I don't see this in the git master branch? Was it rejected? (Perhaps
just missed?)
Thanks
Ed W
On 11/05/2010 13:42, Timo Teräs wrote:
GCC can emit prologue/epilogue code for the functions in various
different cases:
- frame pointers
- PIC build (to load ebx for indirect calls/jumps)
- forced stack smashing protection
If we used jump in such cases, we'd corrupt the call stack and
crash.
Signed-off-by: Timo Teräs<[email protected]>
---
libm/ldouble_wrappers.c | 12 +++++++-----
1 files changed, 7 insertions(+), 5 deletions(-)
diff --git a/libm/ldouble_wrappers.c b/libm/ldouble_wrappers.c
index 7d5af90..5b424dc 100644
--- a/libm/ldouble_wrappers.c
+++ b/libm/ldouble_wrappers.c
@@ -60,7 +60,9 @@ long long func##l(long double x) \
* The return value is returned in st(0) per ABI in both cases (returning
* a long double or returning a double). So we can simply jump to func.
* Using __GI_func in jump to make optimized intra-library jump.
- * gcc will still generate a useless "ret" after asm. Oh well...
+ *
+ * We do need to use call (instead of tail jump) as gcc can create
+ * stack frame, and push/modify/pop ebx during PIC build.
*/
# define WRAPPER1(func) \
long double func##l(long double x) \
@@ -69,7 +71,7 @@ long double func##l(long double x) \
__asm ( \
" fldt %1\n" \
" fstpl %1\n" \
- " jmp " __stringify(__GI_##func) "\n" \
+ " call " __stringify(__GI_##func) "\n" \
: "=t" (st_top) \
: "m" (x) \
); \
@@ -82,7 +84,7 @@ int func##l(long double x) \
__asm ( \
" fldt %1\n" \
" fstpl %1\n" \
- " jmp " __stringify(__GI_##func) "\n" \
+ " call " __stringify(__GI_##func) "\n" \
: "=a" (ret) \
: "m" (x) \
); \
@@ -95,7 +97,7 @@ long func##l(long double x) \
__asm ( \
" fldt %1\n" \
" fstpl %1\n" \
- " jmp " __stringify(__GI_##func) "\n" \
+ " call " __stringify(__GI_##func) "\n" \
: "=a" (ret) \
: "m" (x) \
); \
@@ -108,7 +110,7 @@ long long func##l(long double x) \
__asm ( \
" fldt %1\n" \
" fstpl %1\n" \
- " jmp " __stringify(__GI_##func) "\n" \
+ " call " __stringify(__GI_##func) "\n" \
: "=A" (ret) \
: "m" (x) \
); \
_______________________________________________
uClibc mailing list
[email protected]
http://lists.busybox.net/mailman/listinfo/uclibc
