1. Patchew
  2. linux
  3. View series

[PATCH bpf] selftests/bpf: fix implicit-function-declaration errors

Matthieu Baerts (NGI0) posted 1 patch 2 months, 2 weeks ago 
tools/testing/selftests/bpf/progs/dynptr_success.c | 1 +
1 file changed, 1 insertion(+)
[PATCH bpf] selftests/bpf: fix implicit-function-declaration errors
Posted by Matthieu Baerts (NGI0) 2 months, 2 weeks ago 
When trying to build the latest BPF selftests, with a debug kernel
config, Pahole 1.30 and CLang 20.1.8 (and GCC 15.2), I got these errors:

  progs/dynptr_success.c:579:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
    579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
        |                ^
  progs/dynptr_success.c:579:9: note: did you mean 'bpf_dynptr_size'?
  .virtme/build-debug-btf//tools/include/vmlinux.h:120280:14: note: 'bpf_dynptr_size' declared here
   120280 | extern __u32 bpf_dynptr_size(const struct bpf_dynptr *p) __weak __ksym;
          |              ^
  progs/dynptr_success.c:579:7: error: incompatible integer to pointer conversion assigning to '__u64 *' (aka 'unsigned long long *') from 'int' [-Wint-conversion]
    579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
        |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  progs/dynptr_success.c:596:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
    596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
        |                ^
  progs/dynptr_success.c:596:7: error: incompatible integer to pointer conversion assigning to 'char *' from 'int' [-Wint-conversion]
    596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
        |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I don't have these errors without the debug kernel config from
kernel/configs/debug.config. With the debug kernel, bpf_dynptr_slice()
is not declared in vmlinux.h. It is declared there without debug.config.

The fix is similar to what is done in dynptr_fail.c which is also using
bpf_dynptr_slice(): bpf_kfuncs.h is now included.

Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
---
Notes:
 - This patch looks wrong, I guess bpf_dynptr_slice() should be in
   vmlinux.h even with a "debug" kernel, but it is not:
   $ grep -cw bpf_dynptr_slice .virtme/build-debug-btf/tools/include/vmlinux.h
   0
   $ grep -w bpf_dynptr_slice .virtme/build-btf/tools/include/vmlinux.h
   extern void *bpf_dynptr_slice(...) __weak __ksym;
 - This is on top of bpf/master: commit 63d2247e2e37, tag bpf-fixes.
 - I only see this error when using kernel/configs/debug.config.
 - Because this has not been spot by the BPF CI, I wonder if I'm
   building the BPF selftests properly... Here is what I did:
   $ virtme-configkernel --arch x86_64 --defconfig \
     --custom tools/testing/selftests/net/mptcp/config \
     --custom kernel/configs/debug.config \
     --custom tools/testing/selftests/bpf/config \
     O=${PWD}/.virtme/build-debug-btf
   $ ./scripts/config --file ${PWD}/.virtme/build-debug-btf/.config \
     -e NET_NS_REFCNT_TRACKER -d SLUB_DEBUG_ON \
     -d DEBUG_KMEMLEAK_AUTO_SCAN -e PANIC_ON_OOPS \
     -e SOFTLOCKUP_DETECTOR -e BOOTPARAM_SOFTLOCKUP_PANIC \
     -e HARDLOCKUP_DETECTOR -e BOOTPARAM_HUNG_TASK_PANIC \
     -e DETECT_HUNG_TASK -e BOOTPARAM_HUNG_TASK_PANIC -e DEBUG_INFO \
     -e DEBUG_INFO_DWARF_TOOLCHAIN_DEFAULT -e GDB_SCRIPTS \
     -e DEBUG_INFO_DWARF4 -e DEBUG_INFO_COMPRESSED \
     -e DEBUG_INFO_COMPRESSED_ZLIB -e DEBUG_INFO_BTF_MODULES \
     -e MODULE_ALLOW_BTF_MISMATCH -d IA32_EMULATION -e DYNAMIC_DEBUG \
     --set-val CONSOLE_LOGLEVEL_DEFAULT 8 -e FTRACE -e FUNCTION_TRACER \
     -e DYNAMIC_FTRACE -e FTRACE_SYSCALLS -e HIST_TRIGGERS -e DEBUG_NET \
     -m KUNIT -e KUNIT_DEBUGFS -d KUNIT_ALL_TESTS -m MPTCP_KUNIT_TEST \
     -e BPF_JIT -e BPF_SYSCALL -e TUN -e CRYPTO_USER_API_HASH \
     -e CRYPTO_SHA1 -e NET_SCH_TBF -e BRIDGE -d RETPOLINE -d PCCARD \
     -d MACINTOSH_DRIVERS -d SOUND -d USB_SUPPORT -d NEW_LEDS -d SCSI \
     -d SURFACE_PLATFORMS -d DRM -d FB -d ATA -d MISC_FILESYSTEMS
     # sorry, long list used by the MPTCP CI to accelerate builds, etc.
   $ make O=${PWD}/.virtme/build-debug-btf olddefconfig
   $ make O=${PWD}/.virtme/build-debug-btf -j$(nproc) -l$(nproc)
   $ make O=${PWD}/.virtme/build-debug-btf headers_install \
     INSTALL_HDR_PATH=${PWD}/.virtme/headers
   $ make O=${PWD}/.virtme/build-debug-btf \
     KHDR_INCLUDES=-I${PWD}/.virtme/headers/includes \
     -C tools/testing/selftests/bpf
 - The errors I got should be reproducible using:
   $ docker run -v "${PWD}:${PWD}:rw" -w "${PWD}" --privileged --rm -it \
     -e INPUT_EXTRA_ENV=INPUT_RUN_TESTS_ONLY=bpftest_all \
     --pull always mptcp/mptcp-upstream-virtme-docker:latest \
     auto-btf-debug
 - These issues were originally spot by our MPTCP CI:
   https://github.com/multipath-tcp/mptcp_net-next/actions/runs/18222911614/job/51886811332
 - No errors without kernel/configs/debug.config on the CI and on my side
 - This CI got different issues, and I had to declare more kfuncs there:
   https://github.com/multipath-tcp/mptcp_net-next/commit/4435d4da9f4f
   but this CI is currently on top of 'net', with Jiri's patches from
   https://lore.kernel.org/20251001122223.170830-1-jolsa@kernel.org
 - The builds have been done from a clean build directory each time.
 - Do you think the issue is on my side? Dependences? How the selftests
   are built? I didn't change the way the BPF selftests are built for a
   while. I had other issues with pahole 1.29, but fixed with 1.30.
 - Feel free to discard this patch for a better solution (if any).
 - I don't know which Fixes tag adding, but I doubt this patch is valid.
---
 tools/testing/selftests/bpf/progs/dynptr_success.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tools/testing/selftests/bpf/progs/dynptr_success.c b/tools/testing/selftests/bpf/progs/dynptr_success.c
index 127dea342e5a67dda33e0a39e84d135206d2f3f1..60daf5ce8eb283d8c8bf2d7853eda6313df4fa87 100644
--- a/tools/testing/selftests/bpf/progs/dynptr_success.c
+++ b/tools/testing/selftests/bpf/progs/dynptr_success.c
@@ -6,6 +6,7 @@
 #include <stdbool.h>
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
+#include "bpf_kfuncs.h"
 #include "bpf_misc.h"
 #include "errno.h"
 

---
base-commit: 63d2247e2e37d9c589a0a26aa4e684f736a45e29
change-id: 20251003-bpf-sft-fix-build-err-6-18-6a4c032f680a

Best regards,
-- 
Matthieu Baerts (NGI0) <matttbe@kernel.org>
Re: [PATCH bpf] selftests/bpf: fix implicit-function-declaration errors
Posted by Eduard Zingerman 2 months, 2 weeks ago 
On Fri, 2025-10-03 at 17:24 +0200, Matthieu Baerts (NGI0) wrote:
> When trying to build the latest BPF selftests, with a debug kernel
> config, Pahole 1.30 and CLang 20.1.8 (and GCC 15.2), I got these errors:
> 
>   progs/dynptr_success.c:579:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>     579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
>         |                ^
>   progs/dynptr_success.c:579:9: note: did you mean 'bpf_dynptr_size'?
>   .virtme/build-debug-btf//tools/include/vmlinux.h:120280:14: note: 'bpf_dynptr_size' declared here
>    120280 | extern __u32 bpf_dynptr_size(const struct bpf_dynptr *p) __weak __ksym;
>           |              ^
>   progs/dynptr_success.c:579:7: error: incompatible integer to pointer conversion assigning to '__u64 *' (aka 'unsigned long long *') from 'int' [-Wint-conversion]
>     579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
>         |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>   progs/dynptr_success.c:596:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>     596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
>         |                ^
>   progs/dynptr_success.c:596:7: error: incompatible integer to pointer conversion assigning to 'char *' from 'int' [-Wint-conversion]
>     596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
>         |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 
> I don't have these errors without the debug kernel config from
> kernel/configs/debug.config. With the debug kernel, bpf_dynptr_slice()
> is not declared in vmlinux.h. It is declared there without debug.config.
> 
> The fix is similar to what is done in dynptr_fail.c which is also using
> bpf_dynptr_slice(): bpf_kfuncs.h is now included.
> 
> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
> ---

I can reproduce similar issue when including
kernel/configs/debug.config with my regular dev config, but for
different functions: bpf_rcu_read_{un,}lock().

However, this is not a way to fix this.
Kfuncs are not supposed to just disappear from DWARF.

Running pahole in verbose mode I see the following output:

  $ pahole -V \
      --btf_features=encode_force,var,float,enum64,decl_tag,type_tag,optimized_func,consistent_func,decl_tag_kfuncs \
      --btf_features=attributes \
      --lang_exclude=rust \
      --btf_encode_detached=/dev/null vmlinux
  ...
  matched function 'bpf_rcu_read_lock' with 'bpf_rcu_read_lock.cold'
  ...

Alan, Ihor, does this sound familiar?
Re: [PATCH bpf] selftests/bpf: fix implicit-function-declaration errors
Posted by Matthieu Baerts 2 months, 2 weeks ago 
Hi Eduard,

On 04/10/2025 01:37, Eduard Zingerman wrote:
> On Fri, 2025-10-03 at 17:24 +0200, Matthieu Baerts (NGI0) wrote:
>> When trying to build the latest BPF selftests, with a debug kernel
>> config, Pahole 1.30 and CLang 20.1.8 (and GCC 15.2), I got these errors:
>>
>>   progs/dynptr_success.c:579:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>>     579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
>>         |                ^
>>   progs/dynptr_success.c:579:9: note: did you mean 'bpf_dynptr_size'?
>>   .virtme/build-debug-btf//tools/include/vmlinux.h:120280:14: note: 'bpf_dynptr_size' declared here
>>    120280 | extern __u32 bpf_dynptr_size(const struct bpf_dynptr *p) __weak __ksym;
>>           |              ^
>>   progs/dynptr_success.c:579:7: error: incompatible integer to pointer conversion assigning to '__u64 *' (aka 'unsigned long long *') from 'int' [-Wint-conversion]
>>     579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
>>         |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>   progs/dynptr_success.c:596:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>>     596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
>>         |                ^
>>   progs/dynptr_success.c:596:7: error: incompatible integer to pointer conversion assigning to 'char *' from 'int' [-Wint-conversion]
>>     596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
>>         |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>
>> I don't have these errors without the debug kernel config from
>> kernel/configs/debug.config. With the debug kernel, bpf_dynptr_slice()
>> is not declared in vmlinux.h. It is declared there without debug.config.
>>
>> The fix is similar to what is done in dynptr_fail.c which is also using
>> bpf_dynptr_slice(): bpf_kfuncs.h is now included.
>>
>> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
>> ---
> 
> I can reproduce similar issue when including
> kernel/configs/debug.config with my regular dev config, but for
> different functions: bpf_rcu_read_{un,}lock().

Thank you for having checked! I also had issues with these functions on
my side, when testing on top of 'net'.

> However, this is not a way to fix this.
> Kfuncs are not supposed to just disappear from DWARF.

Indeed. But strange it was fine before. Or fine without debug.config.

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.
Re: [PATCH bpf] selftests/bpf: fix implicit-function-declaration errors
Posted by Ihor Solodrai 2 months, 2 weeks ago 

On 10/3/25 4:37 PM, Eduard Zingerman wrote:
> On Fri, 2025-10-03 at 17:24 +0200, Matthieu Baerts (NGI0) wrote:
>> When trying to build the latest BPF selftests, with a debug kernel
>> config, Pahole 1.30 and CLang 20.1.8 (and GCC 15.2), I got these errors:
>>
>>   progs/dynptr_success.c:579:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>>     579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
>>         |                ^
>>   progs/dynptr_success.c:579:9: note: did you mean 'bpf_dynptr_size'?
>>   .virtme/build-debug-btf//tools/include/vmlinux.h:120280:14: note: 'bpf_dynptr_size' declared here
>>    120280 | extern __u32 bpf_dynptr_size(const struct bpf_dynptr *p) __weak __ksym;
>>           |              ^
>>   progs/dynptr_success.c:579:7: error: incompatible integer to pointer conversion assigning to '__u64 *' (aka 'unsigned long long *') from 'int' [-Wint-conversion]
>>     579 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 1);
>>         |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>   progs/dynptr_success.c:596:9: error: call to undeclared function 'bpf_dynptr_slice'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
>>     596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
>>         |                ^
>>   progs/dynptr_success.c:596:7: error: incompatible integer to pointer conversion assigning to 'char *' from 'int' [-Wint-conversion]
>>     596 |         data = bpf_dynptr_slice(&ptr, 0, NULL, 10);
>>         |              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>
>> I don't have these errors without the debug kernel config from
>> kernel/configs/debug.config. With the debug kernel, bpf_dynptr_slice()
>> is not declared in vmlinux.h. It is declared there without debug.config.
>>
>> The fix is similar to what is done in dynptr_fail.c which is also using
>> bpf_dynptr_slice(): bpf_kfuncs.h is now included.
>>
>> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
>> ---
> 
> I can reproduce similar issue when including
> kernel/configs/debug.config with my regular dev config, but for
> different functions: bpf_rcu_read_{un,}lock().
> 
> However, this is not a way to fix this.
> Kfuncs are not supposed to just disappear from DWARF.
> 
> Running pahole in verbose mode I see the following output:
> 
>   $ pahole -V \
>       --btf_features=encode_force,var,float,enum64,decl_tag,type_tag,optimized_func,consistent_func,decl_tag_kfuncs \
>       --btf_features=attributes \
>       --lang_exclude=rust \
>       --btf_encode_detached=/dev/null vmlinux
>   ...
>   matched function 'bpf_rcu_read_lock' with 'bpf_rcu_read_lock.cold'
>   ...
> 
> Alan, Ihor, does this sound familiar?

This is most likely the issue addressed in this patch:
https://lore.kernel.org/dwarves/f7553b3f-5827-4f50-81a9-9bd0802734b9@linux.dev/

There wasn't a new pahole release with it yet.
Re: [PATCH bpf] selftests/bpf: fix implicit-function-declaration errors
Posted by Ihor Solodrai 2 months, 2 weeks ago 

On 10/3/25 8:08 PM, Ihor Solodrai wrote:
> 
> 
> On 10/3/25 4:37 PM, Eduard Zingerman wrote:
>> On Fri, 2025-10-03 at 17:24 +0200, Matthieu Baerts (NGI0) wrote:
>> [...]
>>
>> Alan, Ihor, does this sound familiar?
> 
> This is most likely the issue addressed in this patch:
> https://lore.kernel.org/dwarves/f7553b3f-5827-4f50-81a9-9bd0802734b9@linux.dev/
> 
> There wasn't a new pahole release with it yet.

Not the best link, sorry. That patch wasn't picked up by lore, only
discussion remains.

Here is the commit pushed to pahole/next:
https://git.kernel.org/pub/scm/devel/pahole/pahole.git/commit/?h=next&id=09c1e9c924da02dc02bba0a3e59490e64449df96
Re: [PATCH bpf] selftests/bpf: fix implicit-function-declaration errors
Posted by Matthieu Baerts 2 months, 2 weeks ago 
Hi Ihor,

On 04/10/2025 05:19, Ihor Solodrai wrote:
> 
> 
> On 10/3/25 8:08 PM, Ihor Solodrai wrote:
>>
>>
>> On 10/3/25 4:37 PM, Eduard Zingerman wrote:
>>> On Fri, 2025-10-03 at 17:24 +0200, Matthieu Baerts (NGI0) wrote:
>>> [...]
>>>
>>> Alan, Ihor, does this sound familiar?
>>
>> This is most likely the issue addressed in this patch:
>> https://lore.kernel.org/dwarves/f7553b3f-5827-4f50-81a9-9bd0802734b9@linux.dev/
>>
>> There wasn't a new pahole release with it yet.
> 
> Not the best link, sorry. That patch wasn't picked up by lore, only
> discussion remains.
> 
> Here is the commit pushed to pahole/next:
> https://git.kernel.org/pub/scm/devel/pahole/pahole.git/commit/?h=next&id=09c1e9c924da02dc02bba0a3e59490e64449df96

Thank you for the links!

OK so I guess to be (fully) able to use BTF in the future v6.18, we
should use the future Pahole v1.31, and not include extra kfuncs. Is
that correct?

If yes, we can drop this patch. (The MPTCP CI will use it internally not
to have to depend on a dev version of Pahole.)

Also, should this dependency be more explicit somehow, to help people
having errors when not using a recent enough Pahole version?

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.

Patchew 2.1-devel-b67e4c2

© 2016 - 2025 Red Hat, Inc.