Date: Fri, 14 Feb 2014 16:33:49 -0800 From: John-Mark Gurney <jmg@funkthat.com> To: "Gumpula, Suresh" <Suresh.Gumpula@netapp.com> Cc: "freebsd-hackers@freebsd.org" <freebsd-hackers@freebsd.org>, Ian Lepore <ian@freebsd.org> Subject: Re: malloc(9) and its alignment Message-ID: <20140215003349.GT34851@funkthat.com> In-Reply-To: <D29CB80EBA4DEA4D91181928AAF51538438F10CC@SACEXCMBX04-PRD.hq.netapp.com> References: <D29CB80EBA4DEA4D91181928AAF51538438EED0A@SACEXCMBX04-PRD.hq.netapp.com> <1392214788.1145.52.camel@revolution.hippie.lan> <D29CB80EBA4DEA4D91181928AAF51538438EF8DC@SACEXCMBX04-PRD.hq.netapp.com> <20140212220705.GV34851@funkthat.com> <D29CB80EBA4DEA4D91181928AAF51538438F10CC@SACEXCMBX04-PRD.hq.netapp.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Gumpula, Suresh wrote this message on Fri, Feb 14, 2014 at 02:07 +0000:
> Thanks for the explanation John. How about porting ARM way of handling required alignment with creation of separate zones
> And allocating with uma_zalloc(zone) for AMD64 too for bus_dmamem_alloc?
It looks like the code in HEAD is different than the code you originally
quoted, this is because HEAD added support for DMAR from Intel VT-d, the
code is now in bounce_bus_dmamem_alloc in x86/x86/busdma_bounce.c, but
it still has the same comment:
398 /*
399 * XXX:
400 * (dmat->alignment < dmat->maxsize) is just a quick hack; the exact
401 * alignment guarantees of malloc need to be nailed down, and the
402 * code below should be rewritten to take that into account.
403 *
404 * In the meantime, we'll warn the user if malloc gets it wrong.
405 */
Per porting arm's way of handling alignment, that makes sense... Though
arm's way forces all allocations to be aligned to the size of a buffer,
but I don't see a better way to handle it..
> -----Original Message-----
> From: John-Mark Gurney [mailto:jmg@funkthat.com]
> Sent: Wednesday, February 12, 2014 5:07 PM
> To: Gumpula, Suresh
> Cc: Ian Lepore; freebsd-hackers@freebsd.org
> Subject: Re: malloc(9) and its alignment
>
> Gumpula, Suresh wrote this message on Wed, Feb 12, 2014 at 19:40 +0000:
> > Thanks Ian for the reply. I will look at the ARM code, but I was thinking why malloc(9) does not return bucket size aligned pointers.
>
> Always returning bucket sizes aligned pointers may not be ideal for a cache.. say you have a buffer of 512 bytes, where often only the first
> 128 bytes are used (but all 512 bytes may be)... If you always align at 512, some cache lines will be more heavily used than others, reducing the effective size of the cache...
>
> This is only one reason not aligning to size may be better...
--
John-Mark Gurney Voice: +1 415 225 5579
"All that I will do, has been done, All that I have, has not."
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20140215003349.GT34851>