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