I was stating to suspect this might be a hardware problem, so I sourced a
similar machine with a Radeon 6770M. Unfortunately the same fault exhibits.

[ 2060.703207] Kernel panic - not syncing: stack-protector: Kernel stack is
corrupted in: radeon_dp_needs_link_train+0x69/0x70 [radeon]
[ 2060.703221] CPU: 2 PID: 1386 Comm: kworker/2:2 Not tainted 5.3.4+ #19
[ 2060.703223] Hardware name: Apple Inc. iMac12,2/Mac-942B59F58194171B, BIOS
82.0.0.0.0 09/25/2018
[ 2060.703238] Workqueue: events radeon_dp_work_func [radeon]
[ 2060.703241] Call Trace:
[ 2060.703248]  dump_stack+0x46/0x5b
[ 2060.703253]  panic+0xf3/0x288
[ 2060.703274]  ? radeon_dp_needs_link_train+0x69/0x70 [radeon]
[ 2060.703278]  __stack_chk_fail+0x10/0x10
[ 2060.703294]  radeon_dp_needs_link_train+0x69/0x70 [radeon]
[ 2060.703308]  radeon_connector_hotplug+0xa4/0xd0 [radeon]
[ 2060.703323]  radeon_dp_work_func+0x28/0x40 [radeon]
[ 2060.703326]  process_one_work+0x1b4/0x330
[ 2060.703329]  worker_thread+0x44/0x3d0
[ 2060.703333]  kthread+0xeb/0x120
[ 2060.703336]  ? process_one_work+0x330/0x330
[ 2060.703338]  ? kthread_park+0xa0/0xa0
[ 2060.703344]  ret_from_fork+0x1f/0x30
[ 2060.703382] Kernel Offset: disabled
[ 2060.703385] ---[ end Kernel panic - not syncing: stack-protector: Kernel
stack is corrupted in: radeon_dp_needs_link_train+0x69/0x70 [radeon] ]---

I'm now at a bit of a loss as to how to debug it. I can reproduce it reliably
by playing a 1080p video on each head with mplayer and triggering the OSD
(volume/brightness) changes. I can't bisect it with an earlier kernel because
it doesn't occur when using the internal panel, one Thunderbolt and one HDMI,
and the dual thunderbolt support only turned up in 5.2.