/*
* nc_ptr_check: validate that instruction pointer is not out of range.
* If not - advance by number of arguments and fetch specified argument.
*/
static int
nc_ptr_check(uintptr_t *iptr, const void *nc, size_t sz,
u_int nargs, uint32_t *val, u_int r)
{
const uint32_t *tptr = (const uint32_t *)*iptr;
u_int i;
KASSERT(ALIGNED_POINTER(*iptr, uint32_t));
KASSERT(nargs > 0);
if ((uintptr_t)tptr < (uintptr_t)nc)
return NPF_ERR_JUMP;
if ((uintptr_t)tptr + (nargs * sizeof(uint32_t)) > (uintptr_t)nc + sz)
return NPF_ERR_RANGE;
for (i = 1; i <= nargs; i++) {
if (val && i == r) {
*val = *tptr;
}
tptr++;
}
*iptr = (uintptr_t)tptr;
return 0;
}
static void
stack_capture(struct stack *st, struct frame *frame)
{
struct frame *fp;
vm_offset_t callpc;
stack_zero(st);
fp = frame;
for (;;) {
if (!INKERNEL((vm_offset_t)fp) ||
!ALIGNED_POINTER(fp, uint64_t))
break;
callpc = fp->fr_pc;
if (!INKERNEL(callpc))
break;
/* Don't bother traversing trap frames. */
if ((callpc > (uint64_t)tl_trap_begin &&
callpc < (uint64_t)tl_trap_end) ||
(callpc > (uint64_t)tl_text_begin &&
callpc < (uint64_t)tl_text_end))
break;
if (stack_put(st, callpc) == -1)
break;
if (v9next_frame(fp) <= fp ||
v9next_frame(fp) >= frame + KSTACK_PAGES * PAGE_SIZE)
break;
fp = v9next_frame(fp);
}
}
TEST dct_speed(const int width)
{
const int size = width * width;
uint64_t call_cnt = 0;
dct_func * tested_func = test_env.strategy->fptr;
KVZ_CLOCK_T clock_now;
KVZ_GET_TIME(&clock_now);
double test_end = KVZ_CLOCK_T_AS_DOUBLE(clock_now) + TIME_PER_TEST;
int16_t _tmp_residual[32 * 32 + SIMD_ALIGNMENT];
int16_t _tmp_coeffs[32 * 32 + SIMD_ALIGNMENT];
int16_t *tmp_residual = ALIGNED_POINTER(_tmp_residual, SIMD_ALIGNMENT);
int16_t *tmp_coeffs = ALIGNED_POINTER(_tmp_coeffs, SIMD_ALIGNMENT);
// Loop until time allocated for test has passed.
for (unsigned i = 0;
test_end > KVZ_CLOCK_T_AS_DOUBLE(clock_now);
++i)
{
int test = i % NUM_TESTS;
uint64_t sum = 0;
for (int offset = 0; offset < NUM_CHUNKS * 64 * 64; offset += NUM_CHUNKS * size) {
// Compare the first chunk against the 35 other chunks to simulate real usage.
for (int chunk = 0; chunk < NUM_CHUNKS; ++chunk) {
kvz_pixel * buf1 = &bufs[test][offset];
kvz_pixel * buf2 = &bufs[test][chunk * size + offset];
for (int p = 0; p < size; ++p) {
tmp_residual[p] = (int16_t)(buf1[p] - buf2[p]);
}
tested_func(8, tmp_residual, tmp_coeffs);
++call_cnt;
sum += tmp_coeffs[0];
}
}
ASSERT(sum > 0);
KVZ_GET_TIME(&clock_now)
}
sprintf(test_env.msg, "%.3fM x %s:%s",
(double)call_cnt / 1000000.0,
test_env.strategy->type,
test_env.strategy->strategy_name);
PASSm(test_env.msg);
}
/*
* nc_fetch_word: fetch a word (32 bits) from the n-code and increase
* instruction pointer by one word.
*/
static inline const void *
nc_fetch_word(const void *iptr, uint32_t *a)
{
const uint32_t *tptr = (const uint32_t *)iptr;
KASSERT(ALIGNED_POINTER(iptr, uint32_t));
*a = *tptr++;
return tptr;
}
/*
* Receive incoming data on our hook. Send it out the socket.
*/
static int
ng_ksocket_rcvdata(hook_p hook, item_p item)
{
struct thread *td = curthread; /* XXX broken */
const node_p node = NG_HOOK_NODE(hook);
const priv_p priv = NG_NODE_PRIVATE(node);
struct socket *const so = priv->so;
struct sockaddr *sa = NULL;
int error;
struct mbuf *m;
#ifdef ALIGNED_POINTER
struct mbuf *n;
#endif /* ALIGNED_POINTER */
struct sa_tag *stag;
/* Extract data */
NGI_GET_M(item, m);
NG_FREE_ITEM(item);
#ifdef ALIGNED_POINTER
if (!ALIGNED_POINTER(mtod(m, caddr_t), uint32_t)) {
n = m_defrag(m, M_NOWAIT);
if (n == NULL) {
m_freem(m);
return (ENOBUFS);
}
m = n;
}
#endif /* ALIGNED_POINTER */
/*
* Look if socket address is stored in packet tags.
* If sockaddr is ours, or provided by a third party (zero id),
* then we accept it.
*/
if (((stag = (struct sa_tag *)m_tag_locate(m, NGM_KSOCKET_COOKIE,
NG_KSOCKET_TAG_SOCKADDR, NULL)) != NULL) &&
(stag->id == NG_NODE_ID(node) || stag->id == 0))
sa = &stag->sa;
/* Reset specific mbuf flags to prevent addressing problems. */
m->m_flags &= ~(M_BCAST|M_MCAST);
/* Send packet */
error = sosend(so, sa, 0, m, 0, 0, td);
return (error);
}
static void setup_tests()
{
for (int test = 0; test < NUM_TESTS; ++test) {
unsigned size = NUM_CHUNKS * 64 * 64;
actual_bufs[test] = malloc(size * sizeof(kvz_pixel) + SIMD_ALIGNMENT);
bufs[test] = ALIGNED_POINTER(actual_bufs[test], SIMD_ALIGNMENT);
}
for (int test = 0; test < NUM_TESTS; ++test) {
for (int chunk = 0; chunk < NUM_CHUNKS; ++chunk) {
const int width = 64;
int x = (test + chunk) % width;
int y = (test + chunk) / width;
init_gradient(width - x, y, width, 255 / width, &bufs[test][chunk * 64*64]);
}
}
}
static void
readisaac (struct isaac *isaac, void *p, size_t size)
{
size_t inbytes = isaac->buffered;
while (true)
{
char *char_p = p;
if (size <= inbytes)
{
memcpy (p, isaac->data.b + ISAAC_BYTES - inbytes, size);
isaac->buffered = inbytes - size;
return;
}
memcpy (p, isaac->data.b + ISAAC_BYTES - inbytes, inbytes);
p = char_p + inbytes;
size -= inbytes;
/* If P is aligned, write to *P directly to avoid the overhead
of copying from the buffer. */
if (ALIGNED_POINTER (p, isaac_word))
{
isaac_word *wp = p;
while (ISAAC_BYTES <= size)
{
isaac_refill (&isaac->state, wp);
wp += ISAAC_WORDS;
size -= ISAAC_BYTES;
if (size == 0)
{
isaac->buffered = 0;
return;
}
}
p = wp;
}
isaac_refill (&isaac->state, isaac->data.w);
inbytes = ISAAC_BYTES;
}
}
/* flags: fseq: contain SEQ field, fack: contain ACK field */
static int
pptp_call_gre_output(pptp_call *_this, int fseq, int fack, u_char *pkt,
int lpkt)
{
int sz;
struct pptp_gre_header *grehdr;
u_char buf[65535], *opkt;
struct sockaddr_storage peer, sock;
#ifndef USE_LIBSOCKUTIL
socklen_t peerlen;
#endif
memset(buf, 0, sizeof(buf));
opkt = buf;
grehdr = (struct pptp_gre_header *)opkt;
opkt += sizeof(struct pptp_gre_header);
/* GRE header */
grehdr->K = 1;
grehdr->ver = PPTP_GRE_VERSION;
grehdr->protocol_type = htons(PPTP_GRE_PROTOCOL_TYPE);
grehdr->payload_length = htons(lpkt);
grehdr->call_id = htons(_this->peers_call_id);
#ifdef PPTP_CALL_DEBUG
if (debuglevel >= 2 && (fseq || fack)) {
pptp_call_log(_this, LOG_DEBUG,
"Sending data packet seq=%u ack=%u",
_this->snd_nxt, _this->rcv_nxt - 1);
}
#endif
PPTP_CALL_ASSERT(ALIGNED_POINTER(opkt, uint32_t));
if (fseq) {
grehdr->S = 1;
*(uint32_t *)opkt = htonl(_this->snd_nxt++);
opkt += 4;
}
if (fack) {
grehdr->A = 1;
_this->rcv_acked = _this->rcv_nxt;
*(uint32_t *)opkt = htonl(_this->rcv_nxt - 1);
opkt += 4;
}
if (lpkt > 0) {
memcpy(opkt, pkt, lpkt);
opkt += lpkt;
}
memcpy(&peer, &_this->ctrl->peer, sizeof(peer));
memcpy(&sock, &_this->ctrl->our, sizeof(sock));
switch (peer.ss_family) {
case AF_INET:
((struct sockaddr_in *)&peer)->sin_port = 0;
((struct sockaddr_in *)&sock)->sin_port = 0;
#ifndef USE_LIBSOCKUTIL
peerlen = sizeof(struct sockaddr_in);
#endif
break;
default:
return 1;
}
if (PPTP_CTRL_CONF(_this->ctrl)->data_out_pktdump != 0) {
pptp_call_log(_this, LOG_DEBUG, "PPTP Data output packet dump");
show_hd(debug_get_debugfp(), buf, opkt - buf);
}
#ifdef USE_LIBSOCKUTIL
sz = sendfromto(pptp_ctrl_sock_gre(_this->ctrl), buf, opkt - buf,
0, (struct sockaddr *)&sock, (struct sockaddr *)&peer);
#else
sz = sendto(pptp_ctrl_sock_gre(_this->ctrl), buf, opkt - buf, 0,
(struct sockaddr *)&peer, peerlen);
#endif
if (sz <= 0)
pptp_call_log(_this, LOG_WARNING, "sendto(%d) failed: %m",
pptp_ctrl_sock_gre(_this->ctrl));
return (sz > 0)? 0 : 1;
}
static void setup_tests()
{
for (int test = 0; test < NUM_TESTS; ++test) {
dct_actual_bufs[test] = malloc(LCU_WIDTH*LCU_WIDTH*sizeof(int16_t) + SIMD_ALIGNMENT);
dct_bufs[test] = ALIGNED_POINTER(dct_actual_bufs[test], SIMD_ALIGNMENT);
}
for (int test = 0; test < NUM_TESTS; ++test) {
const int width = LCU_WIDTH;
init_gradient(width, width, width, 255 / width, dct_bufs[test]);
}
// Select buffer width according to function name for dct function.
int block = 0;
for (int s = 0; s < strategies.count && block < NUM_SIZES; ++s)
{
strategy_t *strat = &strategies.strategies[s];
dct_func* dct_generic = 0;
if (
(
strcmp(strat->type, "fast_forward_dst_4x4") == 0 ||
strcmp(strat->type, "dct_4x4") == 0 ||
strcmp(strat->type, "dct_8x8") == 0 ||
strcmp(strat->type, "dct_16x16") == 0 ||
strcmp(strat->type, "dct_32x32") == 0
)
&&
strcmp(strat->strategy_name, "generic") == 0
)
{
dct_generic = strat->fptr;
dct_generic(KVZ_BIT_DEPTH, dct_bufs[block], dct_result[block]);
++block;
}
}
block = 0;
for (int s = 0; s < strategies.count && block < NUM_SIZES; ++s)
{
strategy_t *strat = &strategies.strategies[s];
dct_func* idct_generic = 0;
if (
(
strcmp(strat->type, "fast_inverse_dst_4x4") == 0 ||
strcmp(strat->type, "idct_4x4") == 0 ||
strcmp(strat->type, "idct_8x8") == 0 ||
strcmp(strat->type, "idct_16x16") == 0 ||
strcmp(strat->type, "idct_32x32") == 0
)
&&
strcmp(strat->strategy_name, "generic") == 0
)
{
idct_generic = strat->fptr;
idct_generic(KVZ_BIT_DEPTH, dct_bufs[block], idct_result[block]);
++block;
}
}
}
