[PATCH v2 05/21] arm: socfpga: Override 'lowlevel_init' to support ATF

[Ang, Chee Hong]

> On 2/20/20 8:05 PM, Ang, Chee Hong wrote:
> >> On 2/20/20 3:27 AM, Ang, Chee Hong wrote:
> >>>> On 2/19/20 1:25 PM, chee.hong.ang at intel.com wrote:
> >>>> [...]
> >>>>> diff --git a/arch/arm/mach-socfpga/lowlevel_init.S
> >>>>> b/arch/arm/mach-socfpga/lowlevel_init.S
> >>>>> new file mode 100644
> >>>>> index 0000000..68053a0
> >>>>> --- /dev/null
> >>>>> +++ b/arch/arm/mach-socfpga/lowlevel_init.S
> >>>>
> >>>> This should be some lowlevel_init_64.S to make it clear it's only
> >>>> for
> >>>> arm64 platforms.
> >>> OK. It makes sense. Thanks.
> >>>>
> >>>>> @@ -0,0 +1,85 @@
> >>>>> +/* SPDX-License-Identifier: GPL-2.0 */
> >>>>> +/*
> >>>>> + * Copyright (C) 2019, Intel Corporation  */
> >>>>> +
> >>>>> +#include <asm-offsets.h>
> >>>>> +#include <config.h>
> >>>>> +#include <linux/linkage.h>
> >>>>> +#include <asm/macro.h>
> >>>>> +
> >>>>> +ENTRY(lowlevel_init)
> >>>>> +	mov	x29, lr			/* Save LR */
> >>>>> +
> >>>>> +#if defined(CONFIG_GICV2) || defined(CONFIG_GICV3) #ifdef
> >>>>> +CONFIG_SPL_ATF
> >>>>> +	branch_if_slave x0, 2f
> >>>>> +#else
> >>>>> +	branch_if_slave x0, 1f
> >>>>> +#endif
> >>>>> +
> >>>>> +	ldr	x0, =GICD_BASE
> >>>>> +	bl	gic_init_secure
> >>>>> +#ifdef CONFIG_SPL_BUILD
> >>>>> +	b	2f
> >>>>> +#else
> >>>>> +	b	3f
> >>>>> +#endif
> >>>>
> >>>> Can't this be done in C code ? Can we reduce the ifdeffery ?
> >>> This lowlevel_init function is shared by SPL and U-Boot and they run
> >>> in slightly different flow.
> >>
> >> What does this 'different flow' mean ?
> > This has something to with multi-cores CPU such as A53.
> > For SPL, we need to make sure the slave CPUs (CPU1/2/3) trapped in a 'place'
> > Where they could be 'activated' by kernel for multi-processor environment.
> > It means the kernel get to 'activate' the slave CPUs from master CPU
> > (CPU0)  U-Boot proper only run on master CPU (CPU0). The rest of slave
> > CPUs are trapped in the beginning of SPL waiting to be 'activated'
> > by kernel.
> 
> OK, so the secondary CPUs are spinning until the kernel releases them.
> 
> > In U-Boot proper, only master CPU gets to run this code and it will
> > just do the basic GIC setup and skip the 'trap'. The 'trap' is to
> > prevent the slave CPUs from running the same SPL, ATF and U-Boot code
> > as the master CPU in parallel. Only single core (maser CPU) is needed for
> bootloaders and firmware.
> 
> I would expect all the other SMP platforms solved this issue with secondary
> CPUs already, so why is agilex special ?
Nothing special. Just like other SMP platforms, I am just telling you the master CPU
always skips the 'spinning trap' in SPL and U-Boot proper.