[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20140520174254.GQ28907@ld-irv-0074>
Date: Tue, 20 May 2014 10:42:54 -0700
From: Brian Norris <computersforpeace@...il.com>
To: Alexey Brodkin <Alexey.Brodkin@...opsys.com>
Cc: linux-mtd@...ts.infradead.org,
Ezequiel Garcia <ezequiel.garcia@...e-electrons.com>,
Vineet Gupta <Vineet.Gupta1@...opsys.com>,
Grant Likely <grant.likely@...aro.org>,
David Woodhouse <dwmw2@...radead.org>,
Francois Bedard <Francois.Bedard@...opsys.com>,
linux-kernel@...r.kernel.org, devicetree@...r.kernel.org
Subject: Re: [PATCH v2] axs_nand - add driver for NAND controller used on
Synopsys AXS dev boards
Hi Alexey,
On Fri, Apr 04, 2014 at 11:38:09AM +0400, Alexey Brodkin wrote:
> Signed-off-by: Alexey Brodkin <abrodkin@...opsys.com>
>
> Reviewed-by: Ezequiel Garcia <ezequiel.garcia@...e-electrons.com>
>
> Cc: Ezequiel Garcia <ezequiel.garcia@...e-electrons.com>
> Cc: Vineet Gupta <vgupta@...opsys.com>
> cc: Brian Norris <computersforpeace@...il.com>
> Cc: Grant Likely <grant.likely@...aro.org>
> Cc: David Woodhouse <dwmw2@...radead.org>
> Cc: Francois Bedard <fbedard@...opsys.com>
> Cc: linux-kernel@...r.kernel.org
> Cc: devicetree@...r.kernel.org
> ---
>
> In this re-spin I addressed Ezequiel comments.
>
> Changes compared to v1:
> * Minor code clean-up
> * Fixed typos
> * axs_flag_is_set() replaced with axs_flag_wait_and_reset()
> * axs_flag_wait_and_reset() has built-in check for timeout instead of endless
> while() loop initially used
>
> Documentation/devicetree/bindings/mtd/axs-nand.txt | 17 +
> drivers/mtd/nand/Kconfig | 6 +
> drivers/mtd/nand/Makefile | 1 +
> drivers/mtd/nand/axs_nand.c | 411 +++++++++++++++++++++
> 4 files changed, 435 insertions(+)
> create mode 100644 Documentation/devicetree/bindings/mtd/axs-nand.txt
> create mode 100644 drivers/mtd/nand/axs_nand.c
>
> diff --git a/Documentation/devicetree/bindings/mtd/axs-nand.txt b/Documentation/devicetree/bindings/mtd/axs-nand.txt
> new file mode 100644
> index 0000000..c522b7f
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/mtd/axs-nand.txt
> @@ -0,0 +1,17 @@
> +* Synopsys AXS NAND controller
> +
> +Required properties:
> + - compatible: must be "snps,axs-nand"
> + - reg: physical base address and size of the registers map
> +
> +The device tree may optionally contain sub-nodes describing partitions of the
> +address space. See partition.txt for more detail.
> +
> +Examples:
> +
> +flash@...6000 {
> + #address-cells = <1>;
> + #size-cells = <1>;
> + compatible = "snps,axs-nand";
> + reg = < 0x16000 0x200 >;
> +};
> diff --git a/drivers/mtd/nand/Kconfig b/drivers/mtd/nand/Kconfig
> index 93ae6a6..3661822 100644
> --- a/drivers/mtd/nand/Kconfig
> +++ b/drivers/mtd/nand/Kconfig
> @@ -510,4 +510,10 @@ config MTD_NAND_XWAY
> Enables support for NAND Flash chips on Lantiq XWAY SoCs. NAND is attached
> to the External Bus Unit (EBU).
>
> +config MTD_NAND_AXS
> + tristate "Support for NAND on Synopsys AXS development board"
> + help
> + Enables support for NAND Flash chips on Synopsys AXS development
> + boards.
> +
> endif # MTD_NAND
> diff --git a/drivers/mtd/nand/Makefile b/drivers/mtd/nand/Makefile
> index 542b568..635a918 100644
> --- a/drivers/mtd/nand/Makefile
> +++ b/drivers/mtd/nand/Makefile
> @@ -49,5 +49,6 @@ obj-$(CONFIG_MTD_NAND_JZ4740) += jz4740_nand.o
> obj-$(CONFIG_MTD_NAND_GPMI_NAND) += gpmi-nand/
> obj-$(CONFIG_MTD_NAND_XWAY) += xway_nand.o
> obj-$(CONFIG_MTD_NAND_BCM47XXNFLASH) += bcm47xxnflash/
> +obj-$(CONFIG_MTD_NAND_AXS) += axs_nand.o
>
> nand-objs := nand_base.o nand_bbt.o
> diff --git a/drivers/mtd/nand/axs_nand.c b/drivers/mtd/nand/axs_nand.c
> new file mode 100644
> index 0000000..9ee21b6
> --- /dev/null
> +++ b/drivers/mtd/nand/axs_nand.c
> @@ -0,0 +1,411 @@
> +/*
> + * Copyright (C) 2014 Synopsys, Inc. (www.synopsys.com)
> + *
> + * Driver for NAND controller on Synopsys AXS development board.
> + *
> + * This file is subject to the terms and conditions of the GNU General Public
> + * License v2. See the file COPYING in the main directory of this archive for
> + * more details.
> + */
> +
> +#include <linux/delay.h>
> +#include <linux/dma-mapping.h>
> +#include <linux/slab.h>
> +#include <linux/module.h>
> +#include <linux/mtd/mtd.h>
> +#include <linux/mtd/nand.h>
> +#include <linux/mtd/partitions.h>
> +#include <linux/of_address.h>
> +#include <linux/of_platform.h>
> +#include <linux/io.h>
> +
> +/*
> + * There's an issue with DMA'd data if data buffer is cached.
> + * So to make NAND storage available for now we'll map data buffer in
> + * uncached memory.
What sort of issue? A hardware bug? I'd prefer to have a single
implementation here if possible.
> + *
> + * As soon as issue with cached buffer is resolved following define to be
> + * removed as well as sources it enables.
> + */
> +#define DATA_BUFFER_UNCACHED
> +
> +#define BUS_WIDTH 8 /* AXI data bus width in bytes */
> +
> +/* DMA buffer descriptor masks */
> +#define BD_STAT_OWN (1 << 31)
> +#define BD_STAT_BD_FIRST (1 << 3)
> +#define BD_STAT_BD_LAST (1 << 2)
> +#define BD_SIZES_BUFFER1_MASK 0xfff
> +
> +#define BD_STAT_BD_COMPLETE (BD_STAT_BD_FIRST | BD_STAT_BD_LAST)
> +
> +/* Controller command types */
> +#define B_CT_ADDRESS (0x0 << 16) /* Address operation */
> +#define B_CT_COMMAND (0x1 << 16) /* Command operation */
> +#define B_CT_WRITE (0x2 << 16) /* Write operation */
> +#define B_CT_READ (0x3 << 16) /* Read operation */
> +
> +/* Controller command options */
> +#define B_WFR (1 << 19) /* 1b - Wait for ready */
> +#define B_LC (1 << 18) /* 1b - Last cycle */
> +#define B_IWC (1 << 13) /* 1b - Interrupt when complete */
> +
> +enum {
> + NAND_ISR_DATAREQUIRED = 0,
> + NAND_ISR_TXUNDERFLOW,
> + NAND_ISR_TXOVERFLOW,
> + NAND_ISR_DATAAVAILABLE,
> + NAND_ISR_RXUNDERFLOW,
> + NAND_ISR_RXOVERFLOW,
> + NAND_ISR_TXDMACOMPLETE,
> + NAND_ISR_RXDMACOMPLETE,
> + NAND_ISR_DESCRIPTORUNAVAILABLE,
> + NAND_ISR_CMDDONE,
> + NAND_ISR_CMDAVAILABLE,
> + NAND_ISR_CMDERROR,
> + NAND_ISR_DATATRANSFEROVER,
> + NAND_ISR_NONE
> +};
> +
> +enum {
> + AC_FIFO = 0, /* Address and command fifo */
> + IDMAC_BDADDR = 0x18, /* IDMAC descriptor list base address */
> + INT_STATUS = 0x118, /* Interrupt status register */
> + INT_CLR_STATUS = 0x120 /* Interrupt clear status register */
> +};
> +
> +#define AXS_BUF_SIZE (PAGE_ALIGN(NAND_MAX_PAGESIZE + NAND_MAX_OOBSIZE))
NAND_MAX_PAGESIZE and NAND_MAX_OOBSIZE have been removed. Please rebase
on l2-mtd.git for testing.
> +
> +struct asx_nand_bd {
> + __le32 status; /* DES0 */
> + __le32 sizes; /* DES1 */
> + dma_addr_t buffer_ptr0; /* DES2 */
> + dma_addr_t buffer_ptr1; /* DES3 */
> +};
> +
> +struct axs_nand_host {
> + struct nand_chip nand_chip;
> + struct mtd_info mtd;
> + void __iomem *io_base;
> + struct device *dev;
> + struct asx_nand_bd *bd; /* Buffer descriptor */
> + dma_addr_t bd_dma; /* DMA handle for buffer descriptor */
> + uint8_t *db; /* Data buffer */
> + dma_addr_t db_dma; /* DMA handle for data buffer */
> +};
> +
> +/**
> + * reg_set - Sets register with provided value.
> + * @host: Pointer to private data structure.
> + * @reg: Register offset from base address.
> + * @value: Value to set in register.
> + */
> +static inline void reg_set(struct axs_nand_host *host, int reg, int value)
> +{
> + iowrite32(value, host->io_base + reg);
> +}
> +
> +/**
> + * reg_get - Gets value of specified register.
> + * @host: Pointer to private data structure.
> + * @reg: Register offset from base address.
> + *
> + * returns: Value of requested register.
> + */
> +static inline unsigned int reg_get(struct axs_nand_host *host, int reg)
> +{
> + return ioread32(host->io_base + reg);
> +}
> +
> +/* Maximum number of milliseconds we wait for flag to appear */
> +#define AXS_FLAG_WAIT_DELAY 1000
> +
> +/**
> + * axs_flag_wait_and_reset - Waits until requested flag in INT_STATUS register
> + * is set by HW and resets it by writing "1" in INT_CLR_STATUS.
> + * @host: Pointer to private data structure.
> + * @flag: Bit/flag offset in INT_STATUS register
> + */
> +static void axs_flag_wait_and_reset(struct axs_nand_host *host, int flag)
> +{
> + unsigned int i;
> +
> + for (i = 0; i < AXS_FLAG_WAIT_DELAY * 100; i++) {
> + unsigned int status = reg_get(host, INT_STATUS);
> +
> + if (status & (1 << flag)) {
> + reg_set(host, INT_CLR_STATUS, 1 << flag);
> + return;
> + }
> +
> + udelay(10);
> + }
> +
> + /*
> + * Since we cannot report this problem any further than
> + * axs_nand_{write|read}_buf() letting user know there's a problem.
> + */
> + dev_err(host->dev, "Waited too long (%d s.) for flag/bit %d\n",
> + AXS_FLAG_WAIT_DELAY, flag);
> +}
I see Ezequiel commented about using this vs. waitfunc. I think your
usage is OK here.
But do you really need a polling loop here? Does your hardware have any
kind of interrupt mechanism for this sort of thing? It is not really
good practice to do I/O this way, as it keeps your CPU busy doing
pointless work. At a minimum, maybe you can use cond_resched()?
> +
> +/**
> + * axs_nand_write_buf - write buffer to chip
> + * @mtd: MTD device structure
> + * @buf: Data buffer
> + * @len: Number of bytes to write
> + */
> +static void axs_nand_write_buf(struct mtd_info *mtd,
> + const uint8_t *buf, int len)
> +{
> + struct nand_chip *this = mtd->priv;
> + struct axs_nand_host *host = this->priv;
> +
> + memcpy(host->db, buf, len);
> +#ifndef DATA_BUFFER_UNCACHED
> + dma_sync_single_for_device(host->dev, host->db_dma, len, DMA_TO_DEVICE);
> +#endif
> +
> + /* Setup buffer descriptor */
> + host->bd->status = BD_STAT_OWN | BD_STAT_BD_COMPLETE;
> + host->bd->sizes = cpu_to_le32(ALIGN(len, BUS_WIDTH) &
> + BD_SIZES_BUFFER1_MASK);
> + host->bd->buffer_ptr0 = cpu_to_le32(host->db_dma);
> + host->bd->buffer_ptr1 = 0;
> +
> + /* Issue "write" command */
> + reg_set(host, AC_FIFO, B_CT_WRITE | B_WFR | B_IWC | B_LC | (len - 1));
> +
> + /* Wait for NAND command and DMA to complete */
> + axs_flag_wait_and_reset(host, NAND_ISR_CMDDONE);
> + axs_flag_wait_and_reset(host, NAND_ISR_TXDMACOMPLETE);
> +}
> +
> +/**
> + * axs_nand_read_buf - read chip data into buffer
> + * @mtd: MTD device structure
> + * @buf: Buffer to store data
> + * @len: Number of bytes to read
> + */
> +static void axs_nand_read_buf(struct mtd_info *mtd, uint8_t *buf, int len)
> +{
> + struct nand_chip *this = mtd->priv;
> + struct axs_nand_host *host = this->priv;
> +
> + /* Setup buffer descriptor */
> + host->bd->status = BD_STAT_OWN | BD_STAT_BD_COMPLETE;
> + host->bd->sizes = cpu_to_le32(ALIGN(len, BUS_WIDTH) &
> + BD_SIZES_BUFFER1_MASK);
> + host->bd->buffer_ptr0 = cpu_to_le32(host->db_dma);
> + host->bd->buffer_ptr1 = 0;
> +
> + /* Issue "read" command */
> + reg_set(host, AC_FIFO, B_CT_READ | B_WFR | B_IWC | B_LC | (len - 1));
> +
> + /* Wait for NAND command and DMA to complete */
> + axs_flag_wait_and_reset(host, NAND_ISR_CMDDONE);
> + axs_flag_wait_and_reset(host, NAND_ISR_RXDMACOMPLETE);
> +
> +#ifndef DATA_BUFFER_UNCACHED
> + dma_sync_single_for_cpu(host->dev, host->db_dma, len, DMA_FROM_DEVICE);
> +#endif
> + if (buf != host->db)
> + memcpy(buf, host->db, len);
> +}
> +
> +/**
> + * axs_nand_read_byte - read one byte from the chip
> + * @mtd: MTD device structure
> + *
> + * returns: read data byte
> + */
> +static uint8_t axs_nand_read_byte(struct mtd_info *mtd)
> +{
> + struct nand_chip *this = mtd->priv;
> + struct axs_nand_host *host = this->priv;
> +
> + axs_nand_read_buf(mtd, host->db, sizeof(uint8_t));
> + return (uint8_t)host->db[0];
> +}
> +
> +/**
> + * axs_nand_read_word - read one word from the chip
> + * @mtd: MTD device structure
> + *
> + * returns: read data word
> + */
> +static uint16_t axs_nand_read_word(struct mtd_info *mtd)
> +{
> + struct nand_chip *this = mtd->priv;
> + struct axs_nand_host *host = this->priv;
> +
> + axs_nand_read_buf(mtd, host->db, sizeof(uint16_t));
> + return (uint16_t)host->db[0];
This doesn't look right. You don't get a "word" access simply by casting
your uint8_t into uint16_t. It's possible you're looking for something
more like this:
return ((uint16_t *)host->db)[0];
But then, you'd need to worry about endianness, and how your controller
handles x16 buswidth devices. Have you tested anything with word access,
or is this just an untested implementation?
> +}
> +
> +/**
> + * axs_nand_cmd_ctrl - hardware specific access to control-lines
> + * @mtd: MTD device structure
> + * @cmd: NAND command
> + * @ctrl: NAND control options
> + */
> +static void axs_nand_cmd_ctrl(struct mtd_info *mtd, int cmd,
> + unsigned int ctrl)
> +{
> + struct nand_chip *nand_chip = mtd->priv;
> + struct axs_nand_host *host = nand_chip->priv;
> +
> + if (cmd == NAND_CMD_NONE)
> + return;
> +
> + cmd = cmd & 0xff;
> +
> + switch (ctrl & (NAND_ALE | NAND_CLE)) {
> + /* Address */
> + case NAND_ALE:
> + cmd |= B_CT_ADDRESS;
> + break;
> +
> + /* Command */
> + case NAND_CLE:
> + cmd |= B_CT_COMMAND | B_WFR;
> +
> + break;
> +
> + default:
> + dev_err(host->dev, "Unknown ctrl %#x\n", ctrl);
> + return;
> + }
> +
> + reg_set(host, AC_FIFO, cmd | B_LC);
> + axs_flag_wait_and_reset(host, NAND_ISR_CMDDONE);
> +}
> +
> +static int axs_nand_probe(struct platform_device *pdev)
> +{
> + struct mtd_part_parser_data ppdata;
> + struct nand_chip *nand_chip;
> + struct axs_nand_host *host;
> + struct resource res_regs;
> + struct mtd_info *mtd;
> + int err;
> +
> + if (!pdev->dev.of_node)
> + return -ENODEV;
If you follow my suggestions below, then this driver won't technically
be dependent on device tree (you could potentially instantiate a
platform_device through some other means), so maybe the above check will
no longer be useful? I'm not saying you would *want* to be writing board
files, but at least I don't think this driver really needs the check, if
it's only handling generic device driver tasks.
> +
> + /* Get registers base address from device tree */
> + err = of_address_to_resource(pdev->dev.of_node, 0, &res_regs);
Can you just use platform_get_resource()? I believe the OF core
automagically converts register resources from device tree into platform
resources for you.
> + if (err) {
> + dev_err(&pdev->dev, "Failed to retrieve registers base from device tree\n");
You won't need this error handling if you keep it close to
devm_ioremap_resource(), which does error checks on the resource for
you. See the comments above devm_ioremap_resource().
> + return -ENODEV;
> + }
> +
> + /* Allocate memory for the device structure (and zero it) */
> + host = kzalloc(sizeof(struct axs_nand_host), GFP_KERNEL);
Try devm_kzalloc()? That will save you some error handling. Also, you
might try the 'sizeof' style from Chapter 14 of
Documentation/CodingStyle:
host = dev_kzalloc(&pdev->dev, sizeof(*host), GFP_KERNEL);
The same 'sizeof' pattern can be used elsewhere.
> + if (!host)
> + return -ENOMEM;
> +
> + host->io_base = devm_ioremap_resource(&pdev->dev, &res_regs);
> + if (IS_ERR(host->io_base)) {
> + err = PTR_ERR(host->io_base);
> + goto out;
> + }
> + dev_dbg(&pdev->dev, "Registers base address is 0x%p\n", host->io_base);
I don't think you need the '0x' prefix to %p, as it's done automatically
for you (did you test the print?). You could also look at using the
custom %pR or %pr format for struct resource. See
Documentation/printk-formats.txt.
> +
> + host->bd = dmam_alloc_coherent(&pdev->dev, sizeof(struct asx_nand_bd),
> + &host->bd_dma, GFP_KERNEL);
Ditto about sizeof.
> + if (!host->bd) {
> + dev_err(&pdev->dev, "Failed to allocate buffer descriptor\n");
> + err = -ENOMEM;
> + goto out;
> + }
> +
> + memset(host->bd, 0, sizeof(struct asx_nand_bd));
Ditto about sizeof.
> +
> +#ifdef DATA_BUFFER_UNCACHED
> + host->db = dmam_alloc_coherent(&pdev->dev, AXS_BUF_SIZE,
> +#else
> + host->db = dmam_alloc_noncoherent(&pdev->dev, AXS_BUF_SIZE,
> +#endif
> + &host->db_dma, GFP_KERNEL);
> + if (!host->db) {
> + dev_err(&pdev->dev, "Failed to allocate data buffer\n");
> + err = -ENOMEM;
> + goto out;
> + }
> + dev_dbg(&pdev->dev, "Data buffer mapped @ %p, DMA @ %x\n", host->db,
> + host->db_dma);
For the 'dma_addr_t', you can try %pad. Again, see
Documentation/printk-formats.txt
> +
> + mtd = &host->mtd;
> + nand_chip = &host->nand_chip;
> + host->dev = &pdev->dev;
> +
> + nand_chip->priv = host;
> + mtd->priv = nand_chip;
> + mtd->name = "axs_nand";
Are you sure there will only be a single 'axs_nand' on a single system?
Or do you want to consider a unique naming system? Some drivers use:
mtd->name = dev_name(&pdev->dev);
But this could be less friendly if you have long device names based on
Device Tree addresses.
> + mtd->owner = THIS_MODULE;
> + mtd->dev.parent = &pdev->dev;
> + ppdata.of_node = pdev->dev.of_node;
> +
> + nand_chip->cmd_ctrl = axs_nand_cmd_ctrl;
> + nand_chip->read_byte = axs_nand_read_byte;
> + nand_chip->read_word = axs_nand_read_word;
> + nand_chip->write_buf = axs_nand_write_buf;
> + nand_chip->read_buf = axs_nand_read_buf;
> + nand_chip->ecc.mode = NAND_ECC_SOFT;
> +
> + dev_set_drvdata(&pdev->dev, host);
> +
> + reg_set(host, IDMAC_BDADDR, host->bd_dma);
> +
> + err = nand_scan(mtd, 1);
> + if (err)
> + goto out1;
> +
> + err = mtd_device_parse_register(mtd, NULL, &ppdata, NULL, 0);
> + if (!err)
> + return err;
> +
> + nand_release(mtd);
> +
> +out1:
> + dev_set_drvdata(&pdev->dev, NULL);
^^ I believe this step is unnecessary now, as it's done by the driver
core.
> +out:
> + kfree(host);
> + return err;
> +}
> +
> +static int axs_nand_remove(struct platform_device *ofdev)
> +{
> + struct axs_nand_host *host = dev_get_drvdata(&ofdev->dev);
> + struct mtd_info *mtd = &host->mtd;
> +
> + nand_release(mtd);
> + dev_set_drvdata(&ofdev->dev, NULL);
^^ Same here
> + kfree(host);
> +
> + return 0;
> +}
> +
> +static const struct of_device_id axs_nand_dt_ids[] = {
> + { .compatible = "snps,axs-nand", },
> + { /* Sentinel */ }
> +};
> +
> +MODULE_DEVICE_TABLE(of, axs_nand_match);
> +
> +static struct platform_driver axs_nand_driver = {
> + .probe = axs_nand_probe,
> + .remove = axs_nand_remove,
> + .driver = {
> + .name = "asx-nand",
> + .owner = THIS_MODULE,
> + .of_match_table = axs_nand_dt_ids,
> + },
> +};
> +
> +module_platform_driver(axs_nand_driver);
> +
> +MODULE_AUTHOR("Alexey Brodkin <abrodkin@...opsys.com>");
> +MODULE_DESCRIPTION("NAND driver for Synopsys AXS development board");
> +MODULE_LICENSE("GPL");
Brian
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists