[PATCH v2] tracing: Improve performance by using do_div()

From: Thorsten Blum
Date: Wed Mar 20 2024 - 18:02:30 EST


Partially revert commit d6cb38e10810 ("tracing: Use div64_u64() instead
of do_div()") and use do_div() again to utilize its faster 64-by-32
division compared to the 64-by-64 division done by div64_u64().

Explicitly cast the divisor bm_cnt to u32 to prevent a Coccinelle
warning reported by do_div.cocci. The warning was removed with commit
d6cb38e10810 ("tracing: Use div64_u64() instead of do_div()").

Using the faster 64-by-32 division and casting bm_cnt to u32 is safe
because we return early from trace_do_benchmark() if bm_cnt > UINT_MAX.
This approach is already used twice in trace_do_benchmark() when
calculating the standard deviation:

do_div(stddev, (u32)bm_cnt);
do_div(stddev, (u32)bm_cnt - 1);

Signed-off-by: Thorsten Blum <thorsten.blum@xxxxxxxxxx>
---
Changes in v2:
- Update patch with latest changes from master
- Update patch title and description
---
kernel/trace/trace_benchmark.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/trace/trace_benchmark.c b/kernel/trace/trace_benchmark.c
index 811b08439406..e19c32f2a938 100644
--- a/kernel/trace/trace_benchmark.c
+++ b/kernel/trace/trace_benchmark.c
@@ -104,7 +104,7 @@ static void trace_do_benchmark(void)
stddev = 0;

delta = bm_total;
- delta = div64_u64(delta, bm_cnt);
+ do_div(delta, (u32)bm_cnt);
avg = delta;

if (stddev > 0) {
--
2.44.0