Re: [PATCH v2] mm/page_isolation: fix a deadlock with printk()

[Michal Hocko]

On Wed 09-10-19 11:08:35, Qian Cai wrote:
> On Wed, 2019-10-09 at 16:34 +0200, Michal Hocko wrote:
> > On Wed 09-10-19 10:19:44, Qian Cai wrote:
> > > On Wed, 2019-10-09 at 15:51 +0200, Michal Hocko wrote:
> > 
> > [...]
> > > > Can you paste the full lock chain graph to be sure we are on the same
> > > > page?
> > > 
> > > WARNING: possible circular locking dependency detected
> > > 5.3.0-next-20190917 #8 Not tainted
> > > ------------------------------------------------------
> > > test.sh/8653 is trying to acquire lock:
> > > ffffffff865a4460 (console_owner){-.-.}, at:
> > > console_unlock+0x207/0x750
> > > 
> > > but task is already holding lock:
> > > ffff88883fff3c58 (&(&zone->lock)->rlock){-.-.}, at:
> > > __offline_isolated_pages+0x179/0x3e0
> > > 
> > > which lock already depends on the new lock.
> > > 
> > > 
> > > the existing dependency chain (in reverse order) is:
> > > 
> > > -> #3 (&(&zone->lock)->rlock){-.-.}:
> > >        __lock_acquire+0x5b3/0xb40
> > >        lock_acquire+0x126/0x280
> > >        _raw_spin_lock+0x2f/0x40
> > >        rmqueue_bulk.constprop.21+0xb6/0x1160
> > >        get_page_from_freelist+0x898/0x22c0
> > >        __alloc_pages_nodemask+0x2f3/0x1cd0
> > >        alloc_pages_current+0x9c/0x110
> > >        allocate_slab+0x4c6/0x19c0
> > >        new_slab+0x46/0x70
> > >        ___slab_alloc+0x58b/0x960
> > >        __slab_alloc+0x43/0x70
> > >        __kmalloc+0x3ad/0x4b0
> > >        __tty_buffer_request_room+0x100/0x250
> > >        tty_insert_flip_string_fixed_flag+0x67/0x110
> > >        pty_write+0xa2/0xf0
> > >        n_tty_write+0x36b/0x7b0
> > >        tty_write+0x284/0x4c0
> > >        __vfs_write+0x50/0xa0
> > >        vfs_write+0x105/0x290
> > >        redirected_tty_write+0x6a/0xc0
> > >        do_iter_write+0x248/0x2a0
> > >        vfs_writev+0x106/0x1e0
> > >        do_writev+0xd4/0x180
> > >        __x64_sys_writev+0x45/0x50
> > >        do_syscall_64+0xcc/0x76c
> > >        entry_SYSCALL_64_after_hwframe+0x49/0xbe
> > 
> > This one looks indeed legit. pty_write is allocating memory from inside
> > the port->lock. But this seems to be quite broken, right? The forward
> > progress depends on GFP_ATOMIC allocation which might fail easily under
> > memory pressure. So the preferred way to fix this should be to change
> > the allocation scheme to use the preallocated buffer and size it from a
> > context when it doesn't hold internal locks. It might be a more complex
> > fix than using printk_deferred or other games but addressing that would
> > make the pty code more robust as well.
> 
> I am not really sure if doing a surgery in pty code is better than fixing the
> memory offline side as a short-term fix.

If this was only about the memory offline code then I would agree. But
we are talking about any printk from the zone->lock context and that is
a bigger deal. Besides that it is quite natural that the printk code
should be more universal and allow to be also called from the MM
contexts as much as possible. If there is any really strong reason this
is not possible then it should be documented at least.

-- 
Michal Hocko
SUSE Labs