1 ============================
2 Subsystem Trace Points: kmem
3 ============================
4
5 The kmem tracing system captures events relate
6 within the kernel. Broadly speaking there are
7
8 - Slab allocation of small objects of unknow
9 - Slab allocation of small objects of known
10 - Page allocation
11 - Per-CPU Allocator Activity
12 - External Fragmentation
13
14 This document describes what each of the trace
15 might be useful.
16
17 1. Slab allocation of small objects of unknown
18 ==============================================
19 ::
20
21 kmalloc call_site=%lx ptr=%p b
22 kmalloc_node call_site=%lx ptr=%p bytes_req
23 kfree call_site=%lx ptr=%p
24
25 Heavy activity for these events may indicate t
26 justified, particularly if kmalloc slab pages
27 internal fragmented as a result of the allocat
28 kmalloc with kfree, it may be possible to iden
29 the allocation sites were.
30
31
32 2. Slab allocation of small objects of known t
33 ==============================================
34 ::
35
36 kmem_cache_alloc call_site=%lx ptr=%p b
37 kmem_cache_alloc_node call_site=%lx ptr=%p b
38 kmem_cache_free call_site=%lx
39
40 These events are similar in usage to the kmall
41 it is likely easier to pin the event down to a
42 of writing, no information is available on wha
43 but the call_site can usually be used to extra
44
45 3. Page allocation
46 ==================
47 ::
48
49 mm_page_alloc page=%p pfn=%lu orde
50 mm_page_alloc_zone_locked page=%p pfn=%lu or
51 mm_page_free page=%p pfn=%lu orde
52 mm_page_free_batched page=%p pfn=%lu orde
53
54 These four events deal with page allocation an
55 a simple indicator of page allocator activity.
56 the per-CPU allocator (high performance) or th
57
58 If pages are allocated directly from the buddy
59 mm_page_alloc_zone_locked event is triggered.
60 amounts of activity imply high activity on the
61 impairs performance by disabling interrupts, d
62 CPUs and serialising many CPUs.
63
64 When a page is freed directly by the caller, t
65 is triggered. Significant amounts of activity
66 callers should be batching their activities.
67
68 When pages are freed in batch, the also mm_pag
69 Broadly speaking, pages are taken off the LRU
70 freed in batch with a page list. Significant a
71 indicate that the system is under memory press
72 contention on the lruvec->lru_lock.
73
74 4. Per-CPU Allocator Activity
75 =============================
76 ::
77
78 mm_page_alloc_zone_locked page=%p pfn=%l
79 mm_page_pcpu_drain page=%p pfn=%l
80
81 In front of the page allocator is a per-cpu pa
82 for order-0 pages, reduces contention on the z
83 amount of writing on struct page.
84
85 When a per-CPU list is empty or pages of the w
86 the zone->lock will be taken once and the per-
87 triggered is mm_page_alloc_zone_locked for eac
88 event indicating whether it is for a percpu_re
89
90 When the per-CPU list is too full, a number of
91 which triggers a mm_page_pcpu_drain event.
92
93 The individual nature of the events is so that
94 between allocation and freeing. A number of dr
95 consecutively imply the zone->lock being taken
96 refills and drains could imply an imbalance be
97 is being concentrated in one place. It could a
98 lists should be a larger size. Finally, large
99 and drains on another could be a factor in cau
100 line bounces due to writes between CPUs and wo
101 can be allocated and freed on the same CPU thr
102
103 5. External Fragmentation
104 =========================
105 ::
106
107 mm_page_alloc_extfrag page=%p pfn=%l
108
109 External fragmentation affects whether a high-
110 successful or not. For some types of hardware,
111 it is avoided where possible. If the system is
112 to be able to resize the pool over the lifetim
113 is important.
114
115 Large numbers of this event implies that memor
116 high-order allocations will start failing at s
117 means of reducing the occurrence of this event
118 min_free_kbytes in increments of 3*pageblock_s
119 pageblock_size is usually the size of the defa
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.