133 lines · plain
1.. SPDX-License-Identifier: GPL-2.02 3.. include:: <isonum.txt>4 5===============================6Bus lock detection and handling7===============================8 9:Copyright: |copy| 2021 Intel Corporation10:Authors: - Fenghua Yu <fenghua.yu@intel.com>11 - Tony Luck <tony.luck@intel.com>12 13Problem14=======15 16A split lock is any atomic operation whose operand crosses two cache lines.17Since the operand spans two cache lines and the operation must be atomic,18the system locks the bus while the CPU accesses the two cache lines.19 20A bus lock is acquired through either split locked access to writeback (WB)21memory or any locked access to non-WB memory. This is typically thousands of22cycles slower than an atomic operation within a cache line. It also disrupts23performance on other cores and brings the whole system to its knees.24 25Detection26=========27 28Intel processors may support either or both of the following hardware29mechanisms to detect split locks and bus locks.30 31#AC exception for split lock detection32--------------------------------------33 34Beginning with the Tremont Atom CPU split lock operations may raise an35Alignment Check (#AC) exception when a split lock operation is attempted.36 37#DB exception for bus lock detection38------------------------------------39 40Some CPUs have the ability to notify the kernel by an #DB trap after a user41instruction acquires a bus lock and is executed. This allows the kernel to42terminate the application or to enforce throttling.43 44Software handling45=================46 47The kernel #AC and #DB handlers handle bus lock based on the kernel48parameter "split_lock_detect". Here is a summary of different options:49 50+------------------+----------------------------+-----------------------+51|split_lock_detect=|#AC for split lock |#DB for bus lock |52+------------------+----------------------------+-----------------------+53|off |Do nothing |Do nothing |54+------------------+----------------------------+-----------------------+55|warn |Kernel OOPs |Warn once per task and |56|(default) |Warn once per task, add a |and continues to run. |57| |delay, add synchronization | |58| |to prevent more than one | |59| |core from executing a | |60| |split lock in parallel. | |61| |sysctl split_lock_mitigate | |62| |can be used to avoid the | |63| |delay and synchronization | |64| |When both features are | |65| |supported, warn in #AC | |66+------------------+----------------------------+-----------------------+67|fatal |Kernel OOPs |Send SIGBUS to user. |68| |Send SIGBUS to user | |69| |When both features are | |70| |supported, fatal in #AC | |71+------------------+----------------------------+-----------------------+72|ratelimit:N |Do nothing |Limit bus lock rate to |73|(0 < N <= 1000) | |N bus locks per second |74| | |system wide and warn on|75| | |bus locks. |76+------------------+----------------------------+-----------------------+77 78Usages79======80 81Detecting and handling bus lock may find usages in various areas:82 83It is critical for real time system designers who build consolidated real84time systems. These systems run hard real time code on some cores and run85"untrusted" user processes on other cores. The hard real time cannot afford86to have any bus lock from the untrusted processes to hurt real time87performance. To date the designers have been unable to deploy these88solutions as they have no way to prevent the "untrusted" user code from89generating split lock and bus lock to block the hard real time code to90access memory during bus locking.91 92It's also useful for general computing to prevent guests or user93applications from slowing down the overall system by executing instructions94with bus lock.95 96 97Guidance98========99off100---101 102Disable checking for split lock and bus lock. This option can be useful if103there are legacy applications that trigger these events at a low rate so104that mitigation is not needed.105 106warn107----108 109A warning is emitted when a bus lock is detected which allows to identify110the offending application. This is the default behavior.111 112fatal113-----114 115In this case, the bus lock is not tolerated and the process is killed.116 117ratelimit118---------119 120A system wide bus lock rate limit N is specified where 0 < N <= 1000. This121allows a bus lock rate up to N bus locks per second. When the bus lock rate122is exceeded then any task which is caught via the buslock #DB exception is123throttled by enforced sleeps until the rate goes under the limit again.124 125This is an effective mitigation in cases where a minimal impact can be126tolerated, but an eventual Denial of Service attack has to be prevented. It127allows to identify the offending processes and analyze whether they are128malicious or just badly written.129 130Selecting a rate limit of 1000 allows the bus to be locked for up to about131seven million cycles each second (assuming 7000 cycles for each bus132lock). On a 2 GHz processor that would be about 0.35% system slowdown.133