linux-kernel - Re: [WARNING] kernel/rcu/tree.c:1058 rcu_irq

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 for Android: free password hash cracker in your pocket

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

Date:   Mon, 5 Oct 2020 09:52:06 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>,
        LKML <linux-kernel@...r.kernel.org>, svens@...ux.ibm.com
Subject: Re: [WARNING] kernel/rcu/tree.c:1058 rcu_irq_enter+0x15/0x20

On Fri, Oct 02, 2020 at 08:13:13PM +0200, Peter Zijlstra wrote:
> > checks within the irq disabling to get rid of the using cpu pointers within
> > preemptable code warnings
> 
> Ah, I think I lost a s/__this_cpu_read/raw_cpu_read/ somewhere. The
> thing is, if we're preemptible/migratable it will be 0 on both CPUs and
> it doesn't matter which 0 we read. If it is !0, IRQs will be disabled
> and we can't get migrated.

Aargh, this isn't in fact correct, and that means I have to revert:

  fddf9055a60d ("lockdep: Use raw_cpu_*() for per-cpu variables")

The trouble is that on a bunch of architectures we can read the other
CPUs variable from the new CPU, long after the old CPU has continued
executing things.

So this really needs to be this_cpu_read(). Luckily Sven has already
fixed s390, but let me go audit all the other archs.