Merge branch 'profiling'

18 files changed, 1097 insertions, 4 deletions

diff --git a/Makefile b/Makefile
index d04b4d4..a12f804 100644
--- a/Makefile
+++ b/Makefile
@@ -27,6 +27,16 @@
 # NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 # SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
+# A couple features that can be enabled at build time, but are not turned on
+# by default:
+# DO_PROFILING: Enable gmon profiling. See libraries/libprof/README.md for
+# more details.
+# DO_TASK_METRICS: Enable task metrics - average/max time between yields. This
+# can be helpful when experimentally adding yields to improve responsiveness.
+#
+# To enable, run `make DO_PROFILING=1 DO_TASK_METRICS=1`
+# (or DO_PROFILING=xyzzy - `make` just cares that the symbol is defined)
+
 # export all variables to child processes by default
 .EXPORT_ALL_VARIABLES:
 
@@ -51,6 +61,9 @@ LIBCLI_BLD = $(LIBS_DIR)/libcli
 LIBTFM_SRC = $(CRYPTECH_ROOT)/sw/thirdparty/libtfm
 LIBTFM_BLD = $(LIBS_DIR)/libtfm
 
+LIBPROF_SRC = $(LIBS_DIR)/libprof
+LIBPROF_BLD = $(LIBS_DIR)/libprof
+
 LIBS = $(MBED_DIR)/libstmf4.a
 
 # linker script
@@ -114,6 +127,9 @@ CFLAGS += -I$(MBED_DIR)/targets/cmsis/TARGET_STM/TARGET_STM32F4
 CFLAGS += -I$(MBED_DIR)/targets/cmsis/TARGET_STM/TARGET_STM32F4/$(BOARD)
 CFLAGS += -I$(MBED_DIR)/targets/hal/TARGET_STM/TARGET_STM32F4
 CFLAGS += -I$(MBED_DIR)/targets/hal/TARGET_STM/TARGET_STM32F4/$(BOARD)
+ifdef DO_TASK_METRICS
+CFLAGS += -DDO_TASK_METRICS
+endif
 
 %.o : %.c
 	$(CC) $(CFLAGS) -c -o $@ $<
@@ -121,7 +137,13 @@ CFLAGS += -I$(MBED_DIR)/targets/hal/TARGET_STM/TARGET_STM32F4/$(BOARD)
 %.o : %.S
 	$(CC) $(CFLAGS) -c -o $@ $<
 
+ifdef DO_PROFILING
+CFLAGS += -pg -DDO_PROFILING
+LIBS += $(LIBPROF_BLD)/libprof.a
+all: hsm
+else
 all: board-test cli-test libhal-test hsm bootloader
+endif
 
 $(MBED_DIR)/libstmf4.a: .FORCE
 	$(MAKE) -C $(MBED_DIR)
@@ -141,6 +163,9 @@ $(LIBHAL_BLD)/libhal.a: $(LIBTFM_BLD)/libtfm.a .FORCE
 $(LIBCLI_BLD)/libcli.a: .FORCE
 	$(MAKE) -C $(LIBCLI_BLD)
 
+$(LIBPROF_BLD)/libprof.a: .FORCE
+	$(MAKE) -C $(LIBPROF_BLD)
+
 libhal-test: $(BOARD_OBJS) $(LIBS) $(LIBHAL_BLD)/libhal.a .FORCE
 	$(MAKE) -C projects/libhal-test
 
@@ -177,3 +202,4 @@ distclean: clean
 	$(MAKE) -C $(MBED_DIR) clean
 	$(MAKE) -C $(LIBTFM_BLD) clean
 	$(MAKE) -C $(LIBCLI_BLD) clean
+	$(MAKE) -C $(LIBPROF_BLD) clean
diff --git a/libraries/libprof/Makefile b/libraries/libprof/Makefile
new file mode 100644
index 0000000..4fe5fb4
--- /dev/null
+++ b/libraries/libprof/Makefile
@@ -0,0 +1,20 @@
+LIB = libprof.a
+
+OBJS = gmon.o profil.o profiler.o
+
+# Don't profile the profiling code, because that way lies madness (and recursion).
+CFLAGS := $(subst -pg,,$(CFLAGS))
+
+all: $(LIB)
+
+%.o : %.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+
+%.o : %.S
+	$(CC) $(CFLAGS) -c -o $@ $<
+
+$(LIB): $(OBJS)
+	$(AR) -r $@ $^
+
+clean:
+	rm -f $(OBJS) $(LIB)
diff --git a/libraries/libprof/README.md b/libraries/libprof/README.md
new file mode 100644
index 0000000..d464644
--- /dev/null
+++ b/libraries/libprof/README.md
@@ -0,0 +1,65 @@
+Profiling the Cryptech Alpha
+============================
+
+Origin
+------
+
+This code was copied from https://github.com/ErichStyger/mcuoneclipse.git,
+directory `Examples/KDS/FRDM-K64F120M/FRDM-K64F_Profiling/Profiling`, commit
+9b7eedddd8b24968128582aedc63be95b61f782c, dated Mon Jan 9 16:56:17 2017 +0100.
+
+References
+----------
+
+I recommend reading both of these to understand how the profiling code works.
+
+1. [Tutorial: Using GNU Profiling (gprof) with ARM Cortex-M](https://mcuoneclipse.com/2015/08/23/tutorial-using-gnu-profiling-gprof-with-arm-cortex-m/)
+
+2. [Semihosting with ARM, GCC, and OpenOCD](http://bgamari.github.io/posts/2014-10-31-semihosting.html)
+
+How to build
+------------
+
+From the top level, run
+
+    $ make DO_PROFILING=1 hsm
+
+By default, all code is profiled, *except* the profiling code itself,
+because that would cause fatal recursion.
+
+How to run
+----------
+
+You need to start OpenOCD on the host, and enable semihosting, at least
+before you try to use it as a remote file system.
+
+I recommend executing the following in the `projects/hsm` directory, so that
+`gmon.out` ends up in the same directory as `hsm.elf`.
+
+Start the debugger:
+
+    $ ../../bin/debug hsm
+
+In another window, connect to OpenOCD:
+
+    $ telnet localhost 4444
+
+In the OpenOCD console, enable semihosting:
+
+    > arm semihosting enable
+    > exit
+
+Then connect to the Cryptech management console:
+
+    $ cryptech_console
+
+In the Cryptech console, type `profile start`, then start the unit test or
+whatever will be exercising the hsm. Afterwards, in the console, type
+`profile stop`.
+
+After invoking `profile stop`, it can take several minutes to write
+`gmon.out` over OpenOCD to the host.
+
+In the `projects/hsm` directory, run `gprof` to analyse the `gmon.out` file:
+
+    $ gprof hsm.elf >gprof.txt
diff --git a/libraries/libprof/gmon.c b/libraries/libprof/gmon.c
new file mode 100644
index 0000000..317a173
--- /dev/null
+++ b/libraries/libprof/gmon.c
@@ -0,0 +1,324 @@
+/*-
+ * Copyright (c) 1983, 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+/*
+ * This file is taken from Cygwin distribution. Please keep it in sync.
+ * The differences should be within __MINGW32__ guard.
+ */
+
+#include <fcntl.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <stdint.h>
+#include "gmon.h"
+#include "profil.h"
+#include <string.h>
+
+#define bzero(ptr,size) memset (ptr, 0, size);
+#define ERR(s) write(2, s, sizeof(s))
+
+struct gmonparam _gmonparam = { GMON_PROF_OFF, NULL, 0, NULL, 0, NULL, 0, 0L, 0, 0, 0};
+static char already_setup = 0; /* flag to indicate if we need to init */
+static int	s_scale;
+/* see profil(2) where this is described (incorrectly) */
+#define		SCALE_1_TO_1	0x10000L
+
+static void moncontrol(int mode);
+
+void monstartup (size_t lowpc, size_t highpc) {
+	register size_t o;
+	char *cp;
+	struct gmonparam *p = &_gmonparam;
+
+	if (already_setup) {
+            /* zero out cp as value will be added there */
+            bzero(p->tos, p->kcountsize + p->fromssize + p->tossize);
+            moncontrol(1); /* start */
+            return;
+        }
+        already_setup = 1;
+
+        /* enable semihosting, for eventual output */
+        extern void initialise_monitor_handles(void);
+        initialise_monitor_handles();
+
+	/*
+	 * round lowpc and highpc to multiples of the density we're using
+	 * so the rest of the scaling (here and in gprof) stays in ints.
+	 */
+	p->lowpc = ROUNDDOWN(lowpc, HISTFRACTION * sizeof(HISTCOUNTER));
+	p->highpc = ROUNDUP(highpc, HISTFRACTION * sizeof(HISTCOUNTER));
+        p->textsize = p->highpc - p->lowpc + 0x20;
+	p->kcountsize = p->textsize / HISTFRACTION;
+	p->fromssize = p->textsize / HASHFRACTION;
+	p->tolimit = p->textsize * ARCDENSITY / 100;
+	if (p->tolimit < MINARCS) {
+		p->tolimit = MINARCS;
+	} else if (p->tolimit > MAXARCS) {
+		p->tolimit = MAXARCS;
+	}
+	p->tossize = p->tolimit * sizeof(struct tostruct);
+
+        extern void *hal_allocate_static_memory(const size_t size);
+        cp = hal_allocate_static_memory(p->kcountsize + p->fromssize + p->tossize);
+	if (cp == NULL) {
+		ERR("monstartup: out of memory\n");
+		return;
+	}
+
+	/* zero out cp as value will be added there */
+	bzero(cp, p->kcountsize + p->fromssize + p->tossize);
+
+	p->tos = (struct tostruct *)cp;
+	cp += p->tossize;
+	p->kcount = (unsigned short *)cp;
+	cp += p->kcountsize;
+	p->froms = (unsigned short *)cp;
+
+	p->tos[0].link = 0;
+
+	o = p->highpc - p->lowpc;
+	if (p->kcountsize < o) {
+#ifndef notdef
+		s_scale = ((float)p->kcountsize / o ) * SCALE_1_TO_1;
+#else /* avoid floating point */
+		int quot = o / p->kcountsize;
+
+		if (quot >= 0x10000)
+			s_scale = 1;
+		else if (quot >= 0x100)
+			s_scale = 0x10000 / quot;
+		else if (o >= 0x800000)
+			s_scale = 0x1000000 / (o / (p->kcountsize >> 8));
+		else
+			s_scale = 0x1000000 / ((o << 8) / p->kcountsize);
+#endif
+	} else {
+		s_scale = SCALE_1_TO_1;
+	}
+	moncontrol(1); /* start */
+}
+
+void _mcleanup(void) {
+	static const char gmon_out[] = "gmon.out";
+	int fd;
+	int fromindex;
+	int endfrom;
+	size_t frompc;
+	int toindex;
+	struct rawarc rawarc;
+	struct gmonparam *p = &_gmonparam;
+	struct gmonhdr gmonhdr = {0}, *hdr;
+	const char *proffile;
+#ifdef DEBUG
+	int log, len;
+	char dbuf[200];
+#endif
+
+	if (p->state == GMON_PROF_ERROR) {
+		ERR("_mcleanup: tos overflow\n");
+	}
+	moncontrol(0); /* stop */
+	proffile = gmon_out;
+	fd = open(proffile , O_CREAT|O_TRUNC|O_WRONLY|O_BINARY, 0666);
+	if (fd < 0) {
+		perror( proffile );
+		return;
+	}
+#ifdef DEBUG
+	log = open("gmon.log", O_CREAT|O_TRUNC|O_WRONLY, 0664);
+	if (log < 0) {
+		perror("mcount: gmon.log");
+		return;
+	}
+	len = sprintf(dbuf, "[mcleanup1] kcount 0x%x ssiz %d\n",
+                      (unsigned int)p->kcount, p->kcountsize);
+	write(log, dbuf, len);
+#endif
+	hdr = (struct gmonhdr *)&gmonhdr;
+	hdr->lpc = p->lowpc;
+	hdr->hpc = p->highpc;
+	hdr->ncnt = p->kcountsize + sizeof(gmonhdr);
+	hdr->version = GMONVERSION;
+	hdr->profrate = PROF_HZ;
+	hdr->spare[0] = hdr->spare[1] = hdr->spare[2] = 0;
+	write(fd, (char *)hdr, sizeof *hdr);
+	write(fd, p->kcount, p->kcountsize);
+	endfrom = p->fromssize / sizeof(*p->froms);
+	for (fromindex = 0; fromindex < endfrom; fromindex++) {
+		if (p->froms[fromindex] == 0) {
+			continue;
+		}
+		frompc = p->lowpc;
+		frompc += fromindex * HASHFRACTION * sizeof(*p->froms);
+		for (toindex = p->froms[fromindex]; toindex != 0; toindex = p->tos[toindex].link) {
+#ifdef DEBUG
+			len = sprintf(dbuf,
+			"[mcleanup2] frompc 0x%x selfpc 0x%x count %ld\n" ,
+				frompc, p->tos[toindex].selfpc,
+				p->tos[toindex].count);
+			write(log, dbuf, len);
+#endif
+			rawarc.raw_frompc = frompc;
+			rawarc.raw_selfpc = p->tos[toindex].selfpc;
+			rawarc.raw_count = p->tos[toindex].count;
+			write(fd, &rawarc, sizeof rawarc);
+		}
+	}
+	close(fd);
+}
+
+/*
+ * Control profiling
+ *	profiling is what mcount checks to see if
+ *	all the data structures are ready.
+ */
+static void moncontrol(int mode) {
+	struct gmonparam *p = &_gmonparam;
+
+	if (mode) {
+		/* start */
+		profil((char *)p->kcount, p->kcountsize, p->lowpc, s_scale);
+		p->state = GMON_PROF_ON;
+	} else {
+		/* stop */
+		profil((char *)0, 0, 0, 0);
+		p->state = GMON_PROF_OFF;
+	}
+}
+
+void _mcount_internal(uint32_t *frompcindex, uint32_t *selfpc) {
+  register struct tostruct	*top;
+  register struct tostruct	*prevtop;
+  register long			toindex;
+  struct gmonparam *p = &_gmonparam;
+
+  /*
+   *	check that we are profiling
+   *	and that we aren't recursively invoked.
+   */
+  if (p->state!=GMON_PROF_ON) {
+    goto out;
+  }
+  p->state++;
+  /*
+   *	check that frompcindex is a reasonable pc value.
+   *	for example:	signal catchers get called from the stack,
+   *			not from text space.  too bad.
+   */
+  frompcindex = (uint32_t*)((long)frompcindex - (long)p->lowpc);
+  if ((unsigned long)frompcindex > p->textsize) {
+      goto done;
+  }
+  frompcindex = (uint32_t*)&p->froms[((long)frompcindex) / (HASHFRACTION * sizeof(*p->froms))];
+  toindex = *((unsigned short*)frompcindex); /* get froms[] value */
+  if (toindex == 0) {
+    /*
+    *	first time traversing this arc
+    */
+    toindex = ++p->tos[0].link; /* the link of tos[0] points to the last used record in the array */
+    if (toindex >= p->tolimit) { /* more tos[] entries than we can handle! */
+	    goto overflow;
+	  }
+    *((unsigned short*)frompcindex) = (unsigned short)toindex; /* store new 'to' value into froms[] */
+    top = &p->tos[toindex];
+    top->selfpc = (size_t)selfpc;
+    top->count = 1;
+    top->link = 0;
+    goto done;
+  }
+  top = &p->tos[toindex];
+  if (top->selfpc == (size_t)selfpc) {
+    /*
+     *	arc at front of chain; usual case.
+     */
+    top->count++;
+    goto done;
+  }
+  /*
+   *	have to go looking down chain for it.
+   *	top points to what we are looking at,
+   *	prevtop points to previous top.
+   *	we know it is not at the head of the chain.
+   */
+  for (; /* goto done */; ) {
+    if (top->link == 0) {
+      /*
+       *	top is end of the chain and none of the chain
+       *	had top->selfpc == selfpc.
+       *	so we allocate a new tostruct
+       *	and link it to the head of the chain.
+       */
+      toindex = ++p->tos[0].link;
+      if (toindex >= p->tolimit) {
+        goto overflow;
+      }
+      top = &p->tos[toindex];
+      top->selfpc = (size_t)selfpc;
+      top->count = 1;
+      top->link = *((unsigned short*)frompcindex);
+      *(unsigned short*)frompcindex = (unsigned short)toindex;
+      goto done;
+    }
+    /*
+     *	otherwise, check the next arc on the chain.
+     */
+    prevtop = top;
+    top = &p->tos[top->link];
+    if (top->selfpc == (size_t)selfpc) {
+      /*
+       *	there it is.
+       *	increment its count
+       *	move it to the head of the chain.
+       */
+      top->count++;
+      toindex = prevtop->link;
+      prevtop->link = top->link;
+      top->link = *((unsigned short*)frompcindex);
+      *((unsigned short*)frompcindex) = (unsigned short)toindex;
+      goto done;
+    }
+  }
+  done:
+    p->state--;
+    /* and fall through */
+  out:
+    return;		/* normal return restores saved registers */
+  overflow:
+    p->state++; /* halt further profiling */
+    #define	TOLIMIT	"mcount: tos overflow\n"
+    write (2, TOLIMIT, sizeof(TOLIMIT));
+  goto out;
+}
+
+void _monInit(void) {
+  _gmonparam.state = GMON_PROF_OFF;
+  already_setup = 0;
+}
diff --git a/libraries/libprof/gmon.h b/libraries/libprof/gmon.h
new file mode 100644
index 0000000..8b5ecf0
--- /dev/null
+++ b/libraries/libprof/gmon.h
@@ -0,0 +1,177 @@
+/*	$OpenBSD: gmon.h,v 1.3 1996/04/21 22:31:46 deraadt Exp $	*/
+/*	$NetBSD: gmon.h,v 1.5 1996/04/09 20:55:30 cgd Exp $	*/
+
+/*-
+ * Copyright (c) 1982, 1986, 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *	@(#)gmon.h	8.2 (Berkeley) 1/4/94
+ */
+
+/*
+ * This file is taken from Cygwin distribution. Please keep it in sync.
+ * The differences should be within __MINGW32__ guard.
+ */
+
+#ifndef _SYS_GMON_H_
+#define _SYS_GMON_H_
+
+#ifndef __P
+#define __P(x) x
+#endif
+
+/* On POSIX systems, profile.h is a KRB5 header.  To avoid collisions, just
+   pull in profile.h's content here.  The profile.h header won't be provided
+   by Mingw-w64 anymore at one point. */
+#if 0
+#include <profile.h>
+#else
+#ifndef _WIN64
+#define _MCOUNT_CALL __attribute__ ((regparm (2)))
+extern void _mcount(void);
+#else
+#define _MCOUNT_CALL
+extern void mcount(void);
+#endif
+#define _MCOUNT_DECL __attribute__((gnu_inline)) __inline__ \
+   void _MCOUNT_CALL _mcount_private
+#define MCOUNT
+#endif
+
+/*
+ * Structure prepended to gmon.out profiling data file.
+ */
+struct gmonhdr {
+	size_t	lpc;		/* base pc address of sample buffer */
+	size_t	hpc;		/* max pc address of sampled buffer */
+	int	ncnt;		    /* size of sample buffer (plus this header) */
+	int	version;	  /* version number */
+	int	profrate;	  /* profiling clock rate */
+	int	spare[3];	  /* reserved */
+};
+#define GMONVERSION	0x00051879
+
+/*
+ * histogram counters are unsigned shorts (according to the kernel).
+ */
+#define	HISTCOUNTER	unsigned short
+
+/*
+ * fraction of text space to allocate for histogram counters here, 1/2
+ */
+#define	HISTFRACTION	2
+
+/*
+ * Fraction of text space to allocate for from hash buckets.
+ * The value of HASHFRACTION is based on the minimum number of bytes
+ * of separation between two subroutine call points in the object code.
+ * Given MIN_SUBR_SEPARATION bytes of separation the value of
+ * HASHFRACTION is calculated as:
+ *
+ *	HASHFRACTION = MIN_SUBR_SEPARATION / (2 * sizeof(short) - 1);
+ *
+ * For example, on the VAX, the shortest two call sequence is:
+ *
+ *	calls	$0,(r0)
+ *	calls	$0,(r0)
+ *
+ * which is separated by only three bytes, thus HASHFRACTION is
+ * calculated as:
+ *
+ *	HASHFRACTION = 3 / (2 * 2 - 1) = 1
+ *
+ * Note that the division above rounds down, thus if MIN_SUBR_FRACTION
+ * is less than three, this algorithm will not work!
+ *
+ * In practice, however, call instructions are rarely at a minimal
+ * distance.  Hence, we will define HASHFRACTION to be 2 across all
+ * architectures.  This saves a reasonable amount of space for
+ * profiling data structures without (in practice) sacrificing
+ * any granularity.
+ */
+#define	HASHFRACTION	2
+
+/*
+ * percent of text space to allocate for tostructs with a minimum.
+ */
+#define ARCDENSITY	2 /* this is in percentage, relative to text size! */
+#define MINARCS		 50
+#define MAXARCS		 ((1 << (8 * sizeof(HISTCOUNTER))) - 2)
+
+struct tostruct {
+	size_t	selfpc; /* callee address/program counter. The caller address is in froms[] array which points to tos[] array */
+	long	count;    /* how many times it has been called */
+	unsigned short link;   /* link to next entry in hash table. For tos[0] this points to the last used entry */
+	unsigned short pad;    /* additional padding bytes, to have entries 4byte aligned */
+};
+
+/*
+ * a raw arc, with pointers to the calling site and
+ * the called site and a count.
+ */
+struct rawarc {
+	size_t	raw_frompc;
+	size_t	raw_selfpc;
+	long	raw_count;
+};
+
+/*
+ * general rounding functions.
+ */
+#define ROUNDDOWN(x,y)	(((x)/(y))*(y))
+#define ROUNDUP(x,y)	  ((((x)+(y)-1)/(y))*(y))
+
+/*
+ * The profiling data structures are housed in this structure.
+ */
+struct gmonparam {
+	int		state;
+	unsigned short	*kcount;    /* histogram PC sample array */
+	size_t		kcountsize; /* size of kcount[] array in bytes */
+	unsigned short	*froms;     /* array of hashed 'from' addresses. The 16bit value is an index into the tos[] array */
+	size_t		fromssize;  /* size of froms[] array in bytes */
+	struct tostruct	*tos; /* to struct, contains histogram counter */
+	size_t		tossize;    /* size of tos[] array in bytes */
+	long		tolimit;
+	size_t		lowpc;      /* low program counter of area */
+	size_t		highpc;     /* high program counter */
+	size_t		textsize;   /* code size */
+};
+extern struct gmonparam _gmonparam;
+
+/*
+ * Possible states of profiling.
+ */
+#define	GMON_PROF_ON	  0
+#define	GMON_PROF_BUSY	1
+#define	GMON_PROF_ERROR	2
+#define	GMON_PROF_OFF	  3
+
+void _mcleanup(void); /* routine to be called to write gmon.out file */
+void _monInit(void); /* initialization routine */
+
+#endif /* !_SYS_GMONH_ */
diff --git a/libraries/libprof/profil.c b/libraries/libprof/profil.c
new file mode 100644
index 0000000..b0d8d55
--- /dev/null
+++ b/libraries/libprof/profil.c
@@ -0,0 +1,96 @@
+/* profil.c -- win32 profil.c equivalent
+
+   Copyright 1998, 1999, 2000, 2001, 2002 Red Hat, Inc.
+
+   This file is part of Cygwin.
+
+   This software is a copyrighted work licensed under the terms of the
+   Cygwin license.  Please consult the file "CYGWIN_LICENSE" for
+   details. */
+
+/*
+ * This file is taken from Cygwin distribution, adapted to be used for bare embedded targets.
+ */
+#include <stdio.h>
+#include <sys/types.h>
+#include <errno.h>
+#include <math.h>
+#include "profil.h"
+#include <string.h>
+
+#include "stm32f4xx_hal.h"      /* __get_MSP */
+
+/* global profinfo for profil() call */
+static struct profinfo prof = {
+  PROFILE_NOT_INIT, 0, 0, 0, 0
+};
+
+/* sample the current program counter */
+void profil_callback(void) {
+  if (prof.state == PROFILE_ON) {
+    /* The interrupt mechanism pushes xPSR, PC, LR, R12, and R3-R0 onto the
+     * stack, so PC is the 6th word from the top at that point. However, the
+     * normal function entry code pushes registers as well, so the stack
+     * offset right now depends on the call tree that got us here.
+     */
+    size_t pc = (size_t)((uint32_t *)__get_MSP())[6 + 6];
+    if (pc >= prof.lowpc && pc < prof.highpc) {
+      size_t idx = PROFIDX (pc, prof.lowpc, prof.scale);
+      prof.counter[idx]++;
+    }
+  }
+}
+
+/* Stop profiling to the profiling buffer pointed to by p. */
+static int profile_off (struct profinfo *p) {
+  p->state = PROFILE_OFF;
+  return 0;
+}
+
+/* Create a timer thread and pass it a pointer P to the profiling buffer. */
+static int profile_on (struct profinfo *p) {
+  p->state = PROFILE_ON;
+  return 0; /* ok */
+}
+
+/*
+ * start or stop profiling
+ *
+ * profiling goes into the SAMPLES buffer of size SIZE (which is treated
+ * as an array of unsigned shorts of size size/2)
+ *
+ * each bin represents a range of pc addresses from OFFSET.  The number
+ * of pc addresses in a bin depends on SCALE.  (A scale of 65536 maps
+ * each bin to two addresses, A scale of 32768 maps each bin to 4 addresses,
+ * a scale of 1 maps each bin to 128k address).  Scale may be 1 - 65536,
+ * or zero to turn off profiling
+ */
+int profile_ctl (struct profinfo *p, char *samples, size_t size, size_t offset, unsigned int scale) {
+  size_t maxbin;
+
+  if (scale > 65536) {
+    errno = EINVAL;
+    return -1;
+  }
+  profile_off(p);
+  if (scale) {
+    memset(samples, 0, size);
+    memset(p, 0, sizeof *p);
+    maxbin = size >> 1;
+    prof.counter = (unsigned short*)samples;
+    prof.lowpc = offset;
+    prof.highpc = PROFADDR(maxbin, offset, scale);
+    prof.scale = scale;
+    return profile_on(p);
+  }
+  return 0;
+}
+
+/* Equivalent to unix profil()
+   Every SLEEPTIME interval, the user's program counter (PC) is examined:
+   offset is subtracted and the result is multiplied by scale.
+   The word pointed to by this address is incremented. */
+int profil (char *samples, size_t size, size_t offset, unsigned int scale) {
+  return profile_ctl (&prof, samples, size, offset, scale);
+}
+
diff --git a/libraries/libprof/profil.h b/libraries/libprof/profil.h
new file mode 100644
index 0000000..c72dc00
--- /dev/null
+++ b/libraries/libprof/profil.h
@@ -0,0 +1,60 @@
+/* profil.h: gprof profiling header file
+
+   Copyright 1998, 1999, 2000, 2001, 2002 Red Hat, Inc.
+
+This file is part of Cygwin.
+
+This software is a copyrighted work licensed under the terms of the
+Cygwin license.  Please consult the file "CYGWIN_LICENSE" for
+details. */
+
+/*
+ * This file is taken from Cygwin distribution. Please keep it in sync.
+ * The differences should be within __MINGW32__ guard.
+ */
+
+#ifndef __PROFIL_H__
+#define __PROFIL_H__
+
+/* profiling frequency.  (No larger than 1000) */
+#define PROF_HZ			1000
+
+/* convert an addr to an index */
+#define PROFIDX(pc, base, scale)	\
+  ({									\
+    size_t i = (pc - base) / 2;				\
+    if (sizeof (unsigned long long int) > sizeof (size_t))		\
+      i = (unsigned long long int) i * scale / 65536;			\
+    else								\
+      i = i / 65536 * scale + i % 65536 * scale / 65536;		\
+    i;									\
+  })
+
+/* convert an index into an address */
+#define PROFADDR(idx, base, scale)		\
+  ((base)					\
+   + ((((unsigned long long)(idx) << 16)	\
+       / (unsigned long long)(scale)) << 1))
+
+/* convert a bin size into a scale */
+#define PROFSCALE(range, bins)		(((bins) << 16) / ((range) >> 1))
+
+typedef void *_WINHANDLE;
+
+typedef enum {
+  PROFILE_NOT_INIT = 0,
+  PROFILE_ON,
+  PROFILE_OFF
+} PROFILE_State;
+
+struct profinfo {
+  PROFILE_State state;		/* profiling state */
+  unsigned short *counter;	/* profiling counters */
+  size_t lowpc, highpc;		/* range to be profiled */
+  unsigned int scale;		/* scale value of bins */
+};
+
+int profile_ctl(struct profinfo *, char *, size_t, size_t, unsigned int);
+int profil(char *, size_t, size_t, unsigned int);
+
+#endif /* __PROFIL_H__ */
diff --git a/libraries/libprof/profiler.S b/libraries/libprof/profiler.S
new file mode 100644
index 0000000..1aa5c97
--- /dev/null
+++ b/libraries/libprof/profiler.S
@@ -0,0 +1,28 @@
+/*
+ * profiler.S
+ *  Implements the gprof profiler arc counting function.
+ *  Created on: 06.08.2015
+ *      Author: Erich Styger
+ */
+
+    .syntax unified
+    .arch armv7-m
+
+.globl __gnu_mcount_nc
+.type __gnu_mcount_nc, %function
+
+__gnu_mcount_nc:
+#if 0 /* dummy version, doing nothing */
+  mov    ip, lr
+  pop    { lr }
+  bx     ip
+#else
+  push {r0, r1, r2, r3, lr}     /* save registers */
+  bic r1, lr, #1                /* R1 contains callee address, with thumb bit cleared */
+  ldr r0, [sp, #20]             /* R0 contains caller address */
+  bic r0, r0, #1                /* clear thumb bit */
+  bl _mcount_internal           /* jump to internal _mcount() implementation */
+  pop {r0, r1, r2, r3, ip, lr}  /* restore saved registers */
+  bx ip                         /* return to caller */
+#endif
+
diff --git a/libraries/libtfm/Makefile b/libraries/libtfm/Makefile
index 57c76a5..b50421c 100644
--- a/libraries/libtfm/Makefile
+++ b/libraries/libtfm/Makefile
@@ -30,7 +30,12 @@ LIB  := tomsfastmath/libtfm.a
 # for now, at least until we confirm that it hasn't already been fixed
 # in a newer version of libtfm.
 
+ifdef DO_PROFILING
+# arm-none-eabi-gcc: error: -pg and -fomit-frame-pointer are incompatible
+STM32_LIBTFM_CFLAGS_OPTIMIZATION := -O3 -funroll-loops
+else
 STM32_LIBTFM_CFLAGS_OPTIMIZATION := -O3 -funroll-loops -fomit-frame-pointer
+endif
 
 CFLAGS := $(subst ${STM32_CFLAGS_OPTIMIZATION},${STM32_LIBTFM_CFLAGS_OPTIMIZATION},${CFLAGS})
 CFLAGS += -DTFM_ARM -Dasm=__asm__ -Wa,-mimplicit-it=thumb
diff --git a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/stm32f4xx_it.c b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/stm32f4xx_it.c
index d3bafb2..81b27cb 100644
--- a/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/stm32f4xx_it.c
+++ b/libraries/mbed/targets/cmsis/TARGET_STM/TARGET_STM32F4/TARGET_CRYPTECH_ALPHA/stm32f4xx_it.c
@@ -64,6 +64,13 @@ void HardFault_Handler(void)
     while (1) { ; }
 }
 
+static void default_SysTick_hook(void) { };
+static void (*SysTick_hook)(void) = default_SysTick_hook;
+void set_SysTick_hook(void (*hook)(void))
+{
+    SysTick_hook = (hook == NULL) ? default_SysTick_hook : hook;
+}
+
 /**
  * @brief  This function handles SysTick Handler.
  * @param  None
diff --git a/memfunc.c b/memfunc.c
new file mode 100644
index 0000000..fd94b28
--- /dev/null
+++ b/memfunc.c
@@ -0,0 +1,101 @@
+#include <stdint.h>
+#include <string.h>
+
+/*
+ * Profilable substitutes for mem*(), lacking libc_p.a
+ *
+ * This code was written with reference to newlib, but does not copy every
+ * quirk and loop-unrolling optimization from newlib. Its only purpose is
+ * to let us figure out who is calling memcpy 2 million times.
+ */
+
+#define is_word_aligned(x) (((size_t)(x) & 3) == 0)
+
+void *memcpy(void *dst, const void *src, size_t n)
+{
+    uint8_t *d8 = (uint8_t *)dst;
+    uint8_t *s8 = (uint8_t *)src;
+
+    if (n >= 4 && is_word_aligned(src) && is_word_aligned(dst)) {
+        uint32_t *d32 = (uint32_t *)dst;
+        uint32_t *s32 = (uint32_t *)src;
+        while (n >= 4) {
+            *d32++ = *s32++;
+            n -= 4;
+        }
+        d8 = (uint8_t *)d32;
+        s8 = (uint8_t *)s32;
+    }
+    while (n-- > 0) {
+        *d8++ = *s8++;
+    }
+
+    return dst;
+}
+
+void *memset(void *dst, int c, size_t n)
+{
+    uint8_t *d8 = (uint8_t *)dst;
+    uint8_t c8 = (uint8_t)c;
+
+    if (n >= 4 && is_word_aligned(dst)) {
+        uint32_t *d32 = (uint32_t *)dst;
+        uint32_t c32 = (c8 << 24) | (c8 << 16) | (c8 << 8) | (c8);
+        while (n >= 4) {
+            *d32++ = c32;
+            n -= 4;
+        }
+        d8 = (uint8_t *)d32;
+    }
+    while (n-- > 0) {
+        *d8++ = c8;
+    }
+
+    return dst;
+}
+
+int memcmp(const void *dst, const void *src, size_t n)
+{
+    uint8_t *d8 = (uint8_t *)dst;
+    uint8_t *s8 = (uint8_t *)src;
+
+    if (n >= 4 && is_word_aligned(src) && is_word_aligned(dst)) {
+        uint32_t *d32 = (uint32_t *)dst;
+        uint32_t *s32 = (uint32_t *)src;
+        while (n >= 4) {
+            if (*d32 != *s32)
+                break;
+            d32++;
+            s32++;
+            n -= 4;
+        }
+        d8 = (uint8_t *)d32;
+        s8 = (uint8_t *)s32;
+    }
+    while (n-- > 0) {
+        if (*d8 != *s8)
+            return (*d8 - *s8);
+        d8++;
+        s8++;
+    }
+
+    return 0;
+}
+
+void *memmove(void *dst, const void *src, size_t n)
+{
+    uint8_t *d8 = (uint8_t *)dst;
+    uint8_t *s8 = (uint8_t *)src;
+
+    if ((s8 < d8) && (d8 < s8 + n)) {
+        /* Destructive overlap...have to copy backwards */
+        s8 += n;
+        d8 += n;
+        while (n-- > 0) {
+            *--d8 = *--s8;
+	}
+        return dst;
+    }
+
+    return memcpy(dst, src, n);
+}
diff --git a/projects/hsm/Makefile b/projects/hsm/Makefile
index 927c9f1..3430e14 100644
--- a/projects/hsm/Makefile
+++ b/projects/hsm/Makefile
@@ -9,9 +9,8 @@ OBJS =	mgmt-cli.o \
 	mgmt-masterkey.o \
 	mgmt-misc.o \
 	mgmt-task.o \
-	log.o
-
-BOARD_OBJS += $(TOPLEVEL)/task.o
+	log.o \
+	$(TOPLEVEL)/task.o
 
 CFLAGS += -DNUM_RPC_TASK=4
 
@@ -21,10 +20,23 @@ CFLAGS += -I$(LIBCLI_SRC)
 LIBS += $(LIBHAL_BLD)/libhal.a $(LIBTFM_BLD)/libtfm.a
 LIBS += $(LIBCLI_BLD)/libcli.a
 
+LDFLAGS += -mcpu=cortex-m4 -mthumb -mlittle-endian -mthumb-interwork
+LDFLAGS += -mfloat-abi=hard -mfpu=fpv4-sp-d16
+LDFLAGS += -Wl,--gc-sections
+
+ifdef DO_PROFILING
+OBJS += $(TOPLEVEL)/memfunc.o
+LDFLAGS += --specs=rdimon.specs -lc -lrdimon
+endif
+
+ifdef DO_TASK_METRICS
+CFLAGS += -DDO_TASK_METRICS
+endif
+
 all: $(PROJ:=.elf)
 
 %.elf: %.o $(BOARD_OBJS) $(OBJS) $(LIBS)
-	$(CC) $(CFLAGS) $^ -o $@ -T$(LDSCRIPT) -g -Wl,-Map=$*.map
+	$(CC) $^ -o $@ -T$(LDSCRIPT) -g -Wl,-Map=$*.map $(LDFLAGS)
 	$(OBJCOPY) -O binary $*.elf $*.bin
 	$(SIZE) $*.elf
 
diff --git a/projects/hsm/hsm.c b/projects/hsm/hsm.c
index 7fc7410..f20ee64 100644
--- a/projects/hsm/hsm.c
+++ b/projects/hsm/hsm.c
@@ -268,6 +268,11 @@ size_t uart_rx_max = 0;
 
 void HAL_SYSTICK_Callback(void)
 {
+#ifdef DO_PROFILING
+    extern void profil_callback(void);
+    profil_callback();
+#endif
+
     size_t count = RINGBUF_COUNT(uart_ringbuf);
     if (uart_rx_max < count) uart_rx_max = count;
 
@@ -393,6 +398,11 @@ void hal_task_yield(void)
     task_yield();
 }
 
+void hal_task_yield_maybe(void)
+{
+    task_yield_maybe();
+}
+
 /* A mutex to arbitrate concurrent access to the keystore.
  */
 task_mutex_t ks_mutex = { 0 };
diff --git a/projects/hsm/mgmt-misc.c b/projects/hsm/mgmt-misc.c
index b06003f..86f1be8 100644
--- a/projects/hsm/mgmt-misc.c
+++ b/projects/hsm/mgmt-misc.c
@@ -113,6 +113,35 @@ int cli_receive_data(struct cli_def *cli, uint8_t *buf, size_t len, cli_data_cal
     return CLI_ERROR;
 }
 
+#ifdef DO_PROFILING
+static int cmd_profile_start(struct cli_def *cli, const char *command, char *argv[], int argc)
+{
+    cli = cli;
+    command = command;
+    argv = argv;
+    argc = argc;
+
+    extern uint32_t CRYPTECH_FIRMWARE_START;
+    extern char __etext; /* end of text/code symbol, defined by linker */
+    extern void monstartup (size_t lowpc, size_t highpc);
+    monstartup((size_t)&CRYPTECH_FIRMWARE_START, (size_t)&__etext);
+    return CLI_OK;
+}
+
+static int cmd_profile_stop(struct cli_def *cli, const char *command, char *argv[], int argc)
+{
+    cli = cli;
+    command = command;
+    argv = argv;
+    argc = argc;
+
+    extern void _mcleanup(void);
+    _mcleanup();
+    return CLI_OK;
+}
+
+#endif
+
 static int cmd_reboot(struct cli_def *cli, const char *command, char *argv[], int argc)
 {
     command = command;
@@ -128,6 +157,15 @@ static int cmd_reboot(struct cli_def *cli, const char *command, char *argv[], in
 
 void configure_cli_misc(struct cli_def *cli)
 {
+#ifdef DO_PROFILING
+    struct cli_command *c_profile = cli_register_command(cli, NULL, "profile", NULL, 0, 0, NULL);
+
+    /* profile start */
+    cli_register_command(cli, c_profile, "start", cmd_profile_start, 0, 0, "Start collecting profiling data");
+
+    /* profile stop */
+    cli_register_command(cli, c_profile, "stop", cmd_profile_stop, 0, 0, "Stop collecting profiling data");
+#endif    
     /* reboot */
     cli_register_command(cli, NULL, "reboot", cmd_reboot, 0, 0, "Reboot the STM32");
 }
diff --git a/projects/hsm/mgmt-task.c b/projects/hsm/mgmt-task.c
index beffc32..c2a3d3f 100644
--- a/projects/hsm/mgmt-task.c
+++ b/projects/hsm/mgmt-task.c
@@ -77,10 +77,54 @@ static int cmd_task_show(struct cli_def *cli, const char *command, char *argv[],
     return CLI_OK;
 }
 
+#ifdef DO_TASK_METRICS
+static int cmd_task_show_metrics(struct cli_def *cli, const char *command, char *argv[], int argc)
+{
+    command = command;
+    argv = argv;
+    argc = argc;
+
+    struct task_metrics tm;
+
+    task_get_metrics(&tm);
+
+    cli_print(cli, "avg time between yields: %ld.%06ld sec", tm.avg.tv_sec, tm.avg.tv_usec);
+    cli_print(cli, "max time between yields: %ld.%06ld sec", tm.max.tv_sec, tm.max.tv_usec);
+
+    return CLI_OK;
+}
+
+static int cmd_task_reset_metrics(struct cli_def *cli, const char *command, char *argv[], int argc)
+{
+    cli = cli;
+    command = command;
+    argv = argv;
+    argc = argc;
+
+    task_reset_metrics();
+
+    return CLI_OK;
+}
+#endif
+
 void configure_cli_task(struct cli_def *cli)
 {
     struct cli_command *c = cli_register_command(cli, NULL, "task", NULL, 0, 0, NULL);
 
     /* task show */
+#ifdef DO_TASK_METRICS
+    struct cli_command *c_show =
+#endif
     cli_register_command(cli, c, "show", cmd_task_show, 0, 0, "Show the active tasks");
+
+#ifdef DO_TASK_METRICS
+    /* task show metrics */
+    cli_register_command(cli, c_show, "metrics", cmd_task_show_metrics, 0, 0, "Show task metrics");
+
+    /* task reset */
+    struct cli_command *c_reset = cli_register_command(cli, c, "reset", NULL, 0, 0, NULL);
+
+    /* task reset metrics */
+    cli_register_command(cli, c_reset, "metrics", cmd_task_reset_metrics, 0, 0, "Reset task metrics");
+#endif
 }
diff --git a/syscalls.c b/syscalls.c
index 2243d0b..fc842c0 100644
--- a/syscalls.c
+++ b/syscalls.c
@@ -48,6 +48,7 @@
 
 /***************************************************************************/
 
+#ifndef DO_PROFILING
 int _read_r (struct _reent *r, int file, char * ptr, int len)
 {
   r = r;
@@ -106,6 +107,7 @@ int _close_r (struct _reent *r, int file)
 
   return 0;
 }
+#endif
 
 /***************************************************************************/
 
@@ -148,6 +150,7 @@ caddr_t _sbrk_r (struct _reent *r, int incr)
 
 /***************************************************************************/
 
+#ifndef DO_PROFILING
 int _fstat_r (struct _reent *r, int file, struct stat * st)
 {
   r = r; 
@@ -186,6 +189,7 @@ int _kill (int a, int b)
   
   return 0;
 }
+#endif
 
 /***************************************************************************/
 
@@ -198,6 +202,7 @@ int _getpid(int a)
 
 /***************************************************************************/
 
+#ifndef DO_PROFILING
 int _open(int a, int b)
 {
   a = a;
@@ -205,5 +210,6 @@ int _open(int a, int b)
   
   return 0;
 }
+#endif
 
 /*** EOF ***/
diff --git a/task.c b/task.c
index 2e2ddec..36e8580 100644
--- a/task.c
+++ b/task.c
@@ -81,6 +81,18 @@ static tcb_t *cur_task = NULL;
 
 #define STACK_GUARD_WORD 0x55AA5A5A
 
+#ifdef DO_TASK_METRICS
+static uint32_t tick_start = 0;
+static uint32_t tick_idle  = 0;
+static uint32_t tick_max   = 0;
+static uint32_t nyield     = 0;
+#endif
+
+static uint32_t tick_prev  = 0;
+#ifndef TASK_YIELD_THRESHOLD
+#define TASK_YIELD_THRESHOLD 100
+#endif
+
 /* Add a task.
  */
 tcb_t *task_add(char *name, funcp_t func, void *cookie, void *stack, size_t stack_len)
@@ -180,6 +192,10 @@ void task_yield(void)
     if (tail == NULL)
 	return;
 
+#ifdef DO_TASK_METRICS
+    uint32_t tick0 = HAL_GetTick();
+#endif
+
     /* Find the next runnable task. Loop if every task is waiting. */
     while (1) {
         next = next_task();
@@ -197,6 +213,22 @@ void task_yield(void)
      * } while (next == NULL);
      */
 
+#ifdef DO_TASK_METRICS
+    uint32_t tick = HAL_GetTick();
+    tick_idle += (tick - tick0);
+    if (tick_start == 0)
+        tick_start = tick;
+    if (tick_prev != 0) {
+        uint32_t duration = tick0 - tick_prev;
+        if (duration > tick_max)
+            tick_max = duration;
+    }
+    tick_prev = tick;
+    ++nyield;
+#else
+    tick_prev = HAL_GetTick();
+#endif
+
     /* If there are no other runnable tasks (and cur_task is runnable),
      * we don't need to context-switch.
      */
@@ -230,6 +262,14 @@ void task_yield(void)
     }
 }
 
+/* Yield if it's been "too long" since the last yield.
+ */
+void task_yield_maybe(void)
+{
+    if (HAL_GetTick() - tick_prev >= TASK_YIELD_THRESHOLD)
+        task_yield();
+}
+
 /* Put the current task to sleep (make it non-runnable).
  */
 void task_sleep(void)
@@ -354,3 +394,25 @@ void task_mutex_unlock(task_mutex_t *mutex)
     if (mutex != NULL)
 	mutex->locked = 0;
 }
+
+#ifdef DO_TASK_METRICS
+void task_get_metrics(struct task_metrics *tm)
+{
+    if (tm != NULL) {
+        tm->avg.tv_sec  = 0;
+        tm->avg.tv_usec = (HAL_GetTick() - tick_start - tick_idle) * 1000 / nyield;
+        if (tm->avg.tv_usec > 1000000) {
+            tm->avg.tv_sec  = tm->avg.tv_usec / 1000000;
+            tm->avg.tv_usec = tm->avg.tv_usec % 1000000;
+        }
+        tm->max.tv_sec  = tick_max / 1000;
+        tm->max.tv_usec = (tick_max % 1000) * 1000;
+    }
+}
+
+void task_reset_metrics(void)
+{
+    tick_start = HAL_GetTick();
+    tick_prev = tick_idle = tick_max = nyield = 0;
+}
+#endif
diff --git a/task.h b/task.h
index 6b45db8..47d62d6 100644
--- a/task.h
+++ b/task.h
@@ -55,6 +55,7 @@ extern tcb_t *task_add(char *name, funcp_t func, void *cookie, void *stack, size
 extern void task_set_idle_hook(funcp_t func);
 
 extern void task_yield(void);
+extern void task_yield_maybe(void);
 extern void task_sleep(void);
 extern void task_wake(tcb_t *t);
 
@@ -73,4 +74,15 @@ extern void task_delay(uint32_t delay);
 extern void task_mutex_lock(task_mutex_t *mutex);
 extern void task_mutex_unlock(task_mutex_t *mutex);
 
+#ifdef DO_TASK_METRICS
+#include <sys/time.h>
+
+struct task_metrics {
+    struct timeval avg, max;
+};
+
+void task_get_metrics(struct task_metrics *tm);
+void task_reset_metrics(void);
+#endif
+
 #endif /* _TASK_H_ */