On Wed, May 30, 2012 at 3:52 PM, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote: > On Wed, May 30, 2012 at 02:42:42PM -0500, Christoph Lameter wrote: >> On Wed, 30 May 2012, Andi Kleen wrote: >> >> > On Wed, May 30, 2012 at 01:50:02PM -0500, Christoph Lameter wrote: >> > > On Wed, 30 May 2012, Andi Kleen wrote: >> > > >> > > > I always regretted that cpusets were no done with custom node lists. >> > > > That would have been much cleaner and also likely faster than what we have. >> > > >> > > Could shared memory policies ignore cpuset constraints? >> > >> > Only if noone uses cpusets as a "security" mechanism, just for a "soft policy" >> > Even with soft policy you could well break someone's setup. >> >> Well at least lets exempt shared memory from memory migration and memory >> policy updates. That seems to be causing many of these issues. > > Migration on the page level is needed for the memory error handling. > > Updates: you mean not allowing to set the policy when there are already > multiple mappers? I could see that causing some unexpected behaviour. Presumably > a standard database will only set it at the beginning, but I don't know > if that would work for all users. We don't need to kill migration core. We only need to kill that mbind(2) updates vma->policy of shmem. page migration for hwpoison is harmless. Because of, an attacker can't inject hwpoison intentntionally on production environment (HWPOISON_INJECTION=N). -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>