From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751879AbbDBTkn (ORCPT ); Thu, 2 Apr 2015 15:40:43 -0400 Received: from mail-la0-f50.google.com ([209.85.215.50]:33499 "EHLO mail-la0-f50.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750781AbbDBTkh (ORCPT ); Thu, 2 Apr 2015 15:40:37 -0400 MIME-Version: 1.0 In-Reply-To: References: <1427985378-4287-1-git-send-email-dvlasenk@redhat.com> <1427985378-4287-3-git-send-email-dvlasenk@redhat.com> <551D5E23.2050002@redhat.com> From: Andy Lutomirski Date: Thu, 2 Apr 2015 12:40:15 -0700 Message-ID: Subject: Re: [PATCH 3/9] x86/asm/entry/64: do not SAVE_EXTRA_REGS in stub_sigreturn To: Brian Gerst Cc: Denys Vlasenko , Ingo Molnar , Linus Torvalds , Steven Rostedt , Borislav Petkov , "H. Peter Anvin" , Oleg Nesterov , Frederic Weisbecker , Alexei Starovoitov , Will Drewry , Kees Cook , "the arch/x86 maintainers" , Linux Kernel Mailing List Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Apr 2, 2015 at 12:10 PM, Brian Gerst wrote: > On Thu, Apr 2, 2015 at 11:20 AM, Denys Vlasenko wrote: >> On 04/02/2015 05:01 PM, Brian Gerst wrote: >>> On Thu, Apr 2, 2015 at 10:36 AM, Denys Vlasenko wrote: >>>> stub_sigreturn ignores old values of pt_regs->REG for all general-purpose >>>> registers, it sets them to values saved on userspace >>>> signal stack. >>>> >>>> Which is hardly surprising - it would be a bug if it would use pt_regs->REG. >>>> sigreturn must restore all registers. >>>> >>>> Therefore, SAVE_EXTRA_REGS in it ought to be redundant. >>>> >>>> It is a leftover from the time SAVE_EXTRA_REGS wasn't only saving registers, >>>> but it also was extending stack to "full" pt_regs. >>>> >>>> Delete this SAVE_EXTRA_REGS. >>>> >>>> Run-tested. >>>> >>>> Signed-off-by: Denys Vlasenko >>>> CC: Linus Torvalds >>>> CC: Steven Rostedt >>>> CC: Ingo Molnar >>>> CC: Borislav Petkov >>>> CC: "H. Peter Anvin" >>>> CC: Andy Lutomirski >>>> CC: Oleg Nesterov >>>> CC: Frederic Weisbecker >>>> CC: Alexei Starovoitov >>>> CC: Will Drewry >>>> CC: Kees Cook >>>> CC: x86@kernel.org >>>> CC: linux-kernel@vger.kernel.org >>>> --- >>>> arch/x86/kernel/entry_64.S | 9 +++++++-- >>>> 1 file changed, 7 insertions(+), 2 deletions(-) >>>> >>>> diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S >>>> index ec51598..1cf245d 100644 >>>> --- a/arch/x86/kernel/entry_64.S >>>> +++ b/arch/x86/kernel/entry_64.S >>>> @@ -447,7 +447,12 @@ ENTRY(stub_rt_sigreturn) >>>> CFI_STARTPROC >>>> addq $8, %rsp >>>> DEFAULT_FRAME 0 >>>> - SAVE_EXTRA_REGS >>>> + /* >>>> + * Despite RESTORE_EXTRA_REGS in return_from_stub, >>>> + * no need to SAVE_EXTRA_REGS here: >>>> + * sys_rt_sigreturn overwrites all general purpose pt_regs->REGs >>>> + * on stack, for RESTORE_{EXTRA,C}_REGS to pick them up. >>>> + */ >>>> call sys_rt_sigreturn >>>> jmp return_from_stub >>>> CFI_ENDPROC >>>> @@ -458,7 +463,7 @@ ENTRY(stub_x32_rt_sigreturn) >>>> CFI_STARTPROC >>>> addq $8, %rsp >>>> DEFAULT_FRAME 0 >>>> - SAVE_EXTRA_REGS >>>> + /* No need to SAVE_EXTRA_REGS */ >>>> call sys32_x32_rt_sigreturn >>>> jmp return_from_stub >>>> CFI_ENDPROC >>> >>> I had the same idea, but determined sigreturn can fault and return an >>> error code without modifying all the registers. This would leak junk >>> from the stack. > > To clarify, I remembered looking at sigreturn possibly faulting from > the 32-bit perspective, where the 6th arg is read from the user stack > and a fault there would return -EFAULT, for any syscall. > >> This still can be made to work by not RESTORE'ing EXTRA_REGS either, >> if there is a way to detect the failure: >> >> call sys_rt_sigreturn >> - jmp return_from_stub >> + testl ??????????? >> + jz return_from_stub >> + ret >> CFI_ENDPROC >> >> But this is not a normal syscall, off-hand I don't see an easy way >> to do the test. sys_rt_sigreturn() on failure runs this code: >> >> ... >> segfault: >> force_sig(SIGSEGV, current); >> return 0; >> } >> >> Help? > > I don't think you can test the return value, because in the success > case it can be any value (the restored RAX value). Given that getting this right is complicated and sigreturn is already really slow, I'm not convinced that this particular optimization is worthwhile. --Andy