mbox series

[v2,00/18] s390x: flic rework, tcg flic support and STSI

Message ID 20180117174047.6382-1-david@redhat.com
Headers show
Series s390x: flic rework, tcg flic support and STSI | expand

Message

David Hildenbrand Jan. 17, 2018, 5:40 p.m. UTC
This patch series implements floating interrupt support for TCG and fixes
STSI so we can remove warnings related to s390x SMP and MTTCG.

KVM code has to be touched in order to factor out the injection routines
into the flic ("s390x/flic: factor out injection of floating interrupts").
Basic testing didn't reveal any problems so far.

With this series I am now able to run fedora 26/27 and Ubuntu 17.10+ with
16 VCPUs (MTTCG) on a 8CPU host, doing a make -j16 in the guest. I got
nasty stalls in the guest beforehand.

Now also included: More flic related caching to avoid recurring
conversions.

Based on: https://github.com/cohuck/qemu.git s390x-next
Available at: https://github.com/davidhildenbrand/qemu.git s390x-queue

v1 -> v2:
- Added due to discussions:
 - "s390x/flic: no need to call s390_io_interrupt() from flic"
 - "s390x/tcg: cache the qemu flic in a central function"
 - "s390x/kvm: cache the kvm flic in a central function"
 - "s390x/flic: cache the common flic class in a central function"
- "s390x/flic: simplify flic initialization"
 -  no need to check for tcg_enabled()
- "s390x/tcg: STSI overhaul"
 - rephrased the comment
- "s390x/tcg: implement TEST PENDING INTERRUPTION"
 - renamed local variable
- "s390x: fix size + content of STSI blocks"
 - added typedefs whiel touching it. Thomas, therefore not added your r-b.
- Minor spelling errors fixed
- Dropped a couple of unneeded OBJECT() conversion


David Hildenbrand (18):
  s390x/tcg: deliver multiple interrupts in a row
  s390x/flic: simplify flic initialization
  s390x/tcg: simplify lookup of flic
  s390x/tcg: simplify machine check handling
  s390x/flic: factor out injection of floating interrupts
  s390x/flic: no need to call s390_io_interrupt() from flic
  s390x/tcg: tolerate wrong wakeups due to floating interrupts
  s390x/flic: make floating interrupts on TCG actually floating
  s390x/tcg: implement TEST PENDING INTERRUPTION
  s390x/flic: implement qemu_s390_clear_io_flic()
  s390x/flic: optimize CPU wakeup for TCG
  s390x: fix size + content of STSI blocks
  s390x/tcg: STSI overhaul
  s390x/tcg: remove SMP warning
  configure: s390x supports mttcg now
  s390x/tcg: cache the qemu flic in a central function
  s390x/kvm: cache the kvm flic in a central function
  s390x/flic: cache the common flic class in a central function

 configure                    |   1 +
 hw/intc/s390_flic.c          | 257 +++++++++++++++++++++++++++++++++++++++--
 hw/intc/s390_flic_kvm.c      |  84 +++++++++++---
 hw/s390x/css.c               |  10 +-
 hw/s390x/s390-virtio-ccw.c   |   4 -
 hw/s390x/virtio-ccw.c        |   4 +-
 include/hw/s390x/s390_flic.h |  57 ++++++++--
 target/s390x/cpu.c           |  10 --
 target/s390x/cpu.h           |  99 ++++++++--------
 target/s390x/excp_helper.c   | 147 +++++++++---------------
 target/s390x/helper.h        |   1 +
 target/s390x/insn-data.def   |   1 +
 target/s390x/internal.h      |   5 -
 target/s390x/interrupt.c     | 100 ++++------------
 target/s390x/kvm-stub.c      |  13 ---
 target/s390x/kvm.c           |  70 +-----------
 target/s390x/kvm_s390x.h     |  10 +-
 target/s390x/misc_helper.c   | 266 +++++++++++++++++++++++++++----------------
 target/s390x/translate.c     |   8 ++
 19 files changed, 680 insertions(+), 467 deletions(-)

Comments

Cornelia Huck Jan. 25, 2018, 10:29 a.m. UTC | #1
On Wed, 17 Jan 2018 18:40:29 +0100
David Hildenbrand <david@redhat.com> wrote:

> This patch series implements floating interrupt support for TCG and fixes
> STSI so we can remove warnings related to s390x SMP and MTTCG.
> 
> KVM code has to be touched in order to factor out the injection routines
> into the flic ("s390x/flic: factor out injection of floating interrupts").
> Basic testing didn't reveal any problems so far.
> 
> With this series I am now able to run fedora 26/27 and Ubuntu 17.10+ with
> 16 VCPUs (MTTCG) on a 8CPU host, doing a make -j16 in the guest. I got
> nasty stalls in the guest beforehand.
> 
> Now also included: More flic related caching to avoid recurring
> conversions.

Basically, seems good to me. Will play with it a bit more.

I'd also love some reviews/acks.
Cornelia Huck Jan. 25, 2018, 12:42 p.m. UTC | #2
On Wed, 17 Jan 2018 18:40:29 +0100
David Hildenbrand <david@redhat.com> wrote:

> This patch series implements floating interrupt support for TCG and fixes
> STSI so we can remove warnings related to s390x SMP and MTTCG.
> 
> KVM code has to be touched in order to factor out the injection routines
> into the flic ("s390x/flic: factor out injection of floating interrupts").
> Basic testing didn't reveal any problems so far.
> 
> With this series I am now able to run fedora 26/27 and Ubuntu 17.10+ with
> 16 VCPUs (MTTCG) on a 8CPU host, doing a make -j16 in the guest. I got
> nasty stalls in the guest beforehand.
> 
> Now also included: More flic related caching to avoid recurring
> conversions.

Hm...

(qemu) device_add virtio-net-ccw,id=xxx
(qemu) **
ERROR:/home/cohuck/git/qemu/target/s390x/excp_helper.c:342:do_mchk_interrupt: assertion failed: (env->cregs[14] & CR0_SERVICE_SC)
Aborted (core dumped)
Cornelia Huck Jan. 25, 2018, 12:47 p.m. UTC | #3
On Thu, 25 Jan 2018 13:42:41 +0100
Cornelia Huck <cohuck@redhat.com> wrote:

> On Wed, 17 Jan 2018 18:40:29 +0100
> David Hildenbrand <david@redhat.com> wrote:
> 
> > This patch series implements floating interrupt support for TCG and fixes
> > STSI so we can remove warnings related to s390x SMP and MTTCG.
> > 
> > KVM code has to be touched in order to factor out the injection routines
> > into the flic ("s390x/flic: factor out injection of floating interrupts").
> > Basic testing didn't reveal any problems so far.
> > 
> > With this series I am now able to run fedora 26/27 and Ubuntu 17.10+ with
> > 16 VCPUs (MTTCG) on a 8CPU host, doing a make -j16 in the guest. I got
> > nasty stalls in the guest beforehand.
> > 
> > Now also included: More flic related caching to avoid recurring
> > conversions.  
> 
> Hm...
> 
> (qemu) device_add virtio-net-ccw,id=xxx
> (qemu) **
> ERROR:/home/cohuck/git/qemu/target/s390x/excp_helper.c:342:do_mchk_interrupt: assertion failed: (env->cregs[14] & CR0_SERVICE_SC)
> Aborted (core dumped)

Works much better if you check for CR14_CHANNEL_REPORT_SC instead :)
David Hildenbrand Jan. 25, 2018, 1:11 p.m. UTC | #4
On 25.01.2018 13:47, Cornelia Huck wrote:
> On Thu, 25 Jan 2018 13:42:41 +0100
> Cornelia Huck <cohuck@redhat.com> wrote:
> 
>> On Wed, 17 Jan 2018 18:40:29 +0100
>> David Hildenbrand <david@redhat.com> wrote:
>>
>>> This patch series implements floating interrupt support for TCG and fixes
>>> STSI so we can remove warnings related to s390x SMP and MTTCG.
>>>
>>> KVM code has to be touched in order to factor out the injection routines
>>> into the flic ("s390x/flic: factor out injection of floating interrupts").
>>> Basic testing didn't reveal any problems so far.
>>>
>>> With this series I am now able to run fedora 26/27 and Ubuntu 17.10+ with
>>> 16 VCPUs (MTTCG) on a 8CPU host, doing a make -j16 in the guest. I got
>>> nasty stalls in the guest beforehand.
>>>
>>> Now also included: More flic related caching to avoid recurring
>>> conversions.  
>>
>> Hm...
>>
>> (qemu) device_add virtio-net-ccw,id=xxx
>> (qemu) **
>> ERROR:/home/cohuck/git/qemu/target/s390x/excp_helper.c:342:do_mchk_interrupt: assertion failed: (env->cregs[14] & CR0_SERVICE_SC)
>> Aborted (core dumped)
> 
> Works much better if you check for CR14_CHANNEL_REPORT_SC instead :)
> 

Thanks for testing and spotting the wrong assert!
Thomas Huth Jan. 25, 2018, 1:13 p.m. UTC | #5
On 25.01.2018 14:11, David Hildenbrand wrote:
> On 25.01.2018 13:47, Cornelia Huck wrote:
>> On Thu, 25 Jan 2018 13:42:41 +0100
>> Cornelia Huck <cohuck@redhat.com> wrote:
>>
>>> On Wed, 17 Jan 2018 18:40:29 +0100
>>> David Hildenbrand <david@redhat.com> wrote:
>>>
>>>> This patch series implements floating interrupt support for TCG and fixes
>>>> STSI so we can remove warnings related to s390x SMP and MTTCG.
>>>>
>>>> KVM code has to be touched in order to factor out the injection routines
>>>> into the flic ("s390x/flic: factor out injection of floating interrupts").
>>>> Basic testing didn't reveal any problems so far.
>>>>
>>>> With this series I am now able to run fedora 26/27 and Ubuntu 17.10+ with
>>>> 16 VCPUs (MTTCG) on a 8CPU host, doing a make -j16 in the guest. I got
>>>> nasty stalls in the guest beforehand.
>>>>
>>>> Now also included: More flic related caching to avoid recurring
>>>> conversions.  
>>>
>>> Hm...
>>>
>>> (qemu) device_add virtio-net-ccw,id=xxx
>>> (qemu) **
>>> ERROR:/home/cohuck/git/qemu/target/s390x/excp_helper.c:342:do_mchk_interrupt: assertion failed: (env->cregs[14] & CR0_SERVICE_SC)
>>> Aborted (core dumped)
>>
>> Works much better if you check for CR14_CHANNEL_REPORT_SC instead :)
> 
> Thanks for testing and spotting the wrong assert!
We definitely need to get tests/virtio-net-test.c running on s390x, too ;-)

 Thomas