On 8/6/20 6:56 PM, Greg Kurz wrote:
> Calls to the KVM XIVE device are guarded by kvm_irqchip_in_kernel(). This
> ensures that QEMU won't try to use the device if KVM is disabled or if
> an in-kernel irqchip isn't required.
>
> When using ic-mode=dual with the pseries machine, we have two possible
> interrupt controllers: XIVE and XICS. The kvm_irqchip_in_kernel() helper
> will return true as soon as any of the KVM device is created. It might
> lure QEMU to think that the other one is also around, while it is not.
> This is exactly what happens with ic-mode=dual at machine init when
> claiming IRQ numbers, which must be done on all possible IRQ backends,
> eg. RTAS event sources or the PHB0 LSI table : only the KVM XICS device
> is active but we end up calling kvmppc_xive_source_reset_one() anyway,
> which fails. This doesn't cause any trouble because of another bug :
> kvmppc_xive_source_reset_one() lacks an error_setg() and callers don't
> see the failure.
>
> Most of the other kvmppc_xive_* functions have similar xive->fd
> checks to filter out the case when KVM XIVE isn't active. It
> might look safer to have idempotent functions but it doesn't
> really help to understand what's going on when debugging.
>
> Since we already have all the kvm_irqchip_in_kernel() in place,
> also have the callers to check xive->fd as well before calling
> KVM XIVE specific code. This is straight-forward for the spapr
> specific XIVE code. Some more care is needed for the platform
> agnostic XIVE code since it cannot access xive->fd directly.
> Introduce new in_kernel() methods in some base XIVE classes
> for this purpose and implement them only in spapr.
>
> In all cases, we still need to call kvm_irqchip_in_kernel() so that
> compilers can optimize the kvmppc_xive_* calls away when CONFIG_KVM
> isn't defined, thus avoiding the need for stubs.
>
> Signed-off-by: Greg Kurz <gr...@kaod.org>
> ---
> v2: Introduce in_kernel() abstract methods in the base XIVE classes
> ---
> hw/intc/spapr_xive.c | 53
> ++++++++++++++++++++++++++++++++++++-------------
> hw/intc/xive.c | 28 +++++++++++++++++++-------
> include/hw/ppc/xive.h | 2 ++
> 3 files changed, 62 insertions(+), 21 deletions(-)
>
> diff --git a/hw/intc/spapr_xive.c b/hw/intc/spapr_xive.c
> index 89c8cd96670b..cd001c580e89 100644
> --- a/hw/intc/spapr_xive.c
> +++ b/hw/intc/spapr_xive.c
> @@ -148,12 +148,19 @@ static void spapr_xive_end_pic_print_info(SpaprXive
> *xive, XiveEND *end,
> xive_end_queue_pic_print_info(end, 6, mon);
> }
>
> +/*
> + * kvm_irqchip_in_kernel() will cause the compiler to turn this
> + * info a nop if CONFIG_KVM isn't defined.
> + */
> +#define spapr_xive_in_kernel(xive) \
> + (kvm_irqchip_in_kernel() && (xive)->fd != -1)
> +
This looks ok. SpaprXive is the userspace frontend device of
the KVM XIVE native device in the hypervisor.
> void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
> {
> XiveSource *xsrc = &xive->source;
> int i;
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> Error *local_err = NULL;
>
> kvmppc_xive_synchronize_state(xive, &local_err);
> @@ -507,8 +514,10 @@ static const VMStateDescription vmstate_spapr_xive_eas =
> {
>
> static int vmstate_spapr_xive_pre_save(void *opaque)
> {
> - if (kvm_irqchip_in_kernel()) {
> - return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
> + SpaprXive *xive = SPAPR_XIVE(opaque);
> +
> + if (spapr_xive_in_kernel(xive)) {
> + return kvmppc_xive_pre_save(xive);
> }
>
> return 0;
> @@ -520,8 +529,10 @@ static int vmstate_spapr_xive_pre_save(void *opaque)
> */
> static int spapr_xive_post_load(SpaprInterruptController *intc, int
> version_id)
> {
> - if (kvm_irqchip_in_kernel()) {
> - return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
> + SpaprXive *xive = SPAPR_XIVE(intc);
> +
> + if (spapr_xive_in_kernel(xive)) {
> + return kvmppc_xive_post_load(xive, version_id);
> }
>
> return 0;
> @@ -564,7 +575,7 @@ static int spapr_xive_claim_irq(SpaprInterruptController
> *intc, int lisn,
> xive_source_irq_set_lsi(xsrc, lisn);
> }
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
> }
>
> @@ -641,7 +652,7 @@ static void spapr_xive_set_irq(SpaprInterruptController
> *intc, int irq, int val)
> {
> SpaprXive *xive = SPAPR_XIVE(intc);
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> kvmppc_xive_source_set_irq(&xive->source, irq, val);
> } else {
> xive_source_set_irq(&xive->source, irq, val);
> @@ -749,11 +760,21 @@ static void
> spapr_xive_deactivate(SpaprInterruptController *intc)
>
> spapr_xive_mmio_set_enabled(xive, false);
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> kvmppc_xive_disconnect(intc);
> }
> }
>
> +static bool spapr_xive_in_kernel_xptr(const XivePresenter *xptr)
> +{
> + return spapr_xive_in_kernel(SPAPR_XIVE(xptr));
> +}
This is mostly OK, a XivePresenter is a part of the XiveRouter.
> +static bool spapr_xive_in_kernel_xn(const XiveNotifier *xn)
> +{
> + return spapr_xive_in_kernel(SPAPR_XIVE(xn));
> +}
This is weird. we have other XiveNotifiers which have no relation
with a kernel backend.
> static void spapr_xive_class_init(ObjectClass *klass, void *data)
> {
> DeviceClass *dc = DEVICE_CLASS(klass);
> @@ -761,6 +782,7 @@ static void spapr_xive_class_init(ObjectClass *klass,
> void *data)
> SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
> XivePresenterClass *xpc = XIVE_PRESENTER_CLASS(klass);
> SpaprXiveClass *sxc = SPAPR_XIVE_CLASS(klass);
> + XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
>
> dc->desc = "sPAPR XIVE Interrupt Controller";
> device_class_set_props(dc, spapr_xive_properties);
> @@ -788,6 +810,9 @@ static void spapr_xive_class_init(ObjectClass *klass,
> void *data)
> sicc->post_load = spapr_xive_post_load;
>
> xpc->match_nvt = spapr_xive_match_nvt;
> + xpc->in_kernel = spapr_xive_in_kernel_xptr;
> +
> + xnc->in_kernel = spapr_xive_in_kernel_xn;
> }
>
> static const TypeInfo spapr_xive_info = {
> @@ -1058,7 +1083,7 @@ static target_ulong h_int_set_source_config(PowerPCCPU
> *cpu,
> new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
> }
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> Error *local_err = NULL;
>
> kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
> @@ -1379,7 +1404,7 @@ static target_ulong h_int_set_queue_config(PowerPCCPU
> *cpu,
> */
>
> out:
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> Error *local_err = NULL;
>
> kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end,
> &local_err);
> @@ -1480,7 +1505,7 @@ static target_ulong h_int_get_queue_config(PowerPCCPU
> *cpu,
> args[2] = 0;
> }
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> Error *local_err = NULL;
>
> kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end,
> &local_err);
> @@ -1642,7 +1667,7 @@ static target_ulong h_int_esb(PowerPCCPU *cpu,
> return H_P3;
> }
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
> flags & SPAPR_XIVE_ESB_STORE);
> } else {
> @@ -1717,7 +1742,7 @@ static target_ulong h_int_sync(PowerPCCPU *cpu,
> * under KVM
> */
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> Error *local_err = NULL;
>
> kvmppc_xive_sync_source(xive, lisn, &local_err);
> @@ -1761,7 +1786,7 @@ static target_ulong h_int_reset(PowerPCCPU *cpu,
>
> device_legacy_reset(DEVICE(xive));
>
> - if (kvm_irqchip_in_kernel()) {
> + if (spapr_xive_in_kernel(xive)) {
> Error *local_err = NULL;
>
> kvmppc_xive_reset(xive, &local_err);
> diff --git a/hw/intc/xive.c b/hw/intc/xive.c
> index 9b55e0356c62..27d27fdc9ee4 100644
> --- a/hw/intc/xive.c
> +++ b/hw/intc/xive.c
> @@ -592,6 +592,17 @@ static const char * const xive_tctx_ring_names[] = {
> "USER", "OS", "POOL", "PHYS",
> };
>
> +/*
> + * kvm_irqchip_in_kernel() will cause the compiler to turn this
> + * info a nop if CONFIG_KVM isn't defined.
> + */
> +#define xive_in_kernel(xptr) \
> + (kvm_irqchip_in_kernel() && \
> + ({ \
> + XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr); \
> + xpc->in_kernel ? xpc->in_kernel(xptr) : false; \
> + }))
> +
>
> void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
> {
> int cpu_index;
> @@ -606,7 +617,7 @@ void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor
> *mon)
>
> cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
>
> - if (kvm_irqchip_in_kernel()) {
> + if (xive_in_kernel(tctx->xptr)) {
> Error *local_err = NULL;
>
> kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
> @@ -671,7 +682,7 @@ static void xive_tctx_realize(DeviceState *dev, Error
> **errp)
> }
>
> /* Connect the presenter to the VCPU (required for CPU hotplug) */
> - if (kvm_irqchip_in_kernel()) {
> + if (xive_in_kernel(tctx->xptr)) {
> kvmppc_xive_cpu_connect(tctx, &local_err);
> if (local_err) {
> error_propagate(errp, local_err);
> @@ -682,10 +693,11 @@ static void xive_tctx_realize(DeviceState *dev, Error
> **errp)
>
> static int vmstate_xive_tctx_pre_save(void *opaque)
> {
> + XiveTCTX *tctx = XIVE_TCTX(opaque);
> Error *local_err = NULL;
>
> - if (kvm_irqchip_in_kernel()) {
> - kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
> + if (xive_in_kernel(tctx->xptr)) {
> + kvmppc_xive_cpu_get_state(tctx, &local_err);
> if (local_err) {
> error_report_err(local_err);
> return -1;
> @@ -697,14 +709,15 @@ static int vmstate_xive_tctx_pre_save(void *opaque)
>
> static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
> {
> + XiveTCTX *tctx = XIVE_TCTX(opaque);
> Error *local_err = NULL;
>
> - if (kvm_irqchip_in_kernel()) {
> + if (xive_in_kernel(tctx->xptr)) {
> /*
> * Required for hotplugged CPU, for which the state comes
> * after all states of the machine.
> */
> - kvmppc_xive_cpu_set_state(XIVE_TCTX(opaque), &local_err);
> + kvmppc_xive_cpu_set_state(tctx, &local_err);
> if (local_err) {
> error_report_err(local_err);
> return -1;
> @@ -1128,6 +1141,7 @@ static void xive_source_reset(void *dev)
> static void xive_source_realize(DeviceState *dev, Error **errp)
> {
> XiveSource *xsrc = XIVE_SOURCE(dev);
> + XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
>
> assert(xsrc->xive);
>
> @@ -1147,7 +1161,7 @@ static void xive_source_realize(DeviceState *dev, Error
> **errp)
> xsrc->status = g_malloc0(xsrc->nr_irqs);
> xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
>
> - if (!kvm_irqchip_in_kernel()) {
> + if (!xnc->in_kernel || !xnc->in_kernel(xsrc->xive)) {
> memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
> &xive_source_esb_ops, xsrc, "xive.esb",
> (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
> diff --git a/include/hw/ppc/xive.h b/include/hw/ppc/xive.h
> index 705cf48176fc..aa46e3fcf512 100644
> --- a/include/hw/ppc/xive.h
> +++ b/include/hw/ppc/xive.h
> @@ -161,6 +161,7 @@ typedef struct XiveNotifier XiveNotifier;
> typedef struct XiveNotifierClass {
> InterfaceClass parent;
> void (*notify)(XiveNotifier *xn, uint32_t lisn);
> + bool (*in_kernel)(const XiveNotifier *xn);
> } XiveNotifierClass;
>
> /*
> @@ -396,6 +397,7 @@ typedef struct XivePresenterClass {
> uint8_t nvt_blk, uint32_t nvt_idx,
> bool cam_ignore, uint8_t priority,
> uint32_t logic_serv, XiveTCTXMatch *match);
> + bool (*in_kernel)(const XivePresenter *xptr);
> } XivePresenterClass;
>
> int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
>
>
It seems redundant. Can we introduce a new XiveBackend QOM interface
which would implement an in_kernel() handler ? and XiveRouter would
inherit from it.
C.
