Re: [PATCH] KVM: s390: reduce number of IO pins to 1

From: David Hildenbrand <david@redhat.com>
To: Christian Borntraeger <borntraeger@de.ibm.com>,
	Janosch Frank <frankja@linux.vnet.ibm.com>
Cc: KVM <kvm@vger.kernel.org>, Cornelia Huck <cohuck@redhat.com>,
	linux-s390 <linux-s390@vger.kernel.org>,
	Paolo Bonzini <pbonzini@redhat.com>
Subject: Re: [PATCH] KVM: s390: reduce number of IO pins to 1
Date: Thu, 18 Jun 2020 11:07:05 +0200	[thread overview]
Message-ID: <66419659-e678-2daf-2e02-4f2158076ddb@redhat.com> (raw)
In-Reply-To: <6953c580-9b99-1c76-b6eb-510dcb70894c@de.ibm.com>

On 17.06.20 13:04, Christian Borntraeger wrote:
> 
> 
> On 17.06.20 12:19, David Hildenbrand wrote:
>> On 17.06.20 10:36, Christian Borntraeger wrote:
>>> The current number of KVM_IRQCHIP_NUM_PINS results in an order 3
>>> allocation (32kb) for each guest start/restart. This can result in OOM
>>> killer activity even with free swap when the memory is fragmented
>>> enough:
>>>
>>> kernel: qemu-system-s39 invoked oom-killer: gfp_mask=0x440dc0(GFP_KERNEL_ACCOUNT|__GFP_COMP|__GFP_ZERO), order=3, oom_score_adj=0
>>> kernel: CPU: 1 PID: 357274 Comm: qemu-system-s39 Kdump: loaded Not tainted 5.4.0-29-generic #33-Ubuntu
>>> kernel: Hardware name: IBM 8562 T02 Z06 (LPAR)
>>> kernel: Call Trace:
>>> kernel: ([<00000001f848fe2a>] show_stack+0x7a/0xc0)
>>> kernel:  [<00000001f8d3437a>] dump_stack+0x8a/0xc0
>>> kernel:  [<00000001f8687032>] dump_header+0x62/0x258
>>> kernel:  [<00000001f8686122>] oom_kill_process+0x172/0x180
>>> kernel:  [<00000001f8686abe>] out_of_memory+0xee/0x580
>>> kernel:  [<00000001f86e66b8>] __alloc_pages_slowpath+0xd18/0xe90
>>> kernel:  [<00000001f86e6ad4>] __alloc_pages_nodemask+0x2a4/0x320
>>> kernel:  [<00000001f86b1ab4>] kmalloc_order+0x34/0xb0
>>> kernel:  [<00000001f86b1b62>] kmalloc_order_trace+0x32/0xe0
>>> kernel:  [<00000001f84bb806>] kvm_set_irq_routing+0xa6/0x2e0
>>> kernel:  [<00000001f84c99a4>] kvm_arch_vm_ioctl+0x544/0x9e0
>>> kernel:  [<00000001f84b8936>] kvm_vm_ioctl+0x396/0x760
>>> kernel:  [<00000001f875df66>] do_vfs_ioctl+0x376/0x690
>>> kernel:  [<00000001f875e304>] ksys_ioctl+0x84/0xb0
>>> kernel:  [<00000001f875e39a>] __s390x_sys_ioctl+0x2a/0x40
>>> kernel:  [<00000001f8d55424>] system_call+0xd8/0x2c8
>>>
>>> As far as I can tell s390x does not use the iopins as we bail our for
>>> anything other than KVM_IRQ_ROUTING_S390_ADAPTER and the chip/pin is
>>> only used for KVM_IRQ_ROUTING_IRQCHIP. So let us use a small number to
>>> reduce the memory footprint.
>>>
>>> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
>>> ---
>>>  arch/s390/include/asm/kvm_host.h | 8 ++++----
>>>  1 file changed, 4 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
>>> index cee3cb6455a2..6ea0820e7c7f 100644
>>> --- a/arch/s390/include/asm/kvm_host.h
>>> +++ b/arch/s390/include/asm/kvm_host.h
>>> @@ -31,12 +31,12 @@
>>>  #define KVM_USER_MEM_SLOTS 32
>>>  
>>>  /*
>>> - * These seem to be used for allocating ->chip in the routing table,
>>> - * which we don't use. 4096 is an out-of-thin-air value. If we need
>>> - * to look at ->chip later on, we'll need to revisit this.
>>> + * These seem to be used for allocating ->chip in the routing table, which we
>>> + * don't use. 1 is as small as we can get to reduce the needed memory. If we
>>> + * need to look at ->chip later on, we'll need to revisit this.
>>>   */
>>>  #define KVM_NR_IRQCHIPS 1
>>> -#define KVM_IRQCHIP_NUM_PINS 4096
>>> +#define KVM_IRQCHIP_NUM_PINS 1
>>>  #define KVM_HALT_POLL_NS_DEFAULT 50000
>>>  
>>>  /* s390-specific vcpu->requests bit members */
>>>
>>
>> Guess it doesn't make sense to wrap all the "->chip" handling in a
>> separate set of defines.
>>
>> Reviewed-by: David Hildenbrand <david@redhat.com>
> 
> I guess this is just the most simple solution. I am asking myself if I should add
> cc stable of Fixes as I was able to trigger this by having several guests with a
> reboot loop and several guests that trigger memory overcommitment.
> 

I don't think this classifies as stable material.

-- 
Thanks,

David / dhildenb