On Thu, Mar 17, 2016 at 16:39 +0100, Mike Belopuhov wrote: > On Wed, Mar 16, 2016 at 23:47 -0700, Mike Larkin wrote: > > On Thu, Mar 17, 2016 at 03:31:45PM +0900, Masao Uebayashi wrote: > > > On Wed, Mar 16, 2016 at 11:26:39PM -0700, Mike Larkin wrote: > > > > On Thu, Mar 17, 2016 at 03:15:07PM +0900, Masao Uebayashi wrote: > > > > > This is the intended use, avoid busy-polling of BIOS PC console if > > > > > running on HV. Avoid 100% CPU usage at boot prompt on hypervisors. > > > > > > > > > > > > > Which hypervisors have you tested this on? > > > > > > VMware ESXi > > > KVM > > > QEMU > > > > While I share some of Theo's concerns about testing this, this particular > > diff seems ok to me. You probably at least need to get someone to test on > > Hyper-V, and run these new bootblocks on a variety of machines (especially > > older non-Intel/AMD i386 machines). > > > > -ml > > > > I've tested the diff and with a small adjustment to the > efiboot/bootia32/Makefile below, I was able to rebuild > stand/ and test the thing (with the follow up diff to > the bioscons.c). > > As I've mentioned on ICB, Xen exposes this issue, while > Hyper-V doesn't. The change doesn't help or harm Xen, no > change on Hyper-V. But I think it's still useful if this > helps KVM and VMware. >
Apparently I missed that glass and serial consoles have different implementations. Duh! So with a "glass" console I can see reduction on both Xen and Hyper-V. On Xen CPU load goes from 100% (according to "xl top") to 0, while on Hyper-V: from 8% to 0. Thanks again for the diff.