Hi,
On 03/16/2016 12:22 AM, Paolo Bonzini wrote:
Why is this necessary? The APIC access page is a peculiarity of Intel
>>processors (and the special memslot for only needs to map 0xfee00000 to
>>0xfee00fff; after that there is the MSI area).
>
>The current lapic regs page is allocated using get_zeroed_page(), which
>can be paged out. If I use these pages for AVIC backing pages, it seems
>to cause VM to slow down quite a bit due to a lot of page faults.
What causes the lapic regs page to be paged out?
This is mainly causing a large number of VMEXIT due to NPF. In my test
running hackbench in the guest. The following are perf result profiling
for 10 seconds in the host in two cases:
CASE1: Using x86_set_memory_region() for AVIC backing page
# ./perf-vmexit.sh 10
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 2.813 MB perf.data.guest (30356 samples) ]
Analyze events for all VMs, all VCPUs:
VM-EXIT Samples Samples% Time% Min Time Max
Time Avg time
interrupt 10042 66.30% 81.33% 0.43us
202.50us 7.43us ( +- 1.20% )
msr 5004 33.04% 15.76% 0.73us
12.21us 2.89us ( +- 0.43% )
pause 58 0.38% 0.18% 0.56us
5.88us 2.92us ( +- 6.43% )
npf 35 0.23% 2.01% 6.41us
207.78us 52.70us ( +- 23.67% )
nmi 4 0.03% 0.02% 2.31us
4.67us 3.49us ( +- 14.26% )
io 3 0.02% 0.70% 82.75us
360.90us 214.28us ( +- 37.64% )
avic_incomp_ipi 1 0.01% 0.00% 2.17us
2.17us 2.17us ( +- 0.00% )
Total Samples:15147, Total events handled time:91715.78us.
CASE2: Using the lapic regs page for AVIC backing page.
# ./perf-vmexit.sh 10
[ perf record: Woken up 255 times to write data ]
[ perf record: Captured and wrote 509.202 MB perf.data.guest (5718856
samples) ]
Analyze events for all VMs, all VCPUs:
VM-EXIT Samples Samples% Time% Min Time Max
Time Avg time
npf 1897710 99.33% 98.08% 1.09us
243.22us 1.67us ( +- 0.04% )
interrupt 7818 0.41% 1.44% 0.44us
216.55us 5.97us ( +- 1.92% )
msr 5001 0.26% 0.45% 0.68us
12.58us 2.89us ( +- 0.50% )
pause 25 0.00% 0.00% 0.71us
4.23us 2.03us ( +- 10.76% )
io 4 0.00% 0.03% 73.91us
337.29us 206.74us ( +- 26.38% )
nmi 1 0.00% 0.00% 5.92us
5.92us 5.92us ( +- 0.00% )
Total Samples:1910559, Total events handled time:3229214.64us.
Thanks,
Suravee
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html