Re: [PATCH v5 08/12] x86/tdx: Add HLT support for TDX guest

From: Borislav Petkov <bp@alien8.de>
To: Kuppuswamy Sathyanarayanan  <sathyanarayanan.kuppuswamy@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Andy Lutomirski <luto@kernel.org>, Peter H Anvin <hpa@zytor.com>,
	Dave Hansen <dave.hansen@intel.com>,
	Tony Luck <tony.luck@intel.com>,
	Dan Williams <dan.j.williams@intel.com>,
	Andi Kleen <ak@linux.intel.com>,
	Kirill Shutemov <kirill.shutemov@linux.intel.com>,
	Sean Christopherson <seanjc@google.com>,
	Kuppuswamy Sathyanarayanan <knsathya@kernel.org>,
	x86@kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v5 08/12] x86/tdx: Add HLT support for TDX guest
Date: Tue, 24 Aug 2021 18:10:41 +0200	[thread overview]
Message-ID: <YSUaAQPiBUqubBHM@zn.tnic> (raw)
In-Reply-To: <20210804181329.2899708-9-sathyanarayanan.kuppuswamy@linux.intel.com>

On Wed, Aug 04, 2021 at 11:13:25AM -0700, Kuppuswamy Sathyanarayanan wrote:
> @@ -240,6 +243,32 @@ SYM_FUNC_START(__tdx_hypercall)
>  
>  	movl $TDVMCALL_EXPOSE_REGS_MASK, %ecx
>  
> +	/*
> +	 * For the idle loop STI needs to be called directly before
> +	 * the TDCALL that enters idle (EXIT_REASON_HLT case). STI
> +	 * enables interrupts only one instruction later. If there
> +	 * are any instructions between the STI and the TDCALL for
> +	 * HLT then an interrupt could happen in that time, but the
> +	 * code would go back to sleep afterwards, which can cause
> +	 * longer delays.

<-- newline in the comment here for better readability.

> There leads to significant difference in

"There leads..." ?

> +	 * network performance benchmarks. So add a special case for
> +	 * EXIT_REASON_HLT to trigger sti before TDCALL. But this
> +	 * change is not required for all HLT cases. So use R15
> +	 * register value to identify the case which needs sti. So,

s/sti/STI/g

> +	 * if R11 is EXIT_REASON_HLT and R15 is 1, then call sti
> +	 * before TDCALL instruction. Note that R15 register is not
> +	 * required by TDCALL ABI when triggering the hypercall for
> +	 * EXIT_REASON_HLT case. So use it in software to select the
> +	 * sti case.
> +	 */
> +	cmpl $EXIT_REASON_HLT, %r11d
> +	jne skip_sti
> +	cmpl $1, %r15d
> +	jne skip_sti
> +	/* Set R15 register to 0, it is unused in EXIT_REASON_HLT case */
> +	xor %r15, %r15
> +	sti
> +skip_sti:
>  	tdcall

...

> +static __cpuidle void tdg_safe_halt(void)
> +{
> +	u64 ret;
> +
> +	/*
> +	 * Enable interrupts next to the TDVMCALL to avoid
> +	 * performance degradation.

That comment needs some more love to say exactly what the problem is.

> +	 */
> +	local_irq_enable();
> +
> +	/* IRQ is enabled, So set R12 as 0 */
> +	ret = _tdx_hypercall(EXIT_REASON_HLT, 0, 0, 0, 1, NULL);
> +
> +	/* It should never fail */
> +	BUG_ON(ret);
> +}

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette