kern/94939: [acpi] [patch] reboot(8) fails on IBM / Intel blades

[Nate Lawson]

The following reply was made to PR kern/94939; it has been noted by GNATS.

From: Nate Lawson <nate at root.org>
To: John Baldwin <jhb at freebsd.org>
Cc: bug-followup at freebsd.org, dodell at ixsystems.com
Subject: Re: kern/94939: [acpi] [patch] reboot(8) fails on IBM / Intel blades
Date: Tue, 28 Mar 2006 11:32:14 -0800

 John Baldwin wrote:
 > On Tuesday 28 March 2006 14:08, Nate Lawson wrote:
 >>  >>>
 >> 4.7.3.6 Reset Register
 >> The optional ACPI reset mechanism specifies a standard mechanism that 
 >> provides a complete system reset.  When implemented, this mechanism must 
 >> reset the entire system. This includes processors, core logic, all
 >> buses, and all peripherals. From an OSPM perspective, asserting the 
 >> reset mechanism is the logical equivalent to power cycling the machine. 
 >> Upon gaining control after a reset, OSPM will perform actions in
 >> like manner to a cold boot.
 >> ...
 >> The system must reset immediately following the write to this register. 
 >> OSPM assumes that the processor will not execute beyond the write 
 >> instruction. OSPM should execute spin loops on the CPUs in the system 
 >> following a write to this register.
 >>  >>>
 >>
 >> So I'm ok with the patch being committed if no other tasks need to 
 >> happen after this shutdown handler is called.  Also, all APs should be 
 >> stopped before this happens and it should only occur once on the BSP.
 > 
 > Does the reset mechanism require that ACPI be "functioning"?  That is,
 > does it have to happen before the call to AcpiTerminate() or can it
 > happen afterwards?  If it can happen afterwards, it should probably be
 > moved to be part of cpu_reset_real().
 
 It *probably* works without acpi enabled because on x86 at least, it's 
 just a write to the SMM io port, which triggers an SMI and the handler 
 resets the machine.  SMM is present whether in legacy mode or acpi mode. 
   However, I don't want to put acpi-specific resets in cpu_reset_real() 
 because then acpi is mandatory for linking the kernel.  Let's just try 
 it in the place the patch puts it for now and see if there are any problems.
 
 The patch has some other major problems that should be addressed before 
 committing.  It should not manually be parsing the GAS and mapping 
 memory etc.  Instead, it should just use AcpiHwLowLevelWrite():
 
 ACPI_STATUS
 AcpiHwLowLevelWrite (
      UINT32                  Width,
      UINT32                  Value,
      ACPI_GENERIC_ADDRESS    *Reg);
 
 Width should be 8, value should be the reset value in the FADT, and Reg 
 should be the FADT GAS struct.
 
 -- 
 Nate