Re: virtio,iommu_platform=on

[Alexey Kardashevskiy]

On 14/11/2019 19:42, Michael S. Tsirkin wrote:
> On Thu, Nov 14, 2019 at 09:58:47AM +1100, Alexey Kardashevskiy wrote:
>>
>>
>> On 13/11/2019 21:00, Michael S. Tsirkin wrote:
>>> On Wed, Nov 13, 2019 at 03:44:28PM +1100, Alexey Kardashevskiy wrote:
>>>>
>>>>
>>>> On 12/11/2019 18:08, Michael S. Tsirkin wrote:
>>>>> On Tue, Nov 12, 2019 at 02:53:49PM +1100, Alexey Kardashevskiy wrote:
>>>>>> Hi!
>>>>>>
>>>>>> I am enabling IOMMU for virtio in the pseries firmware (SLOF) and seeing
>>>>>> problems, one of them is SLOF does SCSI bus scan, then it stops the
>>>>>> virtio-scsi by clearing MMIO|IO|BUSMASTER from PCI_COMMAND (as SLOF
>>>>>> stopped using the devices) and when this happens, I see unassigned
>>>>>> memory access (see below) which happens because disabling busmaster
>>>>>> disables IOMMU and QEMU cannot access the rings to do some shutdown. And
>>>>>> when this happens, the device does not come back even if SLOF re-enables 
>>>>>> it.
>>>>>
>>>>> In fact clearing bus master should disable ring access even
>>>>> without the IOMMU.
>>>>> Once you do this you should not wait for rings to be processed,
>>>>> it is safe to assume they won't be touched again and just
>>>>> free up any buffers that have not been used.
>>>>>
>>>>> Why don't you see this without IOMMU?
>>>>
>>>> Because without IOMMU, virtio can always access rings, it does not need
>>>> bus master address space for that.
>>>
>>> Right and that's a bug in virtio scsi. E.g. virtio net checks
>>> bus mastering before each access.
>>
>> You have to be specific - virtio scsi in the guest or in QEMU?
> 
> If a device accesses memory with bus master on, it's buggy.
> 
>>
>>> Which is all well and good, but we can't just break the world
>>> so I guess we first need to fix SLOF, and then add
>>> a compat property. And maybe keep it broken for
>>> legacy ...
>>>
>>>>
>>>>> It's a bug I think, probably there to work around buggy guests.
>>>>>
>>>>> So pls fix this in SLOF and then hopefully we can drop the
>>>>> work arounds and have clearing bus master actually block DMA.
>>>>
>>>>
>>>> Laszlo suggested writing 0 to the status but this does not seem helping,
>>>> with both ioeventfd=true/false. It looks like setting/clearing busmaster
>>>> bit confused memory region caches in QEMU's virtio. I am confused which
>>>> direction to keep digging to, any suggestions? Thanks,
>>>>
>>>
>>> to clarify you reset after setting bus master? right?
>>
>>
>> I was talking about clearing the bus master, and where I call that
>> virtio reset does not matter. Thanks,
>>
>>
> 
> so
> 
> bus master =0
> reset
> bus master =1
> 
> device does not recover?


No, it does not. With my recent changes, it seems that only virtio-block
cannot recover; virtio-net and virtio-scsi are fine. Thanks,


> 
> 
>>>
>>>
>>>>
>>>>>
>>>>>> Hacking SLOF to not clear BUSMASTER makes virtio-scsi work but it is
>>>>>> hardly a right fix.
>>>>>>
>>>>>> Is this something expected? Thanks,
>>>>>>
>>>>>>
>>>>>> Here is the exact command line:
>>>>>>
>>>>>> /home/aik/pbuild/qemu-garrison2-ppc64/ppc64-softmmu/qemu-system-ppc64 \
>>>>>>
>>>>>> -nodefaults \
>>>>>>
>>>>>> -chardev stdio,id=STDIO0,signal=off,mux=on \
>>>>>>
>>>>>> -device spapr-vty,id=svty0,reg=0x71000110,chardev=STDIO0 \
>>>>>>
>>>>>> -mon id=MON0,chardev=STDIO0,mode=readline \
>>>>>>
>>>>>> -nographic \
>>>>>>
>>>>>> -vga none \
>>>>>>
>>>>>> -enable-kvm \
>>>>>> -m 2G \
>>>>>>
>>>>>> -device
>>>>>> virtio-scsi-pci,id=vscsi0,iommu_platform=on,disable-modern=off,disable-legacy=on
>>>>>> \
>>>>>> -drive id=DRIVE0,if=none,file=img/u1804-64le.qcow2,format=qcow2 \
>>>>>>
>>>>>> -device scsi-disk,id=scsi-disk0,drive=DRIVE0 \
>>>>>>
>>>>>> -snapshot \
>>>>>>
>>>>>> -smp 1 \
>>>>>>
>>>>>> -machine pseries \
>>>>>>
>>>>>> -L /home/aik/t/qemu-ppc64-bios/ \
>>>>>>
>>>>>> -trace events=qemu_trace_events \
>>>>>>
>>>>>> -d guest_errors \
>>>>>>
>>>>>> -chardev socket,id=SOCKET0,server,nowait,path=qemu.mon.ssh59518 \
>>>>>>
>>>>>> -mon chardev=SOCKET0,mode=control
>>>>>>
>>>>>>
>>>>>>
>>>>>> Here is the backtrace:
>>>>>>
>>>>>> Thread 5 "qemu-system-ppc" hit Breakpoint 8, unassigned_mem_accepts
>>>>>> (opaque=0x0, addr=0x5802, size=0x2, is_write=0x0, attrs=...) at /home/
>>>>>> aik/p/qemu/memory.c:1275
>>>>>> 1275        return false;
>>>>>> #0  unassigned_mem_accepts (opaque=0x0, addr=0x5802, size=0x2,
>>>>>> is_write=0x0, attrs=...) at /home/aik/p/qemu/memory.c:1275
>>>>>> #1  0x00000000100a8ac8 in memory_region_access_valid (mr=0x1105c230
>>>>>> <io_mem_unassigned>, addr=0x5802, size=0x2, is_write=0x0, attrs=...) at
>>>>>> /home/aik/p/qemu/memory.c:1377
>>>>>> #2  0x00000000100a8c88 in memory_region_dispatch_read (mr=0x1105c230
>>>>>> <io_mem_unassigned>, addr=0x5802, pval=0x7ffff550d410, op=MO_16,
>>>>>> attrs=...) at /home/aik/p/qemu/memory.c:1418
>>>>>> #3  0x000000001001cad4 in address_space_lduw_internal_cached_slow
>>>>>> (cache=0x7fff68036fa0, addr=0x2, attrs=..., result=0x0,
>>>>>> endian=DEVICE_LITTLE_ENDIAN) at /home/aik/p/qemu/memory_ldst.inc.c:211
>>>>>> #4  0x000000001001cc84 in address_space_lduw_le_cached_slow
>>>>>> (cache=0x7fff68036fa0, addr=0x2, attrs=..., result=0x0) at
>>>>>> /home/aik/p/qemu/memory_ldst.inc.c:249
>>>>>> #5  0x000000001019bd80 in address_space_lduw_le_cached
>>>>>> (cache=0x7fff68036fa0, addr=0x2, attrs=..., result=0x0) at
>>>>>> /home/aik/p/qemu/include/exec/memory_ldst_cached.inc.h:56
>>>>>> #6  0x000000001019c10c in lduw_le_phys_cached (cache=0x7fff68036fa0,
>>>>>> addr=0x2) at /home/aik/p/qemu/include/exec/memory_ldst_phys.inc.h:91
>>>>>> #7  0x000000001019d86c in virtio_lduw_phys_cached (vdev=0x118b9110,
>>>>>> cache=0x7fff68036fa0, pa=0x2) at
>>>>>> /home/aik/p/qemu/include/hw/virtio/virtio-access.h:166
>>>>>> #8  0x000000001019e648 in vring_avail_idx (vq=0x118c2720) at
>>>>>> /home/aik/p/qemu/hw/virtio/virtio.c:302
>>>>>> #9  0x000000001019f5bc in virtio_queue_split_empty (vq=0x118c2720) at
>>>>>> /home/aik/p/qemu/hw/virtio/virtio.c:581
>>>>>> #10 0x000000001019f838 in virtio_queue_empty (vq=0x118c2720) at
>>>>>> /home/aik/p/qemu/hw/virtio/virtio.c:612
>>>>>> #11 0x00000000101a8fa8 in virtio_queue_host_notifier_aio_poll
>>>>>> (opaque=0x118c2798) at /home/aik/p/qemu/hw/virtio/virtio.c:3389
>>>>>> #12 0x000000001092679c in run_poll_handlers_once (ctx=0x11212e40,
>>>>>> timeout=0x7ffff550d7d8) at /home/aik/p/qemu/util/aio-posix.c:520
>>>>>> #13 0x0000000010926aec in try_poll_mode (ctx=0x11212e40,
>>>>>> timeout=0x7ffff550d7d8) at /home/aik/p/qemu/util/aio-posix.c:607
>>>>>> #14 0x0000000010926c2c in aio_poll (ctx=0x11212e40, blocking=0x1) at
>>>>>> /home/aik/p/qemu/util/aio-posix.c:639
>>>>>> #15 0x000000001091fe0c in aio_wait_bh_oneshot (ctx=0x11212e40,
>>>>>> cb=0x1016f35c <virtio_scsi_dataplane_stop_bh>, opaque=0x118b9110) at
>>>>>> /home/aik/p/qemu/util/aio-wait.c:71
>>>>>> #16 0x000000001016fa60 in virtio_scsi_dataplane_stop (vdev=0x118b9110)
>>>>>> at /home/aik/p/qemu/hw/scsi/virtio-scsi-dataplane.c:211
>>>>>> #17 0x0000000010684740 in virtio_bus_stop_ioeventfd (bus=0x118b9098) at
>>>>>> /home/aik/p/qemu/hw/virtio/virtio-bus.c:245
>>>>>> #18 0x0000000010688290 in virtio_pci_stop_ioeventfd (proxy=0x118b0fa0)
>>>>>> at /home/aik/p/qemu/hw/virtio/virtio-pci.c:292
>>>>>> #19 0x00000000106891e8 in virtio_write_config (pci_dev=0x118b0fa0,
>>>>>> address=0x4, val=0x100100, len=0x4) at
>>>>>> /home/aik/p/qemu/hw/virtio/virtio-pci.c:613
>>>>>> #20 0x00000000105b7228 in pci_host_config_write_common
>>>>>> (pci_dev=0x118b0fa0, addr=0x4, limit=0x100, val=0x100100, len=0x4) at
>>>>>> /home/aik/p/qemu/hw/pci/pci_host.c:81
>>>>>> #21 0x00000000101f7bdc in finish_write_pci_config (spapr=0x11217200,
>>>>>> buid=0x800000020000000, addr=0x4, size=0x4, val=0x100100,
>>>>>> rets=0x7e7533e0) at /home/aik/p/qemu/hw/ppc/spapr_pci.c:192
>>>>>> #22 0x00000000101f7cec in rtas_ibm_write_pci_config (cpu=0x11540df0,
>>>>>> spapr=0x11217200, token=0x2017, nargs=0x5, args=0x7e7533cc, nret=0x1,
>>>>>> rets=0x7e7533e0) at /home/aik/p/qemu/hw/ppc/spapr_pci.c:216
>>>>>> #23 0x00000000101f5860 in spapr_rtas_call (cpu=0x11540df0,
>>>>>> spapr=0x11217200, token=0x2017, nargs=0x5, args=0x7e7533cc, nret=0x1,
>>>>>> rets=0x7e7533e0) at /home/aik/p/qemu/hw/ppc/spapr_rtas.c:416
>>>>>> #24 0x00000000101ee214 in h_rtas (cpu=0x11540df0, spapr=0x11217200,
>>>>>> opcode=0xf000, args=0x7ffff4cf0030) at
>>>>>> /home/aik/p/qemu/hw/ppc/spapr_hcall.c:1214
>>>>>> #25 0x00000000101f0524 in spapr_hypercall (cpu=0x11540df0,
>>>>>> opcode=0xf000, args=0x7ffff4cf0030) at
>>>>>> /home/aik/p/qemu/hw/ppc/spapr_hcall.c:2014
>>>>>> #26 0x000000001033bff0 in kvm_arch_handle_exit (cs=0x11540df0,
>>>>>> run=0x7ffff4cf0000) at /home/aik/p/qemu/target/ppc/kvm.c:1684
>>>>>> #27 0x00000000100cc7c8 in kvm_cpu_exec (cpu=0x11540df0) at
>>>>>> /home/aik/p/qemu/accel/kvm/kvm-all.c:2391
>>>>>> #28 0x000000001008edf8 in qemu_kvm_cpu_thread_fn (arg=0x11540df0) at
>>>>>> /home/aik/p/qemu/cpus.c:1318
>>>>>> #29 0x000000001092c704 in qemu_thread_start (args=0x11588d90) at
>>>>>> /home/aik/p/qemu/util/qemu-thread-posix.c:519
>>>>>> #30 0x00007ffff7b58070 in start_thread (arg=0x7ffff550ebf0) at
>>>>>> pthread_create.c:335
>>>>>> #31 0x00007ffff7aa3a70 in clone () at
>>>>>> ../sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S:96
>>>>>> (gdb)
>>>>>>
>>>>>> -- 
>>>>>> Alexey
>>>>>
>>>>
>>>> -- 
>>>> Alexey
>>>
>>
>> -- 
>> Alexey
> 

-- 
Alexey