Re: [lwip-users] PBUF_RAM allocation

[address@hidden]

On 28.06.2018 15:10, Amena El Homsi wrote:

I went over LwIP code and the only memory error I saw is that when p->payload ptr < pbuf ptr. Since our data memory addresses are always less than the frame memory addresses we will not face this issue because p->payload ptr will always be greater than pbuf ptr.

For PBUF_POOL and PBUF_RAM pbufs, pbuf_header() (etc.) will happily add more header room unless p->payload will grow down into struct pbuf. I told you it won't work like you said it would. But you can of course still try if you want.

In a separate note, even if I used LwIP without any change some frames are not contiguous

Wait, you seem to misunderstand this. Pbufs have a payload area and a payload pointer. For PBUF_POOL and PBUF_RAM, the payload area begins right behind the struct pbuf. The payload pointer can point at any point in the payload area. You can never anticipate it by adding some magic counter to the struct pbuf pointer.

But you are right that pbuf allocation for UDP wastes some space because room for TCP headers is reserved. This is on my todo list. Also, in dual stack mode, IPv4 has the same problem as room for IPv6 headers is reserved.

In addition, LwIP doesn't consider the retransmission at the MAC layer (if 802.11 is used). LwIP calls low_level_output() and directly calls pbuf_free() which will deallocates the frame. That's why I have to add a minor modification to pbuf_free() function.

No. The short form is: you need to call pbuf_ref() if transmission is *not* synchronously finished before low_level_output() returns. Then you can later call pbuf_free() when you are done with the pbuf. To be on the safe side, you might still need to duplicate the pbuf if PBUF_NEEDS_COPY() returns true (for at least one of the pbufs in a chain).

You're working on the oldest open task here (task #7896, "Support zero-copy drivers"), so you're right that this is not working out of the box. However, I'd like that to be supported, but I want to wait with changes until 2.1.0 has been released.

Thanks,
Simon