| Message ID | 20190626153611.10170-1-ckeepax@opensource.cirrus.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [1/2] device property: Add new array helper | expand |
On Wed, Jun 26, 2019 at 04:36:10PM +0100, Charles Keepax wrote: > It is fairly common to want to read an integer array property > that is composed of an unknown number of fixed size integer > groups. For example, say each group consists of three values > which correspond to the settings for one input on the device > and the driver supports several chips with different numbers > of inputs. > > Add a new helper function to provide this functionality, it > differs for the existing helpers in that it allows reading a > smaller number of values than the full array size and checks > that the number of values read is a multiple of the group size. > > Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com> > --- > drivers/base/property.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++ > include/linux/property.h | 2 ++ > 2 files changed, 50 insertions(+) Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
On Thu, Jun 27, 2019 at 2:15 AM Greg KH <gregkh@linuxfoundation.org> wrote: > > On Wed, Jun 26, 2019 at 04:36:10PM +0100, Charles Keepax wrote: > > It is fairly common to want to read an integer array property > > that is composed of an unknown number of fixed size integer > > groups. For example, say each group consists of three values > > which correspond to the settings for one input on the device > > and the driver supports several chips with different numbers > > of inputs. > > > > Add a new helper function to provide this functionality, it > > differs for the existing helpers in that it allows reading a > > smaller number of values than the full array size and checks > > that the number of values read is a multiple of the group size. > > > > Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com> > > --- > > drivers/base/property.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++ > > include/linux/property.h | 2 ++ > > 2 files changed, 50 insertions(+) > > Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Well, I'm kind of not convinced about this one. More comments in direct reply to the patch message.
On Wed, Jun 26, 2019 at 5:36 PM Charles Keepax <ckeepax@opensource.cirrus.com> wrote: > > It is fairly common to want to read an integer array property > that is composed of an unknown number of fixed size integer > groups. For example, say each group consists of three values > which correspond to the settings for one input on the device > and the driver supports several chips with different numbers > of inputs. > > Add a new helper function to provide this functionality, it > differs for the existing helpers in that it allows reading a > smaller number of values than the full array size and checks > that the number of values read is a multiple of the group size. I'm not convinced. > Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com> > --- > drivers/base/property.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++ > include/linux/property.h | 2 ++ > 2 files changed, 50 insertions(+) > > diff --git a/drivers/base/property.c b/drivers/base/property.c > index 348b37e64944c..656d21e01a648 100644 > --- a/drivers/base/property.c > +++ b/drivers/base/property.c > @@ -133,6 +133,54 @@ int device_property_read_u32_array(struct device *dev, const char *propname, > EXPORT_SYMBOL_GPL(device_property_read_u32_array); > > /** > + * device_property_read_u32_2darray - return a 2d u32 array property of a device > + * @dev: Device to get the property of > + * @propname: Name of the property > + * @val: The values are stored here or %NULL to return the number of values > + * @nval: Size of the @val array > + * @multiple: Number of entries in each block of data > + * > + * Function reads an array of u32 properties split up into fixed size > + * sub-groups, with @propname from the device firmware description and > + * stores them to @val if found. > + * > + * Return: Number of values read > + * %0 if the property was not found, > + * %-EINVAL if given arguments are not valid, > + * %-ENODATA if the property does not have a value, > + * %-EPROTO if the property is not an array of numbers, > + * %-EOVERFLOW if the size of the property is not as expected. > + * %-ENXIO if no suitable firmware interface is present. > + */ > +int device_property_read_u32_2darray(struct device *dev, const char *propname, > + u32 *val, size_t nval, int multiple) > +{ > + int n, ret; > + > + n = device_property_read_u32_array(dev, propname, NULL, 0); > + if (n == -EINVAL) { > + return 0; /* missing, ignore */ Why can't the caller use the (scheduled for merging in the 5.3 cycle) new device_property_count_u32() to get the size of the array? > + } else if (n < 0) { > + dev_warn(dev, "%s malformed (%d)\n", propname, n); Why dev_warn()? Is there any reason real for anything higher-level that dev_dbg() here? > + return n; > + } else if ((n % multiple) != 0) { I guess the reason why this matters is that the caller expects a certain number of full "rows" and n values are read. Why not to discard the extra values instead of returning an error here? > + dev_warn(dev, "%s not a multiple of %d entries\n", > + propname, multiple); > + return -EOVERFLOW; Why this error code? > + } > + > + if (n > nval) > + n = nval; > + > + ret = device_property_read_u32_array(dev, propname, val, n); So this reads "copy at most nval values from the array property". If that's really what you need, it can be done in two lines of code in prospective callers of this wrapper. > + if (ret < 0) > + return ret; > + else > + return n; > +} > +EXPORT_SYMBOL_GPL(device_property_read_u32_2darray); > + > +/** > * device_property_read_u64_array - return a u64 array property of a device > * @dev: Device to get the property of > * @propname: Name of the property > diff --git a/include/linux/property.h b/include/linux/property.h > index e9caa290cda52..5ab0b4a7d34a2 100644 > --- a/include/linux/property.h > +++ b/include/linux/property.h > @@ -40,6 +40,8 @@ int device_property_read_u16_array(struct device *dev, const char *propname, > u16 *val, size_t nval); > int device_property_read_u32_array(struct device *dev, const char *propname, > u32 *val, size_t nval); > +int device_property_read_u32_2darray(struct device *dev, const char *propname, > + u32 *val, size_t nval, int multiple); > int device_property_read_u64_array(struct device *dev, const char *propname, > u64 *val, size_t nval); > int device_property_read_string_array(struct device *dev, const char *propname, > -- I don't see much value in this change, sorry.
On Thu, Jun 27, 2019 at 11:39:10AM +0200, Rafael J. Wysocki wrote: > On Wed, Jun 26, 2019 at 5:36 PM Charles Keepax > <ckeepax@opensource.cirrus.com> wrote: > > + n = device_property_read_u32_array(dev, propname, NULL, 0); > > + if (n == -EINVAL) { > > + return 0; /* missing, ignore */ > > Why can't the caller use the (scheduled for merging in the 5.3 cycle) > new device_property_count_u32() to get the size of the array? > I wasn't aware of it, am now. > > + } else if (n < 0) { > > + dev_warn(dev, "%s malformed (%d)\n", propname, n); > > Why dev_warn()? Is there any reason real for anything higher-level > that dev_dbg() here? > Nice to know that your DT wasn't valid, but could be left to the caller I guess. > > + return n; > > + } else if ((n % multiple) != 0) { > > I guess the reason why this matters is that the caller expects a > certain number of full "rows" and n values are read. Why not to > discard the extra values instead of returning an error here? > No reason really why it couldn't. Although my expectation would generally be this helper is for reading a variable number of fixed size groups. As in each group represents a "whole" item but you don't know how many of those you have. > > + dev_warn(dev, "%s not a multiple of %d entries\n", > > + propname, multiple); > > + return -EOVERFLOW; > > Why this error code? > As that is the error code all the device_property functions return when the size is not as expected. > > + if (n > nval) > > + n = nval; > > + > > + ret = device_property_read_u32_array(dev, propname, val, n); > > So this reads "copy at most nval values from the array property". > > If that's really what you need, it can be done in two lines of code in > prospective callers of this wrapper. > Indeed the helper here is basically exactly what would be done in the caller if no helper existed. > > +int device_property_read_u32_2darray(struct device *dev, const char *propname, > > + u32 *val, size_t nval, int multiple); > > int device_property_read_u64_array(struct device *dev, const char *propname, > > u64 *val, size_t nval); > > int device_property_read_string_array(struct device *dev, const char *propname, > > -- > > I don't see much value in this change, sorry. That is fine, I don't have any problem with the helper living within our driver instead. Basically the issue from my side is I need to read 6 different device tree properties all of which require this behaviour, ie. read a variable number of fixed groups and check I have whole groups. Open coding this for each call is indeed only going to be 5-10 lines of code for each one but since there are 6 of them it makes sense to put those 5-10 lines into a helper and have 5-10 lines not 30-60 lines. Seemed the helper might be generally more useful, but if it is not then it can go back into the driver. Thanks, Charles
On Thu, Jun 27, 2019 at 12:23 PM Charles Keepax <ckeepax@opensource.cirrus.com> wrote: > > On Thu, Jun 27, 2019 at 11:39:10AM +0200, Rafael J. Wysocki wrote: > > On Wed, Jun 26, 2019 at 5:36 PM Charles Keepax > > <ckeepax@opensource.cirrus.com> wrote: > > > + n = device_property_read_u32_array(dev, propname, NULL, 0); > > > + if (n == -EINVAL) { > > > + return 0; /* missing, ignore */ > > > > Why can't the caller use the (scheduled for merging in the 5.3 cycle) > > new device_property_count_u32() to get the size of the array? > > > > I wasn't aware of it, am now. > > > > + } else if (n < 0) { > > > + dev_warn(dev, "%s malformed (%d)\n", propname, n); > > > > Why dev_warn()? Is there any reason real for anything higher-level > > that dev_dbg() here? > > > > Nice to know that your DT wasn't valid, but could be left to the > caller I guess. Right. And only print the message when the caller really is going to fail. > > > + return n; > > > + } else if ((n % multiple) != 0) { > > > > I guess the reason why this matters is that the caller expects a > > certain number of full "rows" and n values are read. Why not to > > discard the extra values instead of returning an error here? > > > > No reason really why it couldn't. Although my expectation would > generally be this helper is for reading a variable number of > fixed size groups. As in each group represents a "whole" item but > you don't know how many of those you have. That really depends on how the property is defined and which is what the caller knows. > > > + dev_warn(dev, "%s not a multiple of %d entries\n", > > > + propname, multiple); > > > + return -EOVERFLOW; > > > > Why this error code? > > > > As that is the error code all the device_property functions > return when the size is not as expected. To be precise, when there is a mismatch between the storage space supplied by the caller and the property size. That does not seem to be the case here. It is rather about the property format being not as expected and that would be -EPROTO. > > > + if (n > nval) > > > + n = nval; > > > + > > > + ret = device_property_read_u32_array(dev, propname, val, n); > > > > So this reads "copy at most nval values from the array property". > > > > If that's really what you need, it can be done in two lines of code in > > prospective callers of this wrapper. > > > > Indeed the helper here is basically exactly what would be done in > the caller if no helper existed. > > > > +int device_property_read_u32_2darray(struct device *dev, const char *propname, > > > + u32 *val, size_t nval, int multiple); > > > int device_property_read_u64_array(struct device *dev, const char *propname, > > > u64 *val, size_t nval); > > > int device_property_read_string_array(struct device *dev, const char *propname, > > > -- > > > > I don't see much value in this change, sorry. > > That is fine, I don't have any problem with the helper living > within our driver instead. Basically the issue from my side is I > need to read 6 different device tree properties all of which > require this behaviour, ie. read a variable number of fixed > groups and check I have whole groups. Open coding this for each > call is indeed only going to be 5-10 lines of code Exactly two: n = device_property_count_u32(dev, name); ret = device_property_read_u32_array(dev, propname, val, n > nval ? nval : n); And I would be fine with adding wrappers like this (and for the other data types too for that matter). It would take more lines if you wanted to complain about the format, but as pointed out above, that would need to be done in the caller anyway. > for each one but since there are 6 of them it makes sense to put those 5-10 > lines into a helper and have 5-10 lines not 30-60 lines. Seemed > the helper might be generally more useful, but if it is not then > it can go back into the driver.
On Thu, Jun 27, 2019 at 01:02:32PM +0200, Rafael J. Wysocki wrote: > On Thu, Jun 27, 2019 at 12:23 PM Charles Keepax > <ckeepax@opensource.cirrus.com> wrote: > > On Thu, Jun 27, 2019 at 11:39:10AM +0200, Rafael J. Wysocki wrote: > > > On Wed, Jun 26, 2019 at 5:36 PM Charles Keepax > > > <ckeepax@opensource.cirrus.com> wrote: > > That is fine, I don't have any problem with the helper living > > within our driver instead. Basically the issue from my side is I > > need to read 6 different device tree properties all of which > > require this behaviour, ie. read a variable number of fixed > > groups and check I have whole groups. Open coding this for each > > call is indeed only going to be 5-10 lines of code > > Exactly two: > > n = device_property_count_u32(dev, name); > ret = device_property_read_u32_array(dev, propname, val, n > nval ? nval : n); > > And I would be fine with adding wrappers like this (and for the other > data types too for that matter). > > It would take more lines if you wanted to complain about the format, > but as pointed out above, that would need to be done in the caller > anyway. > Ok I think that helps me to follow the situation. If device_property_count cuts down the code required and leaves the majority of the code as printing the messages which then wants to live in the end driver anyways it probably isn't worth adding a core helper for this. Thank you for the review and the explanation. I will update the driver patches to use the new function and resend those. Thanks, Charles > > for each one but since there are 6 of them it makes sense to put those 5-10 > > lines into a helper and have 5-10 lines not 30-60 lines. Seemed > > the helper might be generally more useful, but if it is not then > > it can go back into the driver.
diff --git a/drivers/base/property.c b/drivers/base/property.c index 348b37e64944c..656d21e01a648 100644 --- a/drivers/base/property.c +++ b/drivers/base/property.c @@ -133,6 +133,54 @@ int device_property_read_u32_array(struct device *dev, const char *propname, EXPORT_SYMBOL_GPL(device_property_read_u32_array); /** + * device_property_read_u32_2darray - return a 2d u32 array property of a device + * @dev: Device to get the property of + * @propname: Name of the property + * @val: The values are stored here or %NULL to return the number of values + * @nval: Size of the @val array + * @multiple: Number of entries in each block of data + * + * Function reads an array of u32 properties split up into fixed size + * sub-groups, with @propname from the device firmware description and + * stores them to @val if found. + * + * Return: Number of values read + * %0 if the property was not found, + * %-EINVAL if given arguments are not valid, + * %-ENODATA if the property does not have a value, + * %-EPROTO if the property is not an array of numbers, + * %-EOVERFLOW if the size of the property is not as expected. + * %-ENXIO if no suitable firmware interface is present. + */ +int device_property_read_u32_2darray(struct device *dev, const char *propname, + u32 *val, size_t nval, int multiple) +{ + int n, ret; + + n = device_property_read_u32_array(dev, propname, NULL, 0); + if (n == -EINVAL) { + return 0; /* missing, ignore */ + } else if (n < 0) { + dev_warn(dev, "%s malformed (%d)\n", propname, n); + return n; + } else if ((n % multiple) != 0) { + dev_warn(dev, "%s not a multiple of %d entries\n", + propname, multiple); + return -EOVERFLOW; + } + + if (n > nval) + n = nval; + + ret = device_property_read_u32_array(dev, propname, val, n); + if (ret < 0) + return ret; + else + return n; +} +EXPORT_SYMBOL_GPL(device_property_read_u32_2darray); + +/** * device_property_read_u64_array - return a u64 array property of a device * @dev: Device to get the property of * @propname: Name of the property diff --git a/include/linux/property.h b/include/linux/property.h index e9caa290cda52..5ab0b4a7d34a2 100644 --- a/include/linux/property.h +++ b/include/linux/property.h @@ -40,6 +40,8 @@ int device_property_read_u16_array(struct device *dev, const char *propname, u16 *val, size_t nval); int device_property_read_u32_array(struct device *dev, const char *propname, u32 *val, size_t nval); +int device_property_read_u32_2darray(struct device *dev, const char *propname, + u32 *val, size_t nval, int multiple); int device_property_read_u64_array(struct device *dev, const char *propname, u64 *val, size_t nval); int device_property_read_string_array(struct device *dev, const char *propname,
It is fairly common to want to read an integer array property that is composed of an unknown number of fixed size integer groups. For example, say each group consists of three values which correspond to the settings for one input on the device and the driver supports several chips with different numbers of inputs. Add a new helper function to provide this functionality, it differs for the existing helpers in that it allows reading a smaller number of values than the full array size and checks that the number of values read is a multiple of the group size. Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com> --- drivers/base/property.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++ include/linux/property.h | 2 ++ 2 files changed, 50 insertions(+)