David
/
Foundry
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Foundry/tt/tests/test_tensix_claim.c

759 lines
31 KiB
C
Raw Permalink Blame History

/* SPDX-License-Identifier: AGPL-3.0-or-later */
/* Copyright (C) 2026 David (Corey) Herrera */
/**
* @file test_tensix_claim.c
* @brief Gate for ft_tt_tensix_claim.c — the per-CORE ownership that lets
* correctness work run in parallel instead of serialising behind the
* whole-card lock.
*
* ---------------------------------------------------------------------------
* WHAT THIS SUITE IS BUILT TO CATCH, AND HOW IT AVOIDS BEING VACUOUS
* ---------------------------------------------------------------------------
* Two rules shape every stage here.
*
* **The lock is checked directly, never through the claim layer.** A claim
* verified by asking `ft_tt_tensix_claim_acquire()` again proves only that the
* module agrees with itself; a build whose acquire is a no-op passes that. So
* `lock_file_is_free()` opens `<root>/<card>/tensix/<index>.lock` and takes the
* `flock` itself. That is the kernel's state, and nothing in the module can make
* it lie. Same reasoning as test_udrv_claim.c's forged-record stage.
*
* **The decisive case is two processes racing for one core**, and it is run in
* two shapes because they catch different bugs. Stage 4 is DETERMINISTIC — a
* helper holds the middle core of a three-core request, so the refusal and the
* roll-back happen at a known point and the "nothing was half-claimed" assertion
* is not a timing accident. Stage 5 is a REAL RACE — four forked children fire at
* one core off a shared starting gun, and exactly one may win.
*
* @status Off-hardware. Filesystem and fork only; no device is opened, no NoC
* transaction is possible from anything this links.
* @issues None
* @todo None
*/
/* flock(2) is BSD, not POSIX, so a bare _POSIX_C_SOURCE hides it on both
* platforms this code is built on (DEV_NOTES §5, the _DEFAULT_SOURCE note). */
#if defined(__APPLE__)
#define _DARWIN_C_SOURCE
#else
#define _DEFAULT_SOURCE
#endif
#include <errno.h>
#include <fcntl.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/file.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>
#include "ft_tt_tensix_claim.h"
#include "ft_tt_udrv_claim.h"
#include "ft_tt_test_report.h"
#include "ft_tt_udrv.h"
#define TEST_CARD "0000:12:00.0"
#define RACE_CHILD_COUNT 4u
#define RACE_CORE 9u
static char scratch_root[160];
static void build_scratch_root(void)
{
(void)snprintf(scratch_root, sizeof(scratch_root), "/tmp/ft_tt_tensix_claim_test.%ld",
(long)getpid());
}
static void remove_scratch_root(void)
{
char command[256];
(void)snprintf(command, sizeof(command), "rm -rf '%s'", scratch_root);
if (system(command) != 0)
(void)fprintf(stderr, "fixture: cleanup of %s failed\n", scratch_root);
}
/**
* THE INDEPENDENT INSTRUMENT: is the kernel's lock on this core free?
*
* Deliberately does not call into ft_tt_tensix_claim or ft_tt_udrv_claim at all.
* It opens the lock file by path and takes the flock, so a module that forgets to
* lock, forgets to release, or leaks a descriptor cannot hide from it.
*/
static bool lock_file_is_free(uint32_t index)
{
char path[256];
int fd;
bool free_now;
(void)snprintf(path, sizeof(path), "%s/%s/%s/%u.lock", scratch_root, TEST_CARD,
FT_TT_TENSIX_CLAIM_NAMESPACE, index);
fd = open(path, O_RDWR | O_CREAT | O_CLOEXEC, 0666);
if (fd < 0)
return true; /* no file at all means nobody ever claimed it */
free_now = (flock(fd, LOCK_EX | LOCK_NB) == 0);
if (free_now)
(void)flock(fd, LOCK_UN);
(void)close(fd);
return free_now;
}
/** Non-NULL with a count of zero: an EXPLICITLY empty reserved set. NULL means
* "read the environment", and a suite whose result depends on the shell it was
* launched from is a suite that lies. */
static const uint32_t k_no_reserved[1] = {0u};
/** A request against the scratch root with no reserved set and no retirement. */
static void build_request(ft_tt_tensix_claim_request *request, const uint32_t *indices,
size_t count, const char *reason)
{
memset(request, 0, sizeof(*request));
request->root = scratch_root;
request->card = TEST_CARD;
request->indices = indices;
request->count = count;
request->reason = reason;
request->reserved = k_no_reserved;
request->reserved_count = 0u;
}
static void stage_parse_index_lists(void)
{
uint32_t out[16];
size_t count = 0u;
ft_tt_test_step("Stage 1 — the core list parser, which decides what a sweep may name");
ft_tt_test_check_fmt("a single index parses",
ft_tt_tensix_claim_parse_indices("71", out, 16u, &count) ==
FT_TT_UDRV_OK &&
count == 1u && out[0] == 71u,
"count=%zu", count);
ft_tt_test_check_fmt("a comma list keeps its order",
ft_tt_tensix_claim_parse_indices("3,1,17", out, 16u, &count) ==
FT_TT_UDRV_OK &&
count == 3u && out[0] == 3u && out[1] == 1u && out[2] == 17u,
"count=%zu", count);
ft_tt_test_check_fmt("an inclusive range expands",
ft_tt_tensix_claim_parse_indices("30-33", out, 16u, &count) ==
FT_TT_UDRV_OK &&
count == 4u && out[0] == 30u && out[3] == 33u,
"count=%zu", count);
ft_tt_test_check_fmt("ranges and indices mix",
ft_tt_tensix_claim_parse_indices("1,4-6,9", out, 16u, &count) ==
FT_TT_UDRV_OK &&
count == 5u && out[1] == 4u && out[3] == 6u && out[4] == 9u,
"count=%zu", count);
ft_tt_test_check("an empty string is zero indices, not an error",
ft_tt_tensix_claim_parse_indices("", out, 16u, &count) == FT_TT_UDRV_OK &&
count == 0u,
"unset means no reserved cores");
ft_tt_test_check("a NULL string is zero indices",
ft_tt_tensix_claim_parse_indices(NULL, out, 16u, &count) == FT_TT_UDRV_OK &&
count == 0u,
"");
ft_tt_test_check("an inverted range is refused",
ft_tt_tensix_claim_parse_indices("9-3", out, 16u, &count) ==
FT_TT_UDRV_EINVAL,
"");
ft_tt_test_check("trailing junk is refused",
ft_tt_tensix_claim_parse_indices("1,2x", out, 16u, &count) ==
FT_TT_UDRV_EINVAL,
"a typo must not silently claim fewer cores");
ft_tt_test_check("a space is refused",
ft_tt_tensix_claim_parse_indices("1, 2", out, 16u, &count) ==
FT_TT_UDRV_EINVAL,
"more likely a quoting mistake than an intent");
ft_tt_test_check("an index at the namespace bound is refused",
ft_tt_tensix_claim_parse_indices("256", out, 16u, &count) ==
FT_TT_UDRV_EINVAL,
"");
ft_tt_test_check("overflowing the caller's array is refused, not truncated",
ft_tt_tensix_claim_parse_indices("0-99", out, 16u, &count) ==
FT_TT_UDRV_EINVAL,
"a truncated set is a set that claims the wrong cores");
}
static void stage_set_ownership(void)
{
static const uint32_t wanted[] = {40u, 41u, 42u};
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
ft_tt_test_step("Stage 2 — a set is claimed, and the KERNEL agrees it is claimed");
ft_tt_test_check("all three cores are free before we start",
lock_file_is_free(40u) && lock_file_is_free(41u) && lock_file_is_free(42u),
"checked by flock, not by asking the module");
build_request(&request, wanted, 3u, "stage 2 set");
ft_tt_test_check("the set is granted",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_OK, "");
ft_tt_test_check_fmt("it holds three cores", ft_tt_tensix_claim_count(set) == 3u, "%zu",
ft_tt_tensix_claim_count(set));
ft_tt_test_check_fmt("...in the order asked for",
ft_tt_tensix_claim_index_at(set, 0u) == 40u &&
ft_tt_tensix_claim_index_at(set, 2u) == 42u,
"%u..%u", ft_tt_tensix_claim_index_at(set, 0u),
ft_tt_tensix_claim_index_at(set, 2u));
ft_tt_test_check("the kernel says all three locks are taken",
!lock_file_is_free(40u) && !lock_file_is_free(41u) && !lock_file_is_free(42u),
"flock refused on each");
ft_tt_test_check("a core NOT in the set is untouched", lock_file_is_free(43u), "");
ft_tt_tensix_claim_release(set);
ft_tt_test_check("releasing gives all three back",
lock_file_is_free(40u) && lock_file_is_free(41u) && lock_file_is_free(42u),
"flock granted on each");
ft_tt_tensix_claim_release(NULL);
ft_tt_test_check("releasing NULL is a no-op", true, "returned");
}
static void stage_bad_requests(void)
{
static const uint32_t twice[] = {7u, 12u, 7u};
static const uint32_t out_of_range[] = {FT_TT_TENSIX_CLAIM_INDEX_COUNT};
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
ft_tt_test_step("Stage 3 — malformed requests are refused BEFORE anything is locked");
build_request(&request, twice, 3u, "duplicate");
ft_tt_test_check("a core named twice is refused",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_EINVAL,
"it would self-deadlock and report itself as the holder");
ft_tt_test_check_fmt("...and the refusal names the duplicate",
refusal.outcome == FT_TT_TENSIX_CLAIM_DUPLICATE && refusal.index == 7u,
"index %u", refusal.index);
ft_tt_test_check("...and nothing was locked", lock_file_is_free(7u) && lock_file_is_free(12u),
"");
build_request(&request, out_of_range, 1u, "oob");
ft_tt_test_check("an index at the namespace bound is refused",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_EINVAL,
"");
build_request(&request, twice, 0u, "empty");
ft_tt_test_check("an empty request is refused",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_EINVAL,
"claiming nothing is not a claim");
}
/** The other job: claim `index`, say so down the pipe, then wait to be killed
* WITHOUT releasing anything. Never returns. */
static void holder_child(uint32_t index, const char *reason, int ready_pipe[2])
{
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
uint32_t held_index = index;
char signal_byte = 1;
(void)close(ready_pipe[0]);
build_request(&request, &held_index, 1u, reason);
if (ft_tt_tensix_claim_acquire(&request, &set, &refusal) != FT_TT_UDRV_OK)
_exit(3);
if (write(ready_pipe[1], &signal_byte, 1u) != 1)
_exit(4);
for (;;)
pause();
}
/**
* Fork the holder above.
*
* A pipe handshake rather than a poll loop: a parent that polls by ATTEMPTING the
* claim can win the race itself, and the stage then goes red for a reason that
* has nothing to do with the mechanism (test_udrv_claim.c learned this one).
*
* @return the child pid, or -1
*/
static pid_t fork_holder(uint32_t index, const char *reason, int ready_pipe[2])
{
pid_t child;
if (pipe(ready_pipe) != 0)
return -1;
child = fork();
if (child == 0)
holder_child(index, reason, ready_pipe);
(void)close(ready_pipe[1]);
return child;
}
static void stage_no_half_claim(void)
{
static const uint32_t wanted[] = {20u, 21u, 22u};
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
int ready_pipe[2];
char ready = 0;
pid_t holder;
int status = 0;
ft_tt_test_step("Stage 4 — THE DECISIVE ONE: a refused set leaves NOTHING half-claimed");
holder = fork_holder(21u, "the other job", ready_pipe);
if (holder < 0) {
ft_tt_test_check("fork succeeded", false, "fork failed");
return;
}
ft_tt_test_check("the other process holds core 21",
read(ready_pipe[0], &ready, 1u) == 1 && ready == 1, "handshake");
(void)close(ready_pipe[0]);
ft_tt_test_check("...and the kernel agrees", !lock_file_is_free(21u), "flock refused");
/* 20 is taken first, 21 refuses, 22 is never reached. The roll-back is what
* must give 20 back — and it is checked by flock, not by re-asking. */
build_request(&request, wanted, 3u, "the sweep that loses");
ft_tt_test_check("the whole set is refused",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_EAGAIN,
"one core of three was taken");
ft_tt_test_check_fmt("the refusal names the core it lost on",
refusal.outcome == FT_TT_TENSIX_CLAIM_HELD && refusal.index == 21u,
"outcome=%d index=%u", (int)refusal.outcome, refusal.index);
ft_tt_test_check_fmt("...and names the process holding it", refusal.pid == holder,
"pid %ld vs holder %ld", (long)refusal.pid, (long)holder);
ft_tt_test_check_fmt("...and the reason that process registered",
strcmp(refusal.reason, "the other job") == 0, "%s", refusal.reason);
ft_tt_test_check("core 20 — claimed then rolled back — is FREE", lock_file_is_free(20u),
"the kernel's answer, not the module's");
ft_tt_test_check("core 22 — never reached — is FREE", lock_file_is_free(22u), "");
(void)kill(holder, SIGKILL);
(void)waitpid(holder, &status, 0);
ft_tt_test_check("SIGKILL alone returns core 21", lock_file_is_free(21u),
"no cleanup path of ours ran");
ft_tt_test_check("...so the same set now succeeds",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_OK, "");
ft_tt_tensix_claim_release(set);
}
/** One racer: wait for the gun, claim RACE_CORE, report the outcome down a pipe. */
static void race_child(int gate_read, int report_write)
{
static uint32_t index = RACE_CORE;
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
char gun = 0;
char line[64];
int rc;
int written;
if (read(gate_read, &gun, 1u) != 1)
_exit(9);
build_request(&request, &index, 1u, "racer");
rc = ft_tt_tensix_claim_acquire(&request, &set, &refusal);
written = snprintf(line, sizeof(line), "%ld %d %ld\n", (long)getpid(),
rc == FT_TT_UDRV_OK ? 0 : (int)refusal.outcome, (long)refusal.pid);
if (written > 0) {
/* glibc marks write() warn_unused_result, and a (void) cast does NOT silence that
* under gcc -Werror=unused-result (it does under clang, which is why this built on
* the Mac and broke Makefile.udrv on the card host). This is a child reporting its verdict
* down a pipe on a path where there is nothing useful to do about a short write. */
ssize_t reported = write(report_write, line, (size_t)written);
(void)reported;
}
/* A winner must hold the claim long enough for every loser to be refused,
* otherwise the race is decided by scheduling rather than by flock. */
if (rc == FT_TT_UDRV_OK) {
sleep(1u);
ft_tt_tensix_claim_release(set);
}
_exit(0);
}
static void stage_real_race(void)
{
int gate[2];
int report[2];
pid_t children[RACE_CHILD_COUNT];
char gun[RACE_CHILD_COUNT];
char buffer[512];
ssize_t got;
size_t total = 0u;
unsigned granted = 0u;
unsigned held = 0u;
unsigned other = 0u;
unsigned named_the_winner = 0u;
long winner_pid = 0;
unsigned child;
char *cursor;
ft_tt_test_step("Stage 5 — a REAL race: %u processes, one core, one winner",
RACE_CHILD_COUNT);
if (pipe(gate) != 0 || pipe(report) != 0) {
ft_tt_test_check("pipes created", false, "pipe failed");
return;
}
for (child = 0u; child < RACE_CHILD_COUNT; child++) {
children[child] = fork();
if (children[child] == 0) {
(void)close(gate[1]);
(void)close(report[0]);
race_child(gate[0], report[1]);
}
}
(void)close(gate[0]);
(void)close(report[1]);
memset(gun, 1, sizeof(gun));
ft_tt_test_check("the starting gun fires for all of them at once",
write(gate[1], gun, sizeof(gun)) == (ssize_t)sizeof(gun), "one write");
(void)close(gate[1]);
while (total < sizeof(buffer) - 1u &&
(got = read(report[0], buffer + total, sizeof(buffer) - 1u - total)) > 0)
total += (size_t)got;
buffer[total] = '\0';
(void)close(report[0]);
for (child = 0u; child < RACE_CHILD_COUNT; child++) {
int status = 0;
(void)waitpid(children[child], &status, 0);
}
/* First pass finds the winner, second pass checks every loser named it. */
for (cursor = buffer; *cursor != '\0';) {
long reporter = 0;
int outcome = -1;
long holder = 0;
if (sscanf(cursor, "%ld %d %ld", &reporter, &outcome, &holder) == 3) {
if (outcome == (int)FT_TT_TENSIX_CLAIM_GRANTED) {
granted++;
winner_pid = reporter;
} else if (outcome == (int)FT_TT_TENSIX_CLAIM_HELD) {
held++;
} else {
other++;
}
}
cursor = strchr(cursor, '\n');
if (!cursor)
break;
cursor++;
}
for (cursor = buffer; *cursor != '\0';) {
long reporter = 0;
int outcome = -1;
long holder = 0;
if (sscanf(cursor, "%ld %d %ld", &reporter, &outcome, &holder) == 3 &&
outcome == (int)FT_TT_TENSIX_CLAIM_HELD && holder == winner_pid)
named_the_winner++;
cursor = strchr(cursor, '\n');
if (!cursor)
break;
cursor++;
}
ft_tt_test_check_fmt("exactly one process won", granted == 1u, "%u granted", granted);
ft_tt_test_check_fmt("every other process was told the core is HELD",
held == RACE_CHILD_COUNT - 1u && other == 0u, "%u held, %u other", held,
other);
ft_tt_test_check_fmt("...and every refusal named the winner by pid",
named_the_winner == RACE_CHILD_COUNT - 1u, "%u of %u named pid %ld",
named_the_winner, RACE_CHILD_COUNT - 1u, winner_pid);
ft_tt_test_check("the core is free once the winner exits", lock_file_is_free(RACE_CORE), "");
}
static void stage_reserved_set(void)
{
static const uint32_t wanted[] = {30u, 31u};
static const uint32_t reserved[] = {31u};
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
size_t parsed_count = 0u;
uint32_t parsed[32];
ft_tt_test_step("Stage 6 — reserved cores: another family's, and a live serve's");
build_request(&request, wanted, 2u, "sweep into reserved");
request.reserved = reserved;
request.reserved_count = 1u;
ft_tt_test_check("a request naming a reserved core is refused",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_EAGAIN,
"");
ft_tt_test_check_fmt("...and says RESERVED, not HELD",
refusal.outcome == FT_TT_TENSIX_CLAIM_RESERVED && refusal.index == 31u,
"outcome=%d index=%u", (int)refusal.outcome, refusal.index);
/* The point of checking 30: reserved is evaluated for the WHOLE request
* before any lock is taken, so even the core that WOULD have been fine is
* never touched. */
ft_tt_test_check("...and core 30, which comes first, was never locked",
lock_file_is_free(30u), "reserved is checked before anything is claimed");
request.reserved_count = 0u;
ft_tt_test_check("with nothing reserved the same request succeeds",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_OK, "");
ft_tt_tensix_claim_release(set);
(void)setenv(FT_TT_TENSIX_CLAIM_RESERVED_ENVIRONMENT, "0-9,14,15,28", 1);
ft_tt_test_check_fmt("the environment reserved set parses",
ft_tt_tensix_claim_reserved_from_environment(parsed, 32u,
&parsed_count) ==
FT_TT_UDRV_OK &&
parsed_count == 13u,
"%zu cores", parsed_count);
{
static const uint32_t into_environment_reserved[] = {14u};
ft_tt_tensix_claim_request env_request;
build_request(&env_request, into_environment_reserved, 1u, "env reserved");
env_request.reserved = NULL; /* NULL is what reads the environment */
ft_tt_test_check("a NULL reserved set reads the environment and refuses",
ft_tt_tensix_claim_acquire(&env_request, &set, &refusal) ==
FT_TT_UDRV_EAGAIN,
"");
ft_tt_test_check_fmt("...as RESERVED", refusal.outcome == FT_TT_TENSIX_CLAIM_RESERVED,
"outcome=%d", (int)refusal.outcome);
}
(void)unsetenv(FT_TT_TENSIX_CLAIM_RESERVED_ENVIRONMENT);
}
/* Recovery-on-acquire bookkeeping for stage 7. */
static uint32_t g_retired[8];
static size_t g_retired_count;
static uint32_t g_retire_fails_on = UINT32_MAX;
static int record_retirement(void *context, uint32_t core_index)
{
(void)context;
if (g_retired_count < 8u)
g_retired[g_retired_count++] = core_index;
return (core_index == g_retire_fails_on) ? -7 : 0;
}
static void stage_recovery_on_acquire(void)
{
static const uint32_t wanted[] = {50u, 51u, 52u};
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
ft_tt_test_step("Stage 7 — recovery-on-acquire: a released claim is not a retired core");
g_retired_count = 0u;
g_retire_fails_on = UINT32_MAX;
build_request(&request, wanted, 3u, "retire ok");
request.retire = record_retirement;
ft_tt_test_check("the set is granted",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_OK, "");
ft_tt_test_check_fmt("every claimed core was offered for retirement, in order",
g_retired_count == 3u && g_retired[0] == 50u && g_retired[1] == 51u &&
g_retired[2] == 52u,
"%zu retirements", g_retired_count);
ft_tt_tensix_claim_release(set);
/* A retirement that refuses must release the WHOLE set — a job that cannot
* get one core into a known state must not run on the others. */
g_retired_count = 0u;
g_retire_fails_on = 51u;
ft_tt_test_check("a set whose middle core cannot be retired is refused",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_EAGAIN,
"");
ft_tt_test_check_fmt("...reported as UNRETIRED, naming that core",
refusal.outcome == FT_TT_TENSIX_CLAIM_UNRETIRED && refusal.index == 51u,
"outcome=%d index=%u", (int)refusal.outcome, refusal.index);
ft_tt_test_check("...and ALL THREE locks are back, not just the failing one",
lock_file_is_free(50u) && lock_file_is_free(51u) && lock_file_is_free(52u),
"checked by flock");
ft_tt_test_check_fmt("...and retirement stopped at the refusal", g_retired_count == 2u,
"%zu retirements", g_retired_count);
g_retire_fails_on = UINT32_MAX;
}
/**
* The hazard this project actually has: a harness that `_Exit()`s.
*
* `_Exit` runs no atexit handler and no destructor, so nothing of ours releases
* anything — and the claim comes back anyway, because the kernel closes the
* descriptor. That is the property flock was chosen for. It is also exactly why
* `retire` has to exist: the LOCK came back and the CORE did not.
*/
static void stage_exit_without_release(void)
{
static uint32_t index = 60u;
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
pid_t child;
int status = 0;
ft_tt_test_step("Stage 8 — a harness that _Exit()s: the LOCK returns, the CORE does not");
child = fork();
if (child == 0) {
ft_tt_tensix_claim_request child_request;
ft_tt_tensix_claim_refusal child_refusal;
ft_tt_tensix_claim_set *child_set = NULL;
build_request(&child_request, &index, 1u, "harness that _Exit()s");
if (ft_tt_tensix_claim_acquire(&child_request, &child_set, &child_refusal) !=
FT_TT_UDRV_OK)
_exit(3);
_exit(0); /* no release, no cleanup, no unwind */
}
if (child < 0) {
ft_tt_test_check("fork succeeded", false, "fork failed");
return;
}
(void)waitpid(child, &status, 0);
ft_tt_test_check_fmt("the child claimed and then _Exit()ed",
WIFEXITED(status) && WEXITSTATUS(status) == 0, "status %d", status);
ft_tt_test_check("the lock came back with no cleanup code of ours running",
lock_file_is_free(60u), "the kernel closed the descriptor");
g_retired_count = 0u;
g_retire_fails_on = UINT32_MAX;
build_request(&request, &index, 1u, "the next claimant");
request.retire = record_retirement;
ft_tt_test_check("the next claimant gets the core",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_OK, "");
ft_tt_test_check_fmt("...and is made to retire it before using it — that is the whole point",
g_retired_count == 1u && g_retired[0] == 60u, "%zu retirements",
g_retired_count);
ft_tt_tensix_claim_release(set);
}
/**
* A whole-card operation must be able to SEE per-core claims before it resets.
*
* The card lock only ever saw processes that registered a shared claim, so once
* ordinary work moved to per-core claims a reset could report "nobody is using
* the card" while wiping eight live suites. `ft_tt_core` closes that with a
* census over this namespace — and the census is only true if it resolves the
* root and card exactly as a claimant does, which is what this stage pins.
*/
static void stage_card_wide_census_sees_core_claims(void)
{
static const uint32_t wanted[] = {33u};
ft_tt_tensix_claim_request request;
ft_tt_tensix_claim_refusal refusal;
ft_tt_tensix_claim_set *set = NULL;
ft_tt_udrv_claim_namespace census;
ft_tt_udrv_claim_info *info;
uint32_t held = 0u;
ft_tt_test_step("Stage 9 — a card-wide operation can see, and name, per-core claims");
(void)setenv(FT_TT_TENSIX_CLAIM_ROOT_ENVIRONMENT, scratch_root, 1);
/* NULL-guarded: an unset root is a LEGAL answer from this resolver (it means
* "ft_tt_udrv_claim's own default"), so a mutation that stops reading the
* environment must go red here rather than segfault a stage later. */
{
const char *resolved = ft_tt_tensix_claim_resolve_root(NULL);
ft_tt_test_check_fmt("the root resolver honours the environment",
resolved != NULL && strcmp(resolved, scratch_root) == 0, "%s",
resolved ? resolved : "(null)");
if (!resolved) {
(void)unsetenv(FT_TT_TENSIX_CLAIM_ROOT_ENVIRONMENT);
return; /* the census below would coordinate outside the scratch root */
}
}
ft_tt_test_check("an explicit root still wins over the environment",
strcmp(ft_tt_tensix_claim_resolve_root("/explicit"), "/explicit") == 0, "");
ft_tt_test_check("the card resolver falls back to the single-card default",
strcmp(ft_tt_tensix_claim_resolve_card(NULL),
FT_TT_TENSIX_CLAIM_CARD_DEFAULT) == 0,
"");
memset(&request, 0, sizeof(request));
request.root = NULL; /* NULL is what makes the resolver load-bearing here */
request.card = TEST_CARD;
request.indices = wanted;
request.count = 1u;
request.reason = "a live suite";
request.reserved = k_no_reserved;
ft_tt_test_check("a core is claimed through the resolved root",
ft_tt_tensix_claim_acquire(&request, &set, &refusal) == FT_TT_UDRV_OK, "");
/* The census, built the way ft_tt_core builds it — a SEPARATE namespace
* object over the lower layer, not a second call into ft_tt_tensix_claim. */
(void)ft_tt_udrv_claim_namespace_init(&census, ft_tt_tensix_claim_resolve_root(NULL),
ft_tt_tensix_claim_resolve_card(TEST_CARD),
FT_TT_TENSIX_CLAIM_NAMESPACE,
FT_TT_TENSIX_CLAIM_INDEX_COUNT);
info = calloc((size_t)census.count, sizeof(*info));
if (!info) {
ft_tt_test_check("census buffer allocates", false, "out of memory");
ft_tt_tensix_claim_release(set);
return;
}
(void)ft_tt_udrv_claim_scan(&census, info, (size_t)census.count, &held);
ft_tt_test_check_fmt("the census sees exactly the one claimed core", held == 1u, "%u held",
held);
ft_tt_test_check_fmt("...at the right index, naming this process",
info[33].held && info[33].pid == getpid(), "index 33 pid %ld",
(long)info[33].pid);
ft_tt_test_check_fmt("...with the reason it registered",
strcmp(info[33].reason, "a live suite") == 0, "%s", info[33].reason);
free(info);
ft_tt_tensix_claim_release(set);
(void)unsetenv(FT_TT_TENSIX_CLAIM_ROOT_ENVIRONMENT);
}
int main(void)
{
build_scratch_root();
/* Nothing here may inherit the operator's reserved set: a suite whose result
* depends on the shell it was launched from is a suite that lies. */
(void)unsetenv(FT_TT_TENSIX_CLAIM_RESERVED_ENVIRONMENT);
stage_parse_index_lists();
stage_set_ownership();
stage_bad_requests();
stage_no_half_claim();
stage_real_race();
stage_reserved_set();
stage_recovery_on_acquire();
stage_exit_without_release();
stage_card_wide_census_sees_core_claims();
remove_scratch_root();
return ft_tt_test_finish();
}
Reference in New Issue View Git Blame Copy Permalink