r336876 breaks sysutils/acpi_call

[Konstantin Belousov]

On Sat, Aug 25, 2018 at 11:51:21PM -0400, Theron wrote:
> 
> A recent change in CURRENT has sysutils/acpi_call reliably cause a 
> kernel panic when run on a Dell XPS laptop system.  I bisected this to 
> r336876: Use SMAP on amd64.  I would have thought that this is a simple 
> compatibility problem requiring only a port update, except that the same 
> kernel and acpi_call on different hardware are not affected.  On the 
> problematic system, the kernel module loads without incident; it is when 
> executing ACPI commands, even normally harmless operations such as 
> requesting read-only constants, that the system freeze occurs.  ACPI 
> functionality seems otherwise unaffected.
Does system freeze or panic ?  I would expect the later.

The fact that it occurs on some system and not another is encouraging and
if my expectations are right, it fails on the system where SMAP is supported
by hardware.  If true, this means that this kernel module is already broken
and accesses userspace directly without using copyin(9).  Detecting such
situations and stopping the system is the whole point of SMAP.

> 
> Kernel debugging console and crash dumps are also broken on this system 
> (I suspect due to Intel graphics) however it is an unrelated problem, 
> and is only an excuse for my inability to provide any further crash 
> information.
> 
> Having already bisected to the breaking commit, is there anything else I 
> should do to improve the chances this problem gets fixed, or are there 
> any hardware compatibility notes I may have missed?