IDE 48 bit addressing causes data corruption

IDE 48 bit addressing causes data corruption

[Felix Domke]

Hi,

i'm having linuxppc-embedded based system (2.4.21-pre, but 2.4.20 shows 
same results) with a normal ATAPI-5-styled IDE controller.

When using a 200GB Maxtor HDD, there are strange effects. When writing 
just the sector number to every sector (or every 100MB is enough), and 
reading back, data from incorrect sectors is read. The upper 24bit of 
the LBA seem to be invalid, shows like the upper 24bit aren't updated in 
the 2 cycle LBA-write in the ide-disk.c. For example, reading from 
0x946000000 (LBA 0x4A30000) will read from 0x2f46000000 (LBA 0x17A30000).
note that i lineary fill the disk, and 0x17xxxxxx is the highest 
possible LBA, so the last upper bit from the write won't be actualized 
any more when reading back from lower addresses.

However, reading them back (using the HOB-bit) will work and give 
correct results. I don't really know whats going wrong, except that 
there are addressing-faults.

Is this a known bug of the Maxtor 6Y200L maybe? the 160GB version showed 
the same effect.

Is it possible that the IDE-Controller causes the fault? normal 28bit 
HDDs work just fine.


Can somebody please confirm again that i don't need an atapi-6 (ATA133) 
controller to use LBA48 ?

Regulary some people are stating this, and regulary some people tell 
that these people are wrong.


felix

Re: IDE 48 bit addressing causes data corruption

[Alan Cox]

On Tue, 2003-03-18 at 14:30, Felix Domke wrote:
> Can somebody please confirm again that i don't need an atapi-6 (ATA133) 
> controller to use LBA48 ?
> 
> Regulary some people are stating this, and regulary some people tell 
> that these people are wrong.

LBA48 support and UDMA100/133 support are unrelated to one another.
There are controllers with one or the other, eg the older ALi can do
UDMA133 but not LBA48

Consult your IDE controller vendor

Re: IDE 48 bit addressing causes data corruption

[Felix Domke]

Alan Cox wrote:
> LBA48 support and UDMA100/133 support are unrelated to one another.
> There are controllers with one or the other, eg the older ALi can do
> UDMA133 but not LBA48

OK, looks bad for me.

Just for my interest:

Why does a certain IDE controller not support LBA48?

I always thought an IDE controller isn't more than some ISA-styled bus 
with 3 address lines, 2 chip  selects and special stuff for 
DMA-transfers, together with very special timing generators for PIO-modes.

Whats the problem with these controllers? LBA48, i thought, isn't more 
than writing the LBA-registers twice (because of the FIFO), and using 
different commands for reading/writing (the _EXT functions).

felix

Re: IDE 48 bit addressing causes data corruption

[Alan Cox]

On Tue, 2003-03-18 at 18:20, Felix Domke wrote:
> Why does a certain IDE controller not support LBA48?

In DMA mode some of them have state machines that know too much.
The double pump of the control registers confuses them.