spi: add driver for MTK SPI NAND Flash Interface
authorChuanhong Guo <gch981213@gmail.com>
Sun, 24 Apr 2022 03:25:24 +0000 (11:25 +0800)
committerMark Brown <broonie@kernel.org>
Wed, 27 Apr 2022 16:22:53 +0000 (17:22 +0100)
This driver implements support for the SPI-NAND mode of MTK NAND Flash
Interface as a SPI-MEM controller with pipelined ECC capability.

Signed-off-by: Chuanhong Guo <gch981213@gmail.com>
Tested-by: Daniel Golle <daniel@makrotopia.org>
Link: https://lore.kernel.org/r/20220424032527.673605-3-gch981213@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
drivers/spi/Kconfig
drivers/spi/Makefile
drivers/spi/spi-mtk-snfi.c [new file with mode: 0644]

index cca92a8..9f09aff 100644 (file)
@@ -590,6 +590,16 @@ config SPI_MTK_NOR
          SPI interface as well as several SPI NOR specific instructions
          via SPI MEM interface.
 
+config SPI_MTK_SNFI
+       tristate "MediaTek SPI NAND Flash Interface"
+       depends on ARCH_MEDIATEK || COMPILE_TEST
+       depends on MTD_NAND_ECC_MEDIATEK
+       help
+         This enables support for SPI-NAND mode on the MediaTek NAND
+         Flash Interface found on MediaTek ARM SoCs. This controller
+         is implemented as a SPI-MEM controller with pipelined ECC
+         capcability.
+
 config SPI_NPCM_FIU
        tristate "Nuvoton NPCM FLASH Interface Unit"
        depends on ARCH_NPCM || COMPILE_TEST
index 3aa28ed..51541ff 100644 (file)
@@ -76,6 +76,7 @@ obj-$(CONFIG_SPI_MPC52xx)             += spi-mpc52xx.o
 obj-$(CONFIG_SPI_MT65XX)                += spi-mt65xx.o
 obj-$(CONFIG_SPI_MT7621)               += spi-mt7621.o
 obj-$(CONFIG_SPI_MTK_NOR)              += spi-mtk-nor.o
+obj-$(CONFIG_SPI_MTK_SNFI)             += spi-mtk-snfi.o
 obj-$(CONFIG_SPI_MXIC)                 += spi-mxic.o
 obj-$(CONFIG_SPI_MXS)                  += spi-mxs.o
 obj-$(CONFIG_SPI_NPCM_FIU)             += spi-npcm-fiu.o
diff --git a/drivers/spi/spi-mtk-snfi.c b/drivers/spi/spi-mtk-snfi.c
new file mode 100644 (file)
index 0000000..2c556e3
--- /dev/null
@@ -0,0 +1,1470 @@
+// SPDX-License-Identifier: GPL-2.0
+//
+// Driver for the SPI-NAND mode of Mediatek NAND Flash Interface
+//
+// Copyright (c) 2022 Chuanhong Guo <gch981213@gmail.com>
+//
+// This driver is based on the SPI-NAND mtd driver from Mediatek SDK:
+//
+// Copyright (C) 2020 MediaTek Inc.
+// Author: Weijie Gao <weijie.gao@mediatek.com>
+//
+// This controller organize the page data as several interleaved sectors
+// like the following: (sizeof(FDM + ECC) = snf->nfi_cfg.spare_size)
+// +---------+------+------+---------+------+------+-----+
+// | Sector1 | FDM1 | ECC1 | Sector2 | FDM2 | ECC2 | ... |
+// +---------+------+------+---------+------+------+-----+
+// With auto-format turned on, DMA only returns this part:
+// +---------+---------+-----+
+// | Sector1 | Sector2 | ... |
+// +---------+---------+-----+
+// The FDM data will be filled to the registers, and ECC parity data isn't
+// accessible.
+// With auto-format off, all ((Sector+FDM+ECC)*nsectors) will be read over DMA
+// in it's original order shown in the first table. ECC can't be turned on when
+// auto-format is off.
+//
+// However, Linux SPI-NAND driver expects the data returned as:
+// +------+-----+
+// | Page | OOB |
+// +------+-----+
+// where the page data is continuously stored instead of interleaved.
+// So we assume all instructions matching the page_op template between ECC
+// prepare_io_req and finish_io_req are for page cache r/w.
+// Here's how this spi-mem driver operates when reading:
+//  1. Always set snf->autofmt = true in prepare_io_req (even when ECC is off).
+//  2. Perform page ops and let the controller fill the DMA bounce buffer with
+//     de-interleaved sector data and set FDM registers.
+//  3. Return the data as:
+//     +---------+---------+-----+------+------+-----+
+//     | Sector1 | Sector2 | ... | FDM1 | FDM2 | ... |
+//     +---------+---------+-----+------+------+-----+
+//  4. For other matching spi_mem ops outside a prepare/finish_io_req pair,
+//     read the data with auto-format off into the bounce buffer and copy
+//     needed data to the buffer specified in the request.
+//
+// Write requests operates in a similar manner.
+// As a limitation of this strategy, we won't be able to access any ECC parity
+// data at all in Linux.
+//
+// Here's the bad block mark situation on MTK chips:
+// In older chips like mt7622, MTK uses the first FDM byte in the first sector
+// as the bad block mark. After de-interleaving, this byte appears at [pagesize]
+// in the returned data, which is the BBM position expected by kernel. However,
+// the conventional bad block mark is the first byte of the OOB, which is part
+// of the last sector data in the interleaved layout. Instead of fixing their
+// hardware, MTK decided to address this inconsistency in software. On these
+// later chips, the BootROM expects the following:
+// 1. The [pagesize] byte on a nand page is used as BBM, which will appear at
+//    (page_size - (nsectors - 1) * spare_size) in the DMA buffer.
+// 2. The original byte stored at that position in the DMA buffer will be stored
+//    as the first byte of the FDM section in the last sector.
+// We can't disagree with the BootROM, so after de-interleaving, we need to
+// perform the following swaps in read:
+// 1. Store the BBM at [page_size - (nsectors - 1) * spare_size] to [page_size],
+//    which is the expected BBM position by kernel.
+// 2. Store the page data byte at [pagesize + (nsectors-1) * fdm] back to
+//    [page_size - (nsectors - 1) * spare_size]
+// Similarly, when writing, we need to perform swaps in the other direction.
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/device.h>
+#include <linux/mutex.h>
+#include <linux/clk.h>
+#include <linux/interrupt.h>
+#include <linux/dma-mapping.h>
+#include <linux/iopoll.h>
+#include <linux/of_platform.h>
+#include <linux/mtd/nand-ecc-mtk.h>
+#include <linux/spi/spi.h>
+#include <linux/spi/spi-mem.h>
+#include <linux/mtd/nand.h>
+
+// NFI registers
+#define NFI_CNFG 0x000
+#define CNFG_OP_MODE_S 12
+#define CNFG_OP_MODE_CUST 6
+#define CNFG_OP_MODE_PROGRAM 3
+#define CNFG_AUTO_FMT_EN BIT(9)
+#define CNFG_HW_ECC_EN BIT(8)
+#define CNFG_DMA_BURST_EN BIT(2)
+#define CNFG_READ_MODE BIT(1)
+#define CNFG_DMA_MODE BIT(0)
+
+#define NFI_PAGEFMT 0x0004
+#define NFI_SPARE_SIZE_LS_S 16
+#define NFI_FDM_ECC_NUM_S 12
+#define NFI_FDM_NUM_S 8
+#define NFI_SPARE_SIZE_S 4
+#define NFI_SEC_SEL_512 BIT(2)
+#define NFI_PAGE_SIZE_S 0
+#define NFI_PAGE_SIZE_512_2K 0
+#define NFI_PAGE_SIZE_2K_4K 1
+#define NFI_PAGE_SIZE_4K_8K 2
+#define NFI_PAGE_SIZE_8K_16K 3
+
+#define NFI_CON 0x008
+#define CON_SEC_NUM_S 12
+#define CON_BWR BIT(9)
+#define CON_BRD BIT(8)
+#define CON_NFI_RST BIT(1)
+#define CON_FIFO_FLUSH BIT(0)
+
+#define NFI_INTR_EN 0x010
+#define NFI_INTR_STA 0x014
+#define NFI_IRQ_INTR_EN BIT(31)
+#define NFI_IRQ_CUS_READ BIT(8)
+#define NFI_IRQ_CUS_PG BIT(7)
+
+#define NFI_CMD 0x020
+#define NFI_CMD_DUMMY_READ 0x00
+#define NFI_CMD_DUMMY_WRITE 0x80
+
+#define NFI_STRDATA 0x040
+#define STR_DATA BIT(0)
+
+#define NFI_STA 0x060
+#define NFI_NAND_FSM GENMASK(28, 24)
+#define NFI_FSM GENMASK(19, 16)
+#define READ_EMPTY BIT(12)
+
+#define NFI_FIFOSTA 0x064
+#define FIFO_WR_REMAIN_S 8
+#define FIFO_RD_REMAIN_S 0
+
+#define NFI_ADDRCNTR 0x070
+#define SEC_CNTR GENMASK(16, 12)
+#define SEC_CNTR_S 12
+#define NFI_SEC_CNTR(val) (((val)&SEC_CNTR) >> SEC_CNTR_S)
+
+#define NFI_STRADDR 0x080
+
+#define NFI_BYTELEN 0x084
+#define BUS_SEC_CNTR(val) (((val)&SEC_CNTR) >> SEC_CNTR_S)
+
+#define NFI_FDM0L 0x0a0
+#define NFI_FDM0M 0x0a4
+#define NFI_FDML(n) (NFI_FDM0L + (n)*8)
+#define NFI_FDMM(n) (NFI_FDM0M + (n)*8)
+
+#define NFI_DEBUG_CON1 0x220
+#define WBUF_EN BIT(2)
+
+#define NFI_MASTERSTA 0x224
+#define MAS_ADDR GENMASK(11, 9)
+#define MAS_RD GENMASK(8, 6)
+#define MAS_WR GENMASK(5, 3)
+#define MAS_RDDLY GENMASK(2, 0)
+#define NFI_MASTERSTA_MASK_7622 (MAS_ADDR | MAS_RD | MAS_WR | MAS_RDDLY)
+
+// SNFI registers
+#define SNF_MAC_CTL 0x500
+#define MAC_XIO_SEL BIT(4)
+#define SF_MAC_EN BIT(3)
+#define SF_TRIG BIT(2)
+#define WIP_READY BIT(1)
+#define WIP BIT(0)
+
+#define SNF_MAC_OUTL 0x504
+#define SNF_MAC_INL 0x508
+
+#define SNF_RD_CTL2 0x510
+#define DATA_READ_DUMMY_S 8
+#define DATA_READ_MAX_DUMMY 0xf
+#define DATA_READ_CMD_S 0
+
+#define SNF_RD_CTL3 0x514
+
+#define SNF_PG_CTL1 0x524
+#define PG_LOAD_CMD_S 8
+
+#define SNF_PG_CTL2 0x528
+
+#define SNF_MISC_CTL 0x538
+#define SW_RST BIT(28)
+#define FIFO_RD_LTC_S 25
+#define PG_LOAD_X4_EN BIT(20)
+#define DATA_READ_MODE_S 16
+#define DATA_READ_MODE GENMASK(18, 16)
+#define DATA_READ_MODE_X1 0
+#define DATA_READ_MODE_X2 1
+#define DATA_READ_MODE_X4 2
+#define DATA_READ_MODE_DUAL 5
+#define DATA_READ_MODE_QUAD 6
+#define PG_LOAD_CUSTOM_EN BIT(7)
+#define DATARD_CUSTOM_EN BIT(6)
+#define CS_DESELECT_CYC_S 0
+
+#define SNF_MISC_CTL2 0x53c
+#define PROGRAM_LOAD_BYTE_NUM_S 16
+#define READ_DATA_BYTE_NUM_S 11
+
+#define SNF_DLY_CTL3 0x548
+#define SFCK_SAM_DLY_S 0
+
+#define SNF_STA_CTL1 0x550
+#define CUS_PG_DONE BIT(28)
+#define CUS_READ_DONE BIT(27)
+#define SPI_STATE_S 0
+#define SPI_STATE GENMASK(3, 0)
+
+#define SNF_CFG 0x55c
+#define SPI_MODE BIT(0)
+
+#define SNF_GPRAM 0x800
+#define SNF_GPRAM_SIZE 0xa0
+
+#define SNFI_POLL_INTERVAL 1000000
+
+static const u8 mt7622_spare_sizes[] = { 16, 26, 27, 28 };
+
+struct mtk_snand_caps {
+       u16 sector_size;
+       u16 max_sectors;
+       u16 fdm_size;
+       u16 fdm_ecc_size;
+       u16 fifo_size;
+
+       bool bbm_swap;
+       bool empty_page_check;
+       u32 mastersta_mask;
+
+       const u8 *spare_sizes;
+       u32 num_spare_size;
+};
+
+static const struct mtk_snand_caps mt7622_snand_caps = {
+       .sector_size = 512,
+       .max_sectors = 8,
+       .fdm_size = 8,
+       .fdm_ecc_size = 1,
+       .fifo_size = 32,
+       .bbm_swap = false,
+       .empty_page_check = false,
+       .mastersta_mask = NFI_MASTERSTA_MASK_7622,
+       .spare_sizes = mt7622_spare_sizes,
+       .num_spare_size = ARRAY_SIZE(mt7622_spare_sizes)
+};
+
+static const struct mtk_snand_caps mt7629_snand_caps = {
+       .sector_size = 512,
+       .max_sectors = 8,
+       .fdm_size = 8,
+       .fdm_ecc_size = 1,
+       .fifo_size = 32,
+       .bbm_swap = true,
+       .empty_page_check = false,
+       .mastersta_mask = NFI_MASTERSTA_MASK_7622,
+       .spare_sizes = mt7622_spare_sizes,
+       .num_spare_size = ARRAY_SIZE(mt7622_spare_sizes)
+};
+
+struct mtk_snand_conf {
+       size_t page_size;
+       size_t oob_size;
+       u8 nsectors;
+       u8 spare_size;
+};
+
+struct mtk_snand {
+       struct spi_controller *ctlr;
+       struct device *dev;
+       struct clk *nfi_clk;
+       struct clk *pad_clk;
+       void __iomem *nfi_base;
+       int irq;
+       struct completion op_done;
+       const struct mtk_snand_caps *caps;
+       struct mtk_ecc_config *ecc_cfg;
+       struct mtk_ecc *ecc;
+       struct mtk_snand_conf nfi_cfg;
+       struct mtk_ecc_stats ecc_stats;
+       struct nand_ecc_engine ecc_eng;
+       bool autofmt;
+       u8 *buf;
+       size_t buf_len;
+};
+
+static struct mtk_snand *nand_to_mtk_snand(struct nand_device *nand)
+{
+       struct nand_ecc_engine *eng = nand->ecc.engine;
+
+       return container_of(eng, struct mtk_snand, ecc_eng);
+}
+
+static inline int snand_prepare_bouncebuf(struct mtk_snand *snf, size_t size)
+{
+       if (snf->buf_len >= size)
+               return 0;
+       kfree(snf->buf);
+       snf->buf = kmalloc(size, GFP_KERNEL);
+       if (!snf->buf)
+               return -ENOMEM;
+       snf->buf_len = size;
+       memset(snf->buf, 0xff, snf->buf_len);
+       return 0;
+}
+
+static inline u32 nfi_read32(struct mtk_snand *snf, u32 reg)
+{
+       return readl(snf->nfi_base + reg);
+}
+
+static inline void nfi_write32(struct mtk_snand *snf, u32 reg, u32 val)
+{
+       writel(val, snf->nfi_base + reg);
+}
+
+static inline void nfi_write16(struct mtk_snand *snf, u32 reg, u16 val)
+{
+       writew(val, snf->nfi_base + reg);
+}
+
+static inline void nfi_rmw32(struct mtk_snand *snf, u32 reg, u32 clr, u32 set)
+{
+       u32 val;
+
+       val = readl(snf->nfi_base + reg);
+       val &= ~clr;
+       val |= set;
+       writel(val, snf->nfi_base + reg);
+}
+
+static void nfi_read_data(struct mtk_snand *snf, u32 reg, u8 *data, u32 len)
+{
+       u32 i, val = 0, es = sizeof(u32);
+
+       for (i = reg; i < reg + len; i++) {
+               if (i == reg || i % es == 0)
+                       val = nfi_read32(snf, i & ~(es - 1));
+
+               *data++ = (u8)(val >> (8 * (i % es)));
+       }
+}
+
+static int mtk_nfi_reset(struct mtk_snand *snf)
+{
+       u32 val, fifo_mask;
+       int ret;
+
+       nfi_write32(snf, NFI_CON, CON_FIFO_FLUSH | CON_NFI_RST);
+
+       ret = readw_poll_timeout(snf->nfi_base + NFI_MASTERSTA, val,
+                                !(val & snf->caps->mastersta_mask), 0,
+                                SNFI_POLL_INTERVAL);
+       if (ret) {
+               dev_err(snf->dev, "NFI master is still busy after reset\n");
+               return ret;
+       }
+
+       ret = readl_poll_timeout(snf->nfi_base + NFI_STA, val,
+                                !(val & (NFI_FSM | NFI_NAND_FSM)), 0,
+                                SNFI_POLL_INTERVAL);
+       if (ret) {
+               dev_err(snf->dev, "Failed to reset NFI\n");
+               return ret;
+       }
+
+       fifo_mask = ((snf->caps->fifo_size - 1) << FIFO_RD_REMAIN_S) |
+                   ((snf->caps->fifo_size - 1) << FIFO_WR_REMAIN_S);
+       ret = readw_poll_timeout(snf->nfi_base + NFI_FIFOSTA, val,
+                                !(val & fifo_mask), 0, SNFI_POLL_INTERVAL);
+       if (ret) {
+               dev_err(snf->dev, "NFI FIFOs are not empty\n");
+               return ret;
+       }
+
+       return 0;
+}
+
+static int mtk_snand_mac_reset(struct mtk_snand *snf)
+{
+       int ret;
+       u32 val;
+
+       nfi_rmw32(snf, SNF_MISC_CTL, 0, SW_RST);
+
+       ret = readl_poll_timeout(snf->nfi_base + SNF_STA_CTL1, val,
+                                !(val & SPI_STATE), 0, SNFI_POLL_INTERVAL);
+       if (ret)
+               dev_err(snf->dev, "Failed to reset SNFI MAC\n");
+
+       nfi_write32(snf, SNF_MISC_CTL,
+                   (2 << FIFO_RD_LTC_S) | (10 << CS_DESELECT_CYC_S));
+
+       return ret;
+}
+
+static int mtk_snand_mac_trigger(struct mtk_snand *snf, u32 outlen, u32 inlen)
+{
+       int ret;
+       u32 val;
+
+       nfi_write32(snf, SNF_MAC_CTL, SF_MAC_EN);
+       nfi_write32(snf, SNF_MAC_OUTL, outlen);
+       nfi_write32(snf, SNF_MAC_INL, inlen);
+
+       nfi_write32(snf, SNF_MAC_CTL, SF_MAC_EN | SF_TRIG);
+
+       ret = readl_poll_timeout(snf->nfi_base + SNF_MAC_CTL, val,
+                                val & WIP_READY, 0, SNFI_POLL_INTERVAL);
+       if (ret) {
+               dev_err(snf->dev, "Timed out waiting for WIP_READY\n");
+               goto cleanup;
+       }
+
+       ret = readl_poll_timeout(snf->nfi_base + SNF_MAC_CTL, val, !(val & WIP),
+                                0, SNFI_POLL_INTERVAL);
+       if (ret)
+               dev_err(snf->dev, "Timed out waiting for WIP cleared\n");
+
+cleanup:
+       nfi_write32(snf, SNF_MAC_CTL, 0);
+
+       return ret;
+}
+
+static int mtk_snand_mac_io(struct mtk_snand *snf, const struct spi_mem_op *op)
+{
+       u32 rx_len = 0;
+       u32 reg_offs = 0;
+       u32 val = 0;
+       const u8 *tx_buf = NULL;
+       u8 *rx_buf = NULL;
+       int i, ret;
+       u8 b;
+
+       if (op->data.dir == SPI_MEM_DATA_IN) {
+               rx_len = op->data.nbytes;
+               rx_buf = op->data.buf.in;
+       } else {
+               tx_buf = op->data.buf.out;
+       }
+
+       mtk_snand_mac_reset(snf);
+
+       for (i = 0; i < op->cmd.nbytes; i++, reg_offs++) {
+               b = (op->cmd.opcode >> ((op->cmd.nbytes - i - 1) * 8)) & 0xff;
+               val |= b << (8 * (reg_offs % 4));
+               if (reg_offs % 4 == 3) {
+                       nfi_write32(snf, SNF_GPRAM + reg_offs - 3, val);
+                       val = 0;
+               }
+       }
+
+       for (i = 0; i < op->addr.nbytes; i++, reg_offs++) {
+               b = (op->addr.val >> ((op->addr.nbytes - i - 1) * 8)) & 0xff;
+               val |= b << (8 * (reg_offs % 4));
+               if (reg_offs % 4 == 3) {
+                       nfi_write32(snf, SNF_GPRAM + reg_offs - 3, val);
+                       val = 0;
+               }
+       }
+
+       for (i = 0; i < op->dummy.nbytes; i++, reg_offs++) {
+               if (reg_offs % 4 == 3) {
+                       nfi_write32(snf, SNF_GPRAM + reg_offs - 3, val);
+                       val = 0;
+               }
+       }
+
+       if (op->data.dir == SPI_MEM_DATA_OUT) {
+               for (i = 0; i < op->data.nbytes; i++, reg_offs++) {
+                       val |= tx_buf[i] << (8 * (reg_offs % 4));
+                       if (reg_offs % 4 == 3) {
+                               nfi_write32(snf, SNF_GPRAM + reg_offs - 3, val);
+                               val = 0;
+                       }
+               }
+       }
+
+       if (reg_offs % 4)
+               nfi_write32(snf, SNF_GPRAM + (reg_offs & ~3), val);
+
+       for (i = 0; i < reg_offs; i += 4)
+               dev_dbg(snf->dev, "%d: %08X", i,
+                       nfi_read32(snf, SNF_GPRAM + i));
+
+       dev_dbg(snf->dev, "SNF TX: %u RX: %u", reg_offs, rx_len);
+
+       ret = mtk_snand_mac_trigger(snf, reg_offs, rx_len);
+       if (ret)
+               return ret;
+
+       if (!rx_len)
+               return 0;
+
+       nfi_read_data(snf, SNF_GPRAM + reg_offs, rx_buf, rx_len);
+       return 0;
+}
+
+static int mtk_snand_setup_pagefmt(struct mtk_snand *snf, u32 page_size,
+                                  u32 oob_size)
+{
+       int spare_idx = -1;
+       u32 spare_size, spare_size_shift, pagesize_idx;
+       u32 sector_size_512;
+       u8 nsectors;
+       int i;
+
+       // skip if it's already configured as required.
+       if (snf->nfi_cfg.page_size == page_size &&
+           snf->nfi_cfg.oob_size == oob_size)
+               return 0;
+
+       nsectors = page_size / snf->caps->sector_size;
+       if (nsectors > snf->caps->max_sectors) {
+               dev_err(snf->dev, "too many sectors required.\n");
+               goto err;
+       }
+
+       if (snf->caps->sector_size == 512) {
+               sector_size_512 = NFI_SEC_SEL_512;
+               spare_size_shift = NFI_SPARE_SIZE_S;
+       } else {
+               sector_size_512 = 0;
+               spare_size_shift = NFI_SPARE_SIZE_LS_S;
+       }
+
+       switch (page_size) {
+       case SZ_512:
+               pagesize_idx = NFI_PAGE_SIZE_512_2K;
+               break;
+       case SZ_2K:
+               if (snf->caps->sector_size == 512)
+                       pagesize_idx = NFI_PAGE_SIZE_2K_4K;
+               else
+                       pagesize_idx = NFI_PAGE_SIZE_512_2K;
+               break;
+       case SZ_4K:
+               if (snf->caps->sector_size == 512)
+                       pagesize_idx = NFI_PAGE_SIZE_4K_8K;
+               else
+                       pagesize_idx = NFI_PAGE_SIZE_2K_4K;
+               break;
+       case SZ_8K:
+               if (snf->caps->sector_size == 512)
+                       pagesize_idx = NFI_PAGE_SIZE_8K_16K;
+               else
+                       pagesize_idx = NFI_PAGE_SIZE_4K_8K;
+               break;
+       case SZ_16K:
+               pagesize_idx = NFI_PAGE_SIZE_8K_16K;
+               break;
+       default:
+               dev_err(snf->dev, "unsupported page size.\n");
+               goto err;
+       }
+
+       spare_size = oob_size / nsectors;
+       // If we're using the 1KB sector size, HW will automatically double the
+       // spare size. We should only use half of the value in this case.
+       if (snf->caps->sector_size == 1024)
+               spare_size /= 2;
+
+       for (i = snf->caps->num_spare_size - 1; i >= 0; i--) {
+               if (snf->caps->spare_sizes[i] <= spare_size) {
+                       spare_size = snf->caps->spare_sizes[i];
+                       if (snf->caps->sector_size == 1024)
+                               spare_size *= 2;
+                       spare_idx = i;
+                       break;
+               }
+       }
+
+       if (spare_idx < 0) {
+               dev_err(snf->dev, "unsupported spare size: %u\n", spare_size);
+               goto err;
+       }
+
+       nfi_write32(snf, NFI_PAGEFMT,
+                   (snf->caps->fdm_ecc_size << NFI_FDM_ECC_NUM_S) |
+                           (snf->caps->fdm_size << NFI_FDM_NUM_S) |
+                           (spare_idx << spare_size_shift) |
+                           (pagesize_idx << NFI_PAGE_SIZE_S) |
+                           sector_size_512);
+
+       snf->nfi_cfg.page_size = page_size;
+       snf->nfi_cfg.oob_size = oob_size;
+       snf->nfi_cfg.nsectors = nsectors;
+       snf->nfi_cfg.spare_size = spare_size;
+
+       dev_dbg(snf->dev, "page format: (%u + %u) * %u\n",
+               snf->caps->sector_size, spare_size, nsectors);
+       return snand_prepare_bouncebuf(snf, page_size + oob_size);
+err:
+       dev_err(snf->dev, "page size %u + %u is not supported\n", page_size,
+               oob_size);
+       return -EOPNOTSUPP;
+}
+
+static int mtk_snand_ooblayout_ecc(struct mtd_info *mtd, int section,
+                                  struct mtd_oob_region *oobecc)
+{
+       // ECC area is not accessible
+       return -ERANGE;
+}
+
+static int mtk_snand_ooblayout_free(struct mtd_info *mtd, int section,
+                                   struct mtd_oob_region *oobfree)
+{
+       struct nand_device *nand = mtd_to_nanddev(mtd);
+       struct mtk_snand *ms = nand_to_mtk_snand(nand);
+
+       if (section >= ms->nfi_cfg.nsectors)
+               return -ERANGE;
+
+       oobfree->length = ms->caps->fdm_size - 1;
+       oobfree->offset = section * ms->caps->fdm_size + 1;
+       return 0;
+}
+
+static const struct mtd_ooblayout_ops mtk_snand_ooblayout = {
+       .ecc = mtk_snand_ooblayout_ecc,
+       .free = mtk_snand_ooblayout_free,
+};
+
+static int mtk_snand_ecc_init_ctx(struct nand_device *nand)
+{
+       struct mtk_snand *snf = nand_to_mtk_snand(nand);
+       struct nand_ecc_props *conf = &nand->ecc.ctx.conf;
+       struct nand_ecc_props *reqs = &nand->ecc.requirements;
+       struct nand_ecc_props *user = &nand->ecc.user_conf;
+       struct mtd_info *mtd = nanddev_to_mtd(nand);
+       int step_size = 0, strength = 0, desired_correction = 0, steps;
+       bool ecc_user = false;
+       int ret;
+       u32 parity_bits, max_ecc_bytes;
+       struct mtk_ecc_config *ecc_cfg;
+
+       ret = mtk_snand_setup_pagefmt(snf, nand->memorg.pagesize,
+                                     nand->memorg.oobsize);
+       if (ret)
+               return ret;
+
+       ecc_cfg = kzalloc(sizeof(*ecc_cfg), GFP_KERNEL);
+       if (!ecc_cfg)
+               return -ENOMEM;
+
+       nand->ecc.ctx.priv = ecc_cfg;
+
+       if (user->step_size && user->strength) {
+               step_size = user->step_size;
+               strength = user->strength;
+               ecc_user = true;
+       } else if (reqs->step_size && reqs->strength) {
+               step_size = reqs->step_size;
+               strength = reqs->strength;
+       }
+
+       if (step_size && strength) {
+               steps = mtd->writesize / step_size;
+               desired_correction = steps * strength;
+               strength = desired_correction / snf->nfi_cfg.nsectors;
+       }
+
+       ecc_cfg->mode = ECC_NFI_MODE;
+       ecc_cfg->sectors = snf->nfi_cfg.nsectors;
+       ecc_cfg->len = snf->caps->sector_size + snf->caps->fdm_ecc_size;
+
+       // calculate the max possible strength under current page format
+       parity_bits = mtk_ecc_get_parity_bits(snf->ecc);
+       max_ecc_bytes = snf->nfi_cfg.spare_size - snf->caps->fdm_size;
+       ecc_cfg->strength = max_ecc_bytes * 8 / parity_bits;
+       mtk_ecc_adjust_strength(snf->ecc, &ecc_cfg->strength);
+
+       // if there's a user requested strength, find the minimum strength that
+       // meets the requirement. Otherwise use the maximum strength which is
+       // expected by BootROM.
+       if (ecc_user && strength) {
+               u32 s_next = ecc_cfg->strength - 1;
+
+               while (1) {
+                       mtk_ecc_adjust_strength(snf->ecc, &s_next);
+                       if (s_next >= ecc_cfg->strength)
+                               break;
+                       if (s_next < strength)
+                               break;
+                       s_next = ecc_cfg->strength - 1;
+               }
+       }
+
+       mtd_set_ooblayout(mtd, &mtk_snand_ooblayout);
+
+       conf->step_size = snf->caps->sector_size;
+       conf->strength = ecc_cfg->strength;
+
+       if (ecc_cfg->strength < strength)
+               dev_warn(snf->dev, "unable to fulfill ECC of %u bits.\n",
+                        strength);
+       dev_info(snf->dev, "ECC strength: %u bits per %u bytes\n",
+                ecc_cfg->strength, snf->caps->sector_size);
+
+       return 0;
+}
+
+static void mtk_snand_ecc_cleanup_ctx(struct nand_device *nand)
+{
+       struct mtk_ecc_config *ecc_cfg = nand_to_ecc_ctx(nand);
+
+       kfree(ecc_cfg);
+}
+
+static int mtk_snand_ecc_prepare_io_req(struct nand_device *nand,
+                                       struct nand_page_io_req *req)
+{
+       struct mtk_snand *snf = nand_to_mtk_snand(nand);
+       struct mtk_ecc_config *ecc_cfg = nand_to_ecc_ctx(nand);
+       int ret;
+
+       ret = mtk_snand_setup_pagefmt(snf, nand->memorg.pagesize,
+                                     nand->memorg.oobsize);
+       if (ret)
+               return ret;
+       snf->autofmt = true;
+       snf->ecc_cfg = ecc_cfg;
+       return 0;
+}
+
+static int mtk_snand_ecc_finish_io_req(struct nand_device *nand,
+                                      struct nand_page_io_req *req)
+{
+       struct mtk_snand *snf = nand_to_mtk_snand(nand);
+       struct mtd_info *mtd = nanddev_to_mtd(nand);
+
+       snf->ecc_cfg = NULL;
+       snf->autofmt = false;
+       if ((req->mode == MTD_OPS_RAW) || (req->type != NAND_PAGE_READ))
+               return 0;
+
+       if (snf->ecc_stats.failed)
+               mtd->ecc_stats.failed += snf->ecc_stats.failed;
+       mtd->ecc_stats.corrected += snf->ecc_stats.corrected;
+       return snf->ecc_stats.failed ? -EBADMSG : snf->ecc_stats.bitflips;
+}
+
+static struct nand_ecc_engine_ops mtk_snfi_ecc_engine_ops = {
+       .init_ctx = mtk_snand_ecc_init_ctx,
+       .cleanup_ctx = mtk_snand_ecc_cleanup_ctx,
+       .prepare_io_req = mtk_snand_ecc_prepare_io_req,
+       .finish_io_req = mtk_snand_ecc_finish_io_req,
+};
+
+static void mtk_snand_read_fdm(struct mtk_snand *snf, u8 *buf)
+{
+       u32 vall, valm;
+       u8 *oobptr = buf;
+       int i, j;
+
+       for (i = 0; i < snf->nfi_cfg.nsectors; i++) {
+               vall = nfi_read32(snf, NFI_FDML(i));
+               valm = nfi_read32(snf, NFI_FDMM(i));
+
+               for (j = 0; j < snf->caps->fdm_size; j++)
+                       oobptr[j] = (j >= 4 ? valm : vall) >> ((j % 4) * 8);
+
+               oobptr += snf->caps->fdm_size;
+       }
+}
+
+static void mtk_snand_write_fdm(struct mtk_snand *snf, const u8 *buf)
+{
+       u32 fdm_size = snf->caps->fdm_size;
+       const u8 *oobptr = buf;
+       u32 vall, valm;
+       int i, j;
+
+       for (i = 0; i < snf->nfi_cfg.nsectors; i++) {
+               vall = 0;
+               valm = 0;
+
+               for (j = 0; j < 8; j++) {
+                       if (j < 4)
+                               vall |= (j < fdm_size ? oobptr[j] : 0xff)
+                                       << (j * 8);
+                       else
+                               valm |= (j < fdm_size ? oobptr[j] : 0xff)
+                                       << ((j - 4) * 8);
+               }
+
+               nfi_write32(snf, NFI_FDML(i), vall);
+               nfi_write32(snf, NFI_FDMM(i), valm);
+
+               oobptr += fdm_size;
+       }
+}
+
+static void mtk_snand_bm_swap(struct mtk_snand *snf, u8 *buf)
+{
+       u32 buf_bbm_pos, fdm_bbm_pos;
+
+       if (!snf->caps->bbm_swap || snf->nfi_cfg.nsectors == 1)
+               return;
+
+       // swap [pagesize] byte on nand with the first fdm byte
+       // in the last sector.
+       buf_bbm_pos = snf->nfi_cfg.page_size -
+                     (snf->nfi_cfg.nsectors - 1) * snf->nfi_cfg.spare_size;
+       fdm_bbm_pos = snf->nfi_cfg.page_size +
+                     (snf->nfi_cfg.nsectors - 1) * snf->caps->fdm_size;
+
+       swap(snf->buf[fdm_bbm_pos], buf[buf_bbm_pos]);
+}
+
+static void mtk_snand_fdm_bm_swap(struct mtk_snand *snf)
+{
+       u32 fdm_bbm_pos1, fdm_bbm_pos2;
+
+       if (!snf->caps->bbm_swap || snf->nfi_cfg.nsectors == 1)
+               return;
+
+       // swap the first fdm byte in the first and the last sector.
+       fdm_bbm_pos1 = snf->nfi_cfg.page_size;
+       fdm_bbm_pos2 = snf->nfi_cfg.page_size +
+                      (snf->nfi_cfg.nsectors - 1) * snf->caps->fdm_size;
+       swap(snf->buf[fdm_bbm_pos1], snf->buf[fdm_bbm_pos2]);
+}
+
+static int mtk_snand_read_page_cache(struct mtk_snand *snf,
+                                    const struct spi_mem_op *op)
+{
+       u8 *buf = snf->buf;
+       u8 *buf_fdm = buf + snf->nfi_cfg.page_size;
+       // the address part to be sent by the controller
+       u32 op_addr = op->addr.val;
+       // where to start copying data from bounce buffer
+       u32 rd_offset = 0;
+       u32 dummy_clk = (op->dummy.nbytes * BITS_PER_BYTE / op->dummy.buswidth);
+       u32 op_mode = 0;
+       u32 dma_len = snf->buf_len;
+       int ret = 0;
+       u32 rd_mode, rd_bytes, val;
+       dma_addr_t buf_dma;
+
+       if (snf->autofmt) {
+               u32 last_bit;
+               u32 mask;
+
+               dma_len = snf->nfi_cfg.page_size;
+               op_mode = CNFG_AUTO_FMT_EN;
+               if (op->data.ecc)
+                       op_mode |= CNFG_HW_ECC_EN;
+               // extract the plane bit:
+               // Find the highest bit set in (pagesize+oobsize).
+               // Bits higher than that in op->addr are kept and sent over SPI
+               // Lower bits are used as an offset for copying data from DMA
+               // bounce buffer.
+               last_bit = fls(snf->nfi_cfg.page_size + snf->nfi_cfg.oob_size);
+               mask = (1 << last_bit) - 1;
+               rd_offset = op_addr & mask;
+               op_addr &= ~mask;
+
+               // check if we can dma to the caller memory
+               if (rd_offset == 0 && op->data.nbytes >= snf->nfi_cfg.page_size)
+                       buf = op->data.buf.in;
+       }
+       mtk_snand_mac_reset(snf);
+       mtk_nfi_reset(snf);
+
+       // command and dummy cycles
+       nfi_write32(snf, SNF_RD_CTL2,
+                   (dummy_clk << DATA_READ_DUMMY_S) |
+                           (op->cmd.opcode << DATA_READ_CMD_S));
+
+       // read address
+       nfi_write32(snf, SNF_RD_CTL3, op_addr);
+
+       // Set read op_mode
+       if (op->data.buswidth == 4)
+               rd_mode = op->addr.buswidth == 4 ? DATA_READ_MODE_QUAD :
+                                                  DATA_READ_MODE_X4;
+       else if (op->data.buswidth == 2)
+               rd_mode = op->addr.buswidth == 2 ? DATA_READ_MODE_DUAL :
+                                                  DATA_READ_MODE_X2;
+       else
+               rd_mode = DATA_READ_MODE_X1;
+       rd_mode <<= DATA_READ_MODE_S;
+       nfi_rmw32(snf, SNF_MISC_CTL, DATA_READ_MODE,
+                 rd_mode | DATARD_CUSTOM_EN);
+
+       // Set bytes to read
+       rd_bytes = (snf->nfi_cfg.spare_size + snf->caps->sector_size) *
+                  snf->nfi_cfg.nsectors;
+       nfi_write32(snf, SNF_MISC_CTL2,
+                   (rd_bytes << PROGRAM_LOAD_BYTE_NUM_S) | rd_bytes);
+
+       // NFI read prepare
+       nfi_write16(snf, NFI_CNFG,
+                   (CNFG_OP_MODE_CUST << CNFG_OP_MODE_S) | CNFG_DMA_BURST_EN |
+                           CNFG_READ_MODE | CNFG_DMA_MODE | op_mode);
+
+       nfi_write32(snf, NFI_CON, (snf->nfi_cfg.nsectors << CON_SEC_NUM_S));
+
+       buf_dma = dma_map_single(snf->dev, buf, dma_len, DMA_FROM_DEVICE);
+       if (dma_mapping_error(snf->dev, buf_dma)) {
+               dev_err(snf->dev, "DMA mapping failed.\n");
+               goto cleanup;
+       }
+       nfi_write32(snf, NFI_STRADDR, buf_dma);
+       if (op->data.ecc) {
+               snf->ecc_cfg->op = ECC_DECODE;
+               ret = mtk_ecc_enable(snf->ecc, snf->ecc_cfg);
+               if (ret)
+                       goto cleanup_dma;
+       }
+       // Prepare for custom read interrupt
+       nfi_write32(snf, NFI_INTR_EN, NFI_IRQ_INTR_EN | NFI_IRQ_CUS_READ);
+       reinit_completion(&snf->op_done);
+
+       // Trigger NFI into custom mode
+       nfi_write16(snf, NFI_CMD, NFI_CMD_DUMMY_READ);
+
+       // Start DMA read
+       nfi_rmw32(snf, NFI_CON, 0, CON_BRD);
+       nfi_write16(snf, NFI_STRDATA, STR_DATA);
+
+       if (!wait_for_completion_timeout(
+                   &snf->op_done, usecs_to_jiffies(SNFI_POLL_INTERVAL))) {
+               dev_err(snf->dev, "DMA timed out for reading from cache.\n");
+               ret = -ETIMEDOUT;
+               goto cleanup;
+       }
+
+       // Wait for BUS_SEC_CNTR returning expected value
+       ret = readl_poll_timeout(snf->nfi_base + NFI_BYTELEN, val,
+                                BUS_SEC_CNTR(val) >= snf->nfi_cfg.nsectors, 0,
+                                SNFI_POLL_INTERVAL);
+       if (ret) {
+               dev_err(snf->dev, "Timed out waiting for BUS_SEC_CNTR\n");
+               goto cleanup2;
+       }
+
+       // Wait for bus becoming idle
+       ret = readl_poll_timeout(snf->nfi_base + NFI_MASTERSTA, val,
+                                !(val & snf->caps->mastersta_mask), 0,
+                                SNFI_POLL_INTERVAL);
+       if (ret) {
+               dev_err(snf->dev, "Timed out waiting for bus becoming idle\n");
+               goto cleanup2;
+       }
+
+       if (op->data.ecc) {
+               ret = mtk_ecc_wait_done(snf->ecc, ECC_DECODE);
+               if (ret) {
+                       dev_err(snf->dev, "wait ecc done timeout\n");
+                       goto cleanup2;
+               }
+               // save status before disabling ecc
+               mtk_ecc_get_stats(snf->ecc, &snf->ecc_stats,
+                                 snf->nfi_cfg.nsectors);
+       }
+
+       dma_unmap_single(snf->dev, buf_dma, dma_len, DMA_FROM_DEVICE);
+
+       if (snf->autofmt) {
+               mtk_snand_read_fdm(snf, buf_fdm);
+               if (snf->caps->bbm_swap) {
+                       mtk_snand_bm_swap(snf, buf);
+                       mtk_snand_fdm_bm_swap(snf);
+               }
+       }
+
+       // copy data back
+       if (nfi_read32(snf, NFI_STA) & READ_EMPTY) {
+               memset(op->data.buf.in, 0xff, op->data.nbytes);
+               snf->ecc_stats.bitflips = 0;
+               snf->ecc_stats.failed = 0;
+               snf->ecc_stats.corrected = 0;
+       } else {
+               if (buf == op->data.buf.in) {
+                       u32 cap_len = snf->buf_len - snf->nfi_cfg.page_size;
+                       u32 req_left = op->data.nbytes - snf->nfi_cfg.page_size;
+
+                       if (req_left)
+                               memcpy(op->data.buf.in + snf->nfi_cfg.page_size,
+                                      buf_fdm,
+                                      cap_len < req_left ? cap_len : req_left);
+               } else if (rd_offset < snf->buf_len) {
+                       u32 cap_len = snf->buf_len - rd_offset;
+
+                       if (op->data.nbytes < cap_len)
+                               cap_len = op->data.nbytes;
+                       memcpy(op->data.buf.in, snf->buf + rd_offset, cap_len);
+               }
+       }
+cleanup2:
+       if (op->data.ecc)
+               mtk_ecc_disable(snf->ecc);
+cleanup_dma:
+       // unmap dma only if any error happens. (otherwise it's done before
+       // data copying)
+       if (ret)
+               dma_unmap_single(snf->dev, buf_dma, dma_len, DMA_FROM_DEVICE);
+cleanup:
+       // Stop read
+       nfi_write32(snf, NFI_CON, 0);
+       nfi_write16(snf, NFI_CNFG, 0);
+
+       // Clear SNF done flag
+       nfi_rmw32(snf, SNF_STA_CTL1, 0, CUS_READ_DONE);
+       nfi_write32(snf, SNF_STA_CTL1, 0);
+
+       // Disable interrupt
+       nfi_read32(snf, NFI_INTR_STA);
+       nfi_write32(snf, NFI_INTR_EN, 0);
+
+       nfi_rmw32(snf, SNF_MISC_CTL, DATARD_CUSTOM_EN, 0);
+       return ret;
+}
+
+static int mtk_snand_write_page_cache(struct mtk_snand *snf,
+                                     const struct spi_mem_op *op)
+{
+       // the address part to be sent by the controller
+       u32 op_addr = op->addr.val;
+       // where to start copying data from bounce buffer
+       u32 wr_offset = 0;
+       u32 op_mode = 0;
+       int ret = 0;
+       u32 wr_mode = 0;
+       u32 dma_len = snf->buf_len;
+       u32 wr_bytes, val;
+       size_t cap_len;
+       dma_addr_t buf_dma;
+
+       if (snf->autofmt) {
+               u32 last_bit;
+               u32 mask;
+
+               dma_len = snf->nfi_cfg.page_size;
+               op_mode = CNFG_AUTO_FMT_EN;
+               if (op->data.ecc)
+                       op_mode |= CNFG_HW_ECC_EN;
+
+               last_bit = fls(snf->nfi_cfg.page_size + snf->nfi_cfg.oob_size);
+               mask = (1 << last_bit) - 1;
+               wr_offset = op_addr & mask;
+               op_addr &= ~mask;
+       }
+       mtk_snand_mac_reset(snf);
+       mtk_nfi_reset(snf);
+
+       if (wr_offset)
+               memset(snf->buf, 0xff, wr_offset);
+
+       cap_len = snf->buf_len - wr_offset;
+       if (op->data.nbytes < cap_len)
+               cap_len = op->data.nbytes;
+       memcpy(snf->buf + wr_offset, op->data.buf.out, cap_len);
+       if (snf->autofmt) {
+               if (snf->caps->bbm_swap) {
+                       mtk_snand_fdm_bm_swap(snf);
+                       mtk_snand_bm_swap(snf, snf->buf);
+               }
+               mtk_snand_write_fdm(snf, snf->buf + snf->nfi_cfg.page_size);
+       }
+
+       // Command
+       nfi_write32(snf, SNF_PG_CTL1, (op->cmd.opcode << PG_LOAD_CMD_S));
+
+       // write address
+       nfi_write32(snf, SNF_PG_CTL2, op_addr);
+
+       // Set read op_mode
+       if (op->data.buswidth == 4)
+               wr_mode = PG_LOAD_X4_EN;
+
+       nfi_rmw32(snf, SNF_MISC_CTL, PG_LOAD_X4_EN,
+                 wr_mode | PG_LOAD_CUSTOM_EN);
+
+       // Set bytes to write
+       wr_bytes = (snf->nfi_cfg.spare_size + snf->caps->sector_size) *
+                  snf->nfi_cfg.nsectors;
+       nfi_write32(snf, SNF_MISC_CTL2,
+                   (wr_bytes << PROGRAM_LOAD_BYTE_NUM_S) | wr_bytes);
+
+       // NFI write prepare
+       nfi_write16(snf, NFI_CNFG,
+                   (CNFG_OP_MODE_PROGRAM << CNFG_OP_MODE_S) |
+                           CNFG_DMA_BURST_EN | CNFG_DMA_MODE | op_mode);
+
+       nfi_write32(snf, NFI_CON, (snf->nfi_cfg.nsectors << CON_SEC_NUM_S));
+       buf_dma = dma_map_single(snf->dev, snf->buf, dma_len, DMA_TO_DEVICE);
+       if (dma_mapping_error(snf->dev, buf_dma)) {
+               dev_err(snf->dev, "DMA mapping failed.\n");
+               goto cleanup;
+       }
+       nfi_write32(snf, NFI_STRADDR, buf_dma);
+       if (op->data.ecc) {
+               snf->ecc_cfg->op = ECC_ENCODE;
+               ret = mtk_ecc_enable(snf->ecc, snf->ecc_cfg);
+               if (ret)
+                       goto cleanup_dma;
+       }
+       // Prepare for custom write interrupt
+       nfi_write32(snf, NFI_INTR_EN, NFI_IRQ_INTR_EN | NFI_IRQ_CUS_PG);
+       reinit_completion(&snf->op_done);
+       ;
+
+       // Trigger NFI into custom mode
+       nfi_write16(snf, NFI_CMD, NFI_CMD_DUMMY_WRITE);
+
+       // Start DMA write
+       nfi_rmw32(snf, NFI_CON, 0, CON_BWR);
+       nfi_write16(snf, NFI_STRDATA, STR_DATA);
+
+       if (!wait_for_completion_timeout(
+                   &snf->op_done, usecs_to_jiffies(SNFI_POLL_INTERVAL))) {
+               dev_err(snf->dev, "DMA timed out for program load.\n");
+               ret = -ETIMEDOUT;
+               goto cleanup_ecc;
+       }
+
+       // Wait for NFI_SEC_CNTR returning expected value
+       ret = readl_poll_timeout(snf->nfi_base + NFI_ADDRCNTR, val,
+                                NFI_SEC_CNTR(val) >= snf->nfi_cfg.nsectors, 0,
+                                SNFI_POLL_INTERVAL);
+       if (ret)
+               dev_err(snf->dev, "Timed out waiting for NFI_SEC_CNTR\n");
+
+cleanup_ecc:
+       if (op->data.ecc)
+               mtk_ecc_disable(snf->ecc);
+cleanup_dma:
+       dma_unmap_single(snf->dev, buf_dma, dma_len, DMA_TO_DEVICE);
+cleanup:
+       // Stop write
+       nfi_write32(snf, NFI_CON, 0);
+       nfi_write16(snf, NFI_CNFG, 0);
+
+       // Clear SNF done flag
+       nfi_rmw32(snf, SNF_STA_CTL1, 0, CUS_PG_DONE);
+       nfi_write32(snf, SNF_STA_CTL1, 0);
+
+       // Disable interrupt
+       nfi_read32(snf, NFI_INTR_STA);
+       nfi_write32(snf, NFI_INTR_EN, 0);
+
+       nfi_rmw32(snf, SNF_MISC_CTL, PG_LOAD_CUSTOM_EN, 0);
+
+       return ret;
+}
+
+/**
+ * mtk_snand_is_page_ops() - check if the op is a controller supported page op.
+ * @op spi-mem op to check
+ *
+ * Check whether op can be executed with read_from_cache or program_load
+ * mode in the controller.
+ * This controller can execute typical Read From Cache and Program Load
+ * instructions found on SPI-NAND with 2-byte address.
+ * DTR and cmd buswidth & nbytes should be checked before calling this.
+ *
+ * Return: true if the op matches the instruction template
+ */
+static bool mtk_snand_is_page_ops(const struct spi_mem_op *op)
+{
+       if (op->addr.nbytes != 2)
+               return false;
+
+       if (op->addr.buswidth != 1 && op->addr.buswidth != 2 &&
+           op->addr.buswidth != 4)
+               return false;
+
+       // match read from page instructions
+       if (op->data.dir == SPI_MEM_DATA_IN) {
+               // check dummy cycle first
+               if (op->dummy.nbytes * BITS_PER_BYTE / op->dummy.buswidth >
+                   DATA_READ_MAX_DUMMY)
+                       return false;
+               // quad io / quad out
+               if ((op->addr.buswidth == 4 || op->addr.buswidth == 1) &&
+                   op->data.buswidth == 4)
+                       return true;
+
+               // dual io / dual out
+               if ((op->addr.buswidth == 2 || op->addr.buswidth == 1) &&
+                   op->data.buswidth == 2)
+                       return true;
+
+               // standard spi
+               if (op->addr.buswidth == 1 && op->data.buswidth == 1)
+                       return true;
+       } else if (op->data.dir == SPI_MEM_DATA_OUT) {
+               // check dummy cycle first
+               if (op->dummy.nbytes)
+                       return false;
+               // program load quad out
+               if (op->addr.buswidth == 1 && op->data.buswidth == 4)
+                       return true;
+               // standard spi
+               if (op->addr.buswidth == 1 && op->data.buswidth == 1)
+                       return true;
+       }
+       return false;
+}
+
+static bool mtk_snand_supports_op(struct spi_mem *mem,
+                                 const struct spi_mem_op *op)
+{
+       if (!spi_mem_default_supports_op(mem, op))
+               return false;
+       if (op->cmd.nbytes != 1 || op->cmd.buswidth != 1)
+               return false;
+       if (mtk_snand_is_page_ops(op))
+               return true;
+       return ((op->addr.nbytes == 0 || op->addr.buswidth == 1) &&
+               (op->dummy.nbytes == 0 || op->dummy.buswidth == 1) &&
+               (op->data.nbytes == 0 || op->data.buswidth == 1));
+}
+
+static int mtk_snand_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
+{
+       struct mtk_snand *ms = spi_controller_get_devdata(mem->spi->master);
+       // page ops transfer size must be exactly ((sector_size + spare_size) *
+       // nsectors). Limit the op size if the caller requests more than that.
+       // exec_op will read more than needed and discard the leftover if the
+       // caller requests less data.
+       if (mtk_snand_is_page_ops(op)) {
+               size_t l;
+               // skip adjust_op_size for page ops
+               if (ms->autofmt)
+                       return 0;
+               l = ms->caps->sector_size + ms->nfi_cfg.spare_size;
+               l *= ms->nfi_cfg.nsectors;
+               if (op->data.nbytes > l)
+                       op->data.nbytes = l;
+       } else {
+               size_t hl = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
+
+               if (hl >= SNF_GPRAM_SIZE)
+                       return -EOPNOTSUPP;
+               if (op->data.nbytes > SNF_GPRAM_SIZE - hl)
+                       op->data.nbytes = SNF_GPRAM_SIZE - hl;
+       }
+       return 0;
+}
+
+static int mtk_snand_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
+{
+       struct mtk_snand *ms = spi_controller_get_devdata(mem->spi->master);
+
+       dev_dbg(ms->dev, "OP %02x ADDR %08llX@%d:%u DATA %d:%u", op->cmd.opcode,
+               op->addr.val, op->addr.buswidth, op->addr.nbytes,
+               op->data.buswidth, op->data.nbytes);
+       if (mtk_snand_is_page_ops(op)) {
+               if (op->data.dir == SPI_MEM_DATA_IN)
+                       return mtk_snand_read_page_cache(ms, op);
+               else
+                       return mtk_snand_write_page_cache(ms, op);
+       } else {
+               return mtk_snand_mac_io(ms, op);
+       }
+}
+
+static const struct spi_controller_mem_ops mtk_snand_mem_ops = {
+       .adjust_op_size = mtk_snand_adjust_op_size,
+       .supports_op = mtk_snand_supports_op,
+       .exec_op = mtk_snand_exec_op,
+};
+
+static const struct spi_controller_mem_caps mtk_snand_mem_caps = {
+       .ecc = true,
+};
+
+static irqreturn_t mtk_snand_irq(int irq, void *id)
+{
+       struct mtk_snand *snf = id;
+       u32 sta, ien;
+
+       sta = nfi_read32(snf, NFI_INTR_STA);
+       ien = nfi_read32(snf, NFI_INTR_EN);
+
+       if (!(sta & ien))
+               return IRQ_NONE;
+
+       nfi_write32(snf, NFI_INTR_EN, 0);
+       complete(&snf->op_done);
+       return IRQ_HANDLED;
+}
+
+static const struct of_device_id mtk_snand_ids[] = {
+       { .compatible = "mediatek,mt7622-snand", .data = &mt7622_snand_caps },
+       { .compatible = "mediatek,mt7629-snand", .data = &mt7629_snand_caps },
+       {},
+};
+
+MODULE_DEVICE_TABLE(of, mtk_snand_ids);
+
+static int mtk_snand_enable_clk(struct mtk_snand *ms)
+{
+       int ret;
+
+       ret = clk_prepare_enable(ms->nfi_clk);
+       if (ret) {
+               dev_err(ms->dev, "unable to enable nfi clk\n");
+               return ret;
+       }
+       ret = clk_prepare_enable(ms->pad_clk);
+       if (ret) {
+               dev_err(ms->dev, "unable to enable pad clk\n");
+               goto err1;
+       }
+       return 0;
+err1:
+       clk_disable_unprepare(ms->nfi_clk);
+       return ret;
+}
+
+static void mtk_snand_disable_clk(struct mtk_snand *ms)
+{
+       clk_disable_unprepare(ms->pad_clk);
+       clk_disable_unprepare(ms->nfi_clk);
+}
+
+static int mtk_snand_probe(struct platform_device *pdev)
+{
+       struct device_node *np = pdev->dev.of_node;
+       const struct of_device_id *dev_id;
+       struct spi_controller *ctlr;
+       struct mtk_snand *ms;
+       int ret;
+
+       dev_id = of_match_node(mtk_snand_ids, np);
+       if (!dev_id)
+               return -EINVAL;
+
+       ctlr = devm_spi_alloc_master(&pdev->dev, sizeof(*ms));
+       if (!ctlr)
+               return -ENOMEM;
+       platform_set_drvdata(pdev, ctlr);
+
+       ms = spi_controller_get_devdata(ctlr);
+
+       ms->ctlr = ctlr;
+       ms->caps = dev_id->data;
+
+       ms->ecc = of_mtk_ecc_get(np);
+       if (IS_ERR(ms->ecc))
+               return PTR_ERR(ms->ecc);
+       else if (!ms->ecc)
+               return -ENODEV;
+
+       ms->nfi_base = devm_platform_ioremap_resource(pdev, 0);
+       if (IS_ERR(ms->nfi_base)) {
+               ret = PTR_ERR(ms->nfi_base);
+               goto release_ecc;
+       }
+
+       ms->dev = &pdev->dev;
+
+       ms->nfi_clk = devm_clk_get(&pdev->dev, "nfi_clk");
+       if (IS_ERR(ms->nfi_clk)) {
+               ret = PTR_ERR(ms->nfi_clk);
+               dev_err(&pdev->dev, "unable to get nfi_clk, err = %d\n", ret);
+               goto release_ecc;
+       }
+
+       ms->pad_clk = devm_clk_get(&pdev->dev, "pad_clk");
+       if (IS_ERR(ms->pad_clk)) {
+               ret = PTR_ERR(ms->pad_clk);
+               dev_err(&pdev->dev, "unable to get pad_clk, err = %d\n", ret);
+               goto release_ecc;
+       }
+
+       ret = mtk_snand_enable_clk(ms);
+       if (ret)
+               goto release_ecc;
+
+       init_completion(&ms->op_done);
+
+       ms->irq = platform_get_irq(pdev, 0);
+       if (ms->irq < 0) {
+               ret = ms->irq;
+               goto disable_clk;
+       }
+       ret = devm_request_irq(ms->dev, ms->irq, mtk_snand_irq, 0x0,
+                              "mtk-snand", ms);
+       if (ret) {
+               dev_err(ms->dev, "failed to request snfi irq\n");
+               goto disable_clk;
+       }
+
+       ret = dma_set_mask(ms->dev, DMA_BIT_MASK(32));
+       if (ret) {
+               dev_err(ms->dev, "failed to set dma mask\n");
+               goto disable_clk;
+       }
+
+       // switch to SNFI mode
+       nfi_write32(ms, SNF_CFG, SPI_MODE);
+
+       // setup an initial page format for ops matching page_cache_op template
+       // before ECC is called.
+       ret = mtk_snand_setup_pagefmt(ms, ms->caps->sector_size,
+                                     ms->caps->spare_sizes[0]);
+       if (ret) {
+               dev_err(ms->dev, "failed to set initial page format\n");
+               goto disable_clk;
+       }
+
+       // setup ECC engine
+       ms->ecc_eng.dev = &pdev->dev;
+       ms->ecc_eng.integration = NAND_ECC_ENGINE_INTEGRATION_PIPELINED;
+       ms->ecc_eng.ops = &mtk_snfi_ecc_engine_ops;
+       ms->ecc_eng.priv = ms;
+
+       ret = nand_ecc_register_on_host_hw_engine(&ms->ecc_eng);
+       if (ret) {
+               dev_err(&pdev->dev, "failed to register ecc engine.\n");
+               goto disable_clk;
+       }
+
+       ctlr->num_chipselect = 1;
+       ctlr->mem_ops = &mtk_snand_mem_ops;
+       ctlr->mem_caps = &mtk_snand_mem_caps;
+       ctlr->bits_per_word_mask = SPI_BPW_MASK(8);
+       ctlr->mode_bits = SPI_RX_DUAL | SPI_RX_QUAD | SPI_TX_DUAL | SPI_TX_QUAD;
+       ctlr->dev.of_node = pdev->dev.of_node;
+       ret = spi_register_controller(ctlr);
+       if (ret) {
+               dev_err(&pdev->dev, "spi_register_controller failed.\n");
+               goto disable_clk;
+       }
+
+       return 0;
+disable_clk:
+       mtk_snand_disable_clk(ms);
+release_ecc:
+       mtk_ecc_release(ms->ecc);
+       return ret;
+}
+
+static int mtk_snand_remove(struct platform_device *pdev)
+{
+       struct spi_controller *ctlr = platform_get_drvdata(pdev);
+       struct mtk_snand *ms = spi_controller_get_devdata(ctlr);
+
+       spi_unregister_controller(ctlr);
+       mtk_snand_disable_clk(ms);
+       mtk_ecc_release(ms->ecc);
+       kfree(ms->buf);
+       return 0;
+}
+
+static struct platform_driver mtk_snand_driver = {
+       .probe = mtk_snand_probe,
+       .remove = mtk_snand_remove,
+       .driver = {
+               .name = "mtk-snand",
+               .of_match_table = mtk_snand_ids,
+       },
+};
+
+module_platform_driver(mtk_snand_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Chuanhong Guo <gch981213@gmail.com>");
+MODULE_DESCRIPTION("MeidaTek SPI-NAND Flash Controller Driver");