Re: [PULL v3 05/19] hw/acpi/ich9: Set ACPI PCI hot-plug as default on Q35

[Laurent Vivier <lvivier@redhat.com>]

On 21/07/2021 18:37, Michael S. Tsirkin wrote:
> On Wed, Jul 21, 2021 at 06:27:33PM +0200, Igor Mammedov wrote:
>> On Wed, 21 Jul 2021 12:09:01 -0400
>> "Michael S. Tsirkin" <mst@redhat.com> wrote:
>>
>>> On Wed, Jul 21, 2021 at 05:49:16PM +0200, Laurent Vivier wrote:
>>>> On 21/07/2021 16:59, Igor Mammedov wrote:  
>>>>> On Tue, 20 Jul 2021 14:56:06 +0200
>>>>> Laurent Vivier <lvivier@redhat.com> wrote:
>>>>>   
>>>>>> On 20/07/2021 13:38, Laurent Vivier wrote:  
>>>>>>> On 16/07/2021 17:15, Michael S. Tsirkin wrote:    
>>>>>>>> From: Julia Suvorova <jusual@redhat.com>
>>>>>>>>
>>>>>>>> Q35 has three different types of PCI devices hot-plug: PCIe Native,
>>>>>>>> SHPC Native and ACPI hot-plug. This patch changes the default choice
>>>>>>>> for cold-plugged bridges from PCIe Native to ACPI Hot-plug with
>>>>>>>> ability to use SHPC and PCIe Native for hot-plugged bridges.
>>>>>>>>
>>>>>>>> This is a list of the PCIe Native hot-plug issues that led to this
>>>>>>>> change:
>>>>>>>>     * no racy behavior during boot (see 110c477c2ed)
>>>>>>>>     * no delay during deleting - after the actual power off software
>>>>>>>>       must wait at least 1 second before indicating about it. This case
>>>>>>>>       is quite important for users, it even has its own bug:
>>>>>>>>           https://bugzilla.redhat.com/show_bug.cgi?id=1594168
>>>>>>>>     * no timer-based behavior - in addition to the previous example,
>>>>>>>>       the attention button has a 5-second waiting period, during which
>>>>>>>>       the operation can be canceled with a second press. While this
>>>>>>>>       looks fine for manual button control, automation will result in
>>>>>>>>       the need to queue or drop events, and the software receiving
>>>>>>>>       events in all sort of unspecified combinations of attention/power
>>>>>>>>       indicator states, which is racy and uppredictable.
>>>>>>>>     * fixes:
>>>>>>>>         * https://bugzilla.redhat.com/show_bug.cgi?id=1752465
>>>>>>>>         * https://bugzilla.redhat.com/show_bug.cgi?id=1690256
>>>>>>>>
>>>>>>>> To return to PCIe Native hot-plug:
>>>>>>>>     -global ICH9-LPC.acpi-pci-hotplug-with-bridge-support=off
>>>>>>>>
>>>>>>>> Known issue: older linux guests need the following flag
>>>>>>>> to allow hotplugged pci express devices to use io:
>>>>>>>>         -device pcie-root-port,io-reserve=4096.
>>>>>>>> io is unusual for pci express so this seems minor.
>>>>>>>> We'll fix this by a follow up patch.
>>>>>>>>
>>>>>>>> Signed-off-by: Julia Suvorova <jusual@redhat.com>
>>>>>>>> Reviewed-by: Igor Mammedov <imammedo@redhat.com>
>>>>>>>> Message-Id: <20210713004205.775386-6-jusual@redhat.com>
>>>>>>>> Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
>>>>>>>> Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
>>>>>>>> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
>>>>>>>> ---
>>>>>>>>  hw/acpi/ich9.c | 2 +-
>>>>>>>>  hw/i386/pc.c   | 1 +
>>>>>>>>  2 files changed, 2 insertions(+), 1 deletion(-)
>>>>>>>>
>>>>>>>> diff --git a/hw/acpi/ich9.c b/hw/acpi/ich9.c
>>>>>>>> index 2f4eb453ac..778e27b659 100644
>>>>>>>> --- a/hw/acpi/ich9.c
>>>>>>>> +++ b/hw/acpi/ich9.c
>>>>>>>> @@ -427,7 +427,7 @@ void ich9_pm_add_properties(Object *obj, ICH9LPCPMRegs *pm)
>>>>>>>>      pm->disable_s3 = 0;
>>>>>>>>      pm->disable_s4 = 0;
>>>>>>>>      pm->s4_val = 2;
>>>>>>>> -    pm->use_acpi_hotplug_bridge = false;
>>>>>>>> +    pm->use_acpi_hotplug_bridge = true;
>>>>>>>>  
>>>>>>>>      object_property_add_uint32_ptr(obj, ACPI_PM_PROP_PM_IO_BASE,
>>>>>>>>                                     &pm->pm_io_base, OBJ_PROP_FLAG_READ);
>>>>>>>> diff --git a/hw/i386/pc.c b/hw/i386/pc.c
>>>>>>>> index aa79c5e0e6..f4c7a78362 100644
>>>>>>>> --- a/hw/i386/pc.c
>>>>>>>> +++ b/hw/i386/pc.c
>>>>>>>> @@ -99,6 +99,7 @@ GlobalProperty pc_compat_6_0[] = {
>>>>>>>>      { "qemu64" "-" TYPE_X86_CPU, "model", "6" },
>>>>>>>>      { "qemu64" "-" TYPE_X86_CPU, "stepping", "3" },
>>>>>>>>      { TYPE_X86_CPU, "x-vendor-cpuid-only", "off" },
>>>>>>>> +    { "ICH9-LPC", "acpi-pci-hotplug-with-bridge-support", "off" },
>>>>>>>>  };
>>>>>>>>  const size_t pc_compat_6_0_len = G_N_ELEMENTS(pc_compat_6_0);
>>>>>>>>  
>>>>>>>>    
>>>>>>>
>>>>>>> There is an issue with this patch.
>>>>>>>
>>>>>>> When I try to unplug a VFIO device I have the following error and the device is not unplugged:
>>>>>>>
>>>>>>> (qemu) device_del hostdev0
>>>>>>>
>>>>>>> [   34.116714] ACPI BIOS Error (bug): Could not resolve symbol [^S0B.PCNT], AE_NOT_FOUND
>>>>>>> (20201113/psargs-330)
>>>>>>> [   34.117987] ACPI Error: Aborting method \_SB.PCI0.PCNT due to previous error
>>>>>>> (AE_NOT_FOUND) (20201113/psparse-531)
>>>>>>> [   34.119318] ACPI Error: Aborting method \_GPE._E01 due to previous error (AE_NOT_FOUND)
>>>>>>> (20201113/psparse-531)
>>>>>>> [   34.120600] ACPI Error: AE_NOT_FOUND, while evaluating GPE method [_E01]
>>>>>>> (20201113/evgpe-515)
>>>>>>>
>>>>>>> We can see device is not unplugged (03:00.0)
>>>>>>>
>>>>>>> # lspci -v -s 03:00.0
>>>>>>> 03:00.0 Ethernet controller: Intel Corporation Ethernet Virtual Function 700 Series (rev 02)
>>>>>>> 	Subsystem: Intel Corporation Device 0000
>>>>>>> 	Flags: bus master, fast devsel, latency 0
>>>>>>> 	Memory at fe800000 (64-bit, prefetchable) [size=64K]
>>>>>>> 	Memory at fe810000 (64-bit, prefetchable) [size=16K]
>>>>>>> 	Capabilities: [70] MSI-X: Enable+ Count=5 Masked-
>>>>>>> 	Capabilities: [a0] Express Endpoint, MSI 00
>>>>>>> 	Capabilities: [100] Advanced Error Reporting
>>>>>>> 	Capabilities: [1a0] Transaction Processing Hints
>>>>>>> 	Capabilities: [1d0] Access Control Services
>>>>>>> 	Kernel driver in use: iavf
>>>>>>> 	Kernel modules: iavf
>>>>>>>
>>>>>>> My guest kernel is from RHEL 8.5 (4.18.0-310.el8.x86_64) and my command line is:
>>>>>>>
>>>>>>> $QEMU \
>>>>>>> -L .../pc-bios \
>>>>>>> -nodefaults \
>>>>>>> -nographic \
>>>>>>> -machine q35 \
>>>>>>> -device pcie-root-port,id=pcie-root-port-0,multifunction=on,bus=pcie.0,addr=0x1,chassis=1 \
>>>>>>> -device pcie-pci-bridge,id=pcie-pci-bridge-0,addr=0x0,bus=pcie-root-port-0  \
>>>>>>> -device pcie-root-port,id=pcie-root-port-1,port=0x1,addr=0x1.0x1,bus=pcie.0,chassis=2 \
>>>>>>> -device pcie-root-port,id=pcie-root-port-2,port=0x2,addr=0x1.0x2,bus=pcie.0,chassis=3 \
>>>>>>> -device pcie-root-port,id=pcie-root-port-3,port=0x3,addr=0x1.0x3,bus=pcie.0,chassis=4 \
>>>>>>> -device
>>>>>>> pcie-root-port,id=pcie_extra_root_port_0,multifunction=on,bus=pcie.0,addr=0x3,chassis=5 \
>>>>>>> -nodefaults \
>>>>>>> -m 4066  \
>>>>>>> -smp 4 \
>>>>>>> -device virtio-scsi-pci,id=virtio_scsi_pci0,bus=pcie-root-port-2,addr=0x0 \
>>>>>>> -blockdev
>>>>>>> node-name=file_image1,driver=file,auto-read-only=on,discard=unmap,aio=threads,filename=$IMAGE,cache.direct=on,cache.no-fl\
>>>>>>> -blockdev
>>>>>>> node-name=drive_image1,driver=qcow2,read-only=off,cache.direct=on,cache.no-flush=off,file=file_image1
>>>>>>> \
>>>>>>> -device scsi-hd,id=image1,drive=drive_image1,write-cache=on \
>>>>>>> -enable-kvm \
>>>>>>> -serial mon:stdio \
>>>>>>> -device vfio-pci,host=04:02.0,bus=pcie-root-port-1,addr=0x0,id=hostdev0
>>>>>>>
>>>>>>> PCI 04:02.0 is:
>>>>>>>
>>>>>>> $ lspci -v -s 04:02.0
>>>>>>> 04:02.0 Ethernet controller: Intel Corporation Ethernet Virtual Function 700 Series (rev 02)
>>>>>>> 	Subsystem: Intel Corporation Device 0000
>>>>>>> 	Flags: fast devsel, NUMA node 0, IOMMU group 53
>>>>>>> 	Memory at 92400000 (64-bit, prefetchable) [virtual] [size=64K]
>>>>>>> 	Memory at 92910000 (64-bit, prefetchable) [virtual] [size=16K]
>>>>>>> 	Capabilities: <access denied>
>>>>>>> 	Kernel driver in use: vfio-pci
>>>>>>> 	Kernel modules: iavf
>>>>>>>
>>>>>>> Any idea?    
>>>>>>
>>>>>> It also happens with non-VFIO device like e1000e:
>>>>>>
>>>>>> ...
>>>>>> -device e1000e,bus=pcie-root-port-1,addr=0x0,id=hostdev0 \  
>>>>>                      ^^^^^^^^^^^^^
>>>>> ACPI hotplug operates on slot level, so functions greater than 0 are not considered,
>>>>> hence unexpected ACPI error. For above CLI, setting 'addr' on root-ports to dedicated slots
>>>>> should fix issue.
>>>>>   
>>>>
>>>> Thank you for your answer.
>>>>
>>>> It works well with something like this:
>>>>
>>>> ...
>>>> -device pcie-root-port,id=pcie-root-port-0,addr=0x1,bus=pcie.0,chassis=1 \
>>>> -device pcie-root-port,id=pcie-root-port-1,addr=0x2,bus=pcie.0,chassis=2 \
>>>> -device pcie-root-port,id=pcie-root-port-2,addr=0x3,bus=pcie.0,chassis=3 \
>>>> -device pcie-root-port,id=pcie-root-port-3,addr=0x4,bus=pcie.0,chassis=4 \
>>>> -device e1000e,mac=52:54:00:12:34:56,id=hostdev0,bus=pcie-root-port-1 \
>>>> ...
>>>>
>>>> Is this what you meant?
>> yep
>>
>>>>
>>>> On an other hand, the previous configuration worked well before this patch, can we see
>>>> that as a regression?
>>
>> Maybe for 6.1 we should flip default back to native (revert 17858a16950860),
>> until we sort out multifunction issues.
> 
> Revert had advantages and disadvantages as usual. Let's see what the fix
> is, then we can decide.

This patch breaks also virtio-net failover when the migration is canceled: the unplugged
card is not plugged back.

Thanks,
Laurent