During development of a new hardware an issue was seen where writes to
UART_LCR can result in characters that are currently held in the TX FIFO
being lost rather than sent, even if the userspace process has attempted to
flush them.
This is most visible when using the "resize" command (tested on Busybox),
where we have observed the escape code for restoring cursor position
becoming mangled.
This is the same issue as addressed in
commit 914eaf935ec7 ("serial: 8250_dw: Allow TX FIFO to drain before
writing to UART_LCR")
for the Armada 38x serial port. This patch makes the fix more generic
allowing it to be applied on any similar UART.
Richard Laing (2):
serial: 8250_dw: Allow TX FIFO to drain before writing to UART_LCR An
issue has been observed on the Broadcom BCM56160 serial port which
appears closely related to a similar issue on the Marvell Armada 38x
serial port.
dt-bindings: serial: snps-dw-apb-uart: Add property to drain TX FIFO
.../bindings/serial/snps-dw-apb-uart.yaml | 6 ++++++
drivers/tty/serial/8250/8250_dw.c | 18 ++++++++++++++++++
drivers/tty/serial/8250/8250_dwlib.h | 1 +
3 files changed, 25 insertions(+)
--
2.42.0
Writes to UART_LCR can result in characters that are currently held in the
TX FIFO being lost rather than sent, even if the userspace process has
attempted to flush them.
This is most visible when using the "resize" command (tested on Busybox),
where we have observed the escape code for restoring cursor position
becoming mangled.
Since this appears to be a more common problem add a new driver option
to flush the TX FIFO before writing to the UART_LCR.
Signed-off-by: Richard Laing <[email protected]>
---
drivers/tty/serial/8250/8250_dw.c | 18 ++++++++++++++++++
drivers/tty/serial/8250/8250_dwlib.h | 1 +
2 files changed, 19 insertions(+)
diff --git a/drivers/tty/serial/8250/8250_dw.c b/drivers/tty/serial/8250/8250_dw.c
index f4cafca1a7da..17ee824294c7 100644
--- a/drivers/tty/serial/8250/8250_dw.c
+++ b/drivers/tty/serial/8250/8250_dw.c
@@ -161,6 +161,10 @@ static void dw8250_serial_out(struct uart_port *p, int offset, int value)
{
struct dw8250_data *d = to_dw8250_data(p->private_data);
+ /* Allow the TX to drain before we reconfigure */
+ if (offset == UART_LCR && d->drain_before_lcr_change)
+ dw8250_tx_wait_empty(p);
+
writeb(value, p->membase + (offset << p->regshift));
if (offset == UART_LCR && !d->uart_16550_compatible)
@@ -197,6 +201,10 @@ static void dw8250_serial_outq(struct uart_port *p, int offset, int value)
{
struct dw8250_data *d = to_dw8250_data(p->private_data);
+ /* Allow the TX to drain before we reconfigure */
+ if (offset == UART_LCR && d->drain_before_lcr_change)
+ dw8250_tx_wait_empty(p);
+
value &= 0xff;
__raw_writeq(value, p->membase + (offset << p->regshift));
/* Read back to ensure register write ordering. */
@@ -211,6 +219,10 @@ static void dw8250_serial_out32(struct uart_port *p, int offset, int value)
{
struct dw8250_data *d = to_dw8250_data(p->private_data);
+ /* Allow the TX to drain before we reconfigure */
+ if (offset == UART_LCR && d->drain_before_lcr_change)
+ dw8250_tx_wait_empty(p);
+
writel(value, p->membase + (offset << p->regshift));
if (offset == UART_LCR && !d->uart_16550_compatible)
@@ -228,6 +240,10 @@ static void dw8250_serial_out32be(struct uart_port *p, int offset, int value)
{
struct dw8250_data *d = to_dw8250_data(p->private_data);
+ /* Allow the TX to drain before we reconfigure */
+ if (offset == UART_LCR && d->drain_before_lcr_change)
+ dw8250_tx_wait_empty(p);
+
iowrite32be(value, p->membase + (offset << p->regshift));
if (offset == UART_LCR && !d->uart_16550_compatible)
@@ -597,6 +613,8 @@ static int dw8250_probe(struct platform_device *pdev)
/* Always ask for fixed clock rate from a property. */
device_property_read_u32(dev, "clock-frequency", &p->uartclk);
+ data->drain_before_lcr_change = device_property_read_bool(dev, "drain-before-lcr-change");
+
/* If there is separate baudclk, get the rate from it. */
data->clk = devm_clk_get_optional(dev, "baudclk");
if (data->clk == NULL)
diff --git a/drivers/tty/serial/8250/8250_dwlib.h b/drivers/tty/serial/8250/8250_dwlib.h
index f13e91f2cace..f7d88fa8f058 100644
--- a/drivers/tty/serial/8250/8250_dwlib.h
+++ b/drivers/tty/serial/8250/8250_dwlib.h
@@ -45,6 +45,7 @@ struct dw8250_data {
unsigned int skip_autocfg:1;
unsigned int uart_16550_compatible:1;
+ unsigned int drain_before_lcr_change:1;
};
void dw8250_do_set_termios(struct uart_port *p, struct ktermios *termios, const struct ktermios *old);
--
2.42.0
An issue has been observed on the Broadcom BCM56160 serial port which
appears closely related to a similar issue on the Marvell Armada 38x
serial port.
Add a new property to force the TX FIFO to be drained before
changing the UART_LCR.
Signed-off-by: Richard Laing <[email protected]>
---
.../devicetree/bindings/serial/snps-dw-apb-uart.yaml | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml b/Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml
index 17c553123f96..4266ef96832c 100644
--- a/Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml
+++ b/Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml
@@ -114,6 +114,12 @@ properties:
register. Define this if your serial port does not use this pin.
type: boolean
+ drain-before-lcr-change:
+ description: Force TX buffer flush before LCR change. Make sure all
+ characters in the buffer are sent before reconfiguring. Define this if
+ the UART drops its FIFO when reconfiguring.
+ type: boolean
+
required:
- compatible
- reg
--
2.42.0
On Mon, 16 Oct 2023 14:32:07 +1300, Richard Laing wrote:
> An issue has been observed on the Broadcom BCM56160 serial port which
> appears closely related to a similar issue on the Marvell Armada 38x
> serial port.
>
> Add a new property to force the TX FIFO to be drained before
> changing the UART_LCR.
>
> Signed-off-by: Richard Laing <[email protected]>
> ---
> .../devicetree/bindings/serial/snps-dw-apb-uart.yaml | 6 ++++++
> 1 file changed, 6 insertions(+)
>
My bot found errors running 'make DT_CHECKER_FLAGS=-m dt_binding_check'
on your patch (DT_CHECKER_FLAGS is new in v5.13):
yamllint warnings/errors:
./Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml:120:5: [error] syntax error: could not find expected ':' (syntax)
dtschema/dtc warnings/errors:
make[2]: *** Deleting file 'Documentation/devicetree/bindings/serial/snps-dw-apb-uart.example.dts'
Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml:120:5: could not find expected ':'
make[2]: *** [Documentation/devicetree/bindings/Makefile:26: Documentation/devicetree/bindings/serial/snps-dw-apb-uart.example.dts] Error 1
make[2]: *** Waiting for unfinished jobs....
./Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml:120:5: could not find expected ':'
/builds/robherring/dt-review-ci/linux/Documentation/devicetree/bindings/serial/snps-dw-apb-uart.yaml: ignoring, error parsing file
make[1]: *** [/builds/robherring/dt-review-ci/linux/Makefile:1427: dt_binding_check] Error 2
make: *** [Makefile:234: __sub-make] Error 2
doc reference errors (make refcheckdocs):
See https://patchwork.ozlabs.org/project/devicetree-bindings/patch/[email protected]
The base for the series is generally the latest rc1. A different dependency
should be noted in *this* patch.
If you already ran 'make dt_binding_check' and didn't see the above
error(s), then make sure 'yamllint' is installed and dt-schema is up to
date:
pip3 install dtschema --upgrade
Please check and re-submit after running the above command yourself. Note
that DT_SCHEMA_FILES can be set to your schema file to speed up checking
your schema. However, it must be unset to test all examples with your schema.
On Mon, 16 Oct 2023, Richard Laing wrote:
Your subject line is way too long. If you refer to some other issue,
please link to it properly with commit id and/or with Link: tags.
> Writes to UART_LCR can result in characters that are currently held in the
> TX FIFO being lost rather than sent, even if the userspace process has
> attempted to flush them.
>
> This is most visible when using the "resize" command (tested on Busybox),
> where we have observed the escape code for restoring cursor position
> becoming mangled.
>
> Since this appears to be a more common problem add a new driver option
> to flush the TX FIFO before writing to the UART_LCR.
This looks like a problem we already have solution for, the userspace can
use TCSADRAIN/FLUSH to indicate what kind of flushing it wants for Tx
when it makes the tcsetattr() call. Thus, userspace can avoid the Tx side
corruption as long as its behavior is sane and doesn't e.g. try to race
writes with tcsetattr() call as mentioned in commit 094fb49a2d0d ("tty:
Prevent writing chars during tcsetattr TCSADRAIN/FLUSH").
Have you tried to use the userspace solution? Isn't it working for some
reason?
--
i.
>
> Signed-off-by: Richard Laing <[email protected]>
> ---
> drivers/tty/serial/8250/8250_dw.c | 18 ++++++++++++++++++
> drivers/tty/serial/8250/8250_dwlib.h | 1 +
> 2 files changed, 19 insertions(+)
>
> diff --git a/drivers/tty/serial/8250/8250_dw.c b/drivers/tty/serial/8250/8250_dw.c
> index f4cafca1a7da..17ee824294c7 100644
> --- a/drivers/tty/serial/8250/8250_dw.c
> +++ b/drivers/tty/serial/8250/8250_dw.c
> @@ -161,6 +161,10 @@ static void dw8250_serial_out(struct uart_port *p, int offset, int value)
> {
> struct dw8250_data *d = to_dw8250_data(p->private_data);
>
> + /* Allow the TX to drain before we reconfigure */
> + if (offset == UART_LCR && d->drain_before_lcr_change)
> + dw8250_tx_wait_empty(p);
> +
> writeb(value, p->membase + (offset << p->regshift));
>
> if (offset == UART_LCR && !d->uart_16550_compatible)
> @@ -197,6 +201,10 @@ static void dw8250_serial_outq(struct uart_port *p, int offset, int value)
> {
> struct dw8250_data *d = to_dw8250_data(p->private_data);
>
> + /* Allow the TX to drain before we reconfigure */
> + if (offset == UART_LCR && d->drain_before_lcr_change)
> + dw8250_tx_wait_empty(p);
> +
> value &= 0xff;
> __raw_writeq(value, p->membase + (offset << p->regshift));
> /* Read back to ensure register write ordering. */
> @@ -211,6 +219,10 @@ static void dw8250_serial_out32(struct uart_port *p, int offset, int value)
> {
> struct dw8250_data *d = to_dw8250_data(p->private_data);
>
> + /* Allow the TX to drain before we reconfigure */
> + if (offset == UART_LCR && d->drain_before_lcr_change)
> + dw8250_tx_wait_empty(p);
> +
> writel(value, p->membase + (offset << p->regshift));
>
> if (offset == UART_LCR && !d->uart_16550_compatible)
> @@ -228,6 +240,10 @@ static void dw8250_serial_out32be(struct uart_port *p, int offset, int value)
> {
> struct dw8250_data *d = to_dw8250_data(p->private_data);
>
> + /* Allow the TX to drain before we reconfigure */
> + if (offset == UART_LCR && d->drain_before_lcr_change)
> + dw8250_tx_wait_empty(p);
> +
> iowrite32be(value, p->membase + (offset << p->regshift));
>
> if (offset == UART_LCR && !d->uart_16550_compatible)
> @@ -597,6 +613,8 @@ static int dw8250_probe(struct platform_device *pdev)
> /* Always ask for fixed clock rate from a property. */
> device_property_read_u32(dev, "clock-frequency", &p->uartclk);
>
> + data->drain_before_lcr_change = device_property_read_bool(dev, "drain-before-lcr-change");
> +
> /* If there is separate baudclk, get the rate from it. */
> data->clk = devm_clk_get_optional(dev, "baudclk");
> if (data->clk == NULL)
> diff --git a/drivers/tty/serial/8250/8250_dwlib.h b/drivers/tty/serial/8250/8250_dwlib.h
> index f13e91f2cace..f7d88fa8f058 100644
> --- a/drivers/tty/serial/8250/8250_dwlib.h
> +++ b/drivers/tty/serial/8250/8250_dwlib.h
> @@ -45,6 +45,7 @@ struct dw8250_data {
>
> unsigned int skip_autocfg:1;
> unsigned int uart_16550_compatible:1;
> + unsigned int drain_before_lcr_change:1;
> };
>
> void dw8250_do_set_termios(struct uart_port *p, struct ktermios *termios, const struct ktermios *old);
On Mon, 16 Oct 2023, Ilpo J?rvinen wrote:
> On Mon, 16 Oct 2023, Richard Laing wrote:
>
> Your subject line is way too long. If you refer to some other issue,
> please link to it properly with commit id and/or with Link: tags.
>
> > Writes to UART_LCR can result in characters that are currently held in the
> > TX FIFO being lost rather than sent, even if the userspace process has
> > attempted to flush them.
> >
> > This is most visible when using the "resize" command (tested on Busybox),
> > where we have observed the escape code for restoring cursor position
> > becoming mangled.
> >
> > Since this appears to be a more common problem add a new driver option
> > to flush the TX FIFO before writing to the UART_LCR.
>
> This looks like a problem we already have solution for, the userspace can
> use TCSADRAIN/FLUSH to indicate what kind of flushing it wants for Tx
> when it makes the tcsetattr() call. Thus, userspace can avoid the Tx side
> corruption as long as its behavior is sane and doesn't e.g. try to race
> writes with tcsetattr() call as mentioned in commit 094fb49a2d0d ("tty:
> Prevent writing chars during tcsetattr TCSADRAIN/FLUSH").
I'm sorry, it was actually mentioned in commit 146a37e05d62 ("serial:
8250: Fix serial8250_tx_empty() race with DMA Tx") although that
094fb49a2d0d is also related to the draining.
> Have you tried to use the userspace solution? Isn't it working for some
> reason?
--
i.