[PATCH] xen: Enable interrupts when calling _cond_resched()

[PATCH] xen: Enable interrupts when calling _cond_resched()

[Thomas Gleixner]

xen_maybe_preempt_hcall() is called from the exception entry point
xen_do_hypervisor_callback with interrupts disabled.

_cond_resched() evades the might_sleep() check in cond_resched() which
would have caught that and schedule_debug() unfortunately lacks a check
for irqs_disabled().

Enable interrupts around the call and use cond_resched() to catch future
issues.

Fixes: fdfd811ddde3 ("x86/xen: allow privcmd hypercalls to be preempted")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 drivers/xen/preempt.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

--- a/drivers/xen/preempt.c
+++ b/drivers/xen/preempt.c
@@ -33,8 +33,10 @@ asmlinkage __visible void xen_maybe_pree
 		 * cpu.
 		 */
 		__this_cpu_write(xen_in_preemptible_hcall, false);
-		_cond_resched();
+		local_irq_enable();
+		cond_resched();
 		__this_cpu_write(xen_in_preemptible_hcall, true);
+		local_irq_disable();
 	}
 }
 #endif /* CONFIG_PREEMPTION */

Re: [PATCH] xen: Enable interrupts when calling _cond_resched()

[Jürgen Groß]

On 19.02.20 12:01, Thomas Gleixner wrote:
> xen_maybe_preempt_hcall() is called from the exception entry point
> xen_do_hypervisor_callback with interrupts disabled.
> 
> _cond_resched() evades the might_sleep() check in cond_resched() which
> would have caught that and schedule_debug() unfortunately lacks a check
> for irqs_disabled().
> 
> Enable interrupts around the call and use cond_resched() to catch future
> issues.
> 
> Fixes: fdfd811ddde3 ("x86/xen: allow privcmd hypercalls to be preempted")
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
>   drivers/xen/preempt.c |    4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> --- a/drivers/xen/preempt.c
> +++ b/drivers/xen/preempt.c
> @@ -33,8 +33,10 @@ asmlinkage __visible void xen_maybe_pree
>   		 * cpu.
>   		 */
>   		__this_cpu_write(xen_in_preemptible_hcall, false);
> -		_cond_resched();
> +		local_irq_enable();
> +		cond_resched();
>   		__this_cpu_write(xen_in_preemptible_hcall, true);
> +		local_irq_disable();

Could you please put the call of local_irq_disable() directly after the
cond_resched() call to make the result symmetric regarding writing of
xen_in_preemptible_hcall and irq enable/disable?


Juergen

Re: [PATCH] xen: Enable interrupts when calling _cond_resched()

[Thomas Gleixner]

Jürgen Groß <jgross@suse.com> writes:

> On 19.02.20 12:01, Thomas Gleixner wrote:
>> xen_maybe_preempt_hcall() is called from the exception entry point
>> xen_do_hypervisor_callback with interrupts disabled.
>> 
>> _cond_resched() evades the might_sleep() check in cond_resched() which
>> would have caught that and schedule_debug() unfortunately lacks a check
>> for irqs_disabled().
>> 
>> Enable interrupts around the call and use cond_resched() to catch future
>> issues.
>> 
>> Fixes: fdfd811ddde3 ("x86/xen: allow privcmd hypercalls to be preempted")
>> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
>> ---
>>   drivers/xen/preempt.c |    4 +++-
>>   1 file changed, 3 insertions(+), 1 deletion(-)
>> 
>> --- a/drivers/xen/preempt.c
>> +++ b/drivers/xen/preempt.c
>> @@ -33,8 +33,10 @@ asmlinkage __visible void xen_maybe_pree
>>   		 * cpu.
>>   		 */
>>   		__this_cpu_write(xen_in_preemptible_hcall, false);
>> -		_cond_resched();
>> +		local_irq_enable();
>> +		cond_resched();
>>   		__this_cpu_write(xen_in_preemptible_hcall, true);
>> +		local_irq_disable();
>
> Could you please put the call of local_irq_disable() directly after the
> cond_resched() call to make the result symmetric regarding writing of
> xen_in_preemptible_hcall and irq enable/disable?

Darn. Of course does this need to be there. Stupid me, let me resend.

[PATCH V2] xen: Enable interrupts when calling _cond_resched()

[Thomas Gleixner]

xen_maybe_preempt_hcall() is called from the exception entry point
xen_do_hypervisor_callback with interrupts disabled.

_cond_resched() evades the might_sleep() check in cond_resched() which
would have caught that and schedule_debug() unfortunately lacks a check
for irqs_disabled().

Enable interrupts around the call and use cond_resched() to catch future
issues.

Fixes: fdfd811ddde3 ("x86/xen: allow privcmd hypercalls to be preempted")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Put local_irq_disable() where it belongs.
---
 drivers/xen/preempt.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

--- a/drivers/xen/preempt.c
+++ b/drivers/xen/preempt.c
@@ -33,7 +33,9 @@ asmlinkage __visible void xen_maybe_pree
 		 * cpu.
 		 */
 		__this_cpu_write(xen_in_preemptible_hcall, false);
-		_cond_resched();
+		local_irq_enable();
+		cond_resched();
+		local_irq_disable();
 		__this_cpu_write(xen_in_preemptible_hcall, true);
 	}
 }

Re: [PATCH V2] xen: Enable interrupts when calling _cond_resched()

[Jürgen Groß]

On 19.02.20 18:30, Thomas Gleixner wrote:
> xen_maybe_preempt_hcall() is called from the exception entry point
> xen_do_hypervisor_callback with interrupts disabled.
> 
> _cond_resched() evades the might_sleep() check in cond_resched() which
> would have caught that and schedule_debug() unfortunately lacks a check
> for irqs_disabled().
> 
> Enable interrupts around the call and use cond_resched() to catch future
> issues.
> 
> Fixes: fdfd811ddde3 ("x86/xen: allow privcmd hypercalls to be preempted")
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Reviewed-by: Juergen Gross <jgross@suse.com>


Juergen