Re: About improving the qlge Ethernet driver by following drivers/staging/qlge/TODO

[Benjamin Poirier <benjamin.poirier@gmail.com>]

On 2021-05-09 07:27 +0800, Coiby Xu wrote:
> On Fri, May 07, 2021 at 09:32:39AM +0800, Coiby Xu wrote:
> > On Wed, May 05, 2021 at 05:59:46PM +0900, Benjamin Poirier wrote:
> > > On 2021-05-04 21:14 +0800, Coiby Xu wrote:
> > > > Hi Benjamin,
> > > > 
> > > > As you have known, I'm working on improving drivers/staging/qlge. I'm
> > > > not sure if I correctly understand some TODO items. Since you wrote the TODO
> > > > list, could you explain some of the items or comment on the
> > > > corresponding fix for me?
> > > > 
> [...]
> > > 
> > > However, in the same area, there is also
> > > 			skb = netdev_alloc_skb(qdev->ndev, length);
> > > 			[...]
> > > 			skb_fill_page_desc(skb, 0, lbq_desc->p.pg_chunk.page,
> > > 					   lbq_desc->p.pg_chunk.offset,
> > > 					   length);
> > > 
> > > Why is the skb allocated with "length" size? Something like
> > > 	skb = napi_alloc_skb(&rx_ring->napi, SMALL_BUFFER_SIZE);
> > > would be better I think. The head only needs enough space for the
> > > subsequent hlen pull.
> > 
> > Thanks for the explanation! I think this place needs to modified. I'll
> > try to figure out how to reach this part of code so I can make sure the
> > change wouldn't introduce an issue.
> 
> After failing to reach to this part of code, it occurred to me this
> may be what the first TODO item meant by "dead code" that handle
> non-split case,
> 
> > * commit 7c734359d350 ("qlge: Size RX buffers based on MTU.", v2.6.33-rc1)
> >   introduced dead code in the receive routines, which should be rewritten
> >   anyways by the admission of the author himself, see the comment above
> >   ql_build_rx_skb(). That function is now used exclusively to handle packets
> >   that underwent header splitting but it still contains code to handle non
> >   split cases.
> 
> Do you think so? 

Yes

> Btw, I think you meant commit 4f848c0a9c265cb3457fbf842dbffd28e82a44fd
> ("qlge: Add RX frame handlers for non-split frames") here. Because it was in this
> commit where the ql_process_mac_split_rx_intr was first introduced,
> 
>     -static void ql_process_mac_rx_intr(struct ql_adapter *qdev,
>     +static void ql_process_mac_split_rx_intr(struct ql_adapter *qdev,
>                                        struct rx_ring *rx_ring,
>     -                                  struct ib_mac_iocb_rsp *ib_mac_rsp)
>     +                                  struct ib_mac_iocb_rsp *ib_mac_rsp,
>     +                                  u16 vlan_id)

It's possible that I referenced the wrong commit in the TODO. Clearly
there is dead code after commit 4f848c0a9c26 ("qlge: Add RX frame
handlers for non-split frames.") like you say. I don't remember for sure
if I had found some before even before that.

> 
> Another TODO item I don't understand is as follows,
> > * the driver has a habit of using runtime checks where compile time checks are
> >  possible (ex. ql_free_rx_buffers(), ql_alloc_rx_buffers())
> 
> Could be more specific about which runtime checks are used in
> ql_free_rx_buffers and ql_alloc_rx_buffers?

This specific example was fixed in commit
e4c911a73c89 ("staging: qlge: Remove rx_ring.type")

I forgot to update the TODO when making that commit.

Here are other examples:
a68a5b2fd3a2 ("staging: qlge: Remove bq_desc.maplen")
16714d98bf63 ("staging: qlge: Remove rx_ring.sbq_buf_size")
ec705b983b46 ("staging: qlge: Remove qlge_bq.len & size")

I don't remember of remaining examples to point you to. Maybe there
aren't but given that there were indeed quite a few, I would suggest
that you look at those commits and keep this item in mind as you work on
the other items earlier in the list. If at the end you think that this
is no longer a problem, then remove it from the list.