Skip to contentSkip to conversation

perf: ignore deprecation warning when using libbpf's btf__get_from_id()

PATCH In review Andrii Nakryiko 5 participants 7 messages Sep 14, 2021 – Sep 15, 2021

Subsystems (2)
lore
Share & tools
lore

Andrii Nakryiko

author maintainer
#0 lore raw

PATCH 1/1 perf: ignore deprecation warning when using libbpf's btf__get_from_id()

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as a weak function, presumably to dynamically link against old version of libbpf shared library. Unfortunately this causes compilation warning when perf is compiled against libbpf v0.6+.

For now, just ignore deprecation warning, but there might be a better solution, depending on perf's needs.

---
 tools/perf/util/bpf-event.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/bpf-event.c +3 -0
index 683f6d63560e..1a7112a87736 100644
@@ -24,7 +24,10 @@
 struct btf * __weak btf__load_from_kernel_by_id(__u32 id)
 {
        struct btf *btf;
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
        int err = btf__get_from_id(id, &btf);
+#pragma GCC diagnostic pop
 
        return err ? ERR_PTR(err) : btf;
 }
Signature · 1 line
2.30.2
#1 lore raw
Quoted 7 lines · Andrii Nakryiko

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as a weak function, presumably to dynamically link against old version of libbpf shared library. Unfortunately this causes compilation warning when perf is compiled against libbpf v0.6+.

For now, just ignore deprecation warning, but there might be a better solution, depending on perf's needs.

HI,
the problem we tried to solve is when perf is using symbols which are not yet available in released libbpf.. but it all linkes in default perf build because it's linked statically libbpf.a in the tree

so now we have weak function with that warning disabled locally, which I guess could work? also for future cases like that

jirka

Quoted 24 lines
---
 tools/perf/util/bpf-event.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/bpf-event.c +3 -0
index 683f6d63560e..1a7112a87736 100644
@@ -24,7 +24,10 @@
 struct btf * __weak btf__load_from_kernel_by_id(__u32 id)
 {
        struct btf *btf;
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
        int err = btf__get_from_id(id, &btf);
+#pragma GCC diagnostic pop
 
        return err ? ERR_PTR(err) : btf;
 }
Signature · 1 line
2.30.2
#2 lore raw
Quoted 7 lines · Jiri Olsa
Quoted 5 lines · Andrii Nakryiko

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as a weak function, presumably to dynamically link against old version of libbpf shared library. Unfortunately this causes compilation warning when perf is compiled against libbpf v0.6+.

Quoted 9 lines
Quoted 2 lines

For now, just ignore deprecation warning, but there might be a better solution, depending on perf's needs.

HI,
the problem we tried to solve is when perf is using symbols which are not yet available in released libbpf.. but it all linkes in default perf build because it's linked statically libbpf.a in the tree

If you are always statically linking libbpf into perf, there is no need to implement this __weak shim. Libbpf is never going to deprecate an API if a new/replacement API hasn't been at least in a previous released version. So in this case btf__load_from_kernel_by_id() was added in libbpf 0.5, and btf__get_from_id() was marked deprecated in libbpf 0.6 (not yet released, of course). So with that, do you still think we need this __weak re-implementation?

I was wondering if this was done to make latest perf code compile against some old libbpf source code or dynamically linked against old libbpf. But if that's not the case, the fix should be a removal of __weak btf__load_from_kernel_by_id().

Quoted 30 lines

so now we have weak function with that warning disabled locally, which I guess could work? also for future cases like that

jirka

Quoted 24 lines
---
 tools/perf/util/bpf-event.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/bpf-event.c +3 -1
index 683f6d63560e..1a7112a87736 100644
@@ -24,7 +24,10 @@
 struct btf * __weak btf__load_from_kernel_by_id(__u32 id)
 {
        struct btf *btf;
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
        int err = btf__get_from_id(id, &btf);
+#pragma GCC diagnostic pop
 
        return err ? ERR_PTR(err) : btf;
 }
--

2.30.2

#3 lore raw
Quoted 18 lines · Andrii Nakryiko
Quoted 7 lines · Jiri Olsa
Quoted 5 lines · Andrii Nakryiko

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as a weak function, presumably to dynamically link against old version of libbpf shared library. Unfortunately this causes compilation warning when perf is compiled against libbpf v0.6+.

Quoted 9 lines
Quoted 2 lines

For now, just ignore deprecation warning, but there might be a better solution, depending on perf's needs.

HI,
the problem we tried to solve is when perf is using symbols which are not yet available in released libbpf.. but it all linkes in default perf build because it's linked statically libbpf.a in the tree

Quoted 7 lines

If you are always statically linking libbpf into perf, there is no need to implement this __weak shim. Libbpf is never going to deprecate an API if a new/replacement API hasn't been at least in a previous released version. So in this case btf__load_from_kernel_by_id() was added in libbpf 0.5, and btf__get_from_id() was marked deprecated in libbpf 0.6 (not yet released, of course). So with that, do you still think we need this __weak re-implementation?

we package/build perf to dynamically link to libbpf, so there's time window where perf already uses new libbpf function that has not been released yet in libbpf and perf build fails

Arnaldo has another solution using feature detection and have ifdefs to take care about that, but having the weak functions is less code and seems more manageable

jirka

Quoted 37 lines

I was wondering if this was done to make latest perf code compile against some old libbpf source code or dynamically linked against old libbpf. But if that's not the case, the fix should be a removal of __weak btf__load_from_kernel_by_id().

Quoted 30 lines

so now we have weak function with that warning disabled locally, which I guess could work? also for future cases like that

jirka

Quoted 24 lines
---
 tools/perf/util/bpf-event.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/bpf-event.c +3 -1
index 683f6d63560e..1a7112a87736 100644
@@ -24,7 +24,10 @@
 struct btf * __weak btf__load_from_kernel_by_id(__u32 id)
 {
        struct btf *btf;
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
        int err = btf__get_from_id(id, &btf);
+#pragma GCC diagnostic pop
 
        return err ? ERR_PTR(err) : btf;
 }
--

2.30.2

#4 lore raw
In reply to Andrii Nakryiko

Em Tue, Sep 14, 2021 at 11:28:28AM -0700, Andrii Nakryiko escreveu:

Quoted 26 lines · Jiri Olsa
Quoted 7 lines · Andrii Nakryiko
Quoted 5 lines

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as a weak function, presumably to dynamically link against old version of libbpf shared library. Unfortunately this causes compilation warning when perf is compiled against libbpf v0.6+.

Quoted 9 lines
Quoted 2 lines

For now, just ignore deprecation warning, but there might be a better solution, depending on perf's needs.

HI,
the problem we tried to solve is when perf is using symbols which are not yet available in released libbpf.. but it all linkes in default perf build because it's linked statically libbpf.a in the tree

If you are always statically linking libbpf into perf, there is no need to implement this __weak shim. Libbpf is never going to deprecate an API if a new/replacement API hasn't been at least in a previous released version. So in this case btf__load_from_kernel_by_id() was added in libbpf 0.5, and btf__get_from_id() was marked deprecated in libbpf 0.6 (not yet released, of course). So with that, do you still think we need this __weak re-implementation?

Quoted 4 lines

I was wondering if this was done to make latest perf code compile against some old libbpf source code or dynamically linked against old libbpf. But if that's not the case, the fix should be a removal of __weak btf__load_from_kernel_by_id().

It was made to build against the libbpf that comes with fedora 34, the distro I'm using, which is:

⬢[acme@toolbox perf]$ sudo dnf install libbpf-devel Package libbpf-devel-2:0.4.0-1.fc34.x86_64 is already installed. Dependencies resolved.
Nothing to do.
Complete!
⬢[acme@toolbox perf]$ cat /etc/redhat-release
Fedora release 34 (Thirty Four)

And we have 'make -C tools/perf build-test' that has one entry to build with LIBBPF_EXTERNAL=1, i.e. using whatever libbpf-devel package is installed in the distro, in addtion to statically linking with the libbpf in the kernel sources.

That is done because several distros are linking perf with the libbpf they ship.

When I merged the latest upstream this test failed, and I realized that some files in tools/perf/ had changed to make use of a new function and that was the reason for the build test failure.

So I tried to provide a transition help for these cases, initially as a feature test that would look if that new function was available and if not, provide the fallback, but then ended up following Jiri's suggestion for a __weak function, as that involved less coding.

- Arnaldo

Quoted 30 lines
Quoted 30 lines

so now we have weak function with that warning disabled locally, which I guess could work? also for future cases like that

jirka

Quoted 24 lines
---
 tools/perf/util/bpf-event.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/bpf-event.c +3 -1
index 683f6d63560e..1a7112a87736 100644
@@ -24,7 +24,10 @@
 struct btf * __weak btf__load_from_kernel_by_id(__u32 id)
 {
        struct btf *btf;
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
        int err = btf__get_from_id(id, &btf);
+#pragma GCC diagnostic pop
 
        return err ? ERR_PTR(err) : btf;
 }
--

2.30.2

Signature · 1 line
- Arnaldo
#5 lore raw
Quoted 52 lines · Arnaldo Carvalho de Melo

Em Tue, Sep 14, 2021 at 11:28:28AM -0700, Andrii Nakryiko escreveu:

Quoted 26 lines · Jiri Olsa
Quoted 7 lines · Andrii Nakryiko
Quoted 5 lines

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as a weak function, presumably to dynamically link against old version of libbpf shared library. Unfortunately this causes compilation warning when perf is compiled against libbpf v0.6+.

Quoted 9 lines
Quoted 2 lines

For now, just ignore deprecation warning, but there might be a better solution, depending on perf's needs.

HI,
the problem we tried to solve is when perf is using symbols which are not yet available in released libbpf.. but it all linkes in default perf build because it's linked statically libbpf.a in the tree

If you are always statically linking libbpf into perf, there is no need to implement this __weak shim. Libbpf is never going to deprecate an API if a new/replacement API hasn't been at least in a previous released version. So in this case btf__load_from_kernel_by_id() was added in libbpf 0.5, and btf__get_from_id() was marked deprecated in libbpf 0.6 (not yet released, of course). So with that, do you still think we need this __weak re-implementation?

Quoted 4 lines

I was wondering if this was done to make latest perf code compile against some old libbpf source code or dynamically linked against old libbpf. But if that's not the case, the fix should be a removal of __weak btf__load_from_kernel_by_id().

It was made to build against the libbpf that comes with fedora 34, the distro I'm using, which is:

⬢[acme@toolbox perf]$ sudo dnf install libbpf-devel Package libbpf-devel-2:0.4.0-1.fc34.x86_64 is already installed. Dependencies resolved.
Nothing to do.
Complete!
⬢[acme@toolbox perf]$ cat /etc/redhat-release
Fedora release 34 (Thirty Four)

And we have 'make -C tools/perf build-test' that has one entry to build with LIBBPF_EXTERNAL=1, i.e. using whatever libbpf-devel package is installed in the distro, in addtion to statically linking with the libbpf in the kernel sources.

That is done because several distros are linking perf with the libbpf they ship.

Quoted 9 lines

When I merged the latest upstream this test failed, and I realized that some files in tools/perf/ had changed to make use of a new function and that was the reason for the build test failure.

So I tried to provide a transition help for these cases, initially as a feature test that would look if that new function was available and if not, provide the fallback, but then ended up following Jiri's suggestion for a __weak function, as that involved less coding.

Ok, that's cool, then my "fix" should be fine for now. Can you please land it in perf/core to unblock Stephen's (cc'ed) build failure when merging perf and bpf-next trees?

Also it's good to keep in mind that libbpf is now providing LIBBPF_MAJOR_VERSION and LIBBPF_MINOR_VERSION macro, so when statically linking you should be able to use that to detect libbpf version. For shared library cases we should probably also add runtime APIs (e.g., int libbpf_major_version(void), int libbpf_minor_version(void), const char *libbpf_version(void)) so that you can do more detection based on libbpf version at runtime. Let me know if it's something that would be helpful.

Quoted 36 lines

- Arnaldo

Quoted 24 lines
Quoted 24 lines

so now we have weak function with that warning disabled locally, which I guess could work? also for future cases like that

jirka

Quoted 19 lines
---
 tools/perf/util/bpf-event.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/bpf-event.c +3 -0
index 683f6d63560e..1a7112a87736 100644
@@ -24,7 +24,10 @@
 struct btf * __weak btf__load_from_kernel_by_id(__u32 id)
 {
        struct btf *btf;
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
        int err = btf__get_from_id(id, &btf);
+#pragma GCC diagnostic pop
 
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
tools/perf/util/bpf-event.c +0 -1
        return err ? ERR_PTR(err) : btf;
 }
--

2.30.2

--

- Arnaldo

#6 lore raw
Quoted 68 lines · Andrii Nakryiko
Quoted 52 lines · Arnaldo Carvalho de Melo

Em Tue, Sep 14, 2021 at 11:28:28AM -0700, Andrii Nakryiko escreveu:

Quoted 26 lines · Jiri Olsa
Quoted 7 lines · Andrii Nakryiko
Quoted 5 lines

Perf code re-implements libbpf's btf__load_from_kernel_by_id() API as a weak function, presumably to dynamically link against old version of libbpf shared library. Unfortunately this causes compilation warning when perf is compiled against libbpf v0.6+.

Quoted 9 lines
Quoted 2 lines

For now, just ignore deprecation warning, but there might be a better solution, depending on perf's needs.

HI,
the problem we tried to solve is when perf is using symbols which are not yet available in released libbpf.. but it all linkes in default perf build because it's linked statically libbpf.a in the tree

If you are always statically linking libbpf into perf, there is no need to implement this __weak shim. Libbpf is never going to deprecate an API if a new/replacement API hasn't been at least in a previous released version. So in this case btf__load_from_kernel_by_id() was added in libbpf 0.5, and btf__get_from_id() was marked deprecated in libbpf 0.6 (not yet released, of course). So with that, do you still think we need this __weak re-implementation?

Quoted 4 lines

I was wondering if this was done to make latest perf code compile against some old libbpf source code or dynamically linked against old libbpf. But if that's not the case, the fix should be a removal of __weak btf__load_from_kernel_by_id().

It was made to build against the libbpf that comes with fedora 34, the distro I'm using, which is:

⬢[acme@toolbox perf]$ sudo dnf install libbpf-devel Package libbpf-devel-2:0.4.0-1.fc34.x86_64 is already installed. Dependencies resolved.
Nothing to do.
Complete!
⬢[acme@toolbox perf]$ cat /etc/redhat-release
Fedora release 34 (Thirty Four)

And we have 'make -C tools/perf build-test' that has one entry to build with LIBBPF_EXTERNAL=1, i.e. using whatever libbpf-devel package is installed in the distro, in addtion to statically linking with the libbpf in the kernel sources.

That is done because several distros are linking perf with the libbpf they ship.

Quoted 9 lines

When I merged the latest upstream this test failed, and I realized that some files in tools/perf/ had changed to make use of a new function and that was the reason for the build test failure.

So I tried to provide a transition help for these cases, initially as a feature test that would look if that new function was available and if not, provide the fallback, but then ended up following Jiri's suggestion for a __weak function, as that involved less coding.

Ok, that's cool, then my "fix" should be fine for now. Can you please land it in perf/core to unblock Stephen's (cc'ed) build failure when merging perf and bpf-next trees?

Quoted 8 lines

Also it's good to keep in mind that libbpf is now providing LIBBPF_MAJOR_VERSION and LIBBPF_MINOR_VERSION macro, so when statically linking you should be able to use that to detect libbpf version. For shared library cases we should probably also add runtime APIs (e.g., int libbpf_major_version(void), int libbpf_minor_version(void), const char *libbpf_version(void)) so that you can do more detection based on libbpf version at runtime. Let me know if it's something that would be helpful.

Yes, please! We're currently using this horror to be able to print the libbpf version being used by xdp-tools:

https://github.com/xdp-project/xdp-tools/blob/master/lib/util/util.c#L100

Would be awesome to have an API function we could just call instead :)

-Toke