Re: [PATCH] docs: fusa: Add Assumption of Use (AOU)

[Julien Grall <julien@xxxxxxx>]

On 09/09/2024 10:50, Ayan Kumar Halder wrote:
> On 09/09/2024 10:11, Julien Grall wrote:
> > On 09/09/2024 09:56, Michal Orzel wrote:
> > > Hi Julien,
> > >
> > > On 08/09/2024 23:05, Julien Grall wrote:
> > > > Hi Ayan,
> > > >
> > > > On 06/09/2024 11:13, Ayan Kumar Halder wrote:
> > > > > From: Michal Orzel <michal.orzel@xxxxxxx>
> > > > >
> > > > > AOU are the assumptions Xen relies on other components (eg 
> > > > > platform, domains)
> > > > Searching online, I think the abbrevition is AoU rather than AOU. This
> > > > would also match how we abbreviate in Xen (IOW if we use a lower case
> > > > letter from the expanded name, then the letter in the acronym is also
> > > > lower case).
> > > >
> > > > > to fulfill its requirements. In our case, platform means a 
> > > > > combination of
> > > > > hardware, firmware and bootloader.
> > > > >
> > > > > We have defined AOU in the intro.rst and added AOU for the generic 
> > > > > timer.
> > > > > Signed-off-by: Michal Orzel <michal.orzel@xxxxxxx>
> > > > > Signed-off-by: Ayan Kumar Halder <ayan.kumar.halder@xxxxxxx>
> > > > > ---
> > > > >    .../reqs/design-reqs/arm64/generic-timer.rst  | 19 +++++++++++++ 
> > > > > ++++++
> > > > >    docs/fusa/reqs/intro.rst                      | 10 ++++++++++
> > > > >    2 files changed, 29 insertions(+)
> > > > >
> > > > > diff --git a/docs/fusa/reqs/design-reqs/arm64/generic-timer.rst b/ 
> > > > > docs/fusa/reqs/design-reqs/arm64/generic-timer.rst
> > > > > index f2a0cd7fb8..9df87cf4e0 100644
> > > > > --- a/docs/fusa/reqs/design-reqs/arm64/generic-timer.rst
> > > > > +++ b/docs/fusa/reqs/design-reqs/arm64/generic-timer.rst
> > > > > @@ -116,6 +116,25 @@ Rationale:
> > > > >
> > > > >    Comments:
> > > > >
> > > > > +Covers:
> > > > > + - `XenProd~emulated_timer~1`
> > > > > +
> > > > > +Assumption of Use on the Platform
> > > > > +=================================
> > > > > +
> > > > > +Expose system timer frequency via register
> > > > > +------------------------------------------
> > > > > +
> > > > > +`XenSwdgn~arm64_generic_timer_pf_program_cntfrq_el0~1`
> > > > > +
> > > > > +Description:
> > > > > +Underlying platform shall ensure that CNTFRQ_EL0 register contains 
> > > > > the system
> > > > > +timer frequency.
> > > > The wording in [1] (not yet merged) implies that CNTFRQ_EL0 may be
> > > It is merged:
> > > https://xenbits.xen.org/gitweb/? 
> > > p=xen.git;a=commit;h=51ad2c57a2d21b583a5944a0dc21c709af022f3c
> > > > invalid. This seems to contradict the Assumption of Use. Can you 
> > > > explain
> > > > the difference?
> > > The requirement you refer to is written from a domain perspective and 
> > > is about Xen exposing the frequency
> > > to domains via CNTFRQ and/or dt property. In case of a presence of dt 
> > > property in the host dtb, Xen could for instance decide
> > > to emulate CNTFRQ instead of relying on the domain to parse the dt at 
> > > runtime.
> > AFAICT, you can't trap CNTFRQ access. So what you suggest would not work.
> >
> > > The AoU on the platform (hw/firmware/bootloader) is written from Xen 
> > > perspective and is about the platform
> > > exposing the correct frequency via register. This is Xen expected 
> > > behavior on the platform. In other words, the platform should
> > > expose the correct frequency via register.
> > Xen is able to deal with broken CNTFRQ_EL0. 
> Yes, this is correct if the user provides "clock-frequency" dt property.
> > So I don't understand why we we would want to make an assumption that 
> > it shall not be broken. What do you gain?
> Refer https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/ 
> linux.git/tree/Documentation/devicetree/bindings/timer/arm,arch_timer.yaml
> ```
>
> Use of this property is strongly discouraged; fix your firmware unless 
> absolutely impossible.
> ```
>
> We wish to put the onus on the platform/firmware provider to program the 
> register correctly. Otherwise, we will have to document it somewhere 
> (may be safety manual) that User needs to provide the "clock-frequency" 
> dt property.
I think you will have to. The integrator may not have the possibility to 
modify the firmware.
> We wish to put as little responsibility on the user as 
> possible(especially when the dt documentation discourages it as well).
I think there are some contradiction with the requirements. I understand 
they are for the guest interface only but, if I am not mistaken, the 
only way the "clock-frequency" can be visible to the guest is because it 
is part of the host DT.
Furthermore, as soon as the property is specified, then Xen will ignore 
the value from CNTFRQ_EL0. So if we want to push the firmware vendor  to 
expose a valid CNTFREQ_EL0, then I don't understand why we would want to 
mention the property in the requirements.
Cheers,

--
Julien Grall