Bug 57427
| Summary: | ARM redboot_linux_exec should be changed to avoid miscompilation | ||
|---|---|---|---|
| Product: | [Retired] eCos | Reporter: | Jonathan Larmour <jlarmour> |
| Component: | HAL | Assignee: | Jesper Skov <jskov> |
| Status: | CLOSED WONTFIX | QA Contact: | Nick Garnett <nickg> |
| Severity: | medium | Docs Contact: | |
| Priority: | medium | ||
| Version: | CVS | CC: | gthomas, hmt, jlarmour, jskov |
| Target Milestone: | --- | ||
| Target Release: | --- | ||
| Hardware: | other | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | Bug Fix | |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2003-06-20 16:08:17 UTC | Type: | --- |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
Unfortunately, this doesn't work for the other cases where we use &&,
such as in the GDB locking macros. This is a grep for &&[^ \t] where I've pruned
out false hits:
./devs/flash/arm/mahwah/current/src/flash_query.c:87: __mem_fault_handler =
&&no_device;
./hal/arm/arch/current/include/arm_stub.h:129:
cyg_hal_gdb_place_break((target_register_t)((unsigned
long)&&cyg_hal_gdb_break_place + 1));\
./hal/arm/arch/current/include/arm_stub.h:137:
cyg_hal_gdb_place_break((target_register_t)&&cyg_hal_gdb_break_place ); \
./hal/arm/arch/current/include/arm_stub.h:146: cyg_hal_gdb_remove_break(
(target_register_t)&&cyg_hal_gdb_break_place ); \
./hal/arm/arm9/excalibur/current/src/redboot_linux_exec.c:307: ip = (unsigned
long *)&&lab1;
./hal/arm/arm9/excalibur/current/src/redboot_linux_exec.c:312: for (i = 0; i
< (unsigned long)&&end1-(unsigned long)&&lab1; i++) {
./hal/arm/arm9/aaed2000/current/src/redboot_linux_exec.c:305: ip = (unsigned
long *)&&lab1;
./hal/arm/arm9/aaed2000/current/src/redboot_linux_exec.c:310: for (i = 0; i
< (unsigned long)&&end1-(unsigned long)&&lab1; i++) {
./hal/arm/ebsa285/current/src/redboot_cmds.c:132: ip = (unsigned long *)&&lab1;
./hal/arm/ebsa285/current/src/redboot_cmds.c:134: for (i = 0; i < (unsigned
long)&&end1-(unsigned long)&&lab1; i++) {
./hal/arm/sa110/mahwah/current/src/redboot_cmds.c:132: ip = (unsigned long
*)&&lab1;
./hal/arm/sa110/mahwah/current/src/redboot_cmds.c:134: for (i = 0; i <
(unsigned long)&&end1-(unsigned long)&&lab1; i++) {
./hal/arm/sa110/var/current/src/redboot_cmds.c:89: ip = (unsigned long *)&&lab1;
./hal/arm/sa110/var/current/src/redboot_cmds.c:90: for (i = 0; i < (unsigned
long)&&end1-(unsigned long)&&lab1; i++) {
./hal/arm/sa11x0/var/current/src/redboot_linux_exec.c:302: ip = (unsigned
long *)&&lab1;
./hal/arm/sa11x0/var/current/src/redboot_linux_exec.c:307: for (i = 0; i <
(unsigned long)&&end1-(unsigned long)&&lab1; i++) {
./hal/common/current/include/hal_stub.h:229: cyg_hal_gdb_place_break(
(target_register_t)&&cyg_hal_gdb_break_place ); \
./hal/common/current/include/hal_stub.h:237: cyg_hal_gdb_remove_break(
(target_register_t)&&cyg_hal_gdb_break_place ); \
./hal/common/current/src/hal_misc.c:64: (_x_) =
(CYG_ADDRWORD)&&__backup_return_address
./hal/common/current/src/hal_stub.c:766: __mem_fault_handler = &&err;
./hal/common/current/src/hal_stub.c:847: __mem_fault_handler = &&err;
./hal/common/current/src/hal_stub.c:901: __mem_fault_handler = &&err;
./hal/i386/arch/current/src/i386_stub.c:280: __mem_fault_handler = &&err;
./hal/i386/arch/current/src/i386_stub.c:306: __mem_fault_handler = &&err;
./hal/frv/arch/current/include/frv_stub.h:135:
cyg_hal_gdb_place_break((target_register_t)&&cyg_hal_gdb_break_place ); \
./kernel/current/tests/kcache2.c:645: *start = (unsigned long) &&label_start;
./kernel/current/tests/kcache2.c:646: *end = (unsigned long) &&label_end;
This has nothing to do with me! Jesper was the one who "owned" this.... This bug has moved to http://bugs.ecos.sourceware.org/show_bug.cgi?id=57427 |
The redboot_linux_exec.c's for arm9 excalibur, aaed2000, ebsa285, mahwah, sa110/var and sa11x0/var (at least right now) contain stuff that copies around assembler fragments using labels e.g. &&end1 However as revealed in newer compilers, gcc can move these labels around if they are not referenced to by gotos. A thread called "Re: asm volatile bug" on gcc-local revealed that this was not going to change, so eCos should be fixed. The aaed2000 added the workaround: #if 1 // This is here to work around a arm-linux-gnu-gcc compiler bug if (entry) goto lab1; else goto end1; #endif at the end. The addition of the gotos fixes the labels. This could be added for the other targets. The correct fix though is as rth wrote: This code fragment should be written asm (" asm_start: code asm_end: "); with this whole bit outside _any_ function. Then you write extern unsigned long asm_start[]; extern unsigned long asm_end[]; _prg = (unsigned long *) 0x1F00; n = asm_end - asm_start; for (i = 0; i < n; ++i) *_prg[i] = asm_start[i]; _fn = (void (*)(void)) _prg; _fn(); Not sure what to do about the "returns to function's return address", since I don't know if you can properly unwind the function's stack frame, or if that even matters in this context.