How to find memory data width (x16 or x32?)

Configure it for 32 bits. Do a test write of 0xaaaabbbb. Look at the first 4 bytes, if they're not aaaabbbb then it's not 32 bits wide.

I am sorry, I did not give proper informatio in my mail. I have already tried this, but what happens is that the data gets written even thogh it is 16bits wide. The memory controller seems to perform 2 writes of 16 bits. Anyway, I need to figure out how many Rows and Column lines should be used to address the memory, and this configuration depends on whether memory is 16 bits or 32 bits wide, and also on the size of the memory (as per JEDEC standard for LPDDR).

Please take this information also into account.

Thanks.

Please elaborate about the problem.

You say that "the memory controller seems to perform 2 writes of 16 bits". However you also say that you're the one to "configure the system memory controller".

Can't you just configure it for 32 bit access, and then do the suggested 0xaaaabbbb test in 32 bit mode, and then configure the controller AGAIN according to the test result?

Or, if it comes (magically) preconfigured, can't you just read the configuration registers and thus know what type of memory is currently configured?

Marc

I kind of gave up after reading the OP's second response. If the memory controller is splitting one 32-bit access into two 16-bit accesses, then either it's configured for 16 bit mode or it's broken.

Wait, what kind of embedded device don't you already know what the data width of the memory is? Don't you have a line item on a BOM somewhere that tells you everything you need to know about the RAM?

--
Rob Gaddi, Highland Technology
Email address is currently out of order

-----------------^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

That's the key. *EXPLICITLY* set it to 32b wide. Probe the memory (taking care to understand how a "floating bus" will be read!!) and determine if it is, in fact, 32b wide.

If not, configure for 16b.

I suspect the OP's problem is the controller defaults to

16b mode. So, the 32b write gets broken into two contiguous 16b writes. Regardless of the actual RAM installed, both of these writes will succeed -- and read back properly.

BECAUSE THERE IS ALWAYS AT LEAST 16b WIDE MEMORY PRESENT!

The OP needs to *deliberately* force the controller to operate in 32b mode *before* making his first probe...

I often design flexible memory configurations in products and probe for "what's available" during bootstrap. I had one design that supported single *bit* increments in memory widths! :>

Maybe the guy writing the software doesn't know who is buying the hardware next year?

Rick

e

Lot of replies!! The reason why I want to know about data width of the memory is because I need to setup number of Rows and Columns in the memory controller. And these settings depend on the data width of the memory. It can still work if I setup a bigger value of rows and columns, but It would create memory aliasing problem (or feature!), which I want to avoid.

And I want to write a generic code which can be used for different memory sizes and data widths, since my code is going to run on the same platform with different memory configurations. I already have devices with different memory configuration on different chip selects, and there might be more such devices. FYI, I am writing code for Qualcomm qsd8250 platform.

SO CONFIGURE FOR THE WIDEST POSSIBILITY AND TEST FOR THE MEMORY ALIASING. Sheesh, this is what I and others have been telling you.

Thanks all for the help. I am now able to correct memory width and size configuration. I still have one problem. The approach suggested here works only if there is a memory connected on given chip select.

Is it possible to find if the memory is actually present on a given chip select? I have a device which has 256MB SDRAM on cs0, but there is no memory on cs1. I tried reading some status registers for cs1, but they look the same as cs0. So I can't determine if the memory is actually present there.

would

Walking 1 Data Test.

Ref:

--------------------------------------- Posted through

Thanks for this link. I have tried the test given there to check for missing memory chip. But in my case, the controller hangs if I try to access the memory which is not connected. When I am using JTAG to debug this, it shows "emulation running" message.

Any inputs on this?