From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753513AbaIHKjl (ORCPT ); Mon, 8 Sep 2014 06:39:41 -0400 Received: from cam-admin0.cambridge.arm.com ([217.140.96.50]:51532 "EHLO cam-admin0.cambridge.arm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752950AbaIHKjk (ORCPT ); Mon, 8 Sep 2014 06:39:40 -0400 Date: Mon, 8 Sep 2014 11:39:59 +0100 From: Will Deacon To: Kees Cook Cc: "linux-kernel@vger.kernel.org" , Rabin Vincent , Laura Abbott , Rob Herring , Leif Lindholm , "msalter@redhat.com" , Liu hua , Nikolay Borisov , Nicolas Pitre , Doug Anderson , Jason Wessel , Catalin Marinas , Russell King - ARM Linux , "linux-arm-kernel@lists.infradead.org" Subject: Re: [PATCH v5 3/8] arm: fixmap: implement __set_fixmap() Message-ID: <20140908103959.GD26030@arm.com> References: <1409781429-27593-1-git-send-email-keescook@chromium.org> <1409781429-27593-4-git-send-email-keescook@chromium.org> <20140904170349.GL7156@arm.com> <20140904172748.GO7156@arm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, Sep 05, 2014 at 08:41:08PM +0100, Kees Cook wrote: > On Thu, Sep 4, 2014 at 10:27 AM, Will Deacon wrote: > > On Thu, Sep 04, 2014 at 06:23:42PM +0100, Kees Cook wrote: > >> On Thu, Sep 4, 2014 at 10:03 AM, Will Deacon wrote: > >> > Hi Kees, > >> > > >> > On Wed, Sep 03, 2014 at 10:57:04PM +0100, Kees Cook wrote: > >> >> This is used from set_fixmap() and clear_fixmap() via asm-generic/fixmap.h. > >> >> Also makes sure that the fixmap allocation fits into the expected range. > >> >> > >> >> Based on patch by Rabin Vincent. > >> > > >> > [...] > >> > > >> >> +void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) > >> >> +{ > >> >> + unsigned long vaddr = __fix_to_virt(idx); > >> >> + pte_t *pte = pte_offset_kernel(pmd_off_k(vaddr), vaddr); > >> >> + > >> >> + /* Make sure fixmap region does not exceed available allocation. */ > >> >> + BUILD_BUG_ON(FIXADDR_START + (__end_of_fixed_addresses * PAGE_SIZE) > > >> >> + FIXADDR_END); > >> >> + BUG_ON(idx >= __end_of_fixed_addresses); > >> >> + > >> >> + if (pgprot_val(prot)) > >> >> + set_pte_at(NULL, vaddr, pte, > >> >> + pfn_pte(phys >> PAGE_SHIFT, prot)); > >> >> + else > >> >> + pte_clear(NULL, vaddr, pte); > >> >> + > >> >> + /* > >> >> + * Given the potential a15 tlbi errata, we can only do tlb flushes > >> >> + * with interrupts disabled. Callers must have taken care of this. > >> >> + */ > >> >> + WARN_ON_ONCE(!irqs_disabled()); > >> >> + flush_tlb_kernel_range(vaddr, vaddr + PAGE_SIZE); > >> > > >> > Aha, this explains why we were confusing each other! The issue is that > >> > interrupts must be *enabled*, so this code does the exact opposite of > >> > what we need. > >> > > >> > I think this got lost in a sea of double negatives during the last round > >> > of review. > >> > >> Ah! If this is the case, perhaps we can get away with > >> local_flush_tlb_kernel_range() then? > > > > That's a bit tricky, since you need to ensure that preemption is disabled > > until the mapping is put back like it was. > > Even with CONFIG_ARM_ERRATA_798181 enabled, I don't see a problem here > using flush_tlb_kernel_range(). When doing the ftrace work, this was > trivial to trip over, so I think we're in a good state here. > > I've tested this on real hardware now too, and nothing falls over. > I'll get rid of the comment and the WARN_ON_ONCE, but AFAICT, this is > safe. But was that hardware actually affected by the erratum? There is a runtime check which will avoid the cross-call if not. > Do you have a suggestion about what needs fixing? The easiest thing to do would be ensuring that __set_fixmap is always called with interrupts enabled. If you can guarantee that, then you need to rework the locking so that interrupts aren't disabled. I admit that I've lost a fair amount of the context here, but that's the fundamental issue. Will From mboxrd@z Thu Jan 1 00:00:00 1970 From: will.deacon@arm.com (Will Deacon) Date: Mon, 8 Sep 2014 11:39:59 +0100 Subject: [PATCH v5 3/8] arm: fixmap: implement __set_fixmap() In-Reply-To: References: <1409781429-27593-1-git-send-email-keescook@chromium.org> <1409781429-27593-4-git-send-email-keescook@chromium.org> <20140904170349.GL7156@arm.com> <20140904172748.GO7156@arm.com> Message-ID: <20140908103959.GD26030@arm.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Fri, Sep 05, 2014 at 08:41:08PM +0100, Kees Cook wrote: > On Thu, Sep 4, 2014 at 10:27 AM, Will Deacon wrote: > > On Thu, Sep 04, 2014 at 06:23:42PM +0100, Kees Cook wrote: > >> On Thu, Sep 4, 2014 at 10:03 AM, Will Deacon wrote: > >> > Hi Kees, > >> > > >> > On Wed, Sep 03, 2014 at 10:57:04PM +0100, Kees Cook wrote: > >> >> This is used from set_fixmap() and clear_fixmap() via asm-generic/fixmap.h. > >> >> Also makes sure that the fixmap allocation fits into the expected range. > >> >> > >> >> Based on patch by Rabin Vincent. > >> > > >> > [...] > >> > > >> >> +void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot) > >> >> +{ > >> >> + unsigned long vaddr = __fix_to_virt(idx); > >> >> + pte_t *pte = pte_offset_kernel(pmd_off_k(vaddr), vaddr); > >> >> + > >> >> + /* Make sure fixmap region does not exceed available allocation. */ > >> >> + BUILD_BUG_ON(FIXADDR_START + (__end_of_fixed_addresses * PAGE_SIZE) > > >> >> + FIXADDR_END); > >> >> + BUG_ON(idx >= __end_of_fixed_addresses); > >> >> + > >> >> + if (pgprot_val(prot)) > >> >> + set_pte_at(NULL, vaddr, pte, > >> >> + pfn_pte(phys >> PAGE_SHIFT, prot)); > >> >> + else > >> >> + pte_clear(NULL, vaddr, pte); > >> >> + > >> >> + /* > >> >> + * Given the potential a15 tlbi errata, we can only do tlb flushes > >> >> + * with interrupts disabled. Callers must have taken care of this. > >> >> + */ > >> >> + WARN_ON_ONCE(!irqs_disabled()); > >> >> + flush_tlb_kernel_range(vaddr, vaddr + PAGE_SIZE); > >> > > >> > Aha, this explains why we were confusing each other! The issue is that > >> > interrupts must be *enabled*, so this code does the exact opposite of > >> > what we need. > >> > > >> > I think this got lost in a sea of double negatives during the last round > >> > of review. > >> > >> Ah! If this is the case, perhaps we can get away with > >> local_flush_tlb_kernel_range() then? > > > > That's a bit tricky, since you need to ensure that preemption is disabled > > until the mapping is put back like it was. > > Even with CONFIG_ARM_ERRATA_798181 enabled, I don't see a problem here > using flush_tlb_kernel_range(). When doing the ftrace work, this was > trivial to trip over, so I think we're in a good state here. > > I've tested this on real hardware now too, and nothing falls over. > I'll get rid of the comment and the WARN_ON_ONCE, but AFAICT, this is > safe. But was that hardware actually affected by the erratum? There is a runtime check which will avoid the cross-call if not. > Do you have a suggestion about what needs fixing? The easiest thing to do would be ensuring that __set_fixmap is always called with interrupts enabled. If you can guarantee that, then you need to rework the locking so that interrupts aren't disabled. I admit that I've lost a fair amount of the context here, but that's the fundamental issue. Will