Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
        id S1755537AbcJZPvx (ORCPT <rfc822;w@1wt.eu>);
        Wed, 26 Oct 2016 11:51:53 -0400
Received: from mail-ua0-f179.google.com ([209.85.217.179]:36823 "EHLO
        mail-ua0-f179.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S1752176AbcJZPvv (ORCPT
        <rfc822;linux-kernel@vger.kernel.org>);
        Wed, 26 Oct 2016 11:51:51 -0400
MIME-Version: 1.0
In-Reply-To: <CAHc6FU4e5sueLi7pfeXnSbuuvnc5PaU3xo5Hnn=SvzmQ+ZOEeg@mail.gmail.com>
References: <CAHc6FU4e5sueLi7pfeXnSbuuvnc5PaU3xo5Hnn=SvzmQ+ZOEeg@mail.gmail.com>
From: Andy Lutomirski <luto@amacapital.net>
Date: Wed, 26 Oct 2016 08:51:29 -0700
Message-ID: <CALCETrUt+4ojyscJT1AFN5Zt3mKY0rrxcXMBOUUJzzLMWXFXHg@mail.gmail.com>
Subject: Re: CONFIG_VMAP_STACK, on-stack struct, and wake_up_bit
To: Andreas Gruenbacher <agruenba@redhat.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Peter Zijlstra <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>, LKML <linux-kernel@vger.kernel.org>,
        Bob Peterson <rpeterso@redhat.com>,
        Steven Whitehouse <swhiteho@redhat.com>
Content-Type: text/plain; charset=UTF-8
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
Content-Length: 1318
Lines: 46

On Wed, Oct 26, 2016 at 5:51 AM, Andreas Gruenbacher
<agruenba@redhat.com> wrote:
> Hi,
>
> CONFIG_VMAP_STACK has broken gfs2 and I'm trying to figure out what's
> going on. What I'm seeing is the following: on a fresh gfs2 filesystem
> created with:
>
>   mkfs.gfs2 -p lock_nolock $DEVICE
>
> I get the following BUG with 4.9-rc2, CONFIG_VMAP_STACK and
> CONFIG_DEBUG_VIRTUAL turned on:
>
>   kernel BUG at arch/x86/mm/physaddr.c:26!
>
> Stack of kernel thread:
>
>   __phys_addr(x)
>   bit_waitqueue(word, bit)
>   wake_up_bit(word = &gh->gh_iflags, bit = HIF_WAIT)
>   gfs2_holder_wake(gh)

It's this:

const struct zone *zone = page_zone(virt_to_page(word));

If the stack is vmalloced, then you can't find the page's zone like
that.  We could look it up the slow way (ick!), but maybe another
solution would be to do:

wait_queue_head_t *wait_table;
if (virt_addr_valid(word))
  wait_table = page_zone(virt_to_page(word))->wait_table;
else
  wait_table = funny_wait_table;

where funny_wait_table is an extra wait table just for funny addresses.

This will scale poorly on very large NUMA systems where many zones are
simultaneously using on-stack wait_bit bits, but I suspect this is a
very rare use case.

>
> Is accessing a struct on another kernel thread's stack no longer working?

That part should be fine.