Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Bean Huo <huobean@gmail.com>
To:     miquel.raynal@bootlin.com, richard@nod.at, vigneshr@ti.com,
        s.hauer@pengutronix.de, boris.brezillon@collabora.com,
        derosier@gmail.com
Cc:     linux-mtd@lists.infradead.org, linux-kernel@vger.kernel.org,
        huobean@gmail.com, Bean Huo <beanhuo@micron.com>
Subject: [PATCH v4 0/5] Micron SLC NAND filling block
Date:   Mon, 18 May 2020 15:59:38 +0200
Message-Id: <20200518135943.11749-1-huobean@gmail.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

From: Bean Huo <beanhuo@micron.com>

Hi,

on some legacy planar 2D Micron NAND devices when a block erase command is
issued, occasionally even though a block erase operation completes and returns
a pass status, the flash block may not be completely erased. Subsequent
operations to this block on very rare cases can result in subtle failures or
corruption. These extremely rare cases should nevertheless be considered. This
patchset is to address this potential issue.

After submission of patch V1 [1] and V2 [2], we stopped its update since we get
stuck in the solution on how to avoid the power-loss issue in case power-cut
hits the block filling. In the v1 and v2, to avoid this issue, we always damaged
page0, page1, this's based on the hypothesis that NAND FS is UBIFS. This
FS-specifical code is unacceptable in the MTD layer. Also, it cannot cover all
NAND based file system. Based on the current discussion, seems that re-write all
first 15 page from page0 is a satisfactory solution.

Meanwhile, I borrowed one idea from Miquel Raynal patchset [3], in which keeps
a recode of programmed pages, base on it, for most of the cases, we don't need
to read every page to see if current erasing block is a partially programmed
block.

Changelog:

v3 - v4:
    1. In the patch 4/5, change to directly use ecc.strength to judge the page
       is a empty page or not, rather than max_bitflips < mtd->bitflip_threshold
    2. In the patch 5/5, for the powerloss case, from the next time boot up,
       lots of page will be programmed from >page15 address, if still using
       first_p as GENMASK() bitmask starting position, writtenp will be always 0,
       fix it by changing its bitmask starting at bit position 0.

v2 - v3:
    1. Rebase patch to the latest MTD git tree
    2. Add a record that keeps tracking the programmed pages in the first 16
       pages
    3. Change from program odd pages, damage page 0 and page 1, to program all
       first 15 pages
    4. Address issues which exist in the V2.

v1 - v2:
    1. Rebased V1 to latest Linux kernel.
    2. Add erase preparation function pointer in nand_manufacturer_ops.


[1] https://www.spinics.net/lists/linux-mtd/msg04112.html
[2] https://www.spinics.net/lists/linux-mtd/msg04450.html
[3] https://www.spinics.net/lists/linux-mtd/msg13083.html


Bean Huo (5):
  mtd: rawnand: group all NAND specific ops into new nand_chip_ops
  mtd: rawnand: Add {pre,post}_erase hooks in nand_chip_ops
  mtd: rawnand: Add write_oob hook in nand_chip_ops
  mtd: rawnand: Introduce a new function nand_check_is_erased_page()
  mtd: rawnand: micron: Micron SLC NAND filling block

 drivers/mtd/nand/raw/internals.h     |   3 +-
 drivers/mtd/nand/raw/nand_base.c     |  88 +++++++++++++++++++----
 drivers/mtd/nand/raw/nand_hynix.c    |   2 +-
 drivers/mtd/nand/raw/nand_macronix.c |  10 +--
 drivers/mtd/nand/raw/nand_micron.c   | 104 ++++++++++++++++++++++++++-
 include/linux/mtd/rawnand.h          |  40 +++++++----
 6 files changed, 212 insertions(+), 35 deletions(-)

-- 
2.17.1