From 68ad4a3304335358f95a417f2a2b0c909e5119c4 Mon Sep 17 00:00:00 2001 From: "Uladzislau Rezki (Sony)" Date: Fri, 17 May 2019 14:31:31 -0700 Subject: mm/vmalloc.c: keep track of free blocks for vmap allocation Patch series "improve vmap allocation", v3. Objective --------- Please have a look for the description at: https://lkml.org/lkml/2018/10/19/786 but let me also summarize it a bit here as well. The current implementation has O(N) complexity. Requests with different permissive parameters can lead to long allocation time. When i say "long" i mean milliseconds. Description ----------- This approach organizes the KVA memory layout into free areas of the 1-ULONG_MAX range, i.e. an allocation is done over free areas lookups, instead of finding a hole between two busy blocks. It allows to have lower number of objects which represent the free space, therefore to have less fragmented memory allocator. Because free blocks are always as large as possible. It uses the augment tree where all free areas are sorted in ascending order of va->va_start address in pair with linked list that provides O(1) access to prev/next elements. Since the tree is augment, we also maintain the "subtree_max_size" of VA that reflects a maximum available free block in its left or right sub-tree. Knowing that, we can easily traversal toward the lowest (left most path) free area. Allocation: ~O(log(N)) complexity. It is sequential allocation method therefore tends to maximize locality. The search is done until a first suitable block is large enough to encompass the requested parameters. Bigger areas are split. I copy paste here the description of how the area is split, since i described it in https://lkml.org/lkml/2018/10/19/786 A free block can be split by three different ways. Their names are FL_FIT_TYPE, LE_FIT_TYPE/RE_FIT_TYPE and NE_FIT_TYPE, i.e. they correspond to how requested size and alignment fit to a free block. FL_FIT_TYPE - in this case a free block is just removed from the free list/tree because it fully fits. Comparing with current design there is an extra work with rb-tree updating. LE_FIT_TYPE/RE_FIT_TYPE - left/right edges fit. In this case what we do is just cutting a free block. It is as fast as a current design. Most of the vmalloc allocations just end up with this case, because the edge is always aligned to 1. NE_FIT_TYPE - Is much less common case. Basically it happens when requested size and alignment does not fit left nor right edges, i.e. it is between them. In this case during splitting we have to build a remaining left free area and place it back to the free list/tree. Comparing with current design there are two extra steps. First one is we have to allocate a new vmap_area structure. Second one we have to insert that remaining free block to the address sorted list/tree. In order to optimize a first case there is a cache with free_vmap objects. Instead of allocating from slab we just take an object from the cache and reuse it. Second one is pretty optimized. Since we know a start point in the tree we do not do a search from the top. Instead a traversal begins from a rb-tree node we split. De-allocation. ~O(log(N)) complexity. An area is not inserted straight away to the tree/list, instead we identify the spot first, checking if it can be merged around neighbors. The list provides O(1) access to prev/next, so it is pretty fast to check it. Summarizing. If merged then large coalesced areas are created, if not the area is just linked making more fragments. There is one more thing that i should mention here. After modification of VA node, its subtree_max_size is updated if it was/is the biggest area in its left or right sub-tree. Apart of that it can also be populated back to upper levels to fix the tree. For more details please have a look at the __augment_tree_propagate_from() function and the description. Tests and stressing ------------------- I use the "test_vmalloc.sh" test driver available under "tools/testing/selftests/vm/" since 5.1-rc1 kernel. Just trigger "sudo ./test_vmalloc.sh" to find out how to deal with it. Tested on different platforms including x86_64/i686/ARM64/x86_64_NUMA. Regarding last one, i do not have any physical access to NUMA system, therefore i emulated it. The time of stressing is days. If you run the test driver in "stress mode", you also need the patch that is in Andrew's tree but not in Linux 5.1-rc1. So, please apply it: http://git.cmpxchg.org/cgit.cgi/linux-mmotm.git/commit/?id=e0cf7749bade6da318e98e934a24d8b62fab512c After massive testing, i have not identified any problems like memory leaks, crashes or kernel panics. I find it stable, but more testing would be good. Performance analysis -------------------- I have used two systems to test. One is i5-3320M CPU @ 2.60GHz and another is HiKey960(arm64) board. i5-3320M runs on 4.20 kernel, whereas Hikey960 uses 4.15 kernel. I have both system which could run on 5.1-rc1 as well, but the results have not been ready by time i an writing this. Currently it consist of 8 tests. There are three of them which correspond to different types of splitting(to compare with default). We have 3 ones(see above). Another 5 do allocations in different conditions. a) sudo ./test_vmalloc.sh performance When the test driver is run in "performance" mode, it runs all available tests pinned to first online CPU with sequential execution test order. We do it in order to get stable and repeatable results. Take a look at time difference in "long_busy_list_alloc_test". It is not surprising because the worst case is O(N). # i5-3320M How many cycles all tests took: CPU0=646919905370(default) cycles vs CPU0=193290498550(patched) cycles # See detailed table with results here: ftp://vps418301.ovh.net/incoming/vmap_test_results_v2/i5-3320M_performance_default.txt ftp://vps418301.ovh.net/incoming/vmap_test_results_v2/i5-3320M_performance_patched.txt # Hikey960 8x CPUs How many cycles all tests took: CPU0=3478683207 cycles vs CPU0=463767978 cycles # See detailed table with results here: ftp://vps418301.ovh.net/incoming/vmap_test_results_v2/HiKey960_performance_default.txt ftp://vps418301.ovh.net/incoming/vmap_test_results_v2/HiKey960_performance_patched.txt b) time sudo ./test_vmalloc.sh test_repeat_count=1 With this configuration, all tests are run on all available online CPUs. Before running each CPU shuffles its tests execution order. It gives random allocation behaviour. So it is rough comparison, but it puts in the picture for sure. # i5-3320M vs real 101m22.813s real 0m56.805s user 0m0.011s user 0m0.015s sys 0m5.076s sys 0m0.023s # See detailed table with results here: ftp://vps418301.ovh.net/incoming/vmap_test_results_v2/i5-3320M_test_repeat_count_1_default.txt ftp://vps418301.ovh.net/incoming/vmap_test_results_v2/i5-3320M_test_repeat_count_1_patched.txt # Hikey960 8x CPUs vs real unknown real 4m25.214s user unknown user 0m0.011s sys unknown sys 0m0.670s I did not manage to complete this test on "default Hikey960" kernel version. After 24 hours it was still running, therefore i had to cancel it. That is why real/user/sys are "unknown". This patch (of 3): Currently an allocation of the new vmap area is done over busy list iteration(complexity O(n)) until a suitable hole is found between two busy areas. Therefore each new allocation causes the list being grown. Due to over fragmented list and different permissive parameters an allocation can take a long time. For example on embedded devices it is milliseconds. This patch organizes the KVA memory layout into free areas of the 1-ULONG_MAX range. It uses an augment red-black tree that keeps blocks sorted by their offsets in pair with linked list keeping the free space in order of increasing addresses. Nodes are augmented with the size of the maximum available free block in its left or right sub-tree. Thus, that allows to take a decision and traversal toward the block that will fit and will have the lowest start address, i.e. it is sequential allocation. Allocation: to allocate a new block a search is done over the tree until a suitable lowest(left most) block is large enough to encompass: the requested size, alignment and vstart point. If the block is bigger than requested size - it is split. De-allocation: when a busy vmap area is freed it can either be merged or inserted to the tree. Red-black tree allows efficiently find a spot whereas a linked list provides a constant-time access to previous and next blocks to check if merging can be done. In case of merging of de-allocated memory chunk a large coalesced area is created. Complexity: ~O(log(N)) [urezki@gmail.com: v3] Link: http://lkml.kernel.org/r/20190402162531.10888-2-urezki@gmail.com [urezki@gmail.com: v4] Link: http://lkml.kernel.org/r/20190406183508.25273-2-urezki@gmail.com Link: http://lkml.kernel.org/r/20190321190327.11813-2-urezki@gmail.com Signed-off-by: Uladzislau Rezki (Sony) Reviewed-by: Roman Gushchin Cc: Michal Hocko Cc: Matthew Wilcox Cc: Thomas Garnier Cc: Oleksiy Avramchenko Cc: Steven Rostedt Cc: Joel Fernandes Cc: Thomas Gleixner Cc: Ingo Molnar Cc: Tejun Heo Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/vmalloc.c | 1004 ++++++++++++++++++++++++++++++++++++++++++++-------------- 1 file changed, 758 insertions(+), 246 deletions(-) (limited to 'mm') diff --git a/mm/vmalloc.c b/mm/vmalloc.c index 67bbb8d2a0a8..f8f61ff3235b 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -32,6 +32,7 @@ #include #include #include +#include #include #include @@ -332,14 +333,67 @@ static DEFINE_SPINLOCK(vmap_area_lock); LIST_HEAD(vmap_area_list); static LLIST_HEAD(vmap_purge_list); static struct rb_root vmap_area_root = RB_ROOT; +static bool vmap_initialized __read_mostly; -/* The vmap cache globals are protected by vmap_area_lock */ -static struct rb_node *free_vmap_cache; -static unsigned long cached_hole_size; -static unsigned long cached_vstart; -static unsigned long cached_align; +/* + * This kmem_cache is used for vmap_area objects. Instead of + * allocating from slab we reuse an object from this cache to + * make things faster. Especially in "no edge" splitting of + * free block. + */ +static struct kmem_cache *vmap_area_cachep; + +/* + * This linked list is used in pair with free_vmap_area_root. + * It gives O(1) access to prev/next to perform fast coalescing. + */ +static LIST_HEAD(free_vmap_area_list); + +/* + * This augment red-black tree represents the free vmap space. + * All vmap_area objects in this tree are sorted by va->va_start + * address. It is used for allocation and merging when a vmap + * object is released. + * + * Each vmap_area node contains a maximum available free block + * of its sub-tree, right or left. Therefore it is possible to + * find a lowest match of free area. + */ +static struct rb_root free_vmap_area_root = RB_ROOT; + +static __always_inline unsigned long +va_size(struct vmap_area *va) +{ + return (va->va_end - va->va_start); +} + +static __always_inline unsigned long +get_subtree_max_size(struct rb_node *node) +{ + struct vmap_area *va; + + va = rb_entry_safe(node, struct vmap_area, rb_node); + return va ? va->subtree_max_size : 0; +} -static unsigned long vmap_area_pcpu_hole; +/* + * Gets called when remove the node and rotate. + */ +static __always_inline unsigned long +compute_subtree_max_size(struct vmap_area *va) +{ + return max3(va_size(va), + get_subtree_max_size(va->rb_node.rb_left), + get_subtree_max_size(va->rb_node.rb_right)); +} + +RB_DECLARE_CALLBACKS(static, free_vmap_area_rb_augment_cb, + struct vmap_area, rb_node, unsigned long, subtree_max_size, + compute_subtree_max_size) + +static void purge_vmap_area_lazy(void); +static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); +static unsigned long lazy_max_pages(void); static struct vmap_area *__find_vmap_area(unsigned long addr) { @@ -360,41 +414,522 @@ static struct vmap_area *__find_vmap_area(unsigned long addr) return NULL; } -static void __insert_vmap_area(struct vmap_area *va) -{ - struct rb_node **p = &vmap_area_root.rb_node; - struct rb_node *parent = NULL; - struct rb_node *tmp; +/* + * This function returns back addresses of parent node + * and its left or right link for further processing. + */ +static __always_inline struct rb_node ** +find_va_links(struct vmap_area *va, + struct rb_root *root, struct rb_node *from, + struct rb_node **parent) +{ + struct vmap_area *tmp_va; + struct rb_node **link; + + if (root) { + link = &root->rb_node; + if (unlikely(!*link)) { + *parent = NULL; + return link; + } + } else { + link = &from; + } - while (*p) { - struct vmap_area *tmp_va; + /* + * Go to the bottom of the tree. When we hit the last point + * we end up with parent rb_node and correct direction, i name + * it link, where the new va->rb_node will be attached to. + */ + do { + tmp_va = rb_entry(*link, struct vmap_area, rb_node); - parent = *p; - tmp_va = rb_entry(parent, struct vmap_area, rb_node); - if (va->va_start < tmp_va->va_end) - p = &(*p)->rb_left; - else if (va->va_end > tmp_va->va_start) - p = &(*p)->rb_right; + /* + * During the traversal we also do some sanity check. + * Trigger the BUG() if there are sides(left/right) + * or full overlaps. + */ + if (va->va_start < tmp_va->va_end && + va->va_end <= tmp_va->va_start) + link = &(*link)->rb_left; + else if (va->va_end > tmp_va->va_start && + va->va_start >= tmp_va->va_end) + link = &(*link)->rb_right; else BUG(); + } while (*link); + + *parent = &tmp_va->rb_node; + return link; +} + +static __always_inline struct list_head * +get_va_next_sibling(struct rb_node *parent, struct rb_node **link) +{ + struct list_head *list; + + if (unlikely(!parent)) + /* + * The red-black tree where we try to find VA neighbors + * before merging or inserting is empty, i.e. it means + * there is no free vmap space. Normally it does not + * happen but we handle this case anyway. + */ + return NULL; + + list = &rb_entry(parent, struct vmap_area, rb_node)->list; + return (&parent->rb_right == link ? list->next : list); +} + +static __always_inline void +link_va(struct vmap_area *va, struct rb_root *root, + struct rb_node *parent, struct rb_node **link, struct list_head *head) +{ + /* + * VA is still not in the list, but we can + * identify its future previous list_head node. + */ + if (likely(parent)) { + head = &rb_entry(parent, struct vmap_area, rb_node)->list; + if (&parent->rb_right != link) + head = head->prev; } - rb_link_node(&va->rb_node, parent, p); - rb_insert_color(&va->rb_node, &vmap_area_root); + /* Insert to the rb-tree */ + rb_link_node(&va->rb_node, parent, link); + if (root == &free_vmap_area_root) { + /* + * Some explanation here. Just perform simple insertion + * to the tree. We do not set va->subtree_max_size to + * its current size before calling rb_insert_augmented(). + * It is because of we populate the tree from the bottom + * to parent levels when the node _is_ in the tree. + * + * Therefore we set subtree_max_size to zero after insertion, + * to let __augment_tree_propagate_from() puts everything to + * the correct order later on. + */ + rb_insert_augmented(&va->rb_node, + root, &free_vmap_area_rb_augment_cb); + va->subtree_max_size = 0; + } else { + rb_insert_color(&va->rb_node, root); + } - /* address-sort this list */ - tmp = rb_prev(&va->rb_node); - if (tmp) { - struct vmap_area *prev; - prev = rb_entry(tmp, struct vmap_area, rb_node); - list_add_rcu(&va->list, &prev->list); - } else - list_add_rcu(&va->list, &vmap_area_list); + /* Address-sort this list */ + list_add(&va->list, head); } -static void purge_vmap_area_lazy(void); +static __always_inline void +unlink_va(struct vmap_area *va, struct rb_root *root) +{ + /* + * During merging a VA node can be empty, therefore + * not linked with the tree nor list. Just check it. + */ + if (!RB_EMPTY_NODE(&va->rb_node)) { + if (root == &free_vmap_area_root) + rb_erase_augmented(&va->rb_node, + root, &free_vmap_area_rb_augment_cb); + else + rb_erase(&va->rb_node, root); -static BLOCKING_NOTIFIER_HEAD(vmap_notify_list); + list_del(&va->list); + RB_CLEAR_NODE(&va->rb_node); + } +} + +/* + * This function populates subtree_max_size from bottom to upper + * levels starting from VA point. The propagation must be done + * when VA size is modified by changing its va_start/va_end. Or + * in case of newly inserting of VA to the tree. + * + * It means that __augment_tree_propagate_from() must be called: + * - After VA has been inserted to the tree(free path); + * - After VA has been shrunk(allocation path); + * - After VA has been increased(merging path). + * + * Please note that, it does not mean that upper parent nodes + * and their subtree_max_size are recalculated all the time up + * to the root node. + * + * 4--8 + * /\ + * / \ + * / \ + * 2--2 8--8 + * + * For example if we modify the node 4, shrinking it to 2, then + * no any modification is required. If we shrink the node 2 to 1 + * its subtree_max_size is updated only, and set to 1. If we shrink + * the node 8 to 6, then its subtree_max_size is set to 6 and parent + * node becomes 4--6. + */ +static __always_inline void +augment_tree_propagate_from(struct vmap_area *va) +{ + struct rb_node *node = &va->rb_node; + unsigned long new_va_sub_max_size; + + while (node) { + va = rb_entry(node, struct vmap_area, rb_node); + new_va_sub_max_size = compute_subtree_max_size(va); + + /* + * If the newly calculated maximum available size of the + * subtree is equal to the current one, then it means that + * the tree is propagated correctly. So we have to stop at + * this point to save cycles. + */ + if (va->subtree_max_size == new_va_sub_max_size) + break; + + va->subtree_max_size = new_va_sub_max_size; + node = rb_parent(&va->rb_node); + } +} + +static void +insert_vmap_area(struct vmap_area *va, + struct rb_root *root, struct list_head *head) +{ + struct rb_node **link; + struct rb_node *parent; + + link = find_va_links(va, root, NULL, &parent); + link_va(va, root, parent, link, head); +} + +static void +insert_vmap_area_augment(struct vmap_area *va, + struct rb_node *from, struct rb_root *root, + struct list_head *head) +{ + struct rb_node **link; + struct rb_node *parent; + + if (from) + link = find_va_links(va, NULL, from, &parent); + else + link = find_va_links(va, root, NULL, &parent); + + link_va(va, root, parent, link, head); + augment_tree_propagate_from(va); +} + +/* + * Merge de-allocated chunk of VA memory with previous + * and next free blocks. If coalesce is not done a new + * free area is inserted. If VA has been merged, it is + * freed. + */ +static __always_inline void +merge_or_add_vmap_area(struct vmap_area *va, + struct rb_root *root, struct list_head *head) +{ + struct vmap_area *sibling; + struct list_head *next; + struct rb_node **link; + struct rb_node *parent; + bool merged = false; + + /* + * Find a place in the tree where VA potentially will be + * inserted, unless it is merged with its sibling/siblings. + */ + link = find_va_links(va, root, NULL, &parent); + + /* + * Get next node of VA to check if merging can be done. + */ + next = get_va_next_sibling(parent, link); + if (unlikely(next == NULL)) + goto insert; + + /* + * start end + * | | + * |<------VA------>|<-----Next----->| + * | | + * start end + */ + if (next != head) { + sibling = list_entry(next, struct vmap_area, list); + if (sibling->va_start == va->va_end) { + sibling->va_start = va->va_start; + + /* Check and update the tree if needed. */ + augment_tree_propagate_from(sibling); + + /* Remove this VA, it has been merged. */ + unlink_va(va, root); + + /* Free vmap_area object. */ + kmem_cache_free(vmap_area_cachep, va); + + /* Point to the new merged area. */ + va = sibling; + merged = true; + } + } + + /* + * start end + * | | + * |<-----Prev----->|<------VA------>| + * | | + * start end + */ + if (next->prev != head) { + sibling = list_entry(next->prev, struct vmap_area, list); + if (sibling->va_end == va->va_start) { + sibling->va_end = va->va_end; + + /* Check and update the tree if needed. */ + augment_tree_propagate_from(sibling); + + /* Remove this VA, it has been merged. */ + unlink_va(va, root); + + /* Free vmap_area object. */ + kmem_cache_free(vmap_area_cachep, va); + + return; + } + } + +insert: + if (!merged) { + link_va(va, root, parent, link, head); + augment_tree_propagate_from(va); + } +} + +static __always_inline bool +is_within_this_va(struct vmap_area *va, unsigned long size, + unsigned long align, unsigned long vstart) +{ + unsigned long nva_start_addr; + + if (va->va_start > vstart) + nva_start_addr = ALIGN(va->va_start, align); + else + nva_start_addr = ALIGN(vstart, align); + + /* Can be overflowed due to big size or alignment. */ + if (nva_start_addr + size < nva_start_addr || + nva_start_addr < vstart) + return false; + + return (nva_start_addr + size <= va->va_end); +} + +/* + * Find the first free block(lowest start address) in the tree, + * that will accomplish the request corresponding to passing + * parameters. + */ +static __always_inline struct vmap_area * +find_vmap_lowest_match(unsigned long size, + unsigned long align, unsigned long vstart) +{ + struct vmap_area *va; + struct rb_node *node; + unsigned long length; + + /* Start from the root. */ + node = free_vmap_area_root.rb_node; + + /* Adjust the search size for alignment overhead. */ + length = size + align - 1; + + while (node) { + va = rb_entry(node, struct vmap_area, rb_node); + + if (get_subtree_max_size(node->rb_left) >= length && + vstart < va->va_start) { + node = node->rb_left; + } else { + if (is_within_this_va(va, size, align, vstart)) + return va; + + /* + * Does not make sense to go deeper towards the right + * sub-tree if it does not have a free block that is + * equal or bigger to the requested search length. + */ + if (get_subtree_max_size(node->rb_right) >= length) { + node = node->rb_right; + continue; + } + + /* + * OK. We roll back and find the fist right sub-tree, + * that will satisfy the search criteria. It can happen + * only once due to "vstart" restriction. + */ + while ((node = rb_parent(node))) { + va = rb_entry(node, struct vmap_area, rb_node); + if (is_within_this_va(va, size, align, vstart)) + return va; + + if (get_subtree_max_size(node->rb_right) >= length && + vstart <= va->va_start) { + node = node->rb_right; + break; + } + } + } + } + + return NULL; +} + +enum fit_type { + NOTHING_FIT = 0, + FL_FIT_TYPE = 1, /* full fit */ + LE_FIT_TYPE = 2, /* left edge fit */ + RE_FIT_TYPE = 3, /* right edge fit */ + NE_FIT_TYPE = 4 /* no edge fit */ +}; + +static __always_inline enum fit_type +classify_va_fit_type(struct vmap_area *va, + unsigned long nva_start_addr, unsigned long size) +{ + enum fit_type type; + + /* Check if it is within VA. */ + if (nva_start_addr < va->va_start || + nva_start_addr + size > va->va_end) + return NOTHING_FIT; + + /* Now classify. */ + if (va->va_start == nva_start_addr) { + if (va->va_end == nva_start_addr + size) + type = FL_FIT_TYPE; + else + type = LE_FIT_TYPE; + } else if (va->va_end == nva_start_addr + size) { + type = RE_FIT_TYPE; + } else { + type = NE_FIT_TYPE; + } + + return type; +} + +static __always_inline int +adjust_va_to_fit_type(struct vmap_area *va, + unsigned long nva_start_addr, unsigned long size, + enum fit_type type) +{ + struct vmap_area *lva; + + if (type == FL_FIT_TYPE) { + /* + * No need to split VA, it fully fits. + * + * | | + * V NVA V + * |---------------| + */ + unlink_va(va, &free_vmap_area_root); + kmem_cache_free(vmap_area_cachep, va); + } else if (type == LE_FIT_TYPE) { + /* + * Split left edge of fit VA. + * + * | | + * V NVA V R + * |-------|-------| + */ + va->va_start += size; + } else if (type == RE_FIT_TYPE) { + /* + * Split right edge of fit VA. + * + * | | + * L V NVA V + * |-------|-------| + */ + va->va_end = nva_start_addr; + } else if (type == NE_FIT_TYPE) { + /* + * Split no edge of fit VA. + * + * | | + * L V NVA V R + * |---|-------|---| + */ + lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT); + if (unlikely(!lva)) + return -1; + + /* + * Build the remainder. + */ + lva->va_start = va->va_start; + lva->va_end = nva_start_addr; + + /* + * Shrink this VA to remaining size. + */ + va->va_start = nva_start_addr + size; + } else { + return -1; + } + + if (type != FL_FIT_TYPE) { + augment_tree_propagate_from(va); + + if (type == NE_FIT_TYPE) + insert_vmap_area_augment(lva, &va->rb_node, + &free_vmap_area_root, &free_vmap_area_list); + } + + return 0; +} + +/* + * Returns a start address of the newly allocated area, if success. + * Otherwise a vend is returned that indicates failure. + */ +static __always_inline unsigned long +__alloc_vmap_area(unsigned long size, unsigned long align, + unsigned long vstart, unsigned long vend, int node) +{ + unsigned long nva_start_addr; + struct vmap_area *va; + enum fit_type type; + int ret; + + va = find_vmap_lowest_match(size, align, vstart); + if (unlikely(!va)) + return vend; + + if (va->va_start > vstart) + nva_start_addr = ALIGN(va->va_start, align); + else + nva_start_addr = ALIGN(vstart, align); + + /* Check the "vend" restriction. */ + if (nva_start_addr + size > vend) + return vend; + + /* Classify what we have found. */ + type = classify_va_fit_type(va, nva_start_addr, size); + if (WARN_ON_ONCE(type == NOTHING_FIT)) + return vend; + + /* Update the free vmap_area. */ + ret = adjust_va_to_fit_type(va, nva_start_addr, size, type); + if (ret) + return vend; + + return nva_start_addr; +} /* * Allocate a region of KVA of the specified size and alignment, within the @@ -406,18 +941,19 @@ static struct vmap_area *alloc_vmap_area(unsigned long size, int node, gfp_t gfp_mask) { struct vmap_area *va; - struct rb_node *n; unsigned long addr; int purged = 0; - struct vmap_area *first; BUG_ON(!size); BUG_ON(offset_in_page(size)); BUG_ON(!is_power_of_2(align)); + if (unlikely(!vmap_initialized)) + return ERR_PTR(-EBUSY); + might_sleep(); - va = kmalloc_node(sizeof(struct vmap_area), + va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask & GFP_RECLAIM_MASK, node); if (unlikely(!va)) return ERR_PTR(-ENOMEM); @@ -430,87 +966,20 @@ static struct vmap_area *alloc_vmap_area(unsigned long size, retry: spin_lock(&vmap_area_lock); - /* - * Invalidate cache if we have more permissive parameters. - * cached_hole_size notes the largest hole noticed _below_ - * the vmap_area cached in free_vmap_cache: if size fits - * into that hole, we want to scan from vstart to reuse - * the hole instead of allocating above free_vmap_cache. - * Note that __free_vmap_area may update free_vmap_cache - * without updating cached_hole_size or cached_align. - */ - if (!free_vmap_cache || - size < cached_hole_size || - vstart < cached_vstart || - align < cached_align) { -nocache: - cached_hole_size = 0; - free_vmap_cache = NULL; - } - /* record if we encounter less permissive parameters */ - cached_vstart = vstart; - cached_align = align; - - /* find starting point for our search */ - if (free_vmap_cache) { - first = rb_entry(free_vmap_cache, struct vmap_area, rb_node); - addr = ALIGN(first->va_end, align); - if (addr < vstart) - goto nocache; - if (addr + size < addr) - goto overflow; - - } else { - addr = ALIGN(vstart, align); - if (addr + size < addr) - goto overflow; - - n = vmap_area_root.rb_node; - first = NULL; - - while (n) { - struct vmap_area *tmp; - tmp = rb_entry(n, struct vmap_area, rb_node); - if (tmp->va_end >= addr) { - first = tmp; - if (tmp->va_start <= addr) - break; - n = n->rb_left; - } else - n = n->rb_right; - } - - if (!first) - goto found; - } - /* from the starting point, walk areas until a suitable hole is found */ - while (addr + size > first->va_start && addr + size <= vend) { - if (addr + cached_hole_size < first->va_start) - cached_hole_size = first->va_start - addr; - addr = ALIGN(first->va_end, align); - if (addr + size < addr) - goto overflow; - - if (list_is_last(&first->list, &vmap_area_list)) - goto found; - - first = list_next_entry(first, list); - } - -found: /* - * Check also calculated address against the vstart, - * because it can be 0 because of big align request. + * If an allocation fails, the "vend" address is + * returned. Therefore trigger the overflow path. */ - if (addr + size > vend || addr < vstart) + addr = __alloc_vmap_area(size, align, vstart, vend, node); + if (unlikely(addr == vend)) goto overflow; va->va_start = addr; va->va_end = addr + size; va->flags = 0; - __insert_vmap_area(va); - free_vmap_cache = &va->rb_node; + insert_vmap_area(va, &vmap_area_root, &vmap_area_list); + spin_unlock(&vmap_area_lock); BUG_ON(!IS_ALIGNED(va->va_start, align)); @@ -539,7 +1008,8 @@ overflow: if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) pr_warn("vmap allocation for size %lu failed: use vmalloc= to increase size\n", size); - kfree(va); + + kmem_cache_free(vmap_area_cachep, va); return ERR_PTR(-EBUSY); } @@ -559,35 +1029,16 @@ static void __free_vmap_area(struct vmap_area *va) { BUG_ON(RB_EMPTY_NODE(&va->rb_node)); - if (free_vmap_cache) { - if (va->va_end < cached_vstart) { - free_vmap_cache = NULL; - } else { - struct vmap_area *cache; - cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node); - if (va->va_start <= cache->va_start) { - free_vmap_cache = rb_prev(&va->rb_node); - /* - * We don't try to update cached_hole_size or - * cached_align, but it won't go very wrong. - */ - } - } - } - rb_erase(&va->rb_node, &vmap_area_root); - RB_CLEAR_NODE(&va->rb_node); - list_del_rcu(&va->list); - /* - * Track the highest possible candidate for pcpu area - * allocation. Areas outside of vmalloc area can be returned - * here too, consider only end addresses which fall inside - * vmalloc area proper. + * Remove from the busy tree/list. */ - if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END) - vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end); + unlink_va(va, &vmap_area_root); - kfree_rcu(va, rcu_head); + /* + * Merge VA with its neighbors, otherwise just add it. + */ + merge_or_add_vmap_area(va, + &free_vmap_area_root, &free_vmap_area_list); } /* @@ -794,8 +1245,6 @@ static struct vmap_area *find_vmap_area(unsigned long addr) #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE) -static bool vmap_initialized __read_mostly = false; - struct vmap_block_queue { spinlock_t lock; struct list_head free; @@ -1256,12 +1705,58 @@ void __init vm_area_register_early(struct vm_struct *vm, size_t align) vm_area_add_early(vm); } +static void vmap_init_free_space(void) +{ + unsigned long vmap_start = 1; + const unsigned long vmap_end = ULONG_MAX; + struct vmap_area *busy, *free; + + /* + * B F B B B F + * -|-----|.....|-----|-----|-----|.....|- + * | The KVA space | + * |<--------------------------------->| + */ + list_for_each_entry(busy, &vmap_area_list, list) { + if (busy->va_start - vmap_start > 0) { + free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); + if (!WARN_ON_ONCE(!free)) { + free->va_start = vmap_start; + free->va_end = busy->va_start; + + insert_vmap_area_augment(free, NULL, + &free_vmap_area_root, + &free_vmap_area_list); + } + } + + vmap_start = busy->va_end; + } + + if (vmap_end - vmap_start > 0) { + free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); + if (!WARN_ON_ONCE(!free)) { + free->va_start = vmap_start; + free->va_end = vmap_end; + + insert_vmap_area_augment(free, NULL, + &free_vmap_area_root, + &free_vmap_area_list); + } + } +} + void __init vmalloc_init(void) { struct vmap_area *va; struct vm_struct *tmp; int i; + /* + * Create the cache for vmap_area objects. + */ + vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC); + for_each_possible_cpu(i) { struct vmap_block_queue *vbq; struct vfree_deferred *p; @@ -1276,16 +1771,21 @@ void __init vmalloc_init(void) /* Import existing vmlist entries. */ for (tmp = vmlist; tmp; tmp = tmp->next) { - va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT); + va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT); + if (WARN_ON_ONCE(!va)) + continue; + va->flags = VM_VM_AREA; va->va_start = (unsigned long)tmp->addr; va->va_end = va->va_start + tmp->size; va->vm = tmp; - __insert_vmap_area(va); + insert_vmap_area(va, &vmap_area_root, &vmap_area_list); } - vmap_area_pcpu_hole = VMALLOC_END; - + /* + * Now we can initialize a free vmap space. + */ + vmap_init_free_space(); vmap_initialized = true; } @@ -2477,81 +2977,64 @@ static struct vmap_area *node_to_va(struct rb_node *n) } /** - * pvm_find_next_prev - find the next and prev vmap_area surrounding @end - * @end: target address - * @pnext: out arg for the next vmap_area - * @pprev: out arg for the previous vmap_area + * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to + * @addr: target address * - * Returns: %true if either or both of next and prev are found, - * %false if no vmap_area exists - * - * Find vmap_areas end addresses of which enclose @end. ie. if not - * NULL, *pnext->va_end > @end and *pprev->va_end <= @end. + * Returns: vmap_area if it is found. If there is no such area + * the first highest(reverse order) vmap_area is returned + * i.e. va->va_start < addr && va->va_end < addr or NULL + * if there are no any areas before @addr. */ -static bool pvm_find_next_prev(unsigned long end, - struct vmap_area **pnext, - struct vmap_area **pprev) +static struct vmap_area * +pvm_find_va_enclose_addr(unsigned long addr) { - struct rb_node *n = vmap_area_root.rb_node; - struct vmap_area *va = NULL; + struct vmap_area *va, *tmp; + struct rb_node *n; + + n = free_vmap_area_root.rb_node; + va = NULL; while (n) { - va = rb_entry(n, struct vmap_area, rb_node); - if (end < va->va_end) - n = n->rb_left; - else if (end > va->va_end) + tmp = rb_entry(n, struct vmap_area, rb_node); + if (tmp->va_start <= addr) { + va = tmp; + if (tmp->va_end >= addr) + break; + n = n->rb_right; - else - break; + } else { + n = n->rb_left; + } } - if (!va) - return false; - - if (va->va_end > end) { - *pnext = va; - *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); - } else { - *pprev = va; - *pnext = node_to_va(rb_next(&(*pprev)->rb_node)); - } - return true; + return va; } /** - * pvm_determine_end - find the highest aligned address between two vmap_areas - * @pnext: in/out arg for the next vmap_area - * @pprev: in/out arg for the previous vmap_area - * @align: alignment - * - * Returns: determined end address + * pvm_determine_end_from_reverse - find the highest aligned address + * of free block below VMALLOC_END + * @va: + * in - the VA we start the search(reverse order); + * out - the VA with the highest aligned end address. * - * Find the highest aligned address between *@pnext and *@pprev below - * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned - * down address is between the end addresses of the two vmap_areas. - * - * Please note that the address returned by this function may fall - * inside *@pnext vmap_area. The caller is responsible for checking - * that. + * Returns: determined end address within vmap_area */ -static unsigned long pvm_determine_end(struct vmap_area **pnext, - struct vmap_area **pprev, - unsigned long align) +static unsigned long +pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align) { - const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); + unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); unsigned long addr; - if (*pnext) - addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end); - else - addr = vmalloc_end; - - while (*pprev && (*pprev)->va_end > addr) { - *pnext = *pprev; - *pprev = node_to_va(rb_prev(&(*pnext)->rb_node)); + if (likely(*va)) { + list_for_each_entry_from_reverse((*va), + &free_vmap_area_list, list) { + addr = min((*va)->va_end & ~(align - 1), vmalloc_end); + if ((*va)->va_start < addr) + return addr; + } } - return addr; + return 0; } /** @@ -2571,12 +3054,12 @@ static unsigned long pvm_determine_end(struct vmap_area **pnext, * to gigabytes. To avoid interacting with regular vmallocs, these * areas are allocated from top. * - * Despite its complicated look, this allocator is rather simple. It - * does everything top-down and scans areas from the end looking for - * matching slot. While scanning, if any of the areas overlaps with - * existing vmap_area, the base address is pulled down to fit the - * area. Scanning is repeated till all the areas fit and then all - * necessary data structures are inserted and the result is returned. + * Despite its complicated look, this allocator is rather simple. It + * does everything top-down and scans free blocks from the end looking + * for matching base. While scanning, if any of the areas do not fit the + * base address is pulled down to fit the area. Scanning is repeated till + * all the areas fit and then all necessary data structures are inserted + * and the result is returned. */ struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, const size_t *sizes, int nr_vms, @@ -2584,11 +3067,12 @@ struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, { const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align); const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1); - struct vmap_area **vas, *prev, *next; + struct vmap_area **vas, *va; struct vm_struct **vms; int area, area2, last_area, term_area; - unsigned long base, start, end, last_end; + unsigned long base, start, size, end, last_end; bool purged = false; + enum fit_type type; /* verify parameters and allocate data structures */ BUG_ON(offset_in_page(align) || !is_power_of_2(align)); @@ -2624,7 +3108,7 @@ struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets, goto err_free2; for (area = 0; area < nr_vms; area++) { - vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL); + vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL); vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL); if (!vas[area] || !vms[area]) goto err_free; @@ -2637,49 +3121,29 @@ retry: start = offsets[area]; end = start + sizes[area]; - if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) { - base = vmalloc_end - last_end; - goto found; - } - base = pvm_determine_end(&next, &prev, align) - end; + va = pvm_find_va_enclose_addr(vmalloc_end); + base = pvm_determine_end_from_reverse(&va, align) - end; while (true) { - BUG_ON(next && next->va_end <= base + end); - BUG_ON(prev && prev->va_end > base + end); - /* * base might have underflowed, add last_end before * comparing. */ - if (base + last_end < vmalloc_start + last_end) { - spin_unlock(&vmap_area_lock); - if (!purged) { - purge_vmap_area_lazy(); - purged = true; - goto retry; - } - goto err_free; - } + if (base + last_end < vmalloc_start + last_end) + goto overflow; /* - * If next overlaps, move base downwards so that it's - * right below next and then recheck. + * Fitting base has not been found. */ - if (next && next->va_start < base + end) { - base = pvm_determine_end(&next, &prev, align) - end; - term_area = area; - continue; - } + if (va == NULL) + goto overflow; /* - * If prev overlaps, shift down next and prev and move - * base so that it's right below new next and then - * recheck. + * If this VA does not fit, move base downwards and recheck. */ - if (prev && prev->va_end > base + start) { - next = prev; - prev = node_to_va(rb_prev(&next->rb_node)); - base = pvm_determine_end(&next, &prev, align) - end; + if (base + start < va->va_start || base + end > va->va_end) { + va = node_to_va(rb_prev(&va->rb_node)); + base = pvm_determine_end_from_reverse(&va, align) - end; term_area = area; continue; } @@ -2691,21 +3155,40 @@ retry: area = (area + nr_vms - 1) % nr_vms; if (area == term_area) break; + start = offsets[area]; end = start + sizes[area]; - pvm_find_next_prev(base + end, &next, &prev); + va = pvm_find_va_enclose_addr(base + end); } -found: + /* we've found a fitting base, insert all va's */ for (area = 0; area < nr_vms; area++) { - struct vmap_area *va = vas[area]; + int ret; - va->va_start = base + offsets[area]; - va->va_end = va->va_start + sizes[area]; - __insert_vmap_area(va); - } + start = base + offsets[area]; + size = sizes[area]; - vmap_area_pcpu_hole = base + offsets[last_area]; + va = pvm_find_va_enclose_addr(start); + if (WARN_ON_ONCE(va == NULL)) + /* It is a BUG(), but trigger recovery instead. */ + goto recovery; + + type = classify_va_fit_type(va, start, size); + if (WARN_ON_ONCE(type == NOTHING_FIT)) + /* It is a BUG(), but trigger recovery instead. */ + goto recovery; + + ret = adjust_va_to_fit_type(va, start, size, type); + if (unlikely(ret)) + goto recovery; + + /* Allocated area. */ + va = vas[area]; + va->va_start = start; + va->va_end = start + size; + + insert_vmap_area(va, &vmap_area_root, &vmap_area_list); + } spin_unlock(&vmap_area_lock); @@ -2717,9 +3200,38 @@ found: kfree(vas); return vms; +recovery: + /* Remove previously inserted areas. */ + while (area--) { + __free_vmap_area(vas[area]); + vas[area] = NULL; + } + +overflow: + spin_unlock(&vmap_area_lock); + if (!purged) { + purge_vmap_area_lazy(); + purged = true; + + /* Before "retry", check if we recover. */ + for (area = 0; area < nr_vms; area++) { + if (vas[area]) + continue; + + vas[area] = kmem_cache_zalloc( + vmap_area_cachep, GFP_KERNEL); + if (!vas[area]) + goto err_free; + } + + goto retry; + } + err_free: for (area = 0; area < nr_vms; area++) { - kfree(vas[area]); + if (vas[area]) + kmem_cache_free(vmap_area_cachep, vas[area]); + kfree(vms[area]); } err_free2: -- cgit From bb850f4dae4abb18c5ee727bb2d6df9ca47ede49 Mon Sep 17 00:00:00 2001 From: "Uladzislau Rezki (Sony)" Date: Fri, 17 May 2019 14:31:34 -0700 Subject: mm/vmap: add DEBUG_AUGMENT_PROPAGATE_CHECK macro This macro adds some debug code to check that the augment tree is maintained correctly, meaning that every node contains valid subtree_max_size value. By default this option is set to 0 and not active. It requires recompilation of the kernel to activate it. Set to 1, compile the kernel. [urezki@gmail.com: v4] Link: http://lkml.kernel.org/r/20190406183508.25273-3-urezki@gmail.com Link: http://lkml.kernel.org/r/20190402162531.10888-3-urezki@gmail.com Signed-off-by: Uladzislau Rezki (Sony) Reviewed-by: Roman Gushchin Cc: Ingo Molnar Cc: Joel Fernandes Cc: Matthew Wilcox Cc: Michal Hocko Cc: Oleksiy Avramchenko Cc: Steven Rostedt Cc: Tejun Heo Cc: Thomas Garnier Cc: Thomas Gleixner Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/vmalloc.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 48 insertions(+) (limited to 'mm') diff --git a/mm/vmalloc.c b/mm/vmalloc.c index f8f61ff3235b..aac34c2a410b 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -325,6 +325,8 @@ EXPORT_SYMBOL(vmalloc_to_pfn); /*** Global kva allocator ***/ +#define DEBUG_AUGMENT_PROPAGATE_CHECK 0 + #define VM_LAZY_FREE 0x02 #define VM_VM_AREA 0x04 @@ -539,6 +541,48 @@ unlink_va(struct vmap_area *va, struct rb_root *root) } } +#if DEBUG_AUGMENT_PROPAGATE_CHECK +static void +augment_tree_propagate_check(struct rb_node *n) +{ + struct vmap_area *va; + struct rb_node *node; + unsigned long size; + bool found = false; + + if (n == NULL) + return; + + va = rb_entry(n, struct vmap_area, rb_node); + size = va->subtree_max_size; + node = n; + + while (node) { + va = rb_entry(node, struct vmap_area, rb_node); + + if (get_subtree_max_size(node->rb_left) == size) { + node = node->rb_left; + } else { + if (va_size(va) == size) { + found = true; + break; + } + + node = node->rb_right; + } + } + + if (!found) { + va = rb_entry(n, struct vmap_area, rb_node); + pr_emerg("tree is corrupted: %lu, %lu\n", + va_size(va), va->subtree_max_size); + } + + augment_tree_propagate_check(n->rb_left); + augment_tree_propagate_check(n->rb_right); +} +#endif + /* * This function populates subtree_max_size from bottom to upper * levels starting from VA point. The propagation must be done @@ -588,6 +632,10 @@ augment_tree_propagate_from(struct vmap_area *va) va->subtree_max_size = new_va_sub_max_size; node = rb_parent(&va->rb_node); } + +#if DEBUG_AUGMENT_PROPAGATE_CHECK + augment_tree_propagate_check(free_vmap_area_root.rb_node); +#endif } static void -- cgit From a6cf4e0fe3e740ed7af39fdda721e1ac12247dd3 Mon Sep 17 00:00:00 2001 From: "Uladzislau Rezki (Sony)" Date: Fri, 17 May 2019 14:31:37 -0700 Subject: mm/vmap: add DEBUG_AUGMENT_LOWEST_MATCH_CHECK macro This macro adds some debug code to check that vmap allocations are happened in ascending order. By default this option is set to 0 and not active. It requires recompilation of the kernel to activate it. Set to 1, compile the kernel. [urezki@gmail.com: v4] Link: http://lkml.kernel.org/r/20190406183508.25273-4-urezki@gmail.com Link: http://lkml.kernel.org/r/20190402162531.10888-4-urezki@gmail.com Signed-off-by: Uladzislau Rezki (Sony) Reviewed-by: Roman Gushchin Cc: Ingo Molnar Cc: Joel Fernandes Cc: Matthew Wilcox Cc: Michal Hocko Cc: Oleksiy Avramchenko Cc: Steven Rostedt Cc: Tejun Heo Cc: Thomas Garnier Cc: Thomas Gleixner Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/vmalloc.c | 43 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 43 insertions(+) (limited to 'mm') diff --git a/mm/vmalloc.c b/mm/vmalloc.c index aac34c2a410b..c42872ed82ac 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -326,6 +326,7 @@ EXPORT_SYMBOL(vmalloc_to_pfn); /*** Global kva allocator ***/ #define DEBUG_AUGMENT_PROPAGATE_CHECK 0 +#define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0 #define VM_LAZY_FREE 0x02 #define VM_VM_AREA 0x04 @@ -834,6 +835,44 @@ find_vmap_lowest_match(unsigned long size, return NULL; } +#if DEBUG_AUGMENT_LOWEST_MATCH_CHECK +#include + +static struct vmap_area * +find_vmap_lowest_linear_match(unsigned long size, + unsigned long align, unsigned long vstart) +{ + struct vmap_area *va; + + list_for_each_entry(va, &free_vmap_area_list, list) { + if (!is_within_this_va(va, size, align, vstart)) + continue; + + return va; + } + + return NULL; +} + +static void +find_vmap_lowest_match_check(unsigned long size) +{ + struct vmap_area *va_1, *va_2; + unsigned long vstart; + unsigned int rnd; + + get_random_bytes(&rnd, sizeof(rnd)); + vstart = VMALLOC_START + rnd; + + va_1 = find_vmap_lowest_match(size, 1, vstart); + va_2 = find_vmap_lowest_linear_match(size, 1, vstart); + + if (va_1 != va_2) + pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n", + va_1, va_2, vstart); +} +#endif + enum fit_type { NOTHING_FIT = 0, FL_FIT_TYPE = 1, /* full fit */ @@ -976,6 +1015,10 @@ __alloc_vmap_area(unsigned long size, unsigned long align, if (ret) return vend; +#if DEBUG_AUGMENT_LOWEST_MATCH_CHECK + find_vmap_lowest_match_check(size); +#endif + return nva_start_addr; } -- cgit From 60fce36afa9c77c7ccbf980c4f670f3be3651fce Mon Sep 17 00:00:00 2001 From: Mel Gorman Date: Fri, 17 May 2019 14:31:41 -0700 Subject: mm/compaction.c: correct zone boundary handling when isolating pages from a pageblock syzbot reported the following error from a tree with a head commit of baf76f0c58ae ("slip: make slhc_free() silently accept an error pointer") BUG: unable to handle kernel paging request at ffffea0003348000 #PF error: [normal kernel read fault] PGD 12c3f9067 P4D 12c3f9067 PUD 12c3f8067 PMD 0 Oops: 0000 [#1] PREEMPT SMP KASAN CPU: 1 PID: 28916 Comm: syz-executor.2 Not tainted 5.1.0-rc6+ #89 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:constant_test_bit arch/x86/include/asm/bitops.h:314 [inline] RIP: 0010:PageCompound include/linux/page-flags.h:186 [inline] RIP: 0010:isolate_freepages_block+0x1c0/0xd40 mm/compaction.c:579 Code: 01 d8 ff 4d 85 ed 0f 84 ef 07 00 00 e8 29 00 d8 ff 4c 89 e0 83 85 38 ff ff ff 01 48 c1 e8 03 42 80 3c 38 00 0f 85 31 0a 00 00 <4d> 8b 2c 24 31 ff 49 c1 ed 10 41 83 e5 01 44 89 ee e8 3a 01 d8 ff RSP: 0018:ffff88802b31eab8 EFLAGS: 00010246 RAX: 1ffffd4000669000 RBX: 00000000000cd200 RCX: ffffc9000a235000 RDX: 000000000001ca5e RSI: ffffffff81988cc7 RDI: 0000000000000001 RBP: ffff88802b31ebd8 R08: ffff88805af700c0 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000000 R12: ffffea0003348000 R13: 0000000000000000 R14: ffff88802b31f030 R15: dffffc0000000000 FS: 00007f61648dc700(0000) GS:ffff8880ae900000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: ffffea0003348000 CR3: 0000000037c64000 CR4: 00000000001426e0 Call Trace: fast_isolate_around mm/compaction.c:1243 [inline] fast_isolate_freepages mm/compaction.c:1418 [inline] isolate_freepages mm/compaction.c:1438 [inline] compaction_alloc+0x1aee/0x22e0 mm/compaction.c:1550 There is no reproducer and it is difficult to hit -- 1 crash every few days. The issue is very similar to the fix in commit 6b0868c820ff ("mm/compaction.c: correct zone boundary handling when resetting pageblock skip hints"). When isolating free pages around a target pageblock, the boundary handling is off by one and can stray into the next pageblock. Triggering the syzbot error requires that the end of pageblock is section or zone aligned, and that the next section is unpopulated. A more subtle consequence of the bug is that pageblocks were being improperly used as migration targets which potentially hurts fragmentation avoidance in the long-term one page at a time. A debugging patch revealed that it's definitely possible to stray outside of a pageblock which is not intended. While syzbot cannot be used to verify this patch, it was confirmed that the debugging warning no longer triggers with this patch applied. It has also been confirmed that the THP allocation stress tests are not degraded by this patch. Link: http://lkml.kernel.org/r/20190510182124.GI18914@techsingularity.net Fixes: e332f741a8dd ("mm, compaction: be selective about what pageblocks to clear skip hints") Signed-off-by: Mel Gorman Reported-by: syzbot+d84c80f9fe26a0f7a734@syzkaller.appspotmail.com Cc: Dmitry Vyukov Cc: Andrey Ryabinin Cc: Qian Cai Cc: Michal Hocko Cc: Vlastimil Babka Cc: # v5.1+ Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/compaction.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'mm') diff --git a/mm/compaction.c b/mm/compaction.c index cbac7277978a..9febc8cc84e7 100644 --- a/mm/compaction.c +++ b/mm/compaction.c @@ -1230,7 +1230,7 @@ fast_isolate_around(struct compact_control *cc, unsigned long pfn, unsigned long /* Pageblock boundaries */ start_pfn = pageblock_start_pfn(pfn); - end_pfn = min(start_pfn + pageblock_nr_pages, zone_end_pfn(cc->zone)); + end_pfn = min(pageblock_end_pfn(pfn), zone_end_pfn(cc->zone)) - 1; /* Scan before */ if (start_pfn != pfn) { @@ -1241,7 +1241,7 @@ fast_isolate_around(struct compact_control *cc, unsigned long pfn, unsigned long /* Scan after */ start_pfn = pfn + nr_isolated; - if (start_pfn != end_pfn) + if (start_pfn < end_pfn) isolate_freepages_block(cc, &start_pfn, end_pfn, &cc->freepages, 1, false); /* Skip this pageblock in the future as it's full or nearly full */ -- cgit