Kernel: Instrument syscalls with their process big lock requirements

Currently all syscalls run under the Process:m_big_lock, which is an
obvious bottleneck. Long term we would like to remove the big lock and
replace it with the required fine grained locking.

To facilitate this goal we need a way of gradually decomposing the big
lock into the all of the required fine grained locks. This commit
introduces instrumentation to the syscall table, allowing the big lock
requirement to be toggled on/off per syscall.

Eventually when we are finished, no syscall will required the big lock,
and we'll be able to remove all of this instrumentation.

Kernel/API/Syscall.h

@@ -28,165 +28,170 @@ typedef u32 socklen_t;
 
 namespace Kernel {
 
-#define ENUMERATE_SYSCALLS(S)     \
-    S(yield)                      \
-    S(open)                       \
-    S(close)                      \
-    S(read)                       \
-    S(lseek)                      \
-    S(kill)                       \
-    S(getuid)                     \
-    S(exit)                       \
-    S(geteuid)                    \
-    S(getegid)                    \
-    S(getgid)                     \
-    S(getpid)                     \
-    S(getppid)                    \
-    S(getresuid)                  \
-    S(getresgid)                  \
-    S(waitid)                     \
-    S(mmap)                       \
-    S(munmap)                     \
-    S(get_dir_entries)            \
-    S(getcwd)                     \
-    S(gettimeofday)               \
-    S(gethostname)                \
-    S(sethostname)                \
-    S(chdir)                      \
-    S(uname)                      \
-    S(set_mmap_name)              \
-    S(readlink)                   \
-    S(write)                      \
-    S(ttyname)                    \
-    S(stat)                       \
-    S(getsid)                     \
-    S(setsid)                     \
-    S(getpgid)                    \
-    S(setpgid)                    \
-    S(getpgrp)                    \
-    S(fork)                       \
-    S(execve)                     \
-    S(dup2)                       \
-    S(sigaction)                  \
-    S(umask)                      \
-    S(getgroups)                  \
-    S(setgroups)                  \
-    S(sigreturn)                  \
-    S(sigprocmask)                \
-    S(sigpending)                 \
-    S(pipe)                       \
-    S(killpg)                     \
-    S(seteuid)                    \
-    S(setegid)                    \
-    S(setuid)                     \
-    S(setgid)                     \
-    S(setreuid)                   \
-    S(setresuid)                  \
-    S(setresgid)                  \
-    S(alarm)                      \
-    S(fstat)                      \
-    S(access)                     \
-    S(fcntl)                      \
-    S(ioctl)                      \
-    S(mkdir)                      \
-    S(times)                      \
-    S(utime)                      \
-    S(sync)                       \
-    S(ptsname)                    \
-    S(select)                     \
-    S(unlink)                     \
-    S(poll)                       \
-    S(rmdir)                      \
-    S(chmod)                      \
-    S(socket)                     \
-    S(bind)                       \
-    S(accept4)                    \
-    S(listen)                     \
-    S(connect)                    \
-    S(link)                       \
-    S(chown)                      \
-    S(fchmod)                     \
-    S(symlink)                    \
-    S(sendmsg)                    \
-    S(recvmsg)                    \
-    S(getsockopt)                 \
-    S(setsockopt)                 \
-    S(create_thread)              \
-    S(gettid)                     \
-    S(rename)                     \
-    S(ftruncate)                  \
-    S(exit_thread)                \
-    S(mknod)                      \
-    S(writev)                     \
-    S(beep)                       \
-    S(getsockname)                \
-    S(getpeername)                \
-    S(socketpair)                 \
-    S(sched_setparam)             \
-    S(sched_getparam)             \
-    S(fchown)                     \
-    S(halt)                       \
-    S(reboot)                     \
-    S(mount)                      \
-    S(umount)                     \
-    S(dump_backtrace)             \
-    S(dbgputch)                   \
-    S(dbgputstr)                  \
-    S(create_inode_watcher)       \
-    S(inode_watcher_add_watch)    \
-    S(inode_watcher_remove_watch) \
-    S(mprotect)                   \
-    S(realpath)                   \
-    S(get_process_name)           \
-    S(fchdir)                     \
-    S(getrandom)                  \
-    S(getkeymap)                  \
-    S(setkeymap)                  \
-    S(clock_gettime)              \
-    S(clock_settime)              \
-    S(clock_nanosleep)            \
-    S(join_thread)                \
-    S(module_load)                \
-    S(module_unload)              \
-    S(detach_thread)              \
-    S(set_thread_name)            \
-    S(get_thread_name)            \
-    S(madvise)                    \
-    S(purge)                      \
-    S(profiling_enable)           \
-    S(profiling_disable)          \
-    S(profiling_free_buffer)      \
-    S(futex)                      \
-    S(chroot)                     \
-    S(pledge)                     \
-    S(unveil)                     \
-    S(perf_event)                 \
-    S(shutdown)                   \
-    S(get_stack_bounds)           \
-    S(ptrace)                     \
-    S(sendfd)                     \
-    S(recvfd)                     \
-    S(sysconf)                    \
-    S(set_process_name)           \
-    S(disown)                     \
-    S(adjtime)                    \
-    S(allocate_tls)               \
-    S(prctl)                      \
-    S(mremap)                     \
-    S(set_coredump_metadata)      \
-    S(anon_create)                \
-    S(msyscall)                   \
-    S(readv)                      \
-    S(emuctl)                     \
-    S(statvfs)                    \
-    S(fstatvfs)                   \
-    S(kill_thread)
+enum class NeedsBigProcessLock {
+    Yes,
+    No
+};
+
+#define ENUMERATE_SYSCALLS(S)                               \
+    S(yield, NeedsBigProcessLock::Yes)                      \
+    S(open, NeedsBigProcessLock::Yes)                       \
+    S(close, NeedsBigProcessLock::Yes)                      \
+    S(read, NeedsBigProcessLock::Yes)                       \
+    S(lseek, NeedsBigProcessLock::Yes)                      \
+    S(kill, NeedsBigProcessLock::Yes)                       \
+    S(getuid, NeedsBigProcessLock::Yes)                     \
+    S(exit, NeedsBigProcessLock::Yes)                       \
+    S(geteuid, NeedsBigProcessLock::Yes)                    \
+    S(getegid, NeedsBigProcessLock::Yes)                    \
+    S(getgid, NeedsBigProcessLock::Yes)                     \
+    S(getpid, NeedsBigProcessLock::Yes)                     \
+    S(getppid, NeedsBigProcessLock::Yes)                    \
+    S(getresuid, NeedsBigProcessLock::Yes)                  \
+    S(getresgid, NeedsBigProcessLock::Yes)                  \
+    S(waitid, NeedsBigProcessLock::Yes)                     \
+    S(mmap, NeedsBigProcessLock::Yes)                       \
+    S(munmap, NeedsBigProcessLock::Yes)                     \
+    S(get_dir_entries, NeedsBigProcessLock::Yes)            \
+    S(getcwd, NeedsBigProcessLock::Yes)                     \
+    S(gettimeofday, NeedsBigProcessLock::Yes)               \
+    S(gethostname, NeedsBigProcessLock::Yes)                \
+    S(sethostname, NeedsBigProcessLock::Yes)                \
+    S(chdir, NeedsBigProcessLock::Yes)                      \
+    S(uname, NeedsBigProcessLock::Yes)                      \
+    S(set_mmap_name, NeedsBigProcessLock::Yes)              \
+    S(readlink, NeedsBigProcessLock::Yes)                   \
+    S(write, NeedsBigProcessLock::Yes)                      \
+    S(ttyname, NeedsBigProcessLock::Yes)                    \
+    S(stat, NeedsBigProcessLock::Yes)                       \
+    S(getsid, NeedsBigProcessLock::Yes)                     \
+    S(setsid, NeedsBigProcessLock::Yes)                     \
+    S(getpgid, NeedsBigProcessLock::Yes)                    \
+    S(setpgid, NeedsBigProcessLock::Yes)                    \
+    S(getpgrp, NeedsBigProcessLock::Yes)                    \
+    S(fork, NeedsBigProcessLock::Yes)                       \
+    S(execve, NeedsBigProcessLock::Yes)                     \
+    S(dup2, NeedsBigProcessLock::Yes)                       \
+    S(sigaction, NeedsBigProcessLock::Yes)                  \
+    S(umask, NeedsBigProcessLock::Yes)                      \
+    S(getgroups, NeedsBigProcessLock::Yes)                  \
+    S(setgroups, NeedsBigProcessLock::Yes)                  \
+    S(sigreturn, NeedsBigProcessLock::Yes)                  \
+    S(sigprocmask, NeedsBigProcessLock::Yes)                \
+    S(sigpending, NeedsBigProcessLock::Yes)                 \
+    S(pipe, NeedsBigProcessLock::Yes)                       \
+    S(killpg, NeedsBigProcessLock::Yes)                     \
+    S(seteuid, NeedsBigProcessLock::Yes)                    \
+    S(setegid, NeedsBigProcessLock::Yes)                    \
+    S(setuid, NeedsBigProcessLock::Yes)                     \
+    S(setgid, NeedsBigProcessLock::Yes)                     \
+    S(setreuid, NeedsBigProcessLock::Yes)                   \
+    S(setresuid, NeedsBigProcessLock::Yes)                  \
+    S(setresgid, NeedsBigProcessLock::Yes)                  \
+    S(alarm, NeedsBigProcessLock::Yes)                      \
+    S(fstat, NeedsBigProcessLock::Yes)                      \
+    S(access, NeedsBigProcessLock::Yes)                     \
+    S(fcntl, NeedsBigProcessLock::Yes)                      \
+    S(ioctl, NeedsBigProcessLock::Yes)                      \
+    S(mkdir, NeedsBigProcessLock::Yes)                      \
+    S(times, NeedsBigProcessLock::Yes)                      \
+    S(utime, NeedsBigProcessLock::Yes)                      \
+    S(sync, NeedsBigProcessLock::Yes)                       \
+    S(ptsname, NeedsBigProcessLock::Yes)                    \
+    S(select, NeedsBigProcessLock::Yes)                     \
+    S(unlink, NeedsBigProcessLock::Yes)                     \
+    S(poll, NeedsBigProcessLock::Yes)                       \
+    S(rmdir, NeedsBigProcessLock::Yes)                      \
+    S(chmod, NeedsBigProcessLock::Yes)                      \
+    S(socket, NeedsBigProcessLock::Yes)                     \
+    S(bind, NeedsBigProcessLock::Yes)                       \
+    S(accept4, NeedsBigProcessLock::Yes)                    \
+    S(listen, NeedsBigProcessLock::Yes)                     \
+    S(connect, NeedsBigProcessLock::Yes)                    \
+    S(link, NeedsBigProcessLock::Yes)                       \
+    S(chown, NeedsBigProcessLock::Yes)                      \
+    S(fchmod, NeedsBigProcessLock::Yes)                     \
+    S(symlink, NeedsBigProcessLock::Yes)                    \
+    S(sendmsg, NeedsBigProcessLock::Yes)                    \
+    S(recvmsg, NeedsBigProcessLock::Yes)                    \
+    S(getsockopt, NeedsBigProcessLock::Yes)                 \
+    S(setsockopt, NeedsBigProcessLock::Yes)                 \
+    S(create_thread, NeedsBigProcessLock::Yes)              \
+    S(gettid, NeedsBigProcessLock::Yes)                     \
+    S(rename, NeedsBigProcessLock::Yes)                     \
+    S(ftruncate, NeedsBigProcessLock::Yes)                  \
+    S(exit_thread, NeedsBigProcessLock::Yes)                \
+    S(mknod, NeedsBigProcessLock::Yes)                      \
+    S(writev, NeedsBigProcessLock::Yes)                     \
+    S(beep, NeedsBigProcessLock::Yes)                       \
+    S(getsockname, NeedsBigProcessLock::Yes)                \
+    S(getpeername, NeedsBigProcessLock::Yes)                \
+    S(socketpair, NeedsBigProcessLock::Yes)                 \
+    S(sched_setparam, NeedsBigProcessLock::Yes)             \
+    S(sched_getparam, NeedsBigProcessLock::Yes)             \
+    S(fchown, NeedsBigProcessLock::Yes)                     \
+    S(halt, NeedsBigProcessLock::Yes)                       \
+    S(reboot, NeedsBigProcessLock::Yes)                     \
+    S(mount, NeedsBigProcessLock::Yes)                      \
+    S(umount, NeedsBigProcessLock::Yes)                     \
+    S(dump_backtrace, NeedsBigProcessLock::Yes)             \
+    S(dbgputch, NeedsBigProcessLock::Yes)                   \
+    S(dbgputstr, NeedsBigProcessLock::Yes)                  \
+    S(create_inode_watcher, NeedsBigProcessLock::Yes)       \
+    S(inode_watcher_add_watch, NeedsBigProcessLock::Yes)    \
+    S(inode_watcher_remove_watch, NeedsBigProcessLock::Yes) \
+    S(mprotect, NeedsBigProcessLock::Yes)                   \
+    S(realpath, NeedsBigProcessLock::Yes)                   \
+    S(get_process_name, NeedsBigProcessLock::Yes)           \
+    S(fchdir, NeedsBigProcessLock::Yes)                     \
+    S(getrandom, NeedsBigProcessLock::Yes)                  \
+    S(getkeymap, NeedsBigProcessLock::Yes)                  \
+    S(setkeymap, NeedsBigProcessLock::Yes)                  \
+    S(clock_gettime, NeedsBigProcessLock::Yes)              \
+    S(clock_settime, NeedsBigProcessLock::Yes)              \
+    S(clock_nanosleep, NeedsBigProcessLock::Yes)            \
+    S(join_thread, NeedsBigProcessLock::Yes)                \
+    S(module_load, NeedsBigProcessLock::Yes)                \
+    S(module_unload, NeedsBigProcessLock::Yes)              \
+    S(detach_thread, NeedsBigProcessLock::Yes)              \
+    S(set_thread_name, NeedsBigProcessLock::Yes)            \
+    S(get_thread_name, NeedsBigProcessLock::Yes)            \
+    S(madvise, NeedsBigProcessLock::Yes)                    \
+    S(purge, NeedsBigProcessLock::Yes)                      \
+    S(profiling_enable, NeedsBigProcessLock::Yes)           \
+    S(profiling_disable, NeedsBigProcessLock::Yes)          \
+    S(profiling_free_buffer, NeedsBigProcessLock::Yes)      \
+    S(futex, NeedsBigProcessLock::Yes)                      \
+    S(chroot, NeedsBigProcessLock::Yes)                     \
+    S(pledge, NeedsBigProcessLock::Yes)                     \
+    S(unveil, NeedsBigProcessLock::Yes)                     \
+    S(perf_event, NeedsBigProcessLock::Yes)                 \
+    S(shutdown, NeedsBigProcessLock::Yes)                   \
+    S(get_stack_bounds, NeedsBigProcessLock::Yes)           \
+    S(ptrace, NeedsBigProcessLock::Yes)                     \
+    S(sendfd, NeedsBigProcessLock::Yes)                     \
+    S(recvfd, NeedsBigProcessLock::Yes)                     \
+    S(sysconf, NeedsBigProcessLock::Yes)                    \
+    S(set_process_name, NeedsBigProcessLock::Yes)           \
+    S(disown, NeedsBigProcessLock::Yes)                     \
+    S(adjtime, NeedsBigProcessLock::Yes)                    \
+    S(allocate_tls, NeedsBigProcessLock::Yes)               \
+    S(prctl, NeedsBigProcessLock::Yes)                      \
+    S(mremap, NeedsBigProcessLock::Yes)                     \
+    S(set_coredump_metadata, NeedsBigProcessLock::Yes)      \
+    S(anon_create, NeedsBigProcessLock::Yes)                \
+    S(msyscall, NeedsBigProcessLock::Yes)                   \
+    S(readv, NeedsBigProcessLock::Yes)                      \
+    S(emuctl, NeedsBigProcessLock::Yes)                     \
+    S(statvfs, NeedsBigProcessLock::Yes)                    \
+    S(fstatvfs, NeedsBigProcessLock::Yes)                   \
+    S(kill_thread, NeedsBigProcessLock::Yes)
 
 namespace Syscall {
 
 enum Function {
 #undef __ENUMERATE_SYSCALL
-#define __ENUMERATE_SYSCALL(x) SC_##x,
+#define __ENUMERATE_SYSCALL(sys_call, needs_lock) SC_##sys_call,
     ENUMERATE_SYSCALLS(__ENUMERATE_SYSCALL)
 #undef __ENUMERATE_SYSCALL
         __Count
@@ -196,9 +201,9 @@ constexpr const char* to_string(Function function)
 {
     switch (function) {
 #undef __ENUMERATE_SYSCALL
-#define __ENUMERATE_SYSCALL(x) \
-    case SC_##x:               \
-        return #x;
+#define __ENUMERATE_SYSCALL(sys_call, needs_lock) \
+    case SC_##sys_call:                           \
+        return #sys_call;
         ENUMERATE_SYSCALLS(__ENUMERATE_SYSCALL)
 #undef __ENUMERATE_SYSCALL
     default:
@@ -522,7 +527,7 @@ inline uintptr_t invoke(Function function, T1 arg1, T2 arg2, T3 arg3)
 }
 
 #undef __ENUMERATE_SYSCALL
-#define __ENUMERATE_SYSCALL(x) using Syscall::SC_##x;
+#define __ENUMERATE_SYSCALL(sys_call, needs_lock) using Syscall::SC_##sys_call;
 ENUMERATE_SYSCALLS(__ENUMERATE_SYSCALL)
 #undef __ENUMERATE_SYSCALL
 

Kernel/Process.h

@@ -882,6 +882,12 @@ inline ProcessID Thread::pid() const
     } while (0)
 }
 
+#define VERIFY_PROCESS_BIG_LOCK_ACQUIRED(process) \
+    VERIFY(process->big_lock().own_lock());
+
+#define VERIFY_NO_PROCESS_BIG_LOCK(process) \
+    VERIFY(!process->big_lock().own_lock());
+
 inline static String copy_string_from_user(const Kernel::Syscall::StringArgument& string)
 {
     return copy_string_from_user(string.characters, string.length);

Kernel/Syscall.cpp

@@ -90,8 +90,13 @@ UNMAP_AFTER_INIT void initialize()
 #pragma GCC diagnostic ignored "-Wcast-function-type"
 typedef KResultOr<FlatPtr> (Process::*Handler)(FlatPtr, FlatPtr, FlatPtr);
 typedef KResultOr<FlatPtr> (Process::*HandlerWithRegisterState)(RegisterState&);
-#define __ENUMERATE_SYSCALL(x) reinterpret_cast<Handler>(&Process::sys$##x),
-static const Handler s_syscall_table[] = {
+struct HandlerMetadata {
+    Handler handler;
+    NeedsBigProcessLock needs_lock;
+};
+
+#define __ENUMERATE_SYSCALL(sys_call, needs_lock) { reinterpret_cast<Handler>(&Process::sys$##sys_call), needs_lock },
+static const HandlerMetadata s_syscall_table[] = {
     ENUMERATE_SYSCALLS(__ENUMERATE_SYSCALL)
 };
 #undef __ENUMERATE_SYSCALL
@@ -126,7 +131,7 @@ KResultOr<FlatPtr> handle(RegisterState& regs, FlatPtr function, FlatPtr arg1, F
 
     if (function == SC_fork || function == SC_sigreturn) {
         // These syscalls want the RegisterState& rather than individual parameters.
-        auto handler = (HandlerWithRegisterState)s_syscall_table[function];
+        auto handler = (HandlerWithRegisterState)s_syscall_table[function].handler;
         return (process.*(handler))(regs);
     }
 
@@ -135,12 +140,12 @@ KResultOr<FlatPtr> handle(RegisterState& regs, FlatPtr function, FlatPtr arg1, F
         return ENOSYS;
     }
 
-    if (s_syscall_table[function] == nullptr) {
+    if (s_syscall_table[function].handler == nullptr) {
         dbgln("Null syscall {} requested, you probably need to rebuild this program!", function);
         return ENOSYS;
     }
 
-    return (process.*(s_syscall_table[function]))(arg1, arg2, arg3);
+    return (process.*(s_syscall_table[function].handler))(arg1, arg2, arg3);
 }
 
 }