arm: adapt to being built as multilib #409
Conversation
Unit Test Results7 958 tests 7 434 ✅ 40m 41s ⏱️ Results for commit 920a997. ♻️ This comment has been updated with latest results. |
arch/arm/Makefile
Outdated
| @@ -14,9 +14,19 @@ else ifneq (,$(findstring v8m,$(TARGET_SUFF))) | |||
| include arch/arm/v8m/Makefile | |||
| else ifneq (,$(findstring v8r,$(TARGET_SUFF))) | |||
| include arch/arm/v8r/Makefile | |||
| else ifneq (,$(findstring v8r,$(TARGET_SUFF))) | |||
arch/arm/Makefile
Outdated
| else ifneq (,$(findstring v8r,$(TARGET_SUFF))) | ||
| include arch/arm/v8r/Makefile | ||
| else ifeq ($(TARGET_SUBFAMILY),multilib) | ||
| # TODO: Try to include correct Makefile if possible in case of a multilib. |
There was a problem hiding this comment.
I'm a little bit afraid that we will build non-fuctioning subvariants for supported targets while building the toolchain - eg. which implementation of setjmp will be used for armv7m7-imx117x with FDPIC enabled? (it's not a standard phoenix target, so it will be built only by multilib option?
Maybe we should try to inspect CFLAGS (march/mcpu options)? Or try using output of $(CC) $(CFALGS) -print-multi-directory?
There was a problem hiding this comment.
That would be the go to solution, however, I don't think I have time to fix it
There was a problem hiding this comment.
Note, however, that when we create such a configuration and build simply libphoenix during building the target then correct implementation will be used, as install is make target is called
| char *strncpy(char *dest, const char *src, size_t n) | ||
| { | ||
| size_t i = 0; | ||
| while (((i < n) && (src[i] != '\0'))) { |
There was a problem hiding this comment.
this is not a valid implementation of strncpy (zero-fill all of the remaining place, don't add `'\0' if it doesn't fit)
| } | ||
|
|
||
|
|
||
| int strncmp(const char *s1, const char *s2, size_t count) |
There was a problem hiding this comment.
invalid - strncmp("a", "ab", 2) returns 0 (should return negative value)
| } | ||
|
|
||
|
|
||
| int strcmp(const char *s1, const char *s2) |
There was a problem hiding this comment.
invalid - strcmp("a", "ab") returns 0 (should return negative value)
arch/arm/multilib/string.c
Outdated
| { | ||
| size_t i = 0; | ||
| while ((src[i] != '\0')) { | ||
| ((char *)dest)[i] = ((const char *)src)[i]; |
There was a problem hiding this comment.
why do we need explicit casts here?
There was a problem hiding this comment.
copied from memcpy
badochov
force-pushed
the
badochov/multilib
branch
from
317e341 to
7d4d2a9
Compare
badochov
force-pushed
the
badochov/multilib
branch
from
7d4d2a9 to
26fe01e
Compare
| .globl sym; \ | ||
| .type sym, %function; \ | ||
| sym: \ | ||
| bl abort; \ |
There was a problem hiding this comment.
bl as b may have to short range
| .globl _signal_trampoline | ||
| .type _signal_trampoline, %function | ||
| _signal_trampoline: | ||
| bl abort |
There was a problem hiding this comment.
bl as b may have to short range
JIRA: RTOS-1028
badochov
force-pushed
the
badochov/multilib
branch
from
26fe01e to
920a997
Compare
JIRA: RTOS-1028
Description
Multilib support is very basic and the multilibs are virtually useless as the syscalls are not implemented. To add proper support the ASM code should be chosen for each version based on GCC flags not target.
Motivation and Context
Types of changes
How Has This Been Tested?
Checklist:
Special treatment