--- /dev/null
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Marvell NAND flash controller driver
+ *
+ * Copyright (C) 2017 Marvell
+ * Author: Miquel RAYNAL <miquel.raynal@free-electrons.com>
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/clk.h>
+#include <linux/mtd/rawnand.h>
+#include <linux/of_platform.h>
+#include <linux/iopoll.h>
+#include <linux/interrupt.h>
+#include <linux/slab.h>
+#include <linux/mfd/syscon.h>
+#include <linux/regmap.h>
+#include <asm/unaligned.h>
+
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/dma/pxa-dma.h>
+#include <linux/platform_data/mtd-nand-pxa3xx.h>
+
+/* Data FIFO granularity, FIFO reads/writes must be a multiple of this length */
+#define FIFO_DEPTH             8
+#define FIFO_REP(x)            (x / sizeof(u32))
+#define BCH_SEQ_READS          (32 / FIFO_DEPTH)
+/* NFC does not support transfers of larger chunks at a time */
+#define MAX_CHUNK_SIZE         2112
+/* NFCv1 cannot read more that 7 bytes of ID */
+#define NFCV1_READID_LEN       7
+/* Polling is done at a pace of POLL_PERIOD us until POLL_TIMEOUT is reached */
+#define POLL_PERIOD            0
+#define POLL_TIMEOUT           100000
+/* Interrupt maximum wait period in ms */
+#define IRQ_TIMEOUT            1000
+/* Latency in clock cycles between SoC pins and NFC logic */
+#define MIN_RD_DEL_CNT         3
+/* Maximum number of contiguous address cycles */
+#define MAX_ADDRESS_CYC_NFCV1  5
+#define MAX_ADDRESS_CYC_NFCV2  7
+/* System control registers/bits to enable the NAND controller on some SoCs */
+#define GENCONF_SOC_DEVICE_MUX 0x208
+#define GENCONF_SOC_DEVICE_MUX_NFC_EN BIT(0)
+#define GENCONF_SOC_DEVICE_MUX_ECC_CLK_RST BIT(20)
+#define GENCONF_SOC_DEVICE_MUX_ECC_CORE_RST BIT(21)
+#define GENCONF_SOC_DEVICE_MUX_NFC_INT_EN BIT(25)
+#define GENCONF_CLK_GATING_CTRL        0x220
+#define GENCONF_CLK_GATING_CTRL_ND_GATE BIT(2)
+#define GENCONF_ND_CLK_CTRL    0x700
+#define GENCONF_ND_CLK_CTRL_EN BIT(0)
+
+/* NAND controller data flash control register */
+#define NDCR                   0x00
+#define NDCR_ALL_INT           GENMASK(11, 0)
+#define NDCR_CS1_CMDDM         BIT(7)
+#define NDCR_CS0_CMDDM         BIT(8)
+#define NDCR_RDYM              BIT(11)
+#define NDCR_ND_ARB_EN         BIT(12)
+#define NDCR_RA_START          BIT(15)
+#define NDCR_RD_ID_CNT(x)      (min_t(unsigned int, x, 0x7) << 16)
+#define NDCR_PAGE_SZ(x)                (x >= 2048 ? BIT(24) : 0)
+#define NDCR_DWIDTH_M          BIT(26)
+#define NDCR_DWIDTH_C          BIT(27)
+#define NDCR_ND_RUN            BIT(28)
+#define NDCR_DMA_EN            BIT(29)
+#define NDCR_ECC_EN            BIT(30)
+#define NDCR_SPARE_EN          BIT(31)
+#define NDCR_GENERIC_FIELDS_MASK (~(NDCR_RA_START | NDCR_PAGE_SZ(2048) | \
+                                   NDCR_DWIDTH_M | NDCR_DWIDTH_C))
+
+/* NAND interface timing parameter 0 register */
+#define NDTR0                  0x04
+#define NDTR0_TRP(x)           ((min_t(unsigned int, x, 0xF) & 0x7) << 0)
+#define NDTR0_TRH(x)           (min_t(unsigned int, x, 0x7) << 3)
+#define NDTR0_ETRP(x)          ((min_t(unsigned int, x, 0xF) & 0x8) << 3)
+#define NDTR0_SEL_NRE_EDGE     BIT(7)
+#define NDTR0_TWP(x)           (min_t(unsigned int, x, 0x7) << 8)
+#define NDTR0_TWH(x)           (min_t(unsigned int, x, 0x7) << 11)
+#define NDTR0_TCS(x)           (min_t(unsigned int, x, 0x7) << 16)
+#define NDTR0_TCH(x)           (min_t(unsigned int, x, 0x7) << 19)
+#define NDTR0_RD_CNT_DEL(x)    (min_t(unsigned int, x, 0xF) << 22)
+#define NDTR0_SELCNTR          BIT(26)
+#define NDTR0_TADL(x)          (min_t(unsigned int, x, 0x1F) << 27)
+
+/* NAND interface timing parameter 1 register */
+#define NDTR1                  0x0C
+#define NDTR1_TAR(x)           (min_t(unsigned int, x, 0xF) << 0)
+#define NDTR1_TWHR(x)          (min_t(unsigned int, x, 0xF) << 4)
+#define NDTR1_TRHW(x)          (min_t(unsigned int, x / 16, 0x3) << 8)
+#define NDTR1_PRESCALE         BIT(14)
+#define NDTR1_WAIT_MODE                BIT(15)
+#define NDTR1_TR(x)            (min_t(unsigned int, x, 0xFFFF) << 16)
+
+/* NAND controller status register */
+#define NDSR                   0x14
+#define NDSR_WRCMDREQ          BIT(0)
+#define NDSR_RDDREQ            BIT(1)
+#define NDSR_WRDREQ            BIT(2)
+#define NDSR_CORERR            BIT(3)
+#define NDSR_UNCERR            BIT(4)
+#define NDSR_CMDD(cs)          BIT(8 - cs)
+#define NDSR_RDY(rb)           BIT(11 + rb)
+#define NDSR_ERRCNT(x)         ((x >> 16) & 0x1F)
+
+/* NAND ECC control register */
+#define NDECCCTRL              0x28
+#define NDECCCTRL_BCH_EN       BIT(0)
+
+/* NAND controller data buffer register */
+#define NDDB                   0x40
+
+/* NAND controller command buffer 0 register */
+#define NDCB0                  0x48
+#define NDCB0_CMD1(x)          ((x & 0xFF) << 0)
+#define NDCB0_CMD2(x)          ((x & 0xFF) << 8)
+#define NDCB0_ADDR_CYC(x)      ((x & 0x7) << 16)
+#define NDCB0_ADDR_GET_NUM_CYC(x) (((x) >> 16) & 0x7)
+#define NDCB0_DBC              BIT(19)
+#define NDCB0_CMD_TYPE(x)      ((x & 0x7) << 21)
+#define NDCB0_CSEL             BIT(24)
+#define NDCB0_RDY_BYP          BIT(27)
+#define NDCB0_LEN_OVRD         BIT(28)
+#define NDCB0_CMD_XTYPE(x)     ((x & 0x7) << 29)
+
+/* NAND controller command buffer 1 register */
+#define NDCB1                  0x4C
+#define NDCB1_COLS(x)          ((x & 0xFFFF) << 0)
+#define NDCB1_ADDRS_PAGE(x)    (x << 16)
+
+/* NAND controller command buffer 2 register */
+#define NDCB2                  0x50
+#define NDCB2_ADDR5_PAGE(x)    (((x >> 16) & 0xFF) << 0)
+#define NDCB2_ADDR5_CYC(x)     ((x & 0xFF) << 0)
+
+/* NAND controller command buffer 3 register */
+#define NDCB3                  0x54
+#define NDCB3_ADDR6_CYC(x)     ((x & 0xFF) << 16)
+#define NDCB3_ADDR7_CYC(x)     ((x & 0xFF) << 24)
+
+/* NAND controller command buffer 0 register 'type' and 'xtype' fields */
+#define TYPE_READ              0
+#define TYPE_WRITE             1
+#define TYPE_ERASE             2
+#define TYPE_READ_ID           3
+#define TYPE_STATUS            4
+#define TYPE_RESET             5
+#define TYPE_NAKED_CMD         6
+#define TYPE_NAKED_ADDR                7
+#define TYPE_MASK              7
+#define XTYPE_MONOLITHIC_RW    0
+#define XTYPE_LAST_NAKED_RW    1
+#define XTYPE_FINAL_COMMAND    3
+#define XTYPE_READ             4
+#define XTYPE_WRITE_DISPATCH   4
+#define XTYPE_NAKED_RW         5
+#define XTYPE_COMMAND_DISPATCH 6
+#define XTYPE_MASK             7
+
+/**
+ * Marvell ECC engine works differently than the others, in order to limit the
+ * size of the IP, hardware engineers chose to set a fixed strength at 16 bits
+ * per subpage, and depending on a the desired strength needed by the NAND chip,
+ * a particular layout mixing data/spare/ecc is defined, with a possible last
+ * chunk smaller that the others.
+ *
+ * @writesize:         Full page size on which the layout applies
+ * @chunk:             Desired ECC chunk size on which the layout applies
+ * @strength:          Desired ECC strength (per chunk size bytes) on which the
+ *                     layout applies
+ * @nchunks:           Total number of chunks
+ * @full_chunk_cnt:    Number of full-sized chunks, which is the number of
+ *                     repetitions of the pattern:
+ *                     (data_bytes + spare_bytes + ecc_bytes).
+ * @data_bytes:                Number of data bytes per chunk
+ * @spare_bytes:       Number of spare bytes per chunk
+ * @ecc_bytes:         Number of ecc bytes per chunk
+ * @last_data_bytes:   Number of data bytes in the last chunk
+ * @last_spare_bytes:  Number of spare bytes in the last chunk
+ * @last_ecc_bytes:    Number of ecc bytes in the last chunk
+ */
+struct marvell_hw_ecc_layout {
+       /* Constraints */
+       int writesize;
+       int chunk;
+       int strength;
+       /* Corresponding layout */
+       int nchunks;
+       int full_chunk_cnt;
+       int data_bytes;
+       int spare_bytes;
+       int ecc_bytes;
+       int last_data_bytes;
+       int last_spare_bytes;
+       int last_ecc_bytes;
+};
+
+#define MARVELL_LAYOUT(ws, dc, ds, nc, fcc, db, sb, eb, ldb, lsb, leb) \
+       {                                                               \
+               .writesize = ws,                                        \
+               .chunk = dc,                                            \
+               .strength = ds,                                         \
+               .nchunks = nc,                                          \
+               .full_chunk_cnt = fcc,                                  \
+               .data_bytes = db,                                       \
+               .spare_bytes = sb,                                      \
+               .ecc_bytes = eb,                                        \
+               .last_data_bytes = ldb,                                 \
+               .last_spare_bytes = lsb,                                \
+               .last_ecc_bytes = leb,                                  \
+       }
+
+/* Layouts explained in AN-379_Marvell_SoC_NFC_ECC */
+static const struct marvell_hw_ecc_layout marvell_nfc_layouts[] = {
+       MARVELL_LAYOUT(  512,   512,  1,  1,  1,  512,  8,  8,  0,  0,  0),
+       MARVELL_LAYOUT( 2048,   512,  1,  1,  1, 2048, 40, 24,  0,  0,  0),
+       MARVELL_LAYOUT( 2048,   512,  4,  1,  1, 2048, 32, 30,  0,  0,  0),
+       MARVELL_LAYOUT( 4096,   512,  4,  2,  2, 2048, 32, 30,  0,  0,  0),
+       MARVELL_LAYOUT( 4096,   512,  8,  5,  4, 1024,  0, 30,  0, 64, 30),
+};
+
+/**
+ * The Nand Flash Controller has up to 4 CE and 2 RB pins. The CE selection
+ * is made by a field in NDCB0 register, and in another field in NDCB2 register.
+ * The datasheet describes the logic with an error: ADDR5 field is once
+ * declared at the beginning of NDCB2, and another time at its end. Because the
+ * ADDR5 field of NDCB2 may be used by other bytes, it would be more logical
+ * to use the last bit of this field instead of the first ones.
+ *
+ * @cs:                        Wanted CE lane.
+ * @ndcb0_csel:                Value of the NDCB0 register with or without the flag
+ *                     selecting the wanted CE lane. This is set once when
+ *                     the Device Tree is probed.
+ * @rb:                        Ready/Busy pin for the flash chip
+ */
+struct marvell_nand_chip_sel {
+       unsigned int cs;
+       u32 ndcb0_csel;
+       unsigned int rb;
+};
+
+/**
+ * NAND chip structure: stores NAND chip device related information
+ *
+ * @chip:              Base NAND chip structure
+ * @node:              Used to store NAND chips into a list
+ * @layout             NAND layout when using hardware ECC
+ * @ndcr:              Controller register value for this NAND chip
+ * @ndtr0:             Timing registers 0 value for this NAND chip
+ * @ndtr1:             Timing registers 1 value for this NAND chip
+ * @selected_die:      Current active CS
+ * @nsels:             Number of CS lines required by the NAND chip
+ * @sels:              Array of CS lines descriptions
+ */
+struct marvell_nand_chip {
+       struct nand_chip chip;
+       struct list_head node;
+       const struct marvell_hw_ecc_layout *layout;
+       u32 ndcr;
+       u32 ndtr0;
+       u32 ndtr1;
+       int addr_cyc;
+       int selected_die;
+       unsigned int nsels;
+       struct marvell_nand_chip_sel sels[0];
+};
+
+static inline struct marvell_nand_chip *to_marvell_nand(struct nand_chip *chip)
+{
+       return container_of(chip, struct marvell_nand_chip, chip);
+}
+
+static inline struct marvell_nand_chip_sel *to_nand_sel(struct marvell_nand_chip
+                                                       *nand)
+{
+       return &nand->sels[nand->selected_die];
+}
+
+/**
+ * NAND controller capabilities for distinction between compatible strings
+ *
+ * @max_cs_nb:         Number of Chip Select lines available
+ * @max_rb_nb:         Number of Ready/Busy lines available
+ * @need_system_controller: Indicates if the SoC needs to have access to the
+ *                      system controller (ie. to enable the NAND controller)
+ * @legacy_of_bindings:        Indicates if DT parsing must be done using the old
+ *                     fashion way
+ * @is_nfcv2:          NFCv2 has numerous enhancements compared to NFCv1, ie.
+ *                     BCH error detection and correction algorithm,
+ *                     NDCB3 register has been added
+ * @use_dma:           Use dma for data transfers
+ */
+struct marvell_nfc_caps {
+       unsigned int max_cs_nb;
+       unsigned int max_rb_nb;
+       bool need_system_controller;
+       bool legacy_of_bindings;
+       bool is_nfcv2;
+       bool use_dma;
+};
+
+/**
+ * NAND controller structure: stores Marvell NAND controller information
+ *
+ * @controller:                Base controller structure
+ * @dev:               Parent device (used to print error messages)
+ * @regs:              NAND controller registers
+ * @ecc_clk:           ECC block clock, two times the NAND controller clock
+ * @complete:          Completion object to wait for NAND controller events
+ * @assigned_cs:       Bitmask describing already assigned CS lines
+ * @chips:             List containing all the NAND chips attached to
+ *                     this NAND controller
+ * @caps:              NAND controller capabilities for each compatible string
+ * @dma_chan:          DMA channel (NFCv1 only)
+ * @dma_buf:           32-bit aligned buffer for DMA transfers (NFCv1 only)
+ */
+struct marvell_nfc {
+       struct nand_hw_control controller;
+       struct device *dev;
+       void __iomem *regs;
+       struct clk *ecc_clk;
+       struct completion complete;
+       unsigned long assigned_cs;
+       struct list_head chips;
+       struct nand_chip *selected_chip;
+       const struct marvell_nfc_caps *caps;
+
+       /* DMA (NFCv1 only) */
+       bool use_dma;
+       struct dma_chan *dma_chan;
+       u8 *dma_buf;
+};
+
+static inline struct marvell_nfc *to_marvell_nfc(struct nand_hw_control *ctrl)
+{
+       return container_of(ctrl, struct marvell_nfc, controller);
+}
+
+/**
+ * NAND controller timings expressed in NAND Controller clock cycles
+ *
+ * @tRP:               ND_nRE pulse width
+ * @tRH:               ND_nRE high duration
+ * @tWP:               ND_nWE pulse time
+ * @tWH:               ND_nWE high duration
+ * @tCS:               Enable signal setup time
+ * @tCH:               Enable signal hold time
+ * @tADL:              Address to write data delay
+ * @tAR:               ND_ALE low to ND_nRE low delay
+ * @tWHR:              ND_nWE high to ND_nRE low for status read
+ * @tRHW:              ND_nRE high duration, read to write delay
+ * @tR:                        ND_nWE high to ND_nRE low for read
+ */
+struct marvell_nfc_timings {
+       /* NDTR0 fields */
+       unsigned int tRP;
+       unsigned int tRH;
+       unsigned int tWP;
+       unsigned int tWH;
+       unsigned int tCS;
+       unsigned int tCH;
+       unsigned int tADL;
+       /* NDTR1 fields */
+       unsigned int tAR;
+       unsigned int tWHR;
+       unsigned int tRHW;
+       unsigned int tR;
+};
+
+/**
+ * Derives a duration in numbers of clock cycles.
+ *
+ * @ps: Duration in pico-seconds
+ * @period_ns:  Clock period in nano-seconds
+ *
+ * Convert the duration in nano-seconds, then divide by the period and
+ * return the number of clock periods.
+ */
+#define TO_CYCLES(ps, period_ns) (DIV_ROUND_UP(ps / 1000, period_ns))
+
+/**
+ * NAND driver structure filled during the parsing of the ->exec_op() subop
+ * subset of instructions.
+ *
+ * @ndcb:              Array of values written to NDCBx registers
+ * @cle_ale_delay_ns:  Optional delay after the last CMD or ADDR cycle
+ * @rdy_timeout_ms:    Timeout for waits on Ready/Busy pin
+ * @rdy_delay_ns:      Optional delay after waiting for the RB pin
+ * @data_delay_ns:     Optional delay after the data xfer
+ * @data_instr_idx:    Index of the data instruction in the subop
+ * @data_instr:                Pointer to the data instruction in the subop
+ */
+struct marvell_nfc_op {
+       u32 ndcb[4];
+       unsigned int cle_ale_delay_ns;
+       unsigned int rdy_timeout_ms;
+       unsigned int rdy_delay_ns;
+       unsigned int data_delay_ns;
+       unsigned int data_instr_idx;
+       const struct nand_op_instr *data_instr;
+};
+
+/*
+ * Internal helper to conditionnally apply a delay (from the above structure,
+ * most of the time).
+ */
+static void cond_delay(unsigned int ns)
+{
+       if (!ns)
+               return;
+
+       if (ns < 10000)
+               ndelay(ns);
+       else
+               udelay(DIV_ROUND_UP(ns, 1000));
+}
+
+/*
+ * The controller has many flags that could generate interrupts, most of them
+ * are disabled and polling is used. For the very slow signals, using interrupts
+ * may relax the CPU charge.
+ */
+static void marvell_nfc_disable_int(struct marvell_nfc *nfc, u32 int_mask)
+{
+       u32 reg;
+
+       /* Writing 1 disables the interrupt */
+       reg = readl_relaxed(nfc->regs + NDCR);
+       writel_relaxed(reg | int_mask, nfc->regs + NDCR);
+}
+
+static void marvell_nfc_enable_int(struct marvell_nfc *nfc, u32 int_mask)
+{
+       u32 reg;
+
+       /* Writing 0 enables the interrupt */
+       reg = readl_relaxed(nfc->regs + NDCR);
+       writel_relaxed(reg & ~int_mask, nfc->regs + NDCR);
+}
+
+static void marvell_nfc_clear_int(struct marvell_nfc *nfc, u32 int_mask)
+{
+       writel_relaxed(int_mask, nfc->regs + NDSR);
+}
+
+static void marvell_nfc_force_byte_access(struct nand_chip *chip,
+                                         bool force_8bit)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       u32 ndcr;
+
+       /*
+        * Callers of this function do not verify if the NAND is using a 16-bit
+        * an 8-bit bus for normal operations, so we need to take care of that
+        * here by leaving the configuration unchanged if the NAND does not have
+        * the NAND_BUSWIDTH_16 flag set.
+        */
+       if (!(chip->options & NAND_BUSWIDTH_16))
+               return;
+
+       ndcr = readl_relaxed(nfc->regs + NDCR);
+
+       if (force_8bit)
+               ndcr &= ~(NDCR_DWIDTH_M | NDCR_DWIDTH_C);
+       else
+               ndcr |= NDCR_DWIDTH_M | NDCR_DWIDTH_C;
+
+       writel_relaxed(ndcr, nfc->regs + NDCR);
+}
+
+static int marvell_nfc_wait_ndrun(struct nand_chip *chip)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       u32 val;
+       int ret;
+
+       /*
+        * The command is being processed, wait for the ND_RUN bit to be
+        * cleared by the NFC. If not, we must clear it by hand.
+        */
+       ret = readl_relaxed_poll_timeout(nfc->regs + NDCR, val,
+                                        (val & NDCR_ND_RUN) == 0,
+                                        POLL_PERIOD, POLL_TIMEOUT);
+       if (ret) {
+               dev_err(nfc->dev, "Timeout on NAND controller run mode\n");
+               writel_relaxed(readl(nfc->regs + NDCR) & ~NDCR_ND_RUN,
+                              nfc->regs + NDCR);
+               return ret;
+       }
+
+       return 0;
+}
+
+/*
+ * Any time a command has to be sent to the controller, the following sequence
+ * has to be followed:
+ * - call marvell_nfc_prepare_cmd()
+ *      -> activate the ND_RUN bit that will kind of 'start a job'
+ *      -> wait the signal indicating the NFC is waiting for a command
+ * - send the command (cmd and address cycles)
+ * - enventually send or receive the data
+ * - call marvell_nfc_end_cmd() with the corresponding flag
+ *      -> wait the flag to be triggered or cancel the job with a timeout
+ *
+ * The following helpers are here to factorize the code a bit so that
+ * specialized functions responsible for executing the actual NAND
+ * operations do not have to replicate the same code blocks.
+ */
+static int marvell_nfc_prepare_cmd(struct nand_chip *chip)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       u32 ndcr, val;
+       int ret;
+
+       /* Poll ND_RUN and clear NDSR before issuing any command */
+       ret = marvell_nfc_wait_ndrun(chip);
+       if (ret) {
+               dev_err(nfc->dev, "Last operation did not suceed\n");
+               return ret;
+       }
+
+       ndcr = readl_relaxed(nfc->regs + NDCR);
+       writel_relaxed(readl(nfc->regs + NDSR), nfc->regs + NDSR);
+
+       /* Assert ND_RUN bit and wait the NFC to be ready */
+       writel_relaxed(ndcr | NDCR_ND_RUN, nfc->regs + NDCR);
+       ret = readl_relaxed_poll_timeout(nfc->regs + NDSR, val,
+                                        val & NDSR_WRCMDREQ,
+                                        POLL_PERIOD, POLL_TIMEOUT);
+       if (ret) {
+               dev_err(nfc->dev, "Timeout on WRCMDRE\n");
+               return -ETIMEDOUT;
+       }
+
+       /* Command may be written, clear WRCMDREQ status bit */
+       writel_relaxed(NDSR_WRCMDREQ, nfc->regs + NDSR);
+
+       return 0;
+}
+
+static void marvell_nfc_send_cmd(struct nand_chip *chip,
+                                struct marvell_nfc_op *nfc_op)
+{
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+
+       dev_dbg(nfc->dev, "\nNDCR:  0x%08x\n"
+               "NDCB0: 0x%08x\nNDCB1: 0x%08x\nNDCB2: 0x%08x\nNDCB3: 0x%08x\n",
+               (u32)readl_relaxed(nfc->regs + NDCR), nfc_op->ndcb[0],
+               nfc_op->ndcb[1], nfc_op->ndcb[2], nfc_op->ndcb[3]);
+
+       writel_relaxed(to_nand_sel(marvell_nand)->ndcb0_csel | nfc_op->ndcb[0],
+                      nfc->regs + NDCB0);
+       writel_relaxed(nfc_op->ndcb[1], nfc->regs + NDCB0);
+       writel(nfc_op->ndcb[2], nfc->regs + NDCB0);
+
+       /*
+        * Write NDCB0 four times only if LEN_OVRD is set or if ADDR6 or ADDR7
+        * fields are used (only available on NFCv2).
+        */
+       if (nfc_op->ndcb[0] & NDCB0_LEN_OVRD ||
+           NDCB0_ADDR_GET_NUM_CYC(nfc_op->ndcb[0]) >= 6) {
+               if (!WARN_ON_ONCE(!nfc->caps->is_nfcv2))
+                       writel(nfc_op->ndcb[3], nfc->regs + NDCB0);
+       }
+}
+
+static int marvell_nfc_end_cmd(struct nand_chip *chip, int flag,
+                              const char *label)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       u32 val;
+       int ret;
+
+       ret = readl_relaxed_poll_timeout(nfc->regs + NDSR, val,
+                                        val & flag,
+                                        POLL_PERIOD, POLL_TIMEOUT);
+
+       if (ret) {
+               dev_err(nfc->dev, "Timeout on %s (NDSR: 0x%08x)\n",
+                       label, val);
+               if (nfc->dma_chan)
+                       dmaengine_terminate_all(nfc->dma_chan);
+               return ret;
+       }
+
+       /*
+        * DMA function uses this helper to poll on CMDD bits without wanting
+        * them to be cleared.
+        */
+       if (nfc->use_dma && (readl_relaxed(nfc->regs + NDCR) & NDCR_DMA_EN))
+               return 0;
+
+       writel_relaxed(flag, nfc->regs + NDSR);
+
+       return 0;
+}
+
+static int marvell_nfc_wait_cmdd(struct nand_chip *chip)
+{
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       int cs_flag = NDSR_CMDD(to_nand_sel(marvell_nand)->ndcb0_csel);
+
+       return marvell_nfc_end_cmd(chip, cs_flag, "CMDD");
+}
+
+static int marvell_nfc_wait_op(struct nand_chip *chip, unsigned int timeout_ms)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       int ret;
+
+       /* Timeout is expressed in ms */
+       if (!timeout_ms)
+               timeout_ms = IRQ_TIMEOUT;
+
+       init_completion(&nfc->complete);
+
+       marvell_nfc_enable_int(nfc, NDCR_RDYM);
+       ret = wait_for_completion_timeout(&nfc->complete,
+                                         msecs_to_jiffies(timeout_ms));
+       marvell_nfc_disable_int(nfc, NDCR_RDYM);
+       marvell_nfc_clear_int(nfc, NDSR_RDY(0) | NDSR_RDY(1));
+       if (!ret) {
+               dev_err(nfc->dev, "Timeout waiting for RB signal\n");
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
+static void marvell_nfc_select_chip(struct mtd_info *mtd, int die_nr)
+{
+       struct nand_chip *chip = mtd_to_nand(mtd);
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       u32 ndcr_generic;
+
+       if (chip == nfc->selected_chip && die_nr == marvell_nand->selected_die)
+               return;
+
+       if (die_nr < 0 || die_nr >= marvell_nand->nsels) {
+               nfc->selected_chip = NULL;
+               marvell_nand->selected_die = -1;
+               return;
+       }
+
+       /*
+        * Do not change the timing registers when using the DT property
+        * marvell,nand-keep-config; in that case ->ndtr0 and ->ndtr1 from the
+        * marvell_nand structure are supposedly empty.
+        */
+       writel_relaxed(marvell_nand->ndtr0, nfc->regs + NDTR0);
+       writel_relaxed(marvell_nand->ndtr1, nfc->regs + NDTR1);
+
+       /*
+        * Reset the NDCR register to a clean state for this particular chip,
+        * also clear ND_RUN bit.
+        */
+       ndcr_generic = readl_relaxed(nfc->regs + NDCR) &
+                      NDCR_GENERIC_FIELDS_MASK & ~NDCR_ND_RUN;
+       writel_relaxed(ndcr_generic | marvell_nand->ndcr, nfc->regs + NDCR);
+
+       /* Also reset the interrupt status register */
+       marvell_nfc_clear_int(nfc, NDCR_ALL_INT);
+
+       nfc->selected_chip = chip;
+       marvell_nand->selected_die = die_nr;
+}
+
+static irqreturn_t marvell_nfc_isr(int irq, void *dev_id)
+{
+       struct marvell_nfc *nfc = dev_id;
+       u32 st = readl_relaxed(nfc->regs + NDSR);
+       u32 ien = (~readl_relaxed(nfc->regs + NDCR)) & NDCR_ALL_INT;
+
+       /*
+        * RDY interrupt mask is one bit in NDCR while there are two status
+        * bit in NDSR (RDY[cs0/cs2] and RDY[cs1/cs3]).
+        */
+       if (st & NDSR_RDY(1))
+               st |= NDSR_RDY(0);
+
+       if (!(st & ien))
+               return IRQ_NONE;
+
+       marvell_nfc_disable_int(nfc, st & NDCR_ALL_INT);
+
+       if (!(st & (NDSR_RDDREQ | NDSR_WRDREQ | NDSR_WRCMDREQ)))
+               complete(&nfc->complete);
+
+       return IRQ_HANDLED;
+}
+
+/* HW ECC related functions */
+static void marvell_nfc_enable_hw_ecc(struct nand_chip *chip)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       u32 ndcr = readl_relaxed(nfc->regs + NDCR);
+
+       if (!(ndcr & NDCR_ECC_EN)) {
+               writel_relaxed(ndcr | NDCR_ECC_EN, nfc->regs + NDCR);
+
+               /*
+                * When enabling BCH, set threshold to 0 to always know the
+                * number of corrected bitflips.
+                */
+               if (chip->ecc.algo == NAND_ECC_BCH)
+                       writel_relaxed(NDECCCTRL_BCH_EN, nfc->regs + NDECCCTRL);
+       }
+}
+
+static void marvell_nfc_disable_hw_ecc(struct nand_chip *chip)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       u32 ndcr = readl_relaxed(nfc->regs + NDCR);
+
+       if (ndcr & NDCR_ECC_EN) {
+               writel_relaxed(ndcr & ~NDCR_ECC_EN, nfc->regs + NDCR);
+               if (chip->ecc.algo == NAND_ECC_BCH)
+                       writel_relaxed(0, nfc->regs + NDECCCTRL);
+       }
+}
+
+/* DMA related helpers */
+static void marvell_nfc_enable_dma(struct marvell_nfc *nfc)
+{
+       u32 reg;
+
+       reg = readl_relaxed(nfc->regs + NDCR);
+       writel_relaxed(reg | NDCR_DMA_EN, nfc->regs + NDCR);
+}
+
+static void marvell_nfc_disable_dma(struct marvell_nfc *nfc)
+{
+       u32 reg;
+
+       reg = readl_relaxed(nfc->regs + NDCR);
+       writel_relaxed(reg & ~NDCR_DMA_EN, nfc->regs + NDCR);
+}
+
+/* Read/write PIO/DMA accessors */
+static int marvell_nfc_xfer_data_dma(struct marvell_nfc *nfc,
+                                    enum dma_data_direction direction,
+                                    unsigned int len)
+{
+       unsigned int dma_len = min_t(int, ALIGN(len, 32), MAX_CHUNK_SIZE);
+       struct dma_async_tx_descriptor *tx;
+       struct scatterlist sg;
+       dma_cookie_t cookie;
+       int ret;
+
+       marvell_nfc_enable_dma(nfc);
+       /* Prepare the DMA transfer */
+       sg_init_one(&sg, nfc->dma_buf, dma_len);
+       dma_map_sg(nfc->dma_chan->device->dev, &sg, 1, direction);
+       tx = dmaengine_prep_slave_sg(nfc->dma_chan, &sg, 1,
+                                    direction == DMA_FROM_DEVICE ?
+                                    DMA_DEV_TO_MEM : DMA_MEM_TO_DEV,
+                                    DMA_PREP_INTERRUPT);
+       if (!tx) {
+               dev_err(nfc->dev, "Could not prepare DMA S/G list\n");
+               return -ENXIO;
+       }
+
+       /* Do the task and wait for it to finish */
+       cookie = dmaengine_submit(tx);
+       ret = dma_submit_error(cookie);
+       if (ret)
+               return -EIO;
+
+       dma_async_issue_pending(nfc->dma_chan);
+       ret = marvell_nfc_wait_cmdd(nfc->selected_chip);
+       dma_unmap_sg(nfc->dma_chan->device->dev, &sg, 1, direction);
+       marvell_nfc_disable_dma(nfc);
+       if (ret) {
+               dev_err(nfc->dev, "Timeout waiting for DMA (status: %d)\n",
+                       dmaengine_tx_status(nfc->dma_chan, cookie, NULL));
+               dmaengine_terminate_all(nfc->dma_chan);
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
+static int marvell_nfc_xfer_data_in_pio(struct marvell_nfc *nfc, u8 *in,
+                                       unsigned int len)
+{
+       unsigned int last_len = len % FIFO_DEPTH;
+       unsigned int last_full_offset = round_down(len, FIFO_DEPTH);
+       int i;
+
+       for (i = 0; i < last_full_offset; i += FIFO_DEPTH)
+               ioread32_rep(nfc->regs + NDDB, in + i, FIFO_REP(FIFO_DEPTH));
+
+       if (last_len) {
+               u8 tmp_buf[FIFO_DEPTH];
+
+               ioread32_rep(nfc->regs + NDDB, tmp_buf, FIFO_REP(FIFO_DEPTH));
+               memcpy(in + last_full_offset, tmp_buf, last_len);
+       }
+
+       return 0;
+}
+
+static int marvell_nfc_xfer_data_out_pio(struct marvell_nfc *nfc, const u8 *out,
+                                        unsigned int len)
+{
+       unsigned int last_len = len % FIFO_DEPTH;
+       unsigned int last_full_offset = round_down(len, FIFO_DEPTH);
+       int i;
+
+       for (i = 0; i < last_full_offset; i += FIFO_DEPTH)
+               iowrite32_rep(nfc->regs + NDDB, out + i, FIFO_REP(FIFO_DEPTH));
+
+       if (last_len) {
+               u8 tmp_buf[FIFO_DEPTH];
+
+               memcpy(tmp_buf, out + last_full_offset, last_len);
+               iowrite32_rep(nfc->regs + NDDB, tmp_buf, FIFO_REP(FIFO_DEPTH));
+       }
+
+       return 0;
+}
+
+static void marvell_nfc_check_empty_chunk(struct nand_chip *chip,
+                                         u8 *data, int data_len,
+                                         u8 *spare, int spare_len,
+                                         u8 *ecc, int ecc_len,
+                                         unsigned int *max_bitflips)
+{
+       struct mtd_info *mtd = nand_to_mtd(chip);
+       int bf;
+
+       /*
+        * Blank pages (all 0xFF) that have not been written may be recognized
+        * as bad if bitflips occur, so whenever an uncorrectable error occurs,
+        * check if the entire page (with ECC bytes) is actually blank or not.
+        */
+       if (!data)
+               data_len = 0;
+       if (!spare)
+               spare_len = 0;
+       if (!ecc)
+               ecc_len = 0;
+
+       bf = nand_check_erased_ecc_chunk(data, data_len, ecc, ecc_len,
+                                        spare, spare_len, chip->ecc.strength);
+       if (bf < 0) {
+               mtd->ecc_stats.failed++;
+               return;
+       }
+
+       /* Update the stats and max_bitflips */
+       mtd->ecc_stats.corrected += bf;
+       *max_bitflips = max_t(unsigned int, *max_bitflips, bf);
+}
+
+/*
+ * Check a chunk is correct or not according to hardware ECC engine.
+ * mtd->ecc_stats.corrected is updated, as well as max_bitflips, however
+ * mtd->ecc_stats.failure is not, the function will instead return a non-zero
+ * value indicating that a check on the emptyness of the subpage must be
+ * performed before declaring the subpage corrupted.
+ */
+static int marvell_nfc_hw_ecc_correct(struct nand_chip *chip,
+                                     unsigned int *max_bitflips)
+{
+       struct mtd_info *mtd = nand_to_mtd(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       int bf = 0;
+       u32 ndsr;
+
+       ndsr = readl_relaxed(nfc->regs + NDSR);
+
+       /* Check uncorrectable error flag */
+       if (ndsr & NDSR_UNCERR) {
+               writel_relaxed(ndsr, nfc->regs + NDSR);
+
+               /*
+                * Do not increment ->ecc_stats.failed now, instead, return a
+                * non-zero value to indicate that this chunk was apparently
+                * bad, and it should be check to see if it empty or not. If
+                * the chunk (with ECC bytes) is not declared empty, the calling
+                * function must increment the failure count.
+                */
+               return -EBADMSG;
+       }
+
+       /* Check correctable error flag */
+       if (ndsr & NDSR_CORERR) {
+               writel_relaxed(ndsr, nfc->regs + NDSR);
+
+               if (chip->ecc.algo == NAND_ECC_BCH)
+                       bf = NDSR_ERRCNT(ndsr);
+               else
+                       bf = 1;
+       }
+
+       /* Update the stats and max_bitflips */
+       mtd->ecc_stats.corrected += bf;
+       *max_bitflips = max_t(unsigned int, *max_bitflips, bf);
+
+       return 0;
+}
+
+/* Hamming read helpers */
+static int marvell_nfc_hw_ecc_hmg_do_read_page(struct nand_chip *chip,
+                                              u8 *data_buf, u8 *oob_buf,
+                                              bool raw, int page)
+{
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       struct marvell_nfc_op nfc_op = {
+               .ndcb[0] = NDCB0_CMD_TYPE(TYPE_READ) |
+                          NDCB0_ADDR_CYC(marvell_nand->addr_cyc) |
+                          NDCB0_DBC |
+                          NDCB0_CMD1(NAND_CMD_READ0) |
+                          NDCB0_CMD2(NAND_CMD_READSTART),
+               .ndcb[1] = NDCB1_ADDRS_PAGE(page),
+               .ndcb[2] = NDCB2_ADDR5_PAGE(page),
+       };
+       unsigned int oob_bytes = lt->spare_bytes + (raw ? lt->ecc_bytes : 0);
+       int ret;
+
+       /* NFCv2 needs more information about the operation being executed */
+       if (nfc->caps->is_nfcv2)
+               nfc_op.ndcb[0] |= NDCB0_CMD_XTYPE(XTYPE_MONOLITHIC_RW);
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_end_cmd(chip, NDSR_RDDREQ,
+                                 "RDDREQ while draining FIFO (data/oob)");
+       if (ret)
+               return ret;
+
+       /*
+        * Read the page then the OOB area. Unlike what is shown in current
+        * documentation, spare bytes are protected by the ECC engine, and must
+        * be at the beginning of the OOB area or running this driver on legacy
+        * systems will prevent the discovery of the BBM/BBT.
+        */
+       if (nfc->use_dma) {
+               marvell_nfc_xfer_data_dma(nfc, DMA_FROM_DEVICE,
+                                         lt->data_bytes + oob_bytes);
+               memcpy(data_buf, nfc->dma_buf, lt->data_bytes);
+               memcpy(oob_buf, nfc->dma_buf + lt->data_bytes, oob_bytes);
+       } else {
+               marvell_nfc_xfer_data_in_pio(nfc, data_buf, lt->data_bytes);
+               marvell_nfc_xfer_data_in_pio(nfc, oob_buf, oob_bytes);
+       }
+
+       ret = marvell_nfc_wait_cmdd(chip);
+
+       return ret;
+}
+
+static int marvell_nfc_hw_ecc_hmg_read_page_raw(struct mtd_info *mtd,
+                                               struct nand_chip *chip, u8 *buf,
+                                               int oob_required, int page)
+{
+       return marvell_nfc_hw_ecc_hmg_do_read_page(chip, buf, chip->oob_poi,
+                                                  true, page);
+}
+
+static int marvell_nfc_hw_ecc_hmg_read_page(struct mtd_info *mtd,
+                                           struct nand_chip *chip,
+                                           u8 *buf, int oob_required,
+                                           int page)
+{
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       unsigned int full_sz = lt->data_bytes + lt->spare_bytes + lt->ecc_bytes;
+       int max_bitflips = 0, ret;
+       u8 *raw_buf;
+
+       marvell_nfc_enable_hw_ecc(chip);
+       marvell_nfc_hw_ecc_hmg_do_read_page(chip, buf, chip->oob_poi, false,
+                                           page);
+       ret = marvell_nfc_hw_ecc_correct(chip, &max_bitflips);
+       marvell_nfc_disable_hw_ecc(chip);
+
+       if (!ret)
+               return max_bitflips;
+
+       /*
+        * When ECC failures are detected, check if the full page has been
+        * written or not. Ignore the failure if it is actually empty.
+        */
+       raw_buf = kmalloc(full_sz, GFP_KERNEL);
+       if (!raw_buf)
+               return -ENOMEM;
+
+       marvell_nfc_hw_ecc_hmg_do_read_page(chip, raw_buf, raw_buf +
+                                           lt->data_bytes, true, page);
+       marvell_nfc_check_empty_chunk(chip, raw_buf, full_sz, NULL, 0, NULL, 0,
+                                     &max_bitflips);
+       kfree(raw_buf);
+
+       return max_bitflips;
+}
+
+/*
+ * Spare area in Hamming layouts is not protected by the ECC engine (even if
+ * it appears before the ECC bytes when reading), the ->read_oob_raw() function
+ * also stands for ->read_oob().
+ */
+static int marvell_nfc_hw_ecc_hmg_read_oob_raw(struct mtd_info *mtd,
+                                              struct nand_chip *chip, int page)
+{
+       /* Invalidate page cache */
+       chip->pagebuf = -1;
+
+       return marvell_nfc_hw_ecc_hmg_do_read_page(chip, chip->data_buf,
+                                                  chip->oob_poi, true, page);
+}
+
+/* Hamming write helpers */
+static int marvell_nfc_hw_ecc_hmg_do_write_page(struct nand_chip *chip,
+                                               const u8 *data_buf,
+                                               const u8 *oob_buf, bool raw,
+                                               int page)
+{
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       struct marvell_nfc_op nfc_op = {
+               .ndcb[0] = NDCB0_CMD_TYPE(TYPE_WRITE) |
+                          NDCB0_ADDR_CYC(marvell_nand->addr_cyc) |
+                          NDCB0_CMD1(NAND_CMD_SEQIN) |
+                          NDCB0_CMD2(NAND_CMD_PAGEPROG) |
+                          NDCB0_DBC,
+               .ndcb[1] = NDCB1_ADDRS_PAGE(page),
+               .ndcb[2] = NDCB2_ADDR5_PAGE(page),
+       };
+       unsigned int oob_bytes = lt->spare_bytes + (raw ? lt->ecc_bytes : 0);
+       int ret;
+
+       /* NFCv2 needs more information about the operation being executed */
+       if (nfc->caps->is_nfcv2)
+               nfc_op.ndcb[0] |= NDCB0_CMD_XTYPE(XTYPE_MONOLITHIC_RW);
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_end_cmd(chip, NDSR_WRDREQ,
+                                 "WRDREQ while loading FIFO (data)");
+       if (ret)
+               return ret;
+
+       /* Write the page then the OOB area */
+       if (nfc->use_dma) {
+               memcpy(nfc->dma_buf, data_buf, lt->data_bytes);
+               memcpy(nfc->dma_buf + lt->data_bytes, oob_buf, oob_bytes);
+               marvell_nfc_xfer_data_dma(nfc, DMA_TO_DEVICE, lt->data_bytes +
+                                         lt->ecc_bytes + lt->spare_bytes);
+       } else {
+               marvell_nfc_xfer_data_out_pio(nfc, data_buf, lt->data_bytes);
+               marvell_nfc_xfer_data_out_pio(nfc, oob_buf, oob_bytes);
+       }
+
+       ret = marvell_nfc_wait_cmdd(chip);
+       if (ret)
+               return ret;
+
+       ret = marvell_nfc_wait_op(chip,
+                                 chip->data_interface.timings.sdr.tPROG_max);
+       return ret;
+}
+
+static int marvell_nfc_hw_ecc_hmg_write_page_raw(struct mtd_info *mtd,
+                                                struct nand_chip *chip,
+                                                const u8 *buf,
+                                                int oob_required, int page)
+{
+       return marvell_nfc_hw_ecc_hmg_do_write_page(chip, buf, chip->oob_poi,
+                                                   true, page);
+}
+
+static int marvell_nfc_hw_ecc_hmg_write_page(struct mtd_info *mtd,
+                                            struct nand_chip *chip,
+                                            const u8 *buf,
+                                            int oob_required, int page)
+{
+       int ret;
+
+       marvell_nfc_enable_hw_ecc(chip);
+       ret = marvell_nfc_hw_ecc_hmg_do_write_page(chip, buf, chip->oob_poi,
+                                                  false, page);
+       marvell_nfc_disable_hw_ecc(chip);
+
+       return ret;
+}
+
+/*
+ * Spare area in Hamming layouts is not protected by the ECC engine (even if
+ * it appears before the ECC bytes when reading), the ->write_oob_raw() function
+ * also stands for ->write_oob().
+ */
+static int marvell_nfc_hw_ecc_hmg_write_oob_raw(struct mtd_info *mtd,
+                                               struct nand_chip *chip,
+                                               int page)
+{
+       /* Invalidate page cache */
+       chip->pagebuf = -1;
+
+       memset(chip->data_buf, 0xFF, mtd->writesize);
+
+       return marvell_nfc_hw_ecc_hmg_do_write_page(chip, chip->data_buf,
+                                                   chip->oob_poi, true, page);
+}
+
+/* BCH read helpers */
+static int marvell_nfc_hw_ecc_bch_read_page_raw(struct mtd_info *mtd,
+                                               struct nand_chip *chip, u8 *buf,
+                                               int oob_required, int page)
+{
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       u8 *oob = chip->oob_poi;
+       int chunk_size = lt->data_bytes + lt->spare_bytes + lt->ecc_bytes;
+       int ecc_offset = (lt->full_chunk_cnt * lt->spare_bytes) +
+               lt->last_spare_bytes;
+       int data_len = lt->data_bytes;
+       int spare_len = lt->spare_bytes;
+       int ecc_len = lt->ecc_bytes;
+       int chunk;
+
+       if (oob_required)
+               memset(chip->oob_poi, 0xFF, mtd->oobsize);
+
+       nand_read_page_op(chip, page, 0, NULL, 0);
+
+       for (chunk = 0; chunk < lt->nchunks; chunk++) {
+               /* Update last chunk length */
+               if (chunk >= lt->full_chunk_cnt) {
+                       data_len = lt->last_data_bytes;
+                       spare_len = lt->last_spare_bytes;
+                       ecc_len = lt->last_ecc_bytes;
+               }
+
+               /* Read data bytes*/
+               nand_change_read_column_op(chip, chunk * chunk_size,
+                                          buf + (lt->data_bytes * chunk),
+                                          data_len, false);
+
+               /* Read spare bytes */
+               nand_read_data_op(chip, oob + (lt->spare_bytes * chunk),
+                                 spare_len, false);
+
+               /* Read ECC bytes */
+               nand_read_data_op(chip, oob + ecc_offset +
+                                 (ALIGN(lt->ecc_bytes, 32) * chunk),
+                                 ecc_len, false);
+       }
+
+       return 0;
+}
+
+static void marvell_nfc_hw_ecc_bch_read_chunk(struct nand_chip *chip, int chunk,
+                                             u8 *data, unsigned int data_len,
+                                             u8 *spare, unsigned int spare_len,
+                                             int page)
+{
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       int i, ret;
+       struct marvell_nfc_op nfc_op = {
+               .ndcb[0] = NDCB0_CMD_TYPE(TYPE_READ) |
+                          NDCB0_ADDR_CYC(marvell_nand->addr_cyc) |
+                          NDCB0_LEN_OVRD,
+               .ndcb[1] = NDCB1_ADDRS_PAGE(page),
+               .ndcb[2] = NDCB2_ADDR5_PAGE(page),
+               .ndcb[3] = data_len + spare_len,
+       };
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return;
+
+       if (chunk == 0)
+               nfc_op.ndcb[0] |= NDCB0_DBC |
+                                 NDCB0_CMD1(NAND_CMD_READ0) |
+                                 NDCB0_CMD2(NAND_CMD_READSTART);
+
+       /*
+        * Trigger the naked read operation only on the last chunk.
+        * Otherwise, use monolithic read.
+        */
+       if (lt->nchunks == 1 || (chunk < lt->nchunks - 1))
+               nfc_op.ndcb[0] |= NDCB0_CMD_XTYPE(XTYPE_MONOLITHIC_RW);
+       else
+               nfc_op.ndcb[0] |= NDCB0_CMD_XTYPE(XTYPE_LAST_NAKED_RW);
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+
+       /*
+        * According to the datasheet, when reading from NDDB
+        * with BCH enabled, after each 32 bytes reads, we
+        * have to make sure that the NDSR.RDDREQ bit is set.
+        *
+        * Drain the FIFO, 8 32-bit reads at a time, and skip
+        * the polling on the last read.
+        *
+        * Length is a multiple of 32 bytes, hence it is a multiple of 8 too.
+        */
+       for (i = 0; i < data_len; i += FIFO_DEPTH * BCH_SEQ_READS) {
+               marvell_nfc_end_cmd(chip, NDSR_RDDREQ,
+                                   "RDDREQ while draining FIFO (data)");
+               marvell_nfc_xfer_data_in_pio(nfc, data,
+                                            FIFO_DEPTH * BCH_SEQ_READS);
+               data += FIFO_DEPTH * BCH_SEQ_READS;
+       }
+
+       for (i = 0; i < spare_len; i += FIFO_DEPTH * BCH_SEQ_READS) {
+               marvell_nfc_end_cmd(chip, NDSR_RDDREQ,
+                                   "RDDREQ while draining FIFO (OOB)");
+               marvell_nfc_xfer_data_in_pio(nfc, spare,
+                                            FIFO_DEPTH * BCH_SEQ_READS);
+               spare += FIFO_DEPTH * BCH_SEQ_READS;
+       }
+}
+
+static int marvell_nfc_hw_ecc_bch_read_page(struct mtd_info *mtd,
+                                           struct nand_chip *chip,
+                                           u8 *buf, int oob_required,
+                                           int page)
+{
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       int data_len = lt->data_bytes, spare_len = lt->spare_bytes, ecc_len;
+       u8 *data = buf, *spare = chip->oob_poi, *ecc;
+       int max_bitflips = 0;
+       u32 failure_mask = 0;
+       int chunk, ecc_offset_in_page, ret;
+
+       /*
+        * With BCH, OOB is not fully used (and thus not read entirely), not
+        * expected bytes could show up at the end of the OOB buffer if not
+        * explicitly erased.
+        */
+       if (oob_required)
+               memset(chip->oob_poi, 0xFF, mtd->oobsize);
+
+       marvell_nfc_enable_hw_ecc(chip);
+
+       for (chunk = 0; chunk < lt->nchunks; chunk++) {
+               /* Update length for the last chunk */
+               if (chunk >= lt->full_chunk_cnt) {
+                       data_len = lt->last_data_bytes;
+                       spare_len = lt->last_spare_bytes;
+               }
+
+               /* Read the chunk and detect number of bitflips */
+               marvell_nfc_hw_ecc_bch_read_chunk(chip, chunk, data, data_len,
+                                                 spare, spare_len, page);
+               ret = marvell_nfc_hw_ecc_correct(chip, &max_bitflips);
+               if (ret)
+                       failure_mask |= BIT(chunk);
+
+               data += data_len;
+               spare += spare_len;
+       }
+
+       marvell_nfc_disable_hw_ecc(chip);
+
+       if (!failure_mask)
+               return max_bitflips;
+
+       /*
+        * Please note that dumping the ECC bytes during a normal read with OOB
+        * area would add a significant overhead as ECC bytes are "consumed" by
+        * the controller in normal mode and must be re-read in raw mode. To
+        * avoid dropping the performances, we prefer not to include them. The
+        * user should re-read the page in raw mode if ECC bytes are required.
+        *
+        * However, for any subpage read error reported by ->correct(), the ECC
+        * bytes must be read in raw mode and the full subpage must be checked
+        * to see if it is entirely empty of if there was an actual error.
+        */
+       for (chunk = 0; chunk < lt->nchunks; chunk++) {
+               /* No failure reported for this chunk, move to the next one */
+               if (!(failure_mask & BIT(chunk)))
+                       continue;
+
+               /* Derive ECC bytes positions (in page/buffer) and length */
+               ecc = chip->oob_poi +
+                       (lt->full_chunk_cnt * lt->spare_bytes) +
+                       lt->last_spare_bytes +
+                       (chunk * ALIGN(lt->ecc_bytes, 32));
+               ecc_offset_in_page =
+                       (chunk * (lt->data_bytes + lt->spare_bytes +
+                                 lt->ecc_bytes)) +
+                       (chunk < lt->full_chunk_cnt ?
+                        lt->data_bytes + lt->spare_bytes :
+                        lt->last_data_bytes + lt->last_spare_bytes);
+               ecc_len = chunk < lt->full_chunk_cnt ?
+                       lt->ecc_bytes : lt->last_ecc_bytes;
+
+               /* Do the actual raw read of the ECC bytes */
+               nand_change_read_column_op(chip, ecc_offset_in_page,
+                                          ecc, ecc_len, false);
+
+               /* Derive data/spare bytes positions (in buffer) and length */
+               data = buf + (chunk * lt->data_bytes);
+               data_len = chunk < lt->full_chunk_cnt ?
+                       lt->data_bytes : lt->last_data_bytes;
+               spare = chip->oob_poi + (chunk * (lt->spare_bytes +
+                                                 lt->ecc_bytes));
+               spare_len = chunk < lt->full_chunk_cnt ?
+                       lt->spare_bytes : lt->last_spare_bytes;
+
+               /* Check the entire chunk (data + spare + ecc) for emptyness */
+               marvell_nfc_check_empty_chunk(chip, data, data_len, spare,
+                                             spare_len, ecc, ecc_len,
+                                             &max_bitflips);
+       }
+
+       return max_bitflips;
+}
+
+static int marvell_nfc_hw_ecc_bch_read_oob_raw(struct mtd_info *mtd,
+                                              struct nand_chip *chip, int page)
+{
+       /* Invalidate page cache */
+       chip->pagebuf = -1;
+
+       return chip->ecc.read_page_raw(mtd, chip, chip->data_buf, true, page);
+}
+
+static int marvell_nfc_hw_ecc_bch_read_oob(struct mtd_info *mtd,
+                                          struct nand_chip *chip, int page)
+{
+       /* Invalidate page cache */
+       chip->pagebuf = -1;
+
+       return chip->ecc.read_page(mtd, chip, chip->data_buf, true, page);
+}
+
+/* BCH write helpers */
+static int marvell_nfc_hw_ecc_bch_write_page_raw(struct mtd_info *mtd,
+                                                struct nand_chip *chip,
+                                                const u8 *buf,
+                                                int oob_required, int page)
+{
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       int full_chunk_size = lt->data_bytes + lt->spare_bytes + lt->ecc_bytes;
+       int data_len = lt->data_bytes;
+       int spare_len = lt->spare_bytes;
+       int ecc_len = lt->ecc_bytes;
+       int oob_len = spare_len + ecc_len;
+       int spare_offset = 0;
+       int ecc_offset = (lt->full_chunk_cnt * lt->spare_bytes) +
+               lt->last_spare_bytes;
+       int chunk;
+
+       nand_prog_page_begin_op(chip, page, 0, NULL, 0);
+
+       for (chunk = 0; chunk < lt->nchunks; chunk++) {
+               if (chunk >= lt->full_chunk_cnt) {
+                       data_len = lt->last_data_bytes;
+                       spare_len = lt->last_spare_bytes;
+                       ecc_len = lt->last_ecc_bytes;
+                       oob_len = spare_len + ecc_len;
+               }
+
+               /* Point to the column of the next chunk */
+               nand_change_write_column_op(chip, chunk * full_chunk_size,
+                                           NULL, 0, false);
+
+               /* Write the data */
+               nand_write_data_op(chip, buf + (chunk * lt->data_bytes),
+                                  data_len, false);
+
+               if (!oob_required)
+                       continue;
+
+               /* Write the spare bytes */
+               if (spare_len)
+                       nand_write_data_op(chip, chip->oob_poi + spare_offset,
+                                          spare_len, false);
+
+               /* Write the ECC bytes */
+               if (ecc_len)
+                       nand_write_data_op(chip, chip->oob_poi + ecc_offset,
+                                          ecc_len, false);
+
+               spare_offset += spare_len;
+               ecc_offset += ALIGN(ecc_len, 32);
+       }
+
+       return nand_prog_page_end_op(chip);
+}
+
+static int
+marvell_nfc_hw_ecc_bch_write_chunk(struct nand_chip *chip, int chunk,
+                                  const u8 *data, unsigned int data_len,
+                                  const u8 *spare, unsigned int spare_len,
+                                  int page)
+{
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       int ret;
+       struct marvell_nfc_op nfc_op = {
+               .ndcb[0] = NDCB0_CMD_TYPE(TYPE_WRITE) | NDCB0_LEN_OVRD,
+               .ndcb[3] = data_len + spare_len,
+       };
+
+       /*
+        * First operation dispatches the CMD_SEQIN command, issue the address
+        * cycles and asks for the first chunk of data.
+        * All operations in the middle (if any) will issue a naked write and
+        * also ask for data.
+        * Last operation (if any) asks for the last chunk of data through a
+        * last naked write.
+        */
+       if (chunk == 0) {
+               nfc_op.ndcb[0] |= NDCB0_CMD_XTYPE(XTYPE_WRITE_DISPATCH) |
+                                 NDCB0_ADDR_CYC(marvell_nand->addr_cyc) |
+                                 NDCB0_CMD1(NAND_CMD_SEQIN);
+               nfc_op.ndcb[1] |= NDCB1_ADDRS_PAGE(page);
+               nfc_op.ndcb[2] |= NDCB2_ADDR5_PAGE(page);
+       } else if (chunk < lt->nchunks - 1) {
+               nfc_op.ndcb[0] |= NDCB0_CMD_XTYPE(XTYPE_NAKED_RW);
+       } else {
+               nfc_op.ndcb[0] |= NDCB0_CMD_XTYPE(XTYPE_LAST_NAKED_RW);
+       }
+
+       /* Always dispatch the PAGEPROG command on the last chunk */
+       if (chunk == lt->nchunks - 1)
+               nfc_op.ndcb[0] |= NDCB0_CMD2(NAND_CMD_PAGEPROG) | NDCB0_DBC;
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_end_cmd(chip, NDSR_WRDREQ,
+                                 "WRDREQ while loading FIFO (data)");
+       if (ret)
+               return ret;
+
+       /* Transfer the contents */
+       iowrite32_rep(nfc->regs + NDDB, data, FIFO_REP(data_len));
+       iowrite32_rep(nfc->regs + NDDB, spare, FIFO_REP(spare_len));
+
+       return 0;
+}
+
+static int marvell_nfc_hw_ecc_bch_write_page(struct mtd_info *mtd,
+                                            struct nand_chip *chip,
+                                            const u8 *buf,
+                                            int oob_required, int page)
+{
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+       const u8 *data = buf;
+       const u8 *spare = chip->oob_poi;
+       int data_len = lt->data_bytes;
+       int spare_len = lt->spare_bytes;
+       int chunk, ret;
+
+       /* Spare data will be written anyway, so clear it to avoid garbage */
+       if (!oob_required)
+               memset(chip->oob_poi, 0xFF, mtd->oobsize);
+
+       marvell_nfc_enable_hw_ecc(chip);
+
+       for (chunk = 0; chunk < lt->nchunks; chunk++) {
+               if (chunk >= lt->full_chunk_cnt) {
+                       data_len = lt->last_data_bytes;
+                       spare_len = lt->last_spare_bytes;
+               }
+
+               marvell_nfc_hw_ecc_bch_write_chunk(chip, chunk, data, data_len,
+                                                  spare, spare_len, page);
+               data += data_len;
+               spare += spare_len;
+
+               /*
+                * Waiting only for CMDD or PAGED is not enough, ECC are
+                * partially written. No flag is set once the operation is
+                * really finished but the ND_RUN bit is cleared, so wait for it
+                * before stepping into the next command.
+                */
+               marvell_nfc_wait_ndrun(chip);
+       }
+
+       ret = marvell_nfc_wait_op(chip,
+                                 chip->data_interface.timings.sdr.tPROG_max);
+
+       marvell_nfc_disable_hw_ecc(chip);
+
+       if (ret)
+               return ret;
+
+       return 0;
+}
+
+static int marvell_nfc_hw_ecc_bch_write_oob_raw(struct mtd_info *mtd,
+                                               struct nand_chip *chip,
+                                               int page)
+{
+       /* Invalidate page cache */
+       chip->pagebuf = -1;
+
+       memset(chip->data_buf, 0xFF, mtd->writesize);
+
+       return chip->ecc.write_page_raw(mtd, chip, chip->data_buf, true, page);
+}
+
+static int marvell_nfc_hw_ecc_bch_write_oob(struct mtd_info *mtd,
+                                           struct nand_chip *chip, int page)
+{
+       /* Invalidate page cache */
+       chip->pagebuf = -1;
+
+       memset(chip->data_buf, 0xFF, mtd->writesize);
+
+       return chip->ecc.write_page(mtd, chip, chip->data_buf, true, page);
+}
+
+/* NAND framework ->exec_op() hooks and related helpers */
+static void marvell_nfc_parse_instructions(struct nand_chip *chip,
+                                          const struct nand_subop *subop,
+                                          struct marvell_nfc_op *nfc_op)
+{
+       const struct nand_op_instr *instr = NULL;
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       bool first_cmd = true;
+       unsigned int op_id;
+       int i;
+
+       /* Reset the input structure as most of its fields will be OR'ed */
+       memset(nfc_op, 0, sizeof(struct marvell_nfc_op));
+
+       for (op_id = 0; op_id < subop->ninstrs; op_id++) {
+               unsigned int offset, naddrs;
+               const u8 *addrs;
+               int len = nand_subop_get_data_len(subop, op_id);
+
+               instr = &subop->instrs[op_id];
+
+               switch (instr->type) {
+               case NAND_OP_CMD_INSTR:
+                       if (first_cmd)
+                               nfc_op->ndcb[0] |=
+                                       NDCB0_CMD1(instr->ctx.cmd.opcode);
+                       else
+                               nfc_op->ndcb[0] |=
+                                       NDCB0_CMD2(instr->ctx.cmd.opcode) |
+                                       NDCB0_DBC;
+
+                       nfc_op->cle_ale_delay_ns = instr->delay_ns;
+                       first_cmd = false;
+                       break;
+
+               case NAND_OP_ADDR_INSTR:
+                       offset = nand_subop_get_addr_start_off(subop, op_id);
+                       naddrs = nand_subop_get_num_addr_cyc(subop, op_id);
+                       addrs = &instr->ctx.addr.addrs[offset];
+
+                       nfc_op->ndcb[0] |= NDCB0_ADDR_CYC(naddrs);
+
+                       for (i = 0; i < min_t(unsigned int, 4, naddrs); i++)
+                               nfc_op->ndcb[1] |= addrs[i] << (8 * i);
+
+                       if (naddrs >= 5)
+                               nfc_op->ndcb[2] |= NDCB2_ADDR5_CYC(addrs[4]);
+                       if (naddrs >= 6)
+                               nfc_op->ndcb[3] |= NDCB3_ADDR6_CYC(addrs[5]);
+                       if (naddrs == 7)
+                               nfc_op->ndcb[3] |= NDCB3_ADDR7_CYC(addrs[6]);
+
+                       nfc_op->cle_ale_delay_ns = instr->delay_ns;
+                       break;
+
+               case NAND_OP_DATA_IN_INSTR:
+                       nfc_op->data_instr = instr;
+                       nfc_op->data_instr_idx = op_id;
+                       nfc_op->ndcb[0] |= NDCB0_CMD_TYPE(TYPE_READ);
+                       if (nfc->caps->is_nfcv2) {
+                               nfc_op->ndcb[0] |=
+                                       NDCB0_CMD_XTYPE(XTYPE_MONOLITHIC_RW) |
+                                       NDCB0_LEN_OVRD;
+                               nfc_op->ndcb[3] |= round_up(len, FIFO_DEPTH);
+                       }
+                       nfc_op->data_delay_ns = instr->delay_ns;
+                       break;
+
+               case NAND_OP_DATA_OUT_INSTR:
+                       nfc_op->data_instr = instr;
+                       nfc_op->data_instr_idx = op_id;
+                       nfc_op->ndcb[0] |= NDCB0_CMD_TYPE(TYPE_WRITE);
+                       if (nfc->caps->is_nfcv2) {
+                               nfc_op->ndcb[0] |=
+                                       NDCB0_CMD_XTYPE(XTYPE_MONOLITHIC_RW) |
+                                       NDCB0_LEN_OVRD;
+                               nfc_op->ndcb[3] |= round_up(len, FIFO_DEPTH);
+                       }
+                       nfc_op->data_delay_ns = instr->delay_ns;
+                       break;
+
+               case NAND_OP_WAITRDY_INSTR:
+                       nfc_op->rdy_timeout_ms = instr->ctx.waitrdy.timeout_ms;
+                       nfc_op->rdy_delay_ns = instr->delay_ns;
+                       break;
+               }
+       }
+}
+
+static int marvell_nfc_xfer_data_pio(struct nand_chip *chip,
+                                    const struct nand_subop *subop,
+                                    struct marvell_nfc_op *nfc_op)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       const struct nand_op_instr *instr = nfc_op->data_instr;
+       unsigned int op_id = nfc_op->data_instr_idx;
+       unsigned int len = nand_subop_get_data_len(subop, op_id);
+       unsigned int offset = nand_subop_get_data_start_off(subop, op_id);
+       bool reading = (instr->type == NAND_OP_DATA_IN_INSTR);
+       int ret;
+
+       if (instr->ctx.data.force_8bit)
+               marvell_nfc_force_byte_access(chip, true);
+
+       if (reading) {
+               u8 *in = instr->ctx.data.buf.in + offset;
+
+               ret = marvell_nfc_xfer_data_in_pio(nfc, in, len);
+       } else {
+               const u8 *out = instr->ctx.data.buf.out + offset;
+
+               ret = marvell_nfc_xfer_data_out_pio(nfc, out, len);
+       }
+
+       if (instr->ctx.data.force_8bit)
+               marvell_nfc_force_byte_access(chip, false);
+
+       return ret;
+}
+
+static int marvell_nfc_monolithic_access_exec(struct nand_chip *chip,
+                                             const struct nand_subop *subop)
+{
+       struct marvell_nfc_op nfc_op;
+       bool reading;
+       int ret;
+
+       marvell_nfc_parse_instructions(chip, subop, &nfc_op);
+       reading = (nfc_op.data_instr->type == NAND_OP_DATA_IN_INSTR);
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_end_cmd(chip, NDSR_RDDREQ | NDSR_WRDREQ,
+                                 "RDDREQ/WRDREQ while draining raw data");
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.cle_ale_delay_ns);
+
+       if (reading) {
+               if (nfc_op.rdy_timeout_ms) {
+                       ret = marvell_nfc_wait_op(chip, nfc_op.rdy_timeout_ms);
+                       if (ret)
+                               return ret;
+               }
+
+               cond_delay(nfc_op.rdy_delay_ns);
+       }
+
+       marvell_nfc_xfer_data_pio(chip, subop, &nfc_op);
+       ret = marvell_nfc_wait_cmdd(chip);
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.data_delay_ns);
+
+       if (!reading) {
+               if (nfc_op.rdy_timeout_ms) {
+                       ret = marvell_nfc_wait_op(chip, nfc_op.rdy_timeout_ms);
+                       if (ret)
+                               return ret;
+               }
+
+               cond_delay(nfc_op.rdy_delay_ns);
+       }
+
+       /*
+        * NDCR ND_RUN bit should be cleared automatically at the end of each
+        * operation but experience shows that the behavior is buggy when it
+        * comes to writes (with LEN_OVRD). Clear it by hand in this case.
+        */
+       if (!reading) {
+               struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+
+               writel_relaxed(readl(nfc->regs + NDCR) & ~NDCR_ND_RUN,
+                              nfc->regs + NDCR);
+       }
+
+       return 0;
+}
+
+static int marvell_nfc_naked_access_exec(struct nand_chip *chip,
+                                        const struct nand_subop *subop)
+{
+       struct marvell_nfc_op nfc_op;
+       int ret;
+
+       marvell_nfc_parse_instructions(chip, subop, &nfc_op);
+
+       /*
+        * Naked access are different in that they need to be flagged as naked
+        * by the controller. Reset the controller registers fields that inform
+        * on the type and refill them according to the ongoing operation.
+        */
+       nfc_op.ndcb[0] &= ~(NDCB0_CMD_TYPE(TYPE_MASK) |
+                           NDCB0_CMD_XTYPE(XTYPE_MASK));
+       switch (subop->instrs[0].type) {
+       case NAND_OP_CMD_INSTR:
+               nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_NAKED_CMD);
+               break;
+       case NAND_OP_ADDR_INSTR:
+               nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_NAKED_ADDR);
+               break;
+       case NAND_OP_DATA_IN_INSTR:
+               nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_READ) |
+                                 NDCB0_CMD_XTYPE(XTYPE_LAST_NAKED_RW);
+               break;
+       case NAND_OP_DATA_OUT_INSTR:
+               nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_WRITE) |
+                                 NDCB0_CMD_XTYPE(XTYPE_LAST_NAKED_RW);
+               break;
+       default:
+               /* This should never happen */
+               break;
+       }
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+
+       if (!nfc_op.data_instr) {
+               ret = marvell_nfc_wait_cmdd(chip);
+               cond_delay(nfc_op.cle_ale_delay_ns);
+               return ret;
+       }
+
+       ret = marvell_nfc_end_cmd(chip, NDSR_RDDREQ | NDSR_WRDREQ,
+                                 "RDDREQ/WRDREQ while draining raw data");
+       if (ret)
+               return ret;
+
+       marvell_nfc_xfer_data_pio(chip, subop, &nfc_op);
+       ret = marvell_nfc_wait_cmdd(chip);
+       if (ret)
+               return ret;
+
+       /*
+        * NDCR ND_RUN bit should be cleared automatically at the end of each
+        * operation but experience shows that the behavior is buggy when it
+        * comes to writes (with LEN_OVRD). Clear it by hand in this case.
+        */
+       if (subop->instrs[0].type == NAND_OP_DATA_OUT_INSTR) {
+               struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+
+               writel_relaxed(readl(nfc->regs + NDCR) & ~NDCR_ND_RUN,
+                              nfc->regs + NDCR);
+       }
+
+       return 0;
+}
+
+static int marvell_nfc_naked_waitrdy_exec(struct nand_chip *chip,
+                                         const struct nand_subop *subop)
+{
+       struct marvell_nfc_op nfc_op;
+       int ret;
+
+       marvell_nfc_parse_instructions(chip, subop, &nfc_op);
+
+       ret = marvell_nfc_wait_op(chip, nfc_op.rdy_timeout_ms);
+       cond_delay(nfc_op.rdy_delay_ns);
+
+       return ret;
+}
+
+static int marvell_nfc_read_id_type_exec(struct nand_chip *chip,
+                                        const struct nand_subop *subop)
+{
+       struct marvell_nfc_op nfc_op;
+       int ret;
+
+       marvell_nfc_parse_instructions(chip, subop, &nfc_op);
+       nfc_op.ndcb[0] &= ~NDCB0_CMD_TYPE(TYPE_READ);
+       nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_READ_ID);
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_end_cmd(chip, NDSR_RDDREQ,
+                                 "RDDREQ while reading ID");
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.cle_ale_delay_ns);
+
+       if (nfc_op.rdy_timeout_ms) {
+               ret = marvell_nfc_wait_op(chip, nfc_op.rdy_timeout_ms);
+               if (ret)
+                       return ret;
+       }
+
+       cond_delay(nfc_op.rdy_delay_ns);
+
+       marvell_nfc_xfer_data_pio(chip, subop, &nfc_op);
+       ret = marvell_nfc_wait_cmdd(chip);
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.data_delay_ns);
+
+       return 0;
+}
+
+static int marvell_nfc_read_status_exec(struct nand_chip *chip,
+                                       const struct nand_subop *subop)
+{
+       struct marvell_nfc_op nfc_op;
+       int ret;
+
+       marvell_nfc_parse_instructions(chip, subop, &nfc_op);
+       nfc_op.ndcb[0] &= ~NDCB0_CMD_TYPE(TYPE_READ);
+       nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_STATUS);
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_end_cmd(chip, NDSR_RDDREQ,
+                                 "RDDREQ while reading status");
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.cle_ale_delay_ns);
+
+       if (nfc_op.rdy_timeout_ms) {
+               ret = marvell_nfc_wait_op(chip, nfc_op.rdy_timeout_ms);
+               if (ret)
+                       return ret;
+       }
+
+       cond_delay(nfc_op.rdy_delay_ns);
+
+       marvell_nfc_xfer_data_pio(chip, subop, &nfc_op);
+       ret = marvell_nfc_wait_cmdd(chip);
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.data_delay_ns);
+
+       return 0;
+}
+
+static int marvell_nfc_reset_cmd_type_exec(struct nand_chip *chip,
+                                          const struct nand_subop *subop)
+{
+       struct marvell_nfc_op nfc_op;
+       int ret;
+
+       marvell_nfc_parse_instructions(chip, subop, &nfc_op);
+       nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_RESET);
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_wait_cmdd(chip);
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.cle_ale_delay_ns);
+
+       ret = marvell_nfc_wait_op(chip, nfc_op.rdy_timeout_ms);
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.rdy_delay_ns);
+
+       return 0;
+}
+
+static int marvell_nfc_erase_cmd_type_exec(struct nand_chip *chip,
+                                          const struct nand_subop *subop)
+{
+       struct marvell_nfc_op nfc_op;
+       int ret;
+
+       marvell_nfc_parse_instructions(chip, subop, &nfc_op);
+       nfc_op.ndcb[0] |= NDCB0_CMD_TYPE(TYPE_ERASE);
+
+       ret = marvell_nfc_prepare_cmd(chip);
+       if (ret)
+               return ret;
+
+       marvell_nfc_send_cmd(chip, &nfc_op);
+       ret = marvell_nfc_wait_cmdd(chip);
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.cle_ale_delay_ns);
+
+       ret = marvell_nfc_wait_op(chip, nfc_op.rdy_timeout_ms);
+       if (ret)
+               return ret;
+
+       cond_delay(nfc_op.rdy_delay_ns);
+
+       return 0;
+}
+
+static const struct nand_op_parser marvell_nfcv2_op_parser = NAND_OP_PARSER(
+       /* Monolithic reads/writes */
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_monolithic_access_exec,
+               NAND_OP_PARSER_PAT_CMD_ELEM(false),
+               NAND_OP_PARSER_PAT_ADDR_ELEM(true, MAX_ADDRESS_CYC_NFCV2),
+               NAND_OP_PARSER_PAT_CMD_ELEM(true),
+               NAND_OP_PARSER_PAT_WAITRDY_ELEM(true),
+               NAND_OP_PARSER_PAT_DATA_IN_ELEM(false, MAX_CHUNK_SIZE)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_monolithic_access_exec,
+               NAND_OP_PARSER_PAT_CMD_ELEM(false),
+               NAND_OP_PARSER_PAT_ADDR_ELEM(false, MAX_ADDRESS_CYC_NFCV2),
+               NAND_OP_PARSER_PAT_DATA_OUT_ELEM(false, MAX_CHUNK_SIZE),
+               NAND_OP_PARSER_PAT_CMD_ELEM(true),
+               NAND_OP_PARSER_PAT_WAITRDY_ELEM(true)),
+       /* Naked commands */
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_naked_access_exec,
+               NAND_OP_PARSER_PAT_CMD_ELEM(false)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_naked_access_exec,
+               NAND_OP_PARSER_PAT_ADDR_ELEM(false, MAX_ADDRESS_CYC_NFCV2)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_naked_access_exec,
+               NAND_OP_PARSER_PAT_DATA_IN_ELEM(false, MAX_CHUNK_SIZE)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_naked_access_exec,
+               NAND_OP_PARSER_PAT_DATA_OUT_ELEM(false, MAX_CHUNK_SIZE)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_naked_waitrdy_exec,
+               NAND_OP_PARSER_PAT_WAITRDY_ELEM(false)),
+       );
+
+static const struct nand_op_parser marvell_nfcv1_op_parser = NAND_OP_PARSER(
+       /* Naked commands not supported, use a function for each pattern */
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_read_id_type_exec,
+               NAND_OP_PARSER_PAT_CMD_ELEM(false),
+               NAND_OP_PARSER_PAT_ADDR_ELEM(false, MAX_ADDRESS_CYC_NFCV1),
+               NAND_OP_PARSER_PAT_DATA_IN_ELEM(false, 8)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_erase_cmd_type_exec,
+               NAND_OP_PARSER_PAT_CMD_ELEM(false),
+               NAND_OP_PARSER_PAT_ADDR_ELEM(false, MAX_ADDRESS_CYC_NFCV1),
+               NAND_OP_PARSER_PAT_CMD_ELEM(false),
+               NAND_OP_PARSER_PAT_WAITRDY_ELEM(false)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_read_status_exec,
+               NAND_OP_PARSER_PAT_CMD_ELEM(false),
+               NAND_OP_PARSER_PAT_DATA_IN_ELEM(false, 1)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_reset_cmd_type_exec,
+               NAND_OP_PARSER_PAT_CMD_ELEM(false),
+               NAND_OP_PARSER_PAT_WAITRDY_ELEM(false)),
+       NAND_OP_PARSER_PATTERN(
+               marvell_nfc_naked_waitrdy_exec,
+               NAND_OP_PARSER_PAT_WAITRDY_ELEM(false)),
+       );
+
+static int marvell_nfc_exec_op(struct nand_chip *chip,
+                              const struct nand_operation *op,
+                              bool check_only)
+{
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+
+       if (nfc->caps->is_nfcv2)
+               return nand_op_parser_exec_op(chip, &marvell_nfcv2_op_parser,
+                                             op, check_only);
+       else
+               return nand_op_parser_exec_op(chip, &marvell_nfcv1_op_parser,
+                                             op, check_only);
+}
+
+/*
+ * Layouts were broken in old pxa3xx_nand driver, these are supposed to be
+ * usable.
+ */
+static int marvell_nand_ooblayout_ecc(struct mtd_info *mtd, int section,
+                                     struct mtd_oob_region *oobregion)
+{
+       struct nand_chip *chip = mtd_to_nand(mtd);
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+
+       if (section)
+               return -ERANGE;
+
+       oobregion->length = (lt->full_chunk_cnt * lt->ecc_bytes) +
+                           lt->last_ecc_bytes;
+       oobregion->offset = mtd->oobsize - oobregion->length;
+
+       return 0;
+}
+
+static int marvell_nand_ooblayout_free(struct mtd_info *mtd, int section,
+                                      struct mtd_oob_region *oobregion)
+{
+       struct nand_chip *chip = mtd_to_nand(mtd);
+       const struct marvell_hw_ecc_layout *lt = to_marvell_nand(chip)->layout;
+
+       if (section)
+               return -ERANGE;
+
+       /*
+        * Bootrom looks in bytes 0 & 5 for bad blocks for the
+        * 4KB page / 4bit BCH combination.
+        */
+       if (mtd->writesize == SZ_4K && lt->data_bytes == SZ_2K)
+               oobregion->offset = 6;
+       else
+               oobregion->offset = 2;
+
+       oobregion->length = (lt->full_chunk_cnt * lt->spare_bytes) +
+                           lt->last_spare_bytes - oobregion->offset;
+
+       return 0;
+}
+
+static const struct mtd_ooblayout_ops marvell_nand_ooblayout_ops = {
+       .ecc = marvell_nand_ooblayout_ecc,
+       .free = marvell_nand_ooblayout_free,
+};
+
+static int marvell_nand_hw_ecc_ctrl_init(struct mtd_info *mtd,
+                                        struct nand_ecc_ctrl *ecc)
+{
+       struct nand_chip *chip = mtd_to_nand(mtd);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       const struct marvell_hw_ecc_layout *l;
+       int i;
+
+       if (!nfc->caps->is_nfcv2 &&
+           (mtd->writesize + mtd->oobsize > MAX_CHUNK_SIZE)) {
+               dev_err(nfc->dev,
+                       "NFCv1: writesize (%d) cannot be bigger than a chunk (%d)\n",
+                       mtd->writesize, MAX_CHUNK_SIZE - mtd->oobsize);
+               return -ENOTSUPP;
+       }
+
+       to_marvell_nand(chip)->layout = NULL;
+       for (i = 0; i < ARRAY_SIZE(marvell_nfc_layouts); i++) {
+               l = &marvell_nfc_layouts[i];
+               if (mtd->writesize == l->writesize &&
+                   ecc->size == l->chunk && ecc->strength == l->strength) {
+                       to_marvell_nand(chip)->layout = l;
+                       break;
+               }
+       }
+
+       if (!to_marvell_nand(chip)->layout ||
+           (!nfc->caps->is_nfcv2 && ecc->strength > 1)) {
+               dev_err(nfc->dev,
+                       "ECC strength %d at page size %d is not supported\n",
+                       ecc->strength, mtd->writesize);
+               return -ENOTSUPP;
+       }
+
+       mtd_set_ooblayout(mtd, &marvell_nand_ooblayout_ops);
+       ecc->steps = l->nchunks;
+       ecc->size = l->data_bytes;
+
+       if (ecc->strength == 1) {
+               chip->ecc.algo = NAND_ECC_HAMMING;
+               ecc->read_page_raw = marvell_nfc_hw_ecc_hmg_read_page_raw;
+               ecc->read_page = marvell_nfc_hw_ecc_hmg_read_page;
+               ecc->read_oob_raw = marvell_nfc_hw_ecc_hmg_read_oob_raw;
+               ecc->read_oob = ecc->read_oob_raw;
+               ecc->write_page_raw = marvell_nfc_hw_ecc_hmg_write_page_raw;
+               ecc->write_page = marvell_nfc_hw_ecc_hmg_write_page;
+               ecc->write_oob_raw = marvell_nfc_hw_ecc_hmg_write_oob_raw;
+               ecc->write_oob = ecc->write_oob_raw;
+       } else {
+               chip->ecc.algo = NAND_ECC_BCH;
+               ecc->strength = 16;
+               ecc->read_page_raw = marvell_nfc_hw_ecc_bch_read_page_raw;
+               ecc->read_page = marvell_nfc_hw_ecc_bch_read_page;
+               ecc->read_oob_raw = marvell_nfc_hw_ecc_bch_read_oob_raw;
+               ecc->read_oob = marvell_nfc_hw_ecc_bch_read_oob;
+               ecc->write_page_raw = marvell_nfc_hw_ecc_bch_write_page_raw;
+               ecc->write_page = marvell_nfc_hw_ecc_bch_write_page;
+               ecc->write_oob_raw = marvell_nfc_hw_ecc_bch_write_oob_raw;
+               ecc->write_oob = marvell_nfc_hw_ecc_bch_write_oob;
+       }
+
+       return 0;
+}
+
+static int marvell_nand_ecc_init(struct mtd_info *mtd,
+                                struct nand_ecc_ctrl *ecc)
+{
+       struct nand_chip *chip = mtd_to_nand(mtd);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       int ret;
+
+       if (ecc->mode != NAND_ECC_NONE && (!ecc->size || !ecc->strength)) {
+               if (chip->ecc_step_ds && chip->ecc_strength_ds) {
+                       ecc->size = chip->ecc_step_ds;
+                       ecc->strength = chip->ecc_strength_ds;
+               } else {
+                       dev_info(nfc->dev,
+                                "No minimum ECC strength, using 1b/512B\n");
+                       ecc->size = 512;
+                       ecc->strength = 1;
+               }
+       }
+
+       switch (ecc->mode) {
+       case NAND_ECC_HW:
+               ret = marvell_nand_hw_ecc_ctrl_init(mtd, ecc);
+               if (ret)
+                       return ret;
+               break;
+       case NAND_ECC_NONE:
+       case NAND_ECC_SOFT:
+               if (!nfc->caps->is_nfcv2 && mtd->writesize != SZ_512 &&
+                   mtd->writesize != SZ_2K) {
+                       dev_err(nfc->dev, "NFCv1 cannot write %d bytes pages\n",
+                               mtd->writesize);
+                       return -EINVAL;
+               }
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static u8 bbt_pattern[] = {'M', 'V', 'B', 'b', 't', '0' };
+static u8 bbt_mirror_pattern[] = {'1', 't', 'b', 'B', 'V', 'M' };
+
+static struct nand_bbt_descr bbt_main_descr = {
+       .options = NAND_BBT_LASTBLOCK | NAND_BBT_CREATE | NAND_BBT_WRITE |
+                  NAND_BBT_2BIT | NAND_BBT_VERSION,
+       .offs = 8,
+       .len = 6,
+       .veroffs = 14,
+       .maxblocks = 8, /* Last 8 blocks in each chip */
+       .pattern = bbt_pattern
+};
+
+static struct nand_bbt_descr bbt_mirror_descr = {
+       .options = NAND_BBT_LASTBLOCK | NAND_BBT_CREATE | NAND_BBT_WRITE |
+                  NAND_BBT_2BIT | NAND_BBT_VERSION,
+       .offs = 8,
+       .len = 6,
+       .veroffs = 14,
+       .maxblocks = 8, /* Last 8 blocks in each chip */
+       .pattern = bbt_mirror_pattern
+};
+
+static int marvell_nfc_setup_data_interface(struct mtd_info *mtd, int chipnr,
+                                           const struct nand_data_interface
+                                           *conf)
+{
+       struct nand_chip *chip = mtd_to_nand(mtd);
+       struct marvell_nand_chip *marvell_nand = to_marvell_nand(chip);
+       struct marvell_nfc *nfc = to_marvell_nfc(chip->controller);
+       unsigned int period_ns = 1000000000 / clk_get_rate(nfc->ecc_clk) * 2;
+       const struct nand_sdr_timings *sdr;
+       struct marvell_nfc_timings nfc_tmg;
+       int read_delay;
+
+       sdr = nand_get_sdr_timings(conf);
+       if (IS_ERR(sdr))
+               return PTR_ERR(sdr);
+
+       /*
+        * SDR timings are given in pico-seconds while NFC timings must be
+        * expressed in NAND controller clock cycles, which is half of the
+        * frequency of the accessible ECC clock retrieved by clk_get_rate().
+        * This is not written anywhere in the datasheet but was observed
+        * with an oscilloscope.
+        *
+        * NFC datasheet gives equations from which thoses calculations
+        * are derived, they tend to be slightly more restrictives than the
+        * given core timings and may improve the overall speed.
+        */
+       nfc_tmg.tRP = TO_CYCLES(DIV_ROUND_UP(sdr->tRC_min, 2), period_ns) - 1;
+       nfc_tmg.tRH = nfc_tmg.tRP;
+       nfc_tmg.tWP = TO_CYCLES(DIV_ROUND_UP(sdr->tWC_min, 2), period_ns) - 1;
+       nfc_tmg.tWH = nfc_tmg.tWP;
+       nfc_tmg.tCS = TO_CYCLES(sdr->tCS_min, period_ns);
+       nfc_tmg.tCH = TO_CYCLES(sdr->tCH_min, period_ns) - 1;
+       nfc_tmg.tADL = TO_CYCLES(sdr->tADL_min, period_ns);
+       /*
+        * Read delay is the time of propagation from SoC pins to NFC internal
+        * logic. With non-EDO timings, this is MIN_RD_DEL_CNT clock cycles. In
+        * EDO mode, an additional delay of tRH must be taken into account so
+        * the data is sampled on the falling edge instead of the rising edge.
+        */
+       read_delay = sdr->tRC_min >= 30000 ?
+               MIN_RD_DEL_CNT : MIN_RD_DEL_CNT + nfc_tmg.tRH;
+
+       nfc_tmg.tAR = TO_CYCLES(sdr->tAR_min, period_ns);
+       /*
+        * tWHR and tRHW are supposed to be read to write delays (and vice
+        * versa) but in some cases, ie. when doing a change column, they must
+        * be greater than that to be sure tCCS delay is respected.
+        */
+       nfc_tmg.tWHR = TO_CYCLES(max_t(int, sdr->tWHR_min, sdr->tCCS_min),
+                                period_ns) - 2,
+       nfc_tmg.tRHW = TO_CYCLES(max_t(int, sdr->tRHW_min, sdr->tCCS_min),
+                                period_ns);
+
+       /* Use WAIT_MODE (wait for RB line) instead of only relying on delays */
+       nfc_tmg.tR = TO_CYCLES(sdr->tWB_max, period_ns);
+
+       if (chipnr < 0)
+               return 0;
+
+       marvell_nand->ndtr0 =
+               NDTR0_TRP(nfc_tmg.tRP) |
+               NDTR0_TRH(nfc_tmg.tRH) |
+               NDTR0_ETRP(nfc_tmg.tRP) |
+               NDTR0_TWP(nfc_tmg.tWP) |
+               NDTR0_TWH(nfc_tmg.tWH) |
+               NDTR0_TCS(nfc_tmg.tCS) |
+               NDTR0_TCH(nfc_tmg.tCH) |
+               NDTR0_RD_CNT_DEL(read_delay) |
+               NDTR0_SELCNTR |
+               NDTR0_TADL(nfc_tmg.tADL);
+
+       marvell_nand->ndtr1 =
+               NDTR1_TAR(nfc_tmg.tAR) |
+               NDTR1_TWHR(nfc_tmg.tWHR) |
+               NDTR1_TRHW(nfc_tmg.tRHW) |
+               NDTR1_WAIT_MODE |
+               NDTR1_TR(nfc_tmg.tR);
+
+       return 0;
+}
+
+static int marvell_nand_chip_init(struct device *dev, struct marvell_nfc *nfc,
+                                 struct device_node *np)
+{
+       struct pxa3xx_nand_platform_data *pdata = dev_get_platdata(dev);
+       struct marvell_nand_chip *marvell_nand;
+       struct mtd_info *mtd;
+       struct nand_chip *chip;
+       int nsels, ret, i;
+       u32 cs, rb;
+
+       /*
+        * The legacy "num-cs" property indicates the number of CS on the only
+        * chip connected to the controller (legacy bindings does not support
+        * more than one chip). CS are only incremented one by one while the RB
+        * pin is always the #0.
+        *
+        * When not using legacy bindings, a couple of "reg" and "nand-rb"
+        * properties must be filled. For each chip, expressed as a subnode,
+        * "reg" points to the CS lines and "nand-rb" to the RB line.
+        */
+       if (pdata) {
+               nsels = 1;
+       } else if (nfc->caps->legacy_of_bindings &&
+                  !of_get_property(np, "num-cs", &nsels)) {
+               dev_err(dev, "missing num-cs property\n");
+               return -EINVAL;
+       } else if (!of_get_property(np, "reg", &nsels)) {
+               dev_err(dev, "missing reg property\n");
+               return -EINVAL;
+       }
+
+       if (!pdata)
+               nsels /= sizeof(u32);
+       if (!nsels) {
+               dev_err(dev, "invalid reg property size\n");
+               return -EINVAL;
+       }
+
+       /* Alloc the nand chip structure */
+       marvell_nand = devm_kzalloc(dev, sizeof(*marvell_nand) +
+                                   (nsels *
+                                    sizeof(struct marvell_nand_chip_sel)),
+                                   GFP_KERNEL);
+       if (!marvell_nand) {
+               dev_err(dev, "could not allocate chip structure\n");
+               return -ENOMEM;
+       }
+
+       marvell_nand->nsels = nsels;
+       marvell_nand->selected_die = -1;
+
+       for (i = 0; i < nsels; i++) {
+               if (pdata || nfc->caps->legacy_of_bindings) {
+                       /*
+                        * Legacy bindings use the CS lines in natural
+                        * order (0, 1, ...)
+                        */
+                       cs = i;
+               } else {
+                       /* Retrieve CS id */
+                       ret = of_property_read_u32_index(np, "reg", i, &cs);
+                       if (ret) {
+                               dev_err(dev, "could not retrieve reg property: %d\n",
+                                       ret);
+                               return ret;
+                       }
+               }
+
+               if (cs >= nfc->caps->max_cs_nb) {
+                       dev_err(dev, "invalid reg value: %u (max CS = %d)\n",
+                               cs, nfc->caps->max_cs_nb);
+                       return -EINVAL;
+               }
+
+               if (test_and_set_bit(cs, &nfc->assigned_cs)) {
+                       dev_err(dev, "CS %d already assigned\n", cs);
+                       return -EINVAL;
+               }
+
+               /*
+                * The cs variable represents the chip select id, which must be
+                * converted in bit fields for NDCB0 and NDCB2 to select the
+                * right chip. Unfortunately, due to a lack of information on
+                * the subject and incoherent documentation, the user should not
+                * use CS1 and CS3 at all as asserting them is not supported in
+                * a reliable way (due to multiplexing inside ADDR5 field).
+                */
+               marvell_nand->sels[i].cs = cs;
+               switch (cs) {
+               case 0:
+               case 2:
+                       marvell_nand->sels[i].ndcb0_csel = 0;
+                       break;
+               case 1:
+               case 3:
+                       marvell_nand->sels[i].ndcb0_csel = NDCB0_CSEL;
+                       break;
+               default:
+                       return -EINVAL;
+               }
+
+               /* Retrieve RB id */
+               if (pdata || nfc->caps->legacy_of_bindings) {
+                       /* Legacy bindings always use RB #0 */
+                       rb = 0;
+               } else {
+                       ret = of_property_read_u32_index(np, "nand-rb", i,
+                                                        &rb);
+                       if (ret) {
+                               dev_err(dev,
+                                       "could not retrieve RB property: %d\n",
+                                       ret);
+                               return ret;
+                       }
+               }
+
+               if (rb >= nfc->caps->max_rb_nb) {
+                       dev_err(dev, "invalid reg value: %u (max RB = %d)\n",
+                               rb, nfc->caps->max_rb_nb);
+                       return -EINVAL;
+               }
+
+               marvell_nand->sels[i].rb = rb;
+       }
+
+       chip = &marvell_nand->chip;
+       chip->controller = &nfc->controller;
+       nand_set_flash_node(chip, np);
+
+       chip->exec_op = marvell_nfc_exec_op;
+       chip->select_chip = marvell_nfc_select_chip;
+       if (nfc->caps->is_nfcv2 &&
+           !of_property_read_bool(np, "marvell,nand-keep-config"))
+               chip->setup_data_interface = marvell_nfc_setup_data_interface;
+
+       mtd = nand_to_mtd(chip);
+       mtd->dev.parent = dev;
+
+       /*
+        * Default to HW ECC engine mode. If the nand-ecc-mode property is given
+        * in the DT node, this entry will be overwritten in nand_scan_ident().
+        */
+       chip->ecc.mode = NAND_ECC_HW;
+
+       /*
+        * Save a reference value for timing registers before
+        * ->setup_data_interface() is called.
+        */
+       marvell_nand->ndtr0 = readl_relaxed(nfc->regs + NDTR0);
+       marvell_nand->ndtr1 = readl_relaxed(nfc->regs + NDTR1);
+
+       chip->options |= NAND_BUSWIDTH_AUTO;
+       ret = nand_scan_ident(mtd, marvell_nand->nsels, NULL);
+       if (ret) {
+               dev_err(dev, "could not identify the nand chip\n");
+               return ret;
+       }
+
+       if (pdata && pdata->flash_bbt)
+               chip->bbt_options |= NAND_BBT_USE_FLASH;
+
+       if (chip->bbt_options & NAND_BBT_USE_FLASH) {
+               /*
+                * We'll use a bad block table stored in-flash and don't
+                * allow writing the bad block marker to the flash.
+                */
+               chip->bbt_options |= NAND_BBT_NO_OOB_BBM;
+               chip->bbt_td = &bbt_main_descr;
+               chip->bbt_md = &bbt_mirror_descr;
+       }
+
+       /* Save the chip-specific fields of NDCR */
+       marvell_nand->ndcr = NDCR_PAGE_SZ(mtd->writesize);
+       if (chip->options & NAND_BUSWIDTH_16)
+               marvell_nand->ndcr |= NDCR_DWIDTH_M | NDCR_DWIDTH_C;
+
+       /*
+        * On small page NANDs, only one cycle is needed to pass the
+        * column address.
+        */
+       if (mtd->writesize <= 512) {
+               marvell_nand->addr_cyc = 1;
+       } else {
+               marvell_nand->addr_cyc = 2;
+               marvell_nand->ndcr |= NDCR_RA_START;
+       }
+
+       /*
+        * Now add the number of cycles needed to pass the row
+        * address.
+        *
+        * Addressing a chip using CS 2 or 3 should also need the third row
+        * cycle but due to inconsistance in the documentation and lack of
+        * hardware to test this situation, this case is not supported.
+        */
+       if (chip->options & NAND_ROW_ADDR_3)
+               marvell_nand->addr_cyc += 3;
+       else
+               marvell_nand->addr_cyc += 2;
+
+       if (pdata) {
+               chip->ecc.size = pdata->ecc_step_size;
+               chip->ecc.strength = pdata->ecc_strength;
+       }
+
+       ret = marvell_nand_ecc_init(mtd, &chip->ecc);
+       if (ret) {
+               dev_err(dev, "ECC init failed: %d\n", ret);
+               return ret;
+       }
+
+       if (chip->ecc.mode == NAND_ECC_HW) {
+               /*
+                * Subpage write not available with hardware ECC, prohibit also
+                * subpage read as in userspace subpage access would still be
+                * allowed and subpage write, if used, would lead to numerous
+                * uncorrectable ECC errors.
+                */
+               chip->options |= NAND_NO_SUBPAGE_WRITE;
+       }
+
+       if (pdata || nfc->caps->legacy_of_bindings) {
+               /*
+                * We keep the MTD name unchanged to avoid breaking platforms
+                * where the MTD cmdline parser is used and the bootloader
+                * has not been updated to use the new naming scheme.
+                */
+               mtd->name = "pxa3xx_nand-0";
+       } else if (!mtd->name) {
+               /*
+                * If the new bindings are used and the bootloader has not been
+                * updated to pass a new mtdparts parameter on the cmdline, you
+                * should define the following property in your NAND node, ie:
+                *
+                *      label = "main-storage";
+                *
+                * This way, mtd->name will be set by the core when
+                * nand_set_flash_node() is called.
+                */
+               mtd->name = devm_kasprintf(nfc->dev, GFP_KERNEL,
+                                          "%s:nand.%d", dev_name(nfc->dev),
+                                          marvell_nand->sels[0].cs);
+               if (!mtd->name) {
+                       dev_err(nfc->dev, "Failed to allocate mtd->name\n");
+                       return -ENOMEM;
+               }
+       }
+
+       ret = nand_scan_tail(mtd);
+       if (ret) {
+               dev_err(dev, "nand_scan_tail failed: %d\n", ret);
+               return ret;
+       }
+
+       if (pdata)
+               /* Legacy bindings support only one chip */
+               ret = mtd_device_register(mtd, pdata->parts[0],
+                                         pdata->nr_parts[0]);
+       else
+               ret = mtd_device_register(mtd, NULL, 0);
+       if (ret) {
+               dev_err(dev, "failed to register mtd device: %d\n", ret);
+               nand_release(mtd);
+               return ret;
+       }
+
+       list_add_tail(&marvell_nand->node, &nfc->chips);
+
+       return 0;
+}
+
+static int marvell_nand_chips_init(struct device *dev, struct marvell_nfc *nfc)
+{
+       struct device_node *np = dev->of_node;
+       struct device_node *nand_np;
+       int max_cs = nfc->caps->max_cs_nb;
+       int nchips;
+       int ret;
+
+       if (!np)
+               nchips = 1;
+       else
+               nchips = of_get_child_count(np);
+
+       if (nchips > max_cs) {
+               dev_err(dev, "too many NAND chips: %d (max = %d CS)\n", nchips,
+                       max_cs);
+               return -EINVAL;
+       }
+
+       /*
+        * Legacy bindings do not use child nodes to exhibit NAND chip
+        * properties and layout. Instead, NAND properties are mixed with the
+        * controller ones, and partitions are defined as direct subnodes of the
+        * NAND controller node.
+        */
+       if (nfc->caps->legacy_of_bindings) {
+               ret = marvell_nand_chip_init(dev, nfc, np);
+               return ret;
+       }
+
+       for_each_child_of_node(np, nand_np) {
+               ret = marvell_nand_chip_init(dev, nfc, nand_np);
+               if (ret) {
+                       of_node_put(nand_np);
+                       return ret;
+               }
+       }
+
+       return 0;
+}
+
+static void marvell_nand_chips_cleanup(struct marvell_nfc *nfc)
+{
+       struct marvell_nand_chip *entry, *temp;
+
+       list_for_each_entry_safe(entry, temp, &nfc->chips, node) {
+               nand_release(nand_to_mtd(&entry->chip));
+               list_del(&entry->node);
+       }
+}
+
+static int marvell_nfc_init_dma(struct marvell_nfc *nfc)
+{
+       struct platform_device *pdev = container_of(nfc->dev,
+                                                   struct platform_device,
+                                                   dev);
+       struct dma_slave_config config = {};
+       struct resource *r;
+       dma_cap_mask_t mask;
+       struct pxad_param param;
+       int ret;
+
+       if (!IS_ENABLED(CONFIG_PXA_DMA)) {
+               dev_warn(nfc->dev,
+                        "DMA not enabled in configuration\n");
+               return -ENOTSUPP;
+       }
+
+       ret = dma_set_mask_and_coherent(nfc->dev, DMA_BIT_MASK(32));
+       if (ret)
+               return ret;
+
+       r = platform_get_resource(pdev, IORESOURCE_DMA, 0);
+       if (!r) {
+               dev_err(nfc->dev, "No resource defined for data DMA\n");
+               return -ENXIO;
+       }
+
+       param.drcmr = r->start;
+       param.prio = PXAD_PRIO_LOWEST;
+       dma_cap_zero(mask);
+       dma_cap_set(DMA_SLAVE, mask);
+       nfc->dma_chan =
+               dma_request_slave_channel_compat(mask, pxad_filter_fn,
+                                                ¶m, nfc->dev,
+                                                "data");
+       if (!nfc->dma_chan) {
+               dev_err(nfc->dev,
+                       "Unable to request data DMA channel\n");
+               return -ENODEV;
+       }
+
+       r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!r)
+               return -ENXIO;
+
+       config.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
+       config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
+       config.src_addr = r->start + NDDB;
+       config.dst_addr = r->start + NDDB;
+       config.src_maxburst = 32;
+       config.dst_maxburst = 32;
+       ret = dmaengine_slave_config(nfc->dma_chan, &config);
+       if (ret < 0) {
+               dev_err(nfc->dev, "Failed to configure DMA channel\n");
+               return ret;
+       }
+
+       /*
+        * DMA must act on length multiple of 32 and this length may be
+        * bigger than the destination buffer. Use this buffer instead
+        * for DMA transfers and then copy the desired amount of data to
+        * the provided buffer.
+        */
+       nfc->dma_buf = kmalloc(MAX_CHUNK_SIZE, GFP_DMA);
+       if (!nfc->dma_buf)
+               return -ENOMEM;
+
+       nfc->use_dma = true;
+
+       return 0;
+}
+
+static int marvell_nfc_init(struct marvell_nfc *nfc)
+{
+       struct device_node *np = nfc->dev->of_node;
+
+       /*
+        * Some SoCs like A7k/A8k need to enable manually the NAND
+        * controller, gated clocks and reset bits to avoid being bootloader
+        * dependent. This is done through the use of the System Functions
+        * registers.
+        */
+       if (nfc->caps->need_system_controller) {
+               struct regmap *sysctrl_base =
+                       syscon_regmap_lookup_by_phandle(np,
+                                                       "marvell,system-controller");
+               u32 reg;
+
+               if (IS_ERR(sysctrl_base))
+                       return PTR_ERR(sysctrl_base);
+
+               reg = GENCONF_SOC_DEVICE_MUX_NFC_EN |
+                     GENCONF_SOC_DEVICE_MUX_ECC_CLK_RST |
+                     GENCONF_SOC_DEVICE_MUX_ECC_CORE_RST |
+                     GENCONF_SOC_DEVICE_MUX_NFC_INT_EN;
+               regmap_write(sysctrl_base, GENCONF_SOC_DEVICE_MUX, reg);
+
+               regmap_read(sysctrl_base, GENCONF_CLK_GATING_CTRL, ®);
+               reg |= GENCONF_CLK_GATING_CTRL_ND_GATE;
+               regmap_write(sysctrl_base, GENCONF_CLK_GATING_CTRL, reg);
+
+               regmap_read(sysctrl_base, GENCONF_ND_CLK_CTRL, ®);
+               reg |= GENCONF_ND_CLK_CTRL_EN;
+               regmap_write(sysctrl_base, GENCONF_ND_CLK_CTRL, reg);
+       }
+
+       /* Configure the DMA if appropriate */
+       if (!nfc->caps->is_nfcv2)
+               marvell_nfc_init_dma(nfc);
+
+       /*
+        * ECC operations and interruptions are only enabled when specifically
+        * needed. ECC shall not be activated in the early stages (fails probe).
+        * Arbiter flag, even if marked as "reserved", must be set (empirical).
+        * SPARE_EN bit must always be set or ECC bytes will not be at the same
+        * offset in the read page and this will fail the protection.
+        */
+       writel_relaxed(NDCR_ALL_INT | NDCR_ND_ARB_EN | NDCR_SPARE_EN |
+                      NDCR_RD_ID_CNT(NFCV1_READID_LEN), nfc->regs + NDCR);
+       writel_relaxed(0xFFFFFFFF, nfc->regs + NDSR);
+       writel_relaxed(0, nfc->regs + NDECCCTRL);
+
+       return 0;
+}
+
+static int marvell_nfc_probe(struct platform_device *pdev)
+{
+       struct device *dev = &pdev->dev;
+       struct resource *r;
+       struct marvell_nfc *nfc;
+       int ret;
+       int irq;
+
+       nfc = devm_kzalloc(&pdev->dev, sizeof(struct marvell_nfc),
+                          GFP_KERNEL);
+       if (!nfc)
+               return -ENOMEM;
+
+       nfc->dev = dev;
+       nand_hw_control_init(&nfc->controller);
+       INIT_LIST_HEAD(&nfc->chips);
+
+       r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       nfc->regs = devm_ioremap_resource(dev, r);
+       if (IS_ERR(nfc->regs))
+               return PTR_ERR(nfc->regs);
+
+       irq = platform_get_irq(pdev, 0);
+       if (irq < 0) {
+               dev_err(dev, "failed to retrieve irq\n");
+               return irq;
+       }
+
+       nfc->ecc_clk = devm_clk_get(&pdev->dev, NULL);
+       if (IS_ERR(nfc->ecc_clk))
+               return PTR_ERR(nfc->ecc_clk);
+
+       ret = clk_prepare_enable(nfc->ecc_clk);
+       if (ret)
+               return ret;
+
+       marvell_nfc_disable_int(nfc, NDCR_ALL_INT);
+       marvell_nfc_clear_int(nfc, NDCR_ALL_INT);
+       ret = devm_request_irq(dev, irq, marvell_nfc_isr,
+                              0, "marvell-nfc", nfc);
+       if (ret)
+               goto unprepare_clk;
+
+       /* Get NAND controller capabilities */
+       if (pdev->id_entry)
+               nfc->caps = (void *)pdev->id_entry->driver_data;
+       else
+               nfc->caps = of_device_get_match_data(&pdev->dev);
+
+       if (!nfc->caps) {
+               dev_err(dev, "Could not retrieve NFC caps\n");
+               ret = -EINVAL;
+               goto unprepare_clk;
+       }
+
+       /* Init the controller and then probe the chips */
+       ret = marvell_nfc_init(nfc);
+       if (ret)
+               goto unprepare_clk;
+
+       platform_set_drvdata(pdev, nfc);
+
+       ret = marvell_nand_chips_init(dev, nfc);
+       if (ret)
+               goto unprepare_clk;
+
+       return 0;
+
+unprepare_clk:
+       clk_disable_unprepare(nfc->ecc_clk);
+
+       return ret;
+}
+
+static int marvell_nfc_remove(struct platform_device *pdev)
+{
+       struct marvell_nfc *nfc = platform_get_drvdata(pdev);
+
+       marvell_nand_chips_cleanup(nfc);
+
+       if (nfc->use_dma) {
+               dmaengine_terminate_all(nfc->dma_chan);
+               dma_release_channel(nfc->dma_chan);
+       }
+
+       clk_disable_unprepare(nfc->ecc_clk);
+
+       return 0;
+}
+
+static const struct marvell_nfc_caps marvell_armada_8k_nfc_caps = {
+       .max_cs_nb = 4,
+       .max_rb_nb = 2,
+       .need_system_controller = true,
+       .is_nfcv2 = true,
+};
+
+static const struct marvell_nfc_caps marvell_armada370_nfc_caps = {
+       .max_cs_nb = 4,
+       .max_rb_nb = 2,
+       .is_nfcv2 = true,
+};
+
+static const struct marvell_nfc_caps marvell_pxa3xx_nfc_caps = {
+       .max_cs_nb = 2,
+       .max_rb_nb = 1,
+       .use_dma = true,
+};
+
+static const struct marvell_nfc_caps marvell_armada_8k_nfc_legacy_caps = {
+       .max_cs_nb = 4,
+       .max_rb_nb = 2,
+       .need_system_controller = true,
+       .legacy_of_bindings = true,
+       .is_nfcv2 = true,
+};
+
+static const struct marvell_nfc_caps marvell_armada370_nfc_legacy_caps = {
+       .max_cs_nb = 4,
+       .max_rb_nb = 2,
+       .legacy_of_bindings = true,
+       .is_nfcv2 = true,
+};
+
+static const struct marvell_nfc_caps marvell_pxa3xx_nfc_legacy_caps = {
+       .max_cs_nb = 2,
+       .max_rb_nb = 1,
+       .legacy_of_bindings = true,
+       .use_dma = true,
+};
+
+static const struct platform_device_id marvell_nfc_platform_ids[] = {
+       {
+               .name = "pxa3xx-nand",
+               .driver_data = (kernel_ulong_t)&marvell_pxa3xx_nfc_legacy_caps,
+       },
+       { /* sentinel */ },
+};
+MODULE_DEVICE_TABLE(platform, marvell_nfc_platform_ids);
+
+static const struct of_device_id marvell_nfc_of_ids[] = {
+       {
+               .compatible = "marvell,armada-8k-nand-controller",
+               .data = &marvell_armada_8k_nfc_caps,
+       },
+       {
+               .compatible = "marvell,armada370-nand-controller",
+               .data = &marvell_armada370_nfc_caps,
+       },
+       {
+               .compatible = "marvell,pxa3xx-nand-controller",
+               .data = &marvell_pxa3xx_nfc_caps,
+       },
+       /* Support for old/deprecated bindings: */
+       {
+               .compatible = "marvell,armada-8k-nand",
+               .data = &marvell_armada_8k_nfc_legacy_caps,
+       },
+       {
+               .compatible = "marvell,armada370-nand",
+               .data = &marvell_armada370_nfc_legacy_caps,
+       },
+       {
+               .compatible = "marvell,pxa3xx-nand",
+               .data = &marvell_pxa3xx_nfc_legacy_caps,
+       },
+       { /* sentinel */ },
+};
+MODULE_DEVICE_TABLE(of, marvell_nfc_of_ids);
+
+static struct platform_driver marvell_nfc_driver = {
+       .driver = {
+               .name           = "marvell-nfc",
+               .of_match_table = marvell_nfc_of_ids,
+       },
+       .id_table = marvell_nfc_platform_ids,
+       .probe = marvell_nfc_probe,
+       .remove = marvell_nfc_remove,
+};
+module_platform_driver(marvell_nfc_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Marvell NAND controller driver");