Power Management/lang-es

Administración de potencia

Introducción y material relacionado

Careful stewardship of battery power is critical.

Esta pagina es un trbajao en proceso que recopila información relacionada a la administración de potencia de OLPC.

• Hardware Power Domains
• Power Management Software
• XO Power Management Tips and Tricks
• Misc. Papers on Energy Efficiency in Computing and Networking

Ademas de manejar efectivamente la potencia que hay en la batería también la pagina Battery and power discute muchas fuentes alternas de potencia que pueden ser usadas para complementar el actual cargador.

Enfoque de Linux al control de potencia

Linux is highly cross platform, running on just about all significant architectures, including many that are used for battery powered embedded systems. The infrastructure for power management has therefore become quite sophisticated over the last years, though it is still maturing. This means that the facilities are general, and not tied to any particular architecture. The generation one OLPC system, being from the x86 part of the world, is therefore similar and fundamentally different from other x86 based systems, for reasons that will become clear in the discussion below.

Linux is not dependent on ACPI or the older APM power management systems, which are x86 specific. As such, Linux's design has always done its power control in the operating system, and ACPI and the like are considered "platform dependent".

Innovaciones de OLPC

The DCON chip lets us take over refresh of our very low power flat panel and therefore completely power off the processor board. Given our flat panel is usable in gray scale mode at .1 watt, you can see that the leakage currents and power supply power consumption of the system board can dominate power consumption easily.

We are also able to leave the Marvell wireless module to operate independently, forwarding packets in the mesh while possibly everything else is powered down.

Firmware (BIOS en PC's convencionales)

Uso del Open Firmware

Rather than using a conventional BIOS and boot loader, as of the "C" series of our firmware we are using Open Firmware. We no longer use LinuxBIOS for the setup of our systems. This is a result of having implemented fast past resume from RAM; once you have implemented this, setup of the system initially is almost identical.

We have also removed VSA and VESA support (the Geode Virtual Systems Support) from our firmware. Since our PCI bus is fixed, we have no need for PCI configuration registers. Similarly, using the fbdev driver on Linux, we have no need for VESA emulation; while the PCI bus emulation was free software (AMD had generously made it available), the VESA emulation was the one part of VSA that they did not own, and so we did not have source for it. We did not want an unmaintainable binary blob in our firmware that we did not need anyway, and saves space in the flash for other purposes.

Fast Resume

Resume on our system is extremely fast: even without any serious attempt to optimize resume, we can resume from RAM in 160 milliseconds (mid-April). We believe the hardware limitations for resume are about 63 milliseconds on the B2 and before systems; B3 and later are probably similar. We will work in the future to further speed resume. Note that for most uses, 100ms is considered at the edge of human perception (e.g. typing).

Identificación del sistema

The Manufacturing Data page documents the Model ID string, part number, localization information, factory, BIOS version, and many other pieces of data.

Quiet Boot

Boot is not "quiet" at this time. Linux has facilities to make a splash screen at boot overlay the boot messages, but OLPC has not implemented this as yet. Bug #1394 tracks this issue for eventual resolution.

POST Message

At this moment (prior to Q2C10, at a minimum), if any key is pressed during the initial countdown after the initial message, the boot sequence is stopped and OFW's command interpreter invoked. OFW gives a cheery "ok" message to indicate it is ready for a command.

Open Firmware Command Prompt

The OFW FAQ answers some of the most common questions of how to interact with the OLPC OFW firmware.

Power Management Support

As discussed above, Linux does not depend on ACPI. To meet our fast resume goals and transparency into the firmware, we do not use ACPI, which would significantly slow our resume from suspend while adding no benefit. In this regard, we differ significantly from other x86 systems. This is the normal case for Linux on other architectures, so should not be regarded as unusual for Linux overall.

Botón de Prendido/Apagado

The power button on OLPC serves as a power button.

Momentary Button Push

The system will suspend to RAM after the button is pressed momentarily. Wireless will be left operational when suspended this way. (Prior to deployment of suspend/resume, this button currently performs a clean Linux shutdown). See bug #1396 for more information.

Four Second Button Push

Pressing the power button for four seconds does a hard reset of the system and all state is lost.

Estados de Administración de Potencia

The following are the major operating states of the system. For simplicity's sake in using commonly understood terminology, see Wikipedia's ACPI article.

Powered Down

In this state, (G3 is this state in ACPI). Everything is off and the battery can be swapped. The operating system will have to be booted to start operation; the RAM is not preserved. On our hardware, if power is available, the EC will be powered up and potentially charging the battery.

Suspended, with Mesh Active, No screen

A common mode of use will be the system not being used, but still active forwarding in the mesh network on behalf of others in the mesh, but otherwise unused. This differs from our powered down state by the fact the Marvell wireless will be powered up, and active. In ACPI terminology, the closest match is G1/S3. The processor is suspended to RAM (in self-refresh).

Suspended, with Mesh Active, Screen Enabled

Another common mode of use is sometimes called "ebook mode". Both the screen and the Marvell wireless are left operational: the screen by use of the DCON chip. This differs from our powered down state by the fact the Marvell wireless will be powered up, and active along with the display. In ACPI terminology, the closest match is G1/S3. The processor is suspended to RAM (in self-refresh). Note that the DCON has facilities to implement a "screen saver" where it can disable itself and the backlight after a preset time without requiring the system to be resumed from RAM.

System fully operational

In this state, the system is available for normal use. The ACPI processor states that this corresponds to are C0 and C1 (note that C1 is not useful on a GX, but does save power on the LX). Linux is working very hard to remove "ticks"; as of this writing, the kernel is now "tickless" and this is operational on OLPC, meaning that it no longer uses a periodic timer clock interrupt to drive the scheduling of processes (which had caused 250 interrupts and transitions from C1 to C0 per second). The OLPC has been observed at 40 per second. Work is underway in user space to abolish polling of hardware that might force wakeups, and private communications are that a full Gnome environment has been seen as low as only a few wakeups/second.

Note that in this state, Linux may have many parts of the system powered down: e.g. audio, GPU, etc. as described in detail below.

Special Function Keys

System Management BIOS Interface

Not Supported