linux-kernel - Re: [PATCH v3] arm64: Cleanup __cpu_set_tcr

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20240410161217.GB25225@willie-the-truck>
Date: Wed, 10 Apr 2024 17:12:17 +0100
From: Will Deacon <will@...nel.org>
To: Seongsu Park <sgsu.park@...sung.com>
Cc: catalin.marinas@....com, mark.rutland@....com,
	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
	Leem ChaeHoon <infinite.run@...il.com>,
	Gyeonggeon Choi <gychoi@...dent.42seoul.kr>,
	Soomin Cho <to.soomin@...il.com>, DaeRo Lee <skseofh@...il.com>,
	kmasta <kmasta.study@...il.com>
Subject: Re: [PATCH v3] arm64: Cleanup __cpu_set_tcr_t0sz()

On Mon, Apr 08, 2024 at 11:40:16AM +0900, Seongsu Park wrote:
> In cpu_set_default_tcr_t0sz(), it is an error to shift TCR_T0SZ_OFFSET
> twice form TCR_T0SZ() and __cpu_set_tcr_t0sz().
> Since TCR_T0SZ_OFFSET is 0, no error occurred.
> We need to clarify whether the parameter of __cpu_set_tcr_t0sz is a
> shifted value or an unshifted value.
> 
> We have already shifted the value of t0sz in TCR_T0SZ by TCR_T0SZ_OFFSET.
> This is necessary for consistency with TCR_T1SZ.
> Therefore, the parameter of __cpu_set_tcr_t0sz is clarified as a shifted
> value.

This commit message needs reworking. I would suggest something like:

  The T0SZ field of TCR_EL1 occupies bits 0-5 of the register and
  encodes the virtual address space translated by TTBR0_EL1. When
  updating the field (for example, because we are switching to/from
  the idmap page-table), __cpu_set_tcr_t0sz() erroneously treats its
  't0sz' argument as unshifted, resulting in harmless but confusing
  double shifts by 0 in the code.

  Remove the unnecessary shifts.

> Co-developed-by: Leem ChaeHoon <infinite.run@...il.com>
> Signed-off-by: Leem ChaeHoon <infinite.run@...il.com>
> Co-developed-by: Gyeonggeon Choi <gychoi@...dent.42seoul.kr>
> Signed-off-by: Gyeonggeon Choi <gychoi@...dent.42seoul.kr>
> Co-developed-by: Soomin Cho <to.soomin@...il.com>
> Signed-off-by: Soomin Cho <to.soomin@...il.com>
> Co-developed-by: DaeRo Lee <skseofh@...il.com>
> Signed-off-by: DaeRo Lee <skseofh@...il.com>
> Co-developed-by: kmasta <kmasta.study@...il.com>
> Signed-off-by: kmasta <kmasta.study@...il.com>
> Signed-off-by: Seongsu Park <sgsu.park@...sung.com>

Honestly, although it's great that you all meet up to look at the kernel,
this long list of credits is a little absurd for a trivial patch like
this. Please can you decide who did the most work and give them the
credit? Hopefully there will be future opportunities for you all to
contribute!

> ---
> 
> v2:
>  - Condition is updated
> v3:
>  - Commit message is updated
>  - cpu_set_tcr_t0sz macro is added
> 
> ---
>  arch/arm64/include/asm/mmu_context.h | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/mmu_context.h b/arch/arm64/include/asm/mmu_context.h
> index c768d16b81a4..fb603ec7f61f 100644
> --- a/arch/arm64/include/asm/mmu_context.h
> +++ b/arch/arm64/include/asm/mmu_context.h
> @@ -72,15 +72,16 @@ static inline void __cpu_set_tcr_t0sz(unsigned long t0sz)
>  {
>  	unsigned long tcr = read_sysreg(tcr_el1);
>  
> -	if ((tcr & TCR_T0SZ_MASK) >> TCR_T0SZ_OFFSET == t0sz)
> +	if ((tcr & TCR_T0SZ_MASK) == t0sz)
>  		return;
>  
>  	tcr &= ~TCR_T0SZ_MASK;
> -	tcr |= t0sz << TCR_T0SZ_OFFSET;
> +	tcr |= t0sz;
>  	write_sysreg(tcr, tcr_el1);
>  	isb();
>  }
>  
> +#define cpu_set_tcr_t0sz(t0sz)		__cpu_set_tcr_t0sz(TCR_T0SZ(t0sz))
>  #define cpu_set_default_tcr_t0sz()	__cpu_set_tcr_t0sz(TCR_T0SZ(vabits_actual))
>  #define cpu_set_idmap_tcr_t0sz()	__cpu_set_tcr_t0sz(idmap_t0sz)
>  
> @@ -134,7 +135,7 @@ static inline void cpu_install_ttbr0(phys_addr_t ttbr0, unsigned long t0sz)
>  {
>  	cpu_set_reserved_ttbr0();
>  	local_flush_tlb_all();
> -	__cpu_set_tcr_t0sz(t0sz);
> +	cpu_set_tcr_t0sz(t0sz);

Sorry, but this is wrong. Please have a look at how cpu_install_ttbr0()
is called; specifically how trans_pgd_idmap_page() sets up 't0sz'.

So I don't think you should change cpu_install_ttbr0() at all and adding
a cpu_set_tcr_t0sz() macro which calls TCR_T0SZ on the 't0sz' argument
is a mistake.

Will