PR/60780 CVS commit: src/sys/kern

["Taylor R Campbell" <riastradh%netbsd.org@localhost>]

The following reply was made to PR port-mips/60780; it has been noted by GNATS.

From: "Taylor R Campbell" <riastradh%netbsd.org@localhost>
To: gnats-bugs%gnats.NetBSD.org@localhost
Cc: 
Subject: PR/60780 CVS commit: src/sys/kern
Date: Sat, 3 Oct 2026 00:13:34 +0000

 Module Name:	src
 Committed By:	riastradh
 Date:		Sat Oct  3 00:13:33 UTC 2026
 
 Modified Files:
 	src/sys/kern: kern_ntptime.c
 
 Log Message:
 ntp(9): Avoid more left shift of negative UB.
 
 This logic is, presumably, intended to compute integer arithmetic, so
 just write it as *16 instead of <<4.  If there's an advantage to
 using a machine shift instruction to get the same semantics, the
 compiler can do that for us.
 
 Also avoid arithmetic overflow.  If set a few lines above,
 time_monitor can lie anywhere in the interval [-MAXPHASE,MAXPHASE] =
 [-500e6,500e6].  Multiplying by sixteen can therefore overflow the
 bounds [-2.2e9,2.2e9] of long on LP32 platforms by a factor of four.
 But mtemp >= 256 here, so even if time_monitor*16 overflows the
 signed 32-bit range, the result (time_monitor*16)/mtemp will not.
 Hence: cast to int64_t for the intermediate computation of
 time_monitor*16.
 
 This isn't the end of the analysis: time_monitor can also be set in
 hardpps(9) to something else whose bounds aren't as clear to me, but
 that only applies under `options PPS_SYNC' which is usually not set.
 Perhaps we need to enforce bounds on that too.
 
 PR port-mips/60780: UBSan complains about left-shifting outside of
 int type range
 
 
 To generate a diff of this commit:
 cvs rdiff -u -r1.64 -r1.65 src/sys/kern/kern_ntptime.c
 
 Please note that diffs are not public domain; they are subject to the
 copyright notices on the relevant files.