Re: [PATCH] Xen backend support for paged out grant targets.

[Andres Lagar-Cavilla <andres.lagarcavilla@gmail.com>]

On Sep 5, 2012, at 12:27 PM, Konrad Rzeszutek Wilk wrote:

> On Fri, Aug 31, 2012 at 05:10:56PM +0100, David Vrabel wrote:
>> On 31/08/12 16:42, Andres Lagar-Cavilla wrote:
>>> Actually acted upon your feedback ipso facto:
>>> 
>>> commit d5fab912caa1f0cf6be0a6773f502d3417a207b6
>>> Author: Andres Lagar-Cavilla <andres@lagarcavilla.org>
>>> Date:   Sun Aug 26 09:45:57 2012 -0400
>>> 
>>>    Xen backend support for paged out grant targets.
>> 
>> This looks mostly fine expect for the #define instead of inline functions.
>> 
>>> +#define gnttab_map_grant_no_eagain(_gop)                                    \
>> 
>> This name tripped me up previously. As I read this as:
>> 
>> gnttab_map_grant_no_[retries_for]_eagain().
>> 
>> Perhaps gnttab_map_grant_with_retries() ? Or similar?
> 
> gnttab_map_grant_retry ?
> 
> Besides that, it looks good to me. Ian Campbell needs to Ack so that
> David Miller (the network maintainer) can pick it up. Please CC them
> both and also LKML.

Sure will do. However, I need to bring attention to the following hunk:
diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
index 682633b..5610fd8 100644
--- a/drivers/net/xen-netback/netback.c
+++ b/drivers/net/xen-netback/netback.c
@@ -635,9 +635,7 @@ static void xen_netbk_rx_action(struct xen_netbk *netbk)
		return;

	BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
-	ret = HYPERVISOR_grant_table_op(GNTTABOP_copy, &netbk->grant_copy_op,
-					npo.copy_prod);
-	BUG_ON(ret != 0);
+	gnttab_batch_copy_no_eagain(netbk->grant_copy_op, npo.copy_prod);

It seems like the (extant) code passes a struct gnttab_copy ** to the grant table hypercall (&netbk->grant_copy_op, defined as struct gnttab_copy [])

I "fixed" it to pass a struct gnttab_copy * (netbk->grant_copy_op, no '&'). This matches the other invocation of the grant copy hyper call (in netbk_tx_action). Did I fix it, though? I am confused as to how this could have ever worked.

(cc'ing Ian Campbell for more insight) 
Andres
>> 
>> David