macOS support for sim #179

yamt · 2020-01-29T07:05:16Z

No description provided.

I guess missing kconfig-conf is a common mistake for newbies like me. Bail out earlier instead of producing random errors during a build.

Namely this changes HOSTOS for macOS from "Other" to "Darwin". Also, suppress the following harmless messages during a "make": (Missing redirect in tools/Makefile.host) uname: illegal option -- o usage: uname [-amnprsv]

Mach-O assembler doesn't have them either.

My copy of GNU ld documentation is saying that -g is ignored. macOS's ld doesn't know the option and bails out. % ld -g ld: unknown option: -g %

Namely for Mach-O. Leave __CYGWIN__ case as it is.

The recent x86-64 convention requires 16-byte alignment before (not after) calling a function. This fixes snprintf crash I observed on macOS while saving XMM registers.

It seems that "ld -r" on macOS doesn't include objects from libraries for common symbols. Because of that, sim build ends up with undefined references to globals like g_binfmts and g_mmheap. @(#)PROGRAM:ld PROJECT:ld64-530 BUILD 18:57:17 Dec 13 2019 configured to support archs: armv6 armv7 armv7s arm64 arm64e arm64_32 i386 x86_64 x86_64h armv6m armv7k armv7m armv7em LTO support using: LLVM version 11.0.0, (clang-1100.0.33.17) (static support for 23, runtime is 23) TAPI support using: Apple TAPI version 11.0.0 (tapi-1100.0.11)

On some platforms like macOS, we can't create an empty archive.

yamt · 2020-01-29T07:12:15Z

apps part of this: apache/nuttx-apps#35

* ld doesn't have --start-groupi/--end-group things. As far as I know, it works that way by default. * objcopy with Mach-O support is not widely available. (GNU binutils seem to claim the support but it didn't actually work for me. llvm-objcopy --redefine-syms explicitly rejects Mach-O.) Instead, use -unexported_symbols_list linker flag to hide symbols to avoid conflicts with host symbols.

We usually don't remove input files like $OBJS.

xiaoxiang781216 · 2020-01-29T07:42:35Z

tools/sethost.sh

+  make olddefconfig V=1 || { echo "ERROR: failed to refresh"; exit 1; }
 else
-  make olddefconfig 1>/dev/null 2>&1
+  make olddefconfig 1>/dev/null || { echo "ERROR: failed to refresh"; exit 1; }


need keep 2>&1?

it's intentional to expose the errors.

xiaoxiang781216 · 2020-01-29T07:45:36Z

boards/sim/sim/sim/configs/cxxtest/Make.defs

 include ${TOPDIR}/tools/Config.mk

-HOSTOS = ${shell uname -o 2>/dev/null || echo "Other"}
+HOSTOS = ${shell uname -o 2>/dev/null || uname -s 2>/dev/null || echo "Other"}


should we fix the same issue in:
tools/define.sh
tools/incdir.sh
boards/arm/str71x/olimex-strp711/scripts/oocd.sh
arch/x86/src/Makefile

i guess it doesn't actually matter as they seem just checking if cygwin or not.

Yes, you are right.

xiaoxiang781216 · 2020-01-29T07:57:56Z

arch/sim/src/sim/up_initialstate.c

@@ -67,6 +67,6 @@
 void up_initial_state(struct tcb_s *tcb)
 {
  memset(&tcb->xcp, 0, sizeof(struct xcptcontext));


How about we add a dummy field in xcptcontext instead?

i don't understand your suggestion. can you explain?

Like the below dummy field:
struct xcptcontext
{
void sigdeliver; / Actual type is sig_deliver_t */
xcpt_reg_t dummy;
xcpt_reg_t regs[XCPTCONTEXT_REGS];
};

this change is about the actual stack alignment. that is, the value of RSP register.
not about the context structure.

My previous comment is wrong, but all places(up_use_stack/up_stack_frame) which modify adj_stack_ptr already align this pointer to 16B boundary, the extra subtraction just make the situation even worse, I can't understand why this change fix your problem.

the convention is, align before a function call.
that is, from POV of the function being called, it will be aligned after pushing RBP.
note that up_longjmp jumps into the function directly. (without pushing the return address)
RSP at that point should not be aligned.

xiaoxiang781216 · 2020-01-29T08:09:33Z

arch/sim/src/Makefile

+ifeq ($(HOSTOS),Darwin)
+  LDSTARTGROUP ?=
+  LDENDGROUP ?=
+  LDUNEXPORTSYMBOLS ?= -unexported_symbols_list $(HOSTOS)-names.dat


If so, why we don't update other platform to use -unexported_symbols_list? to simplify the logic in Makefile.

do gnu ld have the option? my copy of its documentation doesn't have it.

Emm, unexported_symbols_list doesn't support on Linux, but after searching we can install binutils and use gobjcopy? even for ld so we don't modify Makefile at all.

as i mentioned in the commit logs, gnu binutils didn't actually work for me.
objcopy runs without errors, producing a corrupted object.

xiaoxiang781216 · 2020-01-29T08:14:10Z

binfmt/binfmt_globals.c

 */

-FAR struct binfmt_s *g_binfmts;
+FAR struct binfmt_s *g_binfmts = NULL;


should we use -fno-common option instead? There are many place which don't initialize global variables explicitly.

i think it works. do you prefer it?

There are many places which don't set the global variable to NULL, the same error will report again and again if we turn more option in defconfig. The right fix is to add -fno-common.

ok, let me try the -fno-common approach later.

xiaoxiang781216 · 2020-01-29T08:22:33Z

arch/sim/src/sim/up_setjmp32.S

 #ifdef __CYGWIN__
 # define SYMBOL(s) _##s
 #else
+#ifdef __ELF__


The logic is wrong, should it be #ifndef?

oops, you are right.

yamt added 9 commits January 29, 2020 14:31

sethost.sh: Don't ignore "make olddefconfig" failure

a607853

I guess missing kconfig-conf is a common mistake for newbies like me. Bail out earlier instead of producing random errors during a build.

Use "uname -s" where "uname -o" is not available

ecb9cbf

Namely this changes HOSTOS for macOS from "Other" to "Darwin". Also, suppress the following harmless messages during a "make": (Missing redirect in tools/Makefile.host) uname: illegal option -- o usage: uname [-amnprsv]

sim: Check __ELF__ for .type/.size directives

86bd529

Mach-O assembler doesn't have them either.

sim: Fix typos in README.txt

ed2c615

sim: Remove -g from LDFLAGS/LDLINKFLAGS

b32483e

My copy of GNU ld documentation is saying that -g is ignored. macOS's ld doesn't know the option and bails out. % ld -g ld: unknown option: -g %

sim: Prefix symbols with _ for non-ELF

32d6387

Namely for Mach-O. Leave __CYGWIN__ case as it is.

sim: Fix stack alignment

a3d2a4a

The recent x86-64 convention requires 16-byte alignment before (not after) calling a function. This fixes snprintf crash I observed on macOS while saving XMM registers.

sim: Fix ostest build on macOS

23ed146

On some platforms like macOS, we can't create an empty archive.

yamt added 3 commits January 29, 2020 16:21

sim: Update documentation for macOS

faa921f

sim: Don't bother to remove nuttx.rel after consuming

a3f5a22

We usually don't remove input files like $OBJS.

yamt force-pushed the osx-sim branch from 1ea9929 to a3f5a22 Compare January 29, 2020 07:25

jerpelea merged commit 4174c81 into apache:master Jan 29, 2020

xiaoxiang781216 reviewed Jan 29, 2020

View reviewed changes

jerpelea mentioned this pull request Jan 29, 2020

Revert "macOS support for sim" #180

Closed

macOS support for sim #179

macOS support for sim #179

Uh oh!

Conversation

yamt commented Jan 29, 2020

Uh oh!

yamt commented Jan 29, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

xiaoxiang781216 Jan 29, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

xiaoxiang781216 Jan 29, 2020 •

edited

Loading