Re: [PATCH 01/11] irqchip/gic-v3-its: Free collection mapping on device teardown

[Zenghui Yu]

Hi Marc,

On 2019/11/6 0:22, Marc Zyngier wrote:
> Somehow, we forgot to free the collection mapping when tearing down
> a device, hence slowly leaking mapping arrays as devices get removed
> from the system. That is, almost never.
>
> Just to be safe, properly free the array on device teardown.
>
> Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
> ---
>   drivers/irqchip/irq-gic-v3-its.c | 1 +
>   1 file changed, 1 insertion(+)
>
> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> index 787e8eec9a7f..07d0bde60e16 100644
> --- a/drivers/irqchip/irq-gic-v3-its.c
> +++ b/drivers/irqchip/irq-gic-v3-its.c
> @@ -2471,6 +2471,7 @@ static void its_free_device(struct its_device *its_dev)
>   	raw_spin_lock_irqsave(&its_dev->its->lock, flags);
>   	list_del(&its_dev->entry);
>   	raw_spin_unlock_irqrestore(&its_dev->its->lock, flags);
> +	kfree(its_dev->event_map.col_map);

I agreed that this is the appropriate place to free the collection
mapping (act as the counterpart of the allocation which happened in
its_create_device).  But as pointed out by Heyi [1], this will
introduce a double free issue.  We'd better also drop the kfree()
in its_irq_domain_free() in this patch?

(I find that it had been dropped in the last patch in your
irq/gic-5.5-wip branch, but maybe better here.)


[1] https://lkml.org/lkml/2019/7/15/329


Thanks,
Zenghui

>   	kfree(its_dev->itt);
>   	kfree(its_dev);
>   }