Hi all,
I would like to know whether Linux on PowerPC puts the processor on
low power mode during idle state. Most PowerPC processors support a
low power mode. I am looking to add support for low-power mode in
Linux for AMCC 4xx processors.
My questions are the following:
1) Is tickless kernel integrated into the PowerPC tree?
2) Does Linux kernel puts the processor in low power mode if cpu is
idle, and takes it out of that state on wakeup (triggered by
interrupt/WOL etc)?
Any suggestions/comments would be welcome.
Thanks,
Prodyut
Hi all,
I would like to know whether Linux on PowerPC puts the processor on
low power mode during idle state. Most PowerPC processors support a
low power mode. I am looking to add support for low-power mode in
Linux for AMCC 4xx processors.
My questions are the following:
1) Is tickless kernel integrated into the PowerPC tree?
2) Does Linux kernel puts the processor in low power mode if cpu is
idle, and takes it out of that state on wakeup (triggered by
interrupt/WOL etc)?
Any suggestions/comments would be welcome.
The 4xx cores themselves are designed to have power savings without
explicit software management. The only sort of "NAP/DOZE" possible
from software on them is to use the Wait Enable state in the idle loop
to stall the CPU pipelines and wait for an exception. This is already
done in more recent kernels, and it doesn't particularly save a lot of
power as the clocks and other logic are still all active.
Some chips may have an external clock and power management ASIC that
can be leverage but it will be SoC specific. You could start there for
a particular board.
josh
But a lot of SoC like 440GT/EX support two modes of power save -
- Cut off power to other cores like PCIExpress/USB/MAC/UART etc
- Lower the CPU frequency
Is it possible to scale down CPU freq or cut off power to unused cores
while the CPU is idle? The ideal way would be that the SoC registers a
set of handlers to the Kernel that can power off/up the SoC components
or reduce/restore CPU frequency. Does any such framework exist for
PowerPC processors, because most PowerPC based SoC (FreeScale/AMCC
etc) support the above two modes of power save.
On 7/3/08, Josh Boyer [off-list ref] wrote:
Hi all,
I would like to know whether Linux on PowerPC puts the processor on
low power mode during idle state. Most PowerPC processors support a
low power mode. I am looking to add support for low-power mode in
Linux for AMCC 4xx processors.
My questions are the following:
1) Is tickless kernel integrated into the PowerPC tree?
2) Does Linux kernel puts the processor in low power mode if cpu is
idle, and takes it out of that state on wakeup (triggered by
interrupt/WOL etc)?
Any suggestions/comments would be welcome.
The 4xx cores themselves are designed to have power savings without
explicit software management. The only sort of "NAP/DOZE" possible
from software on them is to use the Wait Enable state in the idle loop
to stall the CPU pipelines and wait for an exception. This is already
done in more recent kernels, and it doesn't particularly save a lot of
power as the clocks and other logic are still all active.
Some chips may have an external clock and power management ASIC that
can be leverage but it will be SoC specific. You could start there for
a particular board.
josh
From: Benjamin Herrenschmidt <benh@kernel.crashing.org> Date: 2008-07-03 23:44:23
On Thu, 2008-07-03 at 12:52 -0700, prodyut hazarika wrote:
But a lot of SoC like 440GT/EX support two modes of power save -
- Cut off power to other cores like PCIExpress/USB/MAC/UART etc
- Lower the CPU frequency
Is it possible to scale down CPU freq or cut off power to unused cores
while the CPU is idle? The ideal way would be that the SoC registers a
set of handlers to the Kernel that can power off/up the SoC components
or reduce/restore CPU frequency. Does any such framework exist for
PowerPC processors, because most PowerPC based SoC (FreeScale/AMCC
etc) support the above two modes of power save.
Cutting off power to individual cores should be done by the drivers for
those cores (let's call the devices). For example, the USB driver could
cut of power when nothing is connected, though I'm not sure if it can do
that without losing the ability to detect new connections, but then, USB
has a suspend state that should make it mostly unnecessary. For the
UART, you may lose the ability to receive.
At the end of the day, if you don't use at all a core, disable it from
your firmware and set it as unused in the device-tree.
Ben.
On Fri, 04 Jul 2008 09:42:51 +1000
Benjamin Herrenschmidt [off-list ref] wrote:
On Thu, 2008-07-03 at 12:52 -0700, prodyut hazarika wrote:
quoted
But a lot of SoC like 440GT/EX support two modes of power save -
- Cut off power to other cores like PCIExpress/USB/MAC/UART etc
- Lower the CPU frequency
Is it possible to scale down CPU freq or cut off power to unused cores
while the CPU is idle? The ideal way would be that the SoC registers a
set of handlers to the Kernel that can power off/up the SoC components
or reduce/restore CPU frequency. Does any such framework exist for
PowerPC processors, because most PowerPC based SoC (FreeScale/AMCC
etc) support the above two modes of power save.
Cutting off power to individual cores should be done by the drivers for
those cores (let's call the devices). For example, the USB driver could
cut of power when nothing is connected, though I'm not sure if it can do
that without losing the ability to detect new connections, but then, USB
has a suspend state that should make it mostly unnecessary. For the
UART, you may lose the ability to receive.
At the end of the day, if you don't use at all a core, disable it from
your firmware and set it as unused in the device-tree.
Right.
To answer the "can the CPU frequency be scaled" question, the answer is
"maybe". You certainly want to make sure that timebase stays constant
in those transitions, which is easiest to do if timebase is driven by
an external clock and not the CPU clock.
There are also PLB/OPB divisor ratios that have to be taken into
account when changing the CPU frequency. If the hardware can handle
this for you, then it should work well enough. I know of at least one
4xx chip that has had CPU frequency scaling working.
If you pursue this at all, you will want to do it as a cpufreq driver.
That will allow you to use the generic cpufreq infrastructure in the
kernel, including the governors.
josh