From: Michael Kelley <redacted> Sent: Sunday, June 13, 2021 7:42 PM

From: Mark Rutland <mark.rutland@arm.com> Sent: Thursday, June 10, 2021 9:45 AM

quoted

Hi Michael,

[trimming the bulk of the thrread]

On Tue, Jun 08, 2021 at 03:36:06PM +0000, Michael Kelley wrote:

quoted

I've had a couple rounds of discussions with the Hyper-V team.   For
the clocksource we've agreed to table the live migration discussion, and
I'll resubmit the code so that arm_arch_timer.c provides the
standard arch_sys_counter clocksource.  As noted previously, this just
works for a Hyper-V guest.  The live migration discussion may come
back later after a deeper investigation by Hyper-V.

Great; thanks for this!

quoted

For clockevents, there's not a near term fix.  It's more than just plumbing
an interrupt for Hyper-V to virtualize the ARM64 arch timer in a guest VM.
From their perspective there's also benefit in having a timer abstraction
that's independent of the architecture, and in the Linux guest, the STIMER
code is common across x86/x64 and ARM64.  It follows the standard Linux
clockevents model, as it should. The code is already in use in out-of-tree
builds in the Linux VMs included in Windows 10 on ARM64 as part of the
so-called "Windows Subsystem for Linux".

So I'm hoping we can get this core support for ARM64 guests on Hyper-V
into upstream using the existing STIMER support.  At some point, Hyper-V
will do the virtualization of the ARM64 arch timer, but we don't want to
have to stay out-of-tree until after that happens.

My main concern here is making sure that we can rely on architected
properties, and don't have to special-case architected bits for hyperv
(or any other hypervisor), since that inevitably causes longer-term
pain.

While in abstract I'm not as worried about using the timer
clock_event_device specifically, that same driver provides the
clocksource and the event stream, and I want those to work as usual,
without being tied into the hyperv code. IIUC that will require some
work, since the driver won't register if the GTDT is missing timer
interrupts (or if there is no GTDT).

I think it really depends on what that looks like.

Mark,

Here are the details:

The existing initialization and registration code in arm_arch_timer.c
works in a Hyper-V guest with no changes.  As previously mentioned,
the GTDT exists and is correctly populated.  Even though it isn't used,
there's a PPI INTID specified for the virtual timer, just so
the "arm_sys_timer" clockevent can be initialized and registered.
The IRQ shows up in the output of "cat /proc/interrupts" with zero counts
for all CPUs since no interrupts are ever generated.  The EL1 virtual
timer registers (CNTV_CVAL_EL0, CNTV_TVAL_EL0, and CNTV_CTL_EL0)
are accessible in the VM.  The "arm_sys_timer" clockevent is left in
a shutdown state with CNTV_CTL_EL0.ENABLE set to zero when the
Hyper-V STIMER clockevent is registered with a higher rating.

Event streams are initialized and the __delay() implementation
for ARM64 inside the kernel works.  However, on the Ampere
eMAG hardware I'm using for testing, the WFE instruction returns
more quickly than it should even though the event stream fields in
CNTKCTL_EL1 are correct.  I have a query in to the Hyper-V team
to see if they are trapping WFE and just returning, vs. perhaps the
eMAG processor takes the easy way out and has WFE just return
immediately.  I'm not knowledgeable about other uses of timer
event streams, so let me know if there are other usage scenarios
I should check.

I confirmed that Hyper-V is not trapping the WFE instruction.  And
on a Marvell TX2 and on an Ampere Altra, the counter event stream
and WFE provide the expected delay.  Evidently WFE on the eMAG
doesn't actually delay.  Bottom line:  event streams work as expected
in a Hyper-V VM.  No changes needed to arm_arch_timer.[ch].

Michael

Finally, the "arch_sys_counter" clocksource gets initialized and
setup correctly.  If the Hyper-V clocksource is also initialized,
you can flip between the two clocksources at runtime as expected.
If the Hyper-V clocksource is not setup, then Linux in the VM runs
fine with the "arch_sys_counter" clocksource.

Michael