From: Jim Mott <jim@mellanox.com>
Date: Tue, 23 Oct 2007 17:58:27 +0000 (-0700)
Subject: SDP - Method used to allocate socket buffers may cause node to hang
X-Git-Tag: v4.1.12-92~264^2~5^2~341
X-Git-Url: https://www.infradead.org/git/?a=commitdiff_plain;h=7194597b9430ba64cd44f5c80940027d3d842622;p=users%2Fjedix%2Flinux-maple.git

SDP - Method used to allocate socket buffers may cause node to hang

The problem we are seeing is that if a node is under load, and
a memory allocation fails (say in sock_sendmsg()), the kernel will
use the allocation policy to decide how to proceed with the allocation.
If GFP_KERNEL is specified, then the kernel may attempt to free pages
through the iSCSI block device that is making the socket call, which
would result in a deadlock.  Use of GFP_NOIO should prevent the kernel
from using the IO backend to free memory resources.

Each kernel level socket has an allocation flag to specify the
memory allocation policy for socket buffers, the default is GFP_ATOMIC
(or GFP_KERNEL for SDP).  If the caller creates a socket with the
policy set to GFP_NOFS or GFP_NOIO this should be the allocation
policy used by the SDP layer.

Signed-off-by: Jim Mott <jim@mellanox.com>
---

diff --git a/drivers/infiniband/ulp/sdp/sdp_bcopy.c b/drivers/infiniband/ulp/sdp/sdp_bcopy.c
index feccc31157cc1..963105c0f93dc 100644
--- a/drivers/infiniband/ulp/sdp/sdp_bcopy.c
+++ b/drivers/infiniband/ulp/sdp/sdp_bcopy.c
@@ -257,16 +257,26 @@ static void sdp_post_recv(struct sdp_sock *ssk)
 	skb_frag_t *frag;
 	struct sdp_bsdh *h;
 	int id = ssk->rx_head;
+	unsigned int gfp_page;
 
 	/* Now, allocate and repost recv */
 	/* TODO: allocate from cache */
-	skb = sk_stream_alloc_skb(&ssk->isk.sk, SDP_HEAD_SIZE,
-				  GFP_KERNEL);
+
+	if (unlikely(ssk->isk.sk.sk_allocation)) {
+		skb = sk_stream_alloc_skb(&ssk->isk.sk, SDP_HEAD_SIZE,
+					  ssk->isk.sk.sk_allocation);
+		gfp_page = ssk->isk.sk.sk_allocation | __GFP_HIGHMEM;
+	} else {
+		skb = sk_stream_alloc_skb(&ssk->isk.sk, SDP_HEAD_SIZE,
+					  GFP_KERNEL);
+		gfp_page = GFP_HIGHUSER;
+	}
+
 	/* FIXME */
 	BUG_ON(!skb);
 	h = (struct sdp_bsdh *)skb->head;
 	for (i = 0; i < ssk->recv_frags; ++i) {
-		page = alloc_pages(GFP_HIGHUSER, 0);
+		page = alloc_pages(gfp_page, 0);
 		BUG_ON(!page);
 		frag = &skb_shinfo(skb)->frags[i];
 		frag->page                = page;
@@ -425,6 +435,7 @@ void sdp_post_sends(struct sdp_sock *ssk, int nonagle)
 	/* TODO: nonagle? */
 	struct sk_buff *skb;
 	int c;
+	int gfp_page;
 
 	if (unlikely(!ssk->id)) {
 		if (ssk->isk.sk.sk_send_head) {
@@ -436,6 +447,11 @@ void sdp_post_sends(struct sdp_sock *ssk, int nonagle)
 		return;
 	}
 
+	if (unlikely(ssk->isk.sk.sk_allocation))
+		gfp_page = ssk->isk.sk.sk_allocation;
+	else
+		gfp_page = GFP_KERNEL;
+
 	if (ssk->recv_request &&
 	    ssk->rx_tail >= ssk->recv_request_head &&
 	    ssk->bufs >= SDP_MIN_BUFS &&
@@ -445,7 +461,7 @@ void sdp_post_sends(struct sdp_sock *ssk, int nonagle)
 		skb = sk_stream_alloc_skb(&ssk->isk.sk,
 					  sizeof(struct sdp_bsdh) +
 					  sizeof(*resp_size),
-					  GFP_KERNEL);
+					  gfp_page);
 		/* FIXME */
 		BUG_ON(!skb);
 		resp_size = (struct sdp_chrecvbuf *)skb_put(skb, sizeof *resp_size);
@@ -470,7 +486,7 @@ void sdp_post_sends(struct sdp_sock *ssk, int nonagle)
 		skb = sk_stream_alloc_skb(&ssk->isk.sk,
 					  sizeof(struct sdp_bsdh) +
 					  sizeof(*req_size),
-					  GFP_KERNEL);
+					  gfp_page);
 		/* FIXME */
 		BUG_ON(!skb);
 		ssk->sent_request = SDP_MAX_SEND_SKB_FRAGS * PAGE_SIZE;
@@ -501,7 +517,7 @@ void sdp_post_sends(struct sdp_sock *ssk, int nonagle)
 		ssk->bufs > (ssk->remote_credits >= ssk->rx_head - ssk->rx_tail)) {
 		skb = sk_stream_alloc_skb(&ssk->isk.sk,
 					  sizeof(struct sdp_bsdh),
-					  GFP_KERNEL);
+					  gfp_page);
 		/* FIXME */
 		BUG_ON(!skb);
 		sdp_post_send(ssk, skb, SDP_MID_DISCONN);