• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1====================
2The robust futex ABI
3====================
4
5:Author: Started by Paul Jackson <pj@sgi.com>
6
7
8Robust_futexes provide a mechanism that is used in addition to normal
9futexes, for kernel assist of cleanup of held locks on task exit.
10
11The interesting data as to what futexes a thread is holding is kept on a
12linked list in user space, where it can be updated efficiently as locks
13are taken and dropped, without kernel intervention.  The only additional
14kernel intervention required for robust_futexes above and beyond what is
15required for futexes is:
16
17 1) a one time call, per thread, to tell the kernel where its list of
18    held robust_futexes begins, and
19 2) internal kernel code at exit, to handle any listed locks held
20    by the exiting thread.
21
22The existing normal futexes already provide a "Fast Userspace Locking"
23mechanism, which handles uncontested locking without needing a system
24call, and handles contested locking by maintaining a list of waiting
25threads in the kernel.  Options on the sys_futex(2) system call support
26waiting on a particular futex, and waking up the next waiter on a
27particular futex.
28
29For robust_futexes to work, the user code (typically in a library such
30as glibc linked with the application) has to manage and place the
31necessary list elements exactly as the kernel expects them.  If it fails
32to do so, then improperly listed locks will not be cleaned up on exit,
33probably causing deadlock or other such failure of the other threads
34waiting on the same locks.
35
36A thread that anticipates possibly using robust_futexes should first
37issue the system call::
38
39    asmlinkage long
40    sys_set_robust_list(struct robust_list_head __user *head, size_t len);
41
42The pointer 'head' points to a structure in the threads address space
43consisting of three words.  Each word is 32 bits on 32 bit arch's, or 64
44bits on 64 bit arch's, and local byte order.  Each thread should have
45its own thread private 'head'.
46
47If a thread is running in 32 bit compatibility mode on a 64 native arch
48kernel, then it can actually have two such structures - one using 32 bit
49words for 32 bit compatibility mode, and one using 64 bit words for 64
50bit native mode.  The kernel, if it is a 64 bit kernel supporting 32 bit
51compatibility mode, will attempt to process both lists on each task
52exit, if the corresponding sys_set_robust_list() call has been made to
53setup that list.
54
55  The first word in the memory structure at 'head' contains a
56  pointer to a single linked list of 'lock entries', one per lock,
57  as described below.  If the list is empty, the pointer will point
58  to itself, 'head'.  The last 'lock entry' points back to the 'head'.
59
60  The second word, called 'offset', specifies the offset from the
61  address of the associated 'lock entry', plus or minus, of what will
62  be called the 'lock word', from that 'lock entry'.  The 'lock word'
63  is always a 32 bit word, unlike the other words above.  The 'lock
64  word' holds 3 flag bits in the upper 3 bits, and the thread id (TID)
65  of the thread holding the lock in the bottom 29 bits.  See further
66  below for a description of the flag bits.
67
68  The third word, called 'list_op_pending', contains transient copy of
69  the address of the 'lock entry', during list insertion and removal,
70  and is needed to correctly resolve races should a thread exit while
71  in the middle of a locking or unlocking operation.
72
73Each 'lock entry' on the single linked list starting at 'head' consists
74of just a single word, pointing to the next 'lock entry', or back to
75'head' if there are no more entries.  In addition, nearby to each 'lock
76entry', at an offset from the 'lock entry' specified by the 'offset'
77word, is one 'lock word'.
78
79The 'lock word' is always 32 bits, and is intended to be the same 32 bit
80lock variable used by the futex mechanism, in conjunction with
81robust_futexes.  The kernel will only be able to wakeup the next thread
82waiting for a lock on a threads exit if that next thread used the futex
83mechanism to register the address of that 'lock word' with the kernel.
84
85For each futex lock currently held by a thread, if it wants this
86robust_futex support for exit cleanup of that lock, it should have one
87'lock entry' on this list, with its associated 'lock word' at the
88specified 'offset'.  Should a thread die while holding any such locks,
89the kernel will walk this list, mark any such locks with a bit
90indicating their holder died, and wakeup the next thread waiting for
91that lock using the futex mechanism.
92
93When a thread has invoked the above system call to indicate it
94anticipates using robust_futexes, the kernel stores the passed in 'head'
95pointer for that task.  The task may retrieve that value later on by
96using the system call::
97
98    asmlinkage long
99    sys_get_robust_list(int pid, struct robust_list_head __user **head_ptr,
100                        size_t __user *len_ptr);
101
102It is anticipated that threads will use robust_futexes embedded in
103larger, user level locking structures, one per lock.  The kernel
104robust_futex mechanism doesn't care what else is in that structure, so
105long as the 'offset' to the 'lock word' is the same for all
106robust_futexes used by that thread.  The thread should link those locks
107it currently holds using the 'lock entry' pointers.  It may also have
108other links between the locks, such as the reverse side of a double
109linked list, but that doesn't matter to the kernel.
110
111By keeping its locks linked this way, on a list starting with a 'head'
112pointer known to the kernel, the kernel can provide to a thread the
113essential service available for robust_futexes, which is to help clean
114up locks held at the time of (a perhaps unexpectedly) exit.
115
116Actual locking and unlocking, during normal operations, is handled
117entirely by user level code in the contending threads, and by the
118existing futex mechanism to wait for, and wakeup, locks.  The kernels
119only essential involvement in robust_futexes is to remember where the
120list 'head' is, and to walk the list on thread exit, handling locks
121still held by the departing thread, as described below.
122
123There may exist thousands of futex lock structures in a threads shared
124memory, on various data structures, at a given point in time. Only those
125lock structures for locks currently held by that thread should be on
126that thread's robust_futex linked lock list a given time.
127
128A given futex lock structure in a user shared memory region may be held
129at different times by any of the threads with access to that region. The
130thread currently holding such a lock, if any, is marked with the threads
131TID in the lower 29 bits of the 'lock word'.
132
133When adding or removing a lock from its list of held locks, in order for
134the kernel to correctly handle lock cleanup regardless of when the task
135exits (perhaps it gets an unexpected signal 9 in the middle of
136manipulating this list), the user code must observe the following
137protocol on 'lock entry' insertion and removal:
138
139On insertion:
140
141 1) set the 'list_op_pending' word to the address of the 'lock entry'
142    to be inserted,
143 2) acquire the futex lock,
144 3) add the lock entry, with its thread id (TID) in the bottom 29 bits
145    of the 'lock word', to the linked list starting at 'head', and
146 4) clear the 'list_op_pending' word.
147
148On removal:
149
150 1) set the 'list_op_pending' word to the address of the 'lock entry'
151    to be removed,
152 2) remove the lock entry for this lock from the 'head' list,
153 3) release the futex lock, and
154 4) clear the 'lock_op_pending' word.
155
156On exit, the kernel will consider the address stored in
157'list_op_pending' and the address of each 'lock word' found by walking
158the list starting at 'head'.  For each such address, if the bottom 29
159bits of the 'lock word' at offset 'offset' from that address equals the
160exiting threads TID, then the kernel will do two things:
161
162 1) if bit 31 (0x80000000) is set in that word, then attempt a futex
163    wakeup on that address, which will waken the next thread that has
164    used to the futex mechanism to wait on that address, and
165 2) atomically set  bit 30 (0x40000000) in the 'lock word'.
166
167In the above, bit 31 was set by futex waiters on that lock to indicate
168they were waiting, and bit 30 is set by the kernel to indicate that the
169lock owner died holding the lock.
170
171The kernel exit code will silently stop scanning the list further if at
172any point:
173
174 1) the 'head' pointer or an subsequent linked list pointer
175    is not a valid address of a user space word
176 2) the calculated location of the 'lock word' (address plus
177    'offset') is not the valid address of a 32 bit user space
178    word
179 3) if the list contains more than 1 million (subject to
180    future kernel configuration changes) elements.
181
182When the kernel sees a list entry whose 'lock word' doesn't have the
183current threads TID in the lower 29 bits, it does nothing with that
184entry, and goes on to the next entry.
185
186Bit 29 (0x20000000) of the 'lock word' is reserved for future use.
187