diff mbox series

[mm] arm64: kasan: fix MTE symbols exports

Message ID dd36936c3d99582a623c8f01345f618ed4c036dd.1612884525.git.andreyknvl@google.com (mailing list archive)
State New, archived
Headers show
Series [mm] arm64: kasan: fix MTE symbols exports | expand

Commit Message

Andrey Konovalov Feb. 9, 2021, 3:32 p.m. UTC
Only export MTE symbols when KASAN-KUnit tests are enabled.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>

---

Andrew, please squash this into:
"arm64: kasan: export MTE symbols for KASAN tests"
---
 arch/arm64/kernel/mte.c | 4 ++++
 1 file changed, 4 insertions(+)

Comments

Catalin Marinas Feb. 9, 2021, 5:02 p.m. UTC | #1
On Tue, Feb 09, 2021 at 04:32:30PM +0100, Andrey Konovalov wrote:
> diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c
> index a66c2806fc4d..788ef0c3a25e 100644
> --- a/arch/arm64/kernel/mte.c
> +++ b/arch/arm64/kernel/mte.c
> @@ -113,13 +113,17 @@ void mte_enable_kernel(void)
>  	sysreg_clear_set(sctlr_el1, SCTLR_ELx_TCF_MASK, SCTLR_ELx_TCF_SYNC);
>  	isb();
>  }
> +#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
>  EXPORT_SYMBOL_GPL(mte_enable_kernel);
> +#endif
>  
>  void mte_set_report_once(bool state)
>  {
>  	WRITE_ONCE(report_fault_once, state);
>  }
> +#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
>  EXPORT_SYMBOL_GPL(mte_set_report_once);
> +#endif

Do we actually care about exporting them when KASAN_KUNIT_TEST=n? It
looks weird to have these #ifdefs in the arch code. Either the
arch-kasan API requires these symbols to be exported to modules or not.
I'm not keen on such kasan internals trickling down into the arch code.

If you don't want to export them in the KASAN_KUNIT_TEST=n case, add a
wrapper in the kasan built-in code (e.g. kasan_test_enable_tagging,
kasan_test_set_report_once) and conditionally compile them based on
KASAN_KUNIT_TEST.
Andrew Morton Feb. 9, 2021, 6:45 p.m. UTC | #2
On Tue, 9 Feb 2021 17:02:56 +0000 Catalin Marinas <catalin.marinas@arm.com> wrote:

> On Tue, Feb 09, 2021 at 04:32:30PM +0100, Andrey Konovalov wrote:
> > diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c
> > index a66c2806fc4d..788ef0c3a25e 100644
> > --- a/arch/arm64/kernel/mte.c
> > +++ b/arch/arm64/kernel/mte.c
> > @@ -113,13 +113,17 @@ void mte_enable_kernel(void)
> >  	sysreg_clear_set(sctlr_el1, SCTLR_ELx_TCF_MASK, SCTLR_ELx_TCF_SYNC);
> >  	isb();
> >  }
> > +#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
> >  EXPORT_SYMBOL_GPL(mte_enable_kernel);
> > +#endif
> >  
> >  void mte_set_report_once(bool state)
> >  {
> >  	WRITE_ONCE(report_fault_once, state);
> >  }
> > +#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
> >  EXPORT_SYMBOL_GPL(mte_set_report_once);
> > +#endif
> 
> Do we actually care about exporting them when KASAN_KUNIT_TEST=n? It
> looks weird to have these #ifdefs in the arch code. Either the
> arch-kasan API requires these symbols to be exported to modules or not.
> I'm not keen on such kasan internals trickling down into the arch code.
> 
> If you don't want to export them in the KASAN_KUNIT_TEST=n case, add a
> wrapper in the kasan built-in code (e.g. kasan_test_enable_tagging,
> kasan_test_set_report_once) and conditionally compile them based on
> KASAN_KUNIT_TEST.

In other words, the patch's changelog was poor!  It told us what the
patch does (which is often obvious from the code) but it failed to
explain why the patch does what it does.

The same goes for code comments, folks - please explain "why it does
this" rather than "what it does".
Andrey Konovalov Feb. 11, 2021, 5:30 p.m. UTC | #3
On Tue, Feb 9, 2021 at 7:45 PM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Tue, 9 Feb 2021 17:02:56 +0000 Catalin Marinas <catalin.marinas@arm.com> wrote:
>
> > On Tue, Feb 09, 2021 at 04:32:30PM +0100, Andrey Konovalov wrote:
> > > diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c
> > > index a66c2806fc4d..788ef0c3a25e 100644
> > > --- a/arch/arm64/kernel/mte.c
> > > +++ b/arch/arm64/kernel/mte.c
> > > @@ -113,13 +113,17 @@ void mte_enable_kernel(void)
> > >     sysreg_clear_set(sctlr_el1, SCTLR_ELx_TCF_MASK, SCTLR_ELx_TCF_SYNC);
> > >     isb();
> > >  }
> > > +#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
> > >  EXPORT_SYMBOL_GPL(mte_enable_kernel);
> > > +#endif
> > >
> > >  void mte_set_report_once(bool state)
> > >  {
> > >     WRITE_ONCE(report_fault_once, state);
> > >  }
> > > +#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
> > >  EXPORT_SYMBOL_GPL(mte_set_report_once);
> > > +#endif
> >
> > Do we actually care about exporting them when KASAN_KUNIT_TEST=n? It
> > looks weird to have these #ifdefs in the arch code. Either the
> > arch-kasan API requires these symbols to be exported to modules or not.
> > I'm not keen on such kasan internals trickling down into the arch code.

Understood.

> > If you don't want to export them in the KASAN_KUNIT_TEST=n case, add a
> > wrapper in the kasan built-in code (e.g. kasan_test_enable_tagging,
> > kasan_test_set_report_once) and conditionally compile them based on
> > KASAN_KUNIT_TEST.

This might be a better approach indeed.

> In other words, the patch's changelog was poor!  It told us what the
> patch does (which is often obvious from the code) but it failed to
> explain why the patch does what it does.
>
> The same goes for code comments, folks - please explain "why it does
> this" rather than "what it does".

I'm sorry, Andrew.

Could you please drop the "arm64: kasan: export MTE symbols for KASAN
tests" patch from the mm tree (but keep the rest of that series)?

I'll post a separate patch with a fix.

Thanks!
diff mbox series

Patch

diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c
index a66c2806fc4d..788ef0c3a25e 100644
--- a/arch/arm64/kernel/mte.c
+++ b/arch/arm64/kernel/mte.c
@@ -113,13 +113,17 @@  void mte_enable_kernel(void)
 	sysreg_clear_set(sctlr_el1, SCTLR_ELx_TCF_MASK, SCTLR_ELx_TCF_SYNC);
 	isb();
 }
+#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
 EXPORT_SYMBOL_GPL(mte_enable_kernel);
+#endif
 
 void mte_set_report_once(bool state)
 {
 	WRITE_ONCE(report_fault_once, state);
 }
+#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
 EXPORT_SYMBOL_GPL(mte_set_report_once);
+#endif
 
 bool mte_report_once(void)
 {