Re: [BUG] Possible unsafe page_pool usage in octeontx2

[Jesper Dangaard Brouer <jbrouer@redhat.com>]

On 28/08/2023 13.07, Alexander Lobakin wrote:
>> This can be a workaround fix:
>>
>> $ git diff
>> diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
>> b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
>> index dce3cea00032..ab7ca146fddf 100644
>> --- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
>> +++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
>> @@ -578,6 +578,10 @@ int otx2_alloc_buffer(struct otx2_nic *pfvf, struct
>> otx2_cq_queue *cq,
>>                  struct refill_work *work;
>>                  struct delayed_work *dwork;
>>
>> +               /* page_pool alloc API cannot be used from WQ */
>> +               if (cq->rbpool->page_pool)
>> +                       return -ENOMEM;
> I believe that breaks the driver?
> 

Why would that break the driver?

AFAIK returning 0 here will break the driver.
We need to return something non-zero, see otx2_refill_pool_ptrs() 
copy-pasted below signature.


>> +
>>                  work = &pfvf->refill_wrk[cq->cq_idx];
>>                  dwork = &work->pool_refill_work;
>>                  /* Schedule a task if no other task is running */


--Jesper

  void otx2_refill_pool_ptrs(void *dev, struct otx2_cq_queue *cq)
  {
	struct otx2_nic *pfvf = dev;
	dma_addr_t bufptr;

	while (cq->pool_ptrs) {
		if (otx2_alloc_buffer(pfvf, cq, &bufptr))
			break;
		otx2_aura_freeptr(pfvf, cq->cq_idx, bufptr + OTX2_HEAD_ROOM);
		cq->pool_ptrs--;
	}
  }