On Fri, Jun 1, 2018 at 6:01 AM, Kees Cook [off-list ref] wrote:

On Thu, May 31, 2018 at 6:56 PM, Masahiro Yamada
[off-list ref] wrote:

quoted

2018-05-31 12:53 GMT+09:00 Kees Cook [off-list ref]:

quoted

On Wed, May 30, 2018 at 6:26 PM, Kees Cook [off-list ref] wrote:

quoted

On Wed, May 30, 2018 at 6:12 PM, Masahiro Yamada

quoted

quoted

quoted

quoted

This has been triggered by the following commit:


commit 0e461945f3504e09b8ecf947b6398adce1287a28
Author: Masahiro Yamada [off-list ref]
Date:   Mon May 28 18:22:07 2018 +0900

    gcc-plugins: allow to enable GCC_PLUGINS for COMPILE_TEST



CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL was previously disabled
for COMPILE_TEST, which is now enabled.

For the moment, can you add "depends on !COMPILE_TEST" to
CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL in your tree and I'll continue
to figure out what's happening?

I ran into the same thing indepently and bisected it (which led me to
arrive at this thread).
One additional bit of information I have is that this happens with all
versions of
gcc-7 for me, but not gcc-6.3 or older.

Another finding was the particular instance I noticed:

fs/ext4/inode.c: In function 'ext4_inode_csum':
fs/ext4/inode.c:83:1: warning: the frame size of 1688 bytes is larger
than 500 bytes [-Wframe-larger-than=]

comes from inlining the same function multiple times; ext4_inode_csum()
repeatedly calls ext4_chksum(), which has a struct on the stack. Apparently
this normally only takes up stack space only once, but when initializing it
to zero, each instance takes an additional two CRYPTO_MINALIGN bytes
of stack space (the size of the locally defined structure).


       Arnd