void oval_probe_session_destroy(oval_probe_session_t *sess)
{
if (sess == NULL) {
dE("Invalid session (NULL)");
return;
}
oval_phtbl_free(sess->ph);
oval_pext_free(sess->pext);
oscap_free(sess);
}
int oval_probe_session_abort(oval_probe_session_t *sess)
{
oval_ph_t *ph;
if ((ph = oval_probe_handler_get(sess->ph, OVAL_SUBTYPE_ALL) ) == NULL) {
dE("No probe handler for OVAL_SUBTYPE_ALL");
return (-1);
}
return ph->func(OVAL_SUBTYPE_ALL, ph->uptr, PROBE_HANDLER_ACT_ABORT);
}
/**
* Collect an item
* This function adds an item the collected object assosiated
* with the given probe context.
*
* Returns:
* 0 ... the item was succesfully added to the collected object
* 1 ... the item was filtered out
* 2 ... the item was not added because of memory constraints
* and the collected object was flagged as incomplete
*-1 ... unexpected/internal error
*
* The caller must not free the item, it's freed automatically
* by this function or by the icache worker thread.
*/
int probe_item_collect(struct probe_ctx *ctx, SEXP_t *item)
{
SEXP_t *cobj_content;
size_t cobj_itemcnt;
assume_d(ctx != NULL, -1);
assume_d(ctx->probe_out != NULL, -1);
assume_d(item != NULL, -1);
cobj_content = SEXP_listref_nth(ctx->probe_out, 3);
cobj_itemcnt = SEXP_list_length(cobj_content);
SEXP_free(cobj_content);
if (probe_cobj_memcheck(cobj_itemcnt) != 0) {
/*
* Don't set the message again if the collected object is
* already flagged as incomplete.
*/
if (probe_cobj_get_flag(ctx->probe_out) != SYSCHAR_FLAG_INCOMPLETE) {
SEXP_t *msg;
/*
* Sync with the icache thread before modifying the
* collected object.
*/
if (probe_icache_nop(ctx->icache) != 0)
return -1;
msg = probe_msg_creat(OVAL_MESSAGE_LEVEL_WARNING,
"Object is incomplete due to memory constraints.");
probe_cobj_add_msg(ctx->probe_out, msg);
probe_cobj_set_flag(ctx->probe_out, SYSCHAR_FLAG_INCOMPLETE);
SEXP_free(msg);
}
return 2;
}
if (ctx->filters != NULL && probe_item_filtered(item, ctx->filters)) {
SEXP_free(item);
return (1);
}
if (probe_icache_add(ctx->icache, ctx->probe_out, item) != 0) {
dE("Can't add item (%p) to the item cache (%p)", item, ctx->icache);
SEXP_free(item);
return (-1);
}
return (0);
}
int probe_main (probe_ctx *ctx, void *arg)
{
SEXP_t *probe_in, *name_ent, *file_ent, *bh_ent;
char file[PATH_MAX];
size_t file_len = sizeof file;
char name[64];
size_t name_len = sizeof name;
oval_operation_t name_op, file_op;
uint64_t collect_flags = 0;
unsigned int i;
// If probe_init() failed it's because there was no rpm config files
if (arg == NULL) {
probe_cobj_set_flag(probe_ctx_getresult(ctx), SYSCHAR_FLAG_NOT_APPLICABLE);
return 0;
}
/*
* Get refs to object entities
*/
probe_in = probe_ctx_getobject(ctx);
name_ent = probe_obj_getent(probe_in, "name", 1);
file_ent = probe_obj_getent(probe_in, "filepath", 1);
if (name_ent == NULL || file_ent == NULL) {
dE("Missing \"name\" (%p) or \"filepath\" (%p) entity", name_ent, file_ent);
SEXP_free(name_ent);
SEXP_free(file_ent);
return (PROBE_ENOENT);
}
/*
* Extract the requested operation for each entity
*/
name_op = probe_ent_getoperation(name_ent, OVAL_OPERATION_EQUALS);
file_op = probe_ent_getoperation(file_ent, OVAL_OPERATION_EQUALS);
if (name_op == OVAL_OPERATION_UNKNOWN ||
file_op == OVAL_OPERATION_UNKNOWN)
{
SEXP_free(name_ent);
SEXP_free(file_ent);
return (PROBE_EINVAL);
}
/*
* Extract entity values
*/
PROBE_ENT_STRVAL(name_ent, name, name_len, /* void */, strcpy(name, ""););
int probe_icache_nop(probe_icache_t *cache)
{
pthread_cond_t cond;
dD("NOP");
if (pthread_mutex_lock(&cache->queue_mutex) != 0) {
dE("An error ocured while locking the queue mutex: %u, %s",
errno, strerror(errno));
return (-1);
}
if (pthread_cond_init(&cond, NULL) != 0) {
dE("Can't initialize icache queue condition variable (NOP): %u, %s",
errno, strerror(errno));
return (-1);
}
if (__probe_icache_add_nolock(cache, NULL, NULL, &cond) != 0) {
if (pthread_mutex_unlock(&cache->queue_mutex) != 0) {
dE("An error ocured while unlocking the queue mutex: %u, %s",
errno, strerror(errno));
abort();
}
pthread_cond_destroy(&cond);
return (-1);
}
dD("Signaling `notempty'");
if (pthread_cond_signal(&cache->queue_notempty) != 0) {
dE("An error ocured while signaling the `notempty' condition: %u, %s",
errno, strerror(errno));
pthread_cond_destroy(&cond);
return (-1);
}
dD("Waiting for icache worker to handle the NOP");
if (pthread_cond_wait(&cond, &cache->queue_mutex) != 0) {
dE("An error ocured while waiting for the `NOP' queue condition: %u, %s",
errno, strerror(errno));
return (-1);
}
dD("Sync");
if (pthread_mutex_unlock(&cache->queue_mutex) != 0) {
dE("An error ocured while unlocking the queue mutex: %u, %s",
errno, strerror(errno));
abort();
}
pthread_cond_destroy(&cond);
return (0);
}
static SEXP_t *create_item(const char *path, const char *filename, char *pattern,
int instance, char **substrs, int substr_cnt, oval_schema_version_t over)
{
int i;
SEXP_t *item;
SEXP_t *r0;
SEXP_t *se_instance, *se_filepath;
char *text;
if (strlen(path) + strlen(filename) + 1 > PATH_MAX) {
dE("path+filename too long");
return (NULL);
}
if (oval_schema_version_cmp(over, OVAL_SCHEMA_VERSION(5.4)) < 0) {
pattern = text = NULL;
se_instance = NULL;
} else {
text = substrs[0];
se_instance = SEXP_number_newu_64((int64_t) instance);
}
if (oval_schema_version_cmp(over, OVAL_SCHEMA_VERSION(5.6)) < 0) {
se_filepath = NULL;
} else {
const size_t path_len = strlen(path);
/* Avoid 2 slashes */
if (path_len >= 1 && path[path_len - 1] == FILE_SEPARATOR) {
se_filepath = SEXP_string_newf("%s%s", path, filename);
} else {
se_filepath = SEXP_string_newf("%s%c%s", path, FILE_SEPARATOR, filename);
}
}
item = probe_item_create(OVAL_INDEPENDENT_TEXT_FILE_CONTENT, NULL,
"filepath", OVAL_DATATYPE_SEXP, se_filepath,
"path", OVAL_DATATYPE_STRING, path,
"filename", OVAL_DATATYPE_STRING, filename,
"pattern", OVAL_DATATYPE_STRING, pattern,
"instance", OVAL_DATATYPE_SEXP, se_instance,
"line", OVAL_DATATYPE_STRING, pattern,
"text", OVAL_DATATYPE_STRING, substrs[0],
NULL);
for (i = 1; i < substr_cnt; ++i) {
probe_item_ent_add (item, "subexpression", NULL, r0 = SEXP_string_new (substrs[i], strlen (substrs[i])));
SEXP_free (r0);
}
return item;
}
static void probe_icache_item_setID(SEXP_t *item, SEXP_ID_t item_ID)
{
SEXP_t *name_ref, *prev_id;
SEXP_t uniq_id;
uint32_t local_id;
/* ((foo_item :id "<int>") ... ) */
assume_d(item != NULL, /* void */);
assume_d(SEXP_listp(item), /* void */);
#if defined(HAVE_ATOMIC_FUNCTIONS)
local_id = __sync_fetch_and_add(&next_ID, 1);
#else
if (pthread_mutex_lock(&next_ID_mutex) != 0) {
dE("Can't lock the next_ID_mutex: %u, %s", errno, strerror(errno));
abort();
}
local_id = ++next_ID;
if (pthread_mutex_unlock(&next_ID_mutex) != 0) {
dE("Can't unlock the next_ID_mutex: %u, %s", errno, strerror(errno));
abort();
}
#endif
SEXP_string_newf_r(&uniq_id, "1%05u%u", getpid(), local_id);
name_ref = SEXP_listref_first(item);
prev_id = SEXP_list_replace(name_ref, 3, &uniq_id);
SEXP_free(prev_id);
SEXP_free_r(&uniq_id);
SEXP_free(name_ref);
return;
}
static int oval_probe_cmd_init(oval_pext_t *pext)
{
if (pext == NULL) {
return -1;
}
if (SEAP_cmd_register(pext->pdtbl->ctx, PROBECMD_OBJ_EVAL, SEAP_CMDREG_USEARG,
&oval_probe_cmd_obj_eval, (void *)pext) != 0)
{
dE("Can't register command: %s: errno=%u, %s.", "obj_eval", errno, strerror(errno));
return (-1);
}
if (SEAP_cmd_register(pext->pdtbl->ctx, PROBECMD_STE_FETCH, SEAP_CMDREG_USEARG,
&oval_probe_cmd_ste_fetch, (void *)pext) != 0) {
dE("Can't register command: %s: errno=%u, %s.", "ste_fetch", errno, strerror(errno));
/* FIXME: unregister the first command */
return (-1);
}
return (0);
}
struct oval_record_field *oval_record_field_clone(struct oval_record_field *old_rf)
{
struct oval_record_field *new_rf;
switch (old_rf->record_field_type) {
case OVAL_RECORD_FIELD_STATE:
{
struct oval_record_field_STATE *new_rfs, *old_rfs;
new_rfs = oscap_alloc(sizeof(*new_rfs));
if (new_rfs == NULL)
return NULL;
old_rfs = (struct oval_record_field_STATE *) old_rf;
new_rfs->operation = old_rfs->operation;
new_rfs->variable = old_rfs->variable;
new_rfs->var_check = old_rfs->var_check;
new_rfs->ent_check = old_rfs->ent_check;
new_rf = (struct oval_record_field *) new_rfs;
break;
}
case OVAL_RECORD_FIELD_ITEM:
{
struct oval_record_field_ITEM *new_rfi, *old_rfi;
new_rfi = oscap_alloc(sizeof(*new_rfi));
if (new_rfi == NULL)
return NULL;
old_rfi = (struct oval_record_field_ITEM *) old_rf;
new_rfi->status = old_rfi->status;
new_rf = (struct oval_record_field *) new_rfi;
break;
}
default:
dE("Unsupported record field type: %d.\n", old_rf->record_field_type);
return NULL;
}
new_rf->record_field_type = old_rf->record_field_type;
new_rf->name = oscap_strdup(old_rf->name);
new_rf->value = oscap_strdup(old_rf->value);
new_rf->datatype = old_rf->datatype;
new_rf->mask = old_rf->mask;
return new_rf;
}
void
HHV4Vector::CalcCov()
{
m_cov_manually_set = false;
if (P()>0){
//NOTE: only dE is used for Cov calculation!
Double_t dp = E()/P()*dE(); // error propagation p=sqrt(e^2-m^2)
Double_t dpt = sin(Theta())*dp;
m_cov_transversal(0,0) = pow(cos(Phi())*dpt,2);
m_cov_transversal(1,1) = pow(sin(Phi())*dpt,2);
m_cov_transversal(0,1) = sin(Phi())*cos(Phi())*dpt*dpt;
m_cov_transversal(1,0) = sin(Phi())*cos(Phi())*dpt*dpt;
}
}
int oval_probe_session_reset(oval_probe_session_t *sess, struct oval_syschar_model *sysch)
{
oval_ph_t *ph;
if ((ph = oval_probe_handler_get(sess->ph, OVAL_SUBTYPE_ALL)) == NULL) {
dE("No probe handler for OVAL_SUBTYPE_ALL");
return (-1);
}
if (ph->func(OVAL_SUBTYPE_ALL, ph->uptr, PROBE_HANDLER_ACT_RESET) != 0) {
return(-1);
}
if (sysch != NULL)
sess->sys_model = sysch;
return(0);
}
struct oval_record_field *oval_record_field_new(oval_record_field_type_t type)
{
struct oval_record_field *rf;
switch (type) {
case OVAL_RECORD_FIELD_STATE:
{
struct oval_record_field_STATE *rfs;
rfs = oscap_alloc(sizeof(*rfs));
if (rfs == NULL)
return NULL;
rfs->operation = OVAL_OPERATION_UNKNOWN;
rfs->variable = NULL;
rfs->var_check = OVAL_CHECK_UNKNOWN;
rfs->ent_check = OVAL_CHECK_UNKNOWN;
rf = (struct oval_record_field *) rfs;
break;
}
case OVAL_RECORD_FIELD_ITEM:
{
struct oval_record_field_ITEM *rfi;
rfi = oscap_alloc(sizeof(*rfi));
if (rfi == NULL)
return NULL;
rfi->status = SYSCHAR_STATUS_UNKNOWN;
rf = (struct oval_record_field *) rfi;
break;
}
default:
dE("Unsupported record field type: %d.\n", type);
return NULL;
}
rf->record_field_type = type;
rf->name = NULL;
rf->value = NULL;
rf->datatype = OVAL_DATATYPE_UNKNOWN;
rf->mask = 0;
return rf;
}
static int file_cb (const char *p, const char *f, void *ptr)
{
char path_buffer[PATH_MAX];
SEXP_t *item, xattr_name;
struct cbargs *args = (struct cbargs *) ptr;
const char *st_path;
ssize_t xattr_count = -1;
char *xattr_buf = NULL;
size_t xattr_buflen = 0, i;
if (f == NULL) {
st_path = p;
} else {
snprintf (path_buffer, sizeof path_buffer, "%s/%s", p, f);
st_path = path_buffer;
}
SEXP_init(&xattr_name);
do {
/* estimate the size of the buffer */
xattr_count = llistxattr(st_path, NULL, 0);
if (xattr_count == 0)
return (0);
if (xattr_count < 0) {
dI("FAIL: llistxattr(%s, %p, %zu): errno=%u, %s.", errno, strerror(errno));
return 0;
}
/* allocate space for xattr names */
xattr_buflen = xattr_count;
xattr_buf = oscap_realloc(xattr_buf, sizeof(char) * xattr_buflen);
/* fill the buffer */
xattr_count = llistxattr(st_path, xattr_buf, xattr_buflen);
/* check & retry if needed */
} while (errno == ERANGE);
if (xattr_count < 0) {
dI("FAIL: llistxattr(%s, %p, %zu): errno=%u, %s.", errno, strerror(errno));
oscap_free(xattr_buf);
}
/* update lastpath if needed */
if (!SEXP_emptyp(&gr_lastpath)) {
if (SEXP_strcmp(&gr_lastpath, p) != 0) {
SEXP_free_r(&gr_lastpath);
SEXP_string_new_r(&gr_lastpath, p, strlen(p));
}
} else
SEXP_string_new_r(&gr_lastpath, p, strlen(p));
i = 0;
/* collect */
do {
SEXP_string_new_r(&xattr_name, xattr_buf + i, strlen(xattr_buf +i));
if (probe_entobj_cmp(args->attr_ent, &xattr_name) == OVAL_RESULT_TRUE) {
ssize_t xattr_vallen = -1;
char *xattr_val = NULL;
xattr_vallen = lgetxattr(st_path, xattr_buf + i, NULL, 0);
retry_value:
if (xattr_vallen >= 0) {
// Check possible buffer overflow
if (sizeof(char) * (xattr_vallen + 1) <= sizeof(char) * xattr_vallen) {
dE("Attribute is too long.");
abort();
}
// Allocate buffer, '+1' is for trailing '\0'
xattr_val = oscap_realloc(xattr_val, sizeof(char) * (xattr_vallen + 1));
// we don't want to override space for '\0' by call of 'lgetxattr'
// we pass only 'xattr_vallen' instead of 'xattr_vallen + 1'
xattr_vallen = lgetxattr(st_path, xattr_buf + i, xattr_val, xattr_vallen);
if (xattr_vallen < 0 || errno == ERANGE)
goto retry_value;
xattr_val[xattr_vallen] = '\0';
item = probe_item_create(OVAL_UNIX_FILEEXTENDEDATTRIBUTE, NULL,
"filepath", OVAL_DATATYPE_STRING, f == NULL ? NULL : st_path,
"path", OVAL_DATATYPE_SEXP, &gr_lastpath,
"filename", OVAL_DATATYPE_STRING, f == NULL ? "" : f,
"attribute_name", OVAL_DATATYPE_SEXP, &xattr_name,
"value", OVAL_DATATYPE_STRING, xattr_val,
NULL);
oscap_free(xattr_val);
} else {
dI("FAIL: lgetxattr(%s, %s, NULL, 0): errno=%u, %s.", errno, strerror(errno));
item = probe_item_create(OVAL_UNIX_FILEEXTENDEDATTRIBUTE, NULL, NULL);
probe_item_setstatus(item, SYSCHAR_STATUS_ERROR);
if (xattr_val != NULL)
oscap_free(xattr_val);
}
probe_item_collect(args->ctx, item); /* XXX: handle ENOMEM */
}
SEXP_free_r(&xattr_name);
/* skip to next name */
while (i < xattr_buflen && xattr_buf[i] != '\0')
++i;
++i;
} while (xattr_buf + i < xattr_buf + xattr_buflen - 1);
oscap_free(xattr_buf);
return (0);
}
int routingtable_probe_main(probe_ctx *ctx, void *arg)
{
SEXP_t *probe_in, *dst_ent;
FILE *fp;
char *line_buf;
size_t line_len;
struct route_info rt;
int probe_ret = 0;
probe_in = probe_ctx_getobject(ctx);
dst_ent = probe_obj_getent(probe_in, "destination", 1);
if (dst_ent == NULL)
return (PROBE_ENOENT);
rt.ip_dst_ent = dst_ent;
line_len = 0;
line_buf = NULL;
fp = NULL;
switch(probe_ent_getdatatype(dst_ent)) {
case OVAL_DATATYPE_IPV4ADDR:
fp = fopen("/proc/net/route", "r");
/* Skip the header line */
if (getline(&line_buf, &line_len, fp) != -1) {
while(getline(&line_buf, &line_len, fp) != -1) {
if (process_line_ip4(line_buf, &rt) != 0)
break;
if (collect_item(&rt, ctx) != 0)
break;
}
}
if (!feof(fp)) {
/* error */
dE("An error ocured while reading /proc/net/route: %s", strerror(errno));
}
break;
case OVAL_DATATYPE_IPV6ADDR:
fp = fopen("/proc/net/ipv6_route", "r");
while(getline(&line_buf, &line_len, fp) != -1) {
if (process_line_ip6(line_buf, &rt) != 0)
break;
if (collect_item(&rt, ctx) != 0)
break;
}
if (!feof(fp)) {
/* error */
dE("An error ocured while reading /proc/net/ipv6_route: %s", strerror(errno));
}
break;
default:
probe_ret = EINVAL;
}
if (fp != NULL)
fclose(fp);
if (line_buf != NULL)
free(line_buf);
SEXP_free(dst_ent);
return (probe_ret);
}
int oval_probe_session_sethandler(oval_probe_session_t *sess, oval_subtype_t type, oval_probe_handler_t handler, void *ptr)
{
dE("Operation not supported");
return(-1);
}
extern "C" magma_int_t
magma_sbulge_applyQ_v2(char side,
magma_int_t NE, magma_int_t N,
magma_int_t NB, magma_int_t Vblksiz,
float *dE, magma_int_t ldde,
float *V, magma_int_t ldv,
float *T, magma_int_t ldt,
magma_int_t *info)
{
//%===========================
//% local variables
//%===========================
magma_int_t Vm, Vn, mt, nt;
magma_int_t myrow, mycol, blkj, blki;
magma_int_t blkid,vpos,tpos;
magma_int_t firstrow, nbcolinvolvd;
magma_int_t versionL = 113;
magma_int_t versionR = 92;
magma_int_t Vchunksiz = 10;
*info=0;
/* Quick return */
if ( NE == 0 ) {
return MAGMA_SUCCESS;
}
if ( N == 0 ) {
return MAGMA_SUCCESS;
}
if ( NB == 0 ) {
return MAGMA_SUCCESS;
}
/* ==========================================
* some infos for developer
* Initialisation and checking nb of cores
* ==========================================*/
/* we have 2 algo for left (113 114) and 2 algo for right (91 92)
* which correspond to versionL versionR.
* They are very similar (detail explained in tech report and matlab code)
* however version 114 and 92 improve locality.
* while version 113 is used in case WNATZ=1 (construct Q2) which allow
* the construction to be done in an optimized way taking into
* consideration that the matrix is Identity so making less flops.
*
*/
// Initialize streaming and events
cudaDeviceSynchronize();
magma_queue_t cstream;
magmablasGetKernelStream(&cstream);
magma_queue_t stream[2];
magma_queue_create( &stream[0] );
magma_queue_create( &stream[1] );
magma_event_t myevent[2];
cudaEventCreateWithFlags(&myevent[0],cudaEventDisableTiming);
cudaEventCreateWithFlags(&myevent[1],cudaEventDisableTiming);
// Azzam 21/11/2012
// NOTE THAT dwork was of size 2*NE*Vblksiz+...
// but I am thinking why not modifing it to NE*Vblksiz+...
// BUT NO because the 2* is used because of making 2 streams working and so
// they might be using dwork in parallel
float *dwork, *dwork0, *dwork1, *dwvt0, *dwvt1;
float *dT0, *dV0, *dT1, *dV1;
magma_int_t lddv = ldv;
magma_int_t lddt = ldt;
magma_int_t lddw = 0;
magma_int_t lddwork = ((NE+31)/32)*32;
magma_int_t dwVTsiz = lddv*Vblksiz; // lddv*lddv + lddv*lddwork;(v2) // lddv*Vblksiz; (v1,v3)
magma_int_t dworksiz = lddwork*Vblksiz; // lddv*Vblksiz; (v2) // NE*Vblksiz=lddwork*Vblksiz; (v1,v3)
if(MAGMA_SUCCESS != magma_smalloc( &dwork, 2*dworksiz + 2*dwVTsiz + 2*Vchunksiz* (Vblksiz* (lddv+lddt)) )) {
printf ("!!!! magma_sbulge_applyQ magma_alloc failed for: dwork\n" );
exit(-1);
}
dwork0 = dwork; // size = dworksiz;
dwork1 = dwork0 + dworksiz; // size = dworksiz;
dwvt0 = dwork + 2*dworksiz; // size = dwVTsiz;
dwvt1 = dwvt0 + dwVTsiz; // size = dwVTsiz;
dV0 = dwork + 2*dworksiz + 2*dwVTsiz;
dT0 = dV0 + Vchunksiz*Vblksiz*lddv;
dV1 = dT0 + Vchunksiz*Vblksiz*lddt;
dT1 = dV1 + Vchunksiz*Vblksiz*lddv;
// make overlapped copy
magma_int_t ncpy = 0;
magma_int_t copyed=0, copyst=0;
magma_int_t blkcnt,nothing, mysiz, flip, vld,tld, locpos;
findVTsiz(N, NB, Vblksiz, &blkcnt, ¬hing);
flip = 0;
// performance loss if the reflector are applied to a big number of eigenvectors (~10000)
// => apply the reflectors to blocks of eigenvectors.
//magma_int_t nr_bl = magma_ceildiv(NE,10000); //nr of blocks
magma_int_t sz_bl = NE; //magma_ceildiv(NE,nr_bl*64)*64; //maximum size of blocks (to have blocks of around the same size and multiple of 64)
magma_int_t ib; //size of current block
/* SIDE LEFT meaning apply E = Q*E = (q_1*q_2*.....*q_n) * E ==> so traverse Vs in reverse order (forward) from q_n to q_1
* Also E is splitten by row meaning each apply consist in a block of row (horizontal block) */
/* SIDE RIGHT meaning apply E = E*Q = E * (q_1*q_2*.....*q_n) ==> so tarverse Vs in normal order (forward) from q_1 to q_n
* Also E is splitten by col meaning each apply consist in a block of col (vertical block) */
#ifdef ENABLE_DEBUG
printf(" APPLY Q_v22 GPU with N %d, NE %d, NB %d, Vblksiz %d, versionL %d versionR %d SIDE %c \n",
N, NE, NB, Vblksiz, versionL, versionR, side);
#endif
/*
* MagmamaLeft
*/
if(side=='L'){
/*
* Version 113:
* loop over the block_col (nt) and for each find the
* number of tiles (mt) in this block_col. then loop over mt, find
* the size of the V's(Vm,Vn) and apply it to the corresponding
* portion of E.
*/
if( versionL == 113 ) {
nt = magma_ceildiv((N-1),Vblksiz);
for (blkj=nt-1; blkj>=0; blkj--) {
/* the index of the first row on the top of block (blkj) */
firstrow = blkj * Vblksiz + 1;
/*find the number of tile for this block */
if( blkj == nt-1 )
mt = magma_ceildiv( N - firstrow, NB);
else
mt = magma_ceildiv( N - (firstrow+1), NB);
/*loop over the tiles find the size of the Vs and apply it */
for (blki=mt; blki>0; blki--) {
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + (mt-blki)*NB;
mycol = blkj*Vblksiz;
Vm = min( NB+Vblksiz-1, N-myrow);
if( ( blkj == nt-1 ) && ( blki == mt ) ){
Vn = min (Vblksiz, Vm);
} else {
Vn = min (Vblksiz, Vm-1);
}
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
//printf("voici blkj %d blki %d Vm %d Vn %d mycol %d vpos %d \n",blkj,blki,Vm, Vn,mycol,vpos);
magma_bulge_findpos113(N, NB, Vblksiz, mycol, myrow, &blkid);
// COPY Vchunksiz Vs and Vchunksiz Ts to GPU and store it in dV0/dV1 and dT0/dT1
if(ncpy==0){
// flip = 1 for this.
copyst = 0; // meaning that copy will start copying from blkid =copyst
copyed = min(copyst+Vchunksiz, blkcnt); // meaning that copy will end copying at blkid =copyed-1==> next copy had to start at copyed
mysiz = copyed-copyst; // the size of the chunk to be copied
if(mysiz>0){
ncpy = 1;
flip = 1;
vpos = copyst*Vblksiz*ldv;
tpos = copyst*Vblksiz*ldt;
vld = mysiz * ldv;
tld = mysiz * ldt;
magmablasSetKernelStream(stream[1]);
magma_ssetmatrix_async(vld, Vblksiz, V(vpos), vld, dV1, vld, stream[1]);
magma_ssetmatrix_async(tld, Vblksiz, T(tpos), tld, dT1, tld, stream[1]);
//printf("doing the first copy of mysiz %2d copyst %2d copyed %2d vpos %8d tpos %8d into dV1 dT1\n",mysiz,copyst,copyed,vpos,tpos);
}
}
if(blkid == copyst){
flip = ncpy % 2;
copyst = copyed; // meaning that copy will start copying from blkid =copyst
copyed = min(copyst+Vchunksiz, blkcnt); // meaning that copy will end copying at blkid =copyed-1==> next copy had to start at copyed
mysiz = copyed-copyst; // the size of the chunk to be copied
//printf(" get to copy blkid %d blkid+(2*Vchunksiz) %d copyst %d copyed %d\n",blkid,blkid+(Vchunksiz),copyst,copyed);
if(mysiz>0){
ncpy = ncpy + 1;
vpos = copyst*Vblksiz*ldv;
tpos = copyst*Vblksiz*ldt;
vld = mysiz * ldv;
tld = mysiz * ldt;
if(flip==0){ // now I am working on dV0 so copy the next and put it on dV1
//printf("doing overlapping copy of mysiz %2d copyst %2d copyed %2d vpos %8d tpos %8d into dV1 dT1\n",mysiz,copyst,copyed,vpos,tpos);
magmablasSetKernelStream(stream[1]);
magma_ssetmatrix_async(vld, Vblksiz, V(vpos), vld, dV1, vld, stream[1]);
magma_ssetmatrix_async(tld, Vblksiz, T(tpos), tld, dT1, tld, stream[1]);
}else{ // now I am working on dV1 so copy the next and put it on dV0
//printf("doing overlapping copy of mysiz %2d copyst %2d copyed %2d vpos %8d tpos %8d into dV0 dT0\n",mysiz,copyst,copyed,vpos,tpos);
magmablasSetKernelStream(stream[0]);
magma_ssetmatrix_async(vld, Vblksiz, V(vpos), vld, dV0, vld, stream[0]);
magma_ssetmatrix_async(tld, Vblksiz, T(tpos), tld, dT0, tld, stream[0]);
}
}
}
if((Vm>0)&&(Vn>0)){
locpos = blkid%Vchunksiz;
magma_int_t lcvpos = locpos*Vblksiz*lddv;
magma_int_t lctpos = locpos*Vblksiz*lddt;
//printf("voici blkj %d blki %d Vm %d Vn %d mycol %d locvpos %5d loctpos %5d blkid %2d using data in dV%1d dT%1d \n",blkj,blki,Vm, Vn,mycol,lcvpos,lctpos, blkid,flip,flip);
if(flip==0){
magmablasSetKernelStream(stream[0]);
cudaStreamWaitEvent(stream[0], myevent[1], 0);
for(magma_int_t i=0; i<NE; i+= sz_bl){
ib = min(sz_bl, NE-i);
lddw = min(lddwork,sz_bl);
//magma_slarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV0+lcvpos, lddv, dT0+lctpos, lddt, dE(myrow,i), ldde, dwork0, lddw);
magma_slarfb_gpu_gemm( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV0+lcvpos, lddv, dT0+lctpos, lddt, dE(myrow,i), ldde, dwork0, lddw, dwvt0, lddv);
}
cudaEventRecord(myevent[0], stream[0]);
}else{
magmablasSetKernelStream(stream[1]);
cudaStreamWaitEvent(stream[1], myevent[0], 0);
for(magma_int_t i=0; i<NE; i+= sz_bl){
ib = min(sz_bl, NE-i);
lddw = min(lddwork,sz_bl);
//magma_slarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV1+lcvpos, lddv, dT1+lctpos, lddt, dE(myrow,i), ldde, dwork1, lddw);
magma_slarfb_gpu_gemm( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV1+lcvpos, lddv, dT1+lctpos, lddt, dE(myrow,i), ldde, dwork1, lddw, dwvt1, lddv);
}
cudaEventRecord(myevent[1], stream[1]);
}
} // end for (Vm &Vn) > 0
} // end for blki
} // end for blkj
} // end if version=113
/*
* Version 114:
* loop over the block_row (mt) and for each find diagonally the
* number of tiles (nt) in this block_row. then loop over nt, find
* the size of the V's(Vm,Vn) and apply it to the corresponding
* portion of E.
*/
else {
mt = magma_ceildiv((N-1),NB);
for (blki = mt; blki>0; blki--) {
/* nbcolinvolvd = number of column corresponding to this block_row (blki) */
nbcolinvolvd = min(N-1, blki*NB);
/*find the number of tile for this block (diagonal row of tiles) */
nt = magma_ceildiv(nbcolinvolvd,Vblksiz);
/*loop over the tiles find the size of the Vs and apply it */
for (blkj = nt-1; blkj>=0; blkj--) {
/* the index of the first row of the first col meaning
* the block on the top left (blki) */
firstrow = (mt-blki)*NB+1;
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + blkj*Vblksiz;
mycol = blkj*Vblksiz;
Vm = min( NB+Vblksiz-1, N-myrow);
if( ( blkj == nt-1 ) && ( blki == mt ) ){
Vn = min (Vblksiz, Vm);
}else{
Vn = min (Vblksiz, Vm-1);
}
if((Vm>0)&&(Vn>0)){
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
magma_bulge_findVTpos(N, NB, Vblksiz, mycol, myrow, ldv, ldt, &vpos, &tpos);
magma_ssetmatrix_async(Vm, Vn, V(vpos), ldv, dV0, lddv, NULL);
magma_ssetmatrix_async(Vn, Vn, T(tpos), ldt, dT0, lddt, NULL);
//printf("voici blki %d rownbm %d mycol %d coled %d blkid %d vpos %d tpos %d\n", blki, rownbm, mycol, coled, blkid, vpos, tpos);
for(magma_int_t i=0; i<NE; i+= sz_bl){
ib = min(sz_bl, NE-i);
magma_slarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV0, lddv, dT0, lddt, dE(myrow,i), ldde, dwork, NE);
}
} // end for (Vm &Vn) > 0
} // end for blkj
} // end for blki
} // end version 114
} // end LEFT
/*
* MagmaRight
*/
else {
/*
* Version 91:
*/
if( versionR == 91 ) {
nt = magma_ceildiv((N-1),Vblksiz);
for (blkj=0; blkj<nt; blkj++) {
/* the index of the first myrow on the top of block (blkj) */
firstrow = blkj * Vblksiz + 1;
/*find the number of tile for this block */
if( blkj == nt-1 )
mt = magma_ceildiv( N - firstrow, NB);
else
mt = magma_ceildiv( N - (firstrow+1), NB);
/*loop over the tiles find the size of the Vs and apply it */
for (blki=1; blki<=mt; blki++) {
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + (mt-blki)*NB;
Vm = min( NB+Vblksiz-1, N-myrow);
if( ( blkj == nt-1 ) && ( blki == mt ) )
{
Vn = min (Vblksiz, Vm);
}else{
Vn = min (Vblksiz, Vm-1);
}
mycol = blkj*Vblksiz;
if((Vm>0)&&(Vn>0)){
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
magma_bulge_findVTpos(N, NB, Vblksiz ,mycol, myrow, ldv, ldt, &vpos, &tpos);
magma_ssetmatrix_async(Vm, Vn, V(vpos), ldv, dV0, lddv, NULL);
magma_ssetmatrix_async(Vn, Vn, T(tpos), ldt, dT0, lddt, NULL);
magma_slarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, NE, Vm, Vn, dV0, lddv, dT0, lddt, dE(0, myrow), ldde, dwork, NE);
} // end for (Vm &Vn) > 0
} // end for blki
} // end fo blkj
} // end of version 91
/*
* Version 92:
*/
else {
mt = magma_ceildiv((N-1),NB);
for (blki = 1; blki<=mt; blki++) {
/* nbcolinvolvd = number of column corresponding to this block_row (blki) */
nbcolinvolvd = min(N-1, blki*NB);
/*find the number of tile for this block (diagonal row of tiles) */
nt = magma_ceildiv(nbcolinvolvd,Vblksiz);
/*loop over the tiles find the size of the Vs and apply it */
for (blkj = 0; blkj<nt; blkj++) {
/* the index of the first row of the first col meaning
* the block on the top left (blki) */
firstrow = (mt-blki)*NB+1;
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + blkj*Vblksiz;
mycol = blkj*Vblksiz;
Vm = min( NB+Vblksiz-1, N-myrow);
if( ( blkj == nt-1 ) && ( blki == mt ) ){
Vn = min (Vblksiz, Vm);
}else{
Vn = min (Vblksiz, Vm-1);
}
if((Vm>0)&&(Vn>0)){
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
magma_bulge_findVTpos(N, NB, Vblksiz ,mycol, myrow, ldv, ldt, &vpos, &tpos);
magma_ssetmatrix_async(Vm, Vn, V(vpos), ldv, dV0, lddv, NULL);
magma_ssetmatrix_async(Vn, Vn, T(tpos), ldt, dT0, lddt, NULL);
magma_slarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, NE, Vm, Vn, dV0, lddv, dT0, lddt, dE(0, myrow), ldde, dwork, NE);
} // end for (Vm &Vn) > 0
} //end for blkj
} // end for blki
} //end of version 92
} // end RIGHT
cudaDeviceSynchronize();
magmablasSetKernelStream(cstream);
cudaEventDestroy(myevent[0]);
cudaEventDestroy(myevent[1]);
magma_queue_destroy( stream[0] );
magma_queue_destroy( stream[1] );
magma_free(dwork);
return MAGMA_SUCCESS;
}
void
HHV4Vector::PrintErrors() const
{
std::cout << Name() << " ";
Printf(" (Errors) %8.2f %8.4f %8.4f ", dE(), dEta(), dPhi());
}
static int oval_varref_attr_to_sexp(void *sess, struct oval_entity *entity, struct oval_syschar *syschar, SEXP_t **out_sexp)
{
unsigned int val_cnt = 0;
SEXP_t *val_lst, *val_sexp, *varref, *id_sexp, *val_cnt_sexp;
oval_datatype_t dt;
struct oval_variable *var;
struct oval_value_iterator *vit;
struct oval_value *val;
oval_syschar_collection_flag_t flag;
char msg[100];
int ret = 0;
var = oval_entity_get_variable(entity);
if (oval_probe_query_variable(sess, var) != 0) {
dE("Can't convert variable reference to SEXP.");
return -1;
}
flag = oval_variable_get_collection_flag(var);
switch (flag) {
case SYSCHAR_FLAG_COMPLETE:
case SYSCHAR_FLAG_INCOMPLETE:
vit = oval_variable_get_values(var);
if (oval_value_iterator_has_more(vit))
break;
oval_value_iterator_free(vit);
/* fall through */
case SYSCHAR_FLAG_DOES_NOT_EXIST:
snprintf(msg, sizeof(msg), "Referenced variable has no values (%s).", oval_variable_get_id(var));
dI("%s", msg);
ret = 1;
break;
default:
snprintf(msg, sizeof(msg), "There was a problem processing referenced variable (%s).", oval_variable_get_id(var));
dW("%s", msg);
ret = 1;
}
if (ret) {
oval_syschar_add_new_message(syschar, msg, OVAL_MESSAGE_LEVEL_WARNING);
oval_syschar_set_flag(syschar, SYSCHAR_FLAG_DOES_NOT_EXIST);
return ret;
}
val_lst = SEXP_list_new(NULL);
while (oval_value_iterator_has_more(vit)) {
val = oval_value_iterator_next(vit);
dt = oval_entity_get_datatype(entity);
val_sexp = oval_value_to_sexp(val, dt);
if (val_sexp == NULL) {
oval_syschar_add_new_message(syschar, "Failed to convert variable value.", OVAL_MESSAGE_LEVEL_ERROR);
oval_syschar_set_flag(syschar, SYSCHAR_FLAG_ERROR);
SEXP_free(val_lst);
oval_value_iterator_free(vit);
return -1;
}
SEXP_list_add(val_lst, val_sexp);
SEXP_free(val_sexp);
++val_cnt;
}
oval_value_iterator_free(vit);
id_sexp = SEXP_string_newf("%s", oval_variable_get_id(var));
val_cnt_sexp = SEXP_number_newu(val_cnt);
varref = SEXP_list_new(id_sexp, val_cnt_sexp, val_lst, NULL);
SEXP_free(id_sexp);
SEXP_free(val_cnt_sexp);
SEXP_free(val_lst);
*out_sexp = varref;
return 0;
}
void *probe_worker_runfn(void *arg)
{
probe_pwpair_t *pair = (probe_pwpair_t *)arg;
SEXP_t *probe_res, *obj, *oid;
int probe_ret;
#if defined(HAVE_PTHREAD_SETNAME_NP)
pthread_setname_np(pthread_self(), "probe_worker");
#endif
dD("handling SEAP message ID %u", pair->pth->sid);
//
probe_ret = -1;
probe_res = pair->pth->msg_handler(pair->probe, pair->pth->msg, &probe_ret);
//
dD("handler result = %p, return code = %d", probe_res, probe_ret);
/* Assuming that the red-black tree API is doing locking for us... */
if (rbt_i32_del(pair->probe->workers, pair->pth->sid, NULL) != 0) {
dW("thread not found in the probe thread tree, probably canceled by an external signal");
/*
* XXX: this is a possible deadlock; we can't send anything from
* here because the signal handler replied to the message
*/
arg = NULL;
SEAP_msg_free(pair->pth->msg);
SEXP_free(probe_res);
oscap_free(pair);
return (NULL);
} else {
SEXP_t *items;
dD("probe thread deleted");
obj = SEAP_msg_get(pair->pth->msg);
oid = probe_obj_getattrval(obj, "id");
items = probe_cobj_get_items(probe_res);
if (items != NULL) {
SEXP_list_sort(items, SEXP_refcmp);
SEXP_free(items);
}
if (probe_rcache_sexp_add(pair->probe->rcache, oid, probe_res) != 0) {
/* TODO */
abort();
}
SEXP_vfree(obj, oid, NULL);
}
if (probe_ret != 0) {
/*
* Something bad happened. A hint of the cause is stored as a error code in
* probe_ret (should be). We'll send it to the library using a SEAP error packet.
*/
if (SEAP_replyerr(pair->probe->SEAP_ctx, pair->probe->sd, pair->pth->msg, probe_ret) == -1) {
int ret = errno;
dE("An error ocured while sending error status. errno=%u, %s.", errno, strerror(errno));
SEXP_free(probe_res);
/* FIXME */
exit(ret);
}
SEXP_free(probe_res);
} else {
SEAP_msg_t *seap_reply;
/*
* OK, the probe actually returned something, let's send it to the library.
*/
seap_reply = SEAP_msg_new();
SEAP_msg_set(seap_reply, probe_res);
if (SEAP_reply(pair->probe->SEAP_ctx, pair->probe->sd, seap_reply, pair->pth->msg) == -1) {
int ret = errno;
SEAP_msg_free(seap_reply);
SEXP_free(probe_res);
exit(ret);
}
SEAP_msg_free(seap_reply);
SEXP_free(probe_res);
}
SEAP_msg_free(pair->pth->msg);
oscap_free(pair->pth);
oscap_free(pair);
pthread_detach(pthread_self());
return (NULL);
}
extern "C" void magma_zbulge_applyQ(
magma_int_t WANTZ, magma_side_t SIDE, magma_int_t NE, magma_int_t N, magma_int_t NB,
magma_int_t Vblksiz, magmaDoubleComplex *E, magma_int_t LDE,
magmaDoubleComplex *V, magmaDoubleComplex *TAU, magmaDoubleComplex *T,
magma_int_t *INFO, magmaDoubleComplex *dV, magmaDoubleComplex *dT,
magmaDoubleComplex *dE, magma_int_t copytype )
{
//%===========================
//% local variables
//%===========================
magmaDoubleComplex c_zero = MAGMA_Z_ZERO;
magmaDoubleComplex c_one = MAGMA_Z_ONE;
magma_int_t LDT, LDV, firstcolj;
magma_int_t bg, nbGblk, rownbm, k, m, n;
magma_int_t st, ed, fst, vlen, vnb, colj, len;
magma_int_t blkid, vpos, taupos, tpos;
//magmaDoubleComplex *WORK;
magma_int_t LWORK;
magma_int_t cur_blksiz, avai_blksiz, ncolinvolvd;
magma_int_t nbgr, colst, coled, versionL, versionR;
magma_int_t blkcnt=-1;
magma_queue_t orig_stream;
magmablasGetKernelStream( &orig_stream );
*INFO=0;
versionL = 113;
versionR = 92;
LDT = Vblksiz;
LDV = NB+Vblksiz-1;
//blklen = LDV*Vblksiz;
nbGblk = plasma_ceildiv((N-1), Vblksiz);
//magma_zmalloc_cpu( &WORK, LWORK );
/* find the size of the matrix T V*/
findVTsiz(N, NB, Vblksiz, &blkcnt, &LDV);
/* Copy E & V & T to the GPU in dE and dV and dT
* depending on copytype:
* 1: mean copy only V
* 2: mean copy V and T
* 3: mean copy V, T and E
* */
if (copytype > 0) magma_zsetmatrix( LDV, blkcnt*Vblksiz, V, LDV, dV, LDV );
if (copytype > 1) magma_zsetmatrix( LDT, blkcnt*Vblksiz, T, LDT, dT, LDT );
if (copytype > 2) magma_zsetmatrix( N, NE, E, N, dE, N );
magmaDoubleComplex *dwork;
//ldwork = NE;
LWORK = 2*N*max(Vblksiz, 64);
if (MAGMA_SUCCESS != magma_zmalloc( &dwork, LWORK )) {
printf ("!!!! magma_zbulge_applyQ magma_alloc failed for: dwork\n" );
exit(-1);
}
/* SIDE LEFT meaning apply E = Q*E = (q_1*q_2*.....*q_n) * E ==> so traverse Vs in reverse order (forward) from q_n to q_1
* Also E is splitten by row meaning each apply consist in a block of row (horizontal block) */
/* SIDE RIGHT meaning apply E = E*Q = E * (q_1*q_2*.....*q_n) ==> so tarverse Vs in normal order (forward) from q_1 to q_n
* Also E is splitten by col meaning each apply consist in a block of col (vertical block) */
/* WANTZ = 1 meaning E is IDENTITY so form Q using optimized update.
* So we use the reverse order from small q to large one,
* so from q_n to q_1 so Left update to Identity.
* Use versionL 113 because in 114 we need to update the whole matrix and not in icreasing order.
* WANTZ = 2 meaning E is a full matrix and need to be updated from Left or Right so use normal update
* */
if (WANTZ == 1) {
versionL=113;
SIDE = MagmaLeft;
//set the matrix to Identity here to avoid copying it from the CPU
magmablas_zlaset( MagmaFull, N, N, c_zero, c_one, dE, N );
}
printf(" APPLY Q_v115 GPU with N %d NB %d Vblksiz %d SIDE %c versionL %d versionR %d WANTZ %d \n",
(int) N, (int) NB, (int) Vblksiz, SIDE, (int) versionL, (int) versionR, (int) WANTZ);
#if defined(USESTREAM)
magma_int_t N2=N/2;
magma_int_t N1=N-N2;
printf("using stream\n");
magma_queue_t stream[2];
magma_queue_create( &stream[0] );
magma_queue_create( &stream[1] );
#endif
if (SIDE == MagmaLeft) {
if (versionL == 113) {
for (bg = nbGblk; bg > 0; bg--) {
firstcolj = (bg-1)*Vblksiz + 1;
if (bg == nbGblk)
rownbm = plasma_ceildiv((N-(firstcolj)), NB); // last blk has size=1 used for complex to handle A(N,N-1)
else
rownbm = plasma_ceildiv((N-(firstcolj+1)), NB);
for (m = rownbm; m > 0; m--) {
vlen = 0;
vnb = 0;
colj = (bg-1)*Vblksiz; // for k=0; I compute the fst and then can remove it from the loop
fst = (rownbm -m)*NB+colj +1;
for (k=0; k < Vblksiz; k++) {
colj = (bg-1)*Vblksiz + k;
st = (rownbm -m)*NB+colj +1;
ed = min(st+NB-1, N-1);
if (st > ed) break;
if ((st == ed) && (colj != N-2)) break;
vlen=ed-fst+1;
vnb=k+1;
}
colst = (bg-1)*Vblksiz;
findVTpos(N, NB, Vblksiz, colst, fst, &vpos, &taupos, &tpos, &blkid);
printf("voici bg %d m %d vlen %d vnb %d fcolj %d vpos %d taupos %d \n", (int) bg, (int) m, (int) vlen, (int) vnb, (int) colst+1, (int) vpos+1, (int) taupos+1);
if ((vlen > 0) && (vnb > 0)) {
if (WANTZ == 1) {
len = N-colst;
magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, vlen, len, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(fst,colst), LDE, dwork, len);
} else {
magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, vlen, NE, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(fst,0), LDE, dwork, NE);
}
}
}
}
} else if (versionL == 114) {
rownbm = plasma_ceildiv((N-1), NB);
for (m = rownbm; m > 0; m--) {
ncolinvolvd = min(N-1, m*NB);
avai_blksiz=min(Vblksiz, ncolinvolvd);
nbgr = plasma_ceildiv(ncolinvolvd, avai_blksiz);
for (n = nbgr; n > 0; n--) {
vlen = 0;
vnb = 0;
cur_blksiz = min(ncolinvolvd-(n-1)*avai_blksiz, avai_blksiz);
colst = (n-1)*avai_blksiz;
coled = colst + cur_blksiz -1;
fst = (rownbm -m)*NB+colst +1;
for (colj=colst; colj <= coled; colj++) {
st = (rownbm -m)*NB+colj +1;
ed = min(st+NB-1, N-1);
if (st > ed) break;
if ((st == ed) && (colj != N-2)) break;
vlen=ed-fst+1;
vnb=vnb+1;
}
findVTpos(N, NB, Vblksiz, colst, fst, &vpos, &taupos, &tpos, &blkid);
//printf("voici bg %d m %d vlen %d vnb %d fcolj %d vpos %d taupos %d \n", bg, m, vlen, vnb, colst+1, vpos+1, taupos+1);
if ((vlen > 0) && (vnb > 0)) {
#if defined(USESTREAM)
magmablasSetKernelStream(stream[0]);
magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, vlen, N1, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(fst,0), LDE, dwork, N1);
magmablasSetKernelStream(stream[1]);
magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, vlen, N2, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(fst,N1), LDE, &dwork[N1*Vblksiz], N2);
#else
magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, vlen, NE, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(fst,0), LDE, dwork, NE);
#endif
}
}
}
}
} else if (SIDE == MagmaRight) {
if (versionR == 91) {
for (bg =1; bg <= nbGblk; bg++) {
firstcolj = (bg-1)*Vblksiz + 1;
rownbm = plasma_ceildiv((N-(firstcolj+1)), NB);
if (bg == nbGblk) rownbm = plasma_ceildiv((N-(firstcolj)), NB); // last blk has size=1 used for complex to handle A(N,N-1)
for (m = 1; m <= rownbm; m++) {
vlen = 0;
vnb = 0;
// for k=0; I compute the fst and then can remove it from the loop
colj = (bg-1)*Vblksiz;
fst = (rownbm -m)*NB+colj +1;
for (k=0; k < Vblksiz; k++) {
colj = (bg-1)*Vblksiz + k;
st = (rownbm -m)*NB+colj +1;
ed = min(st+NB-1, N-1);
if (st > ed) break;
if ((st == ed) && (colj != N-2)) break;
vlen=ed-fst+1;
vnb=k+1;
}
colj = (bg-1)*Vblksiz;
findVTpos(N, NB, Vblksiz, colj, fst, &vpos, &taupos, &tpos, &blkid);
//printf("voici bg %d m %d vlen %d vnb %d fcolj %d vpos %d taupos %d \n", bg, m, vlen, vnb, colj, vpos, taupos);
if ((vlen > 0) && (vnb > 0)) {
#if defined(USESTREAM)
magmablasSetKernelStream(stream[0]);
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, N1, vlen, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(0, fst), LDE, dwork, N1);
magmablasSetKernelStream(stream[1]);
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, N2, vlen, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(N1, fst), LDE, &dwork[N1*Vblksiz], N2);
#else
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, NE, vlen, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(0, fst), LDE, dwork, NE);
#endif
}
}
}
} else if (versionR == 92) {
rownbm = plasma_ceildiv((N-1), NB);
for (m = 1; m <= rownbm; m++) {
ncolinvolvd = min(N-1, m*NB);
avai_blksiz=min(Vblksiz, ncolinvolvd);
nbgr = plasma_ceildiv(ncolinvolvd, avai_blksiz);
for (n = 1; n <= nbgr; n++) {
vlen = 0;
vnb = 0;
cur_blksiz = min(ncolinvolvd-(n-1)*avai_blksiz, avai_blksiz);
colst = (n-1)*avai_blksiz;
coled = colst + cur_blksiz -1;
fst = (rownbm -m)*NB+colst +1;
for (colj=colst; colj <= coled; colj++) {
st = (rownbm -m)*NB+colj +1;
ed = min(st+NB-1, N-1);
if (st > ed) break;
if ((st == ed) && (colj != N-2)) break;
vlen=ed-fst+1;
vnb=vnb+1;
}
findVTpos(N, NB, Vblksiz, colst, fst, &vpos, &taupos, &tpos, &blkid);
if ((vlen > 0) && (vnb > 0)) {
#if defined(USESTREAM)
magmablasSetKernelStream(stream[0]);
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, N1, vlen, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(0, fst), LDE, dwork, N1);
magmablasSetKernelStream(stream[1]);
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, N2, vlen, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(N1, fst), LDE, &dwork[N1*Vblksiz], N2);
#else
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, NE, vlen, vnb, dV(vpos), LDV, dT(tpos), LDT, dE(0, fst), LDE, dwork, NE);
#endif
}
}
}
}
} else {
printf("ERROR SIDE %d\n", SIDE);
}
#if defined(USESTREAM)
magma_queue_destroy( stream[0] );
magma_queue_destroy( stream[1] );
#endif
magmablasSetKernelStream( orig_stream );
}
ssize_t strbuf_write (strbuf_t *buf, int fd)
{
struct strblk *cur;
ssize_t rsize, wsize;
struct iovec *iov;
int ioc; /* helper/index */
int iot; /* total number of buffers */
int iow; /* total number of buffer that will be passed to the next writev */
rsize = 0;
iov = NULL;
cur = buf->beg;
iot = (buf->size / buf->blkmax) + 1;
dI("total I/O vectors = %d\n", iot);
while (iot > 0) {
/*
* Prepare I/O vector
*/
if (iot > IOV_MAX) {
dI("iot (%d) > IOV_MAX (%d)\n", iot, IOV_MAX);
iow = IOV_MAX;
iot -= IOV_MAX;
} else {
dI("iot (%d) < IOV_MAX (%d)\n", iot, IOV_MAX);
iow = iot;
iot = 0;
}
iov = realloc (iov, sizeof (struct iovec) * iow);
ioc = 0;
while (cur != NULL && ioc < iow) {
iov[ioc].iov_base = cur->data;
iov[ioc].iov_len = cur->size;
++ioc;
cur = cur->next;
}
dI("ioc = %d\n", ioc);
/*
* Write
*/
wsize = writev (fd, iov, ioc);
if (wsize < 0) {
dE("writev(%d, %p, %d) failed: %u, %s.\n", fd, iov, ioc, errno, strerror (errno));
free(iov);
return (-1);
}
rsize += wsize;
}
free (iov);
dI("total bytes written: %zu\n", (size_t)rsize);
return (rsize);
}
int oval_object_to_sexp(void *sess, const char *typestr, struct oval_syschar *syschar, SEXP_t **out_sexp)
{
unsigned int ent_cnt, varref_cnt;
int ret;
SEXP_t *obj_sexp, *elm, *varrefs, *ent_lst, *lst, *stmp;
SEXP_t *r0, *r1, *r2, *obj_attr, sm0, sm1;
struct oval_object *object;
struct oval_object_content_iterator *cit;
struct oval_behavior_iterator *bit;
struct oval_object_content *content;
struct oval_entity *entity;
char obj_name[128];
const char *obj_id;
object = oval_syschar_get_object(syschar);
/*
* Object name & attributes (id)
*/
ret = snprintf(obj_name, sizeof obj_name, "%s_object", typestr);
if (ret<0 || (unsigned int) ret > sizeof obj_name) {
dE("obj_name length too short");
return -1;
}
// even though it returns const char* it has to be freed :-(
char *obj_over = (char*)oval_schema_version_to_cstr(oval_object_get_platform_schema_version(object));
obj_id = oval_object_get_id(object);
obj_attr = probe_attr_creat("id", SEXP_string_new_r(&sm0, obj_id, strlen(obj_id)),
"oval_version", SEXP_string_new_r(&sm1, obj_over, strlen(obj_over)),
NULL);
free(obj_over);
obj_sexp = probe_obj_new(obj_name, obj_attr);
SEXP_free_r(&sm0);
SEXP_free_r(&sm1);
SEXP_free(obj_attr);
/*
* Object content
*/
ent_lst = SEXP_list_new(NULL);
varrefs = NULL;
ent_cnt = varref_cnt = 0;
cit = oval_object_get_object_contents(object);
while (oval_object_content_iterator_has_more(cit)) {
oval_check_t ochk;
oval_entity_varref_type_t vr_type;
content = oval_object_content_iterator_next(cit);
elm = NULL;
lst = ent_lst;
switch (oval_object_content_get_type(content)) {
case OVAL_OBJECTCONTENT_ENTITY:
entity = oval_object_content_get_entity(content);
elm = oval_entity_to_sexp(entity);
if (elm == NULL)
break;
ochk = oval_object_content_get_varCheck(content);
if (ochk != OVAL_CHECK_UNKNOWN) {
probe_ent_attr_add(elm, "var_check",
r0 = SEXP_number_newu_32(ochk));
SEXP_free(r0);
}
ret = 0;
vr_type = oval_entity_get_varref_type(entity);
if (vr_type == OVAL_ENTITY_VARREF_ATTRIBUTE) {
const char *var_id = oval_variable_get_id(oval_entity_get_variable(entity));
const char *field_name = oval_object_content_get_field_name(content);
dI("Object '%s' references variable '%s' in '%s' field.", obj_id, var_id, field_name);
ret = oval_varref_attr_to_sexp(sess, entity, syschar, &stmp);
if (ret == 0) {
if (varrefs == NULL)
varrefs = SEXP_list_new(NULL);
SEXP_list_add(varrefs, stmp);
SEXP_free(stmp);
// todo: don't add duplicates
++varref_cnt;
lst = obj_sexp;
++ent_cnt;
}
} else if (vr_type == OVAL_ENTITY_VARREF_ELEMENT) {
SEXP_t *val_lst;
struct oval_variable *var;
oval_datatype_t dt;
var = oval_entity_get_variable(entity);
dt = oval_entity_get_datatype(entity);
ret = oval_varref_elm_to_sexp(sess, var, dt, &val_lst, syschar);
if (ret == 0) {
SEXP_list_add(elm, val_lst);
SEXP_free(val_lst);
}
}
if (ret != 0) {
SEXP_t s_flag;
SEXP_number_newi_32_r(&s_flag, SYSCHAR_FLAG_DOES_NOT_EXIST);
probe_item_attr_add(obj_sexp, "skip_eval", &s_flag);
SEXP_free_r(&s_flag);
SEXP_free(elm);
SEXP_free(ent_lst);
if (varrefs != NULL)
SEXP_free(varrefs);
oval_object_content_iterator_free(cit);
*out_sexp = obj_sexp;
return (0);
}
break;
case OVAL_OBJECTCONTENT_SET:
elm = oval_set_to_sexp(oval_object_content_get_setobject(content));
break;
case OVAL_OBJECTCONTENT_FILTER: {
struct oval_filter *filter = oval_object_content_get_filter(content);
struct oval_state *ste = oval_filter_get_state(filter);
const char *ste_id = oval_state_get_id(ste);
oval_filter_action_t action = oval_filter_get_filter_action(filter);
const char *action_text = oval_filter_action_get_text(action);
dI("Object '%s' has a filter that %ss items conforming to state '%s'.",
obj_id, action_text, ste_id);
elm = oval_filter_to_sexp(filter);
}
break;
case OVAL_OBJECTCONTENT_UNKNOWN:
break;
}
if (elm == NULL) {
SEXP_free(obj_sexp);
SEXP_free(ent_lst);
if (varrefs != NULL)
SEXP_free(varrefs);
oval_object_content_iterator_free(cit);
return -1;
}
SEXP_list_add(lst, elm);
SEXP_free(elm);
}
if (varrefs != NULL) {
// todo: SEXP_list_push()
stmp = SEXP_list_new(r0 = SEXP_string_new("varrefs", 7),
r1 = SEXP_number_newu(varref_cnt), r2 = SEXP_number_newu(ent_cnt), NULL);
SEXP_vfree(r0, r1, r2, NULL);
r0 = SEXP_list_join(stmp, varrefs);
SEXP_list_add(obj_sexp, r0);
SEXP_vfree(stmp, varrefs, r0, NULL);
}
stmp = SEXP_list_join(obj_sexp, ent_lst);
SEXP_free(obj_sexp);
SEXP_free(ent_lst);
obj_sexp = stmp;
oval_object_content_iterator_free(cit);
/*
* Object behaviors
*/
bit = oval_object_get_behaviors(object);
if (oval_behavior_iterator_has_more(bit)) {
elm = oval_behaviors_to_sexp(bit);
SEXP_list_add(obj_sexp, elm);
SEXP_free(elm);
}
oval_behavior_iterator_free(bit);
*out_sexp = obj_sexp;
return (0);
}
oval_result_t oval_ipaddr_cmp(int af, const char *s1, const char *s2, oval_operation_t op)
{
oval_result_t result = OVAL_RESULT_ERROR;
uint32_t mask1 = 0, mask2 = 0;
char addr1[INET6_ADDRSTRLEN];
char addr2[INET6_ADDRSTRLEN];
if (ipaddr_parse(af, s1, &mask1, &addr1) || ipaddr_parse(af, s2, &mask2, &addr2)) {
return result;
}
switch (op) {
case OVAL_OPERATION_EQUALS:
if (!ipaddr_cmp(af, &addr1, &addr2) && mask1 == mask2)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
case OVAL_OPERATION_NOT_EQUAL:
if (ipaddr_cmp(af, &addr1, &addr2) || mask1 != mask2)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
case OVAL_OPERATION_SUBSET_OF:
/* This asserts that every IP address in the set of IP addresses
* on the system (add2, mask2) must be present in the set of IP
* addresses defined in the stated entity (addr1, mask1). */
if (mask1 > mask2) {
/* The bigger the netmask (IPv4) or prefix-length (IPv6) is
* the less IP addresses there are in the range. */
result = OVAL_RESULT_FALSE;
break;
}
/* Otherwise, compare the first bits defined by mask1 */
ipaddr_mask(af, &addr1, mask1);
ipaddr_mask(af, &addr2, mask1);
if (ipaddr_cmp(af, &addr1, &addr2) == 0)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
case OVAL_OPERATION_GREATER_THAN:
ipaddr_mask(af, &addr1, mask1);
ipaddr_mask(af, &addr2, mask2);
if (ipaddr_cmp(af, &addr1, &addr2) < 0)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
case OVAL_OPERATION_GREATER_THAN_OR_EQUAL:
ipaddr_mask(af, &addr1, mask1);
ipaddr_mask(af, &addr2, mask2);
if (ipaddr_cmp(af, &addr1, &addr2) <= 0)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
case OVAL_OPERATION_SUPERSET_OF:
/* This asserts that every IP address in the set of IP addresses defined in
* the stated entity (addr1, mask1) is present in the set of IP addresses
* on the system. (addr2, mask2). */
if (mask1 < mask2) {
/* The smaller the netmask (IPv4) or prefix-length (IPv6) is
* the more IP addresses there are in the range */
result = OVAL_RESULT_FALSE;
break;
}
/* Otherwise, compare the first bits defined by mask2 */
ipaddr_mask(af, &addr1, mask2);
ipaddr_mask(af, &addr2, mask2);
if (ipaddr_cmp(af, &addr1, &addr2) == 0)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
case OVAL_OPERATION_LESS_THAN:
ipaddr_mask(af, &addr1, mask1);
ipaddr_mask(af, &addr2, mask2);
if (ipaddr_cmp(af, &addr1, &addr2) > 0)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
case OVAL_OPERATION_LESS_THAN_OR_EQUAL:
ipaddr_mask(af, &addr1, mask1);
ipaddr_mask(af, &addr2, mask2);
if (ipaddr_cmp(af, &addr1, &addr2) >= 0)
result = OVAL_RESULT_TRUE;
else
result = OVAL_RESULT_FALSE;
break;
default:
dE("Unexpected compare operation: %d.\n", op);
assert(false);
}
return result;
}
void *probe_signal_handler(void *arg)
{
probe_t *probe = (probe_t *)arg;
siginfo_t siinf;
sigset_t siset;
#if defined(HAVE_PTHREAD_SETNAME_NP)
# if defined(__APPLE__)
pthread_setname_np("signal_handler");
# else
pthread_setname_np(pthread_self(), "signal_handler");
# endif
#endif
sigemptyset(&siset);
sigaddset(&siset, SIGHUP);
sigaddset(&siset, SIGUSR1);
sigaddset(&siset, SIGUSR2);
sigaddset(&siset, SIGINT);
sigaddset(&siset, SIGTERM);
sigaddset(&siset, SIGQUIT);
sigaddset(&siset, SIGPIPE);
#if defined(__linux__)
if (prctl(PR_SET_PDEATHSIG, SIGTERM) != 0)
dW("prctl(PR_SET_PDEATHSIG, SIGTERM) failed");
#endif
dD("Signal handler ready");
switch (errno = pthread_barrier_wait(&OSCAP_GSYM(th_barrier)))
{
case 0:
case PTHREAD_BARRIER_SERIAL_THREAD:
break;
default:
dE("pthread_barrier_wait: %d, %s.", errno, strerror(errno));
return (NULL);
}
while (sigwaitinfo(&siset, &siinf) != -1) {
dD("Received signal %d from %u (%s)",
siinf.si_signo, (unsigned int)siinf.si_pid,
getppid() == siinf.si_pid ? "parent" : "not my parent");
#if defined(PROBE_SIGNAL_PARENTONLY)
/* Listen only to signals sent from the parent process */
if (getppid() != siinf.si_pid)
continue;
#endif
switch(siinf.si_signo) {
case SIGUSR1:/* probe abort */
probe->probe_exitcode = ECONNABORTED;
/* FALLTHROUGH */
case SIGINT:
case SIGTERM:
case SIGQUIT:
case SIGPIPE:
{
__thr_collection coll;
coll.thr = NULL;
coll.cnt = 0;
pthread_cancel(probe->th_input);
/* collect IDs and cancel threads */
rbt_walk_inorder2(probe->workers, __abort_cb, &coll, 0);
/*
* Wait till all threads are canceled (they may temporarily disable
* cancelability), but at most 60 seconds per thread.
*/
for (; coll.cnt > 0; --coll.cnt) {
probe_worker_t *thr = coll.thr[coll.cnt - 1];
#if defined(HAVE_PTHREAD_TIMEDJOIN_NP) && defined(HAVE_CLOCK_GETTIME)
struct timespec j_tm;
if (clock_gettime(CLOCK_REALTIME, &j_tm) == -1) {
dE("clock_gettime(CLOCK_REALTIME): %d, %s.", errno, strerror(errno));
continue;
}
j_tm.tv_sec += 60;
if ((errno = pthread_timedjoin_np(thr->tid, NULL, &j_tm)) != 0) {
dE("[%llu] pthread_timedjoin_np: %d, %s.", (uint64_t)thr->sid, errno, strerror(errno));
/*
* Memory will be leaked here by continuing to the next thread. However, we are in the
* process of shutting down the whole probe. We're just nice and gave the probe_main()
* thread a chance to finish it's critical section which shouldn't take that long...
*/
continue;
}
#else
if ((errno = pthread_join(thr->tid, NULL)) != 0) {
dE("pthread_join: %d, %s.", errno, strerror(errno));
continue;
}
#endif
SEAP_msg_free(coll.thr[coll.cnt - 1]->msg);
oscap_free(coll.thr[coll.cnt - 1]);
}
oscap_free(coll.thr);
goto exitloop;
}
case SIGUSR2:
case SIGHUP:
/* ignore */
break;
}
}
exitloop:
return (NULL);
}
static int read_environment(SEXP_t *pid_ent, SEXP_t *name_ent, probe_ctx *ctx)
{
int err = 1, pid, fd;
bool empty;
size_t env_name_size;
SEXP_t *env_name, *env_value, *item, *pid_sexp;
DIR *d;
struct dirent *d_entry;
char *buffer, env_file[256], *null_char;
ssize_t buffer_used;
size_t buffer_size;
d = opendir("/proc");
if (d == NULL) {
dE("Can't read /proc: errno=%d, %s.\n", errno, strerror (errno));
return PROBE_EACCESS;
}
if ((buffer = oscap_realloc(NULL, BUFFER_SIZE)) == NULL) {
dE("Can't allocate memory");
closedir(d);
return PROBE_EFAULT;
}
buffer_size = BUFFER_SIZE;
while ((d_entry = readdir(d))) {
if (strspn(d_entry->d_name, "0123456789") != strlen(d_entry->d_name))
continue;
pid = atoi(d_entry->d_name);
pid_sexp = SEXP_number_newi_32(pid);
if (probe_entobj_cmp(pid_ent, pid_sexp) != OVAL_RESULT_TRUE) {
SEXP_free(pid_sexp);
continue;
}
SEXP_free(pid_sexp);
sprintf(env_file, "/proc/%d/environ", pid);
if ((fd = open(env_file, O_RDONLY)) == -1) {
dE("Can't open \"%s\": errno=%d, %s.\n", env_file, errno, strerror (errno));
item = probe_item_create(
OVAL_INDEPENDENT_ENVIRONMENT_VARIABLE58, NULL,
"pid", OVAL_DATATYPE_INTEGER, (int64_t)pid,
NULL
);
probe_item_setstatus(item, SYSCHAR_STATUS_ERROR);
probe_item_add_msg(item, OVAL_MESSAGE_LEVEL_ERROR,
"Can't open \"%s\": errno=%d, %s.", env_file, errno, strerror (errno));
probe_item_collect(ctx, item);
continue;
}
empty = true;
if ((buffer_used = read(fd, buffer, buffer_size - 1)) > 0) {
empty = false;
}
while (! empty) {
while (! (null_char = memchr(buffer, 0, buffer_used))) {
ssize_t s;
if ((size_t)buffer_used >= buffer_size) {
buffer_size += BUFFER_SIZE;
buffer = oscap_realloc(buffer, buffer_size);
if (buffer == NULL) {
dE("Can't allocate memory");
exit(ENOMEM);
}
}
s = read(fd, buffer + buffer_used, buffer_size - buffer_used);
if (s <= 0) {
empty = true;
buffer[buffer_used++] = 0;
}
else {
buffer_used += s;
}
}
do {
char *eq_char = strchr(buffer, '=');
if (eq_char == NULL) {
/* strange but possible:
* $ strings /proc/1218/environ
/dev/input/event0 /dev/input/event1 /dev/input/event4 /dev/input/event3
*/
buffer_used -= null_char + 1 - buffer;
memmove(buffer, null_char + 1, buffer_used);
continue;
}
env_name_size = eq_char - buffer;
env_name = SEXP_string_new(buffer, env_name_size);
env_value = SEXP_string_newf("%s", buffer + env_name_size + 1);
if (probe_entobj_cmp(name_ent, env_name) == OVAL_RESULT_TRUE) {
item = probe_item_create(
OVAL_INDEPENDENT_ENVIRONMENT_VARIABLE58, NULL,
"pid", OVAL_DATATYPE_INTEGER, (int64_t)pid,
"name", OVAL_DATATYPE_SEXP, env_name,
"value", OVAL_DATATYPE_SEXP, env_value,
NULL);
probe_item_collect(ctx, item);
err = 0;
}
SEXP_free(env_name);
SEXP_free(env_value);
buffer_used -= null_char + 1 - buffer;
memmove(buffer, null_char + 1, buffer_used);
} while ((null_char = memchr(buffer, 0, buffer_used)));
}
close(fd);
}
closedir(d);
oscap_free(buffer);
if (err) {
SEXP_t *msg = probe_msg_creatf(OVAL_MESSAGE_LEVEL_ERROR,
"Can't find process with requested PID.");
probe_cobj_add_msg(probe_ctx_getresult(ctx), msg);
SEXP_free(msg);
err = 0;
}
return err;
}
static void *probe_icache_worker(void *arg)
{
probe_icache_t *cache = (probe_icache_t *)(arg);
probe_iqpair_t *pair, pair_mem;
SEXP_ID_t item_ID;
assume_d(cache != NULL, NULL);
pthread_setname_np(pthread_self(), "icache_worker");
if (pthread_mutex_lock(&cache->queue_mutex) != 0) {
dE("An error ocured while locking the queue mutex: %u, %s",
errno, strerror(errno));
return (NULL);
}
pair = &pair_mem;
dD("icache worker ready");
switch (errno = pthread_barrier_wait(&OSCAP_GSYM(th_barrier)))
{
case 0:
case PTHREAD_BARRIER_SERIAL_THREAD:
break;
default:
dE("pthread_barrier_wait: %d, %s.",
errno, strerror(errno));
pthread_mutex_unlock(&cache->queue_mutex);
return (NULL);
}
while(pthread_cond_wait(&cache->queue_notempty, &cache->queue_mutex) == 0) {
assume_d(cache->queue_cnt > 0, NULL);
next:
dD("Extracting item from the cache queue: cnt=%"PRIu16", beg=%"PRIu16"", cache->queue_cnt, cache->queue_beg);
/*
* Extract an item from the queue and update queue beg, end & cnt
*/
pair_mem = cache->queue[cache->queue_beg];
#ifndef NDEBUG
memset(cache->queue + cache->queue_beg, 0, sizeof(probe_iqpair_t));
#endif
--cache->queue_cnt;
++cache->queue_beg;
if (cache->queue_beg == cache->queue_max)
cache->queue_beg = 0;
assume_d(cache->queue_cnt == 0 ?
cache->queue_end == cache->queue_beg :
cache->queue_end != cache->queue_beg, NULL);
/*
* Release the mutex
*/
if (pthread_mutex_unlock(&cache->queue_mutex) != 0) {
dE("An error ocured while unlocking the queue mutex: %u, %s",
errno, strerror(errno));
abort();
}
dD("Signaling `notfull'");
if (pthread_cond_signal(&cache->queue_notfull) != 0) {
dE("An error ocured while signaling the `notfull' condition: %u, %s",
errno, strerror(errno));
abort();
}
if (pair->cobj == NULL) {
/*
* Handle NOP case (synchronization)
*/
assume_d(pair->p.cond != NULL, NULL);
dD("Handling NOP");
if (pthread_cond_signal(pair->p.cond) != 0) {
dE("An error ocured while signaling NOP condition: %u, %s",
errno, strerror(errno));
abort();
}
} else {
probe_citem_t *cached = NULL;
dD("Handling cache request");
/*
* Compute item ID
*/
item_ID = SEXP_ID_v(pair->p.item);
dD("item ID=%"PRIu64"", item_ID);
/*
* Perform cache lookup
*/
if (rbt_i64_get(cache->tree, (int64_t)item_ID, (void *)&cached) == 0) {
register uint16_t i;
SEXP_t rest1, rest2;
/*
* Maybe a cache HIT
*/
dD("cache HIT #1");
for (i = 0; i < cached->count; ++i) {
if (SEXP_deepcmp(SEXP_list_rest_r(&rest1, pair->p.item),
SEXP_list_rest_r(&rest2, cached->item[i])))
{
SEXP_free_r(&rest1);
SEXP_free_r(&rest2);
break;
}
SEXP_free_r(&rest1);
SEXP_free_r(&rest2);
}
if (i == cached->count) {
/*
* Cache MISS
*/
dD("cache MISS");
cached->item = oscap_realloc(cached->item, sizeof(SEXP_t *) * ++cached->count);
cached->item[cached->count - 1] = pair->p.item;
/* Assign an unique item ID */
probe_icache_item_setID(pair->p.item, item_ID);
} else {
/*
* Cache HIT
*/
dD("cache HIT #2 -> real HIT");
SEXP_free(pair->p.item);
pair->p.item = cached->item[i];
}
} else {
/*
* Cache MISS
*/
dD("cache MISS");
cached = oscap_talloc(probe_citem_t);
cached->item = oscap_talloc(SEXP_t *);
cached->item[0] = pair->p.item;
cached->count = 1;
/* Assign an unique item ID */
probe_icache_item_setID(pair->p.item, item_ID);
if (rbt_i64_add(cache->tree, (int64_t)item_ID, (void *)cached, NULL) != 0) {
dE("Can't add item (k=%"PRIi64" to the cache (%p)", (int64_t)item_ID, cache->tree);
oscap_free(cached->item);
oscap_free(cached);
/* now what? */
abort();
}
}
if (probe_cobj_add_item(pair->cobj, pair->p.item) != 0) {
dW("An error ocured while adding the item to the collected object");
}
}
if (pthread_mutex_lock(&cache->queue_mutex) != 0) {
dE("An error ocured while re-locking the queue mutex: %u, %s",
errno, strerror(errno));
abort();
}
if (cache->queue_cnt > 0)
goto next;
}
return (NULL);
}
static inline oval_result_t _evaluate_sysent_with_variable(struct oval_syschar_model *syschar_model, struct oval_entity *state_entity, struct oval_sysent *item_entity, oval_operation_t state_entity_operation, struct oval_state_content *content)
{
oval_syschar_collection_flag_t flag;
oval_result_t ent_val_res;
struct oval_variable *state_entity_var;
if ((state_entity_var = oval_entity_get_variable(state_entity)) == NULL) {
oscap_seterr(OSCAP_EFAMILY_OVAL, "OVAL internal error: found NULL variable");
return -1;
}
if (0 != oval_syschar_model_compute_variable(syschar_model, state_entity_var)) {
return -1;
}
flag = oval_variable_get_collection_flag(state_entity_var);
switch (flag) {
case SYSCHAR_FLAG_COMPLETE:
case SYSCHAR_FLAG_INCOMPLETE:{
struct oresults var_ores;
struct oval_value_iterator *val_itr;
ores_clear(&var_ores);
val_itr = oval_variable_get_values(state_entity_var);
while (oval_value_iterator_has_more(val_itr)) {
struct oval_value *var_val;
char *state_entity_val_text = NULL;
oval_result_t var_val_res;
var_val = oval_value_iterator_next(val_itr);
state_entity_val_text = oval_value_get_text(var_val);
if (state_entity_val_text == NULL) {
dE("Found NULL variable value text.");
ores_add_res(&var_ores, OVAL_RESULT_ERROR);
break;
}
oval_datatype_t state_entity_val_datatype = oval_value_get_datatype(var_val);
var_val_res = oval_ent_cmp_str(state_entity_val_text, state_entity_val_datatype, item_entity, state_entity_operation);
if (var_val_res == OVAL_RESULT_ERROR) {
dE("Error occured when comparing a variable '%s' value '%s' with collected item entity = '%s'",
oval_variable_get_id(state_entity_var), state_entity_val_text, oval_sysent_get_value(item_entity));
}
ores_add_res(&var_ores, var_val_res);
}
oval_value_iterator_free(val_itr);
oval_check_t var_check = oval_state_content_get_var_check(content);
ent_val_res = ores_get_result_bychk(&var_ores, var_check);
} break;
case SYSCHAR_FLAG_ERROR:
case SYSCHAR_FLAG_DOES_NOT_EXIST:
case SYSCHAR_FLAG_NOT_COLLECTED:
case SYSCHAR_FLAG_NOT_APPLICABLE:
ent_val_res = OVAL_RESULT_ERROR;
break;
default:
ent_val_res = -1;
}
return ent_val_res;
}
static int rpmverify_collect(probe_ctx *ctx,
const char *name, oval_operation_t name_op,
const char *file, oval_operation_t file_op,
SEXP_t *name_ent, SEXP_t *filepath_ent,
uint64_t flags,
void (*callback)(probe_ctx *, struct rpmverify_res *))
{
rpmdbMatchIterator match;
rpmVerifyAttrs omit = (rpmVerifyAttrs)(flags & RPMVERIFY_RPMATTRMASK);
Header pkgh;
pcre *re = NULL;
int ret = -1;
/* pre-compile regex if needed */
if (file_op == OVAL_OPERATION_PATTERN_MATCH) {
const char *errmsg;
int erroff;
re = pcre_compile(file, PCRE_UTF8, &errmsg, &erroff, NULL);
if (re == NULL) {
/* TODO */
return (-1);
}
}
RPMVERIFY_LOCK;
switch (name_op) {
case OVAL_OPERATION_EQUALS:
match = rpmtsInitIterator (g_rpm.rpmts, RPMTAG_NAME, (const void *)name, 0);
if (match == NULL) {
ret = 0;
goto ret;
}
ret = rpmdbGetIteratorCount (match);
break;
case OVAL_OPERATION_NOT_EQUAL:
match = rpmtsInitIterator (g_rpm.rpmts, RPMDBI_PACKAGES, NULL, 0);
if (match == NULL) {
ret = 0;
goto ret;
}
if (rpmdbSetIteratorRE (match, RPMTAG_NAME, RPMMIRE_GLOB, "*") != 0)
{
ret = -1;
goto ret;
}
break;
case OVAL_OPERATION_PATTERN_MATCH:
match = rpmtsInitIterator (g_rpm.rpmts, RPMDBI_PACKAGES, NULL, 0);
if (match == NULL) {
ret = 0;
goto ret;
}
if (rpmdbSetIteratorRE (match, RPMTAG_NAME, RPMMIRE_REGEX,
(const char *)name) != 0)
{
ret = -1;
goto ret;
}
break;
default:
/* not supported */
dE("package name: operation not supported");
ret = -1;
goto ret;
}
assume_d(RPMTAG_BASENAMES != 0, -1);
assume_d(RPMTAG_DIRNAMES != 0, -1);
while ((pkgh = rpmdbNextIterator (match)) != NULL) {
rpmfi fi;
rpmTag tag[2] = { RPMTAG_BASENAMES, RPMTAG_DIRNAMES };
struct rpmverify_res res;
errmsg_t rpmerr;
int i;
SEXP_t *name_sexp;
res.name = headerFormat(pkgh, "%{NAME}", &rpmerr);
name_sexp = SEXP_string_newf("%s", res.name);
if (probe_entobj_cmp(name_ent, name_sexp) != OVAL_RESULT_TRUE) {
SEXP_free(name_sexp);
continue;
}
SEXP_free(name_sexp);
/*
* Inspect package files & directories
*/
for (i = 0; i < 2; ++i) {
fi = rpmfiNew(g_rpm.rpmts, pkgh, tag[i], 1);
while (rpmfiNext(fi) != -1) {
SEXP_t *filepath_sexp;
res.fflags = rpmfiFFlags(fi);
res.oflags = omit;
if (((res.fflags & RPMFILE_CONFIG) && (flags & RPMVERIFY_SKIP_CONFIG)) ||
((res.fflags & RPMFILE_GHOST) && (flags & RPMVERIFY_SKIP_GHOST)))
continue;
res.file = strdup(rpmfiFN(fi));
filepath_sexp = SEXP_string_newf("%s", res.file);
if (probe_entobj_cmp(filepath_ent, filepath_sexp) != OVAL_RESULT_TRUE) {
SEXP_free(filepath_sexp);
free(res.file);
continue;
}
SEXP_free(filepath_sexp);
if (rpmVerifyFile(g_rpm.rpmts, fi, &res.vflags, omit) != 0)
res.vflags = RPMVERIFY_FAILURES;
callback(ctx, &res);
free(res.file);
}
rpmfiFree(fi);
}
}
match = rpmdbFreeIterator (match);
ret = 0;
ret:
if (re != NULL)
pcre_free(re);
RPMVERIFY_UNLOCK;
return (ret);
}
extern "C" magma_int_t
magma_zbulge_applyQ_v2_m(
magma_int_t ngpu,
magma_side_t side,
magma_int_t NE, magma_int_t N,
magma_int_t NB, magma_int_t Vblksiz,
magmaDoubleComplex *E, magma_int_t lde,
magmaDoubleComplex *V, magma_int_t ldv,
magmaDoubleComplex *T, magma_int_t ldt,
magma_int_t *info)
{
//%===========================
//% local variables
//%===========================
magma_int_t Vm, Vn, mt, nt;
magma_int_t myrow, mycol, blkj, blki;
magma_int_t blkid,vpos,tpos;
magma_int_t firstrow, nbcolinvolvd;
magma_int_t versionL = 113;
magma_int_t versionR = 92;
magma_int_t Vchunksiz = 10;
*info=0;
/* Quick return */
if ( NE == 0 ) {
return *info;
}
if ( N == 0 ) {
return *info;
}
if ( NB == 0 ) {
return *info;
}
/* ==========================================
* some infos for developer
* Initialisation and checking nb of cores
* ==========================================*/
/* we have 2 algo for left (113 114) and 2 algo for right (91 92)
* which correspond to versionL versionR.
* They are very similar (detail explained in tech report and matlab code)
* however version 114 and 92 improve locality.
* while version 113 is used in case WNATZ=1 (construct Q2) which allow
* the construction to be done in an optimized way taking into
* consideration that the matrix is Identity so making less flops.
*
*/
// Initialize streaming and events
magma_device_sync();
magma_device_t orig_dev;
magma_getdevice( &orig_dev );
magma_queue_t orig_stream;
magmablasGetKernelStream( &orig_stream );
magma_int_t nbevents =2, nstream=2;
magma_queue_t streams[MagmaMaxGPUs][20];
magma_event_t myevent[MagmaMaxGPUs][20];
for( magma_int_t dev = 0; dev < ngpu; ++dev ) {
magma_setdevice( dev );
for( magma_int_t i = 0; i < nstream; ++i ) {
magma_queue_create( &streams[dev][i] );
}
for( magma_int_t i = 0; i < nbevents; ++i ) {
cudaEventCreateWithFlags(&myevent[dev][i],cudaEventDisableTiming);
}
}
// Azzam 21/11/2012
// NOTE THAT dwork was of size 2*NE*Vblksiz+...
// but I am thinking why not modifing it to NE*Vblksiz+...
// BUT NO because the 2* is used because of making 2 streams working and so
// they might be using dwork in parallel
magmaDoubleComplex *dE[MagmaMaxGPUs];
magmaDoubleComplex *dwork[MagmaMaxGPUs], *dwork0[MagmaMaxGPUs], *dwork1[MagmaMaxGPUs];
//magmaDoubleComplex *dwvt[MagmaMaxGPUs];
magmaDoubleComplex *dwvt0[MagmaMaxGPUs], *dwvt1[MagmaMaxGPUs];
magmaDoubleComplex *dT0[MagmaMaxGPUs], *dV0[MagmaMaxGPUs], *dT1[MagmaMaxGPUs], *dV1[MagmaMaxGPUs];
magma_int_t dev;
magma_int_t ldde = N;
magma_int_t lddv = ldv;
magma_int_t lddt = ldt;
magma_int_t ne_loc = magma_ceildiv(NE, ngpu);
if (ne_loc < 256)
ne_loc=256;
magma_int_t dwVTsiz = lddv*Vblksiz; // lddv*lddv + lddv*NE; // lddv*Vblksiz;
magma_int_t dworksiz = ne_loc*Vblksiz; // lddv*Vblksiz; // NE*Vblksiz;
ngpu = min(ngpu, magma_ceildiv(NE,ne_loc)); // Don't use GPU that will not have data.
// copy dE to GPUs
for (dev=0; dev < ngpu; ++dev) {
magma_setdevice( dev );
if (MAGMA_SUCCESS != magma_zmalloc( &dE[dev], ldde * ne_loc)) {
printf ("!!!! magma_zbulge_applyQ magma_alloc failed for: dE\n" );
exit(-1);
}
if (MAGMA_SUCCESS != magma_zmalloc( &dwork[dev], 2*dworksiz + 2*dwVTsiz + 2*Vchunksiz* (Vblksiz* (lddv+lddt)) )) {
printf ("!!!! magma_zbulge_applyQ magma_alloc failed for: dwork\n" );
exit(-1);
}
dwork0[dev] = dwork[dev]; // size = dworksiz;
dwork1[dev] = dwork0[dev] + dworksiz; // size = dworksiz;
dwvt0[dev] = dwork[dev] + 2*dworksiz; // size = dwVTsiz;
dwvt1[dev] = dwvt0[dev] + dwVTsiz; // size = dwVTsiz;
dV0[dev] = dwork[dev] + 2*dworksiz + 2*dwVTsiz;
dT0[dev] = dV0[dev] + Vchunksiz*Vblksiz*lddv;
dV1[dev] = dT0[dev] + Vchunksiz*Vblksiz*lddt;
dT1[dev] = dV1[dev] + Vchunksiz*Vblksiz*lddv;
magma_int_t ie_loc = min(ne_loc, NE - ne_loc*dev);
magma_zsetmatrix_async( N, ie_loc, E+lde*ne_loc*dev, lde, dE(dev, 0, 0), ldde, streams[dev][1] );
}
// make overlapped copy
magma_int_t ncpy = 0;
magma_int_t copyed=0, copyst=0;
magma_int_t blkcnt,nothing, mysiz, flip, vld,tld, locpos;
findVTsiz(N, NB, Vblksiz, &blkcnt, ¬hing);
flip = 0;
/* SIDE LEFT meaning apply E = Q*E = (q_1*q_2*.....*q_n) * E ==> so traverse Vs in reverse order (forward) from q_n to q_1
* Also E is splitten by row meaning each apply consist in a block of row (horizontal block) */
/* SIDE RIGHT meaning apply E = E*Q = E * (q_1*q_2*.....*q_n) ==> so tarverse Vs in normal order (forward) from q_1 to q_n
* Also E is splitten by col meaning each apply consist in a block of col (vertical block) */
#ifdef ENABLE_DEBUG
printf(" APPLY Q_v22_m GPU with NGPU %d N %d, NE %d, NB %d, Vblksiz %d, versionL %d versionR %d SIDE %c \n",
ngpu, N, NE, NB, Vblksiz, versionL, versionR, side);
#endif
/*
* MagmamaLeft
*/
if (side == MagmaLeft) {
/*
* Version 113:
* loop over the block_col (nt) and for each find the
* number of tiles (mt) in this block_col. then loop over mt, find
* the size of the V's(Vm,Vn) and apply it to the corresponding
* portion of E.
*/
if ( versionL == 113 ) {
nt = magma_ceildiv((N-1),Vblksiz);
for (blkj=nt-1; blkj >= 0; blkj--) {
/* the index of the first row on the top of block (blkj) */
firstrow = blkj * Vblksiz + 1;
/*find the number of tile for this block */
if ( blkj == nt-1 )
mt = magma_ceildiv( N - firstrow, NB);
else
mt = magma_ceildiv( N - (firstrow+1), NB);
/*loop over the tiles find the size of the Vs and apply it */
for (blki=mt; blki > 0; blki--) {
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + (mt-blki)*NB;
mycol = blkj*Vblksiz;
Vm = min( NB+Vblksiz-1, N-myrow);
if ( ( blkj == nt-1 ) && ( blki == mt ) ) {
Vn = min (Vblksiz, Vm);
} else {
Vn = min (Vblksiz, Vm-1);
}
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
//printf("voici blkj %d blki %d Vm %d Vn %d mycol %d vpos %d \n",blkj,blki,Vm, Vn,mycol,vpos);
magma_bulge_findpos113(N, NB, Vblksiz, mycol, myrow, &blkid);
// COPY Vchunksiz Vs and Vchunksiz Ts to GPU and store it in dV0/dV1 and dT0/dT1
if (ncpy == 0) {
// flip = 1 for this.
copyst = 0; // meaning that copy will start copying from blkid =copyst
copyed = min(copyst+Vchunksiz, blkcnt); // meaning that copy will end copying at blkid =copyed-1==> next copy had to start at copyed
mysiz = copyed-copyst; // the size of the chunk to be copied
if (mysiz > 0) {
ncpy = 1;
flip = 1;
vpos = copyst*Vblksiz*ldv;
tpos = copyst*Vblksiz*ldt;
vld = mysiz * ldv;
tld = mysiz * ldt;
for( dev = 0; dev < ngpu; ++dev ) {
magma_setdevice( dev );
magmablasSetKernelStream( streams[ dev ][ 1 ] );
magma_zsetmatrix_async(vld, Vblksiz, V(vpos), vld, dV1[dev], vld, streams[dev][1]);
magma_zsetmatrix_async(tld, Vblksiz, T(tpos), tld, dT1[dev], tld, streams[dev][1]);
}
//printf("doing the first copy of mysiz %2d copyst %2d copyed %2d vpos %8d tpos %8d into dV1 dT1\n",mysiz,copyst,copyed,vpos,tpos);
}
}
if (blkid == copyst) {
flip = ncpy % 2;
copyst = copyed; // meaning that copy will start copying from blkid =copyst
copyed = min(copyst+Vchunksiz, blkcnt); // meaning that copy will end copying at blkid =copyed-1==> next copy had to start at copyed
mysiz = copyed-copyst; // the size of the chunk to be copied
//printf(" get to copy blkid %d blkid+(2*Vchunksiz) %d copyst %d copyed %d\n",blkid,blkid+(Vchunksiz),copyst,copyed);
if (mysiz > 0) {
ncpy = ncpy + 1;
vpos = copyst*Vblksiz*ldv;
tpos = copyst*Vblksiz*ldt;
vld = mysiz * ldv;
tld = mysiz * ldt;
if (flip == 0) { // now I am working on dV0 so copy the next and put it on dV1
//printf("doing overlapping copy of mysiz %2d copyst %2d copyed %2d vpos %8d tpos %8d into dV1 dT1\n",mysiz,copyst,copyed,vpos,tpos);
for( dev = 0; dev < ngpu; ++dev ) {
magma_setdevice( dev );
magmablasSetKernelStream( streams[ dev ][ 1 ] );
magma_zsetmatrix_async(vld, Vblksiz, V(vpos), vld, dV1[dev], vld, streams[dev][1]);
magma_zsetmatrix_async(tld, Vblksiz, T(tpos), tld, dT1[dev], tld, streams[dev][1]);
}
} else { // now I am working on dV1 so copy the next and put it on dV0
//printf("doing overlapping copy of mysiz %2d copyst %2d copyed %2d vpos %8d tpos %8d into dV0 dT0\n",mysiz,copyst,copyed,vpos,tpos);
for( dev = 0; dev < ngpu; ++dev ) {
magma_setdevice( dev );
magmablasSetKernelStream( streams[ dev ][ 0 ] );
magma_zsetmatrix_async(vld, Vblksiz, V(vpos), vld, dV0[dev], vld, streams[dev][0]);
magma_zsetmatrix_async(tld, Vblksiz, T(tpos), tld, dT0[dev], tld, streams[dev][0]);
}
}
}
}
if ((Vm > 0) && (Vn > 0)) {
locpos = blkid%Vchunksiz;
magma_int_t lcvpos = locpos*Vblksiz*lddv;
magma_int_t lctpos = locpos*Vblksiz*lddt;
//printf("voici blkj %d blki %d Vm %d Vn %d mycol %d locvpos %5d loctpos %5d blkid %2d using data in dV%1d dT%1d \n",blkj,blki,Vm, Vn,mycol,lcvpos,lctpos, blkid,flip,flip);
if (flip == 0) {
for( dev = 0; dev < ngpu; ++dev ) {
magma_int_t ie_loc = min(ne_loc, NE - ne_loc*dev);
magma_int_t nr_bl = magma_ceildiv(ie_loc,10000); //nr of blocks
magma_int_t sz_bl = magma_ceildiv(ie_loc,nr_bl*64)*64; //maximum size of blocks (to have blocks of around the same size and multiple of 64)
magma_int_t ib; //size of current block
magma_setdevice( dev );
magmablasSetKernelStream(streams[dev][0]);
magma_queue_wait_event( streams[dev][0], myevent[dev][1] );
for (magma_int_t i=0; i < ie_loc; i += sz_bl) {
ib = min(sz_bl, ie_loc-i);
//magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV0[dev]+lcvpos, lddv, dT0[dev]+lctpos, lddt, dE(dev,myrow,i), ldde, dwork0[dev], ib);
magma_zlarfb_gpu_gemm( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV0[dev]+lcvpos, lddv, dT0[dev]+lctpos, lddt, dE(dev,myrow,i), ldde, dwork0[dev], ib, dwvt0[dev], Vm);
}
magma_event_record( myevent[dev][0], streams[dev][0] );
}
} else {
for( dev = 0; dev < ngpu; ++dev ) {
magma_int_t ie_loc = min(ne_loc, NE - ne_loc*dev);
magma_int_t nr_bl = magma_ceildiv(ie_loc,10000); //nr of blocks
magma_int_t sz_bl = magma_ceildiv(ie_loc,nr_bl*64)*64; //maximum size of blocks (to have blocks of around the same size and multiple of 64)
magma_int_t ib; //size of current block
magma_setdevice( dev );
magmablasSetKernelStream(streams[dev][1]);
magma_queue_wait_event( streams[dev][1], myevent[dev][0] );
for (magma_int_t i=0; i < ie_loc; i += sz_bl) {
ib = min(sz_bl, ie_loc-i);
//magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV1[dev]+lcvpos, lddv, dT1[dev]+lctpos, lddt, dE(dev,myrow,i), ldde, dwork1[dev], ib);
magma_zlarfb_gpu_gemm( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV1[dev]+lcvpos, lddv, dT1[dev]+lctpos, lddt, dE(dev,myrow,i), ldde, dwork1[dev], ib, dwvt1[dev], Vm);
}
magma_event_record( myevent[dev][1], streams[dev][1] );
}
}
} // end for (Vm &Vn) > 0
} // end for blki
} // end for blkj
} // end if version=113
/*
* Version 114:
* loop over the block_row (mt) and for each find diagonally the
* number of tiles (nt) in this block_row. then loop over nt, find
* the size of the V's(Vm,Vn) and apply it to the corresponding
* portion of E.
*/
else {
printf("versionL 114 not implemented in zbulge_applyQ_v2_m\n");
exit(-1);
mt = magma_ceildiv((N-1),NB);
for (blki = mt; blki > 0; blki--) {
/* nbcolinvolvd = number of column corresponding to this block_row (blki) */
nbcolinvolvd = min(N-1, blki*NB);
/*find the number of tile for this block (diagonal row of tiles) */
nt = magma_ceildiv(nbcolinvolvd,Vblksiz);
/*loop over the tiles find the size of the Vs and apply it */
for (blkj = nt-1; blkj >= 0; blkj--) {
/* the index of the first row of the first col meaning
* the block on the top left (blki) */
firstrow = (mt-blki)*NB+1;
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + blkj*Vblksiz;
mycol = blkj*Vblksiz;
Vm = min( NB+Vblksiz-1, N-myrow);
if ( ( blkj == nt-1 ) && ( blki == mt ) ) {
Vn = min (Vblksiz, Vm);
} else {
Vn = min (Vblksiz, Vm-1);
}
if ((Vm > 0) && (Vn > 0)) {
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
/*
magma_bulge_findVTpos(N, NB, Vblksiz, mycol, myrow, ldv, ldt, &vpos, &tpos);
magma_zsetmatrix_async(Vm, Vn, V(vpos), ldv, dV0, lddv, NULL);
magma_zsetmatrix_async(Vn, Vn, T(tpos), ldt, dT0, lddt, NULL);
//printf("voici blki %d rownbm %d mycol %d coled %d blkid %d vpos %d tpos %d\n", blki, rownbm, mycol, coled, blkid, vpos, tpos);
for (magma_int_t i=0; i < NE; i += sz_bl) {
ib = min(sz_bl, NE-i);
magma_zlarfb_gpu( MagmaLeft, MagmaNoTrans, MagmaForward, MagmaColumnwise, Vm, ib, Vn, dV0, lddv, dT0, lddt, dE(myrow,i), ldde, dwork, NE);
}
*/
} // end for (Vm &Vn) > 0
} // end for blkj
} // end for blki
} // end version 114
} // end LEFT
/*
* MagmaRight
*/
else {
printf("Side 'R' not implemented in zbulge_applyQ_v2_m\n");
exit(-1);
/*
* Version 91:
*/
if ( versionR == 91 ) {
nt = magma_ceildiv((N-1),Vblksiz);
for (blkj=0; blkj < nt; blkj++) {
/* the index of the first myrow on the top of block (blkj) */
firstrow = blkj * Vblksiz + 1;
/*find the number of tile for this block */
if ( blkj == nt-1 )
mt = magma_ceildiv( N - firstrow, NB);
else
mt = magma_ceildiv( N - (firstrow+1), NB);
/*loop over the tiles find the size of the Vs and apply it */
for (blki=1; blki <= mt; blki++) {
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + (mt-blki)*NB;
Vm = min( NB+Vblksiz-1, N-myrow);
if ( (blkj == nt-1) && (blki == mt) ) {
Vn = min (Vblksiz, Vm);
} else {
Vn = min (Vblksiz, Vm-1);
}
mycol = blkj*Vblksiz;
if ((Vm > 0) && (Vn > 0)) {
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
/*
magma_bulge_findVTpos(N, NB, Vblksiz, mycol, myrow, ldv, ldt, &vpos, &tpos);
magma_zsetmatrix_async(Vm, Vn, V(vpos), ldv, dV0, lddv, NULL);
magma_zsetmatrix_async(Vn, Vn, T(tpos), ldt, dT0, lddt, NULL);
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, NE, Vm, Vn, dV0, lddv, dT0, lddt, dE(0, myrow), ldde, dwork, NE);
*/
} // end for (Vm &Vn) > 0
} // end for blki
} // end fo blkj
} // end of version 91
/*
* Version 92:
*/
else {
mt = magma_ceildiv((N-1),NB);
for (blki = 1; blki <= mt; blki++) {
/* nbcolinvolvd = number of column corresponding to this block_row (blki) */
nbcolinvolvd = min(N-1, blki*NB);
/*find the number of tile for this block (diagonal row of tiles) */
nt = magma_ceildiv(nbcolinvolvd,Vblksiz);
/*loop over the tiles find the size of the Vs and apply it */
for (blkj = 0; blkj < nt; blkj++) {
/* the index of the first row of the first col meaning
* the block on the top left (blki) */
firstrow = (mt-blki)*NB+1;
/*calculate the size of each losange of Vs= (Vm,Vn)*/
myrow = firstrow + blkj*Vblksiz;
mycol = blkj*Vblksiz;
Vm = min( NB+Vblksiz-1, N-myrow);
if ( ( blkj == nt-1 ) && ( blki == mt ) ) {
Vn = min (Vblksiz, Vm);
} else {
Vn = min (Vblksiz, Vm-1);
}
if ((Vm > 0) && (Vn > 0)) {
/*calculate the pointer to the Vs and the Ts.
* Note that Vs and Ts have special storage done
* by the bulgechasing function*/
/*
magma_bulge_findVTpos(N, NB, Vblksiz, mycol, myrow, ldv, ldt, &vpos, &tpos);
magma_zsetmatrix_async(Vm, Vn, V(vpos), ldv, dV0, lddv, NULL);
magma_zsetmatrix_async(Vn, Vn, T(tpos), ldt, dT0, lddt, NULL);
magma_zlarfb_gpu( MagmaRight, MagmaNoTrans, MagmaForward, MagmaColumnwise, NE, Vm, Vn, dV0, lddv, dT0, lddt, dE(0, myrow), ldde, dwork, NE);
*/
} // end for (Vm &Vn) > 0
} //end for blkj
} // end for blki
} //end of version 92
} // end RIGHT
// copy back the dE form each GPU
for( dev = 0; dev < ngpu; ++dev ) {
magma_setdevice( dev );
magmablasSetKernelStream(streams[dev][0]);
magma_queue_wait_event( streams[dev][0], myevent[dev][1] );
magma_queue_wait_event( streams[dev][0], myevent[dev][0] );
magma_int_t ie_loc = min(ne_loc, NE - ne_loc*dev);
magma_zgetmatrix_async( N, ie_loc, dE(dev, 0, 0), ldde, E+lde*ne_loc*dev, lde, streams[dev][0] );
magma_event_record( myevent[dev][0], streams[dev][0] );
}
for( magma_int_t dev = 0; dev < ngpu; ++dev ) {
magma_setdevice( dev );
magmablasSetKernelStream(streams[dev][0]);
magma_queue_wait_event( streams[dev][0], myevent[dev][0] );
magma_device_sync(); // no need for synchronize
magma_free(dwork[dev]);
magma_free(dE[dev]);
for( magma_int_t i = 0; i < nbevents; ++i ) {
magma_event_destroy( myevent[dev][i] );
}
for( magma_int_t i = 0; i < nstream; ++i ) {
magma_queue_destroy( streams[dev][i] );
}
}
magma_setdevice( orig_dev );
magmablasSetKernelStream( orig_stream );
return *info;
}
int probe_main(probe_ctx *ctx, void *mutex)
{
LDAP *ldp;
LDAPMessage *ldpres, *entry;
SEXP_t *se_ldap_behaviors = NULL, *se_relative_dn = NULL;
SEXP_t *se_suffix = NULL, *se_attribute = NULL;
SEXP_t *sa_scope, *sv_op;
SEXP_t *item;
SEXP_t *probe_in;
char *relative_dn = NULL;
char *suffix = NULL, *xattribute = NULL;
char *uri_list, *uri, *uri_save, *attr;
int scope;
char base[2048];
char *attrs[3];
bool a_pattern_match = false, rdn_pattern_match = false;
/* runtime */
#if defined(PROBE_LDAP_MUTEX)
assume_r(mutex != NULL, PROBE_EINIT);
#endif
probe_in = probe_ctx_getobject(ctx);
se_ldap_behaviors = probe_obj_getent(probe_in, "behaviors", 1);
if (se_ldap_behaviors != NULL) {
sa_scope = probe_ent_getattrval(se_ldap_behaviors, "scope");
SEXP_free(se_ldap_behaviors);
if (sa_scope == NULL) {
dE("Atrribute `scope' is missing!");
return (PROBE_ENOATTR);
}
if (!SEXP_stringp(sa_scope)) {
dE("Invalid value type of the `scope' attribute.");
SEXP_free(sa_scope);
return (PROBE_EINVAL);
}
if (SEXP_strcmp(sa_scope, "ONE") == 0)
scope = LDAP_SCOPE_ONELEVEL;
else if (SEXP_strcmp(sa_scope, "BASE") == 0)
scope = LDAP_SCOPE_BASE;
else if (SEXP_strcmp(sa_scope, "SUBTREE") == 0)
scope = LDAP_SCOPE_SUBTREE;
else {
dE("Invalid value of the `scope' attribute.");
SEXP_free(sa_scope);
return (PROBE_EINVAL);
}
SEXP_free(sa_scope);
} else
scope = LDAP_SCOPE_BASE;
#define get_string(dst, se_dst, obj, ent_name) \
do { \
SEXP_t *__sval; \
\
__sval = probe_obj_getentval (obj, ent_name, 1); \
\
if (__sval != NULL) { \
(dst) = SEXP_string_cstr (__sval); \
\
if ((dst) == NULL) { \
SEXP_free(__sval); \
return (PROBE_EINVAL); \
} \
\
(se_dst) = __sval; \
} else { \
return (PROBE_ENOATTR); \
} \
} while (0)
get_string(suffix, se_suffix, probe_in, "suffix");
get_string(relative_dn, se_relative_dn, probe_in, "relative_dn");
get_string(xattribute, se_attribute, probe_in, "attribute");
if ((sv_op = probe_ent_getattrval(se_relative_dn, "operation")) != NULL) {
if (SEXP_number_geti_32(sv_op) == OVAL_OPERATION_PATTERN_MATCH)
rdn_pattern_match = true;
SEXP_free(sv_op);
}
if ((sv_op = probe_ent_getattrval(se_attribute, "operation")) != NULL) {
if (SEXP_number_geti_32(sv_op) == OVAL_OPERATION_PATTERN_MATCH)
a_pattern_match = true;
SEXP_free(sv_op);
}
/*
* Construct the attribute array for ldap_search_*
*
* nil -> "1.1"
* .* -> "*"
* "foo" -> "foo"
*/
attrs[0] = "objectClass";
if (xattribute == NULL)
attrs[1] = strdup("1.1"); /* no attibutes */
else if (a_pattern_match)
attrs[1] = strdup("*"); /* collect all, we'll filter them afterwards */
else
attrs[1] = xattribute; /* no pattern match, use the string directly */
attrs[2] = NULL;
/*
* Construct `base'
*/
assume_r(((relative_dn ? strlen(relative_dn) : 0) +
( suffix ? strlen(suffix) : 0) + 2) < (sizeof base/sizeof(char)),
PROBE_ERANGE);
if (relative_dn != NULL) {
strcpy(base, relative_dn);
strcat(base, ",");
strcat(base, suffix);
} else
strcpy(base, suffix);
/*
* Get URIs
*/
if (ldap_get_option(NULL, LDAP_OPT_URI, &uri_list) != LDAP_OPT_SUCCESS) {
item = probe_item_creat("ldap57_item", NULL, NULL);
probe_item_setstatus(item, SYSCHAR_STATUS_ERROR);
probe_item_collect(ctx, item);
dE("ldap_get_option failed");
goto fail0;
}
/*
* Query each URI
*/
for (;;) {
char *entry_dn = NULL;
if ((uri = strtok_r(uri_list, " ,", &uri_save)) == NULL)
break;
ldp = NULL;
if (ldap_initialize(&ldp, uri) != LDAP_SUCCESS)
continue;
if (ldap_search_ext_s(ldp, base, scope, NULL, attrs, 0,
NULL /* serverctrls */,
NULL /* clientctrls */,
NULL /* timeout */,
0, &ldpres) != LDAP_SUCCESS)
{
item = probe_item_creat("ldap57_item", NULL, NULL);
probe_item_setstatus(item, SYSCHAR_STATUS_ERROR);
probe_item_collect(ctx, item);
dE("ldap_search_ext_s failed");
goto fail0;
}
entry = ldap_first_entry(ldp, ldpres);
entry_dn = ldap_get_dn(ldp, entry);
while (entry != NULL) {
BerElement *berelm = NULL;
attr = ldap_first_attribute(ldp, entry, &berelm);
/* XXX: pattern match filter */
while (attr != NULL) {
SEXP_t *se_value = NULL;
ber_tag_t bertag = LBER_DEFAULT;
ber_len_t berlen = 0;
Sockbuf *berbuf = NULL;
SEXP_t se_tmp_mem;
berbuf = ber_sockbuf_alloc();
/*
* Prepare the value (record) entity. Collect only
* primitive (i.e. simple) types.
*/
se_value = probe_ent_creat1("value", NULL, NULL);
probe_ent_setdatatype(se_value, OVAL_DATATYPE_RECORD);
/*
* XXX: does ber_get_next() return LBER_ERROR after the last value?
*/
while ((bertag = ber_get_next(berbuf, &berlen, berelm)) != LBER_ERROR) {
SEXP_t *field = NULL;
oval_datatype_t field_type = OVAL_DATATYPE_UNKNOWN;
switch(bertag & LBER_ENCODING_MASK) {
case LBER_PRIMITIVE:
dI("Found primitive value, bertag = %u", bertag);
break;
case LBER_CONSTRUCTED:
dW("Don't know how to handle LBER_CONSTRUCTED values");
/* FALLTHROUGH */
default:
dW("Skipping attribute value, bertag = %u", bertag);
continue;
}
assume_d(bertag & LBER_PRIMITIVE, NULL);
switch(bertag & LBER_BIG_TAG_MASK) {
case LBER_BOOLEAN:
{ /* LDAPTYPE_BOOLEAN */
ber_int_t val = -1;
if (ber_get_boolean(berelm, &val) == LBER_ERROR) {
dW("ber_get_boolean: LBER_ERROR");
/* XXX: set error status on field */
continue;
}
assume_d(val != -1, NULL);
field = probe_ent_creat1("field", NULL, SEXP_number_newb_r(&se_tmp_mem, (bool)val));
field_type = OVAL_DATATYPE_BOOLEAN;
SEXP_free_r(&se_tmp_mem);
} break;
case LBER_INTEGER:
{ /* LDAPTYPE_INTEGER */
ber_int_t val = -1;
if (ber_get_int(berelm, &val) == LBER_ERROR) {
dW("ber_get_int: LBER_ERROR");
/* XXX: set error status on field */
continue;
}
field = probe_ent_creat1("field", NULL, SEXP_number_newi_r(&se_tmp_mem, (int)val));
field_type = OVAL_DATATYPE_INTEGER;
SEXP_free_r(&se_tmp_mem);
} break;
case LBER_BITSTRING:
/* LDAPTYPE_BIT_STRING */
dW("LBER_BITSTRING: not implemented");
continue;
case LBER_OCTETSTRING:
{ /*
* LDAPTYPE_PRINTABLE_STRING
* LDAPTYPE_NUMERIC_STRING
* LDAPTYPE_DN_STRING
* LDAPTYPE_BINARY (?)
*/
char *val = NULL;
if (ber_get_stringa(berelm, &val) == LBER_ERROR) {
dW("ber_get_stringa: LBER_ERROR");
/* XXX: set error status on field */
continue;
}
assume_d(val != NULL, NULL);
field = probe_ent_creat1("field", NULL, SEXP_string_new_r(&se_tmp_mem, val, strlen(val)));
field_type = OVAL_DATATYPE_STRING;
SEXP_free_r(&se_tmp_mem);
ber_memfree(val);
} break;
case LBER_NULL:
/* XXX: no equivalent LDAPTYPE_? or empty */
dI("LBER_NULL: skipped");
continue;
case LBER_ENUMERATED:
/* XXX: no equivalent LDAPTYPE_? */
dW("Don't know how to handle LBER_ENUMERATED type");
continue;
default:
dW("Unknown attribute value type, bertag = %u", bertag);
continue;
}
if (field != NULL) {
assume_d(field_type != OVAL_DATATYPE_UNKNOWN, NULL);
probe_ent_setdatatype(field, field_type);
probe_ent_attr_add(field, "name", SEXP_string_new_r(&se_tmp_mem, attr, strlen(attr)));
SEXP_list_add(se_value, field);
SEXP_free_r(&se_tmp_mem);
SEXP_free(field);
}
}
ber_sockbuf_free(berbuf);
/*
* Create the item
*/
item = probe_item_create(OVAL_INDEPENDENT_LDAP57, NULL,
"suffix", OVAL_DATATYPE_STRING, suffix,
"relative_dn", OVAL_DATATYPE_STRING, relative_dn, /* XXX: pattern match */
"attribute", OVAL_DATATYPE_STRING, attr,
"object_class", OVAL_DATATYPE_STRING, "",
"ldaptype", OVAL_DATATYPE_STRING, "",
NULL);
SEXP_list_add(item, se_value);
SEXP_free(se_value);
probe_item_collect(ctx, item);
attr = ldap_next_attribute(ldp, entry, berelm);
}
ber_free(berelm, 0);
ldap_memfree(entry_dn);
entry = ldap_next_entry(ldp, entry);
entry_dn = ldap_get_dn(ldp, entry);
}
/*
* Close the LDAP connection and free resources
*/
ldap_unbind_ext_s(ldp, NULL, NULL);
}
ldap_memfree(uri_list);
fail0:
SEXP_free(se_suffix);
SEXP_free(se_relative_dn);
SEXP_free(se_attribute);
free(suffix);
free(relative_dn);
free(attrs[1]); /* attribute */
return (0);
}
