xref: /openbmc/phosphor-pid-control/pid/pidloop.cpp (revision bd1d1024654b893b4f25729f8fc124446cf2411f)
1 /**
2  * Copyright 2017 Google Inc.
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *     http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include "pidloop.hpp"
18 
19 #include "pid/pidcontroller.hpp"
20 #include "pid/tuning.hpp"
21 #include "pid/zone_interface.hpp"
22 #include "sensors/sensor.hpp"
23 
24 #include <boost/asio/steady_timer.hpp>
25 
26 #include <chrono>
27 #include <map>
28 #include <memory>
29 #include <sstream>
30 #include <vector>
31 
32 namespace pid_control
33 {
34 
processThermals(std::shared_ptr<ZoneInterface> zone)35 static void processThermals(std::shared_ptr<ZoneInterface> zone)
36 {
37     // Get the latest margins.
38     zone->updateSensors();
39     // Zero out the set point goals.
40     zone->clearSetPoints();
41     zone->clearRPMCeilings();
42     // Run the margin PIDs.
43     zone->processThermals();
44     // Get the maximum RPM setpoint.
45     zone->determineMaxSetPointRequest();
46 }
47 
pidControlLoop(std::shared_ptr<ZoneInterface> zone,std::shared_ptr<boost::asio::steady_timer> timer,const bool * isCanceling,bool first,uint64_t cycleCnt)48 void pidControlLoop(std::shared_ptr<ZoneInterface> zone,
49                     std::shared_ptr<boost::asio::steady_timer> timer,
50                     const bool* isCanceling, bool first, uint64_t cycleCnt)
51 {
52     if (*isCanceling)
53         return;
54 
55     std::chrono::steady_clock::time_point nextTime;
56 
57     if (first)
58     {
59         if (loggingEnabled)
60         {
61             zone->initializeLog();
62         }
63 
64         zone->initializeCache();
65         processThermals(zone);
66 
67         nextTime = std::chrono::steady_clock::now();
68     }
69     else
70     {
71         nextTime = timer->expiry();
72     }
73 
74     uint64_t msPerFanCycle = zone->getCycleIntervalTime();
75 
76     // Push forward the original expiration time of timer, instead of just
77     // resetting it with expires_after() from now, to make sure the interval
78     // is of the expected duration, and not stretched out by CPU time taken.
79     nextTime += std::chrono::milliseconds(msPerFanCycle);
80     timer->expires_at(nextTime);
81     timer->async_wait([zone, timer, cycleCnt, isCanceling, msPerFanCycle](
82                           const boost::system::error_code& ec) mutable {
83         if (ec == boost::asio::error::operation_aborted)
84         {
85             return; // timer being canceled, stop loop
86         }
87 
88         /*
89          * This should sleep on the conditional wait for the listen thread
90          * to tell us it's in sync.  But then we also need a timeout option
91          * in case phosphor-hwmon is down, we can go into some weird failure
92          * more.
93          *
94          * Another approach would be to start all sensors in worst-case
95          * values, and fail-safe mode and then clear out of fail-safe mode
96          * once we start getting values.  Which I think it is a solid
97          * approach.
98          *
99          * For now this runs before it necessarily has any sensor values.
100          * For the host sensors they start out in fail-safe mode.  For the
101          * fans, they start out as 0 as input and then are adjusted once
102          * they have values.
103          *
104          * If a fan has failed, it's value will be whatever we're told or
105          * however we retrieve it.  This program disregards fan values of 0,
106          * so any code providing a fan speed can set to 0 on failure and
107          * that fan value will be effectively ignored.  The PID algorithm
108          * will be unhappy but nothing bad will happen.
109          *
110          * TODO(venture): If the fan value is 0 should that loop just be
111          * skipped? Right now, a 0 value is ignored in
112          * FanController::inputProc()
113          */
114 
115         // Check if we should just go back to sleep.
116         if (zone->getManualMode())
117         {
118             pidControlLoop(zone, timer, isCanceling, false, cycleCnt);
119             return;
120         }
121 
122         // Get the latest fan speeds.
123         zone->updateFanTelemetry();
124 
125         uint64_t msPerThermalCycle = zone->getUpdateThermalsCycle();
126 
127         // Process thermal cycles at a rate that is less often than fan
128         // cycles. If thermal time is not an exact multiple of fan time,
129         // there will be some remainder left over, to keep the timing
130         // correct, as the intervals are staggered into one another.
131         if (cycleCnt >= msPerThermalCycle)
132         {
133             cycleCnt -= msPerThermalCycle;
134 
135             processThermals(zone);
136         }
137 
138         // Run the fan PIDs every iteration.
139         zone->processFans();
140 
141         if (loggingEnabled)
142         {
143             std::ostringstream out;
144             out << "," << zone->getFailSafeMode() << std::endl;
145             zone->writeLog(out.str());
146         }
147 
148         // Count how many milliseconds have elapsed, so we can know when
149         // to perform thermal cycles, in proper ratio with fan cycles.
150         cycleCnt += msPerFanCycle;
151 
152         pidControlLoop(zone, timer, isCanceling, false, cycleCnt);
153     });
154 }
155 
156 } // namespace pid_control
157