Re: [PATCH] drivers: net: cpsw: fix kmemleak false-positive reports for sk buffers

[Mugunthan V N]

On Tuesday 09 August 2016 05:39 PM, Grygorii Strashko wrote:
> 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:
> [] __kmalloc_track_caller+0x1a4/0x230
> [] __alloc_skb+0x68/0x16c
> [] __netdev_alloc_skb+0x40/0x104
> [] cpsw_ndo_open+0x374/0x670 [ti_cpsw]
> [] __dev_open+0xb0/0x114
> [] __dev_change_flags+0x9c/0x14c
> [] dev_change_flags+0x20/0x50
> [] do_setlink+0x2cc/0x78c
> [] rtnl_setlink+0xcc/0x100
> [] rtnetlink_rcv_msg+0x184/0x224
> [] netlink_rcv_skb+0xa8/0xc4
> [] rtnetlink_rcv+0x2c/0x34
> [] netlink_unicast+0x16c/0x1f8
> [] netlink_sendmsg+0x334/0x348
> [] sock_sendmsg+0x1c/0x2c
> [] 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:
> [] kmem_cache_alloc+0x190/0x208
> [] __build_skb+0x30/0x98
> [] __netdev_alloc_skb+0xb8/0x104
> [] cpsw_rx_handler+0x68/0x1e4 [ti_cpsw]
> [] __cpdma_chan_free+0xa8/0xc4 [davinci_cpdma]
> [] __cpdma_chan_process+0x14c/0x16c [davinci_cpdma]
> [] cpdma_chan_process+0x44/0x5c [davinci_cpdma]
> [] cpsw_rx_poll+0x1c/0x9c [ti_cpsw]
> [] net_rx_action+0x1f0/0x2ec
> [] __do_softirq+0x134/0x258
> [] do_softirq+0x68/0x70
> [] __local_bh_enable_ip+0xd4/0xe8
> [] _raw_spin_unlock_bh+0x30/0x34
> [] igmp6_group_added+0x4c/0x1bc
> [] ipv6_dev_mc_inc+0x398/0x434
> [] 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 
> Signed-off-by: Grygorii Strashko 

Reviewed-by: Mugunthan V N 

Regards
Mugunthan V N

Re: [PATCH] drivers: net: cpsw: fix kmemleak false-positive reports for sk buffers

[Catalin Marinas]

On Tue, Aug 09, 2016 at 03:09:44PM +0300, Grygorii Strashko wrote:
> diff --git a/drivers/net/ethernet/ti/cpsw.c b/drivers/net/ethernet/ti/cpsw.c
> index 0805855..0456e0e 100644
> --- a/drivers/net/ethernet/ti/cpsw.c
> +++ b/drivers/net/ethernet/ti/cpsw.c
> @@ -732,6 +732,7 @@ 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;
> @@ -1323,6 +1324,7 @@ static int cpsw_ndo_open(struct net_device *ndev)
>   kfree_skb(skb);
>   goto err_cleanup;
>   }
> + kmemleak_not_leak(skb);

Nitpick: I would add a short comment above the kmemleak_not_leak() calls
on why there is a false positive. Otherwise:

Acked-by: Catalin Marinas 

[PATCH] drivers: net: cpsw: fix kmemleak false-positive reports for sk buffers

[Grygorii Strashko]

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:
[] __kmalloc_track_caller+0x1a4/0x230
[] __alloc_skb+0x68/0x16c
[] __netdev_alloc_skb+0x40/0x104
[] cpsw_ndo_open+0x374/0x670 [ti_cpsw]
[] __dev_open+0xb0/0x114
[] __dev_change_flags+0x9c/0x14c
[] dev_change_flags+0x20/0x50
[] do_setlink+0x2cc/0x78c
[] rtnl_setlink+0xcc/0x100
[] rtnetlink_rcv_msg+0x184/0x224
[] netlink_rcv_skb+0xa8/0xc4
[] rtnetlink_rcv+0x2c/0x34
[] netlink_unicast+0x16c/0x1f8
[] netlink_sendmsg+0x334/0x348
[] sock_sendmsg+0x1c/0x2c
[] 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:
[] kmem_cache_alloc+0x190/0x208
[] __build_skb+0x30/0x98
[] __netdev_alloc_skb+0xb8/0x104
[] cpsw_rx_handler+0x68/0x1e4 [ti_cpsw]
[] __cpdma_chan_free+0xa8/0xc4 [davinci_cpdma]
[] __cpdma_chan_process+0x14c/0x16c [davinci_cpdma]
[] cpdma_chan_process+0x44/0x5c [davinci_cpdma]
[] cpsw_rx_poll+0x1c/0x9c [ti_cpsw]
[] net_rx_action+0x1f0/0x2ec
[] __do_softirq+0x134/0x258
[] do_softirq+0x68/0x70
[] __local_bh_enable_ip+0xd4/0xe8
[] _raw_spin_unlock_bh+0x30/0x34
[] igmp6_group_added+0x4c/0x1bc
[] ipv6_dev_mc_inc+0x398/0x434
[] 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 
Signed-off-by: Grygorii Strashko 
---
 drivers/net/ethernet/ti/cpsw.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/ethernet/ti/cpsw.c b/drivers/net/ethernet/ti/cpsw.c
index 0805855..0456e0e 100644
--- a/drivers/net/ethernet/ti/cpsw.c
+++ b/drivers/net/ethernet/ti/cpsw.c
@@ -732,6 +732,7 @@ 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;
@@ -1323,6 +1324,7 @@ 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
-- 
2.9.2