185 lines · plain
1====================2The robust futex ABI3====================4 5:Author: Started by Paul Jackson <pj@sgi.com>6 7 8Robust_futexes provide a mechanism that is used in addition to normal9futexes, 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 a12linked list in user space, where it can be updated efficiently as locks13are taken and dropped, without kernel intervention. The only additional14kernel intervention required for robust_futexes above and beyond what is15required for futexes is:16 17 1) a one time call, per thread, to tell the kernel where its list of18 held robust_futexes begins, and19 2) internal kernel code at exit, to handle any listed locks held20 by the exiting thread.21 22The existing normal futexes already provide a "Fast Userspace Locking"23mechanism, which handles uncontested locking without needing a system24call, and handles contested locking by maintaining a list of waiting25threads in the kernel. Options on the sys_futex(2) system call support26waiting on a particular futex, and waking up the next waiter on a27particular futex.28 29For robust_futexes to work, the user code (typically in a library such30as glibc linked with the application) has to manage and place the31necessary list elements exactly as the kernel expects them. If it fails32to do so, then improperly listed locks will not be cleaned up on exit,33probably causing deadlock or other such failure of the other threads34waiting on the same locks.35 36A thread that anticipates possibly using robust_futexes should first37issue the system call::38 39 asmlinkage long40 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 space43consisting of three words. Each word is 32 bits on 32 bit arch's, or 6444bits on 64 bit arch's, and local byte order. Each thread should have45its own thread private 'head'.46 47If a thread is running in 32 bit compatibility mode on a 64 native arch48kernel, then it can actually have two such structures - one using 32 bit49words for 32 bit compatibility mode, and one using 64 bit words for 6450bit native mode. The kernel, if it is a 64 bit kernel supporting 32 bit51compatibility mode, will attempt to process both lists on each task52exit, if the corresponding sys_set_robust_list() call has been made to53setup that list.54 55 The first word in the memory structure at 'head' contains a56 pointer to a single linked list of 'lock entries', one per lock,57 as described below. If the list is empty, the pointer will point58 to itself, 'head'. The last 'lock entry' points back to the 'head'.59 60 The second word, called 'offset', specifies the offset from the61 address of the associated 'lock entry', plus or minus, of what will62 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 'lock64 word' holds 2 flag bits in the upper 2 bits, and the thread id (TID)65 of the thread holding the lock in the bottom 30 bits. See further66 below for a description of the flag bits.67 68 The third word, called 'list_op_pending', contains transient copy of69 the address of the 'lock entry', during list insertion and removal,70 and is needed to correctly resolve races should a thread exit while71 in the middle of a locking or unlocking operation.72 73Each 'lock entry' on the single linked list starting at 'head' consists74of just a single word, pointing to the next 'lock entry', or back to75'head' if there are no more entries. In addition, nearby to each 'lock76entry', 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 bit80lock variable used by the futex mechanism, in conjunction with81robust_futexes. The kernel will only be able to wakeup the next thread82waiting for a lock on a threads exit if that next thread used the futex83mechanism to register the address of that 'lock word' with the kernel.84 85For each futex lock currently held by a thread, if it wants this86robust_futex support for exit cleanup of that lock, it should have one87'lock entry' on this list, with its associated 'lock word' at the88specified 'offset'. Should a thread die while holding any such locks,89the kernel will walk this list, mark any such locks with a bit90indicating their holder died, and wakeup the next thread waiting for91that lock using the futex mechanism.92 93When a thread has invoked the above system call to indicate it94anticipates using robust_futexes, the kernel stores the passed in 'head'95pointer for that task. The task may retrieve that value later on by96using the system call::97 98 asmlinkage long99 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 in103larger, user level locking structures, one per lock. The kernel104robust_futex mechanism doesn't care what else is in that structure, so105long as the 'offset' to the 'lock word' is the same for all106robust_futexes used by that thread. The thread should link those locks107it currently holds using the 'lock entry' pointers. It may also have108other links between the locks, such as the reverse side of a double109linked 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 the113essential service available for robust_futexes, which is to help clean114up locks held at the time of (a perhaps unexpectedly) exit.115 116Actual locking and unlocking, during normal operations, is handled117entirely by user level code in the contending threads, and by the118existing futex mechanism to wait for, and wakeup, locks. The kernels119only essential involvement in robust_futexes is to remember where the120list 'head' is, and to walk the list on thread exit, handling locks121still held by the departing thread, as described below.122 123There may exist thousands of futex lock structures in a threads shared124memory, on various data structures, at a given point in time. Only those125lock structures for locks currently held by that thread should be on126that thread's robust_futex linked lock list a given time.127 128A given futex lock structure in a user shared memory region may be held129at different times by any of the threads with access to that region. The130thread currently holding such a lock, if any, is marked with the threads131TID in the lower 30 bits of the 'lock word'.132 133When adding or removing a lock from its list of held locks, in order for134the kernel to correctly handle lock cleanup regardless of when the task135exits (perhaps it gets an unexpected signal 9 in the middle of136manipulating this list), the user code must observe the following137protocol 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 30 bits145 of the 'lock word', to the linked list starting at 'head', and146 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, and154 4) clear the 'lock_op_pending' word.155 156On exit, the kernel will consider the address stored in157'list_op_pending' and the address of each 'lock word' found by walking158the list starting at 'head'. For each such address, if the bottom 30159bits of the 'lock word' at offset 'offset' from that address equals the160exiting threads TID, then the kernel will do two things:161 162 1) if bit 31 (0x80000000) is set in that word, then attempt a futex163 wakeup on that address, which will waken the next thread that has164 used to the futex mechanism to wait on that address, and165 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 indicate168they were waiting, and bit 30 is set by the kernel to indicate that the169lock owner died holding the lock.170 171The kernel exit code will silently stop scanning the list further if at172any point:173 174 1) the 'head' pointer or an subsequent linked list pointer175 is not a valid address of a user space word176 2) the calculated location of the 'lock word' (address plus177 'offset') is not the valid address of a 32 bit user space178 word179 3) if the list contains more than 1 million (subject to180 future kernel configuration changes) elements.181 182When the kernel sees a list entry whose 'lock word' doesn't have the183current threads TID in the lower 30 bits, it does nothing with that184entry, and goes on to the next entry.185