A typical pseries VM with 16 vCPUs, one disk, one network adapater
uses less than 100 interrupts but the whole IRQ number space of the
QEMU machine is allocated at reset time and it is 8K wide. This is
wasting a considerable amount of interrupt numbers in the global IRQ
space which has 1M interrupts per socket on a POWER9.
To optimise the HW resources, only request at the KVM level interrupts
which have been claimed by the guest. This will help to increase the
maximum number of VMs per system and also help supporting nested guests
using the XIVE interrupt mode.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
---
hw/intc/spapr_xive_kvm.c | 29 ++++++++++++++++++++++++++---
hw/intc/xics_kvm.c | 8 ++++++++
2 files changed, 34 insertions(+), 3 deletions(-)
diff --git a/hw/intc/spapr_xive_kvm.c b/hw/intc/spapr_xive_kvm.c
index 17af4d19f54e..71b88d7797bc 100644
--- a/hw/intc/spapr_xive_kvm.c
+++ b/hw/intc/spapr_xive_kvm.c
@@ -255,11 +255,16 @@ void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp)
static void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
{
+ SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
int i;
for (i = 0; i < xsrc->nr_irqs; i++) {
Error *local_err = NULL;
+ if (!xive_eas_is_valid(&xive->eat[i])) {
+ continue;
+ }
+
kvmppc_xive_source_reset_one(xsrc, i, &local_err);
if (local_err) {
error_propagate(errp, local_err);
@@ -328,11 +333,18 @@ uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
static void kvmppc_xive_source_get_state(XiveSource *xsrc)
{
+ SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
int i;
for (i = 0; i < xsrc->nr_irqs; i++) {
+ uint8_t pq;
+
+ if (!xive_eas_is_valid(&xive->eat[i])) {
+ continue;
+ }
+
/* Perform a load without side effect to retrieve the PQ bits */
- uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
+ pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
/* and save PQ locally */
xive_source_esb_set(xsrc, i, pq);
@@ -521,9 +533,14 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
*/
if (running) {
for (i = 0; i < xsrc->nr_irqs; i++) {
- uint8_t pq = xive_source_esb_get(xsrc, i);
+ uint8_t pq;
uint8_t old_pq;
+ if (!xive_eas_is_valid(&xive->eat[i])) {
+ continue;
+ }
+
+ pq = xive_source_esb_get(xsrc, i);
old_pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8));
/*
@@ -545,7 +562,13 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
* migration is in progress.
*/
for (i = 0; i < xsrc->nr_irqs; i++) {
- uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
+ uint8_t pq;
+
+ if (!xive_eas_is_valid(&xive->eat[i])) {
+ continue;
+ }
+
+ pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
/*
* PQ is set to PENDING to possibly catch a triggered
diff --git a/hw/intc/xics_kvm.c b/hw/intc/xics_kvm.c
index a4d2e876cc5f..ba90d6dc966c 100644
--- a/hw/intc/xics_kvm.c
+++ b/hw/intc/xics_kvm.c
@@ -190,6 +190,10 @@ void ics_get_kvm_state(ICSState *ics)
for (i = 0; i < ics->nr_irqs; i++) {
ICSIRQState *irq = &ics->irqs[i];
+ if (ics_irq_free(ics, i)) {
+ continue;
+ }
+
kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
i + ics->offset, &state, false, &error_fatal);
@@ -301,6 +305,10 @@ int ics_set_kvm_state(ICSState *ics, Error **errp)
Error *local_err = NULL;
int ret;
+ if (ics_irq_free(ics, i)) {
+ continue;
+ }
+
ret = ics_set_kvm_state_one(ics, i, &local_err);
if (ret < 0) {
error_propagate(errp, local_err);
--
2.21.0
On Wed, Sep 11, 2019 at 03:39:37PM +0200, Cédric Le Goater wrote:
> A typical pseries VM with 16 vCPUs, one disk, one network adapater
> uses less than 100 interrupts but the whole IRQ number space of the
> QEMU machine is allocated at reset time and it is 8K wide. This is
> wasting a considerable amount of interrupt numbers in the global IRQ
> space which has 1M interrupts per socket on a POWER9.
>
> To optimise the HW resources, only request at the KVM level interrupts
> which have been claimed by the guest. This will help to increase the
> maximum number of VMs per system and also help supporting nested guests
> using the XIVE interrupt mode.
>
> Signed-off-by: Cédric Le Goater <clg@kaod.org>
Applied to ppc-for-4.2, thanks.
> ---
> hw/intc/spapr_xive_kvm.c | 29 ++++++++++++++++++++++++++---
> hw/intc/xics_kvm.c | 8 ++++++++
> 2 files changed, 34 insertions(+), 3 deletions(-)
>
> diff --git a/hw/intc/spapr_xive_kvm.c b/hw/intc/spapr_xive_kvm.c
> index 17af4d19f54e..71b88d7797bc 100644
> --- a/hw/intc/spapr_xive_kvm.c
> +++ b/hw/intc/spapr_xive_kvm.c
> @@ -255,11 +255,16 @@ void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp)
>
> static void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
> {
> + SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
> int i;
>
> for (i = 0; i < xsrc->nr_irqs; i++) {
> Error *local_err = NULL;
>
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> kvmppc_xive_source_reset_one(xsrc, i, &local_err);
> if (local_err) {
> error_propagate(errp, local_err);
> @@ -328,11 +333,18 @@ uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
>
> static void kvmppc_xive_source_get_state(XiveSource *xsrc)
> {
> + SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
> int i;
>
> for (i = 0; i < xsrc->nr_irqs; i++) {
> + uint8_t pq;
> +
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> /* Perform a load without side effect to retrieve the PQ bits */
> - uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> + pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
>
> /* and save PQ locally */
> xive_source_esb_set(xsrc, i, pq);
> @@ -521,9 +533,14 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
> */
> if (running) {
> for (i = 0; i < xsrc->nr_irqs; i++) {
> - uint8_t pq = xive_source_esb_get(xsrc, i);
> + uint8_t pq;
> uint8_t old_pq;
>
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> + pq = xive_source_esb_get(xsrc, i);
> old_pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8));
>
> /*
> @@ -545,7 +562,13 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
> * migration is in progress.
> */
> for (i = 0; i < xsrc->nr_irqs; i++) {
> - uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> + uint8_t pq;
> +
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> + pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
>
> /*
> * PQ is set to PENDING to possibly catch a triggered
> diff --git a/hw/intc/xics_kvm.c b/hw/intc/xics_kvm.c
> index a4d2e876cc5f..ba90d6dc966c 100644
> --- a/hw/intc/xics_kvm.c
> +++ b/hw/intc/xics_kvm.c
> @@ -190,6 +190,10 @@ void ics_get_kvm_state(ICSState *ics)
> for (i = 0; i < ics->nr_irqs; i++) {
> ICSIRQState *irq = &ics->irqs[i];
>
> + if (ics_irq_free(ics, i)) {
> + continue;
> + }
> +
> kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
> i + ics->offset, &state, false, &error_fatal);
>
> @@ -301,6 +305,10 @@ int ics_set_kvm_state(ICSState *ics, Error **errp)
> Error *local_err = NULL;
> int ret;
>
> + if (ics_irq_free(ics, i)) {
> + continue;
> + }
> +
> ret = ics_set_kvm_state_one(ics, i, &local_err);
> if (ret < 0) {
> error_propagate(errp, local_err);
--
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
On Mon, 16 Sep 2019 10:44:32 +1000
David Gibson <david@gibson.dropbear.id.au> wrote:
> On Wed, Sep 11, 2019 at 03:39:37PM +0200, Cédric Le Goater wrote:
> > A typical pseries VM with 16 vCPUs, one disk, one network adapater
> > uses less than 100 interrupts but the whole IRQ number space of the
> > QEMU machine is allocated at reset time and it is 8K wide. This is
> > wasting a considerable amount of interrupt numbers in the global IRQ
> > space which has 1M interrupts per socket on a POWER9.
> >
> > To optimise the HW resources, only request at the KVM level interrupts
> > which have been claimed by the guest. This will help to increase the
> > maximum number of VMs per system and also help supporting nested guests
> > using the XIVE interrupt mode.
> >
> > Signed-off-by: Cédric Le Goater <clg@kaod.org>
>
> Applied to ppc-for-4.2, thanks.
>
While experimenting 4.1->4.2 migration with your irq cleanup series,
I've hit this:
qemu-system-ppc64: KVM_SET_DEVICE_ATTR failed: Group 3 attr 0x0000000000001300: Invalid argument
qemu-system-ppc64: error while loading state for instance 0x0 of device 'spapr'
qemu-system-ppc64: load of migration failed: Operation not permitted
Failing to restore the source config with EINVAL seems to come from the
following check in kvmppc_xive_native_set_source_config():
if (!state->valid)
return -EINVAL;
which makes sense since we haven't requested any interrupt yet.
We should hence do it at post load before restoring the source
config.
I'll send a patch ASAP.
> > ---
> > hw/intc/spapr_xive_kvm.c | 29 ++++++++++++++++++++++++++---
> > hw/intc/xics_kvm.c | 8 ++++++++
> > 2 files changed, 34 insertions(+), 3 deletions(-)
> >
> > diff --git a/hw/intc/spapr_xive_kvm.c b/hw/intc/spapr_xive_kvm.c
> > index 17af4d19f54e..71b88d7797bc 100644
> > --- a/hw/intc/spapr_xive_kvm.c
> > +++ b/hw/intc/spapr_xive_kvm.c
> > @@ -255,11 +255,16 @@ void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp)
> >
> > static void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
> > {
> > + SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
> > int i;
> >
> > for (i = 0; i < xsrc->nr_irqs; i++) {
> > Error *local_err = NULL;
> >
> > + if (!xive_eas_is_valid(&xive->eat[i])) {
> > + continue;
> > + }
> > +
> > kvmppc_xive_source_reset_one(xsrc, i, &local_err);
> > if (local_err) {
> > error_propagate(errp, local_err);
> > @@ -328,11 +333,18 @@ uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
> >
> > static void kvmppc_xive_source_get_state(XiveSource *xsrc)
> > {
> > + SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
> > int i;
> >
> > for (i = 0; i < xsrc->nr_irqs; i++) {
> > + uint8_t pq;
> > +
> > + if (!xive_eas_is_valid(&xive->eat[i])) {
> > + continue;
> > + }
> > +
> > /* Perform a load without side effect to retrieve the PQ bits */
> > - uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> > + pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> >
> > /* and save PQ locally */
> > xive_source_esb_set(xsrc, i, pq);
> > @@ -521,9 +533,14 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
> > */
> > if (running) {
> > for (i = 0; i < xsrc->nr_irqs; i++) {
> > - uint8_t pq = xive_source_esb_get(xsrc, i);
> > + uint8_t pq;
> > uint8_t old_pq;
> >
> > + if (!xive_eas_is_valid(&xive->eat[i])) {
> > + continue;
> > + }
> > +
> > + pq = xive_source_esb_get(xsrc, i);
> > old_pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8));
> >
> > /*
> > @@ -545,7 +562,13 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
> > * migration is in progress.
> > */
> > for (i = 0; i < xsrc->nr_irqs; i++) {
> > - uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> > + uint8_t pq;
> > +
> > + if (!xive_eas_is_valid(&xive->eat[i])) {
> > + continue;
> > + }
> > +
> > + pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> >
> > /*
> > * PQ is set to PENDING to possibly catch a triggered
> > diff --git a/hw/intc/xics_kvm.c b/hw/intc/xics_kvm.c
> > index a4d2e876cc5f..ba90d6dc966c 100644
> > --- a/hw/intc/xics_kvm.c
> > +++ b/hw/intc/xics_kvm.c
> > @@ -190,6 +190,10 @@ void ics_get_kvm_state(ICSState *ics)
> > for (i = 0; i < ics->nr_irqs; i++) {
> > ICSIRQState *irq = &ics->irqs[i];
> >
> > + if (ics_irq_free(ics, i)) {
> > + continue;
> > + }
> > +
> > kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
> > i + ics->offset, &state, false, &error_fatal);
> >
> > @@ -301,6 +305,10 @@ int ics_set_kvm_state(ICSState *ics, Error **errp)
> > Error *local_err = NULL;
> > int ret;
> >
> > + if (ics_irq_free(ics, i)) {
> > + continue;
> > + }
> > +
> > ret = ics_set_kvm_state_one(ics, i, &local_err);
> > if (ret < 0) {
> > error_propagate(errp, local_err);
>
On Wed, 11 Sep 2019 15:39:37 +0200
Cédric Le Goater <clg@kaod.org> wrote:
> A typical pseries VM with 16 vCPUs, one disk, one network adapater
> uses less than 100 interrupts but the whole IRQ number space of the
> QEMU machine is allocated at reset time and it is 8K wide. This is
> wasting a considerable amount of interrupt numbers in the global IRQ
> space which has 1M interrupts per socket on a POWER9.
>
> To optimise the HW resources, only request at the KVM level interrupts
> which have been claimed by the guest. This will help to increase the
> maximum number of VMs per system and also help supporting nested guests
> using the XIVE interrupt mode.
>
> Signed-off-by: Cédric Le Goater <clg@kaod.org>
> ---
Reviewed-by: Greg Kurz <groug@kaod.org>
> hw/intc/spapr_xive_kvm.c | 29 ++++++++++++++++++++++++++---
> hw/intc/xics_kvm.c | 8 ++++++++
> 2 files changed, 34 insertions(+), 3 deletions(-)
>
> diff --git a/hw/intc/spapr_xive_kvm.c b/hw/intc/spapr_xive_kvm.c
> index 17af4d19f54e..71b88d7797bc 100644
> --- a/hw/intc/spapr_xive_kvm.c
> +++ b/hw/intc/spapr_xive_kvm.c
> @@ -255,11 +255,16 @@ void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp)
>
> static void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
> {
> + SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
> int i;
>
> for (i = 0; i < xsrc->nr_irqs; i++) {
> Error *local_err = NULL;
>
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> kvmppc_xive_source_reset_one(xsrc, i, &local_err);
> if (local_err) {
> error_propagate(errp, local_err);
> @@ -328,11 +333,18 @@ uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
>
> static void kvmppc_xive_source_get_state(XiveSource *xsrc)
> {
> + SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
> int i;
>
> for (i = 0; i < xsrc->nr_irqs; i++) {
> + uint8_t pq;
> +
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> /* Perform a load without side effect to retrieve the PQ bits */
> - uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> + pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
>
> /* and save PQ locally */
> xive_source_esb_set(xsrc, i, pq);
> @@ -521,9 +533,14 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
> */
> if (running) {
> for (i = 0; i < xsrc->nr_irqs; i++) {
> - uint8_t pq = xive_source_esb_get(xsrc, i);
> + uint8_t pq;
> uint8_t old_pq;
>
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> + pq = xive_source_esb_get(xsrc, i);
> old_pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8));
>
> /*
> @@ -545,7 +562,13 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
> * migration is in progress.
> */
> for (i = 0; i < xsrc->nr_irqs; i++) {
> - uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
> + uint8_t pq;
> +
> + if (!xive_eas_is_valid(&xive->eat[i])) {
> + continue;
> + }
> +
> + pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
>
> /*
> * PQ is set to PENDING to possibly catch a triggered
> diff --git a/hw/intc/xics_kvm.c b/hw/intc/xics_kvm.c
> index a4d2e876cc5f..ba90d6dc966c 100644
> --- a/hw/intc/xics_kvm.c
> +++ b/hw/intc/xics_kvm.c
> @@ -190,6 +190,10 @@ void ics_get_kvm_state(ICSState *ics)
> for (i = 0; i < ics->nr_irqs; i++) {
> ICSIRQState *irq = &ics->irqs[i];
>
> + if (ics_irq_free(ics, i)) {
> + continue;
> + }
> +
> kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
> i + ics->offset, &state, false, &error_fatal);
>
> @@ -301,6 +305,10 @@ int ics_set_kvm_state(ICSState *ics, Error **errp)
> Error *local_err = NULL;
> int ret;
>
> + if (ics_irq_free(ics, i)) {
> + continue;
> + }
> +
> ret = ics_set_kvm_state_one(ics, i, &local_err);
> if (ret < 0) {
> error_propagate(errp, local_err);
© 2016 - 2025 Red Hat, Inc.