Re: [Qemu-ppc] [RFC PATCH v2 04/21] ppc/xive: provide a link to the sPAPR ICS object under XIVE

[David Gibson]

On Mon, Sep 11, 2017 at 07:12:18PM +0200, Cédric Le Goater wrote:
> The sPAPR machine first starts with a XICS interrupt model and
> depending on the guest capabilities, the XIVE exploitation mode is
> negotiated during CAS. A reset should then be performed to rebuild the
> device tree but the same IRQ numbers which were allocated by the
> devices prior to reset, when the XICS model was operating, are still
> in use.
> 
> For this purpose, we need a common IRQ number allocator for both the
> interrupt models: XICS legacy or XIVE exploitation. This is what the
> ICSIRQState array of the XICS interrupt source is used for. It also
> contains the LSI/MSI flag of an interrupt which will we need later on.
> 
> So, let's provide a link to the sPAPR ICS object under XIVE to make
> use of it.

Blech, please don't.  The XIVE code absolutely shouldn't be
referencing XICS objects, it's a recipe for trouble down the line.

If we have to have some sort of abstract "spapr interrupt source"
object that could map to either an ICS irq, or a XIVE source then we
can do that, but don't directly link XIVE and XICS.  *Especially* not
new-in-terms-of-old like this, rather than old-in-terms-of-new.


> 
> Signed-off-by: Cédric Le Goater <address@hidden>
> ---
>  hw/intc/spapr_xive.c        | 12 ++++++++++++
>  include/hw/ppc/spapr_xive.h |  4 ++++
>  2 files changed, 16 insertions(+)
> 
> diff --git a/hw/intc/spapr_xive.c b/hw/intc/spapr_xive.c
> index 6d98528fae68..1681affb0848 100644
> --- a/hw/intc/spapr_xive.c
> +++ b/hw/intc/spapr_xive.c
> @@ -56,6 +56,8 @@ void spapr_xive_reset(void *dev)
>  static void spapr_xive_realize(DeviceState *dev, Error **errp)
>  {
>      sPAPRXive *xive = SPAPR_XIVE(dev);
> +    Object *obj;
> +    Error *err = NULL;
>  
>      if (!xive->nr_targets) {
>          error_setg(errp, "Number of interrupt targets needs to be greater 
> 0");
> @@ -68,6 +70,16 @@ static void spapr_xive_realize(DeviceState *dev, Error 
> **errp)
>          return;
>      }
>  
> +    /* Retrieve SPAPR ICS source to share the IRQ number allocator */

This really suggests we need to move the irq number allocator out of
XICS and into the general spapr code.  Or get rid of it entirely
(using a more static irq mapping) if possible.

> +    obj = object_property_get_link(OBJECT(dev), "ics", &err);
> +    if (!obj) {
> +        error_setg(errp, "%s: required link 'ics' not found: %s",
> +                   __func__, error_get_pretty(err));
> +        return;
> +    }
> +
> +    xive->ics = ICS_BASE(obj);
> +
>      /* Allocate SBEs (State Bit Entry). 2 bits, so 4 entries per byte */
>      xive->sbe_size = DIV_ROUND_UP(xive->nr_irqs, 4);
>      xive->sbe = g_malloc0(xive->sbe_size);
> diff --git a/include/hw/ppc/spapr_xive.h b/include/hw/ppc/spapr_xive.h
> index b17dd4f17b0b..29112589b37f 100644
> --- a/include/hw/ppc/spapr_xive.h
> +++ b/include/hw/ppc/spapr_xive.h
> @@ -24,6 +24,7 @@
>  typedef struct sPAPRXive sPAPRXive;
>  typedef struct XiveIVE XiveIVE;
>  typedef struct XiveEQ XiveEQ;
> +typedef struct ICSState ICSState;
>  
>  #define TYPE_SPAPR_XIVE "spapr-xive"
>  #define SPAPR_XIVE(obj) OBJECT_CHECK(sPAPRXive, (obj), TYPE_SPAPR_XIVE)
> @@ -35,6 +36,9 @@ struct sPAPRXive {
>      uint32_t     nr_targets;
>      uint32_t     nr_irqs;
>  
> +    /* IRQ */
> +    ICSState     *ics;  /* XICS source inherited from the SPAPR machine */
> +
>      /* XIVE internal tables */
>      uint8_t      *sbe;
>      uint32_t     sbe_size;

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature