Re: [BUG REPORT] use of unreachable() masks uninitialized variables warnings

From: Theodore Ts'o
Date: Thu Feb 11 2016 - 09:06:46 EST


On Wed, Feb 10, 2016 at 08:13:09PM -0700, Jeff Merkey wrote:
> Here are the sources of several bugs I have seen recently in ext4 I am
> pretty sure with a null bh. One good check is to set the BUG() macro
> NOT TO call unreachable() as a build test since the compiler will
> ignore uninitialized variables in a function if someone calls BUG()
> even conditionally, and never report them during build.
>
> The following are from v4.4.1 with a BUG() macro with the call to
> unreachable() removed:

I checked all of the fs/ext4 warnings you listed and they are all
false positives.

> In file included from fs/ext4/file.c:30:0:
> fs/ext4/ext4_jbd2.h: In function âext4_inode_journal_modeâ:
> fs/ext4/ext4_jbd2.h:409:1: warning: control reaches end of non-void
> function [-Wreturn-type]
> }

This is from a:

if (foo) {
...
return foobie;
} else if (bar) {
...
return barbie;
} else {
BUG();
}

construct.

> fs/ext4/inode.c: In function âext4_map_blocksâ:
> fs/ext4/inode.c:548:5: warning: âretvalâ may be used uninitialized in
> this function [-Wmaybe-uninitialized]
> if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
> ^
> fs/ext4/extents.c:2305:14: warning: âlenâ may be used uninitialized in
> this function [-Wmaybe-uninitialized]
> ext4_lblk_t len;
> ^

All of the may be used uninitialized warnings are from a:

if (foo) {
...
retval = xxx;
} else if (bar) {
...
retval = yyy;
} else {
BUG();
}

construct.

It may be that there are some false warnings, but they certainly weren't
from warnings you've listed from ext4.

Cheers,

- Ted