From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751030AbcFRGMk (ORCPT ); Sat, 18 Jun 2016 02:12:40 -0400 Received: from shards.monkeyblade.net ([149.20.54.216]:57459 "EHLO shards.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750879AbcFRGMj (ORCPT ); Sat, 18 Jun 2016 02:12:39 -0400 Date: Fri, 17 Jun 2016 23:12:34 -0700 (PDT) Message-Id: <20160617.231234.1295900275353757289.davem@davemloft.net> To: hikanda@zlab.co.jp Cc: sparclinux@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH] sparc64: Swap registers for fault code and address in mna trap From: David Miller In-Reply-To: References: X-Mailer: Mew version 6.7 on Emacs 24.5 / Mule 6.0 (HANACHIRUSATO) Mime-Version: 1.0 Content-Type: Text/Plain; charset=iso-2022-jp Content-Transfer-Encoding: 7bit X-Greylist: Sender succeeded SMTP AUTH, not delayed by milter-greylist-4.5.12 (shards.monkeyblade.net [149.20.54.216]); Fri, 17 Jun 2016 23:12:39 -0700 (PDT) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org From: 神田　尚 Date: Fri, 17 Jun 2016 01:49:11 +0000 > This bug may occur in the following. > > user_rtt_fill_64bit <= If mna trap occurred, call do_mna > +-> do_mna <= Mistake storing registers for fault code and address > +-> winfix_mna > +-> user_rtt_fill_fixup <= Put fault address into thread_info->flag's TI_FAULT_CODE > +-> do_sparc64_fault() <= If fault address has FAULT_CODE_ITLB and FAULT_CODE_DTLB bits, call BUG() > +-> BUG() We should not be invoking do_sparc64_fault() in this case. Instead, we call either sun4v_do_mna() or mem_address_unaligned(). Neither of which care about the values stored in the thread's fault address and code. If you can really trigger this code path, please post the kernel log backtrace that happens when the BUG() triggers. That way we can figure out what the real problem is. Your patch is also wrong for other reasons, it would break the unaligned trap code paths that don't go via user_rtt_fill_64bit fixups.