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A newer version of this document is available.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.
PCI Express* Power Management
Software initiates the transition to S3/S4/S5 by performing an I/O write to the Power Management Control register in the SoC. After the I/O write completion has been returned to the processor, the Power Management Controller will signal each root port to send a PME_Turn_Off message on the downstream link. The device attached to the link will eventually respond with a PME_TO_Ack followed by sending a PM_Enter_L23 DLLP (Data Link Layer Packet) request to enter L23. The Express ports and Power Management Controller take no action upon receiving a PME_TO_Ack. When all the Express port links are in state L23, the Power Management Controller will proceed with the entry into S3/S4/S5.
Prior to entering S3, software is required to put each device into D3HOT. When a device is put into D3HOT, it will initiate entry into a L1 link state by sending a PM_Enter_L1 DLLP. Under normal operating conditions when the root ports sends the PME_Turn_Off message, the link will be in state L1. However, when the root port is instructed to send the PME_Turn_Off message, it will send it whether or not the link was in L1. Endpoints attached to the PCH can make no assumptions about the state of the link prior to receiving a PME_Turn_Off message.
Device Initiated PM_PME Message
When the system has returned to a working state from a previous low power state, a device requesting service will send a PM_PME message continuously, until acknowledged by the root port. The root port will take different actions depending upon whether this is the first PM_PME that has been received, or whether a previous message has been received but not yet serviced by the operating system.
If this is the first message received (RSTS.PS), the root port will set RSTS.PS, and log the PME Requester ID into RSTS.RID. If an interrupt is enabled using RCTL.PIE, an interrupt will be generated. This interrupt can be either a pin or an MSI if MSI is enabled using MC.MSIE.
If this is a subsequent message received (RSTS.PS is already set), the root port will set RSTS.PP. No other action will be taken.
When the first PME event is cleared by software clearing RSTS.PS, the root port will set RSTS.PS, clear RSTS.PP, and move the requester ID into RSTS.RID.
If RCTL.PIE is set, an interrupt will be generated. If RCTL.PIE is not set, a message will be sent to the power management controller so that a GPE can be set. If messages have been logged (RSTS.PS is set), and RCTL.PIE is later written from a 0b to a 1b, an interrupt will be generated. This last condition handles the case where the message was received prior to the operating system re-enabling interrupts after resuming from a low power state.
Interrupts for power management events are not supported on legacy operating systems. To support power management on non-PCI Express aware operating systems, PM events can be routed to generate SCI. To generate SCI, MPC.PMCE must be set. When set, a power management event will cause SMSCS.PMCS to be set.
Additionally, BIOS workarounds for power management can be supported by setting MPC.PMME. When this bit is set, power management events will set SMSCS.PMMS, and SMI# will be generated. This bit will be set regardless of whether interrupts or SCI is enabled. The SMI# may occur concurrently with an interrupt or SCI.
Latency Tolerance Reporting (LTR)
The root port supports the extended Latency Tolerance Reporting (LTR) capability. LTR provides a means for device endpoints to dynamically report their service latency requirements for memory access to the root port. Endpoint devices should transmit a new LTR message to the root port each time its latency tolerance changes (and initially during boot). The PCH uses the information to make better power management decisions. The processor uses the worst case tolerance value communicated by the PCH to optimize C-state transitions. This results in better platform power management without impacting endpoint functionality.