> > the lru_lock of struct mem_group_per_zone is used to avoid preemption > > during the mem_cgroup_charge_statistics function. this does not work > > correctly, when CONFIG_PREEMPT_RT is enabled. > > therefore, the preemption is disabled using the preempt_disable_rt macro > > in these cases. > > > > Signed-off-by: Tim Blechmann <tim@xxxxxxxxxx> > > Sorry, memcg's code this function in mmotm kernel is now, following. > Please give me advice if some more thinking is necessary for RT. using get_cpu/put_cpu, the preemption is disabled ... so mmotm should not require any changes. my patch would only be required for an rt-patch, based on a current kernel version ... instead of applying my patch, one could also use the specific part of mmotm for the RT tree ... tim > static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, > struct page_cgroup *pc, > bool charge) > { > int val = (charge)? 1 : -1; > struct mem_cgroup_stat *stat = &mem->stat; > struct mem_cgroup_stat_cpu *cpustat; > int cpu = get_cpu(); > > cpustat = &stat->cpustat[cpu]; > if (PageCgroupCache(pc)) > __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); > else > __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); > > if (charge) > __mem_cgroup_stat_add_safe(cpustat, > MEM_CGROUP_STAT_PGPGIN_COUNT, 1); > else > __mem_cgroup_stat_add_safe(cpustat, > MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); > put_cpu(); > } -- tim@xxxxxxxxxx http://tim.klingt.org Silence is only frightening to people who are compulsively verbalizing. William S. Burroughs
Attachment:
signature.asc
Description: This is a digitally signed message part