From mboxrd@z Thu Jan  1 00:00:00 1970
From: Yang Shi <yang.shi@windriver.com>
Subject: Re: [PATCH v3.14-rt] mips: rt: Replace pagefault_* to raw version
Date: Tue, 17 Feb 2015 13:40:01 -0800
Message-ID: <54E3B531.8050600@windriver.com>
References: <1415317527-15266-1-git-send-email-yang.shi@windriver.com> <20150217135404.GM26177@linutronix.de>
Mime-Version: 1.0
Content-Type: text/plain; charset="utf-8"; format=flowed
Content-Transfer-Encoding: 7bit
Cc: <tglx@linutronix.de>, <rostedt@goodmis.org>,
	<linux-rt-users@vger.kernel.org>, <paul.gortmaker@windriver.com>
To: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Return-path: <linux-rt-users-owner@vger.kernel.org>
Received: from mail.windriver.com ([147.11.1.11]:36389 "EHLO
	mail.windriver.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1750846AbbBQVkQ (ORCPT
	<rfc822;linux-rt-users@vger.kernel.org>);
	Tue, 17 Feb 2015 16:40:16 -0500
In-Reply-To: <20150217135404.GM26177@linutronix.de>
Sender: linux-rt-users-owner@vger.kernel.org
List-ID: <linux-rt-users.vger.kernel.org>

On 2/17/2015 5:54 AM, Sebastian Andrzej Siewior wrote:
> * Yang Shi | 2014-11-06 15:45:27 [-0800]:
>
>> In k{un}map_coherent, pagefault_disable and pagefault_enable are called
>> respectively, but k{un}map_coherent needs preempt disabled according to
>> commit f8829caee311207afbc882794bdc5aa0db5caf33 ("[MIPS] Fix aliasing bug
>> in copy_to_user_page / copy_from_user_page") to avoid dcache alias on COW.
>>
>> k{un}map_coherent are just called when cpu_has_dc_aliases == 1 with VIPT cache.
>> However, actually, the most modern MIPS processors have PIPT dcache without
>> dcache alias issue. In such case, k{un}map_atomic will be called with preempt
>> enabled.
>>
>> To fix this, we replace pagefault_* to raw version in k{un}map_coherent, which
>> disables preempt, otherwise the following kernel panic may be caught:
> And this works? One thing that confuses me is that you talk about kmap()
> while there is no HIGHMEM support in MIPS. The other thing is that I had
> a longson a while ago and I don't remember this. But it could have no
> dcache alias problem :)

Yes, it works. I found it on mips64 target.

>
> Your registers look 64bit wide so this highpage thingy is probably
> just to deal with the dcache alias. So if this is the only problem you
> have I would be happy to take your patch. Could you please rebase it
> against v3.18? The code changed slightly and I would not want to touch
> and break it :)

Yes, the highpage is just used for dealing with MIPS dcache alias issue.

I'm going to rebase it onto 3.18.

Thanks,
Yang

>
> Sebastian