counter: microchip-tcb-capture: Fix DT channel validation
Subsystems (3)
PATCH 1/1 counter: microchip-tcb-capture: Fix DT channel validation
mchp_tc_probe() reads the devicetree "reg" cell - a u32, per the API contract of of_property_read_u32_index() - into a signed int, so the bounds check "channel > 2" fails to reject cell values at or above 0x80000000: reinterpreted as a negative int, they compare below 2 and pass validation.
A malformed devicetree can therefore drive a negative channel into the ATMEL_TC_REG() offset arithmetic, making the driver access syscon regmap offsets outside the TC block's register window, and into the "t%d_clk" clock-name formatting, where it truncates clk_name (sized for "t0_clk".."t2_clk").
Declare channel as u32, matching the API contract; the unsigned comparison then rejects everything except channels 0..2. Adjust the format specifier to %u accordingly, which also resolves the W=1 warning that exposed the gap:
microchip-tcb-capture.c:520:56: warning: '%d' directive output may
be truncated writing between 1 and 11 bytes into a region of size 6 [-Wformat-truncation=]
note: directive argument in the range [-2147483648, 2]
No behavior change for well-formed devicetrees: channels 0..2 take identical paths before and after.
---
Note: struct mchp_tc_data's channel[2] member stays int — after this fix it can only ever hold 0..2, so converting it (and the QDEC-mode comparisons reading it) would be churn beyond the minimal fix. Happy to do that conversion as a follow-up if preferred.
drivers/counter/microchip-tcb-capture.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/counter/microchip-tcb-capture.c +2 -2
@@ -483,7 +483,7 @@ static int mchp_tc_probe(struct platform_device *pdev)char clk_name[7 ];struct regmap *regmap;struct clk *clk[3 ]; -int channel; +u32 channel;int ret, i; counter =devm_counter_alloc (&pdev->dev,sizeof (*priv));
@@ -517,7 +517,7 @@ static int mchp_tc_probe(struct platform_device *pdev) priv->channel[i] = channel; -snprintf (clk_name,sizeof (clk_name),"t%d_clk" , channel); +snprintf (clk_name,sizeof (clk_name),"t%u_clk" , channel); clk[i] =of_clk_get_by_name (np->parent, clk_name);if (IS_ERR (clk[i])) {
base-commit: 3b029c035b34bbc693405ddf759f0e9b920c27f1
Signature · 1 line
2.43.0
Hi William,
A gentle ping on this one - it has been a couple of weeks without any comments, and I could not find a competing fix on the list or in linux-next.
No urgency from my side; I mainly want to check the patch has not simply been missed. Happy to respin if you would prefer a different shape, or to drop it if you would rather leave the validation as it is.
For convenience:
https://lore.kernel.org/all/[email protected]/
Thanks,
Baban
Quoted 60 lines · On Tue, 14 Jul 2026 04:29:10 +0000 Babanpreet Singh <[email protected]>…
mchp_tc_probe()reads the devicetree "reg" cell - a u32, per the API contract ofof_property_read_u32_index()- into a signed int, so the bounds check "channel > 2" fails to reject cell values at or above0x80000000: reinterpreted as a negative int, they compare below 2 and pass validation.A malformed devicetree can therefore drive a negative channel into the
ATMEL_TC_REG()offset arithmetic, making the driver access syscon regmap offsets outside the TC block's register window, and into the "t%d_clk" clock-name formatting, where it truncatesclk_name(sized for "t0_clk".."t2_clk").Declare channel as u32, matching the API contract; the unsigned comparison then rejects everything except channels 0..2. Adjust the format specifier to %u accordingly, which also resolves the W=1 warning that exposed the gap:
microchip-tcb-capture.c:520:56: warning: '%d' directive output may
be truncated writing between 1 and 11 bytes into a region of size 6 [-Wformat-truncation=]
note: directive argument in the range [-2147483648, 2]No behavior change for well-formed devicetrees: channels 0..2 take identical paths before and after.
Fixes 106b104137fd ("counter: Add microchip TCB capture counter")Assisted-by Claude:claude-fable-5 [gcc W=1]Signed-off-by Babanpreet Singh <[email protected]>---
Note: struct mchp_tc_data's channel[2] member stays int — after this fix it can only ever hold 0..2, so converting it (and the QDEC-mode comparisons reading it) would be churn beyond the minimal fix. Happy to do that conversion as a follow-up if preferred.drivers/counter/microchip-tcb-capture.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)drivers/counter/microchip-tcb-capture.c +2
-2@@ -483,7 +483,7 @@ static int mchp_tc_probe(struct platform_device *pdev)char clk_name[7 ];struct regmap *regmap;struct clk *clk[3 ]; -int channel; +u32 channel;int ret, i; counter =devm_counter_alloc (&pdev->dev,sizeof (*priv));@@ -517,7 +517,7 @@ static int mchp_tc_probe(struct platform_device *pdev) priv->channel[i] = channel; -snprintf (clk_name,sizeof (clk_name),"t%d_clk" , channel); +snprintf (clk_name,sizeof (clk_name),"t%u_clk" , channel); clk[i] =of_clk_get_by_name (np->parent, clk_name);if (IS_ERR (clk[i])) {
Quoted 1 line
base-commit:
3b029c035b34bbc693405ddf759f0e9b920c27f1
Makes sense.
Signature · 2 lines
Kind regards, Joshua Crofts
William Breathitt Gray
maintainerQuoted 10 lines · Babanpreet Singh
Hi William,
A gentle ping on this one - it has been a couple of weeks without any comments, and I could not find a competing fix on the list or in linux-next.
No urgency from my side; I mainly want to check the patch has not simply been missed. Happy to respin if you would prefer a different shape, or to drop it if you would rather leave the validation as it is.
Quoted 5 lines
For convenience:
https://lore.kernel.org/all/[email protected]/Thanks,
Baban
This patch looks sensible. Do you have any objections Kamel?
William Breathitt Gray
Quoted 6 lines · Babanpreet Singh
mchp_tc_probe()reads the devicetree "reg" cell - a u32, per the API contract ofof_property_read_u32_index()- into a signed int, so the bounds check "channel > 2" fails to reject cell values at or above0x80000000: reinterpreted as a negative int, they compare below 2 and pass validation.
Quoted 7 lines
A malformed devicetree can therefore drive a negative channel into the
ATMEL_TC_REG()offset arithmetic, making the driver access syscon regmap offsets outside the TC block's register window, and into the "t%d_clk" clock-name formatting, where it truncatesclk_name(sized for "t0_clk".."t2_clk").[...]
Applied, thanks!
[1/1] counter: microchip-tcb-capture: Fix DT channel validation
commit: f1a3a9946aab611dd2200c01ff122f64b033dad2
Best regards,
Signature · 1 line
William Breathitt Gray <[email protected]>