On 1.8.2018 20:23, Stefan Herbrechtsmeier wrote:
Am 30.07.2018 um 10:00 schrieb Michal Simek:
On 27.7.2018 11:14, Stefan Herbrechtsmeier wrote:
Am 27.07.2018 um 08:42 schrieb Michal Simek:
On 26.7.2018 22:04, Stefan Herbrechtsmeier wrote:
Am 26.07.2018 um 10:22 schrieb Michal Simek:
On 25.7.2018 21:17, Stefan Herbrechtsmeier wrote: > Am 25.07.2018 um 08:07 schrieb Michal Simek: >> On 24.7.2018 21:39, Stefan Herbrechtsmeier wrote: >>> Am 24.07.2018 um 10:37 schrieb Michal Simek: >>>> On 23.7.2018 20:29, Stefan Herbrechtsmeier wrote: >>>>> Am 23.07.2018 um 11:08 schrieb Michal Simek: >>>>>> On 20.7.2018 21:31, Stefan Herbrechtsmeier wrote: >>>>>>> Am 12.07.2018 um 16:04 schrieb Michal Simek: >>>>>>>> There should be proper bank name setup to distiguish between >>>>>>>> different >>>>>>>> gpio drivers. Use dev->name for it. >>>>>>>> >>>>>>>> Signed-off-by: Michal Simek michal.simek@xilinx.com >>>>>>>> --- >>>>>>>> >>>>>>>> drivers/gpio/zynq_gpio.c | 2 ++ >>>>>>>> 1 file changed, 2 insertions(+) >>>>>>>> >>>>>>>> diff --git a/drivers/gpio/zynq_gpio.c >>>>>>>> b/drivers/gpio/zynq_gpio.c >>>>>>>> index 26f69b1a713f..f793ee5754a8 100644 >>>>>>>> --- a/drivers/gpio/zynq_gpio.c >>>>>>>> +++ b/drivers/gpio/zynq_gpio.c >>>>>>>> @@ -337,6 +337,8 @@ static int zynq_gpio_probe(struct udevice >>>>>>>> *dev) >>>>>>>> struct zynq_gpio_privdata *priv = >>>>>>>> dev_get_priv(dev); >>>>>>>> struct gpio_dev_priv *uc_priv = >>>>>>>> dev_get_uclass_priv(dev); >>>>>>>> + uc_priv->bank_name = dev->name; >>>>>>>> + >>>>>>>> if (priv->p_data) >>>>>>>> uc_priv->gpio_count = priv->p_data->ngpio; >>>>>>>> >>>>>>> Does this not lead to ugly names because the gpio number is >>>>>>> append to >>>>>>> the bank_name? Have you check the "gpio status -a" output? >>>>>> Yes I was checking it. Names are composed together but also >>>>>> just >>>>>> numbers >>>>>> works as before. >>>>>> >>>>>> gpio@ff0a00000: input: 0 [ ] >>>>>> gpio@ff0a00001: input: 0 [ ] >>>>>> gpio@ff0a00002: input: 0 [ ] >>>>>> gpio@ff0a00003: input: 0 [ ] >>>>>> gpio@ff0a00004: input: 0 [ ] >>>>>> gpio@ff0a00005: input: 0 [ ] >>>>>> gpio@ff0a00006: input: 0 [ ] >>>>>> gpio@ff0a00007: input: 0 [ ] >>>>>> gpio@ff0a00008: input: 0 [ ] >>>>>> gpio@ff0a00009: input: 0 [ ] >>>>> Do you think that this are meaningful names? It isn't >>>>> possible to >>>>> separate the device and pin number as well as it mix hex and >>>>> decimal >>>>> numbers. >>>>> >>>>>> If you know better way how to setup a bank name please let me >>>>>> know >>>>>> but I >>>>>> need to distinguish ps gpio from pl one and for pl we need to >>>>>> know >>>>>> the >>>>>> address. >>>>> I know the use case. >>>>> >>>>> A lot of drivers use the bank_name from the device tree, some >>>>> drivers >>>>> append an underscore to the bank name and others add the >>>>> req_seq of >>>>> the >>>>> device to an alphabetic character. >>>>> >>>>>>> Other drivers use the gpio-bank-name from the device tree. >>>>>> I can't see this property inside Linux kernel. If this has been >>>>>> reviewed >>>>>> by dt guys please let me know. >>>>> This property is only used by u-boot. I think it isn't needed >>>>> by the >>>>> Linux kernel. >>>> I am happy to use consistent solution but what's that? >>> Consistent solution between what? >> all drivers. Name should be composed consistently among all >> drivers. >> It means drivers shouldn't add additional "_" in driver code for >> example. >> >>>> Mixing name with hex and int is not nice but adding "_" or >>>> something >>>> else is just a pain in driver code. If this is done in core I am >>>> fine >>>> with that but adding this code to all drivers don't look like >>>> generic >>>> solution at all. >>> Normally the bank name is an alphabetic character or string. >>> Maybe we >>> could add the device name to the gpio_lookup_name function and >>> add an >>> additional optional device name parameter to the gpio command. >>> >>>> Using additional u-boot property is not good too. >>>> >>>> I have mentioned in "gpio: xilinx: Convert driver to DM" >>>> (sha1:10441ec9224d0d269dc512819a32c0785a6338d3) >>>> that uc-priv->name is completely unused. Maybe this should be >>>> dev->name >>>> and bank_name should be really used for banks. >>> Isn't the uc-priv->name used for the label of the request? >> Last time when I looked at it and I didn't see this name listed >> anywhere >> in output. >> >> >>>> Then in gpio status -a can be >>>> >>>> Device gpio@a0001000: >>>> Bank: >>>> ... >>>> >>>> but not sure how gpio commands will work to address exact pin >>>> from >>>> prompt. Because this is normally working >>>> gpio toggle gpio@a00010001 >>> With an optional device name this would be: >>> gpio toggle gpio@a0001000 1 >>> >>> Alternative the gpio command could support the requested labels: >>> gpio toggle second-gpio >> I am open to see this solution. Feel free to invest your time >> support >> this but I have no time for that. > What does this mean? > > I understand that you don't have the time to develop a new common > solution. > > But the discussion disappoints me. You misuse the bank name in a > poor > way and decline alternative solutions with additional requirements > even > if they are already used in u-boot. > > The name "gpio@a000100011" for pin 11 of the device gpio@a0001000 > isn't > consistent between the u-boot drivers nor is the name used in > Linux. A > bank-name from the device tree is used by several u-boot drivers > even if > it isn't consistent between the drivers. I am sorry that you feel like that. It is not about declining alternative solution. I want to know what's the right solution is.
Thanks a lot for taking the time to write a detail explanation.
Using bank-name/gpio-bank-name via DT is something what is IMHO not correct. The first thing is if this is used just by u-boot it should have at least u-boot prefix. It means u-boot,bank-name, u-boot,gpio-bank-name. (Even I am not sure if u-boot prefix is properly allocated and can be allocated via sort of vendor-prefix).
I agree with you.
The second thing is that I don't think it is good to have two different dts files. One in the kernel and second in u-boot. Even I know we have problem with that but we are trying to sync it as much as possible.
Does the kernel accept properly allocated but not used entries?
kernel like a code doesn't care and ignores what it is not used. But DT maintainers as far as I know don't like these options.
Regarding others options:
at91_gpio: plat->bank_name - which is not filled for OF case. (2 chars via platdata) da8xx - plat->port_name - which is not filled for OF case and also no user
altera_pio, hsdk, msm, pm8916, sandbox: gpio-bank-name intel_broadwell, intel_ich6: bank-name:
pcf8575 - gpio-bank-name or fdt_get_name
atmel_pio4, s5p, vybrid: fdt_get_name bcm6345, rcar: dev->name
hi6220, imx, mxc, omap: "GPIO%d_" plat->bank_index or banknum mpc8xxx: "MPC@%lx_" data->addr pca953x: "%s@%x_", label, info->addr or "gpio@%x_", info->addr axp_gpio : AXP0- hardcoded - "-" prefix bcm2835 - GPIO - without anything sunxi: PA + bank tegra, tegra186: 2char names via list mvebu: 'A' + dev->req_seq pic32, rk: 'A' + bank + some ligc around dev->name
stm32f7: st,bank-name
################################################################### Sum: 2 are not OF 1 is using one prefix (st) 7,5 are using gpio-bank-name or bank-name 5.5 are using dev->name (+2 xilinx which are not listed now) 14 are using own prefixes - made or hardcoded 6 of them are ending with _ 1 ends with - 7 don't end with - or _
Cases with gpio-bank-name, bank-name I have described above.
In case of "_" or "-" suffix Bank name will be listed with this suffix which also doesn't look good. GPIO names below with appended number looks good.
ZynqMP> gpio status -a Bank GPIO_: <================ HERE GPIO_0: input: 0 [ ] GPIO_1: input: 0 [ ] GPIO_2: input: 0 [ ] GPIO_3: input: 0 [ ] GPIO_4: input: 0 [ ] GPIO_5: input: 0 [ ] GPIO_6: input: 0 [ ] GPIO_7: input: 0 [ ] GPIO_8: input: 0 [ ]
This doesn't look good but therefore the pin name looks okay.
yes.
And I hope it is clear that I can't make this bank_name empty or we will end up with this when PL gpios are included which is total mess.
I have the same problem.
ok.
It means what I have used and send patch for is used in 5,5 other cases and they could have the same issue which we are talking about.
The problem I see is that you introduce a suboptimal API which make it hard to changed later or do you accept incompatible changes?
I didn't introduce any API. API was there and I haven't done any change there. I just setup a name which the whole interface expects. gpio numbers are still working without any change. It means gpio toggle 170 just works without any breakage.
But we couldn't change the bank name after the release.
It will be the best to have one stable - yes.
This change is new and will be new in 2018.09 release and we haven't reached rc1 yet. It means we still have enough time to keep this, revert this, find a better solution or use temporary solution (with that _ for example). The same stuff is used by zynq and axi gpio drivers.
Okay
If you think that we should append _ in the driver then I would expect that we should also remove _ it from name when Bank XXX_ is listed.
This would be okay for me.
But maybe there is a better solution. Would it be okay to add an device tree alias for the gpios?
I have convinced Linus to accept dt alias wiring for zynq gpio added by https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/dr...
Okay
but in general statement from the is that gpios don't need to use aliases. In this case alias ID setup base chip ID for sysfs interface which is deprecated.
This means the alias is okay but shouldn't be used from user space?
In this case we could use the seq number to select the device:
gpio set 0 2 gpio set 1 6
The first number would be the seq and the second the offset. This would make the bank name obsolete and could be backward compatible implemented in the gpio command.
seq number is setup based on aliases(if enabled - and in this gpio case not recommended) or based on bind/probe order.
Why isn't it recommended?
Feel free to check with them but my understanding is that if alias is not used by core/drivers then it shouldn't be listed because it is not clear what it is for. Rob was asking me to remove these gpio aliases ,which I had in xilinx tree, when I was pushing zynqmp dts to mainline.
aliases - we know that this is not good to do probe order - depends on DT, clock, etc
But isn't this widely used in u-boot?
Some uclasses enables that options. In u-boot seq alias is present for gpio uclass. Also keep in your mind that it depends on option which can be disabled (DM_SEQ_ALIAS)
I don't think it is a good idea to use seq in this case.
I don't understand why it is widely used if it isn't a good idea.
At the moment the gpio framework use a bank name to distinguish between different controllers of the same type. We need to distinguish between different gpio controllers and hardware configurations. The right place for hardware configuration is the device tree. This means we have to add additional information into the device tree. This could be a bank-name per controller or an device tree alias. The later approach is already used by different other frameworks. It implements a low level interface for u-boot and makes the bank name obsolete.
It is about the flow and recent discussion with had in connection to CMD_DM. To find which seq is which controller you need to run dm command (which is for debug only) or list fdt(where u-boot dt address is not setup by default and also it doesn't reflect if driver was binded). Maybe it is not issue for others but looking for information which controller is which one seems to me additional step. Feel free to enable this option. dev->seq is filled by core automatically.
Thanks for the informations.
Regarding dt property. I have no problem to use property which are already the part of binding. It means using label which describe what it is for seems to be the best candidate to use (used by pca953x_gpio in one case). But again we have the same issue with this approach when name ends with hex number and normally the name won't contain "_" suffix.
Anyway I have not a problem with this change.
diff --git a/drivers/gpio/zynq_gpio.c b/drivers/gpio/zynq_gpio.c index 6fbaafb3fa3c..aa8d51e2f709 100644 --- a/drivers/gpio/zynq_gpio.c +++ b/drivers/gpio/zynq_gpio.c @@ -336,8 +336,18 @@ static int zynq_gpio_probe(struct udevice *dev) { struct zynq_gpio_platdata *platdata = dev_get_platdata(dev); struct gpio_dev_priv *uc_priv = dev_get_uclass_priv(dev);
- uc_priv->bank_name = dev->name; + const void *label_ptr; + void *label_c; + int size;
+ label_ptr = dev_read_prop(dev, "label", &size); + if (label_ptr) { + label_c = calloc(1, size); + memcpy(label_c, label_ptr, size); + uc_priv->bank_name = label_c; + } else { + uc_priv->bank_name = dev->name; + }
if (platdata->p_data) uc_priv->gpio_count = platdata->p_data->ngpio;
This would be okay for me.
Ok. I have sent regular patch for that.
Thanks, Michal