thermal/powerclamp: Prevent division by zero when counting interval
authorPetr Mladek <pmladek@suse.com>
Fri, 5 Aug 2016 13:20:41 +0000 (15:20 +0200)
committerZhang Rui <rui.zhang@intel.com>
Mon, 8 Aug 2016 02:54:55 +0000 (10:54 +0800)
commit70c50ee72eebb3d1e3aae7450269fd8d8074c610
tree802b3644ea140258bd0e1ffad6f2cd5c87fd3462
parent176b1ec21306f0775e6d7a03f42d82a0b144ef0e
thermal/powerclamp: Prevent division by zero when counting interval

I have got a zero division error when disabling the forced
idle injection from the intel powerclamp. I did

  echo 0 >/sys/class/thermal/cooling_device48/cur_state

and got

[  986.072632] divide error: 0000 [#1] PREEMPT SMP
[  986.078989] Modules linked in:
[  986.083618] CPU: 17 PID: 24967 Comm: kidle_inject/17 Not tainted 4.7.0-1-default+ #3055
[  986.093781] Hardware name: Intel Corporation S2600CP/S2600CP, BIOS RMLSDP.86I.R3.27.D685.1305151734 05/15/2013
[  986.106227] task: ffff880430e1c080 task.stack: ffff880427ef0000
[  986.114122] RIP: 0010:[<ffffffff81794859>]  [<ffffffff81794859>] clamp_thread+0x1d9/0x600
[  986.124609] RSP: 0018:ffff880427ef3e20  EFLAGS: 00010246
[  986.131860] RAX: 0000000000000258 RBX: 0000000000000006 RCX: 0000000000000001
[  986.141179] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000018
[  986.150478] RBP: ffff880427ef3ec8 R08: ffff880427ef0000 R09: 0000000000000002
[  986.159779] R10: 0000000000003df2 R11: 0000000000000018 R12: 0000000000000002
[  986.169089] R13: 0000000000000000 R14: ffff880427ef0000 R15: ffff880427ef0000
[  986.178388] FS:  0000000000000000(0000) GS:ffff880435940000(0000) knlGS:0000000000000000
[  986.188785] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  986.196559] CR2: 00007f1d0caf0000 CR3: 0000000002006000 CR4: 00000000001406e0
[  986.205909] Stack:
[  986.209524]  ffff8802be897b00 ffff880430e1c080 0000000000000011 0000006a35959780
[  986.219236]  0000000000000011 ffff880427ef0008 0000000000000000 ffff8804359503d0
[  986.228966]  0000000100029d93 ffffffff81794140 0000000000000000 ffffffff05000011
[  986.238686] Call Trace:
[  986.242825]  [<ffffffff81794140>] ? pkg_state_counter+0x80/0x80
[  986.250866]  [<ffffffff81794680>] ? powerclamp_set_cur_state+0x180/0x180
[  986.259797]  [<ffffffff8111d1a9>] kthread+0xc9/0xe0
[  986.266682]  [<ffffffff8193d69f>] ret_from_fork+0x1f/0x40
[  986.274142]  [<ffffffff8111d0e0>] ? kthread_create_on_node+0x180/0x180
[  986.282869] Code: d1 ea 48 89 d6 80 3d 6a d0 d4 00 00 ba 64 00 00 00 89 d8 41 0f 45 f5 0f af c2 42 8d 14 2e be 31 00 00 00 83 fa 31 0f 42 f2 31 d2 <f7> f6 48 8b 15 9e 07 87 00 48 8b 3d 97 07 87 00 48 63 f0 83 e8
[  986.307806] RIP  [<ffffffff81794859>] clamp_thread+0x1d9/0x600
[  986.315871]  RSP <ffff880427ef3e20>

RIP points to the following lines:

compensation = get_compensation(target_ratio);
interval = duration_jiffies*100/(target_ratio+compensation);

A solution would be to switch the following two commands in
powerclamp_set_cur_state():

set_target_ratio = 0;
end_power_clamp();

But I think that the zero division might happen also when target_ratio
is non-zero because the compensation might be negative. Therefore
we also check the sum of target_ratio and compensation explicitly.

Also the compensated_ratio variable is always set. Therefore there
is no need to initialize it.

Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
drivers/thermal/intel_powerclamp.c
This page took 0.03192 seconds and 5 git commands to generate.