What acpi power state describes when the cpu is off, but the ram is refreshed?

4.8 Platform-level power management

ACPI

The Advanced Configuration and Power Interface (ACPI) [Int96; ACP13][Int96][ACP13] is an open industry standard for power management services. Initially targeted to PCs, it is designed to be compatible with a wide variety of operating systems. The role of ACPI in the system is illustrated in Fig. 4.33. ACPI provides some basic power management facilities and abstracts the hardware layer, the operating system has its own power management module that determines the policy, and the operating system then uses ACPI to send the required controls to the hardware and to observe the hardware's state as input to the power manager.

ACPI supports several basic global power states:

G3, the mechanical off state, in which the system consumes no power.

G2, the soft off state, which requires a full operating system reboot to restore the machine to working condition.

G1, the sleeping state, in which the system appears to be off and the time required to return to working condition is inversely proportional to power consumption.

G0, the working state, in which the system is fully usable.

S4 is a nonvolatile sleep in which the system state is written to nonvolatile memory for later restoration.

The legacy state, in which the system does not comply with ACPI.

The power manager typically includes an observer, which receives messages through the ACPI interface that describe the system behavior. It also includes a decision module that determines power management actions based on those observations.

Processor Power States (Cx States)

ACPI allows processors to sleep while in the G0 working state. The processor power states, the C states, only apply when the global state is G0. The C states are categorized as either active (C0) or sleeping (C1, C2, …). In the ACPI specification, only C0 and C1 are required; the other states are optional. In the sleeping states, no instructions are executed and the processor is expected to consume less power. Of course, the deeper the sleep state, the longer the latency required to awaken from that state.

Transitions from C0 to C1 or other C states are initiated by the operating system during idle periods. The operating system periodically checks an ACPI counter to observe what fraction of its time has recently been spent in the idle loop.

C0. In C0, the processor executes instructions and operates normally.

C1. Of the C sleep states, C1 has the lowest transition latency. In fact, it must be low enough as to be negligible and therefore not an input to the operating system’s decision to transition to this state. Transition to C1 is not apparent to application software and does not otherwise alter system operation. This state is supported through a native instruction of the processor (HLT or MWAIT for IA32 processors) and assumes no hardware support is needed from the chipset.

C2. C2 is a lower-power sleep state, as compared to C1, and its worst-case transition latency can be found in the ACPI system firmware. The operating system does consider the transition latency when determining the benefits of transitioning to this state rather than another. Transition to C2 is not apparent to software and does not otherwise alter system operation.

C3. The C3 state offers great power reduction at the cost of an increased transition latency, which, like C2, is part of the explicit evaluation made by the operating system when making sleep state transition decisions. Additionally, in C3, processor caches maintain their state but do not emit cache coherence traffic; operating system software must ensure cache coherence when a processor resides in C3 by, for example, flushing and invalidating caches prior to state entry.

C4…Cn. ACPI, Revision 2.0 introduced optional additional power states. The specific entry and exit semantics are the responsibility of the vendor, but the principles that define the relationships between the first four states—that higher state numbers imply higher transition latency and greater reductions in power consumption—apply among these states as well.

List of Abbreviations

ACPI

Advanced Configuration and Power Interface

Active Energy Manager

Application Programming Interface

Application-Specific Integrated Circuit

Address Translation Unit

Central Procesing Unit

Data Center Infrastructure Efficiency

Dynamic Concurrency Throttling

Dynamic Exascale-Entry Platform

Direct Memory Access

Dynamic Voltage and Frequency Scaling

Dynamic Voltage Scaling

Energy-Delay-Product

Energy-Efficient High Performance Computing Working Group

First In, First Out

Floating-Point OPeration

Field-Programmable Gate Array

General Purpose Graphics Processing Unit

Graphics Processing Unit

Hard Disk Drive

Hybrid Memory Cube

High Performance Computing

High Performance Linpack

Input/Output Operations Per Second

Joule

Juelich Supercomputing Centre

Lawrence Livermore National Lab

Many Integrated Cores

Million Instructions Per Second

Message passing Interface

National Center for Supercomputing Applications

Network-on-Chip

Near-Threshold Computing

Program Activity Graph

Performance Application Programming Interface

Package Control Unit

Phase-Change Memory

Power Distribution Unit

Partitioned Global Address Space

Power Processing Element

Power Usage Effectiveness

Quantum ChromoDynamics

Random Access Memory

Remote Direct Memory Access

Remote Memory Access

Synergistic Processing Element

Standard Performance Evaluation Corporation

Simultaneous MultiThreading

System-on-a-Chip

Solid State Drive

Total Cost of Ownership

Translation Lookaside Buffer

Thermal and Power Management Device

Virtualized Engine for Low Overhead

Watt

Processor states

ACPI defines the power state of system processors while in the G0 working state as being either active (executing) or sleeping (not executing). Processor power states are designated C0, C1, C2, C3, … Cn. The C0 power state is an active power state where the CPU executes instructions. The C1 through Cn power states are processor sleeping states where the processor consumes less power and dissipates less heat than when the processor in the C0 state. While in a sleeping state, the processor does not execute any instructions. Each processor sleeping state has a latency associated with entering and exiting that corresponds to the power savings. In general, the longer the entry/exit latency, the greater the power savings is for the state. To conserve power, OSPM places the processor into one of its supported sleeping states when idle. While in the C0 state, ACPI allows the performance of the processor to be altered through a defined “throttling” process and through transitions into multiple performance states (P-states). A diagram of processor power states is provided in Figure 2.8.

Now is the right time to ask that question: How do the low power interfaces reduce or optimize power consumption? The answer is simple: As we discussed earlier when introducing power consumption and strategies for power savings, the low power interfaces use the same fundamental mechanism in a way suitable for them to reduce power consumption; for example, idle detection and suspension or power gating/clock gating. In the forthcoming chapters we will discuss how these generic strategies are implemented in specific ways for specific interfaces/controllers/subsystems, based on the suitability. However, before we get there, we will discuss the functional aspect of various subsystems of a system in the very next chapter. That will be followed by the implementation details of each of the subsystem.

1) Industrial Design Standards

Industrial standards have been proposed to facilitate the development of operating system-based power management. Intel, Microsoft and Toshiba proposed the open standard called advanced configuration and power interface (ACPI) [21]. ACPI provides an OS-independent power management and configuration standard. It provides for an orderly transition from legacy hardware to ACPI-compliant hardware. Although this initiative targets personal computers (PC's), it contains useful guidelines for a more general class of systems. The main goals of ACPI are to: 1) enable all PC's to implement motherboard dynamic configuration and power management; 2) enhance power management features and the robustness of power-managed systems; and 3) accelerate implementation of power-managed computers, reduce costs and time to market.

The ACPI specification defines most interfaces between OS software and hardware. The software and hardware components relevant to ACPI are shown in Fig. 11. Applications interact with the OS kernel through application programming interfaces (API's). A module of the OS implements the power management policies. The power management module interacts with the hardware through kernel services (system calls). The kernel interacts with the hardware using device drivers. The front-end of the ACPI interface is the ACPI driver. The driver is OS-specific, it maps kernel requests to ACPI commands, and ACPI responses/messages to kernel signals/interrupts. Notice that the kernel may also interact with non-ACPI-compliant hardware through other device drivers.

At the bottom of Fig. 11 the hardware platform is shown. Although it is represented as a monolithic block, it is useful to distinguish three types of hardware components. First, hardware resources (or devices) are the system components that provide some kind of specialized functionality (e.g., video controllers, modems, bus controllers). Second, the CPU can be seen as a specialized resource that need to be active for the OS (and the ACPI interface layer) to run. Finally, the chipset (also called core logic) is the motherboard logic that controls the most basic hardware functionalities (such as real-time clocks, interrupt signals, processor busses) and interfaces the CPU with all other devices. Although the CPU runs the OS, no system activity could be performed without the chipset. From the power management standpoint, the chipset, or a critical part of it, should always be active because the system relies on it to exit from sleep states.

It is important to notice that ACPI specifies neither how to implement hardware devices nor how to realize power management in the operating system. No constraints are imposed on implementation styles for hardware and on power management policies. Implementation of ACPI-compliant hardware can leverage any technology or architectural optimization as long as the power-managed device is controllable by the standard interface specified by ACPI.

In ACPI, the system has five global power states. Namely, the following.

Mechanical off state G3, with no power consumption.

Soft off state G2 (also called S5). A full OS reboot is needed to restore the working state.

Sleeping state G1. The system appears to be off and power consumption is reduced. The system returns to the working state in an amount of time which grows with the inverse of the power consumption.

Working state G0, where the system is On and fully usable.

Legacy state, which is entered when the system does not comply with ACPI.

The global states are shown in Fig. 12(a). They are ordered from top to bottom by increasing power dissipation.

The ACPI specification refines the classification of global system states by defining four sleeping states within state G1, as shown in Fig. 12(b).

S1 is a sleeping state with low wake-up latency. No system context is lost in the CPU or the chipset.

S2 is a low wake-up latency sleeping state. This state is similar to the S1 sleeping state with the exception that the CPU and system cache context is lost.

S3 is another low wake-up latency sleeping state where all system context is lost except system memory.

S4 is the sleeping state with the lowest power and longest wake-up latency. To reduce power to a minimum, all devices are powered off.

Additionally, the ACPI specification defines states for system components. There are two types of system components, devices and processor, for which power states are specified. Devices are abstract representations of the hardware resources in the system. Four states are defined for devices, as shown in Fig. 12(c). In contrast with global power states, device power states are not visible to the user. For instance, some devices can be in an inactive state, but the system appears to be in a working state. Furthermore, state transitions for different devices can be controlled by different power management schemes. The processor is the central processing unit that controls the entire PC platform. The processor has its own power states, as shown in Fig. 12(d). Notice the intrinsic asymmetry of the ACPI model. The central role of the CPU is recognized, and the processor is not treated as a simple resource.

4.1.2 Advanced Configuration and Power Interface

The Advanced Configuration and Power Interface (ACPI) [3] is a specification of different power states of computer systems. It spans individual components as well as whole computer systems. Figure 6 shows the different ACPI states.

G-, S-, and D-states. Following a top–down analysis of the ACPI tree, the first states to examine are Gx. These states correspond to the mechanical states of an entire computer, and range from G0—Running to G3—Mechanical off. The states G1 and G2 describe two intermediate states in which power is supplied to the computer, however only in state G0 it is possible to actively provide computing power on the system. While state S0 is equal to G0 and S5 equal to G2, states S1 to S4 provide a more fine granular power management inside G1. In S1, while being in sleeping mode, the system can easily be woken up by a simple interrupt command. S2 additionally deactivates the CPU clock and caches. S3 and S4 work similarly: The current system state is saved and reloaded when starting up again. However, in S3, this state is saved to RAM (suspend to RAM), therefore the RAM modules of the system have to continuously be supplied with power not to lose information during sleep. In S4, the state is saved to disk (suspend to disk), which enables a complete shutdown of the system, but at the price of a significantly higher resume time (as RAM contents have to be restored from disk). The states D0 to D3 define different sleep states of devices like modems and network interface cards [3].

C-states. The CPU is the biggest consumer of dynamic power (that is, power consumed only while the system is being utilized) inside a computer [4]. Therefore, it has been a major target for optimization regarding energy efficiency. Early CPUs were based on—from today’s perspective—coarse transistors and ran on high operating voltages. Apart from that, a major drawback of these processors was the inability to scale to different workload intensities. There was little difference between phases of complete idleness and high load as all CPU parts were active. Especially the constraints in energy consumption imposed by the advent of mobile computing increased the need to save energy. This led to the invention of first energy saving techniques, called C-states. C-states are sleep states which are ordered according to the extent of energy savings. These sleep states are only assumed in case of processor idleness, and not in partly utilized states. Current C-states range from C0 to C3, according to the ACPI specification, however it is mentioned in [3] that further C-states exist that enable even deeper sleep states. Especially with the growing popularity of multi-core CPUs, the introduction of deeper sleep states of individual CPU cores is reasonable, as often one single CPU core is able to supply the needed computing power while the remaining cores can remain in sleep mode. According to the ACPI specification, all CPUs must at least support the states C0 and C1.

P-states. Another, more recent power saving technique are so-called performance states (P-states) or more generally dynamic voltage and frequency scaling (DVFS), which serve the adaptation to partly utilized CPU conditions. P-states are implemented as a table of voltage/frequency pairs which are advertised by the CPU itself. In [5] an example of a P-state table is given, which can be seen in Table I.

Table I. P-states of the Intel Pentium M Processor.

P-stateP0P1P2P3P4P5Frequency (MHz)1600140012001000800600Voltage (V)1.4841.4201.2761.1641.0360.956

These voltage/frequency pairs represent possible operating conditions of the CPU circuits. A possible operating condition in this case means a combination of voltage and frequency which enables a stable operation of the processor. Stability can only be guaranteed if both parameters are switched together. Again we explain the reasons for this without going into too much detail. The higher the frequency the processor runs at, the faster the transistors have to switch states. Switching a transistor state in turn requires a change of its gate charge. To speed up this charge change, the circuit voltage has to be raised. While higher voltage is required to speed up the charge change, in case of lower frequency, the voltage can be decreased. P-states are ordered by the degree of performance reduction, from P0 (maximum frequency) to Pn (minimum frequency). The exact number of states is not fixed and left to the manufacturer, however there is an upper limit of 16 total P-states. Both major vendors of desktop and server CPUs—AMD and Intel—are currently using this energy saving technique using different brand specific terminology: “Cool’n’Quiet” [6] and “Enhanced Intel SpeedStep Technology” [5], respectively.

Benefits and value

There are many benefits of connected standby over the traditional ACPI Sleep (S3) and Hibernate (S4) states. Following are some of the prominent benefits:

Instant resume from sleep. A connected standby device resumes extremely quickly. The performance of a resume from connected standby is almost always faster than resuming from the traditional Sleep (S3) state and significantly faster than resuming from the Hibernate (S4) or Shutdown (S5) state.

Keeps the Wi-Fi device turned on in a very low-power mode. The Wi-Fi device automatically searches for known access points and will connect to them according to the user’s preference. This feature allows the system to maintain connectivity seamlessly between various locations like home, work, bus, and coffee shop.

The Wi-Fi device is already connected to the network when the user turns on the system. Hence the user does not have to wait to connect to a Wi-Fi access point and then wait for e-mail to sync. Since Wi-Fi is already connected, the e-mail is already synced.

With a constant Wi-Fi connection, a mobile device with connected standby also maintains constant connectivity with the cloud. Communications applications (such as Skype and Lync) notify the user in real time of an incoming request or call while the system is in connected standby. Applications can also deliver push notifications to alert the user to news events, weather alerts, or instant messages.

A connected standby device automatically roams between all available network types and can use the available networking option (mobile broadband (MBB; cellular) connection and wired LAN/Ethernet) that is the cheapest and uses the least power.

Connected standby is the foundation of the modern mobile experience. Users expect all of their electronics devices to instantly turn on, have long battery life, and always be connected to the cloud. All smartphones and the overwhelming majority of tablets support a sleep mode that is always on and always connected.

Cold and Warm Boot

Typically, when referring to cold boot we refer to the sequence of taking the system from the S5 sleep state in ACPI (or soft-off), where power has been supplied into S0 or the fully awake/active state where the system is available to the application.

While a “cold boot and constant run” mode does apply to most embedded applications, it is not an option for Intel platforms. There are two ACPI-defined sleep states, namely, S3 sleep (suspend to RAM) and S4 hibernate (suspend to disk), where the boot times and boot sequences can be dramatically reduced.

In the S3 state, the system is consuming refresh power in the DIMMs and enough power to react to a wake event in the I/O hardware. The boot sequence for Intel platforms from S3 is the basic initialization outlined above and a short boot script to restore the register contents in the I/O that may have been lost. The advanced initialization and device selection process is skipped. The OS drivers need to be restarted (not loaded and started). This results in a several orders of magnitude speedup when compare to a cold start. The applications do not need to be restarted, but are all active within a few hundred milliseconds of the wake event. As always, the system requirements will dictate how power and response time can be optimized.

In S4, the system is consuming near zero power if the wake events are limited to something akin to the S5 boot. The kernel does not have to be fully loaded. The OS drivers do not have to be loaded/started, as is required in the cold boot case. The applications are already running when in the state prior to hibernate. From a boot perspective, it is normally thought that that S4 boot path is equal to the S5 boot path, but in reality most of the advanced initialization steps can be skipped completely.

Parallels Desktop for Mac

In 2006, Parallels was the first software product to bring virtualization mainstream for the Mac OS X platform, as shown in Figure 3.7. It is similar in form and function to the Windows and Linux versions.

Parallels Desktop for Mac is capable of virtualizing a full set of standard PC hardware, such as

A virtualized CPU of the same type as the host's physical processor

Advanced Configuration and Power Interface (ACPI) compliance system

A generic motherboard compatible with the Intel chipset

Up to 8 GB of RAM for guest VMs

Up to 256 MB of video RAM

VGA and SVGA video adapter

A 1.44 MB floppy drive

Virtual CD/DVD-read-only memory ROM drives can be mapped to either physical drives or ISO image files

DVD/CD-ROM “pass-through” access

Serial ports

Parallel ports

Ethernet network cards

USB 2.0 devices and USB 1.1 devices

Sound card

Standard keyboard and wheel mouse

Another feature of Parallels for Mac is called Coherence, which removes the Windows desktop and virtualization frames to create a more seamless desktop environment between Windows and OS X applications. Additionally, Parallels for Mac can boot existing “Boot Camp” Windows XP partitions, which is a utility within OS X that allows users to install Windows XP or Vista on Intel-based Mac computers.

As new versions of Parallels have been released (version 5 is currently the most recent edition), and there have been additional features added that allow for greater use of the Mac platform. Some of those features are as follows:

The capability to start and stop a VM via the iPhone

MacLook, which themes Windows applications and makes them look like a Mac

The capability to use Multi-Touch gestures with the recently released Magic Mouse in Windows applications, as well as using the standard Apple Remote to control Windows applications

Which ACPI state turns the CPU and RAM off and copies the contents of RAM?

What ACPI power state describes when the power is off?

Which ACPI power state provides power to the CPU and RAM but powers down unused devices select one S0 S1 S2 S3 S4?