Re: [RFC PATCH 21/32] x86/fred: FRED entry/exit and dispatch code

From: Peter Zijlstra <peterz@infradead.org>
To: Xin Li <xin3.li@intel.com>
Cc: linux-kernel@vger.kernel.org, x86@kernel.org,
	kvm@vger.kernel.org, tglx@linutronix.de, mingo@redhat.com,
	bp@alien8.de, dave.hansen@linux.intel.com, hpa@zytor.com,
	andrew.cooper3@citrix.com, seanjc@google.com,
	pbonzini@redhat.com, ravi.v.shankar@intel.com
Subject: Re: [RFC PATCH 21/32] x86/fred: FRED entry/exit and dispatch code
Date: Tue, 20 Dec 2022 10:35:32 +0100	[thread overview]
Message-ID: <Y6GB5C1Bx60L9hk3@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <20221220063658.19271-22-xin3.li@intel.com>

On Mon, Dec 19, 2022 at 10:36:47PM -0800, Xin Li wrote:

> +static DEFINE_FRED_HANDLER(fred_exception)
> +{
> +	/*
> +	 * This intentially omits exceptions that cannot happen on FRED h/w:
> +	 * vectors _NOT_ listed are set to NULL.
> +	 */
> +	static const fred_handler exception_handlers[NUM_EXCEPTION_VECTORS] = {
> +		[X86_TRAP_DE] = fred_exc_divide_error,
> +		[X86_TRAP_DB] = fred_exc_debug,
> +		[X86_TRAP_NMI] = NULL, /* A separate event type, not handled here */
> +		[X86_TRAP_BP] = exc_int3,
> +		[X86_TRAP_OF] = fred_exc_overflow,
> +		[X86_TRAP_BR] = fred_exc_bounds,
> +		[X86_TRAP_UD] = exc_invalid_op,
> +		[X86_TRAP_NM] = fred_exc_device_not_available,
> +		[X86_TRAP_DF] = fred_exc_double_fault,
> +		[X86_TRAP_OLD_MF] = NULL, /* 387 only! */
> +		[X86_TRAP_TS] = fred_exc_invalid_tss,
> +		[X86_TRAP_NP] = fred_exc_segment_not_present,
> +		[X86_TRAP_SS] = fred_exc_stack_segment,
> +		[X86_TRAP_GP] = fred_exc_general_protection,
> +		[X86_TRAP_PF] = fred_exc_page_fault,
> +		[X86_TRAP_SPURIOUS] = NULL, /* Interrupts are their own event type */
> +		[X86_TRAP_MF] = fred_exc_coprocessor_error,
> +		[X86_TRAP_AC] = fred_exc_alignment_check,
> +		[X86_TRAP_MC] = fred_exc_machine_check,
> +		[X86_TRAP_XF] = fred_exc_simd_coprocessor_error
> +	};
> +	static const u32 noinstr_mask = BIT(X86_TRAP_DB) | BIT(X86_TRAP_BP) |
> +					BIT(X86_TRAP_DF) | BIT(X86_TRAP_PF) |
> +					BIT(X86_TRAP_MC) | BIT(X86_TRAP_UD);
> +	u8 vector = array_index_nospec((u8)regs->vector, NUM_EXCEPTION_VECTORS);
> +	irqentry_state_t state;
> +
> +	if (likely(exception_handlers[vector])) {

Can't you get rid of this branch by stuffing the exception_handlers[]
table with fred_bad_event?

> +		if (!(BIT(vector) & noinstr_mask)) {
> +			state = irqentry_enter(regs);
> +			instrumentation_begin();
> +		}
> +
> +		exception_handlers[vector](regs);
> +
> +		if (!(BIT(vector) & noinstr_mask)) {
> +			instrumentation_end();
> +			irqentry_exit(regs, state);
> +		}

This noinstr mask is daft; why not have DEFINE_FRED_HANDLER and
DEFINE_FRED_HANDLER_RAW or something, have the normal one include the
irqentry bits and use the _RAW one for the 'funny' ones that need to do
it themselves?

> +	} else {
> +		return fred_bad_event(regs);
> +	}

Then all this becomes:

	exception_handlers[vector](regs);

no branches, no nothing.

> +}