From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from gate.crashing.org (gate.crashing.org [63.228.1.57]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client did not present a certificate) by ozlabs.org (Postfix) with ESMTP id 7680BDDE44 for ; Thu, 18 Oct 2007 12:57:24 +1000 (EST) Subject: Re: [PATCH] synchronize_irq needs a barrier From: Benjamin Herrenschmidt To: Linus Torvalds In-Reply-To: <1192675254.12879.29.camel@pasglop> References: <1192670742.12879.5.camel@pasglop> <1192675254.12879.29.camel@pasglop> Content-Type: text/plain Date: Thu, 18 Oct 2007 12:57:15 +1000 Message-Id: <1192676235.12879.40.camel@pasglop> Mime-Version: 1.0 Cc: linuxppc-dev list , akpm , Linux Kernel list Reply-To: benh@kernel.crashing.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , > > In general, I tend to think that for this function to make any sense > (that is, to synchronize anything at all), it needs a barrier or you are > just making a decision based on a totally random value of desc->status > since it can have been re-ordered, speculatively loaded, pre-fetched, > whatever'ed... :-). Take a real life example: drivers/message/fusion/mptbase.c /* Disable interrupts! */ CHIPREG_WRITE32(&ioc->chip->IntMask, 0xFFFFFFFF); ioc->active = 0; synchronize_irq(pdev->irq); And we aren't in a spinlock here. That's just a random example grepped.... I think I see a few more. Then, some drivers like tg3 actually do an smp_mb() before calling synchronize_irq(). But then, some don't. I think trying to have all drivers be correct here is asking for trouble, we'd rather have synchronize_irq() be uber-safe. It's not like it was used in hot path anyway. Ben.