[patch] Update SJLJ buffer after dynamic stack allocation

Jeff Law law@redhat.com
Thu May 14 15:15:00 GMT 2015


On 05/13/2015 04:45 AM, Eric Botcazou wrote:
> Hi,
>
> the SJLJ EH schemes manipulate a buffer that contains a few pointers and, in
> particular, the stack pointer.  Therefore, when dynamic stack allocation is
> also used, you need to update its value in the buffer, otherwise Bad Things
> will happen if you resume execution where the stack is shifted.
>
> GNAT uses a manual SJLJ EH scheme that is piggybacked on the __builtin_setjmp
> __builtin_longjmp machinery and has this block in gcc-interface/decl.c:
>
> 	/* If we are defining an object with variable size or an object with
> 	   fixed size that will be dynamically allocated, and we are using the
> 	   setjmp/longjmp exception mechanism, update the setjmp buffer.  */
> 	if (definition
> 	    && Exception_Mechanism == Setjmp_Longjmp
> 	    && get_block_jmpbuf_decl ()
> 	    && DECL_SIZE_UNIT (gnu_decl)
> 	    && (TREE_CODE (DECL_SIZE_UNIT (gnu_decl)) != INTEGER_CST
> 		|| (flag_stack_check == GENERIC_STACK_CHECK
> 		    && compare_tree_int (DECL_SIZE_UNIT (gnu_decl),
> 					 STACK_CHECK_MAX_VAR_SIZE) > 0)))
> 	  add_stmt_with_node (build_call_n_expr
> 			      (update_setjmp_buf_decl, 1,
> 			       build_unary_op (ADDR_EXPR, NULL_TREE,
> 					       get_block_jmpbuf_decl ())),
> 			      gnat_entity);
>
> It is necessary if you want to pass the ACATS testsuite.
>
> But the regular SJLJ scheme (--enable-sjlj-exceptions) doesn't do this update
> and, therefore, doesn't play nice with dynamic stack allocation.  It's a long-
> standing issue and you can find messages about it in the archives.  On the
> other hand, this update has always been done for non-local gotos (they also
> use a buffer, the non-local goto save area).
>
> The attached patch implements this update for the regular SJLJ scheme and
> yields a clean ACATS testsuite.  Tested on x86_64-suse-linux with --enable-
> sjlj-exceptions, OK for the mainline?
>
>
> 2015-05-13  Eric Botcazou  <ebotcazou@adacore.com>
>              Tristan Gingold  <gingold@adacore.com>
>
> 	* insn-notes.def (UPDATE_SJLJ_CONTEXT): New note.
> 	* builtins.c (expand_builtin_update_setjmp_buf): Make global.
> 	(expand_stack_restore): Call record_new_stack_level.
> 	(expand_stack_save): Do not call do_pending_stack_adjust.
> 	* builtins.h (expand_builtin_update_setjmp_buf): Declare.
> 	* calls.c (expand_call): Call record_new_stack_level for alloca.
> 	* except.c (sjlj_mark_call_sites): Expand builtin_update_setjmp_buf
> 	wherever a NOTE_INSN_UPDATE_SJLJ_CONTEXT note is present.
> 	(update_sjlj_context): New global function.
> 	* except.h (update_sjlj_context): Declare.
> 	* explow.c (record_new_stack_level): New global function.
> 	(allocate_dynamic_stack_space): Call record_new_stack_level.
> 	* explow.h (record_new_stack_level): Declare.
> 	* final.c (final_scan_insn): Deal with NOTE_INSN_UPDATE_SJLJ_CONTEXT.
> 	* cfgrtl.c (duplicate_insn_chain): Likewise.
OK.  Thanks for tackling this.

jeff



More information about the Gcc-patches mailing list