Priority Inversion on Mars
13 likes39,677 views
The Mars Pathfinder mission successfully demonstrated new landing techniques and returned valuable data from the Martian surface. However, it experienced issues with priority inversion in its VxWorks real-time operating system. The lower priority weather data collection task would occasionally prevent the higher priority communication task from completing before the next cycle began, resetting the system. Engineers traced the problem to the use of VxWorks' select() call to wait for I/O from multiple devices, allowing long-running lower priority tasks to block critical higher priority tasks.
Priority Inversion on Mars
- 1. Priority Inversion on Mars Jim Huang ( 黃敬群 ) <jserv.tw@gmail.com> Mar 15, 2016 / NCKU, Taiwan
- 2. Mars Pathfinder• Mission – Demonstrate new landing techniques: parachute and airbags – Take pictures – Analyze soil samples – Demonstrate mobile robot technology: Sojourner • Major success on all fronts – Returned 2.3 billion bits of information – 16,500 images from the Lander – 550 images from the Rover – 15 chemical analyses of rocks & soil – Lots of weather data – Both Lander and Rover outlived their design life Sojourner Rover Operators: NASA and JPL (Jet Propulsion Laboratory ) Low-cost (~$150 million) planetary discovery mission
- 3. Mars Pathfinder • Mars Pathfinder was originally designed as a technology demonstration of a way to deliver an instrumented lander and a free-ranging robotic rover to the surface of the red planet. • Due to limited funds, Pathfinder’s development had to be dramatically different from the way in which previous spacecraft had been developed. • Instead of the traditional 8- to 10-year schedule and $1-billion-plus budget, Pathfinder was developed in 3 years for less than $150 million = the cost of some Hollywood movies!
- 4. Pictures taken from an early Mars rover
- 5. Mars Pathfinder Timeline • November 16, 1996 – Russian Mars '96 orbiter/landers launched. • November 17, 1996 – Mars '96 fails to achieve insertion into Mars cruise trajectory and re-enters the Earth's atmosphere. • December 4, 1996 – Mars Pathfinder launched. • July 4, 1997 – Mars Pathfinder lands on Mars and begins a successful mission. • September 27, 1997 – last successful data transmission from Mars Pathfinder
- 6. New York Times: July 15, 1997 Mary Beth Murrill, a spokeswoman for NASA's Jet Propulsion Laboratory, said transmission of the panoramic shot took “a lot of processing power.” She likened the data overload to what happens with a personal computer “when we ask it to do too many things at once.” The project manager, Brian Muirhead, said that to prevent a recurrence, controllers would schedule activities one after another, instead of at the same time. It was the second time the Pathfinder's computer had reset itself while trying to carry out several activities at once. In response, controllers reprogrammed the computer over the weekend to slow down the rate of activities and avoid another reset. But today, about an hour into a two-hour transmission session, it happened again.
- 7. Pathfinder Configurations • Spacecraft ran IBM RS6000 processor and WindRiver's VxWorks RTOS – 20 MIPS 128 MB of DRAM for storage of flight software and engineering and science data, including images and rover information. – 6 MB ROM stored flight software and time-critical data. • Hard real-time OS with concurrent execution of thread – Threads have priorities and are preemptible • Tasks on Pathfinder were executed as threads with priorities – that were assigned reflecting the relative urgency of tasks. • Pathfinder contained an "information bus” – a shared memory area used for passing information between different components of the spacecraft.
- 9. Rover Sojourner • The rover, capable of autonomous navigation and performance of tasks, communicated with Earth via the lander. • Sojourner’s control system was built around an Intel 80C85, with a computing speed of 0.1 MIPS and 500 KB of RAM.
- 12. VMEbus CPU RS6000 Radio Camera Mil1553 Cruise stage controls thrusters, valves, a sun sensor, a star scanner Lander interface to accelerometers, a radar altimeter, an instrument for meteorological science known as the ASI/MET Mil1553: specific paradigm: the software will schedule activity at an 8 Hz rate. This **feature** dictated the architecture of the software which controls both the 1553 bus and the devices attached to it. Mil1553: specific paradigm: the software will schedule activity at an 8 Hz rate. This **feature** dictated the architecture of the software which controls both the 1553 bus and the devices attached to it. Mil 1553 bus
- 13. Pathfinder Configurations CPU 1553 bus . Vme bus Interface cards (radio, camera etc.) Cruise part of the equipment Lander part of the equipment
- 14. Pathfinder used VxWorks RTOS • Threads for the 1553 bus for data collection, scheduled on every 1/8th sec cycle. • 3 periodic tasks – Task 1 – Information Bus Thread: Bus Manager high frequency, high priority – Task 2 – Communication Thread medium frequency / priority, high execution time – Task 3 – Weather Thread: Geological Data Gatherer low frequency, low priority • Each checks if the other executed and completed in the previous cycle – If the check fails, this is a violation of a hard real-time guarantee and the system is reset
- 15. NASA Pathfinder • fault-tolerance – a watchdog timer was used to reset the system in the event that the computer / software locks up – essential design feature (no going to Mars to reboot) – “watched” for hang ups on the highest priority task • inter-task communication – a shared resource (memory) was used to pass data from the data gatherer (task 3) to the communicator (task 2) via the bus manager (task 1).
- 16. Pathfinder Problem • Within a few days of landing, when Pathfinder started gathering meteorological data, spacecraft began experiencing total system resets • This resulted in loss of data collected during each cycle • JPL engineers had exact replica of the spacecraft in their lab • They turned on the tracing feature of VxWorks – All system events such as context switches, uses of synchronization objects, and interrupts traced. – Tracing disabled on the actual spacecraft because generates too much data volume. • After 18 hours of execution, early next morning when all but one engineer had gone home, the symptom was reproduced.
- 17. Pathfinder Problem • Most of the time this combination worked fine • However, with the following scenario: – Data gathering occurs, grab the bus – Shared memory buffer full, retrieval to private memory • This is blocked because of the bus mutex – Period communication task is issued • This is preempted because of lower priority – Data gathering task takes its time – Retrieval task time out due to watchdog timer • System reset!
- 18. VxWorks RTOS • Multiple tasks, each with an associated priority – Higher priority tasks get to run before lower-priority tasks • Information bus – shared memory area used by various tasks – Thread must obtain mutex to write data to the info bus – a monitor Information Bus Mutex Weather Data Thread Communication Thread Information Bus Thread Obtain mutex; write data Wait for mutex to read data
- 19. VxWorks RTOS • Multiple tasks, each with an associated priority – Higher priority tasks get to run before lower-priority tasks • Information bus – shared memory area used by various tasks – Thread must obtain mutex to write data to the info bus – a monitor Information Bus Mutex Weather Data Thread Communication Thread Information Bus Thread Free mutex
- 20. VxWorks RTOS • Multiple tasks, each with an associated priority – Higher priority tasks get to run before lower-priority tasks • Information bus – shared memory area used by various tasks – Thread must obtain mutex to write data to the info bus – a monitor Information Bus Mutex Weather Data Thread Communication Thread Information Bus Thread Lock mutex and read data
- 21. Priority Inversion Information Bus Mutex Weather Data Thread Communication Thread Information Bus Thread Low priority Med Priority High priority
- 22. Priority Inversion Information Bus Mutex Weather Data Thread Communication Thread Information Bus Thread Low priority Med Priority High priority Interrupt! Schedule comm thread ... long running operation
- 23. Priority Inversion Information Bus Mutex • What happens when threads have different priorities? • Comm thread runs for a long time • Comm thread has higher priority than weather data thread • But ... the high priority info bus thread is stuck waiting! • This is called priority inversion Weather Data Thread Communication Thread Information Bus Thread Low priority Med Priority High priority
- 24. Pathfinder incident: Priority Inversion Classical priority inversion problem due to shared system bus! task 1 (high) task 2 (med) task 3 (low) task 3 locks S1 task 1 attempts to lock S1 (blocked) task 3 unlocks S1 (task 1 unblocked) task 1 locks S1 task 1 unlocks S1 CS1 ncsp CS1 ncsp CS1 CS1 Z3,1 Z1,1 • tasks 1 and 3 share a resource (S1) • prio(task1) >prio(task2) >prio(task3) • Task 2 can run for any amount of time… it blocks Task 3 from finishing and unlocking resource needed by task 1.
- 25. Priority Inversion Unbounded Priority Inversion
- 26. Variant: Air transportation • Priority(my_plane) < Priority(obama_plane) • I arrived at the airport first and called sem_wait(&runway_mutex). • Obama arrived at the airport after me and called sem_wait(&runway_mutex). • My plane stopped working. • Obama now has to wait for the engineers to fix my plane and for my plane to take off.
- 27. Look into deeper • VxWorks 5.x • 2 tasks to control the 1553 bus and the attached instruments. – bc_sched task (called the bus scheduler) a task controlled the setup of transactions on the 1553 bus – bc_dist task (for distribution) task also referred as the “communication task” • handles the collection of the transaction results i.e. the data.
- 28. 28 Marsrobot general communication pattern time t1 - bus hardware starts via hardware control on the 8 Hz boundary. The transactions for the this cycle had been set up by the previous execution of the bc_sched task. t2 - 1553 traffic is complete and the bc_dist task is awakened. t3 - bc_dist task has completed all of the data distribution t4 - bc_sched task is awakened to setup transactions for the next cycle t5 - bc_sched activity is complete t1 t2 t1t3 t4 t5 125 ms (8 Hz) Spacecraft functions LOW priority bc-dist MEDIUM priority bc-sched HIGH priority Mil 1553 transaction Science functions (ASI/MET, …) LOWEST priority Check order! bc-dist bc-sched
- 29. 1553 communication • Powered 1553 devices deliver data. • Tasks in the system that access the information collected over the 1553 do so via a double buffered shared memory mechanism into which the bc_dist task places the latest data. • The exception to this is the ASI/MET task which is delivered its information via an interprocess communication mechanism (IPC). The IPC mechanism uses the VxWorks pipe() facility.
- 30. VMEbus CPU RS6000 Radio Camera Mil1553 Cruise stage controls thrusters, valves, a sun sensor, a star scanner Lander interface to accelerometers, a radar altimeter, an instrument for meteorological science known as the ASI/MET MEM Packed buffer D-Buffer D-Buffer D-Buffer IPCPIPE File Descriptor List
- 31. Dedicated systems' tasking model P3 Thread P4 started thread Shared data D Mailbox D4 message queue data usage
- 32. bc-dist bc-sched Spacecraft function taks 1 Spacecraft function tasks Science function task 1 Science function tasks Shared data ASI/MET task Shared data Shared data Shared data Mil 1553 trans setup pipe Mil 1553 data setup File Descriptor Table System_mutex
- 33. IPC Mechanism • Tasks wait on one or more IPC "queues" for messages to arrive using the VxWorks select() mechanism to wait for message arrival. • Multiple queues are used when both high and lower priority messages are required. • Most of the IPC traffic in the system is not for the delivery of real-time data. The exception to this is the use of the IPC mechanism with the ASI/MET task. • The cause of the reset on Mars was in the use and configuration of the IPC mechanism.
- 34. VxWorks select() • Pending on multiple file descriptiors: this routine permits a task to pend until one of a set of file descriptors becomes available • Wait for multiple I/O devices (task level and driver level) • file descriptors pReadFds, pWriteFds • Bits set in pReadFds will cause select() to pend until data becomes available on any of the corresponding file descriptors. • Bits set in pWriteFds will cause select() to pend until any of the corresponding file descriptors becomes available.
- 35. Marsrobot Shared ressource for Communication Using select() Thread AThread A Thread BThread B Thread CThread C Low priority thread Lowest priority sporadic meteo thread ASI/MET Middle priority long lasting Comm thread bc_dist Different I/O channels System_mutex
- 36. The problem, again • Priority inversion occurs when a thread of low priority blocks the execution of threads of higher priority. • Two flavours: – bounded priority inversion (common & relatively harmless) – unbounded priority inversion (insidious & potentially disastrous)
- 37. Bounded Priority Inversion • Suppose a high priority thread becomes blocked waiting for an event to happen. A low priority thread then starts to run and in doing so obtains (i.e locks) a mutex for a shared resource. While the mutex is locked by the low priority thread, the event occurs waking up the high priority thread. • Inversion takes place when the high priority thread tries to lock the mutex held by the low priority thread. In effect the high priority thread must wait for the low priority thread to finish. • It is called bounded inversion since the inversion is limited by the duration of the critical section.
- 38. ISR A HIGH: TASK A (40) LOW TASK C (30) Bounded priority inversion time Lock MUTEX (m) Lock MUTEX (m) run blocked ready UnLock MUTEX (m) Bounded inversion time
- 39. Unbounded Priority Inversion • Here the high level thread can be blocked indefinitely by a medium priority thread. • The medium level thread running prevents the low priority thread from releasing the lock. • All that is required for this to happen is that while the low level thread has locked the mutex, the medium level thread becomes unblocked, preempting the low level thread. • The medium level thread then runs indefinitely.
- 40. ISR A HIGH: TASK A (40) LOW TASK C (30) Unbounded priority inversion time Lock MUTEX (m) Lock MUTEX (m) run blocked ready MIDDLE: TASK B (35) ISR B Unbounded inversion time
- 41. Pathfinder Failure • The failure was identified by the spacecraft as a failure of the bc_dist task to complete its execution before the bc_sched task started. • The reaction to this by the spacecraft was to reset the computer. • This reset reinitializes all of the hardware and software. It also terminates the execution of the current ground commanded activities. No science or engineering data is lost that has already been collected (the data in RAM is recovered so long as power is not lost). • The remainder of the activities for that day were not accomplished until the next day.
- 42. Comm thread pre-emption HIGH: Bus thread bc_sched Marsrobot normal operation time Lock SystemMUTEX (m) run blocked ready MIDDLE Comm thread bc_dist Comm thread Pre-emption End of cycle OK! LOWEST Meteo thead LOW Tasks Un-Lock SystemMUTEX (m)
- 43. Comm thread pre-emption HIGH: Bus thread bc_sched LOW Tasks Marsrobot priority inversion time Lock SystemMUTEX (m) run blocked ready MIDDLE Comm thread bc_dist Not OK! End of cycle Comm thread Pre-emption System Reset LOWEST Meteo thead Lock SystemMUTEX (m) Lock SystemMUTEX (m) Un-Lock SystemMUTEX (m)
- 44. Priority Inversion • The higher priority bc_dist task was blocked by the much lower priority ASI/MET task that was holding a shared resource. • The ASI/MET task had acquired this resource and then been preempted by several of the medium priority tasks. • When the bc_sched task was activated, to setup the transactions for the next 1553 bus cycle, it detected that the bc_dist task had not completed its execution. • The resource that caused this problem was a mutex (here called system_mutex) used within the select() mechanism to control access to the list of file descriptors that the select() mechanism was to wait on.
- 45. Priority Inversion • The select() mechanism creates a system_mutex to protect the "wait list" of file descriptors for those devices which support select(). • The VxWorks pipe mechanism is such a device and the IPC mechanism used is based on using pipes. • The ASI/MET task had called select(), which had called pipeIoctl(), which had called selNodeAdd(), which was in the process of giving the system_mutex. • The ASI/ MET task was preempted and semGive() was not completed. • Several medium priority tasks ran until the bc_dist task was activated. • The bc_dist task attempted to send the newest ASI/MET data via the IPC mechanism which called pipeWrite(). • pipeWrite() blocked, taking the system_mutex. More of the medium priority tasks ran, still not allowing the ASI/MET task to run, until the bc_sched task was awakened. • At that point, the bc_sched task determined that the bc_dist task had not completed its cycle (a hard deadline in the system) and declared the error that initiated the reset.
- 47. As suspected, the Pathfinder computer, struggling with several activities at once, reset itself each time it could not carry out low-priority tasks in the allotted time. A reset is a safety feature similar to hitting a reset button on a home computer. The low-priority task that kept tripping it up was the transfer of temperature and wind measurements from sensors to an electronics board and then into the computer. The solution is to raise the task's priority through some reprogramming, Mr. Muirhead said.
- 48. What is the Fix? • Problem with priority inversion: – A high priority thread is stuck waiting for a low priority thread to finish its work – In this case, the (medium priority) thread was holding up the low-priority thread • General solution: Priority inheritance – If waiting for a low priority thread, allow that thread to inherit the higher priority – High priority thread “donates” its priority to the low priority thread • Why can it fix the problem? – Medium priority comm task cannot preempt weather task – Weather task inherits high priority while it is being waited on
- 49. What was the problem fixed? • JPL had a replica of the Pathfinder system on the ground – Special tracing mode maintrains logs of all interesting system events • e.g., context switches, mutex lock/unlock, interrupts – After much testing were able to replicate the problem in the lab • VxWorks mutex objects have an optional priority inheritance flag – Engineers were able to upload a patch to set this flag on the info bus mutex – After the fix, no more system resets occurred • Lessons: – Automatically reset system to “known good” state if things run amuck • Far better than hanging or crashing – Ability to trace execution of complex multithreaded code is useful – Think through all possible thread interactions carefully!!
- 50. Debug the problem • On replica on earth • Total Tracing on – Context switches – Uses of synchronisation objects – Interrupts • Took time to reproduce the error • Trace analyses ==>priority inversion problem Ty Hardware ASI/MET Tx VxWorks 5.x Physical I/O (BIOS) TICKER routine TRACE bc_dist bc_sched
- 51. Bug Detection • The software that flies on Mars Pathfinder has several debug features within it that are used in the lab but are not used on the flight spacecraft (not used because some of them produce more information than we can send back to Earth). • These features remain in the software by design because JPL strongly believes in the "test what you fly and fly what you test" philosophy.
- 52. Deadlock • With priority inversion, eventually the system makes progress – e.g., Comm thread eventually finishes and rest of system proceeds – Pathfinder watchdog timer reset the system too quickly! • A far more serious situation is deadlock – Two (or more) threads waiting for each other – None of the deadlocked threads ever make progress Mutex 1 Thread 1 Thread 2Mutex 2 holds holds waits for waits for
- 53. Deadlock Definition • Two kinds of resources: – Preemptible: Can take away from a thread • e.g., the CPU – Non-preemptible: Can't take away from a thread • e.g., mutex, lock, virtual memory region, etc. • Why isn't it safe to forcibly take a lock away from a thread • Starvation – A thread never makes progress because other threads are using a resource it needs • Deadlock – A circular waiting for resources • Thread A waits for Thread B • Thread B waits for Thread A • Starvation ≠ Deadlock
- 54. Conditions for Deadlock • Limited access to a resource – Means some threads will have to wait to access a shared resource • No preemption – Means resource cannot be forcibly taken away from a thread • Multiple independent requests – Means a thread can wait for some resources while holding others • Circular dependency graph – Just as in previous example • Without all of these conditions, can't have deadlock! – Suggests several ways to get rid of deadlock
- 55. Get rid of deadlock • Unlimited access to a resource? – Requires that all resources allow arbitrary number of concurrent accesses • Probably not too feasible! • Always allow preemption? – Is it safe to let multiple threads into a critical section? • No multiple independent requests? – This might work! – Require that threads grab all resources they need before using any of them! • Not allowed to wait while holding some resources! • No circular chains of requests? – This might work too! – Require threads to grab resources in some predefined order!
- 58. Resource Access Protocols • Critical sections: sections of code at which exclusive access to some resource must be guaranteed. • Can be guaranteed with semaphores S. P(S) V(S) P(S) V(S) P(S) checks semaphore to see if resource is available and if yes, sets S to „used“. Uninterruptable operations! If no, calling task has to wait. V(S): sets S to „unused“ and starts sleeping task (if any). P(S) checks semaphore to see if resource is available and if yes, sets S to „used“. Uninterruptable operations! If no, calling task has to wait. V(S): sets S to „unused“ and starts sleeping task (if any). Exclusive access to resource guarded by S Task 1 Task 2
- 59. Reference • Mars pathfinder failure, Martin TIMMERMAN • Operating Systems, Paulo Marques, Departamento de Eng. Informática, Universidade de Coimbra • Synchronization Problems and Deadlock, Matt Welsh • Mike Jones article "What really happened on Mars?" http://research.microsoft.com/~mbj/Mars_Pathfinder/