Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 15 Jun 2022 19:58:28 +0300
From:   "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
To:     "Edgecombe, Rick P" <rick.p.edgecombe@intel.com>
Cc:     "peterz@infradead.org" <peterz@infradead.org>,
        "Lutomirski, Andy" <luto@kernel.org>,
        "dave.hansen@linux.intel.com" <dave.hansen@linux.intel.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "hjl.tools@gmail.com" <hjl.tools@gmail.com>,
        "linux-mm@kvack.org" <linux-mm@kvack.org>,
        "kcc@google.com" <kcc@google.com>,
        "andreyknvl@gmail.com" <andreyknvl@gmail.com>,
        "ak@linux.intel.com" <ak@linux.intel.com>,
        "dvyukov@google.com" <dvyukov@google.com>,
        "x86@kernel.org" <x86@kernel.org>,
        "ryabinin.a.a@gmail.com" <ryabinin.a.a@gmail.com>,
        "glider@google.com" <glider@google.com>
Subject: Re: [PATCHv3 5/8] x86/uaccess: Provide untagged_addr() and remove
 tags before address check
Message-ID: <20220615165828.5ggwnoxo7zhvmqzt@black.fi.intel.com>
References: <20220610143527.22974-1-kirill.shutemov@linux.intel.com>
 <20220610143527.22974-6-kirill.shutemov@linux.intel.com>
 <c7e2f8fb44da067e7565d091edeac300977b65ed.camel@intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <c7e2f8fb44da067e7565d091edeac300977b65ed.camel@intel.com>
Precedence: bulk

On Mon, Jun 13, 2022 at 05:36:43PM +0000, Edgecombe, Rick P wrote:
> On Fri, 2022-06-10 at 17:35 +0300, Kirill A. Shutemov wrote:
> > +#ifdef CONFIG_X86_64
> > +/*
> > + * Mask out tag bits from the address.
> > + *
> > + * Magic with the 'sign' allows to untag userspace pointer without
> > any branches
> > + * while leaving kernel addresses intact.
> 
> Trying to understand the magic part here. I guess how it works is, when
> the high bit is set, it does the opposite of untagging the addresses by
> setting the tag bits instead of clearing them. So:
>  - For proper canonical kernel addresses (with U57) it leaves them 
>    intact since the tag bits were already set.
>  - For non-canonical kernel-half addresses, it fixes them up. 
>    (0xeffffff000000840->0xfffffff000000840)
>  - For U48 and 5 level paging, it corrupts some normal kernel 
>    addresses. (0xff90ffffffffffff->0xffffffffffffffff)
> 
> I just ported this to userspace and threw some addresses at it to see
> what happened, so hopefully I got that right.

Ouch. Thanks for noticing this. I should have catched this myself. Yes,
this implementation is broken for LAM_U48 on 5-level machine.

What about this:

	#define untagged_addr(mm, addr)	({					\
		u64 __addr = (__force u64)(addr);				\
		s64 sign = (s64)__addr >> 63;					\
		__addr &= (mm)->context.untag_mask | sign;			\
		(__force __typeof__(addr))__addr;				\
	})

It makes mask effectively. all-ones for supervisor addresses. And it is
less magic to my eyes.

The generated code also look sane to me:

    11d0:	48 89 f8             	mov    %rdi,%rax
    11d3:	48 c1 f8 3f          	sar    $0x3f,%rax
    11d7:	48 0b 05 52 2e 00 00 	or     0x2e52(%rip),%rax        # 4030 <untag_mask>
    11de:	48 21 f8             	and    %rdi,%rax

Any comments?

> Is this special kernel address handling only needed because
> copy_to_kernel_nofault(), etc call the user helpers?

I did not have any particular use-case in mind. But just if some kernel
address gets there and bits get cleared we will have very hard to debug
bug.

-- 
 Kirill A. Shutemov