[PATCH] [MTD] MTD driver for LPDDR2-NVM PCM memories
Brian Norris
computersforpeace at gmail.com
Thu Aug 15 03:06:06 EDT 2013
Hi Vincenzo,
My review is below.
BTW, can you prefix your patch subject with a good subsystem prefix?
Like "mtd: lpddr: add driver for LPDDR2-NVM PCM memories".
On Thu, Jun 20, 2013 at 08:37:53PM +0200, Vincenzo Aliberti wrote:
> Added Support for PCM memories with LPDDR2-NVM interface
>
> Signed-off-by: Vincenzo Aliberti <vincenzo.aliberti at gmail.com>
> ---
> drivers/mtd/lpddr/Kconfig | 12 +-
> drivers/mtd/lpddr/Makefile | 1 +
> drivers/mtd/lpddr/lpddr2_nvm.c | 556 ++++++++++++++++++++++++++++++++++++++++
> include/uapi/mtd/mtd-abi.h | 1 +
> 4 files changed, 568 insertions(+), 2 deletions(-)
> create mode 100644 drivers/mtd/lpddr/lpddr2_nvm.c
>
> diff --git a/drivers/mtd/lpddr/Kconfig b/drivers/mtd/lpddr/Kconfig
> index 265f969..468f06d 100644
> --- a/drivers/mtd/lpddr/Kconfig
> +++ b/drivers/mtd/lpddr/Kconfig
> @@ -1,5 +1,5 @@
> -menu "LPDDR flash memory drivers"
> - depends on MTD!=n
> +menu "LPDDR & LPDDR2 PCM memory drivers"
> + depends on MTD
>
> config MTD_LPDDR
> tristate "Support for LPDDR flash chips"
> @@ -17,4 +17,12 @@ config MTD_QINFO_PROBE
> Window QINFO interface, permits software to be used for entire
> families of devices. This serves similar purpose of CFI on legacy
> Flash products
> +
> +config MTD_LPDDR2_NVM
> + depends on MTD
> + tristate "Support for LPDDR2-NVM flash chips"
> + help
> + This option enables support of PCM memories with a LPDDR2-NVM
> + (Low power double data rate 2) interface.
> +
> endmenu
> diff --git a/drivers/mtd/lpddr/Makefile b/drivers/mtd/lpddr/Makefile
> index da48e46..881d440 100644
> --- a/drivers/mtd/lpddr/Makefile
> +++ b/drivers/mtd/lpddr/Makefile
> @@ -4,3 +4,4 @@
>
> obj-$(CONFIG_MTD_QINFO_PROBE) += qinfo_probe.o
> obj-$(CONFIG_MTD_LPDDR) += lpddr_cmds.o
> +obj-$(CONFIG_MTD_LPDDR2_NVM) += lpddr2_nvm.o
> diff --git a/drivers/mtd/lpddr/lpddr2_nvm.c b/drivers/mtd/lpddr/lpddr2_nvm.c
> new file mode 100644
> index 0000000..068f95d
> --- /dev/null
> +++ b/drivers/mtd/lpddr/lpddr2_nvm.c
> @@ -0,0 +1,556 @@
> +/*
> + * LPDDR2-NVM MTD driver. This module provides read, write, erase, lock/unlock
> + * support for LPDDR2-NVM PCM memories
> + *
> + * Copyright © 2012 Micron Technology, Inc.
2013 maybe? But I won't insist, if this was really authored and used out
of tree last year.
> + *
> + * Vincenzo Aliberti <vincenzo.aliberti at gmail.com>
> + * Domenico Manna <domenico.manna at gmail.com>
> + * Many thanks to Andrea Vigilante for initial enabling
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version 2
> + * of the License, or (at your option) any later version.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + * GNU General Public License for more details.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
> + * 02110-1301, USA.
> + */
> +
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/kernel.h>
> +#include <linux/mtd/map.h>
> +#include <linux/mtd/mtd.h>
> +#include <linux/mtd/partitions.h>
> +#include <linux/slab.h>
> +#include <linux/platform_device.h>
> +#include <linux/ioport.h>
> +#include <linux/sched.h>
> +
> +/* Parameters */
> +#define ERASE_BLOCKSIZE (0x00020000/2) /* in Word */
> +#define WRITE_BUFFSIZE (0x00000400/2) /* in Word */
> +#define OW_BASE_ADDRESS 0x00000000 /* OW offset */
> +
> +/* PFOW symbols address offset */
> +#define PFOW_QUERY_STRING_P (0x0000/2) /* in Word */
> +#define PFOW_QUERY_STRING_F (0x0002/2) /* in Word */
> +#define PFOW_QUERY_STRING_O (0x0004/2) /* in Word */
> +#define PFOW_QUERY_STRING_W (0x0006/2) /* in Word */
> +
> +/* OW registers address */
> +#define CMD_CODE_OFS (0x0080/2) /* in Word */
> +#define CMD_DATA_OFS (0x0084/2) /* in Word */
> +#define CMD_ADD_L_OFS (0x0088/2) /* in Word */
> +#define CMD_ADD_H_OFS (0x008A/2) /* in Word */
> +#define MPR_L_OFS (0x0090/2) /* in Word */
> +#define MPR_H_OFS (0x0092/2) /* in Word */
> +#define CMD_EXEC_OFS (0x00C0/2) /* in Word */
> +#define STATUS_REG_OFS (0x00CC/2) /* in Word */
> +#define PRG_BUFFER_OFS (0x0010/2) /* in Word */
> +
> +/* Datamask */
> +#define MR_CFGMASK 0x8000
> +#define SR_OK_DATAMASK 0x0080
> +
> +/* LPDDR2-NVM Commands */
> +#define LPDDR2_NVM_LOCK 0x0061
> +#define LPDDR2_NVM_UNLOCK 0x0062
> +#define LPDDR2_NVM_SW_PROGRAM 0x0041
> +#define LPDDR2_NVM_SW_OVERWRITE 0x0042
> +#define LPDDR2_NVM_BUF_PROGRAM 0x00E9
> +#define LPDDR2_NVM_BUF_OVERWRITE 0x00EA
> +#define LPDDR2_NVM_ERASE 0x0020
> +
> +/*
> + * Internal Type Definitions
> + * pcm_int_data contains memory controller details:
> + * @reg_data : LPDDR2_MODE_REG_DATA register address after remapping
> + * @reg_cfg : LPDDR2_MODE_REG_CFG register address after remapping
> + * &bus_width: memory bus-width (eg: x16 2 Bytes, x32 4 Bytes)
> + */
> +struct pcm_int_data {
> + void __iomem *reg_data;
> + void __iomem *reg_cfg;
> + int bus_width;
> +};
> +
> +static DEFINE_MUTEX(lpdd2_nvm_mutex);
> +
> +/*
> + * Build a map_word starting from an u_long
> + */
> +static inline map_word build_map_word(u_long myword)
> +{
> + map_word val = { {0} };
> + val.x[0] = myword;
> + return val;
> +}
> +
> +/*
> + * Build Mode Register Configuration DataMask based on device bus-width
> + */
> +static inline u_int build_mr_cfgmask(u_int bus_width)
> +{
> + u_int val = MR_CFGMASK;
> +
> + if (bus_width == 0x0004) /* x32 device */
> + val = val<<16;
Add space around the shift operator.
> +
> + return val;
> +}
> +
> +/*
> + * Build Status Register OK DataMask based on device bus-width
> + */
> +static inline u_int build_sr_ok_datamask(u_int bus_width)
> +{
> + u_int val = SR_OK_DATAMASK;
> +
> + if (bus_width == 0x0004) /* x32 device */
> + val = (val<<16)+val;
Ditto.
> +
> + return val;
> +}
> +
> +/*
> + * Evaluates Overlay Window Control Registers address
> + */
> +static inline u_long ow_reg_add(struct map_info *map, u_long offset)
> +{
> + u_long val = 0;
> + struct pcm_int_data *pcm_data = map->fldrv_priv;
> +
> + val = map->pfow_base + offset*pcm_data->bus_width;
> +
> + return val;
> +}
> +
> +/*
> + * Enable lpddr2-nvm Overlay Window
> + * Overlay Window is a memory mapped area containing all LPDDR2-NVM registers
> + * used by device commands as well as uservisible resources like Device Status
> + * Register, Device ID, etc
> + */
> +static inline void ow_enable(struct map_info *map)
> +{
> + struct pcm_int_data *pcm_data = map->fldrv_priv;
> +
> + writel_relaxed(build_mr_cfgmask(pcm_data->bus_width) | 0x18,
> + pcm_data->reg_cfg);
> + writel_relaxed(0x01, pcm_data->reg_data);
Unfortunately, writel_relaxed() is still not available on all
architectures (notably x86). So this driver either should switch to a
supported IO accessor or change its Kconfig so it depends on the
supported architectures.
> +}
> +
> +/*
> + * Disable lpddr2-nvm Overlay Window
> + * Overlay Window is a memory mapped area containing all LPDDR2-NVM registers
> + * used by device commands as well as uservisible resources like Device Status
> + * Register, Device ID, etc
> + */
> +static inline void ow_disable(struct map_info *map)
> +{
> + struct pcm_int_data *pcm_data = map->fldrv_priv;
> +
> + writel_relaxed(build_mr_cfgmask(pcm_data->bus_width) | 0x18,
> + pcm_data->reg_cfg);
> + writel_relaxed(0x02, pcm_data->reg_data);
> +}
> +
> +/*
> + * Execute lpddr2-nvm operations
> + */
> +static int lpddr2_nvm_do_op(struct map_info *map, u_long cmd_code,
> + u_long cmd_data, u_long cmd_add, u_long cmd_mpr, u_char *buf)
> +{
> + map_word add_l = { {0} }, add_h = { {0} }, mpr_l = { {0} },
> + mpr_h = { {0} }, data_l = { {0} }, cmd = { {0} },
> + exec_cmd = { {0} }, sr;
> + map_word data_h = { {0} }; /* only for 2x x16 devices stacked */
> + u_long i, status_reg, prg_buff_ofs;
> + struct pcm_int_data *pcm_data = map->fldrv_priv;
> + u_int sr_ok_datamask = build_sr_ok_datamask(pcm_data->bus_width);
> +
> + /* Builds low and high words for OW Control Registers */
> + add_l.x[0] = cmd_add&0x0000FFFF;
> + add_h.x[0] = (cmd_add>>16)&0x0000FFFF;
Spacing around the shift and the mask operator (&). Same applies
elsewhere.
> + mpr_l.x[0] = cmd_mpr&0x0000FFFF;
> + mpr_h.x[0] = (cmd_mpr>>16)&0x0000FFFF;
> + cmd.x[0] = cmd_code&0x0000FFFF;
> + exec_cmd.x[0] = 0x0001;
> + data_l.x[0] = cmd_data&0x0000FFFF;
> + data_h.x[0] = (cmd_data>>16)&0x0000FFFF; /* only for 2x x16 */
> +
> + /* Set Overlay Window Control Registers */
> + map_write(map, cmd, ow_reg_add(map, CMD_CODE_OFS));
> + map_write(map, data_l, ow_reg_add(map, CMD_DATA_OFS));
> + map_write(map, add_l, ow_reg_add(map, CMD_ADD_L_OFS));
> + map_write(map, add_h, ow_reg_add(map, CMD_ADD_H_OFS));
> + map_write(map, mpr_l, ow_reg_add(map, MPR_L_OFS));
> + map_write(map, mpr_h, ow_reg_add(map, MPR_H_OFS));
> + if (pcm_data->bus_width == 0x0004) { /* 2x16 devices stacked */
> + map_write(map, cmd, ow_reg_add(map, CMD_CODE_OFS) + 2);
> + map_write(map, data_h, ow_reg_add(map, CMD_DATA_OFS) + 2);
> + map_write(map, add_l, ow_reg_add(map, CMD_ADD_L_OFS) + 2);
> + map_write(map, add_h, ow_reg_add(map, CMD_ADD_H_OFS) + 2);
> + map_write(map, mpr_l, ow_reg_add(map, MPR_L_OFS) + 2);
> + map_write(map, mpr_h, ow_reg_add(map, MPR_H_OFS) + 2);
> + }
> +
> + /* Fill Program Buffer */
> + if ((cmd_code == LPDDR2_NVM_BUF_PROGRAM) ||
> + (cmd_code == LPDDR2_NVM_BUF_OVERWRITE)) {
> + prg_buff_ofs = (map_read(map,
> + ow_reg_add(map, PRG_BUFFER_OFS))).x[0];
> + for (i = 0; i < cmd_mpr; i++) {
> + map_write(map, build_map_word(buf[i]), map->pfow_base +
> + prg_buff_ofs + i);
> + }
> + }
> +
> + /* Command Execute */
> + map_write(map, exec_cmd, ow_reg_add(map, CMD_EXEC_OFS));
> + if (pcm_data->bus_width == 0x0004) /* 2x16 devices stacked */
> + map_write(map, exec_cmd, ow_reg_add(map, CMD_EXEC_OFS) + 2);
> +
> + /* Status Register Check */
> + do {
> + sr = map_read(map, ow_reg_add(map, STATUS_REG_OFS));
> + status_reg = sr.x[0];
> + if (pcm_data->bus_width == 0x0004) {/* 2x16 devices stacked */
> + sr = map_read(map, ow_reg_add(map,
> + STATUS_REG_OFS) + 2);
> + status_reg += sr.x[0]<<16;
> + }
> + } while ((status_reg&sr_ok_datamask) != sr_ok_datamask);
> +
> + return (((status_reg&sr_ok_datamask) == sr_ok_datamask) ? 0 : -EIO);
> +}
> +
> +/*
> + * Execute lpddr2-nvm operations @ block level
> + */
> +static int lpddr2_nvm_do_block_op(struct mtd_info *mtd, loff_t start_add,
> + uint64_t len, u_char block_op)
> +{
> + struct map_info *map = mtd->priv;
> + u_long add, end_add;
> + int ret = 0;
> +
> + mutex_lock(&lpdd2_nvm_mutex);
> +
> + ow_enable(map);
> +
> + add = start_add;
> + end_add = add + len;
> +
> + if ((add > mtd->size) || (end_add > mtd->size)) {
I don't think these checks are required, since mtd_unlock(), mtd_lock(),
and mtd_erase() do this check for you. Plus, your checks are not
exhaustive (what about add < 0, end_add < 0?).
> + ret = -EINVAL;
> + goto out;
> + }
> +
> + do {
> + ret = lpddr2_nvm_do_op(map, block_op, 0x00, add, add, NULL);
> + if (ret)
> + goto out;
> + add += mtd->erasesize;
> + } while (add < end_add);
> +
> +out:
> + ow_disable(map);
> + mutex_unlock(&lpdd2_nvm_mutex);
> + return ret;
> +}
> +
> +/*
> + * verify presence of PFOW string
> + */
> +static int lpddr2_nvm_pfow_present(struct map_info *map)
> +{
> + map_word pfow_val[4];
> + unsigned int found = 1;
> +
> + mutex_lock(&lpdd2_nvm_mutex);
> +
> + ow_enable(map);
> +
> + /* Load string from array */
> + pfow_val[0] = map_read(map, ow_reg_add(map, PFOW_QUERY_STRING_P));
> + pfow_val[1] = map_read(map, ow_reg_add(map, PFOW_QUERY_STRING_F));
> + pfow_val[2] = map_read(map, ow_reg_add(map, PFOW_QUERY_STRING_O));
> + pfow_val[3] = map_read(map, ow_reg_add(map, PFOW_QUERY_STRING_W));
> +
> + /* Verify the string loaded vs expected */
> + if (!map_word_equal(map, build_map_word('P'), pfow_val[0]))
> + found = 0;
> + if (!map_word_equal(map, build_map_word('F'), pfow_val[1]))
> + found = 0;
> + if (!map_word_equal(map, build_map_word('O'), pfow_val[2]))
> + found = 0;
> + if (!map_word_equal(map, build_map_word('W'), pfow_val[3]))
> + found = 0;
> +
> + ow_disable(map);
> +
> + mutex_unlock(&lpdd2_nvm_mutex);
> +
> + return found;
> +}
> +
> +/*
> + * lpddr2_nvm driver read method
> + */
> +static int lpddr2_nvm_read(struct mtd_info *mtd, loff_t start_add,
> + size_t len, size_t *retlen, u_char *buf)
> +{
> + struct map_info *map = mtd->priv;
> + int ret = 0;
> +
> + mutex_lock(&lpdd2_nvm_mutex);
> +
> + *retlen = len;
> +
> + if ((start_add + len) > mtd->size) {
This check is covered by mtd_read(). (And again, your checks are not
exhaustive.)
> + pr_alert("lpddr2_nvm: read out of memory range\n");
> + *retlen = 0;
> + ret = -EINVAL;
> + goto out;
> + }
> +
> + map_copy_from(map, buf, start_add, *retlen);
> +
> +out:
> + mutex_unlock(&lpdd2_nvm_mutex);
> + return ret;
> +}
> +
> +/*
> + * lpddr2_nvm driver write method
> + */
> +static int lpddr2_nvm_write(struct mtd_info *mtd, loff_t start_add,
> + size_t len, size_t *retlen, const u_char *buf)
> +{
> + struct map_info *map = mtd->priv;
> + struct pcm_int_data *pcm_data = map->fldrv_priv;
> + u_long add, current_len, tot_len, target_len, my_data;
> + u_char *write_buf = (u_char *)buf;
> + int ret = 0;
> +
> + mutex_lock(&lpdd2_nvm_mutex);
> +
> + ow_enable(map);
> +
> + /* Set start value for the variables */
> + add = start_add;
> + target_len = len;
> + tot_len = 0;
> +
> + if ((start_add + len) > mtd->size) {
Ditto about the range check. You don't need it.
> + pr_alert("lpddr2_nvm: write out of memory range\n");
> + *retlen = 0;
> + ret = -EINVAL;
> + goto out;
> + }
> +
> + while (tot_len < target_len) {
> + if (!(IS_ALIGNED(add, mtd->writesize))) { /* do sw program */
> + my_data = write_buf[tot_len];
> + my_data += (write_buf[tot_len+1])<<8;
> + if (pcm_data->bus_width == 0x0004) {/* 2x16 devices */
> + my_data += (write_buf[tot_len+2])<<16;
> + my_data += (write_buf[tot_len+3])<<24;
> + }
> + ret = lpddr2_nvm_do_op(map, LPDDR2_NVM_SW_OVERWRITE,
> + my_data, add, 0x00, NULL);
> + if (ret)
> + goto out;
> +
> + add += pcm_data->bus_width;
> + tot_len += pcm_data->bus_width;
> + } else { /* do buffer program */
> + current_len = min(target_len - tot_len,
> + (u_long) mtd->writesize);
> + ret = lpddr2_nvm_do_op(map, LPDDR2_NVM_BUF_OVERWRITE,
> + 0x00, add, current_len, write_buf + tot_len);
> + if (ret)
> + goto out;
> +
> + add += current_len;
> + tot_len += current_len;
> + }
> + }
> +
> +out:
> + *retlen = tot_len;
> + ow_disable(map);
> + mutex_unlock(&lpdd2_nvm_mutex);
> + return ret;
> +}
> +
> +/*
> + * lpddr2_nvm driver erase method
> + */
> +static int lpddr2_nvm_erase(struct mtd_info *mtd, struct erase_info *instr)
> +{
> + int ret = lpddr2_nvm_do_block_op(mtd, instr->addr, instr->len,
> + LPDDR2_NVM_ERASE);
> + if (!ret) {
> + instr->state = MTD_ERASE_DONE;
> + mtd_erase_callback(instr);
> + }
> +
> + return ret;
> +}
> +
> +/*
> + * lpddr2_nvm driver unlock method
> + */
> +static int lpddr2_nvm_unlock(struct mtd_info *mtd, loff_t start_add,
> + uint64_t len)
> +{
> + return lpddr2_nvm_do_block_op(mtd, start_add, len, LPDDR2_NVM_UNLOCK);
> +}
> +
> +/*
> + * lpddr2_nvm driver lock method
> + */
> +static int lpddr2_nvm_lock(struct mtd_info *mtd, loff_t start_add,
> + uint64_t len)
> +{
> + return lpddr2_nvm_do_block_op(mtd, start_add, len, LPDDR2_NVM_LOCK);
> +}
> +
> +/*
> + * lpddr2_nvm driver probe method
> + */
> +static int lpddr2_nvm_probe(struct platform_device *pdev)
> +{
> + int ret = 0;
> + struct map_info *lpddr2_nvm_map;
> + struct mtd_info *lpddr2_nvm_mtd;
Can you drop the prefixes for the local variables (i.e., make these
'map' and 'mtd')? They don't fit the style of other MTD drivers.
> + struct resource *address_range = platform_get_resource(pdev,
> + IORESOURCE_MEM, 0);
> + struct resource *control_regs = platform_get_resource(pdev,
> + IORESOURCE_MEM, 1);
> + struct resource *buswidth = platform_get_resource(pdev,
> + IORESOURCE_BUS, 0);
> + struct pcm_int_data *pcm_data;
> +
> + /* Allocate memory control_regs data structures */
> + pcm_data = kzalloc(sizeof(*pcm_data), GFP_KERNEL);
> + if (!control_regs) {
> + pr_alert("lpddr2_nvm_probe: Failed to allocate memory for int_data\n");
These failure messages are unnecessary; k[mz]alloc should complain.
Also, can't you use the devm_* function? e.g., devm_kzalloc(). Then you
don't have to clean up memory after yourself (fortunately, cause you
don't do any proper error handling now!). Take a look at
Documentation/driver-model/devres.txt. There are a few methods that can
help you.
> + ret = 1;
'ret' needs to be a real error code: -ENOMEM.
> + goto out;
Just 'return -ENOMEM'.
> + }
> + pcm_data->reg_data = (void *) control_regs->start;
> + pcm_data->reg_cfg = (void *) control_regs->end;
> + pcm_data->bus_width = (int) buswidth->start;
> +
> + /* Allocate memory for map_info & mtd_info data structures */
> + lpddr2_nvm_map = kzalloc(sizeof(*lpddr2_nvm_map), GFP_KERNEL);
> + if (!lpddr2_nvm_map) {
> + pr_alert("lpddr2_nvm_probe: Failed to allocate memory for map\n");
> + ret = 1;
> + goto out;
> + }
Same here. And other places.
> + lpddr2_nvm_mtd = kzalloc(sizeof(*lpddr2_nvm_mtd), GFP_KERNEL);
> + if (!lpddr2_nvm_mtd) {
> + pr_alert("lpddr2_nvm_probe: Failed to allocate memory for mtd\n");
> + ret = 1;
> + goto out;
return -ENOMEM
> + }
> +
> + /* Reserve lpddr2_nvm address range */
> + if (request_mem_region(address_range->start, (address_range->end -
> + address_range->start + 1), "lpddr2_nvm") == NULL) {
devm_request_mem_region()
> + pr_alert("lpddr2_nvm_probe: Unable to reserve memory address %#010x\n",
> + address_range->start);
This gives me a warning:
drivers/mtd/lpddr/lpddr2_nvm.c:479:3: warning: format ‘%x’ expects argument of type ‘unsigned int’, but argument 2 has type ‘resource_size_t’ [-Wformat]
And for the remaining error messages, you should probably tone down your
pr_alert()'s to just pr_err().
You might also look at defining pr_fmt for your driver, so you can drop
the "lpddr2_...:" prefixes to every print message. You can use it to
automatically add __func__ or KBUILD_MODNAME or some fixed prefix (e.g.,
"lpddr2_nvm: "). But manually coding the function name isn't great
practice.
> + ret = -EBUSY;
> + goto out;
return -EBUSY;
> + }
> +
> + /* Populate map_info data structure */
> + *lpddr2_nvm_map = (struct map_info) {
> + .virt = ioremap_nocache(address_range->start,
> + (address_range->end-address_range->start+1)),
devm_ioremap_nocache()
> + .name = pdev->dev.init_name,
> + .phys = address_range->start,
> + .size = address_range->end - address_range->start + 1,
> + .bankwidth = pcm_data->bus_width/2,
> + .pfow_base = OW_BASE_ADDRESS,
> + .fldrv_priv = pcm_data,
> + };
> +
> + simple_map_init(lpddr2_nvm_map); /* fill with default methods */
> +
> + /* Populate mtd_info data structure */
> + *lpddr2_nvm_mtd = (struct mtd_info) {
I'm a little torn on this struct-style initialization. It certainly
doesn't match the rest of MTD drivers.
> + .name = pdev->dev.init_name,
> + .type = MTD_RAM,
> + .priv = lpddr2_nvm_map,
> + .size = address_range->end - address_range->start + 1,
> + .erasesize = ERASE_BLOCKSIZE * pcm_data->bus_width,
> + .writesize = 1,
> + .writebufsize = WRITE_BUFFSIZE * pcm_data->bus_width,
> + .flags = (MTD_CAP_PCM | MTD_POWERUP_LOCK),
> + ._read = lpddr2_nvm_read,
> + ._read_oob = NULL,
You don't need any of these NULL assignments, since you allocate with
kzalloc().
> + ._write = lpddr2_nvm_write,
> + ._write_oob = NULL,
Ditto. (And more below)
> + ._erase = lpddr2_nvm_erase,
> + ._unlock = lpddr2_nvm_unlock,
> + ._lock = lpddr2_nvm_lock,
> + ._sync = NULL,
> + ._suspend = NULL,
> + ._resume = NULL,
> + ._block_isbad = NULL,
> + ._block_markbad = NULL,
> + };
> +
> + /* Verify the presence of the device looking for PFOW string */
> + if (lpddr2_nvm_pfow_present(lpddr2_nvm_map)) {
> + pr_alert("lpddr2_nvm_probe: device recognized\n");
> + /* Register the mtd_partition structure */
> + if (mtd_add_partition(lpddr2_nvm_mtd, "my_pcm", 0x10000000,
> + 0x00)){
> + pr_alert("lpddr2_nvm_probe: Failed to add mtd partition\n");
> + ret = 1;
> + goto out;
> + }
> + ret = 0;
> + } else {
> + pr_alert("%s: PFOW string not found at address %#010lx\n",
> + lpddr2_nvm_map->name, lpddr2_nvm_map->pfow_base);
> + ret = 1;
> + }
This error block is strange. It doesn't need the whole if/else, and you
need to pick a real return value, not 1. I'd do:
if (!lpddr2_nvm_pfow_present(...)) {
pr_err(...);
return -EINVAL;
}
return mtd_add_partition(...);
> +
> +out:
> + return ret;
This 'out' clause doesn't do anything. (Right now, this is incorrect,
since you will leak a bunch of memory, but with devm_*(), you still
won't need anything here.) So it makes a better probe routine if you
just return directly from the errors instead of using 'goto out'. And be
sure to use the appropriate codes, as mentioned above.
> +}
> +
> +/* Initialize platform_driver data structure for lpddr2_nvm */
> +static struct platform_driver lpddr2_nvm_drv = {
> + .driver = {
> + .name = "lpddr2_nvm",
> + },
> + .probe = lpddr2_nvm_probe,
> + .remove = NULL,
> + .suspend = NULL,
> + .resume = NULL,
I don't think you need to specify NULL pointers for the unused
functions, since the static struct will be automatically zeroed out.
Also, this driver is non-removable? Can you provide a .remove function?
> +};
> +
> +module_platform_driver(lpddr2_nvm_drv);
> +
> +MODULE_LICENSE("GPL");
> +MODULE_AUTHOR("Vincenzo Aliberti <vincenzo.aliberti at gmail.com>");
> +MODULE_DESCRIPTION("MTD driver for LPDDR2-NVM PCM memories");
> +
Get rid of the extra newline.
> diff --git a/include/uapi/mtd/mtd-abi.h b/include/uapi/mtd/mtd-abi.h
> index 36eace0..33f3036 100644
> --- a/include/uapi/mtd/mtd-abi.h
> +++ b/include/uapi/mtd/mtd-abi.h
> @@ -109,6 +109,7 @@ struct mtd_write_req {
> #define MTD_CAP_RAM (MTD_WRITEABLE | MTD_BIT_WRITEABLE | MTD_NO_ERASE)
> #define MTD_CAP_NORFLASH (MTD_WRITEABLE | MTD_BIT_WRITEABLE)
> #define MTD_CAP_NANDFLASH (MTD_WRITEABLE)
> +#define MTD_CAP_PCM (MTD_WRITEABLE | MTD_BIT_WRITEABLE | MTD_NO_ERASE)
Is PCM a generally-useful class type? Or perhaps it could more generally
be called NVRAM? Or should it really just use MTD_CAP_RAM, since it has
the same capabilities? I'm OK with any of these options, just checking
what the rationale is.
>
> /* Obsolete ECC byte placement modes (used with obsolete MEMGETOOBSEL) */
> #define MTD_NANDECC_OFF 0 // Switch off ECC (Not recommended)
Brian
More information about the linux-mtd
mailing list