From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758567Ab2IFPsa (ORCPT ); Thu, 6 Sep 2012 11:48:30 -0400 Received: from mga09.intel.com ([134.134.136.24]:50536 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758524Ab2IFPsY (ORCPT ); Thu, 6 Sep 2012 11:48:24 -0400 Subject: Re: [PATCH] x86, efi: 1:1 pagetable mapping for virtual EFI calls From: Matt Fleming To: Jan Beulich Cc: Vasco Dias , Matthew Garrett , linux-efi@vger.kernel.org, linux-kernel@vger.kernel.org, cJ-ko@zougloub.eu, "H. Peter Anvin" In-Reply-To: <5048D09A020000780009965B@nat28.tlf.novell.com> References: <1346937309-1581-1-git-send-email-matt@console-pimps.org> <5048D09A020000780009965B@nat28.tlf.novell.com> Content-Type: text/plain; charset="UTF-8" Organization: Intel Corporation (UK) Ltd. - Registered No. 1134945 - Pipers Way, Swindon SN3 1RJ Date: Thu, 06 Sep 2012 16:47:45 +0100 Message-ID: <1346946465.4244.107.camel@mfleming-mobl1.ger.corp.intel.com> Mime-Version: 1.0 X-Mailer: Evolution 2.32.3 (2.32.3-1.fc14) Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 2012-09-06 at 15:34 +0100, Jan Beulich wrote: > >>> On 06.09.12 at 15:15, Matt Fleming wrote: > > From: Matt Fleming > > > > Some firmware still needs a 1:1 (virt->phys) mapping even after we've > > called SetVirtualAddressMap(). So install the mapping alongside our > > existing kernel mapping whenever we make EFI calls in virtual mode. > > > > This bug was discovered on ASUS machines where the firmware > > implementation of GetTime() accesses the RTC device via physical > > addresses, even though that's bogus per the UEFI spec since we've > > informed the firmware via SetVirtualAddressMap() that the boottime > > memory map is no longer valid. > > > @@ -741,6 +742,114 @@ static void __init runtime_code_page_mkexec(void) > > } > > } > > > > +#ifdef CONFIG_X86_64 > > +pgd_t *efi_one_to_one_pgd; > > +int efi_one_to_one_index = 0; > > I wasn't able to spot where this variable ever gets set to a > non-zero value. Oops. Yeah, you're right. It seems I lost that code at some point in development. I'll re-add in a v2 of this patch. > > @@ -58,6 +62,47 @@ static void __init early_code_mapping_set_exec(int > > executable) > > } > > } > > > > +pgd_t *efi_call_virt_prelog(void) > > +{ > > + pgd_t *save; > > + int i; > > + > > + save = kmalloc(sizeof(pgd_t) * (efi_one_to_one_index + 1), GFP_KERNEL); > > GFP_ATOMIC perhaps, given that many runtime calls happen > inside spin locks? Yeah, good idea. > > + if (!save) { > > + pr_alert("Unable to save pgd entries\n"); > > + return NULL; > > + } > > + > > + for (i = 0; i <= efi_one_to_one_index; i++) { > > + pgd_t *pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); > > I'd suggest moving this out of the loop (also below), slightly > adjusting the rest of the loop body. Sure. > > + > > + pgd += i; > > + save[i] = *pgd; > > + set_pgd(pgd, efi_one_to_one_pgd[i]); > > + } > > Did you, as an alternative, consider switching to a different > CR3 instead of copying back and forth? I did consider it, but I couldn't convince myself whether or not the EFI pagetable would need to be manually kept in sync with any other pagetables. But now I look at the code a bit harder, it seems that should be taken care of automatically. In fact, the tboot code seems to do something similar. I'll try that approach. > > + > > + __flush_tlb_all(); > > Is it certain you will _never_ hit a global mapping (in which case I > believe the above would be insufficient)? Are you saying that this should be a flush_tlb_all() if we have global mappings? Or that if we are guaranteed to never have global mappings we can opitmise this by simply doing a __flush_tlb()? Thanks Jan!