Re: [PATCH net-next 11/12] net: mvpp2: handle cases where more CPUs are available than s/w threads

[Antoine Tenart <antoine.tenart@bootlin.com>]

Marc,

On Mon, Oct 29, 2018 at 05:35:52PM +0000, Marc Zyngier wrote:
> On 19/09/18 10:27, Antoine Tenart wrote:
> 
> Really??? How on Earth are you testing this code?

Thank you for the feedback.

> I came up with the following fix, which fixes the issue for me.

I did not test your fix, but it looks good and does fix a real issue.
You can send it to netdev.

Antoine

> From ca25785bd1a679e72ed77e939b19360bfd0eecea Mon Sep 17 00:00:00 2001
> From: Marc Zyngier <marc.zyngier@arm.com>
> Date: Mon, 29 Oct 2018 17:07:25 +0000
> Subject: [PATCH] net: mvpp2: Fix affinity hint allocation
> 
> The mvpp2 driver has the curious behaviour of passing a stack variable
> to irq_set_affinity_hint(), which results in the kernel exploding
> the first time anyone accesses this information. News flash: userspace
> does, and irqbalance will happily take the machine down. Great stuff.
> 
> An easy fix is to track the mask within the queue_vector structure,
> and to make sure it has the same lifetime as the interrupt itself.
> 
> Fixes: e531f76757eb ("net: mvpp2: handle cases where more CPUs are available than s/w threads")
> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
> ---
>  drivers/net/ethernet/marvell/mvpp2/mvpp2.h     |  1 +
>  .../net/ethernet/marvell/mvpp2/mvpp2_main.c    | 18 ++++++++++++++----
>  2 files changed, 15 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
> index 176c6b56fdcc..398328f10743 100644
> --- a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
> +++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
> @@ -796,6 +796,7 @@ struct mvpp2_queue_vector {
>  	int nrxqs;
>  	u32 pending_cause_rx;
>  	struct mvpp2_port *port;
> +	struct cpumask *mask;
>  };
>  
>  struct mvpp2_port {
> diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
> index 14f9679c957c..7a37a37e3fb3 100644
> --- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
> +++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
> @@ -3298,24 +3298,30 @@ static int mvpp2_irqs_init(struct mvpp2_port *port)
>  	for (i = 0; i < port->nqvecs; i++) {
>  		struct mvpp2_queue_vector *qv = port->qvecs + i;
>  
> -		if (qv->type == MVPP2_QUEUE_VECTOR_PRIVATE)
> +		if (qv->type == MVPP2_QUEUE_VECTOR_PRIVATE) {
> +			qv->mask = kzalloc(cpumask_size(), GFP_KERNEL);
> +			if (!qv->mask) {
> +				err = -ENOMEM;
> +				goto err;
> +			}
> +
>  			irq_set_status_flags(qv->irq, IRQ_NO_BALANCING);
> +		}
>  
>  		err = request_irq(qv->irq, mvpp2_isr, 0, port->dev->name, qv);
>  		if (err)
>  			goto err;
>  
>  		if (qv->type == MVPP2_QUEUE_VECTOR_PRIVATE) {
> -			unsigned long mask = 0;
>  			unsigned int cpu;
>  
>  			for_each_present_cpu(cpu) {
>  				if (mvpp2_cpu_to_thread(port->priv, cpu) ==
>  				    qv->sw_thread_id)
> -					mask |= BIT(cpu);
> +					cpumask_set_cpu(cpu, qv->mask);
>  			}
>  
> -			irq_set_affinity_hint(qv->irq, to_cpumask(&mask));
> +			irq_set_affinity_hint(qv->irq, qv->mask);
>  		}
>  	}
>  
> @@ -3325,6 +3331,8 @@ static int mvpp2_irqs_init(struct mvpp2_port *port)
>  		struct mvpp2_queue_vector *qv = port->qvecs + i;
>  
>  		irq_set_affinity_hint(qv->irq, NULL);
> +		kfree(qv->mask);
> +		qv->mask = NULL;
>  		free_irq(qv->irq, qv);
>  	}
>  
> @@ -3339,6 +3347,8 @@ static void mvpp2_irqs_deinit(struct mvpp2_port *port)
>  		struct mvpp2_queue_vector *qv = port->qvecs + i;
>  
>  		irq_set_affinity_hint(qv->irq, NULL);
> +		kfree(qv->mask);
> +		qv->mask = NULL;
>  		irq_clear_status_flags(qv->irq, IRQ_NO_BALANCING);
>  		free_irq(qv->irq, qv);
>  	}
> -- 
> 2.19.1
> 
> 
> -- 
> Jazz is not dead. It just smells funny...

-- 
Antoine Ténart, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com