Re: [Qemu-devel] [PATCH 04/13] spapr/xive: introduce a VM state change handler

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 6595 bytes --]

On Mon, Jan 07, 2019 at 07:39:37PM +0100, Cédric Le Goater wrote:
> This handler is in charge of stabilizing the flow of event notifications
> in the XIVE controller before migrating a guest. This is a requirement
> before transferring the guest EQ pages to a destination.
> 
> When the VM is stopped, the handler masks the sources (PQ=01) to stop
> the flow of events and saves their previous state. The XIVE controller
> is then synced through KVM to flush any in-flight event notification
> and to stabilize the EQs. At this stage, the EQ pages are marked dirty
> to make sure the EQ pages are transferred if a migration sequence is
> in progress.
> 
> The previous configuration of the sources is restored when the VM
> resumes, after a migration or a stop.
> 
> Signed-off-by: Cédric Le Goater <clg@kaod.org>

Looks fine apart from interface details discussed on the kernel series.

> ---
>  include/hw/ppc/spapr_xive.h |   1 +
>  hw/intc/spapr_xive_kvm.c    | 111 +++++++++++++++++++++++++++++++++++-
>  2 files changed, 111 insertions(+), 1 deletion(-)
> 
> diff --git a/include/hw/ppc/spapr_xive.h b/include/hw/ppc/spapr_xive.h
> index 02f2de20111c..8815ed5aa372 100644
> --- a/include/hw/ppc/spapr_xive.h
> +++ b/include/hw/ppc/spapr_xive.h
> @@ -39,6 +39,7 @@ typedef struct sPAPRXive {
>      /* KVM support */
>      int           fd;
>      void          *tm_mmap;
> +    VMChangeStateEntry *change;
>  } sPAPRXive;
>  
>  bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
> diff --git a/hw/intc/spapr_xive_kvm.c b/hw/intc/spapr_xive_kvm.c
> index f52bddc92a2a..c7639ffe7758 100644
> --- a/hw/intc/spapr_xive_kvm.c
> +++ b/hw/intc/spapr_xive_kvm.c
> @@ -350,13 +350,119 @@ static void kvmppc_xive_get_eas_state(sPAPRXive *xive, Error **errp)
>      }
>  }
>  
> +/*
> + * Sync the XIVE controller through KVM to flush any in-flight event
> + * notification and stabilize the EQs.
> + */
> + static void kvmppc_xive_sync_all(sPAPRXive *xive, Error **errp)
> +{
> +    XiveSource *xsrc = &xive->source;
> +    Error *local_err = NULL;
> +    int i;
> +
> +    /* Sync the KVM source. This reaches the XIVE HW through OPAL */
> +    for (i = 0; i < xsrc->nr_irqs; i++) {
> +        XiveEAS *eas = &xive->eat[i];
> +
> +        if (!xive_eas_is_valid(eas)) {
> +            continue;
> +        }
> +
> +        kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_SYNC, i, NULL, true,
> +                          &local_err);
> +        if (local_err) {
> +            error_propagate(errp, local_err);
> +            return;
> +        }
> +    }
> +}
> +
> +/*
> + * The primary goal of the XIVE VM change handler is to mark the EQ
> + * pages dirty when all XIVE event notifications have stopped.
> + *
> + * Whenever the VM is stopped, the VM change handler masks the sources
> + * (PQ=01) to stop the flow of events and saves the previous state in
> + * anticipation of a migration. The XIVE controller is then synced
> + * through KVM to flush any in-flight event notification and stabilize
> + * the EQs.
> + *
> + * At this stage, we can mark the EQ page dirty and let a migration
> + * sequence transfer the EQ pages to the destination, which is done
> + * just after the stop state.
> + *
> + * The previous configuration of the sources is restored when the VM
> + * runs again.
> + */
> +static void kvmppc_xive_change_state_handler(void *opaque, int running,
> +                                             RunState state)
> +{
> +    sPAPRXive *xive = opaque;
> +    XiveSource *xsrc = &xive->source;
> +    Error *local_err = NULL;
> +    int i;
> +
> +    /*
> +     * Restore the sources to their initial state. This is called when
> +     * the VM resumes after a stop or a migration.
> +     */
> +    if (running) {
> +        for (i = 0; i < xsrc->nr_irqs; i++) {
> +            uint8_t pq = xive_source_esb_get(xsrc, i);
> +            if (xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8)) != 0x1) {
> +                error_report("XIVE: IRQ %d has an invalid state", i);
> +            }
> +        }
> +
> +        return;
> +    }
> +
> +    /*
> +     * Mask the sources, to stop the flow of event notifications, and
> +     * save the PQs locally in the XiveSource object. The XiveSource
> +     * state will be collected later on by its vmstate handler if a
> +     * migration is in progress.
> +     */
> +    for (i = 0; i < xsrc->nr_irqs; i++) {
> +        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_01);
> +        xive_source_esb_set(xsrc, i, pq);
> +    }
> +
> +    /*
> +     * Sync the XIVE controller in KVM, to flush in-flight event
> +     * notification that should be enqueued in the EQs.
> +     */
> +    kvmppc_xive_sync_all(xive, &local_err);
> +    if (local_err) {
> +        error_report_err(local_err);
> +        return;
> +    }
> +
> +    /*
> +     * Mark the XIVE EQ pages dirty to collect all updates.
> +     */
> +    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_CTRL,
> +                      KVM_DEV_XIVE_SAVE_EQ_PAGES, NULL, true, &local_err);
> +    if (local_err) {
> +        error_report_err(local_err);
> +    }
> +}
> +
>  void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp)
>  {
>      XiveSource *xsrc = &xive->source;
>      CPUState *cs;
>      Error *local_err = NULL;
>  
> -    kvmppc_xive_source_get_state(xsrc);
> +    /*
> +     * When the VM is stopped, the sources are masked and the previous
> +     * state is saved in anticipation of a migration. We should not
> +     * synchronize the source state in that case else we will override
> +     * the saved state.
> +     */
> +    if (runstate_is_running()) {
> +        kvmppc_xive_source_get_state(xsrc);
> +    }
>  
>      kvmppc_xive_get_eas_state(xive, &local_err);
>      if (local_err) {
> @@ -468,6 +574,9 @@ void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
>                                        "xive.tima", tima_len, xive->tm_mmap);
>      sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
>  
> +    xive->change = qemu_add_vm_change_state_handler(
> +        kvmppc_xive_change_state_handler, xive);
> +
>      kvm_kernel_irqchip = true;
>      kvm_msi_via_irqfd_allowed = true;
>      kvm_gsi_direct_mapping = true;

-- 
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

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]