From: Keir Fraser Subject: Re: [PATCH] x86: fix domain cleanup Date: Tue, 28 Oct 2008 11:15:28 +0000 Message-ID: References: Mime-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Jan Beulich Cc: xen-devel@lists.xensource.com List-Id: xen-devel@lists.xenproject.org On 28/10/08 10:49, "Keir Fraser" wrote: >> Will do it that way for submission. In testing it with that code inlined in >> __put_page_type(), I can confirm that this closes the memory leak, but >> it (obviously) doesn't address the crash when encountering a PGT_partial >> page hanging off of a page table being cleaned up by that explicit call >> to free_page_type() getting executed as a side effect of >> DOMAIN_DESTRUCT_AVOID_RECURSION. The question of course really >> is whether it's worthwhile trying to fix that, or rather to do away with it >> altogether by utilizing the 'real' preemption. > > I don't actually understand the AVOID_RECURSION logic in > relinquish_memory(). I'd be delighted to get rid of it. Actually I'm not sure how PGT_partial reference counting works either. Do such pages hold a general reference count? It appears not, so I don't see why for example we couldn't have the put_page() in get_page_and_type_from_pagenr() cause free_domheap_pages() and then BUG_ON() the fact that the type count is non-zero (because it is partially validated). Shouldn't free_domheap_pages() do cleanup for partially validated pages? Is it possible or necessary to tell the difference between a page that is partially alloc_page_type()d versus one which is partially free_page_type()d? (e.g., are the reference counts left in different states in each case, and does that matter?) -- Keir