From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rusty Russell Subject: Re: dev->destructor Date: Mon, 05 May 2003 15:18:29 +1000 Sender: netdev-bounce@oss.sgi.com Message-ID: <20030505051854.5C31F2C05E@lists.samba.org> References: <20030502.204628.35664814.davem@redhat.com> Cc: kuznet@ms2.inr.ac.ru, shemminger@osdl.org, netdev@oss.sgi.com, acme@conectiva.com.br Return-path: To: "David S. Miller" In-reply-to: Your message of "Fri, 02 May 2003 20:46:28 MST." <20030502.204628.35664814.davem@redhat.com> Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org In message <20030502.204628.35664814.davem@redhat.com> you write: > I think it can work Rusty, in short you create 1 freeze thread > per cpu. You wake up all the freeze threads on non-local cpus, > and they indicate their presence via some bitmask. This code is already in module.c. I'm glad you like it though 8). But we disable local irqs as well: this is what I call a "bogolock" (the read-side of a bogolock is prempt_disable()/preempt_enable(): you could temporarily disable preemption and force the scheduler to run every preempted thread, and remove this). > This means the local master cpu executes the unload sequence. It may > sleep in order to yield to, for example, semaphore holders, it may > also sleep to yield to kswapd and friends for the sake of memory > allocation. I mean... consider all the situations and please try to > find some hole in this. We can make all try_to_*() sleep at this > time too... this in particular needs more thought. Well, it's a big task. Holding interrupts disabled for unbounded time on CPUs needs to be thought about, but I think can be fixed. try_xxx can be called from interrupt context: you really want to get rid of interrupts, too... During previous discussions, I called this "return to primordial soup": back to like during init. Ideally, only userspace context (no interrupts, timers, bottom halves), and life is easy. > To make these freeze threads globally useful, we allow them to > run atomicity commands. The two defined commands are "local_irq_*()" > and "local_bh_*()", two bitmasks control this and the freeze threads > check the bits in their spin loops. Something like this? /* Tell all freeze threads to disable bottom halves. */ void global_bh_disable(void); void global_bh_enable(void); /* Tell all freeze threads to disable interrupts halves. */ void global_irq_disable(void); void global_irq_enable(void); > Do you see? Maybe... it is nearly Nirvana! :-))))) Yes, but I worry it might be an illusion 8) > Our ability to implement this changes the rest of the conversation, > so let us resolve this first. Yes, but it's a big IF. I think it might be easier to make all unregistrations runnable in interrupt context 8( Rusty. -- Anyone who quotes me in their sig is an idiot. -- Rusty Russell.