Skip to contentSkip to conversation

xen: cpupool support - hypervisor support of cpupools

PATCH In review Juergen Gross 1 participant 1 message Apr 17, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/6 xen: cpupool support - hypervisor support of cpupools

Signed-off-by [email protected]
Signature · 5 lines
Juergen Gross                 Principal Developer Operating Systems
TSP ES&S SWE OS6                       Telephone: +49 (0) 89 636 47950
Fujitsu Technolgy Solutions               e-mail: [email protected]
Otto-Hahn-Ring 6                        Internet: ts.fujitsu.com
D-81739 Muenchen                 Company details: ts.fujitsu.com/imprint.html
xen/arch/x86/acpi/cpu_idle.c +5 -5
diff -r 655dc3bc1d8e xen/arch/x86/acpi/cpu_idle.c
@@ -198,7 +198,7 @@ static void acpi_processor_idle(void)
 
     cpufreq_dbs_timer_suspend();
 
-    sched_tick_suspend();
+    sched_tick_suspend(smp_processor_id());
     /*
      * sched_tick_suspend may raise TIMER_SOFTIRQ by __stop_timer,
      * which will break the later assumption of no sofirq pending,
@@ -216,7 +216,7 @@ static void acpi_processor_idle(void)
     if ( softirq_pending(smp_processor_id()) )
     {
         local_irq_enable();
-        sched_tick_resume();
+        sched_tick_resume(smp_processor_id());
         cpufreq_dbs_timer_resume();
         return;
     }
@@ -237,7 +237,7 @@ static void acpi_processor_idle(void)
             pm_idle_save();
         else
             acpi_safe_halt();
-        sched_tick_resume();
+        sched_tick_resume(smp_processor_id());
         cpufreq_dbs_timer_resume();
         return;
     }
@@ -345,7 +345,7 @@ static void acpi_processor_idle(void)
 
     default:
         local_irq_enable();
-        sched_tick_resume();
+        sched_tick_resume(smp_processor_id());
         cpufreq_dbs_timer_resume();
         return;
     }
@@ -357,7 +357,7 @@ static void acpi_processor_idle(void)
         cx->time += sleep_ticks;
     }
 
-    sched_tick_resume();
+    sched_tick_resume(smp_processor_id());
     cpufreq_dbs_timer_resume();
 
     if ( cpuidle_current_governor->reflect )
xen/arch/x86/domain.c +28 -3
diff -r 655dc3bc1d8e xen/arch/x86/domain.c
@@ -1412,7 +1412,13 @@ struct migrate_info {
     void (*saved_schedule_tail)(struct vcpu *);
     cpumask_t saved_affinity;
     unsigned int nest;
+    int borrowed;
 };
+
+long continue_hypercall_on_cpu_dummy(void *data)
+{
+    return 0;
+}
 
 static void continue_hypercall_on_cpu_helper(struct vcpu *v)
 {
@@ -1420,8 +1426,16 @@ static void continue_hypercall_on_cpu_he
     struct migrate_info *info = v->arch.continue_info;
     cpumask_t mask = info->saved_affinity;
     void (*saved_schedule_tail)(struct vcpu *) = info->saved_schedule_tail;
+    int cpu = -1;
 
     regs->eax = info->func(info->data);
+
+    if ( (info->nest == 0) && info->borrowed &&
+         (cpu = cpupool_return_cpu(v->domain->cpupool) >= 0) )
+    {
+        continue_hypercall_on_cpu(cpu, continue_hypercall_on_cpu_dummy,
+            info->data);
+    }
 
     if ( info->nest-- == 0 )
     {
@@ -1440,27 +1454,32 @@ int continue_hypercall_on_cpu(int cpu, l
     struct migrate_info *info;
     cpumask_t mask = cpumask_of_cpu(cpu);
     int rc;
+    int borrowed = 0;
 
     if ( cpu == smp_processor_id() )
         return func(data);
 
+    borrowed = cpupool_borrow_cpu(v->domain->cpupool, cpu);
+
     info = v->arch.continue_info;
     if ( info == NULL )
     {
         info = xmalloc(struct migrate_info);
+        rc = -ENOMEM;
         if ( info == NULL )
-            return -ENOMEM;
+            goto out;
 
         rc = vcpu_lock_affinity(v, &mask);
         if ( rc )
         {
             xfree(info);
-            return rc;
+            goto out;
         }
 
         info->saved_schedule_tail = v->arch.schedule_tail;
         info->saved_affinity = mask;
         info->nest = 0;
+        info->borrowed = 0;
 
         v->arch.schedule_tail = continue_hypercall_on_cpu_helper;
         v->arch.continue_info = info;
@@ -1470,16 +1489,22 @@ int continue_hypercall_on_cpu(int cpu, l
         BUG_ON(info->nest != 0);
         rc = vcpu_locked_change_affinity(v, &mask);
         if ( rc )
-            return rc;
+            goto out;
         info->nest++;
     }
 
+    info->borrowed += borrowed;
     info->func = func;
     info->data = data;
 
     /* Dummy return value will be overwritten by new schedule_tail. */
     BUG_ON(!test_bit(SCHEDULE_SOFTIRQ, &softirq_pending(smp_processor_id())));
     return 0;
+
+out:
+    if ( borrowed )
+        cpupool_return_cpu(v->domain->cpupool);
+    return rc;
 }
 
 #define next_arg(fmt, args) ({                                              \
xen/arch/x86/domain_build.c +3 -2
diff -r 655dc3bc1d8e xen/arch/x86/domain_build.c
@@ -9,6 +9,7 @@
 #include <xen/lib.h>
 #include <xen/ctype.h>
 #include <xen/sched.h>
+#include <xen/sched-if.h>
 #include <xen/smp.h>
 #include <xen/delay.h>
 #include <xen/event.h>
@@ -706,13 +707,13 @@ int __init construct_dom0(
         shared_info(d, vcpu_info[i].evtchn_upcall_mask) = 1;
 
     if ( opt_dom0_max_vcpus == 0 )
-        opt_dom0_max_vcpus = num_online_cpus();
+        opt_dom0_max_vcpus = num_cpupool_cpus(cpupool0);
     if ( opt_dom0_max_vcpus > MAX_VIRT_CPUS )
         opt_dom0_max_vcpus = MAX_VIRT_CPUS;
     printk("Dom0 has maximum %u VCPUs\n", opt_dom0_max_vcpus);
 
     for ( i = 1; i < opt_dom0_max_vcpus; i++ )
-        (void)alloc_vcpu(d, i, i % num_online_cpus());
+        (void)alloc_vcpu(d, i, i % num_cpupool_cpus(cpupool0));
 
     /* Set up CR3 value for write_ptbase */
     if ( paging_mode_enabled(d) )
xen/arch/x86/mm.c +2 -2
diff -r 655dc3bc1d8e xen/arch/x86/mm.c
@@ -212,7 +212,7 @@ void __init arch_init_memory(void)
      * Any Xen-heap pages that we will allow to be mapped will have
      * their domain field set to dom_xen.
      */
-    dom_xen = domain_create(DOMID_XEN, DOMCRF_dummy, 0);
+    dom_xen = domain_create(DOMID_XEN, CPUPOOLID_NONE, DOMCRF_dummy, 0);
     BUG_ON(dom_xen == NULL);
 
     /*
@@ -220,7 +220,7 @@ void __init arch_init_memory(void)
      * This domain owns I/O pages that are within the range of the page_info
      * array. Mappings occur at the priv of the caller.
      */
-    dom_io = domain_create(DOMID_IO, DOMCRF_dummy, 0);
+    dom_io = domain_create(DOMID_IO, CPUPOOLID_NONE, DOMCRF_dummy, 0);
     BUG_ON(dom_io == NULL);
 
     /* First 1MB of RAM is historically marked as I/O. */
xen/arch/x86/setup.c +7 -2
diff -r 655dc3bc1d8e xen/arch/x86/setup.c
@@ -2,6 +2,7 @@
 #include <xen/init.h>
 #include <xen/lib.h>
 #include <xen/sched.h>
+#include <xen/sched-if.h>
 #include <xen/domain.h>
 #include <xen/serial.h>
 #include <xen/softirq.h>
@@ -232,7 +233,7 @@ static void __init init_idle_domain(void
     /* Domain creation requires that scheduler structures are initialised. */
     scheduler_init();
 
-    idle_domain = domain_create(IDLE_DOMAIN_ID, 0, 0);
+    idle_domain = domain_create(IDLE_DOMAIN_ID, CPUPOOLID_NONE, 0, 0);
     if ( (idle_domain == NULL) || (alloc_vcpu(idle_domain, 0, 0) == NULL) )
         BUG();
 
@@ -995,8 +996,12 @@ void __init __start_xen(unsigned long mb
     if ( !tboot_protect_mem_regions() )
         panic("Could not protect TXT memory regions\n");
 
+    /* Create initial cpupool 0. */
+    cpupool0 = cpupool_create(0, NULL);
+    if ( (cpupool0 == NULL) || cpupool0_cpu_assign(cpupool0) )
+        panic("Error creating cpupool 0\n");
     /* Create initial domain 0. */
-    dom0 = domain_create(0, DOMCRF_s3_integrity, DOM0_SSIDREF);
+    dom0 = domain_create(0, 0, DOMCRF_s3_integrity, DOM0_SSIDREF);
     if ( (dom0 == NULL) || (alloc_vcpu(dom0, 0, 0) == NULL) )
         panic("Error creating domain 0\n");
 
xen/arch/x86/smpboot.c +3 -2
diff -r 655dc3bc1d8e xen/arch/x86/smpboot.c
@@ -1265,7 +1265,7 @@ int __cpu_disable(void)
 	/* It's now safe to remove this processor from the online map */
 	cpu_clear(cpu, cpu_online_map);
 
-	cpu_disable_scheduler();
+	cpu_disable_scheduler(cpu, 0);
 
 	return 0;
 }
@@ -1299,7 +1299,7 @@ int cpu_down(unsigned int cpu)
 	int err = 0;
 
 	spin_lock(&cpu_add_remove_lock);
-	if (num_online_cpus() == 1) {
+	if (cpupool_cpu_remove(cpu)) {
 		err = -EBUSY;
 		goto out;
 	}
@@ -1451,6 +1451,7 @@ int __devinit __cpu_up(unsigned int cpu)
 		process_pending_timers();
 	}
 
+	cpupool_cpu_add(cpu);
 	cpufreq_add_cpu(cpu);
 	return 0;
 }
xen/common/Makefile +1 -0
diff -r 655dc3bc1d8e xen/common/Makefile
@@ -1,4 +1,5 @@ obj-y += bitmap.o
 obj-y += bitmap.o
+obj-y += cpupool.o
 obj-y += domctl.o
 obj-y += domain.o
 obj-y += event_channel.o
xen/common/domain.c +7 -2
diff -r 655dc3bc1d8e xen/common/domain.c
@@ -187,7 +187,7 @@ struct vcpu *alloc_idle_vcpu(unsigned in
         return v;
 
     d = (vcpu_id == 0) ?
-        domain_create(IDLE_DOMAIN_ID, 0, 0) :
+        domain_create(IDLE_DOMAIN_ID, CPUPOOLID_NONE, 0, 0) :
         idle_vcpu[cpu_id - vcpu_id]->domain;
     BUG_ON(d == NULL);
 
@@ -198,7 +198,7 @@ struct vcpu *alloc_idle_vcpu(unsigned in
 }
 
 struct domain *domain_create(
-    domid_t domid, unsigned int domcr_flags, ssidref_t ssidref)
+    domid_t domid, int poolid, unsigned int domcr_flags, ssidref_t ssidref)
 {
     struct domain *d, **pd;
     enum { INIT_xsm = 1u<<0, INIT_rangeset = 1u<<1, INIT_evtchn = 1u<<2,
@@ -259,6 +259,9 @@ struct domain *domain_create(
     d->iomem_caps = rangeset_new(d, "I/O Memory", RANGESETF_prettyprint_hex);
     d->irq_caps   = rangeset_new(d, "Interrupts", 0);
     if ( (d->iomem_caps == NULL) || (d->irq_caps == NULL) )
+        goto fail;
+
+    if ( cpupool_add_domain(d, poolid) != 0 )
         goto fail;
 
     if ( sched_init_domain(d) != 0 )
@@ -564,6 +567,8 @@ static void complete_domain_destroy(stru
 
     sched_destroy_domain(d);
 
+    cpupool_rm_domain(d);
+
     /* Free page used by xen oprofile buffer. */
     free_xenoprof_pages(d);
 
xen/common/domctl.c +24 -6
diff -r 655dc3bc1d8e xen/common/domctl.c
@@ -11,6 +11,7 @@
 #include <xen/lib.h>
 #include <xen/mm.h>
 #include <xen/sched.h>
+#include <xen/sched-if.h>
 #include <xen/domain.h>
 #include <xen/event.h>
 #include <xen/domain_page.h>
@@ -138,15 +139,18 @@ void getdomaininfo(struct domain *d, str
     info->max_pages         = d->max_pages;
     info->shared_info_frame = mfn_to_gmfn(d, __pa(d->shared_info)>>PAGE_SHIFT);
 
+    info->cpupool = d->cpupool ? d->cpupool->cpupool_id : CPUPOOLID_NONE;
+
     memcpy(info->handle, d->handle, sizeof(xen_domain_handle_t));
 }
 
-static unsigned int default_vcpu0_location(void)
+static unsigned int default_vcpu0_location(struct domain *dom)
 {
     struct domain *d;
     struct vcpu   *v;
     unsigned int   i, cpu, nr_cpus, *cnt;
     cpumask_t      cpu_exclude_map;
+    cpumask_t      online;
 
     /* Do an initial CPU placement. Pick the least-populated CPU. */
     nr_cpus = last_cpu(cpu_possible_map) + 1;
@@ -171,7 +175,8 @@ static unsigned int default_vcpu0_locati
     if ( cpus_weight(cpu_sibling_map[0]) > 1 )
         cpu = next_cpu(cpu, cpu_sibling_map[0]);
     cpu_exclude_map = cpu_sibling_map[0];
-    for_each_online_cpu ( i )
+    online = (dom->cpupool == NULL) ? cpu_online_map : dom->cpupool->cpu_valid;
+    for_each_cpu_mask(i, online)
     {
         if ( cpu_isset(i, cpu_exclude_map) )
             continue;
@@ -366,12 +371,13 @@ long do_domctl(XEN_GUEST_HANDLE(xen_domc
         domid_t        dom;
         static domid_t rover = 0;
         unsigned int domcr_flags;
+        int            pool = 0;
 
         ret = -EINVAL;
         if ( supervisor_mode_kernel ||
              (op->u.createdomain.flags &
              ~(XEN_DOMCTL_CDF_hvm_guest | XEN_DOMCTL_CDF_hap |
-               XEN_DOMCTL_CDF_s3_integrity)) )
+               XEN_DOMCTL_CDF_s3_integrity | XEN_DOMCTL_CDF_pool)) )
             break;
 
         dom = op->domain;
@@ -405,9 +411,11 @@ long do_domctl(XEN_GUEST_HANDLE(xen_domc
             domcr_flags |= DOMCRF_hap;
         if ( op->u.createdomain.flags & XEN_DOMCTL_CDF_s3_integrity )
             domcr_flags |= DOMCRF_s3_integrity;
+        if ( op->u.createdomain.flags & XEN_DOMCTL_CDF_pool )
+            pool = op->u.createdomain.cpupool;
 
         ret = -ENOMEM;
-        d = domain_create(dom, domcr_flags, op->u.createdomain.ssidref);
+        d = domain_create(dom, pool, domcr_flags, op->u.createdomain.ssidref);
         if ( d == NULL )
             break;
 
@@ -426,6 +434,7 @@ long do_domctl(XEN_GUEST_HANDLE(xen_domc
     {
         struct domain *d;
         unsigned int i, max = op->u.max_vcpus.max, cpu;
+        cpumask_t online;
 
         ret = -ESRCH;
         if ( (d = rcu_lock_domain_by_id(op->domain)) == NULL )
@@ -455,14 +464,15 @@ long do_domctl(XEN_GUEST_HANDLE(xen_domc
             goto maxvcpu_out;
 
         ret = -ENOMEM;
+        online = (d->cpupool == NULL) ? cpu_online_map : d->cpupool->cpu_valid;
         for ( i = 0; i < max; i++ )
         {
             if ( d->vcpu[i] != NULL )
                 continue;
 
             cpu = (i == 0) ?
-                default_vcpu0_location() :
-                cycle_cpu(d->vcpu[i-1]->processor, cpu_online_map);
+                default_vcpu0_location(d) :
+                cycle_cpu(d->vcpu[i-1]->processor, online);
 
             if ( alloc_vcpu(d, i, cpu) == NULL )
                 goto maxvcpu_out;
@@ -890,6 +900,14 @@ long do_domctl(XEN_GUEST_HANDLE(xen_domc
     }
     break;
 
+    case XEN_DOMCTL_cpupool_op:
+    {
+        ret = cpupool_do_domctl(op);
+        if ( (ret == 0) && copy_to_guest(u_domctl, op, 1) )
+            ret = -EFAULT;
+    }
+    break;
+
     default:
         ret = arch_do_domctl(op, u_domctl);
         break;
xen/common/sched_credit.c +295 -136

853 lines not on this page · raw message

xen/common/sched_sedf.c +99 -43

245 lines not on this page · raw message

xen/common/schedule.c +222 -62

433 lines not on this page · raw message

xen/include/public/domctl.h +30 -0
diff -r 655dc3bc1d8e xen/include/public/domctl.h
@@ -59,7 +59,11 @@ struct xen_domctl_createdomain {
  /* Should domain memory integrity be verifed by tboot during Sx? */
 #define _XEN_DOMCTL_CDF_s3_integrity  2
 #define XEN_DOMCTL_CDF_s3_integrity   (1U<<_XEN_DOMCTL_CDF_s3_integrity)
+ /* cpupool is specified (0 otherwise) */
+#define _XEN_DOMCTL_CDF_pool          3
+#define XEN_DOMCTL_CDF_pool           (1U<<_XEN_DOMCTL_CDF_pool)
     uint32_t flags;
+    uint32_t cpupool;
 };
 typedef struct xen_domctl_createdomain xen_domctl_createdomain_t;
 DEFINE_XEN_GUEST_HANDLE(xen_domctl_createdomain_t);
@@ -109,6 +113,7 @@ struct xen_domctl_getdomaininfo {
     uint32_t max_vcpu_id;        /* Maximum VCPUID in use by this domain. */
     uint32_t ssidref;
     xen_domain_handle_t handle;
+    uint32_t cpupool;
 };
 typedef struct xen_domctl_getdomaininfo xen_domctl_getdomaininfo_t;
 DEFINE_XEN_GUEST_HANDLE(xen_domctl_getdomaininfo_t);
@@ -645,6 +650,30 @@ typedef struct xen_domctl_hvmcontext_par
     XEN_GUEST_HANDLE_64(uint8) buffer;  /* OUT: buffer to write record into */
 } xen_domctl_hvmcontext_partial_t;
 DEFINE_XEN_GUEST_HANDLE(xen_domctl_hvmcontext_partial_t);
+
+/*
+ * Move domain to specified cpupool.
+ */
+#define XEN_DOMCTL_cpupool_op  56
+#define XEN_DOMCTL_CPUPOOL_OP_CREATE                1  /* C */
+#define XEN_DOMCTL_CPUPOOL_OP_DESTROY               2  /* D */
+#define XEN_DOMCTL_CPUPOOL_OP_INFO                  3  /* I */
+#define XEN_DOMCTL_CPUPOOL_OP_ADDCPU                4  /* A */
+#define XEN_DOMCTL_CPUPOOL_OP_RMCPU                 5  /* R */
+#define XEN_DOMCTL_CPUPOOL_OP_MOVEDOMAIN            6  /* M */
+#define XEN_DOMCTL_CPUPOOL_OP_FREEINFO              7  /* F */
+#define XEN_DOMCTL_CPUPOOL_PAR_ANY     0xFFFFFFFF
+struct xen_domctl_cpupool_op {
+    uint32_t op;          /* IN */
+    uint32_t cpupool_id;  /* IN: CDIARM OUT: CI */
+    uint32_t sched_id;    /* IN: C      OUT: I  */
+    uint32_t domid;       /* IN: M              */
+    uint32_t cpu;         /* IN: AR             */
+    uint32_t n_dom;       /*            OUT: I  */
+    struct xenctl_cpumap cpumap; /*     OUT: IF */
+};
+typedef struct xen_domctl_cpupool_op xen_domctl_cpupool_op_t;
+DEFINE_XEN_GUEST_HANDLE(xen_domctl_cpupool_op_t);
 
 
 struct xen_domctl {
@@ -688,6 +717,7 @@ struct xen_domctl {
         struct xen_domctl_set_target        set_target;
         struct xen_domctl_subscribe         subscribe;
         struct xen_domctl_debug_op          debug_op;
+        struct xen_domctl_cpupool_op        cpupool_op;
 #if defined(__i386__) || defined(__x86_64__)
         struct xen_domctl_cpuid             cpuid;
 #endif
xen/include/xen/sched-if.h +45 -14

88 lines not on this page · raw message

xen/include/xen/sched.h +26 -4
diff -r 655dc3bc1d8e xen/include/xen/sched.h
@@ -182,6 +182,7 @@ struct domain
 
     /* Scheduling. */
     void            *sched_priv;    /* scheduler-specific data */
+    struct cpupool  *cpupool;
 
     struct domain   *next_in_list;
     struct domain   *next_in_hashbucket;
@@ -341,7 +342,7 @@ static inline struct domain *get_current
 }
 
 struct domain *domain_create(
-    domid_t domid, unsigned int domcr_flags, ssidref_t ssidref);
+    domid_t domid, int poolid, unsigned int domcr_flags, ssidref_t ssidref);
  /* DOMCRF_hvm: Create an HVM domain, as opposed to a PV domain. */
 #define _DOMCRF_hvm           0
 #define DOMCRF_hvm            (1U<<_DOMCRF_hvm)
@@ -426,10 +427,11 @@ void sched_destroy_vcpu(struct vcpu *v);
 void sched_destroy_vcpu(struct vcpu *v);
 int  sched_init_domain(struct domain *d);
 void sched_destroy_domain(struct domain *d);
+void sched_move_domain(struct domain *d, struct cpupool *c);
 long sched_adjust(struct domain *, struct xen_domctl_scheduler_op *);
 int  sched_id(void);
-void sched_tick_suspend(void);
-void sched_tick_resume(void);
+void sched_tick_suspend(unsigned int cpu);
+void sched_tick_resume(unsigned int cpu);
 void vcpu_wake(struct vcpu *d);
 void vcpu_sleep_nosync(struct vcpu *d);
 void vcpu_sleep_sync(struct vcpu *d);
@@ -533,8 +535,13 @@ void domain_unpause_by_systemcontroller(
 void domain_unpause_by_systemcontroller(struct domain *d);
 void cpu_init(void);
 
+struct scheduler;
+
+int schedule_init_global(char *name, struct scheduler *sched);
+void schedule_deinit_global(struct scheduler *sched);
+void schedule_cpu_switch(unsigned int cpu, struct cpupool *c);
 void vcpu_force_reschedule(struct vcpu *v);
-void cpu_disable_scheduler(void);
+int cpu_disable_scheduler(unsigned int cpu, int lock);
 int vcpu_set_affinity(struct vcpu *v, cpumask_t *affinity);
 int vcpu_lock_affinity(struct vcpu *v, cpumask_t *affinity);
 int vcpu_locked_change_affinity(struct vcpu *v, cpumask_t *affinity);
@@ -560,6 +567,21 @@ extern enum cpufreq_controller {
 extern enum cpufreq_controller {
     FREQCTL_none, FREQCTL_dom0_kernel, FREQCTL_xen
 } cpufreq_controller;
+
+#define CPUPOOLID_NONE    -1
+
+struct cpupool *cpupool_create(int poolid, char *sched);
+int cpupool_destroy(struct cpupool *c);
+int cpupool0_cpu_assign(struct cpupool *c);
+int cpupool_assign_ncpu(struct cpupool *c, int ncpu);
+void cpupool_cpu_add(unsigned int cpu);
+int cpupool_cpu_remove(unsigned int cpu);
+int cpupool_borrow_cpu(struct cpupool *c, unsigned int cpu);
+int cpupool_return_cpu(struct cpupool *c);
+int cpupool_add_domain(struct domain *d, int poolid);
+void cpupool_rm_domain(struct domain *d);
+int cpupool_do_domctl(struct xen_domctl *op);
+#define num_cpupool_cpus(c) (cpus_weight((c)->cpu_valid))
 
 #endif /* __SCHED_H__ */
 
xen/common/cpupool.c new file +698 -0

698 lines not on this page · raw message

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel