Re: [PATCH] tmpfs: fix uninitialized return value in shmem_link

[Qian Cai <cai@lca.pw>]

On Wed, 2019-02-27 at 09:09 -0500, Qian Cai wrote:
> On Mon, 2019-02-25 at 16:07 -0800, Linus Torvalds wrote:
> > On Mon, Feb 25, 2019 at 4:03 PM Qian Cai <cai@lca.pw> wrote:
> > > > 
> > > > Of course, that's just gcc. I have no idea what llvm ends up doing.
> > > 
> > > Clang 7.0:
> > > 
> > > # clang  -O2 -S -Wall /tmp/test.c
> > > /tmp/test.c:46:6: warning: variable 'ret' is used uninitialized whenever
> > > 'if'
> > > condition is false [-Wsometimes-uninitialized]
> > 
> > Ok, good.
> > 
> > Do we have any clang builds in any of the zero-day robot
> > infrastructure or something? Should we?
> > 
> > And maybe this was how Dan noticed the problem in the first place? Or
> > is it just because of his eagle-eyes?
> > 
> 
> BTW, even clang is able to generate warnings in your sample code, it does not
> generate any warnings when compiling the buggy shmem.o via "make CC=clang".
> Here is the objdump for arm64 (with KASAN_SW_TAGS inline).
> 

Ah, thanks to the commit 6e8d666e9253 ("Disable "maybe-uninitialized" warning
globally"), it will no longer generate this type of warnings until using "make
W=1" due to the commit a76bcf557ef4 ("Kbuild: enable -Wmaybe-uninitialized
warning for 'make W=1'"). Anyway, the generated code is the same using clang
with and without this patch.

    d_instantiate(dentry, inode);
4eec:       94000000        bl      0 <d_instantiate>
            ret = shmem_reserve_inode(inode->i_sb);
4ef0:       2a1f03e0        mov     w0, wzr             <---- ret = 0
    return ret;