Re: [PATCH] mtd: rawnand: call onfi_fill_data_interface() once again after nand_detect

[Masahiro Yamada]

HI Miquel,

On Thu, Feb 7, 2019 at 10:02 PM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
>
> Hi Masahiro,
>
> Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> wrote on Thu, 7 Feb
> 2019 19:46:54 +0900:
>
> > On Thu, Feb 7, 2019 at 7:16 PM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> > >
> > > Hi Masahiro,
> > >
> > > Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> wrote on Thu,  7 Feb
> > > 2019 18:57:56 +0900:
> > >
> > > > nand_scan_ident() calls onfi_fill_data_interface() at its entry
> > > > to set up the initial timing parameters.
> > > >
> > > > The timing parameters are needed not only for ->setup_data_interface(),
> > > > but also for giving the correct delay to NAND_OP_WAIT_RDY, for example.
> > > >
> > > > If the driver sets the NAND_KEEP_TIMINGS flag, or does not support
> > > > ->setup_data_interface() hook, those parameters will never updated.
> > >
> > >                                                             ^ be
> >
> > Will fix (if v2 is welcome)
> >
> >
> > > >
> > > > Before nand_detect(), we never know whether the chip is ONFi or not.
> > > > So, onfi_fill_data_interface() has to assume the worst case, i.e.
> > > > non-ONFi.
> > >
> > > s/ONFi/ONFI/?
> >
> > Will fix.
> >
> > Looks like I was misunderstanding
> > maybe because the letter 'I' in the logo
> > (http://www.onfi.org/)
> > looks like a lowercase...
> >
> >
>
> Oh right. I don't know what's best. Pick your favorite :)
>
> >
> >
> > >
> > > >
> > > > After nand_detect(), if the chip turns out to be ONFi-compliant,
> > > > we can optimize tPROG_max, tBERS_max, etc.
> > > >
> > > > Call onfi_fill_data_interface() once again.
> > >
> > > Sorry but I don't get why this is needed as there is the same call at
> > > the top of this function. Can you be more specific on where/when the
> > > missing call produces a failure?
> >
> >
> > onfi_fill_data_interface() sets different values
> > for tPROG_max, tBER_max, tR_max, tCCS_min
> > depending on whether the chip is ONFI or not.
> >
> > For the first call, onfi_fill_data_interface()
> > chooses the else-part since we never know
> > the chip specification at this point.
> >
> > If we call onfi_fill_data_interface() once again
> > after nand_detect(), it may choose the if-part.
> >
> >
> > If a driver supports ->setup_data_interface(),
> > nand_init_data_interface() will set the optimal
> > timing parameters anyway.
> >
> > But, if a driver does not support ->setup_data_interface(),
> > it will not happen since nand_has_setup_data_iface() returns false.
>
> And I think this is the expected behavior. Calling again
> onfi_fill_data_interface() would probably enhance a bit the timings.
> The effect is that later exchanges with the NAND chip would be just a
> bit faster. But if you care about performance, then why not implementing
> ->setup_data_interface()? Even a dummy implementation would do the
> trick: only accept timing mode 0 without any changes on the controller
> side.


My driver (denali) does implement ->setup_data_interface().

When I was testing this thoroughly on my board,
I noticed the timing parameters were slightly changed
after nand_detect() detected ONFI chip.

> Unless you give me a use case where this is not possible, I don't think
> it is worth changing this path.

Only the use case I can come up with is when NAND_KEEP_TIMINGS was set.
But, it is just a matter of timeout values.

So, please throw away this patch.


>
> Thanks,
> MiquÃl
>
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/



-- 
Best Regards
Masahiro Yamada