Re: [net-next PATCH 4/4] be2net: replace polling with sleeping in the FW completion path

From: Sathya Perla <redacted>
Date: Wed, 27 Jul 2016 13:33:19 +0530

quoted

-----Original Message-----
From: Sathya Perla [mailto:sathya.perla@broadcom.com]

quoted

quoted

quoted

@@ -4477,6 +4551,22 @@ static int be_if_create(struct be_adapter

*adapter)

quoted

quoted

 	u32 cap_flags = be_if_cap_flags(adapter);
 	int status;

+	/* alloc required memory for other filtering fields */
+	adapter->pmac_id = kcalloc(be_max_uc(adapter),
+				   sizeof(*adapter->pmac_id), GFP_KERNEL);
+	if (!adapter->pmac_id)
+		return -ENOMEM;
+
+	adapter->mc_list = kcalloc(be_max_mc(adapter),
+				   sizeof(*adapter->mc_list), GFP_KERNEL);
+	if (!adapter->mc_list)
+		return -ENOMEM;
+
+	adapter->uc_list = kcalloc(be_max_uc(adapter),
+				   sizeof(*adapter->uc_list), GFP_KERNEL);
+	if (!adapter->uc_list)
+		return -ENOMEM;

These error paths are leaking memory, please audit this in the rest
of

your change as well.

David, thanks for catching this; will fix this (and others if any) and
send out a v2.

David, actually, when be_if_create() returns an error, it falls back into
the error path of
be_setup() (the caller) and be_clear() is called. be_clear() calls
be_if_destroy() which frees the allocated memory.
So, this code will not leak memory in the error path.

David, can you accept the v1 patch set as-is, as it doesn't have the memory
leaks that you suspected initially?

Please resubmit, thanks.