Re: problem in nand.c v 1.16

Re: problem in nand.c v 1.16

[Thomas Gleixner]

On Tuesday, 26. February 2002 02:11, Radu Corlan wrote:
> I'm using the nand driver with a Toshiba TC58256 flash chip. We have found
> that the delay at the end of nand_command (currently 15us) is not enough
> for this part (the datasheet calls for 25us). Before making the change, we
> got a 0xff instead of the first byte of the page.
> Please take this into account in the next revision.

Please keep such discussions on the mailing-list !

Maybe we can solve it with a additional entry in the nand_chip structure.
		
-- 
Thomas
__________________________________________________
Thomas Gleixner, autronix automation GmbH
auf dem berg 3, d-88690 uhldingen-muehlhofen
fon: +49 7556 919891 , fax: +49 7556 919886
mail: gleixner@autronix.de, http://www.autronix.de  

Re: Support for parallel access in nand chips

[Thomas Gleixner]

On Tuesday, 26. February 2002 07:30, Radu Corlan wrote:
> On our application (a PPC405 embedded system), we use two TC58256 NAND
> chips conected on a 16-bit bus. we designed it this way in order to
> increase the read/write speed - it would double our bandwidth. The
> problem, of course, is that nand.c only accesses the flash 8 bits at a
> time.
>
> We intend to hack nand.c to use the 16-bit access (and make a special
> entry in nand_chip with a double erase_size), but i'd like to know if
> anybody thought of supporting the parallel access.
No. 
But please don't hack nand.c. Build a nand16.c from nand.c and hack this. 
Else we get a mess in the code, which was cleaned up a few days ago :)
And a lot of overhead of 8/16bit decisions. You just have to provide the same 
functionality as nand.c. You will not need a special entry in nand_chip. The 
information there is enough. 

-- 
Thomas
__________________________________________________
Thomas Gleixner, autronix automation GmbH
auf dem berg 3, d-88690 uhldingen-muehlhofen
fon: +49 7556 919891 , fax: +49 7556 919886
mail: gleixner@autronix.de, http://www.autronix.de  

Re: Support for parallel access in nand chips

[David Woodhouse]

rcorlan@pcnet.ro said:
>  ok - i was going to make a separate source anyway. Do you know if the
> upper layers mind the 32k erase size?

They'll be fine - it's still smaller than the typical erase size on NOR 
flash.

If you want to run JFFS2 on it, the ECC layout needs to be decided.

--
dwmw2

Re: problem in nand.c v 1.16

[Radu Corlan]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 1111 bytes --]

Sure, that would do ok. 

Radu

On Mon, 25 Feb 2002, Thomas Gleixner wrote:

> On Tuesday, 26. February 2002 02:11, Radu Corlan wrote:
> > I'm using the nand driver with a Toshiba TC58256 flash chip. We have found
> > that the delay at the end of nand_command (currently 15us) is not enough
> > for this part (the datasheet calls for 25us). Before making the change, we
> > got a 0xff instead of the first byte of the page.
> > Please take this into account in the next revision.
> 
> Maybe we can solve it with a additional entry in the nand_chip structure.
> 		
> -- 
> Thomas
> __________________________________________________
> Thomas Gleixner, autronix automation GmbH
> auf dem berg 3, d-88690 uhldingen-muehlhofen
> fon: +49 7556 919891 , fax: +49 7556 919886
> mail: gleixner@autronix.de, http://www.autronix.de  
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Support for parallel access in nand chips

[Radu Corlan]

On our application (a PPC405 embedded system), we use two TC58256 NAND
chips conected on a 16-bit bus. we designed it this way in order to
increase the read/write speed - it would double our bandwidth. The
problem, of course, is that nand.c only accesses the flash 8 bits at a
time.

We intend to hack nand.c to use the 16-bit access (and make a special
entry in nand_chip with a double erase_size), but i'd like to know if
anybody thought of supporting the parallel access.

Radu

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Re: Support for parallel access in nand chips

[Radu Corlan]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 1639 bytes --]

On Tue, 26 Feb 2002, Thomas Gleixner wrote:

> On Tuesday, 26. February 2002 07:30, Radu Corlan wrote:
> > On our application (a PPC405 embedded system), we use two TC58256 NAND
> > chips conected on a 16-bit bus. we designed it this way in order to
> > increase the read/write speed - it would double our bandwidth. The
> > problem, of course, is that nand.c only accesses the flash 8 bits at a
> > time.
> >
> > We intend to hack nand.c to use the 16-bit access (and make a special
> > entry in nand_chip with a double erase_size), but i'd like to know if
> > anybody thought of supporting the parallel access.
> No. 
> But please don't hack nand.c. Build a nand16.c from nand.c and hack this. 
> Else we get a mess in the code, which was cleaned up a few days ago :)
> And a lot of overhead of 8/16bit decisions. You just have to provide the same 
> functionality as nand.c. You will not need a special entry in nand_chip. The 
> information there is enough. 

ok - i was going to make a separate source anyway. Do you know if the
upper layers mind the 32k erase size?

Radu

> 
> -- 
> Thomas
> __________________________________________________
> Thomas Gleixner, autronix automation GmbH
> auf dem berg 3, d-88690 uhldingen-muehlhofen
> fon: +49 7556 919891 , fax: +49 7556 919886
> mail: gleixner@autronix.de, http://www.autronix.de  
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Re: Support for parallel access in nand chips

[Thomas Gleixner]

On Tuesday, 26. February 2002 21:24, Radu Corlan wrote:
> But nftl_format complains that the erase size is not 0x2000; Is there a
> newer version than 1.17?
You don't need NTFL on top of it. Just use JFFS2. There are some 
modifications neccecary fo the ECC layout. 

Radu: Can you send me your actual code, then I can give you some pointers

-- 
Thomas
__________________________________________________
Thomas Gleixner, autronix automation GmbH
auf dem berg 3, d-88690 uhldingen-muehlhofen
fon: +49 7556 919891 , fax: +49 7556 919886
mail: gleixner@autronix.de, http://www.autronix.de  

Re: nftl problem

[Thomas Gleixner]

On Wednesday, 27. February 2002 01:46, Radu Corlan wrote:
> Before i gave up on nftl, i found the following problem: nftl accesses the
> flash through the read_ecc and write_ecc function pointers in the slave
> mtd_info structure. However, when the slave structure is initialised, only
> the read and write function pointers are filled, not the read_ecc and
> write_ecc; i got kernel panic out of this, which got away when i filled
> the fields.
>
> Anyway, i gave up on nftl; my mtd seems to run ok now, as i put a ext2
> filesystem on the mtdblock0 device and it runs fine.
>
> But i can;t get jffs2 to work - i used the code from the latest snapshot
> on the infradead site - is it ok? Can i get a 'stable' version somewhere?
> my kernel is 2.4.2.
What does not work on JFFS2? You must setup the ECC config in your nand16 
driver. 
There's a lot of development going on. But try to use the code from CVS. I 
have just updated the nand stuff to support variable command delays and fixed 
a the CONFIG_ECC_LAYOUT stuff in nand.c

-- 
Thomas
__________________________________________________
Thomas Gleixner, autronix automation GmbH
auf dem berg 3, d-88690 uhldingen-muehlhofen
fon: +49 7556 919891 , fax: +49 7556 919886
mail: gleixner@autronix.de, http://www.autronix.de  

Re: nftl problem

[David Woodhouse]

rcorlan@pcnet.ro said:
> Before i gave up on nftl, i found the following problem: nftl accesses
> the flash through the read_ecc and write_ecc function pointers in the
> slave mtd_info structure. However, when the slave structure is
> initialised, only the read and write function pointers are filled, not
> the read_ecc and write_ecc; i got kernel panic out of this, which got
> away when i filled the fields. 

I don't recall receiving your patch for this.


--
dwmw2

Re: nftl problem

[Thomas Gleixner]

On Wednesday, 27. February 2002 02:30, Radu Corlan wrote:

> I didn't do the 16-bit stuff yet - use just one chip at this time. when i
> try to mount the jffs2 fisesystem, i get a loong delay; with debug
> enabled, it seems that it scans the device for freespace markers or
> something; After that, is mounts the filesystem, shows the right amount of
> free space, but when i write to it, the changes are not maintained over a
> remount or reboot.
>
> But i'm just thinking: the ppc is big endian; maybe i must set the option
> on mkfs?
Due to the fact, that we changed the complete layout, there's currently no 
support to use a Image, which was built on your development system. Use the 
raw flash. mount it. copy files to it. Please get the latest source out of 
CVS. There was a change in nand.c concerning the delays and the ECC_LAYOUT.
Enable Use JFFS2 layout for OOB area (CONFIG_MTD_NAND_ECC_JFFS2). 

-- 
Thomas
__________________________________________________
Thomas Gleixner, autronix automation GmbH
auf dem berg 3, d-88690 uhldingen-muehlhofen
fon: +49 7556 919891 , fax: +49 7556 919886
mail: gleixner@autronix.de, http://www.autronix.de  

Re: nftl problem

[David Woodhouse]

rcorlan@pcnet.ro said:
> probably because i didn't send it ;-). 

:)

> Should the patch go to the list, or directly to you?

The list is generally better - people there are more effective at dealing 
with stuff when I drop it, and it's better archived than my inbox :)

Thanks.


--
dwmw2

Re: nftl problem

[Thomas Gleixner]

On Wednesday, 27. February 2002 03:00, Radu Corlan wrote:
>
> Just to make sure: i can mount a freshly erased chip as a jffs2 fs, right?
Yep
-- 
Thomas
__________________________________________________
Thomas Gleixner, autronix automation GmbH
auf dem berg 3, d-88690 uhldingen-muehlhofen
fon: +49 7556 919891 , fax: +49 7556 919886
mail: gleixner@autronix.de, http://www.autronix.de  

Re: Support for parallel access in nand chips

[Radu Corlan]

But nftl_format complains that the erase size is not 0x2000; Is there a
newer version than 1.17?

Radu

On Tue, 26 Feb 2002, David Woodhouse wrote:

> 
> rcorlan@pcnet.ro said:
> >  ok - i was going to make a separate source anyway. Do you know if the
> > upper layers mind the 32k erase size?
> 
> They'll be fine - it's still smaller than the typical erase size on NOR 
> flash.
> 
> If you want to run JFFS2 on it, the ECC layout needs to be decided.
> 
> --
> dwmw2
> 
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Re: Support for parallel access in nand chips

[Radu Corlan]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 1308 bytes --]

it's ok, i modified nftl_setup - just wanted to know of there is a new
one. We are close to getting it work, so we'll keep at it for a while.

But indeed i should probably avoid using nftl. 

Radu

On Tue, 26 Feb 2002, Thomas Gleixner wrote:

> On Tuesday, 26. February 2002 21:24, Radu Corlan wrote:
> > But nftl_format complains that the erase size is not 0x2000; Is there a
> > newer version than 1.17?
> You don't need NTFL on top of it. Just use JFFS2. There are some 
> modifications neccecary fo the ECC layout. 
> 
> Radu: Can you send me your actual code, then I can give you some pointers
> 
> -- 
> Thomas
> __________________________________________________
> Thomas Gleixner, autronix automation GmbH
> auf dem berg 3, d-88690 uhldingen-muehlhofen
> fon: +49 7556 919891 , fax: +49 7556 919886
> mail: gleixner@autronix.de, http://www.autronix.de  
> 
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

nftl problem

[Radu Corlan]

Before i gave up on nftl, i found the following problem: nftl accesses the
flash through the read_ecc and write_ecc function pointers in the slave
mtd_info structure. However, when the slave structure is initialised, only
the read and write function pointers are filled, not the read_ecc and
write_ecc; i got kernel panic out of this, which got away when i filled
the fields.

Anyway, i gave up on nftl; my mtd seems to run ok now, as i put a ext2
filesystem on the mtdblock0 device and it runs fine.

But i can;t get jffs2 to work - i used the code from the latest snapshot
on the infradead site - is it ok? Can i get a 'stable' version somewhere?
my kernel is 2.4.2.

Thanks,

Radu


-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Re: nftl problem

[Radu Corlan]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 1641 bytes --]

> > Anyway, i gave up on nftl; my mtd seems to run ok now, as i put a ext2
> > filesystem on the mtdblock0 device and it runs fine.
> >
> > But i can;t get jffs2 to work - i used the code from the latest snapshot
> > on the infradead site - is it ok? Can i get a 'stable' version somewhere?
> > my kernel is 2.4.2.
> What does not work on JFFS2? You must setup the ECC config in your nand16 
> driver. 

I didn't do the 16-bit stuff yet - use just one chip at this time. when i
try to mount the jffs2 fisesystem, i get a loong delay; with debug
enabled, it seems that it scans the device for freespace markers or
something; After that, is mounts the filesystem, shows the right amount of
free space, but when i write to it, the changes are not maintained over a
remount or reboot.

But i'm just thinking: the ppc is big endian; maybe i must set the option
on mkfs?


Radu

> There's a lot of development going on. But try to use the code from CVS. I 
> have just updated the nand stuff to support variable command delays and fixed 
> a the CONFIG_ECC_LAYOUT stuff in nand.c



> 
> -- 
> Thomas
> __________________________________________________
> Thomas Gleixner, autronix automation GmbH
> auf dem berg 3, d-88690 uhldingen-muehlhofen
> fon: +49 7556 919891 , fax: +49 7556 919886
> mail: gleixner@autronix.de, http://www.autronix.de  
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Re: nftl problem

[Radu Corlan]

> 
> rcorlan@pcnet.ro said:
> > Before i gave up on nftl, i found the following problem: nftl accesses
> > the flash through the read_ecc and write_ecc function pointers in the
> > slave mtd_info structure. However, when the slave structure is
> > initialised, only the read and write function pointers are filled, not
> > the read_ecc and write_ecc; i got kernel panic out of this, which got
> > away when i filled the fields. 
> 
> I don't recall receiving your patch for this.

probably because i didn't send it ;-). Should the patch go to the list, or
directly to you?

I don't have the code here, so i can only send the patch tomorrow.

Radu

> 
> 
> --
> dwmw2
> 
> 
> 
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Re: nftl problem

[Radu Corlan]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 1660 bytes --]

On Tue, 26 Feb 2002, Thomas Gleixner wrote:

> On Wednesday, 27. February 2002 02:30, Radu Corlan wrote:
> 
> > I didn't do the 16-bit stuff yet - use just one chip at this time. when i
> > try to mount the jffs2 fisesystem, i get a loong delay; with debug
> > enabled, it seems that it scans the device for freespace markers or
> > something; After that, is mounts the filesystem, shows the right amount of
> > free space, but when i write to it, the changes are not maintained over a
> > remount or reboot.
> >
> > But i'm just thinking: the ppc is big endian; maybe i must set the option
> > on mkfs?
> Due to the fact, that we changed the complete layout, there's currently no 
> support to use a Image, which was built on your development system. Use the 
> raw flash. mount it. copy files to it. Please get the latest source out of 
> CVS. There was a change in nand.c concerning the delays and the ECC_LAYOUT.
> Enable Use JFFS2 layout for OOB area (CONFIG_MTD_NAND_ECC_JFFS2). 

I don;t care about the image anyway - i thought it was required. I'll try
again tomorrow. Thanks a bunch.

Radu

> 
> -- 
> Thomas
> __________________________________________________
> Thomas Gleixner, autronix automation GmbH
> auf dem berg 3, d-88690 uhldingen-muehlhofen
> fon: +49 7556 919891 , fax: +49 7556 919886
> mail: gleixner@autronix.de, http://www.autronix.de  
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------

Re: nftl problem

[Radu Corlan]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 1168 bytes --]

> Due to the fact, that we changed the complete layout, there's currently no 
> support to use a Image, which was built on your development system. Use the 
> raw flash. mount it. copy files to it. Please get the latest source out of 
> CVS. There was a change in nand.c concerning the delays and the ECC_LAYOUT.
> Enable Use JFFS2 layout for OOB area (CONFIG_MTD_NAND_ECC_JFFS2). 

Just to make sure: i can mount a freshly erased chip as a jffs2 fs, right?

Radu

> 
> -- 
> Thomas
> __________________________________________________
> Thomas Gleixner, autronix automation GmbH
> auf dem berg 3, d-88690 uhldingen-muehlhofen
> fon: +49 7556 919891 , fax: +49 7556 919886
> mail: gleixner@autronix.de, http://www.autronix.de  
> 
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/
> 
> 

-------------
Radu Corlan           Snail Mail: Bucuresti sect. 1, 
rcorlan@pcnet.ro             str. Argentina nr. 28, 71206 Romania

   You can still escape the "Gates" of Hell!   
                 Use Linux!                       
-------------