Re: [linus:master] [eventfs] 852e46e239: BUG:unable_to_handle_page_fault_for_address

[Linus Torvalds]

On Tue, 30 Jan 2024 at 11:37, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
>
> Do you want me to put this in my urgent branch and mark them for stable,
> and then send them for 6.8?

Hmm. I think the only one that fixes a _reported_ bug is that [PTCH
2/6]. And it turns out that while 'ti->private' really is entirely
uninitialized (and I still think it's the cause of the kernel test
robot report that started this thread), the ti->flags field _is_
initialized to zero in tracefs_alloc_inode().

So even in that [PATCH 2/6], these parts:

  -     ti->flags |= TRACEFS_EVENT_INODE;
  +     ti->flags = TRACEFS_EVENT_INODE;

aren't strictly needed (but aren't wrong either).

The 'make sure to initialize ti->private before exposing the dentry'
part *definitely* needs to be part of 6.8, though. That has an
outstanding actually triggered bug report on it.

I do think that tracefs_alloc_inode() should also initialize
ti->private to NULL, but while that would fix the oops that the test
robot reported, it wouldn't fix the data-race on any ti->private
accesses.

So that "ti->private = ei" needs to be done before the d_instantiate()
(that later became a d_add()) regardless. But not having random fields
left uninitialized for future subtle bugs would be a good idea too.

Anyway.

If you do run the full tracefs tests on the whole series, and there
are no other major problems, I'll happily take it all for 6.8. And
yes, even mark it for stable. I think the other bugs are much harder
to hit, but I do think they exist. And code deletion is always good.

So give it the full test attention, and *if* it all still looks good
and there are no new subtle issues that crop up, let's just put this
saga behind us asap.

               Linus