| Message ID | 20240815005520.1192374-10-sboyd@kernel.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | clk: Unit test clk unregistration paths | expand |
diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c index a0c275e156ad..d99d2d4dd411 100644 --- a/drivers/clk/clk.c +++ b/drivers/clk/clk.c @@ -4611,6 +4611,11 @@ void clk_unregister(struct clk *clk) clk_core_evict_parent_cache(clk->core); hlist_del_init(&clk->core->child_node); + /* + * Prevent clk from being reinserted into the clk tree via + * clk_set_parent() + */ + clk->core->num_parents = 0; if (clk->core->prepare_count) pr_warn("%s: unregistering prepared clock: %s\n",
The clk framework removes the clk_core from the clk tree by deleting the 'child_node' hlist for a clk_hw when it is unregistered in clk_unregister(). This removes the clk from the clk tree, but a call to clk_set_parent() will allow the clk to be added right back into the clk tree, undoing all the work done in clk_unregister() to remove it from the tree and orphan all descendants. This results in a double hlist delete as well, because clk_reparent() deletes the 'child_node' hlist before adding the clk to either the orphan list or the clk tree. The 'clk_nodrv_ops' have a set_parent clk_op that returns failure, but that still won't save us even if we fix the clk_reparent() code to operate on an empty list. That's because we'll reparent the clk back to the original parent when the clk_op returns an error, effectively putting the clk back into the clk tree. Force the number of parents to be zero in clk_unregister() so that clk_set_parent() can't get past the part where it figures out which index to use to call the clk_op with. Cc: Nuno Sá <nuno.sa@analog.com> Cc: Sylwester Nawrocki <s.nawrocki@samsung.com> Fixes: fcb0ee6a3d33 ("clk: Implement clk_unregister") Signed-off-by: Stephen Boyd <sboyd@kernel.org> --- drivers/clk/clk.c | 5 +++++ 1 file changed, 5 insertions(+)