• Christoph Lameter's avatar
    net: replace __this_cpu_inc in route.c with raw_cpu_inc · 3ed66e91
    Christoph Lameter authored
    The RT_CACHE_STAT_INC macro triggers the new preemption checks
    for __this_cpu ops.
    
    I do not see any other synchronization that would allow the use of a
    __this_cpu operation here however in commit dbd2915c ("[IPV4]:
    RT_CACHE_STAT_INC() warning fix") Andrew justifies the use of
    raw_smp_processor_id() here because "we do not care" about races.  In
    the past we agreed that the price of disabling interrupts here to get
    consistent counters would be too high.  These counters may be inaccurate
    due to race conditions.
    
    The use of __this_cpu op improves the situation already from what commit
    dbd2915c did since the single instruction emitted on x86 does not
    allow the race to occur anymore.  However, non x86 platforms could still
    experience a race here.
    
    Trace:
    
      __this_cpu_add operation in preemptible [00000000] code: avahi-daemon/1193
      caller is __this_cpu_preempt_check+0x38/0x60
      CPU: 1 PID: 1193 Comm: avahi-daemon Tainted: GF            3.12.0-rc4+ #187
      Call Trace:
        check_preemption_disabled+0xec/0x110
        __this_cpu_preempt_check+0x38/0x60
        __ip_route_output_key+0x575/0x8c0
        ip_route_output_flow+0x27/0x70
        udp_sendmsg+0x825/0xa20
        inet_sendmsg+0x85/0xc0
        sock_sendmsg+0x9c/0xd0
        ___sys_sendmsg+0x37c/0x390
        __sys_sendmsg+0x49/0x90
        SyS_sendmsg+0x12/0x20
        tracesys+0xe1/0xe6
    Signed-off-by: default avatarChristoph Lameter <cl@linux.com>
    Acked-by: default avatarDavid S. Miller <davem@davemloft.net>
    Acked-by: default avatarIngo Molnar <mingo@kernel.org>
    Cc: Eric Dumazet <edumazet@google.com>
    Cc: Tejun Heo <tj@kernel.org>
    Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
    Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
    3ed66e91
route.c 66 KB