[SRU L][PATCH 0/2] OS cannot boot successfully when enabling VMD in UEFI setup

[Jeff Lane]

From: Michael Reed <Michael.Reed at canonical.com>

BugLink: https://bugs.launchpad.net/bugs/2020022

[Impact]
When enabling VMD in UEFI setup, OS cannot boot successfully. And, the panic leads to the system reboot. The following log is shown:

[ 166.605518] DMAR: VT-d detected Invalidation Queue Error: Reason f
[ 166.605522] DMAR: VT-d detected Invalidation Time-out Error: SID ffff
[ 166.612445] DMAR: VT-d detected Invalidation Completion Error: SID ffff
[ 166.612447] DMAR: QI HEAD: UNKNOWN qw0 = 0x0, qw1 = 0x0
[ 166.612449] DMAR: QI PRIOR: UNKNOWN qw0 = 0x0, qw1 = 0x0
...

Additional info:
  * The issue happens on both Lenovo SE350 server and Lenovo SR850 v2 server.

Debugging info and fix commit info:
  * `git bisect` indicates the offending commit is 6aab5622296b ("PCI: vmd: Clean up domain before enumeration"). The root cause is that VMD driver tries to clear a PCI configuration space range when resetting a VMD domain (https://github.com/torvalds/linux/blob/master/drivers/pci/controller/vmd.c#L544), which leads to the failure.

[Fix]
  * Another `git bisect` indicates the fix commit is 20f3337d350c ("x86: don't use REP_GOOD or ERMS for small memory clearing). I confirmed that this commit can fix the issue.

Would it be possible to include the commit 20f3337d350c in Ubuntu 22.04.2/23.10 kernel?

[Test Plan]

Reproduce Step
1.Disable Intel VMD in BIOS settings
   System Settings --> Devices and I/O Ports --> Intel VMD technology --> Enable/Disable Intel VMD : Disabled

2.Install OS

3.Enable Intel VMD in BIOS settings
   System Settings --> Devices and I/O Ports --> Intel VMD technology --> Enable/Disable Intel VMD : Enabled

4.Rebooting will reproduce this issue

[ Where problems could occur ]
* Lenovo SE350 server and Lenovo SR850 v2 server
* The regression leads to the boot failure (cannot boot info OS successfully).

[ Other Info ]

https://code.launchpad.net/~mreed8855/ubuntu/+source/linux/+git/lunar/+ref/enable_vmd_lp_2020022
Jim Mattson (1):
  x86/cpufeatures: Add macros for Intel's new fast rep string features

Linus Torvalds (1):
  x86: don't use REP_GOOD or ERMS for small memory clearing

 arch/x86/include/asm/cpufeatures.h |  3 ++
 arch/x86/lib/memset_64.S           | 47 +++++++-----------------------
 2 files changed, 14 insertions(+), 36 deletions(-)

-- 
2.34.1