linux-kernel - Re: [git pull] drm fixes for 5.7-rc8/final

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAHk-=wgo1HUhSj-kGO8u+iUCxp+QS+rNenbM8gywbF3pdQ_DQA@mail.gmail.com>
Date:   Thu, 28 May 2020 18:49:14 -0700
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Dave Airlie <airlied@...il.com>,
        Chris Wilson <chris@...is-wilson.co.uk>,
        Rodrigo Vivi <rodrigo.vivi@...el.com>
Cc:     Daniel Vetter <daniel.vetter@...ll.ch>,
        dri-devel <dri-devel@...ts.freedesktop.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [git pull] drm fixes for 5.7-rc8/final

On Thu, May 28, 2020 at 5:21 PM Dave Airlie <airlied@...il.com> wrote:
>
> Seems to have wound down nicely, a couple of i915 fixes, amdgpu fixes
> and minor ingenic fixes.

Dave, this doesn't even build. WTF?

In drivers/gpu/drm/i915/gt/selftest_lrc.c, there's a
engine_heartbeat_disable() function that takes two arguments, but the
new "live_timeslice_nopreempt()" gives it only one.

I'd blame a merge problem, since the failure is in new code, but the
problem exists in your top-of-tree, not just my merge.

In fact, it's not even your merge of the i915 tree that is the source
of the problem (although the fact that you clearly didn't _test_ the
end result most definitely is _part_ of the problem!).

Because the problem exists in the commit that introduced that thing:
commit 1f65efb624c4 ("drm/i915/gt: Prevent timeslicing into
unpreemptable requests").

It's garbage, and never compiled.

See here:

  git grep -1wh engine_heartbeat_disable 1f65efb62 \
        drivers/gpu/drm/i915/gt/selftest_lrc.c

and you'll see how the definition of that function looks like this:

  static void engine_heartbeat_disable(struct intel_engine_cs *engine,
                                       unsigned long *saved)

but then in the middle of that grep, you'll find

                engine_heartbeat_disable(engine);

WTF?

That commit seems to be a cherry-pick of another commit, and maybe it
worked in that other context (which I don't see), but it sure doesn't
work in the context it was then cherry-picked into.

So people took that thing, and it went through at least two different
people WHO NEVER EVEN BOTHERED TO TEST IF IT BUILDS!

Christ, people.

This is why I absolutely DO NOT WANT TO SEE random rebases or
cherry-picks and then sending the resulting untested crap on to me.

Because it's exactly that: untested crap.

It doesn't matter *how* well you have tested a commit in some original
context: the moment you rebase it (or cherry-pick it, which is just
another form of rebasing), it's a completely new commit in a
completely new environment, and all your old testing is null and void.

Guys, get your act together! I should not be getting these kinds of
shit pull requests days before a release!

And how the hell did this not get any build testing at any point
before asking me to pull?

                Linus