On 21/11/2017 14:26, Otavio Salvador wrote:

On Tue, Nov 21, 2017 at 10:52 AM, Bas Mevissen <abuse@...> wrote:

On 20/11/2017 19:22, Fabio Berton wrote:

This commit merge tag Linux 4.14.

Signed-off-by: Fabio Berton <fabio.berton@...>
---
recipes-kernel/linux/linux-fslc_4.14.bb | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/recipes-kernel/linux/linux-fslc_4.14.bb
b/recipes-kernel/linux/linux-fslc_4.14.bb
index 6adc12e1..a96880fd 100644
--- a/recipes-kernel/linux/linux-fslc_4.14.bb
+++ b/recipes-kernel/linux/linux-fslc_4.14.bb
@@ -12,6 +12,6 @@ include linux-fslc.inc
PV = "4.14+git${SRCPV}"
SRCBRANCH = "4.14.x+fslc"
-SRCREV = "bd340b0f7370015791413ea458c0f56715f17e19"
+SRCREV = "c4404197b0b22f479d9548e81a8a4ed639d4dd7c"
COMPATIBLE_MACHINE = "(mxs|mx5|mx6|vf|use-mainline-bsp)"

I see a regression in imx SPI with this change:

bd340b0f7370015791413ea458c0f56715f17e19:

[ 8.610321] spi_imx 200c000.ecspi: registered master spi1
[ 8.626154] spi spi1.0: spi_imx_setup: mode 0, 8 bpw, 5000000 hz
[ 8.626207] spi spi1.0: setup mode 0, 8 bits/w, 5000000 Hz max --> 0
[ 8.633467] spi_imx 200c000.ecspi: registered child spi1.0
[ 8.633593] spi_imx 200c000.ecspi: probed


c4404197b0b22f479d9548e81a8a4ed639d4dd7c:

[ 1.196374] spi_imx 200c000.ecspi: No CS GPIOs available
[ 1.201760] spi_imx: probe of 200c000.ecspi failed with error -22

Platform is i.MX6 dual core.

In the latter case, my SPI device is not working.

commit 3e9b36edc01ec3b896bf212a2a7cccc32ed3f047
Author: Trent Piepho <tpiepho@...>
Date: Thu Oct 26 18:08:39 2017 -0700
spi: imx: Fix failure path leak on GPIO request error
If the code that requests any chip select GPIOs fails, the cleanup of
spi_bitbang_start() by calling spi_bitbang_stop() is not done.
Fix this by moving spi_bitbang_start() to after the code that requets
GPIOs. The GPIOs are dev managed and don't need explicit cleanup.
Since spi_bitbang_start() is now the last operation, it doesn't need
to be cleaned up in the failure path.
CC: Shawn Guo <shawnguo@...>
CC: Sascha Hauer <kernel@...>
CC: Fabio Estevam <fabio.estevam@...>
CC: Mark Brown <broonie@...>
Reviewed-by: Oleksij Rempel <o.rempel@...>
Signed-off-by: Trent Piepho <tpiepho@...>
Signed-off-by: Mark Brown <broonie@...>
Seems to be the culprit; I think you might be missing the chip-select pin, no?

That one looked innocent to me, but it is indeed the only change in spi-imx.c between the two commits I mentioned. It feels quite odd that no one would have noticed the regression in a month time.

Just to be sure, the relevant snippet from the device tree file:

&ecspi2 {
pinctrl-names = "default";
pinctrl-0 = <&pinctrl_ecspi2>;
cs-gpios = <&gpio2 26 GPIO_ACTIVE_HIGH>;
status = "okay";
mrf24j40mc@0 {
compatible = "microchip,mrf24j40mc";
spi-max-frequency = <5000000>; /* datasheet: max 10000000 */
reg = <0>;
interrupts = <20 8>;
interrupt-parent = <&gpio5>;
#ifndef SPIDEV_SPI1_ENABLE
status = "okay";
#else
status = "disabled";
#endif
};
/* Test device */
spidev1_0: spi@0 {
compatible = "spidev", "rohm,dh2228fv"; /* spidev driver needs this entry to avoid runtime warning */
reg = <0>;
spi-max-frequency = <5000000>;
#ifndef SPIDEV_SPI1_ENABLE
status = "disabled";
#else
status = "okay";
#endif
};
};

SPIDEV_SPI1_ENABLE is not defined; it is a terrible kludge to produce a separate device tree file where we can have direct SPI access to the microchip radio during test. If someone has a better idea, please let me know. Creating a DT overlay came to mind, but I haven't got around it yet.

&iomuxc {
pinctrl-names = "default";
pinctrl-0 = <&pinctrl_hog>, <&pinctrl_expansion_hdr_gpio>;

(...)

pinctrl_ecspi2: ecspi2grp {
fsl,pins = <
MX6QDL_PAD_EIM_CS1__ECSPI2_MOSI 0x0b0b0 /* MOSI */
MX6QDL_PAD_EIM_OE__ECSPI2_MISO 0x0b0b0 /* MISO */
MX6QDL_PAD_EIM_CS0__ECSPI2_SCLK 0x0b0b0 /* SCK */
MX6QDL_PAD_EIM_RW__GPIO2_IO26 0x0b0b0 /* CS_N */
>;
};

(...)

};