netdev
[Top] [All Lists]

[PATCH 3/3] TCP hashes: NUMA interleaving

To: akpm@xxxxxxxx
Subject: [PATCH 3/3] TCP hashes: NUMA interleaving
From: Brent Casavant <bcasavan@xxxxxxx>
Date: Mon, 20 Dec 2004 17:37:17 -0600
Cc: linux-kernel@xxxxxxxxxxxxxxx, netdev@xxxxxxxxxxx
Organization: "Silicon Graphics, Inc."
Sender: netdev-bounce@xxxxxxxxxxx
Resend: Submitting for inclusion, as this patch series drew no objections
from interested parties, and required no changes.

The following patch against 2.6.10-rc3 modifies the TCP ehash and TCP
bhash to enable the use of vmalloc to alleviate boottime memory allocation
imbalances on NUMA systems, utilizing flags to the alloc_large_system_hash
routine in order to centralize the enabling of this behavior.

This patch (3/3) depends on patch 1/3 in this patch series.  It
does not depend on, nor is depended upon by, patch 2/3 in the
series.

Signed-off-by: Brent Casavant <bcasavan@xxxxxxx>

Index: linux/net/ipv4/tcp.c
===================================================================
--- linux.orig/net/ipv4/tcp.c   2004-12-10 18:09:49.011272637 -0600
+++ linux/net/ipv4/tcp.c        2004-12-10 18:10:52.285839780 -0600
@@ -256,6 +256,7 @@
 #include <linux/smp_lock.h>
 #include <linux/fs.h>
 #include <linux/random.h>
+#include <linux/bootmem.h>
 
 #include <net/icmp.h>
 #include <net/tcp.h>
@@ -2254,7 +2255,6 @@
 void __init tcp_init(void)
 {
        struct sk_buff *skb = NULL;
-       unsigned long goal;
        int order, i;
 
        if (sizeof(struct tcp_skb_cb) > sizeof(skb->cb))
@@ -2287,43 +2287,35 @@
         *
         * The methodology is similar to that of the buffer cache.
         */
-       if (num_physpages >= (128 * 1024))
-               goal = num_physpages >> (21 - PAGE_SHIFT);
-       else
-               goal = num_physpages >> (23 - PAGE_SHIFT);
-
-       if (thash_entries)
-               goal = (thash_entries * sizeof(struct tcp_ehash_bucket)) >> 
PAGE_SHIFT;
-       for (order = 0; (1UL << order) < goal; order++)
-               ;
-       do {
-               tcp_ehash_size = (1UL << order) * PAGE_SIZE /
-                       sizeof(struct tcp_ehash_bucket);
-               tcp_ehash_size >>= 1;
-               while (tcp_ehash_size & (tcp_ehash_size - 1))
-                       tcp_ehash_size--;
-               tcp_ehash = (struct tcp_ehash_bucket *)
-                       __get_free_pages(GFP_ATOMIC, order);
-       } while (!tcp_ehash && --order > 0);
-
-       if (!tcp_ehash)
-               panic("Failed to allocate TCP established hash table\n");
+       tcp_ehash = (struct tcp_ehash_bucket *)
+               alloc_large_system_hash("TCP established",
+                                       sizeof(struct tcp_ehash_bucket),
+                                       thash_entries,
+                                       (num_physpages >= 128 * 1024) ?
+                                               (25 - PAGE_SHIFT) :
+                                               (27 - PAGE_SHIFT),
+                                       HASH_HIGHMEM,
+                                       &tcp_ehash_size,
+                                       NULL,
+                                       0);
+       tcp_ehash_size = (1 << tcp_ehash_size) >> 1;
        for (i = 0; i < (tcp_ehash_size << 1); i++) {
                rwlock_init(&tcp_ehash[i].lock);
                INIT_HLIST_HEAD(&tcp_ehash[i].chain);
        }
 
-       do {
-               tcp_bhash_size = (1UL << order) * PAGE_SIZE /
-                       sizeof(struct tcp_bind_hashbucket);
-               if ((tcp_bhash_size > (64 * 1024)) && order > 0)
-                       continue;
-               tcp_bhash = (struct tcp_bind_hashbucket *)
-                       __get_free_pages(GFP_ATOMIC, order);
-       } while (!tcp_bhash && --order >= 0);
-
-       if (!tcp_bhash)
-               panic("Failed to allocate TCP bind hash table\n");
+       tcp_bhash = (struct tcp_bind_hashbucket *)
+               alloc_large_system_hash("TCP bind",
+                                       sizeof(struct tcp_bind_hashbucket),
+                                       tcp_ehash_size,
+                                       (num_physpages >= 128 * 1024) ?
+                                               (25 - PAGE_SHIFT) :
+                                               (27 - PAGE_SHIFT),
+                                       HASH_HIGHMEM,
+                                       &tcp_bhash_size,
+                                       NULL,
+                                       64 * 1024);
+       tcp_bhash_size = 1 << tcp_bhash_size;
        for (i = 0; i < tcp_bhash_size; i++) {
                spin_lock_init(&tcp_bhash[i].lock);
                INIT_HLIST_HEAD(&tcp_bhash[i].chain);
@@ -2332,6 +2324,10 @@
        /* Try to be a bit smarter and adjust defaults depending
         * on available memory.
         */
+       for (order = 0; ((1 << order) << PAGE_SHIFT) <
+                       (tcp_bhash_size * sizeof(struct tcp_bind_hashbucket));
+                       order++)
+               ;
        if (order > 4) {
                sysctl_local_port_range[0] = 32768;
                sysctl_local_port_range[1] = 61000;

-- 
Brent Casavant                          If you had nothing to fear,
bcasavan@xxxxxxx                        how then could you be brave?
Silicon Graphics, Inc.                    -- Queen Dama, Source Wars

<Prev in Thread] Current Thread [Next in Thread>
  • [PATCH 3/3] TCP hashes: NUMA interleaving, Brent Casavant <=