|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-devel] [PATCH v9 for Xen 4.7 1/4] xen: enable per-VCPU parameter for RTDS
Add XEN_DOMCTL_SCHEDOP_getvcpuinfo and _putvcpuinfo hypercalls
to independently get and set the scheduling parameters of each
vCPU of a domain.
Also fix a bug in XEN_DOMCTL_SCHEDOP_getinfo, where PERIOD and
BUDGET are not divided by MICROSECS(1) before being retruned
to the caller.
Signed-off-by: Chong Li <chong.li@xxxxxxxxx>
Signed-off-by: Meng Xu <mengxu@xxxxxxxxxxxxx>
Signed-off-by: Sisu Xi <xisisu@xxxxxxxxx>
---
Changes on PATCH v8:
1) Move BUG fix description before SoB.
2) Remove the unuseful padding in struct xen_domctl_schedparam_vcpu.
Changes on PATCH v7:
1) A bug in case XEN_DOMCTL_SCHEDOP_getinfo (in Xen 4.6) is fixed:
The default PERIOD or BUDGET should be divided by MICROSECS(1),
before returned to upper caller.
2) In the *cntl() functions in sched_credit.c and sched_credit2.c, remove
two unhandled cases in "switch".
Changes on PATCH v6:
1) Add explain for nr_vcpus in struct xen_domctl_scheduler_op, because it is
used
in both IN and OUT ways.
2) Remove the check and warning for vcpu settings with low budget or budget.
Making this
feature "per-domain" or "per-operation" is one of the future work.
3) In the *cntl() functions in sched_credit.c and sched_credit2.c, change the
"if-else"
structure to "switch" structure.
4) In rt_dom_cntl(), use copy_to_guest* instead of __copy_to_guest*, because
the latter one
requires lock protection.
Changes on PATCH v5:
1) When processing XEN_DOMCTL_SCHEDOP_get/putvcpuinfo, we do
preemption check in a similar way to XEN_SYSCTL_pcitopoinfo
Changes on PATCH v4:
1) Add uint32_t vcpu_index to struct xen_domctl_scheduler_op.
When processing XEN_DOMCTL_SCHEDOP_get/putvcpuinfo, we call
hypercall_preemption_check in case the current hypercall lasts
too long. If we decide to preempt the current hypercall, we record
the index of the most-recent finished vcpu into the vcpu_index of
struct xen_domctl_scheduler_op. So when we resume the hypercall after
preemption, we start processing from the posion specified by vcpu_index,
and don't need to repeat the work that has already been done in the
hypercall before the preemption.
(This design is based on the do_grant_table_op() in grant_table.c)
2) Coding style changes
Changes on PATCH v3:
1) Remove struct xen_domctl_schedparam_t.
2) Change struct xen_domctl_scheduler_op.
3) Check if period/budget is within a validated range
Changes on PATCH v2:
1) Change struct xen_domctl_scheduler_op, for transferring per-vcpu parameters
between libxc and hypervisor.
2) Handler of XEN_DOMCTL_SCHEDOP_getinfo now just returns the default budget
and
period values of RTDS scheduler.
3) Handler of XEN_DOMCTL_SCHEDOP_getvcpuinfo now can return a random subset of
the parameters of the VCPUs of a specific domain
CC: <dario.faggioli@xxxxxxxxxx>
CC: <george.dunlap@xxxxxxxxxxxxx>
CC: <dgolomb@xxxxxxxxxxxxxx>
CC: <mengxu@xxxxxxxxxxxxx>
CC: <jbeulich@xxxxxxxx>
CC: <lichong659@xxxxxxxxx>
---
xen/common/sched_credit.c | 14 +++----
xen/common/sched_credit2.c | 13 ++++---
xen/common/sched_rt.c | 93 ++++++++++++++++++++++++++++++++++++++-------
xen/common/schedule.c | 15 ++++++--
xen/include/public/domctl.h | 63 ++++++++++++++++++++++++------
5 files changed, 156 insertions(+), 42 deletions(-)
diff --git a/xen/common/sched_credit.c b/xen/common/sched_credit.c
index 305889a..e5d15d8 100644
--- a/xen/common/sched_credit.c
+++ b/xen/common/sched_credit.c
@@ -1080,15 +1080,13 @@ csched_dom_cntl(
* lock. Runq lock not needed anywhere in here. */
spin_lock_irqsave(&prv->lock, flags);
- if ( op->cmd == XEN_DOMCTL_SCHEDOP_getinfo )
+ switch ( op->cmd )
{
+ case XEN_DOMCTL_SCHEDOP_getinfo:
op->u.credit.weight = sdom->weight;
op->u.credit.cap = sdom->cap;
- }
- else
- {
- ASSERT(op->cmd == XEN_DOMCTL_SCHEDOP_putinfo);
-
+ break;
+ case XEN_DOMCTL_SCHEDOP_putinfo:
if ( op->u.credit.weight != 0 )
{
if ( !list_empty(&sdom->active_sdom_elem) )
@@ -1101,7 +1099,9 @@ csched_dom_cntl(
if ( op->u.credit.cap != (uint16_t)~0U )
sdom->cap = op->u.credit.cap;
-
+ break;
+ default:
+ return -EINVAL;
}
spin_unlock_irqrestore(&prv->lock, flags);
diff --git a/xen/common/sched_credit2.c b/xen/common/sched_credit2.c
index 7ddad38..d48ed5a 100644
--- a/xen/common/sched_credit2.c
+++ b/xen/common/sched_credit2.c
@@ -1421,14 +1421,12 @@ csched2_dom_cntl(
* runq lock to update csvcs. */
spin_lock_irqsave(&prv->lock, flags);
- if ( op->cmd == XEN_DOMCTL_SCHEDOP_getinfo )
+ switch ( op->cmd )
{
+ case XEN_DOMCTL_SCHEDOP_getinfo:
op->u.credit2.weight = sdom->weight;
- }
- else
- {
- ASSERT(op->cmd == XEN_DOMCTL_SCHEDOP_putinfo);
-
+ break;
+ case XEN_DOMCTL_SCHEDOP_putinfo:
if ( op->u.credit2.weight != 0 )
{
struct vcpu *v;
@@ -1457,6 +1455,9 @@ csched2_dom_cntl(
vcpu_schedule_unlock(lock, svc->vcpu);
}
}
+ break;
+ default:
+ return -EINVAL;
}
spin_unlock_irqrestore(&prv->lock, flags);
diff --git a/xen/common/sched_rt.c b/xen/common/sched_rt.c
index d98bfb6..321b0a5 100644
--- a/xen/common/sched_rt.c
+++ b/xen/common/sched_rt.c
@@ -90,6 +90,22 @@
#define RTDS_DEFAULT_PERIOD (MICROSECS(10000))
#define RTDS_DEFAULT_BUDGET (MICROSECS(4000))
+/*
+ * Max period: max delta of time type, because period is added to the time
+ * a vcpu activates, so this must not overflow.
+ * Min period: 10 us, considering the scheduling overhead (when period is
+ * too low, scheduling is invoked too frequently, causing high overhead).
+ */
+#define RTDS_MAX_PERIOD (STIME_DELTA_MAX)
+#define RTDS_MIN_PERIOD (MICROSECS(10))
+
+/*
+ * Min budget: 10 us, considering the scheduling overhead (when budget is
+ * consumed too fast, scheduling is invoked too frequently, causing
+ * high overhead).
+ */
+#define RTDS_MIN_BUDGET (MICROSECS(10))
+
#define UPDATE_LIMIT_SHIFT 10
/*
@@ -1266,24 +1282,16 @@ rt_dom_cntl(
struct vcpu *v;
unsigned long flags;
int rc = 0;
+ xen_domctl_schedparam_vcpu_t local_sched;
+ s_time_t period, budget;
+ uint32_t index = 0;
switch ( op->cmd )
{
case XEN_DOMCTL_SCHEDOP_getinfo:
- if ( d->max_vcpus > 0 )
- {
- spin_lock_irqsave(&prv->lock, flags);
- svc = rt_vcpu(d->vcpu[0]);
- op->u.rtds.period = svc->period / MICROSECS(1);
- op->u.rtds.budget = svc->budget / MICROSECS(1);
- spin_unlock_irqrestore(&prv->lock, flags);
- }
- else
- {
- /* If we don't have vcpus yet, let's just return the defaults. */
- op->u.rtds.period = RTDS_DEFAULT_PERIOD;
- op->u.rtds.budget = RTDS_DEFAULT_BUDGET;
- }
+ /* Return the default parameters. */
+ op->u.rtds.period = RTDS_DEFAULT_PERIOD / MICROSECS(1);
+ op->u.rtds.budget = RTDS_DEFAULT_BUDGET / MICROSECS(1);
break;
case XEN_DOMCTL_SCHEDOP_putinfo:
if ( op->u.rtds.period == 0 || op->u.rtds.budget == 0 )
@@ -1300,6 +1308,63 @@ rt_dom_cntl(
}
spin_unlock_irqrestore(&prv->lock, flags);
break;
+ case XEN_DOMCTL_SCHEDOP_getvcpuinfo:
+ case XEN_DOMCTL_SCHEDOP_putvcpuinfo:
+ while ( index < op->u.v.nr_vcpus )
+ {
+ if ( copy_from_guest_offset(&local_sched,
+ op->u.v.vcpus, index, 1) )
+ {
+ rc = -EFAULT;
+ break;
+ }
+ if ( local_sched.vcpuid >= d->max_vcpus ||
+ d->vcpu[local_sched.vcpuid] == NULL )
+ {
+ rc = -EINVAL;
+ break;
+ }
+
+ if ( op->cmd == XEN_DOMCTL_SCHEDOP_getvcpuinfo )
+ {
+ spin_lock_irqsave(&prv->lock, flags);
+ svc = rt_vcpu(d->vcpu[local_sched.vcpuid]);
+ local_sched.u.rtds.budget = svc->budget / MICROSECS(1);
+ local_sched.u.rtds.period = svc->period / MICROSECS(1);
+ spin_unlock_irqrestore(&prv->lock, flags);
+
+ if ( copy_to_guest_offset(op->u.v.vcpus, index,
+ &local_sched, 1) )
+ {
+ rc = -EFAULT;
+ break;
+ }
+ }
+ else
+ {
+ period = MICROSECS(local_sched.u.rtds.period);
+ budget = MICROSECS(local_sched.u.rtds.budget);
+ if ( period > RTDS_MAX_PERIOD || budget < RTDS_MIN_BUDGET ||
+ budget > period || period < RTDS_MIN_PERIOD )
+ {
+ rc = -EINVAL;
+ break;
+ }
+
+ spin_lock_irqsave(&prv->lock, flags);
+ svc = rt_vcpu(d->vcpu[local_sched.vcpuid]);
+ svc->period = period;
+ svc->budget = budget;
+ spin_unlock_irqrestore(&prv->lock, flags);
+ }
+ /* Process a most 64 vCPUs without checking for preemptions. */
+ if ( (++index > 63) && hypercall_preempt_check() )
+ break;
+ }
+ if ( !rc )
+ /* notify upper caller how many vcpus have been processed. */
+ op->u.v.nr_vcpus = index;
+ break;
}
return rc;
diff --git a/xen/common/schedule.c b/xen/common/schedule.c
index 0627eb5..b7dee16 100644
--- a/xen/common/schedule.c
+++ b/xen/common/schedule.c
@@ -1235,11 +1235,20 @@ long sched_adjust(struct domain *d, struct
xen_domctl_scheduler_op *op)
if ( ret )
return ret;
- if ( (op->sched_id != DOM2OP(d)->sched_id) ||
- ((op->cmd != XEN_DOMCTL_SCHEDOP_putinfo) &&
- (op->cmd != XEN_DOMCTL_SCHEDOP_getinfo)) )
+ if ( op->sched_id != DOM2OP(d)->sched_id )
return -EINVAL;
+ switch ( op->cmd )
+ {
+ case XEN_DOMCTL_SCHEDOP_putinfo:
+ case XEN_DOMCTL_SCHEDOP_getinfo:
+ case XEN_DOMCTL_SCHEDOP_putvcpuinfo:
+ case XEN_DOMCTL_SCHEDOP_getvcpuinfo:
+ break;
+ default:
+ return -EINVAL;
+ }
+
/* NB: the pluggable scheduler code needs to take care
* of locking by itself. */
if ( (ret = SCHED_OP(DOM2OP(d), adjust, d, op)) == 0 )
diff --git a/xen/include/public/domctl.h b/xen/include/public/domctl.h
index 076b1ae..2457698 100644
--- a/xen/include/public/domctl.h
+++ b/xen/include/public/domctl.h
@@ -350,24 +350,63 @@ DEFINE_XEN_GUEST_HANDLE(xen_domctl_max_vcpus_t);
#define XEN_SCHEDULER_ARINC653 7
#define XEN_SCHEDULER_RTDS 8
-/* Set or get info? */
+typedef struct xen_domctl_sched_credit {
+ uint16_t weight;
+ uint16_t cap;
+} xen_domctl_sched_credit_t;
+
+typedef struct xen_domctl_sched_credit2 {
+ uint16_t weight;
+} xen_domctl_sched_credit2_t;
+
+typedef struct xen_domctl_sched_rtds {
+ uint32_t period;
+ uint32_t budget;
+} xen_domctl_sched_rtds_t;
+
+typedef struct xen_domctl_schedparam_vcpu {
+ union {
+ xen_domctl_sched_credit_t credit;
+ xen_domctl_sched_credit2_t credit2;
+ xen_domctl_sched_rtds_t rtds;
+ } u;
+ uint32_t vcpuid;
+} xen_domctl_schedparam_vcpu_t;
+DEFINE_XEN_GUEST_HANDLE(xen_domctl_schedparam_vcpu_t);
+
+/*
+ * Set or get info?
+ * For schedulers supporting per-vcpu settings (e.g., RTDS):
+ * XEN_DOMCTL_SCHEDOP_putinfo sets params for all vcpus;
+ * XEN_DOMCTL_SCHEDOP_getinfo gets default params;
+ * XEN_DOMCTL_SCHEDOP_put(get)vcpuinfo sets (gets) params of vcpus;
+ *
+ * For schedulers not supporting per-vcpu settings:
+ * XEN_DOMCTL_SCHEDOP_putinfo sets params for all vcpus;
+ * XEN_DOMCTL_SCHEDOP_getinfo gets domain-wise params;
+ * XEN_DOMCTL_SCHEDOP_put(get)vcpuinfo returns error;
+ */
#define XEN_DOMCTL_SCHEDOP_putinfo 0
#define XEN_DOMCTL_SCHEDOP_getinfo 1
+#define XEN_DOMCTL_SCHEDOP_putvcpuinfo 2
+#define XEN_DOMCTL_SCHEDOP_getvcpuinfo 3
struct xen_domctl_scheduler_op {
uint32_t sched_id; /* XEN_SCHEDULER_* */
uint32_t cmd; /* XEN_DOMCTL_SCHEDOP_* */
+ /* IN/OUT */
union {
- struct xen_domctl_sched_credit {
- uint16_t weight;
- uint16_t cap;
- } credit;
- struct xen_domctl_sched_credit2 {
- uint16_t weight;
- } credit2;
- struct xen_domctl_sched_rtds {
- uint32_t period;
- uint32_t budget;
- } rtds;
+ xen_domctl_sched_credit_t credit;
+ xen_domctl_sched_credit2_t credit2;
+ xen_domctl_sched_rtds_t rtds;
+ struct {
+ XEN_GUEST_HANDLE_64(xen_domctl_schedparam_vcpu_t) vcpus;
+ /*
+ * IN: Number of elements in vcpus array.
+ * OUT: Number of processed elements of vcpus array.
+ */
+ uint32_t nr_vcpus;
+ uint32_t padding;
+ } v;
} u;
};
typedef struct xen_domctl_scheduler_op xen_domctl_scheduler_op_t;
--
1.9.1
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel
|
 |
Lists.xenproject.org is hosted with RackSpace, monitoring our |