Re: [PATCH] ACPI: fix Thunderbolt hotplug

From: Andy Lutomirski <luto@kernel.org>
To: Prarit Bhargava <prarit@redhat.com>, linux-kernel@vger.kernel.org
Cc: Robert Moore <robert.moore@intel.com>,
	Lv Zheng <lv.zheng@intel.com>,
	"Rafael J. Wysocki" <rafael.j.wysocki@intel.com>,
	Len Brown <lenb@kernel.org>,
	linux-acpi@vger.kernel.org, Mario_Limonciello@dell.com,
	stable@vger.kernel.org, devel@acpica.org
Subject: Re: [PATCH] ACPI: fix Thunderbolt hotplug
Date: Thu, 28 Apr 2016 11:38:45 -0700	[thread overview]
Message-ID: <b80830ab-20b4-bd75-5331-4ad55cbff355@kernel.org> (raw)
In-Reply-To: <1461760019-12930-1-git-send-email-prarit@redhat.com>

On 04/27/2016 05:26 AM, Prarit Bhargava wrote:
> Rafael, this patch is in the acpica.git tree as 7a3bd2d ("Dispatcher: Update
> thread ID for recursive method calls").  I've had many positive testing
> results from hardware vendors and users with this patch and this resolves
> many of the problems seen here:
>
> https://bugzilla.kernel.org/show_bug.cgi?id=115121
>
> This does not fix the problems with the TB docking station.  Although this
> patch will also be required, the docking station issues require a FW update.
> Updated FW should be coming soon to resolve those problems.
>

No kidding!?!  Will that update be installable from Linux?  If so, could 
you cc me on that?

> P.
>
> ----8<----
>
> The following hung task trace is seen when hotplugging
> an ethernet dongle in a Thunderbolt port on Linux.
>
> INFO: task kworker/0:4:1468 blocked for more than 120 seconds.
>       Tainted: G        W       4.6.0-rc1+ #1
>  "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
>  kworker/0:4     D ffff8802a265ba38 13344  1468      2 0x00000000
>  Workqueue: kacpid acpi_os_execute_deferred
>  ffff8802a265ba38 ffff8802a265ba00 ffffffff81130200 ffffffff81e0d580
>  ffff88029e5eb340 ffff8802a265c000 ffff88029d69d000 ffff88029e5eb340
>  ffffffff818c1b8d ffff8802b64e8758 ffff8802a265ba50 ffffffff818bdfcc
>  Call Trace:
>  [<ffffffff81130200>] ? test_callback+0x10/0x30
>  [<ffffffff818c1b8d>] ? __down_timeout+0x5d/0xd0
>  [<ffffffff818bdfcc>] schedule+0x3c/0x90
>  [<ffffffff818c2d60>] schedule_timeout+0x210/0x360
>  [<ffffffff8103fc89>] ? sched_clock+0x9/0x10
>  [<ffffffff810ee51c>] ? local_clock+0x1c/0x20
>  [<ffffffff81110c16>] ? mark_held_locks+0x76/0xa0
>  [<ffffffff818c3cfc>] ? _raw_spin_unlock_irq+0x2c/0x40
>  [<ffffffff818c1b8d>] ? __down_timeout+0x5d/0xd0
>  [<ffffffff81110d35>] ? trace_hardirqs_on_caller+0xf5/0x1b0
>  [<ffffffff818c1b8d>] ? __down_timeout+0x5d/0xd0
>  [<ffffffff818c1bac>] __down_timeout+0x7c/0xd0
>  [<ffffffff818c44b2>] ? _raw_spin_lock_irqsave+0x82/0x90
>  [<ffffffff8110b87c>] down_timeout+0x4c/0x60
>  [<ffffffff814e3a9c>] acpi_os_wait_semaphore+0xaa/0x16a
>  [<ffffffff81510f37>] acpi_ex_system_wait_mutex+0x81/0xfa
>  [<ffffffff814f8796>] acpi_ds_begin_method_execution+0x25a/0x373
>  [<ffffffff814f8cea>] acpi_ds_call_control_method+0x107/0x2e0
>  [<ffffffff8151ed60>] acpi_ps_parse_aml+0x177/0x495
>  [<ffffffff8151fa46>] acpi_ps_execute_method+0x1f7/0x2b9
>  [<ffffffff81516c9a>] acpi_ns_evaluate+0x2ee/0x435
>  [<ffffffff814ff84a>] acpi_ev_asynch_execute_gpe_method+0xbd/0x159
>  [<ffffffff814e2a69>] acpi_os_execute_deferred+0x17/0x23
>  [<ffffffff810d1fc2>] process_one_work+0x242/0x700
>  [<ffffffff810d1f3a>] ? process_one_work+0x1ba/0x700
>  [<ffffffff810d24ce>] worker_thread+0x4e/0x490
>  [<ffffffff810d2480>] ? process_one_work+0x700/0x700
>  [<ffffffff810d2480>] ? process_one_work+0x700/0x700
>  [<ffffffff810d98b1>] kthread+0x101/0x120
>  [<ffffffff81110d35>] ? trace_hardirqs_on_caller+0xf5/0x1b0
>  [<ffffffff818c4832>] ret_from_fork+0x22/0x50
>  [<ffffffff810d97b0>] ? kthread_create_on_node+0x250/0x250
>  2 locks held by kworker/0:4/1468:
>  #0:  ("kacpid"){.+.+.+}, at: [<ffffffff810d1f3a>] process_one_work+0x1ba/0x700
>  #1:  ((&dpc->work)){+.+.+.}, at: [<ffffffff810d1f3a>] process_one_work+0x1ba/0x700
>
> The issue appears to be that the kworker thread attempts to acquire the
> _E42 method's mutex twice when executing  acpi_ps_execute_method() and
> recursing through the entry method.
>
> The current code does take the possiblity of this recursion into account,
> however, it is only for the case where the walk_state has been populated.
>
> This can be fixed by setting the thread id in the !walk_state case to
> allow for recursion.

Tested-by: Andy Lutomirski <luto@kernel.org> # On a Dell XPS 13 9350

I don't care at all what variant of this patch is applied, but it would 
be nice to get it in for 4.6.  This fixes hotplug for me and I suspect 
it also fixes some occasional suspend/resume failures I've seen.

--Andy