Re: [net-next RFC v3 7/7] net: ravb: Allocate RX buffers via page pool

[Paul Barker]

On 15/04/2024 13:16, Niklas Söderlund wrote:
> Hi Paul,
> 
> I think using page pool is a good idea!
> 
> On 2024-04-15 10:48:04 +0100, Paul Barker wrote:
>> This patch makes multiple changes that can't be separated:
>>
>>   1) Allocate plain RX buffers via a page pool instead of allocating
>>      SKBs, then use build_skb() when a packet is received.
>>   2) For GbEth IP, reduce the RX buffer size to 2kB.
>>   3) For GbEth IP, merge packets which span more than one RX descriptor
>>      as SKB fragments instead of copying data.
>>
>> Implementing (1) without (2) would require the use of an order-1 page
>> pool (instead of an order-0 page pool split into page fragments) for
>> GbEth.
>>
>> Implementing (2) without (3) would leave us no space to re-assemble
>> packets which span more than one RX descriptor.
>>
>> Implementing (3) without (1) would not be possible as the network stack
>> expects to use put_page() or page_pool_put_page() to free SKB fragments
>> after an SKB is consumed.
>>
>> This patch gives the following improvements during testing with iperf3.
>>
>>   * RZ/G2L:
>>     * TCP RX: same bandwidth at -43% CPU load (70% -> 40%)
>>     * UDP RX: same bandwidth at -17% CPU load (88% -> 74%)
>>
>>   * RZ/G2UL:
>>     * TCP RX: +30% bandwidth (726Mbps -> 941Mbps)
>>     * UDP RX: +417% bandwidth (108Mbps -> 558Mbps)
>>
>>   * RZ/G3S:
>>     * TCP RX: +64% bandwidth (562Mbps -> 920Mbps)
>>     * UDP RX: +420% bandwidth (90Mbps -> 468Mbps)
>>
>>   * RZ/Five:
>>     * TCP RX: +217% bandwidth (145Mbps -> 459Mbps)
>>     * UDP RX: +470% bandwidth (20Mbps -> 114Mbps)
>>
>> There is no significant impact on bandwidth or CPU load in testing on
>> RZ/G2H or R-Car M3N.
>>
>> Signed-off-by: Paul Barker <paul.barker.ct@xxxxxxxxxxxxxx>
>> ---
>>  drivers/net/ethernet/renesas/ravb.h      |  10 +-
>>  drivers/net/ethernet/renesas/ravb_main.c | 209 +++++++++++++----------
>>  2 files changed, 128 insertions(+), 91 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/renesas/ravb.h b/drivers/net/ethernet/renesas/ravb.h
>> index 9c6392ade2f1..4348366c3dc7 100644
>> --- a/drivers/net/ethernet/renesas/ravb.h
>> +++ b/drivers/net/ethernet/renesas/ravb.h
>> @@ -1050,8 +1050,8 @@ struct ravb_hw_info {
>>  	netdev_features_t net_features;
>>  	int stats_len;
>>  	u32 tccr_mask;
>> +	u32 rx_buffer_size;
>>  	u32 rx_max_frame_size;
>> -	u32 rx_max_desc_use;
>>  	u32 rx_desc_size;
>>  	unsigned aligned_tx: 1;
>>  	unsigned needs_irq_coalesce:1;	/* Needs software IRQ coalescing */
>> @@ -1071,6 +1071,11 @@ struct ravb_hw_info {
>>  	unsigned half_duplex:1;		/* E-MAC supports half duplex mode */
>>  };
>>  
>> +struct ravb_rx_buffer {
>> +	struct page *page;
>> +	unsigned int offset;
>> +};
>> +
>>  struct ravb_private {
>>  	struct net_device *ndev;
>>  	struct platform_device *pdev;
>> @@ -1094,7 +1099,8 @@ struct ravb_private {
>>  	struct ravb_tx_desc *tx_ring[NUM_TX_QUEUE];
>>  	void *tx_align[NUM_TX_QUEUE];
>>  	struct sk_buff *rx_1st_skb;
>> -	struct sk_buff **rx_skb[NUM_RX_QUEUE];
>> +	struct page_pool *rx_pool;
> 
> Don't we need a page pool per queue? Else multiple calls to 
> ravb_ring_init() and ravb_ring_free() for different queues will 
> otherwise risk allocating over a previous queue and multiple free the 
> same one.

Ack.

> 
>> +	struct ravb_rx_buffer *rx_buffers[NUM_RX_QUEUE];
>>  	struct sk_buff **tx_skb[NUM_TX_QUEUE];
>>  	u32 rx_over_errors;
>>  	u32 rx_fifo_errors;
>> diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c
>> index 7434faf0820c..892a3eadef1e 100644
>> --- a/drivers/net/ethernet/renesas/ravb_main.c
>> +++ b/drivers/net/ethernet/renesas/ravb_main.c
>> @@ -30,6 +30,7 @@
>>  #include <linux/reset.h>
>>  #include <linux/math64.h>
>>  #include <net/ip.h>
>> +#include <net/page_pool/helpers.h>
>>  
>>  #include "ravb.h"
>>  
>> @@ -113,25 +114,6 @@ static void ravb_set_rate_rcar(struct net_device *ndev)
>>  	}
>>  }
>>  
>> -static struct sk_buff *
>> -ravb_alloc_skb(struct net_device *ndev, const struct ravb_hw_info *info,
>> -	       gfp_t gfp_mask)
>> -{
>> -	struct sk_buff *skb;
>> -	u32 reserve;
>> -
>> -	skb = __netdev_alloc_skb(ndev, info->rx_max_frame_size + RAVB_ALIGN - 1,
>> -				 gfp_mask);
>> -	if (!skb)
>> -		return NULL;
>> -
>> -	reserve = (unsigned long)skb->data & (RAVB_ALIGN - 1);
>> -	if (reserve)
>> -		skb_reserve(skb, RAVB_ALIGN - reserve);
>> -
>> -	return skb;
>> -}
>> -
>>  /* Get MAC address from the MAC address registers
>>   *
>>   * Ethernet AVB device doesn't have ROM for MAC address.
>> @@ -257,21 +239,10 @@ static void ravb_rx_ring_free(struct net_device *ndev, int q)
>>  {
>>  	struct ravb_private *priv = netdev_priv(ndev);
>>  	unsigned int ring_size;
>> -	unsigned int i;
>>  
>>  	if (!priv->rx_ring[q].raw)
>>  		return;
>>  
>> -	for (i = 0; i < priv->num_rx_ring[q]; i++) {
>> -		struct ravb_rx_desc *desc = ravb_rx_get_desc(priv, q, i);
>> -
>> -		if (!dma_mapping_error(ndev->dev.parent,
>> -				       le32_to_cpu(desc->dptr)))
>> -			dma_unmap_single(ndev->dev.parent,
>> -					 le32_to_cpu(desc->dptr),
>> -					 priv->info->rx_max_frame_size,
>> -					 DMA_FROM_DEVICE);
>> -	}
>>  	ring_size = priv->info->rx_desc_size * (priv->num_rx_ring[q] + 1);
>>  	dma_free_coherent(ndev->dev.parent, ring_size, priv->rx_ring[q].raw,
>>  			  priv->rx_desc_dma[q]);
>> @@ -298,13 +269,14 @@ static void ravb_ring_free(struct net_device *ndev, int q)
>>  		priv->tx_ring[q] = NULL;
>>  	}
>>  
>> -	/* Free RX skb ringbuffer */
>> -	if (priv->rx_skb[q]) {
>> -		for (i = 0; i < priv->num_rx_ring[q]; i++)
>> -			dev_kfree_skb(priv->rx_skb[q][i]);
>> +	/* Free RX buffers */
>> +	for (i = 0; i < priv->num_rx_ring[q]; i++) {
>> +		if (priv->rx_buffers[q][i].page)
>> +			page_pool_put_page(priv->rx_pool, priv->rx_buffers[q][i].page, 0, true);
>>  	}
>> -	kfree(priv->rx_skb[q]);
>> -	priv->rx_skb[q] = NULL;
>> +	kfree(priv->rx_buffers[q]);
>> +	priv->rx_buffers[q] = NULL;
>> +	page_pool_destroy(priv->rx_pool);
>>  
>>  	/* Free aligned TX buffers */
>>  	kfree(priv->tx_align[q]);
>> @@ -317,35 +289,54 @@ static void ravb_ring_free(struct net_device *ndev, int q)
>>  	priv->tx_skb[q] = NULL;
>>  }
>>  
>> +static int
>> +ravb_alloc_rx_buffer(struct net_device *ndev, int q, u32 entry, gfp_t gfp_mask,
>> +		     __le32 *dptr)
> 
> Why not pass the struct ravb_rx_desc instead of a dptr? Then the 
> function can deal with the error case and fill in rx_desc->dptr and 
> rx_desc->ds_cc directly making the caller simpler.

Ack.

> 
>> +{
>> +	struct ravb_private *priv = netdev_priv(ndev);
>> +	const struct ravb_hw_info *info = priv->info;
>> +	struct ravb_rx_buffer *rx_buff = &priv->rx_buffers[q][entry];
>> +	dma_addr_t dma_addr;
>> +	unsigned int size;
>> +
>> +	size = info->rx_buffer_size;
>> +	rx_buff->page = page_pool_alloc(priv->rx_pool, &rx_buff->offset, &size,
>> +					gfp_mask);
>> +	if (unlikely(!rx_buff->page))
>> +		return -ENOMEM;
>> +
>> +	dma_addr = page_pool_get_dma_addr(rx_buff->page) + rx_buff->offset;
>> +	dma_sync_single_for_device(ndev->dev.parent, dma_addr,
>> +				   info->rx_buffer_size, DMA_FROM_DEVICE);
>> +	*dptr = cpu_to_le32(dma_addr);
>> +	return 0;
>> +}
>> +
>>  static u32
>>  ravb_rx_ring_refill(struct net_device *ndev, int q, u32 count, gfp_t gfp_mask)
>>  {
>>  	struct ravb_private *priv = netdev_priv(ndev);
>>  	const struct ravb_hw_info *info = priv->info;
>>  	struct ravb_rx_desc *rx_desc;
>> -	dma_addr_t dma_addr;
>>  	u32 i, entry;
>>  
>>  	for (i = 0; i < count; i++) {
>>  		entry = (priv->dirty_rx[q] + i) % priv->num_rx_ring[q];
>>  		rx_desc = ravb_rx_get_desc(priv, q, entry);
>> -		rx_desc->ds_cc = cpu_to_le16(info->rx_max_desc_use);
>>  
>> -		if (!priv->rx_skb[q][entry]) {
>> -			priv->rx_skb[q][entry] = ravb_alloc_skb(ndev, info, gfp_mask);
>> -			if (!priv->rx_skb[q][entry])
>> -				break;
>> -			dma_addr = dma_map_single(ndev->dev.parent,
>> -						  priv->rx_skb[q][entry]->data,
>> -						  priv->info->rx_max_frame_size,
>> -						  DMA_FROM_DEVICE);
>> -			skb_checksum_none_assert(priv->rx_skb[q][entry]);
>> -			/* We just set the data size to 0 for a failed mapping
>> -			 * which should prevent DMA from happening...
>> -			 */
>> -			if (dma_mapping_error(ndev->dev.parent, dma_addr))
>> +		if (!priv->rx_buffers[q][entry].page) {
>> +			if (unlikely(ravb_alloc_rx_buffer(ndev, q, entry, gfp_mask,
>> +							  &rx_desc->dptr))) {
>> +				/* We just set the data size to 0 for a failed mapping
>> +				 * which should prevent DMA from happening...
>> +				 */
>>  				rx_desc->ds_cc = cpu_to_le16(0);
>> -			rx_desc->dptr = cpu_to_le32(dma_addr);
>> +				break;
>> +			}
>> +
>> +			rx_desc->ds_cc = cpu_to_le16(info->rx_buffer_size
>> +						     - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
>> +						     - ETH_FCS_LEN + sizeof(__sum16));
> 
> Can a comment be added to why we subtract and add things to the size?

Ack.

I'll address these in v4.

-- 
Paul Barker

Attachment: OpenPGP_0x27F4B3459F002257.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature