modpost warning by default on missing Module.symvers
Hi Masahiro,
I recently noticed that in commit 4475dff55c54 ("kbuild: fix false-positive modpost warning when all symbols are trimmed") [1] you modified the modpost behavior to always warn (by passing -w) when there are missing Module.symver files in order to allow module builds to continue building with warnings instead of errors. I'm curious why you decided to not continue to rely on KBUILD_MODPOST_WARN to enable/disable that functionality?
I personally find it useful to keep these types of warnings as errors in order to catch missing dependencies at build time (ideally by the CI build) instead of at runtime when a module fails to load due to a missing symbol dependency.
Let me know your thoughts on this and I'll try to come up with a solution to factor in any concerns you have.
Thanks,
Will
Quoted 14 lines · William McVicker
Hi Masahiro,
I recently noticed that in commit
4475dff55c54("kbuild: fix false-positive modpost warning when all symbols are trimmed") [1] you modified the modpost behavior to always warn (by passing-w) when there are missing Module.symver files in order to allow module builds to continue building with warnings instead of errors. I'm curious why you decided to not continue to rely onKBUILD_MODPOST_WARNto enable/disable that functionality?I personally find it useful to keep these types of warnings as errors in order to catch missing dependencies at build time (ideally by the CI build) instead of at runtime when a module fails to load due to a missing symbol dependency.
Quoted 7 lines
Let me know your thoughts on this and I'll try to come up with a solution to factor in any concerns you have.
Thanks,
Will
Good point.
I think we can always require KBUILD_MODPOST_WARN=1 explicitly.
Skipping unresolved symbols is not a good idea. Users can proceed if they want,
but they should be aware of what they are doing, at least.
How about something like this?
scripts/Makefile.modpost +0 -0
@@ -121,16 +121,14 @@ modpost-args += -e $(addprefix -i ,
$(KBUILD_EXTRA_SYMBOLS))
endif # ($(KBUILD_EXTMOD),)
-ifneq ($(missing-input),)
-modpost-args += -w
-endif
-
quiet_cmd_modpost = MODPOST $@
cmd_modpost = \
$(if $(missing-input), \
echo >&2 "WARNING: $(missing-input) is missing."; \
echo >&2 " Modules may not have dependenciesor modversions."; \
- echo >&2 " You may get many unresolved symbol
warnings.";) \
+ echo >&2 " You may get many unresolved symbol
errors.";) \
+ echo >&2 " You can set KBUILD_MODPOST_WARN=1
to turn errors into warning"; \
+ echo >&2 " if you know what you are doing."; \
$(MODPOST) $(modpost-args)
targets += $(output-symdump)
--
Best Regards
Masahiro Yamada
Quoted 78 lines · Masahiro Yamada
Quoted 14 lines · William McVicker
Hi Masahiro,
I recently noticed that in commit
4475dff55c54("kbuild: fix false-positive modpost warning when all symbols are trimmed") [1] you modified the modpost behavior to always warn (by passing-w) when there are missing Module.symver files in order to allow module builds to continue building with warnings instead of errors. I'm curious why you decided to not continue to rely onKBUILD_MODPOST_WARNto enable/disable that functionality?I personally find it useful to keep these types of warnings as errors in order to catch missing dependencies at build time (ideally by the CI build) instead of at runtime when a module fails to load due to a missing symbol dependency.
Quoted 7 lines
Let me know your thoughts on this and I'll try to come up with a solution to factor in any concerns you have.
Thanks,
WillGood point.
I think we can always require KBUILD_MODPOST_WARN=1 explicitly.
Skipping unresolved symbols is not a good idea. Users can proceed if they want,
but they should be aware of what they are doing, at least.How about something like this?
scripts/Makefile.modpost +1
-5@@ -121,16 +121,14 @@ modpost-args += -e $(addprefix -i , $(KBUILD_EXTRA_SYMBOLS)) endif # ($(KBUILD_EXTMOD),) -ifneq ($(missing-input),) -modpost-args += -w -endif - quiet_cmd_modpost = MODPOST $@ cmd_modpost = \ $(if $(missing-input), \ echo >&2 "WARNING: $(missing-input) is missing."; \ echo >&2 " Modules may not have dependencies or modversions."; \ - echo >&2 " You may get many unresolved symbol warnings.";) \ + echo >&2 " You may get many unresolved symbolerrors.";) \
+ echo >&2 " You can set KBUILD_MODPOST_WARN=1
to turn errors into warning"; \
+ echo >&2 " if you know what you are doing."; \$(MODPOST) $(modpost-args)targets += $(output-symdump)
Quoted 4 lines
--
Best RegardsMasahiro Yamada
That looks good to me! You do mention in [1] that there's a case where unresolved symbols are expected. Can you clarify that? Why would you want to build a kernel or module with unresolved symbols?
[1] https://lore.kernel.org/all/[email protected]/
Thanks,
Will
Quoted 89 lines · William McVicker
Quoted 78 lines · Masahiro Yamada
Quoted 14 lines · William McVicker
Hi Masahiro,
I recently noticed that in commit
4475dff55c54("kbuild: fix false-positive modpost warning when all symbols are trimmed") [1] you modified the modpost behavior to always warn (by passing-w) when there are missing Module.symver files in order to allow module builds to continue building with warnings instead of errors. I'm curious why you decided to not continue to rely onKBUILD_MODPOST_WARNto enable/disable that functionality?I personally find it useful to keep these types of warnings as errors in order to catch missing dependencies at build time (ideally by the CI build) instead of at runtime when a module fails to load due to a missing symbol dependency.
Quoted 7 lines
Let me know your thoughts on this and I'll try to come up with a solution to factor in any concerns you have.
Thanks,
WillGood point.
I think we can always require KBUILD_MODPOST_WARN=1 explicitly.
Skipping unresolved symbols is not a good idea. Users can proceed if they want,
but they should be aware of what they are doing, at least.How about something like this?
scripts/Makefile.modpost +1
-5@@ -121,16 +121,14 @@ modpost-args += -e $(addprefix -i , $(KBUILD_EXTRA_SYMBOLS)) endif # ($(KBUILD_EXTMOD),) -ifneq ($(missing-input),) -modpost-args += -w -endif - quiet_cmd_modpost = MODPOST $@ cmd_modpost = \ $(if $(missing-input), \ echo >&2 "WARNING: $(missing-input) is missing."; \ echo >&2 " Modules may not have dependencies or modversions."; \ - echo >&2 " You may get many unresolved symbol warnings.";) \ + echo >&2 " You may get many unresolved symbolerrors.";) \
+ echo >&2 " You can set KBUILD_MODPOST_WARN=1
to turn errors into warning"; \
+ echo >&2 " if you know what you are doing."; \$(MODPOST) $(modpost-args)targets += $(output-symdump)
Quoted 4 lines
--
Best RegardsMasahiro Yamada
That looks good to me! You do mention in [1] that there's a case where unresolved symbols are expected. Can you clarify that? Why would you want to build a kernel or module with unresolved symbols?
Quoted 4 lines
[1] https://lore.kernel.org/all/[email protected]/
Thanks,
Will
What I have in mind is the following cases.
We cannot check unresolved symbols due to missing vmlinux.
[1] Build in-tree modules without building vmlinux
$ make defconfig
$ make modulesPerhaps, this is useful for people who are only interested in particular modules, but not the entire kernel?
[2] Build external modules with minimal setups
$ make defconfig
$ make modules_prepare
$ make M=<path/to/eternal/module>This is useful if people want to compile their modules quicily?
[3] Build single *.ko
$ make defconfig $ make <path/to/a/module>.ko
Perhaps, this is useful for people who are only interested in modules they maintain.
I am not a big fan of any of them, but those have been available since before I became the maintainer.
Signature · 2 lines
Best Regards Masahiro Yamada
Quoted 127 lines · Masahiro Yamada
Quoted 89 lines · William McVicker
Quoted 78 lines · Masahiro Yamada
Quoted 14 lines · William McVicker
Hi Masahiro,
I recently noticed that in commit
4475dff55c54("kbuild: fix false-positive modpost warning when all symbols are trimmed") [1] you modified the modpost behavior to always warn (by passing-w) when there are missing Module.symver files in order to allow module builds to continue building with warnings instead of errors. I'm curious why you decided to not continue to rely onKBUILD_MODPOST_WARNto enable/disable that functionality?I personally find it useful to keep these types of warnings as errors in order to catch missing dependencies at build time (ideally by the CI build) instead of at runtime when a module fails to load due to a missing symbol dependency.
Quoted 7 lines
Let me know your thoughts on this and I'll try to come up with a solution to factor in any concerns you have.
Thanks,
WillGood point.
I think we can always require KBUILD_MODPOST_WARN=1 explicitly.
Skipping unresolved symbols is not a good idea. Users can proceed if they want,
but they should be aware of what they are doing, at least.How about something like this?
scripts/Makefile.modpost +1
-5@@ -121,16 +121,14 @@ modpost-args += -e $(addprefix -i , $(KBUILD_EXTRA_SYMBOLS)) endif # ($(KBUILD_EXTMOD),) -ifneq ($(missing-input),) -modpost-args += -w -endif - quiet_cmd_modpost = MODPOST $@ cmd_modpost = \ $(if $(missing-input), \ echo >&2 "WARNING: $(missing-input) is missing."; \ echo >&2 " Modules may not have dependencies or modversions."; \ - echo >&2 " You may get many unresolved symbol warnings.";) \ + echo >&2 " You may get many unresolved symbolerrors.";) \
+ echo >&2 " You can set KBUILD_MODPOST_WARN=1
to turn errors into warning"; \
+ echo >&2 " if you know what you are doing."; \$(MODPOST) $(modpost-args)targets += $(output-symdump)
Quoted 4 lines
--
Best RegardsMasahiro Yamada
That looks good to me! You do mention in [1] that there's a case where unresolved symbols are expected. Can you clarify that? Why would you want to build a kernel or module with unresolved symbols?
Quoted 4 lines
[1] https://lore.kernel.org/all/[email protected]/
Thanks,
WillWhat I have in mind is the following cases.
We cannot check unresolved symbols due to missing vmlinux.[1] Build in-tree modules without building vmlinux
$ make defconfig $ make modulesPerhaps, this is useful for people who are only interested in particular modules, but not the entire kernel?
[2] Build external modules with minimal setups
$ make defconfig $ make modules_prepare $ make M=<path/to/eternal/module>This is useful if people want to compile their modules quicily?
[3] Build single *.ko
$ make defconfig $ make <path/to/a/module>.ko
Quoted 19 lines
Perhaps, this is useful for people who are only interested in modules they maintain.
I am not a big fan of any of them, but those have been available since before I became the maintainer.
Signature · 5 lines
Best Regards Masahiro Yamada -- To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
Sorry for the delay in getting back to you.
This sounds good to me. I personally always build my modules with the vmlinux so that I can detect at build time if there are any issues, e.g. unresolved symbols. For these 3 cases, would your change require devs to set KBUILD_MODPOST_WARN=1 if they want to ignore the unresolved symbols error?
Thanks,
Will