[v3,0/1] Fix broken suspend on some machines

Message

Grzegorz Uriasz June 19, 2020, 10 p.m. UTC

Hi,

The included patch is a small subset of a bigger patch set spanning few
projects aiming to isolate the GPU in Qubes OS to a dedicated security domain.
I'm doing this together with 3 colleagues as part of our Bachelors thesis.
While working on the project we came across 2 machines - newer gaming
laptops on which the suspend functionality on unmodified xen is completely broken.

The affected machines were able to suspend but not always resume. Even
if the resume succeeded then the kernel time was trashed in the dmesg log
and the machine never managed to suspend another time. After changing
the xen clock to hpet, suspend started working again both on stock
xen and Qubes OS - this indicates a bug in the ACPI PM Timer. After
disassembling the FADT ACPI table on the ASUS FX504GM I understood that the
reported bit width is 32 bits but the flags indicate a 24 bit PM timer.

The included patch fixes the suspend feature on ASUS FX504GM and hopefully
other laptops - Probably next week I will test this patch on my
friend's laptop where this issue also occurs(suspend is broken, trashed kernel
time after resume).

Changes in v2:
- Check pm timer access width
- Proper timer width is set when xpm block is not present
- Cleanup timer initialization

Changes in v3:
- Check pm timer bit offset
- Warn about acpi firmware bugs
- Drop int cast in init_pmtimer
- Merge if's in init_pmtimer

Grzegorz Uriasz (1):
  x86/acpi: Use FADT flags to determine the PMTMR width

 xen/arch/x86/acpi/boot.c    | 19 +++++++++++++++----
 xen/arch/x86/time.c         | 12 ++++--------
 xen/include/acpi/acmacros.h |  8 ++++++++
 3 files changed, 27 insertions(+), 12 deletions(-)