[ELDK] mmap: first byte always wrong

Stefano Babic sbabic at denx.de
Fri Nov 26 09:34:28 CET 2010


On 11/26/2010 09:16 AM, Alberto Caballero wrote:
> Hi,
> 

Hi,

> Just a clarification: I used U-Boot to store the image at the Flash memory.
> Then, using the U-Boot command line I am able to see the contents of the
> memory.

At this point we are sure that the flash has the correct contents and
can be accessed. However, we cannot make any assumptions about what
happens in linux.

> So the U-Boot operation has nothing to do with any memory mapping.

Right. The kernel driver is then responsible to work with the flash.
Surely the flash is accessed by MTD when the system boots to recognize
the chip type, and at least the CFI-Query command is sent.

> This is just a way of checking that the bytes are there directly accessing
> the physical memory.
> 
> Although the EP440xS board is currently unsupported, I cannot understand why
> the access to the mmap returned pointer is wrong. Is there a particular mmap
> code for each board type? Would it have something to do with issuing
> commands to the flash memory?

You are thinking that your flash memory works exactly as RAM, and even
with NOR flash this is not always the case. As example, I already told
you that if the MTD driver let the flash in a different status as result
of the CFI-Query command, you do not read the flash contents. And you
are not taking care that the flash cannot be accessed if another
operation (for example, erasing sectors: do you have jffs2 on another
partition ?) is running. The MTD take care with a mutex on a whole chip
if there are operations that do not allow to access the chip, and the
calling process is blocked until the chip becomes available. Maybe there
is a race condition, or maybe support for your board is broken. Reading
the flash via MTD you can check that.

Accessing the flash with mmap on /dev/men is a fast way to get into
trouble, as you have two actors (your code and the MTD driver) working
on the same physical resource, maybe at the same time.

Best regards,
Stefano Babic

-- 
=====================================================================
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80  Email: office at denx.de
=====================================================================


More information about the eldk mailing list