Re: [PATCH 5/7] x86/bugs: Only harden syscalls when needed

From: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
To: Josh Poimboeuf <jpoimboe@kernel.org>
Cc: x86@kernel.org, linux-kernel@vger.kernel.org,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Daniel Sneddon <daniel.sneddon@linux.intel.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Alexandre Chartre <alexandre.chartre@oracle.com>,
	Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Sean Christopherson <seanjc@google.com>,
	Andrew Cooper <andrew.cooper3@citrix.com>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Nikolay Borisov <nik.borisov@suse.com>,
	KP Singh <kpsingh@kernel.org>, Waiman Long <longman@redhat.com>,
	Borislav Petkov <bp@alien8.de>
Subject: Re: [PATCH 5/7] x86/bugs: Only harden syscalls when needed
Date: Thu, 11 Apr 2024 17:15:22 -0700	[thread overview]
Message-ID: <20240412001522.3zp2mzked4ksglkl@desk> (raw)
In-Reply-To: <97befd7c1e008797734dee05181c49056ff6de57.1712813475.git.jpoimboe@kernel.org>

On Wed, Apr 10, 2024 at 10:40:49PM -0700, Josh Poimboeuf wrote:
> Syscall hardening (i.e., converting the syscall indirect branch to a
> series of direct branches) may cause performance regressions in certain
> scenarios.  Only use the syscall hardening when indirect branches are
> considered unsafe.
> 
> Fixes: 1e3ad78334a6 ("x86/syscall: Don't force use of indirect calls for system calls")
> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
> ---
>  arch/x86/entry/common.c            | 30 +++++++++++++++++++++++++---
>  arch/x86/entry/syscall_32.c        | 11 +---------
>  arch/x86/entry/syscall_64.c        |  8 +-------
>  arch/x86/entry/syscall_x32.c       |  7 ++++++-
>  arch/x86/include/asm/cpufeatures.h |  1 +
>  arch/x86/include/asm/syscall.h     |  8 +++++++-
>  arch/x86/kernel/cpu/bugs.c         | 32 +++++++++++++++++++++++++++++-
>  7 files changed, 74 insertions(+), 23 deletions(-)
> 
> diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
> index 6de50b80702e..80d432d2fe44 100644
> --- a/arch/x86/entry/common.c
> +++ b/arch/x86/entry/common.c
> @@ -39,6 +39,28 @@
>  
>  #ifdef CONFIG_X86_64
>  
> +/*
> + * Do either a direct or an indirect call, depending on whether indirect calls
> + * are considered safe.
> + */
> +#define __do_syscall(table, func_direct, nr, regs)			\
> +({									\
> +	unsigned long __rax, __rdi, __rsi;				\
> +									\
> +	asm_inline volatile(						\
> +		ALTERNATIVE("call " __stringify(func_direct) "\n\t",	\
> +			    ANNOTATE_RETPOLINE_SAFE			\
> +			    "call *%[func_ptr]\n\t",			\

This will likely not insert the lfence before the indirect call in
spectre_v2=eibrs,lfence mode. As X86_FEATURE_INDIRECT_SAFE is not
cleared when eIBRS is enabled, this will not be converted to direct
call.

[...]
> @@ -1720,6 +1744,7 @@ static void __init spectre_v2_select_mitigation(void)
>  
>  	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
>  		pr_err(SPECTRE_V2_LFENCE_MSG);
> +		setup_clear_cpu_cap(X86_FEATURE_INDIRECT_SAFE);

I don't know if it intentional, this seems to be the duplicate of
X86_FEATURE_INDIRECT_SAFE clear later in SPECTRE_V2_LFENCE mode. Also it
seems a bit odd to do this here in SPECTRE_V2_CMD handling.

>  		mode = SPECTRE_V2_LFENCE;
>  		break;
>  
> @@ -1772,11 +1797,16 @@ static void __init spectre_v2_select_mitigation(void)
>  		break;
>  
>  	case SPECTRE_V2_LFENCE:
> +		setup_clear_cpu_cap(X86_FEATURE_INDIRECT_SAFE);
> +		fallthrough;
>  	case SPECTRE_V2_EIBRS_LFENCE:
>  		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
> -		fallthrough;
> +		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
> +		break;
>  
>  	case SPECTRE_V2_RETPOLINE:
> +		setup_clear_cpu_cap(X86_FEATURE_INDIRECT_SAFE);
> +		fallthrough;
>  	case SPECTRE_V2_EIBRS_RETPOLINE:
>  		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
>  		break;