Re: Elan SC520 - problem with MTD

Re: Elan SC520 - problem with MTD

[Robert Kaiser]

Hi Alexander,

On Mon, 23 Apr 2001 you wrote:
> Hello Robert!
> Thank You very much for Your help.
> I worked with kernel 2.2.18 but could not achieve a success.
> Now I use Linux kernel 2.4.0 and latest version MTD from CVS. After booting
> in /proc/mtd I see :
> dev:    size   erasesize  name
> mtd0: 00800000 00040000 "SC520CDP Flash Bank #0"
> mtd1: 00800000 00040000 "SC520CDP Flash Bank #1"
> 
> Das ist fantastik!
> It was a good present for me last week ( my son was born last week!).

Well, congratulations to you and your wife then!!

> Now I must place a kernel to a flash and booting from it.
> What steps should be undertaken to reach that result?

One Question: Do you want to keep the BIOS ?

If yes, then I have a working solution, but it is not very nice:

Some Background info: 

In order to persuade the BIOS to boot from the Flash, you have to enable it's
"disk" emulation:  the BIOS can treat each of the flash banks as an emulated
disk. So far so good, but: these "disk" functions are only accessible through
the BIOS int 13h call, thus, you can't access it from Linux.

Worse even, the BIOS insists on doing "wear levelling" (i.e. it tries to
distribute erases evenly over all flash blocks). It does this by remapping
flash blocks when they have seen to many write erases. Therefore,  if you
write a continuous stream of data to the flash using the BIOS int13h function,
the BIOS will scatter your data all over the flash memory in an unpredictable
way. There is no way (other than buying a customized BIOS from General
Software) to disable his wear levelling. So the bottom line is:

* If you want to use the BIOS to boot from Flash, you must enable it's Disk
  emulation.
* If you use the Disk emulation, you must do all accesses to the boot flash
  through the BIOS int 13h.
* Bootloaders can access the Flash through the BIOS, but Linux can't.

Thus you essentially have to sacrifice one of the two banks entirely for booting
a kernel from it. Quite a waste considering it has 8 MB. In order to get a
bootable (via the BIOS) flash "disk", the easiest way is to put a minimal DOS
system on it with your kernel and LOADLIN. You can then run LOADLIN
from AUTOEXEC.BAT to boot your Linux Kernel.

As I said, this is ugly, but it should work.

However, if you want to get rid of the BIOS, you have to write your own startup
code for the board, which is not a trivial task. Actually, I'm currently
working on something like that, but it is not ready for release yet and I'm
making only slow progress as I have about a hundred other things to do.
However, I believe, Vipin Malik has gotten this to work for himself. Maybe if
you ask him nicely...

Cheers

Rob

----------------------------------------------------------------
Robert Kaiser                         email: rkaiser@sysgo.de
SYSGO RTS GmbH
Am Pfaffenstein 14                    phone: (49) 6136 9948-762
D-55270 Klein-Winternheim / Germany   fax:   (49) 6136 9948-10


To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Elan SC520 - problem with MTD

[David Woodhouse]

rob@sysgo.de said:
>  Some Background info: 
> In order to persuade the BIOS to boot from the Flash, you have to
> enable it's "disk" emulation:  the BIOS can treat each of the flash
> banks as an emulated disk. So far so good, but: these "disk" functions
> are only accessible through the BIOS int 13h call, thus, you can't
> access it from Linux.

> Worse even, the BIOS insists on doing "wear levelling" (i.e. it tries
> to distribute erases evenly over all flash blocks). It does this by
> remapping flash blocks when they have seen to many write erases.
> Therefore,  if you write a continuous stream of data to the flash
> using the BIOS int13h function, the BIOS will scatter your data all
> over the flash memory in an unpredictable way. There is no way (other
> than buying a customized BIOS from General Software) to disable his
> wear levelling. So the bottom line is: 

Has nobody tried to reverse-engineer the format they use for this and 
provide a driver in Linux? See the FTL and NFTL drivers for examples and 
ideas on how the BIOS might be doing the translation.

--
dwmw2




To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

RE: Elan SC520 - problem with MTD

[Alex Lennon]

>> However, if you want to get rid of the BIOS, you have to write your own
startup
>> code for the board, which is not a trivial task. Actually, I'm currently
>> working on something like that, but it is not ready for release yet and
I'm
>> making only slow progress as I have about a hundred other things to do.
>> However, I believe, Vipin Malik has gotten this to work for himself.
Maybe if
>> you ask him nicely...

I this BIOS compressed ? If not you should be able to write a simple BIOS
extension
to hook INT13h or INT19h. Then when your BIOS proper executes the interrupt
your
custom BIOS extension code loads your kernel out of linear flash - or do you
see a
problem with this approach ?

Regards,

	Alex




To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

RE: Elan SC520 - problem with MTD

[Robert Kaiser]

On Mon, 23 Apr 2001 you wrote:
> >> However, if you want to get rid of the BIOS, you have to write your own
> startup
> >> code for the board, which is not a trivial task. Actually, I'm currently
> >> working on something like that, but it is not ready for release yet and
> I'm
> >> making only slow progress as I have about a hundred other things to do.
> >> However, I believe, Vipin Malik has gotten this to work for himself.
> Maybe if
> >> you ask him nicely...
> 
> I this BIOS compressed ? If not you should be able to write a simple BIOS
> extension
> to hook INT13h or INT19h. Then when your BIOS proper executes the interrupt
> your
> custom BIOS extension code loads your kernel out of linear flash - or do you
> see a
> problem with this approach ?

Interesting idea -- a bit like the DOC approach  ;-)

Though I'm not sure if I understand that BIOS extension part right. As  far as I
can tell, the BIOS is not compressed, but it occupies the only ROM socket on the
board. A BIOS extension would have to be put in a seperate ROM, along with an
0xaa55 header and a checksum so the BIOS calls it, right ? 

In principle this could work. One would basically have to put part of the
MTD code into the BIOS extension to do the flash accesses. Keep in mind though
that the BIOS must be entered and left in real-mode (i.e. no access beyond 
1MB), so the BIOS extension would have to make the transition to protected mode
and back. I've done this a couple of times but this stuff is sooo ugly
that I always want to forget all about  it immediately after the job is done ;-)

To be honest, I don't think this is worth the trouble, I'd rather dump the BIOS
completely and write my own startup code.


Rob

----------------------------------------------------------------
Robert Kaiser                         email: rkaiser@sysgo.de
SYSGO RTS GmbH
Am Pfaffenstein 14                    phone: (49) 6136 9948-762
D-55270 Klein-Winternheim / Germany   fax:   (49) 6136 9948-10


To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

RE: Elan SC520 - problem with MTD

[Alex Lennon]

Rob,

>>Though I'm not sure if I understand that BIOS extension part right. As
far as I
>>can tell, the BIOS is not compressed, but it occupies the only ROM socket
on the
>>board. A BIOS extension would have to be put in a seperate ROM, along with
an
>>0xaa55 header and a checksum so the BIOS calls it, right ?

Yes indeed - a cut and paste job with an EPROM burner should do the job,
given
there is a couple of Kb free up there :-) Alternatively, and I guess this is
_particularly_
implementation specific, if the flash array is visible at boot time around
0xC000:0 - DFFF:0
you can put the BIOS extension in the flash array where the BIOS should pick
it up

>>In principle this could work. One would basically have to put part of the
>>MTD code into the BIOS extension to do the flash accesses. Keep in mind
though
>>that the BIOS must be entered and left in real-mode (i.e. no access beyond
>>1MB), so the BIOS extension would have to make the transition to protected
mode
>>and back. I've done this a couple of times but this stuff is sooo ugly
>>that I always want to forget all about  it immediately after the job is
done ;-)

I'm looking at this myself at the moment - the simplest solution is to
perform the hook,
load a zImage into low mem and let the Linux startup code handle the nasty
protected
mode bits 'n' pieces. Works fine for me under 2.2. Unfortunately I really
can't seem to
manage to get 2.4 down to < 500Kb so I need bzImage which means that - yes -
there is some protected mode support required. Alternatively it might be
possible to
throw the kernel up to 0x100000 using 32-bit real mode. I need to check on
this....

Regards,

	Alex




To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

RE: Elan SC520 - problem with MTD

[Robert Kaiser]

On Mon, 23 Apr 2001 you wrote: 
> I'm looking at this myself at the moment - the simplest solution is to
> perform the hook,
> load a zImage into low mem and let the Linux startup code handle the nasty
> protected
> mode bits 'n' pieces. Works fine for me under 2.2. Unfortunately I really
> can't seem to
> manage to get 2.4 down to < 500Kb so I need bzImage which means that - yes -
> there is some protected mode support required. Alternatively it might be
> possible to
> throw the kernel up to 0x100000 using 32-bit real mode. I need to check on
> this....

If you're interested in this kind of stuff, have a look at our GPLed ROM loader.

ftp://ftp.sysgo.de/pub/elinos/1.0/updates/rolo_1.1_src.tar.gz

It loads a bzimage using 32-bit real mode. It is written mostly in C, but it
doesn't compile  with gcc (you need bcc). 

Rob

----------------------------------------------------------------
Robert Kaiser                         email: rkaiser@sysgo.de
SYSGO RTS GmbH
Am Pfaffenstein 14                    phone: (49) 6136 9948-762
D-55270 Klein-Winternheim / Germany   fax:   (49) 6136 9948-10


To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Elan SC520 - problem with MTD

[Vipin Malik]

Wow! Where (and why ;) were your hiding this gem for so long?!!!

This is the answer to a lot of maiden's prayers! Robert, are you the "rob" in the
CVS Id in the files?

Vipin

P.S. I saw a couple of mails ago that some folks mentioned that I have successfully

booted the linux kernel directly out of flash and installed the JFFS(2) as the root
fs
(and a ext2 fs as a compressed fs in ramdisk). That is all true.

Expect boot the kernel directly out of flash part, everything else is fully
documented
(some may say "painfully fully" documented) in my HOWTO in the CVS.

I have supplied, on request, the files to boot the kernel out of flash.  The
original source
of the hacked kernel boot code to do this was AMD (and it's open source of course).

Please let me know if you want those. HOWEVER! The solution is not a very elegant
one.
It involves making changes to some startup files (bootset.S and setup.S) in the
Linux
startup code itself.

Furthermore it only works with the 2.2.x series and with zImage (not bzImage).
Hence i've not
put it into CVS.

I think, if I understood the capabilities of the linked ftp files below properly,
this could be the
very elegant solution that we all were looking for. I'll investigate this a bit
further.




Robert Kaiser wrote:

> On Mon, 23 Apr 2001 you wrote:
> > I'm looking at this myself at the moment - the simplest solution is to
> > perform the hook,
> > load a zImage into low mem and let the Linux startup code handle the nasty
> > protected
> > mode bits 'n' pieces. Works fine for me under 2.2. Unfortunately I really
> > can't seem to
> > manage to get 2.4 down to < 500Kb so I need bzImage which means that - yes -
> > there is some protected mode support required. Alternatively it might be
> > possible to
> > throw the kernel up to 0x100000 using 32-bit real mode. I need to check on
> > this....
>
> If you're interested in this kind of stuff, have a look at our GPLed ROM loader.
>
> ftp://ftp.sysgo.de/pub/elinos/1.0/updates/rolo_1.1_src.tar.gz
>
> It loads a bzimage using 32-bit real mode. It is written mostly in C, but it
> doesn't compile  with gcc (you need bcc).
>
> Rob
>
> ----------------------------------------------------------------
> Robert Kaiser                         email: rkaiser@sysgo.de
> SYSGO RTS GmbH
> Am Pfaffenstein 14                    phone: (49) 6136 9948-762
> D-55270 Klein-Winternheim / Germany   fax:   (49) 6136 9948-10
>
> To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org



To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Elan SC520 - problem with MTD

[Robert Kaiser]

On Die, 24 Apr 2001 you wrote:
> Wow! Where (and why ;) were your hiding this gem for so long?!!!

It's been there all the time ....

> 
> This is the answer to a lot of maiden's prayers! Robert, are you the "rob" in the
> CVS Id in the files?

Yep :-)

If there is enough public interest in this, maybe I can setup a home
page/public CVS for it some day. If only I had more time :-(


> P.S. I saw a couple of mails ago that some folks mentioned that I have
> successfully booted the linux kernel directly out of flash and installed
> the JFFS(2) as the root
> fs (and a ext2 fs as a compressed fs in ramdisk). That is all true.
> 
> I have supplied, on request, the files to boot the kernel out of flash.  The
> original source of the hacked kernel boot code to do this was AMD (and it's
> open source of course).
> Please let me know if you want those.

Yes, please !

> HOWEVER! The solution is not a very
> elegant one.
> It involves making changes to some startup files (bootset.S and setup.S) in the
> Linux startup code itself.

I am currently writing code to support booting  the SC520 CDP from ROM
without BIOS using the above mentioned ROM bootloader (BTW, it's name is
"ROLO" :-)).

What I'm mostly interested in is the startup/initialization sequence for that
board. Once it has been properly initialized, the generic ROLO code should
work out of the box.

I have some code that does the basic SC520 initialization right after
reset and I am now trying to initialize the Super I/O chip (I have just
received the chip manual from Acerlabs). After that, I need to figure out a way
to setup the set up the PCI bridge... Another thing that is still missing in my
code is a proper RAM size detection. 

So any working code you might have along these lines -no matter how ugly-
would probably be helpful to me.

Rob

----------------------------------------------------------------
Robert Kaiser                         email: rkaiser@sysgo.de
SYSGO RTS GmbH
Am Pfaffenstein 14                    phone: (49) 6136 9948-762
D-55270 Klein-Winternheim / Germany   fax:   (49) 6136 9948-10


To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Elan SC520 - problem with MTD

[David Woodhouse]

rob@sysgo.de said:
>  I have some code that does the basic SC520 initialization right after
> reset and I am now trying to initialize the Super I/O chip (I have
> just received the chip manual from Acerlabs). After that, I need to
> figure out a way to setup the set up the PCI bridge... Another thing
> that is still missing in my code is a proper RAM size detection. 

> So any working code you might have along these lines -no matter how
> ugly- would probably be helpful to me.

Talk to the LinuxBIOS people. They have much of this working.

--
dwmw2




To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Elan SC520 - problem with MTD

[Ronald G Minnich]

On Tue, 24 Apr 2001, David Woodhouse wrote:

> > just received the chip manual from Acerlabs). After that, I need to
> > figure out a way to setup the set up the PCI bridge... Another thing
> > that is still missing in my code is a proper RAM size detection. 
> 
> > So any working code you might have along these lines -no matter how
> > ugly- would probably be helpful to me.
> 
> Talk to the LinuxBIOS people. They have much of this working.

hi rob, what is it you need to do?

ron



To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Boot kernel from flash on CDPSC520 w/o BIOS!

[Alexander Melichenko]

Hello, Vipin!
To boot kernel from flash I do :
- reset the board (z).
- init it (yi).
- load lstart16.abs to it (ns lstart16.abs ff800, ns lstart16.abs 7ff800).
- load zImage to it with offset 0 (ns zImage 0).
After that I reboot CDP, but nothing is happening.
What step is wrong? 

Alexander Melichenko.




To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Boot kernel from flash on CDPSC520 w/o BIOS!

[Vipin Malik]

Alexander Melichenko wrote:

> Hello, Vipin!
> To boot kernel from flash I do :
> - reset the board (z).
> - init it (yi).
> - load lstart16.abs to it (ns lstart16.abs ff800, ns lstart16.abs 7ff800).

Nope. You need to load the lstart16.abs file to the BIOS FLASH chip, NOT to
the FLASH bank0

>
> - load zImage to it with offset 0 (ns zImage 0).

Where did you init the root fs? Whe  zFlashImage boots, it will look to mount
the root fs. Also you need to
tell zFlah image what the root device is. See the HOWTO for more details.

>
> After that I reboot CDP, but nothing is happening.
> What step is wrong?

This is quite a complicated procedure. Make sure that you can read/write the
/mtd partitions. Read the HOWTO.

Vipin

>
>
> Alexander Melichenko.



To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org

Re: Boot kernel from flash on CDPSC520 w/o BIOS!

[Alexander Melichenko]

On Friday, April 27, Vipin Malic wrote :


> Alexander Melichenko wrote:
>
> > Hello, Vipin!
> > To boot kernel from flash I do :
> > - reset the board (z).
> > - init it (yi).
> > - load lstart16.abs to it (ns lstart16.abs ff800, ns lstart16.abs
7ff800).
>
> Nope. You need to load the lstart16.abs file to the BIOS FLASH chip, NOT
to
> the FLASH bank0
>
> >
> > - load zImage to it with offset 0 (ns zImage 0).
>
> Where did you init the root fs? Whe  zFlashImage boots, it will look to
mount
> the root fs. Also you need to
> tell zFlah image what the root device is. See the HOWTO for more details.
>
> >
> > After that I reboot CDP, but nothing is happening.
> > What step is wrong?
>
> This is quite a complicated procedure. Make sure that you can read/write
the
> /mtd partitions. Read the HOWTO.
>

Hello, Vipin!
Are the next steps correct?
- put lstart16.abs to 29F040 flash with offset 40000 (ns lstart16.abs
40000);
- put zImage to 29LV017 bank0 with offset 0 (???).
- put zFlash (root file system) to 29LV017 bank1 with offset 0 (???).
- rdev /dev/mtdblock??  /dev/mtdblock??.
- rdev -R zImage 0.

I experimented with that and other options, but currently result is bad.

Thank You.
           Alexander.



To unsubscribe, send "unsubscribe mtd" to majordomo@infradead.org