From nobody Mon Nov 25 02:51:59 2024 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 8C3451E0B67; Wed, 30 Oct 2024 08:33:43 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=185.176.79.56 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1730277226; cv=none; b=U7XUUSUtd3RsJAn8IGTmCmMZAgVBYV3Tgzs5xFlU1fjDd0B64poHM8cplsBpD2In1e09jgx7Jvr+CmJhG61VVBKoxqzsrA1vk+TB4ulN8aW3f2rpsWPHlKvOXfdBKkplwCuvDcLktcfl5i/oVWNU63pOe6byPGbRHtCOozw2WQ8= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1730277226; c=relaxed/simple; bh=a9JRVxGCUBas3AV4K19cNFO6TL9a3VpgMsm74fpcCgA=; h=From:To:CC:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version:Content-Type; b=Vha3QPzmjLc4Joefss2HJnNjpa5pamu34mDw+/w4rmOVPYSoRpbiL7ctb7LKX1YbF2zgKGpC3fqIxn0sSgONd49xG+5vHSs4c18xZLZ4QDAB+0TEHZy7OFZsBc+aQl2M7LRhgprbSOZ3q0GinOpXQyxoqVSSkP7CBej01j/5DKg= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=huawei-partners.com; spf=pass smtp.mailfrom=huawei-partners.com; arc=none smtp.client-ip=185.176.79.56 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=huawei-partners.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=huawei-partners.com Received: from mail.maildlp.com (unknown [172.18.186.216]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdgJL0NLLz6GFtv; Wed, 30 Oct 2024 16:28:50 +0800 (CST) Received: from mscpeml500003.china.huawei.com (unknown [7.188.49.51]) by mail.maildlp.com (Postfix) with ESMTPS id D731A140AA7; Wed, 30 Oct 2024 16:33:40 +0800 (CST) Received: from mscphis01197.huawei.com (10.123.65.218) by mscpeml500003.china.huawei.com (7.188.49.51) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.2.1258.34; Wed, 30 Oct 2024 11:33:40 +0300 From: To: , , , , , , , , , , CC: , , , , , , , , , , , Subject: [RFC PATCH 1/3] mm: Add thp_flags control for cgroup Date: Wed, 30 Oct 2024 16:33:09 +0800 Message-ID: <20241030083311.965933-2-gutierrez.asier@huawei-partners.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20241030083311.965933-1-gutierrez.asier@huawei-partners.com> References: <20241030083311.965933-1-gutierrez.asier@huawei-partners.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable X-ClientProxiedBy: mscpeml500003.china.huawei.com (7.188.49.51) To mscpeml500003.china.huawei.com (7.188.49.51) Content-Type: text/plain; charset="utf-8" From: Asier Gutierrez Exposed a new file in memory cgroup called memory.thp_enabled. This file wo= rks in the same way and same format as thp settings in /sys/kernel/mm/transparent_hugepage/enabled. The patch allows to read from = and write to that file, changing effectively the memory cgroup THP policy. New cgroups will inherit the THP policies from their parents. Signed-off-by: Asier Gutierrez Signed-off-by: Anatoly Stepanov Reviewed-by: Alexander Kozhevnikov --- include/linux/huge_mm.h | 5 +++ include/linux/memcontrol.h | 15 +++++++ mm/huge_memory.c | 71 ++++++++++++++++++++----------- mm/memcontrol.c | 86 ++++++++++++++++++++++++++++++++++++++ 4 files changed, 153 insertions(+), 24 deletions(-) diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index e25d9ebfdf89..86c0fb4c0b28 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -53,6 +53,9 @@ enum transparent_hugepage_flag { TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG, }; =20 +#define HUGEPAGE_FLAGS_ENABLED_MASK ((1UL << TRANSPARENT_HUGEPAGE_FLAG) |\ + (1UL << TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)) + struct kobject; struct kobj_attribute; =20 @@ -430,6 +433,8 @@ void split_huge_pmd_locked(struct vm_area_struct *vma, = unsigned long address, bool unmap_huge_pmd_locked(struct vm_area_struct *vma, unsigned long addr, pmd_t *pmdp, struct folio *folio); =20 +int thp_enabled_parse(const char *buf, unsigned long *flags); +const char *thp_enabled_string(unsigned long flags); #else /* CONFIG_TRANSPARENT_HUGEPAGE */ =20 static inline bool folio_test_pmd_mappable(struct folio *folio) diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h index 0e5bf25d324f..87b5fe93e19d 100644 --- a/include/linux/memcontrol.h +++ b/include/linux/memcontrol.h @@ -315,6 +315,12 @@ struct mem_cgroup { spinlock_t event_list_lock; #endif /* CONFIG_MEMCG_V1 */ =20 +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + unsigned long thp_flags; + unsigned long thp_anon_orders_always; + unsigned long thp_anon_orders_madvise; + unsigned long thp_anon_orders_inherit; +#endif struct mem_cgroup_per_node *nodeinfo[]; }; =20 @@ -1615,6 +1621,15 @@ struct sock; bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pag= es, gfp_t gfp_mask); void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_p= ages); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +int memory_thp_enabled_show(struct seq_file *m, void *v); +ssize_t memory_thp_enabled_write(struct kernfs_open_file *of, char *buf, + size_t nbytes, loff_t off); + +int mem_cgroup_thp_flags_update_all(unsigned long flags, unsigned long mas= k); +unsigned long memcg_get_thp_flags_all(unsigned long mask); +unsigned long memcg_get_thp_flags(struct vm_area_struct *vma); +#endif #ifdef CONFIG_MEMCG extern struct static_key_false memcg_sockets_enabled_key; #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_e= nabled_key) diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 67c86a5d64a6..0fbdd8213443 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -46,6 +46,8 @@ #include "internal.h" #include "swap.h" =20 +#include + #define CREATE_TRACE_POINTS #include =20 @@ -287,21 +289,43 @@ static unsigned long shrink_huge_zero_page_scan(struc= t shrinker *shrink, =20 static struct shrinker *huge_zero_page_shrinker; =20 -#ifdef CONFIG_SYSFS -static ssize_t enabled_show(struct kobject *kobj, - struct kobj_attribute *attr, char *buf) +const char *thp_enabled_string(unsigned long flags) { const char *output; =20 - if (test_bit(TRANSPARENT_HUGEPAGE_FLAG, &transparent_hugepage_flags)) + if (test_bit(TRANSPARENT_HUGEPAGE_FLAG, &flags)) output =3D "[always] madvise never"; - else if (test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, - &transparent_hugepage_flags)) + else if (test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &flags)) output =3D "always [madvise] never"; else output =3D "always madvise [never]"; =20 - return sysfs_emit(buf, "%s\n", output); + return output; +} + +int thp_enabled_parse(const char *buf, unsigned long *flags) +{ + if (sysfs_streq(buf, "always")) { + clear_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, flags); + set_bit(TRANSPARENT_HUGEPAGE_FLAG, flags); + } else if (sysfs_streq(buf, "madvise")) { + clear_bit(TRANSPARENT_HUGEPAGE_FLAG, flags); + set_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, flags); + } else if (sysfs_streq(buf, "never")) { + clear_bit(TRANSPARENT_HUGEPAGE_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, flags); + } else + return -EINVAL; + + return 0; +} + +#ifdef CONFIG_SYSFS +static ssize_t enabled_show(struct kobject *kobj, + struct kobj_attribute *attr, char *buf) +{ + unsigned long flags =3D transparent_hugepage_flags; + return sysfs_emit(buf, "%s\n", thp_enabled_string(flags)); } =20 static ssize_t enabled_store(struct kobject *kobj, @@ -309,24 +333,21 @@ static ssize_t enabled_store(struct kobject *kobj, const char *buf, size_t count) { ssize_t ret =3D count; + int err; =20 - if (sysfs_streq(buf, "always")) { - clear_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &transparent_hugepage_flag= s); - set_bit(TRANSPARENT_HUGEPAGE_FLAG, &transparent_hugepage_flags); - } else if (sysfs_streq(buf, "madvise")) { - clear_bit(TRANSPARENT_HUGEPAGE_FLAG, &transparent_hugepage_flags); - set_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &transparent_hugepage_flags); - } else if (sysfs_streq(buf, "never")) { - clear_bit(TRANSPARENT_HUGEPAGE_FLAG, &transparent_hugepage_flags); - clear_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &transparent_hugepage_flag= s); - } else - ret =3D -EINVAL; + ret =3D thp_enabled_parse(buf, &transparent_hugepage_flags) ? : count; + if (ret <=3D 0) + goto out; =20 - if (ret > 0) { - int err =3D start_stop_khugepaged(); - if (err) - ret =3D err; - } + if (IS_ENABLED(CONFIG_MEMCG) && !mem_cgroup_disabled()) + err =3D mem_cgroup_thp_flags_update_all(transparent_hugepage_flags, + HUGEPAGE_FLAGS_ENABLED_MASK); + else + err =3D start_stop_khugepaged(); + + if (err) + ret =3D err; +out: return ret; } =20 @@ -1036,7 +1057,9 @@ static vm_fault_t __do_huge_pmd_anonymous_page(struct= vm_fault *vmf, gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma) { const bool vma_madvised =3D vma && (vma->vm_flags & VM_HUGEPAGE); - +#ifdef CONFIG_MEMCG + unsigned long transparent_hugepage_flags =3D memcg_get_thp_flags(vma); +#endif /* Always do synchronous compaction */ if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepa= ge_flags)) return GFP_TRANSHUGE | (vma_madvised ? 0 : __GFP_NORETRY); diff --git a/mm/memcontrol.c b/mm/memcontrol.c index d563fb515766..2b25c45c85c3 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -970,6 +970,33 @@ struct mem_cgroup *get_mem_cgroup_from_current(void) return memcg; } =20 +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +static inline bool memcg_thp_always_enabled(struct mem_cgroup *memcg) +{ + return test_bit(TRANSPARENT_HUGEPAGE_FLAG, &memcg->thp_flags); +} + +static inline bool memcg_thp_madvise_enabled(struct mem_cgroup *memcg) +{ + return test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &memcg->thp_flags); +} + +unsigned long memcg_get_thp_flags(struct vm_area_struct *vma) +{ + unsigned long flags =3D 0UL; + struct mem_cgroup *memcg =3D get_mem_cgroup_from_mm(vma->vm_mm); + + if (!memcg) + goto out; + + flags =3D READ_ONCE(memcg->thp_flags); +out: + if (memcg) + css_put(&memcg->css); + return flags; +} +#endif + /** * mem_cgroup_iter - iterate over memory cgroup hierarchy * @root: hierarchy root @@ -3625,6 +3652,11 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *par= ent_css) WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable)); page_counter_init(&memcg->kmem, &parent->kmem); page_counter_init(&memcg->tcpmem, &parent->tcpmem); +#endif +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + WRITE_ONCE(memcg->thp_flags, READ_ONCE(parent->thp_flags)); + WRITE_ONCE(memcg->thp_anon_orders_inherit, + READ_ONCE(parent->thp_anon_orders_inherit)); #endif } else { init_memcg_stats(); @@ -3634,6 +3666,17 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *par= ent_css) #ifdef CONFIG_MEMCG_V1 page_counter_init(&memcg->kmem, NULL); page_counter_init(&memcg->tcpmem, NULL); +#endif +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + WRITE_ONCE(memcg->thp_flags, +#ifdef CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS + (1<thp_anon_orders_inherit, BIT(PMD_ORDER)); #endif root_mem_cgroup =3D memcg; return &memcg->css; @@ -4315,6 +4358,19 @@ static ssize_t memory_reclaim(struct kernfs_open_fil= e *of, char *buf, return nbytes; } =20 +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +DEFINE_MUTEX(memcg_thp_flags_mutex); + +int memory_thp_enabled_show(struct seq_file *m, void *v) +{ + struct mem_cgroup *memcg =3D mem_cgroup_from_seq(m); + unsigned long flags =3D READ_ONCE(memcg->thp_flags); + + seq_printf(m, "%s\n", thp_enabled_string(flags)); + return 0; +} +#endif + static struct cftype memory_files[] =3D { { .name =3D "current", @@ -4383,6 +4439,12 @@ static struct cftype memory_files[] =3D { .flags =3D CFTYPE_NS_DELEGATABLE, .write =3D memory_reclaim, }, +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + { + .name =3D "thp_enabled", + .seq_show =3D memory_thp_enabled_show, + }, +#endif { } /* terminate */ }; =20 @@ -4844,6 +4906,30 @@ void mem_cgroup_uncharge_skmem(struct mem_cgroup *me= mcg, unsigned int nr_pages) refill_stock(memcg, nr_pages); } =20 +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +int mem_cgroup_thp_flags_update_all(unsigned long new_flags, unsigned long= mask) +{ + int ret =3D 0; + struct mem_cgroup *iter, *memcg =3D root_mem_cgroup; + unsigned long enabled_mask =3D + (1UL << TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG) | + (1UL << TRANSPARENT_HUGEPAGE_FLAG); + + mutex_lock(&memcg_thp_flags_mutex); + enabled_mask &=3D new_flags; + + for_each_mem_cgroup_tree(iter, memcg) { + unsigned long old_flags =3D iter->thp_flags; + + iter->thp_flags =3D (old_flags & ~mask) | new_flags; + } + + mutex_unlock(&memcg_thp_flags_mutex); + return ret; +} + +#endif + static int __init cgroup_memory(char *s) { char *token; --=20 2.34.1 From nobody Mon Nov 25 02:51:59 2024 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 118AA1E0E0E; Wed, 30 Oct 2024 08:33:43 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=185.176.79.56 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1730277227; cv=none; b=gbF3qUizDX25b10FuGrXTOH1ZNyIXKRxFViz/dMjUcGLcj9NakgO9D7Y9AA4ebyLXfQc8K5+w4RC0yoAm9wqeXyeMFxf9YzbL2TzmyEgAAXJPbUBMycImnKV5UedS232mv/KNFdB2PW4w6rwC+rJ+7snNhXzF2Pbj2ozM/g0EHM= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1730277227; c=relaxed/simple; bh=ZwYlA8wjYksRut5A2NARwGJBHxFeD34uTT4ncvmLdEs=; h=From:To:CC:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version:Content-Type; b=O2nTkGuKTEmozTfgEE9oV8nBngZ0wabEMLZ5j3sJR3Omjq0irqrfqc/2SC6j+RuU90wc/PCacHtXrauT+yU5y24q2LAutOA7iktHdgNh+7iYk+FdSadu1Y7SSC8EvnlFX14XZWjepItpkJd0sA4aGg7wRfM4PoEcd/ueRBuh8bw= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=huawei-partners.com; spf=pass smtp.mailfrom=huawei-partners.com; arc=none smtp.client-ip=185.176.79.56 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=huawei-partners.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=huawei-partners.com Received: from mail.maildlp.com (unknown [172.18.186.216]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdgNR2dfdz6D95c; Wed, 30 Oct 2024 16:32:23 +0800 (CST) Received: from mscpeml500003.china.huawei.com (unknown [7.188.49.51]) by mail.maildlp.com (Postfix) with ESMTPS id ED54F140C72; Wed, 30 Oct 2024 16:33:40 +0800 (CST) Received: from mscphis01197.huawei.com (10.123.65.218) by mscpeml500003.china.huawei.com (7.188.49.51) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.2.1258.34; Wed, 30 Oct 2024 11:33:40 +0300 From: To: , , , , , , , , , , CC: , , , , , , , , , , , Subject: [RFC PATCH 2/3] mm: Support for huge pages in cgroups Date: Wed, 30 Oct 2024 16:33:10 +0800 Message-ID: <20241030083311.965933-3-gutierrez.asier@huawei-partners.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20241030083311.965933-1-gutierrez.asier@huawei-partners.com> References: <20241030083311.965933-1-gutierrez.asier@huawei-partners.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable X-ClientProxiedBy: mscpeml500003.china.huawei.com (7.188.49.51) To mscpeml500003.china.huawei.com (7.188.49.51) Content-Type: text/plain; charset="utf-8" From: Asier Gutierrez This patch adds support for the correct order mask depending on the memory = THP policy. Also, khugepaged lazy collpasing and kernel boot parameter THP over= ride were added. Signed-off-by: Asier Gutierrez Signed-off-by: Anatoly Stepanov Reviewed-by: Alexander Kozhevnikov --- include/linux/huge_mm.h | 10 ++- include/linux/khugepaged.h | 2 +- include/linux/memcontrol.h | 11 +++ mm/huge_memory.c | 22 ++++-- mm/khugepaged.c | 8 +- mm/memcontrol.c | 147 ++++++++++++++++++++++++++++++++++++- 6 files changed, 187 insertions(+), 13 deletions(-) diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index 86c0fb4c0b28..f99ac9b7e5bc 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -207,6 +207,12 @@ static inline unsigned long thp_vma_suitable_orders(st= ruct vm_area_struct *vma, return orders; } =20 +#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_s= truct *vma, if (hugepage_global_always() || ((vm_flags & VM_HUGEPAGE) && hugepage_global_enabled())) mask |=3D 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 &=3D mask; if (!orders) return 0; diff --git a/include/linux/khugepaged.h b/include/linux/khugepaged.h index f68865e19b0b..50cabca48b93 100644 --- a/include/linux/khugepaged.h +++ b/include/linux/khugepaged.h @@ -9,7 +9,7 @@ extern struct attribute_group khugepaged_attr_group; =20 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, diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h index 87b5fe93e19d..d78318782af8 100644 --- a/include/linux/memcontrol.h +++ b/include/linux/memcontrol.h @@ -23,6 +23,7 @@ #include #include #include +#include =20 struct mem_cgroup; struct obj_cgroup; @@ -1069,6 +1070,9 @@ static inline void memcg_memory_event_mm(struct mm_st= ruct *mm, =20 void split_page_memcg(struct page *head, int old_order, int new_order); =20 +bool memcg_thp_vma_allowable_orders(struct vm_area_struct *vma, + unsigned long vm_flags, + unsigned long *res_mask); #else /* CONFIG_MEMCG */ =20 #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 *v= ma, + unsigned long vm_flags, + unsigned long *res_mask) +{ + return false; +} #endif /* CONFIG_MEMCG */ =20 /* diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 0fbdd8213443..fdffdfc8605c 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -172,15 +172,23 @@ unsigned long __thp_vma_allowable_orders(struct vm_ar= ea_struct *vma, } =20 if (!vma_is_anonymous(vma)) { + bool memcg_enabled =3D 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 =3D 0UL; + + memcg_enabled =3D memcg_thp_vma_allowable_orders(vma, vm_flags, &mask); =20 + 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 =3D mem_cgroup_thp_flags_update_all(transparent_hugepage_flags, HUGEPAGE_FLAGS_ENABLED_MASK); else - err =3D start_stop_khugepaged(); + err =3D start_stop_khugepaged(false); =20 if (err) ret =3D err; @@ -539,7 +547,7 @@ static ssize_t thpsize_enabled_store(struct kobject *ko= bj, if (ret > 0) { int err; =20 - err =3D start_stop_khugepaged(); + err =3D start_stop_khugepaged(false); if (err) ret =3D err; } @@ -812,7 +820,7 @@ static int __init hugepage_init(void) return 0; } =20 - err =3D start_stop_khugepaged(); + err =3D start_stop_khugepaged(false); if (err) goto err_khugepaged; =20 diff --git a/mm/khugepaged.c b/mm/khugepaged.c index cdd1d8655a76..ebed9bf8cfb5 100644 --- a/mm/khugepaged.c +++ b/mm/khugepaged.c @@ -415,6 +415,8 @@ static inline int hpage_collapse_test_exit_or_disable(s= truct mm_struct *mm) =20 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 =3D 0; unsigned long recommended_min; =20 - 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(); } =20 -int start_stop_khugepaged(void) +int start_stop_khugepaged(bool force_stop) { int err =3D 0; =20 mutex_lock(&khugepaged_mutex); - if (hugepage_pmd_enabled()) { + if (hugepage_pmd_enabled() && !force_stop) { if (!khugepaged_thread) khugepaged_thread =3D kthread_run(khugepaged, NULL, "khugepaged"); diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 2b25c45c85c3..938e6894c0b3 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -981,6 +981,37 @@ static inline bool memcg_thp_madvise_enabled(struct me= m_cgroup *memcg) return test_bit(TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG, &memcg->thp_flags); } =20 +bool memcg_thp_vma_allowable_orders(struct vm_area_struct *vma, + unsigned long vm_flags, + unsigned long *res_mask) +{ + unsigned long mask =3D 0UL; + + struct mem_cgroup *memcg =3D 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 =3D THP_ORDERS_ALL_FILE_DEFAULT; + else { + mask =3D READ_ONCE(memcg->thp_anon_orders_always); + + if (vm_flags & VM_HUGEPAGE) + mask |=3D READ_ONCE(memcg->thp_anon_orders_madvise); + + mask =3D mask | READ_ONCE(memcg->thp_anon_orders_inherit); + } + } + + css_put(&memcg->css); + *res_mask =3D mask; + return true; +} + unsigned long memcg_get_thp_flags(struct vm_area_struct *vma) { unsigned long flags =3D 0UL; @@ -3986,10 +4017,52 @@ static void mem_cgroup_kmem_attach(struct cgroup_ta= skset *tset) } } =20 +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +static int mem_cgroup_notify_khugepaged_cb(struct task_struct *p, void *ar= g) +{ + struct vm_area_struct *vma =3D NULL; + struct mem_cgroup *memcg =3D arg; + bool is_madvise =3D memcg_thp_madvise_enabled(memcg); + bool is_always =3D 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); } =20 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 =3D 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 =3D css_task_iter_next(&it))) { + if (!task->mm || (task->flags & PF_KTHREAD)) + continue; + + ret =3D 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 =3D nbytes; + struct mem_cgroup *memcg =3D mem_cgroup_from_css(of_css(of)); + + buf =3D strstrip(buf); + + mutex_lock(&memcg_thp_flags_mutex); + ret =3D thp_enabled_parse(buf, &memcg->thp_flags) ? : nbytes; + if (ret > 0) { + err =3D mem_cgroup_notify_khugepaged(memcg); + if (!err) + err =3D start_stop_khugepaged(false); + if (err) + ret =3D err; + } + mutex_unlock(&memcg_thp_flags_mutex); + return ret; +} #endif =20 static struct cftype memory_files[] =3D { @@ -4443,6 +4564,7 @@ static struct cftype memory_files[] =3D { { .name =3D "thp_enabled", .seq_show =3D memory_thp_enabled_show, + .write =3D memory_thp_enabled_write, }, #endif { } /* terminate */ @@ -4909,7 +5031,9 @@ void mem_cgroup_uncharge_skmem(struct mem_cgroup *mem= cg, unsigned int nr_pages) #ifdef CONFIG_TRANSPARENT_HUGEPAGE int mem_cgroup_thp_flags_update_all(unsigned long new_flags, unsigned long= mask) { - int ret =3D 0; + int ret; + struct css_task_iter task_iter; + struct task_struct *task; struct mem_cgroup *iter, *memcg =3D root_mem_cgroup; unsigned long enabled_mask =3D (1UL << TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG) | @@ -4922,8 +5046,18 @@ int mem_cgroup_thp_flags_update_all(unsigned long ne= w_flags, unsigned long mask) unsigned long old_flags =3D iter->thp_flags; =20 iter->thp_flags =3D (old_flags & ~mask) | new_flags; + + css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &task_iter); + while ((task =3D 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); } =20 + ret =3D 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); =20 + +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +static int __init mem_cgroup_thp_init(void) +{ + if (IS_ENABLED(CONFIG_MEMCG)) + root_mem_cgroup->thp_flags =3D transparent_hugepage_flags; + + return 0; +} +subsys_initcall(mem_cgroup_thp_init); +#endif #endif /* CONFIG_SWAP */ --=20 2.34.1 From nobody Mon Nov 25 02:51:59 2024 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 8C3FC1E0E0B; Wed, 30 Oct 2024 08:33:43 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=185.176.79.56 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1730277227; cv=none; b=NRRvNmQzIDXb7iS7CMnVNKnrWX9ljWcmnnYZw3S2W9LhuqA0a+MTanIdStlncBwK8v3dK1n1792AyBgYWcniD3rPJRXodnyyuzsi+LqqjzFOGlda2V8ZAbNq1pI7MB//JVClvlU2iHadm0hH3nljcEM+9xXTPsNE94Tc3lUid3k= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1730277227; c=relaxed/simple; bh=P+Fy4mbrUnx5m73aSP5REkcuhFBJjsHIiadHOOPubCM=; h=From:To:CC:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version:Content-Type; b=U5PdwVSQxajQPgsRDyiJfwaMNHt9TWqtKEUMJeMf6oO3lLA1QaUPmw5Z6Pa7DpZMWfF1R8WHuSvhrBX0n41G+2b0Sa4Cmts/ohilwSMHCXJs1dpveRxKzu+k6Tky4Sa6acARwK38u7S/pPUq+qTOxUVfoAXyF79+F5sPwDwBTP8= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=huawei-partners.com; spf=pass smtp.mailfrom=huawei-partners.com; arc=none smtp.client-ip=185.176.79.56 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=huawei-partners.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=huawei-partners.com Received: from mail.maildlp.com (unknown [172.18.186.216]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4XdgJL2H7Pz6GFtq; Wed, 30 Oct 2024 16:28:50 +0800 (CST) Received: from mscpeml500003.china.huawei.com (unknown [7.188.49.51]) by mail.maildlp.com (Postfix) with ESMTPS id 0FB3A140CB9; Wed, 30 Oct 2024 16:33:41 +0800 (CST) Received: from mscphis01197.huawei.com (10.123.65.218) by mscpeml500003.china.huawei.com (7.188.49.51) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.2.1258.34; Wed, 30 Oct 2024 11:33:40 +0300 From: To: , , , , , , , , , , CC: , , , , , , , , , , , Subject: [RFC PATCH 3/3] mm: Add thp_defrag control for cgroup Date: Wed, 30 Oct 2024 16:33:11 +0800 Message-ID: <20241030083311.965933-4-gutierrez.asier@huawei-partners.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20241030083311.965933-1-gutierrez.asier@huawei-partners.com> References: <20241030083311.965933-1-gutierrez.asier@huawei-partners.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable X-ClientProxiedBy: mscpeml500003.china.huawei.com (7.188.49.51) To mscpeml500003.china.huawei.com (7.188.49.51) Content-Type: text/plain; charset="utf-8" From: Asier Gutierrez This patch exposes a new file in memory cgroups: memory.thp_defrag, which follows the /sys/kernel/mm/transparent_hugepage/defrag style. Support for different defrag THP defrag policies for memory cgroups were also added. Signed-off-by: Asier Gutierrez Signed-off-by: Anatoly Stepanov Reviewed-by: Alexander Kozhevnikov --- include/linux/huge_mm.h | 8 +++ include/linux/memcontrol.h | 4 +- mm/huge_memory.c | 116 ++++++++++++++++++++++--------------- mm/memcontrol.c | 31 ++++++++++ 4 files changed, 112 insertions(+), 47 deletions(-) diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index f99ac9b7e5bc..177c7d3578ed 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -56,6 +56,12 @@ enum transparent_hugepage_flag { #define HUGEPAGE_FLAGS_ENABLED_MASK ((1UL << TRANSPARENT_HUGEPAGE_FLAG) |\ (1UL << TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)) =20 +#define HUGEPAGE_FLAGS_DEFRAG_MASK ((1UL << TRANSPARENT_HUGEPAGE_DEFRAG_DI= RECT_FLAG) |\ + (1UL << TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG) |\ + (1UL << TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG) |\ + (1UL << TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG) |\ + (1UL << TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG)) + struct kobject; struct kobj_attribute; =20 @@ -442,7 +448,9 @@ bool unmap_huge_pmd_locked(struct vm_area_struct *vma, = unsigned long addr, pmd_t *pmdp, struct folio *folio); =20 int thp_enabled_parse(const char *buf, unsigned long *flags); +int thp_defrag_parse(const char *buf, unsigned long *flags); const char *thp_enabled_string(unsigned long flags); +const char *thp_defrag_string(unsigned long flags); #else /* CONFIG_TRANSPARENT_HUGEPAGE */ =20 static inline bool folio_test_pmd_mappable(struct folio *folio) diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h index d78318782af8..a0edf15b3a07 100644 --- a/include/linux/memcontrol.h +++ b/include/linux/memcontrol.h @@ -1634,9 +1634,11 @@ bool mem_cgroup_charge_skmem(struct mem_cgroup *memc= g, unsigned int nr_pages, void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_p= ages); #ifdef CONFIG_TRANSPARENT_HUGEPAGE int memory_thp_enabled_show(struct seq_file *m, void *v); +int memory_thp_defrag_show(struct seq_file *m, void *v); ssize_t memory_thp_enabled_write(struct kernfs_open_file *of, char *buf, size_t nbytes, loff_t off); - +ssize_t memory_thp_defrag_write(struct kernfs_open_file *of, char *buf, + size_t nbytes, loff_t off); int mem_cgroup_thp_flags_update_all(unsigned long flags, unsigned long mas= k); unsigned long memcg_get_thp_flags_all(unsigned long mask); unsigned long memcg_get_thp_flags(struct vm_area_struct *vma); diff --git a/mm/huge_memory.c b/mm/huge_memory.c index fdffdfc8605c..6e1886b220d9 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -311,6 +311,28 @@ const char *thp_enabled_string(unsigned long flags) return output; } =20 +const char *thp_defrag_string(unsigned long flags) +{ + const char *output; + + if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, + &flags)) + output =3D "[always] defer defer+madvise madvise never"; + else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, + &flags)) + output =3D "always [defer] defer+madvise madvise never"; + else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, + &flags)) + output =3D "always defer [defer+madvise] madvise never"; + else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, + &flags)) + output =3D "always defer defer+madvise [madvise] never"; + else + output =3D "always defer defer+madvise madvise [never]"; + + return output; +} + int thp_enabled_parse(const char *buf, unsigned long *flags) { if (sysfs_streq(buf, "always")) { @@ -328,6 +350,39 @@ int thp_enabled_parse(const char *buf, unsigned long *= flags) return 0; } =20 +int thp_defrag_parse(const char *buf, unsigned long *flags) +{ + if (sysfs_streq(buf, "always")) { + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, flags); + set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, flags); + } else if (sysfs_streq(buf, "defer+madvise")) { + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, flags); + set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, flags); + } else if (sysfs_streq(buf, "defer")) { + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, flags); + set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, flags); + } else if (sysfs_streq(buf, "madvise")) { + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, flags); + set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, flags); + } else if (sysfs_streq(buf, "never")) { + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, flags); + clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, flags); + } else + return -EINVAL; + + return 0; +} + #ifdef CONFIG_SYSFS static ssize_t enabled_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf) @@ -394,60 +449,29 @@ ssize_t single_hugepage_flag_store(struct kobject *ko= bj, static ssize_t defrag_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf) { - const char *output; - - if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, - &transparent_hugepage_flags)) - output =3D "[always] defer defer+madvise madvise never"; - else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, - &transparent_hugepage_flags)) - output =3D "always [defer] defer+madvise madvise never"; - else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, - &transparent_hugepage_flags)) - output =3D "always defer [defer+madvise] madvise never"; - else if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, - &transparent_hugepage_flags)) - output =3D "always defer defer+madvise [madvise] never"; - else - output =3D "always defer defer+madvise madvise [never]"; - - return sysfs_emit(buf, "%s\n", output); + unsigned long flags =3D transparent_hugepage_flags; + return sysfs_emit(buf, "%s\n", thp_defrag_string(flags)); } =20 static ssize_t defrag_store(struct kobject *kobj, struct kobj_attribute *attr, const char *buf, size_t count) { - if (sysfs_streq(buf, "always")) { - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, &transparent_= hugepage_flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, &transparent_hugepa= ge_flags); - set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage_f= lags); - } else if (sysfs_streq(buf, "defer+madvise")) { - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, &transparent_hugepa= ge_flags); - set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, &transparent_hu= gepage_flags); - } else if (sysfs_streq(buf, "defer")) { - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, &transparent_= hugepage_flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, &transparent_hugepa= ge_flags); - set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, &transparent_hugepage_f= lags); - } else if (sysfs_streq(buf, "madvise")) { - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, &transparent_= hugepage_flags); - set_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, &transparent_hugepage= _flags); - } else if (sysfs_streq(buf, "never")) { - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_FLAG, &transparent_hugepage= _flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_KSWAPD_OR_MADV_FLAG, &transparent_= hugepage_flags); - clear_bit(TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG, &transparent_hugepa= ge_flags); - } else - return -EINVAL; + ssize_t ret =3D count; + int err; =20 - return count; + ret =3D thp_defrag_parse(buf, &transparent_hugepage_flags) ? : count; + if (ret > 0 && IS_ENABLED(CONFIG_MEMCG) && + !mem_cgroup_disabled()) { + err =3D mem_cgroup_thp_flags_update_all(transparent_hugepage_flags, + HUGEPAGE_FLAGS_DEFRAG_MASK); + if (err) + ret =3D err; + } + + return ret; } + static struct kobj_attribute defrag_attr =3D __ATTR_RW(defrag); =20 static ssize_t use_zero_page_show(struct kobject *kobj, diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 938e6894c0b3..53384f0a69af 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -3706,6 +3706,8 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *pare= nt_css) #ifdef CONFIG_TRANSPARENT_HUGEPAGE_MADVISE (1<thp_anon_orders_inherit, BIT(PMD_ORDER)); #endif @@ -4490,6 +4492,30 @@ ssize_t memory_thp_enabled_write(struct kernfs_open_= file *of, char *buf, mutex_unlock(&memcg_thp_flags_mutex); return ret; } + +int memory_thp_defrag_show(struct seq_file *m, void *v) +{ + struct mem_cgroup *memcg =3D mem_cgroup_from_seq(m); + unsigned long flags =3D READ_ONCE(memcg->thp_flags); + + seq_printf(m, "%s\n", thp_defrag_string(flags)); + return 0; +} + +ssize_t memory_thp_defrag_write(struct kernfs_open_file *of, char *buf, + size_t nbytes, loff_t off) +{ + int ret =3D nbytes; + struct mem_cgroup *memcg =3D mem_cgroup_from_css(of_css(of)); + + buf =3D strstrip(buf); + + mutex_lock(&memcg_thp_flags_mutex); + ret =3D thp_defrag_parse(buf, &memcg->thp_flags) ? : nbytes; + mutex_unlock(&memcg_thp_flags_mutex); + + return ret; +} #endif =20 static struct cftype memory_files[] =3D { @@ -4566,6 +4592,11 @@ static struct cftype memory_files[] =3D { .seq_show =3D memory_thp_enabled_show, .write =3D memory_thp_enabled_write, }, + { + .name =3D "thp_defrag", + .seq_show =3D memory_thp_defrag_show, + .write =3D memory_thp_defrag_write, + }, #endif { } /* terminate */ }; --=20 2.34.1