Re: [PATCH v6] sched: fix llc shared map unreleased during cpu hotplug

[Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>]

(2014/09/24 17:13), Wanpeng Li wrote:
> BUG: unable to handle kernel NULL pointer dereference at 0000000000000004
> IP: [..] find_busiest_group
> PGD 5a9d5067 PUD 13067 PMD 0
> Oops: 0000 [#3] SMP
> [...]
> Call Trace:
> load_balance
> ? _raw_spin_unlock_irqrestore
> idle_balance
> __schedule
> schedule
> schedule_timeout
> ? lock_timer_base
> schedule_timeout_uninterruptible
> msleep
> lock_device_hotplug_sysfs
> online_store
> dev_attr_store
> sysfs_write_file
> vfs_write
> SyS_write
> system_call_fastpath
> 
> This bug can be triggered by hot add and remove large number of xen
> domain0's vcpus repeatly.
> 
> Last level cache shared map is built during cpu up and build sched domain
> routine takes advantage of it to setup sched domain cpu topology, however,
> llc shared map is unreleased during cpu disable which lead to invalid sched
> domain cpu topology. This patch fix it by release llc shared map correctly
> during cpu disable.
> 
> Yasuaki also reported this can happen on their real hardware.
> https://lkml.org/lkml/2014/7/22/1018
> 
> His case is here.
> ==
> Here is a example on my system.
> My system has 4 sockets and each socket has 15 cores and HT is enabled.
> In this case, each core of sockes is numbered as follows:
> 
>            | CPU#
> Socket#0 | 0-14 , 60-74
> Socket#1 | 15-29, 75-89
> Socket#2 | 30-44, 90-104
> Socket#3 | 45-59, 105-119
> Then llc_shared_mask of CPU#30 has 0x3fff80000001fffc0000000.
> It means that last level cache of Socket#2 is shared with
> CPU#30-44 and 90-104.
> When hot-removing socket#2 and #3, each core of sockets is numbered
> as follows:
> 
>            | CPU#
> Socket#0 | 0-14 , 60-74
> Socket#1 | 15-29, 75-89
> But llc_shared_mask is not cleared. So llc_shared_mask of CPU#30 remains
> having 0x3fff80000001fffc0000000.
> After that, when hot-adding socket#2 and #3, each core of sockets is
> numbered as follows:
> 
>            | CPU#
> Socket#0 | 0-14 , 60-74
> Socket#1 | 15-29, 75-89
> Socket#2 | 30-59
> Socket#3 | 90-119
> Then llc_shared_mask of CPU#30 becomes 0x3fff8000fffffffc0000000.
> It means that last level cache of Socket#2 is shared with CPU#30-59
> and 90-104. So the mask has wrong value.

> At first, I cleared hot-removed CPU number's bit from llc_shared_map
> when hot removing CPU. But Borislav suggested that the problem will
> disappear if readded CPU is assigned same CPU number. And llc_shared_map
> must not be changed.

Please remove it. The description is not explanation for your patch.

Thanks,
Yasuaki Ishimatsu

> 
> Reviewed-by: Borislav Petkov <bp@suse.de>
> Reviewed-by: Toshi Kani <toshi.kani@hp.com>
> Reviewed-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
> Tested-by: Linn Crosetto <linn@hp.com>
> Signed-off-by: Wanpeng Li <wanpeng.li@linux.intel.com>
> ---
> v5 -> v6:
>   * add the real-hardware reports to the changelog
> v4 -> v5:
>   * add the description when the bug can occur
> v3 -> v4:
>   * simplify backtrace
> v2 -> v3:
>   * simplify backtrace
> v1 -> v2:
>   * fix subject line
> 
>   arch/x86/kernel/smpboot.c | 3 +++
>   1 file changed, 3 insertions(+)
> 
> diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
> index 5492798..0134ec7 100644
> --- a/arch/x86/kernel/smpboot.c
> +++ b/arch/x86/kernel/smpboot.c
> @@ -1292,6 +1292,9 @@ static void remove_siblinginfo(int cpu)
>   
>   	for_each_cpu(sibling, cpu_sibling_mask(cpu))
>   		cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
> +	for_each_cpu(sibling, cpu_llc_shared_mask(cpu))
> +		cpumask_clear_cpu(cpu, cpu_llc_shared_mask(sibling));
> +	cpumask_clear(cpu_llc_shared_mask(cpu));
>   	cpumask_clear(cpu_sibling_mask(cpu));
>   	cpumask_clear(cpu_core_mask(cpu));
>   	c->phys_proc_id = 0;
>