BUG_ON/panic proliferation.

[Dave Jones <davej@redhat.com>]

On Thu, Sep 27, 2012 at 09:59:13AM -0700, Linus Torvalds wrote:

 > We have too many f*cking BUG_ON's in the kernel, and the fact that one
 > triggers and it has taken a month and a half without it even being
 > resolved is a problem.

This has bothered me for a while.

$ rgrep BUG_ON drivers/ | wc -l
4018
$ rgrep WARN_ON drivers/ | wc -l
2415
$ rgrep panic drivers/ | wc -l
997

$ rgrep BUG_ON fs | wc -l
2792
$ rgrep WARN_ON fs | wc -l
524
$ rgrep panic fs | wc -l
381

The number of 'raw' panic calls makes me wonder if there's perhaps
a lot of people who just don't realise that BUG_ON includes a panic() call.
If we had named it PANIC_ON it would have made it more obvious that
it has more dire consequences than WARN_ON.

Every time I read a "everything locked up while I was in X" bug,
I can't help but think it was one of these.

But even without X, sometimes these are painful.
I even had to change VM_BUG_ON to use WARN instead of BUG a while ago
in my local tree, because the panic locked up the machine before
the console had a chance to finish dumping.

What would be a good way forward though ? With tens of thousands of them in tree,
auditting them one-by-one and replacing them as we go is going to take forever.

checkpatch already has a check for new additions of BUG/BUG_ON, but the
majority of patches go into the tree unchecked, so that's not particularly helpful.

	Dave