[RFC,0/2] Fix Arndale Octa/Peach Pi boot on Audio subsystem clocks

Commit Message

Krzysztof Kozlowski Nov. 25, 2014, 2:22 p.m. UTC

On wto, 2014-11-25 at 14:36 +0100, Javier Martinez Canillas wrote:
> Hello Krzysztof,
> 
> On 11/25/2014 10:22 AM, Krzysztof Kozlowski wrote:
> > 
> > Hmmm, that's strange. I've looked again at your dmesg and it is little
> > different than mine (when system fails).
> > 1. Does your PMIC come up? In your dmesg:
> > [    2.687297] max77802-pmic max77802-pmic: regulator init failed for 0
> > [    2.693539] platform max77802-pmic: Driver max77802-pmic requests probe deferral
> >
> 
> The Peach Pit/Pi DTS define the parent input supplies for the max77802 PMIC
> regulators and these are regulators from the tps65090 PMU but the max77802
> is probed before. When the system does not fail, the probe is deferred and
> succeed later but I don't know when the system fails (more on that below).

OK.

>  
> > 2. In my case booting fails just after disabling unused clocks which is
> > quite early, my dmesg:
> > [   11.700074] input: gpio_keys as /devices/platform/gpio_keys/input/input0
> > [   11.704282] s3c-rtc 101e0000.rtc: setting system clock to 2029-01-11 19:19:17 UTC (1862853557)
> > [   11.770602] mmcblk0: mmc0:0001 M8G1WA 7.28 GiB
> > [   11.776201] mmcblk0boot0: mmc0:0001 M8G1WA partition 1 2.00 MiB
> > [   11.780968] mmcblk0boot1: mmc0:0001 M8G1WA partition 2 2.00 MiB
> > [   11.786823] mmcblk0rpmb: mmc0:0001 M8G1WA partition 3 128 KiB
> > [   11.793815]  mmcblk0: p1 p2 p3 p4
> > [   11.824165] PVDD_G3D_1V0: disabling
> > [   11.830833] PVDD_G3DS_1V0: disabling
> > [   11.837771] PVDD_HSIC_1V8: disabling
> > [   11.841172] PVDD_ABB_1V8: disabling
> > [   11.844866] PVDD_ANAIP_1V8: disabling
> > [   11.847157] usb 5-1.4: new high-speed USB device number 3 using exynos-ehci
> > [   11.848778] PVDD
> > 
> > In your case you have a gap here:
> > [    3.624142] platform 12d10000.adc: Driver exynos-adc requests probe deferral
> > [   24.188722] random: nonblocking pool is initialized
> >
> > Could you try my patches once again but this time with enabled:
> > DEBUG_EXYNOS_UART
> > EARLY_PRINTK
> > DEBUG_S3C_UART3
> > DEBUG_LL
> >
> 
> Yes, there is a gap and I don't know why. Even after enabling all those
> config options, nothing is printed on the serial console after the exynos-adc
> probe deferral. Only that message in drivers/char/random.c is printed but
> nothing else is printed after all...

You may try attached patch (very early printk) if you suspect that some
logs are missing.

> 
> > and with following change:
> > 
> > diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> > index 4896ae9e23da..316856e55784 100644
> > --- a/drivers/clk/clk.c
> > +++ b/drivers/clk/clk.c
> > @@ -478,7 +478,9 @@ static void clk_disable_unused_subtree(struct clk *clk)
> >          * sequence.  call .disable_unused if available, otherwise fall
> >          * back to .disable
> >          */
> > +       pr_err("clk: checking %s\n", clk->name);
> >         if (__clk_is_enabled(clk)) {
> > +               pr_err("clk: enabled %s, disabling\n", clk->name);
> >                 if (clk->ops->disable_unused)
> >                         clk->ops->disable_unused(clk->hw);
> >                 else if (clk->ops->disable)
> > 
> > Without my patch the boot hangs:
> > [   12.230666] clk: checking dout_unipro
> > [   12.234304] clk: checking mout_unipro
> > [   12.237913] clk: checking sclk_usbd300
> > [   12.241638] clk: enabled sclk_usbd300, disabling
> > [   12.246264] clk: checking dout_usbd300
> > [   12.249992] clk: checking sclk_usbd301
> > [   12.253683] clk: enabled sclk_usbd301, disabling
> > [   12.258309] clk: checking dout_usbd301
> > [   12.262007] clk: checking sclk_maudio0
> > [   12.265728] clk: enabled sclk_maudio0, disabling
> > [   12.270355] clk: checking pcm_bus
> > 
> > because audss seems to be gated so kernel cannot access status of
> > pcm_bus clock.
> >
> 
> I added those printk's but clk_disable_unused() is executed after the
> exynos-adc probe function so the logs are not shown on the serial console.

So this would mean that something fails before clk_disable_unused().

> As mentioned previously in the thread, Tushar's fix [0] works for me and with
> that patch I see the clocks disabled by clk_disable_unused_subtree() and the
> kernel is able to access pcm_bus and other audio clocks:
> 
> [    5.007368] clk: checking mout_spdif
> [    5.010908] clk: checking dout_audio0
> [    5.014566] clk: checking mout_audio0
> [    5.018208] clk: checking sclk_maudio0
> [    5.021926] clk: enabled sclk_maudio0, disabling
> [    5.026541] clk: checking pcm_bus
> [    5.029820] clk: enabled pcm_bus, disabling
> [    5.034001] clk: checking sclk_pcm
> [    5.037367] clk: enabled sclk_pcm, disabling
> 
> Any ideas?

Yes, I got. On Peach board the i2s0 is enabled in DTS. Probing it could
fail because it relies on enabling audss clocks (which cannot be
accessed).

I reproduced another hang on Arndale Octa after enabling i2s0 in DTS.
Maybe that is the cause also on Peach.

Best regards,
Krzysztof

Comments

Javier Martinez Canillas Nov. 25, 2014, 2:52 p.m. UTC | #1

Hello Krzysztof,

On 11/25/2014 03:22 PM, Krzysztof Kozlowski wrote:
>> Any ideas?
> 
> Yes, I got. On Peach board the i2s0 is enabled in DTS. Probing it could
> fail because it relies on enabling audss clocks (which cannot be
> accessed).
> 
> I reproduced another hang on Arndale Octa after enabling i2s0 in DTS.
> Maybe that is the cause also on Peach.
> 

You are correct, if I disable i2s0 then I see the logs from
clk_disable_unused_subtree() and boot hangs when accessing the adma clock:

clk: checking dout_fimd1
clk: checking mout_fimd1_final
clk: checking mout_fimd1
clk: checking mout_sclk_rpll
clk: checking fout_rpll
clk: checking adma

With i2s0 disabled, your series are indeed enough to make it boot again but
as you said enabling i2s0 in the DTS makes it hang even with your patches.

Thanks a lot for digging into this!

Best regards,
Javier

Krzysztof Kozlowski Nov. 25, 2014, 2:54 p.m. UTC | #2

On wto, 2014-11-25 at 15:52 +0100, Javier Martinez Canillas wrote:
> Hello Krzysztof,
> 
> On 11/25/2014 03:22 PM, Krzysztof Kozlowski wrote:
> >> Any ideas?
> > 
> > Yes, I got. On Peach board the i2s0 is enabled in DTS. Probing it could
> > fail because it relies on enabling audss clocks (which cannot be
> > accessed).
> > 
> > I reproduced another hang on Arndale Octa after enabling i2s0 in DTS.
> > Maybe that is the cause also on Peach.
> > 
> 
> You are correct, if I disable i2s0 then I see the logs from
> clk_disable_unused_subtree() and boot hangs when accessing the adma clock:
> 
> clk: checking dout_fimd1
> clk: checking mout_fimd1_final
> clk: checking mout_fimd1
> clk: checking mout_sclk_rpll
> clk: checking fout_rpll
> clk: checking adma
> 
> With i2s0 disabled, your series are indeed enough to make it boot again but
> as you said enabling i2s0 in the DTS makes it hang even with your patches.
> 
> Thanks a lot for digging into this!

My patch only fixed the gate clocks but it didn't touched div and mux.
I'll prepare a v2 of it (I got some feedback) which I hope will fix both
cases: i2s and disabling unused clocks.

Best regards,
Krzysztof

Javier Martinez Canillas Nov. 25, 2014, 3:13 p.m. UTC | #3

Hello Krzysztof,

On 11/25/2014 03:54 PM, Krzysztof Kozlowski wrote:
> 
> My patch only fixed the gate clocks but it didn't touched div and mux.
> I'll prepare a v2 of it (I got some feedback) which I hope will fix both
> cases: i2s and disabling unused clocks.
>

Perfect, I will gladly test on Peach boards once you post it.

Thanks a lot for your help.

Best regards,
Javier

Patch

From def8df7c5d8098ec1656c075701e9ba17f0617c9 Mon Sep 17 00:00:00 2001
From: Krzysztof Kozlowski <[e-mail]>
Date: Thu, 6 Nov 2014 12:58:07 +0100
Subject: [PATCH] Very early printk, for next-20141106 (3.19?)

Signed-off-by: Krzysztof Kozlowski <[e-mail]>
---
 kernel/printk/printk.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
index 1b7092dbb590..269318013504 100644
--- a/kernel/printk/printk.c
+++ b/kernel/printk/printk.c
@@ -1708,6 +1708,13 @@  asmlinkage int vprintk_emit(int facility, int level,
 		}
 	}
 
+	if (1) {
+		extern void printascii(const char *);
+		printascii(text);
+		printed_len = 0;
+		goto early_printk_out;
+	}
+
 	if (level == LOGLEVEL_DEFAULT)
 		level = default_message_loglevel;
 
@@ -1754,6 +1761,7 @@  asmlinkage int vprintk_emit(int facility, int level,
 						 dict, dictlen, text, text_len);
 	}
 
+early_printk_out:
 	logbuf_cpu = UINT_MAX;
 	raw_spin_unlock(&logbuf_lock);
 	lockdep_on();
-- 
1.9.1