Date:	Tue, 25 Oct 2011 03:06:15 -0400
From:	Mike Frysinger <vapier@...too.org>
To:	Kyle Moffett <kyle@...fetthome.net>
Cc:	Kyle Moffett <Kyle.D.Moffett@...ing.com>,
	linux-kernel@...r.kernel.org, netdev@...r.kernel.org,
	Randy Dunlap <rdunlap@...otime.net>,
	Stephen Hemminger <shemminger@...tta.com>,
	"David S. Miller" <davem@...emloft.net>,
	Greg Dietsche <Gregory.Dietsche@....edu>,
	Giuseppe Cavallaro <peppe.cavallaro@...com>,
	David Daney <ddaney@...iumnetworks.com>,
	Arnaud Patard <arnaud.patard@...-net.org>,
	Grant Likely <grant.likely@...retlab.ca>,
	Baruch Siach <baruch@...s.co.il>,
	Thorsten Schubert <tshu@...-ge.com>,
	David Decotigny <decot@...gle.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Lucas De Marchi <lucas.demarchi@...fusion.mobi>,
	Marc Kleine-Budde <mkl@...gutronix.de>,
	linux-doc@...r.kernel.org
Subject: Re: [RFC PATCH 05/17] phy_driver: Make .read_status()/.config_aneg() optional

On Fri, Oct 21, 2011 at 01:13, Kyle Moffett wrote:
> On Thu, Oct 20, 2011 at 17:14, Mike Frysinger wrote:
>> On Thursday 20 October 2011 17:10:12 Mike Frysinger wrote:
>>> On Thursday 20 October 2011 17:00:12 Kyle Moffett wrote:
>>> > Approximately 90% of the PHY drivers follow the PHY layer docs and
>>> > simply use &genphy_read_status and &genphy_config_aneg.  There would
>>> > seem to be little point in requiring them all to manually specify those
>>> > functions.
>>>
>>> well, it does make sense if you think about the compile vs build time
>>> overhead.  yes, your patch does make things much nicer to read, and a
>>> little easier to maintain the source.  however, it adds runtime overhead
>>> (checking the func pointers) while the func pointer storage is unchanged
>>> (it's now a NULL pointer instead of pointing to the genphy funcs).
>>> personally, i think the savings in runtime and smaller compiled code is
>>> more important.  so i'm going to NAK this.  sorry.
>>
>> ah, sorry, i was thinking this was u-boot since we were just having
>> conversations there.
>>
>> since this is Linux, and i don't have real standing in the general netdev
>> community, i can't really NAK here.  but i think my comment still stands in
>> that this patch makes things much worse than the minor code style improvement.
>
> I would argue that the PHY layer itself is not remotely a hot-path,
> especially not to the level of caring about an extra if statement.  A
> single phy_read() is a sleeping call which goes over a slow serial
> bus, so code maintainability is much more important.

i disagree.  ignoring that, what you ultimately desire can be
accomplished without bloating the kernel.

option 1: this can be done in the registration func just like the mtd
layer.  if (!func_pointer) func_pointer = default;

option 2: introduce a new macro in the common phy header similar to:
  #define PHY_DRIVER_DEFAULT_FUNCS \
    .config_aneg = genphy_config_aneg, \
    .read_status = genphy_read_status
and then use that in the phy_driver init structs:
  struct phy_driver bcm5411_driver = {
    PHY_DRIVER_DEFAULT_FUNCS,
    .config_aneg = bcm5411_config_aneg,
    ...

however, imo, these func pointer arrays really should be in the
.rodata section with proper const markers.  this would require
breaking out into a new phy_driver_opts struct since the phy_driver
struct has read/write fields (like the list structure).  option 2
should allow this to work.
-mike
--
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

Powered by blists - more mailing lists