[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <c54858ed-e263-fe80-96aa-9ac8b6f85b1c@huawei.com>
Date: Wed, 21 Aug 2024 19:22:54 +0800
From: "lihuisong (C)" <lihuisong@...wei.com>
To: Daniel Lezcano <daniel.lezcano@...aro.org>, "Rafael J. Wysocki"
<rjw@...ysocki.net>
CC: LKML <linux-kernel@...r.kernel.org>, Lukasz Luba <lukasz.luba@....com>,
Zhang Rui <rui.zhang@...el.com>, Linux PM <linux-pm@...r.kernel.org>
Subject: Re: [PATCH v3 02/14] thermal: core: Rearrange checks in
thermal_bind_cdev_to_trip()
在 2024/8/21 18:49, Daniel Lezcano 写道:
> On 21/08/2024 11:44, lihuisong (C) wrote:
>>
>> 在 2024/8/21 17:28, Daniel Lezcano 写道:
>>> On 21/08/2024 10:49, lihuisong (C) wrote:
>>>
>>> [ ... ]
>>>
>>>>> - list_for_each_entry(pos2, &thermal_cdev_list, node) {
>>>>> - if (pos2 == cdev)
>>>>> - break;
>>>>> - }
>>>>> + lockdep_assert_held(&thermal_list_lock);
>>>>> - if (tz != pos1 || cdev != pos2)
>>>>> + if (list_empty(&tz->node) || list_empty(&cdev->node))
>>>> The old verification is ensure that tz and cdev already add to
>>>> thermal_tz_list and thermal_cdev_list,respectively.
>>>> Namely, tz and cdev are definitely registered and intialized.
>>>> The check is ok for all untizalized thermal_zone_device and cooling
>>>> device.
>>>> But the new verification doesn't seem to do that.
>>>
>>> If the tz or the cdev are registered then their "->node" is not
>>> empty because they are linked with the thermal_list and cdev_list
>>>
>>> So either way is browsing the lists to find the tz/cdev or just
>>> check "->node" is not empty. The latter the faster.
>> Assume that tz/cdev isn't intiazlized and registered to
>> thermal_tz_list or thermal_cdev_list. And then directly call this
>> interface.
>
> Then there is a bug in the internal code because the
> thermal_zone_device_register*() and cooling_device_device_register()
> allocate and initialize those structures.
>
> The caller of the function is supposed to use the API provided by the
> thermal framework. It is not possible to plan every stupid things a
> driver can do. In this particular case, very likely the kernel will
> crash immediately which is a sufficient test for me and coercive
> enough to have the API user to put its code in question ;)
A good point. Agree.
Powered by blists - more mailing lists