ia64/linux-2.6.18-xen.hg

view Documentation/lockdep-design.txt @ 854:950b9eb27661

usbback: fix urb interval value for interrupt urbs.

Signed-off-by: Noboru Iwamatsu <n_iwamatsu@jp.fujitsu.com>
author Keir Fraser <keir.fraser@citrix.com>
date Mon Apr 06 13:51:20 2009 +0100 (2009-04-06)
parents 831230e53067
children
line source
1 Runtime locking correctness validator
2 =====================================
4 started by Ingo Molnar <mingo@redhat.com>
5 additions by Arjan van de Ven <arjan@linux.intel.com>
7 Lock-class
8 ----------
10 The basic object the validator operates upon is a 'class' of locks.
12 A class of locks is a group of locks that are logically the same with
13 respect to locking rules, even if the locks may have multiple (possibly
14 tens of thousands of) instantiations. For example a lock in the inode
15 struct is one class, while each inode has its own instantiation of that
16 lock class.
18 The validator tracks the 'state' of lock-classes, and it tracks
19 dependencies between different lock-classes. The validator maintains a
20 rolling proof that the state and the dependencies are correct.
22 Unlike an lock instantiation, the lock-class itself never goes away: when
23 a lock-class is used for the first time after bootup it gets registered,
24 and all subsequent uses of that lock-class will be attached to this
25 lock-class.
27 State
28 -----
30 The validator tracks lock-class usage history into 5 separate state bits:
32 - 'ever held in hardirq context' [ == hardirq-safe ]
33 - 'ever held in softirq context' [ == softirq-safe ]
34 - 'ever held with hardirqs enabled' [ == hardirq-unsafe ]
35 - 'ever held with softirqs and hardirqs enabled' [ == softirq-unsafe ]
37 - 'ever used' [ == !unused ]
39 Single-lock state rules:
40 ------------------------
42 A softirq-unsafe lock-class is automatically hardirq-unsafe as well. The
43 following states are exclusive, and only one of them is allowed to be
44 set for any lock-class:
46 <hardirq-safe> and <hardirq-unsafe>
47 <softirq-safe> and <softirq-unsafe>
49 The validator detects and reports lock usage that violate these
50 single-lock state rules.
52 Multi-lock dependency rules:
53 ----------------------------
55 The same lock-class must not be acquired twice, because this could lead
56 to lock recursion deadlocks.
58 Furthermore, two locks may not be taken in different order:
60 <L1> -> <L2>
61 <L2> -> <L1>
63 because this could lead to lock inversion deadlocks. (The validator
64 finds such dependencies in arbitrary complexity, i.e. there can be any
65 other locking sequence between the acquire-lock operations, the
66 validator will still track all dependencies between locks.)
68 Furthermore, the following usage based lock dependencies are not allowed
69 between any two lock-classes:
71 <hardirq-safe> -> <hardirq-unsafe>
72 <softirq-safe> -> <softirq-unsafe>
74 The first rule comes from the fact the a hardirq-safe lock could be
75 taken by a hardirq context, interrupting a hardirq-unsafe lock - and
76 thus could result in a lock inversion deadlock. Likewise, a softirq-safe
77 lock could be taken by an softirq context, interrupting a softirq-unsafe
78 lock.
80 The above rules are enforced for any locking sequence that occurs in the
81 kernel: when acquiring a new lock, the validator checks whether there is
82 any rule violation between the new lock and any of the held locks.
84 When a lock-class changes its state, the following aspects of the above
85 dependency rules are enforced:
87 - if a new hardirq-safe lock is discovered, we check whether it
88 took any hardirq-unsafe lock in the past.
90 - if a new softirq-safe lock is discovered, we check whether it took
91 any softirq-unsafe lock in the past.
93 - if a new hardirq-unsafe lock is discovered, we check whether any
94 hardirq-safe lock took it in the past.
96 - if a new softirq-unsafe lock is discovered, we check whether any
97 softirq-safe lock took it in the past.
99 (Again, we do these checks too on the basis that an interrupt context
100 could interrupt _any_ of the irq-unsafe or hardirq-unsafe locks, which
101 could lead to a lock inversion deadlock - even if that lock scenario did
102 not trigger in practice yet.)
104 Exception: Nested data dependencies leading to nested locking
105 -------------------------------------------------------------
107 There are a few cases where the Linux kernel acquires more than one
108 instance of the same lock-class. Such cases typically happen when there
109 is some sort of hierarchy within objects of the same type. In these
110 cases there is an inherent "natural" ordering between the two objects
111 (defined by the properties of the hierarchy), and the kernel grabs the
112 locks in this fixed order on each of the objects.
114 An example of such an object hieararchy that results in "nested locking"
115 is that of a "whole disk" block-dev object and a "partition" block-dev
116 object; the partition is "part of" the whole device and as long as one
117 always takes the whole disk lock as a higher lock than the partition
118 lock, the lock ordering is fully correct. The validator does not
119 automatically detect this natural ordering, as the locking rule behind
120 the ordering is not static.
122 In order to teach the validator about this correct usage model, new
123 versions of the various locking primitives were added that allow you to
124 specify a "nesting level". An example call, for the block device mutex,
125 looks like this:
127 enum bdev_bd_mutex_lock_class
128 {
129 BD_MUTEX_NORMAL,
130 BD_MUTEX_WHOLE,
131 BD_MUTEX_PARTITION
132 };
134 mutex_lock_nested(&bdev->bd_contains->bd_mutex, BD_MUTEX_PARTITION);
136 In this case the locking is done on a bdev object that is known to be a
137 partition.
139 The validator treats a lock that is taken in such a nested fasion as a
140 separate (sub)class for the purposes of validation.
142 Note: When changing code to use the _nested() primitives, be careful and
143 check really thoroughly that the hiearchy is correctly mapped; otherwise
144 you can get false positives or false negatives.
146 Proof of 100% correctness:
147 --------------------------
149 The validator achieves perfect, mathematical 'closure' (proof of locking
150 correctness) in the sense that for every simple, standalone single-task
151 locking sequence that occured at least once during the lifetime of the
152 kernel, the validator proves it with a 100% certainty that no
153 combination and timing of these locking sequences can cause any class of
154 lock related deadlock. [*]
156 I.e. complex multi-CPU and multi-task locking scenarios do not have to
157 occur in practice to prove a deadlock: only the simple 'component'
158 locking chains have to occur at least once (anytime, in any
159 task/context) for the validator to be able to prove correctness. (For
160 example, complex deadlocks that would normally need more than 3 CPUs and
161 a very unlikely constellation of tasks, irq-contexts and timings to
162 occur, can be detected on a plain, lightly loaded single-CPU system as
163 well!)
165 This radically decreases the complexity of locking related QA of the
166 kernel: what has to be done during QA is to trigger as many "simple"
167 single-task locking dependencies in the kernel as possible, at least
168 once, to prove locking correctness - instead of having to trigger every
169 possible combination of locking interaction between CPUs, combined with
170 every possible hardirq and softirq nesting scenario (which is impossible
171 to do in practice).
173 [*] assuming that the validator itself is 100% correct, and no other
174 part of the system corrupts the state of the validator in any way.
175 We also assume that all NMI/SMM paths [which could interrupt
176 even hardirq-disabled codepaths] are correct and do not interfere
177 with the validator. We also assume that the 64-bit 'chain hash'
178 value is unique for every lock-chain in the system. Also, lock
179 recursion must not be higher than 20.
181 Performance:
182 ------------
184 The above rules require _massive_ amounts of runtime checking. If we did
185 that for every lock taken and for every irqs-enable event, it would
186 render the system practically unusably slow. The complexity of checking
187 is O(N^2), so even with just a few hundred lock-classes we'd have to do
188 tens of thousands of checks for every event.
190 This problem is solved by checking any given 'locking scenario' (unique
191 sequence of locks taken after each other) only once. A simple stack of
192 held locks is maintained, and a lightweight 64-bit hash value is
193 calculated, which hash is unique for every lock chain. The hash value,
194 when the chain is validated for the first time, is then put into a hash
195 table, which hash-table can be checked in a lockfree manner. If the
196 locking chain occurs again later on, the hash table tells us that we
197 dont have to validate the chain again.