| Message ID | 20240404-dev-add_dev_errp_probe-v1-3-d18e3eb7ec3f@analog.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | dev_printk: add dev_errp_probe() helper | expand |
On Thu, Apr 04, 2024 at 01:06:25PM +0200, Nuno Sa wrote: > Using dev_errp_probe() to simplify the code. ... > + if (IS_ERR(fwnode)) > + return dev_errp_probe(dev, PTR_ERR(fwnode), > + "Cannot get Firmware reference\n"); ERR_CAST() seems quite good candidate to have here. return dev_errp_probe(dev, fwnode, "Cannot get Firmware reference\n"); (Assuming dev_errp_probe() magically understands that, note you may have it as a macro and distinguish parameter type with _Generic() or so and behave differently: ERR_PTR() vs. ERR_CAST(), see acpi_dev_hid_uid_match() implementation, but also keep in mind that it doesn't distinguish NULL/0, there is a patch available in the mailing list to fix that, though.)
On Thu, 2024-04-04 at 15:23 +0300, Andy Shevchenko wrote: > On Thu, Apr 04, 2024 at 01:06:25PM +0200, Nuno Sa wrote: > > Using dev_errp_probe() to simplify the code. > > ... > > > + if (IS_ERR(fwnode)) > > + return dev_errp_probe(dev, PTR_ERR(fwnode), > > + "Cannot get Firmware reference\n"); > > ERR_CAST() seems quite good candidate to have here. > > return dev_errp_probe(dev, fwnode, "Cannot get Firmware > reference\n"); > > (Assuming dev_errp_probe() magically understands that, note you may have it as > a macro and distinguish parameter type with _Generic() or so and behave > differently: ERR_PTR() vs. ERR_CAST(), see acpi_dev_hid_uid_match() > implementation, but also keep in mind that it doesn't distinguish NULL/0, > there > is a patch available in the mailing list to fix that, though.) > Do we care that much for going with that trouble? I understand like this we go PTR_ERR() to then comeback to ERR_PTR() but this for probe() which is not a fastpath. So perhaps we could just keep it simple? - Nuno Sá
On Thu, Apr 04, 2024 at 04:58:27PM +0200, Nuno Sá wrote: > On Thu, 2024-04-04 at 15:23 +0300, Andy Shevchenko wrote: > > On Thu, Apr 04, 2024 at 01:06:25PM +0200, Nuno Sa wrote: > > > Using dev_errp_probe() to simplify the code. ... > > > + if (IS_ERR(fwnode)) > > > + return dev_errp_probe(dev, PTR_ERR(fwnode), > > > + "Cannot get Firmware reference\n"); > > > > ERR_CAST() seems quite good candidate to have here. > > > > return dev_errp_probe(dev, fwnode, "Cannot get Firmware > > reference\n"); > > > > (Assuming dev_errp_probe() magically understands that, note you may have it as > > a macro and distinguish parameter type with _Generic() or so and behave > > differently: ERR_PTR() vs. ERR_CAST(), see acpi_dev_hid_uid_match() > > implementation, but also keep in mind that it doesn't distinguish NULL/0, > > there > > is a patch available in the mailing list to fix that, though.) > > Do we care that much for going with that trouble? I don't think we do. We are not supposed to be called with ret == 0/NULL. That's why I pointed out to the current version. > I understand like this we go > PTR_ERR() to then comeback to ERR_PTR() but this for probe() which is not a > fastpath. So perhaps we could just keep it simple? It's not about performance, it's about readability. See the difference between yours and mine.
On Thu, 4 Apr 2024 18:12:25 +0300 Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote: > On Thu, Apr 04, 2024 at 04:58:27PM +0200, Nuno Sá wrote: > > On Thu, 2024-04-04 at 15:23 +0300, Andy Shevchenko wrote: > > > On Thu, Apr 04, 2024 at 01:06:25PM +0200, Nuno Sa wrote: > > > > Using dev_errp_probe() to simplify the code. > > ... > > > > > + if (IS_ERR(fwnode)) > > > > + return dev_errp_probe(dev, PTR_ERR(fwnode), > > > > + "Cannot get Firmware reference\n"); > > > > > > ERR_CAST() seems quite good candidate to have here. > > > > > > return dev_errp_probe(dev, fwnode, "Cannot get Firmware > > > reference\n"); > > > > > > (Assuming dev_errp_probe() magically understands that, note you may have it as > > > a macro and distinguish parameter type with _Generic() or so and behave > > > differently: ERR_PTR() vs. ERR_CAST(), see acpi_dev_hid_uid_match() > > > implementation, but also keep in mind that it doesn't distinguish NULL/0, > > > there > > > is a patch available in the mailing list to fix that, though.) > > > > Do we care that much for going with that trouble? > > I don't think we do. We are not supposed to be called with ret == 0/NULL. > That's why I pointed out to the current version. > > > I understand like this we go > > PTR_ERR() to then comeback to ERR_PTR() but this for probe() which is not a > > fastpath. So perhaps we could just keep it simple? > > It's not about performance, it's about readability. See the difference between > yours and mine. > You are suggesting making it transparently take an error ptr or an integer? Whilst clever, I'm not seeing that as a good idea for readability / reviewability. I expect something that looks like a function to take the same parameters (other vargs) always. _Generic messes with that. Maybe I just don't like to learn new things! If consensus comes down in favour of _Generic trickery then I'll get used to it eventually. Jonathan
Hi, On Sat, Apr 06, 2024 at 05:07:17PM +0100, Jonathan Cameron wrote: > On Thu, 4 Apr 2024 18:12:25 +0300 > Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote: > > > On Thu, Apr 04, 2024 at 04:58:27PM +0200, Nuno Sá wrote: > > > On Thu, 2024-04-04 at 15:23 +0300, Andy Shevchenko wrote: > > > > On Thu, Apr 04, 2024 at 01:06:25PM +0200, Nuno Sa wrote: > > > > > Using dev_errp_probe() to simplify the code. > > > > ... > > > > > > > + if (IS_ERR(fwnode)) > > > > > + return dev_errp_probe(dev, PTR_ERR(fwnode), > > > > > + "Cannot get Firmware reference\n"); > > > > > > > > ERR_CAST() seems quite good candidate to have here. > > > > > > > > return dev_errp_probe(dev, fwnode, "Cannot get Firmware > > > > reference\n"); > > > > > > > > (Assuming dev_errp_probe() magically understands that, note you may have it as > > > > a macro and distinguish parameter type with _Generic() or so and behave > > > > differently: ERR_PTR() vs. ERR_CAST(), see acpi_dev_hid_uid_match() > > > > implementation, but also keep in mind that it doesn't distinguish NULL/0, > > > > there > > > > is a patch available in the mailing list to fix that, though.) > > > > > > Do we care that much for going with that trouble? > > > > I don't think we do. We are not supposed to be called with ret == 0/NULL. > > That's why I pointed out to the current version. > > > > > I understand like this we go > > > PTR_ERR() to then comeback to ERR_PTR() but this for probe() which is not a > > > fastpath. So perhaps we could just keep it simple? > > > > It's not about performance, it's about readability. See the difference between > > yours and mine. > > > > You are suggesting making it transparently take an error ptr or an integer? > Whilst clever, I'm not seeing that as a good idea for readability / reviewability. > I expect something that looks like a function to take the same parameters (other vargs) > always. _Generic messes with that. > > Maybe I just don't like to learn new things! If consensus comes down in favour > of _Generic trickery then I'll get used to it eventually. the whole point of the dev_err_...() functions is to add trickery in order to reduce code and brackets. The way I see this is to have a combination of functions: - takes integer, returns integer -> dev_err_probe() - takes integer, returns pointer -> dev_errp_probe() (or dev_err_ptr_probe()) - takes pointer, return integer -> ? dev_ptr_err_probe() - takes pointer, returns pointer -> ? dev_ptr_probe() Thoughts? Andi
On Sat, 2024-04-06 at 17:07 +0100, Jonathan Cameron wrote: > On Thu, 4 Apr 2024 18:12:25 +0300 > Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote: > > > On Thu, Apr 04, 2024 at 04:58:27PM +0200, Nuno Sá wrote: > > > On Thu, 2024-04-04 at 15:23 +0300, Andy Shevchenko wrote: > > > > On Thu, Apr 04, 2024 at 01:06:25PM +0200, Nuno Sa wrote: > > > > > Using dev_errp_probe() to simplify the code. > > > > ... > > > > > > > + if (IS_ERR(fwnode)) > > > > > + return dev_errp_probe(dev, PTR_ERR(fwnode), > > > > > + "Cannot get Firmware > > > > > reference\n"); > > > > > > > > ERR_CAST() seems quite good candidate to have here. > > > > > > > > return dev_errp_probe(dev, fwnode, "Cannot get Firmware > > > > reference\n"); > > > > > > > > (Assuming dev_errp_probe() magically understands that, note you may have > > > > it as > > > > a macro and distinguish parameter type with _Generic() or so and behave > > > > differently: ERR_PTR() vs. ERR_CAST(), see acpi_dev_hid_uid_match() > > > > implementation, but also keep in mind that it doesn't distinguish > > > > NULL/0, > > > > there > > > > is a patch available in the mailing list to fix that, though.) > > > > > > Do we care that much for going with that trouble? > > > > I don't think we do. We are not supposed to be called with ret == 0/NULL. > > That's why I pointed out to the current version. > > > > > I understand like this we go > > > PTR_ERR() to then comeback to ERR_PTR() but this for probe() which is not > > > a > > > fastpath. So perhaps we could just keep it simple? > > > > It's not about performance, it's about readability. See the difference > > between > > yours and mine. > > > > You are suggesting making it transparently take an error ptr or an integer? > Whilst clever, I'm not seeing that as a good idea for readability / > reviewability. > I expect something that looks like a function to take the same parameters > (other vargs) > always. _Generic messes with that. > Maybe I just don't like to learn new things! If consensus comes down in > favour > of _Generic trickery then I'll get used to it eventually. > Yeah, I agree with the above. Not fully convinced but for the ERR_CAST() case I would very much prefer to have another explicit helper rather than hiding stuff in the same macro. - Nuno Sá
On Sat, 2024-04-06 at 20:54 +0200, Andi Shyti wrote: > Hi, > > On Sat, Apr 06, 2024 at 05:07:17PM +0100, Jonathan Cameron wrote: > > On Thu, 4 Apr 2024 18:12:25 +0300 > > Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote: > > > > > On Thu, Apr 04, 2024 at 04:58:27PM +0200, Nuno Sá wrote: > > > > On Thu, 2024-04-04 at 15:23 +0300, Andy Shevchenko wrote: > > > > > On Thu, Apr 04, 2024 at 01:06:25PM +0200, Nuno Sa wrote: > > > > > > Using dev_errp_probe() to simplify the code. > > > > > > ... > > > > > > > > > + if (IS_ERR(fwnode)) > > > > > > + return dev_errp_probe(dev, PTR_ERR(fwnode), > > > > > > + "Cannot get Firmware > > > > > > reference\n"); > > > > > > > > > > ERR_CAST() seems quite good candidate to have here. > > > > > > > > > > return dev_errp_probe(dev, fwnode, "Cannot get > > > > > Firmware > > > > > reference\n"); > > > > > > > > > > (Assuming dev_errp_probe() magically understands that, note you may > > > > > have it as > > > > > a macro and distinguish parameter type with _Generic() or so and > > > > > behave > > > > > differently: ERR_PTR() vs. ERR_CAST(), see acpi_dev_hid_uid_match() > > > > > implementation, but also keep in mind that it doesn't distinguish > > > > > NULL/0, > > > > > there > > > > > is a patch available in the mailing list to fix that, though.) > > > > > > > > Do we care that much for going with that trouble? > > > > > > I don't think we do. We are not supposed to be called with ret == 0/NULL. > > > That's why I pointed out to the current version. > > > > > > > I understand like this we go > > > > PTR_ERR() to then comeback to ERR_PTR() but this for probe() which is > > > > not a > > > > fastpath. So perhaps we could just keep it simple? > > > > > > It's not about performance, it's about readability. See the difference > > > between > > > yours and mine. > > > > > > > You are suggesting making it transparently take an error ptr or an integer? > > Whilst clever, I'm not seeing that as a good idea for readability / > > reviewability. > > I expect something that looks like a function to take the same parameters > > (other vargs) > > always. _Generic messes with that. > > > > Maybe I just don't like to learn new things! If consensus comes down in > > favour > > of _Generic trickery then I'll get used to it eventually. > > the whole point of the dev_err_...() functions is to add trickery > in order to reduce code and brackets. > I'm not sure I'm completely convinced on having more helpers but also no strong opinion tbh. But see below... > The way I see this is to have a combination of functions: > > - takes integer, returns integer -> dev_err_probe() > - takes integer, returns pointer -> dev_errp_probe() (or dev_err_ptr_probe()) > - takes pointer, return integer -> ? dev_ptr_err_probe() This is pretty much all the dev_err_probe(dev, PTR_ERR(), ...) we already have out there. Do we really want to have this variant? > - takes pointer, returns pointer -> ? dev_ptr_probe() dev_ptr_probe() misses to be clear about being an error and think this is pretty much the ERR_CAST() case right? Maybe dev_err_cast_ptr_probe()? Or dev_err_cast_probe()? - Nuno Sá
diff --git a/drivers/iio/industrialio-backend.c b/drivers/iio/industrialio-backend.c index 2fea2bbbe47f..e0b08283d667 100644 --- a/drivers/iio/industrialio-backend.c +++ b/drivers/iio/industrialio-backend.c @@ -296,11 +296,9 @@ struct iio_backend *devm_iio_backend_get(struct device *dev, const char *name) } fwnode = fwnode_find_reference(dev_fwnode(dev), "io-backends", index); - if (IS_ERR(fwnode)) { - dev_err_probe(dev, PTR_ERR(fwnode), - "Cannot get Firmware reference\n"); - return ERR_CAST(fwnode); - } + if (IS_ERR(fwnode)) + return dev_errp_probe(dev, PTR_ERR(fwnode), + "Cannot get Firmware reference\n"); guard(mutex)(&iio_back_lock); list_for_each_entry(back, &iio_back_list, entry) {
Using dev_errp_probe() to simplify the code. Signed-off-by: Nuno Sa <nuno.sa@analog.com> --- drivers/iio/industrialio-backend.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-)