[PATCH] arm64: add early fixmap initialization flag

[Itaru Kitayama]

On Tue, Feb 20, 2024 at 11:55:30AM +0000, Mark Rutland wrote:
> On Tue, Feb 20, 2024 at 09:29:14AM +0900, Itaru Kitayama wrote:
> > On Mon, Feb 19, 2024 at 10:48:26AM +0000, Mark Rutland wrote:
> > > On Sat, Feb 17, 2024 at 11:03:26PM +0900, skseofh at gmail.com wrote:
> > > > From: Daero Lee <skseofh at gmail.com>
> > > > 
> > > > early_fixmap_init may be called multiple times. Since there is no
> > > > change in the page table after early fixmap initialization, an
> > > > initialization flag was added.
> > > 
> > > Why is that better?
> > > 
> > > We call early_fixmap_init() in two places:
> > > 
> > > * early_fdt_map()
> > > * setup_arch()
> > > 
> > > ... and to get to setup_arch() we *must* have gone through early_fdt_map(),
> > > since __primary_switched() calls that before going to setup_arch().
> > > 
> > > So AFAICT we can remove the second call to early_fixmap_init() in setup_arch(),
> > > and rely on the earlier one in early_fdt_map().
> > 
> > Removing the second call makes the code base a bit harder to understand
> > as the functions related to DT and ACPI setup are not separated cleanly.
> > I prefer calling the early_fixmap_init() in setup_arch() as well.
> 
> I appreciate what you're saying here, but I don't think that it's better to
> keep the second call in setup_arch().
> 
> We rely on having a (stub) DT in order to find UEFI and ACPI tables, so the DT
> and ACPI setup can never be truly separated. We always need to map that DT in
> order to find the UEFI+ACPI tables, and in order to do that we must initialize
> the fixmap first.

Okay.

> 
> I don't think there's any good reason to keep a redundant call in setup_arch();
> that's just misleading and potentially problematic if we ever change
> early_fixmap_init() so that it's not idempotent.
> 
> I agree it's somewhat a layering violation for  early_fdt_map() to be
> responsible for initialising the fixmap, so how about we just pull that out,
> e.g. as below?
> 
> Mark.
> 
> ---->8----
> From 5f07f9c1d352f760fa7aba97f1b4f42d9cb99496 Mon Sep 17 00:00:00 2001
> From: Mark Rutland <mark.rutland at arm.com>
> Date: Tue, 20 Feb 2024 11:09:17 +0000
> Subject: [PATCH] arm64: clean up fixmap initalization
> 
> Currently we have redundant initialization of the fixmap, first in
> early_fdt_map(), and then again in setup_arch() before we call
> early_ioremap_init(). This redundant initialization isn't harmful, as
> early_fixmap_init() happens to be idempotent, but it's redundant and
> potentially confusing.
> 
> We need to call early_fixmap_init() before we map the FDT and before we
> call early_ioremap_init(). Ideally we'd place early_fixmap_init() and
> early_ioremap_init() in the same caller as early_ioremap_init() is
> somewhat coupled with the fixmap code.
> 
> Clean this up by moving the calls to early_fixmap_init() and
> early_ioremap_init() into a new early_mappings_init() function, and
> calling this once in __primary_switched() before we call
> early_fdt_map(). This means we initialize the fixmap once, and keep
> early_fixmap_init() and early_ioremap_init() together.
> 
> This is a cleanup, not a fix, and does not need to be backported to
> stable kernels.
> 
> Reported-by: Daero Lee <skseofh at gmail.com>
> Signed-off-by: Mark Rutland <mark.rutland at arm.com>
> Cc: Catalin Marinas <catalin.marinas at arm.com>
> Cc: Itaru Kitayama <itaru.kitayama at linux.dev>
> Cc: Will Deacon <will at kernel.org>
> ---
>  arch/arm64/include/asm/setup.h |  1 +
>  arch/arm64/kernel/head.S       |  2 ++
>  arch/arm64/kernel/setup.c      | 11 ++++++-----
>  3 files changed, 9 insertions(+), 5 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/setup.h b/arch/arm64/include/asm/setup.h
> index 2e4d7da74fb87..c8ba018bcc09f 100644
> --- a/arch/arm64/include/asm/setup.h
> +++ b/arch/arm64/include/asm/setup.h
> @@ -9,6 +9,7 @@
>  
>  void *get_early_fdt_ptr(void);
>  void early_fdt_map(u64 dt_phys);
> +void early_mappings_init(void);
>  
>  /*
>   * These two variables are used in the head.S file.
> diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S
> index cab7f91949d8f..c60c5454c5704 100644
> --- a/arch/arm64/kernel/head.S
> +++ b/arch/arm64/kernel/head.S
> @@ -510,6 +510,8 @@ SYM_FUNC_START_LOCAL(__primary_switched)
>  #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
>  	bl	kasan_early_init
>  #endif
> +	bl	early_mappings_init
> +
>  	mov	x0, x21				// pass FDT address in x0
>  	bl	early_fdt_map			// Try mapping the FDT early
>  	mov	x0, x20				// pass the full boot status
> diff --git a/arch/arm64/kernel/setup.c b/arch/arm64/kernel/setup.c
> index 42c690bb2d608..7a539534ced78 100644
> --- a/arch/arm64/kernel/setup.c
> +++ b/arch/arm64/kernel/setup.c
> @@ -176,8 +176,6 @@ void __init *get_early_fdt_ptr(void)
>  asmlinkage void __init early_fdt_map(u64 dt_phys)
>  {
>  	int fdt_size;
> -
> -	early_fixmap_init();
>  	early_fdt_ptr = fixmap_remap_fdt(dt_phys, &fdt_size, PAGE_KERNEL);
>  }
>  
> @@ -290,6 +288,12 @@ u64 cpu_logical_map(unsigned int cpu)
>  	return __cpu_logical_map[cpu];
>  }
>  
> +asmlinkage void __init early_mappings_init(void)
> +{
> +	early_fixmap_init();
> +	early_ioremap_init();
> +}
> +
>  void __init __no_sanitize_address setup_arch(char **cmdline_p)
>  {
>  	setup_initial_init_mm(_stext, _etext, _edata, _end);
> @@ -305,9 +309,6 @@ void __init __no_sanitize_address setup_arch(char **cmdline_p)
>  	 */
>  	arm64_use_ng_mappings = kaslr_requires_kpti();
>  
> -	early_fixmap_init();
> -	early_ioremap_init();
> -
>  	setup_machine_fdt(__fdt_pointer);
>  
>  	/*

Thanks for this. Makes sense to me; would you post a proper patch
so I can build and do a boot test, just to make sure?

Reviewed-by: Itaru Kitayama <itaru.kitayama at fujitsu.com>

Thanks,
Itaru.

> -- 
> 2.30.2
>