[Qemu-devel] KVM serial8250: too much work for irq3

[Qemu-devel] KVM serial8250: too much work for irq3

[Denys Zagorui]

Hello

This is known issue, it was mentioned at first time about 9 years ago.

https://lkml.org/lkml/2008/2/5/401

https://bugzilla.redhat.com/show_bug.cgi?id=986761


But it seems not fixed. I tested on my setup.

My Host machine Debian Stretch (kernel 4.9-30), Guest vanilla kernel 
4.4-76. Tried on Qemu  2.8 and 2.10

qemu-system-x86_64 -name guest=guest,debug-threads=on -S -object 
secret,id=masterKey0,format=raw,file=/var/lib/libvirt/qemu/domain-7-guest/master-key.aes 
-machine pc-i440fx-2.8,accel=kvm,usb=off,dump-guest-core=off -cpu 
SkylakeClient,ds=on,acpi=on,ss=on,ht=on,tm=on,pbe=on,dtes64=on,monitor=on,ds_cpl=on,vmx=on,smx=on,est=on,tm2=on,xtpr=on,pdcm=on,osxsave=on,tsc_adjust=on,clflushopt=on,pdpe1gb=on 
-m 4096 -realtime mlock=off -smp 4,sockets=1,cores=4,threads=1 -uuid 
4537ca29-73b2-40c3-9b43-666de182ba5f -display none -no-user-config 
-nodefaults -chardev 
socket,id=charmonitor,path=/var/lib/libvirt/qemu/domain-7-guest/monitor.sock,server,nowait 
-mon chardev=charmonitor,id=monitor,mode=control -rtc 
base=utc,driftfix=slew -global kvm-pit.lost_tick_policy=delay -no-hpet 
-no-shutdown -global PIIX4_PM.disable_s3=1 -global PIIX4_PM.disable_s4=1 
-boot strict=on -device ich9-usb-ehci1,id=usb,bus=pci.0,addr=0x8.0x7 
-device 
ich9-usb-uhci1,masterbus=usb.0,firstport=0,bus=pci.0,multifunction=on,addr=0x8 
-device 
ich9-usb-uhci2,masterbus=usb.0,firstport=2,bus=pci.0,addr=0x8.0x1 
-device 
ich9-usb-uhci3,masterbus=usb.0,firstport=4,bus=pci.0,addr=0x8.0x2 -drive 
file=/home/dzagorui/csr/csr_disk.qcow2,format=qcow2,if=none,id=drive-ide0-0-0 
-device 
ide-hd,bus=ide.0,unit=0,drive=drive-ide0-0-0,id=ide0-0-0,bootindex=1 
-netdev tap,fd=25,id=hostnet0 -device 
e1000,netdev=hostnet0,id=net0,mac=52:54:00:a9:4c:86,bus=pci.0,addr=0x3 
-netdev tap,fd=27,id=hostnet1 -device 
e1000,netdev=hostnet1,id=net1,mac=52:54:00:d4:7f:fa,bus=pci.0,addr=0x4 
-netdev tap,fd=28,id=hostnet2 -device 
e1000,netdev=hostnet2,id=net2,mac=52:54:00:d6:ec:f6,bus=pci.0,addr=0x5 
-netdev tap,fd=29,id=hostnet3 -device 
e1000,netdev=hostnet3,id=net3,mac=52:54:00:99:2b:46,bus=pci.0,addr=0xa 
-chardev 
socket,id=charserial0,host=127.0.0.1,port=4000,telnet,server,nowait 
-device isa-serial,chardev=charserial0,id=serial0 -chardev 
socket,id=charserial1,host=127.0.0.1,port=4001,telnet,server,nowait 
-device isa-serial,chardev=charserial1,id=serial1 -device 
virtio-balloon-pci,id=balloon0,bus=pci.0,addr=0x9 -msg timestamp=on


With this 100% reproduce:

dd if=/dev/zero of=/dev/ttyS1


And there is fix proposed by H. Peter Anvin in 2008

https://lkml.org/lkml/2008/2/7/485


So maybe it needs to be reworked?


Regards,

Denys

Re: [Qemu-devel] KVM serial8250: too much work for irq3

[Paolo Bonzini]

On 22/09/17 15:28, Denys Zagorui wrote:
> Hello
> 
> This is known issue, it was mentioned at first time about 9 years ago.
> 
> https://lkml.org/lkml/2008/2/5/401
> 
> https://bugzilla.redhat.com/show_bug.cgi?id=986761

The issue is simply that QEMU disregards the baud rate that the guest
programs into the serial port.  H. Peter Anvin's patch, which you
mention, is only a band aid that is geared specifically at avoiding
Linux's error.

It was fixed for real in commit 6936bfe514bbac7af4b24fad9ed9688b78b5be69
by implementing a baud rate timer (May 2008), but then undone in
fcfb4d6aae611d1f804d486d3c998000912c4c81 (March 2013) when the serial
port got flow control support.

Part of the reason may have been laziness, but also there are reasons to
keep it this way.  Basically, the baud rate timer was implemented also
in order to mitigate the problem with losing characters sent by the
guest on the serial console; and thanks to flow control, QEMU will
already slow down the transmit so that the other side never loses
characters.  And because looking at logs on a serial console bps is slow
and mildly annoying even at 115200 bps, it was decided that the "too
much work" error was the lesser evil.

Thanks,

Paolo

Re: [Qemu-devel] KVM serial8250: too much work for irq3

[hpa]

On January 10, 2019 3:44:54 PM PST, Paolo Bonzini <pbonzini@redhat.com> wrote:
>On 22/09/17 15:28, Denys Zagorui wrote:
>> Hello
>> 
>> This is known issue, it was mentioned at first time about 9 years
>ago.
>> 
>> https://lkml.org/lkml/2008/2/5/401
>> 
>> https://bugzilla.redhat.com/show_bug.cgi?id=986761
>
>The issue is simply that QEMU disregards the baud rate that the guest
>programs into the serial port.  H. Peter Anvin's patch, which you
>mention, is only a band aid that is geared specifically at avoiding
>Linux's error.
>
>It was fixed for real in commit
>6936bfe514bbac7af4b24fad9ed9688b78b5be69
>by implementing a baud rate timer (May 2008), but then undone in
>fcfb4d6aae611d1f804d486d3c998000912c4c81 (March 2013) when the serial
>port got flow control support.
>
>Part of the reason may have been laziness, but also there are reasons
>to
>keep it this way.  Basically, the baud rate timer was implemented also
>in order to mitigate the problem with losing characters sent by the
>guest on the serial console; and thanks to flow control, QEMU will
>already slow down the transmit so that the other side never loses
>characters.  And because looking at logs on a serial console bps is
>slow
>and mildly annoying even at 115200 bps, it was decided that the "too
>much work" error was the lesser evil.
>
>Thanks,
>
>Paolo

In the case of the lesser evil it would seem that the hack I did would be good enough.

A hardware high bandwidth UART simulating device would have to do the best same in some way.
-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.