Thread (28 messages) 28 messages, 7 authors, 2011-03-30

Re: kmemleak for MIPS

From: Eric Dumazet <hidden>
Date: 2011-03-30 12:22:10
Also in: linux-mm, lkml

Le mercredi 30 mars 2011 à 14:24 +0300, Daniel Baluta a écrit :
We have:
quoted
UDP hash table entries: 128 (order: 0, 4096 bytes)
CONFIG_BASE_SMALL=0
udp_table_init looks like:

        if (!CONFIG_BASE_SMALL)
                table->hash = alloc_large_system_hash(name, .. &table->mask);
        /*
         * Make sure hash table has the minimum size
         */

Since CONFIG_BASE_SMALL is 0, we are allocating the hash using
alloc_large_system
Then:
        if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
                table->hash = kmalloc();

table->mask is 127, and UDP_HTABLE_SIZE_MIN is 256, so we are allocating again
table->hash without freeing already allocated memory.

We could free table->hash, before allocating the memory with kmalloc.
I don't fully understand the condition table->mask < UDP_HTABLE_SIZE_MIN - 1.

Eric?
There is nothing special. UDP algo needs a minimum hash table that
alloc_large_system_hash() was not able to provide (???)

As you spotted, there is no free_large-system_hash(), so we 'leak' the
small hash table.

If machine has not enough memory to provide such a small hash table, I
suggest using CONFIG_BASE_SMALL, since :

#define UDP_HTABLE_SIZE_MIN (CONFIG_BASE_SMALL ? 128 : 256)
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help