Re: [PATCH v4 5/5] spi: spi-geni-qcom: Don't keep a local state variable

From: Doug Anderson <dianders@chromium.org>
To: Stephen Boyd <swboyd@chromium.org>
Cc: Mark Brown <broonie@kernel.org>,
	Alok Chauhan <alokc@codeaurora.org>,
	skakit@codeaurora.org, Andy Gross <agross@kernel.org>,
	Bjorn Andersson <bjorn.andersson@linaro.org>,
	linux-arm-msm <linux-arm-msm@vger.kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	linux-spi <linux-spi@vger.kernel.org>
Subject: Re: [PATCH v4 5/5] spi: spi-geni-qcom: Don't keep a local state variable
Date: Thu, 18 Jun 2020 15:00:10 -0700	[thread overview]
Message-ID: <CAD=FV=W1y4Z4T13i410zkb27mUxqn+rQE889=ckEEBhbPuci2w@mail.gmail.com> (raw)
In-Reply-To: <159251712230.62212.10744179843753723398@swboyd.mtv.corp.google.com>

Hi,

On Thu, Jun 18, 2020 at 2:52 PM Stephen Boyd <swboyd@chromium.org> wrote:
>
> Quoting Doug Anderson (2020-06-18 13:09:47)
> > On Thu, Jun 18, 2020 at 11:05 AM Stephen Boyd <swboyd@chromium.org> wrote:
> > >
> > > Quoting Douglas Anderson (2020-06-18 08:06:26)
> > > > @@ -126,20 +120,23 @@ static void handle_fifo_timeout(struct spi_master *spi,
> > > >         struct geni_se *se = &mas->se;
> > > >
> > > >         spin_lock_irq(&mas->lock);
> > > > -       reinit_completion(&mas->xfer_done);
> > > > -       mas->cur_mcmd = CMD_CANCEL;
> > > > -       geni_se_cancel_m_cmd(se);
> > > > +       reinit_completion(&mas->cancel_done);
> > > >         writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
> > > > +       mas->cur_xfer = NULL;
> > >
> > > BTW, is this necessary? It's subtlely placed here without a comment why.
> >
> > I believe so.  Now that we don't have the "cur_mcmd" we rely on
> > cur_xfer being NULL to tell the difference between a "done" for chip
> > select vs. a "done" for transfer.
> >
> > * When we start a transfer we set "cur_xfer" to a non-NULL pointer.
> > When the transfer finishes we set it to NULL again.
> >
> > * When we start a chip select transfer we _don't_ explicitly set it to
> > NULL because it should already be NULL.
> >
> > * When we are aborting a transfer we need to NULL so we can handle the
> > chip select that will come next.
> >
> > I suppose it's possible that we could get by without without NULLing
> > it because I believe when the "abort" IRQ finally fires then it will
> > include a "DONE" and that would presumably NULL it out.  ...but I
> > guess if both the cancel and abort timed out and no IRQ ever fired
> > then nothing would have NULLed it and the next chip select would be
> > confused.
>
> I was going to say that we should set it NULL when starting CS but that
> is not as important as clearing it out when a cancel/abort is processing
> so that a stale transfer isn't kept around.
>
> >
> > Prior to getting rid of "cur_mcmd" this all wasn't needed because
> > "cur_xfer" was only ever looked at if "cur_mcmd" was set to
> > "CMD_XFER".
> >
> >
> > One part of my change that is technically not related to the removal
> > of "cur_mcmd" is the part where I do "mas->tx_rem_bytes =
> > mas->rx_rem_bytes = 0;".  I can split that as a separate change if you
> > want but it seemed fine to just clean up this extra bit of state here.
> >
>
> How about a comment like this?
>
> -----8<----
> diff --git a/drivers/spi/spi-geni-qcom.c b/drivers/spi/spi-geni-qcom.c
> index d8f03ffb8594..670f83793aa4 100644
> --- a/drivers/spi/spi-geni-qcom.c
> +++ b/drivers/spi/spi-geni-qcom.c
> @@ -121,6 +121,10 @@ static void handle_fifo_timeout(struct spi_master *spi,
>         spin_lock_irq(&mas->lock);
>         reinit_completion(&mas->cancel_done);
>         writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
> +       /*
> +        * Make sure we don't finalize a spi transfer that timed out but
> +        * came in while cancelling.
> +        */
>         mas->cur_xfer = NULL;
>         mas->tx_rem_bytes = mas->rx_rem_bytes = 0;
>         geni_se_cancel_m_cmd(se);

Sure.  It gets the point across, though
spi_finalize_current_transfer() is actually pretty harmless if you
call it while cancelling.  It just calls a completion.  I'd rather say
something like "If we're here because the SPI controller was calling
handle_err() then the transfer is done and we shouldn't hold onto it
anymore".

-Doug