From mboxrd@z Thu Jan 1 00:00:00 1970 From: Catalin Marinas Subject: Re: [PATCH v3 19/23] arm64: mte: Add PTRACE_{PEEK,POKE}MTETAGS support Date: Tue, 19 May 2020 17:10:58 +0100 Message-ID: <20200519161057.GE20313@gaia> References: <20200421142603.3894-1-catalin.marinas@arm.com> <20200421142603.3894-20-catalin.marinas@arm.com> <20200513104849.GC2719@gaia> <3d2621ac-9d08-53ea-6c22-c62532911377@linaro.org> <20200513141147.GD2719@gaia> <20200518164723.GA5031@arm.com> <55fe4d37-23ae-a6b7-8db1-884aaf4a9b9c@linaro.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from foss.arm.com ([217.140.110.172]:35664 "EHLO foss.arm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728953AbgESQLD (ORCPT ); Tue, 19 May 2020 12:11:03 -0400 Content-Disposition: inline In-Reply-To: <55fe4d37-23ae-a6b7-8db1-884aaf4a9b9c@linaro.org> Sender: linux-arch-owner@vger.kernel.org List-ID: To: Luis Machado Cc: Dave Martin , linux-arch@vger.kernel.org, Richard Earnshaw , Will Deacon , Omair Javaid , Szabolcs Nagy , Kevin Brodsky , linux-mm@kvack.org, Andrey Konovalov , Vincenzo Frascino , Peter Collingbourne , Alan Hayward , linux-arm-kernel@lists.infradead.org On Mon, May 18, 2020 at 02:12:24PM -0300, Luis Machado wrote: > On 5/18/20 1:47 PM, Dave Martin wrote: > > Wrinkle: just because MTE is "off", pages might still be mapped with > > PROT_MTE and have arbitrary tags set on them, and the debugger perhaps > > needs a way to know that. Currently grubbing around in /proc is the > > only way to discover that. Dunno whether it matters. > > That is the sort of thing that may confused the debugger. > > If MTE is "off" (and thus the debugger doesn't need to validate tags), then > the pages mapped with PROT_MTE that show up in /proc//smaps should be > ignored? There is no such thing as global MTE "off". If the HWCAP is present, a user program can map an address with PROT_MTE and access tags. Maybe it uses it for extra storage, you never know, doesn't have to be heap allocation related. > I'm looking for a precise way to tell if MTE is being used or not for a > particular process/thread. This, in turn, will tell debuggers when to look > for PROT_MTE mappings in /proc//smaps and when to validate tagged > addresses. > > So far my assumption was that MTE will always be "on" when HWCAP2_MTE is > present. So having HWCAP2_MTE means we have the NT_ARM_MTE regset and that > PROT_MTE pages have to be checked. Yes. I haven't figured out what to put in the regset yet, most likely the prctl value as it has other software-only controls like the tagged address ABI. -- Catalin