Implement a get_multiple function for gpio-pca953x. If a driver
leaves get_multiple unimplemented then gpio_chip_get_multiple()
in gpiolib.c takes care of it by calling chip->get() as needed.
For i2c chips this is very inefficient. For example if you do an
8-bit read then instead of a single i2c transaction there are
8 transactions reading the same byte!
This has been tested with max7312 chips on a 5.2 kernel.
Signed-off-by: Paul Thomas <[email protected]>
---
changes from v1: rebased to 5.7-rc1
drivers/gpio/gpio-pca953x.c | 37 +++++++++++++++++++++++++++++++++++++
1 file changed, 37 insertions(+)
diff --git a/drivers/gpio/gpio-pca953x.c b/drivers/gpio/gpio-pca953x.c
index 5638b4e5355f..6317510b0dc3 100644
--- a/drivers/gpio/gpio-pca953x.c
+++ b/drivers/gpio/gpio-pca953x.c
@@ -115,6 +115,7 @@ MODULE_DEVICE_TABLE(acpi, pca953x_acpi_ids);
#define MAX_BANK 5
#define BANK_SZ 8
+#define BANK_SFT 3 /* ilog2(BANK_SZ) */
#define MAX_LINE (MAX_BANK * BANK_SZ)
#define NBANK(chip) DIV_ROUND_UP(chip->gpio_chip.ngpio, BANK_SZ)
@@ -466,6 +467,41 @@ static int pca953x_gpio_get_direction(struct gpio_chip *gc, unsigned off)
return GPIO_LINE_DIRECTION_OUT;
}
+static int pca953x_gpio_get_multiple(struct gpio_chip *gc,
+ unsigned long *mask, unsigned long *bits)
+{
+ struct pca953x_chip *chip = gpiochip_get_data(gc);
+ unsigned int reg_val;
+ int offset, value, i, ret = 0;
+ u8 inreg;
+
+ /* Force offset outside the range of i so that
+ * at least the first relevant register is read
+ */
+ offset = gc->ngpio;
+ for_each_set_bit(i, mask, gc->ngpio) {
+ /* whenever i goes into a new bank update inreg
+ * and read the register
+ */
+ if ((offset >> BANK_SFT) != (i >> BANK_SFT)) {
+ offset = i;
+ inreg = pca953x_recalc_addr(chip, chip->regs->input,
+ offset, true, false);
+ mutex_lock(&chip->i2c_lock);
+ ret = regmap_read(chip->regmap, inreg, ®_val);
+ mutex_unlock(&chip->i2c_lock);
+ if (ret < 0)
+ return ret;
+ }
+ /* reg_val is relative to the last read byte,
+ * so only shift the relative bits
+ */
+ value = (reg_val >> (i % 8)) & 0x01;
+ __assign_bit(i, bits, value);
+ }
+ return ret;
+}
+
static void pca953x_gpio_set_multiple(struct gpio_chip *gc,
unsigned long *mask, unsigned long *bits)
{
@@ -551,6 +587,7 @@ static void pca953x_setup_gpio(struct pca953x_chip *chip, int gpios)
gc->get = pca953x_gpio_get_value;
gc->set = pca953x_gpio_set_value;
gc->get_direction = pca953x_gpio_get_direction;
+ gc->get_multiple = pca953x_gpio_get_multiple;
gc->set_multiple = pca953x_gpio_set_multiple;
gc->set_config = pca953x_gpio_set_config;
gc->can_sleep = true;
--
2.17.1
On Tue, Apr 14, 2020 at 5:30 PM Paul Thomas <[email protected]> wrote:
> Implement a get_multiple function for gpio-pca953x. If a driver
> leaves get_multiple unimplemented then gpio_chip_get_multiple()
> in gpiolib.c takes care of it by calling chip->get() as needed.
> For i2c chips this is very inefficient. For example if you do an
> 8-bit read then instead of a single i2c transaction there are
> 8 transactions reading the same byte!
>
> This has been tested with max7312 chips on a 5.2 kernel.
>
> Signed-off-by: Paul Thomas <[email protected]>
> ---
> changes from v1: rebased to 5.7-rc1
Acked-by: Linus Walleij <[email protected]>
Since I know Bartosz is queueing other patches for this driver I
let him pick it up.
Yours,
Linus Walleij
czw., 16 kwi 2020 o 13:22 Linus Walleij <[email protected]> napisał(a):
>
> On Tue, Apr 14, 2020 at 5:30 PM Paul Thomas <[email protected]> wrote:
>
> > Implement a get_multiple function for gpio-pca953x. If a driver
> > leaves get_multiple unimplemented then gpio_chip_get_multiple()
> > in gpiolib.c takes care of it by calling chip->get() as needed.
> > For i2c chips this is very inefficient. For example if you do an
> > 8-bit read then instead of a single i2c transaction there are
> > 8 transactions reading the same byte!
> >
> > This has been tested with max7312 chips on a 5.2 kernel.
> >
> > Signed-off-by: Paul Thomas <[email protected]>
> > ---
> > changes from v1: rebased to 5.7-rc1
>
> Acked-by: Linus Walleij <[email protected]>
>
> Since I know Bartosz is queueing other patches for this driver I
> let him pick it up.
>
> Yours,
> Linus Walleij
Patch applied. I removed the last line of the commit message since I
guess you did test it on v5.7-rc1 after all?
Bart
On Thu, Apr 16, 2020 at 9:21 AM Bartosz Golaszewski
<[email protected]> wrote:
>
> czw., 16 kwi 2020 o 13:22 Linus Walleij <[email protected]> napisał(a):
> >
> > On Tue, Apr 14, 2020 at 5:30 PM Paul Thomas <[email protected]> wrote:
> >
> > > Implement a get_multiple function for gpio-pca953x. If a driver
> > > leaves get_multiple unimplemented then gpio_chip_get_multiple()
> > > in gpiolib.c takes care of it by calling chip->get() as needed.
> > > For i2c chips this is very inefficient. For example if you do an
> > > 8-bit read then instead of a single i2c transaction there are
> > > 8 transactions reading the same byte!
> > >
> > > This has been tested with max7312 chips on a 5.2 kernel.
> > >
> > > Signed-off-by: Paul Thomas <[email protected]>
> > > ---
> > > changes from v1: rebased to 5.7-rc1
> >
> > Acked-by: Linus Walleij <[email protected]>
> >
> > Since I know Bartosz is queueing other patches for this driver I
> > let him pick it up.
> >
> > Yours,
> > Linus Walleij
>
> Patch applied. I removed the last line of the commit message since I
> guess you did test it on v5.7-rc1 after all?
I applied the patch and compiled the kernel, but I didn't test on our
embedded board with the actual max7312 chips, that board has a whole
series of special commits.
-Paul
>
> Bart