mbedtls/tests/suites/test_suite_timing.function

361 lines
14 KiB
Plaintext
Raw Normal View History

/* BEGIN_HEADER */
New timing unit tests New set of unit tests for the timing module, instead of just running the selftest function. The selftest function sometimes fails on a heavily loaded machine (such as a typical continuous integration system). Because of the all-in-one nature of the test and because the exact load pattern can be hard to reproduce, it is difficult to diagnose failures of CI runs with selftest. The new tests are more separated and I strove to point out potential failure modes in comments. * mbedtls_timing_hardclock: not tested. This function gives so few guarantees that there isn't much to test, and it is hard to test reliably because clock cycles don't easily relate to time in any remotely portable way. This function isn't used in the library anyway, it's only there for benchmark programs. * mbedtls_timing_get_timer: tested by setting a timer and verifying that it reaches its target, and by verifying that a timer started later than another always has a smaller elapsed time. * mbedtls_set_alarm: tested by setting an alarm, busy-waiting for it and measuring the elapsed time with a timer. * mbedtls_timing_set_delay, mbedtls_timing_get_delay: tested by setting a delay object and watching it go through its two delay values, using a timer to check that the delays are passed at the expected time. The tests pass under light to moderate load, but some of them can be defeated with sufficiently heavy load. This is unavoidable since the test process to be effectively suspended for any length of time, making us think that a timer has gone on for too long.
2017-10-10 17:56:06 +00:00
/* This test module exercises the timing module. One of the expected failure
modes is for timers to never expire, which could lead to an infinite loop.
The function timing_timer_simple is protected against this failure mode and
checks that timers do expire. Other functions will terminate if their
timers do expire. Therefore it is recommended to run timing_timer_simple
first and run other test functions only if that timing_timer_simple
succeeded. */
#include <limits.h>
#include "mbedtls/timing.h"
New timing unit tests New set of unit tests for the timing module, instead of just running the selftest function. The selftest function sometimes fails on a heavily loaded machine (such as a typical continuous integration system). Because of the all-in-one nature of the test and because the exact load pattern can be hard to reproduce, it is difficult to diagnose failures of CI runs with selftest. The new tests are more separated and I strove to point out potential failure modes in comments. * mbedtls_timing_hardclock: not tested. This function gives so few guarantees that there isn't much to test, and it is hard to test reliably because clock cycles don't easily relate to time in any remotely portable way. This function isn't used in the library anyway, it's only there for benchmark programs. * mbedtls_timing_get_timer: tested by setting a timer and verifying that it reaches its target, and by verifying that a timer started later than another always has a smaller elapsed time. * mbedtls_set_alarm: tested by setting an alarm, busy-waiting for it and measuring the elapsed time with a timer. * mbedtls_timing_set_delay, mbedtls_timing_get_delay: tested by setting a delay object and watching it go through its two delay values, using a timer to check that the delays are passed at the expected time. The tests pass under light to moderate load, but some of them can be defeated with sufficiently heavy load. This is unavoidable since the test process to be effectively suspended for any length of time, making us think that a timer has gone on for too long.
2017-10-10 17:56:06 +00:00
/* Wait this many milliseconds for a short timing test. This duration
should be large enough that, in practice, if you read the timer
value twice in a row, it won't have jumped by that much. */
#define TIMING_SHORT_TEST_MS 100
/* A loop that waits TIMING_SHORT_TEST_MS must not take more than this many
iterations. This value needs to be large enough to accommodate fast
platforms (e.g. at 4GHz and 10 cycles/iteration a CPU can run through 20
million iterations in 50ms). The only motivation to keep this value low is
to avoid having an infinite loop if the timer functions are not implemented
correctly. Ideally this value should be based on the processor speed but we
don't have this information! */
#define TIMING_SHORT_TEST_ITERATIONS_MAX 1e8
/* alarm(0) must fire in no longer than this amount of time. */
#define TIMING_ALARM_0_DELAY_MS TIMING_SHORT_TEST_MS
static int expected_delay_status( uint32_t int_ms, uint32_t fin_ms,
unsigned long actual_ms )
{
return( fin_ms == 0 ? -1 :
actual_ms >= fin_ms ? 2 :
actual_ms >= int_ms ? 1 :
0 );
}
/* END_HEADER */
/* BEGIN_DEPENDENCIES
* depends_on:MBEDTLS_TIMING_C
* END_DEPENDENCIES
*/
New timing unit tests New set of unit tests for the timing module, instead of just running the selftest function. The selftest function sometimes fails on a heavily loaded machine (such as a typical continuous integration system). Because of the all-in-one nature of the test and because the exact load pattern can be hard to reproduce, it is difficult to diagnose failures of CI runs with selftest. The new tests are more separated and I strove to point out potential failure modes in comments. * mbedtls_timing_hardclock: not tested. This function gives so few guarantees that there isn't much to test, and it is hard to test reliably because clock cycles don't easily relate to time in any remotely portable way. This function isn't used in the library anyway, it's only there for benchmark programs. * mbedtls_timing_get_timer: tested by setting a timer and verifying that it reaches its target, and by verifying that a timer started later than another always has a smaller elapsed time. * mbedtls_set_alarm: tested by setting an alarm, busy-waiting for it and measuring the elapsed time with a timer. * mbedtls_timing_set_delay, mbedtls_timing_get_delay: tested by setting a delay object and watching it go through its two delay values, using a timer to check that the delays are passed at the expected time. The tests pass under light to moderate load, but some of them can be defeated with sufficiently heavy load. This is unavoidable since the test process to be effectively suspended for any length of time, making us think that a timer has gone on for too long.
2017-10-10 17:56:06 +00:00
/* BEGIN_CASE */
void timing_timer_simple( )
{
New timing unit tests New set of unit tests for the timing module, instead of just running the selftest function. The selftest function sometimes fails on a heavily loaded machine (such as a typical continuous integration system). Because of the all-in-one nature of the test and because the exact load pattern can be hard to reproduce, it is difficult to diagnose failures of CI runs with selftest. The new tests are more separated and I strove to point out potential failure modes in comments. * mbedtls_timing_hardclock: not tested. This function gives so few guarantees that there isn't much to test, and it is hard to test reliably because clock cycles don't easily relate to time in any remotely portable way. This function isn't used in the library anyway, it's only there for benchmark programs. * mbedtls_timing_get_timer: tested by setting a timer and verifying that it reaches its target, and by verifying that a timer started later than another always has a smaller elapsed time. * mbedtls_set_alarm: tested by setting an alarm, busy-waiting for it and measuring the elapsed time with a timer. * mbedtls_timing_set_delay, mbedtls_timing_get_delay: tested by setting a delay object and watching it go through its two delay values, using a timer to check that the delays are passed at the expected time. The tests pass under light to moderate load, but some of them can be defeated with sufficiently heavy load. This is unavoidable since the test process to be effectively suspended for any length of time, making us think that a timer has gone on for too long.
2017-10-10 17:56:06 +00:00
struct mbedtls_timing_hr_time timer;
unsigned long millis = 0;
unsigned long new_millis = 0;
unsigned long iterations = 0;
/* Start the timer. */
(void) mbedtls_timing_get_timer( &timer, 1 );
/* Busy-wait loop for a few milliseconds. */
do
{
new_millis = mbedtls_timing_get_timer( &timer, 0 );
++iterations;
/* Check that the timer didn't go backwards */
TEST_ASSERT( new_millis >= millis );
millis = new_millis;
}
while( millis < TIMING_SHORT_TEST_MS &&
iterations <= TIMING_SHORT_TEST_ITERATIONS_MAX );
/* The wait duration should have been large enough for at least a
few runs through the loop, even on the slowest realistic platform. */
TEST_ASSERT( iterations >= 2 );
/* The wait duration shouldn't have overflowed the iteration count. */
TEST_ASSERT( iterations < TIMING_SHORT_TEST_ITERATIONS_MAX );
return;
exit:
/* No cleanup needed, but show some diagnostic iterations, because timing
problems can be hard to reproduce. */
mbedtls_fprintf( stdout, " Finished with millis=%lu new_millis=%lu get(timer)<=%lu iterations=%lu\n",
millis, new_millis, mbedtls_timing_get_timer( &timer, 0 ),
iterations );
}
/* END_CASE */
New timing unit tests New set of unit tests for the timing module, instead of just running the selftest function. The selftest function sometimes fails on a heavily loaded machine (such as a typical continuous integration system). Because of the all-in-one nature of the test and because the exact load pattern can be hard to reproduce, it is difficult to diagnose failures of CI runs with selftest. The new tests are more separated and I strove to point out potential failure modes in comments. * mbedtls_timing_hardclock: not tested. This function gives so few guarantees that there isn't much to test, and it is hard to test reliably because clock cycles don't easily relate to time in any remotely portable way. This function isn't used in the library anyway, it's only there for benchmark programs. * mbedtls_timing_get_timer: tested by setting a timer and verifying that it reaches its target, and by verifying that a timer started later than another always has a smaller elapsed time. * mbedtls_set_alarm: tested by setting an alarm, busy-waiting for it and measuring the elapsed time with a timer. * mbedtls_timing_set_delay, mbedtls_timing_get_delay: tested by setting a delay object and watching it go through its two delay values, using a timer to check that the delays are passed at the expected time. The tests pass under light to moderate load, but some of them can be defeated with sufficiently heavy load. This is unavoidable since the test process to be effectively suspended for any length of time, making us think that a timer has gone on for too long.
2017-10-10 17:56:06 +00:00
/* BEGIN_CASE */
void timing_timer_reset( )
{
struct mbedtls_timing_hr_time timer;
unsigned long millis = 0;
unsigned long iterations = 0;
/* Start the timer. Timers are always reset to 0. */
TEST_ASSERT( mbedtls_timing_get_timer( &timer, 1 ) == 0 );
/* Busy-wait loop for a few milliseconds */
do
{
++iterations;
millis = mbedtls_timing_get_timer( &timer, 0 );
}
while( millis < TIMING_SHORT_TEST_MS );
/* Reset the timer and check that it has restarted. */
TEST_ASSERT( mbedtls_timing_get_timer( &timer, 1 ) == 0 );
/* Read the timer immediately after reset. It should be 0 or close
to it. */
TEST_ASSERT( mbedtls_timing_get_timer( &timer, 0 ) < TIMING_SHORT_TEST_MS );
return;
exit:
/* No cleanup needed, but show some diagnostic information, because timing
problems can be hard to reproduce. */
mbedtls_fprintf( stdout, " Finished with millis=%lu get(timer)<=%lu iterations=%lu\n",
millis, mbedtls_timing_get_timer( &timer, 0 ),
iterations );
}
/* END_CASE */
/* BEGIN_CASE */
void timing_two_timers( int delta )
{
struct mbedtls_timing_hr_time timer1, timer2;
unsigned long millis1, millis2;
/* Start the first timer and wait for a short time. */
(void) mbedtls_timing_get_timer( &timer1, 1 );
do
{
millis1 = mbedtls_timing_get_timer( &timer1, 0 );
}
while( millis1 < TIMING_SHORT_TEST_MS );
/* Do a short busy-wait, so that the difference between timer1 and timer2
doesn't practically always end up being very close to a whole number of
milliseconds. */
while( delta > 0 )
--delta;
/* Start the second timer and compare it with the first. */
mbedtls_timing_get_timer( &timer2, 1 );
do
{
millis1 = mbedtls_timing_get_timer( &timer1, 0 );
millis2 = mbedtls_timing_get_timer( &timer2, 0 );
/* The first timer should always be ahead of the first. */
TEST_ASSERT( millis1 > millis2 );
/* The timers shouldn't drift apart, i.e. millis2-millis1 should stay
roughly constant, but this is hard to test reliably, especially in
a busy environment such as an overloaded continuous integration
system, so we don't test it it. */
}
while( millis2 < TIMING_SHORT_TEST_MS );
return;
exit:
/* No cleanup needed, but show some diagnostic iterations, because timing
problems can be hard to reproduce. */
mbedtls_fprintf( stdout, " Finished with millis1=%lu get(timer1)<=%lu millis2=%lu get(timer2)<=%lu\n",
millis1, mbedtls_timing_get_timer( &timer1, 0 ),
millis2, mbedtls_timing_get_timer( &timer2, 0 ) );
}
/* END_CASE */
/* BEGIN_CASE */
void timing_alarm( int seconds )
{
struct mbedtls_timing_hr_time timer;
unsigned long millis = 0;
/* We check that about the desired number of seconds has elapsed. Be
slightly liberal with the lower bound, so as to allow platforms where
the alarm (with second resolution) and the timer (with millisecond
resolution) are based on different clocks. Be very liberal with the
upper bound, because the platform might be busy. */
unsigned long millis_min = ( seconds > 0 ?
seconds * 900 :
0 );
unsigned long millis_max = ( seconds > 0 ?
seconds * 1100 + 400 :
TIMING_ALARM_0_DELAY_MS );
unsigned long iterations = 0;
/* Set an alarm and count how long it takes with a timer. */
(void) mbedtls_timing_get_timer( &timer, 1 );
mbedtls_set_alarm( seconds );
if( seconds > 0 )
{
/* We set the alarm for at least 1 second. It should not have fired
immediately, even on a slow and busy platform. */
TEST_ASSERT( !mbedtls_timing_alarmed );
}
/* A 0-second alarm should fire quickly, but we don't guarantee that it
fires immediately, so mbedtls_timing_alarmed may or may not be set at
this point. */
/* Busy-wait until the alarm rings */
do
{
++iterations;
millis = mbedtls_timing_get_timer( &timer, 0 );
}
while( !mbedtls_timing_alarmed && millis <= millis_max );
TEST_ASSERT( mbedtls_timing_alarmed );
TEST_ASSERT( millis >= millis_min );
TEST_ASSERT( millis <= millis_max );
mbedtls_timing_alarmed = 0;
return;
exit:
/* Show some diagnostic iterations, because timing
problems can be hard to reproduce. */
mbedtls_fprintf( stdout, " Finished with alarmed=%d millis=%lu get(timer)<=%lu iterations=%lu\n",
mbedtls_timing_alarmed,
millis, mbedtls_timing_get_timer( &timer, 0 ),
iterations );
/* Cleanup */
mbedtls_timing_alarmed = 0;
}
/* END_CASE */
/* BEGIN_CASE */
void timing_delay( int int_ms, int fin_ms )
{
/* This function assumes that if int_ms is nonzero then it is large
enough that we have time to read all timers at least once in an
interval of time lasting int_ms milliseconds, and likewise for (fin_ms
- int_ms). So don't call it with arguments that are too small. */
mbedtls_timing_delay_context delay;
struct mbedtls_timing_hr_time timer;
unsigned long delta; /* delay started between timer=0 and timer=delta */
unsigned long before = 0, after = 0;
unsigned long iterations = 0;
int status = -2;
int saw_status_1 = 0;
int warn_inconclusive = 0;
assert( int_ms >= 0 );
assert( fin_ms >= 0 );
/* Start a reference timer. Program a delay, and verify that the status of
the delay is consistent with the time given by the reference timer. */
(void) mbedtls_timing_get_timer( &timer, 1 );
mbedtls_timing_set_delay( &delay, int_ms, fin_ms );
/* Set delta to an upper bound for the interval between the start of timer
and the start of delay. Reading timer after starting delay gives us an
upper bound for the interval, rounded to a 1ms precision. Since this
might have been rounded down, but we need an upper bound, we add 1. */
delta = mbedtls_timing_get_timer( &timer, 0 ) + 1;
status = mbedtls_timing_get_delay( &delay );
if( fin_ms == 0 )
{
/* Cancelled timer. Just check the correct status for this case. */
TEST_ASSERT( status == -1 );
return;
}
/* Initially, none of the delays must be passed yet if they're nonzero.
This could fail for very small values of int_ms and fin_ms, where "very
small" depends how fast and how busy the platform is. */
if( int_ms > 0 )
{
TEST_ASSERT( status == 0 );
}
else
{
TEST_ASSERT( status == 1 );
}
do
{
unsigned long delay_min, delay_max;
int status_min, status_max;
++iterations;
before = mbedtls_timing_get_timer( &timer, 0 );
status = mbedtls_timing_get_delay( &delay );
after = mbedtls_timing_get_timer( &timer, 0 );
/* At a time between before and after, the delay's status was status.
Check that this is consistent given that the delay was started
between times 0 and delta. */
delay_min = ( before > delta ? before - delta : 0 );
status_min = expected_delay_status( int_ms, fin_ms, delay_min );
delay_max = after;
status_max = expected_delay_status( int_ms, fin_ms, delay_max );
TEST_ASSERT( status >= status_min );
TEST_ASSERT( status <= status_max );
if( status == 1 )
saw_status_1 = 1;
}
while ( before <= fin_ms + delta && status != 2 );
/* Since we've waited at least fin_ms, the delay must have fully
expired. */
TEST_ASSERT( status == 2 );
/* If the second delay is more than the first, then there must have been a
point in time when the first delay was passed but not the second delay.
This could fail for very small values of (fin_ms - int_ms), where "very
small" depends how fast and how busy the platform is. In practice, this
is the test that's most likely to fail on a heavily loaded machine. */
if( fin_ms > int_ms )
{
warn_inconclusive = 1;
TEST_ASSERT( saw_status_1 );
}
return;
exit:
/* No cleanup needed, but show some diagnostic iterations, because timing
problems can be hard to reproduce. */
mbedtls_fprintf( stdout, " Finished with delta=%lu before=%lu after=%lu status=%d iterations=%lu\n",
delta, before, after, status, iterations );
if( warn_inconclusive )
mbedtls_fprintf( stdout, " Inconclusive test, try running it on a less heavily loaded machine.\n" );
}
/* END_CASE */
/* BEGIN_CASE */
void timing_hardclock( )
{
/* We make very few guarantees about mbedtls_timing_hardclock: its rate is
platform-dependent, it can wrap around. So there isn't much we can
test. But we do at least test that it doesn't crash, stall or return
completely nonsensical values. */
struct mbedtls_timing_hr_time timer;
unsigned long hardclock0, hardclock1, delta1;
hardclock0 = mbedtls_timing_hardclock( );
/* Wait 2ms to ensure a nonzero delay. Since the timer interface has 1ms
resolution and unspecified precision, waiting 1ms might be a very small
delay that's rounded up. */
(void) mbedtls_timing_get_timer( &timer, 1 );
while( mbedtls_timing_get_timer( &timer, 0 ) < 2 )
/*busy-wait loop*/;
hardclock1 = mbedtls_timing_hardclock( );
/* Although the hardclock counter can wrap around, the difference
(hardclock1 - hardclock0) is taken modulo the type size, so it is
correct as long as the counter only wrapped around at most once. We
further require the difference to be nonzero (after a wait of more than
1ms, the counter must have changed), and not to be overly large (after
a wait of less than 3ms, plus time lost because other processes were
scheduled on the CPU). If the hardclock counter runs at 4GHz, then
1000000000 (which is 1/4 of the counter wraparound on a 32-bit machine)
allows 250ms. */
delta1 = hardclock1 - hardclock0;
TEST_ASSERT( delta1 > 0 );
TEST_ASSERT( delta1 < 1000000000 );
return;
exit:
/* No cleanup needed, but show some diagnostic iterations, because timing
problems can be hard to reproduce. */
mbedtls_fprintf( stdout, " Finished with hardclock=%lu,%lu\n",
hardclock0, hardclock1 );
}
/* END_CASE */