| Message ID | 20240812-chipcap2-probe-improvements-v1-2-3cdff6d16897@gmail.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | hwmon: chipcap2: small improvements in probe function | expand |
On 8/12/24 08:43, Javier Carrasco wrote: > This check is carried out after getting the regulator, and the device > can be disabled if an error occurs. > I do not see a possible path for a call to cc2_enable() at this point, meaning the regulator won't ever be enabled. Please provide a better explanation why this patch would be necessary. Guenter > Signed-off-by: Javier Carrasco <javier.carrasco.cruz@gmail.com> > --- > drivers/hwmon/chipcap2.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/hwmon/chipcap2.c b/drivers/hwmon/chipcap2.c > index 88689f4eb598..02764689ed21 100644 > --- a/drivers/hwmon/chipcap2.c > +++ b/drivers/hwmon/chipcap2.c > @@ -747,7 +747,7 @@ static int cc2_probe(struct i2c_client *client) > ret = cc2_request_ready_irq(data, dev); > if (ret) { > dev_err_probe(dev, ret, "Failed to request ready irq\n"); > - return ret; > + goto disable; > } > > ret = cc2_request_alarm_irqs(data, dev); >
On 12/08/2024 18:49, Guenter Roeck wrote: > On 8/12/24 08:43, Javier Carrasco wrote: >> This check is carried out after getting the regulator, and the device >> can be disabled if an error occurs. >> > > I do not see a possible path for a call to cc2_enable() at this point, > meaning the regulator won't ever be enabled. Please provide a better > explanation why this patch would be necessary. > > Guenter > Hi Guenter, this patch enforces the state where the dedicated regulator is disabled, no matter what the history of the regulator was. If a previous regulator_disable() failed, it would still be desirable that the regulator gets disabled the next time the driver is probed (i.e. a new attempt to disable it on failure). cc2_disable() checks first if the regulator is enabled to avoid any imbalance. Best regards, Javier Carrasco
On 8/12/24 12:59, Javier Carrasco wrote: > On 12/08/2024 18:49, Guenter Roeck wrote: >> On 8/12/24 08:43, Javier Carrasco wrote: >>> This check is carried out after getting the regulator, and the device >>> can be disabled if an error occurs. >>> >> >> I do not see a possible path for a call to cc2_enable() at this point, >> meaning the regulator won't ever be enabled. Please provide a better >> explanation why this patch would be necessary. >> >> Guenter >> > > Hi Guenter, > > this patch enforces the state where the dedicated regulator is disabled, > no matter what the history of the regulator was. If a previous > regulator_disable() failed, it would still be desirable that the > regulator gets disabled the next time the driver is probed (i.e. a new > attempt to disable it on failure). > cc2_disable() checks first if the regulator is enabled to avoid any > imbalance. > That is very theoretic. Sorry, I am not going to accept this patch. Guenter
On 12/08/2024 22:08, Guenter Roeck wrote: > On 8/12/24 12:59, Javier Carrasco wrote: >> On 12/08/2024 18:49, Guenter Roeck wrote: >>> On 8/12/24 08:43, Javier Carrasco wrote: >>>> This check is carried out after getting the regulator, and the device >>>> can be disabled if an error occurs. >>>> >>> >>> I do not see a possible path for a call to cc2_enable() at this point, >>> meaning the regulator won't ever be enabled. Please provide a better >>> explanation why this patch would be necessary. >>> >>> Guenter >>> >> >> Hi Guenter, >> >> this patch enforces the state where the dedicated regulator is disabled, >> no matter what the history of the regulator was. If a previous >> regulator_disable() failed, it would still be desirable that the >> regulator gets disabled the next time the driver is probed (i.e. a new >> attempt to disable it on failure). >> cc2_disable() checks first if the regulator is enabled to avoid any >> imbalance. >> > > That is very theoretic. Sorry, I am not going to accept this patch. > > Guenter > I get your point, but given that this device requires a dedicated regulator, I believe it makes sense that it tries to disable it whenever possible if it's not going to be used. I think that makes more sense that just returning an error value without even making sure that de regulator was disabled, doesn't it? Of course this is not a killer feature, and I don't want to make you waste much time with it. But I think the dedicated regulator should be shut down in all error paths, whatever status it had before. If that does not sound convincing, then I won't argue any longer. Please take a look at the first patch of the series in any case, which is not a killer feature either, but cleaner than the current implementation. Thanks and best regards, Javier Carrasco
On 8/12/24 13:48, Javier Carrasco wrote: > On 12/08/2024 22:08, Guenter Roeck wrote: >> On 8/12/24 12:59, Javier Carrasco wrote: >>> On 12/08/2024 18:49, Guenter Roeck wrote: >>>> On 8/12/24 08:43, Javier Carrasco wrote: >>>>> This check is carried out after getting the regulator, and the device >>>>> can be disabled if an error occurs. >>>>> >>>> >>>> I do not see a possible path for a call to cc2_enable() at this point, >>>> meaning the regulator won't ever be enabled. Please provide a better >>>> explanation why this patch would be necessary. >>>> >>>> Guenter >>>> >>> >>> Hi Guenter, >>> >>> this patch enforces the state where the dedicated regulator is disabled, >>> no matter what the history of the regulator was. If a previous >>> regulator_disable() failed, it would still be desirable that the >>> regulator gets disabled the next time the driver is probed (i.e. a new >>> attempt to disable it on failure). >>> cc2_disable() checks first if the regulator is enabled to avoid any >>> imbalance. >>> >> >> That is very theoretic. Sorry, I am not going to accept this patch. >> >> Guenter >> > > I get your point, but given that this device requires a dedicated > regulator, I believe it makes sense that it tries to disable it whenever > possible if it's not going to be used. I think that makes more sense > that just returning an error value without even making sure that de > regulator was disabled, doesn't it? > No, it doesn't make any sense whatsoever. What are you planning to do, clutter the kernel with code to disable regulators if instantiating a device fails for whatever reason and it turns out that a regulator which should not have been enabled to start with turns out to be enabled anyway ? > Of course this is not a killer feature, and I don't want to make you > waste much time with it. But I think the dedicated regulator should be > shut down in all error paths, whatever status it had before. > I strongly disagree. This can only mess up the kernel all over the place. Maybe you can convince other maintainers to accept such code, but please refrain from doing that in my scope of responsibility. If the regulator subsystem has the habit of leaving regulators enabled even after they have been released, that problem should be fixed in the regulator subsystem and not be worked around in individual drivers. Guenter
On 12/08/2024 23:26, Guenter Roeck wrote: > On 8/12/24 13:48, Javier Carrasco wrote: >> On 12/08/2024 22:08, Guenter Roeck wrote: >>> On 8/12/24 12:59, Javier Carrasco wrote: >>>> On 12/08/2024 18:49, Guenter Roeck wrote: >>>>> On 8/12/24 08:43, Javier Carrasco wrote: >>>>>> This check is carried out after getting the regulator, and the device >>>>>> can be disabled if an error occurs. >>>>>> >>>>> >>>>> I do not see a possible path for a call to cc2_enable() at this point, >>>>> meaning the regulator won't ever be enabled. Please provide a better >>>>> explanation why this patch would be necessary. >>>>> >>>>> Guenter >>>>> >>>> >>>> Hi Guenter, >>>> >>>> this patch enforces the state where the dedicated regulator is >>>> disabled, >>>> no matter what the history of the regulator was. If a previous >>>> regulator_disable() failed, it would still be desirable that the >>>> regulator gets disabled the next time the driver is probed (i.e. a new >>>> attempt to disable it on failure). >>>> cc2_disable() checks first if the regulator is enabled to avoid any >>>> imbalance. >>>> >>> >>> That is very theoretic. Sorry, I am not going to accept this patch. >>> >>> Guenter >>> >> >> I get your point, but given that this device requires a dedicated >> regulator, I believe it makes sense that it tries to disable it whenever >> possible if it's not going to be used. I think that makes more sense >> that just returning an error value without even making sure that de >> regulator was disabled, doesn't it? >> > > No, it doesn't make any sense whatsoever. What are you planning to do, > clutter the kernel with code to disable regulators if instantiating a > device > fails for whatever reason and it turns out that a regulator which should > not have been enabled to start with turns out to be enabled anyway ? > >> Of course this is not a killer feature, and I don't want to make you >> waste much time with it. But I think the dedicated regulator should be >> shut down in all error paths, whatever status it had before. >> > > I strongly disagree. This can only mess up the kernel all over the place. > Maybe you can convince other maintainers to accept such code, but please > refrain from doing that in my scope of responsibility. If the regulator > subsystem has the habit of leaving regulators enabled even after they > have been released, that problem should be fixed in the regulator subsystem > and not be worked around in individual drivers. > > Guenter > In that case the current behavior is wrong in the opposite direction, and disabling the regulator before any call to cc2_enable() follows the same invalid assumption. I will remove the call to cc2_disable() in the probe function, which was there exactly to ensure that the regulator gets disabled under all circumstances. All error paths will just return, letting the regulator untouched if it was not actively enabled. Best regards, Javier Carrasco
diff --git a/drivers/hwmon/chipcap2.c b/drivers/hwmon/chipcap2.c index 88689f4eb598..02764689ed21 100644 --- a/drivers/hwmon/chipcap2.c +++ b/drivers/hwmon/chipcap2.c @@ -747,7 +747,7 @@ static int cc2_probe(struct i2c_client *client) ret = cc2_request_ready_irq(data, dev); if (ret) { dev_err_probe(dev, ret, "Failed to request ready irq\n"); - return ret; + goto disable; } ret = cc2_request_alarm_irqs(data, dev);
This check is carried out after getting the regulator, and the device can be disabled if an error occurs. Signed-off-by: Javier Carrasco <javier.carrasco.cruz@gmail.com> --- drivers/hwmon/chipcap2.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)