2021-08-19 11:57:17

by Maulik Shah

Subject: [PATCH v2 2/3] irqdomain: Fix irq_domain_trim_hierarchy()

Currently tail marker is moving along with parent domain
irq data. Fix this to initialize only once from where all
parent domain needs trimming.

Also correct the valid irq chip check.

Signed-off-by: Maulik Shah <[email protected]>
---
kernel/irq/irqdomain.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
index 19e83e9..9f6187b 100644
--- a/kernel/irq/irqdomain.c
+++ b/kernel/irq/irqdomain.c
@@ -1235,7 +1235,7 @@ static int irq_domain_trim_hierarchy(unsigned int virq)
*/
for (irqd = irq_data->parent_data; irqd; irq_data = irqd, irqd = irqd->parent_data) {
/* Can't have a valid irqchip after a trim marker */
- if (irqd->chip && tail)
+ if (!IS_ERR(irqd->chip) && tail)
return -EINVAL;

/* Can't have an empty irqchip before a trim marker */
@@ -1247,7 +1247,8 @@ static int irq_domain_trim_hierarchy(unsigned int virq)
if (PTR_ERR(irqd->chip) != -ENOTCONN)
return -EINVAL;

- tail = irq_data;
+ if (!tail)
+ tail = irq_data;
}
}

--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation

2021-08-20 15:56:31

by Marc Zyngier

Subject: Re: [PATCH v2 2/3] irqdomain: Fix irq_domain_trim_hierarchy()

On Thu, 19 Aug 2021 12:53:12 +0100,
Maulik Shah <[email protected]> wrote:
>
> Currently tail marker is moving along with parent domain
> irq data. Fix this to initialize only once from where all
> parent domain needs trimming.
>
> Also correct the valid irq chip check.
>
> Signed-off-by: Maulik Shah <[email protected]>
> ---
> kernel/irq/irqdomain.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
> index 19e83e9..9f6187b 100644
> --- a/kernel/irq/irqdomain.c
> +++ b/kernel/irq/irqdomain.c
> @@ -1235,7 +1235,7 @@ static int irq_domain_trim_hierarchy(unsigned int virq)
> */
> for (irqd = irq_data->parent_data; irqd; irq_data = irqd, irqd = irqd->parent_data) {
> /* Can't have a valid irqchip after a trim marker */
> - if (irqd->chip && tail)
> + if (!IS_ERR(irqd->chip) && tail)
> return -EINVAL;
>
> /* Can't have an empty irqchip before a trim marker */
> @@ -1247,7 +1247,8 @@ static int irq_domain_trim_hierarchy(unsigned int virq)
> if (PTR_ERR(irqd->chip) != -ENOTCONN)
> return -EINVAL;
>
> - tail = irq_data;
> + if (!tail)
> + tail = irq_data;
> }
> }

I think you have the wrong end of the stick. 'tail' represent the
*unique* point in the hierarchy where you can have a trim marker:

(1) If there is a valid irqchip after a trim marker, this is wrong
(2) If there is a trim marker after another trim marker, this is wrong
(3) If there is a NULL irqchip before a trim marker, this is wrong
(4) If there is an error that isn't a trim marker, this is wrong

(1) and (2) are captured by:
if (irqd->chip && tail)
(3) is captured by:
if (!irqd->chip && !tail)
(4) is captured by:
if (IS_ERR(irqd->chip)) {
/* Only -ENOTCONN is a valid trim marker */
if (PTR_ERR(irqd->chip) != -ENOTCONN)

The expected usage is that:

- there is a single potential trim marker in the hierarchy
- all the irqd->chip pointers below the marker are NULL
- all the irqd->chip before the marker are neither NULL nor an error

I don't see any bug here.

Thanks,

M.

--
Without deviation from the norm, progress is not possible.