| Message ID | 20181215233823.1042-3-linus.walleij@linaro.org (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | SPI CS using GPIO descriptors | expand |
On Sun, 16 Dec 2018 00:38:18 +0100 Linus Walleij <linus.walleij@linaro.org> wrote: > This converts the ATH79 SPI master driver to use GPIO descriptors > for chip select handling. > > The ATH79 driver was requesting the GPIO and driving it from the > bitbang .chipselect callback. Do not request it anymore as the SPI > core will request it, remove the line inversion semantics for the > GPIO case (handled by gpiolib) and let the SPI core deal with > requesting the GPIO line from the device tree node of the controller. > > This driver can be instantiated from a board file (no device tree) > but the board files only use native CS (no GPIO lines) so we should > be fine just letting the SPI core grab the GPIO from the device. > > The fact that the driver is actively driving the GPIO in the > ath79_spi_chipselect() callback is confusing since the host does > not set SPI_MASTER_GPIO_SS so this should not ever get called when > using GPIO CS. I put in a comment about this. The SPI_MASTER_GPIO_SS flag came after this driver has been written, so I don't think we need to worry about this. > Cc: Felix Fietkau <nbd@nbd.name> > Cc: Alban Bedel <albeu@free.fr> > Cc: Linuxarm <linuxarm@huawei.com> > Signed-off-by: Linus Walleij <linus.walleij@linaro.org> > --- > drivers/spi/spi-ath79.c | 42 > ++++++++++++++--------------------------- 1 file changed, 14 > insertions(+), 28 deletions(-) > > diff --git a/drivers/spi/spi-ath79.c b/drivers/spi/spi-ath79.c > index 3f6b657394de..ed1068ac055f 100644 > --- a/drivers/spi/spi-ath79.c > +++ b/drivers/spi/spi-ath79.c > @@ -21,7 +21,7 @@ > #include <linux/spi/spi.h> > #include <linux/spi/spi_bitbang.h> > #include <linux/bitops.h> > -#include <linux/gpio.h> > +#include <linux/gpio/consumer.h> > #include <linux/clk.h> > #include <linux/err.h> > > @@ -78,9 +78,16 @@ static void ath79_spi_chipselect(struct spi_device > *spi, int is_active) ath79_spi_wr(sp, AR71XX_SPI_REG_IOC, > sp->ioc_base); } > > - if (gpio_is_valid(spi->cs_gpio)) { > - /* SPI is normally active-low */ > - gpio_set_value_cansleep(spi->cs_gpio, cs_high); > + if (spi->cs_gpiod) { > + /* > + * SPI chipselect is normally active-low, but > + * inversion semantics are handled by gpiolib. > + * > + * FIXME: is this ever used? The driver doesn't > + * set SPI_MASTER_GPIO_SS so this callback should not > + * get called if a CS GPIO is found by the SPI core. > + */ > + gpiod_set_value_cansleep(spi->cs_gpiod, is_active); AFAICT this whole block can be removed. > } else { > u32 cs_bit = AR71XX_SPI_IOC_CS(spi->chip_select); > > @@ -118,21 +125,8 @@ static void ath79_spi_disable(struct ath79_spi > *sp) static int ath79_spi_setup_cs(struct spi_device *spi) > { > struct ath79_spi *sp = ath79_spidev_to_sp(spi); > - int status; > > - status = 0; > - if (gpio_is_valid(spi->cs_gpio)) { > - unsigned long flags; > - > - flags = GPIOF_DIR_OUT; > - if (spi->mode & SPI_CS_HIGH) > - flags |= GPIOF_INIT_LOW; > - else > - flags |= GPIOF_INIT_HIGH; > - > - status = gpio_request_one(spi->cs_gpio, flags, > - dev_name(&spi->dev)); > - } else { > + if (!spi->cs_gpiod) { > u32 cs_bit = AR71XX_SPI_IOC_CS(spi->chip_select); > > if (spi->mode & SPI_CS_HIGH) The bitbang code already take care to properly configure CS in spi_bitbang_setup(), if the core now take care of the GPIO setup we can probably remove this whole function. > @@ -143,13 +137,7 @@ static int ath79_spi_setup_cs(struct spi_device > *spi) ath79_spi_wr(sp, AR71XX_SPI_REG_IOC, sp->ioc_base); > } > > - return status; > -} > - > -static void ath79_spi_cleanup_cs(struct spi_device *spi) > -{ > - if (gpio_is_valid(spi->cs_gpio)) > - gpio_free(spi->cs_gpio); > + return 0; > } > > static int ath79_spi_setup(struct spi_device *spi) > @@ -163,15 +151,12 @@ static int ath79_spi_setup(struct spi_device > *spi) } > > status = spi_bitbang_setup(spi); > - if (status && !spi->controller_state) > - ath79_spi_cleanup_cs(spi); > > return status; > } Now if we can remove ath79_spi_setup_cs() and ath79_spi_cleanup_cs() we can also remove ath79_spi_setup() and directly use spi_bitbang_setup(). That would remove quiet a bit of code, very nice! > static void ath79_spi_cleanup(struct spi_device *spi) > { > - ath79_spi_cleanup_cs(spi); > spi_bitbang_cleanup(spi); > } Here we could also remove this function and directly use spi_bitbang_cleanup(). > @@ -225,6 +210,7 @@ static int ath79_spi_probe(struct platform_device > *pdev) > pdata = dev_get_platdata(&pdev->dev); > > + master->use_gpio_descriptors = true; > master->bits_per_word_mask = SPI_BPW_RANGE_MASK(1, 32); > master->setup = ath79_spi_setup; > master->cleanup = ath79_spi_cleanup; I would really like to test this these changes, however I won't be able to do that before next week. Could you please also add me to the CC list in the first patch in the series for the next iteration? That would make testing easier. Alban
On Sun, 16 Dec 2018 00:38:18 +0100 Linus Walleij <linus.walleij@linaro.org> wrote: > This converts the ATH79 SPI master driver to use GPIO descriptors > for chip select handling. > > The ATH79 driver was requesting the GPIO and driving it from the > bitbang .chipselect callback. Do not request it anymore as the SPI > core will request it, remove the line inversion semantics for the > GPIO case (handled by gpiolib) and let the SPI core deal with > requesting the GPIO line from the device tree node of the controller. > > This driver can be instantiated from a board file (no device tree) > but the board files only use native CS (no GPIO lines) so we should > be fine just letting the SPI core grab the GPIO from the device. > > The fact that the driver is actively driving the GPIO in the > ath79_spi_chipselect() callback is confusing since the host does > not set SPI_MASTER_GPIO_SS so this should not ever get called when > using GPIO CS. I put in a comment about this. > > Cc: Felix Fietkau <nbd@nbd.name> > Cc: Alban Bedel <albeu@free.fr> > Cc: Linuxarm <linuxarm@huawei.com> > Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Trivial suggestion inline that may not be worth doing simply because it'll make for a less readable diff. Jonathan > --- > drivers/spi/spi-ath79.c | 42 ++++++++++++++--------------------------- > 1 file changed, 14 insertions(+), 28 deletions(-) > > diff --git a/drivers/spi/spi-ath79.c b/drivers/spi/spi-ath79.c > index 3f6b657394de..ed1068ac055f 100644 > --- a/drivers/spi/spi-ath79.c > +++ b/drivers/spi/spi-ath79.c > @@ -21,7 +21,7 @@ > #include <linux/spi/spi.h> > #include <linux/spi/spi_bitbang.h> > #include <linux/bitops.h> > -#include <linux/gpio.h> > +#include <linux/gpio/consumer.h> > #include <linux/clk.h> > #include <linux/err.h> > > @@ -78,9 +78,16 @@ static void ath79_spi_chipselect(struct spi_device *spi, int is_active) > ath79_spi_wr(sp, AR71XX_SPI_REG_IOC, sp->ioc_base); > } > > - if (gpio_is_valid(spi->cs_gpio)) { > - /* SPI is normally active-low */ > - gpio_set_value_cansleep(spi->cs_gpio, cs_high); > + if (spi->cs_gpiod) { > + /* > + * SPI chipselect is normally active-low, but > + * inversion semantics are handled by gpiolib. > + * > + * FIXME: is this ever used? The driver doesn't > + * set SPI_MASTER_GPIO_SS so this callback should not > + * get called if a CS GPIO is found by the SPI core. > + */ > + gpiod_set_value_cansleep(spi->cs_gpiod, is_active); > } else { > u32 cs_bit = AR71XX_SPI_IOC_CS(spi->chip_select); > > @@ -118,21 +125,8 @@ static void ath79_spi_disable(struct ath79_spi *sp) > static int ath79_spi_setup_cs(struct spi_device *spi) > { > struct ath79_spi *sp = ath79_spidev_to_sp(spi); > - int status; > > - status = 0; > - if (gpio_is_valid(spi->cs_gpio)) { > - unsigned long flags; > - > - flags = GPIOF_DIR_OUT; > - if (spi->mode & SPI_CS_HIGH) > - flags |= GPIOF_INIT_LOW; > - else > - flags |= GPIOF_INIT_HIGH; > - > - status = gpio_request_one(spi->cs_gpio, flags, > - dev_name(&spi->dev)); > - } else { > + if (!spi->cs_gpiod) { It would obviously not be a minimal change, but the resulting code would be slightly nicer if you were to flip this to exit quickly in the case where we do have descriptors. u32 cs_bit; if (spi->cs_gpiod) return 0; cs_bit = ... return 0; > u32 cs_bit = AR71XX_SPI_IOC_CS(spi->chip_select); > > if (spi->mode & SPI_CS_HIGH) > @@ -143,13 +137,7 @@ static int ath79_spi_setup_cs(struct spi_device *spi) > ath79_spi_wr(sp, AR71XX_SPI_REG_IOC, sp->ioc_base); > } > > - return status; > -} > - > -static void ath79_spi_cleanup_cs(struct spi_device *spi) > -{ > - if (gpio_is_valid(spi->cs_gpio)) > - gpio_free(spi->cs_gpio); > + return 0; > } > > static int ath79_spi_setup(struct spi_device *spi) > @@ -163,15 +151,12 @@ static int ath79_spi_setup(struct spi_device *spi) > } > > status = spi_bitbang_setup(spi); > - if (status && !spi->controller_state) > - ath79_spi_cleanup_cs(spi); > > return status; > } > > static void ath79_spi_cleanup(struct spi_device *spi) > { > - ath79_spi_cleanup_cs(spi); > spi_bitbang_cleanup(spi); > } > > @@ -225,6 +210,7 @@ static int ath79_spi_probe(struct platform_device *pdev) > > pdata = dev_get_platdata(&pdev->dev); > > + master->use_gpio_descriptors = true; > master->bits_per_word_mask = SPI_BPW_RANGE_MASK(1, 32); > master->setup = ath79_spi_setup; > master->cleanup = ath79_spi_cleanup;
Hi Alban, thanks a lot for your review! On Mon, Dec 17, 2018 at 10:02 AM Alban <albeu@free.fr> wrote: > AFAICT this whole block can be removed. (...) > > + if (!spi->cs_gpiod) { > > u32 cs_bit = AR71XX_SPI_IOC_CS(spi->chip_select); > > > > if (spi->mode & SPI_CS_HIGH) > > The bitbang code already take care to properly configure CS in > spi_bitbang_setup(), if the core now take care of the GPIO setup we can > probably remove this whole function. (...) > Now if we can remove ath79_spi_setup_cs() and ath79_spi_cleanup_cs() we > can also remove ath79_spi_setup() and directly use spi_bitbang_setup(). > That would remove quiet a bit of code, very nice! (...) > Here we could also remove this function and directly use > spi_bitbang_cleanup(). I see that you spot a whole bunch of improvements/cleanups we can do in this context, it'd be great if you could make a patch on top of this for those things, when you have time! (I don't really dare to since I can't test it on hardware.) Any testing would also be greatly appreciated! We target v4.22 for these initial SPI CS changes. Yours, Linus Walleij
On Mon, Dec 17, 2018 at 12:35 PM Jonathan Cameron <jonathan.cameron@huawei.com> wrote: > > - } else { > > + if (!spi->cs_gpiod) { > > It would obviously not be a minimal change, but the resulting code > would be slightly nicer if you were to flip this to exit quickly in > the case where we do have descriptors. > > u32 cs_bit; > > if (spi->cs_gpiod) > return 0; > > cs_bit = ... > > return 0; Hm I bet we can do that as a separate cleanup if this is even still there after all the cleanups spotted by Alban. Yours, Linus Walleij
diff --git a/drivers/spi/spi-ath79.c b/drivers/spi/spi-ath79.c index 3f6b657394de..ed1068ac055f 100644 --- a/drivers/spi/spi-ath79.c +++ b/drivers/spi/spi-ath79.c @@ -21,7 +21,7 @@ #include <linux/spi/spi.h> #include <linux/spi/spi_bitbang.h> #include <linux/bitops.h> -#include <linux/gpio.h> +#include <linux/gpio/consumer.h> #include <linux/clk.h> #include <linux/err.h> @@ -78,9 +78,16 @@ static void ath79_spi_chipselect(struct spi_device *spi, int is_active) ath79_spi_wr(sp, AR71XX_SPI_REG_IOC, sp->ioc_base); } - if (gpio_is_valid(spi->cs_gpio)) { - /* SPI is normally active-low */ - gpio_set_value_cansleep(spi->cs_gpio, cs_high); + if (spi->cs_gpiod) { + /* + * SPI chipselect is normally active-low, but + * inversion semantics are handled by gpiolib. + * + * FIXME: is this ever used? The driver doesn't + * set SPI_MASTER_GPIO_SS so this callback should not + * get called if a CS GPIO is found by the SPI core. + */ + gpiod_set_value_cansleep(spi->cs_gpiod, is_active); } else { u32 cs_bit = AR71XX_SPI_IOC_CS(spi->chip_select); @@ -118,21 +125,8 @@ static void ath79_spi_disable(struct ath79_spi *sp) static int ath79_spi_setup_cs(struct spi_device *spi) { struct ath79_spi *sp = ath79_spidev_to_sp(spi); - int status; - status = 0; - if (gpio_is_valid(spi->cs_gpio)) { - unsigned long flags; - - flags = GPIOF_DIR_OUT; - if (spi->mode & SPI_CS_HIGH) - flags |= GPIOF_INIT_LOW; - else - flags |= GPIOF_INIT_HIGH; - - status = gpio_request_one(spi->cs_gpio, flags, - dev_name(&spi->dev)); - } else { + if (!spi->cs_gpiod) { u32 cs_bit = AR71XX_SPI_IOC_CS(spi->chip_select); if (spi->mode & SPI_CS_HIGH) @@ -143,13 +137,7 @@ static int ath79_spi_setup_cs(struct spi_device *spi) ath79_spi_wr(sp, AR71XX_SPI_REG_IOC, sp->ioc_base); } - return status; -} - -static void ath79_spi_cleanup_cs(struct spi_device *spi) -{ - if (gpio_is_valid(spi->cs_gpio)) - gpio_free(spi->cs_gpio); + return 0; } static int ath79_spi_setup(struct spi_device *spi) @@ -163,15 +151,12 @@ static int ath79_spi_setup(struct spi_device *spi) } status = spi_bitbang_setup(spi); - if (status && !spi->controller_state) - ath79_spi_cleanup_cs(spi); return status; } static void ath79_spi_cleanup(struct spi_device *spi) { - ath79_spi_cleanup_cs(spi); spi_bitbang_cleanup(spi); } @@ -225,6 +210,7 @@ static int ath79_spi_probe(struct platform_device *pdev) pdata = dev_get_platdata(&pdev->dev); + master->use_gpio_descriptors = true; master->bits_per_word_mask = SPI_BPW_RANGE_MASK(1, 32); master->setup = ath79_spi_setup; master->cleanup = ath79_spi_cleanup;
This converts the ATH79 SPI master driver to use GPIO descriptors for chip select handling. The ATH79 driver was requesting the GPIO and driving it from the bitbang .chipselect callback. Do not request it anymore as the SPI core will request it, remove the line inversion semantics for the GPIO case (handled by gpiolib) and let the SPI core deal with requesting the GPIO line from the device tree node of the controller. This driver can be instantiated from a board file (no device tree) but the board files only use native CS (no GPIO lines) so we should be fine just letting the SPI core grab the GPIO from the device. The fact that the driver is actively driving the GPIO in the ath79_spi_chipselect() callback is confusing since the host does not set SPI_MASTER_GPIO_SS so this should not ever get called when using GPIO CS. I put in a comment about this. Cc: Felix Fietkau <nbd@nbd.name> Cc: Alban Bedel <albeu@free.fr> Cc: Linuxarm <linuxarm@huawei.com> Signed-off-by: Linus Walleij <linus.walleij@linaro.org> --- drivers/spi/spi-ath79.c | 42 ++++++++++++++--------------------------- 1 file changed, 14 insertions(+), 28 deletions(-)