diff --git a/adb/adb.cpp b/adb/adb.cpp
index 4fbfafb..38c11b9 100644
--- a/adb/adb.cpp
+++ b/adb/adb.cpp
@@ -1221,7 +1221,7 @@
     std::string error;
     atransport* t = acquire_one_transport(type, serial, transport_id, nullptr, &error);
     if (!t) {
-        return SendFail(reply_fd, error);
+        return -1;
     }
 
     int ret = handle_forward_request(service, t, reply_fd);
diff --git a/adb/test_adb.py b/adb/test_adb.py
index 3bb433d..363002f 100644
--- a/adb/test_adb.py
+++ b/adb/test_adb.py
@@ -21,9 +21,11 @@
 """
 from __future__ import print_function
 
+import binascii
 import contextlib
 import os
 import random
+import select
 import socket
 import struct
 import subprocess
@@ -33,6 +35,52 @@
 import adb
 
 
+@contextlib.contextmanager
+def fake_adb_server(protocol=socket.AF_INET, port=0):
+    """Creates a fake ADB server that just replies with a CNXN packet."""
+
+    serversock = socket.socket(protocol, socket.SOCK_STREAM)
+    if protocol == socket.AF_INET:
+        serversock.bind(('127.0.0.1', port))
+    else:
+        serversock.bind(('::1', port))
+    serversock.listen(1)
+
+    # A pipe that is used to signal the thread that it should terminate.
+    readpipe, writepipe = os.pipe()
+
+    def _handle():
+        rlist = [readpipe, serversock]
+        while True:
+            ready, _, _ = select.select(rlist, [], [])
+            for r in ready:
+                if r == readpipe:
+                    # Closure pipe
+                    os.close(r)
+                    serversock.shutdown(socket.SHUT_RDWR)
+                    serversock.close()
+                    return
+                elif r == serversock:
+                    # Server socket
+                    conn, _ = r.accept()
+                    rlist.append(conn)
+                else:
+                    # Client socket
+                    data = r.recv(1024)
+                    if not data:
+                        rlist.remove(r)
+
+    port = serversock.getsockname()[1]
+    server_thread = threading.Thread(target=_handle)
+    server_thread.start()
+
+    try:
+        yield port
+    finally:
+        os.close(writepipe)
+        server_thread.join()
+
+
 class NonApiTest(unittest.TestCase):
     """Tests for ADB that aren't a part of the AndroidDevice API."""
 
@@ -211,45 +259,32 @@
 
         Bug: http://b/30313466
         """
-        ipv4 = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
-        ipv4.bind(('127.0.0.1', 0))
-        ipv4.listen(1)
+        for protocol in (socket.AF_INET, socket.AF_INET6):
+            try:
+                with fake_adb_server(protocol=protocol) as port:
+                    output = subprocess.check_output(
+                        ['adb', 'connect', 'localhost:{}'.format(port)])
 
-        ipv6 = socket.socket(socket.AF_INET6, socket.SOCK_STREAM)
-        try:
-            ipv6.bind(('::1', ipv4.getsockname()[1] + 1))
-            ipv6.listen(1)
-        except socket.error:
-            print("IPv6 not available, skipping")
-            return
+                    self.assertEqual(
+                        output.strip(), 'connected to localhost:{}'.format(port))
+            except socket.error:
+                print("IPv6 not available, skipping")
+                continue
 
-        for s in (ipv4, ipv6):
-            port = s.getsockname()[1]
+    def test_already_connected(self):
+        with fake_adb_server() as port:
             output = subprocess.check_output(
                 ['adb', 'connect', 'localhost:{}'.format(port)])
 
             self.assertEqual(
                 output.strip(), 'connected to localhost:{}'.format(port))
-            s.close()
 
-    def test_already_connected(self):
-        s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
-        s.bind(('127.0.0.1', 0))
-        s.listen(2)
+            # b/31250450: this always returns 0 but probably shouldn't.
+            output = subprocess.check_output(
+                ['adb', 'connect', 'localhost:{}'.format(port)])
 
-        port = s.getsockname()[1]
-        output = subprocess.check_output(
-            ['adb', 'connect', 'localhost:{}'.format(port)])
-
-        self.assertEqual(
-            output.strip(), 'connected to localhost:{}'.format(port))
-
-        # b/31250450: this always returns 0 but probably shouldn't.
-        output = subprocess.check_output(
-            ['adb', 'connect', 'localhost:{}'.format(port)])
-
-        self.assertEqual(
-            output.strip(), 'already connected to localhost:{}'.format(port))
+            self.assertEqual(
+                output.strip(), 'already connected to localhost:{}'.format(port))
 
 def main():
     random.seed(0)
diff --git a/adb/test_device.py b/adb/test_device.py
index b1ad043..f995be2 100644
--- a/adb/test_device.py
+++ b/adb/test_device.py
@@ -188,8 +188,6 @@
         finally:
             self.device.reverse_remove_all()
 
-    # Note: If you run this test when adb connect'd to a physical device over
-    # TCP, it will fail in adb reverse due to https://code.google.com/p/android/issues/detail?id=189821
     def test_forward_reverse_echo(self):
         """Send data through adb forward and read it back via adb reverse"""
         forward_port = 12345
diff --git a/adb/transport.cpp b/adb/transport.cpp
index 92c52e2..f5f6d26 100644
--- a/adb/transport.cpp
+++ b/adb/transport.cpp
@@ -974,7 +974,7 @@
             VLOG(TRANSPORT) << "socket transport " << transport->serial
                             << " is already in pending_list and fails to register";
             delete t;
-            return -1;
+            return -EALREADY;
         }
     }
 
@@ -983,7 +983,7 @@
             VLOG(TRANSPORT) << "socket transport " << transport->serial
                             << " is already in transport_list and fails to register";
             delete t;
-            return -1;
+            return -EALREADY;
         }
     }
 
diff --git a/adb/transport_local.cpp b/adb/transport_local.cpp
index c09fcb7..8032421 100644
--- a/adb/transport_local.cpp
+++ b/adb/transport_local.cpp
@@ -101,7 +101,11 @@
     int ret = register_socket_transport(fd, serial.c_str(), port, 0);
     if (ret < 0) {
         adb_close(fd);
-        *response = android::base::StringPrintf("already connected to %s", serial.c_str());
+        if (ret == -EALREADY) {
+            *response = android::base::StringPrintf("already connected to %s", serial.c_str());
+        } else {
+            *response = android::base::StringPrintf("failed to connect to %s", serial.c_str());
+        }
     } else {
         *response = android::base::StringPrintf("connected to %s", serial.c_str());
     }
diff --git a/base/include/android-base/parseint.h b/base/include/android-base/parseint.h
index 2c8570e..1b7cc5f 100644
--- a/base/include/android-base/parseint.h
+++ b/base/include/android-base/parseint.h
@@ -19,6 +19,7 @@
 
 #include <errno.h>
 #include <stdlib.h>
+#include <string.h>
 
 #include <limits>
 #include <string>
@@ -31,14 +32,20 @@
 // otherwise valid values will be rejected. Returns boolean success; 'out'
 // is untouched if parsing fails.
 template <typename T>
-bool ParseUint(const char* s, T* out,
-               T max = std::numeric_limits<T>::max()) {
+bool ParseUint(const char* s, T* out, T max = std::numeric_limits<T>::max(),
+               bool allow_suffixes = false) {
   int base = (s[0] == '0' && (s[1] == 'x' || s[1] == 'X')) ? 16 : 10;
   errno = 0;
   char* end;
   unsigned long long int result = strtoull(s, &end, base);
-  if (errno != 0 || s == end || *end != '\0') {
-    return false;
+  if (errno != 0 || end == s) return false;
+  if (*end != '\0') {
+    const char* suffixes = "bkmgtpe";
+    const char* suffix;
+    if (!allow_suffixes || (suffix = strchr(suffixes, tolower(*end))) == nullptr) return false;
+#if __clang__  // TODO: win32 still builds with GCC :-(
+    if (__builtin_mul_overflow(result, 1ULL << (10 * (suffix - suffixes)), &result)) return false;
+#endif
   }
   if (max < result) {
     return false;
@@ -49,9 +56,20 @@
 
 // TODO: string_view
 template <typename T>
-bool ParseUint(const std::string& s, T* out,
-               T max = std::numeric_limits<T>::max()) {
-  return ParseUint(s.c_str(), out, max);
+bool ParseUint(const std::string& s, T* out, T max = std::numeric_limits<T>::max(),
+               bool allow_suffixes = false) {
+  return ParseUint(s.c_str(), out, max, allow_suffixes);
+}
+
+template <typename T>
+bool ParseByteCount(const char* s, T* out, T max = std::numeric_limits<T>::max()) {
+  return ParseUint(s, out, max, true);
+}
+
+// TODO: string_view
+template <typename T>
+bool ParseByteCount(const std::string& s, T* out, T max = std::numeric_limits<T>::max()) {
+  return ParseByteCount(s.c_str(), out, max);
 }
 
 // Parses the signed decimal integer in the string 's' and sets 'out' to
diff --git a/base/parseint_test.cpp b/base/parseint_test.cpp
index 483b1d3..fb1c339 100644
--- a/base/parseint_test.cpp
+++ b/base/parseint_test.cpp
@@ -96,3 +96,44 @@
   ASSERT_FALSE(android::base::ParseInt("456x", &u));
   ASSERT_EQ(123u, u);
 }
+
+TEST(parseint, ParseByteCount) {
+  uint64_t i = 0;
+  ASSERT_TRUE(android::base::ParseByteCount("123b", &i));
+  ASSERT_EQ(123ULL, i);
+
+  ASSERT_TRUE(android::base::ParseByteCount("8k", &i));
+  ASSERT_EQ(8ULL * 1024, i);
+
+  ASSERT_TRUE(android::base::ParseByteCount("8M", &i));
+  ASSERT_EQ(8ULL * 1024 * 1024, i);
+
+  ASSERT_TRUE(android::base::ParseByteCount("6g", &i));
+  ASSERT_EQ(6ULL * 1024 * 1024 * 1024, i);
+
+  ASSERT_TRUE(android::base::ParseByteCount("1T", &i));
+  ASSERT_EQ(1ULL * 1024 * 1024 * 1024 * 1024, i);
+
+  ASSERT_TRUE(android::base::ParseByteCount("2p", &i));
+  ASSERT_EQ(2ULL * 1024 * 1024 * 1024 * 1024 * 1024, i);
+
+  ASSERT_TRUE(android::base::ParseByteCount("4e", &i));
+  ASSERT_EQ(4ULL * 1024 * 1024 * 1024 * 1024 * 1024 * 1024, i);
+}
+
+TEST(parseint, ParseByteCount_invalid_suffix) {
+  unsigned u;
+  ASSERT_FALSE(android::base::ParseByteCount("1x", &u));
+}
+
+TEST(parseint, ParseByteCount_overflow) {
+  uint64_t u64;
+  ASSERT_FALSE(android::base::ParseByteCount("4294967295E", &u64));
+
+  uint16_t u16;
+  ASSERT_TRUE(android::base::ParseByteCount("63k", &u16));
+  ASSERT_EQ(63U * 1024, u16);
+  ASSERT_TRUE(android::base::ParseByteCount("65535b", &u16));
+  ASSERT_EQ(65535U, u16);
+  ASSERT_FALSE(android::base::ParseByteCount("65k", &u16));
+}
diff --git a/bootstat/bootstat.cpp b/bootstat/bootstat.cpp
index e60e6be..8ce9dfc 100644
--- a/bootstat/bootstat.cpp
+++ b/bootstat/bootstat.cpp
@@ -38,6 +38,7 @@
 #include <android-base/file.h>
 #include <android-base/logging.h>
 #include <android-base/parseint.h>
+#include <android-base/properties.h>
 #include <android-base/strings.h>
 #include <android/log.h>
 #include <cutils/android_reboot.h>
@@ -302,6 +303,9 @@
     {"kernel_panic,init", 158},
     {"kernel_panic,oom", 159},
     {"kernel_panic,stack", 160},
+    {"kernel_panic,sysrq,livelock,alarm", 161},   // llkd
+    {"kernel_panic,sysrq,livelock,driver", 162},  // llkd
+    {"kernel_panic,sysrq,livelock,zombie", 163},  // llkd
 };
 
 // Converts a string value representing the reason the system booted to an
@@ -1096,14 +1100,28 @@
   boot_event_store->AddBootEventWithValue("absolute_boot_time", absolute_total.count());
 }
 
+// Gets the boot time offset. This is useful when Android is running in a
+// container, because the boot_clock is not reset when Android reboots.
+std::chrono::nanoseconds GetBootTimeOffset() {
+  static const int64_t boottime_offset =
+      android::base::GetIntProperty<int64_t>("ro.boot.boottime_offset", 0);
+  return std::chrono::nanoseconds(boottime_offset);
+}
+
+// Returns the current uptime, accounting for any offset in the CLOCK_BOOTTIME
+// clock.
+android::base::boot_clock::duration GetUptime() {
+  return android::base::boot_clock::now().time_since_epoch() - GetBootTimeOffset();
+}
+
 // Records several metrics related to the time it takes to boot the device,
 // including disambiguating boot time on encrypted or non-encrypted devices.
 void RecordBootComplete() {
   BootEventRecordStore boot_event_store;
   BootEventRecordStore::BootEventRecord record;
 
-  auto time_since_epoch = android::base::boot_clock::now().time_since_epoch();
-  auto uptime = std::chrono::duration_cast<std::chrono::seconds>(time_since_epoch);
+  auto uptime_ns = GetUptime();
+  auto uptime_s = std::chrono::duration_cast<std::chrono::seconds>(uptime_ns);
   time_t current_time_utc = time(nullptr);
 
   if (boot_event_store.GetBootEvent("last_boot_time_utc", &record)) {
@@ -1128,19 +1146,20 @@
     // Log the amount of time elapsed until the device is decrypted, which
     // includes the variable amount of time the user takes to enter the
     // decryption password.
-    boot_event_store.AddBootEventWithValue("boot_decryption_complete", uptime.count());
+    boot_event_store.AddBootEventWithValue("boot_decryption_complete", uptime_s.count());
 
     // Subtract the decryption time to normalize the boot cycle timing.
-    std::chrono::seconds boot_complete = std::chrono::seconds(uptime.count() - record.second);
+    std::chrono::seconds boot_complete = std::chrono::seconds(uptime_s.count() - record.second);
     boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_post_decrypt",
                                            boot_complete.count());
   } else {
-    boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_no_encryption", uptime.count());
+    boot_event_store.AddBootEventWithValue(boot_complete_prefix + "_no_encryption",
+                                           uptime_s.count());
   }
 
   // Record the total time from device startup to boot complete, regardless of
   // encryption state.
-  boot_event_store.AddBootEventWithValue(boot_complete_prefix, uptime.count());
+  boot_event_store.AddBootEventWithValue(boot_complete_prefix, uptime_s.count());
 
   RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init");
   RecordInitBootTimeProp(&boot_event_store, "ro.boottime.init.selinux");
@@ -1149,7 +1168,7 @@
   const BootloaderTimingMap bootloader_timings = GetBootLoaderTimings();
   RecordBootloaderTimings(&boot_event_store, bootloader_timings);
 
-  auto uptime_ms = std::chrono::duration_cast<std::chrono::milliseconds>(time_since_epoch);
+  auto uptime_ms = std::chrono::duration_cast<std::chrono::milliseconds>(uptime_ns);
   RecordAbsoluteBootTime(&boot_event_store, bootloader_timings, uptime_ms);
 }
 
diff --git a/fastboot/fastboot.cpp b/fastboot/fastboot.cpp
index 3a3503e..780ff50 100644
--- a/fastboot/fastboot.cpp
+++ b/fastboot/fastboot.cpp
@@ -82,7 +82,7 @@
 // libsparse will support INT_MAX, but this results in large allocations, so
 // let's keep it at 1GB to avoid memory pressure on the host.
 static constexpr int64_t RESPARSE_LIMIT = 1 * 1024 * 1024 * 1024;
-static int64_t sparse_limit = -1;
+static uint64_t sparse_limit = 0;
 static int64_t target_sparse_limit = -1;
 
 static unsigned g_base_addr = 0x10000000;
@@ -375,7 +375,7 @@
             " -w                         Wipe userdata.\n"
             " -s SERIAL                  Specify a USB device.\n"
             " -s tcp|udp:HOST[:PORT]     Specify a network device.\n"
-            " -S SIZE[K|M|G]             Use sparse files above this limit (0 to disable).\n"
+            " -S SIZE[K|M|G]             Break into sparse files no larger than SIZE.\n"
             " --slot SLOT                Use SLOT; 'all' for both slots, 'other' for\n"
             "                            non-current slot (default: current active slot).\n"
             " --set-active[=SLOT]        Sets the active slot before rebooting.\n"
@@ -410,12 +410,6 @@
         if (!g_cmdline.empty()) {
             bootimg_set_cmdline(reinterpret_cast<boot_img_hdr_v1*>(kdata), g_cmdline);
         }
-        uint32_t header_version_existing =
-                reinterpret_cast<boot_img_hdr_v1*>(kdata)->header_version;
-        if (g_boot_img_hdr.header_version != header_version_existing) {
-            die("header version mismatch, expected: %" PRIu32 " found %" PRIu32 "",
-                g_boot_img_hdr.header_version, header_version_existing);
-        }
 
         if (!ramdisk.empty()) die("cannot boot a boot.img *and* ramdisk");
 
@@ -736,13 +730,10 @@
 }
 
 static int64_t get_sparse_limit(Transport* transport, int64_t size) {
-    int64_t limit;
-
-    if (sparse_limit == 0) {
-        return 0;
-    } else if (sparse_limit > 0) {
-        limit = sparse_limit;
-    } else {
+    int64_t limit = sparse_limit;
+    if (limit == 0) {
+        // Unlimited, so see what the target device's limit is.
+        // TODO: shouldn't we apply this limit even if you've used -S?
         if (target_sparse_limit == -1) {
             target_sparse_limit = get_target_sparse_limit(transport);
         }
@@ -1203,47 +1194,6 @@
     fb_queue_command(command, "");
 }
 
-static int64_t parse_num(const char *arg)
-{
-    char *endptr;
-    unsigned long long num;
-
-    num = strtoull(arg, &endptr, 0);
-    if (endptr == arg) {
-        return -1;
-    }
-
-    if (*endptr == 'k' || *endptr == 'K') {
-        if (num >= (-1ULL) / 1024) {
-            return -1;
-        }
-        num *= 1024LL;
-        endptr++;
-    } else if (*endptr == 'm' || *endptr == 'M') {
-        if (num >= (-1ULL) / (1024 * 1024)) {
-            return -1;
-        }
-        num *= 1024LL * 1024LL;
-        endptr++;
-    } else if (*endptr == 'g' || *endptr == 'G') {
-        if (num >= (-1ULL) / (1024 * 1024 * 1024)) {
-            return -1;
-        }
-        num *= 1024LL * 1024LL * 1024LL;
-        endptr++;
-    }
-
-    if (*endptr != '\0') {
-        return -1;
-    }
-
-    if (num > INT64_MAX) {
-        return -1;
-    }
-
-    return num;
-}
-
 static std::string fb_fix_numeric_var(std::string var) {
     // Some bootloaders (angler, for example), send spurious leading whitespace.
     var = android::base::Trim(var);
@@ -1477,8 +1427,9 @@
                     serial = optarg;
                     break;
                 case 'S':
-                    sparse_limit = parse_num(optarg);
-                    if (sparse_limit < 0) die("invalid sparse limit");
+                    if (!android::base::ParseByteCount(optarg, &sparse_limit)) {
+                        die("invalid sparse limit %s", optarg);
+                    }
                     break;
                 case 'v':
                     set_verbose();
diff --git a/init/README.md b/init/README.md
index 59ddd77..b14521c 100644
--- a/init/README.md
+++ b/init/README.md
@@ -282,6 +282,10 @@
   "shutdown critical" will be killed. When the service tagged with "shutdown critical"
   is not running when shut down starts, it will be started.
 
+`sigstop`
+> Send SIGSTOP to the service immediately before exec is called. This is intended for debugging.
+  See the below section on debugging for how this can be used.
+
 `socket <name> <type> <perm> [ <user> [ <group> [ <seclabel> ] ] ]`
 > Create a unix domain socket named /dev/socket/_name_ and pass its fd to the
   launched process.  _type_ must be "dgram", "stream" or "seqpacket".  User and
@@ -708,23 +712,39 @@
 
 Debugging init
 --------------
-By default, programs executed by init will drop stdout and stderr into
-/dev/null. To help with debugging, you can execute your program via the
-Android program logwrapper. This will redirect stdout/stderr into the
-Android logging system (accessed via logcat).
+Launching init services without init is not recommended as init sets up a significant amount of
+environment (user, groups, security label, capabilities, etc) that is hard to replicate manually.
 
-For example
-service akmd /system/bin/logwrapper /sbin/akmd
+If it is required to debug a service from its very start, the `sigstop` service option is added.
+This option will send SIGSTOP to a service immediately before calling exec. This gives a window
+where developers can attach a debugger, strace, etc before continuing the service with SIGCONT.
 
-For quicker turnaround when working on init itself, use:
+This flag can also be dynamically controled via the ctl.sigstop_on and ctl.sigstop_off properties.
 
-    mm -j &&
-    m ramdisk-nodeps &&
-    m bootimage-nodeps &&
-    adb reboot bootloader &&
-    fastboot boot $ANDROID_PRODUCT_OUT/boot.img
+Below is an example of dynamically debugging logd via the above:
 
-Alternatively, use the emulator:
+    stop logd
+    setprop ctl.sigstop_on logd
+    start logd
+    ps -e | grep logd
+    > logd          4343     1   18156   1684 do_signal_stop 538280 T init
+    gdbclient.py -p 4343
+    b main
+    c
+    c
+    c
+    > Breakpoint 1, main (argc=1, argv=0x7ff8c9a488) at system/core/logd/main.cpp:427
 
-    emulator -partition-size 1024 \
-        -verbose -show-kernel -no-window
+Below is an example of doing the same but with strace
+
+    stop logd
+    setprop ctl.sigstop_on logd
+    start logd
+    ps -e | grep logd
+    > logd          4343     1   18156   1684 do_signal_stop 538280 T init
+    strace -p 4343
+
+    (From a different shell)
+    kill -SIGCONT 4343
+
+    > strace runs
diff --git a/init/bootchart.cpp b/init/bootchart.cpp
index 379b4fa..c2cf573 100644
--- a/init/bootchart.cpp
+++ b/init/bootchart.cpp
@@ -32,12 +32,14 @@
 #include <mutex>
 #include <thread>
 
+#include <android-base/chrono_utils.h>
 #include <android-base/file.h>
 #include <android-base/logging.h>
 #include <android-base/properties.h>
 #include <android-base/stringprintf.h>
 
 using android::base::StringPrintf;
+using android::base::boot_clock;
 using namespace std::chrono_literals;
 
 namespace android {
@@ -50,9 +52,9 @@
 static bool g_bootcharting_finished;
 
 static long long get_uptime_jiffies() {
-  std::string uptime;
-  if (!android::base::ReadFileToString("/proc/uptime", &uptime)) return 0;
-  return 100LL * strtod(uptime.c_str(), NULL);
+    constexpr int64_t kNanosecondsPerJiffy = 10000000;
+    boot_clock::time_point uptime = boot_clock::now();
+    return uptime.time_since_epoch().count() / kNanosecondsPerJiffy;
 }
 
 static std::unique_ptr<FILE, decltype(&fclose)> fopen_unique(const char* filename,
diff --git a/init/builtins.cpp b/init/builtins.cpp
index fc74dda..8bd92cc 100644
--- a/init/builtins.cpp
+++ b/init/builtins.cpp
@@ -82,6 +82,7 @@
 static constexpr std::chrono::nanoseconds kCommandRetryTimeout = 5s;
 
 static Result<Success> reboot_into_recovery(const std::vector<std::string>& options) {
+    LOG(ERROR) << "Rebooting into recovery";
     std::string err;
     if (!write_bootloader_message(options, &err)) {
         return Error() << "Failed to set bootloader message: " << err;
@@ -285,11 +286,8 @@
 
     if (e4crypt_is_native()) {
         if (e4crypt_set_directory_policy(args[1].c_str())) {
-            const std::vector<std::string> options = {
-                "--prompt_and_wipe_data",
-                "--reason=set_policy_failed:"s + args[1]};
-            reboot_into_recovery(options);
-            return Success();
+            return reboot_into_recovery(
+                {"--prompt_and_wipe_data", "--reason=set_policy_failed:"s + args[1]});
         }
     }
     return Success();
@@ -493,8 +491,7 @@
         /* Setup a wipe via recovery, and reboot into recovery */
         PLOG(ERROR) << "fs_mgr_mount_all suggested recovery, so wiping data via recovery.";
         const std::vector<std::string> options = {"--wipe_data", "--reason=fs_mgr_mount_all" };
-        reboot_into_recovery(options);
-        return Success();
+        return reboot_into_recovery(options);
         /* If reboot worked, there is no return. */
     } else if (code == FS_MGR_MNTALL_DEV_FILE_ENCRYPTED) {
         if (e4crypt_install_keyring()) {
@@ -522,6 +519,7 @@
         if (e4crypt_install_keyring()) {
             return Error() << "e4crypt_install_keyring() failed";
         }
+        property_set("ro.crypto.state", "encrypted");
         property_set("ro.crypto.type", "file");
 
         // Although encrypted, vold has already set the device up, so we do not need to
@@ -987,6 +985,29 @@
     return android::base::GetProperty("ro.crypto.type", "") == "file";
 }
 
+static Result<Success> ExecWithRebootOnFailure(const std::string& reboot_reason,
+                                               const BuiltinArguments& args) {
+    auto service = Service::MakeTemporaryOneshotService(args.args);
+    if (!service) {
+        return Error() << "Could not create exec service";
+    }
+    service->AddReapCallback([reboot_reason](const siginfo_t& siginfo) {
+        if (siginfo.si_code != CLD_EXITED || siginfo.si_status != 0) {
+            if (e4crypt_is_native()) {
+                LOG(ERROR) << "Rebooting into recovery, reason: " << reboot_reason;
+                reboot_into_recovery({"--prompt_and_wipe_data", "--reason="s + reboot_reason});
+            } else {
+                LOG(ERROR) << "Failure (reboot suppressed): " << reboot_reason;
+            }
+        }
+    });
+    if (auto result = service->ExecStart(); !result) {
+        return Error() << "Could not start exec service: " << result.error();
+    }
+    ServiceList::GetInstance().AddService(std::move(service));
+    return Success();
+}
+
 static Result<Success> do_installkey(const BuiltinArguments& args) {
     if (!is_file_crypto()) return Success();
 
@@ -994,15 +1015,15 @@
     if (!make_dir(unencrypted_dir, 0700) && errno != EEXIST) {
         return ErrnoError() << "Failed to create " << unencrypted_dir;
     }
-    std::vector<std::string> exec_args = {"exec", "/system/bin/vdc", "--wait", "cryptfs",
-                                          "enablefilecrypto"};
-    return do_exec({std::move(exec_args), args.context});
+    return ExecWithRebootOnFailure(
+        "enablefilecrypto_failed",
+        {{"exec", "/system/bin/vdc", "--wait", "cryptfs", "enablefilecrypto"}, args.context});
 }
 
 static Result<Success> do_init_user0(const BuiltinArguments& args) {
-    std::vector<std::string> exec_args = {"exec", "/system/bin/vdc", "--wait", "cryptfs",
-                                          "init_user0"};
-    return do_exec({std::move(exec_args), args.context});
+    return ExecWithRebootOnFailure(
+        "init_user0_failed",
+        {{"exec", "/system/bin/vdc", "--wait", "cryptfs", "init_user0"}, args.context});
 }
 
 // Builtin-function-map start
diff --git a/init/init.cpp b/init/init.cpp
index 2f3b28a..0d5690b 100644
--- a/init/init.cpp
+++ b/init/init.cpp
@@ -238,6 +238,10 @@
 static const std::map<std::string, ControlMessageFunction>& get_control_message_map() {
     // clang-format off
     static const std::map<std::string, ControlMessageFunction> control_message_functions = {
+        {"sigstop_on",        {ControlTarget::SERVICE,
+                               [](auto* service) { service->set_sigstop(true); return Success(); }}},
+        {"sigstop_off",       {ControlTarget::SERVICE,
+                               [](auto* service) { service->set_sigstop(false); return Success(); }}},
         {"start",             {ControlTarget::SERVICE,   DoControlStart}},
         {"stop",              {ControlTarget::SERVICE,   DoControlStop}},
         {"restart",           {ControlTarget::SERVICE,   DoControlRestart}},
@@ -624,6 +628,14 @@
         mknod("/dev/random", S_IFCHR | 0666, makedev(1, 8));
         mknod("/dev/urandom", S_IFCHR | 0666, makedev(1, 9));
 
+        // Mount staging areas for devices managed by vold
+        // See storage config details at http://source.android.com/devices/storage/
+        mount("tmpfs", "/mnt", "tmpfs", MS_NOEXEC | MS_NOSUID | MS_NODEV,
+              "mode=0755,uid=0,gid=1000");
+        // /mnt/vendor is used to mount vendor-specific partitions that can not be
+        // part of the vendor partition, e.g. because they are mounted read-write.
+        mkdir("/mnt/vendor", 0755);
+
         // Now that tmpfs is mounted on /dev and we have /dev/kmsg, we can actually
         // talk to the outside world...
         InitKernelLogging(argv);
diff --git a/init/service.cpp b/init/service.cpp
index 694e5e7..03c2cee 100644
--- a/init/service.cpp
+++ b/init/service.cpp
@@ -155,7 +155,7 @@
     }
 }
 
-static bool ExpandArgsAndExecv(const std::vector<std::string>& args) {
+static bool ExpandArgsAndExecv(const std::vector<std::string>& args, bool sigstop) {
     std::vector<std::string> expanded_args;
     std::vector<char*> c_strings;
 
@@ -169,6 +169,10 @@
     }
     c_strings.push_back(nullptr);
 
+    if (sigstop) {
+        kill(getpid(), SIGSTOP);
+    }
+
     return execv(c_strings[0], c_strings.data()) == 0;
 }
 
@@ -303,7 +307,7 @@
     }
 }
 
-void Service::Reap() {
+void Service::Reap(const siginfo_t& siginfo) {
     if (!(flags_ & SVC_ONESHOT) || (flags_ & SVC_RESTART)) {
         KillProcessGroup(SIGKILL);
     }
@@ -312,6 +316,10 @@
     std::for_each(descriptors_.begin(), descriptors_.end(),
                   std::bind(&DescriptorInfo::Clean, std::placeholders::_1));
 
+    for (const auto& f : reap_callbacks_) {
+        f(siginfo);
+    }
+
     if (flags_ & SVC_EXEC) UnSetExec();
 
     if (flags_ & SVC_TEMPORARY) return;
@@ -578,6 +586,11 @@
     return Success();
 }
 
+Result<Success> Service::ParseSigstop(const std::vector<std::string>& args) {
+    sigstop_ = true;
+    return Success();
+}
+
 Result<Success> Service::ParseSetenv(const std::vector<std::string>& args) {
     environment_vars_.emplace_back(args[1], args[2]);
     return Success();
@@ -700,6 +713,7 @@
         {"seclabel",    {1,     1,    &Service::ParseSeclabel}},
         {"setenv",      {2,     2,    &Service::ParseSetenv}},
         {"shutdown",    {1,     1,    &Service::ParseShutdown}},
+        {"sigstop",     {0,     0,    &Service::ParseSigstop}},
         {"socket",      {3,     6,    &Service::ParseSocket}},
         {"user",        {1,     1,    &Service::ParseUser}},
         {"writepid",    {1,     kMax, &Service::ParseWritepid}},
@@ -858,7 +872,7 @@
         // priority. Aborts on failure.
         SetProcessAttributes();
 
-        if (!ExpandArgsAndExecv(args_)) {
+        if (!ExpandArgsAndExecv(args_, sigstop_)) {
             PLOG(ERROR) << "cannot execve('" << args_[0] << "')";
         }
 
diff --git a/init/service.h b/init/service.h
index d46a413..cf38f69 100644
--- a/init/service.h
+++ b/init/service.h
@@ -17,6 +17,7 @@
 #ifndef _INIT_SERVICE_H
 #define _INIT_SERVICE_H
 
+#include <signal.h>
 #include <sys/resource.h>
 #include <sys/types.h>
 
@@ -81,7 +82,7 @@
     void Stop();
     void Terminate();
     void Restart();
-    void Reap();
+    void Reap(const siginfo_t& siginfo);
     void DumpState() const;
     void SetShutdownCritical() { flags_ |= SVC_SHUTDOWN_CRITICAL; }
     bool IsShutdownCritical() const { return (flags_ & SVC_SHUTDOWN_CRITICAL) != 0; }
@@ -89,6 +90,9 @@
         is_exec_service_running_ = false;
         flags_ &= ~SVC_EXEC;
     }
+    void AddReapCallback(std::function<void(const siginfo_t& siginfo)> callback) {
+        reap_callbacks_.emplace_back(std::move(callback));
+    }
 
     static bool is_exec_service_running() { return is_exec_service_running_; }
 
@@ -114,6 +118,7 @@
     bool is_override() const { return override_; }
     bool process_cgroup_empty() const { return process_cgroup_empty_; }
     unsigned long start_order() const { return start_order_; }
+    void set_sigstop(bool value) { sigstop_ = value; }
     const std::vector<std::string>& args() const { return args_; }
 
   private:
@@ -149,6 +154,7 @@
     Result<Success> ParseSeclabel(const std::vector<std::string>& args);
     Result<Success> ParseSetenv(const std::vector<std::string>& args);
     Result<Success> ParseShutdown(const std::vector<std::string>& args);
+    Result<Success> ParseSigstop(const std::vector<std::string>& args);
     Result<Success> ParseSocket(const std::vector<std::string>& args);
     Result<Success> ParseFile(const std::vector<std::string>& args);
     Result<Success> ParseUser(const std::vector<std::string>& args);
@@ -209,7 +215,11 @@
 
     std::vector<std::pair<int, rlimit>> rlimits_;
 
+    bool sigstop_ = false;
+
     std::vector<std::string> args_;
+
+    std::vector<std::function<void(const siginfo_t& siginfo)>> reap_callbacks_;
 };
 
 class ServiceList {
diff --git a/init/sigchld_handler.cpp b/init/sigchld_handler.cpp
index 3ec76df..0b03324 100644
--- a/init/sigchld_handler.cpp
+++ b/init/sigchld_handler.cpp
@@ -81,16 +81,15 @@
         }
     }
 
-    auto status = siginfo.si_status;
-    if (WIFEXITED(status)) {
-        LOG(INFO) << name << " exited with status " << WEXITSTATUS(status) << wait_string;
-    } else if (WIFSIGNALED(status)) {
-        LOG(INFO) << name << " killed by signal " << WTERMSIG(status) << wait_string;
+    if (siginfo.si_code == CLD_EXITED) {
+        LOG(INFO) << name << " exited with status " << siginfo.si_status << wait_string;
+    } else {
+        LOG(INFO) << name << " received signal " << siginfo.si_status << wait_string;
     }
 
     if (!service) return true;
 
-    service->Reap();
+    service->Reap(siginfo);
 
     if (service->flags() & SVC_TEMPORARY) {
         ServiceList::GetInstance().RemoveService(*service);
diff --git a/libcutils/include/private/android_filesystem_config.h b/libcutils/include/private/android_filesystem_config.h
index 5d17698..3be8ad0 100644
--- a/libcutils/include/private/android_filesystem_config.h
+++ b/libcutils/include/private/android_filesystem_config.h
@@ -129,6 +129,8 @@
 #define AID_STATSD 1066          /* statsd daemon */
 #define AID_INCIDENTD 1067       /* incidentd daemon */
 #define AID_SECURE_ELEMENT 1068  /* secure element subsystem */
+#define AID_LMKD 1069            /* low memory killer daemon */
+#define AID_LLKD 1070            /* live lock daemon */
 /* Changes to this file must be made in AOSP, *not* in internal branches. */
 
 #define AID_SHELL 2000 /* adb and debug shell user */
diff --git a/libmemunreachable/Android.bp b/libmemunreachable/Android.bp
index f872d0f..619ee34 100644
--- a/libmemunreachable/Android.bp
+++ b/libmemunreachable/Android.bp
@@ -46,6 +46,12 @@
             static_libs: ["libunwind_llvm"],
         },
     },
+
+    // TODO(b/78118944), clang lld link flags do not work with special link
+    // rules for libunwind_llvm yet. Linked aosp_arm-eng image failed to
+    // boot up in the emulator.
+    use_clang_lld: false,
+
     export_include_dirs: ["include"],
     local_include_dirs: ["include"],
 }
diff --git a/libziparchive/Android.bp b/libziparchive/Android.bp
index 075fb86..ed1b9bc 100644
--- a/libziparchive/Android.bp
+++ b/libziparchive/Android.bp
@@ -61,6 +61,7 @@
     vndk: {
         enabled: true,
     },
+    double_loadable: true,
 
     defaults: [
         "libziparchive_defaults",
diff --git a/llkd/Android.bp b/llkd/Android.bp
new file mode 100644
index 0000000..a6edd26
--- /dev/null
+++ b/llkd/Android.bp
@@ -0,0 +1,42 @@
+cc_library_headers {
+    name: "llkd_headers",
+
+    export_include_dirs: ["include"],
+}
+
+cc_library_static {
+    name: "libllkd",
+
+    srcs: [
+        "libllkd.cpp",
+    ],
+
+    shared_libs: [
+        "libbase",
+        "libcutils",
+        "liblog",
+    ],
+
+    export_include_dirs: ["include"],
+
+    cflags: ["-Werror"],
+}
+
+cc_binary {
+    name: "llkd",
+
+    srcs: [
+        "llkd.cpp",
+    ],
+    shared_libs: [
+        "libbase",
+        "libcutils",
+        "liblog",
+    ],
+    static_libs: [
+        "libllkd",
+    ],
+    cflags: ["-Werror"],
+
+    init_rc: ["llkd.rc"],
+}
diff --git a/llkd/OWNERS b/llkd/OWNERS
new file mode 100644
index 0000000..b6af537
--- /dev/null
+++ b/llkd/OWNERS
@@ -0,0 +1,2 @@
+salyzyn@google.com
+surenb@google.com
diff --git a/llkd/README.md b/llkd/README.md
new file mode 100644
index 0000000..b2ba2a2
--- /dev/null
+++ b/llkd/README.md
@@ -0,0 +1,117 @@
+Android Live-LocK Daemon
+========================
+
+Introduction
+------------
+
+Android Live-LocK Daemon (llkd) is used to catch kernel deadlocks and mitigate.
+
+Code is structured to allow integration into another service as either as part
+of the main loop, or spun off as a thread should that be necessary.  A default
+standalone implementation is provided by llkd component.
+
+The 'C' interface from libllkd component is thus:
+
+    #include "llkd.h"
+    bool llkInit(const char* threadname) /* return true if enabled */
+    unsigned llkCheckMillseconds(void)   /* ms to sleep for next check */
+
+If a threadname is provided, a thread will be automatically spawned, otherwise
+caller must call llkCheckMilliseconds in its main loop.  Function will return
+the period of time before the next expected call to this handler.
+
+Operations
+----------
+
+If a thread is in D or Z state with no forward progress for longer than
+ro.llk.timeout_ms, or ro.llk.[D|Z].timeout_ms, kill the process or parent
+process respectively.  If another scan shows the same process continues to
+exist, then have a confirmed live-lock condition and need to panic.  Panic
+the kernel in a manner to provide the greatest bugreporting details as to the
+condition.  Add a alarm self watchdog should llkd ever get locked up that is
+double the expected time to flow through the mainloop.  Sampling is every
+ro.llk_sample_ms.
+
+Default will not monitor init, or [kthreadd] and all that [kthreadd] spawns.
+This reduces the effectiveness of llkd by limiting its coverage.  If there is
+value in covering [kthreadd] spawned threads, the requirement will be that
+the drivers not remain in a persistent 'D' state, or that they have mechanisms
+to recover the thread should it be killed externally (this is good driver
+coding hygiene, a common request to add such to publicly reviewed kernel.org
+maintained drivers).  For instance use wait_event_interruptible() instead of
+wait_event().  The blacklists can be adjusted accordingly if these
+conditions are met to cover kernel components.
+
+An accompanying gTest set have been added, and will setup a persistent D or Z
+process, with and without forward progress, but not in a live-lock state
+because that would require a buggy kernel, or a module or kernel modification
+to stimulate.  The test will check that llkd will mitigate first by killing
+the appropriate process.  D state is setup by vfork() waiting for exec() in
+child process.  Z state is setup by fork() and an un-waited for child process.
+Should be noted that both of these conditions should never happen on Android
+on purpose, and llkd effectively sweeps up processes that create these
+conditions.  If the test can, it will reconfigure llkd to expedite the test
+duration by adjusting the ro.llk.* Android properties.  Tests run the D state
+with some scheduling progress to ensure that ABA checking prevents false
+triggers. If 100% reliable ABA on platform, then ro.llk.killtest can be
+set to false; however this will result in some of the unit tests to panic
+kernel instead of deal with more graceful kill operation.
+
+Android Properties
+------------------
+
+Android Properties llkd respond to (<prop>_ms parms are in milliseconds):
+
+#### ro.config.low_ram
+default false, if true do not sysrq t (dump all threads).
+
+#### ro.llk.enable
+default false, allow live-lock daemon to be enabled.
+
+#### llk.enable
+default ro.llk.enable, and evaluated for eng.
+
+#### ro.khungtask.enable
+default false, allow [khungtask] daemon to be enabled.
+
+#### khungtask.enable
+default ro.khungtask.enable and evaluated for eng.
+
+#### ro.llk.mlockall
+default false, enable call to mlockall().
+
+#### ro.khungtask.timeout
+default value 12 minutes, [khungtask] maximum timelimit.
+
+#### ro.llk.timeout_ms
+default 10 minutes, D or Z maximum timelimit, double this value and it sets
+the alarm watchdog for llkd.
+
+#### ro.llk.D.timeout_ms
+default ro.llk.timeout_ms, D maximum timelimit.
+
+#### ro.llk.Z.timeout_ms
+default ro.llk.timeout_ms, Z maximum timelimit.
+
+#### ro.llk.check_ms
+default 2 minutes samples of threads for D or Z.
+
+#### ro.llk.blacklist.process
+default 0,1,2 (kernel, init and [kthreadd]) plus process names
+init,[kthreadd],[khungtaskd],lmkd,lmkd.llkd,llkd,watchdogd,
+[watchdogd],[watchdogd/0],...,[watchdogd/<get_nprocs-1>].
+
+#### ro.llk.blacklist.parent
+default 0,2 (kernel and [kthreadd]).
+
+#### ro.llk.blacklist.uid
+default <empty>, comma separated list of uid numbers or names.
+
+Architectural Concerns
+----------------------
+
+- Create kernel module and associated gTest to actually test panic.
+- Create gTest to test out blacklist (ro.llk.blacklist.<properties> generally
+  not be inputs).  Could require more test-only interfaces to libllkd.
+- Speed up gTest using something else than ro.llk.<properties>, which should
+  not be inputs.
diff --git a/llkd/include/llkd.h b/llkd/include/llkd.h
new file mode 100644
index 0000000..e3ae4bb
--- /dev/null
+++ b/llkd/include/llkd.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright (C) 2018 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef _LLKD_H_
+#define _LLKD_H_
+
+#ifndef LOG_TAG
+#define LOG_TAG "livelock"
+#endif
+
+#include <stdbool.h>
+#include <sys/cdefs.h>
+
+__BEGIN_DECLS
+
+bool llkInit(const char* threadname); /* threadname NULL, not spawned */
+unsigned llkCheckMilliseconds(void);
+
+/* clang-format off */
+#define LLK_ENABLE_WRITEABLE_PROPERTY  "llk.enable"
+#define LLK_ENABLE_PROPERTY            "ro." LLK_ENABLE_WRITEABLE_PROPERTY
+#define LLK_ENABLE_DEFAULT             false /* "eng" and userdebug true */
+#define KHT_ENABLE_WRITEABLE_PROPERTY  "khungtask.enable"
+#define KHT_ENABLE_PROPERTY            "ro." KHT_ENABLE_WRITEABLE_PROPERTY
+#define LLK_MLOCKALL_PROPERTY          "ro.llk.mlockall"
+#define LLK_MLOCKALL_DEFAULT           true
+#define LLK_KILLTEST_PROPERTY          "ro.llk.killtest"
+#define LLK_KILLTEST_DEFAULT           true
+#define LLK_TIMEOUT_MS_PROPERTY        "ro.llk.timeout_ms"
+#define KHT_TIMEOUT_PROPERTY           "ro.khungtask.timeout"
+#define LLK_D_TIMEOUT_MS_PROPERTY      "ro.llk.D.timeout_ms"
+#define LLK_Z_TIMEOUT_MS_PROPERTY      "ro.llk.Z.timeout_ms"
+#define LLK_CHECK_MS_PROPERTY          "ro.llk.check_ms"
+/* LLK_CHECK_MS_DEFAULT = actual timeout_ms / LLK_CHECKS_PER_TIMEOUT_DEFAULT */
+#define LLK_CHECKS_PER_TIMEOUT_DEFAULT 5
+#define LLK_BLACKLIST_PROCESS_PROPERTY "ro.llk.blacklist.process"
+#define LLK_BLACKLIST_PROCESS_DEFAULT  \
+    "0,1,2,init,[kthreadd],[khungtaskd],lmkd,lmkd.llkd,llkd,watchdogd,[watchdogd],[watchdogd/0]"
+#define LLK_BLACKLIST_PARENT_PROPERTY  "ro.llk.blacklist.parent"
+#define LLK_BLACKLIST_PARENT_DEFAULT   "0,2,[kthreadd]"
+#define LLK_BLACKLIST_UID_PROPERTY     "ro.llk.blacklist.uid"
+#define LLK_BLACKLIST_UID_DEFAULT      ""
+/* clang-format on */
+
+__END_DECLS
+
+#ifdef __cplusplus
+extern "C++" { /* In case this included wrapped with __BEGIN_DECLS */
+
+#include <chrono>
+
+__BEGIN_DECLS
+/* C++ code allowed to not specify threadname argument for this C linkage */
+bool llkInit(const char* threadname = nullptr);
+__END_DECLS
+std::chrono::milliseconds llkCheck(bool checkRunning = false);
+
+/* clang-format off */
+#define LLK_TIMEOUT_MS_DEFAULT  std::chrono::duration_cast<milliseconds>(std::chrono::minutes(10))
+#define LLK_TIMEOUT_MS_MINIMUM  std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::seconds(10))
+#define LLK_CHECK_MS_MINIMUM    std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::seconds(1))
+/* clang-format on */
+
+} /* extern "C++" */
+#endif /* __cplusplus */
+
+#endif /* _LLKD_H_ */
diff --git a/llkd/libllkd.cpp b/llkd/libllkd.cpp
new file mode 100644
index 0000000..f357cc2
--- /dev/null
+++ b/llkd/libllkd.cpp
@@ -0,0 +1,1168 @@
+/*
+ * Copyright (C) 2018 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "llkd.h"
+
+#include <ctype.h>
+#include <dirent.h>  // opendir() and readdir()
+#include <errno.h>
+#include <fcntl.h>
+#include <pthread.h>
+#include <pwd.h>  // getpwuid()
+#include <signal.h>
+#include <stdint.h>
+#include <sys/cdefs.h>  // ___STRING, __predict_true() and _predict_false()
+#include <sys/mman.h>   // mlockall()
+#include <sys/prctl.h>
+#include <sys/stat.h>     // lstat()
+#include <sys/syscall.h>  // __NR_getdents64
+#include <sys/sysinfo.h>  // get_nprocs_conf()
+#include <sys/types.h>
+#include <time.h>
+#include <unistd.h>
+
+#include <chrono>
+#include <ios>
+#include <sstream>
+#include <string>
+#include <unordered_map>
+#include <unordered_set>
+
+#include <android-base/file.h>
+#include <android-base/logging.h>
+#include <android-base/parseint.h>
+#include <android-base/properties.h>
+#include <android-base/strings.h>
+#include <cutils/android_get_control_file.h>
+#include <log/log_main.h>
+
+#define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x)))
+
+#define TASK_COMM_LEN 16  // internal kernel, not uapi, from .../linux/include/linux/sched.h
+
+using namespace std::chrono_literals;
+using namespace std::chrono;
+
+namespace {
+
+constexpr pid_t kernelPid = 0;
+constexpr pid_t initPid = 1;
+constexpr pid_t kthreaddPid = 2;
+
+constexpr char procdir[] = "/proc/";
+
+// Configuration
+milliseconds llkUpdate;                              // last check ms signature
+milliseconds llkCycle;                               // ms to next thread check
+bool llkEnable = LLK_ENABLE_DEFAULT;                 // llk daemon enabled
+bool llkRunning = false;                             // thread is running
+bool llkMlockall = LLK_MLOCKALL_DEFAULT;             // run mlocked
+bool llkTestWithKill = LLK_KILLTEST_DEFAULT;         // issue test kills
+milliseconds llkTimeoutMs = LLK_TIMEOUT_MS_DEFAULT;  // default timeout
+enum { llkStateD, llkStateZ, llkNumStates };         // state indexes
+milliseconds llkStateTimeoutMs[llkNumStates];        // timeout override for each detection state
+milliseconds llkCheckMs;                             // checking interval to inspect any
+                                                     // persistent live-locked states
+bool llkLowRam;                                      // ro.config.low_ram
+bool khtEnable = LLK_ENABLE_DEFAULT;                 // [khungtaskd] panic
+// [khungtaskd] should have a timeout beyond the granularity of llkTimeoutMs.
+// Provides a wide angle of margin b/c khtTimeout is also its granularity.
+seconds khtTimeout = duration_cast<seconds>(llkTimeoutMs * (1 + LLK_CHECKS_PER_TIMEOUT_DEFAULT) /
+                                            LLK_CHECKS_PER_TIMEOUT_DEFAULT);
+
+// Blacklist variables, initialized with comma separated lists of high false
+// positive and/or dangerous references, e.g. without self restart, for pid,
+// ppid, name and uid:
+
+// list of pids, or tids or names to skip. kernel pid (0), init pid (1),
+// [kthreadd] pid (2), ourselves, "init", "[kthreadd]", "lmkd", "llkd" or
+// combinations of watchdogd in kernel and user space.
+std::unordered_set<std::string> llkBlacklistProcess;
+// list of parent pids, comm or cmdline names to skip. default:
+// kernel pid (0), [kthreadd] (2), or ourselves, enforced and implied
+std::unordered_set<std::string> llkBlacklistParent;
+// list of uids, and uid names, to skip, default nothing
+std::unordered_set<std::string> llkBlacklistUid;
+
+class dir {
+  public:
+    enum level { proc, task, numLevels };
+
+  private:
+    int fd;
+    size_t available_bytes;
+    dirent* next;
+    // each directory level picked to be just north of 4K in size
+    static constexpr size_t buffEntries = 15;
+    static dirent buff[numLevels][buffEntries];
+
+    bool fill(enum level index) {
+        if (index >= numLevels) return false;
+        if (available_bytes != 0) return true;
+        if (__predict_false(fd < 0)) return false;
+        // getdents64 has no libc wrapper
+        auto rc = TEMP_FAILURE_RETRY(syscall(__NR_getdents64, fd, buff[index], sizeof(buff[0]), 0));
+        if (rc <= 0) return false;
+        available_bytes = rc;
+        next = buff[index];
+        return true;
+    }
+
+  public:
+    dir() : fd(-1), available_bytes(0), next(nullptr) {}
+
+    explicit dir(const char* directory)
+        : fd(__predict_true(directory != nullptr)
+                 ? ::open(directory, O_CLOEXEC | O_DIRECTORY | O_RDONLY)
+                 : -1),
+          available_bytes(0),
+          next(nullptr) {}
+
+    explicit dir(const std::string&& directory)
+        : fd(::open(directory.c_str(), O_CLOEXEC | O_DIRECTORY | O_RDONLY)),
+          available_bytes(0),
+          next(nullptr) {}
+
+    explicit dir(const std::string& directory)
+        : fd(::open(directory.c_str(), O_CLOEXEC | O_DIRECTORY | O_RDONLY)),
+          available_bytes(0),
+          next(nullptr) {}
+
+    // Don't need any copy or move constructors.
+    explicit dir(const dir& c) = delete;
+    explicit dir(dir& c) = delete;
+    explicit dir(dir&& c) = delete;
+
+    ~dir() {
+        if (fd >= 0) {
+            ::close(fd);
+        }
+    }
+
+    operator bool() const { return fd >= 0; }
+
+    void reset(void) {
+        if (fd >= 0) {
+            ::close(fd);
+            fd = -1;
+            available_bytes = 0;
+            next = nullptr;
+        }
+    }
+
+    dir& reset(const char* directory) {
+        reset();
+        // available_bytes will _always_ be zero here as its value is
+        // intimately tied to fd < 0 or not.
+        fd = ::open(directory, O_CLOEXEC | O_DIRECTORY | O_RDONLY);
+        return *this;
+    }
+
+    void rewind(void) {
+        if (fd >= 0) {
+            ::lseek(fd, off_t(0), SEEK_SET);
+            available_bytes = 0;
+            next = nullptr;
+        }
+    }
+
+    dirent* read(enum level index = proc, dirent* def = nullptr) {
+        if (!fill(index)) return def;
+        auto ret = next;
+        available_bytes -= next->d_reclen;
+        next = reinterpret_cast<dirent*>(reinterpret_cast<char*>(next) + next->d_reclen);
+        return ret;
+    }
+} llkTopDirectory;
+
+dirent dir::buff[dir::numLevels][dir::buffEntries];
+
+// helper functions
+
+bool llkIsMissingExeLink(pid_t tid) {
+    char c;
+    // CAP_SYS_PTRACE is required to prevent ret == -1, but ENOENT is signal
+    auto ret = ::readlink((procdir + std::to_string(tid) + "/exe").c_str(), &c, sizeof(c));
+    return (ret == -1) && (errno == ENOENT);
+}
+
+// Common routine where caller accepts empty content as error/passthrough.
+// Reduces the churn of reporting read errors in the callers.
+std::string ReadFile(std::string&& path) {
+    std::string content;
+    if (!android::base::ReadFileToString(path, &content)) {
+        PLOG(DEBUG) << "Read " << path << " failed";
+        content = "";
+    }
+    return content;
+}
+
+std::string llkProcGetName(pid_t tid, const char* node = "/cmdline") {
+    std::string content = ReadFile(procdir + std::to_string(tid) + node);
+    static constexpr char needles[] = " \t\r\n";  // including trailing nul
+    auto pos = content.find_first_of(needles, 0, sizeof(needles));
+    if (pos != std::string::npos) {
+        content.erase(pos);
+    }
+    return content;
+}
+
+uid_t llkProcGetUid(pid_t tid) {
+    // Get the process' uid.  The following read from /status is admittedly
+    // racy, prone to corruption due to shape-changes.  The consequences are
+    // not catastrophic as we sample a few times before taking action.
+    //
+    // If /loginuid worked on reliably, or on Android (all tasks report -1)...
+    // Android lmkd causes /cgroup to contain memory:/<dom>/uid_<uid>/pid_<pid>
+    // which is tighter, but also not reliable.
+    std::string content = ReadFile(procdir + std::to_string(tid) + "/status");
+    static constexpr char Uid[] = "\nUid:";
+    auto pos = content.find(Uid);
+    if (pos == std::string::npos) {
+        return -1;
+    }
+    pos += ::strlen(Uid);
+    while ((pos < content.size()) && ::isblank(content[pos])) {
+        ++pos;
+    }
+    content.erase(0, pos);
+    for (pos = 0; (pos < content.size()) && ::isdigit(content[pos]); ++pos) {
+        ;
+    }
+    // Content of form 'Uid:	0	0	0	0', newline is error
+    if ((pos >= content.size()) || !::isblank(content[pos])) {
+        return -1;
+    }
+    content.erase(pos);
+    uid_t ret;
+    if (!android::base::ParseInt(content, &ret, uid_t(0))) {
+        return -1;
+    }
+    return ret;
+}
+
+struct proc {
+    pid_t tid;                     // monitored thread id (in Z or D state).
+    nanoseconds schedUpdate;       // /proc/<tid>/sched "se.avg.lastUpdateTime",
+    uint64_t nrSwitches;           // /proc/<tid>/sched "nr_switches" for
+                                   // refined ABA problem detection, determine
+                                   // forward scheduling progress.
+    milliseconds update;           // llkUpdate millisecond signature of last.
+    milliseconds count;            // duration in state.
+    pid_t pid;                     // /proc/<pid> before iterating through
+                                   // /proc/<pid>/task/<tid> for threads.
+    pid_t ppid;                    // /proc/<tid>/stat field 4 parent pid.
+    uid_t uid;                     // /proc/<tid>/status Uid: field.
+    unsigned time;                 // sum of /proc/<tid>/stat field 14 utime &
+                                   // 15 stime for coarse ABA problem detection.
+    std::string cmdline;           // cached /cmdline content
+    char state;                    // /proc/<tid>/stat field 3: Z or D
+                                   // (others we do not monitor: S, R, T or ?)
+    char comm[TASK_COMM_LEN + 3];  // space for adding '[' and ']'
+    bool exeMissingValid;          // exeMissing has been cached
+    bool cmdlineValid;             // cmdline has been cached
+    bool updated;                  // cleared before monitoring pass.
+    bool killed;                   // sent a kill to this thread, next panic...
+
+    void setComm(const char* _comm) { strncpy(comm + 1, _comm, sizeof(comm) - 2); }
+
+    proc(pid_t tid, pid_t pid, pid_t ppid, const char* _comm, int time, char state)
+        : tid(tid),
+          schedUpdate(0),
+          nrSwitches(0),
+          update(llkUpdate),
+          count(0),
+          pid(pid),
+          ppid(ppid),
+          uid(-1),
+          time(time),
+          state(state),
+          exeMissingValid(false),
+          cmdlineValid(false),
+          updated(true),
+          killed(!llkTestWithKill) {
+        memset(comm, '\0', sizeof(comm));
+        setComm(_comm);
+    }
+
+    const char* getComm(void) {
+        if (comm[1] == '\0') {  // comm Valid?
+            strncpy(comm + 1, llkProcGetName(tid, "/comm").c_str(), sizeof(comm) - 2);
+        }
+        if (!exeMissingValid) {
+            if (llkIsMissingExeLink(tid)) {
+                comm[0] = '[';
+            }
+            exeMissingValid = true;
+        }
+        size_t len = strlen(comm + 1);
+        if (__predict_true(len < (sizeof(comm) - 1))) {
+            if (comm[0] == '[') {
+                if ((comm[len] != ']') && __predict_true(len < (sizeof(comm) - 2))) {
+                    comm[++len] = ']';
+                    comm[++len] = '\0';
+                }
+            } else {
+                if (comm[len] == ']') {
+                    comm[len] = '\0';
+                }
+            }
+        }
+        return &comm[comm[0] != '['];
+    }
+
+    const char* getCmdline(void) {
+        if (!cmdlineValid) {
+            cmdline = llkProcGetName(tid);
+            cmdlineValid = true;
+        }
+        return cmdline.c_str();
+    }
+
+    uid_t getUid(void) {
+        if (uid <= 0) {  // Churn on root user, because most likely to setuid()
+            uid = llkProcGetUid(tid);
+        }
+        return uid;
+    }
+
+    void reset(void) {  // reset cache, if we detected pid rollover
+        uid = -1;
+        state = '?';
+        cmdline = "";
+        comm[0] = '\0';
+        exeMissingValid = false;
+        cmdlineValid = false;
+    }
+};
+
+std::unordered_map<pid_t, proc> tids;
+
+// Check range and setup defaults, in order of propagation:
+//     llkTimeoutMs
+//     llkCheckMs
+//     ...
+// KISS to keep it all self-contained, and called multiple times as parameters
+// are interpreted so that defaults, llkCheckMs and llkCycle make sense.
+void llkValidate() {
+    if (llkTimeoutMs == 0ms) {
+        llkTimeoutMs = LLK_TIMEOUT_MS_DEFAULT;
+    }
+    llkTimeoutMs = std::max(llkTimeoutMs, LLK_TIMEOUT_MS_MINIMUM);
+    if (llkCheckMs == 0ms) {
+        llkCheckMs = llkTimeoutMs / LLK_CHECKS_PER_TIMEOUT_DEFAULT;
+    }
+    llkCheckMs = std::min(llkCheckMs, llkTimeoutMs);
+
+    for (size_t state = 0; state < ARRAY_SIZE(llkStateTimeoutMs); ++state) {
+        if (llkStateTimeoutMs[state] == 0ms) {
+            llkStateTimeoutMs[state] = llkTimeoutMs;
+        }
+        llkStateTimeoutMs[state] =
+            std::min(std::max(llkStateTimeoutMs[state], LLK_TIMEOUT_MS_MINIMUM), llkTimeoutMs);
+        llkCheckMs = std::min(llkCheckMs, llkStateTimeoutMs[state]);
+    }
+
+    llkCheckMs = std::max(llkCheckMs, LLK_CHECK_MS_MINIMUM);
+    if (llkCycle == 0ms) {
+        llkCycle = llkCheckMs;
+    }
+    llkCycle = std::min(llkCycle, llkCheckMs);
+}
+
+milliseconds llkGetTimespecDiffMs(timespec* from, timespec* to) {
+    return duration_cast<milliseconds>(seconds(to->tv_sec - from->tv_sec)) +
+           duration_cast<milliseconds>(nanoseconds(to->tv_nsec - from->tv_nsec));
+}
+
+std::string llkProcGetName(pid_t tid, const char* comm, const char* cmdline) {
+    if ((cmdline != nullptr) && (*cmdline != '\0')) {
+        return cmdline;
+    }
+    if ((comm != nullptr) && (*comm != '\0')) {
+        return comm;
+    }
+
+    // UNLIKELY! Here because killed before we kill it?
+    // Assume change is afoot, do not call llkTidAlloc
+
+    // cmdline ?
+    std::string content = llkProcGetName(tid);
+    if (content.size() != 0) {
+        return content;
+    }
+    // Comm instead?
+    content = llkProcGetName(tid, "/comm");
+    if (llkIsMissingExeLink(tid) && (content.size() != 0)) {
+        return '[' + content + ']';
+    }
+    return content;
+}
+
+int llkKillOneProcess(pid_t pid, char state, pid_t tid, const char* tcomm = nullptr,
+                      const char* tcmdline = nullptr, const char* pcomm = nullptr,
+                      const char* pcmdline = nullptr) {
+    std::string forTid;
+    if (tid != pid) {
+        forTid = " for '" + llkProcGetName(tid, tcomm, tcmdline) + "' (" + std::to_string(tid) + ")";
+    }
+    LOG(INFO) << "Killing '" << llkProcGetName(pid, pcomm, pcmdline) << "' (" << pid
+              << ") to check forward scheduling progress in " << state << " state" << forTid;
+    // CAP_KILL required
+    errno = 0;
+    auto r = ::kill(pid, SIGKILL);
+    if (r) {
+        PLOG(ERROR) << "kill(" << pid << ")=" << r << ' ';
+    }
+
+    return r;
+}
+
+// Kill one process
+int llkKillOneProcess(pid_t pid, proc* tprocp) {
+    return llkKillOneProcess(pid, tprocp->state, tprocp->tid, tprocp->getComm(),
+                             tprocp->getCmdline());
+}
+
+// Kill one process specified by kprocp
+int llkKillOneProcess(proc* kprocp, proc* tprocp) {
+    if (kprocp == nullptr) {
+        return -2;
+    }
+
+    return llkKillOneProcess(kprocp->tid, tprocp->state, tprocp->tid, tprocp->getComm(),
+                             tprocp->getCmdline(), kprocp->getComm(), kprocp->getCmdline());
+}
+
+// Acquire file descriptor from environment, or open and cache it.
+// NB: cache is unnecessary in our current context, pedantically
+//     required to prevent leakage of file descriptors in the future.
+int llkFileToWriteFd(const std::string& file) {
+    static std::unordered_map<std::string, int> cache;
+    auto search = cache.find(file);
+    if (search != cache.end()) return search->second;
+    auto fd = android_get_control_file(file.c_str());
+    if (fd >= 0) return fd;
+    fd = TEMP_FAILURE_RETRY(::open(file.c_str(), O_WRONLY | O_CLOEXEC));
+    if (fd >= 0) cache.emplace(std::make_pair(file, fd));
+    return fd;
+}
+
+// Wrap android::base::WriteStringToFile to use android_get_control_file.
+bool llkWriteStringToFile(const std::string& string, const std::string& file) {
+    auto fd = llkFileToWriteFd(file);
+    if (fd < 0) return false;
+    return android::base::WriteStringToFd(string, fd);
+}
+
+bool llkWriteStringToFileConfirm(const std::string& string, const std::string& file) {
+    auto fd = llkFileToWriteFd(file);
+    auto ret = (fd < 0) ? false : android::base::WriteStringToFd(string, fd);
+    std::string content;
+    if (!android::base::ReadFileToString(file, &content)) return ret;
+    return android::base::Trim(content) == string;
+}
+
+void llkPanicKernel(bool dump, pid_t tid, const char* state) __noreturn;
+void llkPanicKernel(bool dump, pid_t tid, const char* state) {
+    auto sysrqTriggerFd = llkFileToWriteFd("/proc/sysrq-trigger");
+    if (sysrqTriggerFd < 0) {
+        // DYB
+        llkKillOneProcess(initPid, 'R', tid);
+        // The answer to life, the universe and everything
+        ::exit(42);
+        // NOTREACHED
+    }
+    ::sync();
+    if (dump) {
+        // Show all locks that are held
+        android::base::WriteStringToFd("d", sysrqTriggerFd);
+        // This can trigger hardware watchdog, that is somewhat _ok_.
+        // But useless if pstore configured for <256KB, low ram devices ...
+        if (!llkLowRam) {
+            android::base::WriteStringToFd("t", sysrqTriggerFd);
+        }
+        ::usleep(200000);  // let everything settle
+    }
+    llkWriteStringToFile(std::string("SysRq : Trigger a crash : 'livelock,") + state + "'\n",
+                         "/dev/kmsg");
+    android::base::WriteStringToFd("c", sysrqTriggerFd);
+    // NOTREACHED
+    // DYB
+    llkKillOneProcess(initPid, 'R', tid);
+    // I sat at my desk, stared into the garden and thought '42 will do'.
+    // I typed it out. End of story
+    ::exit(42);
+    // NOTREACHED
+}
+
+void llkAlarmHandler(int) {
+    llkPanicKernel(false, ::getpid(), "alarm");
+}
+
+milliseconds GetUintProperty(const std::string& key, milliseconds def) {
+    return milliseconds(android::base::GetUintProperty(key, static_cast<uint64_t>(def.count()),
+                                                       static_cast<uint64_t>(def.max().count())));
+}
+
+seconds GetUintProperty(const std::string& key, seconds def) {
+    return seconds(android::base::GetUintProperty(key, static_cast<uint64_t>(def.count()),
+                                                  static_cast<uint64_t>(def.max().count())));
+}
+
+proc* llkTidLookup(pid_t tid) {
+    auto search = tids.find(tid);
+    if (search == tids.end()) {
+        return nullptr;
+    }
+    return &search->second;
+}
+
+void llkTidRemove(pid_t tid) {
+    tids.erase(tid);
+}
+
+proc* llkTidAlloc(pid_t tid, pid_t pid, pid_t ppid, const char* comm, int time, char state) {
+    auto it = tids.emplace(std::make_pair(tid, proc(tid, pid, ppid, comm, time, state)));
+    return &it.first->second;
+}
+
+std::string llkFormat(milliseconds ms) {
+    auto sec = duration_cast<seconds>(ms);
+    std::ostringstream s;
+    s << sec.count() << '.';
+    auto f = s.fill('0');
+    auto w = s.width(3);
+    s << std::right << (ms - sec).count();
+    s.width(w);
+    s.fill(f);
+    s << 's';
+    return s.str();
+}
+
+std::string llkFormat(seconds s) {
+    return std::to_string(s.count()) + 's';
+}
+
+std::string llkFormat(bool flag) {
+    return flag ? "true" : "false";
+}
+
+std::string llkFormat(const std::unordered_set<std::string>& blacklist) {
+    std::string ret;
+    for (auto entry : blacklist) {
+        if (ret.size()) {
+            ret += ",";
+        }
+        ret += entry;
+    }
+    return ret;
+}
+
+// We only officially support comma separators, but wetware being what they
+// are will take some liberty and I do not believe they should be punished.
+std::unordered_set<std::string> llkSplit(const std::string& s,
+                                         const std::string& delimiters = ", \t:") {
+    std::unordered_set<std::string> result;
+
+    size_t base = 0;
+    size_t found;
+    while (true) {
+        found = s.find_first_of(delimiters, base);
+        result.emplace(s.substr(base, found - base));
+        if (found == s.npos) break;
+        base = found + 1;
+    }
+    return result;
+}
+
+bool llkSkipName(const std::string& name,
+                 const std::unordered_set<std::string>& blacklist = llkBlacklistProcess) {
+    if ((name.size() == 0) || (blacklist.size() == 0)) {
+        return false;
+    }
+
+    return blacklist.find(name) != blacklist.end();
+}
+
+bool llkSkipPid(pid_t pid) {
+    return llkSkipName(std::to_string(pid), llkBlacklistProcess);
+}
+
+bool llkSkipPpid(pid_t ppid) {
+    return llkSkipName(std::to_string(ppid), llkBlacklistParent);
+}
+
+bool llkSkipUid(uid_t uid) {
+    // Match by number?
+    if (llkSkipName(std::to_string(uid), llkBlacklistUid)) {
+        return true;
+    }
+
+    // Match by name?
+    auto pwd = ::getpwuid(uid);
+    return (pwd != nullptr) && __predict_true(pwd->pw_name != nullptr) &&
+           __predict_true(pwd->pw_name[0] != '\0') && llkSkipName(pwd->pw_name, llkBlacklistUid);
+}
+
+bool getValidTidDir(dirent* dp, std::string* piddir) {
+    if (!::isdigit(dp->d_name[0])) {
+        return false;
+    }
+
+    // Corner case can not happen in reality b/c of above ::isdigit check
+    if (__predict_false(dp->d_type != DT_DIR)) {
+        if (__predict_false(dp->d_type == DT_UNKNOWN)) {  // can't b/c procfs
+            struct stat st;
+            *piddir = procdir;
+            *piddir += dp->d_name;
+            return (lstat(piddir->c_str(), &st) == 0) && (st.st_mode & S_IFDIR);
+        }
+        return false;
+    }
+
+    *piddir = procdir;
+    *piddir += dp->d_name;
+    return true;
+}
+
+bool llkIsMonitorState(char state) {
+    return (state == 'Z') || (state == 'D');
+}
+
+// returns -1 if not found
+long long getSchedValue(const std::string& schedString, const char* key) {
+    auto pos = schedString.find(key);
+    if (pos == std::string::npos) {
+        return -1;
+    }
+    pos = schedString.find(':', pos);
+    if (__predict_false(pos == std::string::npos)) {
+        return -1;
+    }
+    while ((++pos < schedString.size()) && ::isblank(schedString[pos])) {
+        ;
+    }
+    long long ret;
+    if (!android::base::ParseInt(schedString.substr(pos), &ret, static_cast<long long>(0))) {
+        return -1;
+    }
+    return ret;
+}
+
+// Primary ABA mitigation watching last time schedule activity happened
+void llkCheckSchedUpdate(proc* procp, const std::string& piddir) {
+    // Audit finds /proc/<tid>/sched is just over 1K, and
+    // is rarely larger than 2K, even less on Android.
+    // For example, the "se.avg.lastUpdateTime" field we are
+    // interested in typically within the primary set in
+    // the first 1K.
+    //
+    // Proc entries can not be read >1K atomically via libbase,
+    // but if there are problems we assume at least a few
+    // samples of reads occur before we take any real action.
+    std::string schedString = ReadFile(piddir + "/sched");
+    if (schedString.size() == 0) {
+        // /schedstat is not as standardized, but in 3.1+
+        // Android devices, the third field is nr_switches
+        // from /sched:
+        schedString = ReadFile(piddir + "/schedstat");
+        if (schedString.size() == 0) {
+            return;
+        }
+        auto val = static_cast<unsigned long long>(-1);
+        if (((::sscanf(schedString.c_str(), "%*d %*d %llu", &val)) == 1) &&
+            (val != static_cast<unsigned long long>(-1)) && (val != 0) &&
+            (val != procp->nrSwitches)) {
+            procp->nrSwitches = val;
+            procp->count = 0ms;
+            procp->killed = !llkTestWithKill;
+        }
+        return;
+    }
+
+    auto val = getSchedValue(schedString, "\nse.avg.lastUpdateTime");
+    if (val == -1) {
+        val = getSchedValue(schedString, "\nse.svg.last_update_time");
+    }
+    if (val != -1) {
+        auto schedUpdate = nanoseconds(val);
+        if (schedUpdate != procp->schedUpdate) {
+            procp->schedUpdate = schedUpdate;
+            procp->count = 0ms;
+            procp->killed = !llkTestWithKill;
+        }
+    }
+
+    val = getSchedValue(schedString, "\nnr_switches");
+    if (val != -1) {
+        if (static_cast<uint64_t>(val) != procp->nrSwitches) {
+            procp->nrSwitches = val;
+            procp->count = 0ms;
+            procp->killed = !llkTestWithKill;
+        }
+    }
+}
+
+void llkLogConfig(void) {
+    LOG(INFO) << "ro.config.low_ram=" << llkFormat(llkLowRam) << "\n"
+              << LLK_ENABLE_PROPERTY "=" << llkFormat(llkEnable) << "\n"
+              << KHT_ENABLE_PROPERTY "=" << llkFormat(khtEnable) << "\n"
+              << LLK_MLOCKALL_PROPERTY "=" << llkFormat(llkMlockall) << "\n"
+              << LLK_KILLTEST_PROPERTY "=" << llkFormat(llkTestWithKill) << "\n"
+              << KHT_TIMEOUT_PROPERTY "=" << llkFormat(khtTimeout) << "\n"
+              << LLK_TIMEOUT_MS_PROPERTY "=" << llkFormat(llkTimeoutMs) << "\n"
+              << LLK_D_TIMEOUT_MS_PROPERTY "=" << llkFormat(llkStateTimeoutMs[llkStateD]) << "\n"
+              << LLK_Z_TIMEOUT_MS_PROPERTY "=" << llkFormat(llkStateTimeoutMs[llkStateZ]) << "\n"
+              << LLK_CHECK_MS_PROPERTY "=" << llkFormat(llkCheckMs) << "\n"
+              << LLK_BLACKLIST_PROCESS_PROPERTY "=" << llkFormat(llkBlacklistProcess) << "\n"
+              << LLK_BLACKLIST_PARENT_PROPERTY "=" << llkFormat(llkBlacklistParent) << "\n"
+              << LLK_BLACKLIST_UID_PROPERTY "=" << llkFormat(llkBlacklistUid);
+}
+
+void* llkThread(void* obj) {
+    LOG(INFO) << "started";
+
+    std::string name = std::to_string(::gettid());
+    if (!llkSkipName(name)) {
+        llkBlacklistProcess.emplace(name);
+    }
+    name = static_cast<const char*>(obj);
+    prctl(PR_SET_NAME, name.c_str());
+    if (__predict_false(!llkSkipName(name))) {
+        llkBlacklistProcess.insert(name);
+    }
+    // No longer modifying llkBlacklistProcess.
+    llkRunning = true;
+    llkLogConfig();
+    while (llkRunning) {
+        ::usleep(duration_cast<microseconds>(llkCheck(true)).count());
+    }
+    // NOTREACHED
+    LOG(INFO) << "exiting";
+    return nullptr;
+}
+
+}  // namespace
+
+milliseconds llkCheck(bool checkRunning) {
+    if (!llkEnable || (checkRunning != llkRunning)) {
+        return milliseconds::max();
+    }
+
+    // Reset internal watchdog, which is a healthy engineering margin of
+    // double the maximum wait or cycle time for the mainloop that calls us.
+    //
+    // This alarm is effectively the live lock detection of llkd, as
+    // we understandably can not monitor ourselves otherwise.
+    ::alarm(duration_cast<seconds>(llkTimeoutMs * 2).count());
+
+    // kernel jiffy precision fastest acquisition
+    static timespec last;
+    timespec now;
+    ::clock_gettime(CLOCK_MONOTONIC_COARSE, &now);
+    auto ms = llkGetTimespecDiffMs(&last, &now);
+    if (ms < llkCycle) {
+        return llkCycle - ms;
+    }
+    last = now;
+
+    LOG(VERBOSE) << "opendir(\"" << procdir << "\")";
+    if (__predict_false(!llkTopDirectory)) {
+        // gid containing AID_READPROC required
+        llkTopDirectory.reset(procdir);
+        if (__predict_false(!llkTopDirectory)) {
+            // Most likely reason we could be here is a resource limit.
+            // Keep our processing down to a minimum, but not so low that
+            // we do not recover in a timely manner should the issue be
+            // transitory.
+            LOG(DEBUG) << "opendir(\"" << procdir << "\") failed";
+            return llkTimeoutMs;
+        }
+    }
+
+    for (auto& it : tids) {
+        it.second.updated = false;
+    }
+
+    auto prevUpdate = llkUpdate;
+    llkUpdate += ms;
+    ms -= llkCycle;
+    auto myPid = ::getpid();
+    auto myTid = ::gettid();
+    for (auto dp = llkTopDirectory.read(); dp != nullptr; dp = llkTopDirectory.read()) {
+        std::string piddir;
+
+        if (!getValidTidDir(dp, &piddir)) {
+            continue;
+        }
+
+        // Get the process tasks
+        std::string taskdir = piddir + "/task/";
+        int pid = -1;
+        LOG(VERBOSE) << "+opendir(\"" << taskdir << "\")";
+        dir taskDirectory(taskdir);
+        if (__predict_false(!taskDirectory)) {
+            LOG(DEBUG) << "+opendir(\"" << taskdir << "\") failed";
+        }
+        for (auto tp = taskDirectory.read(dir::task, dp); tp != nullptr;
+             tp = taskDirectory.read(dir::task)) {
+            if (!getValidTidDir(tp, &piddir)) {
+                continue;
+            }
+
+            // Get the process stat
+            std::string stat = ReadFile(piddir + "/stat");
+            if (stat.size() == 0) {
+                continue;
+            }
+            unsigned tid = -1;
+            char pdir[TASK_COMM_LEN + 1];
+            char state = '?';
+            unsigned ppid = -1;
+            unsigned utime = -1;
+            unsigned stime = -1;
+            int dummy;
+            pdir[0] = '\0';
+            // tid should not change value
+            auto match = ::sscanf(
+                stat.c_str(),
+                "%u (%" ___STRING(
+                    TASK_COMM_LEN) "[^)]) %c %u %*d %*d %*d %*d %*d %*d %*d %*d %*d %u %u %d",
+                &tid, pdir, &state, &ppid, &utime, &stime, &dummy);
+            if (pid == -1) {
+                pid = tid;
+            }
+            LOG(VERBOSE) << "match " << match << ' ' << tid << " (" << pdir << ") " << state << ' '
+                         << ppid << " ... " << utime << ' ' << stime << ' ' << dummy;
+            if (match != 7) {
+                continue;
+            }
+
+            auto procp = llkTidLookup(tid);
+            if (procp == nullptr) {
+                procp = llkTidAlloc(tid, pid, ppid, pdir, utime + stime, state);
+            } else {
+                // comm can change ...
+                procp->setComm(pdir);
+                procp->updated = true;
+                // pid/ppid/tid wrap?
+                if (((procp->update != prevUpdate) && (procp->update != llkUpdate)) ||
+                    (procp->ppid != ppid) || (procp->pid != pid)) {
+                    procp->reset();
+                } else if (procp->time != (utime + stime)) {  // secondary ABA.
+                    // watching utime+stime granularity jiffy
+                    procp->state = '?';
+                }
+                procp->update = llkUpdate;
+                procp->pid = pid;
+                procp->ppid = ppid;
+                procp->time = utime + stime;
+                if (procp->state != state) {
+                    procp->count = 0ms;
+                    procp->killed = !llkTestWithKill;
+                    procp->state = state;
+                } else {
+                    procp->count += llkCycle;
+                }
+            }
+
+            // Filter checks in intuitive order of CPU cost to evaluate
+            // If tid unique continue, if ppid or pid unique break
+
+            if (pid == myPid) {
+                break;
+            }
+            if (!llkIsMonitorState(state)) {
+                continue;
+            }
+            if ((tid == myTid) || llkSkipPid(tid)) {
+                continue;
+            }
+            if (llkSkipPpid(ppid)) {
+                break;
+            }
+
+            if (llkSkipName(procp->getComm())) {
+                continue;
+            }
+            if (llkSkipName(procp->getCmdline())) {
+                break;
+            }
+
+            auto pprocp = llkTidLookup(ppid);
+            if (pprocp == nullptr) {
+                pprocp = llkTidAlloc(ppid, ppid, 0, "", 0, '?');
+            }
+            if ((pprocp != nullptr) && (llkSkipName(pprocp->getComm(), llkBlacklistParent) ||
+                                        llkSkipName(pprocp->getCmdline(), llkBlacklistParent))) {
+                break;
+            }
+
+            if ((llkBlacklistUid.size() != 0) && llkSkipUid(procp->getUid())) {
+                continue;
+            }
+
+            // ABA mitigation watching last time schedule activity happened
+            llkCheckSchedUpdate(procp, piddir);
+
+            // Can only fall through to here if registered D or Z state !!!
+            if (procp->count < llkStateTimeoutMs[(state == 'Z') ? llkStateZ : llkStateD]) {
+                LOG(VERBOSE) << state << ' ' << llkFormat(procp->count) << ' ' << ppid << "->"
+                             << pid << "->" << tid << ' ' << procp->getComm();
+                continue;
+            }
+
+            // We have to kill it to determine difference between live lock
+            // and persistent state blocked on a resource.  Is there something
+            // wrong with a process that has no forward scheduling progress in
+            // Z or D?  Yes, generally means improper accounting in the
+            // process, but not always ...
+            //
+            // Whomever we hit with a test kill must accept the Android
+            // Aphorism that everything can be burned to the ground and
+            // must survive.
+            if (procp->killed == false) {
+                procp->killed = true;
+                // confirm: re-read uid before committing to a panic.
+                procp->uid = -1;
+                switch (state) {
+                    case 'Z':  // kill ppid to free up a Zombie
+                        // Killing init will kernel panic without diagnostics
+                        // so skip right to controlled kernel panic with
+                        // diagnostics.
+                        if (ppid == initPid) {
+                            break;
+                        }
+                        LOG(WARNING) << "Z " << llkFormat(procp->count) << ' ' << ppid << "->"
+                                     << pid << "->" << tid << ' ' << procp->getComm() << " [kill]";
+                        if ((llkKillOneProcess(pprocp, procp) >= 0) ||
+                            (llkKillOneProcess(ppid, procp) >= 0)) {
+                            continue;
+                        }
+                        break;
+
+                    case 'D':  // kill tid to free up an uninterruptible D
+                        // If ABA is doing its job, we would not need or
+                        // want the following.  Test kill is a Hail Mary
+                        // to make absolutely sure there is no forward
+                        // scheduling progress.  The cost when ABA is
+                        // not working is we kill a process that likes to
+                        // stay in 'D' state, instead of panicing the
+                        // kernel (worse).
+                        LOG(WARNING) << "D " << llkFormat(procp->count) << ' ' << pid << "->" << tid
+                                     << ' ' << procp->getComm() << " [kill]";
+                        if ((llkKillOneProcess(llkTidLookup(pid), procp) >= 0) ||
+                            (llkKillOneProcess(pid, 'D', tid) >= 0) ||
+                            (llkKillOneProcess(procp, procp) >= 0) ||
+                            (llkKillOneProcess(tid, 'D', tid) >= 0)) {
+                            continue;
+                        }
+                        break;
+                }
+            }
+            // We are here because we have confirmed kernel live-lock
+            LOG(ERROR) << state << ' ' << llkFormat(procp->count) << ' ' << ppid << "->" << pid
+                       << "->" << tid << ' ' << procp->getComm() << " [panic]";
+            llkPanicKernel(true, tid, (state == 'Z') ? "zombie" : "driver");
+        }
+        LOG(VERBOSE) << "+closedir()";
+    }
+    llkTopDirectory.rewind();
+    LOG(VERBOSE) << "closedir()";
+
+    // garbage collection of old process references
+    for (auto p = tids.begin(); p != tids.end();) {
+        if (!p->second.updated) {
+            IF_ALOG(LOG_VERBOSE, LOG_TAG) {
+                std::string ppidCmdline = llkProcGetName(p->second.ppid, nullptr, nullptr);
+                if (ppidCmdline.size()) {
+                    ppidCmdline = "(" + ppidCmdline + ")";
+                }
+                std::string pidCmdline;
+                if (p->second.pid != p->second.tid) {
+                    pidCmdline = llkProcGetName(p->second.pid, nullptr, p->second.getCmdline());
+                    if (pidCmdline.size()) {
+                        pidCmdline = "(" + pidCmdline + ")";
+                    }
+                }
+                std::string tidCmdline =
+                    llkProcGetName(p->second.tid, p->second.getComm(), p->second.getCmdline());
+                if (tidCmdline.size()) {
+                    tidCmdline = "(" + tidCmdline + ")";
+                }
+                LOG(VERBOSE) << "thread " << p->second.ppid << ppidCmdline << "->" << p->second.pid
+                             << pidCmdline << "->" << p->second.tid << tidCmdline << " removed";
+            }
+            p = tids.erase(p);
+        } else {
+            ++p;
+        }
+    }
+    if (__predict_false(tids.empty())) {
+        llkTopDirectory.reset();
+    }
+
+    llkCycle = llkCheckMs;
+
+    timespec end;
+    ::clock_gettime(CLOCK_MONOTONIC_COARSE, &end);
+    auto milli = llkGetTimespecDiffMs(&now, &end);
+    LOG((milli > 10s) ? ERROR : (milli > 1s) ? WARNING : VERBOSE) << "sample " << llkFormat(milli);
+
+    // cap to minimum sleep for 1 second since last cycle
+    if (llkCycle < (ms + 1s)) {
+        return 1s;
+    }
+    return llkCycle - ms;
+}
+
+unsigned llkCheckMilliseconds() {
+    return duration_cast<milliseconds>(llkCheck()).count();
+}
+
+bool llkInit(const char* threadname) {
+    llkLowRam = android::base::GetBoolProperty("ro.config.low_ram", false);
+    if (!LLK_ENABLE_DEFAULT && android::base::GetBoolProperty("ro.debuggable", false)) {
+        llkEnable = android::base::GetProperty(LLK_ENABLE_PROPERTY, "eng") == "eng";
+        khtEnable = android::base::GetProperty(KHT_ENABLE_PROPERTY, "eng") == "eng";
+    }
+    llkEnable = android::base::GetBoolProperty(LLK_ENABLE_PROPERTY, llkEnable);
+    if (llkEnable && !llkTopDirectory.reset(procdir)) {
+        // Most likely reason we could be here is llkd was started
+        // incorrectly without the readproc permissions.  Keep our
+        // processing down to a minimum.
+        llkEnable = false;
+    }
+    khtEnable = android::base::GetBoolProperty(KHT_ENABLE_PROPERTY, khtEnable);
+    llkMlockall = android::base::GetBoolProperty(LLK_MLOCKALL_PROPERTY, llkMlockall);
+    llkTestWithKill = android::base::GetBoolProperty(LLK_KILLTEST_PROPERTY, llkTestWithKill);
+    // if LLK_TIMOUT_MS_PROPERTY was not set, we will use a set
+    // KHT_TIMEOUT_PROPERTY as co-operative guidance for the default value.
+    khtTimeout = GetUintProperty(KHT_TIMEOUT_PROPERTY, khtTimeout);
+    if (khtTimeout == 0s) {
+        khtTimeout = duration_cast<seconds>(llkTimeoutMs * (1 + LLK_CHECKS_PER_TIMEOUT_DEFAULT) /
+                                            LLK_CHECKS_PER_TIMEOUT_DEFAULT);
+    }
+    llkTimeoutMs =
+        khtTimeout * LLK_CHECKS_PER_TIMEOUT_DEFAULT / (1 + LLK_CHECKS_PER_TIMEOUT_DEFAULT);
+    llkTimeoutMs = GetUintProperty(LLK_TIMEOUT_MS_PROPERTY, llkTimeoutMs);
+    llkValidate();  // validate llkTimeoutMs, llkCheckMs and llkCycle
+    llkStateTimeoutMs[llkStateD] = GetUintProperty(LLK_D_TIMEOUT_MS_PROPERTY, llkTimeoutMs);
+    llkStateTimeoutMs[llkStateZ] = GetUintProperty(LLK_Z_TIMEOUT_MS_PROPERTY, llkTimeoutMs);
+    llkCheckMs = GetUintProperty(LLK_CHECK_MS_PROPERTY, llkCheckMs);
+    llkValidate();  // validate all (effectively minus llkTimeoutMs)
+    std::string defaultBlacklistProcess(
+        std::to_string(kernelPid) + "," + std::to_string(initPid) + "," +
+        std::to_string(kthreaddPid) + "," + std::to_string(::getpid()) + "," +
+        std::to_string(::gettid()) + "," LLK_BLACKLIST_PROCESS_DEFAULT);
+    if (threadname) {
+        defaultBlacklistProcess += std::string(",") + threadname;
+    }
+    for (int cpu = 1; cpu < get_nprocs_conf(); ++cpu) {
+        defaultBlacklistProcess += ",[watchdog/" + std::to_string(cpu) + "]";
+    }
+    defaultBlacklistProcess =
+        android::base::GetProperty(LLK_BLACKLIST_PROCESS_PROPERTY, defaultBlacklistProcess);
+    llkBlacklistProcess = llkSplit(defaultBlacklistProcess);
+    if (!llkSkipName("[khungtaskd]")) {  // ALWAYS ignore as special
+        llkBlacklistProcess.emplace("[khungtaskd]");
+    }
+    llkBlacklistParent = llkSplit(android::base::GetProperty(
+        LLK_BLACKLIST_PARENT_PROPERTY, std::to_string(kernelPid) + "," + std::to_string(kthreaddPid) +
+                                           "," LLK_BLACKLIST_PARENT_DEFAULT));
+    llkBlacklistUid =
+        llkSplit(android::base::GetProperty(LLK_BLACKLIST_UID_PROPERTY, LLK_BLACKLIST_UID_DEFAULT));
+
+    // internal watchdog
+    ::signal(SIGALRM, llkAlarmHandler);
+
+    // kernel hung task configuration? Otherwise leave it as-is
+    if (khtEnable) {
+        // EUID must be AID_ROOT to write to /proc/sys/kernel/ nodes, there
+        // are no capability overrides.  For security reasons we do not want
+        // to run as AID_ROOT.  We may not be able to write them successfully,
+        // we will try, but the least we can do is read the values back to
+        // confirm expectations and report whether configured or not.
+        auto configured = llkWriteStringToFileConfirm(std::to_string(khtTimeout.count()),
+                                                      "/proc/sys/kernel/hung_task_timeout_secs");
+        if (configured) {
+            llkWriteStringToFile("65535", "/proc/sys/kernel/hung_task_warnings");
+            llkWriteStringToFile("65535", "/proc/sys/kernel/hung_task_check_count");
+            configured = llkWriteStringToFileConfirm("1", "/proc/sys/kernel/hung_task_panic");
+        }
+        if (configured) {
+            LOG(INFO) << "[khungtaskd] configured";
+        } else {
+            LOG(WARNING) << "[khungtaskd] not configurable";
+        }
+    }
+
+    bool logConfig = true;
+    if (llkEnable) {
+        if (llkMlockall &&
+            // MCL_ONFAULT pins pages as they fault instead of loading
+            // everything immediately all at once. (Which would be bad,
+            // because as of this writing, we have a lot of mapped pages we
+            // never use.) Old kernels will see MCL_ONFAULT and fail with
+            // EINVAL; we ignore this failure.
+            //
+            // N.B. read the man page for mlockall. MCL_CURRENT | MCL_ONFAULT
+            // pins ⊆ MCL_CURRENT, converging to just MCL_CURRENT as we fault
+            // in pages.
+
+            // CAP_IPC_LOCK required
+            mlockall(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT) && (errno != EINVAL)) {
+            PLOG(WARNING) << "mlockall failed ";
+        }
+
+        if (threadname) {
+            pthread_attr_t attr;
+
+            if (!pthread_attr_init(&attr)) {
+                sched_param param;
+
+                memset(&param, 0, sizeof(param));
+                pthread_attr_setschedparam(&attr, &param);
+                pthread_attr_setschedpolicy(&attr, SCHED_BATCH);
+                if (!pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED)) {
+                    pthread_t thread;
+                    if (!pthread_create(&thread, &attr, llkThread, const_cast<char*>(threadname))) {
+                        // wait a second for thread to start
+                        for (auto retry = 50; retry && !llkRunning; --retry) {
+                            ::usleep(20000);
+                        }
+                        logConfig = !llkRunning;  // printed in llkd context?
+                    } else {
+                        LOG(ERROR) << "failed to spawn llkd thread";
+                    }
+                } else {
+                    LOG(ERROR) << "failed to detach llkd thread";
+                }
+                pthread_attr_destroy(&attr);
+            } else {
+                LOG(ERROR) << "failed to allocate attibutes for llkd thread";
+            }
+        }
+    } else {
+        LOG(DEBUG) << "[khungtaskd] left unconfigured";
+    }
+    if (logConfig) {
+        llkLogConfig();
+    }
+
+    return llkEnable;
+}
diff --git a/llkd/llkd.cpp b/llkd/llkd.cpp
new file mode 100644
index 0000000..f10253d
--- /dev/null
+++ b/llkd/llkd.cpp
@@ -0,0 +1,50 @@
+/*
+ * Copyright (C) 2018 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "llkd.h"
+
+#include <sched.h>
+#include <unistd.h>
+
+#include <chrono>
+
+#include <android-base/logging.h>
+
+using namespace std::chrono;
+
+int main(int, char**) {
+    LOG(INFO) << "started";
+
+    bool enabled = llkInit();
+
+    // Would like this policy to be automatic as part of libllkd,
+    // but that would be presumptuous and bad side-effect.
+    struct sched_param param;
+    memset(&param, 0, sizeof(param));
+    sched_setscheduler(0, SCHED_BATCH, &param);
+
+    while (true) {
+        if (enabled) {
+            ::usleep(duration_cast<microseconds>(llkCheck()).count());
+        } else {
+            ::pause();
+        }
+    }
+    // NOTREACHED
+
+    LOG(INFO) << "exiting";
+    return 0;
+}
diff --git a/llkd/llkd.rc b/llkd/llkd.rc
new file mode 100644
index 0000000..e538cdb
--- /dev/null
+++ b/llkd/llkd.rc
@@ -0,0 +1,49 @@
+# eng default for ro.llk.enable and ro.khungtask.enable
+on property:ro.debuggable=*
+    setprop llk.enable ${ro.llk.enable:-0}
+    setprop khungtask.enable ${ro.khungtask.enable:-0}
+
+on property:ro.debuggable=1
+    setprop llk.enable ${ro.llk.enable:-1}
+    setprop khungtask.enable ${ro.khungtask.enable:-1}
+
+on property:ro.llk.enable=eng
+    setprop llk.enable ${ro.debuggable:-0}
+
+on property:ro.khungtask.enable=eng
+    setprop khungtask.enable ${ro.debuggable:-0}
+
+on property:llk.enable=1
+    setprop llk.enable true
+
+on property:llk.enable=0
+    setprop llk.enable false
+
+on property:khungtask.enable=1
+    setprop khungtask.enable true
+
+on property:khungtask.enable=0
+    setprop khungtask.enable false
+
+# Configure [khungtaskd]
+on property:khungtask.enable=true
+    write /proc/sys/kernel/hung_task_timeout_secs ${ro.khungtask.timeout:-720}
+    write /proc/sys/kernel/hung_task_warnings 65535
+    write /proc/sys/kernel/hung_task_check_count 65535
+    write /proc/sys/kernel/hung_task_panic 1
+
+on property:khungtask.enable=false
+    write /proc/sys/kernel/hung_task_panic 0
+
+on property:llk.enable=true
+    start llkd
+
+service llkd /system/bin/llkd
+    class late_start
+    disabled
+    user llkd
+    group llkd readproc
+    capabilities KILL IPC_LOCK
+    file /dev/kmsg w
+    file /proc/sysrq-trigger w
+    writepid /dev/cpuset/system-background/tasks
diff --git a/llkd/tests/Android.bp b/llkd/tests/Android.bp
new file mode 100644
index 0000000..6dd5938
--- /dev/null
+++ b/llkd/tests/Android.bp
@@ -0,0 +1,41 @@
+// Copyright (C) 2018 The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+cc_test {
+    name: "llkd_unit_test",
+
+    shared_libs: [
+        "libbase",
+        "liblog",
+    ],
+    header_libs: [
+        "llkd_headers",
+    ],
+
+    target: {
+        android: {
+            srcs: [
+                "llkd_test.cpp",
+            ],
+        },
+    },
+
+    cflags: [
+        "-Wall",
+        "-Wextra",
+        "-Werror",
+    ],
+
+    compile_multilib: "first",
+}
diff --git a/llkd/tests/llkd_test.cpp b/llkd/tests/llkd_test.cpp
new file mode 100644
index 0000000..3a15ff1
--- /dev/null
+++ b/llkd/tests/llkd_test.cpp
@@ -0,0 +1,287 @@
+/*
+ * Copyright (C) 2018 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <signal.h>
+#include <stdint.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+
+#include <chrono>
+#include <iostream>
+#include <string>
+
+#include <android-base/properties.h>
+#include <gtest/gtest.h>
+#include <log/log_time.h>  // for MS_PER_SEC and US_PER_SEC
+
+#include "llkd.h"
+
+using namespace std::chrono;
+using namespace std::chrono_literals;
+
+namespace {
+
+milliseconds GetUintProperty(const std::string& key, milliseconds def) {
+    return milliseconds(android::base::GetUintProperty(key, static_cast<uint64_t>(def.count()),
+                                                       static_cast<uint64_t>(def.max().count())));
+}
+
+seconds GetUintProperty(const std::string& key, seconds def) {
+    return seconds(android::base::GetUintProperty(key, static_cast<uint64_t>(def.count()),
+                                                  static_cast<uint64_t>(def.max().count())));
+}
+
+// GTEST_LOG_(WARNING) output is fugly, this has much less noise
+// ToDo: look into fixing googletest to produce output that matches style of
+//       all the other status messages, and can switch off __line__ and
+//       __function__ noise
+#define GTEST_LOG_WARNING std::cerr << "[ WARNING  ] "
+#define GTEST_LOG_INFO std::cerr << "[   INFO   ] "
+
+// Properties is _not_ a high performance ABI!
+void rest() {
+    usleep(200000);
+}
+
+void execute(const char* command) {
+    if (getuid() || system(command)) {
+        system((std::string("su root ") + command).c_str());
+    }
+}
+
+seconds llkdSleepPeriod(char state) {
+    auto default_eng = android::base::GetProperty(LLK_ENABLE_PROPERTY, "eng") == "eng";
+    auto default_enable = LLK_ENABLE_DEFAULT;
+    if (!LLK_ENABLE_DEFAULT && default_eng &&
+        android::base::GetBoolProperty("ro.debuggable", false)) {
+        default_enable = true;
+    }
+    default_enable = android::base::GetBoolProperty(LLK_ENABLE_PROPERTY, default_enable);
+    if (default_eng) {
+        GTEST_LOG_INFO << LLK_ENABLE_PROPERTY " defaults to \"eng\" thus "
+                       << (default_enable ? "true" : "false") << "\n";
+    }
+    // Hail Mary hope is unconfigured.
+    if ((GetUintProperty(LLK_TIMEOUT_MS_PROPERTY, LLK_TIMEOUT_MS_DEFAULT) !=
+         duration_cast<milliseconds>(120s)) ||
+        (GetUintProperty(LLK_CHECK_MS_PROPERTY,
+                         LLK_TIMEOUT_MS_DEFAULT / LLK_CHECKS_PER_TIMEOUT_DEFAULT) !=
+         duration_cast<milliseconds>(10s))) {
+        execute("stop llkd");
+        rest();
+        std::string setprop("setprop ");
+        execute((setprop + LLK_ENABLE_WRITEABLE_PROPERTY + " false").c_str());
+        rest();
+        execute((setprop + LLK_TIMEOUT_MS_PROPERTY + " 120000").c_str());
+        rest();
+        execute((setprop + KHT_TIMEOUT_PROPERTY + " 130").c_str());
+        rest();
+        execute((setprop + LLK_CHECK_MS_PROPERTY + " 10000").c_str());
+        rest();
+        execute((setprop + LLK_ENABLE_PROPERTY + " true").c_str());
+        rest();
+        execute((setprop + LLK_ENABLE_WRITEABLE_PROPERTY + " true").c_str());
+        rest();
+    }
+    default_enable = LLK_ENABLE_DEFAULT;
+    if (!LLK_ENABLE_DEFAULT && (android::base::GetProperty(LLK_ENABLE_PROPERTY, "eng") == "eng") &&
+        android::base::GetBoolProperty("ro.debuggable", false)) {
+        default_enable = true;
+    }
+    default_enable = android::base::GetBoolProperty(LLK_ENABLE_PROPERTY, default_enable);
+    if (default_enable) {
+        execute("start llkd");
+        rest();
+        GTEST_LOG_INFO << "llkd enabled\n";
+    } else {
+        GTEST_LOG_WARNING << "llkd disabled\n";
+    }
+
+    /* KISS follows llk_init() */
+    milliseconds llkTimeoutMs = LLK_TIMEOUT_MS_DEFAULT;
+    seconds khtTimeout = duration_cast<seconds>(
+        llkTimeoutMs * (1 + LLK_CHECKS_PER_TIMEOUT_DEFAULT) / LLK_CHECKS_PER_TIMEOUT_DEFAULT);
+    khtTimeout = GetUintProperty(KHT_TIMEOUT_PROPERTY, khtTimeout);
+    llkTimeoutMs =
+        khtTimeout * LLK_CHECKS_PER_TIMEOUT_DEFAULT / (1 + LLK_CHECKS_PER_TIMEOUT_DEFAULT);
+    llkTimeoutMs = GetUintProperty(LLK_TIMEOUT_MS_PROPERTY, llkTimeoutMs);
+    if (llkTimeoutMs < LLK_TIMEOUT_MS_MINIMUM) {
+        llkTimeoutMs = LLK_TIMEOUT_MS_MINIMUM;
+    }
+    milliseconds llkCheckMs = llkTimeoutMs / LLK_CHECKS_PER_TIMEOUT_DEFAULT;
+    auto timeout = GetUintProperty(
+        (state == 'Z') ? LLK_Z_TIMEOUT_MS_PROPERTY : LLK_D_TIMEOUT_MS_PROPERTY, llkTimeoutMs);
+    if (timeout < LLK_TIMEOUT_MS_MINIMUM) {
+        timeout = LLK_TIMEOUT_MS_MINIMUM;
+    }
+
+    if (llkCheckMs > timeout) {
+        llkCheckMs = timeout;
+    }
+    llkCheckMs = GetUintProperty(LLK_CHECK_MS_PROPERTY, llkCheckMs);
+    timeout += llkCheckMs;
+    auto sec = duration_cast<seconds>(timeout);
+    if (sec == 0s) {
+        ++sec;
+    } else if (sec > 59s) {
+        GTEST_LOG_WARNING << "llkd is configured for about " << duration_cast<minutes>(sec).count()
+                          << " minutes to react\n";
+    }
+
+    // 33% margin for the test to naturally timeout waiting for llkd to respond
+    return (sec * 4 + 2s) / 3;
+}
+
+inline void waitForPid(pid_t child_pid) {
+    int wstatus;
+    ASSERT_LE(0, waitpid(child_pid, &wstatus, 0));
+    EXPECT_FALSE(WIFEXITED(wstatus)) << "[   INFO   ] exit=" << WEXITSTATUS(wstatus);
+    ASSERT_TRUE(WIFSIGNALED(wstatus));
+    ASSERT_EQ(WTERMSIG(wstatus), SIGKILL);
+}
+
+bool checkKill(const char* reason) {
+    if (android::base::GetBoolProperty(LLK_KILLTEST_PROPERTY, LLK_KILLTEST_DEFAULT)) {
+        return false;
+    }
+    auto bootreason = android::base::GetProperty("sys.boot.reason", "nothing");
+    if (bootreason == reason) {
+        GTEST_LOG_INFO << "Expected test result confirmed " << reason << "\n";
+        return true;
+    }
+    GTEST_LOG_WARNING << "Expected test result is " << reason << "\n";
+
+    // apct adjustment if needed (set LLK_KILLTEST_PROPERTY to "off" to allow test)
+    //
+    // if (android::base::GetProperty(LLK_KILLTEST_PROPERTY, "") == "false") {
+    //     GTEST_LOG_WARNING << "Bypassing test\n";
+    //     return true;
+    // }
+
+    return false;
+}
+
+}  // namespace
+
+// The tests that use this helper are to simulate processes stuck in 'D'
+// state that are experiencing forward scheduled progress. As such the
+// expectation is that llkd will _not_ perform any mitigations. The sleepfor
+// argument helps us set the amount of forward scheduler progress.
+static void llkd_driver_ABA(const microseconds sleepfor) {
+    const auto period = llkdSleepPeriod('D');
+    if (period <= sleepfor) {
+        GTEST_LOG_WARNING << "llkd configuration too short for "
+                          << duration_cast<milliseconds>(sleepfor).count() << "ms work cycle\n";
+        return;
+    }
+
+    auto child_pid = fork();
+    ASSERT_LE(0, child_pid);
+    int wstatus;
+    if (!child_pid) {
+        auto ratio = period / sleepfor;
+        ASSERT_LT(0, ratio);
+        // vfork() parent is uninterruptable D state waiting for child to exec()
+        while (--ratio > 0) {
+            auto driver_pid = vfork();
+            ASSERT_LE(0, driver_pid);
+            if (driver_pid) {  // parent
+                waitpid(driver_pid, &wstatus, 0);
+                if (!WIFEXITED(wstatus)) {
+                    exit(42);
+                }
+                if (WEXITSTATUS(wstatus) != 42) {
+                    exit(42);
+                }
+            } else {
+                usleep(sleepfor.count());
+                exit(42);
+            }
+        }
+        exit(0);
+    }
+    ASSERT_LE(0, waitpid(child_pid, &wstatus, 0));
+    EXPECT_TRUE(WIFEXITED(wstatus));
+    if (WIFEXITED(wstatus)) {
+        EXPECT_EQ(0, WEXITSTATUS(wstatus));
+    }
+    ASSERT_FALSE(WIFSIGNALED(wstatus)) << "[   INFO   ] signo=" << WTERMSIG(wstatus);
+}
+
+TEST(llkd, driver_ABA_fast) {
+    llkd_driver_ABA(5ms);
+}
+
+TEST(llkd, driver_ABA_slow) {
+    llkd_driver_ABA(1s);
+}
+
+TEST(llkd, driver_ABA_glacial) {
+    llkd_driver_ABA(1min);
+}
+
+// Following tests must be last in this file to capture possible errant
+// kernel_panic mitigation failure.
+
+// The following tests simulate processes stick in 'Z' or 'D' state with
+// no forward scheduling progress, but interruptible. As such the expectation
+// is that llkd will perform kill mitigation and not progress to kernel_panic.
+
+TEST(llkd, zombie) {
+    if (checkKill("kernel_panic,sysrq,livelock,zombie")) {
+        return;
+    }
+
+    const auto period = llkdSleepPeriod('Z');
+
+    /* Create a Persistent Zombie Process */
+    pid_t child_pid = fork();
+    ASSERT_LE(0, child_pid);
+    if (!child_pid) {
+        auto zombie_pid = fork();
+        ASSERT_LE(0, zombie_pid);
+        if (!zombie_pid) {
+            sleep(1);
+            exit(0);
+        }
+        sleep(period.count());
+        exit(42);
+    }
+
+    waitForPid(child_pid);
+}
+
+TEST(llkd, driver) {
+    if (checkKill("kernel_panic,sysrq,livelock,driver")) {
+        return;
+    }
+
+    const auto period = llkdSleepPeriod('D');
+
+    /* Create a Persistent Device Process */
+    auto child_pid = fork();
+    ASSERT_LE(0, child_pid);
+    if (!child_pid) {
+        // vfork() parent is uninterruptable D state waiting for child to exec()
+        auto driver_pid = vfork();
+        ASSERT_LE(0, driver_pid);
+        sleep(period.count());
+        exit(driver_pid ? 42 : 0);
+    }
+
+    waitForPid(child_pid);
+}
diff --git a/lmkd/lmkd.c b/lmkd/lmkd.c
index 151e1dc..80711bc 100644
--- a/lmkd/lmkd.c
+++ b/lmkd/lmkd.c
@@ -68,6 +68,7 @@
 #define MEMINFO_PATH "/proc/meminfo"
 #define LINE_MAX 128
 
+/* gid containing AID_SYSTEM required */
 #define INKERNEL_MINFREE_PATH "/sys/module/lowmemorykiller/parameters/minfree"
 #define INKERNEL_ADJ_PATH "/sys/module/lowmemorykiller/parameters/adj"
 
@@ -455,6 +456,9 @@
         return;
     }
 
+    /* gid containing AID_READPROC required */
+    /* CAP_SYS_RESOURCE required */
+    /* CAP_DAC_OVERRIDE required */
     snprintf(path, sizeof(path), "/proc/%d/oom_score_adj", params.pid);
     snprintf(val, sizeof(val), "%d", params.oomadj);
     if (!writefilestring(path, val, false)) {
@@ -496,8 +500,7 @@
         soft_limit_mult = 64;
     }
 
-    snprintf(path, sizeof(path),
-             "/dev/memcg/apps/uid_%d/pid_%d/memory.soft_limit_in_bytes",
+    snprintf(path, sizeof(path), MEMCG_SYSFS_PATH "apps/uid_%d/pid_%d/memory.soft_limit_in_bytes",
              params.uid, params.pid);
     snprintf(val, sizeof(val), "%d", soft_limit_mult * EIGHT_MEGA);
 
@@ -859,6 +862,7 @@
     int total;
     ssize_t ret;
 
+    /* gid containing AID_READPROC required */
     snprintf(path, PATH_MAX, "/proc/%d/statm", pid);
     fd = open(path, O_RDONLY | O_CLOEXEC);
     if (fd == -1)
@@ -882,6 +886,7 @@
     char *cp;
     ssize_t ret;
 
+    /* gid containing AID_READPROC required */
     snprintf(path, PATH_MAX, "/proc/%d/cmdline", pid);
     fd = open(path, O_RDONLY | O_CLOEXEC);
     if (fd == -1)
@@ -949,6 +954,7 @@
 
     TRACE_KILL_START(pid);
 
+    /* CAP_KILL required */
     r = kill(pid, SIGKILL);
     ALOGI(
         "Killing '%s' (%d), uid %d, adj %d\n"
@@ -1267,6 +1273,7 @@
     int level_idx = (int)level;
     const char *levelstr = level_name[level_idx];
 
+    /* gid containing AID_SYSTEM required */
     mpfd = open(MEMCG_SYSFS_PATH "memory.pressure_level", O_RDONLY | O_CLOEXEC);
     if (mpfd < 0) {
         ALOGI("No kernel memory.pressure_level support (errno=%d)", errno);
@@ -1478,11 +1485,15 @@
              * pins ⊆ MCL_CURRENT, converging to just MCL_CURRENT as we fault
              * in pages.
              */
+            /* CAP_IPC_LOCK required */
             if (mlockall(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT) && (errno != EINVAL)) {
                 ALOGW("mlockall failed %s", strerror(errno));
             }
 
-            sched_setscheduler(0, SCHED_FIFO, &param);
+            /* CAP_NICE required */
+            if (sched_setscheduler(0, SCHED_FIFO, &param)) {
+                ALOGW("set SCHED_FIFO failed %s", strerror(errno));
+            }
         }
 
         mainloop();
diff --git a/lmkd/lmkd.rc b/lmkd/lmkd.rc
index 3bb84ab..76b6055 100644
--- a/lmkd/lmkd.rc
+++ b/lmkd/lmkd.rc
@@ -1,6 +1,8 @@
 service lmkd /system/bin/lmkd
     class core
-    group root readproc
+    user lmkd
+    group lmkd system readproc
+    capabilities DAC_OVERRIDE KILL IPC_LOCK SYS_NICE SYS_RESOURCE
     critical
     socket lmkd seqpacket 0660 system system
     writepid /dev/cpuset/system-background/tasks
diff --git a/logcat/event.logtags b/logcat/event.logtags
index 0983676..7c40a77 100644
--- a/logcat/event.logtags
+++ b/logcat/event.logtags
@@ -119,6 +119,9 @@
 70200 aggregation (aggregation time|2|3)
 70201 aggregation_test (field1|1|2),(field2|1|2),(field3|1|2),(field4|1|2),(field5|1|2)
 
+# gms refuses to register this log tag, b/30156345
+70220 gms_unknown
+
 # libc failure logging
 80100 bionic_event_memcpy_buffer_overflow (uid|1)
 80105 bionic_event_strcat_buffer_overflow (uid|1)
diff --git a/rootdir/init.rc b/rootdir/init.rc
index 1462570..d3504ad 100644
--- a/rootdir/init.rc
+++ b/rootdir/init.rc
@@ -21,9 +21,6 @@
     # Set the security context of /adb_keys if present.
     restorecon /adb_keys
 
-    # Shouldn't be necessary, but sdcard won't start without it. http://b/22568628.
-    mkdir /mnt 0775 root system
-
     # Set the security context of /postinstall if present.
     restorecon /postinstall
 
@@ -34,6 +31,9 @@
     # root memory control cgroup, used by lmkd
     mkdir /dev/memcg 0700 root system
     mount cgroup none /dev/memcg nodev noexec nosuid memory
+    # memory.pressure_level used by lmkd
+    chown root system /dev/memcg/memory.pressure_level
+    chmod 0040 /dev/memcg/memory.pressure_level
     # app mem cgroups, used by activity manager, lmkd and zygote
     mkdir /dev/memcg/apps/ 0755 system system
     # cgroup for system_server and surfaceflinger
@@ -80,9 +80,6 @@
     chmod 0664 /dev/stune/top-app/tasks
     chmod 0664 /dev/stune/rt/tasks
 
-    # Mount staging areas for devices managed by vold
-    # See storage config details at http://source.android.com/tech/storage/
-    mount tmpfs tmpfs /mnt nodev noexec nosuid mode=0755,uid=0,gid=1000
     restorecon_recursive /mnt
 
     mount configfs none /config nodev noexec nosuid
@@ -509,6 +506,7 @@
     mkdir /data/ss 0700 system system
 
     mkdir /data/system 0775 system system
+    mkdir /data/system/dropbox 0700 system system
     mkdir /data/system/heapdump 0700 system system
     mkdir /data/system/users 0775 system system
 
diff --git a/toolbox/Android.bp b/toolbox/Android.bp
index ddd95b2..c6d4d9b 100644
--- a/toolbox/Android.bp
+++ b/toolbox/Android.bp
@@ -63,7 +63,6 @@
     static_libs: ["libpropertyinfoparser"],
 
     symlinks: [
-        "dd",
         "getevent",
         "getprop",
         "newfs_msdos",