Re: [PATCH v14 18/30] tracing: Check for undefined symbols in simple_ring_buffer

From: Vincent Donnefort

Date: Thu Mar 12 2026 - 04:55:25 EST


On Wed, Mar 11, 2026 at 03:18:16PM -0700, Nathan Chancellor wrote:
> Hi Vincent,
>
> On Mon, Mar 09, 2026 at 04:25:04PM +0000, Vincent Donnefort wrote:
> > The simple_ring_buffer implementation must remain simple enough to be
> > used by the pKVM hypervisor. Prevent the object build if unresolved
> > symbols are found.
> >
> > Reviewed-by: Steven Rostedt (Google) <rostedt@xxxxxxxxxxx>
> > Signed-off-by: Vincent Donnefort <vdonnefort@xxxxxxxxxx>
> >
> > diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
> > index d106beca8d7f..3182e1bc1cf7 100644
> > --- a/kernel/trace/Makefile
> > +++ b/kernel/trace/Makefile
> > @@ -132,4 +132,20 @@ obj-$(CONFIG_TRACE_REMOTE) += trace_remote.o
> > obj-$(CONFIG_SIMPLE_RING_BUFFER) += simple_ring_buffer.o
> > obj-$(CONFIG_TRACE_REMOTE_TEST) += remote_test.o
> >
> > +#
> > +# simple_ring_buffer is used by the pKVM hypervisor which does not have access
> > +# to all kernel symbols. Fail the build if forbidden symbols are found.
> > +#
> > +UNDEFINED_ALLOWLIST := memset alt_cb_patch_nops __x86 __ubsan __asan __kasan __gcov __aeabi_unwind
> > +UNDEFINED_ALLOWLIST += __stack_chk_fail stackleak_track_stack __ref_stack __sanitizer
> > +UNDEFINED_ALLOWLIST := $(addprefix -e , $(UNDEFINED_ALLOWLIST))
> > +
> > +quiet_cmd_check_undefined = NM $<
> > + cmd_check_undefined = test -z "`$(NM) -u $< | grep -v $(UNDEFINED_ALLOWLIST)`"
>
> This check triggers when building allmodconfig targeting arm, arm64,
> powerpc, and x86_64 (at least, I did not test more at the moment) with
> clang. If this is a hard failure, this really needs to print something
> out to the developer/user to help them debug off the bat, versus having
> to manually dig the $(NM) command out from the .cmd file or V=1. I came
> up with
>
> diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
> index 3182e1bc1cf7..c725b06876bc 100644
> --- a/kernel/trace/Makefile
> +++ b/kernel/trace/Makefile
> @@ -141,7 +141,13 @@ UNDEFINED_ALLOWLIST += __stack_chk_fail stackleak_track_stack __ref_stack __sani
> UNDEFINED_ALLOWLIST := $(addprefix -e , $(UNDEFINED_ALLOWLIST))
>
> quiet_cmd_check_undefined = NM $<
> - cmd_check_undefined = test -z "`$(NM) -u $< | grep -v $(UNDEFINED_ALLOWLIST)`"
> + cmd_check_undefined = \
> + undefsyms=$$($(NM) -u $< | grep -v $(UNDEFINED_ALLOWLIST) || true); \
> + if [ -n "$$undefsyms" ]; then \
> + echo "Unexpected symbols in $<:" >&2; \
> + echo "$$undefsyms" >&2; \
> + false; \
> + fi
>
> $(obj)/%.o.checked: $(obj)/%.o FORCE
> $(call if_changed,check_undefined)
> --
>
> which prints
>
> Unexpected symbols in kernel/trace/simple_ring_buffer.o:
> U llvm_gcda_emit_arcs
> U llvm_gcda_emit_function
> U llvm_gcda_end_file
> U llvm_gcda_start_file
> U llvm_gcda_summary_info
> U llvm_gcov_init
>
> for arm64, which makes sense since these are LLVM specific GCOV symbols,
> so they should probably get the same treatment as the other ones:
>
> diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
> index c725b06876bc..d464e3aa5bdd 100644
> --- a/kernel/trace/Makefile
> +++ b/kernel/trace/Makefile
> @@ -136,8 +136,8 @@ obj-$(CONFIG_TRACE_REMOTE_TEST) += remote_test.o
> # simple_ring_buffer is used by the pKVM hypervisor which does not have access
> # to all kernel symbols. Fail the build if forbidden symbols are found.
> #
> -UNDEFINED_ALLOWLIST := memset alt_cb_patch_nops __x86 __ubsan __asan __kasan __gcov __aeabi_unwind
> -UNDEFINED_ALLOWLIST += __stack_chk_fail stackleak_track_stack __ref_stack __sanitizer
> +UNDEFINED_ALLOWLIST := memset alt_cb_patch_nops __x86 __ubsan __asan __kasan __gcov llvm_gcda llvm_gcov
> +UNDEFINED_ALLOWLIST += __aeabi_unwind __stack_chk_fail stackleak_track_stack __ref_stack __sanitizer
> UNDEFINED_ALLOWLIST := $(addprefix -e , $(UNDEFINED_ALLOWLIST))
>
> quiet_cmd_check_undefined = NM $<
> --
>
> For x86_64, I see
>
> Unexpected symbols in kernel/trace/simple_ring_buffer.o:
> U __clear_pages_unrolled
> U __memmove
> U copy_page
>
> which comes from the use of KCFI_ADDRESSABLE(), since allmodconfig has
> CONFIG_CFI=y.
>
> For powerpc (with both clang and GCC), I see
>
> Unexpected symbols in kernel/trace/simple_ring_buffer.o:
> U .TOC.
>
> For arm (with both clang and GCC), I see
>
> Unexpected symbols in kernel/trace/simple_ring_buffer.o:
> U __stack_chk_guard
> U warn_slowpath_fmt
>
> Presumably adding all of those should be fine as well?
>
> diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
> index d464e3aa5bdd..4f120cb8c79c 100644
> --- a/kernel/trace/Makefile
> +++ b/kernel/trace/Makefile
> @@ -137,7 +137,8 @@ obj-$(CONFIG_TRACE_REMOTE_TEST) += remote_test.o
> # to all kernel symbols. Fail the build if forbidden symbols are found.
> #
> UNDEFINED_ALLOWLIST := memset alt_cb_patch_nops __x86 __ubsan __asan __kasan __gcov llvm_gcda llvm_gcov
> -UNDEFINED_ALLOWLIST += __aeabi_unwind __stack_chk_fail stackleak_track_stack __ref_stack __sanitizer
> +UNDEFINED_ALLOWLIST += __aeabi_unwind __stack_chk_fail __stack_chk_guard stackleak_track_stack __ref_stack __sanitizer
> +UNDEFINED_ALLOWLIST += \.TOC\. __clear_pages_unrolled __memmove copy_page warn_slowpath_fmt
> UNDEFINED_ALLOWLIST := $(addprefix -e , $(UNDEFINED_ALLOWLIST))
>
> quiet_cmd_check_undefined = NM $<
> --
>
> I don't mind sending a series for these, I just wanted to make sure I
> was reasoning about everything correctly.

Yes this is all fine. If you have something already, please send it. Meanwhile, I'll try
something a bit more durable.

Thanks,
Vincent

>
> Cheers,
> Nathan