On Thursday 06 of June 2013 14:31:56 Yadwinder Singh Brar wrote: > On Wed, Jun 5, 2013 at 7:10 PM, Tomasz Figa <t.figa@xxxxxxxxxxx> wrote: > > Hi Yadwinder, > > > > On Tuesday 04 of June 2013 17:02:48 Yadwinder Singh Brar wrote: > >> Hi Tomasz, > >> > >> On Mon, Jun 3, 2013 at 8:55 PM, Tomasz Figa <t.figa@xxxxxxxxxxx> wrote: > >> > Hi Yadwinder, > >> > > >> > On Thursday 30 of May 2013 12:25:40 Yadwinder Singh Brar wrote: > >> >> Hi Doug, > >> >> > >> >> On Thu, May 30, 2013 at 5:14 AM, Doug Anderson > >> >> <dianders@xxxxxxxxxxxx> > >> > > >> > wrote: > >> >> > Vikas / Yadwinder, > >> >> > > >> >> > On Wed, May 29, 2013 at 6:37 AM, Vikas Sajjan > >> >> > <vikas.sajjan@xxxxxxxxxx> > >> > > >> > wrote: > >> >> >> From: Yadwinder Singh Brar <yadi.brar@xxxxxxxxxxx> > >> >> >> > >> >> >> This patch defines a common rate_table which will contain > >> >> >> recommended > >> >> >> p, > >> >> >> m, s, k values for supported rates that needs to be changed for > >> >> >> changing > >> >> >> corresponding PLL's rate. > >> >> >> > >> >> >> Signed-off-by: Yadwinder Singh Brar <yadi.brar@xxxxxxxxxxx> > >> >> >> --- > >> >> >> > >> >> >> drivers/clk/samsung/clk-exynos4.c | 8 ++++---- > >> >> >> drivers/clk/samsung/clk-exynos5250.c | 14 +++++++------- > >> >> >> drivers/clk/samsung/clk-pll.c | 14 ++++++++++++-- > >> >> >> drivers/clk/samsung/clk-pll.h | 33 > >> >> >> +++++++++++++++++++++++++++++++-- 4 files changed, 54 > >> >> >> insertions(+), > >> >> >> 15 > >> >> >> deletions(-) > >> >> > > >> >> > I also reviewed this in our gerrit > >> >> > <https://gerrit.chromium.org/gerrit/#/c/56742/>, but I'll > >> >> > summarize > >> >> > here for the list... > >> >> > > >> >> >> struct clk * __init samsung_clk_register_pll35xx(const char > >> >> >> *name, > >> >> >> > >> >> >> - const char *pname, const void __iomem > >> >> >> *base) > >> >> >> + const char *pname, const void __iomem > >> >> >> *base, > >> >> >> + const struct samsung_pll_rate_table > >> >> >> *rate_table, > >> >> >> + const unsigned int rate_count) > >> >> > > >> >> > Feels like you should document here that rate_table needs to be > >> >> > sorted > >> >> > and the sort order. > >> >> > >> >> sure, we will add comment to sort the table in descending order. > >> >> > >> >> >> +struct samsung_pll_rate_table { > >> >> >> + unsigned int rate; > >> >> > > >> >> > nit: extra space before "int" should be removed. > >> >> > >> >> ok > >> >> > >> >> > Also: you can include rate here if you need a convenient place > >> >> > to > >> >> > store it (which sadly means that this structure can't be const). > >> >> > ...but I do like Tomasz's idea of actually calculating it. You > >> >> > can't > >> >> > know it at compile time since the parent rate comes from the > >> >> > device > >> >> > tree. > >> >> > > >> >> > compatible = "samsung,clock-xxti"; > >> >> > clock-frequency = <24000000>; > >> >> > >> >> Actually this table should contain the recommended values > >> >> and if we see the user manual, the input(parent) rate is also a > >> >> part > >> >> of recommended > >> >> table of different output rate for a particular PLL for that SoC. > >> > > >> > From what I understood in the documentation is that there is a set > >> > of > >> > recommended P, M, S (, K) tuples for each PLL and they are not > >> > dependent on > >> > input frequency - f_in and f_out are provided in the table just for > >> > reference to see the relation between output frequency and input > >> > frequency. > >> > >> If input rate(f_in) gets changed for PLL, then the corresponding > >> output rates(f_out) > >> will change by using same(common) set of recommended P, M, S (, K), > >> and the new set of output rates(f_out) may not be the _required_ set > >> of target rates. > >> So we need different set of P, M, S (,K) values for different f_in. > >> > >> Table should contain set of P, M, S (,K) values for the _required_ > >> target(f_out) rates > >> for a particular input(f_in) rate. > > > > I'm not sure what required rates you are talking about. For most PLLs > > the list of available frequencies seems to be defined by the set of > > supported PMS values and those are the frequencies that can be set by > > consumers, not the other way around. > > > > There are some cases when some particular frequency must be generated, > > like PLLs used for audio or video clock sources, but this just means > > that there must be an entry in PMS table that gives PLL configuration > > which produces such frequency with used PLL input clock. > > Yes, I was talking about same cases. Perhaps,I think it holds true in > general for PLLs, whose set_rate can be called. Not really. For PLLs used for general purpose clocking, like for CPU or GPU, there is usually no need to achieve certain frequency. Actually you might want to have as many as possible intermediate steps to improve frequency scaling efficiency. > >> > I think we should ask some H/W engineer about that to make sure and > >> > choose the proper implementation, which will work properly for > >> > future cases, instead of ending with something that works just > >> > with current cases.>> > >> We already asked hardware engineer about PMS values for PLL, > >> and we got a table containing recommended P, M ,S values for a given > >> f_in(24 MHz) > >> and required f_out rates. > > > > This doesn't answer the question about using those PMS values with > > different input rate. > > > > Kukjin, maybe you can give some information on this or point to a > > person who could? > > > >> Please let me know, why you think we are specific to current cases > >> only ?> > > I don't like the idea of having separate tables for each input rate. > > Just imagine how much data would have to be added if boards with > > several different input rates showed up. > > I think by declaring all these different table as __init, > and creating a copy to be used while registering, we can > save a little memory as well as the unnecessary cost which > we have to bear while doing set_rate() and round_rate() > by using single big table(for some time critical cases). Well, this is not really about memory, but rather the amount of code in the driver. The amount of memory needed for such setting table is rather insignificant. > > Sure, if each input rate requires different set of PMS values by > > hardware design, it can't be helped, but if the same PMS values can > > be reused (even giving different output rates, but this is usually > > not important), one table can be used for all cases. > > Its good to know whether same PMS values can be reused, from hardware > guys. Another thing worth knowing would be whether a different frequency of input clock is really allowed or the SoC supports only 24 MHz. If the latter is the case, then there is no need to bother supporting other frequencies. I'm waiting for some opinion from hardware engineers then. Best regards, Tomasz > Using different tables we can always blindly stick to recommended values > without bothering much every time, whether existing PMS can be reused > or not while adding new cases. > > >> Regards, > >> Yadwinder > >> > >> > Best regards, > >> > -- > >> > Tomasz Figa > >> > Linux Kernel Developer > >> > Samsung R&D Institute Poland > >> > Samsung Electronics > >> > > >> >> So as Tomasz said input(parent) rate may change with board, > >> >> then, do those corresponding recommended p, m, s, k will be valid? > >> >> > >> >> In case, input(parent) rate changes then we may need different set > >> >> of p, m > >> >> ,s, k values recommended for new input rate to get > >> >> required(recommended > >> >> to > >> >> use) output rate. > >> >> > >> >> So, we think its better that the p, m, s and k along with the > >> >> parent > >> >> is known at the compile time ( or DT ?), > >> >> as these p, m, s, k values are very much coupled with the parent > >> >> rate > >> >> to achieve the > >> >> > >> >> required(recommended to use) output rate. > >> >> > >> >> Also, since the sorted table is required (sorted based on "rate"), > >> >> its better to have the rate in a const rate table. > >> >> > >> >> And the whole set of recommended values should come from same > >> >> place(DT > >> >> or static table), > >> >> to keep the things simple and consistent. > >> >> > >> >> Moreover, practically for a particular SoC , we use the > >> >> recommended > >> >> input(parent) rate only for a PLL. > >> >> So we should keep the things simple here presently. > >> >> > >> >> >> + unsigned int pdiv; > >> >> >> + unsigned int mdiv; > >> >> >> + unsigned int sdiv; > >> >> >> + unsigned int kdiv; > >> >> > > >> >> > I think kdiv is signed. > >> >> > >> >> No, as these values should be the recommended values to be written > >> >> in > >> >> corresponding register bits. So it should remain unsigned. > >> >> > >> >> K value should be considered as negative only while recalculating > >> >> rate. > >> >> > >> >> As per exynos5250 user manual's section 7.3.2 : > >> >> " K value description "Postive value (Negative value)": > >> >> Postive values is that you should write EPLLCON/VPLLCON register. > >> >> Negative value is that you can calculate PLL output frequency with > >> >> it." > >> >> > >> >> > -Doug > >> >> > >> >> Regards, > >> >> Yadwinder & Vikas. > >> >> -- > >> >> To unsubscribe from this list: send the line "unsubscribe > >> >> linux-samsung-soc" in the body of a message to > >> >> majordomo@xxxxxxxxxxxxxxx More majordomo info at > >> >> http://vger.kernel.org/majordomo-info.html>> > >> -- > >> To unsubscribe from this list: send the line "unsubscribe > >> linux-samsung-soc" in the body of a message to > >> majordomo@xxxxxxxxxxxxxxx > >> More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- > To unsubscribe from this list: send the line "unsubscribe > linux-samsung-soc" in the body of a message to > majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
 |