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

timers-howto.rst (4052B)


      1===================================================================
      2delays - Information on the various kernel delay / sleep mechanisms
      3===================================================================
      4
      5This document seeks to answer the common question: "What is the
      6RightWay (TM) to insert a delay?"
      7
      8This question is most often faced by driver writers who have to
      9deal with hardware delays and who may not be the most intimately
     10familiar with the inner workings of the Linux Kernel.
     11
     12
     13Inserting Delays
     14----------------
     15
     16The first, and most important, question you need to ask is "Is my
     17code in an atomic context?"  This should be followed closely by "Does
     18it really need to delay in atomic context?" If so...
     19
     20ATOMIC CONTEXT:
     21	You must use the `*delay` family of functions. These
     22	functions use the jiffie estimation of clock speed
     23	and will busy wait for enough loop cycles to achieve
     24	the desired delay:
     25
     26	ndelay(unsigned long nsecs)
     27	udelay(unsigned long usecs)
     28	mdelay(unsigned long msecs)
     29
     30	udelay is the generally preferred API; ndelay-level
     31	precision may not actually exist on many non-PC devices.
     32
     33	mdelay is macro wrapper around udelay, to account for
     34	possible overflow when passing large arguments to udelay.
     35	In general, use of mdelay is discouraged and code should
     36	be refactored to allow for the use of msleep.
     37
     38NON-ATOMIC CONTEXT:
     39	You should use the `*sleep[_range]` family of functions.
     40	There are a few more options here, while any of them may
     41	work correctly, using the "right" sleep function will
     42	help the scheduler, power management, and just make your
     43	driver better :)
     44
     45	-- Backed by busy-wait loop:
     46
     47		udelay(unsigned long usecs)
     48
     49	-- Backed by hrtimers:
     50
     51		usleep_range(unsigned long min, unsigned long max)
     52
     53	-- Backed by jiffies / legacy_timers
     54
     55		msleep(unsigned long msecs)
     56		msleep_interruptible(unsigned long msecs)
     57
     58	Unlike the `*delay` family, the underlying mechanism
     59	driving each of these calls varies, thus there are
     60	quirks you should be aware of.
     61
     62
     63	SLEEPING FOR "A FEW" USECS ( < ~10us? ):
     64		* Use udelay
     65
     66		- Why not usleep?
     67			On slower systems, (embedded, OR perhaps a speed-
     68			stepped PC!) the overhead of setting up the hrtimers
     69			for usleep *may* not be worth it. Such an evaluation
     70			will obviously depend on your specific situation, but
     71			it is something to be aware of.
     72
     73	SLEEPING FOR ~USECS OR SMALL MSECS ( 10us - 20ms):
     74		* Use usleep_range
     75
     76		- Why not msleep for (1ms - 20ms)?
     77			Explained originally here:
     78				https://lore.kernel.org/r/15327.1186166232@lwn.net
     79
     80			msleep(1~20) may not do what the caller intends, and
     81			will often sleep longer (~20 ms actual sleep for any
     82			value given in the 1~20ms range). In many cases this
     83			is not the desired behavior.
     84
     85		- Why is there no "usleep" / What is a good range?
     86			Since usleep_range is built on top of hrtimers, the
     87			wakeup will be very precise (ish), thus a simple
     88			usleep function would likely introduce a large number
     89			of undesired interrupts.
     90
     91			With the introduction of a range, the scheduler is
     92			free to coalesce your wakeup with any other wakeup
     93			that may have happened for other reasons, or at the
     94			worst case, fire an interrupt for your upper bound.
     95
     96			The larger a range you supply, the greater a chance
     97			that you will not trigger an interrupt; this should
     98			be balanced with what is an acceptable upper bound on
     99			delay / performance for your specific code path. Exact
    100			tolerances here are very situation specific, thus it
    101			is left to the caller to determine a reasonable range.
    102
    103	SLEEPING FOR LARGER MSECS ( 10ms+ )
    104		* Use msleep or possibly msleep_interruptible
    105
    106		- What's the difference?
    107			msleep sets the current task to TASK_UNINTERRUPTIBLE
    108			whereas msleep_interruptible sets the current task to
    109			TASK_INTERRUPTIBLE before scheduling the sleep. In
    110			short, the difference is whether the sleep can be ended
    111			early by a signal. In general, just use msleep unless
    112			you know you have a need for the interruptible variant.
    113
    114	FLEXIBLE SLEEPING (any delay, uninterruptible)
    115		* Use fsleep