linux-kernel - Re: [PATCH] Reduce overhead of CONFIG_TIMER

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] [day] [month] [year] [list]

Message-ID: <20071217190339.GE25780@ghostprotocols.net>
Date:	Mon, 17 Dec 2007 17:03:39 -0200
From:	Arnaldo Carvalho de Melo <acme@...hat.com>
To:	Arjan van de Ven <arjan@...radead.org>
Cc:	Arnaldo Carvalho de Melo <acme@...hat.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	linux-kernel@...r.kernel.org, Ingo Molnar <mingo@...e.hu>
Subject: Re: [PATCH] Reduce overhead of CONFIG_TIMER_STATS

Em Mon, Dec 17, 2007 at 09:03:37AM -0800, Arjan van de Ven escreveu:
> On Sun, 16 Dec 2007 19:29:44 -0200
> Arnaldo Carvalho de Melo <acme@...hat.com> wrote:
> 
> > Hi,
> > 	While looking at the pahole output for struct timer_list on
> > recent kernels I noticed that there is a 4 bytes padding on struct
> > timer_list that gets propagated to many structs on 64 bits
> > architectures:
> > 
> > [acme@...pio linux-2.6]$ pahole -C timer_list /tmp/tcp.o.before
> > struct timer_list {
> >         struct list_head      entry;          /*     0    16 */
> >         long unsigned int     expires;        /*    16     8 */
> >         void                  (*function)(long unsigned int); /*
> > 24     8 */ long unsigned int     data;           /*    32     8 */
> >         struct tvec_t_base_s *base;           /*    40     8 */
> >         void *                start_site;     /*    48     8 */
> >         char                  start_comm[16]; /*    56    16 */
> >         /* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
> >         int                   start_pid;      /*    72     4 */
> > 
> >         /* size: 80, cachelines: 2 */
> >         /* padding: 4 */
> >         /* last cacheline: 16 bytes */
> > };
> > [acme@...pio linux-2.6]$
> > 
> > 	So the attached patch reduces the 4 bytes hole overhead of
> > CONFIG_TIMER_STATS on 64bit architectures by shrinking the field for
> > the process name by 4 bytes.
> > 
> > 	Statistically this doesn't affects that many process names as
> > most are less than 12 bytes. As CONFIG_TIMER_STATS is enabled at least
> > on fedora kernels I think that we can, with this patch, still reap the
> > benefits of powertopping.
> 
> I'm still worried that this means that PowerTOP will end up displaying shortened names..
> it's already sometimes tricky to know who's guilty at 16... at 12 I fear it just gets worse.
> Isn't there some other reorder that you can do to still get rid of the hole?

Nope. As shown above there is no other hole that we can conbine with the
4 bytes padding on struct timer_list.

<brainfarting>
What we could do, perhaps, would be to struct start_comm to be a pointer
and store the strings in a list, refcounted, so instead of 16 or 12
bytes we would use 8, but would incur in some CPU overhead definetely
bigger than a simple memcpy, but that would reduce even further the
memory footprint.

But the hole would still be there, as it would be sizeof(void *) +
sizeof(start_pid) :-)
</>

- Arnaldo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/