[RFC PATCH 2/2] nand: cavium: Nand flash controller for Cavium ARM64 SOCs

[Boris Brezillon]

On Mon, 15 May 2017 14:33:37 +0200
Boris Brezillon <boris.brezillon at free-electrons.com> wrote:

> Hi Jan,
> 
> On Sun, 30 Apr 2017 15:01:00 +0200
> Boris Brezillon <boris.brezillon at free-electrons.com> wrote:
> 
> >   
> > > 
> > > How far are you with the new interface, can you share some code?    
> > 
> > I started to rework the NAND framework a while ago [1], but never had
> > time to finish it. I think I was too ambitious, so let's try to be
> > pragmatic this time, and focus on one problem at a time.
> > 
> > Your problem here is the separation of the CMD/ADDR cycles (done in  
> > ->cmdfunc() and/or cmd_ctrl()) and the DATA cycles (done in
> > ->read/write_buf/byte/word()), which complexifies the driver logic.    
> > 
> > What you should look at is defining a proper nand_operation object
> > (here's my initial definition [2], but you may want/need to remove some
> > fields or add new ones) and add a new ->exec_op() hook to nand_chip
> > taking a nand_operation struct (+ a pointer to the nand_chip, of
> > course).
> > 
> > Once you have that, you should patch all accesses from the core to use
> > the new ->exec_op() interface instead of ->cmdfunc() +  
> > ->read/write_xx(). Of course, that means providing a compatibility layer    
> > for all drivers still implementing ->cmdfunc(), which is probably the
> > trickiest part of the job.
> > 
> > You'll have to rework the nand_do_read/write_ops() functions to
> > prevent the separation of the ->cmdfunc() call (done in
> > nand_do_read/write_ops() function) from the data transfer (done in
> > chip->ecc.read/write_page_xxx()).
> > 
> > I'll try to come up with an initial/ugly patch to show you the
> > direction, and I'll let you cleanup/massage the implementation ;-).  
> 
> I finally had time to finish the PoC I had in mind [1], unfortunately I
> didn't have time to provide a reference implementation to show you how
> it is supposed to work from a driver PoV.

One more thing, this is completely untested (only compile-tested), so
you're likely to face bugs when testing it ;-).