#!/bin/sh

# Copyright 2016 The Chromium OS Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.

set -e

JOB=$(basename "$0")

DEVICE_SHUTTING_DOWN=4 # Defined in session_manager_service.h

# Check if file exists and is writeable by us. -w is always true under root.
is_writeable() {
  local file="$1"
  local v=0

  if [ -f "${file}" ]; then
    local wfile=$(ls -l "${file}" | awk '$1 ~ /^.*w.*/ { print $NF; }')

    if [ ! -z "${wfile}" ]; then
      v=1
    fi
  fi

  echo "${v}"
}

# Read chrome OS configurations
read_cros_config() {
  local path="$1"
  local key="$2"
  local def="$3"

  # cros_config exits with non-zero when a key isn't existed,
  cros_config "${path}" "${key}" || echo "${def}"
}

read_cros_config_int() {
  local v=$(read_cros_config "$@")

  # config value should be an integer
  case "${v}" in
    ''|*[!0-9]*)
        logger -t "${JOB}" -p warn \
          "cros_config ${path}/${key} is not an integer! "\
          "Let me use ${def} instead of ${v}."
      echo "${def}"
      ;;
    *)
      echo "${v}"
      ;;
    esac
}

# The ui job can be restarted asynchronously via ui-respawn. This allows the
# possibility that ui can be started while shutdown/reboot is in process.
# Make sure that the reboot/halt jobs are not active at the time we are
# restarting ui.
if ! initctl status reboot | grep -q 'stop/waiting' || \
    ! initctl status halt | grep -q 'stop/waiting'; then
  logger -t "${UPSTART_JOB}" "ERROR: reboot/shutdown in progress, quit."
  stop ui
  exit ${DEVICE_SHUTTING_DOWN}
fi

# Setup drm log tracing if available. This will allow us to access more verbose
# logs to debug display connectivity issues when they arise. The traces are
# accessible via chrome://system and included in feedback/crash reports.
#
# As configured, we won't have a lot of volume in these traces, they will only
# be written in cases where the display configuration changes, or an error
# occurs. As such, there is no performance impact during steady-state operation.
#
# TODO(seanpaul): Once we've finished moving over to the upstream tracefs
#                 implementation, remove the block initializing the downstream
#                 debugfs code. Tracked in b/163580546.
DRM_TRACE_DIR="/sys/kernel/debug/dri"
DRM_TRACE_MASK="${DRM_TRACE_DIR}/trace_mask"
if [ -w "${DRM_TRACE_MASK}" ]; then
  # 64kB per-cpu of drm logs.
  echo 64 > "${DRM_TRACE_DIR}/trace_size_kb"
  # 0x106 corresponds to (DRM_UT_DRIVER | DRM_UT_KMS | DRM_UT_DP).
  echo 0x106 > "${DRM_TRACE_MASK}"
fi
DRM_TRACEFS_INSTANCE="/sys/kernel/debug/tracing/instances/drm"
DRM_TRACEFS_BUF_SIZE="${DRM_TRACEFS_INSTANCE}/buffer_size_kb"
if [ -w "${DRM_TRACEFS_BUF_SIZE}" ]; then
  # 64kB per-cpu of drm logs.
  echo 64 > "${DRM_TRACEFS_BUF_SIZE}"
fi

mkdir -p "${UI_LOG_DIR}"
ln -sf ui.$(date --utc +%Y%m%d-%H%M%S) "${UI_LOG_DIR}/${UI_LOG_FILE}"

mkdir -p "${CONTAINER_ROOT_DIR}"

# Create a top-level cpu cgroup for ui tasks.  This will be used to share cpu
# resources with tasks not managed by the UI.  Android will be placed in a
# parallel cgroup allowing priority to be given to one or the other depending
# on what the user is doing at the time.
if [ ! -d "${UI_CPU_CGROUP_DIR}" ]; then
  mkdir -p "${UI_CPU_CGROUP_DIR}"
fi

# Set up cgroups for chrome. We create two task groups, one for at most one
# foreground renderer and one for all the background renderers and set the
# background group to a very low priority. We specifically do not set it to
# the lowest "2" such that other processes like the update-engine can be even
# lower. The default value is 1024.
CHROME_CPU_CGROUP_DIR=/sys/fs/cgroup/cpu/chrome_renderers
if [ ! -d "${CHROME_CPU_CGROUP_DIR}" ]; then
  mkdir -p "${CHROME_CPU_CGROUP_DIR}/foreground"
  mkdir -p "${CHROME_CPU_CGROUP_DIR}/background"
  echo "10" > "${CHROME_CPU_CGROUP_DIR}/background/cpu.shares"
  chown -R chronos "${CHROME_CPU_CGROUP_DIR}"
fi

# TODO(ikjn): This scheduler settings are a bit complex,
# Let's move these out to an another executable (c++).
# Tracked in b/182958867

# apply ui and chrome foreground cgroup's uclamp.min, default 20%,
v=$(read_cros_config_int "/scheduler-tune" "boost-urgent" "20")
for knob in "${UI_CPU_CGROUP_DIR}/cpu.uclamp.min" \
            "${CHROME_CPU_CGROUP_DIR}/cpu.uclamp.min" \
            "${CHROME_CPU_CGROUP_DIR}/foreground/cpu.uclamp.min"; do
  if [ -f "${knob}" ]; then
    echo "${v}" > "${knob}"
  fi
done

# Apply boosting values for platforms using schedtune
SCHEDTUNE_BOOST_DIR=/sys/fs/cgroup/schedtune/chrome/urgent
if [ -d "${SCHEDTUNE_BOOST_DIR}" ]; then
  echo "${v}" > "${SCHEDTUNE_BOOST_DIR}/schedtune.boost"
  echo "1" > "${SCHEDTUNE_BOOST_DIR}/schedtune.prefer_idle"
fi

# Apply Input boosting
INPUT_BOOST=/sys/module/cpu_boost/parameters/input_boost_freq_percent
INPUT_BOOST_DEFAULT=0
CPUFREQ_CONF=/etc/cpufreq.conf

if [ -f "${INPUT_BOOST}" ]; then
  # Default value is 75 for platforms using EAS
  if [ -f "${CPUFREQ_CONF}" ]; then
    . "${CPUFREQ_CONF}"
    if [ "${CPUFREQ_GOVERNOR}" = "schedutil" ] || \
       [ "${CPUFREQ_GOVERNOR}" = "sched" ]; then
      INPUT_BOOST_DEFAULT=75
    fi
  fi
  read_cros_config_int "/scheduler-tune" "input-boost" \
    "${INPUT_BOOST_DEFAULT}" > "${INPUT_BOOST}"
fi

# Limit non-urgent tasks to small cores
CPUBUSDIR=/sys/bus/cpu/devices
CPUSET_NONURGENT_DIR=/sys/fs/cgroup/cpuset/chrome/non-urgent
if [ -d "${CPUSET_NONURGENT_DIR}" ]; then
  v=$(read_cros_config "/scheduler-tune" "cpuset-nonurgent" "")

  # If a cpuset-nonurgent isn't specified in cros_config,
  # use all little cores as a non-urgent cpuset.
  if [ -z "${v}" ] && grep -q '^big_little$' /etc/ui_use_flags.txt; then
    nr_cpus=0
    nr_small_cpus=0
    min_cap=1025
    if [ -f "${CPUBUSDIR}/cpu0/cpufreq/cpuinfo_max_freq" ]; then
      min_maxfreq=$(cat "${CPUBUSDIR}/cpu0/cpufreq/cpuinfo_max_freq")
    fi
    cpus=""
    prefix="${CPUBUSDIR}/cpu"
    for cpu in ${CPUBUSDIR}/cpu*; do
      if [ -f "${cpu}/cpu_capacity" ]; then
        cap=$(cat "${cpu}/cpu_capacity")
        if [ "${cap}" -lt "${min_cap}" ]; then
          cpus="${cpu##${prefix}}"
          min_cap="${cap}"
          nr_small_cpus=1
        elif [ "${cap}" -eq "${min_cap}" ]; then
          cpus="${cpus},${cpu##${prefix}}"
          nr_small_cpus=$((nr_small_cpus + 1))
        fi
      elif [ -f "${cpu}/cpufreq/cpuinfo_max_freq" ]; then
        maxfreq=$(cat "${cpu}/cpufreq/cpuinfo_max_freq")
        if [ "${maxfreq}" -lt "${min_maxfreq}" ]; then
          cpus="${cpu##${prefix}}"
          min_maxfreq="${maxfreq}"
          nr_small_cpus=1
        elif [ "${maxfreq}" -eq "${min_maxfreq}" ]; then
          cpus="${cpus},${cpu##${prefix}}"
          nr_small_cpus=$((nr_small_cpus + 1))
        fi
      else
        cpus=""
        break;
      fi
      nr_cpus=$((nr_cpus + 1))
    done
    if [ "${nr_small_cpus}" -lt "${nr_cpus}" ]; then
      v="${cpus}"
    fi
  fi

  if [ -n "${v}" ]; then
    echo "${v}" > "${CPUSET_NONURGENT_DIR}/cpus"
  fi
fi

# Set up a folder on the stateful partition for ephemeral encrypted swap
# files.
USERSPACE_SWAP_DIR=/mnt/stateful_partition/unencrypted/userspace_swap.tmp/
if [ ! -d "${USERSPACE_SWAP_DIR}" ]; then
  mkdir -p "${USERSPACE_SWAP_DIR}"
  chown -R chronos "${USERSPACE_SWAP_DIR}"
fi

# Set up the ui freezer cgroup. This is used during suspend to order the
# freezing of processes to avoid FUSE causing freeze to timeout.
if [ ! -d ${UI_FREEZER_CGROUP_DIR} ]; then
  mkdir -p ${UI_FREEZER_CGROUP_DIR}
  chown -R chronos ${UI_FREEZER_CGROUP_DIR}
  chown power:power ${UI_FREEZER_CGROUP_DIR}/freezer.state
fi

# Set up the renderer freezer.  This is used during the suspend/resume cycle
# to freeze all the chrome renderers so that they don't take up a lot of power
# when the system comes up in dark resume
if [ ! -d "${CHROME_FREEZER_CGROUP_DIR}" ]; then
  mkdir -p "${CHROME_FREEZER_CGROUP_DIR}"
  mkdir -p "${CHROME_FREEZER_CGROUP_DIR}/to_be_frozen"
  chown -R chronos "${CHROME_FREEZER_CGROUP_DIR}"
fi

# Set up a cgroup for containers(Android) started by session manager.  To
# start, limit the cpu shares of Android.  This will be reset to the default
# once the user clicks on an Android app.  Keeping the usage low prevents
# Android boot from slowing the login process.
CONTAINER_CGROUP_DIR=/sys/fs/cgroup/cpu/session_manager_containers
if [ -d "${CONTAINER_CGROUP_DIR}" ]; then
  echo "32" > "${CONTAINER_CGROUP_DIR}/cpu.shares"
  echo "-1" > "${CONTAINER_CGROUP_DIR}/cpu.cfs_quota_us"
  if [ -f "${CONTAINER_CGROUP_DIR}/cpu.uclamp.min" ]; then
    echo "100" > "${CONTAINER_CGROUP_DIR}/cpu.uclamp.min"
  fi
fi

# Enable madvise(MADV_HUGEPAGE) for VMs guest mappings
if [ -f "/sys/kernel/mm/transparent_hugepage/shmem_enabled" ]; then
  echo "advise" > "/sys/kernel/mm/transparent_hugepage/shmem_enabled"
fi

# Tweak KVM HLT polling only for x86
if uname -m | grep -q x86; then
  HALT_POLL_NS_SHRINK=/sys/module/kvm/parameters/halt_poll_ns_shrink
  HALT_POLL_NS=/sys/module/kvm/parameters/halt_poll_ns
  # Older kernels (e.g. 4.4) do not permit dynamic halt_poll_ns_shrink and
  # have this attr as read-only.
  v=$(is_writeable "${HALT_POLL_NS_SHRINK}")
  if [ "${v}" != "0" ]; then
    echo 2 > "${HALT_POLL_NS_SHRINK}"
  fi
  # Ditto, just in case
  v=$(is_writeable "${HALT_POLL_NS}")
  if [ "${v}" != "0" ]; then
    echo 160000 > "${HALT_POLL_NS}"
  fi
fi

if crossystem "cros_debug?1"; then
  # Set up the alt-syscall kernel module (on systems with kernel 3.14 or later)
  # to allow developer mode only system calls to be exposed to Android
  # containers (and potentially other containers) when the system is in
  # developer mode.
  DEVMODE_SYSCALL=/proc/sys/kernel/chromiumos/alt_syscall/allow_devmode_syscalls
  if [ -e "${DEVMODE_SYSCALL}" ]; then
    echo "1" > "${DEVMODE_SYSCALL}"
  fi

  # /proc/sys/kernel/perf_event_paranoid controls the use of kernel perf events
  # system by users without the CAP_SYS_ADMIN capability (see
  # kernel/Documentation/sysctl/kernel.txt for details). On systems with kernel
  # 3.14 or later, we enable CONFIG_SECURITY_PERF_EVENTS_RESTRICT in kernel,
  # which sets the default value of perf_event_paranoid to a more restrictive
  # level "3", instead of "1". As alt-syscall exposes perf_event_open() to
  # Android containers when the system is in developer mode, restore
  # perf_event_paranoid to the less restrictive level "1".
  PERF_EVENT_PARANOID=/proc/sys/kernel/perf_event_paranoid
  if [ -e "${PERF_EVENT_PARANOID}" ]; then
    echo "1" > "${PERF_EVENT_PARANOID}"
  fi
fi

# Set up the lucid sleep debug flag.  This file only exists on devices that
# use kernel 3.14.  When this sysfs entry is set to 1, the kernel will treat
# all resumes as dark resumes, relying on chrome and powerd to detect any user
# activity and to transition out of dark resume into regular resume.  Since we
# are currently unable to distinguish the source that woke up the system, this
# is a temporary workaround to allow for apps and extensions to test lucid
# sleep functionality.  Chrome will write 1 to this file only if the
# wake-on-packets flag is set.
#
# TODO(chirantan): Remove this once the firmware change and corresponding
# kernel change for querying the wakeup source is ready (crbug.com/414949).
DARK_RESUME_ALWAYS_FILE=/sys/power/dark_resume_always
if [ -e "${DARK_RESUME_ALWAYS_FILE}" ]; then
  chown chronos "${DARK_RESUME_ALWAYS_FILE}"
fi