Re: [PATCH 6/7] mm: convert generic code to 5-level paging

[Michal Hocko]

On Thu 09-03-17 14:47:16, Kirill A. Shutemov wrote:
> On Thu, Mar 09, 2017 at 10:54:15AM +0100, Michal Hocko wrote:
> > On Wed 08-03-17 18:21:30, Kirill A. Shutemov wrote:
[...]
> > > We can drop the hack once all architectures that support kasan would be
> > > converted to pgtable-nop4d.h -- amd64 and x86 at the moment.
> > 
> > But those architectures even do not enable kasan
> > $ git grep "select *HAVE_ARCH_KASAN"
> > arch/arm64/Kconfig:     select HAVE_ARCH_KASAN if SPARSEMEM_VMEMMAP && !(ARM64_16K_PAGES && ARM64_VA_BITS_48)
> > arch/x86/Kconfig:       select HAVE_ARCH_KASAN                  if X86_64 && SPARSEMEM_VMEMMAP
> > 
> > both arm64 and x86 (64b) do compile fine without the ifdef... So I guess
> > we should be fine without it.
> 
> Have you build the image to the final linking? lm_alias() hides the error
> until later.
> 
> x86-64 allmodconfig without the #ifndef:
> 
>   MODPOST vmlinux.o
> mm/built-in.o: In function `kasan_populate_zero_shadow':
> (.init.text+0xb72b): undefined reference to `kasan_zero_p4d'
> Makefile:983: recipe for target 'vmlinux' failed
> make: *** [vmlinux] Error 1

Interesting
arm64 cross compile:
$ grep CONFIG_KASAN .config
CONFIG_KASAN=y
CONFIG_KASAN_OUTLINE=y
# CONFIG_KASAN_INLINE is not set

Compiling for arm64 with aarch64-linux using gcc 4.9.0
[...]
  LD      vmlinux.o
  MODPOST vmlinux.o
  KSYM    .tmp_kallsyms1.o
  KSYM    .tmp_kallsyms2.o
  LD      vmlinux
  SORTEX  vmlinux
  SYSMAP  System.map

x86_64 crosscompile with the same version to rule out gcc version
changes

$ grep CONFIG_KASAN .config
CONFIG_KASAN_SHADOW_OFFSET=0xdffffc0000000000
CONFIG_KASAN=y
CONFIG_KASAN_OUTLINE=y
# CONFIG_KASAN_INLINE is not set

[...]
  LD      init/built-in.o
  LD      vmlinux.o
  MODPOST vmlinux.o
mm/built-in.o: In function `kasan_populate_zero_shadow':
(.init.text+0x84e5): undefined reference to `kasan_zero_p4d'
Makefile:983: recipe for target 'vmlinux' failed

no idea why arm64 build was OK.

Anyway I am not insisting on removing this ifdef it is just too ugly to
spread __ARCH_HAS_5LEVEL_HACK outside of the arch code. We have few more
in the mm code but those look much more understandable. Maybe a short
comment explaining the ifdef would be better.

Thanks!
-- 
Michal Hocko
SUSE Labs