lkml.org 
[lkml]   [2020]   [Nov]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC 6/9] staging: dpaa2-switch: add .ndo_start_xmit() callback
> +static int dpaa2_switch_build_single_fd(struct ethsw_core *ethsw,
> + struct sk_buff *skb,
> + struct dpaa2_fd *fd)
> +{
> + struct device *dev = ethsw->dev;
> + struct sk_buff **skbh;
> + dma_addr_t addr;
> + u8 *buff_start;
> + void *hwa;
> +
> + buff_start = PTR_ALIGN(skb->data - DPAA2_SWITCH_TX_DATA_OFFSET -
> + DPAA2_SWITCH_TX_BUF_ALIGN,
> + DPAA2_SWITCH_TX_BUF_ALIGN);
> +
> + /* Clear FAS to have consistent values for TX confirmation. It is
> + * located in the first 8 bytes of the buffer's hardware annotation
> + * area
> + */
> + hwa = buff_start + DPAA2_SWITCH_SWA_SIZE;
> + memset(hwa, 0, 8);
> +
> + /* Store a backpointer to the skb at the beginning of the buffer
> + * (in the private data area) such that we can release it
> + * on Tx confirm
> + */
> + skbh = (struct sk_buff **)buff_start;
> + *skbh = skb;

Where is the TX confirm which uses this stored pointer. I don't see it
in this file.

It can be expensive to store pointer like this in buffers used for
DMA. It has to be flushed out of the cache here as part of the
send. Then the TX complete needs to invalidate and then read it back
into the cache. Or you use coherent memory which is just slow.

It can be cheaper to keep a parallel ring in cacheable memory which
never gets flushed.

Andrew

\
 
 \ /
  Last update: 2020-11-05 02:05    [W:0.115 / U:0.196 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site