/*
* Read memory statistics data
*/
static void get_paging_data(struct paging_data *pd)
{
static const char *key_pgpgin = "pgpgin";
static const char *key_nr_free_pages = "nr_free_pages";
static const char parse_error_msg[] = "error parsing /proc/vmstat";
char *cp, *xp;
bool found_pgpgin = false;
bool found_nr_free_pages = false;
int nfound = 0;
const int need_found = 2;
pd->ts = getnow();
pd->mem_pages = sysconf(_SC_PHYS_PAGES);
if (pd->mem_pages <= 0)
fatal_msg("unable to get system memory size");
read_whole_file("/proc/vmstat", &vmstat, &vmstat_size, 4096);
for (cp = vmstat; *cp != '\0';)
{
if (is_key(&cp, key_pgpgin))
{
if (!consume_umax(&cp, &pd->pgpgin))
fatal_msg(parse_error_msg);
if (!found_pgpgin)
{
found_pgpgin = true;
if (need_found == ++nfound)
break;
}
}
else if (is_key(&cp, key_nr_free_pages))
{
if (!consume_umax(&cp, &pd->nr_free_pages))
fatal_msg(parse_error_msg);
if (!found_nr_free_pages)
{
found_nr_free_pages = true;
if (need_found == ++nfound)
break;
}
}
else
{
xp = strchr(cp, '\n');
if (!xp) break;
cp = xp + 1;
}
}
if (need_found != nfound)
fatal_msg(parse_error_msg);
}
static void _handle_touch(struct cxl_afu_h *afu, uint64_t addr, uint8_t size)
{
uint8_t buffer;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_handle_touch");
if (!_testmemaddr((uint8_t *) addr)) {
if (_handle_dsi(afu, addr) < 0) {
perror("DSI Failure");
return;
}
DPRINTF("TOUCH of invalid addr @ 0x%016" PRIx64 "\n", addr);
buffer = (uint8_t) PSLSE_MEM_FAILURE;
if (put_bytes_silent(afu->fd, 1, &buffer) != 1) {
afu->opened = 0;
afu->attached = 0;
}
return;
}
buffer = PSLSE_MEM_SUCCESS;
if (put_bytes_silent(afu->fd, 1, &buffer) != 1) {
afu->opened = 0;
afu->attached = 0;
}
DPRINTF("TOUCH of addr @ 0x%016" PRIx64 "\n", addr);
}
static void _handle_read(struct cxl_afu_h *afu, uint64_t addr, uint8_t size)
{
uint8_t buffer[MAX_LINE_CHARS];
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_handle_read");
if (!_testmemaddr((uint8_t *) addr)) {
if (_handle_dsi(afu, addr) < 0) {
perror("DSI Failure");
return;
}
DPRINTF("READ from invalid addr @ 0x%016" PRIx64 "\n", addr);
buffer[0] = (uint8_t) PSLSE_MEM_FAILURE;
if (put_bytes_silent(afu->fd, 1, buffer) != 1) {
afu->opened = 0;
afu->attached = 0;
}
return;
}
buffer[0] = PSLSE_MEM_SUCCESS;
memcpy(&(buffer[1]), (void *)addr, size);
if (put_bytes_silent(afu->fd, size + 1, buffer) != size + 1) {
afu->opened = 0;
afu->attached = 0;
}
DPRINTF("READ from addr @ 0x%016" PRIx64 "\n", addr);
}
static void _handle_ack(struct cxl_afu_h *afu)
{
uint8_t data[sizeof(uint64_t)];
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_handle_ack");
DPRINTF("MMIO ACK\n");
if ((afu->mmio.type == PSLSE_MMIO_READ64)| (afu->mmio.type == PSLSE_MMIO_EBREAD)) {
if (get_bytes_silent(afu->fd, sizeof(uint64_t), data, 1000, 0) <
0) {
warn_msg("Socket failure getting MMIO Ack");
_all_idle(afu);
afu->mmio.data = 0xFEEDB00FFEEDB00FL;
} else {
memcpy(&(afu->mmio.data), data, sizeof(uint64_t));
afu->mmio.data = ntohll(afu->mmio.data);
}
}
if (afu->mmio.type == PSLSE_MMIO_READ32) {
if (get_bytes_silent(afu->fd, sizeof(uint32_t), data, 1000, 0) <
0) {
warn_msg("Socket failure getting MMIO Read 32 data");
afu->mmio.data = 0xFEEDB00FL;
_all_idle(afu);
} else {
memcpy(&(afu->mmio.data), data, sizeof(uint32_t));
debug_msg("KEM:0x%08x", afu->mmio.data);
afu->mmio.data = ntohl(afu->mmio.data);
debug_msg("KEM:0x%08x", afu->mmio.data);
}
}
afu->mmio.state = LIBCXL_REQ_IDLE;
}
static int _handle_dsi(struct cxl_afu_h *afu, uint64_t addr)
{
uint16_t size;
int i;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_handle_dsi");
// Only track a single DSI at a time
pthread_mutex_lock(&(afu->event_lock));
i = 0;
while (afu->events[i] != NULL) {
if (afu->events[i]->header.type == CXL_EVENT_DATA_STORAGE) {
pthread_mutex_unlock(&(afu->event_lock));
return 0;
}
++i;
}
assert(i < EVENT_QUEUE_MAX);
size = sizeof(struct cxl_event_header) +
sizeof(struct cxl_event_data_storage);
afu->events[i] = (struct cxl_event *)calloc(1, size);
afu->events[i]->header.type = CXL_EVENT_DATA_STORAGE;
afu->events[i]->header.size = size;
afu->events[i]->header.process_element = afu->context;
afu->events[i]->fault.addr = addr & FOURK_MASK;
afu->events[i]->fault.dsisr = DSISR;
do {
i = write(afu->pipe[1], &(afu->events[i]->header.type), 1);
}
while ((i == 0) || (errno == EINTR));
pthread_mutex_unlock(&(afu->event_lock));
return i;
}
static __inline__
void *safe_calloc(unsigned int size, char *file, unsigned int line)
{
void *res = calloc(1, size);
if (res == NULL) fatal_msg("CALLOC failed", file, line);
return res;
}
static void _mmio_read(struct cxl_afu_h *afu)
{
uint8_t *buffer;
uint32_t addr;
int size, offset;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_mmio_read");
size = 1 + sizeof(addr);
buffer = (uint8_t *) malloc(size);
buffer[0] = afu->mmio.type;
offset = 1;
addr = htonl(afu->mmio.addr);
memcpy((char *)&(buffer[offset]), (char *)&addr, sizeof(addr));
if (put_bytes_silent(afu->fd, size, buffer) != size) {
warn_msg("_mmio_read: put_bytes_silent failed");
free(buffer);
close_socket(&(afu->fd));
afu->opened = 0;
afu->attached = 0;
afu->mmio.state = LIBCXL_REQ_IDLE;
afu->mmio.data = 0xFEEDB00FFEEDB00FL;
return;
}
free(buffer);
afu->mmio.state = LIBCXL_REQ_PENDING;
}
/*
* Commit current xenstore transaction.
*
* Returns:
* XSTS_OK success
* XSTS_RETRY retry the transaction
* XSTS_NORETRY out of retry limit (message logged)
* XSTS_FAIL commit failed (message logged)
*/
static XsTransactionStatus commit_xs(int* p_nretries)
{
if (xst == XS_TRANSACTION_NULL)
fatal_msg("bug: commit_xs outside of a transaction");
if (xs_transaction_end(xs, xst, FALSE))
{
debug_msg(15, "committed transaction");
xst = XS_TRANSACTION_NULL;
return XSTS_OK;
}
xst = XS_TRANSACTION_NULL;
if (errno == EAGAIN)
{
if (++(*p_nretries) > max_xs_retries)
{
error_perror("unable to write xenstore (transaction retry limit exceeded)");
return XSTS_NORETRY;
}
retry_wait(*p_nretries);
debug_msg(2, "restarting xenstore transaction");
return XSTS_RETRY;
}
else
{
error_perror("unable to write xenstore (xs_transaction_end)");
return XSTS_FAIL;
}
}
static void _pslse_attach(struct cxl_afu_h *afu)
{
uint8_t *buffer;
uint64_t *wed_ptr;
int size, offset;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_pslse_attach");
size = 1 + sizeof(uint64_t);
buffer = (uint8_t *) malloc(size);
buffer[0] = PSLSE_ATTACH;
offset = 1;
wed_ptr = (uint64_t *) & (buffer[offset]);
*wed_ptr = htonll(afu->attach.wed);
if (put_bytes_silent(afu->fd, size, buffer) != size) {
free(buffer);
close_socket(&(afu->fd));
afu->opened = 0;
afu->attached = 0;
afu->attach.state = LIBCXL_REQ_IDLE;
return;
}
free(buffer);
afu->attach.state = LIBCXL_REQ_PENDING;
}
/*
* Process expected signals
*/
static void handle_signals(struct pollfd *pollfds, bool *p_recheck_time)
{
struct signalfd_siginfo fdsi;
ssize_t sz;
/* SIGTERM - exit gracefully */
if (pollfds[NPFD_SIGTERM].revents & (POLLIN|POLLPRI))
{
shutdown_xs();
notice_msg("terminating...");
exit(EXIT_SUCCESS);
}
/* SIGHUP - reload configuration */
if (pollfds[NPFD_SIGHUP].revents & (POLLIN|POLLPRI))
{
sz = read(fd_sighup, &fdsi, sizeof(fdsi));
if (sz < 0)
fatal_perror("read signalfd");
if (sz != sizeof(fdsi))
fatal_msg("read signalfd");
/* reload configuration -- currently none for memprobed */
if (p_recheck_time)
*p_recheck_time = true;
}
}
static void _mmio_write32(struct cxl_afu_h *afu)
{
uint8_t *buffer;
uint32_t data;
uint32_t addr;
int size, offset;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_mmio_write32");
size = 1 + sizeof(addr) + sizeof(data);
buffer = (uint8_t *) malloc(size);
buffer[0] = PSLSE_MMIO_WRITE32;
offset = 1;
addr = htonl(afu->mmio.addr);
memcpy((char *)&(buffer[offset]), (char *)&addr, sizeof(addr));
offset += sizeof(addr);
data = htonl(afu->mmio.data);
memcpy((char *)&(buffer[offset]), (char *)&data, sizeof(data));
if (put_bytes_silent(afu->fd, size, buffer) != size) {
free(buffer);
close_socket(&(afu->fd));
afu->opened = 0;
afu->attached = 0;
afu->mmio.state = LIBCXL_REQ_IDLE;
return;
}
free(buffer);
afu->mmio.state = LIBCXL_REQ_PENDING;
}
static void _mmio_map(struct cxl_afu_h *afu)
{
uint8_t *buffer;
uint32_t *flags_ptr;
uint32_t flags;
int size;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_mmio_map");
size = 1 + sizeof(uint32_t);
buffer = (uint8_t *) malloc(size);
buffer[0] = PSLSE_MMIO_MAP;
flags = (uint32_t) afu->mmio.data;
flags_ptr = (uint32_t *) & (buffer[1]);
*flags_ptr = htonl(flags);
if (put_bytes_silent(afu->fd, size, buffer) != size) {
free(buffer);
close_socket(&(afu->fd));
afu->opened = 0;
afu->attached = 0;
afu->mmio.state = LIBCXL_REQ_IDLE;
return;
}
free(buffer);
afu->mmio.state = LIBCXL_REQ_PENDING;
}
/*
* Verify we are running under Xen,
* and not in the root domain
*/
static void verify_hypervisor(void)
{
FILE *fp = NULL;
char buf[256];
char *p;
const static char root_uuid[] = "00000000-0000-0000-0000-000000000000";
const char *rp;
/*
* Check that file /sys/hypervisor/type is present
* and contains "xen"
*/
if (access("/sys/hypervisor/type", F_OK) == -1 &&
(errno == ENOENT || errno == ENOTDIR))
fatal_msg("not running under a hypervisor");
fp = fopen("/sys/hypervisor/type", "r");
if (fp == NULL || NULL == fgets(buf, countof(buf), fp))
fatal_msg("unable to read /sys/hypervisor/type");
buf[countof(buf) - 1] = '\0';
p = strchr(buf, '\n');
if (p) *p = '\0';
if (0 != strcasecmp(buf, "xen"))
fatal_msg("hypervisor is not Xen");
fclose(fp);
/*
* Read Xen domain UUID from /sys/hypervisor/uuid
* and check it is not the root domain
*/
fp = fopen("/sys/hypervisor/uuid", "r");
if (fp == NULL || NULL == fgets(buf, countof(buf), fp))
fatal_msg("unable to read /sys/hypervisor/uuid");
buf[countof(buf) - 1] = '\0';
p = strchr(buf, '\n');
if (p) *p = '\0';
if (0 == strcmp(buf, root_uuid))
fatal_msg("may run only in a VM, not root domain");
fclose(fp);
if (strlen(buf) != strlen(root_uuid) /* ||
strlen(buf) != UUID_STRING_SIZE - 1 */)
fatal_msg("unexpected domain uuid");
for (p = buf, rp = root_uuid; *rp; p++, rp++)
{
if (*rp == '-' && *p == '-') continue;
if (isxdigit(*rp) && isxdigit(*p)) continue;
fatal_msg("unexpected domain uuid");
}
// strcpy(vm_uuid, buf);
}
static void _all_idle(struct cxl_afu_h *afu)
{
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_all_idle");
afu->int_req.state = LIBCXL_REQ_IDLE;
afu->open.state = LIBCXL_REQ_IDLE;
afu->attach.state = LIBCXL_REQ_IDLE;
afu->mmio.state = LIBCXL_REQ_IDLE;
afu->mapped = 0;
afu->attached = 0;
afu->opened = 0;
}
static int _handle_afu_error(struct cxl_afu_h *afu)
{
uint64_t error;
uint16_t size;
uint8_t data[sizeof(error)];
int i;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_handle_afu_error");
DPRINTF("AFU ERROR\n");
if (get_bytes_silent(afu->fd, sizeof(error), data, 1000, 0) < 0) {
warn_msg("Socket failure getting AFU ERROR");
_all_idle(afu);
return -1;
}
memcpy(&error, data, sizeof(error));
error = ntohll(error);
// Only track a single AFU error at a time
pthread_mutex_lock(&(afu->event_lock));
i = 0;
while (afu->events[i] != NULL) {
if (afu->events[i]->header.type == CXL_EVENT_AFU_ERROR) {
pthread_mutex_unlock(&(afu->event_lock));
return 0;
}
++i;
}
assert(i < EVENT_QUEUE_MAX);
size = sizeof(struct cxl_event_header) +
sizeof(struct cxl_event_afu_error);
afu->events[i] = (struct cxl_event *)calloc(1, size);
afu->events[i]->header.type = CXL_EVENT_AFU_ERROR;
afu->events[i]->header.size = size;
afu->events[i]->header.process_element = afu->context;
afu->events[i]->afu_error.error = error;
do {
i = write(afu->pipe[1], &(afu->events[i]->header.type), 1);
}
while ((i == 0) || (errno == EINTR));
pthread_mutex_unlock(&(afu->event_lock));
return i;
}
/*
* Select preferred interval clock among available ones
*/
static void select_clk_id(void)
{
struct timespec ts;
clk_id = CLOCK_BOOTTIME;
if (0 == clock_gettime(clk_id, &ts))
return;
clk_id = CLOCK_MONOTONIC_RAW;
if (0 == clock_gettime(clk_id, &ts))
return;
clk_id = CLOCK_MONOTONIC;
if (0 == clock_gettime(clk_id, &ts))
return;
fatal_msg("unable to select clk_id");
}
static int _handle_interrupt(struct cxl_afu_h *afu)
{
uint16_t size, irq;
uint8_t data[sizeof(irq)];
int i;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_handle_interrupt");
DPRINTF("AFU INTERRUPT\n");
if (get_bytes_silent(afu->fd, sizeof(irq), data, 1000, 0) < 0) {
warn_msg("Socket failure getting IRQ");
_all_idle(afu);
return -1;
}
memcpy(&irq, data, sizeof(irq));
irq = ntohs(irq);
// Only track a single interrupt at a time
pthread_mutex_lock(&(afu->event_lock));
i = 0;
while (afu->events[i] != NULL) {
if (afu->events[i]->header.type == CXL_EVENT_AFU_INTERRUPT) {
pthread_mutex_unlock(&(afu->event_lock));
return 0;
}
++i;
}
assert(i < EVENT_QUEUE_MAX);
size = sizeof(struct cxl_event_header) +
sizeof(struct cxl_event_afu_interrupt);
afu->events[i] = (struct cxl_event *)calloc(1, size);
afu->events[i]->header.type = CXL_EVENT_AFU_INTERRUPT;
afu->events[i]->header.size = size;
afu->events[i]->header.process_element = afu->context;
afu->events[i]->irq.irq = irq;
do {
i = write(afu->pipe[1], &(afu->events[i]->header.type), 1);
}
while ((i == 0) || (errno == EINTR));
pthread_mutex_unlock(&(afu->event_lock));
return i;
}
/*
* Start XenStore transaction.
* If successful, return @true.
* On error, return @false and log error message. .
*/
static bool begin_xs(void)
{
int nretries = 0;
if (xst != XS_TRANSACTION_NULL)
fatal_msg("bug: begin_xs inside a transaction");
for (;;)
{
xst = xs_transaction_start(xs);
if (xst != XS_TRANSACTION_NULL)
break;
/*
* ENOSPC means too many concurrent transactions from the domain
*/
if (errno == ENOSPC)
{
if (++nretries > max_xs_retries)
{
error_perror("unable to start xenstore transaction (retry limit exceeded)");
break;
}
retry_wait(nretries);
debug_msg(2, "retrying to start xenstore transaction");
}
else
{
error_perror("unable to start xenstore transaction");
break;
}
}
if (xst != XS_TRANSACTION_NULL)
{
debug_msg(15, "started transaction");
return true;
}
else
{
return false;
}
}
/*
* Socket file with the same name already exists.
* If another instance of membalanced is already running, abort.
* Otherwise delete leftover file, and let the caller retry again.
*/
static void on_addrinuse(const struct sockaddr_un& addr)
{
int fd;
int rc;
DO_RESTARTABLE(fd, socket(AF_UNIX, SOCK_STREAM, 0));
if (fd == -1)
fatal_perror("unable to create RPC socket");
DO_RESTARTABLE(rc, connect(fd, (struct sockaddr*)&addr, sizeof addr));
if (rc == 0)
fatal_msg("membalanced is already running and owns RPC socket");
if (errno != ECONNREFUSED)
fatal_perror("socket file %s in a bad state", socket_path);
/* socket exists, but no one is lisening on it */
if (unlink(socket_path) && errno != ENOENT)
fatal_perror("unable to delete socket file %s", socket_path);
}
static void _release_adapters(struct cxl_adapter_h *adapter)
{
struct cxl_adapter_h *current;
uint8_t rc = PSLSE_DETACH;
if (!adapter)
fatal_msg("NULL adapter passed to libcxl.c:_release_adapters");
_release_afus(adapter->afu_list);
current = adapter;
while (current != NULL) {
adapter = current;
current = current->_next;
// Disconnect from PSLSE
if (adapter->fd) {
put_bytes_silent(adapter->fd, 1, &rc);
close_socket(&(adapter->fd));
}
free(adapter->id);
free(adapter);
}
}
/*
* Log/print error message (incl. errno) and terminate
*/
static void fatal_perror(const char *fmt, ...) /* __noreturn__ __format_printf__ */
{
char *errno_msg;
char errno_msg_buf[128];
char msg[512];
int size = 0;
va_list ap;
errno_msg = strerror_r(errno, errno_msg_buf, countof(errno_msg_buf));
strcpy(msg, "fatal error: ");
size = strlen(msg);
va_start(ap, fmt);
vsnprintf(msg + size, countof(msg) - size, fmt, ap);
va_end(ap);
size = strlen(msg);
snprintf(msg + size, countof(msg) - size, ": %s", errno_msg);
fatal_msg("%s", msg);
}
static void _req_max_int(struct cxl_afu_h *afu)
{
uint8_t *buffer;
int size;
uint16_t value;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_req_max_int");
size = 1 + sizeof(uint16_t);
buffer = (uint8_t *) malloc(size);
buffer[0] = PSLSE_MAX_INT;
value = htons(afu->int_req.max);
memcpy((char *)&(buffer[1]), (char *)&value, sizeof(uint16_t));
if (put_bytes_silent(afu->fd, size, buffer) != size) {
free(buffer);
close_socket(&(afu->fd));
afu->int_req.max = 0;
_all_idle(afu);
return;
}
free(buffer);
afu->int_req.state = LIBCXL_REQ_PENDING;
}
// Send a randomly selected pending response back to AFU
void handle_response(struct cmd *cmd)
{
struct cmd_event **head;
struct cmd_event *event;
struct client *client;
int rc;
// Select a random pending response (or none)
client = NULL;
head = &cmd->list;
while (*head != NULL) {
// Fast track error responses
if (((*head)->resp == PSL_RESPONSE_PAGED) ||
((*head)->resp == PSL_RESPONSE_NRES) ||
((*head)->resp == PSL_RESPONSE_NLOCK) ||
((*head)->resp == PSL_RESPONSE_FAILED) ||
((*head)->resp == PSL_RESPONSE_FLUSHED)) {
event = *head;
goto drive_resp;
}
if (((*head)->state == MEM_DONE) && !allow_reorder(cmd->parms)) {
break;
}
head = &((*head)->_next);
}
// Randomly decide not to drive response yet
event = *head;
if ((event == NULL) || ((event->client_state == CLIENT_VALID)
&& !allow_resp(cmd->parms))) {
return;
}
// Test for client disconnect
if ((event == NULL) || ((client = _get_client(cmd, event)) == NULL))
return;
// Send response, remove command from list and free memory
if ((event->resp == PSL_RESPONSE_PAGED) ||
(event->resp == PSL_RESPONSE_AERROR) ||
(event->resp == PSL_RESPONSE_DERROR)) {
client->flushing = FLUSH_FLUSHING;
_update_pending_resps(cmd, PSL_RESPONSE_FLUSHED);
}
drive_resp:
// Check for pending buffer activity
if (event == cmd->buffer_read) {
fatal_msg("Driving response when buffer read still active");
_print_event(event);
assert(event != cmd->buffer_read);
}
rc = psl_response(cmd->afu_event, event->tag, event->resp, 1, 0, 0);
if (rc == PSL_SUCCESS) {
debug_msg("%s:RESPONSE tag=0x%02x code=0x%x", cmd->afu_name,
event->tag, event->resp);
debug_cmd_response(cmd->dbg_fp, cmd->dbg_id, event->tag);
if ((client != NULL) && (event->command == PSL_COMMAND_RESTART))
client->flushing = FLUSH_NONE;
*head = event->_next;
free(event->data);
free(event->parity);
free(event);
cmd->credits++;
}
}
static void *_psl_loop(void *ptr)
{
struct cxl_afu_h *afu = (struct cxl_afu_h *)ptr;
uint8_t buffer[MAX_LINE_CHARS];
uint8_t size;
uint64_t addr;
uint16_t value;
uint32_t lvalue;
int rc;
if (!afu)
fatal_msg("NULL afu passed to libcxl.c:_psl_loop");
afu->opened = 1;
while (afu->opened) {
_delay_1ms();
// Send any requests to PSLSE over socket
if (afu->int_req.state == LIBCXL_REQ_REQUEST)
_req_max_int(afu);
if (afu->attach.state == LIBCXL_REQ_REQUEST)
_pslse_attach(afu);
if (afu->mmio.state == LIBCXL_REQ_REQUEST) {
switch (afu->mmio.type) {
case PSLSE_MMIO_MAP:
_mmio_map(afu);
break;
case PSLSE_MMIO_WRITE64:
_mmio_write64(afu);
break;
case PSLSE_MMIO_WRITE32:
_mmio_write32(afu);
break;
case PSLSE_MMIO_EBREAD:
case PSLSE_MMIO_READ64:
case PSLSE_MMIO_READ32: /*fall through */
_mmio_read(afu);
break;
default:
break;
}
}
// Process socket input from PSLSE
rc = bytes_ready(afu->fd, 1000, 0);
if (rc == 0)
continue;
if (rc < 0) {
warn_msg("Socket failure testing bytes_ready");
_all_idle(afu);
break;
}
if (get_bytes_silent(afu->fd, 1, buffer, 1000, 0) < 0) {
warn_msg("Socket failure getting PSL event");
_all_idle(afu);
break;
}
DPRINTF("PSL EVENT\n");
switch (buffer[0]) {
case PSLSE_OPEN:
if (get_bytes_silent(afu->fd, 1, buffer, 1000, 0) < 0) {
warn_msg("Socket failure getting OPEN context");
_all_idle(afu);
break;
}
afu->context = (uint16_t) buffer[0];
afu->open.state = LIBCXL_REQ_IDLE;
break;
case PSLSE_ATTACH:
afu->attach.state = LIBCXL_REQ_IDLE;
break;
case PSLSE_DETACH:
info_msg("detach response from from pslse");
afu->mapped = 0;
afu->attached = 0;
afu->opened = 0;
afu->open.state = LIBCXL_REQ_IDLE;
afu->attach.state = LIBCXL_REQ_IDLE;
afu->mmio.state = LIBCXL_REQ_IDLE;
afu->int_req.state = LIBCXL_REQ_IDLE;
break;
case PSLSE_MAX_INT:
size = sizeof(uint16_t);
if (get_bytes_silent(afu->fd, size, buffer, 1000, 0) <
0) {
warn_msg
("Socket failure getting max interrupt acknowledge");
_all_idle(afu);
break;
}
memcpy((char *)&value, (char *)buffer,
sizeof(uint16_t));
afu->irqs_max = ntohs(value);
afu->int_req.state = LIBCXL_REQ_IDLE;
break;
case PSLSE_QUERY: {
size = sizeof(uint16_t) + sizeof(uint16_t) + sizeof(uint16_t) +
sizeof(uint64_t) + sizeof(uint64_t) + sizeof(uint64_t) +
sizeof(uint16_t) + sizeof(uint16_t) + sizeof(uint32_t);
if (get_bytes_silent(afu->fd, size, buffer, 1000, 0) <
0) {
warn_msg("Socket failure getting PSLSE query");
_all_idle(afu);
break;
}
memcpy((char *)&value, (char *)&(buffer[0]), 2);
afu->irqs_min = (long)(value);
memcpy((char *)&value, (char *)&(buffer[2]), 2);
afu->irqs_max = (long)(value);
memcpy((char *)&value, (char *)&(buffer[4]), 2);
afu->modes_supported = (long)(value);
memcpy((char *)&value, (char *)&(buffer[6]), 8);
afu->mmio_len = (long)(value);
memcpy((char *)&value, (char *)&(buffer[14]), 8);
afu->mmio_off = (long)(value);
memcpy((char *)&value, (char *)&(buffer[22]), 8);
//afu->eb_len = (long)ntohll(value);
afu->eb_len = (long)(value);
memcpy((char *)&value, (char *)&(buffer[30]), 2);
afu->cr_device = (long)ntohs(value);
memcpy((char *)&value, (char *)&(buffer[32]), 2);
afu->cr_vendor = (long)ntohs(value);
memcpy((char *)&lvalue, (char *)&(buffer[34]), 4);
afu->cr_class = ntohl(lvalue);
//no better place to put this right now
afu->prefault_mode = CXL_PREFAULT_MODE_NONE;
break;
}
case PSLSE_MEMORY_READ:
DPRINTF("AFU MEMORY READ\n");
if (get_bytes_silent(afu->fd, 1, buffer, 1000, 0) < 0) {
warn_msg
("Socket failure getting memory read size");
_all_idle(afu);
break;
}
size = (uint8_t) buffer[0];
if (get_bytes_silent(afu->fd, sizeof(uint64_t), buffer,
-1, 0) < 0) {
warn_msg
("Socket failure getting memory read addr");
_all_idle(afu);
break;
}
memcpy((char *)&addr, (char *)buffer, sizeof(uint64_t));
addr = ntohll(addr);
_handle_read(afu, addr, size);
break;
case PSLSE_MEMORY_WRITE:
DPRINTF("AFU MEMORY WRITE\n");
if (get_bytes_silent(afu->fd, 1, buffer, 1000, 0) < 0) {
warn_msg
("Socket failure getting memory write size");
_all_idle(afu);
break;
}
size = (uint8_t) buffer[0];
if (get_bytes_silent(afu->fd, sizeof(uint64_t), buffer,
-1, 0) < 0) {
_all_idle(afu);
break;
}
memcpy((char *)&addr, (char *)buffer, sizeof(uint64_t));
addr = ntohll(addr);
if (get_bytes_silent(afu->fd, size, buffer, 1000, 0) <
0) {
warn_msg
("Socket failure getting memory write data");
_all_idle(afu);
break;
}
_handle_write(afu, addr, size, buffer);
break;
case PSLSE_MEMORY_TOUCH:
DPRINTF("AFU MEMORY TOUCH\n");
if (get_bytes_silent(afu->fd, 1, buffer, 1000, 0) < 0) {
warn_msg
("Socket failure getting memory touch size");
_all_idle(afu);
break;
}
size = buffer[0];
if (get_bytes_silent(afu->fd, sizeof(uint64_t), buffer,
-1, 0) < 0) {
warn_msg
("Socket failure getting memory touch addr");
_all_idle(afu);
break;
}
memcpy((char *)&addr, (char *)buffer, sizeof(uint64_t));
addr = ntohll(addr);
_handle_touch(afu, addr, size);
break;
case PSLSE_MMIO_ACK:
_handle_ack(afu);
break;
case PSLSE_INTERRUPT:
if (_handle_interrupt(afu) < 0) {
perror("Interrupt Failure");
goto psl_fail;
}
break;
case PSLSE_AFU_ERROR:
if (_handle_afu_error(afu) < 0) {
perror("AFU ERROR Failure");
goto psl_fail;
}
break;
default:
break;
}
}
psl_fail:
afu->attached = 0;
pthread_exit(NULL);
}
/*
* Start built-in RPC server
*/
void start_rcmd_server(void)
{
struct sockaddr_un addr;
int rc;
if (sock != -1)
return;
/* ensure socket directory exists */
make_membalance_rundir();
for (;;)
{
/* create socket */
if (sock == -1)
{
DO_RESTARTABLE(sock, socket(AF_UNIX, SOCK_STREAM, 0));
if (sock == -1)
fatal_perror("unable to create RPC socket");
}
/* build address block */
if (strlen(socket_path) >= sizeof(addr.sun_path))
fatal_msg("RPC socket path is too long");
memset(&addr, 0, sizeof(addr));
addr.sun_family = AF_UNIX;
strcpy(addr.sun_path, socket_path);
/* try to bind socket to a name (and create socket file in file system) */
DO_RESTARTABLE(rc, bind(sock, (struct sockaddr*)&addr, sizeof addr));
if (rc != 0)
{
/* socket file already exists? */
if (errno == EADDRINUSE)
{
on_addrinuse(addr);
continue;
}
else
{
fatal_perror("unable to bind RPC socket to %s", socket_path);
}
}
/* successfully created socket file */
sock_unlink = true;
if (chmod(socket_path, S_IRWXU))
{
stop_rcmd_server();
fatal_perror("unable to set protection on socket file %s", socket_path);
}
DO_RESTARTABLE(rc, listen(sock, SOMAXCONN));
if (rc == 0)
break;
stop_rcmd_server();
if (errno == EADDRINUSE)
{
on_addrinuse(addr);
continue;
}
else
{
fatal_perror("unable to listen on socket file %s", socket_path);
}
}
if (!svc_registered)
{
if (!svc_register(NULL, RCMD_MEMBALANCED, RCMD_MEMBALANCED_V1, rcmd_membalanced_1, 0))
fatal_msg("unable to register RPC server (svc_register)");
svc_registered = true;
}
debug_msg(7, "registered RPC server");
}
static struct cxl_afu_h *_pslse_open(int *fd, uint16_t afu_map, uint8_t major,
uint8_t minor, char afu_type)
{
struct cxl_afu_h *afu;
uint8_t *buffer;
uint16_t position;
if (!fd)
fatal_msg("NULL fd passed to libcxl.c:_pslse_open");
position = 0x8000;
position >>= 4 * major;
position >>= minor;
if ((afu_map & position) != position) {
warn_msg("open:AFU not in system");
close_socket(fd);
errno = ENODEV;
return NULL;
}
// Create struct for AFU
afu = _new_afu(afu_map, position, *fd);
if (afu == NULL)
return NULL;
buffer = (uint8_t *) calloc(1, MAX_LINE_CHARS);
buffer[0] = (uint8_t) PSLSE_OPEN;
buffer[1] = afu->dbg_id;
buffer[2] = afu_type;
afu->fd = *fd;
if (put_bytes_silent(afu->fd, 3, buffer) != 3) {
warn_msg("open:Failed to write to socket");
free(buffer);
goto open_fail;
}
free(buffer);
afu->_head = afu;
afu->adapter = major;
afu->id = (char *)malloc(7);
afu->open.state = LIBCXL_REQ_PENDING;
// Start thread
if (pthread_create(&(afu->thread), NULL, _psl_loop, afu)) {
perror("pthread_create");
close_socket(&(afu->fd));
goto open_fail;
}
// Wait for open acknowledgement
while (afu->open.state != LIBCXL_REQ_IDLE) /*infinite loop */
_delay_1ms();
if (!afu->opened) {
pthread_join(afu->thread, NULL);
goto open_fail;
}
sprintf(afu->id, "afu%d.%d", major, minor);
return afu;
open_fail:
pthread_mutex_destroy(&(afu->event_lock));
free(afu);
errno = ENODEV;
return NULL;
}
/*
* Commit a pseudo-transaction that involves just one read or write operation
*/
static XsTransactionStatus commit_singleop_xs(int* p_nretries)
{
if (xst != XS_TRANSACTION_NULL)
fatal_msg("bug: commit_singleop_xs inside a real transaction");
return XSTS_OK;
}
/*
* Abort a pseudo-transaction that involves just one read or write operation
*/
static void abort_singleop_xs(void)
{
if (xst != XS_TRANSACTION_NULL)
fatal_msg("bug: abort_singleop_xs inside a real transaction");
}
/*
* Begin a pseudo-transaction that involves just one read or write operation
*/
static bool begin_singleop_xs(void)
{
if (xst != XS_TRANSACTION_NULL)
fatal_msg("bug: begin_singleop_xs inside a transaction");
return true;
}
/*
* Log/print "out of memory" message and terminate
*/
static void out_of_memory(void) /* __noreturn__ __format_printf__ */
{
fatal_msg("out of memory");
}