On Tue, Mar 24, 2015 at 10:30:08PM +0000, Srinivas Kandagatla wrote:
> +static ssize_t bin_attr_eeprom_write(struct file *filp, struct kobject *kobj,
> + struct bin_attribute *attr,
> + char *buf, loff_t offset, size_t count)
> +{
> + struct device *dev = container_of(kobj, struct device, kobj);
> + struct eeprom_device *eeprom = to_eeprom(dev);
> + int rc;
> +
> + if (offset > eeprom->size)
> + return -EINVAL;
> +
> + if (offset + count > eeprom->size)
> + count = eeprom->size - offset;
> +
> + rc = regmap_bulk_write(eeprom->regmap, offset,
> + buf, count/eeprom->stride);
Are you sure that this and the read interface should be using the bulk
interface and not the raw interface - do we want the byte swapping that
the bulk interface provides?
I'm also not entirely able to convince myself that the above error
checks and code line up with what I'd expect the userspace ABI to be, we
seem to be treating offset as both a byte offset into the data (which is
what I'd expect the userspace ABI to do) and a word based index into the
data (which is what the regmap API is doing). For example with 16 bit
words offset 2 will start at the 5th byte of data but if userspace seeks
to offset 5 it will get the 11th byte and onwards.
The stride and the word size are separate, they will frequently line up
for memory mapped devices but typically won't for other devices. I
think you need more data mangling to handle this robustly.
Attachment:
signature.asc
Description: Digital signature
