CVS commit: src/sys/kern

["YAMAMOTO Takashi" <yamt%netbsd.org@localhost>]

Module Name:    src
Committed By:   yamt
Date:           Sun Mar  1 13:56:50 UTC 2026

Modified Files:
        src/sys/kern: sched_4bsd.c

Log Message:
sched_4bsd.c: honor the upper bound of l_estcpu

otherwise, certain threaded workloads can quickly
accumulate too large l_estcpu.

this change fixes mysterious long (eg 10 seconds) pauses
i've seen occasionally with git. git seems to spawn worker
threads internally these days and happens to hit the condition
of this bug. with very large l_estcpu, the lwp gets effective
priority 0 for a long time. on a busy system, such an lwp
basically will never get a chance to run until the estcpu decay
will eventually make it possible to run again.

it seems this bug has been there since 2007.
("Merge scheduler changes from the vmlocking branch")
at that time, while i had been using netbsd heavily for daily
things, i haven't noticed this issue at all.  it seems that
threaded programs got ubiquitous in this decades.


To generate a diff of this commit:
cvs rdiff -u -r1.47 -r1.48 src/sys/kern/sched_4bsd.c

Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.