Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 24 Jun 2020 14:51:12 -0700
From:   Guenter Roeck <linux@roeck-us.net>
To:     alexandru.tachici@analog.com
Cc:     linux-hwmon@vger.kernel.org, linux-kernel@vger.kernel.org,
        devicetree@vger.kernel.org, robh+dt@kernel.org
Subject: Re: [PATCH v4 5/7] hwmon: pmbus: adm1266: read blackbox
Message-ID: <20200624215112.GA75386@roeck-us.net>
References: <20200623173659.41358-1-alexandru.tachici@analog.com>
 <20200623173659.41358-6-alexandru.tachici@analog.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200623173659.41358-6-alexandru.tachici@analog.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Jun 23, 2020 at 08:36:57PM +0300, alexandru.tachici@analog.com wrote:
> From: Alexandru Tachici <alexandru.tachici@analog.com>
> 
> Use the nvmem kernel api to expose the black box
> chip functionality to userspace.
> 
> Signed-off-by: Alexandru Tachici <alexandru.tachici@analog.com>
> ---
>  drivers/hwmon/pmbus/adm1266.c | 134 ++++++++++++++++++++++++++++++++++
>  1 file changed, 134 insertions(+)
> 
> diff --git a/drivers/hwmon/pmbus/adm1266.c b/drivers/hwmon/pmbus/adm1266.c
> index 0960ead8d96a..b9e92ab1e39a 100644
> --- a/drivers/hwmon/pmbus/adm1266.c
> +++ b/drivers/hwmon/pmbus/adm1266.c
> @@ -15,6 +15,8 @@
>  #include <linux/init.h>
>  #include <linux/kernel.h>
>  #include <linux/module.h>
> +#include <linux/nvmem-consumer.h>
> +#include <linux/nvmem-provider.h>
>  #include <linux/proc_fs.h>
>  #include <linux/slab.h>
>  #include <linux/uaccess.h>
> @@ -22,10 +24,13 @@
>  #include <linux/adm1266.h>
>  #include "pmbus.h"
>  
> +#define ADM1266_BLACKBOX_CONFIG	0xD3
>  #define ADM1266_PDIO_CONFIG	0xD4
>  #define ADM1266_GO_COMMAND	0xD8
>  #define ADM1266_READ_STATE	0xD9
> +#define ADM1266_READ_BLACKBOX	0xDE
>  #define ADM1266_GPIO_CONFIG	0xE1
> +#define ADM1266_BLACKBOX_INFO	0xE6
>  #define ADM1266_PDIO_STATUS	0xE9
>  #define ADM1266_GPIO_STATUS	0xEA
>  
> @@ -42,6 +47,9 @@
>  #define ADM1266_PDIO_GLITCH_FILT(x)	FIELD_GET(GENMASK(12, 9), x)
>  #define ADM1266_PDIO_OUT_CFG(x)		FIELD_GET(GENMASK(2, 0), x)
>  
> +#define ADM1266_BLACKBOX_OFFSET		0x7F700
> +#define ADM1266_BLACKBOX_SIZE		64
> +
>  #define ADM1266_PMBUS_BLOCK_MAX		255
>  
>  DECLARE_CRC8_TABLE(pmbus_crc_table);
> @@ -52,6 +60,17 @@ struct adm1266_data {
>  	const char *gpio_names[ADM1266_GPIO_NR + ADM1266_PDIO_NR];
>  	struct i2c_client *client;
>  	struct mutex ioctl_mutex; /* lock ioctl access */
> +	struct nvmem_config nvmem_config;
> +	struct nvmem_device *nvmem;
> +	u8 *dev_mem;
> +};
> +
> +static const struct nvmem_cell_info adm1266_nvmem_cells[] = {
> +	{
> +		.name           = "blackbox",
> +		.offset         = ADM1266_BLACKBOX_OFFSET,
> +		.bytes          = 2048,
> +	},
>  };
>  
>  /* Different from Block Read as it sends data and waits for the slave to
> @@ -404,6 +423,117 @@ static int adm1266_init_procfs(struct adm1266_data *data)
>  	return 0;
>  }
>  
> +static int adm1266_nvmem_read_blackbox(struct adm1266_data *data, u8 *buf)
> +{
> +	u8 read_buf[5];
> +	char index;
> +	int record_count;
> +	int ret;
> +
> +	ret = i2c_smbus_read_block_data(data->client, ADM1266_BLACKBOX_INFO,
> +					read_buf);
> +	if (ret < 0)
> +		return ret;
> +
> +	record_count = read_buf[3];
> +
> +	for (index = 0; index < record_count; index++) {
> +		ret = pmbus_block_xfer(data->client, ADM1266_READ_BLACKBOX, 1,
> +				       &index, buf);
> +		if (ret < 0)
> +			return ret;
> +
> +		buf += ADM1266_BLACKBOX_SIZE;
> +	}
> +
> +	return 0;
> +}
> +
> +static bool adm1266_cell_is_accessed(const struct nvmem_cell_info *mem_cell,
> +				     unsigned int offset, size_t bytes)
> +{
> +	unsigned int start_addr = offset;
> +	unsigned int end_addr = offset + bytes;
> +	unsigned int cell_start = mem_cell->offset;
> +	unsigned int cell_end = mem_cell->offset + mem_cell->bytes;
> +
> +	if (start_addr <= cell_end && cell_start <= end_addr)
> +		return true;
> +
> +	return false;

	return start_addr <= cell_end && cell_start <= end_add;

does the same.

> +}
> +
> +static int adm1266_read_mem_cell(struct adm1266_data *data,
> +				 const struct nvmem_cell_info *mem_cell)
> +{
> +	u8 *mem_offset;
> +	int ret;
> +
> +	switch (mem_cell->offset) {
> +	case ADM1266_BLACKBOX_OFFSET:

How would this ever have a different value ?

> +		mem_offset = data->dev_mem + mem_cell->offset;
> +		ret = adm1266_nvmem_read_blackbox(data, mem_offset);
> +		if (ret)
> +			dev_err(&data->client->dev, "Could not read blackbox!");
> +		return ret;
> +	default:
> +		return -EINVAL;
> +	}
> +}
> +
> +static int adm1266_nvmem_read(void *priv, unsigned int offset, void *val,
> +			      size_t bytes)
> +{
> +	const struct nvmem_cell_info *mem_cell;
> +	struct adm1266_data *data = priv;
> +	int ret;
> +	int i;
> +
> +	for (i = 0; i < data->nvmem_config.ncells; i++) {
> +		mem_cell = &adm1266_nvmem_cells[i];
> +		if (!adm1266_cell_is_accessed(mem_cell, offset, bytes))
> +			continue;
> +
> +		ret = adm1266_read_mem_cell(data, mem_cell);
> +		if (ret < 0)
> +			return ret;
> +	}

I am a bit puzzled about the complexity of this code.
Is there reason to believe that there will ever be more than one cell ?
If that is not the case, I don't see the value in the complexity.
If there is a plan to add more nvram cells later, it should be
mentioned somewhere. As it is, the code is difficult to understand,
and I really only want to spend time analyzing it if it is really
necessary.

> +
> +	memcpy(val, data->dev_mem + offset, bytes);
> +
> +	return 0;
> +}
> +
> +static int adm1266_config_nvmem(struct adm1266_data *data)
> +{
> +	data->nvmem_config.name = dev_name(&data->client->dev);
> +	data->nvmem_config.dev = &data->client->dev;
> +	data->nvmem_config.root_only = true;
> +	data->nvmem_config.read_only = true;
> +	data->nvmem_config.owner = THIS_MODULE;
> +	data->nvmem_config.reg_read = adm1266_nvmem_read;
> +	data->nvmem_config.cells = adm1266_nvmem_cells;
> +	data->nvmem_config.ncells = ARRAY_SIZE(adm1266_nvmem_cells);
> +	data->nvmem_config.priv = data;
> +	data->nvmem_config.stride = 1;
> +	data->nvmem_config.word_size = 1;
> +	data->nvmem_config.size = 0x80000;
> +
> +	data->nvmem = nvmem_register(&data->nvmem_config);

If CONFIG_NVMEM is not enabled, this function will return -EOPNOTSUPP,
and the driver will fail to load. I don't think that is acceptable.

Also, this should really use devm_nvmem_register().

> +	if (IS_ERR(data->nvmem)) {
> +		dev_err(&data->client->dev, "Could not register nvmem!");
> +		return PTR_ERR(data->nvmem);
> +	}
> +
> +	data->dev_mem = devm_kzalloc(&data->client->dev,
> +				     data->nvmem_config.size,
> +				     GFP_KERNEL);

This is at least potentially racy. Presumably nvram can be accessed right
after the call to nvmem_register(). If that happens, dev_mem is not yet
set, and the system will crash.

> +	if (!data->dev_mem)
> +		return -ENOMEM;
> +
> +	return 0;
> +}
> +
>  static int adm1266_probe(struct i2c_client *client,
>  			 const struct i2c_device_id *id)
>  {
> @@ -430,6 +560,10 @@ static int adm1266_probe(struct i2c_client *client,
>  	if (ret < 0)
>  		return ret;
>  
> +	ret = adm1266_config_nvmem(data);
> +	if (ret < 0)
> +		return ret;
> +
>  	info = &data->info;
>  	info->pages = 17;
>  	info->format[PSC_VOLTAGE_OUT] = linear;