netdev - Re: [PATCH net-next 2/2] gianfar: convert u64 status counters to atomic64

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:	Wed, 13 Feb 2013 10:03:44 -0800
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Claudiu Manoil <claudiu.manoil@...escale.com>
Cc:	Paul Gortmaker <paul.gortmaker@...driver.com>,
	David Miller <davem@...emloft.net>, netdev@...r.kernel.org
Subject: Re: [PATCH net-next 2/2] gianfar: convert u64 status counters to
 atomic64_t

On Wed, 2013-02-13 at 19:47 +0200, Claudiu Manoil wrote:
> On 2/13/2013 6:14 PM, Eric Dumazet wrote:
> > On Wed, 2013-02-13 at 17:22 +0200, Claudiu Manoil wrote:
> >
> >> At least it seems that this conversion results in fewer asm
> >> instructions, as apparently addze and the double lwz/stw are
> >> not generated anymore. Hopefully it's faster too :P
> >
> > Strange, could you show us these asm instructions ?
> >
> >
> >
> 
> Ok, I'm looking over gfar_clean_rx_ring's asm code, with and w/o this
> patch. They are difficult to compare as asm code changed considerably,
> the initial version having more lines.
> The first thing I notice is that the initial ver has 13 'addic'
> instructions, and the new version has 7.
> 
> Now taking the code around the last 'addic' instruction (from the
> gfar_clean_rx_ring function):
> 
> Initial version looks like this:
> 
>      5024:	4b ff ff ac 	b       4fd0 <gfar_clean_rx_ring+0x450>
>      5028:	81 5d 06 30 	lwz     r10,1584(r29)
>      502c:	81 7d 06 34 	lwz     r11,1588(r29)
>      5030:	31 6b 00 01 	addic   r11,r11,1
>      5034:	7d 4a 01 94 	addze   r10,r10
>      5038:	91 5d 06 30 	stw     r10,1584(r29)
>      503c:	91 7d 06 34 	stw     r11,1588(r29)
>      5040:	4b ff fe fc 	b       4f3c <gfar_clean_rx_ring+0x3bc>
> 
> New version looks like this:
> 
>      4ff8:	4b ff fd a8 	b       4da0 <gfar_clean_rx_ring+0x1ec>
>      4ffc:	80 1c 00 a0 	lwz     r0,160(r28)
>      5000:	38 60 00 00 	li      r3,0
>      5004:	80 a1 00 18 	lwz     r5,24(r1)
>      5008:	38 80 00 01 	li      r4,1
>      500c:	30 00 00 01 	addic   r0,r0,1
>      5010:	90 1c 00 a0 	stw     r0,160(r28)
>      5014:	48 00 00 01 	bl      5014 <gfar_clean_rx_ring+0x460>
>      5018:	4b ff ff 4c 	b       4f64 <gfar_clean_rx_ring+0x3b0>
> 
> 
> I have the whole function's asm excepts, if needed.
> 

I guess you are not looking at the right spot.

32bit powerpc probably use the generic atomic64

Your kernel should have an atomic64_add() function (in lib/atomic64.c)
and gianfar should call it.

This is certainly expensive, but these counters are not in fast path.

If they were in fast path, include/linux/u64_stats_sync.h would be a
better choice.



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html