Re: [PATCH RFC RFT v2 2/3] clk: divider: add CLK_DIVIDER_INVALID_ZERO flag

From: Jian Hu

Date: Fri Nov 20 2026 - 03:24:12 EST



On 11/6/2026 1:50 AM, Jerome Brunet wrote:
When a divider register reads zero, divider_recalc_rate() currently
emits a warning (unless CLK_DIVIDER_ALLOW_ZERO is set) and returns
the parent rate.

This behaviour assumes a zero divisor acts as a pass-through, which
does not apply to all hardware. Some dividers, for example PLL
pre-dividers on Amlogic SoCs, come out of reset with a zero value
that does not produce a usable clock. Returning the parent rate
hides this fault behind an incorrect, plausible value.

CLK_DIVIDER_ALLOW_ZERO only suppresses the warning while still
reporting the parent rate, so it cannot represent this hardware
behaviour.

Add CLK_DIVIDER_INVALID_ZERO for dividers where a zero divisor value
is invalid. When this flag is set, recalc_rate() returns 0, making
the faulty state visible to the clock framework rather than hiding it
as a pass-through.

Signed-off-by: Jian Hu <jian.hu@xxxxxxxxxxx>
---
drivers/clk/clk-divider.c | 3 +++
include/linux/clk-provider.h | 6 ++++++
2 files changed, 9 insertions(+)

diff --git a/drivers/clk/clk-divider.c b/drivers/clk/clk-divider.c
index f68216c65d3d..6c1523d22477 100644
--- a/drivers/clk/clk-divider.c
+++ b/drivers/clk/clk-divider.c
@@ -143,6 +143,9 @@ unsigned long divider_recalc_rate(struct clk_hw *hw, unsigned long parent_rate,

div = _get_div(table, val, flags, width);
if (!div) {
+ if (flags & CLK_DIVIDER_INVALID_ZERO)
+ return 0;
While I think CLK_DIVIDER_INVALID_ZERO really makes sense for a writeable divider
only, to gracefully handle a reset value at zero that we would later correct,
we should still handle all paths correctly. This include a RO divider and its
determine rate path. I'll try to send something for this soon.


You are right, and I agree: a change in the core divider code has to be
correct for every user, not only for the cases this series happens to use it
for. I understand the RO divider / determine_rate path is the part that still
needs handling.

I will drop the 2/3 patch in v3 since you will submit it separately.

+
WARN(!(flags & CLK_DIVIDER_ALLOW_ZERO),
"%s: Zero divisor and CLK_DIVIDER_ALLOW_ZERO not set\n",
clk_hw_get_name(hw));
diff --git a/include/linux/clk-provider.h b/include/linux/clk-provider.h
index dc8e161f9c2c..fe2dbbba49e3 100644
--- a/include/linux/clk-provider.h
+++ b/include/linux/clk-provider.h
@@ -744,6 +744,11 @@ struct clk_div_table {
* big endian.
* CLK_DIVIDER_EVEN_INTEGERS - clock divisor is 2, 4, 6, 8, 10, etc.
* Formula is 2 * (value read from hardware + 1).
+ * CLK_DIVIDER_INVALID_ZERO - a zero value read from the register is not a
+ * divide by one / bypass but an invalid setting which does not produce
+ * a usable clock. recalc_rate() returns 0 so that the broken state is
+ * reported to the framework instead of the parent rate, without
+ * triggering the CLK_DIVIDER_ALLOW_ZERO warning.

CLK_DIVIDER_INVALID_ZERO - a zero divisor is an invalid setting with no
usable output. recalc_rate() returns 0 instead of the parent rate.
Takes precedence over CLK_DIVIDER_ALLOW_ZERO.


Ok, it is short and clear.

--
Jerome

--

Jian