cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

locktorture.rst (5742B)


      1==================================
      2Kernel Lock Torture Test Operation
      3==================================
      4
      5CONFIG_LOCK_TORTURE_TEST
      6========================
      7
      8The CONFIG LOCK_TORTURE_TEST config option provides a kernel module
      9that runs torture tests on core kernel locking primitives. The kernel
     10module, 'locktorture', may be built after the fact on the running
     11kernel to be tested, if desired. The tests periodically output status
     12messages via printk(), which can be examined via the dmesg (perhaps
     13grepping for "torture").  The test is started when the module is loaded,
     14and stops when the module is unloaded. This program is based on how RCU
     15is tortured, via rcutorture.
     16
     17This torture test consists of creating a number of kernel threads which
     18acquire the lock and hold it for specific amount of time, thus simulating
     19different critical region behaviors. The amount of contention on the lock
     20can be simulated by either enlarging this critical region hold time and/or
     21creating more kthreads.
     22
     23
     24Module Parameters
     25=================
     26
     27This module has the following parameters:
     28
     29
     30Locktorture-specific
     31--------------------
     32
     33nwriters_stress
     34		  Number of kernel threads that will stress exclusive lock
     35		  ownership (writers). The default value is twice the number
     36		  of online CPUs.
     37
     38nreaders_stress
     39		  Number of kernel threads that will stress shared lock
     40		  ownership (readers). The default is the same amount of writer
     41		  locks. If the user did not specify nwriters_stress, then
     42		  both readers and writers be the amount of online CPUs.
     43
     44torture_type
     45		  Type of lock to torture. By default, only spinlocks will
     46		  be tortured. This module can torture the following locks,
     47		  with string values as follows:
     48
     49		     - "lock_busted":
     50				Simulates a buggy lock implementation.
     51
     52		     - "spin_lock":
     53				spin_lock() and spin_unlock() pairs.
     54
     55		     - "spin_lock_irq":
     56				spin_lock_irq() and spin_unlock_irq() pairs.
     57
     58		     - "rw_lock":
     59				read/write lock() and unlock() rwlock pairs.
     60
     61		     - "rw_lock_irq":
     62				read/write lock_irq() and unlock_irq()
     63				rwlock pairs.
     64
     65		     - "mutex_lock":
     66				mutex_lock() and mutex_unlock() pairs.
     67
     68		     - "rtmutex_lock":
     69				rtmutex_lock() and rtmutex_unlock() pairs.
     70				Kernel must have CONFIG_RT_MUTEX=y.
     71
     72		     - "rwsem_lock":
     73				read/write down() and up() semaphore pairs.
     74
     75
     76Torture-framework (RCU + locking)
     77---------------------------------
     78
     79shutdown_secs
     80		  The number of seconds to run the test before terminating
     81		  the test and powering off the system.  The default is
     82		  zero, which disables test termination and system shutdown.
     83		  This capability is useful for automated testing.
     84
     85onoff_interval
     86		  The number of seconds between each attempt to execute a
     87		  randomly selected CPU-hotplug operation.  Defaults
     88		  to zero, which disables CPU hotplugging.  In
     89		  CONFIG_HOTPLUG_CPU=n kernels, locktorture will silently
     90		  refuse to do any CPU-hotplug operations regardless of
     91		  what value is specified for onoff_interval.
     92
     93onoff_holdoff
     94		  The number of seconds to wait until starting CPU-hotplug
     95		  operations.  This would normally only be used when
     96		  locktorture was built into the kernel and started
     97		  automatically at boot time, in which case it is useful
     98		  in order to avoid confusing boot-time code with CPUs
     99		  coming and going. This parameter is only useful if
    100		  CONFIG_HOTPLUG_CPU is enabled.
    101
    102stat_interval
    103		  Number of seconds between statistics-related printk()s.
    104		  By default, locktorture will report stats every 60 seconds.
    105		  Setting the interval to zero causes the statistics to
    106		  be printed -only- when the module is unloaded.
    107
    108stutter
    109		  The length of time to run the test before pausing for this
    110		  same period of time.  Defaults to "stutter=5", so as
    111		  to run and pause for (roughly) five-second intervals.
    112		  Specifying "stutter=0" causes the test to run continuously
    113		  without pausing.
    114
    115shuffle_interval
    116		  The number of seconds to keep the test threads affinitied
    117		  to a particular subset of the CPUs, defaults to 3 seconds.
    118		  Used in conjunction with test_no_idle_hz.
    119
    120verbose
    121		  Enable verbose debugging printing, via printk(). Enabled
    122		  by default. This extra information is mostly related to
    123		  high-level errors and reports from the main 'torture'
    124		  framework.
    125
    126
    127Statistics
    128==========
    129
    130Statistics are printed in the following format::
    131
    132  spin_lock-torture: Writes:  Total: 93746064  Max/Min: 0/0   Fail: 0
    133     (A)		    (B)		   (C)		  (D)	       (E)
    134
    135  (A): Lock type that is being tortured -- torture_type parameter.
    136
    137  (B): Number of writer lock acquisitions. If dealing with a read/write
    138       primitive a second "Reads" statistics line is printed.
    139
    140  (C): Number of times the lock was acquired.
    141
    142  (D): Min and max number of times threads failed to acquire the lock.
    143
    144  (E): true/false values if there were errors acquiring the lock. This should
    145       -only- be positive if there is a bug in the locking primitive's
    146       implementation. Otherwise a lock should never fail (i.e., spin_lock()).
    147       Of course, the same applies for (C), above. A dummy example of this is
    148       the "lock_busted" type.
    149
    150Usage
    151=====
    152
    153The following script may be used to torture locks::
    154
    155	#!/bin/sh
    156
    157	modprobe locktorture
    158	sleep 3600
    159	rmmod locktorture
    160	dmesg | grep torture:
    161
    162The output can be manually inspected for the error flag of "!!!".
    163One could of course create a more elaborate script that automatically
    164checked for such errors.  The "rmmod" command forces a "SUCCESS",
    165"FAILURE", or "RCU_HOTPLUG" indication to be printk()ed.  The first
    166two are self-explanatory, while the last indicates that while there
    167were no locking failures, CPU-hotplug problems were detected.
    168
    169Also see: Documentation/RCU/torture.rst