From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Vrabel Subject: Re: [PATCH 3/3] xen: use maximum reservation to limit dom0 memory Date: Tue, 16 Aug 2011 15:41:03 +0100 Message-ID: <4E4A817F.2040406@citrix.com> References: <1313488838-28809-1-git-send-email-david.vrabel@citrix.com> <1313488838-28809-4-git-send-email-david.vrabel@citrix.com> <20110816135352.GD30261@dumpdata.com> Mime-Version: 1.0 Content-Type: text/plain; charset="ISO-8859-1" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20110816135352.GD30261@dumpdata.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Konrad Rzeszutek Wilk Cc: "xen-devel@lists.xensource.com" List-Id: xen-devel@lists.xenproject.org On 16/08/11 14:53, Konrad Rzeszutek Wilk wrote: > On Tue, Aug 16, 2011 at 11:00:38AM +0100, David Vrabel wrote: >> Use the maximum reservation hypercall to set limit the amount of >> usable dom0 memory. This reduces the size of pages tables etc. if >> dom0 is to use less memory than the maximum available. > > Ok, so it sounds like this patch by itself can fix the "more page tables > than we need" issue. This patch with the Xen patch does, yes. > If so, I would prefer that you stick the tiny piece of code that > calls the xen_get_max_pages() from the setup in this patch. This way > we can backport this particular patch to stable tree without including > the other patchsets you have posted. And it is a nicely contained > one-patch-fixes-the-problem. Does this problem need to be fixed in stable? It has a simple workaround (the 'mem' kernel command line option) and requires an updated Xen. I do think that patches #1 and #2 are useful because they allow 32-bit guests to have more low memory, rather than making all balloon memory high memory. I could rearrange the order. Make #3 first so it can also be applied to 3.0.n and 3.1 and then #1 and #2 could be queued for 3.2. >> Note this requires a patched Xen that sets max_pages when creating dom0. > > Please mention in the description the c/s and the name of the patch. Ok. >> --- a/arch/x86/xen/setup.c >> +++ b/arch/x86/xen/setup.c >> @@ -197,7 +197,12 @@ static unsigned long __init xen_get_max_pages(void) >> unsigned long max_pages = MAX_DOMAIN_PAGES; /* Limited by memory map. */ >> >> if (xen_initial_domain()) { >> - /* FIXME: ask hypervisor for max pages. */ >> + domid_t domid = DOMID_SELF; >> + int ret; >> + >> + ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid); >> + if (ret > 0) >> + max_pages = ret; > Don't you want to clamp it? Say MAX_DOMAIN_PAGES is set to 1GB, and you > set it to 2GB here - that will blow the P2M out. Perhaps It is... > max_pages = min(ret, max_pages); ? > >> } >> >> return min(max_pages, MAX_DOMAIN_PAGES); ... here. David