• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# `lws_sul` scheduler api
2
3Since v3.2 lws no longer requires periodic checking for timeouts and
4other events.  A new system was refactored in where future events are
5scheduled on to a single, unified, sorted linked-list in time order,
6with everything at us resolution.
7
8This makes it very cheap to know when the next scheduled event is
9coming and restrict the poll wait to match, or for event libraries
10set a timer to wake at the earliest event when returning to the
11event loop.
12
13Everything that was checked periodically was converted to use `lws_sul`
14and schedule its own later event.  The end result is when lws is idle,
15it will stay asleep in the poll wait until a network event or the next
16scheduled `lws_sul` event happens, which is optimal for power.
17
18# Side effect for older code
19
20If your older code uses `lws_service_fd()`, it used to be necessary
21to call this with a NULL pollfd periodically to indicate you wanted
22to let the background checks happen.  `lws_sul` eliminates the whole
23concept of periodic checking and NULL is no longer a valid pollfd
24value for this and related apis.
25
26# Using `lws_sul` in user code
27
28See `minimal-http-client-multi` for an example of using the `lws_sul`
29scheduler from your own code; it uses it to spread out connection
30attempts so they are staggered in time.  You must create an
31`lws_sorted_usec_list_t` object somewhere, eg, in you own existing object.
32
33```
34static lws_sorted_usec_list_t sul_stagger;
35```
36
37Create your own callback for the event... the argument points to the sul object
38used when the callback was scheduled.  You can use pointer arithmetic to translate
39that to your own struct when the `lws_sorted_usec_list_t` was a member of the
40same struct.
41
42```
43static void
44stagger_cb(lws_sorted_usec_list_t *sul)
45{
46...
47}
48```
49
50When you want to schedule the callback, use `lws_sul_schedule()`... this will call
51it 10ms in the future
52
53```
54	lws_sul_schedule(context, 0, &sul_stagger, stagger_cb, 10 * LWS_US_PER_MS);
55```
56
57In the case you destroy your object and need to cancel the scheduled callback, use
58
59```
60	lws_sul_schedule(context, 0, &sul_stagger, NULL, LWS_SET_TIMER_USEC_CANCEL);
61```
62
63# lws_sul2 and system suspend
64
65In v4.1, alongside the existing `lws_sul` apis there is a refactor and additional
66functionality aimed at negotiating system suspend, while remaining completely
67backwards-compatible with v3.2+ lws_sul apis.
68
69Devicewide suspend is basically the withdrawal of CPU availability for an unbounded
70amount of time, so what may have been scheduled by the user code may miss its time
71slot because the cpu was down and nothing is getting serviced.  Whether that is
72actively desirable, OK, a big disaster, or a failure that will be corrected at other
73layers at the cost of, eg, some additional latency, depends on the required device
74behaviours and the function of the user code that was scheduled, and its meaning to
75the system.
76
77Before v4.1, lws just offers the same scheduling service for everything both internal
78and arranged by user code, and has no way to know what is critical for the device to
79operate as intended, and so must force wake from suspend, or if for that scheduled
80event 'failure [to get the event] is an option'.
81
82For example locally-initiated periodic keepalive pings not happening may allow
83persistently dead (ie, no longer passing data) connections to remain unrenewed, but
84eventually when suspend ends for another reason, the locally-initiated PING probes
85will resume and it will be discovered and if the connectivity allows, corrected.
86
87If the device's function can handle the latency of there being no connectivity in
88suspend under those conditions until it wakes for another reason, it's OK for these
89kind of timeouts to be suppressed during suspend and basically take the power saving
90instead.  If for a particular device it's intolerable to ever have a silently dead
91connection for more than a very short time compared to suspend durations, then these
92kind of timeouts must have the priority to wake the whole device from suspend so
93they continue to operate unimpeded.
94
95That is just one example, lws offers generic scheduler services the user code can
96exploit for any purpose, including mission-critical ones.  The changes give the user
97code a way to tell lws if a particular scheduled event is important enough to the
98system operation to wake the system from devicewide suspend.
99
100