Re: [MiNT] MiNT, CT60 and power off

["Konrad Kokoszkiewicz" <draco@obta.uw.edu.pl>]

>> 1) Disk caches are built-in from 1.15.0 onwards, IIRC.
>
> Well, but I had problems with 1.16.0alpha that is downloadable
> from you site. Not with the previous versions. Cache were cancelable,
> it is not anymore.

Not true. Please read carefully:

As far as I know, there are two types of filesystem cache operation:
write-back and write-through. The "dangerous" one is the first one. It
should only be active when you explicitly enable it using the WB_ENABLE in
your mint.cnf. Otherwise only the write-through cache is active, which only
serves as read cache and does not affect writes (all data are written
immediately to the media).

Of course there is some probability that there is a bug which allows the
write-back cache to be active even when you haven't enabled it. But I
actually don't believe in it, because the difference between two types of
cache operation can be easily observed on floppies. You enable the cache for
A: and you immediately see the difference in operation. You disable it, the
previous behaviour is back.

Frank Naumann could probably tell more.

> My problem is not only due to the shutdown of the Hades, but
> I'm a programmer and I make various tests. The faintest
> problem with programs, that causes a reset, can lead to a disk
> crash.

Yes. That's why the write-back cache can be disabled. It hasn't been
demonstrated so far that this disable is not effective. But even so you
should shutdown the system before switching the power off.

BTW. memory protection decreases the probability of an accidental reset.

>> 2) No disk caches are effective for TOSFS.
>
> So, it must have changed meanwhile.

No.

> It is you that told me
> that there was a cache on TOSFS, and that it was the only
> difference between previous kernels.

No, I never told you that. "TOSFS" is the filesystem that uses routines
which reside in the ROM GEMDOS. This filesystem is not cached by MiNT. Never
was. Never will be.

The "difference" between previous kernels and 1.16.0 alpha is that MiNT
1.16.0 does not contain TOS FS driver at all. All FAT disks are handled by
the internal filesystem called "FATFS". Of course, this one is fully cached,
as all others, by the same global cache. This cache can be only enabled by
the WB_ENABLE keyword in the mint.cnf file.

>> 3) Filesystem drivers and block cache in 1.16.0 are (IIRC) exactly
>> the same as in 1.15.12.
>
> How do you explain that I had no problems with 1.15.12, apart
> that there is a non cancelable cache with TOSFS in the alpha
> versions ?

Well, have you ever tried to fix a car that is a few thousands kilometers
from you?

I have no clue. I heard about various problems with MiNT, but you are
perhaps the only person who told that filesystems don't work well.

You have replaced the 1.15.12 with 1.16.0, booted successfully, worked for
some time and then switched the power off, right? To my knowledge, the
write-back cache should never become active for you UNLESS you had WB_ENABLE
keyword in your mint.cnf with a list of drives to cache. On 1.15.12 it
didn't work, because TOSFS cannot be cached. But once you loaded 1.16.0 the
cache got activated, and once you switched the power off data loss occurred.

I cannot imagine another scenario especially that I personally never had any
problems with it, neither on Falcon, nor on Aranym, eventhough I keep all fs
caches enabled and eventhough I experienced crashes. Even if there is a bug,
it is hardly probable for me to trace it down, because I don't know what to
search for. Perhaps you try to talk to Frank Naumann, the FATFS author?

> Should I test a more recent kernel (I won't) ?

So why asking?

--
CVV
Konrad M.Kokoszkiewicz, http://draco.atari.org

** Ea natura multitudinis est, aut seruit humiliter, aut superbe dominatur.
** Taka to już natura pospólstwa, albo służalczo się płaszczy,
** albo bezczelnie się panoszy. (T. Liuius XXIV, 25).