Job control, function keys, and real sleeps

- SIGSTOP/SIGTSTP/SIGTTIN/SIGTTOU stop, SIGCONT resumes; the parent gets
  SIGCHLD and wait4 reports WUNTRACED/WCONTINUED; ps shows T. Ctrl-Z works,
  and mc no longer hangs waiting for its subshell to stop.
- Blocking reads (console, pty, pipe) and sleeps take a pending stop and
  carry on after SIGCONT; a caught signal without SA_RESTART ends them
  with EINTR.
- nanosleep reads tv_nsec (sleep 0.5 used to return at once), writes the
  remainder, and clock_nanosleep understands TIMER_ABSTIME.
- rt_sigaction uses the k_sigaction layout for Linux-ABI processes: the old
  action's restorer was written 112 bytes past the caller's struct. Flags
  are kept; SIGKILL and SIGSTOP cannot be caught.
- F1..F12 reach programs, as the Linux console sends them (PS/2, USB, I2C).
- TIOCLINUX subcode 6 answers "no modifiers" instead of echoing the 6 back.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4

include/core/defs.h

@@ -61,6 +61,10 @@ int             pciuser_mmio_ok(uint64 phys, uint64 len);
 void            pciuser_release_all(struct proc*);
 int             pciuser_owner_pid(int idx);
 int             proc_reincarnate(int status);
+int             sig_default_stops(int);
+void            proc_stop(int);
+int             sig_attention(struct proc*);
+int             sig_blocking_check(void);
 void            proc_cputime(uint64 *user, uint64 *sys);
 // kernel/dev/procfs.c: /proc/strace
 int             strace_wanted(const char *name);

include/core/proc.h

@@ -175,6 +175,7 @@ struct procinfo {
   char   name[16];
   char   exe[64];           // full path, for /proc/<pid>/exe
   int    kthread;           // a kernel thread, not a program
+  int    stopped;           // stopped by a signal (ps shows T)
 };
 
 struct proc {
@@ -248,6 +249,7 @@ struct proc {
   // support nested signal delivery).
   uint64 sig_handler[32];
   uint64 sig_restorer[32];
+  uint64 sig_flags[32];             // sa_flags; SA_RESTART is the one read
   struct trapframe saved_sigtf;
   int in_signal;
 
@@ -330,6 +332,13 @@ struct proc {
 
   // umask(2). Inherited across fork; 022 for the first process.
   uint umask;
+
+  // Job control. stopsig is the signal that stopped the process, 0 while it
+  // runs; stop_report and cont_report are what wait(WUNTRACED/WCONTINUED)
+  // has yet to tell the parent.
+  int stopsig;
+  int stop_report;
+  int cont_report;
 };
 #define FD_CLOEXEC 1
 

include/dev/kbd.h

@@ -59,6 +59,9 @@
 #define KBD_PGDN        (KBD_SPECIAL + 0x07)
 #define KBD_INS         (KBD_SPECIAL + 0x08)
 #define KBD_DEL         (KBD_SPECIAL + 0x09)
+// F1..F12, consecutive: KBD_F1 + n is F(n+1).
+#define KBD_F1          (KBD_SPECIAL + 0x10)
+#define KBD_F12         (KBD_SPECIAL + 0x1B)
 
 // Switch virtual terminal.
 //

kernel/core/console.c

@@ -624,6 +624,21 @@ consoleintr(int (*getc)(void))
       case KBD_PGDN: esc = "\033[6~"; break;
       case KBD_INS:  esc = "\033[2~"; break;
       case KBD_DEL:  esc = "\033[3~"; break;
+      // The function keys, as the Linux console sends them -- the strings
+      // the `linux` terminfo entry lists. Nothing sent them at all, so mc's
+      // F10 never quit and htop's F-key bar was decoration.
+      case KBD_F1:      esc = "\033[[A";  break;
+      case KBD_F1 + 1:  esc = "\033[[B";  break;
+      case KBD_F1 + 2:  esc = "\033[[C";  break;
+      case KBD_F1 + 3:  esc = "\033[[D";  break;
+      case KBD_F1 + 4:  esc = "\033[[E";  break;
+      case KBD_F1 + 5:  esc = "\033[17~"; break;
+      case KBD_F1 + 6:  esc = "\033[18~"; break;
+      case KBD_F1 + 7:  esc = "\033[19~"; break;
+      case KBD_F1 + 8:  esc = "\033[20~"; break;
+      case KBD_F1 + 9:  esc = "\033[21~"; break;
+      case KBD_F1 + 10: esc = "\033[23~"; break;
+      case KBD_F12:     esc = "\033[24~"; break;
       }
       if (esc) {
         // Not echoed. In raw mode -- which is where a line editor lives, and
@@ -727,6 +742,16 @@ consoleread(struct inode *ip, uint64 off, char *dst, int n)
         ilock(ip);
         return -EINTR;
       }
+      if (sig_attention(myproc())) {     // Ctrl-Z, or a handler to run
+        release(&input_lock);
+        int e = sig_blocking_check();
+        if (e) {
+          ilock(ip);
+          return e;
+        }
+        acquire(&input_lock);
+        continue;
+      }
       sleep(&v->in.r, &input_lock);
     }
     c = v->in.buf[v->in.r++ % VT_INPUT_BUF];

kernel/dev/keymap.c

@@ -24,6 +24,12 @@
   [0xCB] = KBD_LF,    [0xCD] = KBD_RT,               \
   [0x97] = KBD_HOME,  [0xCF] = KBD_END,              \
   [0xD2] = KBD_INS,   [0xD3] = KBD_DEL,              \
+  [0x3B] = KBD_F1,     [0x3C] = KBD_F1 + 1,          \
+  [0x3D] = KBD_F1 + 2, [0x3E] = KBD_F1 + 3,          \
+  [0x3F] = KBD_F1 + 4, [0x40] = KBD_F1 + 5,          \
+  [0x41] = KBD_F1 + 6, [0x42] = KBD_F1 + 7,          \
+  [0x43] = KBD_F1 + 8, [0x44] = KBD_F1 + 9,          \
+  [0x57] = KBD_F1 + 10, [0x58] = KBD_F12,            \
   [0x9C] = '\n',      /* KP_Enter */                 \
   [0xB5] = '/'        /* KP_Div   */
 

kernel/dev/pty.c

@@ -458,6 +458,16 @@ pts_read(struct inode *ip, uint64 off, char *dst, int n)
         ilock(ip);
         return got ? got : -EINTR;
       }
+      if (sig_attention(myproc())) {
+        release(&ptylock);
+        int e = sig_blocking_check();
+        if (e) {
+          ilock(ip);
+          return got ? got : e;
+        }
+        acquire(&ptylock);
+        continue;
+      }
       sleep(&p->ir, &ptylock);
     }
     c = p->in[p->ir++ % PTY_BUF];

kernel/drivers/usb/hid/usb_hid.c

@@ -168,6 +168,8 @@ hid_special(uchar usage, uchar mods)
   if (usage >= 0x3A && usage < 0x3A + NVT && (mods & HID_MOD_ALT) &&
       !(mods & HID_MOD_SHIFT))
     return KBD_VT_SWITCH + (usage - 0x3A);
+  if (usage >= 0x3A && usage <= 0x45)
+    return KBD_F1 + (usage - 0x3A);
   switch (usage) {
   case 0x49: return KBD_INS;
   case 0x4A: return KBD_HOME;

kernel/fs/pipe.c

@@ -183,6 +183,14 @@ piperead(struct pipe *p, char *addr, int n)
       release(&p->lock);
       return -1;
     }
+    if(sig_attention(myproc())){     // stopped, or a handler wants to run
+      release(&p->lock);
+      int e = sig_blocking_check();
+      if(e)
+        return e;
+      acquire(&p->lock);
+      continue;
+    }
     sleep(&p->nread, &p->lock); //DOC: piperead-sleep
   }
   i = 0;

kernel/fs/procfs.c

@@ -198,7 +198,7 @@ gen_pid_stat(struct sb *s, const struct procinfo *pi)
   sbputs(s, " (");
   sbputs(s, pi->name);
   sbputs(s, ") ");
-  sbputc(s, state_char(pi->state));
+  sbputc(s, pi->stopped ? 'T' : state_char(pi->state));
   sbputc(s, ' ');
   sbputi(s, pi->ppid);   // 4  ppid
   sbputc(s, ' ');
@@ -247,7 +247,7 @@ static void
 gen_pid_status(struct sb *s, const struct procinfo *pi)
 {
   sbputs(s, "Name:\t");   sbputs(s, pi->name);            sbputc(s, '\n');
-  sbputs(s, "State:\t");  sbputc(s, state_char(pi->state));
+  sbputs(s, "State:\t");  sbputc(s, pi->stopped ? 'T' : state_char(pi->state));
   sbputs(s, "\nTgid:\t"); sbputi(s, pi->pid);
   sbputs(s, "\nPid:\t");  sbputi(s, pi->pid);
   sbputs(s, "\nPPid:\t"); sbputi(s, pi->ppid);

kernel/proc/proc.c

@@ -318,6 +318,7 @@ found:
   p->alarm_at = p->alarm_every = 0;   // a child's timers start disarmed
   p->utime = p->stime = p->cutime = p->cstime = 0;
   p->strace = 0;
+  p->stopsig = p->stop_report = p->cont_report = 0;
   p->umask = 022;
 
   // A recycled slot must not inherit a dead process's place in the queue --
@@ -486,6 +487,7 @@ fork(void)
   // aren't inherited. allocproc() already zeroed pending_signals/
   // in_signal/saved_sigtf for this fresh slot.
   memmove(np->sig_handler, myproc()->sig_handler, sizeof(np->sig_handler));
+  memmove(np->sig_flags, myproc()->sig_flags, sizeof(np->sig_flags));
   memmove(np->sig_restorer, myproc()->sig_restorer, sizeof(np->sig_restorer));
   np->signal_mask = myproc()->signal_mask;
 
@@ -779,6 +781,25 @@ waitpid(int pid, int *status, int options, uint64 *ut, uint64 *st)
         if (status) *status = rstatus;
         return rpid;
       }
+      // WUNTRACED and WCONTINUED: a stop or a resume, each reported once.
+      if((options & 2) && p->stop_report){
+        int rpid = p->pid, rstatus = (p->stop_report << 8) | 0x7f;
+        p->stop_report = 0;
+        if (ut) *ut = p->utime;
+        if (st) *st = p->stime;
+        release(&ptable.lock);
+        if (status) *status = rstatus;
+        return rpid;
+      }
+      if((options & 8) && p->cont_report){
+        int rpid = p->pid;
+        p->cont_report = 0;
+        if (ut) *ut = p->utime;
+        if (st) *st = p->stime;
+        release(&ptable.lock);
+        if (status) *status = 0xffff;
+        return rpid;
+      }
     }
 
     // Two different failures, and they used to share one bare -1.
@@ -1645,6 +1666,139 @@ alarm_thread_start(void)
     cprintf("alarm: no kernel thread; alarm() and setitimer() will not fire\n");
 }
 
+// --- Stopping and continuing --------------------------------------------
+//
+// SIGSTOP, and SIGTSTP/SIGTTIN/SIGTTOU with no handler, stop a process; SIGCONT
+// resumes it. Neither existed: Ctrl-Z did nothing, and mc -- which stops its
+// subshell and waits to be told it has stopped -- hung before drawing.
+//
+// The stop itself happens on the way back to user mode, in deliver_signals(),
+// where the process can sleep. SIGCONT acts when it is sent, because a
+// stopped process will not reach that point by itself.
+
+int
+sig_default_stops(int sig)
+{
+  return sig == SIGSTOP || sig == SIGTSTP || sig == SIGTTIN || sig == SIGTTOU;
+}
+
+#define STOP_SIGS ((1ULL << SIGSTOP) | (1ULL << SIGTSTP) | \
+                   (1ULL << SIGTTIN) | (1ULL << SIGTTOU))
+
+// Tell the parent about a stop or a resume: SIGCHLD, and a wake for a wait()
+// or a sigsuspend() that is waiting to hear it. Called with ptable.lock.
+static void
+notify_parent(struct proc *p)
+{
+  struct proc *pp = p->parent;
+  if(pp == 0)
+    return;
+  pp->pending_signals |= 1ULL << SIGCHLD;
+  if(pp->state == SLEEPING){
+    pp->state = RUNNABLE;
+    sched_enqueue(pp);
+  }
+}
+
+// The side effects of sending sig to p, beyond setting it pending. With
+// ptable.lock held.
+static void
+sig_job_control(struct proc *p, int sig)
+{
+  if(sig == SIGCONT){
+    // A resume cancels any stop still on its way, and the other way round.
+    p->pending_signals &= ~STOP_SIGS;
+    if(p->stopsig){
+      p->stopsig = 0;
+      p->stop_report = 0;
+      p->cont_report = 1;
+      notify_parent(p);
+    }
+  } else if(sig_default_stops(sig)){
+    p->pending_signals &= ~(1ULL << SIGCONT);
+  }
+}
+
+// Stop the calling process with sig until SIGCONT or SIGKILL.
+void
+proc_stop(int sig)
+{
+  struct proc *p = myproc();
+
+  acquire(&ptable.lock);
+  p->stopsig = sig;
+  p->stop_report = sig;
+  p->cont_report = 0;
+  notify_parent(p);
+  // Other signals wake it too; only a resume or a kill ends the stop.
+  while(p->stopsig && !p->killed)
+    sleep(&p->stopsig, &ptable.lock);
+  p->stopsig = 0;
+  release(&ptable.lock);
+}
+
+// A handler that ends a blocking call. Installed with SA_RESTART -- which
+// musl's signal() always asks for -- the call goes on waiting instead, and
+// the handler runs when it returns: Linux restarts the call around the
+// handler, and a program that asked for that does not check for EINTR.
+#define SA_RESTART 0x10000000
+static int
+sig_interrupts(struct proc *p, int sig)
+{
+  return p->sig_handler[sig] > 1 && !(p->sig_flags[sig] & SA_RESTART);
+}
+
+// Whether a pending signal should end a blocking wait: one that will stop the
+// process, or one with a handler to run. A lockless look, for the inside of a
+// sleep loop that holds some other lock; sig_blocking_check() does the work.
+int
+sig_attention(struct proc *p)
+{
+  uint64 d = p->pending_signals & ~p->signal_mask;
+  if(d == 0)
+    return 0;
+  for(int sig = 1; sig < 32; sig++)
+    if((d & (1ULL << sig)) &&
+       (sig == SIGSTOP || sig_interrupts(p, sig) ||
+        (sig_default_stops(sig) && p->sig_handler[sig] == 0)))
+      return 1;
+  return 0;
+}
+
+// For a blocking call that saw sig_attention(), with no locks held. A stop is
+// taken here, and the call carries on waiting once continued -- which is what
+// Linux's restart does, and what lets Ctrl-Z stop a `sleep` or a `cat`. A
+// signal with a handler ends the call with -EINTR, as POSIX has it.
+int
+sig_blocking_check(void)
+{
+  struct proc *p = myproc();
+
+  for(;;){
+    int stop = 0, caught = 0;
+    acquire(&ptable.lock);
+    uint64 d = p->pending_signals & ~p->signal_mask;
+    for(int sig = 1; sig < 32 && !stop; sig++){
+      if(!(d & (1ULL << sig)))
+        continue;
+      if(sig == SIGSTOP || (sig_default_stops(sig) && p->sig_handler[sig] == 0)){
+        p->pending_signals &= ~(1ULL << sig);
+        stop = sig;
+      } else if(sig_interrupts(p, sig)){
+        caught = 1;
+      }
+    }
+    release(&ptable.lock);
+    if(stop){
+      proc_stop(stop);
+      if(p->killed)
+        return -EINTR;
+      continue;
+    }
+    return (caught || p->killed) ? -EINTR : 0;
+  }
+}
+
 int
 sigsend(int pid, int sig)
 {
@@ -1661,6 +1815,7 @@ sigsend(int pid, int sig)
         return 0;           // silently dropped, as POSIX requires
       }
       p->pending_signals |= 1ULL << sig;
+      sig_job_control(p, sig);
       if(sig_kills_now(p, sig))
         p->killed = 1;
       if(sig == SIGKILL)
@@ -1705,6 +1860,7 @@ sigsend_pgrp(int pgid, int sig)
     if(init_ignores(p, sig))
       continue;              // see init_ignores(): this is the Ctrl-C case
     p->pending_signals |= 1ULL << sig;
+    sig_job_control(p, sig);
     if(sig_kills_now(p, sig)){
       p->killed = 1;
       if(p->killsig == 0)
@@ -1929,6 +2085,7 @@ proc_snapshot(int idx, struct procinfo *out)
     safestrcpy(out->name, p->name, sizeof(out->name));
     safestrcpy(out->exe, p->exe, sizeof(out->exe));
     out->kthread = p->kfn != 0;
+    out->stopped = p->stopsig != 0;
     release(&ptable.lock);
     return 0;
   }

kernel/sys/sys_fs.c

@@ -1653,6 +1653,20 @@ addr_t sys_ioctl(void)
     // terminal and nothing to choose. It records the choice now, which is what
     // /dev/tty resolves through and what a session leader on tty3 needs so
     // that its /dev/tty is tty3 and not whatever is on screen.
+    // TIOCLINUX. Only subcode 6, the modifier state, is answered -- with
+    // "none held", which is true often enough. mc asks it after every key on
+    // a Linux console, and the ioctl used to succeed without writing: the 6
+    // it passed in came back as Ctrl+AltGr held, and Enter arrived as an
+    // unknown chord printed as a dot.
+    if (request == 0x541C) {
+        uchar sub;
+        if (copy_from_user(&sub, argp, 1) < 0)
+            return -EFAULT;
+        if (sub != 6)
+            return -EINVAL;
+        sub = 0;
+        return copy_to_user(argp, &sub, 1) < 0 ? -EFAULT : 0;
+    }
     if (request == TIOCSCTTY) {
         struct vt *v = vt_for_minor(f->ip->minor);
         myproc()->ctty = vt_index(v) + 1;

kernel/sys/sys_sig.c

@@ -79,13 +79,18 @@ sys_sigwait(void)
     return 0;
 }
 
-// musl's struct sigaction (musl 1.2.5 signal.h) for this target:
+// The native musl hands over its own struct sigaction (musl 1.2.5 signal.h):
-// sa_handler (8 bytes) at offset 0, then a 128-byte sigset_t (musl uses
+// sa_handler at 0, a 128-byte sigset_t, sa_flags (4 bytes) at 136, then
-// 128/sizeof(long) = 16 longs regardless of NSIG), sa_flags (4 bytes) +
+// sa_restorer at 144.
-// padding, then sa_restorer (8 bytes) at offset 144. We only care about
+//
-// those two 8-byte fields.
+// A Linux-ABI process does not. Its musl converts to the kernel's
+// k_sigaction first -- handler at 0, flags at 8, restorer at 16, mask at 24,
+// thirty-two bytes in all -- and this used the native offsets for both, so
+// every Alpine program that asked for the old action had 8 bytes written 112
+// bytes past the end of the struct, into whatever its stack held there.
 #define SA_HANDLER_OFF 0
-#define SA_RESTORER_OFF 144
+#define SA_FLAGS_OFF(p)    ((p)->syscall_map ? 8 : 136)
+#define SA_RESTORER_OFF(p) ((p)->syscall_map ? 16 : 144)
 
 // rt_sigaction: store the handler + restorer so deliver_signals() (called
 // from syscall()'s tail, kernel/sys/syscall.c) can actually jump to user
@@ -108,15 +113,30 @@ sys_rt_sigaction(void)
     argaddr(2, &oldactp);
     if (sig <= 0 || sig >= 32) return -1;
 
+    struct proc *p = myproc();
     if (oldactp) {
-        put_user(myproc()->sig_handler[sig], oldactp + SA_HANDLER_OFF);
+        put_user(p->sig_handler[sig], oldactp + SA_HANDLER_OFF);
-        put_user(myproc()->sig_restorer[sig], oldactp + SA_RESTORER_OFF);
+        put_user(p->sig_restorer[sig], oldactp + SA_RESTORER_OFF(p));
+        if (p->syscall_map) {
+            put_user(p->sig_flags[sig], oldactp + SA_FLAGS_OFF(p));
+        } else {
+            uint32 fl = (uint32)p->sig_flags[sig];
+            put_user(fl, oldactp + SA_FLAGS_OFF(p));
+        }
     }
     if (actp) {
-        copy_from_user(&myproc()->sig_handler[sig], actp + SA_HANDLER_OFF,
+        uint64 h, r, fl = 0;
-                       sizeof(myproc()->sig_handler[sig]));
+        if (copy_from_user(&h, actp + SA_HANDLER_OFF, sizeof(h)) < 0 ||
-        copy_from_user(&myproc()->sig_restorer[sig], actp + SA_RESTORER_OFF,
+            copy_from_user(&r, actp + SA_RESTORER_OFF(p), sizeof(r)) < 0 ||
-                       sizeof(myproc()->sig_restorer[sig]));
+            copy_from_user(&fl, actp + SA_FLAGS_OFF(p),
+                           p->syscall_map ? 8 : 4) < 0)
+            return -EFAULT;
+        // SIGKILL and SIGSTOP cannot be caught or ignored.
+        if (sig == SIGKILL || sig == SIGSTOP)
+            return -EINVAL;
+        p->sig_handler[sig] = h;
+        p->sig_restorer[sig] = r;
+        p->sig_flags[sig] = fl;
     }
     return 0;
 }
@@ -195,6 +215,16 @@ deliver_signals(struct trapframe *tf)
         // Except for a signal that was blocked when it arrived: sigsend()
         // left it pending without acting, and now that it is deliverable its
         // default action happens here.
+        // SIGSTOP cannot be caught or ignored; the other three stop only by
+        // default. Stopped here and resumed here, by SIGCONT or SIGKILL.
+        if (sig == SIGSTOP ||
+            (sig_default_stops(sig) && myproc()->sig_handler[sig] == 0)) {
+            myproc()->pending_signals &= ~(1ULL << sig);
+            proc_stop(sig);
+            if (myproc()->killed)
+                exit(W_EXITCODE(0, myproc()->killsig ? myproc()->killsig : SIGKILL));
+            continue;
+        }
         if (myproc()->sig_handler[sig] <= 1) {
             myproc()->pending_signals &= ~(1ULL << sig);
             if (myproc()->sig_handler[sig] == 0 && sig_default_kills(sig)) {

kernel/sys/sys_time.c

@@ -60,25 +60,67 @@ sys_gettimeofday(void)
     return 0;
 }
 
-addr_t
+// Sleep for a timespec, in ticks, rounded up.
-sys_nanosleep(void)
+//
+// Only tv_sec was ever read: `sleep 0.5` returned at once and every
+// sub-second sleep -- cmatrix's frame delay, a game loop's -- was no sleep at
+// all. And a signal could not end one: Ctrl-Z on `sleep 100` did nothing.
+static addr_t
+sleep_ticks(uint64 want, addr_t remaddr)
 {
-    addr_t tsaddr, raw;
+    struct proc *p = myproc();
-    if (argaddr(0, &tsaddr) < 0 || fetchaddr(tsaddr, &raw) < 0)
-        return -1;
-    int64 sec = (int64)raw;
-    if (sec > 0) {
     acquire(&tickslock);
     uint t0 = ticks;
-        while (ticks - t0 < (uint)(sec * HZ)) {
+    while ((uint64)(uint)(ticks - t0) < want) {
-            if (myproc()->killed) { release(&tickslock); return -EINTR; }
+        if (p->killed || sig_attention(p)) {
+            release(&tickslock);
+            int e = p->killed ? -EINTR : sig_blocking_check();
+            if (e) {
+                if (remaddr) {
+                    uint64 left = want - (uint)(ticks - t0);
+                    struct timespec r = { (long)(left / HZ),
+                                          (long)((left % HZ) * (1000000000 / HZ)) };
+                    copy_to_user(remaddr, &r, sizeof(r));
+                }
+                return e;
+            }
+            acquire(&tickslock);
+            continue;
+        }
         sleep(&ticks, &tickslock);
     }
     release(&tickslock);
-    }
     return 0;
 }
 
+static int
+timespec_ticks(const struct timespec *ts, uint64 *out)
+{
+    if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
+        return -EINVAL;
+    uint64 sec = (uint64)ts->tv_sec;
+    if (sec > 0xFFFFFFFFULL / HZ)
+        sec = 0xFFFFFFFFULL / HZ;          // a few years: forever enough
+    *out = sec * HZ + ((uint64)ts->tv_nsec * HZ + 999999999) / 1000000000;
+    return 0;
+}
+
+addr_t
+sys_nanosleep(void)
+{
+    addr_t tsaddr, remaddr;
+    struct timespec ts;
+    uint64 want;
+    argaddr(0, &tsaddr);
+    argaddr(1, &remaddr);
+    if (copy_from_user(&ts, tsaddr, sizeof(ts)) < 0)
+        return -EFAULT;
+    int e = timespec_ticks(&ts, &want);
+    if (e)
+        return e;
+    return sleep_ticks(want, remaddr);
+}
+
 addr_t
 sys_clock_gettime(void)
 {
@@ -173,20 +215,34 @@ sys_clock_getres(void)
 addr_t
 sys_clock_nanosleep(void)
 {
-    // Treat the same as nanosleep: arg2 is the timespec, arg1 is flags.
+    // clock_nanosleep(clockid, flags, rqtp, rmtp). TIMER_ABSTIME (1) makes
-    // Linux: clock_nanosleep(clockid, flags, rqtp, rmtp)
+    // rqtp a time on that clock rather than an interval.
-    addr_t tsaddr, raw;
+    int clkid, flags;
-    if (argaddr(2, &tsaddr) < 0 || fetchaddr(tsaddr, &raw) < 0)
+    addr_t tsaddr, remaddr;
-        return -1;
+    struct timespec ts;
-    int64 sec = (int64)raw;
+    uint64 want;
-    if (sec > 0) {
+    argint(0, &clkid);
-        acquire(&tickslock);
+    argint(1, &flags);
-        uint t0 = ticks;
+    argaddr(2, &tsaddr);
-        while (ticks - t0 < (uint)(sec * HZ)) {
+    argaddr(3, &remaddr);
-            if (myproc()->killed) { release(&tickslock); return -EINTR; }
+    if (copy_from_user(&ts, tsaddr, sizeof(ts)) < 0)
-            sleep(&ticks, &tickslock);
+        return -EFAULT;
+    if (flags & 1) {
+        uint64 sec, usec;
+        wallclock(&sec, &usec);
+        if (clkid == 1)
+            sec -= boot_epoch;
+        long dsec = ts.tv_sec - (long)sec;
+        long dns  = ts.tv_nsec - (long)(usec * 1000);
+        if (dns < 0) { dns += 1000000000; dsec--; }
+        if (dsec < 0)
+            return 0;                       // already past
+        ts.tv_sec = dsec;
+        ts.tv_nsec = dns;
+        remaddr = 0;                        // never written for an absolute sleep
     }
-        release(&tickslock);
+    int e = timespec_ticks(&ts, &want);
-    }
+    if (e)
-    return 0;
+        return e;
+    return sleep_ticks(want, remaddr);
 }