2024-02-25 23:29:07

by Thorsten Blum

Subject: [PATCH] time: Use div64_long() instead of do_div()

Fixes Coccinelle/coccicheck warning reported by do_div.cocci.

Compared to do_div(), div64_long() does not implicitly cast the divisor and
does not unnecessarily calculate the remainder.

Signed-off-by: Thorsten Blum <[email protected]>
---
kernel/time/jiffies.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/time/jiffies.c b/kernel/time/jiffies.c
index bc4db9e5ab70..9d23178e2b6a 100644
--- a/kernel/time/jiffies.c
+++ b/kernel/time/jiffies.c
@@ -91,7 +91,7 @@ int register_refined_jiffies(long cycles_per_second)
/* shift_hz stores hz<<8 for extra accuracy */
shift_hz = (u64)cycles_per_second << 8;
shift_hz += cycles_per_tick/2;
- do_div(shift_hz, cycles_per_tick);
+ shift_hz = div64_long(shift_hz, cycles_per_tick);
/* Calculate nsec_per_tick using shift_hz */
nsec_per_tick = (u64)NSEC_PER_SEC << 8;
nsec_per_tick += (u32)shift_hz/2;
--
2.43.2

2024-02-26 03:28:55

by Liao, Chang

Subject: Re: [PATCH] time: Use div64_long() instead of do_div()

Thorsten,

在 2024/2/26 7:25, Thorsten Blum 写道:
> Fixes Coccinelle/coccicheck warning reported by do_div.cocci.
>
> Compared to do_div(), div64_long() does not implicitly cast the divisor and
> does not unnecessarily calculate the remainder.
>
> Signed-off-by: Thorsten Blum <[email protected]>
> ---
> kernel/time/jiffies.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/time/jiffies.c b/kernel/time/jiffies.c
> index bc4db9e5ab70..9d23178e2b6a 100644
> --- a/kernel/time/jiffies.c
> +++ b/kernel/time/jiffies.c
> @@ -91,7 +91,7 @@ int register_refined_jiffies(long cycles_per_second)
> /* shift_hz stores hz<<8 for extra accuracy */
> shift_hz = (u64)cycles_per_second << 8;
> shift_hz += cycles_per_tick/2;
> - do_div(shift_hz, cycles_per_tick);
> + shift_hz = div64_long(shift_hz, cycles_per_tick);

I am considering using div64_ul() to calculate the result. as shift_hz is
unsigned long, assume the sign bit of divisor cycles_per_tick never be set
in this context,then div64_long() will do a extra sign extension for result.
Or are there other recommendations?

Thanks.

> /* Calculate nsec_per_tick using shift_hz */
> nsec_per_tick = (u64)NSEC_PER_SEC << 8;
> nsec_per_tick += (u32)shift_hz/2;

--
BR
Liao, Chang

2024-02-26 08:27:52

by Thorsten Blum

Subject: Re: [PATCH] time: Use div64_long() instead of do_div()

> On Feb 26, 2024, at 04:28, Liao, Chang <[email protected]> wrote:
>
> I am considering using div64_ul() to calculate the result. as shift_hz is
> unsigned long, assume the sign bit of divisor cycles_per_tick never be set
> in this context,then div64_long() will do a extra sign extension for result.

Yes, div64_ul() is better and also removes the warning.

I'll submit a v2 shortly.

Thanks,
Thorsten

2024-02-26 09:41:50

by Thorsten Blum

Subject: [PATCH v2] time: Use div64_ul() instead of do_div()

Fixes Coccinelle/coccicheck warning reported by do_div.cocci.

Compared to do_div(), div64_ul() does not implicitly cast the divisor and
does not unnecessarily calculate the remainder.

Signed-off-by: Thorsten Blum <[email protected]>
---
Changes in v2:
- s/div64_long/div64_ul/ as suggested by Chang Liao
---
kernel/time/jiffies.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/time/jiffies.c b/kernel/time/jiffies.c
index bc4db9e5ab70..fccee15e94d5 100644
--- a/kernel/time/jiffies.c
+++ b/kernel/time/jiffies.c
@@ -91,7 +91,7 @@ int register_refined_jiffies(long cycles_per_second)
/* shift_hz stores hz<<8 for extra accuracy */
shift_hz = (u64)cycles_per_second << 8;
shift_hz += cycles_per_tick/2;
- do_div(shift_hz, cycles_per_tick);
+ shift_hz = div64_ul(shift_hz, cycles_per_tick);
/* Calculate nsec_per_tick using shift_hz */
nsec_per_tick = (u64)NSEC_PER_SEC << 8;
nsec_per_tick += (u32)shift_hz/2;
--
2.43.2