Re: kern/41889: st(4) driver splits larger IO request into peaces - compability problem with other OS's

[Wolfgang.Stukenbrock%nagler-company.com@localhost]

Hi again,

I haven't seen 38643 before - OK.

The problem there has the same cause.
A tape device has a little bit different semantic in reading and  
writing as a "normal" block device.
Blocknumber have no meaning for a tape and a tape block is the atom  
for a transfer.
Therefore a write should write exaclty one tape block - or should fail.
And a read should try to read exactly on tape block - or should fail.
A read my succeeed only if the requested size is at least the size of  
the tape block.
There is absolutly no way to get only "a second" half of a tape block  
from a tape.

This problem needs a fix as soon as possible!

My fix will "just" remove the silent misbehaviour for simple read and  
write calls.
A fix fore readv and writev is mentioned, but perhaps someone has a  
better idea ..

W. Stukenbrock

Zitat von "David Holland" <dholland-bugs%NetBSD.org@localhost>:

> The following reply was made to PR kern/41889; it has been noted by GNATS.
>
> From: David Holland <dholland-bugs%netbsd.org@localhost>
> To: gnats-bugs%NetBSD.org@localhost
> Cc:
> Subject: Re: kern/41889: st(4) driver splits larger IO request into peaces
>         - compability problem with other OS's
> Date: Sun, 16 Aug 2009 05:20:37 +0000
>
>  On Fri, Aug 14, 2009 at 11:50:00AM +0000,  
> Wolfgang.Stukenbrock%nagler-company.com@localhost wrote:
>   > The current implementation of physio() (in kern_physio.c)
>   > determines the maximum possible transfer size of the underlying HW
>   > be fist clamping to MAXPHYS and then asking the driver for further
>   > restrictions.
>
>  See also PR 38643. :-|
>
>  --
>  David A. Holland
>  dholland%netbsd.org@localhost