Re: [PATCH] mmc: core: Prevent processing SDIO IRQs when the card is suspended

From: Ulf Hansson <ulf.hansson@linaro.org>
To: Doug Anderson <dianders@chromium.org>
Cc: Linux MMC List <linux-mmc@vger.kernel.org>,
	Adrian Hunter <adrian.hunter@intel.com>,
	Brian Norris <briannorris@chromium.org>,
	Shawn Lin <shawn.lin@rock-chips.com>,
	Guenter Roeck <groeck@chromium.org>,
	Heiko Stuebner <heiko@sntech.de>,
	Kalle Valo <kvalo@codeaurora.org>,
	linux-wireless <linux-wireless@vger.kernel.org>,
	"# 4.0+" <stable@vger.kernel.org>
Subject: Re: [PATCH] mmc: core: Prevent processing SDIO IRQs when the card is suspended
Date: Mon, 17 Jun 2019 11:56:20 +0200	[thread overview]
Message-ID: <CAPDyKFoyq5__e-2ZSZMzMzBg3X=7zou6wQ4J6o57a0CfTvEO7g@mail.gmail.com> (raw)
In-Reply-To: <CAD=FV=Ujb=_jEFMdxLPW6tYwb9DZo5-RZ8BVuyq5DdFQ5jJbQw@mail.gmail.com>

On Fri, 14 Jun 2019 at 17:42, Doug Anderson <dianders@chromium.org> wrote:
>
> Hi,
>
> On Fri, Jun 14, 2019 at 4:56 AM Ulf Hansson <ulf.hansson@linaro.org> wrote:
> >
> > > I was more worried about the safety of mmc_card_set_suspended()
> > > itself.  That is:
> > >
> > > #define mmc_card_set_suspended(c) ((c)->state |= MMC_STATE_SUSPENDED)
> > >
> > > ...so it's doing a read-modify-write of "state".  Is that safe to do
> > > without any type of locking?
> >
> > In this case, yes I think so.
> >
> > The point is, it really doesn't matter if the reader (work or thread),
> > reads a non-updated value, because the synchronization is managed by
> > the later mmc_claim_host() and the cancel_delayed_work_sync().
>
> If this were just an "int" then perhaps, but this is a bitfield.  So
> if someone else updates the bitfield at the same time then we can
> fully clobber their modification or they can clobber ours, right?
>
> task 1: load "state" from memory into CPU register on cpu0
> task 2: load "state" from memory into CPU register on cpu1
> task 1: OR in MMC_CARD_REMOVED
> task 1: write "state" from CPU register on cpu0
> task 2: OR in MMC_STATE_SUSPENDED
> task 2: write "state" from CPU register on cpu1
>
> ...so now we've clobbered MMC_CARD_REMOVED.  ...or am I just being
> paranoid here and everything else in "state" is somehow guaranteed to
> not be touched at the same time this function is running?

I understand your concern. It's not obvious by looking at the code,
but yes, there should be no other writing to the "state" at the same
time mmc_sdio_supend() is running.

MMC_CARD_REMOVED for example, is set from  _mmc_detect_card_removed(),
but because the detect work (mmc_recan()) has been disabled and the
block device driver has been suspended, it can't be called.

Anyway, perhaps we get reasons to add a lock to the card struct when
going forward, but at this point I think we are fine.

Kind regards
Uffe