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

From: Steven Rostedt
Date: Wed Jan 31 2024 - 14:42:15 EST


On Wed, 31 Jan 2024 11:35:18 -0800
Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Wed, 31 Jan 2024 at 07:58, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> >
> > BTW, I ran my full test suite on your patches with the below updates and it
> > all passed.
>
> Those patch updates all look sane to me.
>
> > I can break up and clean up the patches so that they are bisectable, and if
> > that passes the bisectable portion of my tests, I can still send them to
> > you for 6.8.
>
> Ack. That series you posted looks fine. I didn't do any actual testing
> or applying the patches, just looking at them.
>
> The one thing I noticed is that the 'llist' removal still needs to be
> done. The logical point is that "[PATCH v2 7/7]" where the
> eventfs_workfn stuff is ripped out.
>
> And the 'rcu' head should now be a union with something that is no
> longer used after the last kref. The only thing that *is* used after
> the last kref is the "is_freed" bit, so there's lots of choice. Using
> the 'struct list_head listl' that is used for the child list would
> seem to be the obvious choice, but it could be anything (including all
> of the beginning of that eventfs_inode, but then you would need to
> group that as another nested unnamed struct, so picking a "big enough"
> entry like 'list' makes it syntactically simpler.

Yeah, that was what I was talking about in my cover letter with:

Note, there's more clean ups that can happen. One being cleaning up the
eventfs_inode structure. But that's not critical now and can be added
later.

I just want to get the majority of the broken parts done. The clean up of
the eventfs_inode is something that I'd add a separate patch. Not sure that
falls in your "fixes" category for 6.8.

-- Steve