Strange dma map/unmap/memcpy dance in ipoib_cm_handle_rx_wc

Strange dma map/unmap/memcpy dance in ipoib_cm_handle_rx_wc

[Nikolay Borisov]

Hello,

I need a bit of help in understanding the logic behind the following
code sequence in ipoib_cm_handle_rx_wc:

newskb = ipoib_cm_alloc_rx_skb(dev, rx_ring, wr_id, frags,
                                       mapping, GFP_ATOMIC);

so Here a new skb is being allocated and pages constituting its
fragments are being dma mapped.

ipoib_cm_dma_unmap_rx(priv, frags, rx_ring[wr_id].mapping);

immediately after that the very same mappings are being unmapped. And
finally the resulting dma addresses from ipoib_cm_alloc_rx_skb are being
copied via this call to memcpy:

memcpy(rx_ring[wr_id].mapping, mapping, (frags + 1) * sizeof *mapping);

Can someone explain what's the logic of doing the memcpy after the
mappings have been destroyed, doesn't that mean the dma addresses stored
in rx_ring[wr_id].mapping are invalid?

Regards,
Nikolay
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Strange dma map/unmap/memcpy dance in ipoib_cm_handle_rx_wc

[Doug Ledford]

[-- Attachment #1: Type: text/plain, Size: 1533 bytes --]

On 04/27/2016 03:48 AM, Nikolay Borisov wrote:
> Hello,
> 
> I need a bit of help in understanding the logic behind the following
> code sequence in ipoib_cm_handle_rx_wc:
> 
> newskb = ipoib_cm_alloc_rx_skb(dev, rx_ring, wr_id, frags,
>                                        mapping, GFP_ATOMIC);
> 
> so Here a new skb is being allocated and pages constituting its
> fragments are being dma mapped.

Into the local variable mapping.

> ipoib_cm_dma_unmap_rx(priv, frags, rx_ring[wr_id].mapping);
> 
> immediately after that the very same mappings are being unmapped.

No, these are the saved mappings of the skb that has the same ring slot
as the one we are going to take.

> And
> finally the resulting dma addresses from ipoib_cm_alloc_rx_skb are being
> copied via this call to memcpy:
> 
> memcpy(rx_ring[wr_id].mapping, mapping, (frags + 1) * sizeof *mapping);

From our local variable to the saved mapping slot for this ring index.

> Can someone explain what's the logic of doing the memcpy after the
> mappings have been destroyed, doesn't that mean the dma addresses stored
> in rx_ring[wr_id].mapping are invalid?
> 
> Regards,
> Nikolay
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 


-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
              GPG KeyID: 0E572FDD



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]