Reduce scheduler latency for realtime processes (#1638)

* WIP: reduce boardd and other lags

* Copypasta fault

* Silence spurious startup warning

Co-authored-by: Comma Device <device@comma.ai>
old-commit-hash: 68531b071c

common/realtime.py

@@ -6,6 +6,7 @@ import subprocess
 import multiprocessing
 from cffi import FFI
 
+from common.android import ANDROID
 from common.common_pyx import sec_since_boot  # pylint: disable=no-name-in-module, import-error
 
 
@@ -20,11 +21,7 @@ ffi = FFI()
 ffi.cdef("long syscall(long number, ...);")
 libc = ffi.dlopen(None)
 
-
-def set_realtime_priority(level):
-  if os.getuid() != 0:
-    print("not setting priority, not root")
-    return
+def _get_tid():
   if platform.machine() == "x86_64":
     NR_gettid = 186
   elif platform.machine() == "aarch64":
@@ -32,8 +29,25 @@ def set_realtime_priority(level):
   else:
     raise NotImplementedError
 
-  tid = libc.syscall(NR_gettid)
-  return subprocess.call(['chrt', '-f', '-p', str(level), str(tid)])
+  return libc.syscall(NR_gettid)
+
+
+def set_realtime_priority(level):
+  if os.getuid() != 0:
+    print("not setting priority, not root")
+    return
+
+  return subprocess.call(['chrt', '-f', '-p', str(level), str(_get_tid())])
+
+def set_core_affinity(core):
+  if os.getuid() != 0:
+    print("not setting affinity, not root")
+    return
+
+  if ANDROID:
+    return subprocess.call(['taskset', '-p', str(core), str(_get_tid())])
+  else:
+    return -1
 
 
 class Ratekeeper():

launch_chffrplus.sh

@@ -55,18 +55,22 @@ function launch {
     fi
   fi
 
-  # no cpu rationing for now
-  echo 0-3 > /dev/cpuset/background/cpus
-  echo 0-3 > /dev/cpuset/system-background/cpus
-  echo 0-3 > /dev/cpuset/foreground/boost/cpus
-  echo 0-3 > /dev/cpuset/foreground/cpus
-  echo 0-3 > /dev/cpuset/android/cpus
-
-  # change interrupt affinity
-  echo 3 > /proc/irq/6/smp_affinity_list # MDSS
-  echo 1 > /proc/irq/78/smp_affinity_list # Modem, can potentially lock up
-  echo 2 > /proc/irq/733/smp_affinity_list # USB
-  echo 2 > /proc/irq/736/smp_affinity_list # USB
+  # Android and other system processes are not permitted to run on CPU 3
+  # NEOS installed app processes can run anywhere
+  echo 0-2 > /dev/cpuset/background/cpus
+  echo 0-2 > /dev/cpuset/system-background/cpus
+  [ -d "/dev/cpuset/foreground/boost/cpus" ] && echo 0-2 > /dev/cpuset/foreground/boost/cpus  # Not present in < NEOS 15
+  echo 0-2 > /dev/cpuset/foreground/cpus
+  echo 0-2 > /dev/cpuset/android/cpus
+  echo 0-3 > /dev/cpuset/app/cpus
+
+  # Collect RIL and other possibly long-running I/O interrupts onto CPU 1
+  echo 1 > /proc/irq/78/smp_affinity_list # qcom,smd-modem (LTE radio)
+  echo 1 > /proc/irq/33/smp_affinity_list # ufshcd (flash storage)
+  echo 1 > /proc/irq/35/smp_affinity_list # wifi (wlan_pci)
+  # USB traffic needs realtime handling on cpu 3
+  [ -d "/proc/irq/733" ] && echo 3 > /proc/irq/733/smp_affinity_list # USB for LeEco
+  [ -d "/proc/irq/736" ] && echo 3 > /proc/irq/736/smp_affinity_list # USB for OP3T
 
   DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null && pwd )"
 

selfdrive/boardd/boardd.cc

@@ -856,9 +856,11 @@ int main() {
   int err;
   LOGW("starting boardd");
 
-  // set process priority
-  err = set_realtime_priority(4);
+  // set process priority and affinity
+  err = set_realtime_priority(54);
   LOG("set priority returns %d", err);
+  err = set_core_affinity(3);
+  LOG("set affinity returns %d", err);
 
   // check the environment
   if (getenv("STARTED")) {

selfdrive/camerad/main.cc

@@ -342,7 +342,7 @@ void* processing_thread(void *arg) {
 
   set_thread_name("processing");
 
-  err = set_realtime_priority(1);
+  err = set_realtime_priority(51);
   LOG("setpriority returns %d", err);
 
   // init cl stuff
@@ -1203,7 +1203,7 @@ void party(VisionState *s) {
 #endif
 
   // priority for cameras
-  err = set_realtime_priority(1);
+  err = set_realtime_priority(51);
   LOG("setpriority returns %d", err);
 
   cameras_run(&s->cameras);
@@ -1234,7 +1234,7 @@ void party(VisionState *s) {
 
 int main(int argc, char *argv[]) {
   int err;
-  set_realtime_priority(1);
+  set_realtime_priority(51);
 
   zsys_handler_set(NULL);
   signal(SIGINT, (sighandler_t)set_do_exit);

selfdrive/common/util.c

@@ -7,6 +7,7 @@
 #ifdef __linux__
 #include <sys/prctl.h>
 #include <sys/syscall.h>
+#define __USE_GNU
 #include <sched.h>
 #endif
 
@@ -61,3 +62,16 @@ int set_realtime_priority(int level) {
 #endif
 }
 
+int set_core_affinity(int core) {
+#ifdef QCOM
+
+  long tid = syscall(SYS_gettid);
+  cpu_set_t rt_cpu;
+
+  CPU_ZERO(&rt_cpu);
+  CPU_SET(core, &rt_cpu);
+  return sched_setaffinity(tid, sizeof(rt_cpu), &rt_cpu);
+#else
+  return -1;
+#endif
+}

selfdrive/common/util.h

@@ -44,6 +44,7 @@ void* read_file(const char* path, size_t* out_len);
 void set_thread_name(const char* name);
 
 int set_realtime_priority(int level);
+int set_core_affinity(int core);
 
 #ifdef __cplusplus
 }

selfdrive/controls/controlsd.py

@@ -3,7 +3,7 @@ import os
 import gc
 from cereal import car, log
 from common.numpy_fast import clip
-from common.realtime import sec_since_boot, set_realtime_priority, Ratekeeper, DT_CTRL
+from common.realtime import sec_since_boot, set_realtime_priority, set_core_affinity, Ratekeeper, DT_CTRL
 from common.profiler import Profiler
 from common.params import Params, put_nonblocking
 import cereal.messaging as messaging
@@ -39,7 +39,8 @@ EventName = car.CarEvent.EventName
 class Controls:
   def __init__(self, sm=None, pm=None, can_sock=None):
     gc.disable()
-    set_realtime_priority(3)
+    set_realtime_priority(53)
+    set_core_affinity(3)
 
     # Setup sockets
     self.pm = pm

selfdrive/controls/dmonitoringd.py

@@ -12,7 +12,7 @@ def dmonitoringd_thread(sm=None, pm=None):
   gc.disable()
 
   # start the loop
-  set_realtime_priority(3)
+  set_realtime_priority(53)
 
   params = Params()
 

selfdrive/controls/plannerd.py

@@ -15,7 +15,7 @@ def plannerd_thread(sm=None, pm=None):
   gc.disable()
 
   # start the loop
-  set_realtime_priority(2)
+  set_realtime_priority(52)
 
   cloudlog.info("plannerd is waiting for CarParams")
   CP = car.CarParams.from_bytes(Params().get("CarParams", block=True))

selfdrive/controls/radard.py

@@ -174,7 +174,7 @@ class RadarD():
 
 # fuses camera and radar data for best lead detection
 def radard_thread(sm=None, pm=None, can_sock=None):
-  set_realtime_priority(2)
+  set_realtime_priority(52)
 
   # wait for stats about the car to come in from controls
   cloudlog.info("radard is waiting for CarParams")