Currently, building KASAN-KUnit tests as a module fails with:
ERROR: modpost: "mte_enable_kernel" [lib/test_kasan.ko] undefined!
ERROR: modpost: "mte_set_report_once" [lib/test_kasan.ko] undefined!
This change adds KASAN wrappers for mte_enable_kernel() and
mte_set_report_once() and only defines and exports them when KASAN-KUnit
tests are enabled.
The wrappers aren't defined when tests aren't enabled to avoid misuse.
The mte_() functions aren't exported directly to avoid having low-level
KASAN ifdefs in the arch code.
Signed-off-by: Andrey Konovalov <[email protected]>
---
Changes v1->v2:
- Add wrappers instead of exporting MTE symbols directly.
- Only define and export wrappers when KASAN-KUnit tests are enabled.
---
lib/test_kasan.c | 6 +++---
mm/kasan/hw_tags.c | 16 ++++++++++++++++
mm/kasan/kasan.h | 12 ++++++++++++
3 files changed, 31 insertions(+), 3 deletions(-)
diff --git a/lib/test_kasan.c b/lib/test_kasan.c
index 1328c468fdb5..e5647d147b35 100644
--- a/lib/test_kasan.c
+++ b/lib/test_kasan.c
@@ -53,13 +53,13 @@ static int kasan_test_init(struct kunit *test)
}
multishot = kasan_save_enable_multi_shot();
- hw_set_tagging_report_once(false);
+ kasan_set_tagging_report_once(false);
return 0;
}
static void kasan_test_exit(struct kunit *test)
{
- hw_set_tagging_report_once(true);
+ kasan_set_tagging_report_once(true);
kasan_restore_multi_shot(multishot);
}
@@ -97,7 +97,7 @@ static void kasan_test_exit(struct kunit *test)
READ_ONCE(fail_data.report_found)); \
if (IS_ENABLED(CONFIG_KASAN_HW_TAGS)) { \
if (READ_ONCE(fail_data.report_found)) \
- hw_enable_tagging(); \
+ kasan_enable_tagging(); \
migrate_enable(); \
} \
} while (0)
diff --git a/mm/kasan/hw_tags.c b/mm/kasan/hw_tags.c
index 1dfe4f62a89e..2aad21fda156 100644
--- a/mm/kasan/hw_tags.c
+++ b/mm/kasan/hw_tags.c
@@ -185,3 +185,19 @@ struct kasan_track *kasan_get_free_track(struct kmem_cache *cache,
return &alloc_meta->free_track[0];
}
+
+#if IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
+
+void kasan_set_tagging_report_once(bool state)
+{
+ hw_set_tagging_report_once(state);
+}
+EXPORT_SYMBOL_GPL(kasan_set_tagging_report_once);
+
+void kasan_enable_tagging(void)
+{
+ hw_enable_tagging();
+}
+EXPORT_SYMBOL_GPL(kasan_enable_tagging);
+
+#endif
diff --git a/mm/kasan/kasan.h b/mm/kasan/kasan.h
index cc787ba47e1b..3436c6bf7c0c 100644
--- a/mm/kasan/kasan.h
+++ b/mm/kasan/kasan.h
@@ -308,6 +308,18 @@ static inline const void *arch_kasan_set_tag(const void *addr, u8 tag)
#endif /* CONFIG_KASAN_HW_TAGS */
+#if defined(CONFIG_KASAN_HW_TAGS) && IS_ENABLED(CONFIG_KASAN_KUNIT_TEST)
+
+void kasan_set_tagging_report_once(bool state);
+void kasan_enable_tagging(void);
+
+#else /* CONFIG_KASAN_HW_TAGS || CONFIG_KASAN_KUNIT_TEST */
+
+static inline void kasan_set_tagging_report_once(bool state) { }
+static inline void kasan_enable_tagging(void) { }
+
+#endif /* CONFIG_KASAN_HW_TAGS || CONFIG_KASAN_KUNIT_TEST */
+
#ifdef CONFIG_KASAN_SW_TAGS
u8 kasan_random_tag(void);
#elif defined(CONFIG_KASAN_HW_TAGS)
--
2.30.0.478.g8a0d178c01-goog
On Fri, 12 Feb 2021 21:08:52 +0100 Andrey Konovalov <[email protected]> wrote:
> Currently, building KASAN-KUnit tests as a module fails with:
>
> ERROR: modpost: "mte_enable_kernel" [lib/test_kasan.ko] undefined!
> ERROR: modpost: "mte_set_report_once" [lib/test_kasan.ko] undefined!
>
> This change adds KASAN wrappers for mte_enable_kernel() and
> mte_set_report_once() and only defines and exports them when KASAN-KUnit
> tests are enabled.
>
> The wrappers aren't defined when tests aren't enabled to avoid misuse.
> The mte_() functions aren't exported directly to avoid having low-level
> KASAN ifdefs in the arch code.
>
Please confirm that this is applicable to current Linus mainline?
Today is pretty much the last day for getting material into 5.11, and
this patch has been churning somewhat.
So I think it would be better to merge this into 5.12-rc1, with a
cc:stable so it goes into 5.11.1.
For which we'll need a Fixes:, please?
On Fri, Feb 12, 2021 at 9:16 PM Andrew Morton <[email protected]> wrote:
>
> On Fri, 12 Feb 2021 21:08:52 +0100 Andrey Konovalov <[email protected]> wrote:
>
> > Currently, building KASAN-KUnit tests as a module fails with:
> >
> > ERROR: modpost: "mte_enable_kernel" [lib/test_kasan.ko] undefined!
> > ERROR: modpost: "mte_set_report_once" [lib/test_kasan.ko] undefined!
> >
> > This change adds KASAN wrappers for mte_enable_kernel() and
> > mte_set_report_once() and only defines and exports them when KASAN-KUnit
> > tests are enabled.
> >
> > The wrappers aren't defined when tests aren't enabled to avoid misuse.
> > The mte_() functions aren't exported directly to avoid having low-level
> > KASAN ifdefs in the arch code.
> >
>
> Please confirm that this is applicable to current Linus mainline?
It's not applicable. KUnit tests for HW_TAGS aren't supported there,
the patches for that are in mm only. So no need to put it into 5.11.
> Today is pretty much the last day for getting material into 5.11, and
> this patch has been churning somewhat.
>
> So I think it would be better to merge this into 5.12-rc1, with a
> cc:stable so it goes into 5.11.1.
>
> For which we'll need a Fixes:, please?
>
> --
> You received this message because you are subscribed to the Google Groups "kasan-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
> To view this discussion on the web visit https://groups.google.com/d/msgid/kasan-dev/20210212121610.ff05a7bb37f97caef97dc924%40linux-foundation.org.
On Fri, 12 Feb 2021 21:21:39 +0100 Andrey Konovalov <[email protected]> wrote:
> > > The wrappers aren't defined when tests aren't enabled to avoid misuse.
> > > The mte_() functions aren't exported directly to avoid having low-level
> > > KASAN ifdefs in the arch code.
> > >
> >
> > Please confirm that this is applicable to current Linus mainline?
>
> It's not applicable. KUnit tests for HW_TAGS aren't supported there,
> the patches for that are in mm only. So no need to put it into 5.11.
So... which -mm patch does this patch fix?
On Fri, Feb 12, 2021 at 9:54 PM Andrew Morton <[email protected]> wrote:
>
> On Fri, 12 Feb 2021 21:21:39 +0100 Andrey Konovalov <[email protected]> wrote:
>
> > > > The wrappers aren't defined when tests aren't enabled to avoid misuse.
> > > > The mte_() functions aren't exported directly to avoid having low-level
> > > > KASAN ifdefs in the arch code.
> > > >
> > >
> > > Please confirm that this is applicable to current Linus mainline?
> >
> > It's not applicable. KUnit tests for HW_TAGS aren't supported there,
> > the patches for that are in mm only. So no need to put it into 5.11.
>
> So... which -mm patch does this patch fix?
"kasan, arm64: allow using KUnit tests with HW_TAGS mode".
There will be some minor adjacent-line-changed conflicts if you decide
to squash it.
Alternatively, this can go as a separate patch after the tests series
(after "kasan: don't run tests when KASAN is not enabled").
Thanks!
On Fri, 12 Feb 2021 22:01:38 +0100 Andrey Konovalov <[email protected]> wrote:
> On Fri, Feb 12, 2021 at 9:54 PM Andrew Morton <[email protected]> wrote:
> >
> > On Fri, 12 Feb 2021 21:21:39 +0100 Andrey Konovalov <[email protected]> wrote:
> >
> > > > > The wrappers aren't defined when tests aren't enabled to avoid misuse.
> > > > > The mte_() functions aren't exported directly to avoid having low-level
> > > > > KASAN ifdefs in the arch code.
> > > > >
> > > >
> > > > Please confirm that this is applicable to current Linus mainline?
> > >
> > > It's not applicable. KUnit tests for HW_TAGS aren't supported there,
> > > the patches for that are in mm only. So no need to put it into 5.11.
> >
> > So... which -mm patch does this patch fix?
>
> "kasan, arm64: allow using KUnit tests with HW_TAGS mode".
>
> There will be some minor adjacent-line-changed conflicts if you decide
> to squash it.
>
> Alternatively, this can go as a separate patch after the tests series
> (after "kasan: don't run tests when KASAN is not enabled").
Thanks - it wasn't obvious.
I staged it as a fix against "kasan, arm64: allow using KUnit tests
with HW_TAGS mode". To make the series as nice as we can, and to avoid
bisection holes.
On Fri, Feb 12, 2021 at 10:08 PM Andrew Morton
<[email protected]> wrote:
>
> On Fri, 12 Feb 2021 22:01:38 +0100 Andrey Konovalov <[email protected]> wrote:
>
> > On Fri, Feb 12, 2021 at 9:54 PM Andrew Morton <[email protected]> wrote:
> > >
> > > On Fri, 12 Feb 2021 21:21:39 +0100 Andrey Konovalov <[email protected]> wrote:
> > >
> > > > > > The wrappers aren't defined when tests aren't enabled to avoid misuse.
> > > > > > The mte_() functions aren't exported directly to avoid having low-level
> > > > > > KASAN ifdefs in the arch code.
> > > > > >
> > > > >
> > > > > Please confirm that this is applicable to current Linus mainline?
> > > >
> > > > It's not applicable. KUnit tests for HW_TAGS aren't supported there,
> > > > the patches for that are in mm only. So no need to put it into 5.11.
> > >
> > > So... which -mm patch does this patch fix?
> >
> > "kasan, arm64: allow using KUnit tests with HW_TAGS mode".
> >
> > There will be some minor adjacent-line-changed conflicts if you decide
> > to squash it.
> >
> > Alternatively, this can go as a separate patch after the tests series
> > (after "kasan: don't run tests when KASAN is not enabled").
>
> Thanks - it wasn't obvious.
>
> I staged it as a fix against "kasan, arm64: allow using KUnit tests
> with HW_TAGS mode". To make the series as nice as we can, and to avoid
> bisection holes.
Sounds good, thank you!