[PATCH v4 0/6] LSR flag preservation improvements

[PATCH v4 0/6] LSR flag preservation improvements

[Ilpo Järvinen]

Improve LSR flag preservation. Not all devices preserve all LSR flags
on read. Therefore, the non-permanent flags must be stored until
consumed. 8250_port has handled this for some of the reads (but not
all). Drivers not so much but it's unclear to me which of the devices
actually clear flags on read so this series only does per driver fixes
for DW.

I've just put Uwe as Co-dev-by and SoB according to Andy's suggestion
in order to allow forward progress (Uwe indicated he's OK with any
sensible combination of tags anyway). I agree with Uwe's position
though that it would be nice to be able to record Acked-by in some
(probably quite rare) cases even if the person is already SoB but I
don't want to keep dragging this issue on and on over a single line
patch :-).

v2:
- Added more patches

v3:
- Fix Uwe's email and convert it to Co-developed-by

v4:
- Tweak comments
- Fix URL link to earlier discussion & improve changelog

Ilpo Järvinen (6):
  serial: 8250: Store to lsr_save_flags after lsr read
  serial: 8250: Create serial_lsr_in()
  serial: 8250: Get preserved flags using serial_lsr_in()
  serial: 8250: Adjust misleading LSR related comment
  serial: 8250_dw: Use serial_lsr_in() in dw8250_handle_irq()
  serial: 8250_dw: Store LSR into lsr_saved_flags in
    dw8250_tx_wait_empty()

 drivers/tty/serial/8250/8250.h      | 20 ++++++++++++++++++++
 drivers/tty/serial/8250/8250_core.c |  3 +--
 drivers/tty/serial/8250/8250_dw.c   |  7 +++++--
 drivers/tty/serial/8250/8250_port.c | 23 ++++++++++-------------
 4 files changed, 36 insertions(+), 17 deletions(-)

-- 
2.30.2

[PATCH v4 1/6] serial: 8250: Store to lsr_save_flags after lsr read

[Ilpo Järvinen]

Not all LSR register flags are preserved across reads. Therefore, LSR
readers must store the non-preserved bits into lsr_save_flags.

This fix was initially mixed into feature commit f6f586102add ("serial:
8250: Handle UART without interrupt on TEMT using em485"). However,
that feature change had a flaw and it was reverted to make room for
simpler approach providing the same feature. The embedded fix got
reverted with the feature change.

Re-add the lsr_save_flags fix and properly mark it's a fix.

Fixes: e490c9144cfa ("tty: Add software emulated RS485 support for 8250")
Link: https://lore.kernel.org/all/1d6c31d-d194-9e6a-ddf9-5f29af829f3@linux.intel.com/T/#m1737eef986bd20cf19593e344cebd7b0244945fc
Co-developed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
---
 drivers/tty/serial/8250/8250_port.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
index 4998799abae2..c5e0f925f4b6 100644
--- a/drivers/tty/serial/8250/8250_port.c
+++ b/drivers/tty/serial/8250/8250_port.c
@@ -1511,6 +1511,8 @@ static inline void __stop_tx(struct uart_8250_port *p)
 		unsigned char lsr = serial_in(p, UART_LSR);
 		u64 stop_delay = 0;
 
+		p->lsr_saved_flags |= lsr & LSR_SAVE_FLAGS;
+
 		if (!(lsr & UART_LSR_THRE))
 			return;
 		/*
-- 
2.30.2

[PATCH v4 2/6] serial: 8250: Create serial_lsr_in()

[Ilpo Järvinen]

LSR register readers need to be careful in order to not lose bits that
are not preserved across reads. Create a helper that takes care of
storing the non-preserved bits into lsr_save_flags.

Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
---
 drivers/tty/serial/8250/8250.h      | 20 ++++++++++++++++++++
 drivers/tty/serial/8250/8250_core.c |  3 +--
 drivers/tty/serial/8250/8250_port.c | 15 ++++-----------
 3 files changed, 25 insertions(+), 13 deletions(-)

diff --git a/drivers/tty/serial/8250/8250.h b/drivers/tty/serial/8250/8250.h
index 3d9a7e539dea..b120da57c61f 100644
--- a/drivers/tty/serial/8250/8250.h
+++ b/drivers/tty/serial/8250/8250.h
@@ -123,6 +123,26 @@ static inline void serial_out(struct uart_8250_port *up, int offset, int value)
 	up->port.serial_out(&up->port, offset, value);
 }
 
+/**
+ *	serial_lsr_in - Read LSR register and preserve flags across reads
+ *	@up:	uart 8250 port
+ *
+ *	Read LSR register and handle saving non-preserved flags across reads.
+ *	The flags that are not preserved across reads are stored into
+ *	up->lsr_saved_flags.
+ *
+ *	Returns LSR value or'ed with the preserved flags (if any).
+ */
+static inline unsigned int serial_lsr_in(struct uart_8250_port *up)
+{
+	unsigned int lsr = up->lsr_saved_flags;
+
+	lsr |= serial_in(up, UART_LSR);
+	up->lsr_saved_flags = lsr & LSR_SAVE_FLAGS;
+
+	return lsr;
+}
+
 /*
  * For the 16C950
  */
diff --git a/drivers/tty/serial/8250/8250_core.c b/drivers/tty/serial/8250/8250_core.c
index 12c5e5d0ce6d..90ddc8924811 100644
--- a/drivers/tty/serial/8250/8250_core.c
+++ b/drivers/tty/serial/8250/8250_core.c
@@ -276,8 +276,7 @@ static void serial8250_backup_timeout(struct timer_list *t)
 	 * the "Diva" UART used on the management processor on many HP
 	 * ia64 and parisc boxes.
 	 */
-	lsr = serial_in(up, UART_LSR);
-	up->lsr_saved_flags |= lsr & LSR_SAVE_FLAGS;
+	lsr = serial_lsr_in(up);
 	if ((iir & UART_IIR_NO_INT) && (up->ier & UART_IER_THRI) &&
 	    (!uart_circ_empty(&up->port.state->xmit) || up->port.x_char) &&
 	    (lsr & UART_LSR_THRE)) {
diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
index c5e0f925f4b6..ec5abeb638eb 100644
--- a/drivers/tty/serial/8250/8250_port.c
+++ b/drivers/tty/serial/8250/8250_port.c
@@ -1508,11 +1508,9 @@ static inline void __stop_tx(struct uart_8250_port *p)
 	struct uart_8250_em485 *em485 = p->em485;
 
 	if (em485) {
-		unsigned char lsr = serial_in(p, UART_LSR);
+		unsigned char lsr = serial_lsr_in(p);
 		u64 stop_delay = 0;
 
-		p->lsr_saved_flags |= lsr & LSR_SAVE_FLAGS;
-
 		if (!(lsr & UART_LSR_THRE))
 			return;
 		/*
@@ -1567,10 +1565,8 @@ static inline void __start_tx(struct uart_port *port)
 
 	if (serial8250_set_THRI(up)) {
 		if (up->bugs & UART_BUG_TXEN) {
-			unsigned char lsr;
+			unsigned char lsr = serial_lsr_in(up);
 
-			lsr = serial_in(up, UART_LSR);
-			up->lsr_saved_flags |= lsr & LSR_SAVE_FLAGS;
 			if (lsr & UART_LSR_THRE)
 				serial8250_tx_chars(up);
 		}
@@ -2001,8 +1997,7 @@ static unsigned int serial8250_tx_empty(struct uart_port *port)
 	serial8250_rpm_get(up);
 
 	spin_lock_irqsave(&port->lock, flags);
-	lsr = serial_port_in(port, UART_LSR);
-	up->lsr_saved_flags |= lsr & LSR_SAVE_FLAGS;
+	lsr = serial_lsr_in(up);
 	spin_unlock_irqrestore(&port->lock, flags);
 
 	serial8250_rpm_put(up);
@@ -2078,9 +2073,7 @@ static void wait_for_lsr(struct uart_8250_port *up, int bits)
 
 	/* Wait up to 10ms for the character(s) to be sent. */
 	for (;;) {
-		status = serial_in(up, UART_LSR);
-
-		up->lsr_saved_flags |= status & LSR_SAVE_FLAGS;
+		status = serial_lsr_in(up);
 
 		if ((status & bits) == bits)
 			break;
-- 
2.30.2

[PATCH v4 3/6] serial: 8250: Get preserved flags using serial_lsr_in()

[Ilpo Järvinen]

serial8250_handle_irq() assumes it's the first to read LSR register.
However, there are 8250 drivers which perform LSR read in their own irq
handler prior to calling serial8250_handle_irq(). As not all flags are
preserved across LSR reads, use serial_lsr_in() helper to get all the
preserved flags.

This commit might fix other commits too besides the ones for DW UART
mentioned below. It's just not clear to me which of the other devices
clear some of the LSR flags on read. AFAIK, nobody has complained about
this problem (either against DW or other devices) so it might not have
that bad impact in the end.

Fixes: 424d79183af0 ("serial: 8250_dw: Avoid "too much work" from bogus rx timeout interrupt")
Fixes: aa63d786cea2 ("serial: 8250: dw: Add support for DMA flow controlling devices")
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
---
 drivers/tty/serial/8250/8250_port.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
index ec5abeb638eb..a0ea048eb2ad 100644
--- a/drivers/tty/serial/8250/8250_port.c
+++ b/drivers/tty/serial/8250/8250_port.c
@@ -1916,7 +1916,7 @@ int serial8250_handle_irq(struct uart_port *port, unsigned int iir)
 
 	spin_lock_irqsave(&port->lock, flags);
 
-	status = serial_port_in(port, UART_LSR);
+	status = serial_lsr_in(up);
 
 	/*
 	 * If port is stopped and there are no error conditions in the
-- 
2.30.2

[PATCH v4 4/6] serial: 8250: Adjust misleading LSR related comment

[Ilpo Järvinen]

serial8250_rx_chars() has max_count based character limit. If it
triggers, the function returns the old LSR value (and it has never
returned only flags which were not handled). Adjust the comment to
match behavior and warn about which flags can be depended on.

I'd have moved LSR read before LSR read and used serial_lsr_in() also
here but I came across an old discussion about the topic. That
discussion generated commit d22f8f10683c ("serial: 8250: Fix lost rx
state") so I left the code as it is (it works as long as the callers
only use a subset of the LSR flags which holds true today) and changed
the comment instead.

Link: https://www.spinics.net/lists/linux-serial/msg16220.html
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
---
 drivers/tty/serial/8250/8250_port.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
index a0ea048eb2ad..c860f5964138 100644
--- a/drivers/tty/serial/8250/8250_port.c
+++ b/drivers/tty/serial/8250/8250_port.c
@@ -1782,9 +1782,11 @@ void serial8250_read_char(struct uart_8250_port *up, unsigned char lsr)
 EXPORT_SYMBOL_GPL(serial8250_read_char);
 
 /*
- * serial8250_rx_chars: processes according to the passed in LSR
- * value, and returns the remaining LSR bits not handled
- * by this Rx routine.
+ * serial8250_rx_chars - Read characters. The first LSR value must be passed in.
+ *
+ * Returns LSR bits. The caller should rely only on non-Rx related LSR bits
+ * (such as THRE) because the LSR value might come from an already consumed
+ * character.
  */
 unsigned char serial8250_rx_chars(struct uart_8250_port *up, unsigned char lsr)
 {
-- 
2.30.2

[PATCH v4 5/6] serial: 8250_dw: Use serial_lsr_in() in dw8250_handle_irq()

[Ilpo Järvinen]

dw8250_handle_irq() reads LSR under a few conditions, convert both to
use serial_lsr_in() in order to preserve LSR flags properly across
reads.

Cc: Douglas Anderson <dianders@chromium.org>
Cc: Phil Edworthy <phil.edworthy@renesas.com>
Cc: Miquel Raynal <miquel.raynal@bootlin.com>
Fixes: 424d79183af0 ("serial: 8250_dw: Avoid "too much work" from bogus rx timeout interrupt")
Fixes: aa63d786cea2 ("serial: 8250: dw: Add support for DMA flow controlling devices")
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
---
 drivers/tty/serial/8250/8250_dw.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/serial/8250/8250_dw.c b/drivers/tty/serial/8250/8250_dw.c
index f57bbd32ef11..1fae45991812 100644
--- a/drivers/tty/serial/8250/8250_dw.c
+++ b/drivers/tty/serial/8250/8250_dw.c
@@ -253,7 +253,7 @@ static int dw8250_handle_irq(struct uart_port *p)
 	 */
 	if (!up->dma && rx_timeout) {
 		spin_lock_irqsave(&p->lock, flags);
-		status = p->serial_in(p, UART_LSR);
+		status = serial_lsr_in(up);
 
 		if (!(status & (UART_LSR_DR | UART_LSR_BI)))
 			(void) p->serial_in(p, UART_RX);
@@ -263,7 +263,7 @@ static int dw8250_handle_irq(struct uart_port *p)
 
 	/* Manually stop the Rx DMA transfer when acting as flow controller */
 	if (quirks & DW_UART_QUIRK_IS_DMA_FC && up->dma && up->dma->rx_running && rx_timeout) {
-		status = p->serial_in(p, UART_LSR);
+		status = serial_lsr_in(up);
 		if (status & (UART_LSR_DR | UART_LSR_BI)) {
 			dw8250_writel_ext(p, RZN1_UART_RDMACR, 0);
 			dw8250_writel_ext(p, DW_UART_DMASA, 1);
-- 
2.30.2

[PATCH v4 6/6] serial: 8250_dw: Store LSR into lsr_saved_flags in dw8250_tx_wait_empty()

[Ilpo Järvinen]

Make sure LSR flags are preserved in dw8250_tx_wait_empty(). This
function is called from a low-level out function and therefore cannot
call serial_lsr_in() as it would lead to infinite recursion.

It is borderline if the flags need to be saved here at all since this
code relates to writing LCR register which usually implies no important
characters should be arriving.

Fixes: 914eaf935ec7 ("serial: 8250_dw: Allow TX FIFO to drain before writing to UART_LCR")
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
---
 drivers/tty/serial/8250/8250_dw.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/tty/serial/8250/8250_dw.c b/drivers/tty/serial/8250/8250_dw.c
index 1fae45991812..4cc69bb612ab 100644
--- a/drivers/tty/serial/8250/8250_dw.c
+++ b/drivers/tty/serial/8250/8250_dw.c
@@ -122,12 +122,15 @@ static void dw8250_check_lcr(struct uart_port *p, int value)
 /* Returns once the transmitter is empty or we run out of retries */
 static void dw8250_tx_wait_empty(struct uart_port *p)
 {
+	struct uart_8250_port *up = up_to_u8250p(p);
 	unsigned int tries = 20000;
 	unsigned int delay_threshold = tries - 1000;
 	unsigned int lsr;
 
 	while (tries--) {
 		lsr = readb (p->membase + (UART_LSR << p->regshift));
+		up->lsr_saved_flags |= lsr & LSR_SAVE_FLAGS;
+
 		if (lsr & UART_LSR_TEMT)
 			break;
 
-- 
2.30.2

Re: [PATCH v4 1/6] serial: 8250: Store to lsr_save_flags after lsr read

[Andy Shevchenko]

On Wed, Jun 8, 2022 at 12:44 PM Ilpo Järvinen
<ilpo.jarvinen@linux.intel.com> wrote:
>
> Not all LSR register flags are preserved across reads. Therefore, LSR
> readers must store the non-preserved bits into lsr_save_flags.
>
> This fix was initially mixed into feature commit f6f586102add ("serial:
> 8250: Handle UART without interrupt on TEMT using em485"). However,
> that feature change had a flaw and it was reverted to make room for
> simpler approach providing the same feature. The embedded fix got
> reverted with the feature change.
>
> Re-add the lsr_save_flags fix and properly mark it's a fix.

Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>

> Fixes: e490c9144cfa ("tty: Add software emulated RS485 support for 8250")
> Link: https://lore.kernel.org/all/1d6c31d-d194-9e6a-ddf9-5f29af829f3@linux.intel.com/T/#m1737eef986bd20cf19593e344cebd7b0244945fc
> Co-developed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
> ---
>  drivers/tty/serial/8250/8250_port.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
> index 4998799abae2..c5e0f925f4b6 100644
> --- a/drivers/tty/serial/8250/8250_port.c
> +++ b/drivers/tty/serial/8250/8250_port.c
> @@ -1511,6 +1511,8 @@ static inline void __stop_tx(struct uart_8250_port *p)
>                 unsigned char lsr = serial_in(p, UART_LSR);
>                 u64 stop_delay = 0;
>
> +               p->lsr_saved_flags |= lsr & LSR_SAVE_FLAGS;
> +
>                 if (!(lsr & UART_LSR_THRE))
>                         return;
>                 /*
> --
> 2.30.2
>


-- 
With Best Regards,
Andy Shevchenko