differently from a non-relocatable kernel.

Although it's possible we could try and detect this situation and
handle it, it's much more robust to simply make the state of
CONFIG_RELOCATABLE part of the module vermagic.

Fixes: 71810db27c1c ("modversions: treat symbol CRCs as 32 bit quantities")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/include/asm/module.h