2007-11-15 16:13:48

by Alexey Dobriyan

Subject: [PATCH] proc: use BUG_ON() in de_put()

It's much more visible that some printk. I still has an unexplained oops
in proc, so let's leave it for a while.

Signed-off-by: Alexey Dobriyan <[email protected]>
---

fs/proc/inode.c | 7 +------
1 file changed, 1 insertion(+), 6 deletions(-)

--- a/fs/proc/inode.c
+++ b/fs/proc/inode.c
@@ -37,12 +37,7 @@ void de_put(struct proc_dir_entry *de)
{
if (de) {
lock_kernel();
- if (!atomic_read(&de->count)) {
- printk("de_put: entry %s already free!\n", de->name);
- unlock_kernel();
- return;
- }
-
+ BUG_ON(atomic_read(&de->count) == 0);
if (atomic_dec_and_test(&de->count)) {
if (de->deleted) {
printk("de_put: deferred delete of %s\n",

2007-11-16 21:47:48

by Andrew Morton

Subject: Re: [PATCH] proc: use BUG_ON() in de_put()

On Thu, 15 Nov 2007 19:12:49 +0300
Alexey Dobriyan <[email protected]> wrote:

> It's much more visible that some printk. I still has an unexplained oops
> in proc, so let's leave it for a while.
>
> Signed-off-by: Alexey Dobriyan <[email protected]>
> ---
>
> fs/proc/inode.c | 7 +------
> 1 file changed, 1 insertion(+), 6 deletions(-)
>
> --- a/fs/proc/inode.c
> +++ b/fs/proc/inode.c
> @@ -37,12 +37,7 @@ void de_put(struct proc_dir_entry *de)
> {
> if (de) {
> lock_kernel();
> - if (!atomic_read(&de->count)) {
> - printk("de_put: entry %s already free!\n", de->name);
> - unlock_kernel();
> - return;
> - }
> -
> + BUG_ON(atomic_read(&de->count) == 0);
> if (atomic_dec_and_test(&de->count)) {
> if (de->deleted) {
> printk("de_put: deferred delete of %s\n",

I don't see that an error in here _requires_ that we nuke the machine.
Surely we can emit a warning and then recover in some fashion?

2007-11-22 08:49:44

by Alexey Dobriyan

Subject: Re: [PATCH] proc: use BUG_ON() in de_put()

On Fri, Nov 16, 2007 at 01:46:42PM -0800, Andrew Morton wrote:
> On Thu, 15 Nov 2007 19:12:49 +0300
> Alexey Dobriyan <[email protected]> wrote:
>
> > It's much more visible that some printk. I still has an unexplained oops
> > in proc, so let's leave it for a while.

> > --- a/fs/proc/inode.c
> > +++ b/fs/proc/inode.c
> > @@ -37,12 +37,7 @@ void de_put(struct proc_dir_entry *de)
> > {
> > if (de) {
> > lock_kernel();
> > - if (!atomic_read(&de->count)) {
> > - printk("de_put: entry %s already free!\n", de->name);
> > - unlock_kernel();
> > - return;
> > - }
> > -
> > + BUG_ON(atomic_read(&de->count) == 0);
> > if (atomic_dec_and_test(&de->count)) {
> > if (de->deleted) {
> > printk("de_put: deferred delete of %s\n",
>
> I don't see that an error in here _requires_ that we nuke the machine.
> Surely we can emit a warning and then recover in some fashion?

Hmm... atomic_dec_and_test() in -mm already has diagnostics? when doing
0 => -1 transition, google says this check triggered only once. I think
we should just drop it.

? i386 only :^)