Linux clock framework development
 help / color / mirror / Atom feed
* [PATCH] clk: rp1: allow the clock provider to be probed again
@ 2026-11-21 16:39 Joshua Covington
  2026-11-21 16:53 ` sashiko-bot
  0 siblings, 1 reply; 2+ messages in thread
From: Joshua Covington @ 2026-11-21 16:39 UTC (permalink / raw)
  To: Stephen Boyd
  Cc: Brian Masney, Jerome Brunet, Andrea della Porta, linux-clk,
	linux-rpi-kernel, linux-kernel, Joshua Covington

The clock descriptors are static and are registered directly. The
clock framework clears hw.init once a clock has been registered, so
when the RP1 device is unbound and bound again, the second probe
registers clocks without init data and oopses:

  pc : __clk_register+0x5c/0x9dc
  Call trace:
   __clk_register+0x5c/0x9dc (P)
   devm_clk_hw_register+0x5c/0xe0
   rp1_register_pll+0x24/0x40
   rp1_clk_probe+0xc0/0x168

Keep the init data of each descriptor and restore it before the clock
is registered.

Fixes: 6486341721a2 ("clk: rp1: Add support for clocks provided by RP1")
Signed-off-by: Joshua Covington <joshuacov@gmail.com>
---
Tested on a Raspberry Pi CM5 with 6.18.64 and the Raspberry Pi
downstream patches, where binding rp1_pci again after unbinding it hit
this. Applies unchanged to v7.4-rc3 and builds there without warnings.

 drivers/clk/clk-rp1.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/clk/clk-rp1.c b/drivers/clk/clk-rp1.c
index 1dd276dd2..5c3929e93 100644
--- a/drivers/clk/clk-rp1.c
+++ b/drivers/clk/clk-rp1.c
@@ -363,6 +363,7 @@ struct rp1_clk_desc {
 				       struct rp1_clk_desc *desc);
 	const void *data;
 	struct clk_hw hw;
+	const struct clk_init_data *init;
 	struct rp1_clockman *clockman;
 	unsigned long cached_rate;
 	struct clk_divider div;
@@ -2418,8 +2419,15 @@ static int rp1_clk_probe(struct platform_device *pdev)
 
 	for (i = 0; i < asize; i++) {
 		desc = clk_desc_array[i];
-		if (desc && desc->clk_register && desc->data)
-			hws[i] = desc->clk_register(clockman, desc);
+		if (!desc || !desc->clk_register || !desc->data)
+			continue;
+
+		/* The clock core clears hw.init of these static descriptors */
+		if (!desc->init)
+			desc->init = desc->hw.init;
+		desc->hw.init = desc->init;
+
+		hws[i] = desc->clk_register(clockman, desc);
 	}
 
 	clk_audio_core = &pll_audio_core_desc;
-- 
2.56.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] clk: rp1: allow the clock provider to be probed again
  2026-11-21 16:39 [PATCH] clk: rp1: allow the clock provider to be probed again Joshua Covington
@ 2026-11-21 16:53 ` sashiko-bot
  0 siblings, 0 replies; 2+ messages in thread
From: sashiko-bot @ 2026-11-21 16:53 UTC (permalink / raw)
  To: Joshua Covington; +Cc: linux-clk

Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [High] clk: rp1: use-after-free in rp1_clk_probe() from stale hw.core
  on re-probe
--

commit 0c4e0577acfd5e28d80d2ef3e1cc9693b2d23c01
Author: Joshua Covington <joshuacov@gmail.com>

clk: rp1: allow the clock provider to be probed again

This commit saves and restores the clk_init_data pointer on each static
rp1_clk_desc descriptor in rp1_clk_probe() so the driver can register its
clocks again after an unbind and rebind.

> diff --git a/drivers/clk/clk-rp1.c b/drivers/clk/clk-rp1.c
> index 1dd276dd2bb3e..5c3929e935fce 100644
> --- a/drivers/clk/clk-rp1.c
> +++ b/drivers/clk/clk-rp1.c
> @@ -363,6 +363,7 @@ struct rp1_clk_desc {
>  				       struct rp1_clk_desc *desc);
>  	const void *data;
>  	struct clk_hw hw;
> +	const struct clk_init_data *init;
>  	struct rp1_clockman *clockman;
>  	unsigned long cached_rate;
>  	struct clk_divider div;
> @@ -2418,8 +2419,15 @@ static int rp1_clk_probe(struct platform_device *pdev)
>  
>  	for (i = 0; i < asize; i++) {
>  		desc = clk_desc_array[i];
> -		if (desc && desc->clk_register && desc->data)
> -			hws[i] = desc->clk_register(clockman, desc);
> +		if (!desc || !desc->clk_register || !desc->data)
> +			continue;
> +
> +		/* The clock core clears hw.init of these static descriptors */
> +		if (!desc->init)
> +			desc->init = desc->hw.init;
> +		desc->hw.init = desc->init;
> +
> +		hws[i] = desc->clk_register(clockman, desc);
>  	}

[Severity: High]
Can re-probing cause a use-after-free on a freed struct clk_core when a
clock references a parent at a higher index in clk_desc_array?

When the device is unbound, clk_unregister() frees each clock's struct
clk_core without clearing hw->core on the static descriptors.  In
clk_core_fill_parent_index(), the clock core checks entry->hw->core and
assumes a non-NULL pointer is an already-registered parent.

Several descriptors in clk_desc_array reference parents via .hw that sit
at higher array indices, such as clk_mipi0_dpi_desc (RP1_CLK_MIPI0_DPI)
referencing clksrc_mipi0_dsi_byteclk_desc (RP1_CLK_MIPI0_DSI_BYTECLOCK).
Because rp1_clk_probe() registers clocks in a single pass without first
clearing desc->hw.core (and desc->div.hw.core) across clk_desc_array, if
the hardware mux selects one of those higher-indexed parents during
re-probe, __clk_init_parent() -> clk_core_fill_parent_index() returns the
dangling clk_core pointer from the previous bind and __clk_core_init()
links into parent->children on freed memory.

Clearing desc->hw.core and desc->div.hw.core for all descriptors before
the registration loop would avoid this.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261121163915.86681-1-joshuacov@gmail.com?part=1

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-21 16:53 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-21 16:39 [PATCH] clk: rp1: allow the clock provider to be probed again Joshua Covington
2026-11-21 16:53 ` sashiko-bot

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox