Message ID | 79db8e82aadb0e174bc82b9996423c3503c8fb37.1680732084.git.jan.kundrat@cesnet.cz |
---|---|
State | New |
Headers | show |
Series | serial: max310x: fix IO data corruption in batched operations | expand |
On Thu, Apr 6, 2023 at 1:08 AM Jan Kundrát <jan.kundrat@cesnet.cz> wrote: > > After upgrading from 5.16 to 6.1, our board with a MAX14830 started > producing lots of garbage data over UART. Bisection pointed out commit > 285e76fc049c as the culprit. That patch tried to replace hand-written > code which I added in 2b4bac48c1084 ("serial: max310x: Use batched reads > when reasonably safe") with the generic regmap infrastructure for > batched operations. > > Unfortunately, the `regmap_raw_read` and `regmap_raw_write` which were > used are actually functions which perform IO over *multiple* registers. > That's not what is needed for accessing these Tx/Rx FIFOs; the > appropriate functions are the `_noinc_` versions, not the `_raw_` ones. > > Fix this regression by using `regmap_noinc_read()` and > `regmap_noinc_write()` along with the necessary `regmap_config` setup; > with this patch in place, our board communicates happily again. Since > our board uses SPI for talking to this chip, the I2C part is completely > untested. Make sense, thanks for fixing this! Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com> > Fixes: 285e76fc049c serial: max310x: use regmap methods for SPI batch operations > Signed-off-by: Jan Kundrát <jan.kundrat@cesnet.cz> > Cc: stable@vger.kernel.org > --- > drivers/tty/serial/max310x.c | 17 +++++++++++++++-- > 1 file changed, 15 insertions(+), 2 deletions(-) > > diff --git a/drivers/tty/serial/max310x.c b/drivers/tty/serial/max310x.c > index c82391c928cb..47520d4a381f 100644 > --- a/drivers/tty/serial/max310x.c > +++ b/drivers/tty/serial/max310x.c > @@ -529,6 +529,11 @@ static bool max310x_reg_precious(struct device *dev, unsigned int reg) > return false; > } > > +static bool max310x_reg_noinc(struct device *dev, unsigned int reg) > +{ > + return reg == MAX310X_RHR_REG; > +} > + > static int max310x_set_baud(struct uart_port *port, int baud) > { > unsigned int mode = 0, div = 0, frac = 0, c = 0, F = 0; > @@ -658,14 +663,14 @@ static void max310x_batch_write(struct uart_port *port, u8 *txbuf, unsigned int > { > struct max310x_one *one = to_max310x_port(port); > > - regmap_raw_write(one->regmap, MAX310X_THR_REG, txbuf, len); > + regmap_noinc_write(one->regmap, MAX310X_THR_REG, txbuf, len); > } > > static void max310x_batch_read(struct uart_port *port, u8 *rxbuf, unsigned int len) > { > struct max310x_one *one = to_max310x_port(port); > > - regmap_raw_read(one->regmap, MAX310X_RHR_REG, rxbuf, len); > + regmap_noinc_read(one->regmap, MAX310X_RHR_REG, rxbuf, len); > } > > static void max310x_handle_rx(struct uart_port *port, unsigned int rxlen) > @@ -1480,6 +1485,10 @@ static struct regmap_config regcfg = { > .writeable_reg = max310x_reg_writeable, > .volatile_reg = max310x_reg_volatile, > .precious_reg = max310x_reg_precious, > + .writeable_noinc_reg = max310x_reg_noinc, > + .readable_noinc_reg = max310x_reg_noinc, > + .max_raw_read = MAX310X_FIFO_SIZE, > + .max_raw_write = MAX310X_FIFO_SIZE, > }; > > #ifdef CONFIG_SPI_MASTER > @@ -1565,6 +1574,10 @@ static struct regmap_config regcfg_i2c = { > .volatile_reg = max310x_reg_volatile, > .precious_reg = max310x_reg_precious, > .max_register = MAX310X_I2C_REVID_EXTREG, > + .writeable_noinc_reg = max310x_reg_noinc, > + .readable_noinc_reg = max310x_reg_noinc, > + .max_raw_read = MAX310X_FIFO_SIZE, > + .max_raw_write = MAX310X_FIFO_SIZE, > }; > > static const struct max310x_if_cfg max310x_i2c_if_cfg = { > -- > 2.39.2 > >
On Wed, 5 Apr 2023, Jan Kundrát wrote: > After upgrading from 5.16 to 6.1, our board with a MAX14830 started > producing lots of garbage data over UART. Bisection pointed out commit > 285e76fc049c as the culprit. That patch tried to replace hand-written > code which I added in 2b4bac48c1084 ("serial: max310x: Use batched reads > when reasonably safe") with the generic regmap infrastructure for > batched operations. > > Unfortunately, the `regmap_raw_read` and `regmap_raw_write` which were > used are actually functions which perform IO over *multiple* registers. > That's not what is needed for accessing these Tx/Rx FIFOs; the > appropriate functions are the `_noinc_` versions, not the `_raw_` ones. > > Fix this regression by using `regmap_noinc_read()` and > `regmap_noinc_write()` along with the necessary `regmap_config` setup; > with this patch in place, our board communicates happily again. Since > our board uses SPI for talking to this chip, the I2C part is completely > untested. > > Fixes: 285e76fc049c serial: max310x: use regmap methods for SPI batch operations > Signed-off-by: Jan Kundrát <jan.kundrat@cesnet.cz> > Cc: stable@vger.kernel.org > --- > drivers/tty/serial/max310x.c | 17 +++++++++++++++-- > 1 file changed, 15 insertions(+), 2 deletions(-) > > diff --git a/drivers/tty/serial/max310x.c b/drivers/tty/serial/max310x.c > index c82391c928cb..47520d4a381f 100644 > --- a/drivers/tty/serial/max310x.c > +++ b/drivers/tty/serial/max310x.c > @@ -529,6 +529,11 @@ static bool max310x_reg_precious(struct device *dev, unsigned int reg) > return false; > } > > +static bool max310x_reg_noinc(struct device *dev, unsigned int reg) > +{ > + return reg == MAX310X_RHR_REG; I'd either add a comment or just do || reg == MAX310X_THR_REG (which probably gets optimized away by the compiler) to make it look less magic for the Tx side.
diff --git a/drivers/tty/serial/max310x.c b/drivers/tty/serial/max310x.c index c82391c928cb..47520d4a381f 100644 --- a/drivers/tty/serial/max310x.c +++ b/drivers/tty/serial/max310x.c @@ -529,6 +529,11 @@ static bool max310x_reg_precious(struct device *dev, unsigned int reg) return false; } +static bool max310x_reg_noinc(struct device *dev, unsigned int reg) +{ + return reg == MAX310X_RHR_REG; +} + static int max310x_set_baud(struct uart_port *port, int baud) { unsigned int mode = 0, div = 0, frac = 0, c = 0, F = 0; @@ -658,14 +663,14 @@ static void max310x_batch_write(struct uart_port *port, u8 *txbuf, unsigned int { struct max310x_one *one = to_max310x_port(port); - regmap_raw_write(one->regmap, MAX310X_THR_REG, txbuf, len); + regmap_noinc_write(one->regmap, MAX310X_THR_REG, txbuf, len); } static void max310x_batch_read(struct uart_port *port, u8 *rxbuf, unsigned int len) { struct max310x_one *one = to_max310x_port(port); - regmap_raw_read(one->regmap, MAX310X_RHR_REG, rxbuf, len); + regmap_noinc_read(one->regmap, MAX310X_RHR_REG, rxbuf, len); } static void max310x_handle_rx(struct uart_port *port, unsigned int rxlen) @@ -1480,6 +1485,10 @@ static struct regmap_config regcfg = { .writeable_reg = max310x_reg_writeable, .volatile_reg = max310x_reg_volatile, .precious_reg = max310x_reg_precious, + .writeable_noinc_reg = max310x_reg_noinc, + .readable_noinc_reg = max310x_reg_noinc, + .max_raw_read = MAX310X_FIFO_SIZE, + .max_raw_write = MAX310X_FIFO_SIZE, }; #ifdef CONFIG_SPI_MASTER @@ -1565,6 +1574,10 @@ static struct regmap_config regcfg_i2c = { .volatile_reg = max310x_reg_volatile, .precious_reg = max310x_reg_precious, .max_register = MAX310X_I2C_REVID_EXTREG, + .writeable_noinc_reg = max310x_reg_noinc, + .readable_noinc_reg = max310x_reg_noinc, + .max_raw_read = MAX310X_FIFO_SIZE, + .max_raw_write = MAX310X_FIFO_SIZE, }; static const struct max310x_if_cfg max310x_i2c_if_cfg = {
After upgrading from 5.16 to 6.1, our board with a MAX14830 started producing lots of garbage data over UART. Bisection pointed out commit 285e76fc049c as the culprit. That patch tried to replace hand-written code which I added in 2b4bac48c1084 ("serial: max310x: Use batched reads when reasonably safe") with the generic regmap infrastructure for batched operations. Unfortunately, the `regmap_raw_read` and `regmap_raw_write` which were used are actually functions which perform IO over *multiple* registers. That's not what is needed for accessing these Tx/Rx FIFOs; the appropriate functions are the `_noinc_` versions, not the `_raw_` ones. Fix this regression by using `regmap_noinc_read()` and `regmap_noinc_write()` along with the necessary `regmap_config` setup; with this patch in place, our board communicates happily again. Since our board uses SPI for talking to this chip, the I2C part is completely untested. Fixes: 285e76fc049c serial: max310x: use regmap methods for SPI batch operations Signed-off-by: Jan Kundrát <jan.kundrat@cesnet.cz> Cc: stable@vger.kernel.org --- drivers/tty/serial/max310x.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)