From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 23794 invoked by alias); 28 Oct 2013 22:32:57 -0000 Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Archive: List-Post: List-Help: Sender: gcc-bugs-owner@gcc.gnu.org Received: (qmail 23755 invoked by uid 48); 28 Oct 2013 22:32:52 -0000 From: "marcovanotti15+gcc at gmail dot com" To: gcc-bugs@gcc.gnu.org Subject: [Bug target/58744] Illegal Memory Access on 3-byte packed struct ARCH: x86_64 Date: Mon, 28 Oct 2013 22:32:00 -0000 X-Bugzilla-Reason: CC X-Bugzilla-Type: changed X-Bugzilla-Watch-Reason: None X-Bugzilla-Product: gcc X-Bugzilla-Component: target X-Bugzilla-Version: 4.8.1 X-Bugzilla-Keywords: wrong-code X-Bugzilla-Severity: minor X-Bugzilla-Who: marcovanotti15+gcc at gmail dot com X-Bugzilla-Status: NEW X-Bugzilla-Priority: P3 X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org X-Bugzilla-Target-Milestone: --- X-Bugzilla-Flags: X-Bugzilla-Changed-Fields: Message-ID: In-Reply-To: References: Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/ Auto-Submitted: auto-generated MIME-Version: 1.0 X-SW-Source: 2013-10/txt/msg02026.txt.bz2 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58744 --- Comment #3 from Marco Vanotti --- (In reply to Richard Biener from comment #2) > Confirmed. On i?86 we properly do a 16byte and a 8byte access (but we copy > to stack anyway). > Yes, if the value is passed on the stack, it gets copied right. (For example, if it is the seventh parameter of a function, it will be passed on the stack and will be copied right). The thing is that in the x86_64 calling convention it has to be passed on registers while the are available (rdi, rsi, rdx, rcx, r8 and r9). Reading the source code, precisely the gcc/calls.c file: http://gcc.gnu.org/viewcvs/gcc/trunk/gcc/calls.c?revision=203967&view=markup --- The params that are passed on the stack are handled in line 3027, which says: /* Now store (and compute if necessary) all non-register parms. These come before register parms, since they can require block-moves, which could clobber the registers used for register parms. Parms which have partial registers are not stored here, but we do preallocate space here if they want that. */ Assuming that the registers may not require block-moves. It uses the function "store_one_arg" to store the arg in the stack (it doesn't work with a non-partial register). --- After a while, the function "load_register_parameters" (line 1860) is called, in this function, it falls in the case: move_block_to_reg (REGNO (reg), mem, nregs, args[i].mode); where nregs == 1. So a whole register is copied. --- I don't know how this issue should be fixed, should we copy the register into pseudos before the "load_register_parameters" ? Or should we change the move_block_to_reg function to make the right size of move instructions, for x86_64 we don't need "backup-registers", but maybe this bug is also in another arch. size 3: mov di, [rax] sal rdi, 16 mov dil, [rax] --- size 5: mov edi, [rax] sal rdi, 8 mov dil, [rax] --- size 6: mov edi, [rax] sal rdi, 16 mov di, [rax] --- size 7: mov edi, [rax] ;move 4 sal rdi, 24 mov di, [rax] ;move 3 sal rdi, 16 mov dil, [rax] ----------------------------- I would gladly submit a patch if I can get some advice on how this should be fixed :)