Specifications

AMD Geode™ GeodeROM Functional Specification 239
17
Power Management 32087C
17.0Power Management
There are multiple industry-standard power management (PM) methods to be supported, each sharing common PM con-
cepts such as device timeouts, power states, and power state transition diagrams. Power management methods typically
differ primarily in the API and in some of the details, such as the number of power states and the types of power state tran-
sitions supported.
The GeodeROM power management is modularized relative to SMM in a carefully defined manner. For example, the API
for Advanced Power Management (APM) is implemented in an APM-specific VSM. The setting of a particular power state
(e.g. Standby), however, is common among several PM methods. Rather than duplicate power state logic in each VSM,
such common functionality is implemented in a generic PM VSM. Other PM VSMs leverage this common code by commu-
nicating its requirements to the generic PM VSM via virtual registers. This distributed architecture results in smaller overall
code size when multiple PM schemes are supported. In addition, when a change (e.g., a bug fix or feature enhancement) is
made to the generic PM VSM, all of the PM VSMs benefit from such changes, usually without recompilation of the API-spe-
cific VSMs.
The generic PM VSM handles the low-level details of power management. It controls the CPU power state directly. It also
coordinates power management among other VSMs. Specifically, when the power state changes, it broadcasts a
MSG_POWER_STATE message to all other VSMs. The new power state is passed as a parameter. This gives other VSMs
a chance to put their associated hardware, if any, into the new power state. VSMs that do not control actual hardware
would typically ignore these messages. However, VSMs such as the XpressGRAPHICS™ subsystem, XpressAUDIO™
subsystem, a SuperI/O or battery VSM would take appropriate action. As stated earlier, this distributed design provides the
ability for custom VSMs to be written to handle platform-specific power management issues without the need to modify the
basic power management software.
17.1 Power States
Different power management methods define the various power states differently. However, they typically have enough in
common that they can be mapped onto one of the following:
• FULL ON - The processor is running at full speed with all peripherals at a ready state.
• DOZE - Peripherals are at a ready state. The processor is in a clock-throttled mode. Any user input (keyboard or mouse
activity) returns the system to FULL ON in a negligible amount of time. SMIs and IRQs also return the system to FULL
ON for a short duration, then the system is clock-throttled again. This ensures quick interrupt and video servicing. Tran-
sitions in and out of this state usually cannot be noticed by the user.
• STANDBY - Peripherals may be put in a lower power state (minimal recovery time). The processor is in its lowest power
state. The state of DRAM is preserved, the video is blanked and the hard drive is spun down. Any SMI or IRQ defined as
a wake-up event returns the system to FULL ON.
• SUSPEND - This state preserves the entire state of the system, either in a Suspend-to-RAM or Suspend-to-Disk mode.
The processor is off, but the power to the DRAM controller remains on in the case of Suspend-to-RAM.
• OFF - Power to the system is turned off. The system is typically set to this state after a Suspend-to-Disk or after the
operating system has made the system ready for shutdown.