[PATCH 4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[PATCH 4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[Ivan Mikhaylov]

Prevent a kernel panic caused by unintentionally clearing TCR
watchdog bits.  At this point in the kernel boot, the watchdog has
already been enabled by u-boot.  The original code's attempt to
write to the TCR register results in an inadvertent clearing of the
watchdog configuration bits, causing the 476 to reset.
Panic happens in case of error as silently reboot without any outputs
on serial.

Signed-off-by: Ivan Mikhaylov <ivan@de.ibm.com>
---
 arch/powerpc/kernel/time.c |   16 ++++++++++++++++
 1 files changed, 16 insertions(+), 0 deletions(-)

diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index 2b33cfa..f75e512 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -738,12 +738,28 @@ static int __init get_freq(char *name, int cells, unsigned long *val)
 
 static void start_cpu_decrementer(void)
 {
+	unsigned int tcr;
 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
 	/* Clear any pending timer interrupts */
 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
 
+#ifdef CONFIG_FSP2
+	/*
+	 * Prevent a kernel panic caused by unintentionally clearing TCR
+	 * watchdog bits.  At this point in the kernel boot, the watchdog has
+	 * already been enabled by u-boot.  The original code's attempt to
+	 * write to the TCR register results in an inadvertent clearing of the
+	 * watchdog configuration bits, causing the 440 to reset.
+	 */
+	tcr = mfspr(SPRN_TCR);
+	tcr &= TCR_WP_MASK; /* clear all bits except for TCR[WP] */
+	tcr |= TCR_DIE; /* enable decrementer */
+	mtspr(SPRN_TCR, tcr);
+#else
 	/* Enable decrementer interrupt */
 	mtspr(SPRN_TCR, TCR_DIE);
+#endif
+
 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
 }
 
-- 
1.7.1

Re: [PATCH 4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[Michael Ellerman]

Hi Ivan,

Ivan Mikhaylov <ivan@de.ibm.com> writes:
> Prevent a kernel panic caused by unintentionally clearing TCR
> watchdog bits.  At this point in the kernel boot, the watchdog has
> already been enabled by u-boot.  The original code's attempt to
> write to the TCR register results in an inadvertent clearing of the
> watchdog configuration bits, causing the 476 to reset.
> Panic happens in case of error as silently reboot without any outputs
> on serial.

That sounds reasonable.

> diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
> index 2b33cfa..f75e512 100644
> --- a/arch/powerpc/kernel/time.c
> +++ b/arch/powerpc/kernel/time.c
> @@ -738,12 +738,28 @@ static int __init get_freq(char *name, int cells, u=
nsigned long *val)
>=20=20
>  static void start_cpu_decrementer(void)
>  {
> +	unsigned int tcr;
>  #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
>  	/* Clear any pending timer interrupts */
>  	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
>=20=20
> +#ifdef CONFIG_FSP2
> +	/*
> +	 * Prevent a kernel panic caused by unintentionally clearing TCR
> +	 * watchdog bits.  At this point in the kernel boot, the watchdog has
> +	 * already been enabled by u-boot.  The original code's attempt to

Don't refer to "the original code", as it doesn't exist anymore now that
we've patched it. Just say something like ".. so we must not clear the
watchdog configuration bits".

> +	 * write to the TCR register results in an inadvertent clearing of the
> +	 * watchdog configuration bits, causing the 440 to reset.
> +	 */
> +	tcr =3D mfspr(SPRN_TCR);
> +	tcr &=3D TCR_WP_MASK; /* clear all bits except for TCR[WP] */
> +	tcr |=3D TCR_DIE; /* enable decrementer */
> +	mtspr(SPRN_TCR, tcr);
> +#else
>  	/* Enable decrementer interrupt */
>  	mtspr(SPRN_TCR, TCR_DIE);
> +#endif
> +
>  #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */

That breaks the build for other platforms:

  arch/powerpc/kernel/time.c: In function =E2=80=98start_cpu_decrementer=E2=
=80=99:
  arch/powerpc/kernel/time.c:741:15: error: unused variable =E2=80=98tcr=E2=
=80=99 [-Werror=3Dunused-variable]

You can just do something like:

  #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
  	unsigned int tcr;

  #ifdef CONFIG_FSP2
  	tcr =3D mfspr(SPRN_TCR);
  	tcr &=3D TCR_WP_MASK; /* clear all bits except for TCR[WP] */
  #else
  	tcr =3D 0;
  #endif
  	tcr |=3D TCR_DIE; /* enable decrementer */
  	mtspr(SPRN_TCR, TCR_DIE);
  #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */


Or you could possibly just always leave TCR[WP], is there any case where
it would be correct to clear that?

cheers

Re: [PATCH 4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[Ivan Mikhaylov]

Hi Michael,

>> diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
>> index 2b33cfa..f75e512 100644
>> --- a/arch/powerpc/kernel/time.c
>> +++ b/arch/powerpc/kernel/time.c
>> @@ -738,12 +738,28 @@ static int =5F=5Finit get=5Ffreq(char *name, int c=
ells, unsigned long *val)
>> =20
>>  static void start=5Fcpu=5Fdecrementer(void)
>>  {
>> +	unsigned int tcr;
>>  #if defined(CONFIG=5FBOOKE) || defined(CONFIG=5F40x)
>>  	/* Clear any pending timer interrupts */
>>  	mtspr(SPRN=5FTSR, TSR=5FENW | TSR=5FWIS | TSR=5FDIS | TSR=5FFIS);
>> =20
>> +#ifdef CONFIG=5FFSP2
>> +	/*
>> +	 * Prevent a kernel panic caused by unintentionally clearing TCR
>> +	 * watchdog bits.  At this point in the kernel boot, the watchdog has
>> +	 * already been enabled by u-boot.  The original code's attempt to
>
> Don't refer to "the original code", as it doesn't exist anymore now that
> we've patched it. Just say something like ".. so we must not clear the
> watchdog configuration bits".
Ok, got it.

>That breaks the build for other platforms:
>
> arch/powerpc/kernel/time.c: In function =E2=80=98start=5Fcpu=5Fdecremente=
r=E2=80=99:
> arch/powerpc/kernel/time.c:741:15: error: unused variable =E2=80=98tcr=E2=
=80=99 [-Werror=3Dunused-variable]
>
Oops, didn't notice, my fault.=20

>Or you could possibly just always leave TCR[WP], is there any case where
>it would be correct to clear that?
>
>cheers

>From my point of view it's possible. I've checked docu and on idea
it should be possible cause WP is only affecting watchdog ping time.
Which in case of '00' is very small, around ~5 ms.=20
Ben also in next message said about get rid of ifdef for FSP2.

And now patch looks like this, What do you think Michael, Ben?

 arch/powerpc/kernel/time.c |   15 +++++++++++++--
 1 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index bc2e08d..2411c49 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -718,11 +718,22 @@ static int =5F=5Finit get=5Ffreq(char *name, int cell=
s, unsigned long *val)
 static void start=5Fcpu=5Fdecrementer(void)
 {
 #if defined(CONFIG=5FBOOKE) || defined(CONFIG=5F40x)
+	unsigned int tcr;
 	/* Clear any pending timer interrupts */
 	mtspr(SPRN=5FTSR, TSR=5FENW | TSR=5FWIS | TSR=5FDIS | TSR=5FFIS);
=20
-	/* Enable decrementer interrupt */
-	mtspr(SPRN=5FTCR, TCR=5FDIE);
+	tcr =3D mfspr(SPRN=5FTCR);
+	/*
+	 * At this point in the kernel boot, the watchdog has already
+	 * been enabled by u-boot. If we set it this to '00' it may
+	 * trigger watchdog earlier than needed which will cause
+	 * inattentional kernel panic. In this case we leaving TCR[WP]
+	 * bit setting from uboot/bootstrap.
+	 */
+	tcr &=3D TCR=5FWP=5FMASK; /* clear all bits except for TCR[WP] */
+	tcr |=3D TCR=5FDIE; /* enable decrementer */
+	mtspr(SPRN=5FTCR, tcr);
+
 #endif /* defined(CONFIG=5FBOOKE) || defined(CONFIG=5F40x) */
 }
=20
--=20
1.7.1

Re: [PATCH 4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[Michael Ellerman]

Ivan Mikhaylov <ivan@de.ibm.com> writes:
>
> From my point of view it's possible. I've checked docu and on idea
> it should be possible cause WP is only affecting watchdog ping time.

The question is, is there any chance that leaving those bits set on
another platform will cause a problem?

ie. on existing machines we always clear those bits, is it OK that we
will now start leaving those bits with whatever value they had.

cheers

Re: [PATCH 4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[Michael Ellerman]

Michael Ellerman <mpe@ellerman.id.au> writes:

> Ivan Mikhaylov <ivan@de.ibm.com> writes:
>>
>> From my point of view it's possible. I've checked docu and on idea
>> it should be possible cause WP is only affecting watchdog ping time.
>
> The question is, is there any chance that leaving those bits set on
> another platform will cause a problem?
>
> ie. on existing machines we always clear those bits, is it OK that we
> will now start leaving those bits with whatever value they had.

I came up with the patch below (more or less your v2), if it breaks
something we can always fix it.

cheers

commit c80409358a9c91e1f6b18353dad939b851bb3522
Author: Ivan Mikhaylov <ivan@de.ibm.com>
Date:   Fri May 19 18:47:05 2017 +0300

    powerpc/[booke|4xx]: Don't clobber TCR[WP] when setting TCR[DIE]
    
    Prevent a kernel panic caused by unintentionally clearing TCR watchdog
    bits. At this point in the kernel boot, the watchdog may have already
    been enabled by u-boot. The original code's attempt to write to the TCR
    register results in an inadvertent clearing of the watchdog
    configuration bits, causing the 476 to reset.
    
    Signed-off-by: Ivan Mikhaylov <ivan@de.ibm.com>
    Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index 2b33cfaac7b8..60714b8c9a2f 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -739,12 +739,20 @@ static int __init get_freq(char *name, int cells, unsigned long *val)
 static void start_cpu_decrementer(void)
 {
 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
+	unsigned int tcr;
+
 	/* Clear any pending timer interrupts */
 	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
 
-	/* Enable decrementer interrupt */
-	mtspr(SPRN_TCR, TCR_DIE);
-#endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
+	tcr = mfspr(SPRN_TCR);
+	/*
+	 * The watchdog may have already been enabled by u-boot. So leave
+	 * TRC[WP] (Watchdog Period) alone.
+	 */
+	tcr &= TCR_WP_MASK;	/* Clear all bits except for TCR[WP] */
+	tcr |= TCR_DIE;		/* Enable decrementer */
+	mtspr(SPRN_TCR, tcr);
+#endif
 }
 
 void __init generic_calibrate_decr(void)

Re: [4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[Michael Ellerman]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 3138 bytes --]

On Fri, 2017-05-19 at 15:47:05 UTC, Ivan Mikhaylov wrote:
> Hi Michael,
> 
> >> diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
> >> index 2b33cfa..f75e512 100644
> >> --- a/arch/powerpc/kernel/time.c
> >> +++ b/arch/powerpc/kernel/time.c
> >> @@ -738,12 +738,28 @@ static int __init get_freq(char *name, int cells, unsigned long *val)
> >>  
> >>  static void start_cpu_decrementer(void)
> >>  {
> >> +	unsigned int tcr;
> >>  #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
> >>  	/* Clear any pending timer interrupts */
> >>  	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
> >>  
> >> +#ifdef CONFIG_FSP2
> >> +	/*
> >> +	 * Prevent a kernel panic caused by unintentionally clearing TCR
> >> +	 * watchdog bits.  At this point in the kernel boot, the watchdog has
> >> +	 * already been enabled by u-boot.  The original code's attempt to
> >
> > Don't refer to "the original code", as it doesn't exist anymore now that
> > we've patched it. Just say something like ".. so we must not clear the
> > watchdog configuration bits".
> Ok, got it.
> 
> >That breaks the build for other platforms:
> >
> > arch/powerpc/kernel/time.c: In function ‘start_cpu_decrementer’:
> > arch/powerpc/kernel/time.c:741:15: error: unused variable ‘tcr’ [-Werror=unused-variable]
> >
> Oops, didn't notice, my fault. 
> 
> >Or you could possibly just always leave TCR[WP], is there any case where
> >it would be correct to clear that?
> >
> >cheers
> 
> >From my point of view it's possible. I've checked docu and on idea
> it should be possible cause WP is only affecting watchdog ping time.
> Which in case of '00' is very small, around ~5 ms. 
> Ben also in next message said about get rid of ifdef for FSP2.
> 
> And now patch looks like this, What do you think Michael, Ben?
> 
>  arch/powerpc/kernel/time.c |   15 +++++++++++++--
>  1 files changed, 13 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
> index bc2e08d..2411c49 100644
> --- a/arch/powerpc/kernel/time.c
> +++ b/arch/powerpc/kernel/time.c
> @@ -718,11 +718,22 @@ static int __init get_freq(char *name, int cells, unsigned long *val)
>  static void start_cpu_decrementer(void)
>  {
>  #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
> +	unsigned int tcr;
>  	/* Clear any pending timer interrupts */
>  	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
>  
> -	/* Enable decrementer interrupt */
> -	mtspr(SPRN_TCR, TCR_DIE);
> +	tcr = mfspr(SPRN_TCR);
> +	/*
> +	 * At this point in the kernel boot, the watchdog has already
> +	 * been enabled by u-boot. If we set it this to '00' it may
> +	 * trigger watchdog earlier than needed which will cause
> +	 * inattentional kernel panic. In this case we leaving TCR[WP]
> +	 * bit setting from uboot/bootstrap.
> +	 */
> +	tcr &= TCR_WP_MASK; /* clear all bits except for TCR[WP] */
> +	tcr |= TCR_DIE; /* enable decrementer */
> +	mtspr(SPRN_TCR, tcr);
> +
>  #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */
>  }
>  

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/6e2f03e292ef46eed2b31b0a344a91

cheers

Re: [PATCH 4/4] arch/powerpc/44x/fsp2: wdt tcr update instead of whole rewrite

[Benjamin Herrenschmidt]

On Mon, 2017-05-15 at 16:07 +0300, Ivan Mikhaylov wrote:
> +#ifdef CONFIG_FSP2
> +       /*
> +        * Prevent a kernel panic caused by unintentionally clearing TCR
> +        * watchdog bits.  At this point in the kernel boot, the watchdog has
> +        * already been enabled by u-boot.  The original code's attempt to
> +        * write to the TCR register results in an inadvertent clearing of the
> +        * watchdog configuration bits, causing the 440 to reset.
> +        */
> +       tcr = mfspr(SPRN_TCR);
> +       tcr &= TCR_WP_MASK; /* clear all bits except for TCR[WP] */
> +       tcr |= TCR_DIE; /* enable decrementer */
> +       mtspr(SPRN_TCR, tcr);
> +#else

This should be a runtime test, not a compile time option.

Cheers,
Ben.