Re: [dpdk-dev] [PATCH] app/testpmd: remove most uses of rte_eth_devices

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 7/15/2021 2:20 PM, Paulis Gributs wrote:
> This patch removes most uses of the global variable rte_eth_devices
> from testpmd. This was done to avoid using the object directly which
> applications should not do.
> 
> Most uses have been replaced with standard function calls, however
> the use of it in the show_macs function could not be replaced as no
> function call exists to get all mac addresses of a given port.
> 
> Signed-off-by: Paulis Gributs <paulis.gributs@intel.com>
> ---
>  app/test-pmd/testpmd.c | 80 +++++++++++++++++++++++++++++-------------
>  app/test-pmd/txonly.c  | 18 ++++++++--
>  2 files changed, 71 insertions(+), 27 deletions(-)
> 
> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> index 1cdd3cdd12..42d9804dab 100644
> --- a/app/test-pmd/testpmd.c
> +++ b/app/test-pmd/testpmd.c
> @@ -857,16 +857,23 @@ dma_unmap_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
>  	int ret;
>  
>  	RTE_ETH_FOREACH_DEV(pid) {
> -		struct rte_eth_dev *dev =
> -			&rte_eth_devices[pid];
> +		struct rte_eth_dev_info dev_info;
>  
> -		ret = rte_dev_dma_unmap(dev->device, memhdr->addr, 0,
> -					memhdr->len);
> +		ret = eth_dev_info_get_print_err(pid, &dev_info);
> +		if (ret != 0) {
> +			TESTPMD_LOG(DEBUG,
> +				    "unable to get device info for port %d on addr 0x%p,"
> +				    "mempool unmapping will not be performed\n",
> +				    pid, memhdr->addr);
> +			continue;
> +		}
> +
> +		ret = rte_dev_dma_unmap(dev_info.device, memhdr->addr, 0, memhdr->len);
>  		if (ret) {
>  			TESTPMD_LOG(DEBUG,
>  				    "unable to DMA unmap addr 0x%p "
>  				    "for device %s\n",
> -				    memhdr->addr, dev->data->name);
> +				    memhdr->addr, dev_info.device->name);
>  		}
>  	}
>  	ret = rte_extmem_unregister(memhdr->addr, memhdr->len);
> @@ -892,16 +899,22 @@ dma_map_cb(struct rte_mempool *mp __rte_unused, void *opaque __rte_unused,
>  		return;
>  	}
>  	RTE_ETH_FOREACH_DEV(pid) {
> -		struct rte_eth_dev *dev =
> -			&rte_eth_devices[pid];
> +		struct rte_eth_dev_info dev_info;
>  
> -		ret = rte_dev_dma_map(dev->device, memhdr->addr, 0,
> -				      memhdr->len);
> +		ret = eth_dev_info_get_print_err(pid, &dev_info);
> +		if (ret != 0) {
> +			TESTPMD_LOG(DEBUG,
> +				    "unable to get device info for port %d on addr 0x%p,"
> +				    "mempool mapping will not be performed\n",
> +				    pid, memhdr->addr);
> +			continue;
> +		}
> +		ret = rte_dev_dma_map(dev_info.device, memhdr->addr, 0, memhdr->len);
>  		if (ret) {
>  			TESTPMD_LOG(DEBUG,
>  				    "unable to DMA map addr 0x%p "
>  				    "for device %s\n",
> -				    memhdr->addr, dev->data->name);
> +				    memhdr->addr, dev_info.device->name);
>  		}
>  	}
>  }

Hi Shahaf,

These callbacks are used to map/unmap anon memory and added on commit [1].

Can you please elaborate why it is required? And does xmem covers this
functionality already?

The concern I have is, it uses some DPDK details, like rte_device to implement
functionality in a test applications (testpmd). If this is a required
functionality for end user, it is very hard for them to implement this, and
perhaps we should have some APIs/wrappers to help the users in that case.
Or if it is not required, we can perhaps drop from testpmd.

But first I am trying to understand what functionality it brings, if it is
something required by end user or not.

Thanks,
ferruh

[1]
Commit 3a0968c828a6 ("app/testpmd: map anonymous memory for devices")