Reattach/redetect allways connected umass device - is it possible ?

Poul-Henning Kamp phk at phk.freebsd.dk
Mon Mar 28 08:39:43 PST 2005


In message <424830AC.7090309 at samsco.org>, Scott Long writes:

>>>So are you saying that an async notification mechanism needs to be
>>>invented for SCSI, or that all SCSI users should be required to use
>>>SES or SAFTE enclosures for all SCSI devices, or that we should be like
>>>Windows and constantly poll the devices?
>> 
>> I don't care _how_ we make CAM/SCSI behave like the users expect.
>
>Well, I waved my hands at it for a few minutes, but nothing changed... 
>hmm.... =-)
>
>My question to you was partially rhetorical,

That's why you got a terse answer indicating that the problem may
be a disparity between SCSI and users (reasonable IMO) expectations
to how devices act.

I'm not claiming there is any perfect solution, but that doesn't
mean that we shouldn't attempt to create a decent or at least
workable solution.

CD drives are a very specific problem because the ioctls we implement
(and want to keep implementing for compatibility reasons) are defined
without regard to the drive vs. media dictomy.  atapi-cd implments
what looks to be a workable workaround for this.

I can't possibly why disk devices can't just simulate DTRT with
polling.  I even remember trying to implement that in scsi_da and
I probably still have the hole in my mailbox where a response from
the CAM/SCSI clued would have gone, had something like that existed.

I think I ran into this issue well over two years ago, and all my
attempts to get people to look at it failed and I simply parked it
"until somebody gives a shit".

Now that somebody does, can we stop religious slamming ancient and
bogus user interaction models around and work together to find out
how to implement or at least simulate something that makes sense
in this day and time ?

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.


More information about the freebsd-mobile mailing list