Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20201019223257.261223-1-abhishekpandit@chromium.org>
 <20201019153232.1.I797f9874972a07fc381fe586b6748ce71c7b1fda@changeid> <20201020055738.GD3437534@kroah.com>
In-Reply-To: <20201020055738.GD3437534@kroah.com>
From:   Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
Date:   Tue, 20 Oct 2020 20:27:27 -0700
Message-ID: <CANFp7mXd=spL6PAL1cGrdZYBHu4A5uqJhFnu3Fmg9FmmFu6JJw@mail.gmail.com>
Subject: Re: [PATCH 1/1] kobject: Don't emit change events if not in sysfs
To:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc:     Rafael Wysocki <rafael.j.wysocki@intel.com>,
        Stephen Boyd <swboyd@chromium.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Linux PM <linux-pm@vger.kernel.org>,
        "Rafael J. Wysocki" <rafael@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Greg,

I was debugging without a live repro and I was told this patch
improved behavior but it's only by chance (someone bisected a Dell
D6000 dock's displayport issue to this commit and this change seemed
to help; udev logs later shows that's not the case). I took another
look at device_init_wakeup and I can see that
device_set_wakeup_capable does indeed check for device_is_registered
before adding the wakeup attributes so the ordering of events I
suspected cannot occur.

Thanks for pushing back Greg. It made me take a deeper look at an
assumption I hadn't challenged. Please consider this patch abandoned.

Abhishek

On Mon, Oct 19, 2020 at 10:56 PM Greg Kroah-Hartman
<gregkh@linuxfoundation.org> wrote:
>
> On Mon, Oct 19, 2020 at 03:32:57PM -0700, Abhishek Pandit-Subedi wrote:
> > Add a check to make sure the kobj is created and in sysfs before sending
> > a change event notification. Otherwise, udev rules that depend on the
> > change notification may find that the path that changed doesn't actually
> > exist.
>
> Why is the user of the kobject trying to emit a uevent before it is
> registered?  Shouldn't we fix the root problem here instead?  Otherwise
> the event is still "gone", the caller will not know what to do about it.
>
> Please fix the root problem here.
>
> thanks,
>
> greg k-h