[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] Re: [Bochs-developers] [PATCH 1/2] create acpi cpu definiti
[Qemu-devel] Re: [Bochs-developers] [PATCH 1/2] create acpi cpu definitions
Sun, 19 Apr 2009 23:34:43 +0930
On Sun, Apr 19, 2009 at 8:42 PM, Gleb Natapov <address@hidden> wrote:
> On Sun, Apr 19, 2009 at 11:01:09AM +0930, Brendan Trotter wrote:
>> For some reason (unknown to me) some Linux developers made wild
>> assumptions about disabled "Processor Local APIC" entries, and now
>> they're inventing fictitious hardware to support their unfounded
>> Please, correct me if I'm wrong...
> Windows 2008 supports CPU hot plug (but not unplug IIRC). How they do it
> if there is not specification about how it should work on 80x86?
Unfortunately, "no specification" typically means hardware
manufacturers are free to make up their own way of doing it - for all
I know Windows 2008 needs a special driver for each system.
I've been unable to find any 80x86 hardware that supports hot-add and
hot-remove of CPUs, and therefore I haven't been able to attempt to
find out how this hardware notifies the OS of capabilities, changes,
etc. The closest thing I did find is redundant CPUs, but I couldn't
find anything with details on how this works, and don't know if the OS
itself is made aware of it at all (the chipset and firmware might
replaced a failing CPU with a spare/redundant CPU without the OS's
I did find a little information for hot add/remove for Itanium
systems, but nothing very detailed. From
"The NEC Express5800/1000 Series of servers, utilizes the Intel
Itanium Machine Check Architecture (for error handling), and adds
another layer of intelligence through a service processor and related
software called the GlobalMaster (managed via either a GUI or CLI), as
shown in Figure 1. The GlobalMaster also communicates with Microsoft
Windows Server 2008 to convey to the operating system that a hot add
(of processors, memory, or I/O) or hot replace (of processors and
memory) is taking place."
This sounds like a lot more involved than just misappropriating a flag
in an ACPI table...