[PATCH v5 0/2] MTK Smart Device Gen1 NAND Driver
Boris Brezillon
boris.brezillon at free-electrons.com
Mon Jun 20 04:54:11 PDT 2016
On Tue, 14 Jun 2016 11:50:49 -0400
Jorge Ramirez-Ortiz <jorge.ramirez-ortiz at linaro.org> wrote:
> Fixes/Changes on v5 relative to:
> --------------------------------
>
> tree : https://github.com/bbrezillon/linux-0day
> branch : nand/next
> commit :
> 'commit "mtd: nand: sunxi: fix return value check..."'
> Author: Wei Yongjun <yongjun_wei at trendmicro.com.cn>
> Date: Mon Jun 13 14:27:18 2016 +0000
>
> Patch v5
> --------
> mtd: mediatek: device tree bindings for MTK Smart Device Gen1 NAND
> - remove unused property
>
> mtd: mediatek: driver for MTK Smart Device Gen1 NAND
> - clean up macros and clarify semantics.
> - replace codec requests with operations.
> - replace semaphore with mutex on ecc accesses.
> - addtional comments to some functions.
> - explicit error for 16bits buswidth (not supported).
> - resolve endian issues.
> - determine active ecc codec by register inspection.
>
> Patch v4
> --------
> mtd: mediatek: device tree docs for MTK Smart Device Gen1 NAND
> - nand-on-flash-bbt, nand-ecc-mode, nand-ecc-step-size and
> nand-ecc-strength are now optional parameters
> - spare-per-sector removed from DT settings.
> - extend documentation.
>
> mtd: mediatek: driver for MTK Smart Device Gen1 NAND
> - rename functions (semantic issues).
> - add lock to ECC engine.
> - rework ECC interface (single enable/disable function).
> - rework bad block mark (now stored in OOB).
> - fix stat reporting.
> - rework FDM calculation.
> - handle optional DT settings (strength, mode, ecc step size).
> - spare per sector clamped to valid register values (hw spec).
>
> Patch v3
> --------
> mtd: mediatek: device tree docs for MTK Smart Device Gen1 NAND
> - rename nandc (nand controller)
>
> mtd: mediatek: driver for MTK Smart Device Gen1 NAND
> - rename source and header files
> - register definitions move to source files
> - ecc interface dropped
> - blank lines between functions (checkpatch.pl doesn't catch them)
> - ecc_check function renamed ecc_get_stats
> - ecc->dev assigned during probe()
> - generic ecc control function replaced by individual functions
> - required ecc parameters passed in ecc specific structures
> - fix memory corruption (size 0 array)
> - clarify nand offset/ptr function helpers (naming corrected)
> - move ecc config call to select_chip()
> - comment on hw issue workaround during read_byte
> - implement write_buf
> - remove switch_oob
> - fix suspend/resume operation
> - make nand-ecc-mode mandatory in device tree
>
> RFC v2
> ------
>
> mtd: mediatek: device tree docs for MTK Smart Device Gen1 NAND
> - split nand flash interface and ecc engine
>
> mtd: mediatek: driver for MTK Smart Device Gen1 NAND
> - split host and ecc in two separate drivers.
> - expose ecc layout through mtd_ooblayout_ops.
> - fix wait loops.
> - use device tree to describe ecc strength and ecc size.
> - use device tree to describe CS.
> - replace cmdfunc with cmd_ctrl.
> - fix read byte interface.
>
> dont enable controller in mt2701.dtsi
> - requires additional device tree changes. This will be done in
> the near future.
>
> Overview:
> =========
>
> The following patch-set adds support for Mediatek's Smart Device
> Gen1 NAND controller.
>
> * Controller expected layout:
> -----------------------------
>
> The controller expects the OOB information relevant to each sector
> within the page to follow immediately the data field.
>
> For example for a 4KiB page with 1KiB sectors and 224B of OOB information,
> the controller would expect the following layout.
>
> -----------------------------------------------------
> | 1KiB | 56B | 1KiB | 56B | 1KiB | 56B | 1KiB | 56B |
> -----------------------------------------------------
>
> This implies that the following actions are required:
>
> - raw writes:
> to match the controller expected layout, the driver uses the CPU and
> a private bounce buffer to reassemble the page before the actual
> writes.
>
> - non raw writes:
> ecc is hardware generated and data DMA'd in a single transfer.
>
> Bad Block mark
> ==============
> The bad block mark is stored in the OOB region (one byte)
>
> Controller limitations:
> =======================
>
> * dma chaning not supported:
> To guarantee that the pages corresponding to the buffers are contiguous
> in memory the driver enables NAND_USE_BOUNCE_BUFFER.
> * spare per sector: 16B to 128B
> * sector sizes: 512B or 1024B
> * programable page sizes: 512B, 2KB, 4KB, 8K, 16K
>
> Tests:
> ======
>
> * UBIFS support has been validated on 512MiB device (iozone, dd)
> * All drivers/mtd/tests/* pass clean.
> * Speed tests:
> eraseblock write speed: 7563 KiB/s
> eraseblock read speed: 10904 KiB/s
> page write speed: 7497 KiB/s
> page read speed: 10864 KiB/s
> 2 page write speed: 7527 KiB/s
> 2 page read speed: 10881 KiB/s
> erase speed: 103157 KiB/s
> 2x multi-block erase speed: 357445 KiB/s
> 4x multi-block erase speed: 362944 KiB/s
> 8x multi-block erase speed: 363805 KiB/s
> 16x multi-block erase speed: 364237 KiB/s
> 32x multi-block erase speed: 364670 KiB/s
> 64x multi-block erase speed: 365104 KiB/s
Applied after fixing a few coding style issues.
Thanks,
Boris
>
> Jorge Ramirez-Ortiz (2):
> mtd: mediatek: device tree bindings for MTK Smart Device Gen1 NAND
> mtd: mediatek: driver for MTK Smart Device Gen1 NAND
>
> Documentation/devicetree/bindings/mtd/mtk-nand.txt | 160 +++
> drivers/mtd/nand/Kconfig | 7 +
> drivers/mtd/nand/Makefile | 1 +
> drivers/mtd/nand/mtk_ecc.c | 528 +++++++
> drivers/mtd/nand/mtk_ecc.h | 50 +
> drivers/mtd/nand/mtk_nand.c | 1509 ++++++++++++++++++++
> 6 files changed, 2255 insertions(+)
> create mode 100644 Documentation/devicetree/bindings/mtd/mtk-nand.txt
> create mode 100644 drivers/mtd/nand/mtk_ecc.c
> create mode 100644 drivers/mtd/nand/mtk_ecc.h
> create mode 100644 drivers/mtd/nand/mtk_nand.c
>
More information about the linux-mtd
mailing list