Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <7728c846-0077-a34d-7fdc-bab9743edc4f@arm.com>
Date:   Wed, 28 Sep 2022 15:08:47 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux aarch64; rv:91.0) Gecko/20100101
 Thunderbird/91.10.0
Subject: Re: [PATCH] x86/resctrl: Clear the staged configs when destroying
 schemata list
Content-Language: en-GB
To:     Reinette Chatre <reinette.chatre@intel.com>,
        Shawn Wang <shawnwang@linux.alibaba.com>, fenghua.yu@intel.com
Cc:     tglx@linutronix.de, mingo@redhat.com, bp@alien8.de,
        dave.hansen@linux.intel.com, x86@kernel.org, hpa@zytor.com,
        linux-kernel@vger.kernel.org
References: <1664247269-41295-1-git-send-email-shawnwang@linux.alibaba.com>
 <80d6238b-223c-e60a-6930-24a981d9dd0c@arm.com>
 <2728c354-ac75-be4c-66ad-86ebd9c50248@intel.com>
From:   James Morse <james.morse@arm.com>
In-Reply-To: <2728c354-ac75-be4c-66ad-86ebd9c50248@intel.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi Reinette,

On 27/09/2022 22:21, Reinette Chatre wrote:
> On 9/27/2022 6:06 AM, James Morse wrote:
>> On 27/09/2022 03:54, Shawn Wang wrote:

>>> The problem can be reproduced by the following commands:
>>>     # A system with 16 usable closids and mba disabled
>>>     mount -t resctrl resctrl -o cdp /sys/fs/resctrl
>>>     mkdir /sys/fs/resctrl/p{1..7}
>>>     umount /sys/fs/resctrl/
>>>     mount -t resctrl resctrl /sys/fs/resctrl
>>>     mkdir /sys/fs/resctrl/p{1..8}
>>
>> Thanks for the reproducer - but I don't see what could set have_new_ctrl in this sequence.
>> You can't call apply_config() to set CPUs in the mask without that being set.
>>
>> Creating a new control group, (your mkdir step) shouldn't touch the hardware at all, as it
>> should be left in its reset state from the last umount(), or setup.
> 
> There is an attempt to configure the hardware in the mkdir path:
> rdtgroup_mkdir()->rdtgroup_mkdir_ctrl_mon()->rdtgroup_init_alloc()->resctrl_arch_update_domains()

Aha! I'm not sure why my grepping around didn't find this.

This is a path that doesn't memset() the staged config first, so that explains it.

[..]

> What do you think about clearing the staged config within resctrl_arch_update_domains()
> after the configuration is complete and there is no more need for it? That may reduce
> complexity where each caller no longer need to remember to do so.
> I see "staged_config" as a temporary storage and it my help to understand the code better
> if it is treated as such. 

Yup, that would it with the idea of the value being consumed by
resctrl_arch_update_domains(), which is how I think of it.


Thanks,

James