On Sun, 2017-10-29 at 20:50 +0000, Ian Lepore wrote: > Author: ian > Date: Sun Oct 29 20:50:03 2017 > New Revision: 325108 > URL: https://svnweb.freebsd.org/changeset/base/325108 > > Log: > Improve the performance of the hpet timer in bhyve guests by making the > timer frequency a power of two. This changes the frequency from 10 to > 16.7 MHz (2 ^ 24 HZ). Using a power of two avoids roundoff errors when > doing arithmetic in sbintime_t units. > > Testing shows this can fix erratic ntpd behavior in guests using the > hpet timer (which is the default for multicore guests). > > Reported by: bsam@ > > Modified: > head/sys/amd64/vmm/io/vhpet.c > > Modified: head/sys/amd64/vmm/io/vhpet.c > ============================================================================== > --- head/sys/amd64/vmm/io/vhpet.c Sun Oct 29 20:40:56 2017 > (r325107) > +++ head/sys/amd64/vmm/io/vhpet.c Sun Oct 29 20:50:03 2017 > (r325108) > @@ -51,7 +51,7 @@ __FBSDID("$FreeBSD$"); > > static MALLOC_DEFINE(M_VHPET, "vhpet", "bhyve virtual hpet"); > > -#define HPET_FREQ 10000000 /* 10.0 Mhz */ > +#define HPET_FREQ 16777216 /* 16.7 (2^24) Mhz */ > #define FS_PER_S 1000000000000000ul > > /* Timer N Configuration and Capabilities Register */ >
It should be noted that this is really more of a workaround than a fix. I think the right fix might be to use bintime rather than sbintime to avoid the rounding errors. I also suspect this problem affects other emulated timers such as atpit/8254 where there is less freedom to wish away the problem with a power-of-two frequency change. Still, I committed this as-is because it was shown to fix a real-world problem and it doesn't prevent doing a better fix later. -- Ian _______________________________________________ svn-src-head@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/svn-src-head To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"