MIME-Version: 1.0
References: <cover.1639432170.git.andreyknvl@google.com> <af3819749624603ed5cb0cbd869d5e4b3ed116b3.1639432170.git.andreyknvl@google.com>
 <Ybj2zms+c6J3J/pf@elver.google.com>
In-Reply-To: <Ybj2zms+c6J3J/pf@elver.google.com>
From:   Andrey Konovalov <andreyknvl@gmail.com>
Date:   Tue, 14 Dec 2021 23:08:49 +0100
Message-ID: <CA+fCnZeY+AEXrPyuWjq9yQ+HOsDxqqp-gw9scvEdLqV5v7q2dA@mail.gmail.com>
Subject: Re: [PATCH mm v3 29/38] kasan, vmalloc: add vmalloc tagging for HW_TAGS
To:     Marco Elver <elver@google.com>
Cc:     andrey.konovalov@linux.dev,
        Alexander Potapenko <glider@google.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Dmitry Vyukov <dvyukov@google.com>,
        Andrey Ryabinin <ryabinin.a.a@gmail.com>,
        kasan-dev <kasan-dev@googlegroups.com>,
        Linux Memory Management List <linux-mm@kvack.org>,
        Vincenzo Frascino <vincenzo.frascino@arm.com>,
        Catalin Marinas <catalin.marinas@arm.com>,
        Will Deacon <will@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Linux ARM <linux-arm-kernel@lists.infradead.org>,
        Peter Collingbourne <pcc@google.com>,
        Evgenii Stepanov <eugenis@google.com>,
        LKML <linux-kernel@vger.kernel.org>,
        Andrey Konovalov <andreyknvl@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Dec 14, 2021 at 8:56 PM Marco Elver <elver@google.com> wrote:
>
> On Mon, Dec 13, 2021 at 10:54PM +0100, andrey.konovalov@linux.dev wrote:
> [...]
> >
> > +     /*
> > +      * Skip page_alloc poisoning and zeroing for pages backing VM_ALLOC
> > +      * mappings. Only effective in HW_TAGS mode.
> > +      */
> > +     gfp &= __GFP_SKIP_KASAN_UNPOISON & __GFP_SKIP_ZERO;
>
> This will turn gfp == 0 always. Should it have been
>
>         gfp |= __GFP_SKIP_KASAN_UNPOISON | __GFP_SKIP_ZERO

Oh, this is bad. Thanks for noticing! Will fix in v4.

> Also, not sure it matters, but on non-KASAN builds, this will now always
> generate an extra instruction. You could conditionally define GFP_SKIP*
> only in the KASAN modes that need them, otherwise they become 0, so the
> compiler optimizes this out. (Although I think it does does complicate
> GFP_SHIFT a little?)

I can implement this, but I don't think a single extra instruction per
vmalloc() matters.