[PATCHv6+ 01/13] of: introduce of_property_for_earch_phandle_with_args()
From: Grant Likely <hidden>
Date: 2013-12-12 11:34:26
Also in:
linux-devicetree, linux-iommu, linux-tegra, lkml
On Wed, 11 Dec 2013 14:33:38 +0100, Hiroshi Doyu [off-list ref] wrote:
quoted hunk ↗ jump to hunk
Hi Grant, Grant Likely [off-list ref] wrote @ Wed, 11 Dec 2013 14:28:45 +0100:quoted
On Thu, 21 Nov 2013 11:57:00 -0700, Stephen Warren [off-list ref] wrote:quoted
On 11/21/2013 10:17 AM, Hiroshi Doyu wrote:quoted
Iterating over a property containing a list of phandles with arguments is a common operation for device drivers. This patch adds a new of_property_for_each_phandle_with_args() macro to make the iteration simpler. Signed-off-by: Hiroshi Doyu <redacted> --- v6+: Use the description, which Grant Likely proposed, to be full enough that a future reader can figure out why a patch was written. http://lists.linuxfoundation.org/pipermail/iommu/2013-November/007062.html...quoted
That's right, I forgot I said that. Yes please fix the implementation.Here's the latest. I'll include this with the next v7 series. Can I get your Acked-by with this? --8<---- From 8f7c0404aa68f0e8dbe0babc240590f6528ecc1f Mon Sep 17 00:00:00 2001 From: Hiroshi Doyu <redacted> Date: Fri, 15 Nov 2013 10:52:53 +0200 Subject: [PATCH] of: introduce of_property_for_each_phandle_with_args() Iterating over a property containing a list of phandles with arguments is a common operation for device drivers. This patch adds a new of_property_for_each_phandle_with_args() macro to make the iteration simpler. Signed-off-by: Hiroshi Doyu <redacted> Cc: Rob Herring <redacted> --- v7: Fixed some minors pointed by Rob and Stephen. v6++++: Iterate without intrducing a new struct. v6+++: Introduced a new struct "of_phandle_iter" to keep the state when iterating over the list. v6++: Optimized to avoid O(n^2), suggested by Stephen Warren. http://lists.linuxfoundation.org/pipermail/iommu/2013-November/007066.html I didn't introduce any struct to hold params and state here. v6+: Use the description, which Grant Likely proposed, to be full enough that a future reader can figure out why a patch was written. v5: New patch for v5. Signed-off-by: Hiroshi Doyu <redacted> --- drivers/of/base.c | 46 ++++++++++++++++++++++++++++++++++++++++++++++ include/linux/of.h | 32 ++++++++++++++++++++++++++++++++ 2 files changed, 78 insertions(+)diff --git a/drivers/of/base.c b/drivers/of/base.c index f807d0e..cd4ab05 100644 --- a/drivers/of/base.c +++ b/drivers/of/base.c@@ -1201,6 +1201,52 @@ void of_print_phandle_args(const char *msg, const struct of_phandle_args *args) printk("\n"); } +const __be32 *of_phandle_iter_next(const char *cells_name, int cell_count, + const __be32 *cur, const __be32 *end, + struct of_phandle_args *out_args)
Having to pass in cells_name, cell_count, cur and end each time seems a
little odd. Can a state structure be used instead?
struct of_phandle_iter_state {
const char *cells_name;
int cells_count;
const __be32 *cur;
const __be32 *end;
struct of_phandle_args out_args;
}
Make the caller provide one of those and fill it in with the init
function.
quoted hunk ↗ jump to hunk
+{ + struct device_node *dn; + int i; + + if (!cells_name && !cell_count) + return NULL; + + if (!cur || (cur >= end)) + return NULL; + + dn = of_find_node_by_phandle(be32_to_cpup(cur++)); + if (!dn) + return NULL; + + if (cells_name) + if (of_property_read_u32(dn, cells_name, &cell_count)) + return NULL; + + out_args->np = dn; + out_args->args_count = cell_count; + for (i = 0; i < cell_count; i++) + out_args->args[i] = be32_to_cpup(cur++); + + return cur; +} +EXPORT_SYMBOL_GPL(of_phandle_iter_next); + +const __be32 *of_phandle_iter_init(const struct device_node *np, + const char *list_name, + const __be32 **end) +{ + size_t bytes; + const __be32 *cur; + + cur = of_get_property(np, list_name, &bytes); + *end = cur; + if (bytes) + *end += bytes / sizeof(*cur); + + return cur; +} +EXPORT_SYMBOL_GPL(of_phandle_iter_init); + static int __of_parse_phandle_with_args(const struct device_node *np, const char *list_name, const char *cells_name,diff --git a/include/linux/of.h b/include/linux/of.h index 276c546..4345582 100644 --- a/include/linux/of.h +++ b/include/linux/of.h@@ -303,6 +303,14 @@ extern int of_parse_phandle_with_fixed_args(const struct device_node *np, extern int of_count_phandle_with_args(const struct device_node *np, const char *list_name, const char *cells_name); +extern const __be32 *of_phandle_iter_init(const struct device_node *np, + const char *list_name, + const __be32 **end); +extern const __be32 *of_phandle_iter_next(const char *cells_name, + int cell_count, + const __be32 *cur, const __be32 *end, + struct of_phandle_args *out_args); + extern void of_alias_scan(void * (*dt_alloc)(u64 size, u64 align)); extern int of_alias_get_id(struct device_node *np, const char *stem);@@ -527,6 +535,22 @@ static inline int of_count_phandle_with_args(struct device_node *np, return -ENOSYS; } +static inline const __be32 *of_phandle_iter_init(const struct device_node *np, + const char *list_name, + const __be32 **end) +{ + return NULL; +} + +static inline const __be32 *of_phandle_iter_next(const char *cells_name, + int cell_count, + const __be32 *cur, + const __be32 *end, + struct of_phandle_args *out_args); +{ + return NULL; +} + static inline int of_alias_get_id(struct device_node *np, const char *stem) { return -ENOSYS;@@ -613,6 +637,14 @@ static inline int of_property_read_u32(const struct device_node *np, s; \ s = of_prop_next_string(prop, s)) +#define of_property_for_each_phandle_with_args(node, list_name, cells_name, \ + cell_count, out_args, cur, end) \ + for (cur = of_phandle_iter_init(node, list_name, &end), \ + cur = of_phandle_iter_next(cells_name, cell_count, \ + cur, end, &out_args); \
The above construct is a little odd. Why wouldn't the initializer provide the first element (or NULL if empty) right at the start. That in combination with the suggestion I made above would change the macro to be: #define of_property_for_each_phandle_with_args(node, list_name, cells_name, \ cell_count, &iter_state) \ for (cur = of_phandle_iter_init(node, list_name, cells_name, \ cells_count, &iter_state); \ cur; cur = of_phandle_iter_next(&iter_state)) \ Simpler, right? It also means whatever the user passed in for cells_name, cell_count won't get evaluated every time through the loop. g.
+ cur; \ + cur = of_phandle_iter_next(cells_name, cell_count, cur, end, &out_args)) + #if defined(CONFIG_PROC_FS) && defined(CONFIG_PROC_DEVICETREE) extern void proc_device_tree_add_node(struct device_node *, struct proc_dir_entry *); extern void proc_device_tree_add_prop(struct proc_dir_entry *pde, struct property *prop); -- 1.8.1.5