SDIO multiplexing

Carsten Haitzler (The Rasterman) raster at openmoko.org
Fri May 2 04:24:56 CEST 2008


On Thu, 01 May 2008 19:13:24 +0100 Andy Green <andy at openmoko.com> babbled:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Somebody in the thread at some point said:
> 
> | just for the record: my main reason to not like the glamo is that it
> | severely
> | limits I/O passthrough to the microSD, hence effectively forbidding SD
> | and GPU
> | access at the same time.
> 
> Well it's not that bad at all.  Glamo SD works by setting up the
> transaction to the card, and then Linux goes and does something else
> until it gets an interrupt from the Glamo.  Only then it pulls the bulk
> data out of Glamo RAM.  So during the interrupt the CPU pulls SD data;
> the rest of the time, including while it waits for completion, it can do
> the graphics side.

the problem is that the bus is still shared. sharing that bus between IO for SD
and graphics is a killer - it's not fat enough to support graphics, letalone
both SD as well as the graphics. sure. the cpu sleeps waiting on interrupt -
but when we read, and while we read, we block the cpu - and it is artificially
slow on reads due to waiting on the glamo bus thus we loose something
like 80-90% of our cpu in sitting and waiting while this gets done as transfer
from glamo buffer to system ram is still the bottleneck.same with graphics and
writes. we we could write at the same speed we can read/write to ram we
wouldn't be in as bad a situation... but that'd need a bus that is a lot fatter
than the one to the glamo.

so when doing IO and graphics one of these (or both) will lose out as u now
have to share an already limited pipe for 2 data-heavy things. :(

-- 
Carsten Haitzler (The Rasterman) <raster at openmoko.org>




More information about the hardware mailing list