2018-12-20 16:10:10

by Srinivas Ramana

[permalink] [raw]
Subject: [PATCH] genirq: Make sure the affinity to be set is not empty

If the default_affinity is managed/initialized by a system and
all the cpus present in irq_default_affinity are hotplugged out
we may end up passing a cpumask of 0 to irq_do_set_affinity().

Fix this by falling back to cpu_online_mask in case the
calculated affinity becomes zero.

Signed-off-by: Srinivas Ramana <[email protected]>
---
kernel/irq/manage.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index 9dbdccab3b6a..5c0ba5ca5930 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -393,6 +393,9 @@ int irq_setup_affinity(struct irq_desc *desc)
}

cpumask_and(&mask, cpu_online_mask, set);
+ if (cpumask_empty(&mask))
+ cpumask_copy(&mask, cpu_online_mask);
+
if (node != NUMA_NO_NODE) {
const struct cpumask *nodemask = cpumask_of_node(node);

--
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc.,
is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.



2019-01-02 08:05:27

by Srinivas Ramana

[permalink] [raw]
Subject: Re: [PATCH] genirq: Make sure the affinity to be set is not empty

On 12/20/2018 7:05 PM, Srinivas Ramana wrote:
> If the default_affinity is managed/initialized by a system and
> all the cpus present in irq_default_affinity are hotplugged out
> we may end up passing a cpumask of 0 to irq_do_set_affinity().
>
> Fix this by falling back to cpu_online_mask in case the
> calculated affinity becomes zero.
>
> Signed-off-by: Srinivas Ramana <[email protected]>
> ---
> kernel/irq/manage.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> index 9dbdccab3b6a..5c0ba5ca5930 100644
> --- a/kernel/irq/manage.c
> +++ b/kernel/irq/manage.c
> @@ -393,6 +393,9 @@ int irq_setup_affinity(struct irq_desc *desc)
> }
>
> cpumask_and(&mask, cpu_online_mask, set);
> + if (cpumask_empty(&mask))
> + cpumask_copy(&mask, cpu_online_mask);
> +
> if (node != NUMA_NO_NODE) {
> const struct cpumask *nodemask = cpumask_of_node(node);
>
>

Hi,

Can you provide your feedback on this?

Thanks,
-- Srinivas R

--
Qualcomm India Private Limited, on behalf of Qualcomm Innovation
Center, Inc., is a member of Code Aurora Forum, a Linux Foundation
Collaborative Project

Subject: [tip:irq/urgent] genirq: Make sure the initial affinity is not empty

Commit-ID: bddda606ec76550dd63592e32a6e87e7d32583f7
Gitweb: https://git.kernel.org/tip/bddda606ec76550dd63592e32a6e87e7d32583f7
Author: Srinivas Ramana <[email protected]>
AuthorDate: Thu, 20 Dec 2018 19:05:57 +0530
Committer: Thomas Gleixner <[email protected]>
CommitDate: Tue, 15 Jan 2019 11:23:27 +0100

genirq: Make sure the initial affinity is not empty

If all CPUs in the irq_default_affinity mask are offline when an interrupt
is initialized then irq_setup_affinity() can set an empty affinity mask for
a newly allocated interrupt.

Fix this by falling back to cpu_online_mask in case the resulting affinity
mask is zero.

Signed-off-by: Srinivas Ramana <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Cc: [email protected]
Link: https://lkml.kernel.org/r/[email protected]

---
kernel/irq/manage.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index a4888ce4667a..84b54a17b95d 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -393,6 +393,9 @@ int irq_setup_affinity(struct irq_desc *desc)
}

cpumask_and(&mask, cpu_online_mask, set);
+ if (cpumask_empty(&mask))
+ cpumask_copy(&mask, cpu_online_mask);
+
if (node != NUMA_NO_NODE) {
const struct cpumask *nodemask = cpumask_of_node(node);