um: Switch clocksource to hrtimers

UML is using an obsolete itimer call for
all timers and "polls" for kernel space timer firing
in its userspace portion resulting in a long list
of bugs and incorrect behaviour(s). It also uses
ITIMER_VIRTUAL for its timer which results in the
timer being dependent on it running and the cpu
load.

This patch fixes this by moving to posix high resolution
timers firing off CLOCK_MONOTONIC and relaying the timer
correctly to the UML userspace.

Fixes:
 - crashes when hosts suspends/resumes
 - broken userspace timers - effecive ~40Hz instead
   of what they should be. Note - this modifies skas behavior
   by no longer setting an itimer per clone(). Timer events
   are relayed instead.
 - kernel network packet scheduling disciplines
 - tcp behaviour especially under load
 - various timer related corner cases

Finally, overall responsiveness of userspace is better.

Signed-off-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Anton Ivanov <aivanov@brocade.com>
[rw: massaged commit message]
Signed-off-by: Richard Weinberger <richard@nod.at>

6 files changed, 168 insertions, 181 deletions

diff --git a/arch/um/os-Linux/internal.h b/arch/um/os-Linux/internal.h
deleted file mode 100644
index 0dc2c9f135f6..000000000000
--- a/arch/um/os-Linux/internal.h
+++ /dev/null
@@ -1 +0,0 @@
-void alarm_handler(int sig, struct siginfo *unused_si, mcontext_t *mc);
diff --git a/arch/um/os-Linux/main.c b/arch/um/os-Linux/main.c
index df9191acd926..9d499de87e63 100644
--- a/arch/um/os-Linux/main.c
+++ b/arch/um/os-Linux/main.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
  */
@@ -163,13 +164,13 @@ int __init main(int argc, char **argv, char **envp)
 
 	/*
 	 * This signal stuff used to be in the reboot case.  However,
-	 * sometimes a SIGVTALRM can come in when we're halting (reproducably
+	 * sometimes a timer signal can come in when we're halting (reproducably
 	 * when writing out gcov information, presumably because that takes
 	 * some time) and cause a segfault.
 	 */
 
-	/* stop timers and set SIGVTALRM to be ignored */
-	disable_timer();
+	/* stop timers and set timer signal to be ignored */
+	os_timer_disable();
 
 	/* disable SIGIO for the fds and set SIGIO to be ignored */
 	err = deactivate_all_fds();
diff --git a/arch/um/os-Linux/process.c b/arch/um/os-Linux/process.c
index 8408aba915b2..b3e0d40932e1 100644
--- a/arch/um/os-Linux/process.c
+++ b/arch/um/os-Linux/process.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
  * Copyright (C) 2002 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
  */
@@ -89,6 +90,11 @@ int os_process_parent(int pid)
 	return parent;
 }
 
+void os_alarm_process(int pid)
+{
+	kill(pid, SIGALRM);
+}
+
 void os_stop_process(int pid)
 {
 	kill(pid, SIGSTOP);
diff --git a/arch/um/os-Linux/signal.c b/arch/um/os-Linux/signal.c
index 036d0dbc7b52..c211153ca69a 100644
--- a/arch/um/os-Linux/signal.c
+++ b/arch/um/os-Linux/signal.c
@@ -1,4 +1,6 @@
 /*
+ * Copyright (C) 2015 Anton Ivanov (aivanov@{brocade.com,kot-begemot.co.uk})
+ * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
  * Copyright (C) 2004 PathScale, Inc
  * Copyright (C) 2004 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
@@ -13,7 +15,6 @@
 #include <kern_util.h>
 #include <os.h>
 #include <sysdep/mcontext.h>
-#include "internal.h"
 
 void (*sig_info[NSIG])(int, struct siginfo *, struct uml_pt_regs *) = {
 	[SIGTRAP]	= relay_signal,
@@ -23,7 +24,8 @@ void (*sig_info[NSIG])(int, struct siginfo *, struct uml_pt_regs *) = {
 	[SIGBUS]	= bus_handler,
 	[SIGSEGV]	= segv_handler,
 	[SIGIO]		= sigio_handler,
-	[SIGVTALRM]	= timer_handler };
+	[SIGALRM]	= timer_handler
+};
 
 static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
 {
@@ -38,7 +40,7 @@ static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
 	}
 
 	/* enable signals if sig isn't IRQ signal */
-	if ((sig != SIGIO) && (sig != SIGWINCH) && (sig != SIGVTALRM))
+	if ((sig != SIGIO) && (sig != SIGWINCH) && (sig != SIGALRM))
 		unblock_signals();
 
 	(*sig_info[sig])(sig, si, &r);
@@ -55,8 +57,8 @@ static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
 #define SIGIO_BIT 0
 #define SIGIO_MASK (1 << SIGIO_BIT)
 
-#define SIGVTALRM_BIT 1
-#define SIGVTALRM_MASK (1 << SIGVTALRM_BIT)
+#define SIGALRM_BIT 1
+#define SIGALRM_MASK (1 << SIGALRM_BIT)
 
 static int signals_enabled;
 static unsigned int signals_pending;
@@ -78,36 +80,38 @@ void sig_handler(int sig, struct siginfo *si, mcontext_t *mc)
 	set_signals(enabled);
 }
 
-static void real_alarm_handler(mcontext_t *mc)
+static void timer_real_alarm_handler(mcontext_t *mc)
 {
 	struct uml_pt_regs regs;
 
 	if (mc != NULL)
 		get_regs_from_mc(&regs, mc);
-	regs.is_user = 0;
-	unblock_signals();
-	timer_handler(SIGVTALRM, NULL, &regs);
+	timer_handler(SIGALRM, NULL, &regs);
 }
 
-void alarm_handler(int sig, struct siginfo *unused_si, mcontext_t *mc)
+void timer_alarm_handler(int sig, struct siginfo *unused_si, mcontext_t *mc)
 {
 	int enabled;
 
 	enabled = signals_enabled;
 	if (!signals_enabled) {
-		signals_pending |= SIGVTALRM_MASK;
+		signals_pending |= SIGALRM_MASK;
 		return;
 	}
 
 	block_signals();
 
-	real_alarm_handler(mc);
+	timer_real_alarm_handler(mc);
 	set_signals(enabled);
 }
 
-void timer_init(void)
+void deliver_alarm(void) {
+    timer_alarm_handler(SIGALRM, NULL, NULL);
+}
+
+void timer_set_signal_handler(void)
 {
-	set_handler(SIGVTALRM);
+	set_handler(SIGALRM);
 }
 
 void set_sigstack(void *sig_stack, int size)
@@ -131,10 +135,9 @@ static void (*handlers[_NSIG])(int sig, struct siginfo *si, mcontext_t *mc) = {
 
 	[SIGIO] = sig_handler,
 	[SIGWINCH] = sig_handler,
-	[SIGVTALRM] = alarm_handler
+	[SIGALRM] = timer_alarm_handler
 };
 
-
 static void hard_handler(int sig, siginfo_t *si, void *p)
 {
 	struct ucontext *uc = p;
@@ -188,9 +191,9 @@ void set_handler(int sig)
 
 	/* block irq ones */
 	sigemptyset(&action.sa_mask);
-	sigaddset(&action.sa_mask, SIGVTALRM);
 	sigaddset(&action.sa_mask, SIGIO);
 	sigaddset(&action.sa_mask, SIGWINCH);
+	sigaddset(&action.sa_mask, SIGALRM);
 
 	if (sig == SIGSEGV)
 		flags |= SA_NODEFER;
@@ -283,8 +286,8 @@ void unblock_signals(void)
 		if (save_pending & SIGIO_MASK)
 			sig_handler_common(SIGIO, NULL, NULL);
 
-		if (save_pending & SIGVTALRM_MASK)
-			real_alarm_handler(NULL);
+		if (save_pending & SIGALRM_MASK)
+			timer_real_alarm_handler(NULL);
 	}
 }
 
diff --git a/arch/um/os-Linux/skas/process.c b/arch/um/os-Linux/skas/process.c
index d38f4952d748..b856c66ebd3a 100644
--- a/arch/um/os-Linux/skas/process.c
+++ b/arch/um/os-Linux/skas/process.c
@@ -1,4 +1,5 @@
 /*
+ * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
  */
@@ -45,7 +46,7 @@ static int ptrace_dump_regs(int pid)
  * Signals that are OK to receive in the stub - we'll just continue it.
  * SIGWINCH will happen when UML is inside a detached screen.
  */
-#define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
+#define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
 
 /* Signals that the stub will finish with - anything else is an error */
 #define STUB_DONE_MASK (1 << SIGTRAP)
@@ -183,19 +184,13 @@ extern char __syscall_stub_start[];
 static int userspace_tramp(void *stack)
 {
 	void *addr;
-	int err, fd;
+	int fd;
 	unsigned long long offset;
 
 	ptrace(PTRACE_TRACEME, 0, 0, 0);
 
 	signal(SIGTERM, SIG_DFL);
 	signal(SIGWINCH, SIG_IGN);
-	err = set_interval();
-	if (err) {
-		printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
-		       "errno = %d\n", err);
-		exit(1);
-	}
 
 	/*
 	 * This has a pte, but it can't be mapped in with the usual
@@ -286,7 +281,7 @@ int start_userspace(unsigned long stub_stack)
 			       "errno = %d\n", errno);
 			goto out_kill;
 		}
-	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
+	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
 
 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
 		err = -EINVAL;
@@ -319,8 +314,6 @@ int start_userspace(unsigned long stub_stack)
 
 void userspace(struct uml_pt_regs *regs)
 {
-	struct itimerval timer;
-	unsigned long long nsecs, now;
 	int err, status, op, pid = userspace_pid[0];
 	/* To prevent races if using_sysemu changes under us.*/
 	int local_using_sysemu;
@@ -329,13 +322,8 @@ void userspace(struct uml_pt_regs *regs)
 	/* Handle any immediate reschedules or signals */
 	interrupt_end();
 
-	if (getitimer(ITIMER_VIRTUAL, &timer))
-		printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno);
-	nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
-		timer.it_value.tv_usec * UM_NSEC_PER_USEC;
-	nsecs += os_nsecs();
-
 	while (1) {
+
 		/*
 		 * This can legitimately fail if the process loads a
 		 * bogus value into a segment register.  It will
@@ -405,18 +393,7 @@ void userspace(struct uml_pt_regs *regs)
 			case SIGTRAP:
 				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
 				break;
-			case SIGVTALRM:
-				now = os_nsecs();
-				if (now < nsecs)
-					break;
-				block_signals();
-				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
-				unblock_signals();
-				nsecs = timer.it_value.tv_sec *
-					UM_NSEC_PER_SEC +
-					timer.it_value.tv_usec *
-					UM_NSEC_PER_USEC;
-				nsecs += os_nsecs();
+			case SIGALRM:
 				break;
 			case SIGIO:
 			case SIGILL:
@@ -464,7 +441,6 @@ __initcall(init_thread_regs);
 
 int copy_context_skas0(unsigned long new_stack, int pid)
 {
-	struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
 	int err;
 	unsigned long current_stack = current_stub_stack();
 	struct stub_data *data = (struct stub_data *) current_stack;
@@ -476,11 +452,10 @@ int copy_context_skas0(unsigned long new_stack, int pid)
 	 * prepare offset and fd of child's stack as argument for parent's
 	 * and child's mmap2 calls
 	 */
-	*data = ((struct stub_data) { .offset	= MMAP_OFFSET(new_offset),
-				      .fd	= new_fd,
-				      .timer    = ((struct itimerval)
-					           { .it_value = tv,
-						     .it_interval = tv }) });
+	*data = ((struct stub_data) {
+			.offset	= MMAP_OFFSET(new_offset),
+			.fd     = new_fd
+	});
 
 	err = ptrace_setregs(pid, thread_regs);
 	if (err < 0) {
diff --git a/arch/um/os-Linux/time.c b/arch/um/os-Linux/time.c
index e9824d5dd7d5..0e39b9978729 100644
--- a/arch/um/os-Linux/time.c
+++ b/arch/um/os-Linux/time.c
@@ -1,4 +1,7 @@
 /*
+ * Copyright (C) 2015 Anton Ivanov (aivanov@{brocade.com,kot-begemot.co.uk})
+ * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
+ * Copyright (C) 2012-2014 Cisco Systems
  * Copyright (C) 2000 - 2007 Jeff Dike (jdike{addtoit,linux.intel}.com)
  * Licensed under the GPL
  */
@@ -10,177 +13,177 @@
 #include <sys/time.h>
 #include <kern_util.h>
 #include <os.h>
-#include "internal.h"
+#include <string.h>
+#include <timer-internal.h>
 
-int set_interval(void)
-{
-	int usec = UM_USEC_PER_SEC / UM_HZ;
-	struct itimerval interval = ((struct itimerval) { { 0, usec },
-							  { 0, usec } });
-
-	if (setitimer(ITIMER_VIRTUAL, &interval, NULL) == -1)
-		return -errno;
+static timer_t event_high_res_timer = 0;
 
-	return 0;
+static inline long long timeval_to_ns(const struct timeval *tv)
+{
+	return ((long long) tv->tv_sec * UM_NSEC_PER_SEC) +
+		tv->tv_usec * UM_NSEC_PER_USEC;
 }
 
-int timer_one_shot(int ticks)
+static inline long long timespec_to_ns(const struct timespec *ts)
 {
-	unsigned long usec = ticks * UM_USEC_PER_SEC / UM_HZ;
-	unsigned long sec = usec / UM_USEC_PER_SEC;
-	struct itimerval interval;
-
-	usec %= UM_USEC_PER_SEC;
-	interval = ((struct itimerval) { { 0, 0 }, { sec, usec } });
+	return ((long long) ts->tv_sec * UM_NSEC_PER_SEC) +
+		ts->tv_nsec;
+}
 
-	if (setitimer(ITIMER_VIRTUAL, &interval, NULL) == -1)
-		return -errno;
+long long os_persistent_clock_emulation (void) {
+	struct timespec realtime_tp;
 
-	return 0;
+	clock_gettime(CLOCK_REALTIME, &realtime_tp);
+	return timespec_to_ns(&realtime_tp);
 }
 
 /**
- * timeval_to_ns - Convert timeval to nanoseconds
- * @ts:		pointer to the timeval variable to be converted
- *
- * Returns the scalar nanosecond representation of the timeval
- * parameter.
- *
- * Ripped from linux/time.h because it's a kernel header, and thus
- * unusable from here.
+ * os_timer_create() - create an new posix (interval) timer
  */
-static inline long long timeval_to_ns(const struct timeval *tv)
-{
-	return ((long long) tv->tv_sec * UM_NSEC_PER_SEC) +
-		tv->tv_usec * UM_NSEC_PER_USEC;
+int os_timer_create(void* timer) {
+
+	timer_t* t = timer;
+
+	if(t == NULL) {
+		t = &event_high_res_timer;
+	}
+
+	if (timer_create(
+		CLOCK_MONOTONIC,
+		NULL,
+		t) == -1) {
+		return -1;
+	}
+	return 0;
 }
 
-long long disable_timer(void)
+int os_timer_set_interval(void* timer, void* i)
 {
-	struct itimerval time = ((struct itimerval) { { 0, 0 }, { 0, 0 } });
-	long long remain, max = UM_NSEC_PER_SEC / UM_HZ;
+	struct itimerspec its;
+	unsigned long long nsec;
+	timer_t* t = timer;
+	struct itimerspec* its_in = i;
 
-	if (setitimer(ITIMER_VIRTUAL, &time, &time) < 0)
-		printk(UM_KERN_ERR "disable_timer - setitimer failed, "
-		       "errno = %d\n", errno);
+	if(t == NULL) {
+		t = &event_high_res_timer;
+	}
 
-	remain = timeval_to_ns(&time.it_value);
-	if (remain > max)
-		remain = max;
+	nsec = UM_NSEC_PER_SEC / UM_HZ;
 
-	return remain;
-}
+	if(its_in != NULL) {
+		its.it_value.tv_sec = its_in->it_value.tv_sec;
+		its.it_value.tv_nsec = its_in->it_value.tv_nsec;
+	} else {
+		its.it_value.tv_sec = 0;
+		its.it_value.tv_nsec = nsec;
+	}
 
-long long os_nsecs(void)
-{
-	struct timeval tv;
+	its.it_interval.tv_sec = 0;
+	its.it_interval.tv_nsec = nsec;
 
-	gettimeofday(&tv, NULL);
-	return timeval_to_ns(&tv);
-}
+	if(timer_settime(*t, 0, &its, NULL) == -1) {
+		return -errno;
+	}
 
-#ifdef UML_CONFIG_NO_HZ_COMMON
-static int after_sleep_interval(struct timespec *ts)
-{
 	return 0;
 }
 
-static void deliver_alarm(void)
+/**
+ * os_timer_remain() - returns the remaining nano seconds of the given interval
+ *                     timer
+ * Because this is the remaining time of an interval timer, which correspondends
+ * to HZ, this value can never be bigger than one second. Just
+ * the nanosecond part of the timer is returned.
+ * The returned time is relative to the start time of the interval timer.
+ * Return an negative value in an error case.
+ */
+long os_timer_remain(void* timer)
 {
-	alarm_handler(SIGVTALRM, NULL, NULL);
-}
+	struct itimerspec its;
+	timer_t* t = timer;
 
-static unsigned long long sleep_time(unsigned long long nsecs)
-{
-	return nsecs;
-}
+	if(t == NULL) {
+		t = &event_high_res_timer;
+	}
 
-#else
-unsigned long long last_tick;
-unsigned long long skew;
+	if(timer_gettime(t, &its) == -1) {
+		return -errno;
+	}
 
-static void deliver_alarm(void)
-{
-	unsigned long long this_tick = os_nsecs();
-	int one_tick = UM_NSEC_PER_SEC / UM_HZ;
+	return its.it_value.tv_nsec;
+}
 
-	/* Protection against the host's time going backwards */
-	if ((last_tick != 0) && (this_tick < last_tick))
-		this_tick = last_tick;
+int os_timer_one_shot(int ticks)
+{
+	struct itimerspec its;
+	unsigned long long nsec;
+	unsigned long sec;
 
-	if (last_tick == 0)
-		last_tick = this_tick - one_tick;
+    nsec = (ticks + 1);
+    sec = nsec / UM_NSEC_PER_SEC;
+	nsec = nsec % UM_NSEC_PER_SEC;
 
-	skew += this_tick - last_tick;
+	its.it_value.tv_sec = nsec / UM_NSEC_PER_SEC;
+	its.it_value.tv_nsec = nsec;
 
-	while (skew >= one_tick) {
-		alarm_handler(SIGVTALRM, NULL, NULL);
-		skew -= one_tick;
-	}
+	its.it_interval.tv_sec = 0;
+	its.it_interval.tv_nsec = 0; // we cheat here
 
-	last_tick = this_tick;
+	timer_settime(event_high_res_timer, 0, &its, NULL);
+	return 0;
 }
 
-static unsigned long long sleep_time(unsigned long long nsecs)
+/**
+ * os_timer_disable() - disable the posix (interval) timer
+ * Returns the remaining interval timer time in nanoseconds
+ */
+long long os_timer_disable(void)
 {
-	return nsecs > skew ? nsecs - skew : 0;
-}
+	struct itimerspec its;
 
-static inline long long timespec_to_us(const struct timespec *ts)
-{
-	return ((long long) ts->tv_sec * UM_USEC_PER_SEC) +
-		ts->tv_nsec / UM_NSEC_PER_USEC;
+	memset(&its, 0, sizeof(struct itimerspec));
+	timer_settime(event_high_res_timer, 0, &its, &its);
+
+	return its.it_value.tv_sec * UM_NSEC_PER_SEC + its.it_value.tv_nsec;
 }
 
-static int after_sleep_interval(struct timespec *ts)
+long long os_vnsecs(void)
 {
-	int usec = UM_USEC_PER_SEC / UM_HZ;
-	long long start_usecs = timespec_to_us(ts);
-	struct timeval tv;
-	struct itimerval interval;
-
-	/*
-	 * It seems that rounding can increase the value returned from
-	 * setitimer to larger than the one passed in.  Over time,
-	 * this will cause the remaining time to be greater than the
-	 * tick interval.  If this happens, then just reduce the first
-	 * tick to the interval value.
-	 */
-	if (start_usecs > usec)
-		start_usecs = usec;
-
-	start_usecs -= skew / UM_NSEC_PER_USEC;
-	if (start_usecs < 0)
-		start_usecs = 0;
+	struct timespec ts;
 
-	tv = ((struct timeval) { .tv_sec  = start_usecs / UM_USEC_PER_SEC,
-				 .tv_usec = start_usecs % UM_USEC_PER_SEC });
-	interval = ((struct itimerval) { { 0, usec }, tv });
+	clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&ts);
+	return timespec_to_ns(&ts);
+}
 
-	if (setitimer(ITIMER_VIRTUAL, &interval, NULL) == -1)
-		return -errno;
+long long os_nsecs(void)
+{
+	struct timespec ts;
 
-	return 0;
+	clock_gettime(CLOCK_MONOTONIC,&ts);
+	return timespec_to_ns(&ts);
 }
-#endif
 
-void idle_sleep(unsigned long long nsecs)
+/**
+ * os_idle_sleep() - sleep for a given time of nsecs
+ * @nsecs: nanoseconds to sleep
+ */
+void os_idle_sleep(unsigned long long nsecs)
 {
 	struct timespec ts;
 
-	/*
-	 * nsecs can come in as zero, in which case, this starts a
-	 * busy loop.  To prevent this, reset nsecs to the tick
-	 * interval if it is zero.
-	 */
-	if (nsecs == 0)
-		nsecs = UM_NSEC_PER_SEC / UM_HZ;
+	if (nsecs <= 0) {
+		return;
+	}
 
-	nsecs = sleep_time(nsecs);
-	ts = ((struct timespec) { .tv_sec	= nsecs / UM_NSEC_PER_SEC,
-				  .tv_nsec	= nsecs % UM_NSEC_PER_SEC });
+	ts = ((struct timespec) {
+			.tv_sec  = nsecs / UM_NSEC_PER_SEC,
+			.tv_nsec = nsecs % UM_NSEC_PER_SEC
+	});
 
-	if (nanosleep(&ts, &ts) == 0)
+	/*
+	 * Relay the signal if clock_nanosleep is interrupted.
+	 */
+	if (clock_nanosleep(CLOCK_MONOTONIC, 0, &ts, NULL)) {
 		deliver_alarm();
-	after_sleep_interval(&ts);
+	}
 }