98f
Subj : Re: Linux NPTL anomally
To   : comp.programming.threads
From : dmarkh
Date : Wed Sep 28 2005 04:27 am

> Sounds like either a bug in the kernel module or a bug in the way you
>are using it. It is documented to allow it to be called by a different
>thread than cleans it up?

The kernel module is never called by any of the threads.  Functions
within the kernel module are called only by the main part of the
application and by other external applications.

The  kernel  ioctl routine that registers the pid blurp:

       case (LCRSMEM_ID):
        {
                // identify a shared memory partition
                if (copy_from_user( &shm_info, (int *)arg,
                        sizeof(lcrs_shm_info_t)) != 0)
                   return -EFAULT;

                // search for an unused entry to fill
                for (i=0;i<MAX_LCRSMEM_DEV;i++)
                {
                        if ((lcrsmmap[i].key == shm_info.key) &&
(lcrsmmap[i].pid == current->pid))
                        {
                                // duplicate key, stale?
                                break;
                        }
                        else if (lcrsmmap[i].key == 0)
                        {
                                // use first free entry
                                lcrsmmap[i].pid = current->pid;
                                lcrsmmap[i].key = shm_info.key;
                                lcrsmmap[i].vm_start = (unsigned
int)shm_info.address;
                                lcrsmmap[i].type = shm_info.type;
                                lcrsmmap[i].size = shm_info.size;
                                break;
                        }
                }


The cleanup kernel routine blurp:

int lcrsmem_release(struct inode *inode, struct file *filp)
{
    int i;
    for (i=0;i<MAX_LCRSMEM_DEV;i++)
    {
        if (current->pid == lcrsmmap[i].pid)
        {
             .
             .      // code that never gets executed because the wrong
pid is found
             .
         }
    }

I realize there are other ways to do what I am doing but this should
work. The current->pid in the release routine should be the pid of the
main part of the app as IT is the one that calls it.
None of the threads ever calls any functions of this kernel module and
at the release point all the threads have long since exited anyway.

Mark

.

0