Skip to contentSkip to conversation

clk: qcom: clk-rcg2: Fix wrong RCG clock rate for high parent frequencies

PATCH Superseded Devi Priya 4 participants 6 messages Jul 20, 2023 – Aug 30, 2023

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Devi Priya

author
#0 lore raw

PATCH 1/1 clk: qcom: clk-rcg2: Fix wrong RCG clock rate for high parent frequencies

If the parent clock rate is greater than unsigned long max/2 then integer overflow happens when calculating the clock rate on 32-bit systems. As RCG2 uses half integer dividers, the clock rate is first being multiplied by 2 which will overflow the unsigned long max value. So, use unsigned long long for rate computations to avoid overflow.

---
 drivers/clk/qcom/clk-rcg2.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/clk/qcom/clk-rcg2.c +6 -6
index e22baf3a7112..42d00b134975 100644
@@ -156,18 +156,18 @@ static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index)
  *            hid_div       n
  */
 static unsigned long
-calc_rate(unsigned long rate, u32 m, u32 n, u32 mode, u32 hid_div)
+calc_rate(unsigned long parent_rate, u32 m, u32 n, u32 mode, u32 hid_div)
 {
+	u64 rate = parent_rate;
+
 	if (hid_div) {
 		rate *= 2;
-		rate /= hid_div + 1;
+		do_div(rate, hid_div + 1);
 	}
 
 	if (mode) {
-		u64 tmp = rate;
-		tmp *= m;
-		do_div(tmp, n);
-		rate = tmp;
+		rate *= m;
+		do_div(rate, n);
 	}
 
 	return rate;
Signature · 1 line
2.17.1
#1 lore raw
Quoted 16 lines · Devi Priya

If the parent clock rate is greater than unsigned long max/2 then integer overflow happens when calculating the clock rate on 32-bit systems. As RCG2 uses half integer dividers, the clock rate is first being multiplied by 2 which will overflow the unsigned long max value. So, use unsigned long long for rate computations to avoid overflow.

---
 drivers/clk/qcom/clk-rcg2.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/clk/qcom/clk-rcg2.c +0 -0
index e22baf3a7112..42d00b134975 100644

drivers/clk/qcom/clk-rcg2.c static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index) · line 161

Quoted 7 lines
drivers/clk/qcom/clk-rcg2.c +2 -1
@@ -156,18 +156,18 @@ static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index)
  *            hid_div       n
  */
 static unsigned long
-calc_rate(unsigned long rate, u32 m, u32 n, u32 mode, u32 hid_div)
+calc_rate(unsigned long parent_rate, u32 m, u32 n, u32 mode, u32 hid_div)
 {
+	u64 rate = parent_rate;

This should not be necessary.. You're being passed a copy of the original value, which you can operate on.

Otherwise, LGTM

Konrad

Reviewed-by
#2 lore raw
Quoted 27 lines · Konrad Dybcio
Quoted 16 lines · Devi Priya

If the parent clock rate is greater than unsigned long max/2 then integer overflow happens when calculating the clock rate on 32-bit systems. As RCG2 uses half integer dividers, the clock rate is first being multiplied by 2 which will overflow the unsigned long max value. So, use unsigned long long for rate computations to avoid overflow.

---
 drivers/clk/qcom/clk-rcg2.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/clk/qcom/clk-rcg2.c +0 -0
index e22baf3a7112..42d00b134975 100644

drivers/clk/qcom/clk-rcg2.c static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index) · line 161

Quoted 7 lines
drivers/clk/qcom/clk-rcg2.c +2 -1
@@ -156,18 +156,18 @@ static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index)
  *            hid_div       n
  */
 static unsigned long
-calc_rate(unsigned long rate, u32 m, u32 n, u32 mode, u32 hid_div)
+calc_rate(unsigned long parent_rate, u32 m, u32 n, u32 mode, u32 hid_div)
 {
+	u64 rate = parent_rate;

This should not be necessary.. You're being passed a copy of the original value, which you can operate on.

Quoted 1 line

Otherwise, LGTM

Well obviously no, as I hit enter I realized this is of a different type..

Konrad

maintainer
#3 lore raw
In reply to Devi Priya

Quoting Devi Priya (2023-07-20 01:33:04)

Quoted 8 lines

If the parent clock rate is greater than unsigned long max/2 then integer overflow happens when calculating the clock rate on 32-bit systems. As RCG2 uses half integer dividers, the clock rate is first being multiplied by 2 which will overflow the unsigned long max value. So, use unsigned long long for rate computations to avoid overflow.

---

Any Fixes tag?

#4 lore raw
In reply to Devi Priya

Can you retitle this to state "overflow" rather than just "wrong"? That's more descriptive.

E.g "Fix clockrate overflow for high parent frequencies"

Quoted 18 lines · Devi Priya

If the parent clock rate is greater than unsigned long max/2 then integer overflow happens when calculating the clock rate on 32-bit systems. As RCG2 uses half integer dividers, the clock rate is first being multiplied by 2 which will overflow the unsigned long max value. So, use unsigned long long for rate computations to avoid overflow.

---
 drivers/clk/qcom/clk-rcg2.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/clk/qcom/clk-rcg2.c +0 -0
index e22baf3a7112..42d00b134975 100644
@@ -156,18 +156,18 @@ static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index)
  *            hid_div       n
  */

drivers/clk/qcom/clk-rcg2.c static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index) · line 165

Quoted 10 lines
drivers/clk/qcom/clk-rcg2.c +4 -2
 static unsigned long
-calc_rate(unsigned long rate, u32 m, u32 n, u32 mode, u32 hid_div)
+calc_rate(unsigned long parent_rate, u32 m, u32 n, u32 mode, u32 hid_div)
 {
+	u64 rate = parent_rate;
+
 	if (hid_div) {
 		rate *= 2;
-		rate /= hid_div + 1;
+		do_div(rate, hid_div + 1);

I'm pretty sure mult_frac() could have solved this as well, without temporarily going to u64?

    mult_frac(rate, 2, hid_div + 1)

drivers/clk/qcom/clk-rcg2.c static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index) · line 170

Quoted 9 lines
drivers/clk/qcom/clk-rcg2.c +2 -4
 	}
 
 	if (mode) {
-		u64 tmp = rate;
-		tmp *= m;
-		do_div(tmp, n);
-		rate = tmp;
+		rate *= m;
+		do_div(rate, n);

mult_frac(rate, m, n)

Or am I totally wrong?

- Marijn

Quoted 6 lines
drivers/clk/qcom/clk-rcg2.c +0 -0
 	}
 
 	return rate;
Signature · 1 line
2.17.1
author
#5 lore raw
Quoted 4 lines · Marijn Suijten

Can you retitle this to state "overflow" rather than just "wrong"? That's more descriptive.

E.g "Fix clockrate overflow for high parent frequencies"

Sure okay

Quoted 34 lines · Devi Priya
Quoted 18 lines

If the parent clock rate is greater than unsigned long max/2 then integer overflow happens when calculating the clock rate on 32-bit systems. As RCG2 uses half integer dividers, the clock rate is first being multiplied by 2 which will overflow the unsigned long max value. So, use unsigned long long for rate computations to avoid overflow.

---
  drivers/clk/qcom/clk-rcg2.c | 12 ++++++------
  1 file changed, 6 insertions(+), 6 deletions(-)
drivers/clk/qcom/clk-rcg2.c +0 -0
index e22baf3a7112..42d00b134975 100644
@@ -156,18 +156,18 @@ static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index)
   *            hid_div       n
   */

drivers/clk/qcom/clk-rcg2.c static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index) · line 165

Quoted 10 lines
drivers/clk/qcom/clk-rcg2.c +4 -2
  static unsigned long
-calc_rate(unsigned long rate, u32 m, u32 n, u32 mode, u32 hid_div)
+calc_rate(unsigned long parent_rate, u32 m, u32 n, u32 mode, u32 hid_div)
  {
+	u64 rate = parent_rate;
+
  	if (hid_div) {
  		rate *= 2;
-		rate /= hid_div + 1;
+		do_div(rate, hid_div + 1);

I'm pretty sure mult_frac() could have solved this as well, without temporarily going to u64?

Quoted 1 line
     mult_frac(rate, 2, hid_div + 1)

Yes, sure will update

Quoted 12 lines

drivers/clk/qcom/clk-rcg2.c static int clk_rcg2_set_parent(struct clk_hw *hw, u8 index) · line 170

Quoted 10 lines
drivers/clk/qcom/clk-rcg2.c +2 -4
  	}
 
  	if (mode) {
-		u64 tmp = rate;
-		tmp *= m;
-		do_div(tmp, n);
-		rate = tmp;
+		rate *= m;
+		do_div(rate, n);

mult_frac(rate, m, n)

Will update in V2

Thanks,
Devi Priya

Quoted 11 lines

Or am I totally wrong?

- Marijn

Quoted 6 lines
drivers/clk/qcom/clk-rcg2.c +0 -0
  	}
 
  	return rate;
Signature · 1 line
2.17.1