From: Hollis Blanchard Date: Tue, 15 Apr 2008 18:19:10 +0000 Subject: Re: [kvm-ppc-devel] [PATCH 2 of 2] Add PowerPC KVM guest wait Message-Id: <200804151319.10339.hollisb@us.ibm.com> List-Id: References: <21179d0ab8a62ecc24d1.1206740047@thinkpadL> In-Reply-To: <21179d0ab8a62ecc24d1.1206740047@thinkpadL> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: kvm-ppc@vger.kernel.org On Tuesday 15 April 2008 11:14:52 Jerone Young wrote: > Actually there appears to be a real problem with preempt notify in 44x. > I had no gotten a chance to get back with you about it. But I did some > investigation into it last week. We are following the same code paths > (common code) as x86 for preempt initalization. But I ran some tests > using preempt_disable() & preempt_enable() around some places where it > would make since (places where we disable interrupts), but just using > these functions whould cause the kernel to dump sig #11. preempt_disable() and preempt_enable() are completely different, and I've called those in the past and had no problems. > The issue we have using function kvm_vcpu_block() that it is identical > to the code below, BUT it calls vcpu_put which then calls > preempt_notify_unregister() if it is called it will also sig #11. If preempt_notify_unregister() dereferences a bad pointer, that shouldn't be too difficult to track down. > I'm not sure if what is going on honestly. Based on what I found it > should "just work" as we are initializing everything like x86 (we are > calling preempt_notify_init() in the same place). But for 44x any > preempt notfication calls blow up. So it appears calling anything > preempt notify related just blows up. > > This is a much bigger issue. I'm not sure that we honest want to be > stuck on this for long periods of time just to have a function call in a > place where we honestly do not absolutely need to have it at this time. > Plus I'm no expert with these scheduling frameworks. But givin what have > read around the net what we have now "should" work. It just doesn't. > > Something can come back to later. But for now we should just roll with > the working code. No, I don't want to commit the hack. Sorry if I didn't make that clear before. Accessing a bad pointer doesn't seem like a "much bigger issue" to me, so if it's more complicated than that please elaborate. -- Hollis Blanchard IBM Linux Technology Center ------------------------------------------------------------------------- This SF.net email is sponsored by the 2008 JavaOne(SM) Conference Don't miss this year's exciting event. There's still time to save $100. Use priority code J8TL2D2. http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone _______________________________________________ kvm-ppc-devel mailing list kvm-ppc-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/kvm-ppc-devel