[membkdr] Provide a tile adapter to access internal rows

[Razer6]

This PR adds a custom row adapter that is being used by the memory backdoor util to access and manipulate rows of data. Integrators might are using SRAM primitives with a different internal organization than Width x Depth. They can provide a custom row adapter to the memory backdoor utility to for the needs of their memory prims.

[Razer6]

At the moment I am not sure why I get a Null error here

[Razer6]

@rswarbrick I am not sure how to fix that. Can you help me on that?
The constructor has the following statement:

   if (row_adapter != null) begin
      this.row_adapter = row_adapter;
    end else begin
      this.row_adapter = new();
    end

And later on, the row adapter is used like:

this.width                  = (n_bits / depth) - this.row_adapter.get_num_extra_bits();

However, this causes the following error (line 132 is the line from above when calling get_num_extra_bits()) :

Error-[NOA] Null object access
                ../src/lowrisc_dv_mem_bkdr_util_0/mem_bkdr_util.sv, 132
                  The object at dereference depth 1 is being used before it was
                  constructed/allocated.
                  Please make sure that the object is allocated before using it.

[rswarbrick]

Hmm, that seems really odd to me: I'm not sure what's going on. Can I suggest debugging by adding some print statements inside the if/else block (possibly using %p to show the value of this.row_adapter at the end of it?

[Razer6]

Hmm, that seems really odd to me: I'm not sure what's going on. Can I suggest debugging by adding some print statements inside the if/else block (possibly using %p to show the value of this.row_adapter at the end of it?

Stupid me! After the if/else, the default assignment from the parameter to the member variable was still there. This overwrote the previously allocated object with null again. @rswarbrick Fixed now, please take a look.

[rswarbrick]

Ha! I didn't notice it either :-) I'll have a proper look now.

[rswarbrick]

Now I've understood a bit more, I'm not sure this is quite right. I think this is returning several words of data packed together? But the read function is supposed to return just one word, isn't it?

Don't we need to extract just one lane here?

[andreaskurth]

Based on how the functions and variables are currently named, I agree with @rswarbrick. This code seems to decode a row and assign that to data. @Razer6: Doesn't this function need to select a word from the decoded row based on addr?

[Razer6]

I think the later function read8 (which calls read) needs a proper access function to select the right word?

[andreaskurth]

Ah, I see. I think what confused me is the short description of this function ("Read the entire word at the given address."). I suggest we change word to memory row, because it may contain more than one word with parity bits, depending on the memory architecture. Does that make sense?

[rswarbrick]

I think this has the same problem as the read function above. If we've got multiple words in the row, won't this need to do a read-modify-write?

[andreaskurth]

Based on how the functions and variables are currently named, I agree with @rswarbrick. This code seems to decode a row and assign that to data. @Razer6: Doesn't this function need to select a word from the decoded row based on addr?