When traversing all nodes and moving to a new path component, the old
one must be released by calling of_node_put(). Else the refcounts of the
parent node(s) will not be decremented.

Signed-off-by: Geert Uytterhoeven <[email protected]>
---
Background.

While investigating a reference count imbalance issue with
CONFIG_OF_DYNAMIC=y, I wrote the debug code below to validate the
reference counts of the nodes I was interested in.
During the first call of check_refcnts(), it gathers all reference
counts. During a subsequent call, it verifies that they are still the
same.

Surprisingly, lots of reference counts were wrong, and kept incrementing
every time check_refcnts() was called.

I was just wondering whether it would be useful to have a reference
count test in OF_UNITTEST, but now I see the "select OF_DYNAMIC" will go
away?

Feel free to (ab)use the code below and derive a unittest from it...

static struct to_check {
        const char *path;
        int refcnt;
} to_check[] = {
        { "/" },
        { "/cpus/cpu@0" },
        { "/cpus/cpu@1" },
        /* ... other paths I was interested in ... */
};

static void check_refcnts(void)
{
        static bool called;
        unsigned int i;
        const char *path;
        struct device_node *np;
        int refcnt;
        unsigned int errors = 0;

        pr_info("----- %s reference counts -----\n",
                called ? "Checking" : "Saving");
        for (i = 0; i < ARRAY_SIZE(to_check); i++) {
                path = to_check[i].path;
                np = of_find_node_by_path(path);
                if (!np)
                        continue;

                refcnt = atomic_read(&np->kobj.kref.refcount);
                if (!called) {
                        pr_info("%s %d\n", path, refcnt);
                        to_check[i].refcnt = refcnt;
                } else if (refcnt == to_check[i].refcnt) {
                        pr_info("%s %d (OK)\n", path, refcnt);
                } else {
                        pr_info("%s %d (should be %d)\n", path, refcnt,
                                to_check[i].refcnt);
                        errors++;
                }

                of_node_put(np);
        }

        if (called)
                pr_info("----- Checking done (%u errors) -----\n", errors);
        else
                pr_info("----- Saving done -----\n");

        called = true;
}
---
 drivers/of/base.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/of/base.c b/drivers/of/base.c
index 36536b6a8834acd2..f3e346e19c69d1f2 100644
--- a/drivers/of/base.c
+++ b/drivers/of/base.c
@@ -791,8 +791,10 @@ struct device_node *of_find_node_opts_by_path(const char 
*path, const char **opt
        if (!np)
                np = of_node_get(of_root);
        while (np && *path == '/') {
+               struct device_node *parent = np;
                path++; /* Increment past '/' delimiter */
-               np = __of_find_node_by_path(np, path);
+               np = __of_find_node_by_path(parent, path);
+               of_node_put(parent);
                path = strchrnul(path, '/');
        }
        raw_spin_unlock_irqrestore(&devtree_lock, flags);
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to