On 10/09/2020 16:35, Bartosz Golaszewski wrote:
> On Thu, Sep 10, 2020 at 3:43 PM Jon Hunter <[email protected]> wrote:
>>
>> The AT24 EEPROM driver does not initialise the 'id' field of the
>> nvmem_config structure and because the entire structure is not
>> initialised, it ends up with a random value. This causes the NVMEM
>> driver to append the device 'devid' value to name of the NVMEM
>> device. Although this is not a problem per-se, for I2C devices such as
>> the AT24, that already have a device unique name, there does not seem
>> much value in appending an additional 0 to the I2C name. For example,
>> appending a 0 to an I2C device name such as 1-0050 does not seem
>> necessary and maybe even a bit confusing. Therefore, fix this by
>> setting the NVMEM config.id to NVMEM_DEVID_NONE for AT24 EEPROMs.
>>
>> Signed-off-by: Jon Hunter <[email protected]>
>> ---
>> drivers/misc/eeprom/at24.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/misc/eeprom/at24.c b/drivers/misc/eeprom/at24.c
>> index e9df1ca251df..3f7a3bb6a36c 100644
>> --- a/drivers/misc/eeprom/at24.c
>> +++ b/drivers/misc/eeprom/at24.c
>> @@ -715,6 +715,7 @@ static int at24_probe(struct i2c_client *client)
>>
>> nvmem_config.name = dev_name(dev);
>> nvmem_config.dev = dev;
>> + nvmem_config.id = NVMEM_DEVID_NONE;
>> nvmem_config.read_only = !writable;
>> nvmem_config.root_only = !(flags & AT24_FLAG_IRUGO);
>> nvmem_config.owner = THIS_MODULE;
>> --
>> 2.25.1
>>
>
> This patch is correct and thanks for catching it. I vaguely recall
> wondering at some point why the appended 0 in the nvmem name for at24.
> Unfortunately this change would affect how the device is visible in
> user-space in /sys/bus/nvmem/devices/ and this could break existing
> users. Also: there are many in-kernel users that would need to be
> updated. I'm afraid we'll need some sort of backward compatibility.
Thanks, yes that is a problem. I guess for now we could explicitly init
to NVMEM_DEVID_AUTO or maybe just 0 so that it defaults to the same path
in the NVMEM driver. However, I am not sure how we can make allow some
devices to use NVMEM_DEVID_NONE and others use something else. This is
not really something that we can describe in DT because it has nothing
to do with h/w.
Cheers
Jon
--
nvpublic
On 10/09/2020 19:15, Jon Hunter wrote:
>
> On 10/09/2020 16:35, Bartosz Golaszewski wrote:
>> On Thu, Sep 10, 2020 at 3:43 PM Jon Hunter <[email protected]> wrote:
>>>
>>> The AT24 EEPROM driver does not initialise the 'id' field of the
>>> nvmem_config structure and because the entire structure is not
>>> initialised, it ends up with a random value. This causes the NVMEM
>>> driver to append the device 'devid' value to name of the NVMEM
>>> device. Although this is not a problem per-se, for I2C devices such as
>>> the AT24, that already have a device unique name, there does not seem
>>> much value in appending an additional 0 to the I2C name. For example,
>>> appending a 0 to an I2C device name such as 1-0050 does not seem
>>> necessary and maybe even a bit confusing. Therefore, fix this by
>>> setting the NVMEM config.id to NVMEM_DEVID_NONE for AT24 EEPROMs.
>>>
>>> Signed-off-by: Jon Hunter <[email protected]>
>>> ---
>>> drivers/misc/eeprom/at24.c | 1 +
>>> 1 file changed, 1 insertion(+)
>>>
>>> diff --git a/drivers/misc/eeprom/at24.c b/drivers/misc/eeprom/at24.c
>>> index e9df1ca251df..3f7a3bb6a36c 100644
>>> --- a/drivers/misc/eeprom/at24.c
>>> +++ b/drivers/misc/eeprom/at24.c
>>> @@ -715,6 +715,7 @@ static int at24_probe(struct i2c_client *client)
>>>
>>> nvmem_config.name = dev_name(dev);
>>> nvmem_config.dev = dev;
>>> + nvmem_config.id = NVMEM_DEVID_NONE;
>>> nvmem_config.read_only = !writable;
>>> nvmem_config.root_only = !(flags & AT24_FLAG_IRUGO);
>>> nvmem_config.owner = THIS_MODULE;
>>> --
>>> 2.25.1
>>>
>>
>> This patch is correct and thanks for catching it. I vaguely recall
>> wondering at some point why the appended 0 in the nvmem name for at24.
>> Unfortunately this change would affect how the device is visible in
>> user-space in /sys/bus/nvmem/devices/ and this could break existing
>> users. Also: there are many in-kernel users that would need to be
>> updated. I'm afraid we'll need some sort of backward compatibility.
>
>
> Thanks, yes that is a problem. I guess for now we could explicitly init
> to NVMEM_DEVID_AUTO or maybe just 0 so that it defaults to the same path
> in the NVMEM driver. However, I am not sure how we can make allow some
> devices to use NVMEM_DEVID_NONE and others use something else. This is
> not really something that we can describe in DT because it has nothing
> to do with h/w.
Unless we make the configuration of the 'id' dependent on the 'label'
property so something like ...
if (device_property_present(dev, "label")) {
nvmem_config.id = NVMEM_DEVID_NONE;
err = device_property_read_string(dev, "label",
&nvmem_config.name);
if (err)
return err;
} else {
nvmem_config.id = NVMEM_DEVID_AUTO;
nvmem_config.name = dev_name(dev);
}
Cheers
Jon
--
nvpublic
On Thu, Sep 10, 2020 at 8:19 PM Jon Hunter <[email protected]> wrote:
>
>
> On 10/09/2020 19:15, Jon Hunter wrote:
> >
> > On 10/09/2020 16:35, Bartosz Golaszewski wrote:
> >> On Thu, Sep 10, 2020 at 3:43 PM Jon Hunter <[email protected]> wrote:
> >>>
> >>> The AT24 EEPROM driver does not initialise the 'id' field of the
> >>> nvmem_config structure and because the entire structure is not
> >>> initialised, it ends up with a random value. This causes the NVMEM
> >>> driver to append the device 'devid' value to name of the NVMEM
> >>> device. Although this is not a problem per-se, for I2C devices such as
> >>> the AT24, that already have a device unique name, there does not seem
> >>> much value in appending an additional 0 to the I2C name. For example,
> >>> appending a 0 to an I2C device name such as 1-0050 does not seem
> >>> necessary and maybe even a bit confusing. Therefore, fix this by
> >>> setting the NVMEM config.id to NVMEM_DEVID_NONE for AT24 EEPROMs.
> >>>
> >>> Signed-off-by: Jon Hunter <[email protected]>
> >>> ---
> >>> drivers/misc/eeprom/at24.c | 1 +
> >>> 1 file changed, 1 insertion(+)
> >>>
> >>> diff --git a/drivers/misc/eeprom/at24.c b/drivers/misc/eeprom/at24.c
> >>> index e9df1ca251df..3f7a3bb6a36c 100644
> >>> --- a/drivers/misc/eeprom/at24.c
> >>> +++ b/drivers/misc/eeprom/at24.c
> >>> @@ -715,6 +715,7 @@ static int at24_probe(struct i2c_client *client)
> >>>
> >>> nvmem_config.name = dev_name(dev);
> >>> nvmem_config.dev = dev;
> >>> + nvmem_config.id = NVMEM_DEVID_NONE;
> >>> nvmem_config.read_only = !writable;
> >>> nvmem_config.root_only = !(flags & AT24_FLAG_IRUGO);
> >>> nvmem_config.owner = THIS_MODULE;
> >>> --
> >>> 2.25.1
> >>>
> >>
> >> This patch is correct and thanks for catching it. I vaguely recall
> >> wondering at some point why the appended 0 in the nvmem name for at24.
> >> Unfortunately this change would affect how the device is visible in
> >> user-space in /sys/bus/nvmem/devices/ and this could break existing
> >> users. Also: there are many in-kernel users that would need to be
> >> updated. I'm afraid we'll need some sort of backward compatibility.
> >
> >
> > Thanks, yes that is a problem. I guess for now we could explicitly init
> > to NVMEM_DEVID_AUTO or maybe just 0 so that it defaults to the same path
> > in the NVMEM driver. However, I am not sure how we can make allow some
> > devices to use NVMEM_DEVID_NONE and others use something else. This is
> > not really something that we can describe in DT because it has nothing
> > to do with h/w.
>
>
> Unless we make the configuration of the 'id' dependent on the 'label'
> property so something like ...
>
> if (device_property_present(dev, "label")) {
> nvmem_config.id = NVMEM_DEVID_NONE;
> err = device_property_read_string(dev, "label",
> &nvmem_config.name);
> if (err)
> return err;
> } else {
> nvmem_config.id = NVMEM_DEVID_AUTO;
> nvmem_config.name = dev_name(dev);
> }
>
> Cheers
> Jon
>
> --
> nvpublic
Yes, this looks like the best compromise we can get for now. Please
make sure to document why we do this in the code.
Bartosz