cpufreq: exynos: allow disabling cpufreq drivers
PATCH 1/1 cpufreq: exynos: allow disabling cpufreq drivers
Commit 6e6aac7590 "ARM: EXYNOS: Migrate clock support to common clock framework" broke support for the exynos cpufreq drivers. While we're waiting for a fix for this, let's get back to a state where the kernel builds again with the cpufreq subsystem enabled but ARM_EXYNOS_CPUFREQ disabled. I assume that this was the intention behind this Kconfig symbol anyway.
---
Rafael, I'm putting this patch into the arm-soc tree now to avoid a bug based on a patch that I got from Kukjin. If you have any objections, please let me know so I can revert it again.
Everyone else: Why does 6e6aac7590 have your "Tested-by" and "Signed-off-by" tags on it when it's obviously broken? Who is fixing this? Having no working exynos cpufreq support in 3.10 would be a serious regression.
drivers/cpufreq/Kconfig.arm +3 -0
@@ -61,18 +61,21 @@ config ARM_EXYNOS_CPUFREQ config ARM_EXYNOS4210_CPUFREQ def_bool CPU_EXYNOS4210 + depends on ARM_EXYNOS_CPUFREQ help This adds the CPUFreq driver for Samsung EXYNOS4210 SoC (S5PV310 or S5PC210). config ARM_EXYNOS4X12_CPUFREQ def_bool (SOC_EXYNOS4212 || SOC_EXYNOS4412) + depends on ARM_EXYNOS_CPUFREQ help This adds the CPUFreq driver for Samsung EXYNOS4X12 SoC (EXYNOS4212 or EXYNOS4412). config ARM_EXYNOS5250_CPUFREQ def_bool SOC_EXYNOS5250 + depends on ARM_EXYNOS_CPUFREQ help This adds the CPUFreq driver for Samsung EXYNOS5250 SoC.
Quoted 7 lines · Arnd Bergmann
Commit
6e6aac7590"ARM: EXYNOS: Migrate clock support to common clock framework" broke support for the exynos cpufreq drivers. While we're waiting for a fix for this, let's get back to a state where the kernel builds again with the cpufreq subsystem enabled butARM_EXYNOS_CPUFREQdisabled. I assume that this was the intention behind thisKconfigsymbol anyway.
Quoted 7 lines
Signed-off-by Arnd Bergmann <[email protected]>Cc Sylwester Nawrocki <[email protected]>Cc Tomasz Figa <[email protected]>Cc Thomas Abraham <[email protected]>Cc Kukjin Kim <[email protected]>Cc Rafael J. Wysocki <[email protected]>
Hmm, actually this conflicts heavily with the other Kconfig changes in the cpufreq tree. Any other idea how to resolve this?
Arnd
Quoted 9 lines
---
Rafael, I'm putting this patch into the arm-soc tree now to avoid a bug based on a patch that I got from Kukjin. If you have any objections, please let me know so I can revert it again.Everyone else: Why does
6e6aac7590have your "Tested-by" and "Signed-off-by" tags on it when it's obviously broken? Who is fixing this? Having no working exynos cpufreq support in 3.10 would be a serious regression.
Quoted 19 lines · Arnd Bergmann
Commit
6e6aac7590"ARM: EXYNOS: Migrate clock support to common clock framework" broke support for the exynos cpufreq drivers. While we're waiting for a fix for this, let's get back to a state where the kernel builds again with the cpufreq subsystem enabled butARM_EXYNOS_CPUFREQdisabled. I assume that this was the intention behind thisKconfigsymbol anyway.Signed-off-by Arnd Bergmann <[email protected]>Cc Sylwester Nawrocki <[email protected]>Cc Tomasz Figa <[email protected]>Cc Thomas Abraham <[email protected]>Cc Kukjin Kim <[email protected]>Cc Rafael J. Wysocki <[email protected]>---
Rafael, I'm putting this patch into the arm-soc tree now to avoid a bug based on a patch that I got from Kukjin. If you have any objections, please let me know so I can revert it again.
Quoted 4 lines
Everyone else: Why does
6e6aac7590have your "Tested-by" and "Signed-off-by" tags on it when it's obviously broken? Who is fixing this? Having no working exynos cpufreq support in 3.10 would be a serious regression.
Not sure what happened here, but my and Tomasz's Tested-by were given for v5 of the patch series including 6e6aac7590, while it seems v6 has been applied. And the patch is not same in both series.
There could be some differences in the base tree I used when testing this [1] series, as it was hard to determine against what tag it was created exactly. Also I didn't test in on all platforms, only Exynos4412.
And the cpufreq really needs to get fixed, as it now modifies registers which belong to the clock driver...
[1] http://www.spinics.net/lists/arm-kernel/msg214149.html
Thanks,
Sylwester
Quoted 3 lines · Sylwester Nawrocki
Not sure what happened here, but my and Tomasz's Tested-by were given for v5 of the patch series including
6e6aac7590, while it seems v6 has been applied. And the patch is not same in both series.
I see.
Quoted 4 lines
There could be some differences in the base tree I used when testing this [1] series, as it was hard to determine against what tag it was created exactly. Also I didn't test in on all platforms, only Exynos4412.
Maybe you just didn't have CPUFREQ enabled? I also didn't notice this bug at first because it is turned off in the defconfigs.
It could also be working by chance on EXYNOS4, since the register definitions were not actually removed from the regs-clk.h header for those.
Arnd
Quoted 2 lines · Arnd Bergmann
Maybe you just didn't have CPUFREQ enabled? I also didn't notice this bug at first because it is turned off in the defconfigs.
I certainly had it enabled. Since I remember I needed to turn it off temporarily after someone has introduced an incorrect change to the clock tree definition and cpufreq was unexpectedly switching parents of one of the root clocks at run time, which was causing malfunction of the camera image processing chain.
Anyway this wasn't a clean mainline kernel, as there have been recently lot of stuff pending upstream needed to have complete system with dt support. I assume after 3.10 release the situation is going to be much better.
Quoted 3 lines
It could also be working by chance on EXYNOS4, since the register definitions were not actually removed from the
regs-clk.hheader for those.
That's most likely what happened.
Thanks,
Sylwester
Quoted 5 lines · Sylwester Nawrocki
Quoted 3 lines
It could also be working by chance on EXYNOS4, since the register definitions were not actually removed from the
regs-clk.hheader for those.That's most likely what happened.
Ok. I think I see the bigger picture now. The EXYNOS5 cpufreq driver was actually just merged now, so I assume that by the time Thomas did the clock changes, the register definitions for EXYNOS5 were actually unused and he subsequently removed them from the kernel.
Since the driver now depends on them, I think the only way forward for 3.10 is to revert the change that removed the definitions and put them back in place.
We really should use the clock API from all the exynos cpufreq drivers, but it's too late for that change now in 3.10 and we will have to do it for 3.11 then.
This are not looking so good for exynos multiplatform support in 3.10 now, but if we get it in, that will mean we have to disable cpufreq when building multiplatform, but it can remain for the classic single-platform configurations.
Arnd
Quoted 7 lines · Arnd Bergmann
Quoted 5 lines · Sylwester Nawrocki
Quoted 3 lines
It could also be working by chance on EXYNOS4, since the register definitions were not actually removed from the
regs-clk.hheader for those.That's most likely what happened.
Quoted 8 lines
Ok. I think I see the bigger picture now. The EXYNOS5 cpufreq driver was actually just merged now, so I assume that by the time Thomas did the clock changes, the register definitions for EXYNOS5 were actually unused and he subsequently removed them from the kernel.
Since the driver now depends on them, I think the only way forward for 3.10 is to revert the change that removed the definitions and put them back in place.
Yes, it sounds like a best thing to do. I was also considering that.
Quoted 2 lines
We really should use the clock API from all the exynos cpufreq drivers, but it's too late for that change now in 3.10 and we will have to do it for 3.11 then.
Yes, the changes required to make the cpufreq driver using the clocks API would be rather significant.
Quoted 3 lines
This are not looking so good for exynos multiplatform support in 3.10 now, but if we get it in, that will mean we have to disable cpufreq when building multiplatform, but it can remain for the classic single-platform configurations.
Alright, let's hope eventually we get at least a minimum booting system. ;)
Thanks,
Sylwester
PATCH 1/1 ARM: exynos: restore mach/regs-clock.h for exynos5
In reply to Sylwester NawrockiCommit 6e6aac75 "ARM: EXYNOS: Migrate clock support to common clock framework" from Thomas Abraham removed the Exynos5 specific register definitions as they were unused at the time, but the cpufreq driver actually still uses them.
8<----
I'd apply this one on next/drivers to fix the build error, unless someone has concerns.
arch/arm/mach-exynos/include/mach/regs-clock.h +107 -0
@@ -256,6 +256,113 @@#define EXYNOS4_CLKDIV_CAM1_JPEG_SHIFT (0)#define EXYNOS4_CLKDIV_CAM1_JPEG_MASK (0xf << EXYNOS4_CLKDIV_CAM1_JPEG_SHIFT) +/* For EXYNOS5250 */ + +#define EXYNOS5_APLL_LOCK EXYNOS_CLKREG(0x00000) +#define EXYNOS5_APLL_CON0 EXYNOS_CLKREG(0x00100) +#define EXYNOS5_CLKSRC_CPU EXYNOS_CLKREG(0x00200) +#define EXYNOS5_CLKMUX_STATCPU EXYNOS_CLKREG(0x00400) +#define EXYNOS5_CLKDIV_CPU0 EXYNOS_CLKREG(0x00500) +#define EXYNOS5_CLKDIV_CPU1 EXYNOS_CLKREG(0x00504) +#define EXYNOS5_CLKDIV_STATCPU0 EXYNOS_CLKREG(0x00600) +#define EXYNOS5_CLKDIV_STATCPU1 EXYNOS_CLKREG(0x00604) + +#define EXYNOS5_PWR_CTRL1 EXYNOS_CLKREG(0x01020) +#define EXYNOS5_PWR_CTRL2 EXYNOS_CLKREG(0x01024) + +#define EXYNOS5_MPLL_CON0 EXYNOS_CLKREG(0x04100) +#define EXYNOS5_CLKSRC_CORE1 EXYNOS_CLKREG(0x04204) + +#define EXYNOS5_CLKGATE_IP_CORE EXYNOS_CLKREG(0x04900) + +#define EXYNOS5_CLKDIV_ACP EXYNOS_CLKREG(0x08500) + +#define EXYNOS5_EPLL_CON0 EXYNOS_CLKREG(0x10130) +#define EXYNOS5_EPLL_CON1 EXYNOS_CLKREG(0x10134) +#define EXYNOS5_EPLL_CON2 EXYNOS_CLKREG(0x10138) +#define EXYNOS5_VPLL_CON0 EXYNOS_CLKREG(0x10140) +#define EXYNOS5_VPLL_CON1 EXYNOS_CLKREG(0x10144) +#define EXYNOS5_VPLL_CON2 EXYNOS_CLKREG(0x10148) +#define EXYNOS5_CPLL_CON0 EXYNOS_CLKREG(0x10120) + +#define EXYNOS5_CLKSRC_TOP0 EXYNOS_CLKREG(0x10210) +#define EXYNOS5_CLKSRC_TOP1 EXYNOS_CLKREG(0x10214) +#define EXYNOS5_CLKSRC_TOP2 EXYNOS_CLKREG(0x10218) +#define EXYNOS5_CLKSRC_TOP3 EXYNOS_CLKREG(0x1021C) +#define EXYNOS5_CLKSRC_GSCL EXYNOS_CLKREG(0x10220) +#define EXYNOS5_CLKSRC_DISP1_0 EXYNOS_CLKREG(0x1022C) +#define EXYNOS5_CLKSRC_MAUDIO EXYNOS_CLKREG(0x10240) +#define EXYNOS5_CLKSRC_FSYS EXYNOS_CLKREG(0x10244) +#define EXYNOS5_CLKSRC_PERIC0 EXYNOS_CLKREG(0x10250) +#define EXYNOS5_CLKSRC_PERIC1 EXYNOS_CLKREG(0x10254) +#define EXYNOS5_SCLK_SRC_ISP EXYNOS_CLKREG(0x10270) + +#define EXYNOS5_CLKSRC_MASK_TOP EXYNOS_CLKREG(0x10310) +#define EXYNOS5_CLKSRC_MASK_GSCL EXYNOS_CLKREG(0x10320) +#define EXYNOS5_CLKSRC_MASK_DISP1_0 EXYNOS_CLKREG(0x1032C) +#define EXYNOS5_CLKSRC_MASK_MAUDIO EXYNOS_CLKREG(0x10334) +#define EXYNOS5_CLKSRC_MASK_FSYS EXYNOS_CLKREG(0x10340) +#define EXYNOS5_CLKSRC_MASK_PERIC0 EXYNOS_CLKREG(0x10350) +#define EXYNOS5_CLKSRC_MASK_PERIC1 EXYNOS_CLKREG(0x10354) + +#define EXYNOS5_CLKDIV_TOP0 EXYNOS_CLKREG(0x10510) +#define EXYNOS5_CLKDIV_TOP1 EXYNOS_CLKREG(0x10514) +#define EXYNOS5_CLKDIV_GSCL EXYNOS_CLKREG(0x10520) +#define EXYNOS5_CLKDIV_DISP1_0 EXYNOS_CLKREG(0x1052C) +#define EXYNOS5_CLKDIV_GEN EXYNOS_CLKREG(0x1053C) +#define EXYNOS5_CLKDIV_MAUDIO EXYNOS_CLKREG(0x10544) +#define EXYNOS5_CLKDIV_FSYS0 EXYNOS_CLKREG(0x10548) +#define EXYNOS5_CLKDIV_FSYS1 EXYNOS_CLKREG(0x1054C) +#define EXYNOS5_CLKDIV_FSYS2 EXYNOS_CLKREG(0x10550) +#define EXYNOS5_CLKDIV_FSYS3 EXYNOS_CLKREG(0x10554) +#define EXYNOS5_CLKDIV_PERIC0 EXYNOS_CLKREG(0x10558) +#define EXYNOS5_CLKDIV_PERIC1 EXYNOS_CLKREG(0x1055C) +#define EXYNOS5_CLKDIV_PERIC2 EXYNOS_CLKREG(0x10560) +#define EXYNOS5_CLKDIV_PERIC3 EXYNOS_CLKREG(0x10564) +#define EXYNOS5_CLKDIV_PERIC4 EXYNOS_CLKREG(0x10568) +#define EXYNOS5_CLKDIV_PERIC5 EXYNOS_CLKREG(0x1056C) +#define EXYNOS5_SCLK_DIV_ISP EXYNOS_CLKREG(0x10580) + +#define EXYNOS5_CLKGATE_IP_ACP EXYNOS_CLKREG(0x08800) +#define EXYNOS5_CLKGATE_IP_ISP0 EXYNOS_CLKREG(0x0C800) +#define EXYNOS5_CLKGATE_IP_ISP1 EXYNOS_CLKREG(0x0C804) +#define EXYNOS5_CLKGATE_IP_GSCL EXYNOS_CLKREG(0x10920) +#define EXYNOS5_CLKGATE_IP_DISP1 EXYNOS_CLKREG(0x10928) +#define EXYNOS5_CLKGATE_IP_MFC EXYNOS_CLKREG(0x1092C) +#define EXYNOS5_CLKGATE_IP_G3D EXYNOS_CLKREG(0x10930) +#define EXYNOS5_CLKGATE_IP_GEN EXYNOS_CLKREG(0x10934) +#define EXYNOS5_CLKGATE_IP_FSYS EXYNOS_CLKREG(0x10944) +#define EXYNOS5_CLKGATE_IP_GPS EXYNOS_CLKREG(0x1094C) +#define EXYNOS5_CLKGATE_IP_PERIC EXYNOS_CLKREG(0x10950) +#define EXYNOS5_CLKGATE_IP_PERIS EXYNOS_CLKREG(0x10960) +#define EXYNOS5_CLKGATE_BLOCK EXYNOS_CLKREG(0x10980) + +#define EXYNOS5_BPLL_CON0 EXYNOS_CLKREG(0x20110) +#define EXYNOS5_CLKSRC_CDREX EXYNOS_CLKREG(0x20200) +#define EXYNOS5_CLKDIV_CDREX EXYNOS_CLKREG(0x20500) + +#define EXYNOS5_PLL_DIV2_SEL EXYNOS_CLKREG(0x20A24) + +#define EXYNOS5_EPLL_LOCK EXYNOS_CLKREG(0x10030) + +#define EXYNOS5_EPLLCON0_LOCKED_SHIFT (29) + +#define PWR_CTRL1_CORE2_DOWN_RATIO (7 << 28) +#define PWR_CTRL1_CORE1_DOWN_RATIO (7 << 16) +#define PWR_CTRL1_DIV2_DOWN_EN (1 << 9) +#define PWR_CTRL1_DIV1_DOWN_EN (1 << 8) +#define PWR_CTRL1_USE_CORE1_WFE (1 << 5) +#define PWR_CTRL1_USE_CORE0_WFE (1 << 4) +#define PWR_CTRL1_USE_CORE1_WFI (1 << 1) +#define PWR_CTRL1_USE_CORE0_WFI (1 << 0) + +#define PWR_CTRL2_DIV2_UP_EN (1 << 25) +#define PWR_CTRL2_DIV1_UP_EN (1 << 24) +#define PWR_CTRL2_DUR_STANDBY2_VAL (1 << 16) +#define PWR_CTRL2_DUR_STANDBY1_VAL (1 << 8) +#define PWR_CTRL2_CORE2_UP_RATIO (1 << 4) +#define PWR_CTRL2_CORE1_UP_RATIO (1 << 0) +/* Compatibility defines and inclusion */ #include <mach/regs-pmu.h>
Re: [PATCH] ARM: exynos: restore mach/regs-clock.h for exynos5
Quoted 10 lines · Arnd Bergmann
Commit
6e6aac75"ARM: EXYNOS: Migrate clock support to common clock framework" from Thomas Abraham removed the Exynos5 specific register definitions as they were unused at the time, but the cpufreq driver actually still uses them.Cc Sylwester Nawrocki <[email protected]>Cc Tomasz Figa <[email protected]>Cc Thomas Abraham <[email protected]>Cc Kukjin Kim <[email protected]>Signed-off-by Arnd Bergmann <[email protected]>
Quoted 4 lines
8<----
I'd apply this one on next/drivers to fix the build error, unless someone has concerns.