From mboxrd@z Thu Jan 1 00:00:00 1970 From: Matt Fleming Subject: Re: [PATCH] efi: runtime-wrappers: run UEFI Runtime Services with interrupts enabled Date: Thu, 11 Feb 2016 16:04:05 +0000 Message-ID: <20160211160405.GK4134@codeblueprint.co.uk> References: <20160203105851.GA22159@gmail.com> <1454594327-5444-1-git-send-email-ard.biesheuvel@linaro.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org To: Andy Lutomirski Cc: Ard Biesheuvel , "H. Peter Anvin" , "linux-efi@vger.kernel.org" , "linux-kernel@vger.kernel.org" , Thomas Gleixner , Sai Praneeth Prakhya , Ingo Molnar , Linus Torvalds , Peter Zijlstra List-Id: linux-efi@vger.kernel.org On Mon, 08 Feb, at 11:37:58AM, Andy Lutomirski wrote: > On Feb 4, 2016 5:58 AM, "Ard Biesheuvel" wrote: > > > > OK, since Sai has confirmed that Windows leaves interrupts enabled when > > calling the EFI variable store related runtime services, we should be able > > to do the same for Linux, or at least be slightly more confident that we > > won't have to back out this change later. > > Could this use a mutex instead of a spinlock? > > Can someone with a mixed mode setup read a variable in a loop and make > sure it doesn't crash and burn? It should work fine, but explicit > testing would be nice. (It's interesting mainly because doing a mixed > mode call with interrupts on can result in a non-IST CPL0 to CPL0 > exception delivery, which won't result in a stack switch. This could > easily trigger a stack overflow, logic bug, microcode bug, or > as-yet-unknown CPU "feature". I don't have physical hardware for testing mixed mode anymore (that was returned to Intel when I left) but testing with Qemu didn't turn up any problems when running with interrupts enabled. > Hmm. We should also audit the mixed mode entry code to make sure that > the high bits of RSP are explicitly clear before switching into compat > mode. If I had to make a guess about how CPUs behave, I'd guess > pessimistically: Intel CPUs clear the high bits of RSP when switching > into long mode due to interrupt delivery, and AMD CPUs leave them set > just to mess with us. Interesting thought. I'm not aware of anyone testing mixed mode with AMD CPUs, so that would be a good data point. > Also, a WARN_ON(in_interrupt()) somewhere might be a good sanity check. lockdep should catch this kind of stuff pretty quickly since we grab the spinlock in every code path.