Re: [PATCH 2/3] work_on_cpu: Use our own workqueue.

[Andrew Morton]

On Fri, 30 Jan 2009 14:49:35 +0100 Ingo Molnar <mingo@xxxxxxx> wrote:

> > So we still don't have any non-buggy proposal.
> 
> Current upstream code is not pretty (due to the extra workqueue) but not 
> buggy either. You'd be right to point out that it is easy to insert a bug 
> into it and thus it's not pleasant (more of a workaround than a real fix) 
> but if it's outright buggy then please talk up.

OK, so we're not aware of anything in there which will trigger the bug
yet.  Although allocate_threshold_blocks() takes about half the locks in
the kernel - it can run an ext3 commit, it does netlink tx, synchronous
process exec, etc.

Why not give up and kill the whole work_on_cpu() thing?  afaict the
only caller which cannot be immediately switched to use an IPI is
mce_amd_64's allocate_threshold_blocks(), and it looks like that can be
trivially fixed by moving the entire function except for one rdmsr() up
into the caller.  

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/