ONFI timing mode with onfi_set_features unsupported

Boris Brezillon boris.brezillon at free-electrons.com
Tue Nov 22 03:10:47 PST 2016


On Tue, 22 Nov 2016 12:02:27 +0100
Sascha Hauer <s.hauer at pengutronix.de> wrote:

> On Mon, Nov 21, 2016 at 02:51:04PM +0100, Boris Brezillon wrote:
> > Hi Sascha,
> > 
> > On Mon, 21 Nov 2016 14:23:27 +0100
> > Sascha Hauer <s.hauer at pengutronix.de> wrote:
> >   
> > > Hi all,
> > > 
> > > I have a i.MX6ul here with a Cypress S34ML04G2 connected. This Nand chip
> > > has ONFI support, but onfi_set_features is unsupported (opt_cmd is
> > > 0x3b). Currently the gpmi Nand driver calls nand_onfi_set_features(),
> > > gets -EINVAL as error and continues with a very slow default timing.
> > > 
> > > I assume the nand_onfi_set_features() call is just unnecessary for this
> > > chip, if I skip it, the chip works with the fast timing.
> > > 
> > > Any idea how to cope with this situation? I attached the most obvious
> > > patch, but it looks a bit hackish. Any suggestions or is the patch fine
> > > as is?  
> > 
> > It looks good to me. Why do you find this code hackish?
> > Of course, it would be even better to implement the  
> > ->setup_data_interface() method.  
> 
> Indeed, but my current project doesn't allow to spend that much time at
> the moment.

Understood. Let's wait for Han's review.

> 
> > 
> > BTW, can you patch the core to only send the ->set_feature() command
> > (to change the timings mode) when the chip supports it?  
> 
> With hackish I mean that I think the problem should be solved in the
> core. How about returning -EOPNOTSUPP from onfi_set_features() when the
> operation is not supported? The caller could then decide what to do
> without testing for bits in the onfi param page.

The problem is, ->onfi_set_feature() is a method that can be overloaded
by NAND controller drivers. Of course, we could add a wrapper around
->onfi_set_feature() (nand_set_feature()?), but then, the meaning of
-ENOTSUPP is not clear. It could be returned if the
->onfi_set_feature() is NULL or if the requested feature is not
supported.

Another solution would be to add an extra helper to check if a specific
feature is supported:

bool nand_feature_supported(nand, feature_id);




More information about the linux-mtd mailing list