CompactFlash for storage

[ Message cross-posted to comp.arch.embedded and comp.arch.storage ]

Hello everyone,

I have an x86-based system running Linux, and I'm considering using a
CompactFlash card for non-volatile storage, instead of a typical hard
disk drive.

I was wondering if I could just use the same IDE driver I've been using
for hard disk drives.

As far as I understand, if I use a CF-to-IDE adapter, and plug that
adapter into the IDE connector, then there is no problem.

However the motherboard provides a "CF socket", so I was thought I might
not need the CF-to-IDE adapter. If I stick a CF card in the CF socket,
will the regular IDE driver work, or do I need one of the MTD drivers?

Note: the datasheet states "CompactFlash Socket on IDE 0
(Primary/Master)" which seems to indicate the CF socket is just an
embedded CF-to-IDE adapter, and everything should work, right?

Another question: I've been using so-called DOMs (Disk On Module) which
often include some form of wear-leveling to even the wear of different
memory cells. Do typical CF cards include this feature?

How much electronics (other than plain storage) is there in a CF card?

Are features such as DMA support implemented in the card or are they
supported at a different level? (In the IDE controller? Elsewhere?)

Regards.

On Sep 4, 5:51 am, Spoon root@localhost wrote:

However the motherboard provides a "CF socket",

Without exception that I have seen, on-board CF sockets on x86 systems
are already wired in True-IDE mode.

embedded CF-to-IDE adapter, and everything should work, right?

The "adapter" is a single strapped pin that puts the device in True-
IDE mode. Otherwise it's just a connector.

Another question: I've been using so-called DOMs (Disk On Module) which
often include some form of wear-leveling to even the wear of different

All CF cards include wear leveling. However the strategies you use
when running off flash are very different from what you'd use running
off rotating magnetic media. Run some quick back-of-an-envelope
calculations and you'll see that even a seemingly-infrequent regular
write operation can VERY quickly eat the erase lifetime of typical
NAND flash chips.

Also: There are bonus gotchas when using CF media.

* Writes are slower than reads, and much more power-hungry.
* It used to be true - and I believe from recent experiment is still
true - that a power failure during write operations can leave the card
unusable. Not merely corrupt filesystem, but actually a useless card.
* In True-IDE mode, you cannot hot-swap cards.
* {etc...}

How much electronics (other than plain storage) is there in a CF card?

A typical CF card is more or less three chips: voltage regulator, NAND
flash [might be more than one physical chip], and microcontroller. The
performance of the CF card is heavily dependent on the capabilities of
the micro. Some companies even used to advertise "USES FAST H8
MICROCONTROLLER!".

supported at a different level? (In the IDE controller? Elsewhere?)

There is no "IDE controller" per se in a CF card, there's only the
micro.

In article 46DD2AA2.4040205@localhost, Spoon root@localhost wrote:

Another question: I've been using so-called DOMs (Disk On Module) which
often include some form of wear-leveling to even the wear of different
memory cells. Do typical CF cards include this feature?

Typical IDE DOMs and CF cards are the exact same item in a slightly
different shape, particularly if you go with the so-called "industrial"
ones from the same manufacturer. But one gotcha with motherboard CF
sockets (this is what I warned about before) is that some board makers,
for reasons that are unclear to me, do not actually run the signal lines
needed for DMA to the CF socket, so, on the same board, a DOM with the
exact same flash, microcontroller, and firmware _may_ be faster: you
should test for yourself.

--
Thor Lancelot Simon

"The inconsistency is startling, though admittedly, if consistency is to
be abandoned or transcended, there is no problem." - Noam Chomsky