Re: [PATCH 02/11] mtd: nand_bbt: introduce BBT related data structure

[Boris Brezillon]

On Mon, 28 Mar 2016 16:09:44 +0800
Peter Pan <peterpansjtu@xxxxxxxxx> wrote:

> Hi Boris,
> 
> Firstly, thanks a lot for taking time to review my patches.
> 
> On Fri, Mar 25, 2016 at 4:35 PM, Boris Brezillon
> <boris.brezillon@xxxxxxxxxxxxxxxxxx> wrote:
> > Hi Peter,
> >
> > On Mon, 14 Mar 2016 02:47:55 +0000
> > Peter Pan <peterpansjtu@xxxxxxxxx> wrote:
> >
> >> From: Brian Norris <computersforpeace@xxxxxxxxx>
> >>
> >> Currently nand_bbt.c is tied with struct nand_chip, and it makes other
> >> NAND family chips hard to use nand_bbt.c. Maybe it's the reason why
> >> onenand has own bbt(onenand_bbt.c).
> >>
> >> Separate struct nand_chip from BBT code can make current BBT shareable.
> >> We create struct nand_bbt to take place of nand_chip in nand_bbt.c
> >>
> >> Below is mtd folder structure we want:
> >>       drivers/mtd/nand/<all-nand-core-code>
> >>       drivers/mtd/nand/raw/<raw-nand-controller-drivers>
> >>       drivers/mtd/nand/spi/<spi-nand-code>
> >>       drivers/mtd/nand/onenand/<onenand-code>
> >>       drivers/mtd/nand/chips/<manufacturer-spcific-code>
> >>
> >> Of course, nand_bbt.c should be part of <all-nand-core-code>.
> >>
> >> We put every chip layout related information BBT needed into struct
> >> nand_chip_layout_info.
> >>       @numchips:      number of physical chips, required for NAND_BBT_PERCHIP
> >>       @chipsize:      the size of one chip for multichip arrays
> >>       @chip_shift:    number of address bits in one chip
> >>       @bbt_erase_shift:       number of address bits in a bbt entry
> >>       @page_shift:    number of address bits in a page
> >>
> >> We defined a struct nand_bbt_ops for BBT ops. Struct
> >>       @is_bad_bbm:    check if a block is factory bad block
> >>       @erase: erase block bypassing resvered checks
> >>
> >> Struct nand_bbt includes all BBT information:
> >>       @mtd:   pointer to MTD device structure
> >>       @bbt_options:   bad block specific options. All options used
> >>                       here must come from nand_bbt.h.
> >>       @bbt_ops:       struct nand_bbt_ops pointer.
> >>       @info:          struct nand_chip_layout_info pointer.
> >>       @bbt_td:        bad block table descriptor for flash lookup.
> >>       @bbt_md:        bad block table mirror descriptor
> >>       @bbt:           bad block table pointer
> >>
> >> Signed-off-by: Brian Norris <computersforpeace@xxxxxxxxx>
> >> [Peter: 1. correct comment style
> >>       2. introduce struct nand_bbt_ops and nand_chip_layout_info]
> >> Signed-off-by: Peter Pan <peterpandong@xxxxxxxxxx>
> >> ---
> >>  include/linux/mtd/nand_bbt.h | 67 ++++++++++++++++++++++++++++++++++++++++++++
> >>  1 file changed, 67 insertions(+)
> >>
> >> diff --git a/include/linux/mtd/nand_bbt.h b/include/linux/mtd/nand_bbt.h
> >> index 5a65230..cfb22c8 100644
> >> --- a/include/linux/mtd/nand_bbt.h
> >> +++ b/include/linux/mtd/nand_bbt.h
> >> @@ -18,6 +18,8 @@
> >>  #ifndef __LINUX_MTD_NAND_BBT_H
> >>  #define __LINUX_MTD_NAND_BBT_H
> >>
> >> +struct mtd_info;
> >> +
> >>  /* The maximum number of NAND chips in an array */
> >>  #define NAND_MAX_CHIPS               8
> >>
> >> @@ -115,4 +117,69 @@ struct nand_bbt_descr {
> >>  /* The maximum number of blocks to scan for a bbt */
> >>  #define NAND_BBT_SCAN_MAXBLOCKS      4
> >>
> >> +struct nand_bbt;
> >> +
> >> +/**
> >> + * struct nand_bbt_ops - bad block table operations
> >> + * @is_bad_bbm:      check if a block is factory bad block
> >> + * @erase:   erase block bypassing resvered checks
> >> + */
> >> +struct nand_bbt_ops {
> >> +     /*
> >> +      * This is important to abstract out of nand_bbt.c and provide
> >> +      * separately in nand_base.c and spi-nand-base.c -- it's sort of
> >> +      * duplicated in nand_block_bad() (nand_base) and
> >> +      * scan_block_fast() (nand_bbt) right now
> >> +      *
> >> +      * Note that this also means nand_chip.badblock_pattern should
> >> +      * be removed from nand_bbt.c
> >> +      */
> >> +     int (*is_bad_bbm)(struct mtd_info *mtd, loff_t ofs);
> >> +
> >> +     /* Erase a block, bypassing reserved checks */
> >> +     int (*erase)(struct mtd_info *mtd, loff_t ofs);
> >> +};
> >> +
> >> +/**
> >> + * struct nand_chip_layout_info - strucure contains all chip layout
> >> + * information that BBT needed.
> >> + * @numchips:        number of physical chips, required for NAND_BBT_PERCHIP
> >> + * @chipsize:        the size of one chip for multichip arrays
> >> + * @chip_shift:      number of address bits in one chip
> >> + * @bbt_erase_shift: number of address bits in a bbt entry
> >> + * @page_shift:      number of address bits in a page
> >> + */
> >> +struct nand_chip_layout_info {
> >
> > I know I'm the one who suggested this name, but NAND datasheet seems to
> > call it "memory organization", so maybe we should rename this struct
> > nand_memory_organization.
> 
> Fix this in v4
> >
> >> +     int numchips;
> >
> > I would rename it numdies, or ndies. numchips implies you're having
> > several chips, which is not the case.
> 
> Fix this in v4
> >
> >> +     u64 chipsize;
> >
> > Ditto, s/chipsize/diesize/
> 
> Fix this in v4
> >
> >> +     int chip_shift;
> >
> > Ditto.
> 
> Fix this in v4
> >
> >> +     int bbt_erase_shift;
> >
> > Hm, this is not related to the memory organization. I'd prefer moving
> > this one directly in
> 
> Yes, I also realize bbt_erase_shift is not proper. How about just rename it
> to erase_block_shift or block_shift ?

eraseblock_shift sounds good as long as
bbt_erase_shift == eraseblock_shift is always true.

> 
> >
> >> +     int page_shift;
> >> +};
> >
> > The structure should probably contain other info like (oob size, pages
> > per block, blocks per die, ...)
> > I know some of those information are redundant with mtd_info content,
> > but it would be clearer to have everything in a common place.
> >
> > Also, I'd recommend using helpers to access memory organization info.
> > For example nand_get_die_size(mtd), nand_get_page_size(mtd), ...
> >
> > On a more general note, as already said, I'd like to see more
> > generalization across NAND based devices, no matter the interface
> > they're using.
> > Doing that implies forcing all NAND based devices to inherit from a
> > common class. Something like
> >
> > struct nand_device {
> >         struct mtd_info mtd;
> >         struct nand_memory_organization memorg;
> >         /* ... */
> > };
> >
> > /* rawnand_device <-> nand_chip */
> > struct rawnand_device {
> >         struct nand_device base;
> >         /* raw NAND specific fields */
> > }
> >
> > struct spinand_device {
> >         struct nand_device base;
> >         /* SPI NAND specific fields */
> > };
> >
> > struct onenand_device {
> >         struct nand_device base;
> >         /* OneNAND specific fields */
> > };
> >
> > With this design, nand_bbt and nand_bbt_ops could use the generic
> > nand_device instead of directly using the mtd instance.
> >
> > Anyway, that's just a long term goal, and I wanted to share my
> > ideas. I guess your plan is to add support for SPI nand devices, so
> > keep this in mind ;-).
> 
> Acctually your idea is quite good. Actually, struct nand_chip_layout_info
> shouldn't be in nand_bbt.h. It should be in nand.h or nand_base.h and embedded
> in struct nand_chip (or struct nand_deivce as your said).

Yes, I didn't comment on that since I don't want to create a
nand_base.h header file. The idea is to rename nand.h into rawnand.h
and then create a nand.h file containing all interface-independent
stuff (like memory organization info).

> The reason I did't do this is I feel it will be too involved. I need
> to change almost
> all files under mtd/nand/, which generates a larger patch set.

Yes, I know, that's why I'm not asking that right now. But that would
be great if we could prepare things for this move...


-- 
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com