> On Sat, Dec 11, 2010 at 10:37:47PM +0000, Matt Thomas wrote: > > Module Name: src > > Committed By: matt > > Date: Sat Dec 11 22:37:47 UTC 2010 > > > > Modified Files: > > src/sys/netinet: in_offload.c in_offload.h ip_input.c > > src/sys/netinet6: in6_offload.c in6_offload.h > > > > Log Message: > > Add routines to calculate a checkesum if the driver concludes that the > > h/w can't do it. > > The change to ip_input.c: > > 1) looks wrong (ipintr is MPSAFE, so kernel lock should not be held > there to begin with)
it also hangs my SB2500 at mount_nfs time. > 2) breaks all tests that use IP (*) > 3) conflicts with the performance anecdote in rev 1.286 > > Can you address the above points or back out the change? Thanks. > > > *) > http://www.gson.org/netbsd/bugs/build/build/2010.12.13.01.48.50/test.html#failed-tcs-summary > http://www.whooppee.com/~paul/amd64-results/566_atf.html#failed-tcs-summary >
