[Xen-devel] [PATCH v5 for-xen-4.5 1/3] dpci: Replace tasklet with an softirq (v5)

[Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>]

The existing tasklet mechanism has a single global
spinlock that is taken every-time the global list
is touched. And we use this lock quite a lot - when
we call do_tasklet_work which is called via an softirq
and from the idle loop. We take the lock on any
operation on the tasklet_list.

The problem we are facing is that there are quite a lot of
tasklets scheduled. The most common one that is invoked is
the one injecting the VIRQ_TIMER in the guest. Guests
are not insane and don't set the one-shot or periodic
clocks to be in sub 1ms intervals (causing said tasklet
to be scheduled for such small intervalls).

The problem appears when PCI passthrough devices are used
over many sockets and we have an mix of heavy-interrupt
guests and idle guests. The idle guests end up seeing
1/10 of its RUNNING timeslice eaten by the hypervisor
(and 40% steal time).

The mechanism by which we inject PCI interrupts is by
hvm_do_IRQ_dpci which schedules the hvm_dirq_assist
tasklet every time an interrupt is received.
The callchain is:

_asm_vmexit_handler
 -> vmx_vmexit_handler
    ->vmx_do_extint
        -> do_IRQ
            -> __do_IRQ_guest
                -> hvm_do_IRQ_dpci
                   tasklet_schedule(&dpci->dirq_tasklet);
                   [takes lock to put the tasklet on]

[later on the schedule_tail is invoked which is 'vmx_do_resume']

vmx_do_resume
 -> vmx_asm_do_vmentry
        -> call vmx_intr_assist
          -> vmx_process_softirqs
            -> do_softirq
              [executes the tasklet function, takes the
               lock again]

While on other CPUs they might be sitting in a idle loop
and invoked to deliver an VIRQ_TIMER, which also ends
up taking the lock twice: first to schedule the
v->arch.hvm_vcpu.assert_evtchn_irq_tasklet (accounted to
the guests' BLOCKED_state); then to execute it - which is
accounted for in the guest's RUNTIME_state.

The end result is that on a 8 socket machine with
PCI passthrough, where four sockets are busy with interrupts,
and the other sockets have idle guests - we end up with
the idle guests having around 40% steal time and 1/10
of its timeslice (3ms out of 30 ms) being tied up
taking the lock. The latency of the PCI interrupts delieved
to guest is also hindered.

With this patch the problem disappears completly.
That is removing the lock for the PCI passthrough use-case
(the 'hvm_dirq_assist' case) by not using tasklets at all.

The patch is simple - instead of scheduling an tasklet
we schedule our own softirq - HVM_DPCI_SOFTIRQ, which will
take care of running 'hvm_dirq_assist'. The information we need
on each CPU is which 'struct hvm_pirq_dpci'  structure the 'hvm_dirq_assist'
needs to run on. That is simple solved by threading the
'struct hvm_pirq_dpci' through a linked list. The rule of only running
one 'hvm_dirq_assist' for only one 'hvm_pirq_dpci' is also preserved
by having 'schedule_dpci_for' ignore any subsequent calls for
an domain which has already been scheduled.

Since we are using 'hvm_pirq_dpci' it is setup in 'pt_irq_create_bind'
function tore down in 'pt_irq_destroy_bind'. If the OS forgot (or QEMU)
we still clean it up in pci_clean_dpci_irq. We can ignore the
return value of 'dpci_kill' in 'pt_irq_create_bind' for the error
routines as 'flags' is reset - which means that 'pt_pirq_iterate'
will ignore the structure and never call 'hvm_dirq_assist' on
this particular 'hvm_pirq_dpci' structure.

We also add 'dpci_kill' in the cleanup of 'pirq_cleanup_check'.
If the 'dpci_kill' fails we need to inhibit the deletion
of the structure from the 'pirq_tree' - otherwise the
catch-it-all 'pci_clean_dpci_irq' won't be able to properly
cleanup up using the 'dpci_kill'.

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
Suggested-by: Jan Beulich <JBeulich@xxxxxxxx>

---
v2: On top of ref-cnts also have wait loop for the outstanding
    'struct domain' that need to be processed.
v3: Add -ERETRY, fix up StyleGuide issues
v4: Clean it up mode, redo per_cpu, this_cpu logic
v5: Instead of threading struct domain, use hvm_pirq_dpci.
---
 xen/arch/x86/domain.c         |   4 +-
 xen/arch/x86/irq.c            |   6 ++
 xen/drivers/passthrough/io.c  | 175 +++++++++++++++++++++++++++++++++++++++---
 xen/drivers/passthrough/pci.c |  27 +++++--
 xen/include/xen/hvm/irq.h     |   5 +-
 xen/include/xen/pci.h         |   2 +-
 xen/include/xen/softirq.h     |   3 +
 7 files changed, 200 insertions(+), 22 deletions(-)

diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c
index 7b1dfe6..12bf241 100644
--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -1924,7 +1924,9 @@ int domain_relinquish_resources(struct domain *d)
     switch ( d->arch.relmem )
     {
     case RELMEM_not_started:
-        pci_release_devices(d);
+        ret = pci_release_devices(d);
+        if ( ret )
+            return ret;
 
         /* Tear down paging-assistance stuff. */
         paging_teardown(d);
diff --git a/xen/arch/x86/irq.c b/xen/arch/x86/irq.c
index f214072..a4c7053 100644
--- a/xen/arch/x86/irq.c
+++ b/xen/arch/x86/irq.c
@@ -1337,6 +1337,12 @@ void (pirq_cleanup_check)(struct pirq *pirq, struct 
domain *d)
             return;
         if ( !pt_pirq_cleanup_check(&pirq->arch.hvm.dpci) )
             return;
+        /*
+         * Inhibit deletion from 'pirq_tree' so that 'pci_clean_dpci_irq'
+         * can still properly call 'dpci_kill'.
+         */
+        if ( dpci_kill(&pirq->arch.hvm.dpci) )
+            return;
     }
 
     if ( radix_tree_delete(&d->pirq_tree, pirq->pirq) != pirq )
diff --git a/xen/drivers/passthrough/io.c b/xen/drivers/passthrough/io.c
index 4cd32b5..5f82aba 100644
--- a/xen/drivers/passthrough/io.c
+++ b/xen/drivers/passthrough/io.c
@@ -20,14 +20,79 @@
 
 #include <xen/event.h>
 #include <xen/iommu.h>
+#include <xen/cpu.h>
 #include <xen/irq.h>
 #include <asm/hvm/irq.h>
 #include <asm/hvm/iommu.h>
 #include <asm/hvm/support.h>
 #include <xen/hvm/irq.h>
-#include <xen/tasklet.h>
 
-static void hvm_dirq_assist(unsigned long _d);
+static DEFINE_PER_CPU(struct list_head, dpci_list);
+
+/*
+ * These fancy bit manipulations (bit 0 and bit 1) along with using a lock
+ * operation allow us to have four stages in dpci life-time.
+ *   a) 0x0: Completely empty (not scheduled nor running).
+ *   b) 0x1: Scheduled but not running. Used to guard in 'schedule_dpci_for'
+ *      such that we will only schedule one. If it is scheduled and had never
+ *      run (hence never clearing STATE_SCHED bit), dpci_kill will spin
+ *      forever on said hvm_pirq_dpci. However 'schedule_dpci_for' raises the
+ *      softirq  so it will run, albeit there might be a race (dpci_kill
+ *      spinning until the softirq handler runs).
+ *   c) 0x2: it is running (only on one CPU) and can be scheduled on any
+ *      CPU. The bit 0 - scheduled is cleared at this stage allowing
+ *      'schedule_dpci_for' to succesfully schedule.
+ *   d) 0x3: scheduled and running - only possible if the running 
hvm_dirq_assist
+ *      calls 'schedule_dpci_for'. It does not do that, but it could in the
+ *      future.
+ *
+ *  The two bits play a vital role in assuring that the 'hvm_dirq_assist' is
+ *  scheduled once and runs only once for a hvm_pirq_dpci steps are:
+ *
+ *   1) schedule_dpci_for: STATE_SCHED bit set (0x1), added on the per cpu 
list.
+ *   2) dpci_softirq picks one hvm_pirq_dpci from the list. Schedules
+ *   itself later if there are more hvm_pirq_dpci's on it. Tries to set 
STATE_RUN bit
+ *   (0x3) - if it fails adds hvm_pirq_dpci back to the per-cpu list. If it 
succeeds
+ *   clears the STATE_SCHED bit (0x2). Once hvm_dirq_assist for an 
hvm_pirq_dpci
+ *   complets, it unsets STATE_RUN (0x0 or 0x1 if 'schedule_dpci_for' is called
+ *   from 'hvm_dirq_assist').
+ */
+enum {
+    STATE_SCHED, /* Bit 0 */
+    STATE_RUN
+};
+
+static void schedule_dpci_for(struct hvm_pirq_dpci *pirq_dpci)
+{
+    if ( !test_and_set_bit(STATE_SCHED, &pirq_dpci->state) )
+    {
+        unsigned long flags;
+
+        get_knownalive_domain(pirq_dpci->dom); /* To be put by 'dpci_softirq' 
*/
+
+        local_irq_save(flags);
+        list_add_tail(&pirq_dpci->softirq_list, &this_cpu(dpci_list));
+        local_irq_restore(flags);
+
+        raise_softirq(HVM_DPCI_SOFTIRQ);
+    }
+}
+
+int dpci_kill(struct hvm_pirq_dpci *pirq_dpci)
+{
+    if ( test_and_set_bit(STATE_SCHED, &pirq_dpci->state) )
+        return -EAGAIN;
+
+    if ( test_bit(STATE_RUN, &pirq_dpci->state) )
+        return -EAGAIN;
+
+    clear_bit(STATE_SCHED, &pirq_dpci->state);
+
+    /* In case we turn around and call 'schedule_dpci_for' right away. */
+    INIT_LIST_HEAD(&pirq_dpci->softirq_list);
+    return 0;
+}
+
 
 bool_t pt_irq_need_timer(uint32_t flags)
 {
@@ -114,9 +179,7 @@ int pt_irq_create_bind(
             spin_unlock(&d->event_lock);
             return -ENOMEM;
         }
-        softirq_tasklet_init(
-            &hvm_irq_dpci->dirq_tasklet,
-            hvm_dirq_assist, (unsigned long)d);
+
         for ( i = 0; i < NR_HVM_IRQS; i++ )
             INIT_LIST_HEAD(&hvm_irq_dpci->girq[i]);
 
@@ -131,6 +194,16 @@ int pt_irq_create_bind(
     }
     pirq_dpci = pirq_dpci(info);
 
+    /* Something is horrible wrong if it has been scheduled or is running. */
+    ASSERT(pirq_dpci->state == 0);
+
+    /* Guest unbinds (pt_irq_destroy_bind) and rebinds it back. */
+    if ( !pirq_dpci->dom )
+        pirq_dpci->dom = d;
+
+    /* Otherwise 'schedule_dpci_for' will crash. */
+    ASSERT(pirq_dpci->dom == d);
+
     switch ( pt_irq_bind->irq_type )
     {
     case PT_IRQ_TYPE_MSI:
@@ -156,7 +229,9 @@ int pt_irq_create_bind(
             {
                 pirq_dpci->gmsi.gflags = 0;
                 pirq_dpci->gmsi.gvec = 0;
+                pirq_dpci->dom = NULL;
                 pirq_dpci->flags = 0;
+                dpci_kill(pirq_dpci);
                 pirq_cleanup_check(info, d);
                 spin_unlock(&d->event_lock);
                 return rc;
@@ -232,7 +307,6 @@ int pt_irq_create_bind(
         {
             unsigned int share;
 
-            pirq_dpci->dom = d;
             if ( pt_irq_bind->irq_type == PT_IRQ_TYPE_MSI_TRANSLATE )
             {
                 pirq_dpci->flags = HVM_IRQ_DPCI_MAPPED |
@@ -263,6 +337,7 @@ int pt_irq_create_bind(
                 list_del(&digl->list);
                 hvm_irq_dpci->link_cnt[link]--;
                 pirq_dpci->flags = 0;
+                dpci_kill(pirq_dpci);
                 pirq_cleanup_check(info, d);
                 spin_unlock(&d->event_lock);
                 xfree(girq);
@@ -393,6 +468,7 @@ int pt_irq_destroy_bind(
             kill_timer(&pirq_dpci->timer);
         pirq_dpci->dom   = NULL;
         pirq_dpci->flags = 0;
+        dpci_kill(pirq_dpci);
         pirq_cleanup_check(pirq, d);
     }
 
@@ -414,6 +490,8 @@ int pt_irq_destroy_bind(
 void pt_pirq_init(struct domain *d, struct hvm_pirq_dpci *dpci)
 {
     INIT_LIST_HEAD(&dpci->digl_list);
+    INIT_LIST_HEAD(&dpci->softirq_list);
+    dpci->dom = d;
     dpci->gmsi.dest_vcpu_id = -1;
 }
 
@@ -459,7 +537,7 @@ int hvm_do_IRQ_dpci(struct domain *d, struct pirq *pirq)
         return 0;
 
     pirq_dpci->masked = 1;
-    tasklet_schedule(&dpci->dirq_tasklet);
+    schedule_dpci_for(pirq_dpci);
     return 1;
 }
 
@@ -562,11 +640,11 @@ static int _hvm_dirq_assist(struct domain *d, struct 
hvm_pirq_dpci *pirq_dpci,
     return 0;
 }
 
-static void hvm_dirq_assist(unsigned long _d)
+static void hvm_dirq_assist(struct hvm_pirq_dpci *pirq_dpci)
 {
-    struct domain *d = (struct domain *)_d;
+    struct domain *d = pirq_dpci->dom;
 
-    ASSERT(d->arch.hvm_domain.irq.dpci);
+    ASSERT(d && d->arch.hvm_domain.irq.dpci);
 
     spin_lock(&d->event_lock);
     pt_pirq_iterate(d, _hvm_dirq_assist, NULL);
@@ -625,3 +703,80 @@ void hvm_dpci_eoi(struct domain *d, unsigned int guest_gsi,
 unlock:
     spin_unlock(&d->event_lock);
 }
+
+static void dpci_softirq(void)
+{
+    struct hvm_pirq_dpci *pirq_dpci;
+    unsigned int cpu = smp_processor_id();
+    LIST_HEAD(our_list);
+
+    local_irq_disable();
+    list_splice_init(&per_cpu(dpci_list, cpu), &our_list);
+    local_irq_enable();
+
+    while ( !list_empty(&our_list) )
+    {
+        pirq_dpci = list_entry(our_list.next, struct hvm_pirq_dpci, 
softirq_list);
+        list_del(&pirq_dpci->softirq_list);
+
+        if ( !test_and_set_bit(STATE_RUN, &pirq_dpci->state) )
+        {
+            if ( !test_and_clear_bit(STATE_SCHED, &pirq_dpci->state) )
+                BUG();
+            hvm_dirq_assist(pirq_dpci);
+            put_domain(pirq_dpci->dom);
+            clear_bit(STATE_RUN, &pirq_dpci->state);
+            continue;
+        }
+
+        local_irq_disable();
+        list_add_tail(&pirq_dpci->softirq_list, &per_cpu(dpci_list, cpu));
+        local_irq_enable();
+
+        raise_softirq(HVM_DPCI_SOFTIRQ);
+    }
+}
+static int cpu_callback(
+    struct notifier_block *nfb, unsigned long action, void *hcpu)
+{
+    unsigned int cpu = (unsigned long)hcpu;
+
+    switch ( action )
+    {
+    case CPU_UP_PREPARE:
+        INIT_LIST_HEAD(&per_cpu(dpci_list, cpu));
+        break;
+    case CPU_UP_CANCELED:
+    case CPU_DEAD:
+        /*
+         * On CPU_DYING this callback is called (on the CPU that is dying)
+         * with an possible HVM_DPIC_SOFTIRQ pending - at which point we can
+         * clear out any outstanding domains (by the virtue of the idle loop
+         * calling the softirq later). In CPU_DEAD case the CPU is deaf and
+         * there are no pending softirqs for us to handle so we can chill.
+         */
+        ASSERT(list_empty(&per_cpu(dpci_list, cpu)));
+        break;
+    default:
+        break;
+    }
+
+    return NOTIFY_DONE;
+}
+
+static struct notifier_block cpu_nfb = {
+    .notifier_call = cpu_callback,
+};
+
+static int __init setup_dpci_softirq(void)
+{
+    int cpu;
+
+    for_each_online_cpu(cpu)
+        INIT_LIST_HEAD(&per_cpu(dpci_list, cpu));
+
+    open_softirq(HVM_DPCI_SOFTIRQ, dpci_softirq);
+    register_cpu_notifier(&cpu_nfb);
+    return 0;
+}
+__initcall(setup_dpci_softirq);
diff --git a/xen/drivers/passthrough/pci.c b/xen/drivers/passthrough/pci.c
index 1eba833..9be3405 100644
--- a/xen/drivers/passthrough/pci.c
+++ b/xen/drivers/passthrough/pci.c
@@ -767,40 +767,49 @@ static int pci_clean_dpci_irq(struct domain *d,
         xfree(digl);
     }
 
-    return 0;
+    return dpci_kill(pirq_dpci);
 }
 
-static void pci_clean_dpci_irqs(struct domain *d)
+static int pci_clean_dpci_irqs(struct domain *d)
 {
     struct hvm_irq_dpci *hvm_irq_dpci = NULL;
 
     if ( !iommu_enabled )
-        return;
+        return -ESRCH;
 
     if ( !is_hvm_domain(d) )
-        return;
+        return -EINVAL;
 
     spin_lock(&d->event_lock);
     hvm_irq_dpci = domain_get_irq_dpci(d);
     if ( hvm_irq_dpci != NULL )
     {
-        tasklet_kill(&hvm_irq_dpci->dirq_tasklet);
+        int ret = pt_pirq_iterate(d, pci_clean_dpci_irq, NULL);
 
-        pt_pirq_iterate(d, pci_clean_dpci_irq, NULL);
+        if ( ret )
+        {
+            spin_unlock(&d->event_lock);
+            return ret;
+        }
 
         d->arch.hvm_domain.irq.dpci = NULL;
         free_hvm_irq_dpci(hvm_irq_dpci);
     }
     spin_unlock(&d->event_lock);
+    return 0;
 }
 
-void pci_release_devices(struct domain *d)
+int pci_release_devices(struct domain *d)
 {
     struct pci_dev *pdev;
     u8 bus, devfn;
+    int ret;
 
     spin_lock(&pcidevs_lock);
-    pci_clean_dpci_irqs(d);
+    ret = pci_clean_dpci_irqs(d);
+    if ( ret == -EAGAIN )
+        return ret;
+
     while ( (pdev = pci_get_pdev_by_domain(d, -1, -1, -1)) )
     {
         bus = pdev->bus;
@@ -811,6 +820,8 @@ void pci_release_devices(struct domain *d)
                    PCI_SLOT(devfn), PCI_FUNC(devfn));
     }
     spin_unlock(&pcidevs_lock);
+
+    return 0;
 }
 
 #define PCI_CLASS_BRIDGE_HOST    0x0600
diff --git a/xen/include/xen/hvm/irq.h b/xen/include/xen/hvm/irq.h
index c89f4b1..de8ea57 100644
--- a/xen/include/xen/hvm/irq.h
+++ b/xen/include/xen/hvm/irq.h
@@ -88,7 +88,6 @@ struct hvm_irq_dpci {
     DECLARE_BITMAP(isairq_map, NR_ISAIRQS);
     /* Record of mapped Links */
     uint8_t link_cnt[NR_LINK];
-    struct tasklet dirq_tasklet;
 };
 
 /* Machine IRQ to guest device/intx mapping. */
@@ -100,6 +99,8 @@ struct hvm_pirq_dpci {
     struct domain *dom;
     struct hvm_gmsi_info gmsi;
     struct timer timer;
+    struct list_head softirq_list;
+    unsigned long state;
 };
 
 void pt_pirq_init(struct domain *, struct hvm_pirq_dpci *);
@@ -108,7 +109,7 @@ int pt_pirq_iterate(struct domain *d,
                     int (*cb)(struct domain *,
                               struct hvm_pirq_dpci *, void *arg),
                     void *arg);
-
+int dpci_kill(struct hvm_pirq_dpci *);
 /* Modify state of a PCI INTx wire. */
 void hvm_pci_intx_assert(
     struct domain *d, unsigned int device, unsigned int intx);
diff --git a/xen/include/xen/pci.h b/xen/include/xen/pci.h
index 91520bc..5f295f3 100644
--- a/xen/include/xen/pci.h
+++ b/xen/include/xen/pci.h
@@ -99,7 +99,7 @@ struct pci_dev *pci_lock_domain_pdev(
 
 void setup_hwdom_pci_devices(struct domain *,
                             int (*)(u8 devfn, struct pci_dev *));
-void pci_release_devices(struct domain *d);
+int pci_release_devices(struct domain *d);
 int pci_add_segment(u16 seg);
 const unsigned long *pci_get_ro_map(u16 seg);
 int pci_add_device(u16 seg, u8 bus, u8 devfn, const struct pci_dev_info *);
diff --git a/xen/include/xen/softirq.h b/xen/include/xen/softirq.h
index 0895a16..f3da5df 100644
--- a/xen/include/xen/softirq.h
+++ b/xen/include/xen/softirq.h
@@ -8,6 +8,9 @@ enum {
     NEW_TLBFLUSH_CLOCK_PERIOD_SOFTIRQ,
     RCU_SOFTIRQ,
     TASKLET_SOFTIRQ,
+#ifdef HAS_PASSTHROUGH
+    HVM_DPCI_SOFTIRQ,
+#endif
     NR_COMMON_SOFTIRQS
 };
 
-- 
1.9.3


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel