Message ID | 1406805732-17372-5-git-send-email-hsnaves@gmail.com (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
(dropping linux-doc ML and Randy from Cc) On 31/07/14 13:22, Humberto Silva Naves wrote: > This implements the fixed rate clocks generated either inside or > outside the SoC. It also adds a dt-binding constant for the > sclk_hdmiphy clock, which shall be later used by other drivers, > such as the DRM. > > Since the external fixed rate clock fin_pll is now registered by > the clk-exynos5410 file, the bindings with the device tree file have > changed. It is no longer needed to define fin_pll as a fixed clock, > such as in: > > fin_pll: xxti { > compatible = "fixed-clock"; > clock-frequency = <24000000>; > clock-output-names = "fin_pll"; > #clock-cells = <0>; > }; > > The above lines should be replaced by the following lines: > > fixed-rate-clocks { > oscclk { > compatible = "samsung,exynos5410-oscclk"; > clock-frequency = <24000000>; > }; > }; > > This new form of binding was properly documented in the relevant > documentation file. Can you explain what is rationale behind this change ? Is it related to suspend/resume ordering ? Obviously it breaks the kernel/dtb compatibility. We should be moving in opposite direction, i.e. completely remove the custom samsung fixed rate clocks. I've tried to address this with patches [1], [2] but Tomasz wasn't happy with them IIRC and I postponed work on that. [1] http://www.spinics.net/lists/arm-kernel/msg333211.html [2] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-May/258490.html -- Regards, Sylwester -- 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 Humberto, You can find my comments inline. On 31.07.2014 13:22, Humberto Silva Naves wrote: > This implements the fixed rate clocks generated either inside or > outside the SoC. It also adds a dt-binding constant for the > sclk_hdmiphy clock, which shall be later used by other drivers, > such as the DRM. > > Since the external fixed rate clock fin_pll is now registered by > the clk-exynos5410 file, the bindings with the device tree file have > changed. It is no longer needed to define fin_pll as a fixed clock, > such as in: > > fin_pll: xxti { > compatible = "fixed-clock"; > clock-frequency = <24000000>; > clock-output-names = "fin_pll"; > #clock-cells = <0>; > }; > > The above lines should be replaced by the following lines: > > fixed-rate-clocks { > oscclk { > compatible = "samsung,exynos5410-oscclk"; > clock-frequency = <24000000>; > }; > }; > > This new form of binding was properly documented in the relevant > documentation file. In general this is backwards. This Exynos-specific clock binding was invented before generic fixed rate clock binding showed up and so few drivers still use it to maintain DT ABI compatibility. However new drivers are required to use the new generic binding and so does the one for Exynos5410. 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 perfectly see your point. However my question was why you did you decide to postpone Sylwester's? Was there any specific reason? I suppose it would break all the dtb compatibility, but besides that, was there any other reason? Best, Humberto On Thu, Jul 31, 2014 at 2:53 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: >> This implements the fixed rate clocks generated either inside or >> outside the SoC. It also adds a dt-binding constant for the >> sclk_hdmiphy clock, which shall be later used by other drivers, >> such as the DRM. >> >> Since the external fixed rate clock fin_pll is now registered by >> the clk-exynos5410 file, the bindings with the device tree file have >> changed. It is no longer needed to define fin_pll as a fixed clock, >> such as in: >> >> fin_pll: xxti { >> compatible = "fixed-clock"; >> clock-frequency = <24000000>; >> clock-output-names = "fin_pll"; >> #clock-cells = <0>; >> }; >> >> The above lines should be replaced by the following lines: >> >> fixed-rate-clocks { >> oscclk { >> compatible = "samsung,exynos5410-oscclk"; >> clock-frequency = <24000000>; >> }; >> }; >> >> This new form of binding was properly documented in the relevant >> documentation file. > > In general this is backwards. This Exynos-specific clock binding was > invented before generic fixed rate clock binding showed up and so few > drivers still use it to maintain DT ABI compatibility. However new > drivers are required to use the new generic binding and so does the one > for Exynos5410. > > 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 15:23, Humberto Naves wrote: > Hi Tomasz, > > I perfectly see your point. > However my question was why you did you decide to postpone > Sylwester's? Was there any specific reason? > I suppose it would break all the dtb compatibility, but besides that, > was there any other reason? We discussed this in private (we work in the same office) and I pointed out certain issues with Sylwester's proposed implementation and we agreed that one more revision of the patch is needed, but as it happens, higher priority tasks showed up and this one got lost in action. The first version of the patch [1] changed the original behavior breaking DT ABI compatibility and relied on improper assumption that those clocks are always in "fixed-rate-clocks" node. The thing is that no code should rely on DT node naming. Second version [2] was much better in this aspect, but it had some minor implementation issues - custom clk_ops used instead of generic mux clock and chipid block being constantly mapped and unmapped in every call to __fin_pll_mux_get_parent(). I should have reviewed them both on the mailing lists, but at that time there was no major activity related to Exynos4 outside of our office, so it was more convenient to just talk together directly. [1] http://www.spinics.net/lists/arm-kernel/msg333211.html [2] http://lists.infradead.org/pipermail/linux-arm-kernel/2014-May/258490.html Anyway, I don't think this is all relevant to Exynos5410, which just uses the generic fixed rate binding and has the thing done right from the start. Best regards, Tomasz > > Best, > Humberto > > On Thu, Jul 31, 2014 at 2:53 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: >>> This implements the fixed rate clocks generated either inside or >>> outside the SoC. It also adds a dt-binding constant for the >>> sclk_hdmiphy clock, which shall be later used by other drivers, >>> such as the DRM. >>> >>> Since the external fixed rate clock fin_pll is now registered by >>> the clk-exynos5410 file, the bindings with the device tree file have >>> changed. It is no longer needed to define fin_pll as a fixed clock, >>> such as in: >>> >>> fin_pll: xxti { >>> compatible = "fixed-clock"; >>> clock-frequency = <24000000>; >>> clock-output-names = "fin_pll"; >>> #clock-cells = <0>; >>> }; >>> >>> The above lines should be replaced by the following lines: >>> >>> fixed-rate-clocks { >>> oscclk { >>> compatible = "samsung,exynos5410-oscclk"; >>> clock-frequency = <24000000>; >>> }; >>> }; >>> >>> This new form of binding was properly documented in the relevant >>> documentation file. >> >> In general this is backwards. This Exynos-specific clock binding was >> invented before generic fixed rate clock binding showed up and so few >> drivers still use it to maintain DT ABI compatibility. However new >> drivers are required to use the new generic binding and so does the one >> for Exynos5410. >> >> 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 1:45 PM, Sylwester Nawrocki <s.nawrocki@samsung.com> wrote: > Can you explain what is rationale behind this change ? Is it related to > suspend/resume ordering ? I had forgotten, but now remember the reason why I did this. If you see the current implementation of clk-exynos5410, you will notice it heavily depends on the clock "fin_pll". On the other hand, this clock exists because in the current dtb (exynos5410-smdk5410.dts), there is a node fin_pll such as fin_pll: xxti { compatible = "fixed-clock"; clock-frequency = <24000000>; clock-output-names = "fin_pll"; #clock-cells = <0>; }; So far so good. But the real problem comes in when I check the rate of fin_pll to determine if I should install the rate table or not (and I really need this for my patch). More specifically 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; } I *have* to determine if the rate of fin_pll is 24MHz, and this is impossible to do if fin_pll is not available. Moreover, there is no way I can ensure that the fixed clock provider for fin_pll was initialized before mine, so there is chance that _get_rate won't work. The only way I fix that is to set the dependency explicitly in the dtb, by adding the fin_pll clock as required resource. clock: clock-controller@10010000 { compatible = "samsung,exynos5410-clock"; reg = <0x10010000 0x30000>; #clock-cells = <1>; /* Add the parent clock */ clocks = <&fin_pll>; clock-names = "fin_pll"; }; But in any case, the bindings with the DTB must be changed one way or another, because I *really* need to use fin_pll on my driver registration. If you agree with this alternative solution I previously described, I can change that in the next version of the patch series. Best regards, Humberto > Obviously it breaks the kernel/dtb compatibility. We should be moving > in opposite direction, i.e. completely remove the custom samsung fixed > rate clocks. -- 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, On 31.07.2014 23:01, Humberto Naves wrote: > Hi, > > On Thu, Jul 31, 2014 at 1:45 PM, Sylwester Nawrocki > <s.nawrocki@samsung.com> wrote: >> Can you explain what is rationale behind this change ? Is it related to >> suspend/resume ordering ? > > I had forgotten, but now remember the reason why I did this. If you > see the current implementation of clk-exynos5410, you will notice it > heavily depends on the clock "fin_pll". On the other hand, this clock > exists because in the current dtb (exynos5410-smdk5410.dts), there is > a node fin_pll such as > > fin_pll: xxti { > compatible = "fixed-clock"; > clock-frequency = <24000000>; > clock-output-names = "fin_pll"; > #clock-cells = <0>; > }; > > So far so good. But the real problem comes in when I check the rate of > fin_pll to determine if I should install the rate table or not (and I > really need this for my patch). More specifically > > 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; > } > > I *have* to determine if the rate of fin_pll is 24MHz, and this is > impossible to do if fin_pll is not available. Moreover, there is no > way I can ensure that the fixed clock provider for fin_pll was > initialized before mine, so there is chance that _get_rate won't work. > The only way I fix that is to set the dependency explicitly in the > dtb, by adding the fin_pll clock as required resource. > > clock: clock-controller@10010000 { > compatible = "samsung,exynos5410-clock"; > reg = <0x10010000 0x30000>; > #clock-cells = <1>; > /* Add the parent clock */ > clocks = <&fin_pll>; > clock-names = "fin_pll"; > }; This is the correct solution to your problem. The clocks and clock-names properties should have been there from the beginning but apparently this has been missed in review. Also see below. > > But in any case, the bindings with the DTB must be changed one way or > another, because I *really* need to use fin_pll on my driver > registration. This is a backwards compatible change. On DTBs without clocks and clock-names properties the PLL tables simply won't be registered which is exactly the same behavior we have now without any tables in the driver at all. > If you agree with this alternative solution I previously > described, I can change that in the next version of the patch series. Please update the dts instead, in the way you pointed above. 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/Documentation/devicetree/bindings/clock/exynos5410-clock.txt b/Documentation/devicetree/bindings/clock/exynos5410-clock.txt index aeab635..9f4a286 100644 --- a/Documentation/devicetree/bindings/clock/exynos5410-clock.txt +++ b/Documentation/devicetree/bindings/clock/exynos5410-clock.txt @@ -18,12 +18,21 @@ tree sources. External clock: -There is clock that is generated outside the SoC. It -is expected that it is defined using standard clock bindings -with following clock-output-name: - +There is a clock that is generated outside the SoC, namely - "fin_pll" - PLL input clock from XXTI +It is expected that a binding compatible with "samsung,exynos5410-oscclk" +having a populated clock-frequency field such as follows to be used in +order to define this external clock: + + fixed-rate-clocks { + oscclk { + compatible = "samsung,exynos5410-oscclk"; + clock-frequency = <24000000>; + }; + }; + + Example 1: An example of a clock controller node is listed below. clock: clock-controller@0x10010000 { diff --git a/drivers/clk/samsung/clk-exynos5410.c b/drivers/clk/samsung/clk-exynos5410.c index a9c261c..efbe734 100644 --- a/drivers/clk/samsung/clk-exynos5410.c +++ b/drivers/clk/samsung/clk-exynos5410.c @@ -307,6 +307,20 @@ PNAME(group2_p) = { "fin_pll", "fin_pll", "none", "none", "none", "none", "sclk_mpll_bpll", "none", "none", "sclk_cpll" }; +/* fixed rate clocks generated outside the soc */ +static struct samsung_fixed_rate_clock exynos5410_fixed_rate_ext_clks[] __initdata = { + FRATE(CLK_FIN_PLL, "fin_pll", NULL, CLK_IS_ROOT, 0), +}; + +/* fixed rate clocks generated inside the soc */ +static struct samsung_fixed_rate_clock exynos5410_fixed_rate_clks[] __initdata = { + FRATE(CLK_SCLK_HDMIPHY, "sclk_hdmiphy", NULL, CLK_IS_ROOT, 24000000), + FRATE(0, "sclk_hdmi24m", NULL, CLK_IS_ROOT, 24000000), + FRATE(0, "sclk_hdmi27m", NULL, CLK_IS_ROOT, 27000000), + FRATE(0, "sclk_dptxphy", NULL, CLK_IS_ROOT, 24000000), + FRATE(0, "sclk_uhostphy", NULL, CLK_IS_ROOT, 48000000), +}; + 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), @@ -411,6 +425,11 @@ static struct samsung_pll_clock exynos5410_plls[nr_plls] __initdata = { KPLL_CON0, NULL), }; +static const struct of_device_id ext_clk_match[] __initconst = { + { .compatible = "samsung,exynos5410-oscclk", .data = (void *)0, }, + { }, +}; + /* register exynos5410 clocks */ static void __init exynos5410_clk_init(struct device_node *np) { @@ -428,9 +447,14 @@ static void __init exynos5410_clk_init(struct device_node *np) if (!ctx) panic("%s: unable to allocate context.\n", __func__); + 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_pll(ctx, exynos5410_plls, ARRAY_SIZE(exynos5410_plls), reg_base); - + samsung_clk_register_fixed_rate(ctx, exynos5410_fixed_rate_clks, + ARRAY_SIZE(exynos5410_fixed_rate_clks)); samsung_clk_register_mux(ctx, exynos5410_mux_clks, ARRAY_SIZE(exynos5410_mux_clks)); samsung_clk_register_div(ctx, exynos5410_div_clks, diff --git a/include/dt-bindings/clock/exynos5410.h b/include/dt-bindings/clock/exynos5410.h index 9b180f0..3a8da3c 100644 --- a/include/dt-bindings/clock/exynos5410.h +++ b/include/dt-bindings/clock/exynos5410.h @@ -17,6 +17,7 @@ #define CLK_SCLK_MMC0 132 #define CLK_SCLK_MMC1 133 #define CLK_SCLK_MMC2 134 +#define CLK_SCLK_HDMIPHY 135 /* gate clocks */ #define CLK_UART0 257
This implements the fixed rate clocks generated either inside or outside the SoC. It also adds a dt-binding constant for the sclk_hdmiphy clock, which shall be later used by other drivers, such as the DRM. Since the external fixed rate clock fin_pll is now registered by the clk-exynos5410 file, the bindings with the device tree file have changed. It is no longer needed to define fin_pll as a fixed clock, such as in: fin_pll: xxti { compatible = "fixed-clock"; clock-frequency = <24000000>; clock-output-names = "fin_pll"; #clock-cells = <0>; }; The above lines should be replaced by the following lines: fixed-rate-clocks { oscclk { compatible = "samsung,exynos5410-oscclk"; clock-frequency = <24000000>; }; }; This new form of binding was properly documented in the relevant documentation file. Signed-off-by: Humberto Silva Naves <hsnaves@gmail.com> --- .../devicetree/bindings/clock/exynos5410-clock.txt | 17 ++++++++++--- drivers/clk/samsung/clk-exynos5410.c | 26 +++++++++++++++++++- include/dt-bindings/clock/exynos5410.h | 1 + 3 files changed, 39 insertions(+), 5 deletions(-)