netdev - Re: [PATCH net-next] ipv6: Implement optimized IPv6 masked address comparison for ARM64

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Fri, 17 Mar 2017 15:20:06 -0600
From:   Subash Abhinov Kasiviswanathan <subashab@...eaurora.org>
To:     James Greenhalgh <james.greenhalgh@....com>,
        Robin Murphy <robin.murphy@....com>
Cc:     netdev@...r.kernel.org, davem@...emloft.net,
        luke.starrett@...adcom.com, catalin.marinas@....com, nd@....com,
        netdev-owner@...r.kernel.org
Subject: Re: [PATCH net-next] ipv6: Implement optimized IPv6 masked address
 comparison for ARM64

> 
>> That's clearly not right - I'm not sure quite what undefined behaviour
>> assumption convinces GCC to optimise the whole thing away>
> 
> While the pointer casting is a bit ghastly, I don't actually think that
> GCC is taking advantage of undefined behaviour here, rather it looks 
> like
> you have a simple typo on line 3:
> 
>> 	const __uint128_t *ul1 = (const __uint128_t *)a1;
>> 	const __uint128_t *ulm = (const __uint128_t *)m;
>> 	const __uint128_t *ul2 = (const __uint128_t *)a1;
> 
> ul2 = a2, surely?
> 
> As it is (stripping casts) you have a1 ^ a1, which will get you to 0
> pretty quickly. Fixing that up for you;
> 
>   bool
>   ipv6_masked_addr_cmp_new(const struct in6_addr *a1, const struct
>   in6_addr *m,
>   		     const struct in6_addr *a2)
>   {
>   	const __uint128_t *ul1 = (const __uint128_t *)a1;
>   	const __uint128_t *ulm = (const __uint128_t *)m;
>   	const __uint128_t *ul2 = (const __uint128_t *)a2;
> 
>   	return !!((*ul1 ^ *ul2) & *ulm);
>   }
> 
> $ gcc -O2
> 
>   ipv6_masked_addr_cmp_new:
> 	ldp	x4, x3, [x0]
> 	ldp	x5, x2, [x2]
> 	ldp	x0, x1, [x1]
> 	eor	x4, x4, x5
> 	eor	x2, x3, x2
> 	and	x0, x0, x4
> 	and	x1, x1, x2
> 	orr	x0, x0, x1
> 	cmp	x0, 0
> 	cset	w0, ne
> 	ret
> 
> Which at least looks like it might calculate something useful :-)
> 
Hi Robin / James

Thanks for checking and sorry for the confusion. I'll retest this.

--
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a 
Linux Foundation Collaborative Project