On Thursday, October 28, 2010 2:50:42 pm Scott Long wrote:
On Thu, 28 Oct 2010, John Baldwin wrote:
On Thursday, October 28, 2010 1:01:24 pm Scott Long wrote:
On Oct 28, 2010, at 10:54 AM, John Baldwin wrote:
[ cc'ing acpi@ to be safe, but I think the topic warrants the wider
audience
of arch@ ]
I think we should drop support for having acpi load as a module for i386.
It
adds extra complication and hacks to the i386 APIC and interrupt code
that are
gratuitously different from amd64 as a result. Originally it was made a
module so that GENERIC on i386 did not include ACPI by default but would
only
use up memory to hold ACPI-related code if the machine supported ACPI.
Now
that acpi is part of GENERIC on i386 in 8.0 and later this argument is no
longer relevant. I'd like to remove support for ACPI as a module to
remove
the various hacks on i386 and reduce differences with amd64.
Just to be clear, it'll still be an optional kernel device, it just won't
be a KLD anymore, right? If you do that, what will happen with the
evil
bootloader code that gropes around for the AML tables and auto-loads the
module? Is there any reason to keep that around for compatibility?
If it
goes away, don't forget to also update the bootforth code that knows how to
manipulate it.
It already does the right thing in this case (it did regardless, but that
was
part of the testing before enabling 'device acpi' in GENERIC for 8.0). If
we remove the KLD support then we can now remove that code from the loader
and Forth scripts as they will no longer be needed.
You lost me, what is the right thing. What I'm asking is whether there
will be any surprises to people upgrading from 8.0 to 8.x with regard to
the bootloader no longer autoloading acpi.ko, and will there be any
surprises to those who update their bootblocks but maybe switch back and
forth between old and new kernels?
The loader code just sets 'acpi_load=YES'. If acpi is compiled into the
kernel or it is not present it just silently fails. This was already
considered and tested during the 8.0 release cycle.
I am only proposing making this change for 9, FYI, not to MFC it. If we
were to remove the code from the loader that sets acpi_load in 9 and someone
booted an 8.x or 7.x kernel that did not include 'device acpi', then acpi.ko
would not be autoloaded. We could easily leave the code in the loader around
until 10.0 so there is one release branch worth of compatibility, though the
fact that GENERIC i386 in 8 ships with acpi compiled in and not a module on
i386 is probably already giving us a release branch of compatibility as it is.
That is, a GENERIC 8.0 i386 kernel would work fine with a loader that removed
the ACPI bits, only a 7.x GENERIC kernel would fail to autoload acpi.ko with a
modified loader. Given that we don't generally support 7.x - 9.0 upgrades, I
really think it would be ok to remove the loader support from 9.
However, what I really care about are the kernel changes, not the loader
changes.
The loader changes could wait until 10 if really necessary.
--
John Baldwin
___
freebsd-acpi@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-acpi
To unsubscribe, send any mail to freebsd-acpi-unsubscr...@freebsd.org
