Work queues and Interrupts failed

Question

Basically what I'm trying to do is to add a WorkQueue every interrupt, but after exactly 100001 interrupts, the systemm fails and the interrupt doesn't work anymore "irq 63: nobody cared (try booting with the "irqpoll" option)".

This is my code:

Inside.h file:

typedef struct    work_struct     GFA_DMA_QUEUE_Work_Struct_t;

typedef struct
{
  uint32_t      dma_length;
  uint8_t  *    dma_page_pointer;
} GFA_DMA_QUEUE_Work_Data_t;

typedef struct
{
  GFA_DMA_QUEUE_Work_Struct_t       Work;         
  GFA_DMA_QUEUE_Work_Data_t         Data;
  GFA_DMA_QUEUE_Work_t;
}

Inside .c file:

static    struct  workqueue_struct *   GFA_DMA_QUEUE_workqueue;
GFA_DMA_QUEUE_Work_t *      dma_queue_work;

create_task( dma_page_pointer,dma_length)
{
//.....
dma_queue_work = (GFA_DMA_QUEUE_Work_t *) kmalloc( sizeof(GFA_DMA_QUEUE_Work_t), __GFP_NOFAIL | GFP_KERNEL );
dma_queue_work->Data.dma_length = dma_length;
dma_queue_work->Data.dma_page_pointer = dma_page_pointer;
INIT_WORK ( (GFA_DMA_QUEUE_Work_Struct_t *)dma_queue_work, GFA_DMA_QUEUE_queue_handler);
retVal = queue_work (GFA_DMA_QUEUE_workqueue, (GFA_DMA_QUEUE_Work_Struct_t * )dma_queue_work);
}

static   void   GFA_DMA_QUEUE_queue_handler( GFA_DMA_QUEUE_Work_Struct_t *  Dma_Queue_Work )
{
  //.....
  kfree( dma_queue_work );
}

Inside the interrupt:

I just call create_tasks() function with the right parameters.

And after 100001 interrupts, I got "irq 63: nobody cared (try booting with the "irqpoll" option)". The system was working as it should be.

The time that the interrupt needs to create the task in about 6uSecs, so is not a problem.

I unmount the module and mount again and everytime that the interruption arrives to 100001 failed.

Could be a memory leak? Is the a way to check memory leaks in the kernel module?

Do you have and idea that it could be happening?

Thanks,

GFP_KERNEL within interrupt handler? Really? Almost every manual about Linux kernel programming tells that this is prohibited. You need to use GFP_ATOMIC and be ready to allocation failures. — Tsyvarev

Oscar A. Oscar A. · Accepted Answer · 2016-02-18T09:56:01

Thanks for your help, Tsyvarev, I miss that (GFP_ATOMIC).. But after 100000 interrupts something continue wrong. I deleted everything inside the interrupt and the problem continues, the only thing that could be wrong is the return value, and indeed, it was, I never return IRQ_HANDLED. That was the problem.

Work queues and Interrupts failed

1 Answers