Gilmore on USB power

[Mark J. Foster]

Hi, Mitch!

Very interesting discussion!  Your email arrived during my meeting today 
with Quanta on completely revising our USB power architecture.  To be 
brief (I'll go into more detail during the conference call tomorrow 
morning), we now completely understand why we've been seeing USB power 
issues, and the solution I'm proposing will eliminate the problem, while 
also allowing S/W to have much more control over power allocation.

First, we'll add better per-socket power filtering, which we've 
discovered was simply broken in the A-Test layout (that's why you would 
occasionally see devices that would work in one USB socket, but not in 
another socket on the same motherboard).  In addition, we will eliminate 
the USB-specific current limiter that's causing a variety of problems.  
Therefore, we'll be able to supply up to 7.5W to the USB ports.  The 
only H/W limit will be the current limiter on the 5V power supply, which 
drives only USB and the internal speaker amplifier.  It'll be up to 
software to ensure that we don't exceed total system power consumption 
limits, which should be fairly straightforward (I'll explain how in the 
call).

I think folks will like this solution!

Cheers!
MarkF

Mitch Bradley wrote:
> John Gilmore and I were having a conversation about OLPC.  He had some
> interesting comments regarding the power available to USB devices.
> (Forwarded to this list with his approval.)
>
> Mitch:
>
> The crux of the problem is that full support for arbitrary USB devices 
> blows our power budget right out of the water.
>
> John:
>
> OK, but devices should not end up with *flakey* power.  The USB protocol
> has the device specify how much power it needs.  If the system can't
> provide the amount of power that the device requests, it should turn
> off the device and report the error -- rather than providing it "some
> of" the power it requests.
>
> Is the OLPC violating the USB spec by failing to supply the requested
> power?  You may not be able to use the trademarked USB name, logo, or
> patented connector if your device violates their specs.
>
> When the OLPC is plugged into a wall, can it provide more power than
> when it's running from a battery?  Or is the current limitation in
> some intermediate circuit like a voltage regulator?  All the current
> "A" boards are running on wall warts, but reporting problems anyway.  
> Why?
>
> And whose "power budget" is this, anyway?  If the user would rather
> have an hour of usage with a USB device, rather than have a day of
> usage with no USB device, why is the design team making that choice
> for the user?  Pop up a message if you insist -- but don't just say,
> "Sorry, we have the power, but we won't use it for *that*!"  The
> user may need to copy one minute's worth of data to or from an external
> hard drive, or hook to an Ethernet for long enough to handle queued
> emails, or something; the design would deny them that opportunity?
>
> (Of course you'll run into devices that actually draw more than what
> they ask for -- that's the device's bug, and should be reported
> back so that they lose their USB certification if they don't fix it.
> But that problem shouldn't be too widespread, though you won't know
> til you start enforcing it.)
>
>