[PATCH v6 3/6] x86, efi, kasan: #undef memset/memcpy/memmove per arch.

2015-09-29 11:38 GMT+03:00 Ingo Molnar [off-list ref]:

* Andrey Ryabinin [off-list ref] wrote:

quoted

In not-instrumented code KASAN replaces instrumented
memset/memcpy/memmove with not-instrumented analogues
__memset/__memcpy/__memove.
However, on x86 the EFI stub is not linked with the kernel.
It uses not-instrumented mem*() functions from
arch/x86/boot/compressed/string.c
So we don't replace them with __mem*() variants in EFI stub.

On ARM64 the EFI stub is linked with the kernel, so we should
replace mem*() functions with __mem*(), because the EFI stub
runs before KASAN sets up early shadow.

So let's move these #undef mem* into arch's asm/efi.h which is
also included by the EFI stub.

Also, this will fix the warning in 32-bit build reported by
kbuild test robot [off-list ref]:
      efi-stub-helper.c:599:2: warning: implicit declaration of function 'memcpy'

Signed-off-by: Andrey Ryabinin <ryabinin.a.a@gmail.com>
---
 arch/x86/include/asm/efi.h             | 12 ++++++++++++
 drivers/firmware/efi/libstub/efistub.h |  4 ----
 2 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/arch/x86/include/asm/efi.h b/arch/x86/include/asm/efi.h
index 155162e..6db2742 100644
--- a/arch/x86/include/asm/efi.h
+++ b/arch/x86/include/asm/efi.h

@@ -86,6 +86,18 @@ extern u64 asmlinkage efi_call(void *fp, ...);
 extern void __iomem *__init efi_ioremap(unsigned long addr, unsigned long size,
                                      u32 type, u64 attribute);

+/*
+ * CONFIG_KASAN may redefine memset to __memset.
+ * __memset function is present only in kernel binary.
+ * Since the EFI stub linked into a separate binary it
+ * doesn't have __memset(). So we should use standard
+ * memset from arch/x86/boot/compressed/string.c
+ * The same applies to memcpy and memmove.
+ */
+#undef memcpy
+#undef memset
+#undef memmove

Hm, so this hack got upstream via -mm, and it breaks the 64-bit x86 build with
some configs:

 arch/x86/platform/efi/efi.c:673:3: error: implicit declaration of function ?memcpy? [-Werror=implicit-function-declaration]
 arch/x86/platform/efi/efi_64.c:139:2: error: implicit declaration of function ?memcpy? [-Werror=implicit-function-declaration]
 ./arch/x86/include/asm/desc.h:121:2: error: implicit declaration of function ?memcpy? [-Werror=implicit-function-declaration]

I guess it's about EFI=y but KASAN=n. Config attached.

It's actually, it's about KMEMCHECK=y and KASAN=n, because declaration
of memcpy() is hidden under ifndef.

arch/x86/include/asm/string_64.h:
    #ifndef CONFIG_KMEMCHECK
    #if (__GNUC__ == 4 && __GNUC_MINOR__ >= 3) || __GNUC__ > 4
    extern void *memcpy(void *to, const void *from, size_t len);
    #else
    #define memcpy(dst, src, len)                                   \
    .......
    #endif
    #else
    /*
     * kmemcheck becomes very happy if we use the REP instructions
unconditionally,
     * because it means that we know both memory operands in advance.
     */
    #define memcpy(dst, src, len) __inline_memcpy((dst), (src), (len))
    #endif

So it also broke build with GCCs 4.0 - 4.3.
And it also breaks clang build, because AFAIK clang defines GNUC,
GNUC_MINOR as 4.2.

beyond fixing the build bug ... could we also engineer this in a better fashion
than spreading random #undefs across various KASAN unrelated headers?

I think we can add something like -DNOT_KERNEL (anyone has a better name ?)
to the CFLAGS for everything that is not linked with the kernel binary
(efistub, arch/x86/boot)

So, if NOT_KERNEL is defined we will not #define memcpy(), so we won't
need these undefs.

Thanks,

        Ingo