[Tinyos-2.0wg] TinyOS 2.x WG/07.5.2006 Meeting Notes

[Philip Levis]

On Jul 6, 2006, at 2:56 PM, henri dubois-ferriere wrote:

>> Phil:
>>
>>
>> lets talk about ImmediateRequest.
>> AR says not needed.
>>
>> David:
>> Henri is using it.
>>
>> Kevin:
>> I could not understand his code. He is not going back to request  
>> if the
>> first call fails
>>
>
>
> in some cases (in particular in XE1205PhyP) there just isn't much
> reason to do a split-phase request if immediaterequest fails, because
> by the time the granted comes back to give you the bus it will be too
> late to be of any use, i.e., the radio fifo will have over-flowed (in
> rx; in tx it will have underflowed).
>
> The timing constraints are 1 byte time at 76kbps: 100us, which melts
> away pretty quickly once you throw in atomic overhead, latency to
> transfer first byte over SPI, etc.
>
> OT: As a historical aside the radio can in fact do 152kbps and after a
> lot of pain i managed to get the 1.x stack running solidly at that
> speed; but in 2.x (at least when i implemented the driver back in jan)
> overhead of things like SPI or arbiters was had grown just enough that
> i had to put the max bit rate at 76k.

In theory, if a platform *knows* that the bus will not be shared and  
only a single subsystem will use it, it can replace the chip bus  
policies. This would require some work, though. I think the current  
mindset is that the core should be flexible to a lot of platforms,  
sacrificing optimization if need be. But platforms can use special  
optimizations if they want.

Phil