[U-Boot] [PATCH] arm: exynos: Use the generic lowlevel_init instead of the specific one

Tue Nov 15 08:33:15 CET 2016

Hi, Thomas, Alex and York,

Before there are some discussions about this patch, could we make a solution now? Or else, the patches about [PATCH v8 0/3] armv8: Support loading 32-bit OS in AArch32 execution state can't be merged, as the compiling will fail without this patch.

Thomas, is ARMV8_MULTIENTRY enabled on Exynos7420 now? If not, is there a good way to enable ARMV8_MULTIENTRY on Exynos7420 now?

Thanks.

Best Regards,
Alison Wang

> -----Original Message-----
> From: Thomas Abraham [mailto:ta.omasab at gmail.com]
> Sent: Tuesday, September 20, 2016 4:45 PM
> To: Alexander Graf <agraf at suse.de>
> Cc: Alison Wang <b18965 at freescale.com>; thomas.ab at samsung.com; Minkyu
> Kang <mk7.kang at samsung.com>; york sun <york.sun at nxp.com>; U-Boot
> Mailing List <u-boot at lists.denx.de>; Jason Jin <jason.jin at nxp.com>
> Subject: Re: [U-Boot] [PATCH] arm: exynos: Use the generic
> lowlevel_init instead of the specific one
> 
> On Tue, Sep 20, 2016 at 12:03 PM, Alexander Graf <agraf at suse.de> wrote:
> >
> >
> > On 20.09.16 08:25, Thomas Abraham wrote:
> >> On Tue, Sep 20, 2016 at 11:35 AM, Alexander Graf <agraf at suse.de>
> wrote:
> >>>
> >>>
> >>> Am 20.09.2016 um 07:51 schrieb Thomas Abraham <ta.omasab at gmail.com>:
> >>>
> >>> Hi Alison,
> >>>
> >>>
> >>> On Mon, Sep 19, 2016 at 7:38 PM, Alexander Graf <agraf at suse.de>
> wrote:
> >>>
> >>>
> >>>
> >>> On 09.09.16 10:48, Alison Wang wrote:
> >>>
> >>> This patch is to use the the generic lowlevel_init instead of the
> >>>
> >>> specific one.
> >>>
> >>>
> >>> Signed-off-by: Alison Wang <alison.wang at nxp.com>
> >>>
> >>>
> >>> If I had to guess, I'd think they only had their own version
> because the
> >>>
> >>> old one required a GIC.
> >>>
> >>>
> >>> I apologize for the delay.
> >>>
> >>> The reason for using a custom version was to avoid enabling
> >>> ARMV8_MULTIENTRY config option since the Exynos7 code was ready for
> >>> it.
> >>>
> >>>
> >>> Either way, since Samsung doesn't reply, I'm fine potentially
> breaking
> >>>
> >>> their boards if that means that we can make progress for actively
> >>>
> >>> maintained ones:
> >>>
> >>>
> >>>  Reviewed-by: Alexander Graf <agraf at suse.de>
> >>>
> >>>
> >>> This patch without the ARMV8_MULTIENTRY and ARMV8_SWITCH_TO_EL1
> config
> >>> options does not switch the boot CPU from EL3 to EL1. So it would
> be
> >>> preferable to not merge this patch until ARMV8_MULTIENTRY  is
> enabled
> >>> for Exynos7.
> >>>
> >>>
> >>> Why do you want to switch it to EL1 in the first place? Linux is
> very happy
> >>> to live in EL2 - which is what we call it in by default.
> >>
> >> Okay, there is no particular requirement to be in EL1 for Exynos7.
> EL2
> >> would also be fine. But Exynos7 support in u-boot is not yet ready
> for
> >> enabling ARMV8_MULTIENTRY config option. Is there anything be
> blocked
> >> due to Exynos7 using a custom lowlevel_init function?
> >
> > Yes, we're changing the semantics of armv8_switch_to_el2 and
> > armv8_switch_to_el1:
> >
> >   http://lists.denx.de/pipermail/u-boot/2016-September/266217.html
> >
> > which is a prerequisite for AArch32 kernel boot on AArch64 systems.
> 
> Okay.
> 
> >
> > How quickly do you think you could make Exynos7 work with MULTIENTRY?
> >
> 
> Exynos7420 uses CPU 0 of Cluster 1 as boot CPU (master CPU). The macro
> 'branch_if_master' requires all affinity values to be zero for a CPU
> to be identified as a master CPU. And so the boot CPU is incorrectly
> detected as a slave CPU. I have tested with the following temporary
> workaround to enable ARMV8_MULTIENTRY on Exynos7420. If it looks fine,
> this can be merged along with Alison's patch.
> 
> Thomas.
> 
> diff --git a/arch/arm/mach-exynos/Kconfig b/arch/arm/mach-
> exynos/Kconfig
> index ce2a16f..45c5eeb 100644
> --- a/arch/arm/mach-exynos/Kconfig
> +++ b/arch/arm/mach-exynos/Kconfig
> @@ -126,6 +126,8 @@ choice
>  config  TARGET_ESPRESSO7420
>         bool "ESPRESSO7420 board"
>         select ARM64
> +       select ARMV8_MULTIENTRY
> +       select ARMV8_SWITCH_TO_EL1
>         select SUPPORT_SPL
>         select OF_CONTROL
>         select SPL_DISABLE_OF_CONTROL
> diff --git a/arch/arm/mach-exynos/soc.c b/arch/arm/mach-exynos/soc.c
> index f9c7468..6c3ebb0 100644
> --- a/arch/arm/mach-exynos/soc.c
> +++ b/arch/arm/mach-exynos/soc.c
> @@ -9,6 +9,16 @@
>  #include <asm/io.h>
>  #include <asm/system.h>
> 
> +#ifdef CONFIG_TARGET_ESPRESSO7420
> +/*
> + * Exynos7420 uses CPU0 of Cluster-1 as boot CPU. Due to this
> branch_if_master
> + * fails to identify as the master CPU. As temporary workaround, setup
> the
> + * slave CPU boot address as "_main".
> + */
> +extern void _main(void);
> +void *secondary_boot_addr = (void *)_main;
> +#endif /* CONFIG_TARGET_ESPRESSO7420 */
> +
>  void reset_cpu(ulong addr)
>  {
>  #ifdef CONFIG_CPU_V7
> diff --git a/include/configs/exynos7420-common.h
> b/include/configs/exynos7420-common.h
> index 9e03962..6f58aef 100644
> --- a/include/configs/exynos7420-common.h
> +++ b/include/configs/exynos7420-common.h
> @@ -48,6 +48,7 @@
>  #define CONFIG_IRAM_BASE               0x02100000
>  #define CONFIG_IRAM_SIZE               0x58000
>  #define CONFIG_IRAM_END                        (CONFIG_IRAM_BASE +
> CONFIG_IRAM_SIZE)
> +#define CPU_RELEASE_ADDR               secondary_boot_addr
> 
>  /* Number of CPUs available */
>  #define CONFIG_CORE_COUNT              0x8
> 
> >
> > Alex