Dynamic Speed Switching Software for Power Management

The invention relates to speed switching software that responds to requests to slow down the system (typically to save energy) or to restore the system to full speed. More particularly, the invention relates to controlling the operating speed of a system which is capable of operating in a high speed mode and a low speed mode, and coordinating a transition from one speed to the other smoothly so the system can operate without need for resetting the system.

There are many reasons to conserve power in modern electronic devices. Many people are increasingly aware that power is a limited resource and it is advantageous to save power whenever possible. Power management has become an important issue for designers of many electronic devices. This is particularly true for computer systems, but is also useful for devices such as televisions, monitors, stereos and other audio/visual equipment, facsimile machines, telephone answering machines, and the like.

In many instances, such as a facsimile machine or telephone answering machine, very little system activity is required while waiting for a condition of interest, such as a ringing telephone. In a similar way, a computer which is connected to a network may require very little activity while monitoring network traffic, watching for a call to that specific computer. Once the condition of interest (a telephone call or a network call) is detected, the system generally needs additional resources to execute the procedure appropriate for the condition of interest, such as demodulating and processing a facsimile transmission, or providing some sort of file response, e.g. reading or writing a requested file.

In some systems, notably portable devices, the available power is limited and it is advantageous to save every possible milliwatt. This will preserve battery life as much as possible. In other systems, which may be powered by line current, various components of the system consume varying amounts of power. In the example of a computer, there may be a monitor, a hard drive, a CPU, and other ASICs, each of which consume power. To control heat build-up inside a computer housing, it is common to include a fan, which consumes still more power. Taken together, the system may consume more power than is desirable on a long term basis.

Many modern systems implement some form of energy saving procedure, such as turning down or turning off any component which can be safely modified without affecting system performance. One option would be to turn down or turn off any components which are not being used. For example, many computer monitors can respond to a variety of commands to, for example, first reduce the brightness of the display, then, ultimately, effectively turn off the display, yet holding the display in a state that can quickly be restored to normal operating conditions. Screen savers are quite common on modern personal computers. One form of screen saver dims the image on a monitor after some preset period of time or upon activation by a command. This reduces the amount of power consumed by the monitor. One useful program for dimming a monitor screen after a preset time is the Energy Saver control panel from Apple Computer (Cupertino, Calif.), available with Apple's System Software 7.5. Other components may be shut down as well. For example, the CPU Energy Saver control panel (from Apple Computer) can be used to turn off a computer after a preset time. This program is specifically designed to work with Apple computers which include soft power on/off, but one skilled in the art could design a similar system for other systems with appropriate hardware capabilities. Portable computers, such as Apple Computer's 520, 520c or 540c, often include software that will monitor hard disk activity and spin down the hard disk after a preset period of inactivity, thereby saving energy.

Another way of saving power is to run a system at slower clock speeds. A computer running at 25 MHz will consume less power than the same computer running at 33, 50, or 66 MHz. In general, however, users prefer to run their computers as fast as possible in order to run more powerful programs or to compute results more quickly.

One alternative might be to control the speed of the computer so as to run the computer at relatively low speed whenever possible and at higher, usually maximum, speed whenever needed. If the computer is a self-contained or independent system so that all system timings need be maintained constant only relative to various components within the system, this can work well. Even within a computer, several components have timing constraints which may differ. For example, dynamic RAM (DRAM) needs to be refreshed within certain time constraints. A processor (CPU) may be capable of operating at one or more speeds. System bus management may require servicing requests within certain time constraints. Software requests may require some timing constraints, particularly when dealing with time-sensitive events such as real-time sampling or most I/O which involves audio or video. A user will quickly notice if an audio CD does not play back smoothly or if a video recording or playback has timing glitches. Each of these different timing constraints need to be coordinated. The situation becomes even more complicated when the computer is connected to external devices, e.g. a network connection, that cannot or will not change speed synchronously with the computer.

External devices of particular interest include network services such as AppleTalk or EtherNet and communication services such as telephone and facsimile, which might be monitored using GeoPort. It would be advantageous if the computer of interest could monitor these services to detect any signal directed or addressed to the computer, then transition smoothly to a higher, usually full, speed state to service the signal and capture or respond to any attendant communication. One such example would be to monitor a telephone line while in a low power state. If an incoming call is detected, the system could transition to a normal speed state to activate call-responsive services such as answering the telephone, monitoring for facsimile or data modem signals or voice, provide an appropriate response, such as receiving, storing and printing a facsimile or capturing and logging a voice message. Once the communication has been handled, the computer system can remain in the high speed state for a pre-selected period of time, then return to the low speed state.

Another example of a communication is responding to a network services request. A computer may be connected to a network but be needed only for infrequent communication. One such example might be a user's computer which is turned on, but has remained idle for some time-out period, perhaps 15 minutes. The user may be away from her desk but desire to log onto her computer. By using another computer connected to the network, the second computer can send to the primary computer a request for servicing, namely a request for attention to be followed by or integrated with a log in request. The primary computer can remain in a low power state until a network packet addressed to that computer is detected. At that point, the computer can transition to a normal speed state to service the request more effectively.

The speed transition is generally difficult for traditional computers. For example, a portable computer such as a PowerBook might be connected to an EtherNet network. If the portable is capable of going to a slow, low-powered state ("sleep mode") but is not capable of dynamic speed switching, the transition between speeds can be problematic. A typical portable has a preferred system operating speed, e.g. 33 MHz. The relationship of the system operating speed to the speed of another service such as servicing EtherNet packets is generally at a fixed ratio. This ratio will have one value for the slow system operating speed, and a different ratio for the normal system operating speed.

Solutions used in the past provide some benefit but do not go as far as the present invention. Older portable computers, notably older Macintosh PowerBook products, used an "economode" in which the computer could be run at a different speed (frequency) after restarting the machine. The available speeds are generally "hard-coded" into such machines. Newer portable computers, notably newer Macintosh PowerBook products, utilize a technique called "processor cycling" to dynamically change the speed of the processor, thereby reducing power consumption. Some processor cycling implementations actually turn the processor on and off rapidly enough to effect a "slower" CPU speed. When processor cycling, the machine's processor will resume full power operation upon any external interrupt. However, certain types of interrupts may not warrant full power operation. For example, if a floppy disk needs to be ejected, there may be no need to resume full power operation of, for example, the monitor or in some cases even the hard disk.

Since the second service is operating in a single time domain, the computer should transition smoothly between its differing operating speeds. This includes identifying and coordinating any system clock edges. A typical modern computer system may have a primary system clock speed but may have additional internal clocks in various relationships with the system clock. For example, many computer systems include one or more industry standard buses, such as PCI, which operate at a defined frequency (33 MHz for PCI). The system bus connecting the computer's primary CPU to major resources such as memory and a bus bridge to the standard bus typically will operate at a different frequency. Certain processors such as the PowerPC.TM. 601 or 601+ from IBM/Motorola require various secondary clocks at multiples of the system clock. Heretofore, changing system operating speeds has required a system reset or other major system transition in order to avoid introducing glitches such as timing errors which can potentially crash a system or cause data error problems. The recent development of dynamic speed switching hardware provides a means for smoothly transitioning between system operating speeds while maintaining services.

The present invention provides a software program and method to coordinate with hardware which is capable of dynamic speed switching. Dynamic speed switching software is used to reset timing (bus and processor clock) sensitive elements such that machine speed (bus and processor clock frequencies) can be changed dynamically without interruption of I/O services or general OS and application level functions.

The invention includes storing timing constants for different system operating speeds. These timing parameters can be stored, then loaded into low level memory in coordination with certain service and speed changes to effect a smooth speed transition within a minimal number of clock cycles. In a preferred form, the invention includes examining the hardware configuration at system startup and determining timing parameters for the specific hardware installed in the system.

One object of this invention is to provide software to manage dynamic speed switching for an electronic device, preferably including a computer.

Yet another object of this invention is to provide software drivers to support dynamic speed switching hardware.

This and other objects and advantages of the invention, as well as the details of an illustrative embodiment, will be more fully understood from the following specification and drawings.

FIG. 1 illustrates startup code execution, which can be performed during machine boot.

FIG. 2 illustrates power manager idle time execution.

FIG. 3 illustrates the major logical functions of a hardware controller useful in practicing this invention.

FIG. 4 illustrates the relationship of several waveforms during normal operation of a computer under a variety of clocking schemes.

FIG. 5 illustrates the relationship of the waveforms of FIG. 4 during normal operation, at the point of transition to doze mode, and during doze mode.

One preferred implementation of the program and method of this invention is in a computer system which includes dynamic speed switching hardware capabilities. A preferred form of software is a driver named Power Manager. The main events of the Power Manager are illustrated in FIGS. 1 and 2. The related hardware capabilities could be implemented in a variety of ways by one skilled in the art. One preferred hardware implementation uses the "Sparky" chip in the Macintosh 7200.

Referring to FIG. 1, Power Manager initialization is executed during computer startup or booting. Part of this initialization involves checking the particular hardware features of the host computer. If the machine supports dynamic speed switching, the following sequence of events takes place:

1. The initial full speed timing constants for TimeDBRA, TimeSCC, TimeVIA and TimeSCSI are saved in Power Manager low memory globals.

2. The speed switching routine specific to the host computer executes, reducing the processor and computer system bus (e.g. ARBus in a Macintosh 7200) clock frequencies. This switching routine is illustrated in FIG. 2. The switching routine includes copying the full speed timing constants into memory where the constants can be maintained at least while the computer is turned on without resetting or rebooting.

3. The same timing constants are recalculated for this new speed.

4. The low speed (doze mode) constants are saved in Power Manager low memory.

5. The speed switching routine is called to switch back to full speed operation.

6. The full speed timing constants previously saved in Power Manager low memory globals are copied into general low memory.

7. The "boot" process continues.

TimeDBRA is a low memory constant that provides a means of consistent measure of time across machines of different speed. The value is computed based on the speed of the host machine and can be used by software as a timing value.

TimeSCSI is a low memory constant that indicates the delay in accessing SCSI controller registers, the SCSI FIFO queues, etc. It is used primarily in polled SCSI transactions.

TimeVIA is a low memory constant that indicates the delay in accessing VIA ASIC registers. It is used to indicate a known quantity of time.

TimeSCC is a low memory constant that indicates the delay in accessing SCC (serial) ASIC registers. It is used to indicate a known quantity of time.

If at any time during operation of the computer an event occurs to drive the machine into a low power state (e.g. auto sleep is invoked by the Power Manager, perhaps by passage of time without significant activity; the user selects a `sleep` command, e.g. from a Macintosh Finder "Special" menu; or a call is made to the private Power Manager dispatch trap), the following sequence of events take place. Referring to FIG. 2,

1. The machine-specific speed switching routine is called to drive the speed of the machine low for doze or idle operation.

2. The appropriate time constants (calculated and saved during startup) are copied from the Power Manager globals to general low memory.

3. Any queues of pending events with timing sensitivity are examined for possible speed-switch impacts. In particular, the AppleTalk transition queue is traversed, notifying each procedure in the queue that a speed change has occurred and the network timing dependencies should be recalculated using the new timing constants in low memory (essentially re-execute the calibration done at startup/opening of network services using traditional, well known AppleTalk protocols).

Most timing-sensitive elements of the operating system recalculate any timing sensitive information before each transaction. Therefore, by changing the timing constants in low memory, components such as SCC, SCSI and floppy drive recalibrate correctly upon the next transaction involving that device. In a Macintosh computer, AppleTalk is the primary, and perhaps only, component that should be explicitly told to recalibrate.

The events described above in reference to FIG. 2 are repeated when the device is driven to its full speed operational mode. The Power Manager operates a state change notification queue. Operating System level and Application level software can install routines in this queue and obtain a notification message when the speed of the device has changed dynamically. This message can be used to recalibrate any timing-specific code on which the notified software may depend.

The event loop for detecting a need for a speed switch can be managed in a variety of ways well known to one skilled in the art. One method for managing the event loop is to include a test instruction in an operating system main event loop that tests the condition "time to sleep or wake" on a periodic basis. Another potential monitor is interrupt driven, particularly when in the doze or idle state, is to provide a combinatorial event monitor that can monitor the status of one or more conditions and upon any of the conditions changing state to a trigger status, the combination of inputs will signal "time to sleep or wake" and initiate the event loop. In the preferred implementation, the Power Manager patches the SynchIdleTime call and runs it's `IdleMind` code every third call to this routine. The resulting effect is that the Power Manager is tied into the OS event loop.

Also in a preferred implementation, the Power Manager not only checks for "time to wake or sleep" but also for pending sleep/wake events from device controlling software (drivers). A driver can post an event to which the Power Manager will respond accordingly the next time through the IdleMind code (see FIG. 2).

During a speed change, certain system settings need to be adjusted. Referring again to FIG. 1, but noting that these steps are part of the Speed Switching Routine in FIG. 2, during a Call Switching Routine to a low speed,

1. If a level 2 (L2) cache is present, the cache is disabled and invalidated.

2. The speed is switched by writing the new frequency value to a register in the memory controller that controls the mode of the clock generation chip ("Sparky" in one preferred implementation--described in more detail below). This chip switches the computer system bus (ARBus in the Macintosh 7200) and processor frequencies without producing a glitch that would `crash` the computer system.

3. The DRAM refresh interval is reprogrammed to preserve the desired interval at the low clock frequency. In a preferred implementation, the VRAM and ROM timings are also changed. The VRAM refresh is disabled, because the assumption is that the monitor and frame buffer controller are powered down.

Referring again to FIG. 1 and noting FIG. 2, during a Call Switching Routine to a high speed,

1. The full speed mode value is written to the clock generator control register in the memory controller ASIC (e.g. "Sparky").

2. The DRAM refresh interval, together with any changes to VRAM and ROM timings, are restored to the correct value for full speed operation.

3. If a level 2 cache is present, it is enabled.

This speed switching routine is completely dynamic in that it will work with essentially any full and low speed timings as well as with any cache size. The principles could be used to switch to an intermediate speed, if desired for some reason, using the principles described above.

The dynamic speed switching (DSS) code conforms to the Power Manager conventions in that it can be invoked from any computer's primitive table and does not rely on a specific implementation of hardware. Any machine that has the hardware capability to switch speeds dynamically simply sets the appropriate feature bit and includes a hardware specific switching routine. The DSS code can interface as needed with the hardware specific switching routine to initiate any needed operational modes and to store and provide needed information about timings.

Although a variety of hardware can be used with this dynamic speed switching, one preferred embodiment of hardware useful as an adjunct in practicing the present invention is illustrated in FIG. 3.

When the doze mode is entered the 2.times.PClk and /PClkEn clocks are reduced by a factor of 4, the /BClkEn matches the /PClkEn. Doze mode is a glitchless change in the clock waveforms. The doze signal should be sampled at the VCO frequency to synchronize the transition to and from doze mode for 2.times..sub.-- PClk, /PClkEn, /BClkEn, BCLK.times., /BCLK.

Sparky output frequencies shows an example of a VCO constrained to run between 300 MHz and 1 GHz. The implementation is free to choose other VCO frequencies so long as the select lines cause the appropriate outputs and the 2.times.PClk is divided from the VCO. Alternate embodiments may use a lower speed VCO. A limited frequency version allows the maximum 2.times.PCLKEN frequency to be 120 MHz. Otherwise all other specifications need to be met.

The hardware device shown in FIG. 3 features: PowerPC.TM. 601 clocks; fully integrated PLL; single crystal value for all output frequencies (33.3333 MHz); output frequency up to 300 MHz; .+-.150 ps maximum PPC601 cycle-to-cycle jitter; 350 ps maximum PPC601 output-to output skew; doze mode for reduced power (Energy Star); delay to PPC601 clocks vs. BClks; BClk.div.2,3,4 of PClk; PClk 66.66, 80, 100, 120, 133.33, 150 MHz; PCI clock (33.3333 MHz); 3.3V Vcc; and a 32 pin PQFP package.

In this preferred embodiment, a functional definition of output frequencies includes:

C33MXtal=33.3333 MHz (100 ppm)

VCO=C33MXtal*M/N

2.times.PClk=VCO.div.((P*2)*(If(Doze)Then 4 Else 1))

/PClkEn=2.times.PClk.div.2

/BClkEn=/PClkEn.div.(If(Doze)Then 1 Else B)

BClk=/BClkEn

PClClk=C33M

A general description of the device and method of using the present invention as well as a preferred embodiment of the present invention has been set forth above. One skilled in the art will recognize and be able to practice many changes in many aspects of the device and method described above, including variations which fall within the teachings of this invention. The spirit and scope of the invention should be limited only as set forth in the claims which follow.