[Shr-User] uSD hosed, on every unstable.

Gennady Kupava gb at bsdmn.com
Wed Sep 8 21:31:17 CEST 2010


В Срд, 08/09/2010 в 20:48 +0100, Al Johnson пишет:
> On Wednesday 08 September 2010, jeremy jozwik wrote:
> > On Wed, Sep 8, 2010 at 9:50 AM, Gennady Kupava <gb at bsdmn.com> wrote:
> > > SD initalisation in kernel look strange and fact that we need
> > > different rootwait/rootdelay in kernel, while u-boot can read
> > > SD card almost instantly seriously puzzles me.
> > > 
> > > Also i can notice fact that I have 50% probability of successfull mmc
> > > read in u-boot wuth 2-4-2 timings, while 4-4-4 is 100% successfull, may
> > > be only in extremly rare cases. This makes me think that some delay (or
> > > proper sync) is needed in mmc sequence.
> > 
> > are you eluding to u-boot accessing the SD card incorrectly? if so,
> > how is it that previously locally saved u-boot version used to work.
> > and i used to be able to read the SD card within shr.
> 
> I think there are two separate issues here.
> 
> I think Gennady is correct to say there is something wrong in the kernel 
> handling of mmc on the glamo, and probably also something wrong in u-boot. If 
> everything was correct we wouldn't need a list of compatible and incompatible 
> cards, or parameters to fiddle with delays and clock speeds. This may be z
> another case where some study of the glamo documentation and the existing code 
> can help. I have a 4G Kingston card that's on the non-compatible list if 
> someone wants to try this.
> 
> I don't think that is the cause of your issue because your cards used to work, 
> but don't any more, with software that used to work for you and still does for 
> other people.
> 

I can say mean exactly, i know following bugs:

1. u-boot 2-4-2 need fix to read kernel from sd, but kernel works
perfect with this timings (never saw single read failure).
2. kernel can't access sd without large delay and this delay is
different in different versions. but u-boot can access data instantly!

Also, i know following things which is probably bugs:
3. sd card init in kernel look strange. some mess with voltage, some
delays. Need investigation.
4. sd card read speed is somehow 7 times slower than for same sd card on
host. glamo<->sd bus is 4 bit, same to host as far as i understand. no
idea why it should be so much slower. card read speed bottleneck is
_not_ cpu<-> glamo bus or glamo<->sd bus. Need more investigation, but
entirely possible it is just something wrong with driver/hardware setup.
5. yes. supported and unsupported look strange and i do not believe
glamo 'works' with some cards and do not with others. I never saw
devices with such selectivity. Need more inverstigation, look like
something wrong with hardware or may be glamo setup.

Gennady.




More information about the community mailing list