ACPICA: Avoid cache flush inside virtual machines
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Tue, 5 Apr 2022 23:29:38 +0000 (02:29 +0300)
committerDave Hansen <dave.hansen@linux.intel.com>
Thu, 7 Apr 2022 15:27:54 +0000 (08:27 -0700)
commite2efb6359e620521d1e13f69b2257de8ceaa9475
tree6069d7bfca3129c1a9ebc3e5d35170e1d2a0cf45
parentf4c9361f97c40d365c34f9cb8b8bc3eae0ee7778
ACPICA: Avoid cache flush inside virtual machines

While running inside virtual machine, the kernel can bypass cache
flushing. Changing sleep state in a virtual machine doesn't affect the
host system sleep state and cannot lead to data loss.

Before entering sleep states, the ACPI code flushes caches to prevent
data loss using the WBINVD instruction.  This mechanism is required on
bare metal.

But, any use WBINVD inside of a guest is worthless.  Changing sleep
state in a virtual machine doesn't affect the host system sleep state
and cannot lead to data loss, so most hypervisors simply ignore it.
Despite this, the ACPI code calls WBINVD unconditionally anyway.
It's useless, but also normally harmless.

In TDX guests, though, WBINVD stops being harmless; it triggers a
virtualization exception (#VE).  If the ACPI cache-flushing WBINVD
were left in place, TDX guests would need handling to recover from
the exception.

Avoid using WBINVD whenever running under a hypervisor.  This both
removes the useless WBINVDs and saves TDX from implementing WBINVD
handling.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20220405232939.73860-30-kirill.shutemov@linux.intel.com
arch/x86/include/asm/acenv.h