@@ -207,6 +207,12 @@ static inline unsigned long thp_vma_suitable_orders(struct vm_area_struct *vma,
return orders;
}
+#if defined(CONFIG_MEMCG) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
+bool memcg_thp_vma_allowable_orders(struct vm_area_struct *vma,
+ unsigned long vm_flags,
+ unsigned long *res_mask);
+#endif
+
static inline bool file_thp_enabled(struct vm_area_struct *vma)
{
struct inode *inode;
@@ -255,7 +261,9 @@ unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
if (hugepage_global_always() ||
((vm_flags & VM_HUGEPAGE) && hugepage_global_enabled()))
mask |= READ_ONCE(huge_anon_orders_inherit);
-
+#if defined(CONFIG_MEMCG) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
+ memcg_thp_vma_allowable_orders(vma, vm_flags, &mask);
+#endif
orders &= mask;
if (!orders)
return 0;
@@ -9,7 +9,7 @@ extern struct attribute_group khugepaged_attr_group;
extern int khugepaged_init(void);
extern void khugepaged_destroy(void);
-extern int start_stop_khugepaged(void);
+extern int start_stop_khugepaged(bool force_stop);
extern void __khugepaged_enter(struct mm_struct *mm);
extern void __khugepaged_exit(struct mm_struct *mm);
extern void khugepaged_enter_vma(struct vm_area_struct *vma,
@@ -23,6 +23,7 @@
#include <linux/writeback.h>
#include <linux/page-flags.h>
#include <linux/shrinker.h>
+#include <linux/khugepaged.h>
struct mem_cgroup;
struct obj_cgroup;
@@ -1069,6 +1070,9 @@ static inline void memcg_memory_event_mm(struct mm_struct *mm,
void split_page_memcg(struct page *head, int old_order, int new_order);
+bool memcg_thp_vma_allowable_orders(struct vm_area_struct *vma,
+ unsigned long vm_flags,
+ unsigned long *res_mask);
#else /* CONFIG_MEMCG */
#define MEM_CGROUP_ID_SHIFT 0
@@ -1476,6 +1480,13 @@ void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
static inline void split_page_memcg(struct page *head, int old_order, int new_order)
{
}
+
+static inline bool memcg_thp_vma_allowable_orders(struct vm_area_struct *vma,
+ unsigned long vm_flags,
+ unsigned long *res_mask)
+{
+ return false;
+}
#endif /* CONFIG_MEMCG */
/*
@@ -172,15 +172,23 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
}
if (!vma_is_anonymous(vma)) {
+ bool memcg_enabled = false;
/*
* Enforce sysfs THP requirements as necessary. Anonymous vmas
* were already handled in thp_vma_allowable_orders().
*/
- if (enforce_sysfs &&
- (!hugepage_global_enabled() || (!(vm_flags & VM_HUGEPAGE) &&
- !hugepage_global_always())))
- return 0;
+ if (enforce_sysfs) {
+ unsigned long mask = 0UL;
+
+ memcg_enabled = memcg_thp_vma_allowable_orders(vma, vm_flags, &mask);
+ if (memcg_enabled && !mask)
+ return 0;
+ if (!memcg_enabled && (!hugepage_global_enabled() ||
+ (!(vm_flags & VM_HUGEPAGE) &&
+ !hugepage_global_always())))
+ return 0;
+ }
/*
* Trust that ->huge_fault() handlers know what they are doing
* in fault path.
@@ -343,7 +351,7 @@ static ssize_t enabled_store(struct kobject *kobj,
err = mem_cgroup_thp_flags_update_all(transparent_hugepage_flags,
HUGEPAGE_FLAGS_ENABLED_MASK);
else
- err = start_stop_khugepaged();
+ err = start_stop_khugepaged(false);
if (err)
ret = err;
@@ -539,7 +547,7 @@ static ssize_t thpsize_enabled_store(struct kobject *kobj,
if (ret > 0) {
int err;
- err = start_stop_khugepaged();
+ err = start_stop_khugepaged(false);
if (err)
ret = err;
}
@@ -812,7 +820,7 @@ static int __init hugepage_init(void)
return 0;
}
- err = start_stop_khugepaged();
+ err = start_stop_khugepaged(false);
if (err)
goto err_khugepaged;
@@ -415,6 +415,8 @@ static inline int hpage_collapse_test_exit_or_disable(struct mm_struct *mm)
static bool hugepage_pmd_enabled(void)
{
+ if (IS_ENABLED(CONFIG_MEMCG) && !mem_cgroup_disabled())
+ return true;
/*
* We cover both the anon and the file-backed case here; file-backed
* hugepages, when configured in, are determined by the global control.
@@ -2586,7 +2588,7 @@ static void set_recommended_min_free_kbytes(void)
int nr_zones = 0;
unsigned long recommended_min;
- if (!hugepage_pmd_enabled()) {
+ if (!hugepage_pmd_enabled() || !khugepaged_thread) {
calculate_min_free_kbytes();
goto update_wmarks;
}
@@ -2631,12 +2633,12 @@ static void set_recommended_min_free_kbytes(void)
setup_per_zone_wmarks();
}
-int start_stop_khugepaged(void)
+int start_stop_khugepaged(bool force_stop)
{
int err = 0;
mutex_lock(&khugepaged_mutex);
- if (hugepage_pmd_enabled()) {
+ if (hugepage_pmd_enabled() && !force_stop) {
if (!khugepaged_thread)
khugepaged_thread = kthread_run(khugepaged, NULL,
"khugepaged");
@@ -981,6 +981,37 @@ static inline bool memcg_thp_madvise_enabled(struct mem_cgroup *memcg)
return test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &memcg->thp_flags);
}
+bool memcg_thp_vma_allowable_orders(struct vm_area_struct *vma,
+ unsigned long vm_flags,
+ unsigned long *res_mask)
+{
+ unsigned long mask = 0UL;
+
+ struct mem_cgroup *memcg = get_mem_cgroup_from_mm(vma->vm_mm);
+
+ if (!memcg)
+ return false;
+
+ if (memcg_thp_always_enabled(memcg) ||
+ ((vm_flags & VM_HUGEPAGE) &&
+ memcg_thp_madvise_enabled(memcg))) {
+ if (!vma_is_anonymous(vma))
+ mask = THP_ORDERS_ALL_FILE_DEFAULT;
+ else {
+ mask = READ_ONCE(memcg->thp_anon_orders_always);
+
+ if (vm_flags & VM_HUGEPAGE)
+ mask |= READ_ONCE(memcg->thp_anon_orders_madvise);
+
+ mask = mask | READ_ONCE(memcg->thp_anon_orders_inherit);
+ }
+ }
+
+ css_put(&memcg->css);
+ *res_mask = mask;
+ return true;
+}
+
unsigned long memcg_get_thp_flags(struct vm_area_struct *vma)
{
unsigned long flags = 0UL;
@@ -3986,10 +4017,52 @@ static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset)
}
}
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+static int mem_cgroup_notify_khugepaged_cb(struct task_struct *p, void *arg)
+{
+ struct vm_area_struct *vma = NULL;
+ struct mem_cgroup *memcg = arg;
+ bool is_madvise = memcg_thp_madvise_enabled(memcg);
+ bool is_always = memcg_thp_always_enabled(memcg);
+
+ VMA_ITERATOR(vmi, p->mm, 0);
+
+ if (!is_always && !is_madvise) {
+ khugepaged_exit(p->mm);
+ return 0;
+ }
+
+ for_each_vma(vmi, vma) {
+ if (is_madvise && !(vma->vm_flags & VM_HUGEPAGE))
+ continue;
+
+ khugepaged_enter_vma(vma, vma->vm_flags);
+
+ if (test_bit(MMF_VM_HUGEPAGE, &vma->vm_mm->flags))
+ break;
+ }
+
+ return 0;
+}
+
+static void mem_cgroup_thp_attach(struct cgroup_taskset *tset)
+{
+ struct task_struct *task;
+ struct cgroup_subsys_state *css;
+
+ cgroup_taskset_for_each(task, css, tset) {
+ mem_cgroup_notify_khugepaged_cb(task, mem_cgroup_from_css(css));
+ }
+}
+#else
+static void mem_cgroup_thp_attach(struct cgroup_taskset *tset) {}
+#endif
+
static void mem_cgroup_attach(struct cgroup_taskset *tset)
{
mem_cgroup_lru_gen_attach(tset);
mem_cgroup_kmem_attach(tset);
+ mem_cgroup_thp_attach(tset);
}
static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value)
@@ -4369,6 +4442,54 @@ int memory_thp_enabled_show(struct seq_file *m, void *v)
seq_printf(m, "%s\n", thp_enabled_string(flags));
return 0;
}
+
+static int mem_cgroup_notify_khugepaged(struct mem_cgroup *memcg)
+{
+ struct mem_cgroup *iter;
+ int ret = 0;
+
+ for_each_mem_cgroup_tree(iter, memcg) {
+ struct css_task_iter it;
+ struct task_struct *task;
+
+ css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it);
+ while (!ret && (task = css_task_iter_next(&it))) {
+ if (!task->mm || (task->flags & PF_KTHREAD))
+ continue;
+
+ ret = mem_cgroup_notify_khugepaged_cb(task, memcg);
+ }
+ css_task_iter_end(&it);
+ if (ret) {
+ mem_cgroup_iter_break(memcg, iter);
+ break;
+ }
+ }
+
+ return ret;
+}
+
+ssize_t memory_thp_enabled_write(struct kernfs_open_file *of, char *buf,
+ size_t nbytes, loff_t off)
+{
+ int err;
+ int ret = nbytes;
+ struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
+
+ buf = strstrip(buf);
+
+ mutex_lock(&memcg_thp_flags_mutex);
+ ret = thp_enabled_parse(buf, &memcg->thp_flags) ? : nbytes;
+ if (ret > 0) {
+ err = mem_cgroup_notify_khugepaged(memcg);
+ if (!err)
+ err = start_stop_khugepaged(false);
+ if (err)
+ ret = err;
+ }
+ mutex_unlock(&memcg_thp_flags_mutex);
+ return ret;
+}
#endif
static struct cftype memory_files[] = {
@@ -4443,6 +4564,7 @@ static struct cftype memory_files[] = {
{
.name = "thp_enabled",
.seq_show = memory_thp_enabled_show,
+ .write = memory_thp_enabled_write,
},
#endif
{ } /* terminate */
@@ -4909,7 +5031,9 @@ void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
int mem_cgroup_thp_flags_update_all(unsigned long new_flags, unsigned long mask)
{
- int ret = 0;
+ int ret;
+ struct css_task_iter task_iter;
+ struct task_struct *task;
struct mem_cgroup *iter, *memcg = root_mem_cgroup;
unsigned long enabled_mask =
(1UL << TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG) |
@@ -4922,8 +5046,18 @@ int mem_cgroup_thp_flags_update_all(unsigned long new_flags, unsigned long mask)
unsigned long old_flags = iter->thp_flags;
iter->thp_flags = (old_flags & ~mask) | new_flags;
+
+ css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &task_iter);
+ while ((task = css_task_iter_next(&task_iter))) {
+ if (!task->mm || (task->flags & PF_KTHREAD))
+ continue;
+
+ mem_cgroup_notify_khugepaged_cb(task, iter);
+ }
+ css_task_iter_end(&task_iter);
}
+ ret = start_stop_khugepaged(!enabled_mask);
mutex_unlock(&memcg_thp_flags_mutex);
return ret;
}
@@ -5509,4 +5643,15 @@ static int __init mem_cgroup_swap_init(void)
}
subsys_initcall(mem_cgroup_swap_init);
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+static int __init mem_cgroup_thp_init(void)
+{
+ if (IS_ENABLED(CONFIG_MEMCG))
+ root_mem_cgroup->thp_flags = transparent_hugepage_flags;
+
+ return 0;
+}
+subsys_initcall(mem_cgroup_thp_init);
+#endif
#endif /* CONFIG_SWAP */