Hello I'm fallowing a tutorial for pthreading, but something went wrong down the road since can't make the reference to my struct passable by
pthread_create(...,(void* stuctName)) , I'm looking for some advices or fixes since I have no idea where or what I messed up...
code:
#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
#include <unistd.h>
#include <math.h>
struct dataBlock{
struct node *root;
int listSize;
int forIndex;
};
struct node { // std linked list node
int value;
int worker;
struct node *next;
};
int slots = 3; // only 3 threads are allowed to access the list
int availableCheck(){ // check if thread can acces the list
if(slots < 3) return 0;
else return -1;
}
pthread_mutex_t mutp = PTHREAD_MUTEX_INITIALIZER; //condvar mutex
pthread_cond_t condvar = PTHREAD_COND_INITIALIZER; //condvar
void * worker( void *data ){ //WORKER FUNCTION
printf( "* Thread start: ^\n");
struct dataBlock *inData = (struct dataBlock *) data;
struct node *root = data->root;
int listSize = data->listSize;
int forIndex = data ->forIndex;
// printf( " * I am %li _ worker # %li : \n", forIndex, pthread_self() );
pthread_mutex_lock( &mutp );
if(availableCheck() < 0){
printf( " ^^^ List not available yet... \n" );
pthread_cond_wait( &condvar, &mutp );
}
// printf( "* Got data: %lu \n", index );
pthread_cond_signal( &condvar ); //
pthread_mutex_unlock( &mutp );
return NULL;
}
int main( int argc, char *argv[] ){
if ( argc != 3 ){
printf( "Programm must be called with \n NR of elements and NR of workers! \n " );
exit( 1 );
int i;
struct node *root;
struct node *iterator;
//prepare list for task
int listSize = atoi(argv[1]);
int nrWorkers = atoi(argv[2]);
root = malloc(sizeof( struct node) );
root->value = rand() % 100;
root->worker = 0;
iterator = root;
for( i=1; i<listSize; i++ ){
iterator->next = malloc(sizeof(struct node));
iterator = iterator->next;
iterator->value = rand() % 100;
iterator->worker = i % nrWorkers;
printf("node #%d worker: %d value: %d\n", i, iterator->worker,iterator->value);
}
// Create all threads to parse the link list
int ret, *id;
printf("workersInput: %d\n",nrWorkers);
pthread_t w_thread;
pthread_t* w_threads = malloc(nrWorkers * sizeof(w_thread));
struct dataBlock *data = malloc(sizeof(struct dataBlock));
data->root = root;
data->listSize = listSize;
for( i=0; i < nrWorkers; i++ ){ // CREATING THREADS
data->forIndex = i;
ret = pthread_create ( &w_threads[i], NULL, worker, (void *) data );
if( ret ) {
perror("Thread creation fail");
exit(2);
}
}
for ( i = 0; i < nrWorkers; i++){
pthread_join(w_threads[i],NULL);
}
free(root);
free(iterator);
return 0;
}
Compiling(Wall flag)
s.c: In function ‘worker’:
s.c:32:26: warning: dereferencing ‘void *’ pointer [enabled by default]
s.c:32:26: error: request for member ‘root’ in something not a structure or union
s.c:33:22: warning: dereferencing ‘void *’ pointer [enabled by default]
s.c:33:22: error: request for member ‘listSize’ in something not a structure or union
s.c:34:22: warning: dereferencing ‘void *’ pointer [enabled by default]
s.c:34:22: error: request for member ‘forIndex’ in something not a structure or union
s.c:34:6: warning: unused variable ‘forIndex’ [-Wunused-variable]
s.c:33:6: warning: unused variable ‘listSize’ [-Wunused-variable]
s.c:32:15: warning: unused variable ‘root’ [-Wunused-variable]
s.c:31:20: warning: unused variable ‘inData’ [-Wunused-variable]
s.c: In function ‘main’:
s.c:81:12: warning: variable ‘id’ set but not used [-Wunused-but-set-variable]
Use inData->root instead of data->root etc... (or cast explicitly always your data) since data is a void* pointer, that is a pointer to some unspecified data type.
Related
I have try to compile a program which manage userfault-fd to collect some dirty pages on memory. In this program i have used the uffdio_writeprotect structure to survey a memory region where program try to access. I am using Ubuntu 18.04 with linux kernel version 5.14.0. But when i compile a program with the command gcc -o with_userfault -I. with_userfault.c
i still have the same errors such as: with_userfault.c:108:34: error: storage size of ‘wp’ isn’t known
struct uffdio_writeprotect wp;
with_userfault.c:114:21: error: ‘UFFDIO_WRITEPROTECT’ undeclared (first use in this function); did you mean ‘UFFDIO_REGISTER’?
if (ioctl(fd, UFFDIO_WRITEPROTECT, &wp) == -1)
I don't know how to solve this problem. Please I need your help to solve this issue because i don't know what i am suppose to do now. thanks!!!!!
this is a source code:
The header file:
#ifndef __RDTSC_H_DEFINED__
#define __RDTSC_H_DEFINED__
#if defined(__i386__)
static __inline__ unsigned long long rdtsc(void)
{
unsigned long long int x;
__asm__ volatile (".byte 0x0f, 0x31" : "=A" (x));
return x;
}
#elif defined(__x86_64__)
static __inline__ unsigned long long rdtsc(void)
{
unsigned hi, lo;
__asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi));
return ( (unsigned long long)lo)|( ((unsigned long long)hi)<<32 );
}
#elif defined(__powerpc__)
static __inline__ unsigned long long rdtsc(void)
{
unsigned long long int result=0;
unsigned long int upper, lower,tmp;
__asm__ volatile(
"0: \n"
"\tmftbu %0 \n"
"\tmftb %1 \n"
"\tmftbu %2 \n"
"\tcmpw %2,%0 \n"
"\tbne 0b \n"
: "=r"(upper),"=r"(lower),"=r"(tmp)
);
result = upper;
result = result<<32;
result = result|lower;
return(result);
}
#else
#error "No tick counter is available!"
#endif
/* $RCSfile: $ $Author: kazutomo $
* $Revision: 1.6 $ $Date: 2005/04/13 18:49:58 $
*/
#endif
The source file
/*
* Example program about using userfaultfd(2) for garbage collection.
*
* This establishes a couple pages, all of which are filled from
* compressed files on disk when first accessed. For simplicity
* these are
* one file per page. Files are written at the beginning of the
* program.
*
* Later, this program demonstrates the use of write protection to
* get
* a notification on write access, analogous to using
* mprotect(!PROT_WRITE)
* and doing the bookkeeping in a SIGSEGV handler.
*
*/
#include <linux/userfaultfd.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <sys/ioctl.h>
#include <time.h>
#include <math.h>
#include <unistd.h>
#include <strings.h>
#include <string.h>
#include <unistd.h>
#include <asm/unistd.h>
#include <poll.h>
#include <pthread.h>
#include "rdtsc.h"
//#define size 102400
#define FILE_BUF 25
#define PAGE_SIZE sysconf(_SC_PAGE_SIZE)
#define n_pages 102400
//102400
#define iterations 1
pthread_t tracker;
void *tracker_task();
int i;
unsigned long main_tsc_start, main_tsc_end;
// This is doing the work in the uffd handler thread
void *tracker_task(void *data)
{
int fd = *(int *)(data);
for (;;)
{
struct uffd_msg msg;
struct pollfd pollfd;
int pollres, readret;
unsigned long addr, page_begin, whichpage;
unsigned long handler_tsc_start, handler_tsc_end;
pollfd.fd = fd;
pollfd.events = POLLIN;
pollres = poll(&pollfd, 1, -1);
if( pollres == -1 )
perror("poll");
if (pollfd.revents & POLLERR)
{
fprintf(stderr, "POLLERR on userfaultfd\n");
exit(1);
}
readret = read(fd, &msg, sizeof(msg));
if (readret == -1)
perror("read userfaultfd");
if (readret != sizeof(msg))
{
fprintf(stderr, "short read, not expected, exiting\n");
exit(1);
}
/*
* Proper sequence is important here.
*
* For the GC we expect that write-protected pages can only
* be pages already backed by physical pages.
* Regular writes into unprotected pages that come before
* reads need the page be filled.
*
* So we do the WP case first and get it out of the way.
* Then both of the other cases need the page read.
*/
if (msg.arg.pagefault.flags & UFFD_PAGEFAULT_FLAG_WP)
{
// send write unlock
struct uffdio_writeprotect wp;
wp.range.start = msg.arg.pagefault.address;
wp.range.len = PAGE_SIZE;
wp.mode = 0;
//printf("sending !UFFDIO_WRITEPROTECT event to
//userfaultfd\n");
if (ioctl(fd, UFFDIO_WRITEPROTECT, &wp) == -1)
perror("ioctl(UFFDIO_WRITEPROTECT)");
//continue;
}
}
printf("end\n");
return NULL;
}
int main(int argc, char *argv[])
{
unsigned long *region;
int uffd, uffd_flags, expected, t_create;
void *status;
struct uffdio_writeprotect wp;
struct uffdio_api uffdio_api;
struct uffdio_register uffdio_register;
char *clear = "3", *drop_caches_path = "/proc/sys/vm/drop_caches";
FILE *drop_caches_file;
main_tsc_start = rdtsc();
uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
if (uffd == -1)
{
perror("syscall");
exit(2);
}
uffdio_api.api = UFFD_API;
uffdio_api.features = 1;
if (ioctl(uffd, UFFDIO_API, &uffdio_api))
{
fprintf(stderr, "UFFDIO_API\n");
return 1;
}
//printf("Features: 0x%llx\n", uffdio_api.features);
if (uffdio_api.api != UFFD_API)
{
fprintf(stderr, "UFFDIO_API error %Lu\n", uffdio_api.api);
return 1;
}
/* Allocate memory that will be tracked */
region = (unsigned long *) mmap(NULL, PAGE_SIZE * n_pages, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0);
if (!region)
{
perror("mmap");
exit(2);
}
/* Force alignment of contiguous pages */
if (posix_memalign((void **)region, PAGE_SIZE, PAGE_SIZE * (n_pages - 1)))
{
fprintf(stderr, "cannot align by PAGE_SIZE %ld\n", PAGE_SIZE);
exit(1);
}
uffdio_register.range.start = (unsigned long)region;
uffdio_register.range.len = PAGE_SIZE * n_pages;
uffdio_register.mode = UFFDIO_REGISTER_MODE_WP;
//main_tsc_start = rdtsc();
if (ioctl(uffd, UFFDIO_REGISTER, &uffdio_register) == -1)
{
perror("ioctl(UFFDIO_REGISTER)");
exit(1);
}
expected = UFFD_API_RANGE_IOCTLS;
if ((uffdio_register.ioctls & expected) != expected)
{
fprintf(stderr, "ioctl set is incorrect\n");
exit(1);
}
if( (t_create = pthread_create(&tracker, NULL, tracker_task,
&uffd)) != 0 )
{
errno = t_create;
perror("pthread_create");
}
//printf("mainline writing writable pages.\n");
for (unsigned long i = 0; i < n_pages; i++)
{
unsigned long entry = (i * PAGE_SIZE / sizeof(unsigned long)) + ((rand()%10000) % 512);//(i % 512);
region[entry] = i;
}
drop_caches_file = fopen(drop_caches_path, "w");
for (i = 0; i < iterations; i++)//
{
/* Indicate the range of pages to be write-protected */
//sync();
fwrite(clear, sizeof(clear), 1, drop_caches_file);
wp.range.start = (unsigned long long)region;
wp.range.len = PAGE_SIZE * n_pages;
wp.mode = UFFDIO_WRITEPROTECT_MODE_WP;
/* Write-protect pages */
if (ioctl(uffd, UFFDIO_WRITEPROTECT, &wp) == -1)
{
perror("ioctl(UFFDIO_WRITEPROTECT)");
exit(1);
}
/* Now try to "touch" the pages to trigger page faults and handling by the tracker thread */
for (unsigned long i = 0; i < n_pages; i++)
{
unsigned long entry = (i * PAGE_SIZE / sizeof(unsigned long)) + ((rand()%10000) % 512);//(i % 512);
region[entry] = i;
}
}
fclose(drop_caches_file);
if (ioctl(uffd, UFFDIO_UNREGISTER, &uffdio_register.range)) {
fprintf(stderr, "ioctl unregister failure\n");
return 1;
}
main_tsc_end = rdtsc();
printf("Apllication TSC : %lu\n", (unsigned long) (main_tsc_end-main_tsc_start));
return 0;
}
I am learning how to write Linux Device Driver.
I wrote a dummy character device driver, implemented open, release, write, read in fops;
When I read from device , everything was ok;
When I wrote to device by "echo xx > ", the OS was hang.
Even I comment out all codes in write function except pr_alert and return statements, It still hangs;
Could anybody help me figure it out?
#include <linux/init.h>
#include <linux/types.h>
#include <linux/module.h>
#include <linux/mutex.h>
#include <linux/cdev.h>
#include <linux/fs.h>
#include <linux/slab.h>
#include <linux/string.h>
struct hello_dev
{
char *buffer;
int length;
dev_t dev;
struct mutex lock;
struct cdev *pcdev;
};
struct hello_dev *pHelloDev;
int open_device(struct inode *pinode, struct file *filp)
{
filp->private_data = pHelloDev;
return 0;
}
int close_device(struct inode *pinode, struct file *filp)
{
struct hello_dev *pDev = filp->private_data;
if (pDev->buffer != NULL)
kfree(pDev->buffer);
pDev->buffer = NULL;
return 0;
}
ssize_t read_device(struct file *filp, char __user *buffer, size_t len, loff_t *loff)
{
pr_alert("read\n");
struct hello_dev *pDev = filp->private_data;
mutex_lock(&pDev->lock);
if (pDev->buffer == NULL)
{
mutex_unlock(&pDev->lock);
return 0;
}
int length = strlen(pDev->buffer);
// offset max than strlen in buffer, return
if (*loff > (length - 1))
{
mutex_unlock(&pDev->lock);
return 0;
} else {
// available to read
int len2read = length - *loff;
if (len < len2read)
{// buffer length less than available data
len2read = len;
}
int read = copy_to_user(buffer, pDev->buffer + *loff, len2read);
if (read)
{
*loff = *loff + read;
mutex_unlock(&pDev->lock);
return read;
} else {
*loff = *loff + len2read;
mutex_unlock(&pDev->lock);
return len2read;
}
}
}
ssize_t write_device(struct file *filp , const char __user *buffer, size_t len, loff_t* loff) {
pr_alert("write %s\n", buffer);
// struct hello_dev *pDev = filp->private_data;
// mutex_lock(&pDev->lock);
// if(pDev->buffer == NULL) {
// pDev->buffer = kmalloc(100, GFP_KERNEL);
// pDev->length = 100;
// }
// copy_from_user(pDev->buffer, buffer, len);
// *loff = *loff + len;
// mutex_unlock(&pDev->lock);
return len;
}
struct file_operations fops = {
.open = open_device,
.release = close_device,
.read = read_device,
.write = write_device
};
int init_device(void)
{
pr_alert("init device\n");
pHelloDev = kmalloc(sizeof(struct hello_dev), GFP_KERNEL);
pHelloDev->buffer = NULL;
pHelloDev->length = 0;
int ret = alloc_chrdev_region(&pHelloDev->dev, 0, 1, "hello");
if (ret)
goto alloc_error;
if (pHelloDev == NULL)
goto kmalloc_error;
pHelloDev->pcdev = cdev_alloc();
pHelloDev->pcdev->ops = &fops;
mutex_init(&pHelloDev->lock);
ret = cdev_add(pHelloDev->pcdev, pHelloDev->dev, 1);
if (ret)
goto cdev_add_error;
return 0;
alloc_error:
pr_alert("alloc_chrdev_region error, %d\n", ret);
return ret;
kmalloc_error:
pr_alert("alloc struct hello_dev error");
return -ENOMEM;
cdev_add_error:
pr_alert("cdev_add error, %d\n", ret);
return ret;
}
void cleanup_device(void)
{
pr_alert("unload ko\n");
cdev_del(pHelloDev->pcdev);
unregister_chrdev_region(pHelloDev->dev, 1);
}
MODULE_LICENSE("GPL");
module_init(init_device);
module_exit(cleanup_device);
I found why write to device hangs.
//this statements has problem
//maybe there is no \0 in buffer
//so I print it out, it will hang
//I wrote a program to write something to device
//and used strace to trace system call made by this program
//and found it hangs at write(...) system call
//and there was nothing printed out
//so, it must be this statement causing the problem
//when I removed this statement, everything was ok
pr_alert("write %s\n", buffer);
#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
#include <string.h>
struct string_count_struct{
char fname;
char str;
long long count;
};
void* filesearch(void* arg)
{
//get the file name
struct string_count_struct *arg_ptr = (struct string_count_struct*) arg;
int line_num = 1;
int find_result = 0;
char temp[512];
//create a file pointer
FILE *fp;
fp = fopen(arg_ptr -> fname, "r");
//dont forget error handling
if (fp == NULL){
printf("File could not be opened");
return(-1);
}
while (fgets(temp, 512, fp) != NULL) {
if ((strstr(temp, arg_ptr -> str)) != NULL) {
find_result++;
}
line_num++;
}
if(find_result = 0) {
printf("\nSorry, couldn't find a match.\n");
}
arg_ptr -> count = find_result;
//close the file
if (fp){
fclose(fp);
}
pthread_exit(0);
}
int main (int argc, char **argv)
{
if (argc < 3) {
printf("Usage: <file> <string> <arg1> <arg2>...<argN>\n", argv[0]);
exit(-1);
}
int num_args = argc - 2;
struct string_count_struct args[num_args];
//Thread Creation:
pthread_t tids[num_args];
for(int i = 0; i < num_args; i++) {
args[i].fname = atoll(argv[i + 2]);
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_create(&tids[i], &attr, filesearch, &args[i]);
}
//Wait until work is completed
for (int i = 0; i < num_args; i ++){
pthread_join(tids[i], NULL);
printf("blah is blah %lld\n", args[i].count);
}
return 0;
}
Here are my warnings
root#kali:~/Desktop# gcc prog2.c -lbthread
prog2.c: In function ‘filesearch’:
prog2.c:29:13: warning: passing argument 1 of ‘fopen’ makes pointer from integer without a cast [-Wint-conversion]
fp = fopen(arg_ptr -> fname, "r");
^~~~~~~
In file included from prog2.c:1:0:
/usr/include/stdio.h:274:14: note: expected ‘const char * restrict’ but argument is of type ‘char’
extern FILE *fopen (const char *__restrict __filename,
^~~~~
prog2.c:34:8: warning: return makes pointer from integer without a cast [-Wint-conversion]
return(-1);
^
prog2.c:38:21: warning: passing argument 2 of ‘strstr’ makes pointer from integer without a cast [-Wint-conversion]
if ((strstr(temp, arg_ptr -> str)) != NULL) {
^~~~~~~
In file included from prog2.c:4:0:
/usr/include/string.h:337:14: note: expected ‘const char *’ but argument is of type ‘char’
extern char *strstr (const char *__haystack, const char *__needle)
^~~~~~
prog2.c: In function ‘main’:
prog2.c:78:17: error: assignment of read-only member ‘fname’
args[i].fname = atoll(argv[i + 2]);
I am unsure of what I am doing wrong, these errors are preventing my program from correctly reading through the desired files and calculating the # of occurrences of a particular string that the user will select. I have fixed my error but not the warnings.
The program will take a command line argument, create a separate thread for each file to be searched through, search through each file, and then give the results. I plan on using a Mutex for further refinement, but right now I am just trying to solve my I/O issues.
Just addressing some of the warnings, I'm not at all sure if this will make the code work:
In line 29, fopen expects a filename (char *) but fname is just a char in string_count_struct. Make it a char*. In the main function you convert one of the arguments from ASCII to long long and assign it to fname, which shall later be used as a file name for fopen(). That's probably not what you want to do.
In line 34, you return -1 which is not a pointer. You declared the function to be returning a void pointer. Make it return(0) (or return(NULL)).
Same as in line 29 happens in line 38: str in the struct string_count_struct is just a char, but strstr expects a char*. Make it a char*, too.
Your "Usage" is missing a "%s" format string to actually print the argument argv[0].
I'm trying to create a struct that will hold both the host and device arrays in one place and should reside on the host. I later intend expand it to be a element of a linked list. The basic struct looks like this:
typedef struct Data{
double *h;
double *d;
} Data;
Where *h points to an array of doubles on the host and *d points to an array of doubles on the device.
There are various answers on SO about copying whole structs to the the device (CUDA cudaMemcpy Struct of Arrays) but none of them quite do what I need. I have the following code but keep getting illegal memory access errors.
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include "cuda.h"
/*
* CUDA Error stuff
*/
static void HandleError( cudaError_t err,
const char *file,
int line ) {
if (err != cudaSuccess) {
printf( "%s in %s at line %d\n", cudaGetErrorString( err ),
file, line );
exit( EXIT_FAILURE );
}
}
#define HANDLE_ERROR( err ) (HandleError( err, __FILE__, __LINE__ ))
#define HANDLE_NULL( a ) {if (a == NULL) { \
printf( "Host memory failed in %s at line %d\n", \
__FILE__, __LINE__ ); \
exit( EXIT_FAILURE );}}
//malloc error code
int errMsg(const char *message, int errorCode)
{
printf("%s\n", message);
return errorCode;
}
typedef struct Data{
double *h;
double *d;
} Data;
__global__ void kernel(Data *d)
{
int tid = blockIdx.x * blockDim.x + threadIdx.x;
if(tid<100){
d->d[tid] = 2;
}
}
int main()
{
Data *d;
d = (Data*)malloc(sizeof(Data));
d->h = (double*)malloc(sizeof(double)*100);
HANDLE_ERROR( cudaMalloc((void**) &(d->d), 100*sizeof(double)) );
for(int i=0; i<100; i++){
d->h[i] = i;
}
HANDLE_ERROR( cudaMemcpy(d->d, d->h, 100*sizeof(double), cudaMemcpyHostToDevice) );
printf("%f\n", d->h[1]);
kernel<<<1, 102>>>(d);
printf("done\n");
{
cudaError_t cudaerr = cudaDeviceSynchronize();
if (cudaerr != cudaSuccess)
printf("kernel launch failed with error \"%s\"->\n",
cudaGetErrorString(cudaerr));
}
HANDLE_ERROR( cudaMemcpy(d->h, d->d, 100*sizeof(double), cudaMemcpyDeviceToHost) );
printf("%f\n", d->h[99]);
return 0;
}
The output I get is:
1.000000
done
kernel launch failed with error "an illegal memory access was encountered"->
an illegal memory access was encountered in linkedListGPU.cu at line 77
I suspect I have just messed up my pointers a bit. The error handling code is from the Wiley introduction to CUDA book, if there code is not allowed on here I'll remove it.
Thanks.
The problem is d itself is a pointer to a host-allocated struct (where the d and h pointers are contained. When you pass the d struct pointer to the kernel like so:
kernel<<<1, 102>>>(d);
^
this is a pointer to memory on the host
and then attempt to dereference that pointer in device code here:
d->...;
^
This operator dereferences the pointer to the left of it
you get an illegal memory access.
There are at least 2 obvious ways to fix this:
Pass the struct by value instead of by pointer.
Here is an example:
$ cat t1311.cu
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include "cuda.h"
/*
* CUDA Error stuff
*/
static void HandleError( cudaError_t err,
const char *file,
int line ) {
if (err != cudaSuccess) {
printf( "%s in %s at line %d\n", cudaGetErrorString( err ),
file, line );
exit( EXIT_FAILURE );
}
}
#define HANDLE_ERROR( err ) (HandleError( err, __FILE__, __LINE__ ))
#define HANDLE_NULL( a ) {if (a == NULL) { \
printf( "Host memory failed in %s at line %d\n", \
__FILE__, __LINE__ ); \
exit( EXIT_FAILURE );}}
//malloc error code
int errMsg(const char *message, int errorCode)
{
printf("%s\n", message);
return errorCode;
}
typedef struct Data{
double *h;
double *d;
} Data;
__global__ void kernel(Data d)
{
int tid = blockIdx.x * blockDim.x + threadIdx.x;
if(tid<100){
d.d[tid] = 2;
}
}
int main()
{
Data d;
d.h = (double*)malloc(sizeof(double)*100);
HANDLE_ERROR( cudaMalloc((void**) &(d.d), 100*sizeof(double)) );
for(int i=0; i<100; i++){
d.h[i] = i;
}
HANDLE_ERROR( cudaMemcpy(d.d, d.h, 100*sizeof(double), cudaMemcpyHostToDevice) );
printf("%f\n", d.h[1]);
kernel<<<1, 102>>>(d);
printf("done\n");
{
cudaError_t cudaerr = cudaDeviceSynchronize();
if (cudaerr != cudaSuccess)
printf("kernel launch failed with error \"%s\"->\n",
cudaGetErrorString(cudaerr));
}
HANDLE_ERROR( cudaMemcpy(d.h, d.d, 100*sizeof(double), cudaMemcpyDeviceToHost) );
printf("%f\n", d.h[99]);
return 0;
}
$ nvcc -arch=sm_35 -o t1311 t1311.cu
$ cuda-memcheck ./t1311
========= CUDA-MEMCHECK
1.000000
done
2.000000
========= ERROR SUMMARY: 0 errors
$
Make a device copy of the struct that the d host pointer points to:
Here is an example:
$ cat t1311.cu
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include "cuda.h"
/*
* CUDA Error stuff
*/
static void HandleError( cudaError_t err,
const char *file,
int line ) {
if (err != cudaSuccess) {
printf( "%s in %s at line %d\n", cudaGetErrorString( err ),
file, line );
exit( EXIT_FAILURE );
}
}
#define HANDLE_ERROR( err ) (HandleError( err, __FILE__, __LINE__ ))
#define HANDLE_NULL( a ) {if (a == NULL) { \
printf( "Host memory failed in %s at line %d\n", \
__FILE__, __LINE__ ); \
exit( EXIT_FAILURE );}}
//malloc error code
int errMsg(const char *message, int errorCode)
{
printf("%s\n", message);
return errorCode;
}
typedef struct Data{
double *h;
double *d;
} Data;
__global__ void kernel(Data *d)
{
int tid = blockIdx.x * blockDim.x + threadIdx.x;
if(tid<100){
d->d[tid] = 2;
}
}
int main()
{
Data *d, *dev_d;
d = (Data*)malloc(sizeof(Data));
HANDLE_ERROR(cudaMalloc(&dev_d, sizeof(Data)));
d->h = (double*)malloc(sizeof(double)*100);
HANDLE_ERROR( cudaMalloc((void**) &(d->d), 100*sizeof(double)) );
for(int i=0; i<100; i++){
d->h[i] = i;
}
HANDLE_ERROR( cudaMemcpy(d->d, d->h, 100*sizeof(double), cudaMemcpyHostToDevice) );
HANDLE_ERROR(cudaMemcpy(dev_d, d, sizeof(Data), cudaMemcpyHostToDevice));
printf("%f\n", d->h[1]);
kernel<<<1, 102>>>(dev_d);
printf("done\n");
{
cudaError_t cudaerr = cudaDeviceSynchronize();
if (cudaerr != cudaSuccess)
printf("kernel launch failed with error \"%s\"->\n",
cudaGetErrorString(cudaerr));
}
HANDLE_ERROR( cudaMemcpy(d->h, d->d, 100*sizeof(double), cudaMemcpyDeviceToHost) );
printf("%f\n", d->h[99]);
return 0;
}
$ nvcc -arch=sm_35 -o t1311 t1311.cu
$ cuda-memcheck ./t1311
========= CUDA-MEMCHECK
1.000000
done
2.000000
========= ERROR SUMMARY: 0 errors
$
As an aside, you can follow the method outlined here to carry your debug process a bit farther.
I have a bit of a problem in using IPC (inter-process communication) program below.
Please let me explain:
I want to pass Linux commands such as "ls" or "wc file.txt"
from a parent to a child to execute using the message queue, and
then have the child returning the command outputs back to
the parent process using shared memory method.
But this is what I got: The parent process always got the output 1 step behind;
in the following fashion:
Step1) ls file.txt
(Nothing showed up.)
Step2) wc file.txt
(Output of earlier command "ls file.txt" showed up here instead.)
Step 3) cat file.txt
(Output of earlier command "wc file.txt" showed up instead.)
Any help is appreciated.
To compile: gcc -o program ./program.c
To run: -./program -v
Code:
#define BUFSZ 512
#define ERRBUFSZ 512
#define TIMEOUT_TIMEDIO 20
#define SHM_SIZE 5120
#include <stdio.h>
#include <errno.h>
#include <unistd.h>
#include <stdlib.h>
#include <string.h>
#include <signal.h>
#include <setjmp.h>
#include <sys/wait.h>
#include <sys/types.h>
#include <sys/ipc.h>
#include <sys/shm.h>
static sigjmp_buf jmpbuf;
int timed_io(char* buf, int len, FILE* rfp, int sec);
static void sigalrm_handler(int signo);
void do_cmd(char *buf, int len, int linenum, char *errbuf);
int parse_cmd(char *buf, char **vbuf, char *errbuf);
int process_cmd_ipc(char *argv, int linenum, char *errbuf);
struct my_msgbuf {
long mtype;
char mtext[256];
};
static void sigalrm_handler(int signo)
{
siglongjmp(jmpbuf, 1);
}
int timed_io(char* buf, int len, FILE* rfp, int sec)
{
struct sigaction nsigaction[1];
struct sigaction osigaction[1];
int prev_alrm;
int st = 0;
if(sigsetjmp(jmpbuf, 1) == 0)
{
nsigaction->sa_handler = sigalrm_handler;
sigemptyset(&nsigaction->sa_mask);
nsigaction->sa_flags = SA_RESTART;
prev_alrm = alarm(0);
sigaction(SIGALRM, nsigaction, osigaction);
alarm(sec);
if (fgets(buf, len, rfp) == NULL)
st = -1; // EOF
buf[strlen(buf) - 1] = 0;
}
else { st = -2; } // Time-out
alarm(0); // Reset old alarm and handler
sigaction(SIGALRM, osigaction, 0);
return st;
}
int process_cmd_ipc(char *argv, int linenum, char* errbuf)
{
struct my_msgbuf buf;
int msqid, msqid_parent, st, shmid, str_len;
key_t key, key_shm;
char* shared_buf;
FILE *fd;
// create key for shared memory segment
if ((key_shm = ftok("shm_key.txt", 'R')) == -1) {
perror("ftok");
exit(1);
}
// Connect to shared memory segment
if ((shmid = shmget(key_shm, SHM_SIZE, 0644 | IPC_CREAT)) == -1)
{
perror("shmget");
exit(1);
}
// Attach to shared memory segment
shared_buf = shmat(shmid, (void *) 0, 0);
if (shared_buf == (char *) (-1)) {
perror("shmat");
exit(1);
}
// End of shared memory section` //
// Begin: message queue section
pid_t cpid=fork();
if (cpid<0) {
fprintf(stderr,"ERR: \"fork\" error! (Line=%d)\n", linenum);
exit (-1);
} else if (cpid==0) // child process
{ // Begin: message queue
if ((key = ftok("mysh.c", 'B')) == -1) {
perror("ftok");
exit(1);
}
if ((msqid = msgget(key, 0644)) == -1) {
perror("msgget from child");
exit(1);
}
memset(buf.mtext, 0, sizeof(buf.mtext)); // Clear buffer
if(msgrcv(msqid, (struct msgbuf*) &buf, sizeof(buf), 0,0) == -1)
{
perror("msgrcv");
exit(1);
}
// End: message queue
// begin: shared memory segment
memset(shared_buf, 0, SHM_SIZE); // zeroize shared_buf
fd = popen(buf.mtext, "r");
str_len = 0;
while(fgets(shared_buf + str_len, SHM_SIZE, fd) != NULL)
{ str_len = strlen(shared_buf); }
pclose(fd);
// end: shared memory segment
}
else { // parent
// Begin - message queue
if ((key = ftok("mysh.c", 'B')) == -1) {
perror("ftok");
exit(1);
}
if ((msqid_parent = msgget(key, 0644 | IPC_CREAT)) == -1) {
perror("msgget from parent");
exit(1);
}
buf.mtype = 1;
strncpy(buf.mtext, argv, strlen(argv));
if(msgsnd(msqid_parent, (struct my_msgbuf*) &buf, strlen(buf.mtext), 0) == -1)
perror("msgsnd");
// End - message queue
// Begin - shared memory
// usleep(10000);
printf("%s", shared_buf);
// End - shared memory
} // if-else fork
}
int parse_cmd(char *buf, char **vbuf, char *errbuf)
{
int i=0;
char *delim=" ,\t\n";
char *tok;
tok=strtok(buf,delim);
while (tok) {
vbuf[i]=(char *)malloc(BUFSZ*sizeof(char));
strcpy(vbuf[i],tok);
tok=strtok(NULL,delim);
i++;
}
vbuf[i]=0;
return i;
}
void do_cmd(char *buf, int len, int linenum, char *errbuf) {
int i=0; int numargs;
char *vbuf[128];
char* copy = (char *) malloc(strlen(buf) + 1);
int maxargs=sizeof(vbuf)/sizeof(char *);
strcpy(copy, buf);
numargs = parse_cmd(copy,vbuf,errbuf);
process_cmd_ipc(buf,linenum, errbuf);
for (i=0;i<numargs; i++) { free(vbuf[i]); }
free(copy);
copy = NULL;
return;
}
int main(int argc, char **argv)
{
int i; int st; int linenum=0;
char *buf=(char *)malloc(BUFSZ*sizeof(char));
char *errbuf=(char *)malloc(ERRBUFSZ*sizeof(char));
char *mysh = "";
FILE *rfp=stdin;
if (isatty(fileno(rfp))) {
mysh = "mysh (Ctrl-C to exit)>";
fprintf(stderr,"%s",mysh);
}
while(1)
{
st = timed_io(buf, BUFSZ, stdin, TIMEOUT_TIMEDIO);
if (st != 0)
{
fprintf(stderr, "ERR: No input %s (Status=%d)\n", errbuf, st);
return -1;
}
else
{
linenum++;
if (*buf)
{ do_cmd(buf, BUFSZ, linenum,errbuf); }
if (mysh)
fprintf(stderr,"%s",mysh);
}
}
}