Re: idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

Re: idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

[Mikael Pettersson]

On Sat, 1 Sep 2001 11:03:12 EDT, Floydsmith@aol.com wrote:

>Now, I can get everything (my ide ls-120 and ide HP 8Gig tape) to work in 
>...
>If I try not to use SCSI emulation for all 2.4.x kernels (including:
>Kernel 2.4.9-ac5 on  i686
>then
>ide-tape: ht0: I/O error, pc =  8, key =  5, asc = 2c, ascq =  0
>tar: /dev/ht0: Cannot read: Input/output error
>(writes work OK though)
>
>As mentioned above, scsi emulation works for 2.4.4 (reads and writes). But if 
>turned on in 2.4.9-ac5, then I get
>/dev/st0: No such device

FWIW, my Seagate 4/8 GB ATAPI tape drive works just fine in all
2.4.x kernels as /dev/{n,}ht0 -- no SCSI emulation for me.

There are two known problem areas, which may or may not explain
your problems:
- block size: The 2.4 ide-tape driver only works reliably if you
  write data with the correct block size. If you don't write full
  blocks the last block of data may not be readable.
  The driver will log the block size to the kernel log when it's
  initializing, so you can take that value and pass it to your
  backup utility (26KB in my case so I pass -b52 to tar).
- HP's not-quite ATAPI drives: Don't know about your model, but the
  HP 14(?)GB model is believed to deviate from ATAPI standards.

/Mikael

Re: idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

[Pete Zaitcev]

> From: Mikael Pettersson <mikpe@csd.uu.se>
> Date: Fri, 28 Sep 2001 11:15:07 +0200
> Cc: linux-kernel@vger.kernel.org, linux-tape@vger.kernel.org

>  > By the way, why does everyone insist on using ide-tape?
>  > It seems to be broken beyond any repair by injection of
>  > lethal poison marked "OnStream Support" (not that it was brilliant
>  > before, but that was the last nail in the coffin). Just use ide-scsi
>  > and be done with it. I really do not enjoy reading ide-tape.c.
> 
> I agree that ide-tape.c looks like a buggy piece of cr*p, but apart
> from that, what would I gain from using scsi tape on top of ide-scsi?
> Would it magically work on broken Colorados?

Umm... I hope someone else would fix ide-scsi when it breaks :)

I fixed Colorado, but it's certainly not the last bug.
Frist, DMA must be off, and it is often on by default.
Second, Dell QA already filed a new tapemark related bug...

-- Pete

Re: idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

[Mikael Pettersson]

Pete Zaitcev writes:
 > By the way, why does everyone insist on using ide-tape?
 > It seems to be broken beyond any repair by injection of
 > lethal poison marked "OnStream Support" (not that it was brilliant
 > before, but that was the last nail in the coffin). Just use ide-scsi
 > and be done with it. I really do not enjoy reading ide-tape.c.

I agree that ide-tape.c looks like a buggy piece of cr*p, but apart
from that, what would I gain from using scsi tape on top of ide-scsi?
Would it magically work on broken Colorados?

/Mikael

Re: idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

[Pete Zaitcev]

> Date: Wed, 5 Sep 2001 00:34:57 +0200 (MET DST)
> From: Mikael Pettersson <mikpe@csd.uu.se>
> To: zaitcev@redhat.com
> Cc: Floydsmith@aol.com, linux-kernel@vger.kernel.org,
>    linux-tape@vger.kernel.org

> >> - block size: The 2.4 ide-tape driver only works reliably if you
> >>   write data with the correct block size. If you don't write full
> >>   blocks the last block of data may not be readable.
> >
> >I fixed that some time ago, it's in current -ac
> >if not in Linus's tree.
> 
> Sorry, but that's not correct. I just ran a test, and the bug is
> still there in 2.4.9-ac7. Maybe you're thinking of some other bug?

OK, perhaps you are right. I received a credible report from
Ed Tomlinson that the "reading the last block" bug is in 2.4.10.

It seems that either I fixed something else with the same symptoms
or I fixed it improperly. Unfortunately, I cannot reproduce it.

By the way, why does everyone insist on using ide-tape?
It seems to be broken beyond any repair by injection of
lethal poison marked "OnStream Support" (not that it was brilliant
before, but that was the last nail in the coffin). Just use ide-scsi
and be done with it. I really do not enjoy reading ide-tape.c.

-- Pete

Re: idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

[Mikael Pettersson]

On Mon, 3 Sep 2001 20:14:06 -0400, Pete Zaitcev wrote:

>> - block size: The 2.4 ide-tape driver only works reliably if you
>>   write data with the correct block size. If you don't write full
>>   blocks the last block of data may not be readable.
>
>I fixed that some time ago, it's in current -ac
>if not in Linus's tree.

Sorry, but that's not correct. I just ran a test, and the bug is
still there in 2.4.9-ac7. Maybe you're thinking of some other bug?

ide-tape tells me it uses a 14*26KB buffer for my Seagate STT8000A.
If I dd a 39KB (1.5 "buffer units") file with bs=1k to /dev/ht0 it tells
me it wrote 39 blocks. If I then rewind and dd with bs=1k from /dev/ht0
it only reads 26 blocks. The same happens in 2.2 + Hedrick's IDE patch.

2.2 vanilla reads 56 blocks, of which the first 39 are identical to
what I initially wrote. The last 13 contain junk but that's not a big
problem since I back up with tar which writes its own EOF mark.

/Mikael

Re: idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

[Pete Zaitcev]

> >Kernel 2.4.9-ac5 on  i686
> >then
> >ide-tape: ht0: I/O error, pc =  8, key =  5, asc = 2c, ascq =  0
> >tar: /dev/ht0: Cannot read: Input/output error
> >(writes work OK though)

> - block size: The 2.4 ide-tape driver only works reliably if you
>   write data with the correct block size. If you don't write full
>   blocks the last block of data may not be readable.

I fixed that some time ago, it's in current -ac
if not in Linus's tree. The bug has nothing to do
with the ASC=2c problem the original poster complained.

> - HP's not-quite ATAPI drives: Don't know about your model, but the
>   HP 14(?)GB model is believed to deviate from ATAPI standards.

Colorado sucks... I think I have a bug pending about it,
did not look well yet. IIRC it was giving ASC=24, not 2c.
I'll look it up.

-- Pete

idetape broke in 2.4.x-2.4.9-ac5 (write OK but not read) ide-scsi works in 2.4.4

[Floydsmith]

Now, I can get everything (my ide ls-120 and ide HP 8Gig tape) to work in 
2.2.18. I can also get both to work in 2.4.4 (and earlier 2.4.x) but only 
with what I believe is an unnecessay "workarround" - namely by configuring 
IDESCSI into the kernel (SCSI emulation) and then using boot param 
hdc=ide-scsi. However, I feel, that I should not have to do that with any 
kernel (that is, SCSI emulation should be optional for this device) because 
this was the case for 2.2.18.

If I try not to use SCSI emulation for all 2.4.x kernels (including:
Kernel 2.4.9-ac5 on  i686
then
ide-tape: ht0: I/O error, pc =  8, key =  5, asc = 2c, ascq =  0
tar: /dev/ht0: Cannot read: Input/output error
(writes work OK though)

As mentioned above, scsi emulation works for 2.4.4 (reads and writes). But if 
turned on in 2.4.9-ac5, then I get
/dev/st0: No such device
So, with kernels above 2.4.4, I can't use the drive at all (or at least can't 
"read" with it which makes it useless).

The ideal situtation (at least for me) would be to have 2.4.x  kernels behave 
like the 2.2.x ones, that is, scsi emulation would not need to be turned on 
for these ide drives (that is, it would be optional) and I would go back to 
using /dev/ht0.

Floyd,