Re: [PATCH][RFC] kthread abstraction

[Dave Hansen <haveblue@us.ibm.com>]

Christoph Hellwig wrote:
> On Fri, Feb 01, 2002 at 09:21:12AM -0800, Dave Hansen wrote:
>>Things like nfsd are always holding the BKL, only 
>>releasing it on schedule(), and exit.  Is there any compelling reason to 
>>hold the BKL during times other than during the daemonize() process?
>
> In general there is no reason.  If the data the thread accesses is not
> protected by anything but BKL it must hold it - else it seems superflous
> to me.
What do you think about the BKL hold during daemonize()?  Can we expand 
the use of the task lock to keep the BKL from being held?

/* these all use task_lock(): */
         exit_mm(current);
         exit_fs(current);
         exit_files(current);

/* Is there more locking needed for this? */
         current->session = 1;
         current->pgrp = 1;
         current->tty = NULL;
         fs = init_task.fs;
         current->fs = fs;

/* This is already safe: */
         atomic_inc(&fs->count);
         current->files = init_task.files;
         atomic_inc(&current->files->count);


-- 
Dave Hansen
haveblue@us.ibm.com