mail archive of the barebox mailing list
 help / color / mirror / Atom feed
* [PATCH] Arm: i.MX5: Do not disable L2 cache
@ 2020-06-18 12:12 Sascha Hauer
  2020-06-18 12:21 ` Lucas Stach
  0 siblings, 1 reply; 3+ messages in thread
From: Sascha Hauer @ 2020-06-18 12:12 UTC (permalink / raw)
  To: Barebox List

Disabling the L2 cache is not working in imx5_init_lowlevel() because
the necessary cache maintenance operations are missing. This often
results in cache corruption in a chainloaded barebox.
Disabling the cache is unnecessary: when we come from the ROM the L2
cache is disabled anyway, so disabling it is a no-op. When we get here
in a chainloaded barebox the L2 cache is already enabled and correctly
configured. So instead of initializing it again we can take an enabled
L2 cache as a sign to skip initialization and just return from the
function.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
---
 arch/arm/mach-imx/imx5.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/arch/arm/mach-imx/imx5.c b/arch/arm/mach-imx/imx5.c
index 96288f99e0..dd6c079fe3 100644
--- a/arch/arm/mach-imx/imx5.c
+++ b/arch/arm/mach-imx/imx5.c
@@ -37,10 +37,13 @@ void imx5_init_lowlevel(void)
 {
 	u32 r;
 
-	/* ARM errata ID #468414 */
 	__asm__ __volatile__("mrc 15, 0, %0, c1, c0, 1":"=r"(r));
+
+	if (r & (1 << 1))
+		return;
+
+	/* ARM errata ID #468414 */
 	r |= (1 << 5);    /* enable L1NEON bit */
-	r &= ~(1 << 1);   /* explicitly disable L2 cache */
 	__asm__ __volatile__("mcr 15, 0, %0, c1, c0, 1" : : "r"(r));
 
         /* reconfigure L2 cache aux control reg */
-- 
2.27.0


_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH] Arm: i.MX5: Do not disable L2 cache
  2020-06-18 12:12 [PATCH] Arm: i.MX5: Do not disable L2 cache Sascha Hauer
@ 2020-06-18 12:21 ` Lucas Stach
  2020-06-26  4:47   ` Sascha Hauer
  0 siblings, 1 reply; 3+ messages in thread
From: Lucas Stach @ 2020-06-18 12:21 UTC (permalink / raw)
  To: Sascha Hauer, Barebox List

Am Donnerstag, den 18.06.2020, 14:12 +0200 schrieb Sascha Hauer:
> Disabling the L2 cache is not working in imx5_init_lowlevel() because
> the necessary cache maintenance operations are missing. This often
> results in cache corruption in a chainloaded barebox.
> Disabling the cache is unnecessary: when we come from the ROM the L2
> cache is disabled anyway, so disabling it is a no-op. When we get here
> in a chainloaded barebox the L2 cache is already enabled and correctly
> configured. So instead of initializing it again we can take an enabled
> L2 cache as a sign to skip initialization and just return from the
> function.

While this won't hurt much, shouldn't we also disable the L2 cache on
Barebox shutdown, like we do on other SoCs?

Regards,
Lucas

> Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
> ---
>  arch/arm/mach-imx/imx5.c | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm/mach-imx/imx5.c b/arch/arm/mach-imx/imx5.c
> index 96288f99e0..dd6c079fe3 100644
> --- a/arch/arm/mach-imx/imx5.c
> +++ b/arch/arm/mach-imx/imx5.c
> @@ -37,10 +37,13 @@ void imx5_init_lowlevel(void)
>  {
>  	u32 r;
>  
> -	/* ARM errata ID #468414 */
>  	__asm__ __volatile__("mrc 15, 0, %0, c1, c0, 1":"=r"(r));
> +
> +	if (r & (1 << 1))
> +		return;
> +
> +	/* ARM errata ID #468414 */
>  	r |= (1 << 5);    /* enable L1NEON bit */
> -	r &= ~(1 << 1);   /* explicitly disable L2 cache */
>  	__asm__ __volatile__("mcr 15, 0, %0, c1, c0, 1" : : "r"(r));
>  
>          /* reconfigure L2 cache aux control reg */


_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH] Arm: i.MX5: Do not disable L2 cache
  2020-06-18 12:21 ` Lucas Stach
@ 2020-06-26  4:47   ` Sascha Hauer
  0 siblings, 0 replies; 3+ messages in thread
From: Sascha Hauer @ 2020-06-26  4:47 UTC (permalink / raw)
  To: Lucas Stach; +Cc: Barebox List

On Thu, Jun 18, 2020 at 02:21:08PM +0200, Lucas Stach wrote:
> Am Donnerstag, den 18.06.2020, 14:12 +0200 schrieb Sascha Hauer:
> > Disabling the L2 cache is not working in imx5_init_lowlevel() because
> > the necessary cache maintenance operations are missing. This often
> > results in cache corruption in a chainloaded barebox.
> > Disabling the cache is unnecessary: when we come from the ROM the L2
> > cache is disabled anyway, so disabling it is a no-op. When we get here
> > in a chainloaded barebox the L2 cache is already enabled and correctly
> > configured. So instead of initializing it again we can take an enabled
> > L2 cache as a sign to skip initialization and just return from the
> > function.
> 
> While this won't hurt much, shouldn't we also disable the L2 cache on
> Barebox shutdown, like we do on other SoCs?

Documentation/arm/booting.rst clearly states:

Data cache must be off.

So yes, to do it correctly we should disable the L2 cache.

So far I am happy that my QSB board no longer randomly crashes in a
secondary barebox.

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2020-06-26  4:47 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-06-18 12:12 [PATCH] Arm: i.MX5: Do not disable L2 cache Sascha Hauer
2020-06-18 12:21 ` Lucas Stach
2020-06-26  4:47   ` Sascha Hauer

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox