@@ -2030,7 +2030,7 @@ struct memcg_stock_pcp {
local_lock_t stock_lock;
struct mem_cgroup *cached; /* this never be root cgroup */
unsigned int nr_pages;
-#ifndef CONFIG_PREEMPT_RT
+#ifndef CONFIG_PREEMPTION
/* Protects only task_obj */
local_lock_t task_obj_lock;
struct obj_stock task_obj;
@@ -2043,7 +2043,7 @@ struct memcg_stock_pcp {
};
static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = {
.stock_lock = INIT_LOCAL_LOCK(stock_lock),
-#ifndef CONFIG_PREEMPT_RT
+#ifndef CONFIG_PREEMPTION
.task_obj_lock = INIT_LOCAL_LOCK(task_obj_lock),
#endif
};
@@ -2132,7 +2132,7 @@ static void drain_local_stock(struct work_struct *dummy)
* drain_stock races is that we always operate on local CPU stock
* here with IRQ disabled
*/
-#ifndef CONFIG_PREEMPT_RT
+#ifndef CONFIG_PREEMPTION
local_lock(&memcg_stock.task_obj_lock);
old = drain_obj_stock(&this_cpu_ptr(&memcg_stock)->task_obj, NULL);
local_unlock(&memcg_stock.task_obj_lock);
@@ -2741,7 +2741,7 @@ static inline struct obj_stock *get_obj_stock(unsigned long *pflags,
{
struct memcg_stock_pcp *stock;
-#ifndef CONFIG_PREEMPT_RT
+#ifndef CONFIG_PREEMPTION
if (likely(in_task())) {
*pflags = 0UL;
*stock_lock_acquried = false;
@@ -2759,7 +2759,7 @@ static inline struct obj_stock *get_obj_stock(unsigned long *pflags,
static inline void put_obj_stock(unsigned long flags,
bool stock_lock_acquried)
{
-#ifndef CONFIG_PREEMPT_RT
+#ifndef CONFIG_PREEMPTION
if (likely(!stock_lock_acquried)) {
local_unlock(&memcg_stock.task_obj_lock);
return;
@@ -3177,7 +3177,7 @@ static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
{
struct mem_cgroup *memcg;
-#ifndef CONFIG_PREEMPT_RT
+#ifndef CONFIG_PREEMPTION
if (in_task() && stock->task_obj.cached_objcg) {
memcg = obj_cgroup_memcg(stock->task_obj.cached_objcg);
if (memcg && mem_cgroup_is_descendant(memcg, root_memcg))
Based on my understanding the optimisation with task_obj for in_task() mask sense on non-PREEMPTIBLE kernels because preempt_disable()/enable() is optimized away. This could be then restricted to !CONFIG_PREEMPTION kernel instead to only PREEMPT_RT. With CONFIG_PREEMPT_DYNAMIC a non-PREEMPTIBLE kernel can also be configured but these kernels always have preempt_disable()/enable() present so it probably makes no sense here for the optimisation. I did a micro benchmark with disabled interrupts and a loop of 100.000.000 invokcations of kfree(kmalloc()). Based on the results it makes no sense to add an exception based on dynamic preemption. Restrict the optimisation to !CONFIG_PREEMPTION kernels. Link: https://lore.kernel.org/all/YdX+INO9gQje6d0S@linutronix.de Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> --- mm/memcontrol.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)