linux-kernel - Re: [PATCH 2/2] serial: qcom_geni_serial: Always use 4 bytes per TX FIFO word

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Fri, 10 Jul 2020 11:27:53 -0700
From:   Doug Anderson <dianders@...omium.org>
To:     Evan Green <evgreen@...omium.org>
Cc:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Daniel Thompson <daniel.thompson@...aro.org>,
        Akash Asthana <akashast@...eaurora.org>,
        Stephen Boyd <swboyd@...omium.org>,
        kgdb-bugreport@...ts.sourceforge.net,
        linux-arm-msm <linux-arm-msm@...r.kernel.org>,
        Sumit Garg <sumit.garg@...aro.org>,
        Vivek Gautam <vivek.gautam@...eaurora.org>,
        Andy Gross <agross@...nel.org>,
        Bjorn Andersson <bjorn.andersson@...aro.org>,
        Jiri Slaby <jslaby@...e.com>,
        LKML <linux-kernel@...r.kernel.org>, linux-serial@...r.kernel.org
Subject: Re: [PATCH 2/2] serial: qcom_geni_serial: Always use 4 bytes per TX
 FIFO word

Hi,

On Fri, Jul 10, 2020 at 10:46 AM Evan Green <evgreen@...omium.org> wrote:
>
> On Fri, Jun 26, 2020 at 1:01 PM Douglas Anderson <dianders@...omium.org> wrote:
> >
> > The geni serial driver had a rule that we'd only use 1 byte per FIFO
> > word for the TX FIFO if we were being used for the serial console.
> > This is ugly and a bit of a pain.  It's not too hard to fix, so fix
> > it.
> >
> > Signed-off-by: Douglas Anderson <dianders@...omium.org>
> > ---
> >
> >  drivers/tty/serial/qcom_geni_serial.c | 57 +++++++++++++++++----------
> >  1 file changed, 37 insertions(+), 20 deletions(-)
> >
> > diff --git a/drivers/tty/serial/qcom_geni_serial.c b/drivers/tty/serial/qcom_geni_serial.c
> > index 4610e391e886..583d903321b5 100644
> > --- a/drivers/tty/serial/qcom_geni_serial.c
> > +++ b/drivers/tty/serial/qcom_geni_serial.c
> > @@ -103,12 +103,18 @@
> >  #define DEFAULT_IO_MACRO_IO2_IO3_MASK          GENMASK(15, 4)
> >  #define IO_MACRO_IO2_IO3_SWAP          0x4640
> >
> > +/* We always configure 4 bytes per FIFO word */
> > +#define BYTES_PER_FIFO_WORD            4
> > +
> >  struct qcom_geni_private_data {
> >         /* NOTE: earlycon port will have NULL here */
> >         struct uart_driver *drv;
> >
> >         u32 poll_cached_bytes;
> >         unsigned int poll_cached_bytes_cnt;
> > +
> > +       u32 write_cached_bytes;
> > +       unsigned int write_cached_bytes_cnt;
> >  };
> >
> >  struct qcom_geni_serial_port {
> > @@ -121,8 +127,6 @@ struct qcom_geni_serial_port {
> >         bool setup;
> >         int (*handle_rx)(struct uart_port *uport, u32 bytes, bool drop);
> >         unsigned int baud;
> > -       unsigned int tx_bytes_pw;
> > -       unsigned int rx_bytes_pw;
> >         void *rx_fifo;
> >         u32 loopback;
> >         bool brk;
> > @@ -390,13 +394,25 @@ static void qcom_geni_serial_poll_put_char(struct uart_port *uport,
> >  #ifdef CONFIG_SERIAL_QCOM_GENI_CONSOLE
> >  static void qcom_geni_serial_wr_char(struct uart_port *uport, int ch)
> >  {
> > -       writel(ch, uport->membase + SE_GENI_TX_FIFOn);
> > +       struct qcom_geni_private_data *private_data = uport->private_data;
> > +
> > +       private_data->write_cached_bytes =
> > +               (private_data->write_cached_bytes >> 8) | (ch << 24);
> > +       private_data->write_cached_bytes_cnt++;
> > +
> > +       if (private_data->write_cached_bytes_cnt == BYTES_PER_FIFO_WORD) {
> > +               writel(private_data->write_cached_bytes,
> > +                      uport->membase + SE_GENI_TX_FIFOn);
> > +               private_data->write_cached_bytes_cnt = 0;
> > +       }
> >  }
> >
> >  static void
> >  __qcom_geni_serial_console_write(struct uart_port *uport, const char *s,
> >                                  unsigned int count)
> >  {
> > +       struct qcom_geni_private_data *private_data = uport->private_data;
> > +
> >         int i;
> >         u32 bytes_to_send = count;
> >
> > @@ -431,6 +447,15 @@ __qcom_geni_serial_console_write(struct uart_port *uport, const char *s,
> >                                                         SE_GENI_M_IRQ_CLEAR);
> >                 i += chars_to_write;
> >         }
> > +
> > +       if (private_data->write_cached_bytes_cnt) {
> > +               private_data->write_cached_bytes >>= BITS_PER_BYTE *
> > +                       (BYTES_PER_FIFO_WORD - private_data->write_cached_bytes_cnt);
> > +               writel(private_data->write_cached_bytes,
> > +                      uport->membase + SE_GENI_TX_FIFOn);
> > +               private_data->write_cached_bytes_cnt = 0;
> > +       }
>
> How does this not end up sending stray zeros? In other words, how does
> the hardware know which bytes of this word are valid?

We told it how many bytes we wanted to send in
qcom_geni_serial_setup_tx().  If the total number of bytes being sent
is not a multiple of the FIFO word size then it knows that the last
word will be a partial and it'll extract just the number of needed
bytes out of it.

Like receiving, sending bytes out of geni is also packet based.
Though the packets work a little differently for sending vs. receiving
in both cases you are supposed to fully finish a packet before you
send more bytes (you can sorta cancel / start a new packet, but that's
not what we're doing here).  So ahead of time we told it how many
bytes to expect and then we sent them all.

NOTE: if we wanted to simplify this function at the expense of
efficiency, we could change it to always send 1-byte packets.  Then
we'd start a packet, send 1 byte, wait for done, start a new packet,
send 1 byte, wait for done, etc.  In fact, that's how the polling code
does it...


-Doug