From: gabriel huau Date: Wed, 08 Apr 2015 19:45:06 -0700 Subject: [U-Boot] [PATCH] x86: minnowmax: add GPIO mapping support In-Reply-To: References: <1424037328-31636-1-git-send-email-contact@huau-gabriel.fr> <54EDF800.8060907@huau-gabriel.fr> <54F022B9.8060000@huau-gabriel.fr> <551BD476.4070107@huau-gabriel.fr> <5522235F.9070704@huau-gabriel.fr> Message-ID: <5525E7B2.4010804@huau-gabriel.fr> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de Hi Simon, On 04/07/2015 07:03 PM, Simon Glass wrote: > Hi Gabriel, > > On 6 April 2015 at 00:10, Gabriel Huau wrote: >> Hi Simon, >> >> >> On 04/05/2015 11:31 AM, Simon Glass wrote: >>> Hi Gabriel, >>> >>> On 1 April 2015 at 05:20, Gabriel Huau wrote: >>>> Hi Simon, >>>> >>>> >>>> On 03/31/2015 07:32 PM, Simon Glass wrote: >>>>> Hi Gabriel, >>>>> >>>>> On 27 February 2015 at 01:52, Bin Meng wrote: >>>>>> Hi Gabriel, >>>>>> >>>>>> On Fri, Feb 27, 2015 at 3:54 PM, gabriel huau >>>>>> wrote: >>>>>>> Hi Bin, >>>>>>> >>>>>>> >>>>>>> On 02/26/2015 07:30 PM, Bin Meng wrote: >>>>>>>> Hi Gabriel, >>>>>>>> >>>>>>>> On Thu, Feb 26, 2015 at 12:27 AM, Gabriel Huau >>>>>>>> >>>>>>>> wrote: >>>>>>>>> Hi Bin, >>>>>>>>> >>>>>>>>> >>>>>>>>> On 02/24/2015 11:52 PM, Bin Meng wrote: >>>>>>>>>> Hi Gabriel, >>>>>>>>>> >>>>>>>>>> On Mon, Feb 16, 2015 at 5:55 AM, Gabriel Huau >>>>>>>>>> >>>>>>>>>> wrote: >>>>>>>>>>> Configure the pinctrl as it required to make some IO controllers >>>>>>>>>>> working (USB/UART/I2C/...). >>>>>>>>>>> The idea would be in the next version to modify the pch GPIO >>>>>>>>>>> driver >>>>>>>>>>> and >>>>>>>>>>> configure these pins through the device tree. >>>>>>>>>>> >>>>>>>>>>> These modifications are ported from the coreboot project. >>>>>>>>>>> >>>>>>>>>>> Signed-off-by: Gabriel Huau >>>>>>>>>>> --- >>>>>>>>>>> arch/x86/cpu/baytrail/Makefile | 1 + >>>>>>>>>>> arch/x86/cpu/baytrail/gpio.c | 206 >>>>>>>>>>> +++++++++++++++ >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/gpio.h | 364 >>>>>>>>>>> ++++++++++++++++++++++++++ >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/iomap.h | 73 ++++++ >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/irq.h | 119 +++++++++ >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/irqroute.h | 67 +++++ >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/pci_devs.h | 144 >>>>>>>>>>> ++++++++++ >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/pmc.h | 253 >>>>>>>>>>> ++++++++++++++++++ >>>>>>>>>>> board/intel/minnowmax/minnowmax.c | 212 >>>>>>>>>>> +++++++++++++++ >>>>>>>>>>> include/configs/minnowmax.h | 11 + >>>>>>>>>>> 10 files changed, 1450 insertions(+) >>>>>>>>>>> create mode 100644 arch/x86/cpu/baytrail/gpio.c >>>>>>>>>>> create mode 100644 arch/x86/include/asm/arch-baytrail/iomap.h >>>>>>>>>>> create mode 100644 arch/x86/include/asm/arch-baytrail/irq.h >>>>>>>>>>> create mode 100644 >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/irqroute.h >>>>>>>>>>> create mode 100644 >>>>>>>>>>> arch/x86/include/asm/arch-baytrail/pci_devs.h >>>>>>>>>>> create mode 100644 arch/x86/include/asm/arch-baytrail/pmc.h >>>>>>>>>>> >>>>>>>>>> [snip] >>>>>>>>>> >>>>>>>>>>> diff --git a/include/configs/minnowmax.h >>>>>>>>>>> b/include/configs/minnowmax.h >>>>>>>>>>> index 823e051..738c6fa 100644 >>>>>>>>>>> --- a/include/configs/minnowmax.h >>>>>>>>>>> +++ b/include/configs/minnowmax.h >>>>>>>>>>> @@ -69,4 +69,15 @@ >>>>>>>>>>> /* Avoid a warning in the Realtek Ethernet driver */ >>>>>>>>>>> #define CONFIG_SYS_CACHELINE_SIZE 16 >>>>>>>>>>> >>>>>>>>>>> +/* >>>>>>>>>>> + * Baytrail has 3 GPIOs bank over PCI, there is no >>>>>>>>>>> + * driver at the moment so let's disable the command >>>>>>>>>>> + * and the default x86 driver to avoid any collision >>>>>>>>>>> + * with the GPIO mapping code. >>>>>>>>>>> + * @TODO: adding a baytrail-gpio driver and configure >>>>>>>>>>> + * the muxing through the device tree >>>>>>>>>>> + */ >>>>>>>>>>> +#undef CONFIG_INTEL_ICH6_GPIO >>>>>>>>>>> +#undef CONFIG_CMD_GPIO >>>>>>>>>>> + >>>>>>>>>> Why undef these two? The BayTrail SoC does support GPIO banks in >>>>>>>>>> the >>>>>>>>>> legacy bridge. >>>>>>>>> I might misunderstood the GPIO subsystem but I thought there was 2 >>>>>>>>> banks >>>>>>>>> available through the PCU iLB GPIO controller which contains the >>>>>>>>> SCORE >>>>>>>>> and >>>>>>>>> SSUS (102 / 44 pins). >>>>>>>>> The intel_ich6_gpio has a limitation of 32 GPIOs per bank and I >>>>>>>>> thought >>>>>>>>> it >>>>>>>>> was just a different controller from the Baytrail, but if I can use >>>>>>>>> it >>>>>>>>> to >>>>>>>>> control all the GPIOs + doing the IO mapping, I'll be glad to do it! >>>>>>>> I checked the BayTrail datasheet. Its GPIO is in the iLB (legacy >>>>>>>> bridge), which is the same as other IA chipset (Ivybridge, >>>>>>>> TunnelCreek, Quark). It has 4 banks in core domain and 2 banks in sus >>>>>>>> domain. So 6 banks in total. You need define 6 gpio nodes in the >>>>>>>> minnowmax board dts file. You should be able to use the existing gpio >>>>>>>> driver to configure. >>>>>>> >>>>>>> Thanks for the clarification! >>>>>>> Actually, I saw it today when I was doing some tests and I configured >>>>>>> the 6 >>>>>>> banks in the devices tree. I also fixed the GPIO base address to 0x48 >>>>>>> but I >>>>>>> got some issues like the fact I'm reading only 0 from all the >>>>>>> registers. >>>>>> Yep, the offset should be 0x48 for BayTrail. >>>>>> >>>>>>> The registers are configured to be in the IO Space (0x500), I checked >>>>>>> the >>>>>>> PCI configuration space to make sure that everything is enabled >>>>>>> correctly, >>>>>>> but I'm still missing something. >>>>>> I checked the gpio driver codes, and it currently has: >>>>>> >>>>>> /* >>>>>> * Okay, I guess we're looking at the right device. The >>>>>> actual >>>>>> * GPIO registers are in the PCI device's I/O space, starting >>>>>> * at the offset that we just read. Bit 0 indicates that it's >>>>>> * an I/O address, not a memory address, so mask that off. >>>>>> */ >>>>>> gpiobase = tmplong & 0xfffe; >>>>>> >>>>>> This should be changed to >>>>>> >>>>>> gpiobase = tmplong & 0xfffc; >>>>>> >>>>>> as bit1 is the enable bit on BayTrail (Intel changes this GPIO base >>>>>> register again for BayTrail, sigh...) >>>>>> >>>>>>> Once I'll be able to use these GPIOs, I will update the entire patch >>>>>>> to >>>>>>> remove the port from Coreboot as this is not necessary. >>>>>>> >>>>>>>>>>> #endif /* __CONFIG_H */ >>>>>>>>>>> -- >>>>> What is the next step with this patch please? It would be good to >>>>> apply it to with the changes discussed. >>>> >>>> Sorry, actually I was super busy and wasn't able to work on the >>>> minnowboard >>>> max ... I should have some time this week end. >>>> But you can go ahead and drop this patch, I will submit a new one because >>>> most of the modification are actually not needed, we can use the generic >>>> pch_gpio driver already present in u-boot with some modification/fix. >>>> >>>> My only problem at the moment, is to find a solution to declare properly >>>> the >>>> pin muxing for every pins. I don't want to extend the pch_gpio_set* >>>> structures because we have 6 banks on the MNW ... I was thinking to use >>>> the >>>> device tree but I'm not really fan of the idea to change the pin muxing >>>> based on the device tree, it should be only a hardware description. >>>> My idea at the moment, is to do the pin muxing through functions named >>>> "gpio_set_cfg/gpio_set_pull/..." called in the board file. A little bit >>>> like >>>> the driver s5p_gpio.c and board file board/sunxi/board.c. >>>> What is your though about that? >>> I think pin mux is part of the hardware description - and each board >>> can do it differently depending on how the circuit works. I believe >>> device tree is exactly the right place. >>> >>> Is there a suitable binding or do you need to make one up? >>> >>> Regards, >>> Simon >> >> Agreed. >> >> I was thinking to use the gpioX node and create a binding 'pch,pins', for >> example: >> >> gpioa { >> compatible = "intel,ich6-gpio" >> u-boot,dm-pre-reloc; >> reg = <0 0x20> >> bank-name = "A"; >> >> pch,pins = < >> BYT_UART_HD_RSTB MUX_MODE0 >> >; >> } >> >> My problem is the pad value/direction for the GPIO are in another register, >> so I cannot really easily do something like "BYT_UART_HD_RSTB (MUX_MODE0 | >> GPIO_OUTPUT)" or even "BYT_UART_HD_RSTB (MUX_MODE0 | GPIO_OUTPUT_HIGH)". I >> was thinking to something like this: >> >> pch,pins = < >> BYT_UART_HD_RSTB_CFG (MUX_MODE0 | PULL_20K) >> BYT_UART_HD_RSTB_PAD (GPIO_OUTPUT_HIGH) >> >> BYT_UART_HD_RSTC_CFG (MUX_MODE1) >> BYT_UART_HD_RSTC_PAD NONE >> >> ... >> >; >> >> But it means that we will need to use 2 lines to defines a pin, even if the >> second line could define 'nothing', I'm personally fine with that, but not >> sure if this is the best solution. > Are these #defines from a binding file that you would include? > > I don't think 2 lines is a problem, but on the other handle the device > tree binding does not need to match the register layout exactly. In > general you can do things like: > > pch,pins { > pin0 { > reg = <0>; > some-property = <3>; > boolean-prop; > } > pin1 { > ... > }; > > but to be more efficient I understand that bitfields are better. But > having a few properties is fine. The code can translate them. > >> Also, I was going to put the code in drivers/gpios/intel_ich6_gpio.c but the >> GPIO muxing code is called only when we run the command 'gpio', I'll will >> try to find a better solution as it needs to be call as early as possible. > If you do something like: > > struct udevice *dev; > > uclass_get_device(UCLASS_GPIO, 0, &dev); > > it will init the GPIOs. But I think what you really want here is > pinmux. Perhaps add a special function that sets up the pinmux, > unrelated to the GPIOs? Linux has the concept of pinctrl which is not > the same as GPIOs. Pins can be used as GPIOs but also other functions. > > For now, how about just having a function called pinctrl_init() which > reads the config out of the device tree and applies it? If you are > keen, you could turn it into a driver in a new UCLASS_PINCTRL uclass. > > Regards, > Simon Thanks for you feedback! Actually I'm going to try to implement something close to your idea this week end and we can see what's going on :). I think you can close this thread/patch, I will submit a new patch series where we will be able to discuss of the implementation if you are ok with that. Regards, Gabriel