Re: [PATCH v1] dt-bindings: riscv: add SBI PMU event mappings

[Conor Dooley]

On Thu, Dec 22, 2022 at 11:55:29AM -0800, Atish Kumar Patra wrote:
> On Thu, Dec 22, 2022 at 11:24 AM Conor Dooley <conor@xxxxxxxxxx> wrote:
> > On Thu, Dec 22, 2022 at 12:06:27PM -0600, Rob Herring wrote:
> > > On Wed, Dec 21, 2022 at 02:15:49PM +0000, Conor Dooley wrote:
> > > > From: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
> > > >
> > > > The SBI PMU extension requires a firmware to be aware of the event to
> > > > counter/mhpmevent mappings supported by the hardware. OpenSBI may use
> > > > DeviceTree to describe the PMU mappings. This binding is currently
> > > > described in markdown in OpenSBI (since v1.0 in Dec 2021) & used by QEMU
> > > > since v7.2.0.
> > > >
> > > > Import the binding for use while validating dtb dumps from QEMU and
> > > > upcoming hardware (eg JH7110 SoC) that will make use of the event
> > > > mapping.
> > > >
> > > > Link: https://github.com/riscv-software-src/opensbi/blob/master/docs/pmu_support.md
> > > > Co-developed-by: Atish Patra <atishp@xxxxxxxxxxxx>
> > > > Signed-off-by: Atish Patra <atishp@xxxxxxxxxxxx>
> > > > Signed-off-by: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
> > > > +description: |
> > > > +  SBI PMU extension supports allow supervisor software to configure, start &
> > > > +  stop any performance counter at anytime. Thus, a user can leverage full
> > > > +  capability of performance analysis tools such as perf if the SBI PMU
> > > > +  extension is enabled. The OpenSBI implementation makes the following
> > > > +  assumptions about the hardware platform:
> > > > +    MCOUNTINHIBIT CSR must be implemented in the hardware. Otherwise, the SBI
> > > > +    PMU extension will not be enabled.
> > > > +
> 
> This is no longer true since OpenSBI commit b28f070. We should remove
> this from OpenSBI docs as well.

Since you know the detail of it, I volunteer you for that one ;)

> > > > +    The platform must provide information about PMU event to counter mapping
> > > > +    via device tree or platform specific hooks. Otherwise, the SBI PMU
> > > > +    extension will not be enabled.
> > > > +
> > > > +    The platforms should provide information about the PMU event selector
> > > > +    values that should be encoded in the expected value of MHPMEVENTx while
> > > > +    configuring MHPMCOUNTERx for that specific event. This can be done via a
> > > > +    device tree or platform specific hooks. The exact value to be written to
> > > > +    the MHPMEVENTx is completely dependent on the platform. The generic
> > > > +    platform writes the zero-extended event_idx as the expected value for
> > > > +    hardware cache/generic events as suggested by the SBI specification.
> > > > +
> 
> But the larger question is these are OpenSBI implementation specific
> assumptions. Should it be included in
> the DT binding?

Probably not. I'll drop it for v2.

> > > > +properties:
> > > > +  compatible:
> > > > +    const: riscv,pmu
> > > > +
> > > > +  riscv,event-to-mhpmevent:
> > > > +    $ref: /schemas/types.yaml#/definitions/uint32-matrix
> > > > +    description:
> > > > +      Represents an ONE-to-ONE mapping between a PMU event and the event
> > > > +      selector value that platform expects to be written to the MHPMEVENTx CSR
> > > > +      for that event.
> > > > +      The mapping is encoded in an array format where each row represents an
> > >
> > > s/array/matrix/
> > >
> > > > +      event. The first element represents the event idx while the second &
> > > > +      third elements represent the event selector value that should be encoded
> > > > +      in the expected value to be written in MHPMEVENTx.
> > > > +      This property shouldn't encode any raw hardware event.
> > > > +    minItems: 1
> > > > +    maxItems: 255
> >
> > I copied these 255s from dt-schema somewhere as a sane max.
> > Atish, is there a cap here or should we allow as many as someone likes?
> > The md binding doesn't mention any limits - is it in the SBI spec?
> > The strongest wording I saw there was "limited" & event idx is 20 bits
> > wide as per the spec [0]. Does that make 2^20 the hard limit here?
> >
> 
> This is for hardware & cache events. The total number of events
> defined there are 52 (10 HW + 42 Cache) events.
> So 255 is a sane max that provides enough room for expansion in future
> if required.

WFM.

> > > > +  riscv,raw-event-to-mhpmcounters:
> > > > +    $ref: /schemas/types.yaml#/definitions/uint32-matrix
> > > > +    description:
> > > > +      Represents an ONE-to-MANY or MANY-to-MANY mapping between the rawevent(s)
> > > > +      and all the MHPMCOUNTERx in a bitmap format that can be used to monitor
> > > > +      that raw event.
> > > > +      The encoding of the raw events are platform specific. The information is
> > > > +      encoded in an array format where each row represents the specific raw
> > > > +      event(s). The first element is a 64-bit match value where the invariant
> > > > +      bits of range of events are set. The second element is a 64-bit mask that
> > > > +      will have all the variant bits of the range of events cleared. All other
> > > > +      bits should be set in the mask. The third element is a 32-bit value to
> > > > +      represent bitmap of all MHPMCOUNTERx that can monitor these set of
> > > > +      event(s). If a platform directly encodes each raw PMU event as a unique
> > > > +      ID, the value of select_mask must be 0xffffffff_ffffffff.
> > > > +    minItems: 1
> > > > +    maxItems: 255
> > > > +    items:
> > > > +      $ref: /schemas/types.yaml#/definitions/uint32-array
> > > > +      maxItems: 5
> > > > +
> > > > +required:
> > > > +  - compatible
> > >
> > > I assume at least one of the other properties must be present?
> >
> > Atish: (and +CC the OpenSBI list)
> > Which properties are actually needed here? Or are any? The md binding
> > in the OpenSBI sources doesn't seem to require anything other than the
> > compatible?
> >
> 
> That's true. Opensbi allows the platform to define functions
> to provide the mapping. Now that's an OpenSBI implementation thing.
> Other implementations may choose to use it differently.

In the case where the platform can define functions, the "bare" node
with just the compatible is still required though in that case, right?

Thanks,
Conor.

Attachment: signature.asc
Description: PGP signature