by Dmitry Vyukov

[permalink] [raw]

Subject: Re: [PATCH mm v3 18/19] kasan, mm: allow cache merging with no metadata

On Fri, Nov 13, 2020 at 11:20 PM Andrey Konovalov <[email protected]> wrote:
>
> The reason cache merging is disabled with KASAN is because KASAN puts its
> metadata right after the allocated object. When the merged caches have
> slightly different sizes, the metadata ends up in different places, which
> KASAN doesn't support.
>
> It might be possible to adjust the metadata allocation algorithm and make
> it friendly to the cache merging code. Instead this change takes a simpler
> approach and allows merging caches when no metadata is present. Which is
> the case for hardware tag-based KASAN with kasan.mode=prod.
>
> Co-developed-by: Vincenzo Frascino <[email protected]>
> Signed-off-by: Vincenzo Frascino <[email protected]>
> Signed-off-by: Andrey Konovalov <[email protected]>
> Link: https://linux-review.googlesource.com/id/Ia114847dfb2244f297d2cb82d592bf6a07455dba

Somehow gerrit contains an old version... so I was going to
independently propose what Marco already proposed as simplification...
until I looked at the patch in the email :)

Reviewed-by: Dmitry Vyukov <[email protected]>

> ---
> include/linux/kasan.h | 21 +++++++++++++++++++--
> mm/kasan/common.c | 11 +++++++++++
> mm/slab_common.c | 3 ++-
> 3 files changed, 32 insertions(+), 3 deletions(-)
>
> diff --git a/include/linux/kasan.h b/include/linux/kasan.h
> index 16cf53eac29b..173a8e81d001 100644
> --- a/include/linux/kasan.h
> +++ b/include/linux/kasan.h
> @@ -81,17 +81,30 @@ struct kasan_cache {
> };
>
> #ifdef CONFIG_KASAN_HW_TAGS
> +
> DECLARE_STATIC_KEY_FALSE(kasan_flag_enabled);
> +
> static __always_inline bool kasan_enabled(void)
> {
> return static_branch_likely(&kasan_flag_enabled);
> }
> -#else
> +
> +#else /* CONFIG_KASAN_HW_TAGS */
> +
> static inline bool kasan_enabled(void)
> {
> return true;
> }
> -#endif
> +
> +#endif /* CONFIG_KASAN_HW_TAGS */
> +
> +slab_flags_t __kasan_never_merge(void);
> +static __always_inline slab_flags_t kasan_never_merge(void)
> +{
> + if (kasan_enabled())
> + return __kasan_never_merge();
> + return 0;
> +}
>
> void __kasan_unpoison_range(const void *addr, size_t size);
> static __always_inline void kasan_unpoison_range(const void *addr, size_t size)
> @@ -238,6 +251,10 @@ static inline bool kasan_enabled(void)
> {
> return false;
> }
> +static inline slab_flags_t kasan_never_merge(void)
> +{
> + return 0;
> +}
> static inline void kasan_unpoison_range(const void *address, size_t size) {}
> static inline void kasan_alloc_pages(struct page *page, unsigned int order) {}
> static inline void kasan_free_pages(struct page *page, unsigned int order) {}
> diff --git a/mm/kasan/common.c b/mm/kasan/common.c
> index cf874243efab..a5a4dcb1254d 100644
> --- a/mm/kasan/common.c
> +++ b/mm/kasan/common.c
> @@ -87,6 +87,17 @@ asmlinkage void kasan_unpoison_task_stack_below(const void *watermark)
> }
> #endif /* CONFIG_KASAN_STACK */
>
> +/*
> + * Only allow cache merging when stack collection is disabled and no metadata
> + * is present.
> + */
> +slab_flags_t __kasan_never_merge(void)
> +{
> + if (kasan_stack_collection_enabled())
> + return SLAB_KASAN;
> + return 0;
> +}
> +
> void __kasan_alloc_pages(struct page *page, unsigned int order)
> {
> u8 tag;
> diff --git a/mm/slab_common.c b/mm/slab_common.c
> index 0b5ae1819a8b..075b23ce94ec 100644
> --- a/mm/slab_common.c
> +++ b/mm/slab_common.c
> @@ -19,6 +19,7 @@
> #include <linux/seq_file.h>
> #include <linux/proc_fs.h>
> #include <linux/debugfs.h>
> +#include <linux/kasan.h>
> #include <asm/cacheflush.h>
> #include <asm/tlbflush.h>
> #include <asm/page.h>
> @@ -54,7 +55,7 @@ static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
> */
> #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
> SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
> - SLAB_FAILSLAB | SLAB_KASAN)
> + SLAB_FAILSLAB | kasan_never_merge())
>
> #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
> SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
> --
> 2.29.2.299.gdc1121823c-goog
>

2020-11-23 13:58:00

by Andrey Konovalov

[permalink] [raw]

Subject: Re: [PATCH mm v3 18/19] kasan, mm: allow cache merging with no metadata

On Tue, Nov 17, 2020 at 2:25 PM Dmitry Vyukov <[email protected]> wrote:
>
> On Fri, Nov 13, 2020 at 11:20 PM Andrey Konovalov <[email protected]> wrote:
> >
> > The reason cache merging is disabled with KASAN is because KASAN puts its
> > metadata right after the allocated object. When the merged caches have
> > slightly different sizes, the metadata ends up in different places, which
> > KASAN doesn't support.
> >
> > It might be possible to adjust the metadata allocation algorithm and make
> > it friendly to the cache merging code. Instead this change takes a simpler
> > approach and allows merging caches when no metadata is present. Which is
> > the case for hardware tag-based KASAN with kasan.mode=prod.
> >
> > Co-developed-by: Vincenzo Frascino <[email protected]>
> > Signed-off-by: Vincenzo Frascino <[email protected]>
> > Signed-off-by: Andrey Konovalov <[email protected]>
> > Link: https://linux-review.googlesource.com/id/Ia114847dfb2244f297d2cb82d592bf6a07455dba
>
> Somehow gerrit contains an old version... so I was going to
> independently propose what Marco already proposed as simplification...
> until I looked at the patch in the email :)

Ah, this is because I couldn't push next/mm-based changes into Gerrit
without manually adding tags to all of the yet-out-of-tree patches. So
the Gerrit doesn't have the last version of the patchset.

> Reviewed-by: Dmitry Vyukov <[email protected]>

Thanks!