[HEADSUP] big polling changes
Gleb Smirnoff
glebius at FreeBSD.org
Mon Oct 3 13:23:24 PDT 2005
On Tue, Oct 04, 2005 at 12:12:29AM +0400, Yuriy N. Shkandybin wrote:
Y> Hello.
Y>
Y> I've tested nge and confirm that with
Y> sc->nge_ldata = contigmalloc(sizeof(struct nge_list_data), M_DEVBUF,
Y> - M_NOWAIT, 0, 0xffffffff, PAGE_SIZE, 0);
Y> + M_NOWAIT|M_ZERO, 0, 0xffffffff, PAGE_SIZE, 0);
Y>
Y> polling works.
I hope Yar will handle this one.
Y> Also want to notice interrupt load with polling enable.
Y> CPU states: 0.0% user, 0.0% nice, 2.4% system, 26.4% interrupt, 71.2% idle
Y> without polling.
Y> CPU states: 1.0% user, 0.0% nice, 5.8% system, 0.0% interrupt, 93.3% idle
Y> but same behavour i've seen since RELENG_5.
Y>
Y> There is no such load with em driver.
Was there such load before my polling changes?
Y> If additional testing needed i can help and provide access to testbox with 7.0 and nge card.
--
Totus tuus, Glebius.
GLEBIUS-RIPN GLEB-RIPE
More information about the freebsd-current
mailing list