[PATCH 1/3] KVM: arm64: fix include path for ring buffer implementation

[PATCH 1/3] KVM: arm64: fix include path for ring buffer implementation

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

The simple_ring_buffer.c file is in the source tree rather than
generated at build time, so the include path is wrong when using
separate object trees:

arch/arm64/kvm/hyp/nvhe/trace.c:16:10: fatal error: simple_ring_buffer.c: No such file or directory
   16 | #include "simple_ring_buffer.c"

Include it from the source tree instead.

Fixes: 680a04c333fa ("KVM: arm64: Add tracing capability for the nVHE/pKVM hyp")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 arch/arm64/kvm/hyp/nvhe/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/Makefile b/arch/arm64/kvm/hyp/nvhe/Makefile
index 143d55ec7298..3d33fbefdfc1 100644
--- a/arch/arm64/kvm/hyp/nvhe/Makefile
+++ b/arch/arm64/kvm/hyp/nvhe/Makefile
@@ -33,7 +33,7 @@ hyp-obj-$(CONFIG_NVHE_EL2_TRACING) += clock.o trace.o events.o
 hyp-obj-y += $(lib-objs)
 
 # Path to simple_ring_buffer.c
-CFLAGS_trace.nvhe.o += -I$(objtree)/kernel/trace/
+CFLAGS_trace.nvhe.o += -I$(srctree)/kernel/trace/
 
 ##
 ## Build rules for compiling nVHE hyp code
-- 
2.39.5

[PATCH 2/3] tracing: add more symbols to whitelist

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

Randconfig builds show a number of cryptic build errors from
hitting undefined symbols in simple_ring_buffer.o:

make[7]: *** [/home/arnd/arm-soc/kernel/trace/Makefile:147: kernel/trace/simple_ring_buffer.o.checked] Error 1

These happen with CONFIG_TRACE_BRANCH_PROFILING, CONFIG_KASAN_HW_TAGS,
CONFIG_STACKPROTECTOR, CONFIG_DEBUG_IRQFLAGS and indirectly from WARN_ON().

Add exceptions for each one that I have hit so far on arm64, x86_64 and arm
randconfig builds.

Other architectures likely hit additional ones, so it would be nice
to produce a little more verbose output that include the name of the
missing symbols directly.

Fixes: a717943d8ecc ("tracing: Check for undefined symbols in simple_ring_buffer")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 kernel/trace/Makefile | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index 3182e1bc1cf7..e3f8d6e619d2 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -138,6 +138,8 @@ obj-$(CONFIG_TRACE_REMOTE_TEST) += remote_test.o
 #
 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 += ftrace_likely_update __hwasan_load __hwasan_store __hwasan_tag_memory
+UNDEFINED_ALLOWLIST += warn_bogus_irq_restore warn_slowpath_fmt __stack_chk_guard
 UNDEFINED_ALLOWLIST := $(addprefix -e , $(UNDEFINED_ALLOWLIST))
 
 quiet_cmd_check_undefined = NM      $<
-- 
2.39.5

[PATCH 3/3] KVM: arm64: tracing: add ftrace dependency

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

Selecting CONFIG_TRACE_REMOTE causes a build time warning when FTRACE
is disabled:

WARNING: unmet direct dependencies detected for TRACE_REMOTE
  Depends on [n]: FTRACE [=n]
  Selected by [y]:
  - NVHE_EL2_TRACING [=y] && VIRTUALIZATION [=y] && KVM [=y] && NVHE_EL2_DEBUG [=y] && TRACING [=y]

Add this as another dependency to ensure a clean build.

Fixes: 3aed038aac8d ("KVM: arm64: Add trace remote for the nVHE/pKVM hyp")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 arch/arm64/kvm/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm64/kvm/Kconfig b/arch/arm64/kvm/Kconfig
index 32c7a4f85e75..bcc87f9dad33 100644
--- a/arch/arm64/kvm/Kconfig
+++ b/arch/arm64/kvm/Kconfig
@@ -78,7 +78,7 @@ if NVHE_EL2_DEBUG
 
 config NVHE_EL2_TRACING
 	bool
-	depends on TRACING
+	depends on TRACING && FTRACE
 	select TRACE_REMOTE
 	default y
 
-- 
2.39.5

Re: [PATCH 1/3] KVM: arm64: fix include path for ring buffer implementation

[Vincent Donnefort]

On Thu, Mar 12, 2026 at 01:35:42PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> The simple_ring_buffer.c file is in the source tree rather than
> generated at build time, so the include path is wrong when using
> separate object trees:
> 
> arch/arm64/kvm/hyp/nvhe/trace.c:16:10: fatal error: simple_ring_buffer.c: No such file or directory
>    16 | #include "simple_ring_buffer.c"
> 
> Include it from the source tree instead.
> 
> Fixes: 680a04c333fa ("KVM: arm64: Add tracing capability for the nVHE/pKVM hyp")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

I believe Marc has taken a fix already for that one

https://lore.kernel.org/all/20260311164956.1424119-1-vdonnefort@google.com/

> ---
>  arch/arm64/kvm/hyp/nvhe/Makefile | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/kvm/hyp/nvhe/Makefile b/arch/arm64/kvm/hyp/nvhe/Makefile
> index 143d55ec7298..3d33fbefdfc1 100644
> --- a/arch/arm64/kvm/hyp/nvhe/Makefile
> +++ b/arch/arm64/kvm/hyp/nvhe/Makefile
> @@ -33,7 +33,7 @@ hyp-obj-$(CONFIG_NVHE_EL2_TRACING) += clock.o trace.o events.o
>  hyp-obj-y += $(lib-objs)
>  
>  # Path to simple_ring_buffer.c
> -CFLAGS_trace.nvhe.o += -I$(objtree)/kernel/trace/
> +CFLAGS_trace.nvhe.o += -I$(srctree)/kernel/trace/
>  
>  ##
>  ## Build rules for compiling nVHE hyp code
> -- 
> 2.39.5
> 

Re: [PATCH 2/3] tracing: add more symbols to whitelist

[Vincent Donnefort]

On Thu, Mar 12, 2026 at 01:35:43PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> Randconfig builds show a number of cryptic build errors from
> hitting undefined symbols in simple_ring_buffer.o:
> 
> make[7]: *** [/home/arnd/arm-soc/kernel/trace/Makefile:147: kernel/trace/simple_ring_buffer.o.checked] Error 1
> 
> These happen with CONFIG_TRACE_BRANCH_PROFILING, CONFIG_KASAN_HW_TAGS,
> CONFIG_STACKPROTECTOR, CONFIG_DEBUG_IRQFLAGS and indirectly from WARN_ON().
> 
> Add exceptions for each one that I have hit so far on arm64, x86_64 and arm
> randconfig builds.
> 
> Other architectures likely hit additional ones, so it would be nice
> to produce a little more verbose output that include the name of the
> missing symbols directly.
> 
> Fixes: a717943d8ecc ("tracing: Check for undefined symbols in simple_ring_buffer")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  kernel/trace/Makefile | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
> index 3182e1bc1cf7..e3f8d6e619d2 100644
> --- a/kernel/trace/Makefile
> +++ b/kernel/trace/Makefile
> @@ -138,6 +138,8 @@ obj-$(CONFIG_TRACE_REMOTE_TEST) += remote_test.o
>  #
>  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 += ftrace_likely_update __hwasan_load __hwasan_store __hwasan_tag_memory
> +UNDEFINED_ALLOWLIST += warn_bogus_irq_restore warn_slowpath_fmt __stack_chk_guard
>  UNDEFINED_ALLOWLIST := $(addprefix -e , $(UNDEFINED_ALLOWLIST))
>  
>  quiet_cmd_check_undefined = NM      $<
> -- 
> 2.39.5
> 

Thanks for the patch.

Most of those ones are covered here
https://lore.kernel.org/all/20260312113535.2213350-1-vdonnefort@google.com/.

This should fix allmodconfig.

For the ones not covered, I'm working on a follow-up patch using a different
approach which should fix them. I'll be able to share it today or tomorrow. Not
sure if we want to take this temporarily?

Re: [PATCH 3/3] KVM: arm64: tracing: add ftrace dependency

[Vincent Donnefort]

On Thu, Mar 12, 2026 at 01:35:44PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> Selecting CONFIG_TRACE_REMOTE causes a build time warning when FTRACE
> is disabled:
> 
> WARNING: unmet direct dependencies detected for TRACE_REMOTE
>   Depends on [n]: FTRACE [=n]
>   Selected by [y]:
>   - NVHE_EL2_TRACING [=y] && VIRTUALIZATION [=y] && KVM [=y] && NVHE_EL2_DEBUG [=y] && TRACING [=y]
> 
> Add this as another dependency to ensure a clean build.
> 
> Fixes: 3aed038aac8d ("KVM: arm64: Add trace remote for the nVHE/pKVM hyp")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  arch/arm64/kvm/Kconfig | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/kvm/Kconfig b/arch/arm64/kvm/Kconfig
> index 32c7a4f85e75..bcc87f9dad33 100644
> --- a/arch/arm64/kvm/Kconfig
> +++ b/arch/arm64/kvm/Kconfig
> @@ -78,7 +78,7 @@ if NVHE_EL2_DEBUG
>  
>  config NVHE_EL2_TRACING
>  	bool
> -	depends on TRACING
> +	depends on TRACING && FTRACE
>  	select TRACE_REMOTE
>  	default y

Reviewed-by: Vincent Donnefort <vdonnefort@google.com>

>  
> -- 
> 2.39.5
> 

Re: [PATCH 2/3] tracing: add more symbols to whitelist

[Arnd Bergmann]

On Thu, Mar 12, 2026, at 14:40, Vincent Donnefort wrote:
> On Thu, Mar 12, 2026 at 01:35:43PM +0100, Arnd Bergmann wrote:
>> diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile

>> +UNDEFINED_ALLOWLIST += ftrace_likely_update __hwasan_load __hwasan_store __hwasan_tag_memory
>> +UNDEFINED_ALLOWLIST += warn_bogus_irq_restore warn_slowpath_fmt __stack_chk_guard
 
>
> Thanks for the patch.
>
> Most of those ones are covered here
> https://lore.kernel.org/all/20260312113535.2213350-1-vdonnefort@google.com/.
>
> This should fix allmodconfig.

The ones I see in that patch look entirely different from the ones
I found.

> For the ones not covered, I'm working on a follow-up patch using a different
> approach which should fix them. I'll be able to share it today or tomorrow. Not
> sure if we want to take this temporarily?

I can wait, my randconfig setup has my patch applied locally for now,
but I've only fixed the ones I saw on three architectures with gcc-16.0.1,
while Nathan's list seems to be mostly for clang specific symbols.

A lot of them also don't show up in allmodconfig builds, only after a
larger number of randconfig configurations.

      Arnd

Re: [PATCH 2/3] tracing: add more symbols to whitelist

[Marc Zyngier]

On Thu, 12 Mar 2026 14:37:50 +0000,
"Arnd Bergmann" <arnd@arndb.de> wrote:
> 
> On Thu, Mar 12, 2026, at 14:40, Vincent Donnefort wrote:
> > On Thu, Mar 12, 2026 at 01:35:43PM +0100, Arnd Bergmann wrote:
> >> diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
> 
> >> +UNDEFINED_ALLOWLIST += ftrace_likely_update __hwasan_load __hwasan_store __hwasan_tag_memory
> >> +UNDEFINED_ALLOWLIST += warn_bogus_irq_restore warn_slowpath_fmt __stack_chk_guard
>  
> >
> > Thanks for the patch.
> >
> > Most of those ones are covered here
> > https://lore.kernel.org/all/20260312113535.2213350-1-vdonnefort@google.com/.
> >
> > This should fix allmodconfig.
> 
> The ones I see in that patch look entirely different from the ones
> I found.
> 
> > For the ones not covered, I'm working on a follow-up patch using a different
> > approach which should fix them. I'll be able to share it today or tomorrow. Not
> > sure if we want to take this temporarily?
> 
> I can wait, my randconfig setup has my patch applied locally for now,
> but I've only fixed the ones I saw on three architectures with gcc-16.0.1,
> while Nathan's list seems to be mostly for clang specific symbols.
> 
> A lot of them also don't show up in allmodconfig builds, only after a
> larger number of randconfig configurations.

I've decided to take this patch anyway, on top of Vincent's. Should a
more generic solution arise, I can always drop both.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.

Re: [PATCH 2/3] tracing: add more symbols to whitelist

[Arnd Bergmann]

On Thu, Mar 12, 2026, at 16:21, Marc Zyngier wrote:
>
> I've decided to take this patch anyway, on top of Vincent's. Should a
> more generic solution arise, I can always drop both.
>

Ok, thanks!

FWIW, I came across two additional ones by now, and will keep
running more tests to see what else comes up:

--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -140,6 +140,7 @@ UNDEFINED_ALLOWLIST := memset alt_cb_patch_nops __x86 __ubsan __asan __kasan __g
 UNDEFINED_ALLOWLIST += __stack_chk_fail stackleak_track_stack __ref_stack __sanitizer
 UNDEFINED_ALLOWLIST += ftrace_likely_update __hwasan_load __hwasan_store __hwasan_tag_memory
 UNDEFINED_ALLOWLIST += warn_bogus_irq_restore warn_slowpath_fmt __stack_chk_guard
+UNDEFINED_ALLOWLIST += __kcsan __tsan
 UNDEFINED_ALLOWLIST := $(addprefix -e , $(UNDEFINED_ALLOWLIST))
 
 quiet_cmd_check_undefined = NM      $<


      Arnd

Re: (subset) [PATCH 1/3] KVM: arm64: fix include path for ring buffer implementation

[Marc Zyngier]

On Thu, 12 Mar 2026 13:35:42 +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> The simple_ring_buffer.c file is in the source tree rather than
> generated at build time, so the include path is wrong when using
> separate object trees:
> 
> arch/arm64/kvm/hyp/nvhe/trace.c:16:10: fatal error: simple_ring_buffer.c: No such file or directory
>    16 | #include "simple_ring_buffer.c"
> 
> [...]

Applied to next, thanks!

[2/3] tracing: add more symbols to whitelist
      commit: 7e4b6c94300e355a72670c5b896ccc26ac312c63
[3/3] KVM: arm64: tracing: add ftrace dependency
      commit: 9e5dd49de5d82401e92098c03e0a0e978ddd515a

Cheers,

	M.
-- 
Without deviation from the norm, progress is not possible.

Re: (subset) [PATCH 1/3] KVM: arm64: fix include path for ring buffer implementation

[Nathan Chancellor]

On Thu, Mar 12, 2026 at 03:45:26PM +0000, Marc Zyngier wrote:
> Applied to next, thanks!
> 
> [2/3] tracing: add more symbols to whitelist
>       commit: 7e4b6c94300e355a72670c5b896ccc26ac312c63

This dropped the SPDX indentifier line in kernel/trace/Makefile.

Cheers,
Nathan