Re: [PATCH] x86/asm/entry/64: better check for canonical address

From: Ingo Molnar <mingo@kernel.org>
To: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Brian Gerst <brgerst@gmail.com>,
	Andy Lutomirski <luto@amacapital.net>,
	Borislav Petkov <bp@alien8.de>,
	the arch/x86 maintainers <x86@kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [PATCH] x86/asm/entry/64: better check for canonical address
Date: Sat, 28 Mar 2015 10:11:06 +0100	[thread overview]
Message-ID: <20150328091106.GA5361@gmail.com> (raw)
In-Reply-To: <55154DB3.9000008@redhat.com>

* Denys Vlasenko <dvlasenk@redhat.com> wrote:

> On 03/27/2015 01:16 PM, Ingo Molnar wrote:
> >>> Indeed, an IRET ought to be pretty cheap for same-ring interrupt 
> >>> returns in any case.
> >>
> >> Unfortunately, it is not. Try attached program.
> >>
> >> On this CPU, 1 ns ~= 3 cycles.
> >>
> >> $ ./timing_test64 callret
> >> 10000 loops in 0.00008s = 7.87 nsec/loop for callret
> >> 100000 loops in 0.00076s = 7.56 nsec/loop for callret
> >> 1000000 loops in 0.00548s = 5.48 nsec/loop for callret
> >> 10000000 loops in 0.02882s = 2.88 nsec/loop for callret
> >> 100000000 loops in 0.18334s = 1.83 nsec/loop for callret
> >> 200000000 loops in 0.36051s = 1.80 nsec/loop for callret
> >> 400000000 loops in 0.71632s = 1.79 nsec/loop for callret
> >>
> >> Near call + near ret = 5 cycles
> >>
> >> $ ./timing_test64 lret
> >> 10000 loops in 0.00034s = 33.95 nsec/loop for lret
> >> 100000 loops in 0.00328s = 32.83 nsec/loop for lret
> >> 1000000 loops in 0.04541s = 45.41 nsec/loop for lret
> >> 10000000 loops in 0.32130s = 32.13 nsec/loop for lret
> >> 20000000 loops in 0.64191s = 32.10 nsec/loop for lret
> >>
> >> push my_cs + push next_label + far ret = ~90 cycles
> >>
> >> $ ./timing_test64 iret
> >> 10000 loops in 0.00344s = 343.90 nsec/loop for iret
> >> 100000 loops in 0.01890s = 188.97 nsec/loop for iret
> >> 1000000 loops in 0.08228s = 82.28 nsec/loop for iret
> >> 10000000 loops in 0.77910s = 77.91 nsec/loop for iret
> >>
> >> This is the "same-ring interrupt return". ~230 cycles!  :(
> > 
> > Ugh, that's really expensive! Why is that so? Same-ring irqs are 
> > supposedly a lot simpler.
> 
> Descriptor checks for restored CS and SS,
> checking canonical-ness of RIP,
> supporting "return to TSS" (flags.NT bit),
> "return to VM86" (flags.VM bit),
> complex logic around restoring RFLAGS
>   ("don't allow CPL3 to be able to disable interrupts...
>   ...unless their flags.IOPL is 3." Gasp)
> return to 16-bit code ("do not touch high 16 bits")
> 
> All of this is a giant PITA to encode in microcode.

I guess they could optimize it by adding a single "I am a modern OS 
executing regular userspace" flag to the descriptor [or expressing the 
same as a separate instruction], to avoid all that legacy crap that 
won't trigger on like 99.999999% of systems ...

Thanks,

	Ingo