Original version of that patch (commit a89941816726) had to be reverted
due to Xen allocating irqs in its cpu_up ops.
The first patch moves allocations into hotplug notifiers and the second
one restores the original patch (with minor adjustments to new hotplug
framework)
Boris Ostrovsky (2):
xen/x86: Move irq allocation from Xen smp_op.cpu_up()
hotplug: Prevent alloc/free of irq descriptors during cpu up/down
(again)
arch/x86/kernel/smpboot.c | 11 ---------
arch/x86/xen/enlighten.c | 53 +++++++++++++++++++++++++++++++++++++-------
arch/x86/xen/smp.c | 45 +------------------------------------
arch/x86/xen/smp.h | 3 ++
kernel/cpu.c | 8 ++++++
5 files changed, 57 insertions(+), 63 deletions(-)
Commit ce0d3c0a6fb1 ("genirq: Revert sparse irq locking around
__cpu_up() and move it to x86 for now") reverted irq locking
introduced by commit a89941816726 ("hotplug: Prevent alloc/free
of irq descriptors during cpu up/down") because of Xen allocating
irqs in both of its cpu_up ops.
We can move those allocations into CPU notifiers so that original
patch can be reinstated.
Signed-off-by: Boris Ostrovsky <[email protected]>
---
arch/x86/xen/enlighten.c | 53 ++++++++++++++++++++++++++++++++++++++-------
arch/x86/xen/smp.c | 45 +-------------------------------------
arch/x86/xen/smp.h | 3 ++
3 files changed, 49 insertions(+), 52 deletions(-)
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index 2c26108..d1a86db 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -137,6 +137,8 @@ RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
__read_mostly int xen_have_vector_callback;
EXPORT_SYMBOL_GPL(xen_have_vector_callback);
+static struct notifier_block xen_cpu_notifier;
+
/*
* Point at some empty memory to start with. We map the real shared_info
* page as soon as fixmap is up and running.
@@ -1596,6 +1598,7 @@ asmlinkage __visible void __init xen_start_kernel(void)
xen_initial_gdt = &per_cpu(gdt_page, 0);
xen_smp_init();
+ register_cpu_notifier(&xen_cpu_notifier);
#ifdef CONFIG_ACPI_NUMA
/*
@@ -1783,17 +1786,49 @@ static void __init init_hvm_pv_info(void)
xen_domain_type = XEN_HVM_DOMAIN;
}
-static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
- void *hcpu)
+static int xen_cpu_notify(struct notifier_block *self, unsigned long action,
+ void *hcpu)
{
int cpu = (long)hcpu;
+ int rc;
+
switch (action) {
case CPU_UP_PREPARE:
- xen_vcpu_setup(cpu);
- if (xen_have_vector_callback) {
- if (xen_feature(XENFEAT_hvm_safe_pvclock))
- xen_setup_timer(cpu);
+ if (xen_hvm_domain()) {
+ /*
+ * This can happen if CPU was offlined earlier and
+ * offlining timed out in common_cpu_die().
+ */
+ if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
+ xen_smp_intr_free(cpu);
+ xen_uninit_lock_cpu(cpu);
+ }
+
+ xen_vcpu_setup(cpu);
}
+
+ if (xen_pv_domain() ||
+ (xen_have_vector_callback &&
+ xen_feature(XENFEAT_hvm_safe_pvclock)))
+ xen_setup_timer(cpu);
+
+ rc = xen_smp_intr_init(cpu);
+ if (rc) {
+ WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
+ cpu, rc);
+ return NOTIFY_BAD;
+ }
+
+ break;
+ case CPU_ONLINE:
+ xen_init_lock_cpu(cpu);
+ break;
+ case CPU_UP_CANCELED:
+ xen_smp_intr_free(cpu);
+ if (xen_pv_domain() ||
+ (xen_have_vector_callback &&
+ xen_feature(XENFEAT_hvm_safe_pvclock)))
+ xen_teardown_timer(cpu);
break;
default:
break;
@@ -1801,8 +1836,8 @@ static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
return NOTIFY_OK;
}
-static struct notifier_block xen_hvm_cpu_notifier = {
- .notifier_call = xen_hvm_cpu_notify,
+static struct notifier_block xen_cpu_notifier = {
+ .notifier_call = xen_cpu_notify,
};
#ifdef CONFIG_KEXEC_CORE
@@ -1834,7 +1869,7 @@ static void __init xen_hvm_guest_init(void)
if (xen_feature(XENFEAT_hvm_callback_vector))
xen_have_vector_callback = 1;
xen_hvm_smp_init();
- register_cpu_notifier(&xen_hvm_cpu_notifier);
+ register_cpu_notifier(&xen_cpu_notifier);
xen_unplug_emulated_devices();
x86_init.irqs.intr_init = xen_init_IRQ;
xen_hvm_init_time_ops();
diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
index 719cf29..09d5cc0 100644
--- a/arch/x86/xen/smp.c
+++ b/arch/x86/xen/smp.c
@@ -115,7 +115,7 @@ asmlinkage __visible void cpu_bringup_and_idle(int cpu)
cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
}
-static void xen_smp_intr_free(unsigned int cpu)
+void xen_smp_intr_free(unsigned int cpu)
{
if (per_cpu(xen_resched_irq, cpu).irq >= 0) {
unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL);
@@ -159,7 +159,7 @@ static void xen_smp_intr_free(unsigned int cpu)
per_cpu(xen_pmu_irq, cpu).name = NULL;
}
};
-static int xen_smp_intr_init(unsigned int cpu)
+int xen_smp_intr_init(unsigned int cpu)
{
int rc;
char *resched_name, *callfunc_name, *debug_name, *pmu_name;
@@ -468,8 +468,6 @@ static int xen_cpu_up(unsigned int cpu, struct task_struct *idle)
common_cpu_up(cpu, idle);
xen_setup_runstate_info(cpu);
- xen_setup_timer(cpu);
- xen_init_lock_cpu(cpu);
/*
* PV VCPUs are always successfully taken down (see 'while' loop
@@ -488,10 +486,6 @@ static int xen_cpu_up(unsigned int cpu, struct task_struct *idle)
xen_pmu_init(cpu);
- rc = xen_smp_intr_init(cpu);
- if (rc)
- return rc;
-
rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
BUG_ON(rc);
@@ -761,47 +755,12 @@ static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
xen_init_lock_cpu(0);
}
-static int xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle)
-{
- int rc;
-
- /*
- * This can happen if CPU was offlined earlier and
- * offlining timed out in common_cpu_die().
- */
- if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
- xen_smp_intr_free(cpu);
- xen_uninit_lock_cpu(cpu);
- }
-
- /*
- * xen_smp_intr_init() needs to run before native_cpu_up()
- * so that IPI vectors are set up on the booting CPU before
- * it is marked online in native_cpu_up().
- */
- rc = xen_smp_intr_init(cpu);
- WARN_ON(rc);
- if (!rc)
- rc = native_cpu_up(cpu, tidle);
-
- /*
- * We must initialize the slowpath CPU kicker _after_ the native
- * path has executed. If we initialized it before none of the
- * unlocker IPI kicks would reach the booting CPU as the booting
- * CPU had not set itself 'online' in cpu_online_mask. That mask
- * is checked when IPIs are sent (on HVM at least).
- */
- xen_init_lock_cpu(cpu);
- return rc;
-}
-
void __init xen_hvm_smp_init(void)
{
if (!xen_have_vector_callback)
return;
smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
- smp_ops.cpu_up = xen_hvm_cpu_up;
smp_ops.cpu_die = xen_cpu_die;
smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
diff --git a/arch/x86/xen/smp.h b/arch/x86/xen/smp.h
index 963d62a..45faaf3 100644
--- a/arch/x86/xen/smp.h
+++ b/arch/x86/xen/smp.h
@@ -8,6 +8,9 @@ extern void xen_send_IPI_allbutself(int vector);
extern void xen_send_IPI_all(int vector);
extern void xen_send_IPI_self(int vector);
+extern int xen_smp_intr_init(unsigned int cpu);
+extern void xen_smp_intr_free(unsigned int cpu);
+
#ifdef CONFIG_XEN_PVH
extern void xen_pvh_early_cpu_init(int cpu, bool entry);
#else
--
1.7.1
Now that Xen no longer allocates irqs in _cpu_up() we can restore
commit a89941816726 ("hotplug: Prevent alloc/free of irq descriptors
during cpu up/down")
Signed-off-by: Boris Ostrovsky <[email protected]>
---
arch/x86/kernel/smpboot.c | 11 -----------
kernel/cpu.c | 8 ++++++++
2 files changed, 8 insertions(+), 11 deletions(-)
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 643dbdc..cabe21e 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -1083,17 +1083,8 @@ int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
common_cpu_up(cpu, tidle);
- /*
- * We have to walk the irq descriptors to setup the vector
- * space for the cpu which comes online. Prevent irq
- * alloc/free across the bringup.
- */
- irq_lock_sparse();
-
err = do_boot_cpu(apicid, cpu, tidle);
-
if (err) {
- irq_unlock_sparse();
pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
return -EIO;
}
@@ -1111,8 +1102,6 @@ int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
touch_nmi_watchdog();
}
- irq_unlock_sparse();
-
return 0;
}
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 6ea42e8..2ff63b3 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -342,8 +342,16 @@ static int bringup_cpu(unsigned int cpu)
struct task_struct *idle = idle_thread_get(cpu);
int ret;
+ /*
+ * Some architectures have to walk the irq descriptors to
+ * setup the vector space for the cpu which comes online.
+ * Prevent irq alloc/free across the bringup.
+ */
+ irq_lock_sparse();
+
/* Arch-specific enabling code. */
ret = __cpu_up(cpu, idle);
+ irq_unlock_sparse();
if (ret) {
cpu_notify(CPU_UP_CANCELED, cpu);
return ret;
--
1.7.1
On 03/25/2016 11:10 AM, Konrad Rzeszutek Wilk wrote:
> On Thu, Mar 17, 2016 at 09:33:32AM -0400, Boris Ostrovsky wrote:
>> Commit ce0d3c0a6fb1 ("genirq: Revert sparse irq locking around
>> __cpu_up() and move it to x86 for now") reverted irq locking
>> introduced by commit a89941816726 ("hotplug: Prevent alloc/free
>> of irq descriptors during cpu up/down") because of Xen allocating
>> irqs in both of its cpu_up ops.
>>
>> We can move those allocations into CPU notifiers so that original
>> patch can be reinstated.
> Original being "hotplug: Prevent alloc/free..." ?
Yes.
>
> -static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
> - void *hcpu)
> +static int xen_cpu_notify(struct notifier_block *self, unsigned long action,
> + void *hcpu)
> {
> int cpu = (long)hcpu;
> + int rc;
> +
> switch (action) {
> case CPU_UP_PREPARE:
> - xen_vcpu_setup(cpu);
> - if (xen_have_vector_callback) {
> - if (xen_feature(XENFEAT_hvm_safe_pvclock))
> - xen_setup_timer(cpu);
> + if (xen_hvm_domain()) {
> + /*
> + * This can happen if CPU was offlined earlier and
> + * offlining timed out in common_cpu_die().
> + */
> + if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
> + xen_smp_intr_free(cpu);
> + xen_uninit_lock_cpu(cpu);
> + }
> +
> + xen_vcpu_setup(cpu);
> }
> +
> + if (xen_pv_domain() ||
> + (xen_have_vector_callback &&
> + xen_feature(XENFEAT_hvm_safe_pvclock)))
> + xen_setup_timer(cpu);
> +
> + rc = xen_smp_intr_init(cpu);
> + if (rc) {
> + WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
> + cpu, rc);
> + return NOTIFY_BAD;
> + }
> +
> + break;
> + case CPU_ONLINE:
> + xen_init_lock_cpu(cpu);
> + break;
> + case CPU_UP_CANCELED:
> + xen_smp_intr_free(cpu);
> xen_uninit_lock_cpu ?
I don't think this is needed: we initialize lock in CPU_ONLINE notifier
which can only be called after CPU_UP_CANCELED would have run (in which
case we'll never do CPU_ONLINE)
-boris
>
>
>> + if (xen_pv_domain() ||
>> + (xen_have_vector_callback &&
>> + xen_feature(XENFEAT_hvm_safe_pvclock)))
>> + xen_teardown_timer(cpu);
>> break;
>> default:
>> break;
Ping?
On 03/17/2016 09:33 AM, Boris Ostrovsky wrote:
> Original version of that patch (commit a89941816726) had to be reverted
> due to Xen allocating irqs in its cpu_up ops.
>
> The first patch moves allocations into hotplug notifiers and the second
> one restores the original patch (with minor adjustments to new hotplug
> framework)
>
> Boris Ostrovsky (2):
> xen/x86: Move irq allocation from Xen smp_op.cpu_up()
> hotplug: Prevent alloc/free of irq descriptors during cpu up/down
> (again)
>
> arch/x86/kernel/smpboot.c | 11 ---------
> arch/x86/xen/enlighten.c | 53 +++++++++++++++++++++++++++++++++++++-------
> arch/x86/xen/smp.c | 45 +------------------------------------
> arch/x86/xen/smp.h | 3 ++
> kernel/cpu.c | 8 ++++++
> 5 files changed, 57 insertions(+), 63 deletions(-)
>