* Jamal Hadi Salim [off-list ref] 2006-07-09 10:03

On Sun, 2006-09-07 at 15:33 +0200, Thomas Graf wrote:

quoted

That's not gonna work, dev->queue_lock may be held legimitately
by someone else than an underlying dev_queue_xmit() call.

If there is a legitimate reason then it wont work. I cant think of one
though.

See sch_generic.c, it's documented. A simple grep on queue_lock
would have told you the same.

This is also another approach that would work. If you think its simpler
go ahead and shoot a patch.

It's not simpler, it's correct, while your patch is wrong.

A->*->A is a no-no.
And in some cases it is fine to let the user just fsck themselves
because then they will understand it is a bad idea [1] when shit
happens. OTOH, if there was a KISS way of doing it (as in the ifb case,
why not).

I remind you that you started mentioning this A->*->A case while
talking about tx deadlocks that were supposed to be prevented with
the !from check or something along that lines. I can't really tell
because you explain it differently in every posting.

Yes, of course otherwise i wouldnt bother to comment on any patches.

So maintain the code and fix your bugs.