On 06/23/2013 08:33 PM, Sascha Hauer wrote:
On Fri, Jun 21, 2013 at 05:15:18PM +0200, Sebastian Hesselbarth wrote:Since property values can be empty, we need property value pointer to be non-NULL to distinguish those properties from non-existing properties. This adds a static u32 to which empty properties set their value pointer. Also, the value memory is only freed, if property length is not zero. Signed-off-by: Sebastian Hesselbarth<sebastian.hesselbarth@xxxxxxxxx> --- Cc: barebox@xxxxxxxxxxxxxxxxxxx --- drivers/of/base.c | 6 +++++- 1 files changed, 5 insertions(+), 1 deletions(-) diff --git a/drivers/of/base.c b/drivers/of/base.c index 7926d5d..a100a17 100644 --- a/drivers/of/base.c +++ b/drivers/of/base.c @@ -1501,6 +1501,8 @@ struct device_node *of_new_node(struct device_node *parent, const char *name) return node; } +static u32 empty_prop_value; + struct property *of_new_property(struct device_node *node, const char *name, const void *data, int len) { @@ -1515,6 +1517,7 @@ struct property *of_new_property(struct device_node *node, const char *name, goto bail_out; prop->length = len; + prop->value =&empty_prop_value;This at least breaks of_set_property() and of_free(). Both unconditionally do a free(pp->value).
Sascha, this is patch 3/3, the two functions above are taken care of patch 2/3.
Why do we need this anyway? We can always call of_find_property() to see if a property exists.
Actually, I was preparing to import drivers/of/address.c from linux. of_translate_one uses of_get_property on "ranges". This returns the property's value pointer instead of the property itself, which is NULL for an empty ranges property. Now that you point it out, using of_find_property is also an option. Nevertheless, patches 1 (with your comments applied) and 2 seem sensible to me. Patch 3 is an option to keep it in sync with Linux OF API behavior but is optional. Sebastian _______________________________________________ barebox mailing list barebox@xxxxxxxxxxxxxxxxxxx http://lists.infradead.org/mailman/listinfo/barebox
