Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 28 Jan 2022 18:02:36 +0100
From:   Sebastian Andrzej Siewior <bigeasy@linutronix.de>
To:     "Jason A. Donenfeld" <Jason@zx2c4.com>
Cc:     Jonathan =?utf-8?Q?Neusch=C3=A4fer?= <j.neuschaefer@gmx.net>,
        Andy Lutomirski <luto@amacapital.net>,
        LKML <linux-kernel@vger.kernel.org>,
        Theodore Ts'o <tytso@mit.edu>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>, Will Deacon <will@kernel.org>,
        Waiman Long <longman@redhat.com>,
        Boqun Feng <boqun.feng@gmail.com>,
        Thomas Gleixner <tglx@linutronix.de>
Subject: Re: "BUG: Invalid wait context" in invalidate_batched_entropy
Message-ID: <YfQhrA6P15KQTsrS@linutronix.de>
References: <YfMa0QgsjCVdRAvJ@latitude>
 <CAHmME9pb9A4SN6TTjNvvxKqw1L3gXVOX7KKihfEH4AgKGNGZ2A@mail.gmail.com>
 <YfOqsOiNfURyvFRX@linutronix.de>
 <CAHmME9qBj2gL=spp3AUEo-Ri4r0KQq-KrR-f6dG++xXQvzcA7A@mail.gmail.com>
 <YfQXgaAyBBEYSspV@linutronix.de>
 <CAHmME9qhdRbFT26beStH3JYLiE0mpAZz-jp1jJsJdJNZKFpbvA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <CAHmME9qhdRbFT26beStH3JYLiE0mpAZz-jp1jJsJdJNZKFpbvA@mail.gmail.com>
Precedence: bulk

On 2022-01-28 17:28:47 [+0100], Jason A. Donenfeld wrote:
> Hi Sebastian,
Hi Jason,

> Gotcha. Surely, then, Andy's patch at least goes some of the way
> toward fixing this, since it outright _removes_ a spinlock_t. There is
> still the other spinlock_t that you want removed, I realize, though
> this doesn't appear to be the one from Jonathan's bug report. It
> sounds like Andy's patch might be one side of the fix, and your patch
> the other?

Only if we want to keep that lock a raw_spinlock_t. And this change
extends your IRQ-off region. Before I was only worried about that one
lock and all the callers. Now we have a little more possibilities.

From looking at get_random_u32(), the whole worst case includes the
whole of extract_crng(). So we have the possible call chain
crng_reseed() -> crng_finalize_init() and here we have
wake_up_interruptible() and kill_fasync() which both can't be called
with disabled interrupts.

> Jason

Sebastian