W dniu 30.10.2015 o 19:43, Pavel Fedin pisze:
> Hello!
>
>> Please, carefully look at:
>> Documentation/devicetree/bindings/net/gpmc-eth.txt
>> Documentation/devicetree/bindings/bus/ti-gpmc.txt
>>
>> 1. Try to re-use existing bindings. Although I see existing bank-width
>> and gpmc,device-width but yours samsung,srom-data-width seems better.
>> Maybe there are other to re-use?
>
> I've done this and this is what i currently came up with. I'd like to discuss this before respin because nobody needs this
> back-and-forth bouncing.
> --- cut exynos5410.dtsi ---
> sromc: sromc@12250000 {
> #address-cells = <2>;
> #size-cells = <1>;
> ranges = <0 0 0x04000000 0x20000
> 1 0 0x05000000 0x20000
> 2 0 0x06000000 0x20000
> 3 0 0x07000000 0x20000>;
Do you have to use 2 cells for address? Cannot it be:
ranges = <0 0x04000000 0x20000
1 0x05000000 0x20000
2 0x06000000 0x20000
3 0x07000000 0x20000>;
>
> compatible = "samsung,exynos-srom";
> reg = <0x12250000 0x14>;
Put compatible and reg at beginning.
> };
> --- cut exynos5410.dtsi ---
> --- cut exynos5410-smdk5410.dts ---
> &sromc {
> pinctrl-names = "default";
> pinctrl-0 = <&srom_ctl>, <&srom_ebi>;
>
> ethernet@3 {
> compatible = "smsc,lan9115";
> reg = <3 0 0x10000>;
> phy-mode = "mii";
> interrupt-parent = <&gpx0>;
> interrupts = <5 8>;
> reg-io-width = <2>;
> smsc,irq-push-pull;
> smsc,force-internal-phy;
>
> samsung,srom-config = <1 9 12 1 9 1 1>;
> };
> };
> --- cut exynos5410-smdk5410.dts ---
>
> 1. After writing proper ranges i was able to get rid of dedicated bank number property, because now it is the first value in
> device's "reg".
Good, I like your approach.
> 2. I noticed that smsc binding already uses reg-io-width property. I searched for it, and it appears to be reused by many devices.
> So, i decided to use it to specify bank width, since it's already there. ti-gpmc uses bank-width because it supports configurations
> like reg-io-width = 4 && bank-width = 2. I don't know if it's possible to do the same with SROMc, Exynos doc doesn't tell anything
> about 32-bit accesses. But, if you want to support it too, i would suggest the following algorithm:
> if (have bank-width)
> width = bank-width
> else if (have reg-io-width)
> width = reg-io-width
> else
> width = 1 /* default */
Re-using reg-io-width seems fine. In future, if needed, this can be
extended by introducing bank-width but now there is no sense in it.
> 3. About samsung,srom-config array. I have the following reasons to keep if this way:
> - listing every property under own name is just too much typing
> - these values really do not make sense without each other, or partialy. I would say that in array form they are even better
> readable, because it is the same order in which they go into the register.
For timings - OK. PMC is separate. This is not a timing.
> However, it is still a nice idea do document them, but...
> my PDF has "confidential" watermark on it, and this would mean that i essentially copy some information from it into Linux doc. Is
> it okay to do this?
You need to document them. We are not gonna put some data looking like a
vendor blob into the binding. I understand that document has
confidential mark... all of Samsung datasheets I've seen had it. I don't
find it as problem but of course I am not the one to judge here. If you
do not feel comfortable publishing such data, get an approval from your
manager. You can also try to search for this in vendor code
(opensource.samsung.com). If it is published there, then this won't be a
disclosure.
> If you still dislike the array, i'll redo is a set of properties like samsung,srom-tacs, samsung,srom-tcos, etc.
>
Ok, array for timings, but PMC IMHO is different.
BTW, your email client is weird. You are sending non-wrapped emails
without format=flowed in Content-Type. Please stick to mailing list
convention of wrapping at 72-char. It is really difficult to read your
replies.
Best regards,
Krzysztof
Hello!
> > --- cut exynos5410.dtsi ---
> > sromc: sromc@12250000 {
> > #address-cells = <2>;
> > #size-cells = <1>;
> > ranges = <0 0 0x04000000 0x20000
> > 1 0 0x05000000 0x20000
> > 2 0 0x06000000 0x20000
> > 3 0 0x07000000 0x20000>;
>
> Do you have to use 2 cells for address? Cannot it be:
> ranges = <0 0x04000000 0x20000
> 1 0x05000000 0x20000
> 2 0x06000000 0x20000
> 3 0x07000000 0x20000>;
I tried this first, but it didn't work, and ranges translation
gave me something really weird (like addr = 0x80 and
size = 0x04000004). I decided not to dig deeply into
"ranges" processing, but just followed GPMC approach. They say
that first number is range ID and second number is offset within
the range. And it just worked.
> > 3. About samsung,srom-config array. I have the following reasons to keep if this way:
> > - listing every property under own name is just too much typing
> > - these values really do not make sense without each other, or partialy. I would say
> that in array form they are even better
> > readable, because it is the same order in which they go into the register.
>
> For timings - OK. PMC is separate. This is not a timing.
But it's srom-config, and not srom-timing anymore. So do you want
two properties like srom-timing + srom-page-mode (with vendor prefix
of course)?
> You need to document them. We are not gonna put some data looking like a
> vendor blob into the binding. I understand that document has
> confidential mark... all of Samsung datasheets I've seen had it. I don't
> find it as problem but of course I am not the one to judge here. If you
> do not feel comfortable publishing such data, get an approval from your
> manager. You can also try to search for this in vendor code
> (opensource.samsung.com). If it is published there, then this won't be a
> disclosure.
I've seen the actual SROMc settings in some old Android kernels, like 3.0.4.
But they are not documented there, no comments, just #define's.
Ok, i'll try to take a look how to minimize the actual information. :)
> BTW, your email client is weird. You are sending non-wrapped emails
> without format=flowed in Content-Type. Please stick to mailing list
> convention of wrapping at 72-char.
I know, sorry, this is MS Outlook (not Express), and it's impossible
to configure it this way (if i just set up word wrap, it starts to
corrupt patches). I am now trying to wrap the text by hands, i hope
it's better. And i cannot use anything else because... You know why. :)
Kind regards,
Pavel Fedin
Expert Engineer
Samsung Electronics Research center Russia
On 02.11.2015 16:31, Pavel Fedin wrote:
> Hello!
>
>>> --- cut exynos5410.dtsi ---
>>> sromc: sromc@12250000 {
>>> #address-cells = <2>;
>>> #size-cells = <1>;
>>> ranges = <0 0 0x04000000 0x20000
>>> 1 0 0x05000000 0x20000
>>> 2 0 0x06000000 0x20000
>>> 3 0 0x07000000 0x20000>;
>>
>> Do you have to use 2 cells for address? Cannot it be:
>> ranges = <0 0x04000000 0x20000
>> 1 0x05000000 0x20000
>> 2 0x06000000 0x20000
>> 3 0x07000000 0x20000>;
>
> I tried this first, but it didn't work, and ranges translation
> gave me something really weird (like addr = 0x80 and
> size = 0x04000004).
Did you change the address-cells to <1>?
> I decided not to dig deeply into
> "ranges" processing, but just followed GPMC approach. They say
> that first number is range ID and second number is offset within
> the range. And it just worked.
>
>>> 3. About samsung,srom-config array. I have the following reasons to keep if this way:
>>> - listing every property under own name is just too much typing
>>> - these values really do not make sense without each other, or partialy. I would say
>> that in array form they are even better
>>> readable, because it is the same order in which they go into the register.
>>
>> For timings - OK. PMC is separate. This is not a timing.
>
> But it's srom-config, and not srom-timing anymore. So do you want
> two properties like srom-timing + srom-page-mode (with vendor prefix
> of course)?
Yes, I still want separate properties because this too generic. In next
SoC they will extend the register and you will have to extend the
binding. I agree for simplicity reasons to group timings in an array.
But not entire register.
>> You need to document them. We are not gonna put some data looking like a
>> vendor blob into the binding. I understand that document has
>> confidential mark... all of Samsung datasheets I've seen had it. I don't
>> find it as problem but of course I am not the one to judge here. If you
>> do not feel comfortable publishing such data, get an approval from your
>> manager. You can also try to search for this in vendor code
>> (opensource.samsung.com). If it is published there, then this won't be a
>> disclosure.
>
> I've seen the actual SROMc settings in some old Android kernels, like 3.0.4.
> But they are not documented there, no comments, just #define's.
> Ok, i'll try to take a look how to minimize the actual information. :)
Actually 15 secs of search by grep revealed that they are already
documented and published. I looked at first image coming into my mind:
GT-I9300_JB
arch/arm/mach-exynos/include/mach/sromc-exynos4.h
>> BTW, your email client is weird. You are sending non-wrapped emails
>> without format=flowed in Content-Type. Please stick to mailing list
>> convention of wrapping at 72-char.
>
> I know, sorry, this is MS Outlook (not Express), and it's impossible
> to configure it this way (if i just set up word wrap, it starts to
> corrupt patches). I am now trying to wrap the text by hands, i hope
> it's better. And i cannot use anything else because... You know why. :)
Actually no... I don't know why...
All of us are using whatever we want (mutt+MDA, thunderbird, kmail,
etc.). At least in two locations I've been: in Polish R&D and in HQ.
Best regards,
Krzysztof
Hello!
> >>> --- cut exynos5410.dtsi ---
> >>> sromc: sromc@12250000 {
> >>> #address-cells = <2>;
> >>> #size-cells = <1>;
> >>> ranges = <0 0 0x04000000 0x20000
> >>> 1 0 0x05000000 0x20000
> >>> 2 0 0x06000000 0x20000
> >>> 3 0 0x07000000 0x20000>;
> >>
> >> Do you have to use 2 cells for address? Cannot it be:
> >> ranges = <0 0x04000000 0x20000
> >> 1 0x05000000 0x20000
> >> 2 0x06000000 0x20000
> >> 3 0x07000000 0x20000>;
> >
> > I tried this first, but it didn't work, and ranges translation
> > gave me something really weird (like addr = 0x80 and
> > size = 0x04000004).
>
> Did you change the address-cells to <1>?
Of course i did.
I don't quote the rest because i simply agree to what you've said. Will respin with these changes.
Kind regards,
Pavel Fedin
Expert Engineer
Samsung Electronics Research center Russia
On 03.11.2015 15:58, Pavel Fedin wrote:
> Hello!
>
>>>>> --- cut exynos5410.dtsi ---
>>>>> sromc: sromc@12250000 {
>>>>> #address-cells = <2>;
>>>>> #size-cells = <1>;
>>>>> ranges = <0 0 0x04000000 0x20000
>>>>> 1 0 0x05000000 0x20000
>>>>> 2 0 0x06000000 0x20000
>>>>> 3 0 0x07000000 0x20000>;
>>>>
>>>> Do you have to use 2 cells for address? Cannot it be:
>>>> ranges = <0 0x04000000 0x20000
>>>> 1 0x05000000 0x20000
>>>> 2 0x06000000 0x20000
>>>> 3 0x07000000 0x20000>;
>>>
>>> I tried this first, but it didn't work, and ranges translation
>>> gave me something really weird (like addr = 0x80 and
>>> size = 0x04000004).
>>
>> Did you change the address-cells to <1>?
>
> Of course i did.
I saw some other nodes use the ranges without the offset but maybe some
more steps are required for such configuration.
So anyway send the version with the child offset and I will try to
figure out why it is required.
Best regards,
Krzysztof