[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <ZoZ0jSbohdYAqY4E@hovoldconsulting.com>
Date: Thu, 4 Jul 2024 12:08:13 +0200
From: Johan Hovold <johan@...nel.org>
To: Doug Anderson <dianders@...omium.org>
Cc: Johan Hovold <johan+linaro@...nel.org>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Jiri Slaby <jirislaby@...nel.org>,
Konrad Dybcio <konrad.dybcio@...aro.org>,
Bjorn Andersson <andersson@...nel.org>,
linux-arm-msm@...r.kernel.org, linux-serial@...r.kernel.org,
linux-kernel@...r.kernel.org, stable@...r.kernel.org
Subject: Re: [PATCH 2/3] serial: qcom-geni: fix soft lockup on sw flow
control and suspend
On Wed, Jun 26, 2024 at 09:54:40AM +0200, Johan Hovold wrote:
> On Mon, Jun 24, 2024 at 02:58:39PM -0700, Doug Anderson wrote:
> > 1. The function is named qcom_geni_serial_clear_tx_fifo() which
> > implies that when it finishes that the hardware FIFO will have nothing
> > in it. ...but how does your code ensure this?
>
> Yeah, I realised after I sent out the series that this may not be the
> case. I was under the impression that cancelling a command would discard
> the data in the FIFO (e.g. when starting the next command) but that was
> probably an error in my mental model.
I went back and did some more reverse engineering and have now confirmed
that the hardware works as I assumed for v1, that is, that cancelling a
command leaves data in the fifo, which is later discarded when a new
command is issued.
> > 3. On my hardware you're setting the FIFO level to 16 here. The docs I
> > have say that if the FIFO level is "less than" the value you set here
> > then the interrupt will go off and further clarifies that if you set
> > the register to 1 here then you'll get interrupted when the FIFO is
> > empty. So what happens with your solution if the FIFO is completely
> > full? In that case you'd have to set this to 17, right? ...but then I
> > could believe that might confuse the interrupt handler which would get
> > told to start transmitting when there is no room for anything.
>
> Indeed. I may implicitly be relying on the absence of hardware flow
> control as well so that waiting for one character to be sent is what
> makes this work.
I'm keeping the watermark level unchanged in v2 and instead restart tx
by issuing a short transfer command to clear any stale data from the
fifo which could prevent the watermark interrupt from firing.
Johan
Powered by blists - more mailing lists