| Message ID | 1445513420-23527-1-git-send-email-mike.looijmans@topic.nl |
|---|---|
| State | New |
| Headers | show |
Hi Mike, On Thu, 2015-10-22 at 01:30PM +0200, Mike Looijmans wrote: > Supplying pinmux configuration for e.g. gpio pins leads to deferred > probes because the pinctrl device is probed much later than gpio. > Move the init call to a much earlier stage so it probes before the > devices that may need it. > > Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> in general, the change should be OK, but neither on zc702 nor zc706 do I see a difference in respect to deferred probes. With and without the patch I see: root@zynq:~# dmesg | grep -i defer [ 0.097021] zynq-gpio e000a000.gpio: could not find pctldev for node /amba/slcr@f8000000/pinctrl@700/gpio0-default, deferring probe root@zynq:~# If you have a case this patch improves things though, feel free to add my Tested-by: Sören Brinkmann <soren.brinkmann@xilinx.com> Thanks, Sören
On 22-10-15 18:07, Sören Brinkmann wrote: > Hi Mike, > > On Thu, 2015-10-22 at 01:30PM +0200, Mike Looijmans wrote: >> Supplying pinmux configuration for e.g. gpio pins leads to deferred >> probes because the pinctrl device is probed much later than gpio. >> Move the init call to a much earlier stage so it probes before the >> devices that may need it. >> >> Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> > > in general, the change should be OK, but neither on zc702 nor zc706 do I > see a difference in respect to deferred probes. With and without the > patch I see: > root@zynq:~# dmesg | grep -i defer > [ 0.097021] zynq-gpio e000a000.gpio: could not find pctldev for node /amba/slcr@f8000000/pinctrl@700/gpio0-default, deferring probe > root@zynq:~# > > If you have a case this patch improves things though, feel free to add my > Tested-by: Sören Brinkmann <soren.brinkmann@xilinx.com> > On the Florida boards there are i2c controlled clocks, power supplies and reset signals. Replacing the Cadence I2C controller with a GPIO-bitbang controller solved the I2C problems but caused a storm of dozens of deferred probes because of the pinmux driver arriving even after the first probe attempt of the i2c bus driver. Moving the pinmux driver to an earlier stage solved that problem neatly, now the "zynq-pinctrl 700.pinctrl: zynq pinctrl initialized" message appears after the OCM driver. Judging from your comment the GPIO driver still probes earlier (I don't have any GPIO-only pinmuxes yet), so maybe we should amend the patch to probe even earlier. The pinmux driver doesn't depend on anything, so it can potentially probe very early. What do you think? Mike. Kind regards, Mike Looijmans System Expert TOPIC Embedded Products Eindhovenseweg 32-C, NL-5683 KH Best Postbus 440, NL-5680 AK Best Telefoon: +31 (0) 499 33 69 79 Telefax: +31 (0) 499 33 69 70 E-mail: mike.looijmans@topicproducts.com Website: www.topicproducts.com Please consider the environment before printing this e-mail Visit us at : Aerospace Electrical Systems Expo Europe which will be held from 17.11.2015 till 19.11.2015, Findorffstrasse 101 Bremen, Germany, Hall 5, stand number C65 http://www.aesexpo.eu
On 23-10-15 07:31, Mike Looijmans wrote: > On 22-10-15 18:07, Sören Brinkmann wrote: >> Hi Mike, >> >> On Thu, 2015-10-22 at 01:30PM +0200, Mike Looijmans wrote: >>> Supplying pinmux configuration for e.g. gpio pins leads to deferred >>> probes because the pinctrl device is probed much later than gpio. >>> Move the init call to a much earlier stage so it probes before the >>> devices that may need it. >>> >>> Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> >> >> in general, the change should be OK, but neither on zc702 nor zc706 do I >> see a difference in respect to deferred probes. With and without the >> patch I see: >> root@zynq:~# dmesg | grep -i defer >> [ 0.097021] zynq-gpio e000a000.gpio: could not find pctldev for node >> /amba/slcr@f8000000/pinctrl@700/gpio0-default, deferring probe >> root@zynq:~# >> >> If you have a case this patch improves things though, feel free to add my >> Tested-by: Sören Brinkmann <soren.brinkmann@xilinx.com> >> > > On the Florida boards there are i2c controlled clocks, power supplies and > reset signals. Replacing the Cadence I2C controller with a GPIO-bitbang > controller solved the I2C problems but caused a storm of dozens of deferred > probes because of the pinmux driver arriving even after the first probe > attempt of the i2c bus driver. Moving the pinmux driver to an earlier stage > solved that problem neatly, now the "zynq-pinctrl 700.pinctrl: zynq pinctrl > initialized" message appears after the OCM driver. > Judging from your comment the GPIO driver still probes earlier (I don't have > any GPIO-only pinmuxes yet), so maybe we should amend the patch to probe even > earlier. The pinmux driver doesn't depend on anything, so it can potentially > probe very early. What do you think? Minor addition: The gpio-zynq driver probes at "postcore_initcall", so to beat that, the zynq-pinmux driver should move to "core_initcall" (instead of "arch_initcall"). That would make the gpio deferral go away. An alternative would be to move the gpio driver to "arch", then "postcore" would be enough for the pinmux. If the gpio probe gets deferred, it apparently has already been moved to below "subsys" already, with aparently no ill effects. Mike. Kind regards, Mike Looijmans System Expert TOPIC Embedded Products Eindhovenseweg 32-C, NL-5683 KH Best Postbus 440, NL-5680 AK Best Telefoon: +31 (0) 499 33 69 79 Telefax: +31 (0) 499 33 69 70 E-mail: mike.looijmans@topicproducts.com Website: www.topicproducts.com Please consider the environment before printing this e-mail Visit us at : Aerospace Electrical Systems Expo Europe which will be held from 17.11.2015 till 19.11.2015, Findorffstrasse 101 Bremen, Germany, Hall 5, stand number C65 http://www.aesexpo.eu
On Fri, 2015-10-23 at 07:31AM +0200, Mike Looijmans wrote: > On 22-10-15 18:07, Sören Brinkmann wrote: > >Hi Mike, > > > >On Thu, 2015-10-22 at 01:30PM +0200, Mike Looijmans wrote: > >>Supplying pinmux configuration for e.g. gpio pins leads to deferred > >>probes because the pinctrl device is probed much later than gpio. > >>Move the init call to a much earlier stage so it probes before the > >>devices that may need it. > >> > >>Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> > > > >in general, the change should be OK, but neither on zc702 nor zc706 do I > >see a difference in respect to deferred probes. With and without the > >patch I see: > > root@zynq:~# dmesg | grep -i defer > > [ 0.097021] zynq-gpio e000a000.gpio: could not find pctldev for node /amba/slcr@f8000000/pinctrl@700/gpio0-default, deferring probe > > root@zynq:~# > > > >If you have a case this patch improves things though, feel free to add my > >Tested-by: Sören Brinkmann <soren.brinkmann@xilinx.com> > > > > On the Florida boards there are i2c controlled clocks, power supplies and > reset signals. Replacing the Cadence I2C controller with a GPIO-bitbang > controller solved the I2C problems but caused a storm of dozens of deferred > probes because of the pinmux driver arriving even after the first probe > attempt of the i2c bus driver. Moving the pinmux driver to an earlier stage > solved that problem neatly, now the "zynq-pinctrl 700.pinctrl: zynq pinctrl > initialized" message appears after the OCM driver. OK, makes sense. Thanks for the background. > Judging from your comment the GPIO driver still probes earlier (I don't have > any GPIO-only pinmuxes yet), so maybe we should amend the patch to probe > even earlier. The pinmux driver doesn't depend on anything, so it can > potentially probe very early. What do you think? I'm pretty neutral on this one :) Hasn't the probe deferral mechanism been introduced to avoid having to create ordering through the initcall stages? But I agree, having the probe deferral notices is not particularly pretty. So, I'd definitely not oppose changing this. Though, there is one dependency on the SLCR regmap, but that is initialized fairly early. Sören
On Fri, 2015-10-23 at 07:43AM +0200, Mike Looijmans wrote: > On 23-10-15 07:31, Mike Looijmans wrote: > >On 22-10-15 18:07, Sören Brinkmann wrote: > >>Hi Mike, > >> > >>On Thu, 2015-10-22 at 01:30PM +0200, Mike Looijmans wrote: > >>>Supplying pinmux configuration for e.g. gpio pins leads to deferred > >>>probes because the pinctrl device is probed much later than gpio. > >>>Move the init call to a much earlier stage so it probes before the > >>>devices that may need it. > >>> > >>>Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> > >> > >>in general, the change should be OK, but neither on zc702 nor zc706 do I > >>see a difference in respect to deferred probes. With and without the > >>patch I see: > >> root@zynq:~# dmesg | grep -i defer > >> [ 0.097021] zynq-gpio e000a000.gpio: could not find pctldev for node > >>/amba/slcr@f8000000/pinctrl@700/gpio0-default, deferring probe > >> root@zynq:~# > >> > >>If you have a case this patch improves things though, feel free to add my > >>Tested-by: Sören Brinkmann <soren.brinkmann@xilinx.com> > >> > > > >On the Florida boards there are i2c controlled clocks, power supplies and > >reset signals. Replacing the Cadence I2C controller with a GPIO-bitbang > >controller solved the I2C problems but caused a storm of dozens of deferred > >probes because of the pinmux driver arriving even after the first probe > >attempt of the i2c bus driver. Moving the pinmux driver to an earlier stage > >solved that problem neatly, now the "zynq-pinctrl 700.pinctrl: zynq pinctrl > >initialized" message appears after the OCM driver. > >Judging from your comment the GPIO driver still probes earlier (I don't have > >any GPIO-only pinmuxes yet), so maybe we should amend the patch to probe even > >earlier. The pinmux driver doesn't depend on anything, so it can potentially > >probe very early. What do you think? > > Minor addition: > The gpio-zynq driver probes at "postcore_initcall", so to beat that, the > zynq-pinmux driver should move to "core_initcall" (instead of > "arch_initcall"). That would make the gpio deferral go away. > > An alternative would be to move the gpio driver to "arch", then "postcore" > would be enough for the pinmux. If the gpio probe gets deferred, it > apparently has already been moved to below "subsys" already, with aparently > no ill effects. Makes me wonder whether a single deferred probe is worth the risk of breaking something by moving these things around. But probably worth a try. Sören
Hi Linus, On 10/23/2015 07:44 AM, Sören Brinkmann wrote: > On Fri, 2015-10-23 at 07:31AM +0200, Mike Looijmans wrote: >> On 22-10-15 18:07, Sören Brinkmann wrote: >>> Hi Mike, >>> >>> On Thu, 2015-10-22 at 01:30PM +0200, Mike Looijmans wrote: >>>> Supplying pinmux configuration for e.g. gpio pins leads to deferred >>>> probes because the pinctrl device is probed much later than gpio. >>>> Move the init call to a much earlier stage so it probes before the >>>> devices that may need it. >>>> >>>> Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> >>> >>> in general, the change should be OK, but neither on zc702 nor zc706 do I >>> see a difference in respect to deferred probes. With and without the >>> patch I see: >>> root@zynq:~# dmesg | grep -i defer >>> [ 0.097021] zynq-gpio e000a000.gpio: could not find pctldev for node /amba/slcr@f8000000/pinctrl@700/gpio0-default, deferring probe >>> root@zynq:~# >>> >>> If you have a case this patch improves things though, feel free to add my >>> Tested-by: Sören Brinkmann <soren.brinkmann@xilinx.com> >>> >> >> On the Florida boards there are i2c controlled clocks, power supplies and >> reset signals. Replacing the Cadence I2C controller with a GPIO-bitbang >> controller solved the I2C problems but caused a storm of dozens of deferred >> probes because of the pinmux driver arriving even after the first probe >> attempt of the i2c bus driver. Moving the pinmux driver to an earlier stage >> solved that problem neatly, now the "zynq-pinctrl 700.pinctrl: zynq pinctrl >> initialized" message appears after the OCM driver. > > OK, makes sense. Thanks for the background. > >> Judging from your comment the GPIO driver still probes earlier (I don't have >> any GPIO-only pinmuxes yet), so maybe we should amend the patch to probe >> even earlier. The pinmux driver doesn't depend on anything, so it can >> potentially probe very early. What do you think? > > I'm pretty neutral on this one :) Hasn't the probe deferral mechanism > been introduced to avoid having to create ordering through the initcall > stages? But I agree, having the probe deferral notices is not particularly > pretty. So, I'd definitely not oppose changing this. > Though, there is one dependency on the SLCR regmap, but that is initialized > fairly early. Any comment on this one? Thanks, Michal
On Thu, Oct 22, 2015 at 1:30 PM, Mike Looijmans <mike.looijmans@topic.nl> wrote: > Supplying pinmux configuration for e.g. gpio pins leads to deferred > probes because the pinctrl device is probed much later than gpio. > Move the init call to a much earlier stage so it probes before the > devices that may need it. > > Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> Patch applied with Sören's Tested-by tag. I am liberal with initcall reordering because I think the deferred probe mechanism is not really working. Take some time to review Rafael Wysocki's latest device link concept that can get initcall reordering closer to the device core. Yours, Linus Walleij
diff --git a/drivers/pinctrl/pinctrl-zynq.c b/drivers/pinctrl/pinctrl-zynq.c index fd4602d..36984f1 100644 --- a/drivers/pinctrl/pinctrl-zynq.c +++ b/drivers/pinctrl/pinctrl-zynq.c @@ -1173,7 +1173,17 @@ static struct platform_driver zynq_pinctrl_driver = { .remove = zynq_pinctrl_remove, }; -module_platform_driver(zynq_pinctrl_driver); +static int __init zynq_pinctrl_init(void) +{ + return platform_driver_register(&zynq_pinctrl_driver); +} +arch_initcall(zynq_pinctrl_init); + +static void __exit zynq_pinctrl_exit(void) +{ + platform_driver_unregister(&zynq_pinctrl_driver); +} +module_exit(zynq_pinctrl_exit); MODULE_AUTHOR("Sören Brinkmann <soren.brinkmann@xilinx.com>"); MODULE_DESCRIPTION("Xilinx Zynq pinctrl driver");
Supplying pinmux configuration for e.g. gpio pins leads to deferred probes because the pinctrl device is probed much later than gpio. Move the init call to a much earlier stage so it probes before the devices that may need it. Signed-off-by: Mike Looijmans <mike.looijmans@topic.nl> --- drivers/pinctrl/pinctrl-zynq.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)