On Tue, Dec 9, 2008 at 4:03 PM, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:
> Is the put_page() here really correct? The original reference in the skb
> will be still around, won't it?
But you didn't quote the part with the alloc_page call. Here's the
whole function.
> static int fcoe_get_paged_crc_eof(struct sk_buff *skb, int tlen)
> {
> struct fcoe_percpu_s *fps;
> struct page *page;
> int cpu_idx;
>
> cpu_idx = get_cpu();
> fps = fcoe_percpu[cpu_idx];
> page = fps->crc_eof_page;
> if (!page) {
> page = alloc_page(GFP_ATOMIC);
> if (!page) {
> put_cpu();
> return -ENOMEM;
> }
> fps->crc_eof_page = page;
> WARN_ON(fps->crc_eof_offset != 0);
> }
>
> get_page(page);
> skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, page,
> fps->crc_eof_offset, tlen);
> skb->len += tlen;
> skb->data_len += tlen;
> skb->truesize += tlen;
> fps->crc_eof_offset += sizeof(struct fcoe_crc_eof);
>
> if (fps->crc_eof_offset >= PAGE_SIZE) {
> fps->crc_eof_page = NULL;
> fps->crc_eof_offset = 0;
> put_page(page);
> }
> put_cpu();
> return 0;
> }
FCoE needs to attach an 8 byte trailer to all FC frames created by
libfc before transmitting. If the skb is non-linear, say a SCSI write
data frame, the only way to do that is to attach a fragment page to
the skb.
In order to avoid repeated calls to alloc_page when only 8 bytes are
needed, the fcoe module keeps a reference to the page using it
repeatedly until it reaches the end. Then then call to put_page
happens, and a new page is allocated on the next transmit. At this
point the page may still have references from any skbs that are still
outstanding (get_page before skb_fill_page_desc). With 4k pages 512
fragmented frames can be sent before calling alloc_page again.
- Chris
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
