| Message ID | 1406805732-17372-6-git-send-email-hsnaves@gmail.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
Hi Humberto, You can find my comments inline. On 31.07.2014 13:22, Humberto Silva Naves wrote: > Added the remaining PLL clocks, and also added the configuration > tables with the PLL coefficients for the supported frequencies. > These frequency tables are only installed when a 24MHz clock is > supplied as the input clock source. To reflect these changes, new > constants were added to the dt-bindings file. [snip] > +static struct samsung_pll_rate_table apll_24mhz_tbl[] __initdata = { > + /* sorted in descending order */ > + /* PLL_35XX_RATE(rate, m, p, s) */ > + PLL_35XX_RATE(2100000000, 175, 2, 0), > + PLL_35XX_RATE(2000000000, 250, 3, 0), > + PLL_35XX_RATE(1900000000, 475, 6, 0), > + PLL_35XX_RATE(1800000000, 225, 3, 0), > + PLL_35XX_RATE(1700000000, 425, 6, 0), > + PLL_35XX_RATE(1600000000, 200, 3, 0), > + PLL_35XX_RATE(1500000000, 250, 4, 0), > + PLL_35XX_RATE(1400000000, 175, 3, 0), > + PLL_35XX_RATE(1300000000, 325, 6, 0), > + PLL_35XX_RATE(1200000000, 100, 2, 0), > + PLL_35XX_RATE(1100000000, 275, 3, 1), > + PLL_35XX_RATE(1000000000, 250, 3, 1), > + PLL_35XX_RATE(900000000, 150, 2, 1), > + PLL_35XX_RATE(800000000, 200, 3, 1), > + PLL_35XX_RATE(700000000, 175, 3, 1), > + PLL_35XX_RATE(600000000, 100, 2, 1), > + PLL_35XX_RATE(500000000, 250, 3, 2), > + PLL_35XX_RATE(400000000, 200, 3, 2), > + PLL_35XX_RATE(300000000, 100, 2, 2), > + PLL_35XX_RATE(200000000, 200, 3, 3), nit: The numbers could be aligned to the right using spaces (see exynos4.c). > + { }, > +}; > + > +static struct samsung_pll_rate_table cpll_24mhz_tbl[] __initdata = { > + /* sorted in descending order */ > + /* PLL_35XX_RATE(rate, m, p, s) */ > + PLL_35XX_RATE(666000000, 222, 4, 1), > + PLL_35XX_RATE(640000000, 160, 3, 1), > + PLL_35XX_RATE(320000000, 160, 3, 2), > + { }, > +}; > + > +static struct samsung_pll_rate_table dpll_24mhz_tbl[] __initdata = { > + /* sorted in descending order */ > + /* PLL_35XX_RATE(rate, m, p, s) */ > + PLL_35XX_RATE(600000000, 200, 4, 1), > + { }, > +}; > + > +static struct samsung_pll_rate_table epll_24mhz_tbl[] __initdata = { > + /* sorted in descending order */ > + /* PLL_36XX_RATE(rate, m, p, s, k) */ > + PLL_36XX_RATE(600000000, 100, 2, 1, 0), > + PLL_36XX_RATE(400000000, 200, 3, 2, 0), > + PLL_36XX_RATE(200000000, 200, 3, 3, 0), > + PLL_36XX_RATE(180633600, 301, 5, 3, -3670), > + PLL_36XX_RATE( 67737600, 452, 5, 5, -27263), > + PLL_36XX_RATE( 49152000, 197, 3, 5, -25690), > + PLL_36XX_RATE( 45158401, 181, 3, 5, -24012), Have you ensured that the rates specified match the rates calculated using PLL equation? You can find how it is calculated in recalc_rate callback of this particular PLL type in clk-pll.c. As a side note, the PLL registration code should be made a bit more robust and just calculate the rates itself and printing warnings if they don't match the entered ones. I definitely need more hours in a day, so much to do. ;) > + { }, > +}; > + > +static struct samsung_pll_rate_table ipll_24mhz_tbl[] __initdata = { > + /* sorted in descending order */ > + /* PLL_35XX_RATE(rate, m, p, s, k) */ > + PLL_35XX_RATE(864000000, 288, 4, 1), > + PLL_35XX_RATE(666000000, 222, 4, 1), > + PLL_35XX_RATE(432000000, 288, 4, 2), > + { }, > +}; > + > +static struct samsung_pll_rate_table kpll_24mhz_tbl[] __initdata = { > + /* sorted in descending order */ > + /* PLL_35XX_RATE(rate, m, p, s) */ > + PLL_35XX_RATE(1500000000, 250, 4, 0), > + PLL_35XX_RATE(1400000000, 175, 3, 0), > + PLL_35XX_RATE(1300000000, 325, 6, 0), > + PLL_35XX_RATE(1200000000, 100, 2, 0), > + PLL_35XX_RATE(1100000000, 275, 3, 1), > + PLL_35XX_RATE(1000000000, 250, 3, 1), > + PLL_35XX_RATE(900000000, 150, 2, 1), > + PLL_35XX_RATE(800000000, 200, 3, 1), > + PLL_35XX_RATE(700000000, 175, 3, 1), > + PLL_35XX_RATE(600000000, 100, 2, 1), > + PLL_35XX_RATE(500000000, 250, 3, 2), > + PLL_35XX_RATE(400000000, 200, 3, 2), > + PLL_35XX_RATE(300000000, 100, 2, 2), > + PLL_35XX_RATE(200000000, 200, 3, 3), nit: Alignment. Otherwise looks good, thanks. Best regards, Tomasz -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Tomasz, I remember checking these rates on my calculator. You might notice the odd frequency of 45158401Hz (no pun intended) in the EPLL clock. This particular clock frequency was giving me a big headache in a previous project, since it was wrongly listed as 45158400. At first it seems innocuous, but whenever I would ask for one of the child clocks to change the rate, the driver would miscalculate the correct frequencies and errors would propagate up and down the clock tree. Anyway, I would double check these tables. And if you want, I can write a separate patch for the rate table validation. I presume that you would like to add a new field, such as default_base_freq, to the structure samsung_pll_clock, and if that field is non-zero, you perform the validation of the table in _samsung_clk_register_pll? Best, Humberto On Thu, Jul 31, 2014 at 3:07 PM, Tomasz Figa <tomasz.figa@gmail.com> wrote: > Hi Humberto, > > You can find my comments inline. > > On 31.07.2014 13:22, Humberto Silva Naves wrote: >> Added the remaining PLL clocks, and also added the configuration >> tables with the PLL coefficients for the supported frequencies. >> These frequency tables are only installed when a 24MHz clock is >> supplied as the input clock source. To reflect these changes, new >> constants were added to the dt-bindings file. > > [snip] > >> +static struct samsung_pll_rate_table apll_24mhz_tbl[] __initdata = { >> + /* sorted in descending order */ >> + /* PLL_35XX_RATE(rate, m, p, s) */ >> + PLL_35XX_RATE(2100000000, 175, 2, 0), >> + PLL_35XX_RATE(2000000000, 250, 3, 0), >> + PLL_35XX_RATE(1900000000, 475, 6, 0), >> + PLL_35XX_RATE(1800000000, 225, 3, 0), >> + PLL_35XX_RATE(1700000000, 425, 6, 0), >> + PLL_35XX_RATE(1600000000, 200, 3, 0), >> + PLL_35XX_RATE(1500000000, 250, 4, 0), >> + PLL_35XX_RATE(1400000000, 175, 3, 0), >> + PLL_35XX_RATE(1300000000, 325, 6, 0), >> + PLL_35XX_RATE(1200000000, 100, 2, 0), >> + PLL_35XX_RATE(1100000000, 275, 3, 1), >> + PLL_35XX_RATE(1000000000, 250, 3, 1), >> + PLL_35XX_RATE(900000000, 150, 2, 1), >> + PLL_35XX_RATE(800000000, 200, 3, 1), >> + PLL_35XX_RATE(700000000, 175, 3, 1), >> + PLL_35XX_RATE(600000000, 100, 2, 1), >> + PLL_35XX_RATE(500000000, 250, 3, 2), >> + PLL_35XX_RATE(400000000, 200, 3, 2), >> + PLL_35XX_RATE(300000000, 100, 2, 2), >> + PLL_35XX_RATE(200000000, 200, 3, 3), > > nit: The numbers could be aligned to the right using spaces (see exynos4.c). > >> + { }, >> +}; >> + >> +static struct samsung_pll_rate_table cpll_24mhz_tbl[] __initdata = { >> + /* sorted in descending order */ >> + /* PLL_35XX_RATE(rate, m, p, s) */ >> + PLL_35XX_RATE(666000000, 222, 4, 1), >> + PLL_35XX_RATE(640000000, 160, 3, 1), >> + PLL_35XX_RATE(320000000, 160, 3, 2), >> + { }, >> +}; >> + >> +static struct samsung_pll_rate_table dpll_24mhz_tbl[] __initdata = { >> + /* sorted in descending order */ >> + /* PLL_35XX_RATE(rate, m, p, s) */ >> + PLL_35XX_RATE(600000000, 200, 4, 1), >> + { }, >> +}; >> + >> +static struct samsung_pll_rate_table epll_24mhz_tbl[] __initdata = { >> + /* sorted in descending order */ >> + /* PLL_36XX_RATE(rate, m, p, s, k) */ >> + PLL_36XX_RATE(600000000, 100, 2, 1, 0), >> + PLL_36XX_RATE(400000000, 200, 3, 2, 0), >> + PLL_36XX_RATE(200000000, 200, 3, 3, 0), >> + PLL_36XX_RATE(180633600, 301, 5, 3, -3670), >> + PLL_36XX_RATE( 67737600, 452, 5, 5, -27263), >> + PLL_36XX_RATE( 49152000, 197, 3, 5, -25690), >> + PLL_36XX_RATE( 45158401, 181, 3, 5, -24012), > > Have you ensured that the rates specified match the rates calculated > using PLL equation? You can find how it is calculated in recalc_rate > callback of this particular PLL type in clk-pll.c. > > As a side note, the PLL registration code should be made a bit more > robust and just calculate the rates itself and printing warnings if they > don't match the entered ones. I definitely need more hours in a day, so > much to do. ;) > >> + { }, >> +}; >> + >> +static struct samsung_pll_rate_table ipll_24mhz_tbl[] __initdata = { >> + /* sorted in descending order */ >> + /* PLL_35XX_RATE(rate, m, p, s, k) */ >> + PLL_35XX_RATE(864000000, 288, 4, 1), >> + PLL_35XX_RATE(666000000, 222, 4, 1), >> + PLL_35XX_RATE(432000000, 288, 4, 2), >> + { }, >> +}; >> + >> +static struct samsung_pll_rate_table kpll_24mhz_tbl[] __initdata = { >> + /* sorted in descending order */ >> + /* PLL_35XX_RATE(rate, m, p, s) */ >> + PLL_35XX_RATE(1500000000, 250, 4, 0), >> + PLL_35XX_RATE(1400000000, 175, 3, 0), >> + PLL_35XX_RATE(1300000000, 325, 6, 0), >> + PLL_35XX_RATE(1200000000, 100, 2, 0), >> + PLL_35XX_RATE(1100000000, 275, 3, 1), >> + PLL_35XX_RATE(1000000000, 250, 3, 1), >> + PLL_35XX_RATE(900000000, 150, 2, 1), >> + PLL_35XX_RATE(800000000, 200, 3, 1), >> + PLL_35XX_RATE(700000000, 175, 3, 1), >> + PLL_35XX_RATE(600000000, 100, 2, 1), >> + PLL_35XX_RATE(500000000, 250, 3, 2), >> + PLL_35XX_RATE(400000000, 200, 3, 2), >> + PLL_35XX_RATE(300000000, 100, 2, 2), >> + PLL_35XX_RATE(200000000, 200, 3, 3), > > nit: Alignment. > > Otherwise looks good, thanks. > > Best regards, > Tomasz -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 31.07.2014 15:37, Humberto Naves wrote: > Hi Tomasz, > > I remember checking these rates on my calculator. You might notice the > odd frequency of 45158401Hz (no pun intended) in the EPLL clock. This > particular clock frequency was giving me a big headache in a previous > project, since it was wrongly listed as 45158400. At first it seems > innocuous, but whenever I would ask for one of the child clocks to > change the rate, the driver would miscalculate the correct frequencies > and errors would propagate up and down the clock tree. > > Anyway, I would double check these tables. And if you want, I can > write a separate patch for the rate table validation. I presume that > you would like to add a new field, such as default_base_freq, to the > structure samsung_pll_clock, and if that field is non-zero, you > perform the validation of the table in _samsung_clk_register_pll? I'm not sure I get the idea of the field you're suggesting. If I understand correctly, your intention would be to provide a default frequency if there is no table provided. I don't think there is a need for it, because current code can read back current settings from registers and calculate current rate. As for the validation itself, one more thing that needs to be considered is that the rate table must be sorted. We once decided to rely on the fact that tables in SoC drivers have rates explicitly specified and are correctly sorted, but now I'm inclined to reconsider this, based on the fact that those rates often are already incorrectly calculated in vendor code or even datasheets, which are main information sources for patch authors. Before mainlining PLL drivers (which was quite some time ago), we had a bit different implementation in our internal tree, which did not use explicitly specified rates at all (they could have been considered just comments to improve table readability) and the _register_pll() function simply calculated rates for all entries creating new table for internal use of the PLL driver that was in addition explicitly sorted to make sure that the order is correct. This kind of implementation is what I would lean toward today. Best regards, Tomasz -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi, On Thu, Jul 31, 2014 at 5:19 PM, Tomasz Figa <tomasz.figa@gmail.com> wrote: > > I'm not sure I get the idea of the field you're suggesting. If I > understand correctly, your intention would be to provide a default > frequency if there is no table provided. I don't think there is a need > for it, because current code can read back current settings from > registers and calculate current rate. > I think I was not clear enough. I am not trying to provide a default frequency for the clock, but I do want to specify the base frequency on which the rate table was based upon. Let me give you an example that will hopefully clarify the matter. Suppose I want to register my PLL, such as in the 5410. The *current* solution would be like this: static struct samsung_pll_rate_table ipll_24mhz_tbl[] __initdata = { /* PLL_35XX_RATE(rate, m, p, s, k) */ PLL_35XX_RATE(864000000, 288, 4, 1), { }, }; static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { [ipll] = PLL(pll_35xx, CLK_FOUT_IPLL, "fout_ipll", "fin_pll", IPLL_LOCK, IPLL_CON0, NULL), }; And in the driver initialization function, I would add the rate table if the input clock source matches what I expected, in this case 24Mhz: if (_get_rate("fin_pll") == 24 * MHZ) { exynos5410_plls[ipll].rate_table = ipll_24mhz_tbl; } An alternative approach would be as follows, we add a new field (say "base_rate") to the structure samsung_pll_clock, and to the macro PLL, and describe the pll table in a simpler way, such as follows: static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { [ipll] = PLL(pll_35xx, CLK_FOUT_IPLL, "fout_ipll", "fin_pll", IPLL_LOCK, IPLL_CON0, &ipll_24mhz_tbl, 24 * MHZ), }; and in the _samsung_clk_register_pll function, all the validation would be performed. Here I am talking about checking if the parent rate is what is specified on the samsung_pll_clock structure, and if so, to check if the rates on the rate table actually match what the coefficients are telling. > As for the validation itself, one more thing that needs to be considered > is that the rate table must be sorted. The _samsung_clk_register_pll could do that in theory. > > We once decided to rely on the fact that tables in SoC drivers have > rates explicitly specified and are correctly sorted, but now I'm > inclined to reconsider this, based on the fact that those rates often > are already incorrectly calculated in vendor code or even datasheets, > which are main information sources for patch authors. > > Before mainlining PLL drivers (which was quite some time ago), we had a > bit different implementation in our internal tree, which did not use > explicitly specified rates at all (they could have been considered just > comments to improve table readability) and the _register_pll() function > simply calculated rates for all entries creating new table for internal > use of the PLL driver that was in addition explicitly sorted to make > sure that the order is correct. This kind of implementation is what I > would lean toward today. I would strongly object to such as solution. I think that in the table, the frequency *must* be specified. As you said previously, the coefficients should be carefully chosen. We cannot know for sure that the same coefficients that work for a base frequency of 24 Mhz will also work for a different base frequency. So the driver cannot simply compute the frequency from the coefficients, and it must also check that the input rate is correct. This is another reason why I want to add the base_frequency field to that structure. I believe that the the _samsung_clk_register_pll must double-check if the frequencies match what the formula tells, and must drop the entries (or the whole frequency table) that are faulty. And then finally, it should sort the entries in descending order. I hope I made myself clear now. Best regards, Humberto > > Best regards, > Tomasz -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Humberto, [dropping few addresses from Cc as this topic is rather irrelevant for them and adding Mike and Sylwester] On 31.07.2014 23:19, Humberto Naves wrote: > Hi, > > On Thu, Jul 31, 2014 at 5:19 PM, Tomasz Figa <tomasz.figa@gmail.com> wrote: >> >> I'm not sure I get the idea of the field you're suggesting. If I >> understand correctly, your intention would be to provide a default >> frequency if there is no table provided. I don't think there is a need >> for it, because current code can read back current settings from >> registers and calculate current rate. >> > I think I was not clear enough. I am not trying to provide a default > frequency for the clock, but I do want to specify the base frequency > on which the rate table was based upon. Let me give you an example > that will hopefully clarify the matter. Suppose I want to register my > PLL, such as in the 5410. The *current* solution would be like this: > > static struct samsung_pll_rate_table ipll_24mhz_tbl[] __initdata = { > /* PLL_35XX_RATE(rate, m, p, s, k) */ > PLL_35XX_RATE(864000000, 288, 4, 1), > { }, > }; > static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { > [ipll] = PLL(pll_35xx, CLK_FOUT_IPLL, "fout_ipll", "fin_pll", IPLL_LOCK, > IPLL_CON0, NULL), > }; > > And in the driver initialization function, I would add the rate table > if the input clock source matches what I expected, in this case 24Mhz: > > if (_get_rate("fin_pll") == 24 * MHZ) { > exynos5410_plls[ipll].rate_table = ipll_24mhz_tbl; > } > > An alternative approach would be as follows, we add a new field (say > "base_rate") to the structure samsung_pll_clock, and to the macro PLL, > and describe the pll table in a simpler way, such as follows: > > static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { > [ipll] = PLL(pll_35xx, CLK_FOUT_IPLL, "fout_ipll", "fin_pll", IPLL_LOCK, > IPLL_CON0, &ipll_24mhz_tbl, 24 * MHZ), > }; > > and in the _samsung_clk_register_pll function, all the validation > would be performed. Here I am talking about checking if the parent > rate is what is specified on the samsung_pll_clock structure, and if > so, to check if the rates on the rate table actually match what the > coefficients are telling. What if there are multiple sets of tables for different "base rates"? Certain PLLs can run with several reference clock frequencies, e.g. VPLL rates are often specified for 24 MHz and 27 MHz, depending on setting of mout_vpllsrc. > >> As for the validation itself, one more thing that needs to be considered >> is that the rate table must be sorted. > > The _samsung_clk_register_pll could do that in theory. > >> >> We once decided to rely on the fact that tables in SoC drivers have >> rates explicitly specified and are correctly sorted, but now I'm >> inclined to reconsider this, based on the fact that those rates often >> are already incorrectly calculated in vendor code or even datasheets, >> which are main information sources for patch authors. >> >> Before mainlining PLL drivers (which was quite some time ago), we had a >> bit different implementation in our internal tree, which did not use >> explicitly specified rates at all (they could have been considered just >> comments to improve table readability) and the _register_pll() function >> simply calculated rates for all entries creating new table for internal >> use of the PLL driver that was in addition explicitly sorted to make >> sure that the order is correct. This kind of implementation is what I >> would lean toward today. > > I would strongly object to such as solution. I think that in the > table, the frequency *must* be specified. As you said previously, the > coefficients should be carefully chosen. We cannot know for sure that > the same coefficients that work for a base frequency of 24 Mhz will > also work for a different base frequency. So the driver cannot simply > compute the frequency from the coefficients, and it must also check > that the input rate is correct. This is another reason why I want to > add the base_frequency field to that structure. Sorry, I don't understand your concern. Currently it's SoC driver's duty to check which rate table to use depending on input clock rate. If input rate matches any table the driver has, it assigns the table pointer. Otherwise no table is used and the PLL is working in read-only mode. If we leave this as is, then PLL driver will have enough information to calculate PLL rate, because it can retrieve input clock rate by calling clk_get_rate() on its parent clock. No need to specify output_rate in rate table, as it is redundant. However... > > I believe that the the _samsung_clk_register_pll must double-check if > the frequencies match what the formula tells, and must drop the > entries (or the whole frequency table) that are faulty. And then > finally, it should sort the entries in descending order. Based on your proposal, another idea came to my mind. We can add input_rate to rate table instead, so that each entry will have its own input rate specified. Then register_pll() would do following: - validate all the entries by checking if entry.rate == calc_rate(entry.fin, entry.p, entry.m, ...), - discard invalid entries and print a warning, - sort the table. Resulting table would have entries for various input clock rates mixed together, but all sorted according to output rate. Then set_rate() when going through rate table would not only check the output_rate, but also whether input_rate matches and skip entries unsuitable for current setup. This would have the advantage of being able to work with input_rate which can dynamically change, e.g. when mout_vpllsrc setting changes, while no check for input_rate would have to be done in SoC driver at all. However I'm still not sure whether specifying output_rate in rate table has really any benefits. It's something that can be calculated precisely by the driver, so it introduces redundancy and unnecessarily increases error possibility, because you need to precisely calculate those rates yourself according to equation from the datasheet or reverse engineered from the code, because you often get incorrectly rounded values from the vendor. Not even saying that it makes it harder to add more frequencies to the table. I'd like to collect more opinions on this though and so altered Cc list as mentioned at the top. Mike, Sylwester, what do you think? Best regards, Tomasz -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoting Tomasz Figa (2014-07-31 15:17:29) > Humberto, > > [dropping few addresses from Cc as this topic is rather irrelevant for > them and adding Mike and Sylwester] > > On 31.07.2014 23:19, Humberto Naves wrote: > > Hi, > > > > On Thu, Jul 31, 2014 at 5:19 PM, Tomasz Figa <tomasz.figa@gmail.com> wrote: > >> > >> I'm not sure I get the idea of the field you're suggesting. If I > >> understand correctly, your intention would be to provide a default > >> frequency if there is no table provided. I don't think there is a need > >> for it, because current code can read back current settings from > >> registers and calculate current rate. > >> > > I think I was not clear enough. I am not trying to provide a default > > frequency for the clock, but I do want to specify the base frequency > > on which the rate table was based upon. Let me give you an example > > that will hopefully clarify the matter. Suppose I want to register my > > PLL, such as in the 5410. The *current* solution would be like this: > > > > static struct samsung_pll_rate_table ipll_24mhz_tbl[] __initdata = { > > /* PLL_35XX_RATE(rate, m, p, s, k) */ > > PLL_35XX_RATE(864000000, 288, 4, 1), > > { }, > > }; > > static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { > > [ipll] = PLL(pll_35xx, CLK_FOUT_IPLL, "fout_ipll", "fin_pll", IPLL_LOCK, > > IPLL_CON0, NULL), > > }; > > > > And in the driver initialization function, I would add the rate table > > if the input clock source matches what I expected, in this case 24Mhz: > > > > if (_get_rate("fin_pll") == 24 * MHZ) { > > exynos5410_plls[ipll].rate_table = ipll_24mhz_tbl; > > } > > > > An alternative approach would be as follows, we add a new field (say > > "base_rate") to the structure samsung_pll_clock, and to the macro PLL, > > and describe the pll table in a simpler way, such as follows: > > > > static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { > > [ipll] = PLL(pll_35xx, CLK_FOUT_IPLL, "fout_ipll", "fin_pll", IPLL_LOCK, > > IPLL_CON0, &ipll_24mhz_tbl, 24 * MHZ), > > }; > > > > and in the _samsung_clk_register_pll function, all the validation > > would be performed. Here I am talking about checking if the parent > > rate is what is specified on the samsung_pll_clock structure, and if > > so, to check if the rates on the rate table actually match what the > > coefficients are telling. > > What if there are multiple sets of tables for different "base rates"? > Certain PLLs can run with several reference clock frequencies, e.g. VPLL > rates are often specified for 24 MHz and 27 MHz, depending on setting of > mout_vpllsrc. > > > > >> As for the validation itself, one more thing that needs to be considered > >> is that the rate table must be sorted. > > > > The _samsung_clk_register_pll could do that in theory. > > > >> > >> We once decided to rely on the fact that tables in SoC drivers have > >> rates explicitly specified and are correctly sorted, but now I'm > >> inclined to reconsider this, based on the fact that those rates often > >> are already incorrectly calculated in vendor code or even datasheets, > >> which are main information sources for patch authors. > >> > >> Before mainlining PLL drivers (which was quite some time ago), we had a > >> bit different implementation in our internal tree, which did not use > >> explicitly specified rates at all (they could have been considered just > >> comments to improve table readability) and the _register_pll() function > >> simply calculated rates for all entries creating new table for internal > >> use of the PLL driver that was in addition explicitly sorted to make > >> sure that the order is correct. This kind of implementation is what I > >> would lean toward today. > > > > I would strongly object to such as solution. I think that in the > > table, the frequency *must* be specified. As you said previously, the > > coefficients should be carefully chosen. We cannot know for sure that > > the same coefficients that work for a base frequency of 24 Mhz will > > also work for a different base frequency. So the driver cannot simply > > compute the frequency from the coefficients, and it must also check > > that the input rate is correct. This is another reason why I want to > > add the base_frequency field to that structure. > > Sorry, I don't understand your concern. Currently it's SoC driver's duty > to check which rate table to use depending on input clock rate. If input > rate matches any table the driver has, it assigns the table pointer. > Otherwise no table is used and the PLL is working in read-only mode. > > If we leave this as is, then PLL driver will have enough information to > calculate PLL rate, because it can retrieve input clock rate by calling > clk_get_rate() on its parent clock. No need to specify output_rate in > rate table, as it is redundant. However... > > > > > I believe that the the _samsung_clk_register_pll must double-check if > > the frequencies match what the formula tells, and must drop the > > entries (or the whole frequency table) that are faulty. And then > > finally, it should sort the entries in descending order. > > Based on your proposal, another idea came to my mind. > > We can add input_rate to rate table instead, so that each entry will > have its own input rate specified. Then register_pll() would do following: > - validate all the entries by checking if entry.rate == > calc_rate(entry.fin, entry.p, entry.m, ...), > - discard invalid entries and print a warning, > - sort the table. This is how my "coordinated clock rates" approach works. A given coordinated rate table must specify the parent(s) and input rate(s). An assignee at Linaro was working on this but then got shuffled out so unfortunately the work is stalled. My secret plan is to replace the "cpu clocks" type stuff I've seen on the list recently with an approach that is baked into the clock framework core and is re-usable for everybody. Back on topic, the approach of specifying input rate for these patches is sane and helps a lot to sanity check. Regards, Mike > > Resulting table would have entries for various input clock rates mixed > together, but all sorted according to output rate. > > Then set_rate() when going through rate table would not only check the > output_rate, but also whether input_rate matches and skip entries > unsuitable for current setup. > > This would have the advantage of being able to work with input_rate > which can dynamically change, e.g. when mout_vpllsrc setting changes, > while no check for input_rate would have to be done in SoC driver at all. > > However I'm still not sure whether specifying output_rate in rate table > has really any benefits. It's something that can be calculated precisely > by the driver, so it introduces redundancy and unnecessarily increases > error possibility, because you need to precisely calculate those rates > yourself according to equation from the datasheet or reverse engineered > from the code, because you often get incorrectly rounded values from the > vendor. Not even saying that it makes it harder to add more frequencies > to the table. > > I'd like to collect more opinions on this though and so altered Cc list > as mentioned at the top. > > Mike, Sylwester, what do you think? > > Best regards, > Tomasz -- To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/clk/samsung/clk-exynos5410.c b/drivers/clk/samsung/clk-exynos5410.c index efbe734..9a6a371 100644 --- a/drivers/clk/samsung/clk-exynos5410.c +++ b/drivers/clk/samsung/clk-exynos5410.c @@ -157,7 +157,8 @@ /* list of PLLs */ enum exynos5410_plls { apll, cpll, mpll, - bpll, kpll, + bpll, kpll, dpll, + epll, ipll, vpll, nr_plls /* number of PLLs */ }; @@ -302,6 +303,7 @@ PNAME(mout_kfc_p) = { "mout_kpll", "sclk_mpll", }; PNAME(mpll_user_p) = { "fin_pll", "sclk_mpll", }; PNAME(bpll_user_p) = { "fin_pll", "sclk_bpll", }; PNAME(mpll_bpll_p) = { "sclk_mpll_muxed", "sclk_bpll_muxed", }; +PNAME(mout_vpllsrc_p) = { "fin_pll", "sclk_hdmi27m" }; PNAME(group2_p) = { "fin_pll", "fin_pll", "none", "none", "none", "none", "sclk_mpll_bpll", @@ -321,6 +323,10 @@ static struct samsung_fixed_rate_clock exynos5410_fixed_rate_clks[] __initdata = FRATE(0, "sclk_uhostphy", NULL, CLK_IS_ROOT, 48000000), }; +static struct samsung_mux_clock exynos5410_pll_pmux_clks[] __initdata = { + MUX(0, "mout_vpllsrc", mout_vpllsrc_p, SRC_TOP2, 0, 1), +}; + static struct samsung_mux_clock exynos5410_mux_clks[] __initdata = { MUX(0, "mout_apll", apll_p, SRC_CPU, 0, 1), MUX(0, "mout_cpu", mout_cpu_p, SRC_CPU, 16, 1), @@ -412,6 +418,107 @@ static struct samsung_gate_clock exynos5410_gate_clks[] __initdata = { SRC_MASK_PERIC0, 8, CLK_SET_RATE_PARENT, 0), }; +static struct samsung_pll_rate_table apll_24mhz_tbl[] __initdata = { + /* sorted in descending order */ + /* PLL_35XX_RATE(rate, m, p, s) */ + PLL_35XX_RATE(2100000000, 175, 2, 0), + PLL_35XX_RATE(2000000000, 250, 3, 0), + PLL_35XX_RATE(1900000000, 475, 6, 0), + PLL_35XX_RATE(1800000000, 225, 3, 0), + PLL_35XX_RATE(1700000000, 425, 6, 0), + PLL_35XX_RATE(1600000000, 200, 3, 0), + PLL_35XX_RATE(1500000000, 250, 4, 0), + PLL_35XX_RATE(1400000000, 175, 3, 0), + PLL_35XX_RATE(1300000000, 325, 6, 0), + PLL_35XX_RATE(1200000000, 100, 2, 0), + PLL_35XX_RATE(1100000000, 275, 3, 1), + PLL_35XX_RATE(1000000000, 250, 3, 1), + PLL_35XX_RATE(900000000, 150, 2, 1), + PLL_35XX_RATE(800000000, 200, 3, 1), + PLL_35XX_RATE(700000000, 175, 3, 1), + PLL_35XX_RATE(600000000, 100, 2, 1), + PLL_35XX_RATE(500000000, 250, 3, 2), + PLL_35XX_RATE(400000000, 200, 3, 2), + PLL_35XX_RATE(300000000, 100, 2, 2), + PLL_35XX_RATE(200000000, 200, 3, 3), + { }, +}; + +static struct samsung_pll_rate_table cpll_24mhz_tbl[] __initdata = { + /* sorted in descending order */ + /* PLL_35XX_RATE(rate, m, p, s) */ + PLL_35XX_RATE(666000000, 222, 4, 1), + PLL_35XX_RATE(640000000, 160, 3, 1), + PLL_35XX_RATE(320000000, 160, 3, 2), + { }, +}; + +static struct samsung_pll_rate_table dpll_24mhz_tbl[] __initdata = { + /* sorted in descending order */ + /* PLL_35XX_RATE(rate, m, p, s) */ + PLL_35XX_RATE(600000000, 200, 4, 1), + { }, +}; + +static struct samsung_pll_rate_table epll_24mhz_tbl[] __initdata = { + /* sorted in descending order */ + /* PLL_36XX_RATE(rate, m, p, s, k) */ + PLL_36XX_RATE(600000000, 100, 2, 1, 0), + PLL_36XX_RATE(400000000, 200, 3, 2, 0), + PLL_36XX_RATE(200000000, 200, 3, 3, 0), + PLL_36XX_RATE(180633600, 301, 5, 3, -3670), + PLL_36XX_RATE( 67737600, 452, 5, 5, -27263), + PLL_36XX_RATE( 49152000, 197, 3, 5, -25690), + PLL_36XX_RATE( 45158401, 181, 3, 5, -24012), + { }, +}; + +static struct samsung_pll_rate_table ipll_24mhz_tbl[] __initdata = { + /* sorted in descending order */ + /* PLL_35XX_RATE(rate, m, p, s, k) */ + PLL_35XX_RATE(864000000, 288, 4, 1), + PLL_35XX_RATE(666000000, 222, 4, 1), + PLL_35XX_RATE(432000000, 288, 4, 2), + { }, +}; + +static struct samsung_pll_rate_table kpll_24mhz_tbl[] __initdata = { + /* sorted in descending order */ + /* PLL_35XX_RATE(rate, m, p, s) */ + PLL_35XX_RATE(1500000000, 250, 4, 0), + PLL_35XX_RATE(1400000000, 175, 3, 0), + PLL_35XX_RATE(1300000000, 325, 6, 0), + PLL_35XX_RATE(1200000000, 100, 2, 0), + PLL_35XX_RATE(1100000000, 275, 3, 1), + PLL_35XX_RATE(1000000000, 250, 3, 1), + PLL_35XX_RATE(900000000, 150, 2, 1), + PLL_35XX_RATE(800000000, 200, 3, 1), + PLL_35XX_RATE(700000000, 175, 3, 1), + PLL_35XX_RATE(600000000, 100, 2, 1), + PLL_35XX_RATE(500000000, 250, 3, 2), + PLL_35XX_RATE(400000000, 200, 3, 2), + PLL_35XX_RATE(300000000, 100, 2, 2), + PLL_35XX_RATE(200000000, 200, 3, 3), + { }, +}; + +static struct samsung_pll_rate_table vpll_24mhz_tbl[] __initdata = { + /* sorted in descending order */ + /* PLL_36XX_RATE(rate, m, p, s, k) */ + PLL_36XX_RATE(880000000, 220, 3, 1, 0), + PLL_36XX_RATE(640000000, 160, 3, 1, 0), + PLL_36XX_RATE(532000000, 133, 3, 1, 0), + PLL_36XX_RATE(480000000, 240, 3, 2, 0), + PLL_36XX_RATE(440000000, 220, 3, 2, 0), + PLL_36XX_RATE(350000000, 175, 3, 2, 0), + PLL_36XX_RATE(333000000, 111, 2, 2, 0), + PLL_36XX_RATE(266000000, 133, 3, 2, 0), + PLL_36XX_RATE(177000000, 118, 2, 3, 0), + PLL_36XX_RATE(123500000, 330, 4, 4, 0), + PLL_36XX_RATE( 89000000, 178, 3, 4, 0), + { }, +}; + static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { [apll] = PLL(pll_35xx, CLK_FOUT_APLL, "fout_apll", "fin_pll", APLL_LOCK, APLL_CON0, NULL), @@ -423,6 +530,14 @@ static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { BPLL_CON0, NULL), [kpll] = PLL(pll_35xx, CLK_FOUT_KPLL, "fout_kpll", "fin_pll", KPLL_LOCK, KPLL_CON0, NULL), + [dpll] = PLL(pll_35xx, CLK_FOUT_DPLL, "fout_dpll", "fin_pll", DPLL_LOCK, + DPLL_CON0, NULL), + [epll] = PLL(pll_36xx, CLK_FOUT_EPLL, "fout_epll", "fin_pll", EPLL_LOCK, + EPLL_CON0, NULL), + [ipll] = PLL(pll_35xx, CLK_FOUT_IPLL, "fout_ipll", "fin_pll", IPLL_LOCK, + IPLL_CON0, NULL), + [vpll] = PLL(pll_36xx, CLK_FOUT_VPLL, "fout_vpll", "mout_vpllsrc", + VPLL_LOCK, VPLL_CON0, NULL), }; static const struct of_device_id ext_clk_match[] __initconst = { @@ -450,6 +565,19 @@ static void __init exynos5410_clk_init(struct device_node *np) samsung_clk_of_register_fixed_ext(ctx, exynos5410_fixed_rate_ext_clks, ARRAY_SIZE(exynos5410_fixed_rate_ext_clks), ext_clk_match); + samsung_clk_register_mux(ctx, exynos5410_pll_pmux_clks, + ARRAY_SIZE(exynos5410_pll_pmux_clks)); + + if (_get_rate("fin_pll") == 24 * MHZ) { + exynos5410_plls[apll].rate_table = apll_24mhz_tbl; + exynos5410_plls[cpll].rate_table = cpll_24mhz_tbl; + exynos5410_plls[kpll].rate_table = kpll_24mhz_tbl; + exynos5410_plls[dpll].rate_table = dpll_24mhz_tbl; + exynos5410_plls[epll].rate_table = epll_24mhz_tbl; + exynos5410_plls[ipll].rate_table = ipll_24mhz_tbl; + } + if (_get_rate("mout_vpllsrc") == 24 * MHZ) + exynos5410_plls[vpll].rate_table = vpll_24mhz_tbl; samsung_clk_register_pll(ctx, exynos5410_plls, ARRAY_SIZE(exynos5410_plls), reg_base); diff --git a/include/dt-bindings/clock/exynos5410.h b/include/dt-bindings/clock/exynos5410.h index 3a8da3c..7da296c 100644 --- a/include/dt-bindings/clock/exynos5410.h +++ b/include/dt-bindings/clock/exynos5410.h @@ -8,6 +8,10 @@ #define CLK_FOUT_MPLL 4 #define CLK_FOUT_BPLL 5 #define CLK_FOUT_KPLL 6 +#define CLK_FOUT_DPLL 7 +#define CLK_FOUT_EPLL 8 +#define CLK_FOUT_IPLL 9 +#define CLK_FOUT_VPLL 10 /* gate for special clocks (sclk) */ #define CLK_SCLK_UART0 128
Added the remaining PLL clocks, and also added the configuration tables with the PLL coefficients for the supported frequencies. These frequency tables are only installed when a 24MHz clock is supplied as the input clock source. To reflect these changes, new constants were added to the dt-bindings file. Furthermore, the definition of the clock "mout_vpllsrc" was added, as it is required for the VPLL. Signed-off-by: Humberto Silva Naves <hsnaves@gmail.com> --- drivers/clk/samsung/clk-exynos5410.c | 130 +++++++++++++++++++++++++++++++- include/dt-bindings/clock/exynos5410.h | 4 + 2 files changed, 133 insertions(+), 1 deletion(-)