Is OLPC going for ACPI support?

[Mitch Bradley]

Manish Regmi wrote:
> On 3/3/07, Mitch Bradley <wmb at firmworks.com> wrote:
>
>>
>> The OLPC OS does not use the ACPI interface, nor does the OLPC firmware
>> support it .  The OS drivers perform the appropriate power management
>> operations directly, with full knowledge of the behavior of the hardware
>> on which they operate.  The firmware exports a low-level suspend-to-RAM
>> function in the form of a subroutine.  The OS, after doing whatever
>> driver work is necessary, calls that subroutine.  The firmware then
>> saves the core state and puts the hardware into a low-power state with
>> RAM powered in self-refresh mode.  Upon a wakeup event, the firmware
>> restores the core state and returns from the subroutine.
>>
>> The current size of the firmware suspend/resume code is about 1K.  It is
>> written in assembly language, optimized to do its job as quickly as
>> possible.  We hope to be able to do "micro sleeps" so fast that the user
>> will not realize that the system is sleeping.
>>
>> OLPC has special hardware features for retaining the image on the screen
>> and for performing wireless mesh routing while the main processor is
>> powered off.
>
> Thank you very much for such a nice explanation.
> In the code i saw a lot of "magic" port numbers. Is there any
> documentation describing them (esp. ones not found in traditional PC).
>
>
The suspend/resume code uses MSR registers that are documented in the 
AMD chip specs for the Geode GX and 5536 companion chips.  The I/O ports 
are either legacy PC ports (for example the interrupt controller and 
interval timer) or Geode-specific devices documented by the AMD chip 
spec (for example MFGPT and SMBUS registers).  The OLPC wiki has 
pointers to those AMD chip specs.  The base port numbers and addresses 
for the Geode devices are established by MSR registers, as follows:

I/O ports:

18b0 SMBUS
1000 GPIO
1800 MFGPT
1880 IRQ mapper
1400 PM
1841 ACPI
1480 AC-97

Memory space:

fd000000 Frame buffer
fe000000 Graphics processor
fe004000 Video processor
fe008000 Display controller
fe01a000 OHCI (USB1)
fe01b000 EHCI (USB2)
efc00000 UOC (USB option controller)
fe00c000 CaFe NAND
fe010000 CaFe SD
fe014000 CaFe Camera

These addresses can be "discovered" by doing PCI configuration reads to 
PCI device base address registers or by browing the Open Firmware device 
tree.

There are a very small number of registers that the AMD public specs 
list as reserved, but the code has to use anyway.  AMD told us the 
required settings for those registers.  There are really very few such 
registers.  The vast majority of the registers are fully documented in 
AMD's public specs.

AMD employees and former employees, particularly Jordan Crouse and Tom 
Sylla, have helped us a lot.  Even with the extensive documentation, it 
requires a lot of time and experience to understand how everything fits 
together.  Without the extensive handholding that those two people gave 
me, I would not have been able to figure it all out.  Thanks, Jordan and 
Tom!