On Fri 22-06-18 11:49:14, David Rientjes wrote: > On Fri, 22 Jun 2018, Michal Hocko wrote: > > > > > preempt_disable() is required because it calls kvm_kick_many_cpus() with > > > > wait == true because KVM_REQ_APIC_PAGE_RELOAD sets KVM_REQUEST_WAIT and > > > > thus the smp_call_function_many() is going to block until all cpus can run > > > > ack_flush(). > > > > > > I will make sure to talk to the maintainer of the respective code to > > > do the nonblock case correctly. > > > > I've just double checked this particular code and the wait path and this > > one is not a sleep. It is a busy wait for IPI to get handled. So this > > one should be OK AFAICS. Anyway I will send an RFC and involve > > respective maintainers to make sure I am not making any incorrect > > assumptions. > > Do you believe that having the only potential source of memory freeing > busy waiting for all other cpus on the system to run ack_flush() is > particularly dangerous given the fact that they may be allocating > themselves? These are IPIs. How could they depend on a memory allocation? In other words we do rely on the very same mechanism for TLB flushing so this is any different. Maybe I am missing something here though. -- Michal Hocko SUSE Labs
