Re: PCIexpress Slot_Power_Limit and Power budgetting.

[Jesse Barnes <jbarnes@virtuousgeek.org>]

On Thu, 19 Nov 2009 11:20:50 +0100
"Stephane GONAUER" <stephane.gonauer@sofrelog.com> wrote:
> I am a embedded system designer and currently working on the
> preliminary design of a PCIepress expansion board to be used on x86
> motherboard.
> 
> My board is to be an x1 PCie board. Thus according to the PCIe
> electromechanical specifications,  the board may consume 10Watts by
> default, and 25Watts after correct dialog is established with upper
> bridges and the root complex. (which after acknowledging the request
> should send a Slot_Power_Limit message to the board to allow for
> higher power consumption).
> 
> The board is based on a PEX8311 PCIe to local bus chip from PLX tech.
> This component is able to present its request for more power using
> the power budgeting capability as defined by the PCIe standard.
> 
> However this doesn’t seems to have effect on the system (running
> latest ubuntu with 2.6.31 kernel) and from my understanding of the
> kernel sources this is not supported. Is that correct ?
> 
> Are there ways that I’m not aware to request for more power in the
> existing source base ?

Not in PCIe that I'm aware of, aside from the existing ASPM support,
but that's a bit different.

> I’ve been reading the PCIE-port driver source to try to find a way to
> achieve my goal. My understanding is that this driver is very generic
> and as such shouldn’t be able to support power budgeting. I believe
> adding Power budgeting to it would require to match more closely the
> motherboard hardware. Am I correct ?

Yeah, we'd need to get power budget info from the firmware or platform
somehow, which means we'd have to add a platform specific portion to
the PCIe port driver, or at the very least some basic power
initialization routines that could be called from platform specific
code.

> Perhaps you have better ideas on the subject than me.
> 
> I’d be very glad if I could get some comments on this issue. I hope I
> haven’t been writing to the wrong people.

Given how long it's taken me to reply, maybe you have an implementation
already. :)

We'd definitely be interested in adding such a feature if your platform
supports it and has need of it.

-- 
Jesse Barnes, Intel Open Source Technology Center