Re: [RFC PATCH] x86/arch_prctl: Add ARCH_SET_XCR0 to mask XCR0 per-thread

From: Keno Fischer <keno@juliacomputing.com>
To: Andi Kleen <andi@firstfloor.org>
Cc: "Linux Kernel Mailing List" <linux-kernel@vger.kernel.org>,
	"Thomas Gleixner" <tglx@linutronix.de>,
	"Ingo Molnar" <mingo@redhat.com>,
	x86@kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
	"Borislav Petkov" <bp@suse.de>,
	"Dave Hansen" <dave.hansen@linux.intel.com>,
	"Paolo Bonzini" <pbonzini@redhat.com>,
	"Radim Krčmář" <rkrcmar@redhat.com>,
	"Kyle Huey" <khuey@kylehuey.com>,
	"Robert O'Callahan" <robert@ocallahan.org>
Subject: Re: [RFC PATCH] x86/arch_prctl: Add ARCH_SET_XCR0 to mask XCR0 per-thread
Date: Sun, 17 Jun 2018 14:22:01 -0400	[thread overview]
Message-ID: <CABV8kRyuD6h_EGrxxQJ43FEaL9zuh3oYNYbt6ULqLaZZEiyoSg@mail.gmail.com> (raw)
In-Reply-To: <CABV8kRyf2R=a2zezgovByK9Lc8HhDXyfZ4OPE_DHfKKg34bNfw@mail.gmail.com>

> Almost difference in CPU behavior
> between the replayer and the replayee.

Not sure what happened to this sentence.
What I meant to say was:

Almost any difference in CPU behavior between
the replayer and the replayee will cause problems
for the determinism of the trace.

To elaborate on this, even if a register whose content
differs between the recording and the replay, it can
still cause problems down the line, e.g. if it is spilled
to the stack and that stack address is then re-used later.
In order for rr to work, we basically rely on the CPU
behaving *exactly* the same during the record and the
replay (down to counting performance counters the same).
This works well, but there are corner cases (like this XCR0
one).