On Fri, 2005-05-06 at 12:52, Hal Rosenstock wrote:
> On Wed, 2005-05-04 at 02:57, Michael S. Tsirkin wrote:
> > > If this only occurs on machines which are running OpenSM, it could be a
> > > problem with the local MAD handling where ib_free_recv_mad is not called
> > > for some case where it should. I will look into this if this is the
> > > case.
> > >
> > > -- Hal
> > >
> >
> > This is the case for me: the error only occurs on machines which are
> > running OpenSM.
>
> So assuming mthca sets the bits properly for local MAD handling (and I
> have no reason to think otherwise), this is likely a local MAD handling
> error in mad.c where some path does not return an allocation to the MAD
> cache. I inspected the code to try and find it but couldn't. I will need
> to dig deeper. This will take some more time. (I will add some
> accounting in and see what this shows up). Please bear with me.
This patch works for me. Can you try it out ? If it works for you, I
will check it in.
-- Hal
Index: mad.c
===================================================================
-- mad.c (revision 2331)
+++ mad.c (working copy)
@@ -1602,7 +1602,6 @@
INIT_LIST_HEAD(&mad_recv_wc->rmpp_list);
list_add(&mad_recv_wc->recv_buf.list, &mad_recv_wc->rmpp_list);
-
if (mad_agent_priv->agent.rmpp_version) {
mad_recv_wc = ib_process_rmpp_recv_wc(mad_agent_priv,
mad_recv_wc);
@@ -2054,6 +2053,8 @@
/* Empty wait list to prevent receives from finding a request */
list_splice_init(&mad_agent_priv->wait_list, &cancel_list);
+ /* Empty local completion list as well */
+ list_splice_init(&mad_agent_priv->local_list, &cancel_list);
spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
/* Report all cancelled requests */
@@ -2168,6 +2169,7 @@
struct ib_mad_local_private *local;
struct ib_mad_agent_private *recv_mad_agent;
unsigned long flags;
+ int recv = 0;
struct ib_wc wc;
struct ib_mad_send_wc mad_send_wc;
@@ -2183,10 +2185,10 @@
recv_mad_agent = local->recv_mad_agent;
if (!recv_mad_agent) {
printk(KERN_ERR PFX "No receive MAD agent for
local completion\n");
- kmem_cache_free(ib_mad_cache, local->mad_priv);
goto local_send_completion;
}
+ recv = 1;
/*
* Defined behavior is to complete response
* before request
@@ -2199,6 +2201,8 @@
local->mad_priv->header.recv_wc.mad_len =
sizeof(struct ib_mad);
INIT_LIST_HEAD(&local->mad_priv->header.recv_wc.rmpp_list);
+ list_add(&local->mad_priv->header.recv_wc.recv_buf.list,
+ &local->mad_priv->header.recv_wc.rmpp_list);
local->mad_priv->header.recv_wc.recv_buf.grh = NULL;
local->mad_priv->header.recv_wc.recv_buf.mad =
&local->mad_priv->mad.mad;
@@ -2229,6 +2233,8 @@
spin_lock_irqsave(&mad_agent_priv->lock, flags);
list_del(&local->completion_list);
atomic_dec(&mad_agent_priv->refcount);
+ if (!recv)
+ kmem_cache_free(ib_mad_cache, local->mad_priv);
kfree(local);
}
spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
