Re: [PATCH] timer: Fix unsafe cpu variable access in migrate_timers

[Borislav Petkov <bp@alien8.de>]

On Sat, Jun 27, 2015 at 11:55:49AM +0200, Jan Kiszka wrote:
> From: Jan Kiszka <jan.kiszka@siemens.com>
> 
> migrate_timers is invoked with preemption enabled. Therefore we have to
> get/put the cpu-local variable tvec_bases like before commit 0eeda71bc3.
> 
> This fixes
> 
> BUG: using smp_processor_id() in preemptible [00000000] code: bash/4917
> caller is debug_smp_processor_id+0x17/0x19
> CPU: 0 PID: 4917 Comm: bash Not tainted 4.1.0-dbg+ #97
> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.8.1-0-g4adadbd-20150316_085822-nilsson.home.kraxel.org 04/01/2014
>  ffff880038674040 ffff880002e6fb98 ffffffff815356a0 0000000000000002
>  0000000000000000 ffff880002e6fbc8 ffffffff8130c16f ffff88003fd8d500
>  000000000000d500 0000000000000000 0000000000000003 ffff880002e6fbd8
> Call Trace:
>  [<ffffffff815356a0>] dump_stack+0x4f/0x7b
>  [<ffffffff8130c16f>] check_preemption_disabled+0xdd/0xef
>  [<ffffffff8130c198>] debug_smp_processor_id+0x17/0x19
>  [<ffffffff810a9b37>] timer_cpu_notify+0x4f/0x112
>  [<ffffffff8106718b>] notifier_call_chain+0x4c/0x71
>  [<ffffffff810671be>] __raw_notifier_call_chain+0xe/0x10
>  [<ffffffff810484b6>] __cpu_notify+0x20/0x37
>  [<ffffffff810484e0>] cpu_notify+0x13/0x15
>  [<ffffffff81048591>] cpu_notify_nofail+0xe/0x16
>  [<ffffffff8152efb2>] _cpu_down+0x178/0x268
>  [<ffffffff8108bc3a>] ? trace_hardirqs_on+0xd/0xf
>  [<ffffffff8152f0ca>] cpu_down+0x28/0x3c
>  [<ffffffff813cbdb9>] cpu_subsys_offline+0x14/0x16
>  [<ffffffff813c779d>] device_offline+0x7d/0xb1
>  [<ffffffff813c78a2>] online_store+0x48/0x68
>  [<ffffffff813c5544>] dev_attr_store+0x18/0x22
>  [<ffffffff811dac6c>] sysfs_kf_write+0x49/0x51
>  [<ffffffff811da139>] kernfs_fop_write+0x105/0x158
>  [<ffffffff8116c54f>] __vfs_write+0x28/0xbd
>  [<ffffffff812ab014>] ? security_file_permission+0x23/0x90
>  [<ffffffff8116ccb2>] vfs_write+0xb2/0x169
>  [<ffffffff8116d717>] SyS_write+0x4a/0x91
>  [<ffffffff8153d12e>] entry_SYSCALL_64_fastpath+0x12/0x76
> 
> triggered when offlining a CPU, e.g. via sysfs.
> 
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> ---
>  kernel/time/timer.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/time/timer.c b/kernel/time/timer.c
> index 520499d..c826178 100644
> --- a/kernel/time/timer.c
> +++ b/kernel/time/timer.c
> @@ -1566,7 +1566,7 @@ static void migrate_timers(int cpu)
>  
>  	BUG_ON(cpu_online(cpu));
>  	old_base = per_cpu_ptr(&tvec_bases, cpu);
> -	new_base = this_cpu_ptr(&tvec_bases);
> +	new_base = &get_cpu_var(tvec_bases);

Hmm, tglx's version doesn't disable preemtion around it:

http://git.kernel.org/cgit/linux/kernel/git/tip/tip.git/commit/?h=timers/urgent&id=24bfcb100959c8641a627b5604d967243f8f240c

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--