On Fri, Dec 6, 2019 at 7:50 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> I assume it's the added "do_wakeup = 0" (not the spinlock) that ends
> up having some subtle issue.
Ahh, and then later that is removed, but when it is removed it also
remote the wakeup before the pipe_wait(). So whatever issue that
commit introduces ends up remaining.
I wonder if the extra wakeups ended up hiding some other bug. We do
extra wakeups on the write side too, with a "FIXME! Is this really
true?" comment..
Linus
