applied your patch without others, so trace and panic there.
Screenshot attached. Also tested kernels backward and found out that
this trace bug first triggered on 6.6-rc1.
Let me know if you need more assistance with this.
--Ilkka
On Tue, May 28, 2024 at 9:47 PM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
>
> On Tue, 28 May 2024 07:51:30 +0300
> Ilkka Naulapää <digirigawa@xxxxxxxxx> wrote:
>
> > yeah, the cache_from_obj tracing bug (without panic) has been
> > displayed quite some time now - maybe even since 6.7.x or so. I could
> > try checking a few versions back for this and try bisecting it if I
> > can find when this started.
> >
>
> OK, so I don't think the commit your last bisect hit is the cause of
> the bug. It added a delay (via RCU) and is causing the real bug to blow
> up more.
>
> Can you add this patch to v6.9.2 and hopefully it crashes in a better
> location that we can find where the mixup happened.
>
> You may need to add the other commit (too if this doesn't trigger.
>
> Thanks,
>
> -- Steve
>
> diff --git a/fs/tracefs/inode.c b/fs/tracefs/inode.c
> index 417c840e6403..7af3f696696d 100644
> --- a/fs/tracefs/inode.c
> +++ b/fs/tracefs/inode.c
> @@ -50,6 +50,7 @@ static struct inode *tracefs_alloc_inode(struct super_block *sb)
> list_add_rcu(&ti->list, &tracefs_inodes);
> spin_unlock_irqrestore(&tracefs_inode_lock, flags);
>
> + ti->magic = 20240823;
> return &ti->vfs_inode;
> }
>
> @@ -66,6 +67,7 @@ static void tracefs_free_inode(struct inode *inode)
> struct tracefs_inode *ti = get_tracefs(inode);
> unsigned long flags;
>
> + BUG_ON(ti->magic != 20240823);
> spin_lock_irqsave(&tracefs_inode_lock, flags);
> list_del_rcu(&ti->list);
> spin_unlock_irqrestore(&tracefs_inode_lock, flags);
> @@ -271,16 +273,6 @@ static const struct inode_operations tracefs_file_inode_operations = {
> .setattr = tracefs_setattr,
> };
>
> -struct inode *tracefs_get_inode(struct super_block *sb)
> -{
> - struct inode *inode = new_inode(sb);
> - if (inode) {
> - inode->i_ino = get_next_ino();
> - simple_inode_init_ts(inode);
> - }
> - return inode;
> -}
> -
> struct tracefs_mount_opts {
> kuid_t uid;
> kgid_t gid;
> @@ -448,6 +440,17 @@ static const struct super_operations tracefs_super_operations = {
> .show_options = tracefs_show_options,
> };
>
> +struct inode *tracefs_get_inode(struct super_block *sb)
> +{
> + struct inode *inode = new_inode(sb);
> + BUG_ON(sb->s_op != &tracefs_super_operations);
> + if (inode) {
> + inode->i_ino = get_next_ino();
> + simple_inode_init_ts(inode);
> + }
> + return inode;
> +}
> +
> /*
> * It would be cleaner if eventfs had its own dentry ops.
> *
> diff --git a/fs/tracefs/internal.h b/fs/tracefs/internal.h
> index f704d8348357..dda7d2708e30 100644
> --- a/fs/tracefs/internal.h
> +++ b/fs/tracefs/internal.h
> @@ -16,6 +16,7 @@ struct tracefs_inode {
> };
> /* The below gets initialized with memset_after(ti, 0, vfs_inode) */
> struct list_head list;
> + unsigned long magic;
> unsigned long flags;
> void *private;
> };
Attachment:
6.9.2-steve-patch.jpeg
Description: JPEG image
