Intel® Pentium® Silver and Intel® Celeron® Processors Datasheet, Volume 1
A newer version of this document is available. Customers should click here to go to the newest version.
Device Power Down Support
In order to power down peripherals connected to the PCH GSPI bus, the idle configured state of the I/O signals must be retained to avoid transitions on the bus that can affect the connected powered peripheral. Connected devices are allowed to remain in the D0 active or D2 low power states when the bus is powered off (power gated). The PCH HW will prevent any transitions on the serial bus signals during a power gate event.
Latency Tolerance Reporting (LTR)
Latency Tolerance Reporting is used to allow the system to optimize internal power states based on dynamic data, comprehending the current platform activity and service latency requirements. However, the GSPI bus architecture does not provide the architectural means to define dynamic latency tolerance messaging. Therefore, the interface supports this by reporting its service latency requirements to the platform power management controller via LTR registers.
The controller’s latency tolerance reporting can be managed by one of the two following schemes. The platform integrator must choose the correct scheme for managing latency tolerance reporting based on the platform, OS and usage.
- Platform/HW Default Control: This scheme is used for usage models in which the controller’s state correctly informs the platform of the current latency requirements. In this scheme, the latency requirement is a function of the controller state. The latency for transmitting data to/from its connected device at a given rate while the controller is active is representative of the active latency requirements. On the other hand if the device is not transmitting or receiving data and idle, there is no expectation for end to end latency.
- Driver Control: This scheme is used for usage models in which the controller state does not inform the platform correctly of the current latency requirements. If the FIFOs of the connected device are much smaller than the controller FIFOs, or the connected device’s end-to-end traffic assumptions are much smaller than the latency to restore the platform from low power state, driver control should be used.