xen/arm: smpboot: Allocate the CPU sibling/core maps while preparing the CPU

[PATCH] xen/arm: smpboot: Allocate the CPU sibling/core maps while preparing the CPU

Posted by Julien Grall 2 years, 5 months ago

From: Julien Grall <jgrall@amazon.com>

Commit 5047cd1d5dea "xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in
xmalloc()" extended the checks in _xmalloc() to catch any use of the
helpers from context with interrupts disabled.

Unfortunately, the rule is not followed when allocating the CPU
sibling/core maps.

(XEN) Xen call trace:
(XEN)    [<00238a5c>] _xmalloc+0xfc/0x314 (PC)
(XEN)    [<00000000>] 00000000 (LR)
(XEN)    [<00238c8c>] _xzalloc+0x18/0x4c
(XEN)    [<00288cb4>] smpboot.c#setup_cpu_sibling_map+0x38/0x138
(XEN)    [<00289024>] start_secondary+0x1b4/0x270
(XEN)    [<40010170>] 40010170
(XEN)
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 2:
(XEN) Assertion '!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1)' failed at common/xmalloc_tlsf.c:601
(XEN) ****************************************

This is happening because zalloc_cpumask_var() may allocate memory
if NR_CPUS is > 2 * sizeof(unsigned long).

Avoid the problem by allocate the per-CPU IRQs while preparing the
CPU.

This also has the benefit to remove a panic() in the secondary CPU
code.

Signed-off-by: Julien Grall <jgrall@amazon.com>
---
 xen/arch/arm/smpboot.c | 25 ++++++++++++++++++++-----
 1 file changed, 20 insertions(+), 5 deletions(-)

diff --git a/xen/arch/arm/smpboot.c b/xen/arch/arm/smpboot.c
index 4888bcd78a5a..2b0c92cd369b 100644
--- a/xen/arch/arm/smpboot.c
+++ b/xen/arch/arm/smpboot.c
@@ -79,15 +79,17 @@ DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_mask);
 static bool __read_mostly opt_hmp_unsafe = false;
 boolean_param("hmp-unsafe", opt_hmp_unsafe);
 
-static void setup_cpu_sibling_map(int cpu)
+static int setup_cpu_sibling_map(int cpu)
 {
     if ( !zalloc_cpumask_var(&per_cpu(cpu_sibling_mask, cpu)) ||
          !zalloc_cpumask_var(&per_cpu(cpu_core_mask, cpu)) )
-        panic("No memory for CPU sibling/core maps\n");
+        return -ENOMEM;
 
     /* A CPU is a sibling with itself and is always on its own core. */
     cpumask_set_cpu(cpu, per_cpu(cpu_sibling_mask, cpu));
     cpumask_set_cpu(cpu, per_cpu(cpu_core_mask, cpu));
+
+    return 0;
 }
 
 static void remove_cpu_sibling_map(int cpu)
@@ -292,9 +294,14 @@ smp_get_max_cpus (void)
 void __init
 smp_prepare_cpus(void)
 {
+    int rc;
+
     cpumask_copy(&cpu_present_map, &cpu_possible_map);
 
-    setup_cpu_sibling_map(0);
+    rc = setup_cpu_sibling_map(0);
+    if ( rc )
+        panic("Unable to allocate CPU sibling/core maps\n");
+
 }
 
 /* Boot the current CPU */
@@ -361,8 +368,6 @@ void start_secondary(void)
 
     set_current(idle_vcpu[cpuid]);
 
-    setup_cpu_sibling_map(cpuid);
-
     /* Run local notifiers */
     notify_cpu_starting(cpuid);
     /*
@@ -530,9 +535,19 @@ static int cpu_smpboot_callback(struct notifier_block *nfb,
                                 void *hcpu)
 {
     unsigned int cpu = (unsigned long)hcpu;
+    unsigned int rc = 0;
 
     switch ( action )
     {
+    case CPU_UP_PREPARE:
+        rc = setup_cpu_sibling_map(cpu);
+        if ( rc )
+            printk(XENLOG_ERR
+                   "Unable to allocate CPU sibling/core map  for CPU%u\n",
+                   cpu);
+
+        break;
+
     case CPU_DEAD:
         remove_cpu_sibling_map(cpu);
         break;
-- 
2.32.0

Re: [PATCH] xen/arm: smpboot: Allocate the CPU sibling/core maps while preparing the CPU

Posted by Michal Orzel 2 years, 5 months ago

Hi Julien,

On 14.06.2022 11:41, Julien Grall wrote:
> From: Julien Grall <jgrall@amazon.com>
> 
> Commit 5047cd1d5dea "xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in
> xmalloc()" extended the checks in _xmalloc() to catch any use of the
> helpers from context with interrupts disabled.
> 
> Unfortunately, the rule is not followed when allocating the CPU
> sibling/core maps.
> 
> (XEN) Xen call trace:
> (XEN)    [<00238a5c>] _xmalloc+0xfc/0x314 (PC)
> (XEN)    [<00000000>] 00000000 (LR)
> (XEN)    [<00238c8c>] _xzalloc+0x18/0x4c
> (XEN)    [<00288cb4>] smpboot.c#setup_cpu_sibling_map+0x38/0x138
> (XEN)    [<00289024>] start_secondary+0x1b4/0x270
> (XEN)    [<40010170>] 40010170
> (XEN)
> (XEN)
> (XEN) ****************************************
> (XEN) Panic on CPU 2:
> (XEN) Assertion '!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1)' failed at common/xmalloc_tlsf.c:601
> (XEN) ****************************************
> 
> This is happening because zalloc_cpumask_var() may allocate memory
> if NR_CPUS is > 2 * sizeof(unsigned long).
> 
> Avoid the problem by allocate the per-CPU IRQs while preparing the
> CPU.
Shouldn't this be "by allocating the CPU sibling/core maps while ..."
to reflect the commit title and to distinguish between this change and the IRQ one?

> 
> This also has the benefit to remove a panic() in the secondary CPU
> code.
> 
> Signed-off-by: Julien Grall <jgrall@amazon.com>
> ---
>  xen/arch/arm/smpboot.c | 25 ++++++++++++++++++++-----
>  1 file changed, 20 insertions(+), 5 deletions(-)
> 
> diff --git a/xen/arch/arm/smpboot.c b/xen/arch/arm/smpboot.c
> index 4888bcd78a5a..2b0c92cd369b 100644
> --- a/xen/arch/arm/smpboot.c
> +++ b/xen/arch/arm/smpboot.c
> @@ -79,15 +79,17 @@ DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_core_mask);
>  static bool __read_mostly opt_hmp_unsafe = false;
>  boolean_param("hmp-unsafe", opt_hmp_unsafe);
>  
> -static void setup_cpu_sibling_map(int cpu)
> +static int setup_cpu_sibling_map(int cpu)
>  {
>      if ( !zalloc_cpumask_var(&per_cpu(cpu_sibling_mask, cpu)) ||
>           !zalloc_cpumask_var(&per_cpu(cpu_core_mask, cpu)) )
> -        panic("No memory for CPU sibling/core maps\n");
> +        return -ENOMEM;
>  
>      /* A CPU is a sibling with itself and is always on its own core. */
>      cpumask_set_cpu(cpu, per_cpu(cpu_sibling_mask, cpu));
>      cpumask_set_cpu(cpu, per_cpu(cpu_core_mask, cpu));
> +
> +    return 0;
>  }
>  
>  static void remove_cpu_sibling_map(int cpu)
> @@ -292,9 +294,14 @@ smp_get_max_cpus (void)
>  void __init
>  smp_prepare_cpus(void)
>  {
> +    int rc;
Here you are leaving rc uninitialized (which is ok) but ...

> +
>      cpumask_copy(&cpu_present_map, &cpu_possible_map);
>  
> -    setup_cpu_sibling_map(0);
> +    rc = setup_cpu_sibling_map(0);
> +    if ( rc )
> +        panic("Unable to allocate CPU sibling/core maps\n");
> +
>  }
>  
>  /* Boot the current CPU */
> @@ -361,8 +368,6 @@ void start_secondary(void)
>  
>      set_current(idle_vcpu[cpuid]);
>  
> -    setup_cpu_sibling_map(cpuid);
> -
>      /* Run local notifiers */
>      notify_cpu_starting(cpuid);
>      /*
> @@ -530,9 +535,19 @@ static int cpu_smpboot_callback(struct notifier_block *nfb,
>                                  void *hcpu)
>  {
>      unsigned int cpu = (unsigned long)hcpu;
> +    unsigned int rc = 0;
... here you are setting rc to 0 even though it will be reassigned.
Furthermore, if rc is used only in case of CPU_UP_PREPARE, why not moving the definition there?

>  
>      switch ( action )
>      {
> +    case CPU_UP_PREPARE:
> +        rc = setup_cpu_sibling_map(cpu);
> +        if ( rc )
> +            printk(XENLOG_ERR
> +                   "Unable to allocate CPU sibling/core map  for CPU%u\n",
Too many spaces between 'map' and 'for'.

> +                   cpu);
> +
> +        break;
> +
>      case CPU_DEAD:
>          remove_cpu_sibling_map(cpu);
>          break;

Cheers,
Michal

Re: [PATCH] xen/arm: smpboot: Allocate the CPU sibling/core maps while preparing the CPU

Posted by Julien Grall 2 years, 5 months ago


On 14/06/2022 12:02, Michal Orzel wrote:
> Hi Julien,

Hi Michal,

> On 14.06.2022 11:41, Julien Grall wrote:
>> From: Julien Grall <jgrall@amazon.com>
>>
>> Commit 5047cd1d5dea "xen/common: Use enhanced ASSERT_ALLOC_CONTEXT in
>> xmalloc()" extended the checks in _xmalloc() to catch any use of the
>> helpers from context with interrupts disabled.
>>
>> Unfortunately, the rule is not followed when allocating the CPU
>> sibling/core maps.
>>
>> (XEN) Xen call trace:
>> (XEN)    [<00238a5c>] _xmalloc+0xfc/0x314 (PC)
>> (XEN)    [<00000000>] 00000000 (LR)
>> (XEN)    [<00238c8c>] _xzalloc+0x18/0x4c
>> (XEN)    [<00288cb4>] smpboot.c#setup_cpu_sibling_map+0x38/0x138
>> (XEN)    [<00289024>] start_secondary+0x1b4/0x270
>> (XEN)    [<40010170>] 40010170
>> (XEN)
>> (XEN)
>> (XEN) ****************************************
>> (XEN) Panic on CPU 2:
>> (XEN) Assertion '!in_irq() && (local_irq_is_enabled() || num_online_cpus() <= 1)' failed at common/xmalloc_tlsf.c:601
>> (XEN) ****************************************
>>
>> This is happening because zalloc_cpumask_var() may allocate memory
>> if NR_CPUS is > 2 * sizeof(unsigned long).
>>
>> Avoid the problem by allocate the per-CPU IRQs while preparing the
>> CPU.
> Shouldn't this be "by allocating the CPU sibling/core maps while ..."
> to reflect the commit title and to distinguish between this change and the IRQ one?

Yes. I will update it.

[...]

>>   static void remove_cpu_sibling_map(int cpu)
>> @@ -292,9 +294,14 @@ smp_get_max_cpus (void)
>>   void __init
>>   smp_prepare_cpus(void)
>>   {
>> +    int rc;
> Here you are leaving rc uninitialized (which is ok) but ...
> 
>> +
>>       cpumask_copy(&cpu_present_map, &cpu_possible_map);
>>   
>> -    setup_cpu_sibling_map(0);
>> +    rc = setup_cpu_sibling_map(0);
>> +    if ( rc )
>> +        panic("Unable to allocate CPU sibling/core maps\n");
>> +
>>   }
>>   
>>   /* Boot the current CPU */
>> @@ -361,8 +368,6 @@ void start_secondary(void)
>>   
>>       set_current(idle_vcpu[cpuid]);
>>   
>> -    setup_cpu_sibling_map(cpuid);
>> -
>>       /* Run local notifiers */
>>       notify_cpu_starting(cpuid);
>>       /*
>> @@ -530,9 +535,19 @@ static int cpu_smpboot_callback(struct notifier_block *nfb,
>>                                   void *hcpu)
>>   {
>>       unsigned int cpu = (unsigned long)hcpu;
>> +    unsigned int rc = 0;
> ... here you are setting rc to 0 even though it will be reassigned.
> Furthermore, if rc is used only in case of CPU_UP_PREPARE, why not moving the definition there?

Because I forgot to replace "return NOTIFY_DONE;" with :

return !rc ? NOTIFY_DONE : notifier_from_errno(rc);

In this case, we would need to initialize rc to 0 so it doesn't get used 
initialized.

Cheers,

-- 
Julien Grall

Re: [PATCH] xen/arm: smpboot: Allocate the CPU sibling/core maps while preparing the CPU

Posted by Michal Orzel 2 years, 5 months ago


On 14.06.2022 13:08, Julien Grall wrote:
>>> +    unsigned int rc = 0;
>> ... here you are setting rc to 0 even though it will be reassigned.
>> Furthermore, if rc is used only in case of CPU_UP_PREPARE, why not moving the definition there?
> 
> Because I forgot to replace "return NOTIFY_DONE;" with :
> 
> return !rc ? NOTIFY_DONE : notifier_from_errno(rc);
That is what I thought.
With these fixes you can add my Rb.

> 
> In this case, we would need to initialize rc to 0 so it doesn't get used initialized.
> 
> Cheers,
> 

Cheers,
Michal

Re: [PATCH] xen/arm: smpboot: Allocate the CPU sibling/core maps while preparing the CPU

Posted by Stefano Stabellini 2 years, 5 months ago

On Tue, 14 Jun 2022, Michal Orzel wrote:
> On 14.06.2022 13:08, Julien Grall wrote:
> >>> +    unsigned int rc = 0;
> >> ... here you are setting rc to 0 even though it will be reassigned.
> >> Furthermore, if rc is used only in case of CPU_UP_PREPARE, why not moving the definition there?
> > 
> > Because I forgot to replace "return NOTIFY_DONE;" with :
> > 
> > return !rc ? NOTIFY_DONE : notifier_from_errno(rc);
> That is what I thought.
> With these fixes you can add my Rb.

And also my

Acked-by: Stefano Stabellini <sstabellini@kernel.org>

Re: [PATCH] xen/arm: smpboot: Allocate the CPU sibling/core maps while preparing the CPU

Posted by Julien Grall 2 years, 5 months ago


On 15/06/2022 01:23, Stefano Stabellini wrote:
> On Tue, 14 Jun 2022, Michal Orzel wrote:
>> On 14.06.2022 13:08, Julien Grall wrote:
>>>>> +    unsigned int rc = 0;
>>>> ... here you are setting rc to 0 even though it will be reassigned.
>>>> Furthermore, if rc is used only in case of CPU_UP_PREPARE, why not moving the definition there?
>>>
>>> Because I forgot to replace "return NOTIFY_DONE;" with :
>>>
>>> return !rc ? NOTIFY_DONE : notifier_from_errno(rc);
>> That is what I thought.
>> With these fixes you can add my Rb.
> 
> And also my
> 
> Acked-by: Stefano Stabellini <sstabellini@kernel.org>

Thanks. I have committed this patch.

Cheers,

-- 
Julien Grall