Re: CVS commit: src/sys/kern

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Thu, Feb 21, 2008 at 12:36:33PM +0100, Matthias Drochner wrote:
> -As said, eg. in the RB_HALT case on i386 we want the keyboard
>  to respond at the "press any key" prompt.

This is the same assumption as being able to use the keyboard during
early boot. It is not that different at all. Depending on the type of
hardware, the parents has to be in certain states. David's idea of
explicitly detaching all devices and restoring a sane state is good
enough for that. It is much less a hack than what you just introduced.

> -On platforms where the firmware doesn't deal with PCI power
>  management, it might not get the hardware out of D3
>  on reboot. In case it does a hardware reset, it is fine,
>  otherwise not. mi code can't know.

Hardware is not explict put into D3 on reboot or shutdown.

> -PCI devices in extender boxes can't be reset by the BIOS,
>  so they would stay dead forever.

They can. The BIOS is broken otherwise.

> -People tend to reboot the box if there are symptoms of
>  beginning memory corruption. Thus only code which is
>  really necessary should be called in the shutdown path,
>  and complex data structures avoided if possible.

You are arguing with broken BIOSes. The same kind of firmware doesn't
stop e.g. bus master devices from randomly corrupting system memory.
That has been a known issue for *long* time. Most of the old shutdown
hooks tried to do that and nothing else.

That said I request you to back this change out.

Joerg