From mboxrd@z Thu Jan 1 00:00:00 1970 From: "H. Peter Anvin" Subject: Re: [GIT PULL] EFI changes for v3.16 Date: Mon, 19 May 2014 15:47:31 -0700 Message-ID: <537A8A03.8060604@zytor.com> References: <20140503130447.GW26088@console-pimps.org> <20140519105129.GF4798@console-pimps.org> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20140519105129.GF4798-HNK1S37rvNbeXh+fF434Mdi2O/JbrIOy@public.gmane.org> Sender: linux-efi-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org To: Matt Fleming , Ingo Molnar Cc: linux-efi-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Borislav Petkov , "Luck, Tony" List-Id: linux-efi@vger.kernel.org On 05/19/2014 03:51 AM, Matt Fleming wrote: > On Sat, 03 May, at 02:04:47PM, Matt Fleming wrote: >> Folks, please queue the following change for v3.16 from Borislav that >> uses the more strict kernel_fpu_{begin,end}() instead of the __* >> verisons that won't catch buggy use in interrupt context. >> >> The following changes since commit e33655a386ed3b26ad36fb97a47ebb1c2ca1e928: >> >> efivars: Add compatibility code for compat tasks (2014-04-17 13:53:53 +0100) >> >> are available in the git repository at: >> >> git://git.kernel.org/pub/scm/linux/kernel/git/mfleming/efi.git tags/efi-next >> >> for you to fetch changes up to baa916f39b50ad91661534652110df40396acda0: >> >> x86/efi: Check for unsafe dealing with FPU state in irq ctxt (2014-05-03 06:39:25 +0100) >> >> ---------------------------------------------------------------- >> * Use the more strict FPU handling functions before invoking EFI >> services to catch misuse in irq context - Borislav Petkov > > Ping? > > efi_call can happen in an irq context (pstore) and there we really need > to make sure we're not scribbling over FPU state while we've interrupted > a thread or kernel mode with a live FPU state. Therefore, use the > kernel_fpu_begin/end() variants which do that check. How on earth does this solve anything? The only thing we add here is a WARN_ON_ONCE()... but the above text already tells us we have a problem. It seems, rather, that we need to figure out how to deal with a pstore in this case. There are a few possibilities: 1. We could keep an XSAVE buffer area around for this particular use. I am *assuming* we don't let more than one CPU into EFI, because I cannot for my life imagine that this is safe in typical CPUs. 2. Drop the pstore on the floor if !irq_fpu_usable(). 3. Allow the pstore, then die (on the assumption that we're dead anyway.) Comments? -hpa