On Fri, Dec 6, 2019 at 12:28 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> I think I found it.
>
> TOTALLY UNTESTED patch appended. It's whitespace-damaged and may be
> completely wrong. And might not fix it.
Well, it compiles, and it's obviously correct, so I've committed it.
It doesn't fix my "kernel compiles go single-threaded" issue. Which is
not surprising - make doesn't use splice(), it just reads and writes
single characters (the main make server writes a "+" character for
each available job, and parallel sub-makes will read one for each job
they start and write one when done - or something very close to that).
I think that is related to the other pipe changes, though - there were
some wakeup changes in there too. The btrfs problem was bisected to
the original commit, which is what I think my patch fixes.
Linus
