[Hyper-V] x86/hyperv: Stop suppressing X86_FEATURE_PCID

[Joshua R. Poulson]

Anywhere where Meltdown fixes have gone in have likely also picked up upstream work on TLB flushes. This is an important performance change.

x86/hyperv: Stop suppressing X86_FEATURE_PCIDx86/hyperv

When hypercall-based TLB flush was enabled for Hyper-V guests PCID feature
was deliberately suppressed as a precaution: back then PCID was never
exposed to Hyper-V guests and it wasn't clear what will happen if some day
it becomes available. The day came and PCID/INVPCID features are already
exposed on certain Hyper-V hosts.

From TLFS (as of 5.0b) it is unclear how TLB flush hypercalls combine with
PCID. In particular the usage of PCID is per-cpu based: the same mm gets
different CR3 values on different CPUs. If the hypercall does exact
matching this will fail. However, this is not the case. David Zhang
explains:

 "In practice, the AddressSpace argument is ignored on any VM that supports
  PCIDs.

  Architecturally, the AddressSpace argument must match the CR3 with PCID
  bits stripped out (i.e., the low 12 bits of AddressSpace should be 0 in
  long mode). The flush hypercalls flush all PCIDs for the specified
  AddressSpace."

With this, PCID can be enabled.

Upstream commit link: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/commit/?id=04651dd978a8749e59065df14b970a127f219ac2

[Marcelo Cerri (mhcerri)]

Josh. Is that intended to be included in both the custom kernels as well in the regular kernels?

[Marcelo Cerri (mhcerri)]

For linux-azure: https://lists.ubuntu.com/archives/kernel-team/2018-January/089627.html

[Joshua R. Poulson (jrp)]

Anywhere where you've patched for Meltdown. If PCID is not visible to the guest, there's a TLB flush on every syscall, even when it's not necessary.

[Marcelo Cerri (mhcerri)]

Will this patch land for 4.15? Otherwise I will include it to bionic too.

[Marcelo Cerri (mhcerri)]

Xenial and trusty kernels do not support TLB flush via Hyper-V hypercalls, thus the fix is not necessary on those kernels.

[Marcelo Cerri (mhcerri)]

The same is true for artful.

[Launchpad Janitor (janitor)]

This bug was fixed in the package linux-azure - 4.13.0-1009.12

---------------
linux-azure (4.13.0-1009.12) xenial; urgency=low

  * linux-azure: 4.13.0-1009.12 -proposed tracker (LP: #1746726)

  * libata: apply MAX_SEC_1024 to all LITEON EP1 series devices (LP: #1743053)
    - libata: apply MAX_SEC_1024 to all LITEON EP1 series devices

linux-azure (4.13.0-1008.11) xenial; urgency=low

  * linux-azure: 4.13.0-1008.11 -proposed tracker (LP: #1745669)

  * [Hyper-V] x86/hyperv: Stop suppressing X86_FEATURE_PCID (LP: #1745247)
    - x86/hyperv: Stop suppressing X86_FEATURE_PCID

  * [Hyper-V] scsi: storvsc: Spread interrupts when picking a channel for I/O
    requests (LP: #1745260)
    - SAUCE: scsi: storvsc: Spread interrupts when picking a channel for I/O
      requests

  * [Hyper-V] scsi: storvsc: Increase cmd_per_lun for higher speed devices
    (LP: #1745261)
    - SAUCE: scsi: storvsc: Increase cmd_per_lun for higher speed devices
    - [Config] CONFIG_SCSI_MQ_DEFAULT=y

 -- Marcelo Henrique Cerri <email address hidden> Thu, 01 Feb 2018 11:13:09 -0200