Message-ID: <ee217f34-539f-2759-e4ac-5098e3923555@linux.ibm.com>
Date:   Thu, 16 Dec 2021 09:51:22 -0500
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.3.0
Subject: Re: [PATCH 23/32] KVM: s390: pci: handle refresh of PCI translations
Content-Language: en-US
To:     Niklas Schnelle <schnelle@linux.ibm.com>,
        Pierre Morel <pmorel@linux.ibm.com>, linux-s390@vger.kernel.org
Cc:     alex.williamson@redhat.com, cohuck@redhat.com,
        farman@linux.ibm.com, borntraeger@linux.ibm.com, hca@linux.ibm.com,
        gor@linux.ibm.com, gerald.schaefer@linux.ibm.com,
        agordeev@linux.ibm.com, frankja@linux.ibm.com, david@redhat.com,
        imbrenda@linux.ibm.com, vneethv@linux.ibm.com,
        oberpar@linux.ibm.com, freude@linux.ibm.com, thuth@redhat.com,
        pasic@linux.ibm.com, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <20211207205743.150299-1-mjrosato@linux.ibm.com>
 <20211207205743.150299-24-mjrosato@linux.ibm.com>
 <d2af697e-bf48-e78b-eed6-766f0790232f@linux.ibm.com>
 <a963388d-b13e-07c5-c256-c91671b3aa73@linux.ibm.com>
 <9a953a7938218afed246e93995d22ee7d09a81f3.camel@linux.ibm.com>
From:   Matthew Rosato <mjrosato@linux.ibm.com>
In-Reply-To: <9a953a7938218afed246e93995d22ee7d09a81f3.camel@linux.ibm.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 12/16/21 9:39 AM, Niklas Schnelle wrote:
> On Tue, 2021-12-14 at 12:54 -0500, Matthew Rosato wrote:
>> On 12/14/21 11:59 AM, Pierre Morel wrote:
>>>
>>> On 12/7/21 21:57, Matthew Rosato wrote:
>>>> Add a routine that will perform a shadow operation between a guest
>>>> and host IOAT.  A subsequent patch will invoke this in response to
>>>> an 04 RPCIT instruction intercept.
>>>>
>>>> Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
>>>> ---
>>>>    arch/s390/include/asm/kvm_pci.h |   1 +
>>>>    arch/s390/include/asm/pci_dma.h |   1 +
>>>>    arch/s390/kvm/pci.c             | 191 ++++++++++++++++++++++++++++++++
>>>>    arch/s390/kvm/pci.h             |   4 +-
>>>>    4 files changed, 196 insertions(+), 1 deletion(-)
>>>>
> ---8<---
>>>
>>>> +
>>>> +int kvm_s390_pci_refresh_trans(struct kvm_vcpu *vcpu, unsigned long req,
>>>> +                   unsigned long start, unsigned long size)
>>>> +{
>>>> +    struct zpci_dev *zdev;
>>>> +    u32 fh;
>>>> +    int rc;
>>>> +
>>>> +    /* If the device has a SHM bit on, let userspace take care of
>>>> this */
>>>> +    fh = req >> 32;
>>>> +    if ((fh & aift.mdd) != 0)
>>>> +        return -EOPNOTSUPP;
>>>
>>> I think you should make this check in the caller.
>>
>> OK
>>
>>>> +
>>>> +    /* Make sure this is a valid device associated with this guest */
>>>> +    zdev = get_zdev_by_fh(fh);
>>>> +    if (!zdev || !zdev->kzdev || zdev->kzdev->kvm != vcpu->kvm)
>>>> +        return -EINVAL;
>>>> +
>>>> +    /* Only proceed if the device is using the assist */
>>>> +    if (zdev->kzdev->ioat.head[0] == 0)
>>>> +        return -EOPNOTSUPP;
>>>
>>> Using the assist means using interpretation over using interception and
>>> legacy vfio-pci. right?
>>
>> Right - more specifically that the IOAT assist feature was never set via
>> the vfio feature ioctl, so we can't handle the RPCIT for this device and
>> so throw to userspace.
>>
>> The way the QEMU series is being implemented, a device using
>> interpretation will always have the IOAT feature set on.
>>
>>>> +
>>>> +    rc = dma_table_shadow(vcpu, zdev, start, size);
>>>> +    if (rc > 0)
>>>> +        rc = zpci_refresh_trans((u64) zdev->fh << 32, start, size);
>>>
>>> Here you lose the status reported by the hardware.
>>> You should directly use __rpcit(fn, addr, range, &status);
>>
>> OK, I can have a look at doing this.
>>
>> @Niklas thoughts on how you would want this exported.  Renamed to
>> zpci_rpcit or so?
> 
> Hmm with using __rpcit() directly we would lose the error reporting in
> s390dbf and this ist still kind of a RPCIT in the host. How about we
> add the status as an out parameter to zpci_refresh_trans()? But yes if

Another advantage of doing this would be that we then also keep the cc2 
retry logic in zpci_refresh_trans(), which would be nice.

However we do still lose the returned CC value from the instruction. 
But I think we can infer a CC1 from a nonzero status and a CC3 from a 
zero status so maybe this is OK too.

I think I will add the status parm to zpci_refresh_trans().

FWIW, I do also think it is likely we will end up with a s390dbf for 
kvm-pci at some point after this initial series.


> you prefer to use __rpcit() directly I would rename it to zpci_rpcit().
> 

>>
> 
> ---8<---
>