x86: Mishandling of SYSCALL singlestep during emulation

ISSUE DESCRIPTION

The typical behaviour of singlestepping exceptions is determined at the start of the instruction, with a #DB trap being raised at the end of the instruction.
SYSCALL (and SYSRET, although we don’t implement it) behave differently because the typical behaviour allows userspace to escalate its privilege. (This difference in behaviour seems to be undocumented.)
Xen wrongly raised the exception based on the flags at the start of the instruction.

IMPACT

Guest userspace which can invoke the instruction emulator can use this flaw to escalate its privilege to that of the guest kernel.

VULNERABLE SYSTEMS

All Xen versions are affected.
The vulnerability is only exposed to 64-bit x86 HVM guests.
On Xen 4.6 and earlier the vulnerability is exposed to all guest user processes, including unprivileged processes, in such guests.
On Xen 4.7 and later, the vulnerability is exposed only to guest user processes granted a degree of privilege (such as direct hardware access) by the guest administrator; or, to all user processes when the VM has been explicitly configured with a non-default cpu vendor string (in xm/xl, this would be done with a `cpuid=’ domain config option).
A 64-bit guest kernel which uses an IST for #DB handling will most likely mitigate the issue, but will have a single unexpected #DB exception frame to deal with. This in practice means that Linux is not vulnerable.
The vulnerability is not exposed to 32-bit HVM guests. This is because the emulation bug also matches real hardware behaviour, and a 32-bit guest kernel using SYSCALL will already have to be using a Task Gate for handling #DB to avoid being susceptible to an escalation of privilege.
The vulnerability is not exposed to PV guests.
ARM systems are not vulnerable.