]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
drivers: net: cpsw: fix kmemleak false-positive reports for sk buffers
authorGrygorii Strashko <grygorii.strashko@ti.com>
Tue, 9 Aug 2016 12:09:44 +0000 (15:09 +0300)
committerDavid S. Miller <davem@davemloft.net>
Tue, 9 Aug 2016 21:54:59 +0000 (14:54 -0700)
Kmemleak reports following false positive memory leaks for each sk
buffers allocated by CPSW (__netdev_alloc_skb_ip_align()) in
cpsw_ndo_open() and cpsw_rx_handler():

unreferenced object 0xea915000 (size 2048):
  comm "systemd-network", pid 713, jiffies 4294938323 (age 102.180s)
  hex dump (first 32 bytes):
    00 58 91 ea ff ff ff ff ff ff ff ff ff ff ff ff  .X..............
    ff ff ff ff ff ff fd 0f 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<c0108680>] __kmalloc_track_caller+0x1a4/0x230
    [<c0529eb4>] __alloc_skb+0x68/0x16c
    [<c052c884>] __netdev_alloc_skb+0x40/0x104
    [<bf1ad29c>] cpsw_ndo_open+0x374/0x670 [ti_cpsw]
    [<c053c3d4>] __dev_open+0xb0/0x114
    [<c053c690>] __dev_change_flags+0x9c/0x14c
    [<c053c760>] dev_change_flags+0x20/0x50
    [<c054bdcc>] do_setlink+0x2cc/0x78c
    [<c054c358>] rtnl_setlink+0xcc/0x100
    [<c054b34c>] rtnetlink_rcv_msg+0x184/0x224
    [<c056467c>] netlink_rcv_skb+0xa8/0xc4
    [<c054b1c0>] rtnetlink_rcv+0x2c/0x34
    [<c0564018>] netlink_unicast+0x16c/0x1f8
    [<c0564498>] netlink_sendmsg+0x334/0x348
    [<c052015c>] sock_sendmsg+0x1c/0x2c
    [<c05213e0>] SyS_sendto+0xc0/0xe8

unreferenced object 0xec861780 (size 192):
  comm "softirq", pid 0, jiffies 4294938759 (age 109.540s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 b0 5a ed 00 00 00 00 00 00 00 00  ......Z.........
  backtrace:
    [<c0107830>] kmem_cache_alloc+0x190/0x208
    [<c052c768>] __build_skb+0x30/0x98
    [<c052c8fc>] __netdev_alloc_skb+0xb8/0x104
    [<bf1abc54>] cpsw_rx_handler+0x68/0x1e4 [ti_cpsw]
    [<bf11aa30>] __cpdma_chan_free+0xa8/0xc4 [davinci_cpdma]
    [<bf11ab98>] __cpdma_chan_process+0x14c/0x16c [davinci_cpdma]
    [<bf11abfc>] cpdma_chan_process+0x44/0x5c [davinci_cpdma]
    [<bf1adc78>] cpsw_rx_poll+0x1c/0x9c [ti_cpsw]
    [<c0539180>] net_rx_action+0x1f0/0x2ec
    [<c003881c>] __do_softirq+0x134/0x258
    [<c0038a00>] do_softirq+0x68/0x70
    [<c0038adc>] __local_bh_enable_ip+0xd4/0xe8
    [<c0640994>] _raw_spin_unlock_bh+0x30/0x34
    [<c05f4e9c>] igmp6_group_added+0x4c/0x1bc
    [<c05f6600>] ipv6_dev_mc_inc+0x398/0x434
    [<c05dba74>] addrconf_dad_work+0x224/0x39c

This happens because CPSW allocates SK buffers and then passes
pointers on them in CPDMA where they stored in internal CPPI RAM
(SRAM) which belongs to DEV MMIO space. Kmemleak does not scan IO
memory and so reports memory leaks.

Hence, mark allocated sk buffers as false positive explicitly.

Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/ti/cpsw.c

index c51f34693eae40440604c4090e23dcf6201cd21e..f85d605e45606f1b3e8493dfd1e1b3383005d7b4 100644 (file)
@@ -734,6 +734,7 @@ static void cpsw_rx_handler(void *token, int len, int status)
                netif_receive_skb(skb);
                ndev->stats.rx_bytes += len;
                ndev->stats.rx_packets++;
+               kmemleak_not_leak(new_skb);
        } else {
                ndev->stats.rx_dropped++;
                new_skb = skb;
@@ -1325,6 +1326,7 @@ static int cpsw_ndo_open(struct net_device *ndev)
                                kfree_skb(skb);
                                goto err_cleanup;
                        }
+                       kmemleak_not_leak(skb);
                }
                /* continue even if we didn't manage to submit all
                 * receive descs