clk: Evict unregistered clks from parent caches

We leave a dangling pointer in each clk_core::parents array that has an unregistered clk as a potential parent when that clk_core pointer is freed by clk{_hw}_unregister(). It is impossible for the true parent of a clk to be set with clk_set_parent() once the dangling pointer is left in the cache because we compare parent pointers in clk_fetch_parent_index() instead of checking for a matching clk name or clk_hw pointer.

Before commit ede77858473a ("clk: Remove global clk traversal on fetch parent index"), we would check clk_hw pointers, which has a higher chance of being the same between registration and unregistration, but it can still be allocated and freed by the clk provider. In fact, this has been a long standing problem since commit da0f0b2c3ad2 ("clk: Correct lookup logic in clk_fetch_parent_index()") where we stopped trying to compare clk names and skipped over entries in the cache that weren't NULL.

There are good (performance) reasons to not do the global tree lookup in cases where the cache holds dangling pointers to parents that have been unregistered. Let's take the performance hit on the uncommon registration path instead. Loop through all the clk_core::parents arrays when a clk is unregistered and set the entry to NULL when the parent cache entry and clk being unregistered are the same pointer. This will fix this problem and avoid the overhead for the "normal" case.

Based on a patch by Bjorn Andersson.

Fixes da0f0b2c3ad2 ("clk: Correct lookup logic in clk_fetch_parent_index()")

Cc Bjorn Andersson <[email protected]>

Cc Sai Prakash Ranjan <[email protected]>

Signed-off-by Stephen Boyd <[email protected]>

---
 drivers/clk/clk.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

drivers/clk/clk.c +30 -0

index c0990703ce54..f3982bfa39d6 100644
@@ -3737,6 +3737,34 @@ static const struct clk_ops clk_nodrv_ops = {
 	.set_parent	= clk_nodrv_set_parent,
 };
 
+static void clk_core_evict_parent_cache_subtree(struct clk_core *root,
+						struct clk_core *target)
+{
+	int i;
+	struct clk_core *child;
+
+	for (i = 0; i < root->num_parents; i++)
+		if (root->parents[i].core == target)
+			root->parents[i].core = NULL;
+
+	hlist_for_each_entry(child, &root->children, child_node)
+		clk_core_evict_parent_cache_subtree(child, target);
+}
+
+/* Remove this clk from all parent caches */
+static void clk_core_evict_parent_cache(struct clk_core *core)
+{
+	struct hlist_head **lists;
+	struct clk_core *root;
+
+	lockdep_assert_held(&prepare_lock);
+
+	for (lists = all_lists; *lists; lists++)
+		hlist_for_each_entry(root, *lists, child_node)
+			clk_core_evict_parent_cache_subtree(root, core);
+
+}
+
 /**
  * clk_unregister - unregister a currently registered clock
  * @clk: clock to unregister

@@ -3775,6 +3803,8 @@ void clk_unregister(struct clk *clk)
 			clk_core_set_parent_nolock(child, NULL);
 	}
 
+	clk_core_evict_parent_cache(clk->core);
+
 	hlist_del_init(&clk->core->child_node);
 
 	if (clk->core->prepare_count)

Signature · 1 line

Sent by a computer through tubes

We leave a dangling pointer in each clk_core::parents array that has an unregistered clk as a potential parent when that clk_core pointer is freed by clk{_hw}_unregister(). It is impossible for the true parent of a clk to be set with clk_set_parent() once the dangling pointer is left in the cache because we compare parent pointers in clk_fetch_parent_index() instead of checking for a matching clk name or clk_hw pointer.

Before commit ede77858473a ("clk: Remove global clk traversal on fetch parent index"), we would check clk_hw pointers, which has a higher chance of being the same between registration and unregistration, but it can still be allocated and freed by the clk provider. In fact, this has been a long standing problem since commit da0f0b2c3ad2 ("clk: Correct lookup logic in clk_fetch_parent_index()") where we stopped trying to compare clk names and skipped over entries in the cache that weren't NULL.

There are good (performance) reasons to not do the global tree lookup in cases where the cache holds dangling pointers to parents that have been unregistered. Let's take the performance hit on the uncommon registration path instead. Loop through all the clk_core::parents arrays when a clk is unregistered and set the entry to NULL when the parent cache entry and clk being unregistered are the same pointer. This will fix this problem and avoid the overhead for the "normal" case.

Based on a patch by Bjorn Andersson.

Fixes da0f0b2c3ad2 ("clk: Correct lookup logic in clk_fetch_parent_index()")

Cc Bjorn Andersson <[email protected]>

Cc Sai Prakash Ranjan <[email protected]>

Signed-off-by Stephen Boyd <[email protected]>

---
 drivers/clk/clk.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

drivers/clk/clk.c +28 -0

index c0990703ce54..f3982bfa39d6 100644
@@ -3737,6 +3737,34 @@ static const struct clk_ops clk_nodrv_ops = {
 	.set_parent	= clk_nodrv_set_parent,
 };
 
+static void clk_core_evict_parent_cache_subtree(struct clk_core *root,
+						struct clk_core *target)
+{
+	int i;
+	struct clk_core *child;
+
+	for (i = 0; i < root->num_parents; i++)
+		if (root->parents[i].core == target)
+			root->parents[i].core = NULL;
+
+	hlist_for_each_entry(child, &root->children, child_node)
+		clk_core_evict_parent_cache_subtree(child, target);
+}
+
+/* Remove this clk from all parent caches */
+static void clk_core_evict_parent_cache(struct clk_core *core)
+{
+	struct hlist_head **lists;
+	struct clk_core *root;
+
+	lockdep_assert_held(&prepare_lock);
+
+	for (lists = all_lists; *lists; lists++)
+		hlist_for_each_entry(root, *lists, child_node)
+			clk_core_evict_parent_cache_subtree(root, core);
+
+}
+
 /**
  * clk_unregister - unregister a currently registered clock
  * @clk: clock to unregister

Hi Stephen,

I love your patch! Yet something to improve:

[auto build test ERROR on linus/master]
[cannot apply to v5.3-rc6 next-20190827]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url: https://github.com/0day-ci/linux/commits/Stephen-Boyd/clk-Evict-unregistered-clks-from-parent-caches/20190827-165138
config: mips-allnoconfig (attached as .config)
compiler: mips-linux-gcc (GCC) 7.4.0
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
GCC_VERSION=7.4.0 make.cross ARCH=mips

If you fix the issue, kindly add following tag

Reported-by kbuild test robot <[email protected]>

All errors (new ones prefixed by >>):

drivers/clk/clk.c: In function 'clk_core_evict_parent_cache':

Quoted 1 line

Quoted 1 line

drivers/clk/clk.c:3785:15: error: 'all_lists' undeclared (first use in this function); did you mean 'lists'?

for (lists = all_lists; *lists; lists++)
^~~~~~~~~
lists
drivers/clk/clk.c:3785:15: note: each undeclared identifier is reported only once for each function it appears in