| Message ID | 20201001002710.748166-5-guro@fb.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | mm: allow mapping accounted kernel pages to userspace | expand |
On Wed, Sep 30, 2020 at 05:27:10PM -0700, Roman Gushchin wrote: > PageKmemcg flag is currently defined as a page type (like buddy, > offline, table and guard). Semantically it means that the page > was accounted as a kernel memory by the page allocator and has > to be uncharged on the release. > > As a side effect of defining the flag as a page type, the accounted > page can't be mapped to userspace (look at page_has_type() and > comments above). In particular, this blocks the accounting of > vmalloc-backed memory used by some bpf maps, because these maps > do map the memory to userspace. > > One option is to fix it by complicating the access to page->mapcount, > which provides some free bits for page->page_type. > > But it's way better to move this flag into page->memcg_data flags. > Indeed, the flag makes no sense without enabled memory cgroups > and memory cgroup pointer set in particular. > > This commit replaces PageKmemcg() and __SetPageKmemcg() with > PageMemcgKmem() and an open-coded OR operation setting the memcg > pointer with the MEMCG_DATA_KMEM bit. __ClearPageKmemcg() can be > simple deleted, as the whole memcg_data is zeroed at once. > > As a bonus, on !CONFIG_MEMCG build the PageMemcgKmem() check will > be compiled out. > > Signed-off-by: Roman Gushchin <guro@fb.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Hi. On Wed, Sep 30, 2020 at 05:27:10PM -0700, Roman Gushchin <guro@fb.com> wrote: > @@ -369,8 +371,12 @@ enum page_memcg_data_flags { > */ > static inline struct mem_cgroup *page_memcg(struct page *page) > { > + unsigned long memcg_data = page->memcg_data; > + > VM_BUG_ON_PAGE(PageSlab(page), page); > - return (struct mem_cgroup *)page->memcg_data; > + VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); > + > + return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); > } Shouldn't this change go also into page_memcg_rcu()? (I don't think the current single user (workingset_activation() would pass a non-slab kernel page but for consistency sake.) Alternatively, I'm thinking why (in its single use) is there page_memcg_rcu() a separate function to page_memcg() (cross memcg page migration?). Regards, Michal
On Thu, Oct 01, 2020 at 07:00:36PM +0200, Michal Koutný wrote: > Hi. > > On Wed, Sep 30, 2020 at 05:27:10PM -0700, Roman Gushchin <guro@fb.com> wrote: > > @@ -369,8 +371,12 @@ enum page_memcg_data_flags { > > */ > > static inline struct mem_cgroup *page_memcg(struct page *page) > > { > > + unsigned long memcg_data = page->memcg_data; > > + > > VM_BUG_ON_PAGE(PageSlab(page), page); > > - return (struct mem_cgroup *)page->memcg_data; > > + VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); > > + > > + return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); > > } > Shouldn't this change go also into page_memcg_rcu()? (I don't think the > current single user (workingset_activation() would pass a non-slab > kernel page but for consistency sake.) +1 > Alternatively, I'm thinking why (in its single use) is there > page_memcg_rcu() a separate function to page_memcg() (cross memcg page > migration?). It goes back to commit 55779ec759ccc3c12b917b3712a7716e1140c652. The activation code is the only path where page migration is not excluded. Because unlike with page state statistics, we don't really mind a race when counting an activation event. I do think there is a bug, though: mem_cgroup_move_account() should use WRITE_ONCE() on page->mem_cgroup.
On Thu, Oct 01, 2020 at 07:00:36PM +0200, Michal Koutny wrote: > Hi. > > On Wed, Sep 30, 2020 at 05:27:10PM -0700, Roman Gushchin <guro@fb.com> wrote: > > @@ -369,8 +371,12 @@ enum page_memcg_data_flags { > > */ > > static inline struct mem_cgroup *page_memcg(struct page *page) > > { > > + unsigned long memcg_data = page->memcg_data; > > + > > VM_BUG_ON_PAGE(PageSlab(page), page); > > - return (struct mem_cgroup *)page->memcg_data; > > + VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); > > + > > + return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); > > } > Shouldn't this change go also into page_memcg_rcu()? (I don't think the > current single user (workingset_activation() would pass a non-slab > kernel page but for consistency sake.) Yes, a good idea, I will add it. Thanks!
On Thu, Oct 01, 2020 at 01:27:13PM -0400, Johannes Weiner <hannes@cmpxchg.org> wrote: > The activation code is the only path where page migration is not > excluded. Because unlike with page state statistics, we don't really > mind a race when counting an activation event. Thanks for the explanation. I see why the accessor trio is justified. > I do think there is a bug, though: mem_cgroup_move_account() should > use WRITE_ONCE() on page->mem_cgroup. If this were a bug, wouldn't be the proper approach rcu_assign_pointer()/rcu_dereference() pair? Michal
On Fri, Oct 02, 2020 at 12:03:50PM +0200, Michal Koutný wrote: > On Thu, Oct 01, 2020 at 01:27:13PM -0400, Johannes Weiner <hannes@cmpxchg.org> wrote: > > I do think there is a bug, though: mem_cgroup_move_account() should > > use WRITE_ONCE() on page->mem_cgroup. > If this were a bug, wouldn't be the proper approach > rcu_assign_pointer()/rcu_dereference() pair? Yes, you're right. I'll double check these code paths and follow up with a patch if necessary.
diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h index 5e05599e1f74..8c15978294b2 100644 --- a/include/linux/memcontrol.h +++ b/include/linux/memcontrol.h @@ -346,8 +346,10 @@ extern struct mem_cgroup *root_mem_cgroup; enum page_memcg_data_flags { /* page->memcg_data is a pointer to an objcgs vector */ MEMCG_DATA_OBJCGS = (1UL << 0), + /* page has been accounted as a non-slab kernel page */ + MEMCG_DATA_KMEM = (1UL << 1), /* the next bit after the last actual flag */ - __NR_MEMCG_DATA_FLAGS = (1UL << 1), + __NR_MEMCG_DATA_FLAGS = (1UL << 2), }; #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1) @@ -369,8 +371,12 @@ enum page_memcg_data_flags { */ static inline struct mem_cgroup *page_memcg(struct page *page) { + unsigned long memcg_data = page->memcg_data; + VM_BUG_ON_PAGE(PageSlab(page), page); - return (struct mem_cgroup *)page->memcg_data; + VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); + + return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); } /* @@ -416,7 +422,7 @@ static inline struct mem_cgroup *page_memcg_check(struct page *page) if (memcg_data & MEMCG_DATA_OBJCGS) return NULL; - return (struct mem_cgroup *)memcg_data; + return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); } /* @@ -438,6 +444,20 @@ static inline void set_page_memcg(struct page *page, struct mem_cgroup *memcg) } +/* + * PageMemcgKmem - check if the page has MemcgKmem flag set + * @page: a pointer to the page struct + * + * Checks if the page has MemcgKmem flag set. The caller must ensure that + * the page has an associated memory cgroup. It's not safe to call this function + * against some types of pages, e.g. slab pages. + */ +static inline bool PageMemcgKmem(struct page *page) +{ + VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page); + return page->memcg_data & MEMCG_DATA_KMEM; +} + #ifdef CONFIG_MEMCG_KMEM /* * page_objcgs - get the object cgroups vector associated with a page @@ -453,8 +473,8 @@ static inline struct obj_cgroup **page_objcgs(struct page *page) { unsigned long memcg_data = READ_ONCE(page->memcg_data); - VM_BUG_ON_PAGE(memcg_data && - !(memcg_data & MEMCG_DATA_OBJCGS), page); + VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page); + VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); } @@ -471,9 +491,11 @@ static inline struct obj_cgroup **page_objcgs_check(struct page *page) { unsigned long memcg_data = READ_ONCE(page->memcg_data); - if (memcg_data && (memcg_data & MEMCG_DATA_OBJCGS)) + if (memcg_data && (memcg_data & MEMCG_DATA_OBJCGS)) { + VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); return (struct obj_cgroup **) (memcg_data & ~MEMCG_DATA_FLAGS_MASK); + } return NULL; } @@ -1138,6 +1160,11 @@ static inline void set_page_memcg(struct page *page, { } +static inline bool PageMemcgKmem(struct page *page) +{ + return false; +} + static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) { return true; diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h index 4f6ba9379112..fc0e1bd48e73 100644 --- a/include/linux/page-flags.h +++ b/include/linux/page-flags.h @@ -715,9 +715,8 @@ PAGEFLAG_FALSE(DoubleMap) #define PAGE_MAPCOUNT_RESERVE -128 #define PG_buddy 0x00000080 #define PG_offline 0x00000100 -#define PG_kmemcg 0x00000200 -#define PG_table 0x00000400 -#define PG_guard 0x00000800 +#define PG_table 0x00000200 +#define PG_guard 0x00000400 #define PageType(page, flag) \ ((page->page_type & (PAGE_TYPE_BASE | flag)) == PAGE_TYPE_BASE) @@ -768,12 +767,6 @@ PAGE_TYPE_OPS(Buddy, buddy) */ PAGE_TYPE_OPS(Offline, offline) -/* - * If kmemcg is enabled, the buddy allocator will set PageKmemcg() on - * pages allocated with __GFP_ACCOUNT. It gets cleared on page free. - */ -PAGE_TYPE_OPS(Kmemcg, kmemcg) - /* * Marks pages in use as page tables. */ diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 7e690424a917..ff9bb3e8333a 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -3090,8 +3090,8 @@ int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) if (memcg && !mem_cgroup_is_root(memcg)) { ret = __memcg_kmem_charge(memcg, gfp, 1 << order); if (!ret) { - set_page_memcg(page, memcg); - __SetPageKmemcg(page); + page->memcg_data = (unsigned long)memcg | + MEMCG_DATA_KMEM; return 0; } css_put(&memcg->css); @@ -3116,10 +3116,6 @@ void __memcg_kmem_uncharge_page(struct page *page, int order) __memcg_kmem_uncharge(memcg, nr_pages); page->memcg_data = 0; css_put(&memcg->css); - - /* slab pages do not have PageKmemcg flag set */ - if (PageKmemcg(page)) - __ClearPageKmemcg(page); } static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) @@ -6863,12 +6859,10 @@ static void uncharge_page(struct page *page, struct uncharge_gather *ug) nr_pages = compound_nr(page); ug->nr_pages += nr_pages; - if (!PageKmemcg(page)) { - ug->pgpgout++; - } else { + if (PageMemcgKmem(page)) ug->nr_kmem += nr_pages; - __ClearPageKmemcg(page); - } + else + ug->pgpgout++; ug->dummy_page = page; page->memcg_data = 0; diff --git a/mm/page_alloc.c b/mm/page_alloc.c index e9f0fe4a143e..5ebd50183d93 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -1179,7 +1179,7 @@ static __always_inline bool free_pages_prepare(struct page *page, * Do not let hwpoison pages hit pcplists/buddy * Untie memcg state and reset page's owner */ - if (memcg_kmem_enabled() && PageKmemcg(page)) + if (memcg_kmem_enabled() && PageMemcgKmem(page)) __memcg_kmem_uncharge_page(page, order); reset_page_owner(page, order); return false; @@ -1209,7 +1209,7 @@ static __always_inline bool free_pages_prepare(struct page *page, } if (PageMappingFlags(page)) page->mapping = NULL; - if (memcg_kmem_enabled() && PageKmemcg(page)) + if (memcg_kmem_enabled() && PageMemcgKmem(page)) __memcg_kmem_uncharge_page(page, order); if (check_free) bad += check_free_page(page);
PageKmemcg flag is currently defined as a page type (like buddy, offline, table and guard). Semantically it means that the page was accounted as a kernel memory by the page allocator and has to be uncharged on the release. As a side effect of defining the flag as a page type, the accounted page can't be mapped to userspace (look at page_has_type() and comments above). In particular, this blocks the accounting of vmalloc-backed memory used by some bpf maps, because these maps do map the memory to userspace. One option is to fix it by complicating the access to page->mapcount, which provides some free bits for page->page_type. But it's way better to move this flag into page->memcg_data flags. Indeed, the flag makes no sense without enabled memory cgroups and memory cgroup pointer set in particular. This commit replaces PageKmemcg() and __SetPageKmemcg() with PageMemcgKmem() and an open-coded OR operation setting the memcg pointer with the MEMCG_DATA_KMEM bit. __ClearPageKmemcg() can be simple deleted, as the whole memcg_data is zeroed at once. As a bonus, on !CONFIG_MEMCG build the PageMemcgKmem() check will be compiled out. Signed-off-by: Roman Gushchin <guro@fb.com> --- include/linux/memcontrol.h | 39 ++++++++++++++++++++++++++++++++------ include/linux/page-flags.h | 11 ++--------- mm/memcontrol.c | 16 +++++----------- mm/page_alloc.c | 4 ++-- 4 files changed, 42 insertions(+), 28 deletions(-)