On 07.12.22 11:26, Jan Beulich wrote:
> On 07.12.2022 11:18, Juergen Gross wrote:
>> On 07.12.22 10:25, Jan Beulich wrote:
>>> On 07.12.2022 08:23, Juergen Gross wrote:
>>>> Commit ad7f402ae4f4 ("xen/netback: Ensure protocol headers don't fall in
>>>> the non-linear area") introduced a (valid) build warning.
>>>>
>>>> Fix it.
>>>>
>>>> Fixes: ad7f402ae4f4 ("xen/netback: Ensure protocol headers don't fall in the non-linear area")
>>>> Signed-off-by: Juergen Gross <jgross@suse.com>
>>>
>>> Reviewed-by: Jan Beulich <jbeulich@suse.com>
>>>
>>>> --- a/drivers/net/xen-netback/netback.c
>>>> +++ b/drivers/net/xen-netback/netback.c
>>>> @@ -530,7 +530,7 @@ static int xenvif_tx_check_gop(struct xenvif_queue *queue,
>>>>    	const bool sharedslot = nr_frags &&
>>>>    				frag_get_pending_idx(&shinfo->frags[0]) ==
>>>>    				    copy_pending_idx(skb, copy_count(skb) - 1);
>>>> -	int i, err;
>>>> +	int i, err = 0;
>>>>    
>>>>    	for (i = 0; i < copy_count(skb); i++) {
>>>>    		int newerr;
>>>
>>> I'm afraid other logic (below here) is now slightly wrong as well, in
>>> particular
>>>
>>> 				/* If the mapping of the first frag was OK, but
>>> 				 * the header's copy failed, and they are
>>> 				 * sharing a slot, send an error
>>> 				 */
>>> 				if (i == 0 && !first_shinfo && sharedslot)
>>> 					xenvif_idx_release(queue, pending_idx,
>>> 							   XEN_NETIF_RSP_ERROR);
>>> 				else
>>> 					xenvif_idx_release(queue, pending_idx,
>>> 							   XEN_NETIF_RSP_OKAY);
>>>
>>> which looks to be intended to deal with _only_ failure of the one shared
>>> part of the header, whereas "err" now can indicate an error on any earlier
>>> part as well.
>>
>> The comment at the end of that loop seems to imply this is the desired
>> behavior:
>>
>> 		/* Remember the error: invalidate all subsequent fragments. */
>> 		err = newerr;
>> 	}
> 
> This says "subsequent", whereas I was describing a situation where e.g.
> the first piece of header copying failed, the 2nd (shared part) succeeded,
> and the mapping of the rest of the shared part also succeeded. At the
> very least the comment in the code fragment I did quote then has become
> stale. Furthermore, if "all subsequent" really meant all, then in the
> new first loop this isn't followed either - an error response is sent
> only for the pieces where copying failed.

Having stared at the code for quite some time now, I think there is some
room for confusion: "invalidating" the frags seems not to be the same as
setting the related idx to have an error.

XEN_NETIF_RSP_ERROR seems to be set only for the idx which really had an
error, while if any of them had one, all idx-es must be unmapped, have a
status set, and returned to the frontend.

And I think the code is doing this quite fine.

The comments _could_ need some improvements, though.

And some more restructuring could help, too (at least I think that the
"goto check_frags" is a rather clumsy construct - I'd prefer splitting
xenvif_tx_check_gop() into some helper functions and a rather small
body calling those with e.g. different shinfo values).


Juergen