[v10,0/4] This suspend patch is only support cut off the power of cpu and some external
mbox series

Message ID 1417423940-1669-1-git-send-email-zyw@rock-chips.com
Headers show
Series
  • This suspend patch is only support cut off the power of cpu and some external
Related show

Message

Chris Zhong Dec. 1, 2014, 8:52 a.m. UTC
devices, since we still lack power_domain driver, so the other power rail
of rk3288 need keep power on.
I have tested it on rk3288-evb board, atop next-20141112. goto suspend by type
"echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so it can be
determined in sleep mode, then press power button to wakeup it.

Changes in v10:
- remove regulator-suspend-mem-enabled and regulator-suspend-mem-microvolt
- enable the lcd, codec, sdmmc power during suspend

Changes in v9:
- fold Doug's patches
- modify some print log
- update the subject and description

Changes in v8:
- use enum for define sleep mode
- move rk3288_config_bootdata to the front of sram memcpy
- add ddr pinctrl for suspend
- keep all except cpu&tp power rail on during suspend
- add regulator-on-in-suspend before set suspend voltage
- add a reference of ddrio_pwroff and ddr0_retention

Changes in v7:
- get rid all of unused code
- add regulator-state-mem sub node for suspend

Changes in v6:
- get rid of the save/restore of SRAM
- doing the copy of resume code once at init time
- remove ROCKCHIP_ARM_OFF_LOGIC_DEEP from rk3288_fill_in_bootram
- add of_platform_populate in rockchip_dt_init
- change pmu_intmem@ff720000 to sram@ff720000
- change pmu_intmem@ff720000 to sram@ff720000

Changes in v5:
- use rk3288_bootram_sz for memcpy size
- fixed error of sram save and restore
- change the size of sram in example
- change size to 4k

Changes in v4:
- remove grf regmap

Changes in v3:
- move the pinmux of gpio6_c6 save and restore to pinctrl-rockchip

Changes in v2:
- add the regulator calls in prepare and finish.
- add the pinmux of gpio6_c6 save and restore
- put "rockchip,rk3288-pmu-sram" to first

Chris Zhong (4):
  ARM: rockchip: add suspend and resume for RK3288
  ARM: rockchip: Add pmu-sram binding
  ARM: dts: add RK3288 suspend support
  ARM: dts: rockchip: add suspend settings for rk3288-evb-rk808

 .../devicetree/bindings/arm/rockchip/pmu-sram.txt  |  16 ++
 arch/arm/boot/dts/rk3288-evb-rk808.dts             |  53 ++++-
 arch/arm/boot/dts/rk3288.dtsi                      |  23 ++
 arch/arm/mach-rockchip/Makefile                    |   1 +
 arch/arm/mach-rockchip/pm.c                        | 260 +++++++++++++++++++++
 arch/arm/mach-rockchip/pm.h                        |  99 ++++++++
 arch/arm/mach-rockchip/rockchip.c                  |   2 +
 arch/arm/mach-rockchip/sleep.S                     |  73 ++++++
 8 files changed, 526 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/devicetree/bindings/arm/rockchip/pmu-sram.txt
 create mode 100644 arch/arm/mach-rockchip/pm.c
 create mode 100644 arch/arm/mach-rockchip/pm.h
 create mode 100644 arch/arm/mach-rockchip/sleep.S

Comments

Kevin Hilman Dec. 1, 2014, 7:51 p.m. UTC | #1
Chris Zhong <zyw@rock-chips.com> writes:

> devices, since we still lack power_domain driver, so the other power rail
> of rk3288 need keep power on.
> I have tested it on rk3288-evb board, atop next-20141112. goto suspend by type
> "echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so it can be
> determined in sleep mode, then press power button to wakeup it.

I tested this on top of today's linux-next (next-20141201) and it
suspends, but doesn't wake up from any of the button presses.  What
wakeup sources are configured for the rk3288-evb-rk808?

Kevin
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Douglas Anderson Dec. 1, 2014, 10:08 p.m. UTC | #2
Hi,

On Mon, Dec 1, 2014 at 11:51 AM, Kevin Hilman <khilman@kernel.org> wrote:
> Chris Zhong <zyw@rock-chips.com> writes:
>
>> devices, since we still lack power_domain driver, so the other power rail
>> of rk3288 need keep power on.
>> I have tested it on rk3288-evb board, atop next-20141112. goto suspend by type
>> "echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so it can be
>> determined in sleep mode, then press power button to wakeup it.
>
> I tested this on top of today's linux-next (next-20141201) and it
> suspends, but doesn't wake up from any of the button presses.  What
> wakeup sources are configured for the rk3288-evb-rk808?

Just to close the loop (I talked with Kevin over IM about this, too):

I have a huge description of how I tested this as part of my patch at
<https://patchwork.kernel.org/patch/5414941/>.  Chris:  I think Kevin
has asked you several times to include information like this in your
cover letter.  Please, please, please can you try to remember to do
this?

For those that don't want to click on my link, I'll include the
relevant bits here:

---

Total patches atop that version of Linux were:

1. https://patchwork.kernel.org/patch/5051881/ - clocksource:
   arch_timer: Allow the device tree to specify uninitialized timer
   registers

2. https://patchwork.kernel.org/patch/5363671/ - clocksource:
   arch_timer: Fix code to use physical timers when requested

3. https://patchwork.kernel.org/patch/5382141/ - ARM: dts: rk3288: add
   arm,cpu-registers-not-fw-configured

4. Revert (b77d439 ARM: dts: rockchip: temporarily disable smp on
   rk3288)

5. https://patchwork.kernel.org/patch/5325111/ - usb: dwc2: resume
   root hub when device detect with suspend state

6. https://patchwork.kernel.org/patch/5410611/ - ARM: rockchip: add
   suspend and resume for RK3288

7. https://patchwork.kernel.org/patch/5410621/ - ARM: rockchip: Add
   pmu-sram binding

8. https://patchwork.kernel.org/patch/5410631/ - ARM: dts: add RK3288
   suspend support

9. https://patchwork.kernel.org/patch/5410641/ - ARM: dts: rockchip:
   add suspend settings for rk3288-evb-rk808

It looks like my pinctrl patches might be dropped due to cross
dependency problems, so tomorrow's linux-next will probably also need
(https://patchwork.kernel.org/patch/5344551/ - pinctrl: rockchip:
Handle wakeup pins).

I've also got a local hack to the Rockchip "pm.c" to replace the usage
of "PMU_ARMINT_WAKEUP_EN" with 0x0e.  There seems to be some sort of
ARM Interrupt waking us up all the time right when we go to sleep and
the above will hack it so that only GPIOs + SDMMC Card Detect can wake
us up.  Someone should track down what's going on there, but for now
I've used the hack to prove that the basic code actually works.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Douglas Anderson Dec. 2, 2014, 1:07 a.m. UTC | #3
Hi,

On Mon, Dec 1, 2014 at 2:08 PM, Doug Anderson <dianders@chromium.org> wrote:
> Hi,
>
> On Mon, Dec 1, 2014 at 11:51 AM, Kevin Hilman <khilman@kernel.org> wrote:
>> Chris Zhong <zyw@rock-chips.com> writes:
>>
>>> devices, since we still lack power_domain driver, so the other power rail
>>> of rk3288 need keep power on.
>>> I have tested it on rk3288-evb board, atop next-20141112. goto suspend by type
>>> "echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so it can be
>>> determined in sleep mode, then press power button to wakeup it.
>>
>> I tested this on top of today's linux-next (next-20141201) and it
>> suspends, but doesn't wake up from any of the button presses.  What
>> wakeup sources are configured for the rk3288-evb-rk808?
>
> Just to close the loop (I talked with Kevin over IM about this, too):
>
> I have a huge description of how I tested this as part of my patch at
> <https://patchwork.kernel.org/patch/5414941/>.  Chris:  I think Kevin
> has asked you several times to include information like this in your
> cover letter.  Please, please, please can you try to remember to do
> this?

Talked to Chris offline.  He said that in his tests the other patches
weren't needed, so he didn't list any other patches.  Things just
worked for him.  ...so I guess he did post the instructions that
worked for him.  Sorry for the complaint.  Possibly things are
different on "next-20141112" and that's where Chris said he tested.

I know that I personally needed some of the extra patches.  I guess
the USB one wasn't truly needed (only needed for hotplug), but for me
things were unhappy without SMP.  It was hanging when trying to turn
off secondary CPUs.  I didn't dig, though.  The clocksource patches
are needed for me because I'm using an old bootloader, but I think
they're also relevant for S2R because (I'm told) we can lose the
virtual offset at suspend time in certain modes.

-Doug

.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Chris Zhong Dec. 2, 2014, 1:18 a.m. UTC | #4
On 12/02/2014 06:08 AM, Doug Anderson wrote:
> Hi,
>
> On Mon, Dec 1, 2014 at 11:51 AM, Kevin Hilman <khilman@kernel.org> wrote:
>> Chris Zhong <zyw@rock-chips.com> writes:
>>
>>> devices, since we still lack power_domain driver, so the other power rail
>>> of rk3288 need keep power on.
>>> I have tested it on rk3288-evb board, atop next-20141112. goto suspend by type
>>> "echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so it can be
>>> determined in sleep mode, then press power button to wakeup it.
>> I tested this on top of today's linux-next (next-20141201) and it
>> suspends, but doesn't wake up from any of the button presses.  What
>> wakeup sources are configured for the rk3288-evb-rk808?
> Just to close the loop (I talked with Kevin over IM about this, too):
>
> I have a huge description of how I tested this as part of my patch at
> <https://patchwork.kernel.org/patch/5414941/>.  Chris:  I think Kevin
> has asked you several times to include information like this in your
> cover letter.  Please, please, please can you try to remember to do
> this?

I wrote it to v8 cover letter, but I found that they are all landed on 
linux-next now, so I remove they from v9, v10.

Based on:
[v5,1/6] ARM: rockchip: convert to regmap and use pmu syscon if available
https://patchwork.kernel.org/patch/5086341/
[v5,2/6] ARM: rockchip: add option to access the pmu via a phandle in smp_operations
https://patchwork.kernel.org/patch/5086441/
[v5,3/6] ARM: dts: rockchip: add pmu references to cpus nodes
https://patchwork.kernel.org/patch/5086351/
[v5,4/6] ARM: rockchip: add basic smp support for rk3288
https://patchwork.kernel.org/patch/5086371/
[v5,5/6] ARM: dts: rockchip: add intmem node for rk3288 smp support
https://patchwork.kernel.org/patch/5086361/
[v5,6/6] ARM: dts: rockchip: add reset for CPU nodes
https://patchwork.kernel.org/patch/5086381/
[v3] usb: dwc2: add bus suspend/resume for dwc2
https://patchwork.kernel.org/patch/5266281/


But I missed this 3 patches, since I thought they just for fix timer 
disorder when I use coreboot.
Doug said they are related to SMP and related to suspend/resume.

1. https://patchwork.kernel.org/patch/5051881/ - clocksource:
    arch_timer: Allow the device tree to specify uninitialized timer
    registers

2. https://patchwork.kernel.org/patch/5363671/ - clocksource:
    arch_timer: Fix code to use physical timers when requested

3. https://patchwork.kernel.org/patch/5382141/ - ARM: dts: rk3288: add
    arm,cpu-registers-not-fw-configured


I'm not sure whether we need these 2 patches.

4. Revert (b77d439 ARM: dts: rockchip: temporarily disable smp on
    rk3288)

5. https://patchwork.kernel.org/patch/5325111/ - usb: dwc2: resume
    root hub when device detect with suspend state



>
> For those that don't want to click on my link, I'll include the
> relevant bits here:
>
> ---
>
> Total patches atop that version of Linux were:
>
> 1. https://patchwork.kernel.org/patch/5051881/ - clocksource:
>     arch_timer: Allow the device tree to specify uninitialized timer
>     registers
>
> 2. https://patchwork.kernel.org/patch/5363671/ - clocksource:
>     arch_timer: Fix code to use physical timers when requested
>
> 3. https://patchwork.kernel.org/patch/5382141/ - ARM: dts: rk3288: add
>     arm,cpu-registers-not-fw-configured
>
> 4. Revert (b77d439 ARM: dts: rockchip: temporarily disable smp on
>     rk3288)
>
> 5. https://patchwork.kernel.org/patch/5325111/ - usb: dwc2: resume
>     root hub when device detect with suspend state
>
> 6. https://patchwork.kernel.org/patch/5410611/ - ARM: rockchip: add
>     suspend and resume for RK3288
>
> 7. https://patchwork.kernel.org/patch/5410621/ - ARM: rockchip: Add
>     pmu-sram binding
>
> 8. https://patchwork.kernel.org/patch/5410631/ - ARM: dts: add RK3288
>     suspend support
>
> 9. https://patchwork.kernel.org/patch/5410641/ - ARM: dts: rockchip:
>     add suspend settings for rk3288-evb-rk808
>
> It looks like my pinctrl patches might be dropped due to cross
> dependency problems, so tomorrow's linux-next will probably also need
> (https://patchwork.kernel.org/patch/5344551/ - pinctrl: rockchip:
> Handle wakeup pins).
>
> I've also got a local hack to the Rockchip "pm.c" to replace the usage
> of "PMU_ARMINT_WAKEUP_EN" with 0x0e.  There seems to be some sort of
> ARM Interrupt waking us up all the time right when we go to sleep and
> the above will hack it so that only GPIOs + SDMMC Card Detect can wake
> us up.  Someone should track down what's going on there, but for now
> I've used the hack to prove that the basic code actually works.
>
>
>


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Kevin Hilman Dec. 2, 2014, 1:26 a.m. UTC | #5
Doug Anderson <dianders@chromium.org> writes:

> Hi,
>
> On Mon, Dec 1, 2014 at 2:08 PM, Doug Anderson <dianders@chromium.org> wrote:
>> Hi,
>>
>> On Mon, Dec 1, 2014 at 11:51 AM, Kevin Hilman <khilman@kernel.org> wrote:
>>> Chris Zhong <zyw@rock-chips.com> writes:
>>>
>>>> devices, since we still lack power_domain driver, so the other power rail
>>>> of rk3288 need keep power on.
>>>> I have tested it on rk3288-evb board, atop next-20141112. goto suspend by type
>>>> "echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so it can be
>>>> determined in sleep mode, then press power button to wakeup it.
>>>
>>> I tested this on top of today's linux-next (next-20141201) and it
>>> suspends, but doesn't wake up from any of the button presses.  What
>>> wakeup sources are configured for the rk3288-evb-rk808?
>>
>> Just to close the loop (I talked with Kevin over IM about this, too):
>>
>> I have a huge description of how I tested this as part of my patch at
>> <https://patchwork.kernel.org/patch/5414941/>.  Chris:  I think Kevin
>> has asked you several times to include information like this in your
>> cover letter.  Please, please, please can you try to remember to do
>> this?
>
> Talked to Chris offline.  He said that in his tests the other patches
> weren't needed, so he didn't list any other patches.  Things just
> worked for him.  ...so I guess he did post the instructions that
> worked for him.  Sorry for the complaint.  Possibly things are
> different on "next-20141112" and that's where Chris said he tested.

This series doesn't apply cleanly to next-20141112.  Manually applying
(with fuzz), it boots but I have the same results: it suspends, but none
of the buttons wake it up.

Kevin
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Chris Zhong Dec. 3, 2014, 1:55 p.m. UTC | #6
On 12/02/2014 09:26 AM, Kevin Hilman wrote:
> Doug Anderson <dianders@chromium.org> writes:
>
>> Hi,
>>
>> On Mon, Dec 1, 2014 at 2:08 PM, Doug Anderson <dianders@chromium.org> wrote:
>>> Hi,
>>>
>>> On Mon, Dec 1, 2014 at 11:51 AM, Kevin Hilman <khilman@kernel.org> wrote:
>>>> Chris Zhong <zyw@rock-chips.com> writes:
>>>>
>>>>> devices, since we still lack power_domain driver, so the other power rail
>>>>> of rk3288 need keep power on.
>>>>> I have tested it on rk3288-evb board, atop next-20141112. goto suspend by type
>>>>> "echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so it can be
>>>>> determined in sleep mode, then press power button to wakeup it.
>>>> I tested this on top of today's linux-next (next-20141201) and it
>>>> suspends, but doesn't wake up from any of the button presses.  What
>>>> wakeup sources are configured for the rk3288-evb-rk808?
>>> Just to close the loop (I talked with Kevin over IM about this, too):
>>>
>>> I have a huge description of how I tested this as part of my patch at
>>> <https://patchwork.kernel.org/patch/5414941/>.  Chris:  I think Kevin
>>> has asked you several times to include information like this in your
>>> cover letter.  Please, please, please can you try to remember to do
>>> this?
>> Talked to Chris offline.  He said that in his tests the other patches
>> weren't needed, so he didn't list any other patches.  Things just
>> worked for him.  ...so I guess he did post the instructions that
>> worked for him.  Sorry for the complaint.  Possibly things are
>> different on "next-20141112" and that's where Chris said he tested.
> This series doesn't apply cleanly to next-20141112.  Manually applying
> (with fuzz), it boots but I have the same results: it suspends, but none
> of the buttons wake it up.
>
> Kevin
>
Hi, Kevin

I have test these patches on evb board base on next-20141128 with a 
defconfig[0], and with u-boot[1].
As Doug said, we need below 3 patches for resume.

1.https://patchwork.kernel.org/patch/5051881/  - clocksource:
    arch_timer: Allow the device tree to specify uninitialized timer
    registers

2.https://patchwork.kernel.org/patch/5363671/  - clocksource:
    arch_timer: Fix code to use physical timers when requested

3.https://patchwork.kernel.org/patch/5382141/  - ARM: dts: rk3288: add
    arm,cpu-registers-not-fw-configured


And it will auto wakeup, as Heiko said in v8.  But I have never notice 
before, since the u-boot never enable edp,
and I use the coreboot with edp display.
Actually it is a bug in rk3288, the rk3288 have not 27Mhz clock source, 
but the edp initially set to this
non-existent clock. At this time, edp is working on a unknown state, and 
it always bring a interrupt, this
interrupt avoid system enter suspend. So if we want to enter suspend 
normally, the edp_24m_sel(bit 15) of
CRU_CLKSEL28_CON(0xff7600d0) must be set to 1.

[0] 
https://github.com/mmind/linux-rockchip/blob/devel/workbench/arch/arm/configs/rk3288_defconfig 

[1] https://githubremotes/origin/u-boot-rk3288

here is my local  work around:

diff --git a/arch//cpu/armv7/rk32xx/clock-rk3288.c 
b/arch/arm/cpu/armv7/rk32xx/clock-rk3288.c
index cfd0acd..3df0900 100755
--- a/arch/arm/cpu/armv7/rk32xx/clock-rk3288.c
+++ b/arch/arm/cpu/armv7/rk32xx/clock-rk3288.c
@@ -1233,7 +1233,6 @@ int rkclk_lcdc_clk_set(uint32 lcdc_id, uint32 dclk_hz)
         }
  }

-
  /*
   * rkplat set nandc clock div
   * nandc_id:   nandc id
@@ -1270,6 +1269,11 @@ int rkclk_set_nandc_div(uint32 nandc_id, uint32 
pllsrc, uint32 freq)
         return 0;
  }

+void rkclk_init_edp_source(void)
+{
+       cru_writel(1<<15 | 1<<31, CRU_CLKSELS_CON(28));
+}
+
  /*
   * rkplat set sd clock src
   * 0: codec pll; 1: general pll; 2: 24M
diff --git a/board/rockchip/rk32xx/rk32xx.c b/board/rockchip/rk32xx/rk32xx.c
index bfdcf0e..3e19f5d 100755
--- a/board/rockchip/rk32xx/rk32xx.c
+++ b/board/rockchip/rk32xx/rk32xx.c
@@ -114,7 +114,7 @@ void rk_backlight_ctrl(int brightness)

  void rk_fb_init(unsigned int onoff)
  {
-
+       rkclk_init_edp_source();
  #ifdef CONFIG_OF_LIBFDT
         if (lcd_node == 0) rk_lcd_parse_dt(gd->fdt_blob);











--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Kevin Hilman Dec. 3, 2014, 7:23 p.m. UTC | #7
Chris Zhong <zyw@rock-chips.com> writes:

[...]

> I have test these patches on evb board base on next-20141128 with a
> defconfig[0], and with u-boot[1].
> As Doug said, we need below 3 patches for resume.
>
> 1.https://patchwork.kernel.org/patch/5051881/  - clocksource:
>    arch_timer: Allow the device tree to specify uninitialized timer
>    registers
>
> 2.https://patchwork.kernel.org/patch/5363671/  - clocksource:
>    arch_timer: Fix code to use physical timers when requested
>
> 3.https://patchwork.kernel.org/patch/5382141/  - ARM: dts: rk3288: add
>    arm,cpu-registers-not-fw-configured
>
>
> And it will auto wakeup, as Heiko said in v8.  

OK, with your series plus those 3 patches on top of next-20141128, I'm
now seeing it auto-wakeup, either with multi_v7_defconfig or Heiko's
rk3288_defconfig.

> But I have never notice before, since the u-boot never enable edp, and
> I use the coreboot with edp display.  Actually it is a bug in rk3288,
> the rk3288 have not 27Mhz clock source, but the edp initially set to
> this non-existent clock. At this time, edp is working on a unknown
> state, and it always bring a interrupt, this interrupt avoid system
> enter suspend. 

I see, good find!

> So if we want to enter suspend normally, the
> edp_24m_sel(bit 15) of CRU_CLKSEL28_CON(0xff7600d0) must be set to 1.

I didn't try your u-boot fix, but it sounds like there should be a kernel
fix for this.  Why doesn't the disabling of unused clocks put the EDP
into a safe/disabled state?

Kevin

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Heiko Stübner Dec. 7, 2014, 11:46 p.m. UTC | #8
Hi,

Am Mittwoch, 3. Dezember 2014, 11:23:26 schrieb Kevin Hilman:
> Chris Zhong <zyw@rock-chips.com> writes:
> 
> [...]
> 
> > I have test these patches on evb board base on next-20141128 with a
> > defconfig[0], and with u-boot[1].
> > As Doug said, we need below 3 patches for resume.
> > 
> > 1.https://patchwork.kernel.org/patch/5051881/  - clocksource:
> >    arch_timer: Allow the device tree to specify uninitialized timer
> >    registers
> > 
> > 2.https://patchwork.kernel.org/patch/5363671/  - clocksource:
> >    arch_timer: Fix code to use physical timers when requested
> > 
> > 3.https://patchwork.kernel.org/patch/5382141/  - ARM: dts: rk3288: add
> > 
> >    arm,cpu-registers-not-fw-configured
> > 
> > And it will auto wakeup, as Heiko said in v8.
> 
> OK, with your series plus those 3 patches on top of next-20141128, I'm
> now seeing it auto-wakeup, either with multi_v7_defconfig or Heiko's
> rk3288_defconfig.
> 
> > But I have never notice before, since the u-boot never enable edp, and
> > I use the coreboot with edp display.  Actually it is a bug in rk3288,
> > the rk3288 have not 27Mhz clock source, but the edp initially set to
> > this non-existent clock. At this time, edp is working on a unknown
> > state, and it always bring a interrupt, this interrupt avoid system
> > enter suspend.
> 
> I see, good find!

I think the "problem" might be a different one.

With Chris' short patch [0] enabling a bit of gic debug output on resume I was
able to track down where my wakeup comes from and it seems to be the
sdmmc_detect_n interrupt. When I have a sd card inserted it wakes up again
directly and when I remove it it stays asleep till I wake it with the power-
button.

Relevant output would be:
	gic_cpu_restore add = f0000000, 0, 0, 0, 0, 8

I have verified that at least for me the sclk_edp_24m has no influence on this,
by using an uboot that does not reparent the edp clock (it stays an orphan)
and an uboot that reparents it to xin24m.

In both cases the system wakes up directly when I have a sd card in the slot
and stays asleep (till pressing the power-key) when I remove the card.

So the auto-wakeup is not the fault of the new suspend-code, but we should
find a way to tame sdmmc_detect_n :-)


Now I only need to check why I get a
	rk3288_suspend_finish: Suspend finish failed
on every resume [1], but that is something for tomorrow.


Heiko


[0]
diff --git a/drivers/irqchip/irq-gic.c b/drivers/irqchip/irq-gic.c
index d617ee5..37d5ce0 100644
--- a/drivers/irqchip/irq-gic.c
+++ b/drivers/irqchip/irq-gic.c
@@ -538,6 +538,7 @@ static void gic_cpu_restore(unsigned int gic_nr)
 	u32 *ptr;
 	void __iomem *dist_base;
 	void __iomem *cpu_base;
+	u32 reg[5];
 
 	if (gic_nr >= MAX_GIC_NR)
 		BUG();
@@ -562,6 +563,10 @@ static void gic_cpu_restore(unsigned int gic_nr)
 
 	writel_relaxed(GICC_INT_PRI_THRESHOLD, cpu_base + GIC_CPU_PRIMASK);
 	gic_cpu_if_up();
+
+	for (i = 0; i < 5; i++)
+		reg[i] = readl_relaxed(dist_base + 0x200 + i * 4);
+	printk("%s add = %x, %x, %x, %x, %x, %x\n",__func__, dist_base, reg[0], reg[1], reg[2], reg[3], reg[4]);
 }
 
 static int gic_notifier(struct notifier_block *self, unsigned long cmd,	void *v)

[1]
Disabling non-boot CPUs ...
CPU1: shutdown
CPU2: shutdown
CPU3: shutdown
gic_cpu_restore add = f0000000, 0, 0, 0, 20000, 0
Enabling non-boot CPUs ...
CPU1 is up
CPU2 is up
CPU3 is up
PM: noirq resume of devices complete after 0.892 msecs
PM: early resume of devices complete after 0.991 msecs
rk3288_suspend_finish: Suspend finish failed
PM: resume of devices complete after 2.496 msecs

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Heiko Stübner Jan. 2, 2015, 8:57 p.m. UTC | #9
Am Montag, 1. Dezember 2014, 16:52:16 schrieb Chris Zhong:
> devices, since we still lack power_domain driver, so the other power rail
> of rk3288 need keep power on.
> I have tested it on rk3288-evb board, atop next-20141112. goto suspend by
> type "echo mem > /sys/power/state", vdd_cpu is about 0mv by measuring, so
> it can be determined in sleep mode, then press power button to wakeup it.

I've tested this series again on 3.19-rc1. It still sucessfully suspends and 
wakes a rk3288-evb-rk808 system on power-key presses.
[except the sdmmc detect interrupt going off when a card is inserted on 
suspend]

So I've applied this series to appropriate soc and dts branches for 3.20


Heiko
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/