Re: [PATCH RFC 2/3] mtd: rawnand: Add support Macronix Block Protection function

[Boris Brezillon <boris.brezillon@collabora.com>]

On Mon, 21 Oct 2019 16:40:57 +0800
masonccyang@mxic.com.tw wrote:

> Hi Boris,
> 
>  
> > > > > > > Then fill-in these two hooks from the manufacturer code,   
> without 
> > > the  
> > > > > > > postponed init.
> > > > > > >   
> > > > > > 
> > > > > > But in the final of nand_scan_tail(), mtd->_lock/_unlock will be
> > > > > > filled by NULL, right ?   
> > > > > 
> > > > > The NAND core should set mtd->_lock/_unlock() to NAND specific   
> hooks 
> > > so  
> > > > > that the MTD layer is abstracted and and drivers do not see it.   
> Then,
> > > > > in the NAND helper, either there is no specific hook defined by a
> > > > > manufacturer driver and you return -ENOTSUPP, or you execute the
> > > > > defined hook.   
> > > > 
> > > > okay, patch specific manufacturer _lock/_unlock driver
> > > > in nand_manufacturer_init();
> > > > 
> > > > and in the final of nand_scan_tail()
> > > > if (!mtd->_lock)
> > > >  mtd->_lock = NULL;
> > > > if (!mtd->_unlock)
> > > >  mtd->_unlock = NULL;   
> > > 
> > > 
> > > I'm still considering of post_init() in nand_scan_tail() for
> > > MTD layer default call-back function replacement because
> > > there would be more call-back functions need it.
> > > i.e., 
> > > MTD->_lock/_unlokc
> > > MTD->_suspend/_resume  
> > 
> > Again, that's something that needs to be abstracted so that both the
> > NAND manufacturer driver and the NAND controller driver can take
> > appropriate actions on suspend/resume operations.
> >   
> > > NTD->_point/_unpoint  
> >   
> > ->_point/_unpoint() are irrelevant for a NAND chip.  
> >   
> > > ...
> > > 
> > > 
> > > actually, my patch series are including MTD->_locl/_unlock and 
> > > MTD->_suspend/_resume. how do you think ?  
> > 
> > Miquel was suggesting to add nand_chip->{lock,unlock,is_locked}()
> > methods that would be implemented by the NAND manufacturer drivers, and
> > have generic wrappers implemented in nand_base.c:
> > 
> > static int nand_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
> > {
> >    struct nand_chip *chip = mtd_to_nand(mtd);
> > 
> >    if (!chip->lock)
> >       return -ENOTSUPP;
> > 
> >    return chip->lock(chip, ofs, len);
> > }
> > 
> > ...
> > 
> > If you do that, you won't need this post_init() hook.  
> 
> got it, but ... 
> user space program flash_lock/flash_unlock are calling 
> mtd_lock() & mtd_unlock().
> i.e.,
> int mtd_lock(struct mtd_info *mtd, loff_t ofs, uint64_t len)
> {
>          if (!mtd->_lock)
>                  return -EOPNOTSUPP;
>          if (ofs < 0 || ofs >= mtd->size || len > mtd->size - ofs)
>                  return -EINVAL;
>          if (!len)
>                  return 0;
>          return mtd->_lock(mtd, ofs, len);
> }
> 

Assign mtd lock/unlock/is_locked hooks to the generic wrappers in
nand_scan_tail():

	mtd->_lock = nand_lock;
	mtd->_unlock = nand_unlock;
	mtd->_is_locked = nand_is_locked;

Seriously, we've almost implemented the thing for you with all the
details we've given. At some point you have to look more closely at how
things are done/designed in the NAND framework if you want to
contribute core changes. I'm fine giving hints but we're far beyond
that point here.

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