Re: [RFC] tcp md5 use of alloc_percpu
From: Eric Dumazet <hidden>
Date: 2014-10-23 14:43:27
Subsystem:
networking [general], networking [tcp], the rest · Maintainers:
"David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Neal Cardwell, Linus Torvalds
On Thu, 2014-10-23 at 06:21 -0700, Eric Dumazet wrote:
On Thu, 2014-10-23 at 02:58 -0400, Jonathan Toppins wrote:quoted
quoted
+ if (!pool) { + pool = kzalloc_node(sizeof(*pool), GFP_KERNEL,GFP_DMA | GFP_KERNEL This memory will possibly be used in a DMA correct? (thinking crypto hardware offload)I am not sure this is the case, but this certainly can be added.
Yes, this is not the case. The real problem is because sg_set_buf() does the following : sg_set_page(sg, virt_to_page(buf), buflen, offset_in_page(buf)); So it assumes a memory range is not spanning multiple pages. So maybe a simpler patch would be :
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 1bec4e76d88c5852d8ba3392b22aa58d6453ab4d..53e355199437b00836635c5919f2f1a1ae237271 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c@@ -2886,10 +2886,17 @@ static void __tcp_free_md5sig_pool(struct tcp_md5sig_pool __percpu *pool) static void __tcp_alloc_md5sig_pool(void) { - int cpu; struct tcp_md5sig_pool __percpu *pool; + size_t sz; + int cpu; - pool = alloc_percpu(struct tcp_md5sig_pool); + /* sg_set_buf() assumes a contiguous memory area, but alloc_percpu() + * can use vmalloc(). So make sure we ask an alignment so that + * tcp_md5sig_pool is located in a single page. + */ + BUILD_BUG_ON(sizeof(struct tcp_md5sig_pool) > PAGE_SIZE); + sz = roundup_pow_of_two(sizeof(struct tcp_md5sig_pool)); + pool = __alloc_percpu(sz, sz); if (!pool) return;