> Subject: Re: [PATCH] ARM: Add support for Realtek SOC
>
> On Thu, Sep 5, 2019 at 7:48 AM <jamestai....@gmail.com> wrote:
> >
> > From: "james.tai" <james....@realtek.com>
> >
> > This patch adds the basic machine file for the Realtek RTD16XX
> > platform.
> >
> > Signed-off-by: james.tai <james....@realtek.com>
>
> Hi James,
>
> Thanks a lot for your submission! I'm glad to see interest in upstream support
> for this SoC family. I have a few small comments on details, mostly where I
> would either like to see an explanation in the patch description, or things
> that
> looks like they can be left out from the patch.
>
Thanks for your reply.
> > index 33b00579beff..c7c9a3662eb7 100644
> > --- a/arch/arm/Kconfig
> > +++ b/arch/arm/Kconfig
> > @@ -836,6 +836,8 @@ source "arch/arm/mach-zx/Kconfig"
> >
> > source "arch/arm/mach-zynq/Kconfig"
> >
> > +source "arch/arm/mach-realtek/Kconfig"
> > +
> > diff --git a/arch/arm/mach-realtek/Kconfig
> b/arch/arm/mach-realtek/Kconfig
> > @@ -225,6 +226,7 @@ machine-$(CONFIG_ARCH_VT8500)
> += vt8500
> > machine-$(CONFIG_ARCH_W90X900) += w90x900
> > machine-$(CONFIG_ARCH_ZX) += zx
> > machine-$(CONFIG_ARCH_ZYNQ) += zynq
> > +machine-$(CONFIG_ARCH_REALTEK) += realtek
> > machine-$(CONFIG_PLAT_SPEAR) += spear
> >
> > # Platform directory name. This list is sorted alphanumerically
>
> Please keep these lists in alphabetical order.
>
I will keep these lists in alphabetical order in new version patch.
> > # ARMv7-M architecture
> > config ARCH_EFM32
> > bool "Energy Micro efm32"
> > diff --git a/arch/arm/Makefile b/arch/arm/Makefile index
> > c3624ca6c0bc..1f0926449d47 100644
> > --- a/arch/arm/Makefile
> > +++ b/arch/arm/Makefile
> > @@ -148,6 +148,7 @@ endif
> > textofs-$(CONFIG_ARCH_MSM8X60) := 0x00208000
> > textofs-$(CONFIG_ARCH_MSM8960) := 0x00208000
> > textofs-$(CONFIG_ARCH_MESON) := 0x00208000
> > +textofs-$(CONFIG_ARCH_REALTEK) := 0x00208000
> > textofs-$(CONFIG_ARCH_AXXIA) := 0x00308000
>
> Can you explain why this is needed for your platform?
>
We need to reserve memory (0x00000000 ~ 0x001B0000) for rom and boot code.
> > # Machine directory name. This list is sorted alphanumerically new
> > file mode 100644 index 000000000000..a8269964dbdb
> > --- /dev/null
> > +++ b/arch/arm/mach-realtek/Kconfig
> > @@ -0,0 +1,32 @@
> > +# SPDX-License-Identifier: GPL-2.0-only menuconfig ARCH_REALTEK
> > + bool "Realtek SoCs"
>
> Please add "depends on ARCH_MULTI_V7" to avoid compile time issues when
> selecting it on an earlier architecture.
>
I will add "depends on ARCH_MULTI_V7" in new version patch.
> > + select ARM_GLOBAL_TIMER
> > + select CLKDEV_LOOKUP
> > + select HAVE_SMP
> > + select HAVE_MACH_CLKDEV
> > + select GENERIC_CLOCKEVENTS
> > + select HAVE_SCHED_CLOCK
> > + select ARCH_HAS_CPUFREQ
> > + select CLKSRC_OF
> > + select ARCH_REQUIRE_GPIOLIB
> > + select GENERIC_IRQ_CHIP
> > + select IRQ_DOMAIN
> > + select PINCTRL
> > + select COMMON_CLK
> > + select ARCH_HAS_BARRIERS
> > + select SPARSE_IRQ
> > + select PM_OPP
> > + select ARM_HAS_SG_CHAIN
> > + select ARM_PATCH_PHYS_VIRT
> > + select AUTO_ZRELADDR
> > + select MIGHT_HAVE_PCI
> > + select MULTI_IRQ_HANDLER
> > + select PCI_DOMAINS if PCI
> > + select USE_OF
>
> Almost all of the symbols above are implied by
> ARCH_MULTI_V7 and should not be selected separately.
>
OK, I understand.
> > +config ARCH_RTD16XX
> > + bool "Enable support for RTD1619"
> > + depends on ARCH_REALTEK
> > + select ARM_GIC_V3
> > + select ARM_PSCI
>
> As I understand, this chip uses a Cortex-A55. What is the reason for adding
> support only to the 32-bit ARM architecture rather than 64-bit?
The RTD16XX platform also support the 64-bit ARM architecture.
I will add the 64-bit ARM architecture in new version patch.
> Most 64-bit SoCs are only supported with arch/arm64, but generally speaking
> that is not a requirement. My rule of thumb is that on systems with 1GB of
> RAM or more, one would want to run a 64-bit kernel, while systems with less
> than that are better off with a 32-bit one, but that is clearly not the only
> reason
> for picking one over the other.
>
Support 32-bit ARM architecture is for application compatibility.
> > +
> > +static int rtk_boot_secondary(unsigned int cpu, struct task_struct
> > +*idle) {
> > + unsigned long entry_pa = __pa_symbol(secondary_startup);
> > +
> > + writel_relaxed(entry_pa | (cpu << CPUID), cpu_release_virt);
> > +
> > + arch_send_wakeup_ipi_mask(cpumask_of(cpu));
> > +
> > + return 0;
> > +}
>
> It's very unusual to see custom smp operations on an ARMv8 system, as we
> normally use PSCI here. Can you explain what is going on here? Are you able to
> use a boot wrapper that implements these in psci instead?
>
The smp operations is porting form other Realtek platform.
Currently, The RTD16XX platform can use the PSCI method.
I will add PSCI method in new version patch.
> > +
> > +#include "platsmp.h"
> > +
> > +#define RBUS_BASE_PHYS (0x98000000)
> > +#define RBUS_BASE_VIRT (0xfe000000)
> > +#define RBUS_BASE_SIZE (0x00100000)
> > +
> > +static struct map_desc rtk_io_desc[] __initdata = {
> > + {
> > + .virtual = (unsigned long) IOMEM(RBUS_BASE_VIRT),
> > + .pfn = __phys_to_pfn(RBUS_BASE_PHYS),
> > + .length = RBUS_BASE_SIZE,
> > + .type = MT_DEVICE,
> > + },
> > +};
>
> This needs a comment: Why do you require a static mapping for
> "RBUS_BASE_PHYS"? Normally device drivers should just use
> ioremap() for mapping whichever registers they want to access.
>
The static mapping is for old Realtek devices driver.
I will 'use ioremap()' to replace with static mapping.
> > +static void __init rtk_dt_init(void)
> > +{
> > + of_platform_populate(NULL, of_default_bus_match_table, NULL,
> > +NULL); }
>
> This should be taken care of by the
> of_platform_default_populate_init() and can be dropped.
>
I will remove rtk_dt_init() in new version patch.
> > +static void __init rtk_timer_init(void) { #ifdef CONFIG_COMMON_CLK
> > + of_clk_init(NULL);
> > +#endif
>
> COMMON_CLK is implied by ARCH_MULTI_V7, so the #ifdef can be dropped.
>
OK, I understand.
> > + timer_probe();
> > + tick_setup_hrtimer_broadcast(); }
>
> What do you need tick_setup_hrtimer_broadcast() for? I don't see any other
> platform calling this.
>
I want to initialize the HR timer.
> > +bool __init rtk_smp_init_ops(void)
> > +{
> > + smp_set_ops(smp_ops(rtk_smp_ops));
> > +
> > + return true;
> > +}
>
> I think this can also be dropped, as you set the smp_ops in the machine
> descriptor.
>
OK, I understand.
> Arnd
>
> ------Please consider the environment before printing this e-mail.
