[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <54e2dac4-3f8e-4dad-d71c-09572039bd39@huawei.com>
Date: Fri, 22 Dec 2017 17:06:46 +0800
From: Yisheng Xie <xieyisheng1@...wei.com>
To: Greg KH <gregkh@...uxfoundation.org>
CC: <thomas.lendacky@....com>, <lorenzo.pieralisi@....com>,
<bp@...e.de>, <tglx@...utronix.de>, <kstewart@...uxfoundation.org>,
<linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2] devres: use MACRO instead of function for devm_ioremap
Hi Greg,
On 2017/12/21 23:08, Greg KH wrote:
> On Thu, Dec 21, 2017 at 07:50:16PM +0800, Yisheng Xie wrote:
>> Hi Greg,
>>
>> On 2017/12/19 18:52, Yisheng Xie wrote:
>>> Hi Greg,
>>>
>>> On 2017/12/19 16:46, Greg KH wrote:
>>>> On Sat, Nov 25, 2017 at 05:23:33PM +0800, Yisheng Xie wrote:
>>>>> Default ioremap is ioremap_nocache, so devm_ioremap has the same function
>>>>> with devm_ioremap_nocache, which may just be killed. However, there are
>>>>> many places which use devm_ioremap_nocache, while many use devm_ioremap.
>>>>>
>>>>> This patch is to use MACRO for devm_ioremap, which will reduce the size of
>>>>> devres.o from 206824 Bytes to 203768 Bytes.
>>>>
>>>> Ok, the idea is good, but why not just get rid of the callers of
>>>> devm_ioremap_nocache() instead and have them call devm_ioremap() if they
>>>> really are the same thing? No need to keep a macro around for the
>>>> duplicate thing, just delete the one and things are much better.
>>>
>>> Yeah, I thought someone may still want to use devm_ioremap_nocache().
>>>
>>> I will take your suggestion in next version and kill devm_ioremap_nocache().
>>
>> I am trying to kill devm_ioremap_nocache() as your suggestion, however, if
>> put it as a single patch it will be a patch really big (the patch file may
>> have more than 1k lines), for many places will use devm_ioremap_nocache().
>> But it also makes me fell uncomfortable to separate it to so many litter
>> patchs for a litter optimize. :)
>>
>> So I put the v3 patch in the attachment, please let me know if I should
>> separate it to litter patchs :)
>
> Yes, of course you need to break it up into smaller pieces.
>
> One per subsystem, make a large patch series, send it out, get a few
> merged, refresh, send again, and keep going until all of the users are
> out of the tree and then drop the old api call. That's how we do this
> all the time, should take about 1 kernel release if you are quick and
> lucky :)
>
> hope this helps,
Yes, it do help.
Thanks for your help.
Yisheng Xie
>
> greg k-h
>
> .
>
Powered by blists - more mailing lists