Message-ID: <CAJOA=zNteC+24oWpqkQJRnL8oqAJe2NcsjHNLbJb1MGN--VEsw@mail.gmail.com>
Date:	Tue, 20 Mar 2012 22:45:36 -0700
From:	"Turquette, Mike" <mturquette@...com>
To:	Saravana Kannan <skannan@...eaurora.org>
Cc:	Sascha Hauer <s.hauer@...gutronix.de>,
	Andrew Lunn <andrew@...n.ch>,
	Grant Likely <grant.likely@...retlab.ca>,
	Jamie Iles <jamie@...ieiles.com>,
	Jeremy Kerr <jeremy.kerr@...onical.com>,
	Magnus Damm <magnus.damm@...il.com>,
	Deepak Saxena <dsaxena@...aro.org>,
	linux-arm-kernel@...ts.infradead.org,
	Arnd Bergman <arnd.bergmann@...aro.org>,
	linux-arm-msm@...r.kernel.org,
	Rob Herring <rob.herring@...xeda.com>,
	Russell King <linux@....linux.org.uk>,
	Thomas Gleixner <tglx@...utronix.de>,
	Richard Zhao <richard.zhao@...aro.org>,
	Shawn Guo <shawn.guo@...escale.com>,
	Paul Walmsley <paul@...an.com>,
	Linus Walleij <linus.walleij@...ricsson.com>,
	Mark Brown <broonie@...nsource.wolfsonmicro.com>,
	Stephen Boyd <sboyd@...eaurora.org>,
	linux-kernel@...r.kernel.org,
	Amit Kucheria <amit.kucheria@...aro.org>
Subject: Re: [PATCH 1/2] clk: Fix error handling in fixed clock hardware type
 register fn

On Tue, Mar 20, 2012 at 7:32 PM, Saravana Kannan <skannan@...eaurora.org> wrote:
> On 03/20/2012 05:13 PM, Turquette, Mike wrote:
>>
>> On Tue, Mar 20, 2012 at 12:46 AM, Saravana Kannan
>> <skannan@...eaurora.org>  wrote:
>>>
>>> On Tue, March 20, 2012 12:19 am, Sascha Hauer wrote:
>>>>
>>>> On Mon, Mar 19, 2012 at 08:38:25PM -0700, Saravana Kannan wrote:
>>>>>
>>>>> If memory allocation for the parents array or the parent string fails,
>>>>> then
>>>>> fail the registration immediately instead of calling clk_register and
>>>>> hoping it fails there.
>>>>>
>>>>> Return -ENOMEM on failure.
>>>>>
>>>>> Signed-off-by: Saravana Kannan<skannan@...eaurora.org>
>>>>> Cc: Mike Turquette<mturquette@...aro.org>
>>>>> Cc: Andrew Lunn<andrew@...n.ch>
>>>>> Cc: Rob Herring<rob.herring@...xeda.com>
>>>>> Cc: Russell King<linux@....linux.org.uk>
>>>>> Cc: Jeremy Kerr<jeremy.kerr@...onical.com>
>>>>> Cc: Thomas Gleixner<tglx@...utronix.de>
>>>>> Cc: Arnd Bergman<arnd.bergmann@...aro.org>
>>>>> Cc: Paul Walmsley<paul@...an.com>
>>>>> Cc: Shawn Guo<shawn.guo@...escale.com>
>>>>> Cc: Sascha Hauer<s.hauer@...gutronix.de>
>>>>> Cc: Jamie Iles<jamie@...ieiles.com>
>>>>> Cc: Richard Zhao<richard.zhao@...aro.org>
>>>>> Cc: Saravana Kannan<skannan@...eaurora.org>
>>>>> Cc: Magnus Damm<magnus.damm@...il.com>
>>>>> Cc: Mark Brown<broonie@...nsource.wolfsonmicro.com>
>>>>> Cc: Linus Walleij<linus.walleij@...ricsson.com>
>>>>> Cc: Stephen Boyd<sboyd@...eaurora.org>
>>>>> Cc: Amit Kucheria<amit.kucheria@...aro.org>
>>>>> Cc: Deepak Saxena<dsaxena@...aro.org>
>>>>> Cc: Grant Likely<grant.likely@...retlab.ca>
>>>>> ---
>>>>> There are still some memory free issues when clk_register() fails, but
>>>>> I
>>>>> will
>>>>> fix it when I fixed the other register() fns to return ENOMEM of alloc
>>>>> failure instead of a NULL.
>>>>>
>>>>>  drivers/clk/clk-fixed-rate.c |   10 +++++++---
>>>>>  1 files changed, 7 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/drivers/clk/clk-fixed-rate.c
>>>>> b/drivers/clk/clk-fixed-rate.c
>>>>> index 90c79fb..6423ae9 100644
>>>>> --- a/drivers/clk/clk-fixed-rate.c
>>>>> +++ b/drivers/clk/clk-fixed-rate.c
>>>>> @@ -61,22 +61,26 @@ struct clk *clk_register_fixed_rate(struct device
>>>>> *dev, const char *name,
>>>>>              parent_names = kmalloc(sizeof(char *), GFP_KERNEL);
>>>>>
>>>>>              if (! parent_names)
>>>>> -                    goto out;
>>>>> +                    goto fail_ptr;
>>>>>
>>>>>              len = sizeof(char) * strlen(parent_name);
>>>>>
>>>>>              parent_names[0] = kmalloc(len, GFP_KERNEL);
>>>>>
>>>>>              if (!parent_names[0])
>>>>> -                    goto out;
>>>>> +                    goto fail_str;
>>>>>
>>>>>              strncpy(parent_names[0], parent_name, len);
>>>>>      }
>>>>
>>>>
>>>> It's easier to add a char *parent to struct clk_fixed and pass it to
>>>> clk_register with&fixed->parent. This saves you a kmalloc call and
>>>>
>>>> makes the error path simpler. It's the same way already done in the
>>>> divider.
>>
>>
>> I thought I had done this for v7... hmm looks like one got left out.
>> I'll line up a patch to get it in sync with the others as part of my
>> fixes.
>>
>>> I thought about that since I saw the same was done for gated and divider
>>> (I think). Here is my guess at Mike's reasoning for this:
>>>
>>> Gated and divider clocks have to have a parent. There's nothing to gate
>>> otherwise. But fixed rate clocks might not have a parent. It could be
>>> XO's
>>> or PLLs running off of always on XOs not controlled by the SoC. So, it's
>>> arguable to not have a parent. I don't have a strong opinion on this --
>>> since Mike took the time to write it, it left it to his subjective
>>> preference.
>>
>>
>> I appreciate the thoughtfulness.  Re-using the same type of mechanism
>> as the divider and gate clocks will still allow the fixed-rate clock
>> to be parentless, and it makes for cleaner code, one less allocation
>> and lines up with how the other single-parent basic clocks are done,
>> so I'll take that method in instead of your patch.
>
>
> No problem, go for it.
>
>
>>
>>> I sent this patch first since it was around the place I was cleaning up.
>>> I
>>> didn't want to actually just shuffle around a bug. As I mentioned, this
>>> patch still leaves a bug open -- what if clk_register() fails. I plan to
>>> fix that once my two patches are picked up (hopefully).
>>
>>
>> Do you still find it useful to return -ENOMEM from the registration
>> function instead of a NULL clock?  I'm always worried that people
>> don't check for error codes on pointers in their platform code and
>> only check for NULL...
>
>
> The last discussion I remember, NULL was considered a valid clock. So, I
> think on error, we shouldn't ever return NULL when the return type is struct
> clk *.

IIRC, that discussion was with respect to the .parent member of struct
clk.  It was decided that having .parent = NULL does not imply that a
clock is a root clock, but instead we rely on the CLK_IS_ROOT flag.  I
can't think of any other instance where a NULL clk when returned from
a registration function would be useful.

Regards,
Mike

> Thanks,
> Saravana
--
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/

Powered by blists - more mailing lists