static void
meminfo_display(void)
{
printf("----------- MEMORY_SEGMENTS -----------\n");
rte_dump_physmem_layout(stdout);
printf("--------- END_MEMORY_SEGMENTS ---------\n");
printf("------------ MEMORY_ZONES -------------\n");
rte_memzone_dump(stdout);
printf("---------- END_MEMORY_ZONES -----------\n");
printf("------------- TAIL_QUEUES -------------\n");
rte_dump_tailq(stdout);
printf("---------- END_TAIL_QUEUES ------------\n");
}
static void
spdk_nvmf_startup(spdk_event_t event)
{
int rc;
/* start the rdma poller that will listen
on all available ports */
rc = nvmf_acceptor_start();
if (rc < 0) {
SPDK_ERRLOG("nvmf_acceptor_start() failed\n");
goto initialize_error;
}
if (getenv("MEMZONE_DUMP") != NULL) {
rte_memzone_dump(stdout);
fflush(stdout);
}
return;
initialize_error:
spdk_app_stop(rc);
}
int
main(int argc, char **argv)
{
int ret;
int i;
char c_flag[] = "-c1";
char n_flag[] = "-n4";
char mp_flag[] = "--proc-type=secondary";
char *argp[argc + 3];
argp[0] = argv[0];
argp[1] = c_flag;
argp[2] = n_flag;
argp[3] = mp_flag;
for(i = 1; i < argc; i++) {
argp[i + 3] = argv[i];
}
argc += 3;
ret = rte_eal_init(argc, argp);
if (ret < 0)
rte_panic("Cannot init EAL\n");
printf("----------- MEMORY_SEGMENTS -----------\n");
rte_dump_physmem_layout(stdout);
printf("--------- END_MEMORY_SEGMENTS ---------\n");
printf("------------ MEMORY_ZONES -------------\n");
rte_memzone_dump(stdout);
printf("---------- END_MEMORY_ZONES -----------\n");
printf("------------- TAIL_QUEUES -------------\n");
rte_dump_tailq(stdout);
printf("---------- END_TAIL_QUEUES ------------\n");
return 0;
}
static void
dpdk_init__(const struct smap *ovs_other_config)
{
char **argv = NULL, **argv_to_release = NULL;
int result;
int argc, argc_tmp;
bool auto_determine = true;
int err = 0;
cpu_set_t cpuset;
char *sock_dir_subcomponent;
if (process_vhost_flags("vhost-sock-dir", xstrdup(ovs_rundir()),
NAME_MAX, ovs_other_config,
&sock_dir_subcomponent)) {
struct stat s;
if (!strstr(sock_dir_subcomponent, "..")) {
vhost_sock_dir = xasprintf("%s/%s", ovs_rundir(),
sock_dir_subcomponent);
err = stat(vhost_sock_dir, &s);
if (err) {
VLOG_ERR("vhost-user sock directory '%s' does not exist.",
vhost_sock_dir);
}
} else {
vhost_sock_dir = xstrdup(ovs_rundir());
VLOG_ERR("vhost-user sock directory request '%s/%s' has invalid"
"characters '..' - using %s instead.",
ovs_rundir(), sock_dir_subcomponent, ovs_rundir());
}
free(sock_dir_subcomponent);
} else {
vhost_sock_dir = sock_dir_subcomponent;
}
argv = grow_argv(&argv, 0, 1);
argc = 1;
argv[0] = xstrdup(ovs_get_program_name());
argc_tmp = get_dpdk_args(ovs_other_config, &argv, argc);
while (argc_tmp != argc) {
if (!strcmp("-c", argv[argc]) || !strcmp("-l", argv[argc])) {
auto_determine = false;
break;
}
argc++;
}
argc = argc_tmp;
/**
* NOTE: This is an unsophisticated mechanism for determining the DPDK
* lcore for the DPDK Master.
*/
if (auto_determine) {
int i;
/* Get the main thread affinity */
CPU_ZERO(&cpuset);
err = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t),
&cpuset);
if (!err) {
for (i = 0; i < CPU_SETSIZE; i++) {
if (CPU_ISSET(i, &cpuset)) {
argv = grow_argv(&argv, argc, 2);
argv[argc++] = xstrdup("-c");
argv[argc++] = xasprintf("0x%08llX", (1ULL<<i));
i = CPU_SETSIZE;
}
}
} else {
VLOG_ERR("Thread getaffinity error %d. Using core 0x1", err);
/* User did not set dpdk-lcore-mask and unable to get current
* thread affintity - default to core 0x1 */
argv = grow_argv(&argv, argc, 2);
argv[argc++] = xstrdup("-c");
argv[argc++] = xasprintf("0x%X", 1);
}
}
argv = grow_argv(&argv, argc, 1);
argv[argc] = NULL;
optind = 1;
if (VLOG_IS_INFO_ENABLED()) {
struct ds eal_args;
int opt;
ds_init(&eal_args);
ds_put_cstr(&eal_args, "EAL ARGS:");
for (opt = 0; opt < argc; ++opt) {
ds_put_cstr(&eal_args, " ");
ds_put_cstr(&eal_args, argv[opt]);
}
VLOG_INFO("%s", ds_cstr_ro(&eal_args));
ds_destroy(&eal_args);
}
argv_to_release = grow_argv(&argv_to_release, 0, argc);
for (argc_tmp = 0; argc_tmp < argc; ++argc_tmp) {
argv_to_release[argc_tmp] = argv[argc_tmp];
}
/* Make sure things are initialized ... */
result = rte_eal_init(argc, argv);
if (result < 0) {
ovs_abort(result, "Cannot init EAL");
}
argv_release(argv, argv_to_release, argc);
/* Set the main thread affinity back to pre rte_eal_init() value */
if (auto_determine && !err) {
err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t),
&cpuset);
if (err) {
VLOG_ERR("Thread setaffinity error %d", err);
}
}
rte_memzone_dump(stdout);
/* We are called from the main thread here */
RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID;
#ifdef DPDK_PDUMP
VLOG_INFO("DPDK pdump packet capture enabled");
err = rte_pdump_init(ovs_rundir());
if (err) {
VLOG_INFO("Error initialising DPDK pdump");
rte_pdump_uninit();
} else {
char *server_socket_path;
server_socket_path = xasprintf("%s/%s", ovs_rundir(),
"pdump_server_socket");
fatal_signal_add_file_to_unlink(server_socket_path);
free(server_socket_path);
}
#endif
/* Finally, register the dpdk classes */
netdev_dpdk_register();
}
void
spdk_memzone_dump(FILE *f)
{
rte_memzone_dump(f);
}
static int
test_memzone_reserve_max_aligned(void)
{
const struct rte_memzone *mz;
const struct rte_config *config;
const struct rte_memseg *ms;
int memseg_idx = 0;
int memzone_idx = 0;
uintptr_t addr_offset;
size_t len = 0;
void* last_addr;
size_t maxlen = 0;
/* random alignment */
rte_srand((unsigned)rte_rdtsc());
const unsigned align = 1 << ((rte_rand() % 8) + 5); /* from 128 up to 4k alignment */
/* get pointer to global configuration */
config = rte_eal_get_configuration();
ms = rte_eal_get_physmem_layout();
addr_offset = 0;
for (memseg_idx = 0; memseg_idx < RTE_MAX_MEMSEG; memseg_idx++){
/* ignore smaller memsegs as they can only get smaller */
if (ms[memseg_idx].len < maxlen)
continue;
/* align everything */
last_addr = RTE_PTR_ALIGN_CEIL(ms[memseg_idx].addr, RTE_CACHE_LINE_SIZE);
len = ms[memseg_idx].len - RTE_PTR_DIFF(last_addr, ms[memseg_idx].addr);
len &= ~((size_t) RTE_CACHE_LINE_MASK);
/* cycle through all memzones */
for (memzone_idx = 0; memzone_idx < RTE_MAX_MEMZONE; memzone_idx++) {
/* stop when reaching last allocated memzone */
if (config->mem_config->memzone[memzone_idx].addr == NULL)
break;
/* check if the memzone is in our memseg and subtract length */
if ((config->mem_config->memzone[memzone_idx].addr >=
ms[memseg_idx].addr) &&
(config->mem_config->memzone[memzone_idx].addr <
(RTE_PTR_ADD(ms[memseg_idx].addr, ms[memseg_idx].len)))) {
/* since the zones can now be aligned and occasionally skip
* some space, we should calculate the length based on
* reported length and start addresses difference.
*/
len -= (uintptr_t) RTE_PTR_SUB(
config->mem_config->memzone[memzone_idx].addr,
(uintptr_t) last_addr);
len -= config->mem_config->memzone[memzone_idx].len;
last_addr =
RTE_PTR_ADD(config->mem_config->memzone[memzone_idx].addr,
(size_t) config->mem_config->memzone[memzone_idx].len);
}
}
/* make sure we get the alignment offset */
if (len > maxlen) {
addr_offset = RTE_PTR_ALIGN_CEIL((uintptr_t) last_addr, align) - (uintptr_t) last_addr;
maxlen = len;
}
}
if (maxlen == 0 || maxlen == addr_offset) {
printf("There is no space left for biggest %u-aligned memzone!\n", align);
return 0;
}
maxlen -= addr_offset;
mz = rte_memzone_reserve_aligned("max_zone_aligned", 0,
SOCKET_ID_ANY, 0, align);
if (mz == NULL){
printf("Failed to reserve a big chunk of memory\n");
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
if (mz->len != maxlen) {
printf("Memzone reserve with 0 size and alignment %u did not return"
" bigest block\n", align);
printf("Expected size = %zu, actual size = %zu\n",
maxlen, mz->len);
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
return 0;
}
static int
test_memzone_reserve_max(void)
{
const struct rte_memzone *mz;
const struct rte_config *config;
const struct rte_memseg *ms;
int memseg_idx = 0;
int memzone_idx = 0;
size_t len = 0;
void* last_addr;
size_t maxlen = 0;
/* get pointer to global configuration */
config = rte_eal_get_configuration();
ms = rte_eal_get_physmem_layout();
for (memseg_idx = 0; memseg_idx < RTE_MAX_MEMSEG; memseg_idx++){
/* ignore smaller memsegs as they can only get smaller */
if (ms[memseg_idx].len < maxlen)
continue;
/* align everything */
last_addr = RTE_PTR_ALIGN_CEIL(ms[memseg_idx].addr, RTE_CACHE_LINE_SIZE);
len = ms[memseg_idx].len - RTE_PTR_DIFF(last_addr, ms[memseg_idx].addr);
len &= ~((size_t) RTE_CACHE_LINE_MASK);
/* cycle through all memzones */
for (memzone_idx = 0; memzone_idx < RTE_MAX_MEMZONE; memzone_idx++) {
/* stop when reaching last allocated memzone */
if (config->mem_config->memzone[memzone_idx].addr == NULL)
break;
/* check if the memzone is in our memseg and subtract length */
if ((config->mem_config->memzone[memzone_idx].addr >=
ms[memseg_idx].addr) &&
(config->mem_config->memzone[memzone_idx].addr <
(RTE_PTR_ADD(ms[memseg_idx].addr, ms[memseg_idx].len)))) {
/* since the zones can now be aligned and occasionally skip
* some space, we should calculate the length based on
* reported length and start addresses difference. Addresses
* are allocated sequentially so we don't need to worry about
* them being in the right order.
*/
len -= RTE_PTR_DIFF(
config->mem_config->memzone[memzone_idx].addr,
last_addr);
len -= config->mem_config->memzone[memzone_idx].len;
last_addr = RTE_PTR_ADD(config->mem_config->memzone[memzone_idx].addr,
(size_t) config->mem_config->memzone[memzone_idx].len);
}
}
/* we don't need to calculate offset here since length
* is always cache-aligned */
if (len > maxlen)
maxlen = len;
}
if (maxlen == 0) {
printf("There is no space left!\n");
return 0;
}
mz = rte_memzone_reserve("max_zone", 0, SOCKET_ID_ANY, 0);
if (mz == NULL){
printf("Failed to reserve a big chunk of memory\n");
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
if (mz->len != maxlen) {
printf("Memzone reserve with 0 size did not return bigest block\n");
printf("Expected size = %zu, actual size = %zu\n",
maxlen, mz->len);
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
return 0;
}
static int
test_memzone(void)
{
const struct rte_memzone *memzone1;
const struct rte_memzone *memzone2;
const struct rte_memzone *memzone3;
const struct rte_memzone *memzone4;
const struct rte_memzone *mz;
memzone1 = rte_memzone_reserve("testzone1", 100,
SOCKET_ID_ANY, 0);
memzone2 = rte_memzone_reserve("testzone2", 1000,
0, 0);
memzone3 = rte_memzone_reserve("testzone3", 1000,
1, 0);
memzone4 = rte_memzone_reserve("testzone4", 1024,
SOCKET_ID_ANY, 0);
/* memzone3 may be NULL if we don't have NUMA */
if (memzone1 == NULL || memzone2 == NULL || memzone4 == NULL)
return -1;
rte_memzone_dump(stdout);
/* check cache-line alignments */
printf("check alignments and lengths\n");
if ((memzone1->phys_addr & RTE_CACHE_LINE_MASK) != 0)
return -1;
if ((memzone2->phys_addr & RTE_CACHE_LINE_MASK) != 0)
return -1;
if (memzone3 != NULL && (memzone3->phys_addr & RTE_CACHE_LINE_MASK) != 0)
return -1;
if ((memzone1->len & RTE_CACHE_LINE_MASK) != 0 || memzone1->len == 0)
return -1;
if ((memzone2->len & RTE_CACHE_LINE_MASK) != 0 || memzone2->len == 0)
return -1;
if (memzone3 != NULL && ((memzone3->len & RTE_CACHE_LINE_MASK) != 0 ||
memzone3->len == 0))
return -1;
if (memzone4->len != 1024)
return -1;
/* check that zones don't overlap */
printf("check overlapping\n");
if (is_memory_overlap(memzone1->phys_addr, memzone1->len,
memzone2->phys_addr, memzone2->len))
return -1;
if (memzone3 != NULL &&
is_memory_overlap(memzone1->phys_addr, memzone1->len,
memzone3->phys_addr, memzone3->len))
return -1;
if (memzone3 != NULL &&
is_memory_overlap(memzone2->phys_addr, memzone2->len,
memzone3->phys_addr, memzone3->len))
return -1;
printf("check socket ID\n");
/* memzone2 must be on socket id 0 and memzone3 on socket 1 */
if (memzone2->socket_id != 0)
return -1;
if (memzone3 != NULL && memzone3->socket_id != 1)
return -1;
printf("test zone lookup\n");
mz = rte_memzone_lookup("testzone1");
if (mz != memzone1)
return -1;
printf("test duplcate zone name\n");
mz = rte_memzone_reserve("testzone1", 100,
SOCKET_ID_ANY, 0);
if (mz != NULL)
return -1;
printf("test reserving memzone with bigger size than the maximum\n");
if (test_memzone_reserving_zone_size_bigger_than_the_maximum() < 0)
return -1;
printf("test reserving memory in smallest segments\n");
if (test_memzone_reserve_memory_in_smallest_segment() < 0)
return -1;
printf("test reserving memory in segments with smallest offsets\n");
if (test_memzone_reserve_memory_with_smallest_offset() < 0)
return -1;
printf("test memzone_reserve flags\n");
if (test_memzone_reserve_flags() < 0)
return -1;
printf("test alignment for memzone_reserve\n");
if (test_memzone_aligned() < 0)
return -1;
printf("test boundary alignment for memzone_reserve\n");
if (test_memzone_bounded() < 0)
return -1;
printf("test invalid alignment for memzone_reserve\n");
if (test_memzone_invalid_alignment() < 0)
return -1;
printf("test reserving amounts of memory equal to segment's length\n");
if (test_memzone_reserve_remainder() < 0)
return -1;
printf("test reserving the largest size memzone possible\n");
if (test_memzone_reserve_max() < 0)
return -1;
printf("test reserving the largest size aligned memzone possible\n");
if (test_memzone_reserve_max_aligned() < 0)
return -1;
return 0;
}
