YAMAMOTO Takashi wrote:
>>> it's better to distinguish them from normal specfs operations.
>>> ie. another REQ_xxx, rather than using NULL/NODEV.
>> Oh, in that case, I can revert the previous change; though you'll have
>> to help a bit with that as I'm unfamiliar with them.
> 
> i'm not familiar with details of each drivers either...
> let's ask authers. :)  it seems that ad@ wrore or ported many of them.

Ah. :) I already did, a while ago, and the impression I received was
that what these "passthru" calls do is basically raw access to kernel
memory and/or disk in some of the cases.

>> What requests are we looking at? IIUC you can't really tell what kind
>> of request it is, so we should probably have a single request to
>> identify these, or do you have a better idea?
> 
> i think that a single request with dev_t of a controller as a context
> can cover the most cases.

We'll have to look into that. I think in some of the cases we can
distinguish between memory/disk access, etc., but perhaps asking ad@
what to do for these cases would be the best.

Andy?

-e.

-- 
Elad Efrat