| Message ID | 4CA47011.4010400@codesourcery.com |
|---|---|
| State | New |
| Headers | show |
On Thu, Sep 30, 2010 at 1:10 PM, Yao Qi <yao@codesourcery.com> wrote: > > We created a struct type for va_list to follow ARM ABI, but TREE_CHAIN > (va_list_type) is *not* set properly. This problem causes segmentation > fault reported in PR45447. > > This patch is to address this problem. Regression test for > arm-none-linux-gnueabi is still running. OK to trunk/4.5/4.4 if no > regression? Err - why do you think assigning to TREE_CHAIN makes sense? I think you want to use TYPE_STUB_DECL instead. > -- > Yao Qi > CodeSourcery > yao@codesourcery.com > (650) 331-3385 x739 >
Richard Guenther wrote: > On Thu, Sep 30, 2010 at 1:10 PM, Yao Qi <yao@codesourcery.com> wrote: >> We created a struct type for va_list to follow ARM ABI, but TREE_CHAIN >> (va_list_type) is *not* set properly. This problem causes segmentation >> fault reported in PR45447. >> >> This patch is to address this problem. Regression test for >> arm-none-linux-gnueabi is still running. OK to trunk/4.5/4.4 if no >> regression? > > Err - why do you think assigning to TREE_CHAIN makes sense? Because 1) TYPE_STUB_DECL and TREE_CHAIN is the same thing, defined in tree.h, #define TYPE_STUB_DECL(NODE) TREE_CHAIN (NODE) 2) TREE_CHAIN is used also for the similar purpose in i386.c: ix86_build_builtin_va_list_abi() TREE_CHAIN (record) = type_decl; > I think you want to use TYPE_STUB_DECL instead. > Are they different? or I may miss something.
On Thu, Sep 30, 2010 at 1:30 PM, Yao Qi <yao@codesourcery.com> wrote: > Richard Guenther wrote: >> On Thu, Sep 30, 2010 at 1:10 PM, Yao Qi <yao@codesourcery.com> wrote: >>> We created a struct type for va_list to follow ARM ABI, but TREE_CHAIN >>> (va_list_type) is *not* set properly. This problem causes segmentation >>> fault reported in PR45447. >>> >>> This patch is to address this problem. Regression test for >>> arm-none-linux-gnueabi is still running. OK to trunk/4.5/4.4 if no >>> regression? >> >> Err - why do you think assigning to TREE_CHAIN makes sense? > > Because 1) TYPE_STUB_DECL and TREE_CHAIN is the same thing, defined in > tree.h, > > #define TYPE_STUB_DECL(NODE) TREE_CHAIN (NODE) > > 2) TREE_CHAIN is used also for the similar purpose in i386.c: > ix86_build_builtin_va_list_abi() > TREE_CHAIN (record) = type_decl; > >> I think you want to use TYPE_STUB_DECL instead. >> > > Are they different? or I may miss something. TREE_CHAIN is non-descriptive, it shouldn't be used if there is a more descriptive macro available. > -- > Yao Qi > CodeSourcery > yao@codesourcery.com > (650) 331-3385 x739 >
gcc/ 2010-09-30 Yao Qi <yao@codesourcery.com> PR target/45447 * config/arm/arm.c (arm_build_builtin_va_list): Assign va_list_name to TREE_CHAIN (va_list_type). gcc/testsuite/ 2010-09-30 Yao Qi <yao@codesourcery.com> PR target/45447 * gcc.target/arm/pr45447.c: New test. diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index 2ff010b..f61006a 100644 --- a/gcc/config/arm/arm.c +++ b/gcc/config/arm/arm.c @@ -1205,6 +1205,7 @@ arm_build_builtin_va_list (void) va_list_type); DECL_ARTIFICIAL (va_list_name) = 1; TYPE_NAME (va_list_type) = va_list_name; + TREE_CHAIN (va_list_type) = va_list_name; /* Create the __ap field. */ ap_field = build_decl (BUILTINS_LOCATION, FIELD_DECL, diff --git a/gcc/testsuite/gcc.target/arm/pr45447.c b/gcc/testsuite/gcc.target/arm/pr45447.c new file mode 100644 index 0000000..cb4a442 --- /dev/null +++ b/gcc/testsuite/gcc.target/arm/pr45447.c @@ -0,0 +1,3 @@ +/* { dg-do compile } */ +/* { dg-options "-g -femit-struct-debug-baseonly" } */ +typedef __builtin_va_list x;