Before commit
f8441069c1d5 ("clk: Allow parents to be specified without
string names") child clks couldn't find their parent until the parent
clk was added to a list in __clk_core_init(). After that commit, child
clks can reference their parent clks directly via a clk_hw pointer, or
they can lookup that clk_hw pointer via DT if the parent clk is
registered with an OF clk provider.
The common clk framework treats hw->core being non-NULL as "the clk is
registered" per the logic within clk_core_fill_parent_index():
parent = entry->hw->core;
/*
* We have a direct reference but it isn't registered yet?
* Orphan it and let clk_reparent() update the orphan status
* when the parent is registered.
*/
if (!parent)
Therefore we need to be extra careful to not set hw->core until the clk
is fully registered with the clk framework. Otherwise we can get into a
situation where a child finds a parent clk and we move the child clk off
the orphan list when the parent isn't actually registered, wrecking our
enable accounting and breaking critical clks.
Consider the following scenario:
CPU0 CPU1
---- ----
struct clk_hw clkBad;
struct clk_hw clkA;
clkA.init.parent_hws = { &clkBad };
clk_hw_register(&clkA) clk_hw_register(&clkBad)
... __clk_register()
hw->core = core
...
__clk_register()
__clk_core_init()
clk_prepare_lock()
__clk_init_parent()
clk_core_get_parent_by_index()
clk_core_fill_parent_index()
if (entry->hw) {
parent = entry->hw->core;
At this point, 'parent' points to clkBad even though clkBad hasn't been
fully registered yet. Ouch! A similar problem can happen if a clk
controller registers orphan clks that are referenced in the DT node of
another clk controller.
Let's fix all this by only setting the hw->core pointer underneath the
clk prepare lock in __clk_core_init(). This way we know that
clk_core_fill_parent_index() can't see hw->core be non-NULL until the
clk is fully registered.
Fixes: f8441069c1d5 ("clk: Allow parents to be specified without string names")
Signed-off-by: Mike Tipton <quic_mdtipton@quicinc.com>
Link: https://lore.kernel.org/r/20211109043438.4639-1-quic_mdtipton@quicinc.com
[sboyd@kernel.org: Reword commit text, update comment]
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
clk_prepare_lock();
+ /*
+ * Set hw->core after grabbing the prepare_lock to synchronize with
+ * callers of clk_core_fill_parent_index() where we treat hw->core
+ * being NULL as the clk not being registered yet. This is crucial so
+ * that clks aren't parented until their parent is fully registered.
+ */
+ core->hw->core = core;
+
ret = clk_pm_runtime_get(core);
if (ret)
goto unlock;
out:
clk_pm_runtime_put(core);
unlock:
- if (ret)
+ if (ret) {
hlist_del_init(&core->child_node);
+ core->hw->core = NULL;
+ }
clk_prepare_unlock();
core->num_parents = init->num_parents;
core->min_rate = 0;
core->max_rate = ULONG_MAX;
- hw->core = core;
ret = clk_core_populate_parent_map(core, init);
if (ret)
goto fail_create_clk;
}
- clk_core_link_consumer(hw->core, hw->clk);
+ clk_core_link_consumer(core, hw->clk);
ret = __clk_core_init(core);
if (!ret)