On Tue, 2004-11-02 at 23:00, Shirley Ma wrote:
> Here is the patch. Please review it.
Yes, these are memory leaks in agent_mad_send but it would be better to
fix them with the deallocation being done at the function level where
the allocation is done. Hence rather than agent_send_mad returning void,
it should return int, etc. I will post a patch for this shortly.
-- Hal
>
> diff -urN access/agent.c access.patch5/agent.c
> --- access/agent.c 2004-11-02 17:40:06.000000000 -0800
> +++ access.patch5/agent.c 2004-11-02 18:43:47.534608536 -0800
> @@ -357,12 +357,16 @@
> if (!port_priv) {
> printk(KERN_ERR SPFX "agent_mad_send: no matching MAD agent %p\n",
> mad_agent);
> + kfree(mad);
> return;
> }
>
> agent_send_wr = kmalloc(sizeof(*agent_send_wr), GFP_KERNEL);
> - if (!agent_send_wr)
> + if (!agent_send_wr) {
> + printk(KERN_ERR SPFX "No memory for agent work request\n");
> + kfree(mad);
> return;
> + }
> agent_send_wr->mad = mad;
>
> /* PCI mapping */
> @@ -407,6 +411,7 @@
> if (IS_ERR(agent_send_wr->ah)) {
> printk(KERN_ERR SPFX "No memory for address handle\n");
> kfree(mad);
> + kfree(agent_send_wr);
> return;
> }
>
> @@ -432,6 +437,8 @@
> sizeof(struct ib_mad),
> PCI_DMA_TODEVICE);
> ib_destroy_ah(agent_send_wr->ah);
> + kfree(mad);
> + kfree(agent_send_wr);
> } else {
> list_add_tail(&agent_send_wr->send_list,
> &port_priv->send_posted_list);
>
>
_______________________________________________
openib-general mailing list
[EMAIL PROTECTED]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
