cpu/native: fix build with musl #18942

maribu · 2022-11-21T11:29:17Z

Contribution description

This changes a bunch of things that allows building with BOARD=native on 32-bit arches with the musl C lib, provided that libucontext-dev and pkg-config is installed.

Note that installing libucontext makes absolutely zero sense on C libs that do natively provide this deprecated System V API, such as glibc. Hence, it no sane glibc setup is expected to ever have libucontext installed.

Testing procedure

make BOARD=native -C examples/default

On a 32 bit system with musl as C lib and pkg-config as well as libucontext-dev installed should now succeed. The generated binary on glibc based 32 bit or multiarch systems should not change compared to master.

Issues/PRs references

Fixes #18937

chrysn

Works for me in a docker run --rm -it --network=host --platform linux/i386 alpine:edge container.

On master:

/tmp/RIOT # make -C examples/hello-world
make: Entering directory '/tmp/RIOT/examples/hello-world'
musl libc (i386)
Version 1.2.3
Dynamic Program Loader
Usage: /lib/ld-musl-i386.so.1 [options] [--] pathname
Building application "hello-world" for "native" with MCU "native".

"make" -C /tmp/RIOT/boards/common/init
"make" -C /tmp/RIOT/boards/native
"make" -C /tmp/RIOT/boards/native/drivers
"make" -C /tmp/RIOT/core
"make" -C /tmp/RIOT/core/lib
"make" -C /tmp/RIOT/cpu/native
/tmp/RIOT/cpu/native/irq_cpu.c: In function 'native_isr_entry':
/tmp/RIOT/cpu/native/irq_cpu.c:365:68: error: 'REG_EIP' undeclared (first use in this function)
  365 |     _native_saved_eip = ((ucontext_t *)context)->uc_mcontext.gregs[REG_EIP];
      |                                                                    ^~~~~~~
/tmp/RIOT/cpu/native/irq_cpu.c:365:68: note: each undeclared identifier is reported only once for each function it appears in
make[2]: *** [/tmp/RIOT/Makefile.base:146: /tmp/RIOT/examples/hello-world/bin/native/cpu/irq_cpu.o] Error 1
make[1]: *** [/tmp/RIOT/Makefile.base:31: ALL--/tmp/RIOT/cpu/native] Error 2
make: *** [/tmp/RIOT/examples/hello-world/../../Makefile.include:739: application_hello-world.module] Error 2
make: Leaving directory '/tmp/RIOT/examples/hello-world'

On this branch but without libucontext-dev (because maybe there is a cheap way to make pkgconfig err fatally in this case -- otherwise just so that it can be found through full-text search in the issue tracker):

/tmp/RIOT # make -C examples/hello-world
make: Entering directory '/tmp/RIOT/examples/hello-world'
musl libc (i386)
Version 1.2.3
Dynamic Program Loader
Usage: /lib/ld-musl-i386.so.1 [options] [--] pathname
Building application "hello-world" for "native" with MCU "native".

"make" -C /tmp/RIOT/boards/common/init
"make" -C /tmp/RIOT/boards/native
"make" -C /tmp/RIOT/boards/native/drivers
"make" -C /tmp/RIOT/core
"make" -C /tmp/RIOT/core/lib
"make" -C /tmp/RIOT/cpu/native
"make" -C /tmp/RIOT/cpu/native/periph
"make" -C /tmp/RIOT/cpu/native/stdio_native
"make" -C /tmp/RIOT/drivers
"make" -C /tmp/RIOT/drivers/periph_common
"make" -C /tmp/RIOT/sys
"make" -C /tmp/RIOT/sys/auto_init
"make" -C /tmp/RIOT/sys/libc
"make" -C /tmp/RIOT/sys/preprocessor
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: warning: /tmp/RIOT/examples/hello-world/bin/native/hello-world.elf has a LOAD segment with RWX permissions
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/irq_cpu.o: in function `native_isr_entry':
/tmp/RIOT/cpu/native/irq_cpu.c:340: undefined reference to `makecontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/irq_cpu.o: in function `native_interrupt_init':
/tmp/RIOT/cpu/native/irq_cpu.c:520: undefined reference to `getcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/cpu/native/irq_cpu.c:538: undefined reference to `makecontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/native_cpu.o: in function `isr_cpu_switch_context_exit':
/tmp/RIOT/cpu/native/native_cpu.c:175: undefined reference to `setcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/native_cpu.o: in function `isr_thread_yield':
/tmp/RIOT/cpu/native/native_cpu.c:227: undefined reference to `setcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/native_cpu.o: in function `thread_stack_init':
/tmp/RIOT/cpu/native/native_cpu.c:140: undefined reference to `getcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/cpu/native/native_cpu.c:153: undefined reference to `makecontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/native_cpu.o: in function `cpu_switch_context_exit':
/tmp/RIOT/cpu/native/native_cpu.c:196: undefined reference to `makecontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/cpu/native/native_cpu.c:197: undefined reference to `setcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/native_cpu.o: in function `thread_yield_higher':
/tmp/RIOT/cpu/native/native_cpu.c:245: undefined reference to `makecontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/cpu/native/native_cpu.c:246: undefined reference to `swapcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/native_cpu.o: in function `native_cpu_init':
/tmp/RIOT/cpu/native/native_cpu.c:255: undefined reference to `getcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/cpu/native/native_cpu.c:262: undefined reference to `makecontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/syscalls.o: in function `_native_syscall_leave':
/tmp/RIOT/cpu/native/syscalls.c:142: undefined reference to `makecontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/cpu/native/syscalls.c:143: undefined reference to `swapcontext'
/usr/lib/gcc/i586-alpine-linux-musl/12.2.1/../../../../i586-alpine-linux-musl/bin/ld: /tmp/RIOT/examples/hello-world/bin/native/cpu/tramp.o: in function `_native_sig_leave_tramp':
/tmp/RIOT/cpu/native/tramp.S:105: undefined reference to `swapcontext'
collect2: error: ld returned 1 exit status
make: *** [/tmp/RIOT/examples/hello-world/../../Makefile.include:730: /tmp/RIOT/examples/hello-world/bin/native/hello-world.elf] Error 1
make: Leaving directory '/tmp/RIOT/examples/hello-world'

On this branch with libucontext-dev installed: hello-world builds and runs (it terminates after the hello-world message; that may just be due to the container's terminal settings). Other programs still don't work (eg. filesystem and gcoap have trouble around struct timeval), but hey it's a start to get it working on more platforms.

chrysn · 2022-11-21T13:50:57Z

cpu/native/irq_cpu.c

@@ -13,10 +13,16 @@
 * @author  Ludwig Knüpfer <[email protected]>
 */

+/* __USE_GNU for gregs[REG_EIP] access under glibc
+ * _GNU_SOURCE for REG_EIP and strsignal() under musl */
+#define __USE_GNU


Pulling these to the top is a good call anyway (and when done in a .c file it can even work, distinct from what I tried in #18681).

riot-ci · 2022-11-21T16:44:07Z

Murdock results

✔️ PASSED

e93b5e4 core/thread: fix thread_measure_stack_free()

Success	Failures	Total	Runtime
10160	0	10161	17m:22s

Artifacts

Documentation preview

Teufelchen1 · 2024-01-24T15:36:22Z

What do you need to get this merged @maribu? 🙂

maribu · 2024-02-06T20:00:08Z

cpu/native/startup.c

+#if __GLIBC__
+static const bool _is_glibc = true;
+#else
+static const bool _is_glibc = false;
+#endif


This may look like a good use case for IS_ACTIVE(). However, IS_ACTIVE only works if the macro is either 1, 0, or undefined. __GLIBC__ however is 2 on current glibc based systems, so it just won't work.

maribu · 2024-02-06T20:03:53Z

cpu/native/native_cpu.c

@@ -262,7 +267,8 @@ void native_cpu_init(void)
        err(EXIT_FAILURE, "native_cpu_init: getcontext");
    }

-    end_context.uc_stack.ss_sp = __end_stack;
+    end_context.uc_stack.ss_sp = malloc(SIGSTKSZ);


Why no longer statically allocating? Because

# define SIGSTKSZ sysconf (_SC_SIGSTKSZ)

which is not a constant number, but function call issued at runtime. There still is a legacy compat mode in glibc that just defines this as a large constant, but moving the _GNU_SOURCE up disabled this compat.

benpicco · 2024-02-07T10:11:49Z

Looks like native64 is crashing in CI now 😕

This changes a bunch of things that allows building with the musl C lib, provided that `libucontext-dev` and `pkg-config` are installed. Note that installing libucontext makes absolutely zero sense on C libs that do natively provide this deprecated System V API, such as glibc. Hence, it no sane glibc setup is expected to ever have libucontext installed. A main pain point was that argv and argc are expected to be passed to init_fini handlers, but that is actually a glibc extension. This just parses `/proc/self/cmdline` by hand to populate argv and argc during startup, unless running on glibc.

chrysn · 2024-05-31T13:03:23Z

core/include/thread.h

 */
-uintptr_t thread_measure_stack_free(const char *stack);
+uintptr_t thread_measure_stack_free(const char *stack, size_t size);


Why not just pass a thread struct pointer here instead? That'd make it a typed pointer that's easier to document and harder to mix up. (Either works for RIOT-OS/rust-riot-wrappers#96 which is where I'm coming from)

That was also my first take. But ESP is using this function to measure the stack usage of the ISR stack, for which no thread_t is existing.

It is also a bit cumbersome for the idle thread statistics, as no thread_t pointer is available there. It would be possible to use thread_get(1), but that would bake in an assumption that no threads are created before kernel_init (e.g. not in periph_init()), which currently would work provided the "don't yield" flag is used.

OK, thanks for the explanation. If this were a higher level API I'd suggest having an internal low-level tool and high-level functions for "of thread *thread", "of the ISR thread" and "of the idle pseudothread", but as this is mostly used internally, fine.

Maybe I should rename it to measure_stack_free() and provide a thread_measure_stack_free() as static inline wrapper for convenience?

OK, did just that. The last commit points rust_riot_wrappers to the PR so that I can see if this now still segfaults on non-musl systems.

I intend to drop that commit, as pointing to some github.com/maribu/* is IMO not ideal. Who is that maribu, anyway, and how could we trust his personal repos?

Wonderful :) With the rust-riot-wrappers working with both variants, I reordered the commits to bump that first so that git biset is more fun :)

See-Also: RIOT-OS/RIOT#18942

maribu · 2024-05-31T17:11:43Z

Can I haz ACKs?

chrysn

ACK reaffirmed. I didn't test things again, trusting your testing. The rust-riot-wrappers update just pulls in the one fix that is needed for the API change. The API change is a pretty one now (programs will not accidentally compile because pointer content doesn't match).

chrysn · 2024-05-31T17:33:43Z

core/include/thread.h

 */
-uintptr_t thread_measure_stack_free(const char *stack);
+uintptr_t internal_measure_stack_free(const char *stack, size_t size);


I'd probably have gone with thread_measure_stack_free_internal so that the (publicly visible) symbol is still easily attributed – but either works, and I don't think we have a precise policy on internal-but-visible symbols.

Comprise: Let's go for measure_stack_free_internal().

The thread_ prefix would indicate it is not suitable for ISR stacks, but that is point of having this second flavour.

`thread_measure_stack_free()` previously assumed that reading past the stack is safe. When the stack was indeed part of a thread, the `thread_t` structure is put after the stack, increasing the odds of this assumption to hold. However, `thread_measure_stack_free()` could also be used on the ISR stack, which may be allocated at the end of SRAM. A second parameter had to be added to indicate the stack size, so that reading past the stack can now be prevented. This also makes valgrind happy on `native`/`native64`.

maribu · 2024-06-01T06:53:28Z

Hooray 🎉 Thx a bunch!

benpicco · 2024-06-04T14:33:10Z

This breaks the stack free measurement :(

> ps
2024-06-04 16:27:07,709 # 	pid | name                 | state    Q | pri | stack  ( used) ( free) | base addr  | current     
2024-06-04 16:27:07,718 # 	  - | isr_stack            | -        - |   - |    512 (  172) (  340) | 0x20000000 | 0x200001c8
2024-06-04 16:27:07,726 # 	  1 | main                 | running  Q |   7 |   1536 ( 1536) (    0) | 0x20000340 | 0x20000854 
2024-06-04 16:27:07,736 # 	  2 | pktdump              | bl rx    _ |   6 |   1472 ( 1472) (    0) | 0x20002fb0 | 0x200034c4 
2024-06-04 16:27:07,744 # 	  3 | ipv6                 | bl rx    _ |   4 |    960 (  960) (    0) | 0x20000a5c | 0x20000d54 
2024-06-04 16:27:07,752 # 	  4 | udp                  | bl rx    _ |   5 |    448 (  448) (    0) | 0x20003bfc | 0x20003cfc 
2024-06-04 16:27:07,761 # 	  5 | sam0_eth             | bl anyfl _ |   2 |    896 (  896) (    0) | 0x200013b8 | 0x2000167c 
2024-06-04 16:27:07,770 # 	  6 | RPL                  | bl rx    _ |   5 |   1024 ( 1024) (    0) | 0x200037b8 | 0x20003aec 
2024-06-04 16:27:07,776 # 	    | SUM                  |            |     |   6848 ( 6508) (  340)

github-actions bot added Area: cpu Area: CPU/MCU ports Platform: native Platform: This PR/issue effects the native platform labels Nov 21, 2022

benpicco added the CI: ready for build If set, CI server will compile all applications for all available boards for the labeled PR label Nov 21, 2022

chrysn approved these changes Nov 21, 2022

View reviewed changes

maribu enabled auto-merge November 21, 2022 15:35

benpicco added CI: ready for build If set, CI server will compile all applications for all available boards for the labeled PR and removed CI: ready for build If set, CI server will compile all applications for all available boards for the labeled PR labels Aug 23, 2023

auto-merge was automatically disabled November 8, 2023 12:48
Merge queue setting changed

maribu force-pushed the cpu/native branch from efa251b to 53de305 Compare February 6, 2024 19:53

maribu commented Feb 6, 2024

View reviewed changes

maribu force-pushed the cpu/native branch from 53de305 to ef93040 Compare February 6, 2024 20:05

maribu force-pushed the cpu/native branch 2 times, most recently from 5321059 to 9aa9df8 Compare May 31, 2024 07:42

maribu force-pushed the cpu/native branch from 9aa9df8 to be3e194 Compare May 31, 2024 10:18

maribu requested review from gschorcht and kaspar030 as code owners May 31, 2024 10:18

github-actions bot added Area: core Area: RIOT kernel. Handle PRs marked with this with care! Platform: ESP Platform: This PR/issue effects ESP-based platforms Area: sys Area: System labels May 31, 2024

maribu force-pushed the cpu/native branch from be3e194 to 1e72775 Compare May 31, 2024 11:13

maribu mentioned this pull request May 31, 2024

thread: adapt to API change of thread_measure_stack_free() RIOT-OS/rust-riot-wrappers#96

Merged

maribu requested review from leandrolanzieri, aabadie and MichelRottleuthner as code owners May 31, 2024 12:33

github-actions bot added the Area: tests Area: tests and testing framework label May 31, 2024

github-actions bot added the Area: examples Area: Example Applications label May 31, 2024

chrysn reviewed May 31, 2024

View reviewed changes

maribu force-pushed the cpu/native branch 2 times, most recently from 829cf29 to dc79930 Compare May 31, 2024 14:37

chrysn added a commit to RIOT-OS/rust-riot-wrappers that referenced this pull request May 31, 2024

thread: accept either version of thread_measure_stack_free()

d0e0ff8

See-Also: RIOT-OS/RIOT#18942

*/Cargo.lock: bump rust-riot-wrappers

835eaee

maribu force-pushed the cpu/native branch 2 times, most recently from 3164824 to 8b6192a Compare May 31, 2024 16:35

maribu added Process: API change Integration Process: PR contains or issue proposes an API change. Should be handled with care. Process: needs >1 ACK Integration Process: This PR requires more than one ACK labels May 31, 2024

github-actions bot added the Process: missing approvals Integration Process: PR needs more ACKS (handled by action) label May 31, 2024

chrysn approved these changes May 31, 2024

View reviewed changes

maribu force-pushed the cpu/native branch from 8b6192a to e93b5e4 Compare May 31, 2024 17:54

github-actions bot removed the Process: missing approvals Integration Process: PR needs more ACKS (handled by action) label May 31, 2024

maribu added this pull request to the merge queue May 31, 2024

Merged via the queue into RIOT-OS:master with commit 886c6a2 May 31, 2024
25 checks passed

maribu deleted the cpu/native branch June 1, 2024 06:53

maribu mentioned this pull request Jun 4, 2024

cpu/native: fix compilation with GCC 14.1 #20722

Merged

benpicco mentioned this pull request Jun 4, 2024

measure_stack_free_internal(): don't try to align end of stack #20723

Merged

chrysn mentioned this pull request Jun 4, 2024

Do not count multiple reviews from the same user RIOT-OS/check-labels-action#2

Merged

mguetschow added this to the Release 2024.07 milestone Jun 11, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

cpu/native: fix build with musl #18942

cpu/native: fix build with musl #18942

maribu commented Nov 21, 2022 •

edited

Loading

chrysn left a comment

chrysn Nov 21, 2022

riot-ci commented Nov 21, 2022 •

edited

Loading

Teufelchen1 commented Jan 24, 2024

maribu Feb 6, 2024

maribu Feb 6, 2024

benpicco commented Feb 7, 2024

chrysn May 31, 2024

maribu May 31, 2024

chrysn May 31, 2024

maribu May 31, 2024

maribu May 31, 2024

maribu May 31, 2024

maribu commented May 31, 2024

chrysn left a comment

chrysn May 31, 2024

maribu May 31, 2024

maribu commented Jun 1, 2024

benpicco commented Jun 4, 2024 •

edited

Loading

cpu/native: fix build with musl #18942

cpu/native: fix build with musl #18942

Conversation

maribu commented Nov 21, 2022 • edited Loading

Contribution description

Testing procedure

Issues/PRs references

chrysn left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

riot-ci commented Nov 21, 2022 • edited Loading

Murdock results

Artifacts

Teufelchen1 commented Jan 24, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

benpicco commented Feb 7, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

maribu commented May 31, 2024

chrysn left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

maribu commented Jun 1, 2024

benpicco commented Jun 4, 2024 • edited Loading

maribu commented Nov 21, 2022 •

edited

Loading

riot-ci commented Nov 21, 2022 •

edited

Loading

benpicco commented Jun 4, 2024 •

edited

Loading