b...@freebsd.org presented the problem that this change fixed. I think it
was in a driver. It may have been a case in which the macro argument
had side effects, and having it appear twice in the macro made those
side effects happen twice. But I have not preserved the original report
from bz. I regret that I did not adequately describe the problem being
fixed in the commit message. Perhaps bz can shed some light on the matter.
Doug
On 2/10/25 21:30, Zhenlei Huang wrote:
On Feb 10, 2025, at 7:15 PM, Doug Moore <do...@freebsd.org> wrote:
The branch stable/14 has been updated by dougm:
URL:
https://cgit.FreeBSD.org/src/commit/?id=7bcc7a0b88ccb5e1fe31de88ab9a17e46859318b
commit 7bcc7a0b88ccb5e1fe31de88ab9a17e46859318b
Author: Doug Moore <do...@freebsd.org>
AuthorDate: 2024-09-27 23:43:07 +0000
Commit: Doug Moore <do...@freebsd.org>
CommitDate: 2025-02-10 10:30:05 +0000
libkern: avoid local var in order_base_2()
order_base_2(n) is implemented with a variable, which keeps it from
being used at file scope. Implement it instead as ilog2(2*n-1), which
produces a different result when 2*n overflows, which appears unlikely
in practice.
Reviewed by: bz
Differential Revision: https://reviews.freebsd.org/D46826
(cherry picked from commit b7cbf741d55468ba34305a14ac3acc1c286af034)
---
sys/sys/libkern.h | 5 +----
1 file changed, 1 insertion(+), 4 deletions(-)
diff --git a/sys/sys/libkern.h b/sys/sys/libkern.h
index a10289d72dc7..835e5ffaf0b7 100644
--- a/sys/sys/libkern.h
+++ b/sys/sys/libkern.h
@@ -229,10 +229,7 @@ ilog2_long_long(long long n)
#define ilog2(n) (__builtin_constant_p(n) ? ilog2_const(n) : ilog2_var(n))
#define rounddown_pow_of_two(n) ((__typeof(n))1 << ilog2(n))
-#define order_base_2(n) ({ \
- __typeof(n) _n = (n); \
This local var `_n` is within the scope of the macro `order_base_2`, it is
surrounded with
{} and is harmless, so it will not pollute the file scoped variables.
Am I reading the commit message wrong ?
Best regards,
Zhenlei
- _n == 1 ? 0 : 1 + ilog2(_n - 1); \
-})
+#define order_base_2(n) ilog2(2*(n)-1)
#define roundup_pow_of_two(n) ((__typeof(n))1 << order_base_2(n))
#define bitcount64(x) __bitcount64((uint64_t)(x))
