On Mon, Jun 04, 2012 at 03:21:06AM -0700, Sagar Dharia wrote:

quoted

The enumeration slim_ch_proto is incorrect. It declares 2 transport
protocols which do not exist in the specification: SLIM_HARD_ISO;
SLIM_AUTO_ISO.

The enums are more SW representation (and not 1-1 mapping). Difference
between HARD_ISO and AUTO_ISO with example:
Let's say the root frequency is 24.576MHz. then all 4K family channels
(sample rate multiple of 4K) can run isochronously, and all 11.025KHz
channel can run with good efficiency.
So if a client wants 11.025KHz and is does not want to do flow-control at
this root frequency, then the client can specify AUTO_ISO to get the next
available isochronous frequency.

I understand it is not a 1-to-1 mapping. However it *is* used as such:

 wbuf[2] = (u8)((segdist & 0xF00) >> 8) | (slc->prop.prot << 4);

which results in NEXT_DEFINE_CHANNEL messages with an invalid TP
field.

More importantly I think it makes it harder to understand the
framework. The concept of HARD_ISO and AUTO_ISO are extensions, and
should really look like such, say perhaps:

enum {
       SLIM_ISO,
       SLIM_PUSH,
       SLIM_PULL,
       SLIM_LOCKED,
       SLIM_ASYNC_SMPLX,
       SLIM_ASYNC_HALF_DUP,
       SLIM_EXT_SMPLX, 
       SLIM_EXT_HALF_DUP,
       SLIM_AUTO_ISO,
       SLIM_USER_DEF_1 = 14,
       SLIM_USER_DEF_2 = 15,
       /* extensions */
       SLIM_HARD_ISO,
       SLIM_AUTO_ISO
};       

quoted

b) Similarly to (a) the driver may be probed before the device has
been given a logical address (LA). This makes sense in the case of
driver that turns on the device (say via gpio) once the bus has
booted. However, the driver then needs to sit and poll
slim_get_logical_addr() until the logical address.

This is not the case anymore.
While taking care of the comments for RFC, I have introduced a completion
that will be signalled when LA is given to the device. The driver can wait
on that completion (wait_enum) instead of polling.

Yes, my mistake. The driver wouldn't have to poll if there was another
callback. So I don't see how the completion mechanism is superior: it
forces a synchronous interface to asynchronous events, or the driver
developer has to work around it.

-- 
If you wake up and you're not in pain, you know you're dead.
(Russian Proverb)