RE: [PATCH] clockevents/drivers/i8253: Do not zero timer counter in shutdown

[David Woodhouse <dwmw2@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 2161 bytes --]




On 2023-02-08 at 01:04:19 +0000, "Michael Kelley (LINUX)" <mikelley@microsoft.com> said:
> From: lirongqing@baidu.com <lirongqing@baidu.com> Sent: Monday, February 6, 2023 5:15 PM
> > 
> > Zeroing the counter register in pit_shutdown() isn't actually supposed to
> > stop it from counting,  will causes the PIT to start running again,
> > From the spec:
> > 
> >   The largest possible initial count is 0; this is equivalent to 216 for
> >   binary counting and 104 for BCD counting.
> > 
> >   The Counter does not stop when it reaches zero. In Modes 0, 1, 4, and 5 the
> >   Counter "wraps around" to the highest count, either FFFF hex for binary
> >   count- ing or 9999 for BCD counting, and continues counting.
> > 
> >   Mode 0 is typically used for event counting. After the Control Word is
> >   written, OUT is initially low, and will remain low until the Counter
> >   reaches zero. OUT then goes high and remains high until a new count or a
> >   new Mode 0 Control Word is written into the Counter.
> > 
> > Hyper-V and KVM follow the spec, the issue that 35b69a42 "(clockevents/drivers/
> > i8253: Add support for PIT shutdown quirk") fixed is in i8253 drivers, not Hyper-v,
> > so delete the zero timer counter register in shutdown, and delete PIT shutdown
> > quirk for Hyper-v
> 
> From the standpoint of Hyper-V, I'm good with this change.  But there's a
> risk that old hardware might not be compliant with the spec, and needs the
> zero'ing for some reason. The experts in the x86 space will be in the best
> position to assess the risk.  At the time, the quirk approach was taken so
> the change applied only to Hyper-V, and any such risk was avoided.
> 
> For Hyper-V,
> Reviewed-by: Michael Kelley <mikelley@microsoft.com>

It's not entirely clear why we zero it at all. What was it supposed to
achieve?

I suppose if we want to be ultra-careful, perhaps we could do the
zeroing only on real hardware, if hypervisor_is_type(X86_HYPER_NATIVE)?

This does seem to be making VMMs spend a bunch of time stealing cycles
from the guest to emulate an unwanted timer, so it would be good to fix
it.

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 5965 bytes --]